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 struct xattr_handler *ocfs2_xattr_handler(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;
 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_entries(struct inode *inode,
 456                                     struct ocfs2_xattr_header *header,
 457                                     char *buffer, size_t buffer_size)
 458 {
 459         size_t rest = buffer_size;
 460         int i;
 461
 462         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 463                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 464                 struct xattr_handler *handler =
 465                         ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
 466
 467                 if (handler) {
 468                         size_t size = handler->list(inode, buffer, rest,
 469                                         ((char *)header +
 470                                         le16_to_cpu(entry->xe_name_offset)),
 471                                         entry->xe_name_len);
 472                         if (buffer) {
 473                                 if (size > rest)
 474                                         return -ERANGE;
 475                                 buffer += size;
 476                         }
 477                         rest -= size;
 478                 }
 479         }
 480
 481         return buffer_size - rest;
 482 }
 483
 484 static int ocfs2_xattr_ibody_list(struct inode *inode,
 485                                   struct ocfs2_dinode *di,
 486                                   char *buffer,
 487                                   size_t buffer_size)
 488 {
 489         struct ocfs2_xattr_header *header = NULL;
 490         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 491         int ret = 0;
 492
 493         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 494                 return ret;
 495
 496         header = (struct ocfs2_xattr_header *)
 497                  ((void *)di + inode->i_sb->s_blocksize -
 498                  le16_to_cpu(di->i_xattr_inline_size));
 499
 500         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 501
 502         return ret;
 503 }
 504
 505 static int ocfs2_xattr_block_list(struct inode *inode,
 506                                   struct ocfs2_dinode *di,
 507                                   char *buffer,
 508                                   size_t buffer_size)
 509 {
 510         struct buffer_head *blk_bh = NULL;
 511         struct ocfs2_xattr_block *xb;
 512         int ret = 0;
 513
 514         if (!di->i_xattr_loc)
 515                 return ret;
 516
 517         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
 518                                le64_to_cpu(di->i_xattr_loc),
 519                                &blk_bh, OCFS2_BH_CACHED, inode);
 520         if (ret < 0) {
 521                 mlog_errno(ret);
 522                 return ret;
 523         }
 524         /*Verify the signature of xattr block*/
 525         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
 526                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
 527                 ret = -EFAULT;
 528                 goto cleanup;
 529         }
 530
 531         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 532
 533         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 534                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 535                 ret = ocfs2_xattr_list_entries(inode, header,
 536                                                buffer, buffer_size);
 537         } else {
 538                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
 539                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
 540                                                    buffer, buffer_size);
 541         }
 542 cleanup:
 543         brelse(blk_bh);
 544
 545         return ret;
 546 }
 547
 548 ssize_t ocfs2_listxattr(struct dentry *dentry,
 549                         char *buffer,
 550                         size_t size)
 551 {
 552         int ret = 0, i_ret = 0, b_ret = 0;
 553         struct buffer_head *di_bh = NULL;
 554         struct ocfs2_dinode *di = NULL;
 555         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
 556
 557         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
 558                 return -EOPNOTSUPP;
 559
 560         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 561                 return ret;
 562
 563         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
 564         if (ret < 0) {
 565                 mlog_errno(ret);
 566                 return ret;
 567         }
 568
 569         di = (struct ocfs2_dinode *)di_bh->b_data;
 570
 571         down_read(&oi->ip_xattr_sem);
 572         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
 573         if (i_ret < 0)
 574                 b_ret = 0;
 575         else {
 576                 if (buffer) {
 577                         buffer += i_ret;
 578                         size -= i_ret;
 579                 }
 580                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
 581                                                buffer, size);
 582                 if (b_ret < 0)
 583                         i_ret = 0;
 584         }
 585         up_read(&oi->ip_xattr_sem);
 586         ocfs2_inode_unlock(dentry->d_inode, 0);
 587
 588         brelse(di_bh);
 589
 590         return i_ret + b_ret;
 591 }
 592
 593 static int ocfs2_xattr_find_entry(int name_index,
 594                                   const char *name,
 595                                   struct ocfs2_xattr_search *xs)
 596 {
 597         struct ocfs2_xattr_entry *entry;
 598         size_t name_len;
 599         int i, cmp = 1;
 600
 601         if (name == NULL)
 602                 return -EINVAL;
 603
 604         name_len = strlen(name);
 605         entry = xs->here;
 606         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
 607                 cmp = name_index - ocfs2_xattr_get_type(entry);
 608                 if (!cmp)
 609                         cmp = name_len - entry->xe_name_len;
 610                 if (!cmp)
 611                         cmp = memcmp(name, (xs->base +
 612                                      le16_to_cpu(entry->xe_name_offset)),
 613                                      name_len);
 614                 if (cmp == 0)
 615                         break;
 616                 entry += 1;
 617         }
 618         xs->here = entry;
 619
 620         return cmp ? -ENODATA : 0;
 621 }
 622
 623 static int ocfs2_xattr_get_value_outside(struct inode *inode,
 624                                          struct ocfs2_xattr_value_root *xv,
 625                                          void *buffer,
 626                                          size_t len)
 627 {
 628         u32 cpos, p_cluster, num_clusters, bpc, clusters;
 629         u64 blkno;
 630         int i, ret = 0;
 631         size_t cplen, blocksize;
 632         struct buffer_head *bh = NULL;
 633         struct ocfs2_extent_list *el;
 634
 635         el = &xv->xr_list;
 636         clusters = le32_to_cpu(xv->xr_clusters);
 637         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 638         blocksize = inode->i_sb->s_blocksize;
 639
 640         cpos = 0;
 641         while (cpos < clusters) {
 642                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 643                                                &num_clusters, el);
 644                 if (ret) {
 645                         mlog_errno(ret);
 646                         goto out;
 647                 }
 648
 649                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 650                 /* Copy ocfs2_xattr_value */
 651                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 652                         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
 653                                                &bh, OCFS2_BH_CACHED, inode);
 654                         if (ret) {
 655                                 mlog_errno(ret);
 656                                 goto out;
 657                         }
 658
 659                         cplen = len >= blocksize ? blocksize : len;
 660                         memcpy(buffer, bh->b_data, cplen);
 661                         len -= cplen;
 662                         buffer += cplen;
 663
 664                         brelse(bh);
 665                         bh = NULL;
 666                         if (len == 0)
 667                                 break;
 668                 }
 669                 cpos += num_clusters;
 670         }
 671 out:
 672         return ret;
 673 }
 674
 675 static int ocfs2_xattr_ibody_get(struct inode *inode,
 676                                  int name_index,
 677                                  const char *name,
 678                                  void *buffer,
 679                                  size_t buffer_size,
 680                                  struct ocfs2_xattr_search *xs)
 681 {
 682         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 683         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
 684         struct ocfs2_xattr_value_root *xv;
 685         size_t size;
 686         int ret = 0;
 687
 688         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 689                 return -ENODATA;
 690
 691         xs->end = (void *)di + inode->i_sb->s_blocksize;
 692         xs->header = (struct ocfs2_xattr_header *)
 693                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
 694         xs->base = (void *)xs->header;
 695         xs->here = xs->header->xh_entries;
 696
 697         ret = ocfs2_xattr_find_entry(name_index, name, xs);
 698         if (ret)
 699                 return ret;
 700         size = le64_to_cpu(xs->here->xe_value_size);
 701         if (buffer) {
 702                 if (size > buffer_size)
 703                         return -ERANGE;
 704                 if (ocfs2_xattr_is_local(xs->here)) {
 705                         memcpy(buffer, (void *)xs->base +
 706                                le16_to_cpu(xs->here->xe_name_offset) +
 707                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
 708                 } else {
 709                         xv = (struct ocfs2_xattr_value_root *)
 710                                 (xs->base + le16_to_cpu(
 711                                  xs->here->xe_name_offset) +
 712                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
 713                         ret = ocfs2_xattr_get_value_outside(inode, xv,
 714                                                             buffer, size);
 715                         if (ret < 0) {
 716                                 mlog_errno(ret);
 717                                 return ret;
 718                         }
 719                 }
 720         }
 721
 722         return size;
 723 }
 724
 725 static int ocfs2_xattr_block_get(struct inode *inode,
 726                                  int name_index,
 727                                  const char *name,
 728                                  void *buffer,
 729                                  size_t buffer_size,
 730                                  struct ocfs2_xattr_search *xs)
 731 {
 732         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
 733         struct buffer_head *blk_bh = NULL;
 734         struct ocfs2_xattr_block *xb;
 735         struct ocfs2_xattr_value_root *xv;
 736         size_t size;
 737         int ret = -ENODATA, name_offset, name_len, block_off, i;
 738
 739         if (!di->i_xattr_loc)
 740                 return ret;
 741
 742         memset(&xs->bucket, 0, sizeof(xs->bucket));
 743
 744         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
 745                                le64_to_cpu(di->i_xattr_loc),
 746                                &blk_bh, OCFS2_BH_CACHED, inode);
 747         if (ret < 0) {
 748                 mlog_errno(ret);
 749                 return ret;
 750         }
 751         /*Verify the signature of xattr block*/
 752         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
 753                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
 754                 ret = -EFAULT;
 755                 goto cleanup;
 756         }
 757
 758         xs->xattr_bh = blk_bh;
 759         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 760
 761         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 762                 xs->header = &xb->xb_attrs.xb_header;
 763                 xs->base = (void *)xs->header;
 764                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
 765                 xs->here = xs->header->xh_entries;
 766
 767                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
 768         } else
 769                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
 770                                                    name_index,
 771                                                    name, xs);
 772
 773         if (ret)
 774                 goto cleanup;
 775         size = le64_to_cpu(xs->here->xe_value_size);
 776         if (buffer) {
 777                 ret = -ERANGE;
 778                 if (size > buffer_size)
 779                         goto cleanup;
 780
 781                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
 782                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
 783                 i = xs->here - xs->header->xh_entries;
 784
 785                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
 786                         ret = ocfs2_xattr_bucket_get_name_value(inode,
 787                                                                 xs->bucket.xh,
 788                                                                 i,
 789                                                                 &block_off,
 790                                                                 &name_offset);
 791                         xs->base = xs->bucket.bhs[block_off]->b_data;
 792                 }
 793                 if (ocfs2_xattr_is_local(xs->here)) {
 794                         memcpy(buffer, (void *)xs->base +
 795                                name_offset + name_len, size);
 796                 } else {
 797                         xv = (struct ocfs2_xattr_value_root *)
 798                                 (xs->base + name_offset + name_len);
 799                         ret = ocfs2_xattr_get_value_outside(inode, xv,
 800                                                             buffer, size);
 801                         if (ret < 0) {
 802                                 mlog_errno(ret);
 803                                 goto cleanup;
 804                         }
 805                 }
 806         }
 807         ret = size;
 808 cleanup:
 809         for (i = 0; i < OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET; i++)
 810                 brelse(xs->bucket.bhs[i]);
 811         memset(&xs->bucket, 0, sizeof(xs->bucket));
 812
 813         brelse(blk_bh);
 814         return ret;
 815 }
 816
 817 /* ocfs2_xattr_get()
 818  *
 819  * Copy an extended attribute into the buffer provided.
 820  * Buffer is NULL to compute the size of buffer required.
 821  */
 822 int ocfs2_xattr_get(struct inode *inode,
 823                     int name_index,
 824                     const char *name,
 825                     void *buffer,
 826                     size_t buffer_size)
 827 {
 828         int ret;
 829         struct ocfs2_dinode *di = NULL;
 830         struct buffer_head *di_bh = NULL;
 831         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 832         struct ocfs2_xattr_search xis = {
 833                 .not_found = -ENODATA,
 834         };
 835         struct ocfs2_xattr_search xbs = {
 836                 .not_found = -ENODATA,
 837         };
 838
 839         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
 840                 return -EOPNOTSUPP;
 841
 842         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 843                 ret = -ENODATA;
 844
 845         ret = ocfs2_inode_lock(inode, &di_bh, 0);
 846         if (ret < 0) {
 847                 mlog_errno(ret);
 848                 return ret;
 849         }
 850         xis.inode_bh = xbs.inode_bh = di_bh;
 851         di = (struct ocfs2_dinode *)di_bh->b_data;
 852
 853         down_read(&oi->ip_xattr_sem);
 854         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
 855                                     buffer_size, &xis);
 856         if (ret == -ENODATA)
 857                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
 858                                             buffer_size, &xbs);
 859         up_read(&oi->ip_xattr_sem);
 860         ocfs2_inode_unlock(inode, 0);
 861
 862         brelse(di_bh);
 863
 864         return ret;
 865 }
 866
 867 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
 868                                            struct ocfs2_xattr_value_root *xv,
 869                                            const void *value,
 870                                            int value_len)
 871 {
 872         int ret = 0, i, cp_len, credits;
 873         u16 blocksize = inode->i_sb->s_blocksize;
 874         u32 p_cluster, num_clusters;
 875         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 876         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
 877         u64 blkno;
 878         struct buffer_head *bh = NULL;
 879         handle_t *handle;
 880
 881         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
 882
 883         credits = clusters * bpc;
 884         handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits);
 885         if (IS_ERR(handle)) {
 886                 ret = PTR_ERR(handle);
 887                 mlog_errno(ret);
 888                 goto out;
 889         }
 890
 891         while (cpos < clusters) {
 892                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 893                                                &num_clusters, &xv->xr_list);
 894                 if (ret) {
 895                         mlog_errno(ret);
 896                         goto out_commit;
 897                 }
 898
 899                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 900
 901                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 902                         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
 903                                                &bh, OCFS2_BH_CACHED, inode);
 904                         if (ret) {
 905                                 mlog_errno(ret);
 906                                 goto out_commit;
 907                         }
 908
 909                         ret = ocfs2_journal_access(handle,
 910                                                    inode,
 911                                                    bh,
 912                                                    OCFS2_JOURNAL_ACCESS_WRITE);
 913                         if (ret < 0) {
 914                                 mlog_errno(ret);
 915                                 goto out_commit;
 916                         }
 917
 918                         cp_len = value_len > blocksize ? blocksize : value_len;
 919                         memcpy(bh->b_data, value, cp_len);
 920                         value_len -= cp_len;
 921                         value += cp_len;
 922                         if (cp_len < blocksize)
 923                                 memset(bh->b_data + cp_len, 0,
 924                                        blocksize - cp_len);
 925
 926                         ret = ocfs2_journal_dirty(handle, bh);
 927                         if (ret < 0) {
 928                                 mlog_errno(ret);
 929                                 goto out_commit;
 930                         }
 931                         brelse(bh);
 932                         bh = NULL;
 933
 934                         /*
 935                          * XXX: do we need to empty all the following
 936                          * blocks in this cluster?
 937                          */
 938                         if (!value_len)
 939                                 break;
 940                 }
 941                 cpos += num_clusters;
 942         }
 943 out_commit:
 944         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 945 out:
 946         brelse(bh);
 947
 948         return ret;
 949 }
 950
 951 static int ocfs2_xattr_cleanup(struct inode *inode,
 952                                struct ocfs2_xattr_info *xi,
 953                                struct ocfs2_xattr_search *xs,
 954                                size_t offs)
 955 {
 956         handle_t *handle = NULL;
 957         int ret = 0;
 958         size_t name_len = strlen(xi->name);
 959         void *val = xs->base + offs;
 960         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 961
 962         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
 963                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
 964         if (IS_ERR(handle)) {
 965                 ret = PTR_ERR(handle);
 966                 mlog_errno(ret);
 967                 goto out;
 968         }
 969         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
 970                                    OCFS2_JOURNAL_ACCESS_WRITE);
 971         if (ret) {
 972                 mlog_errno(ret);
 973                 goto out_commit;
 974         }
 975         /* Decrease xattr count */
 976         le16_add_cpu(&xs->header->xh_count, -1);
 977         /* Remove the xattr entry and tree root which has already be set*/
 978         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
 979         memset(val, 0, size);
 980
 981         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
 982         if (ret < 0)
 983                 mlog_errno(ret);
 984 out_commit:
 985         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 986 out:
 987         return ret;
 988 }
 989
 990 static int ocfs2_xattr_update_entry(struct inode *inode,
 991                                     struct ocfs2_xattr_info *xi,
 992                                     struct ocfs2_xattr_search *xs,
 993                                     size_t offs)
 994 {
 995         handle_t *handle = NULL;
 996         int ret = 0;
 997
 998         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
 999                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1000         if (IS_ERR(handle)) {
1001                 ret = PTR_ERR(handle);
1002                 mlog_errno(ret);
1003                 goto out;
1004         }
1005         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1006                                    OCFS2_JOURNAL_ACCESS_WRITE);
1007         if (ret) {
1008                 mlog_errno(ret);
1009                 goto out_commit;
1010         }
1011
1012         xs->here->xe_name_offset = cpu_to_le16(offs);
1013         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1014         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1015                 ocfs2_xattr_set_local(xs->here, 1);
1016         else
1017                 ocfs2_xattr_set_local(xs->here, 0);
1018         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1019
1020         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1021         if (ret < 0)
1022                 mlog_errno(ret);
1023 out_commit:
1024         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1025 out:
1026         return ret;
1027 }
1028
1029 /*
1030  * ocfs2_xattr_set_value_outside()
1031  *
1032  * Set large size value in B tree.
1033  */
1034 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1035                                          struct ocfs2_xattr_info *xi,
1036                                          struct ocfs2_xattr_search *xs,
1037                                          size_t offs)
1038 {
1039         size_t name_len = strlen(xi->name);
1040         void *val = xs->base + offs;
1041         struct ocfs2_xattr_value_root *xv = NULL;
1042         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1043         int ret = 0;
1044
1045         memset(val, 0, size);
1046         memcpy(val, xi->name, name_len);
1047         xv = (struct ocfs2_xattr_value_root *)
1048                 (val + OCFS2_XATTR_SIZE(name_len));
1049         xv->xr_clusters = 0;
1050         xv->xr_last_eb_blk = 0;
1051         xv->xr_list.l_tree_depth = 0;
1052         xv->xr_list.l_count = cpu_to_le16(1);
1053         xv->xr_list.l_next_free_rec = 0;
1054
1055         ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1056                                          xi->value_len);
1057         if (ret < 0) {
1058                 mlog_errno(ret);
1059                 return ret;
1060         }
1061         ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value,
1062                                               xi->value_len);
1063         if (ret < 0) {
1064                 mlog_errno(ret);
1065                 return ret;
1066         }
1067         ret = ocfs2_xattr_update_entry(inode, xi, xs, offs);
1068         if (ret < 0)
1069                 mlog_errno(ret);
1070
1071         return ret;
1072 }
1073
1074 /*
1075  * ocfs2_xattr_set_entry_local()
1076  *
1077  * Set, replace or remove extended attribute in local.
1078  */
1079 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1080                                         struct ocfs2_xattr_info *xi,
1081                                         struct ocfs2_xattr_search *xs,
1082                                         struct ocfs2_xattr_entry *last,
1083                                         size_t min_offs)
1084 {
1085         size_t name_len = strlen(xi->name);
1086         int i;
1087
1088         if (xi->value && xs->not_found) {
1089                 /* Insert the new xattr entry. */
1090                 le16_add_cpu(&xs->header->xh_count, 1);
1091                 ocfs2_xattr_set_type(last, xi->name_index);
1092                 ocfs2_xattr_set_local(last, 1);
1093                 last->xe_name_len = name_len;
1094         } else {
1095                 void *first_val;
1096                 void *val;
1097                 size_t offs, size;
1098
1099                 first_val = xs->base + min_offs;
1100                 offs = le16_to_cpu(xs->here->xe_name_offset);
1101                 val = xs->base + offs;
1102
1103                 if (le64_to_cpu(xs->here->xe_value_size) >
1104                     OCFS2_XATTR_INLINE_SIZE)
1105                         size = OCFS2_XATTR_SIZE(name_len) +
1106                                 OCFS2_XATTR_ROOT_SIZE;
1107                 else
1108                         size = OCFS2_XATTR_SIZE(name_len) +
1109                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1110
1111                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1112                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1113                         /* The old and the new value have the
1114                            same size. Just replace the value. */
1115                         ocfs2_xattr_set_local(xs->here, 1);
1116                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1117                         /* Clear value bytes. */
1118                         memset(val + OCFS2_XATTR_SIZE(name_len),
1119                                0,
1120                                OCFS2_XATTR_SIZE(xi->value_len));
1121                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1122                                xi->value,
1123                                xi->value_len);
1124                         return;
1125                 }
1126                 /* Remove the old name+value. */
1127                 memmove(first_val + size, first_val, val - first_val);
1128                 memset(first_val, 0, size);
1129                 xs->here->xe_name_hash = 0;
1130                 xs->here->xe_name_offset = 0;
1131                 ocfs2_xattr_set_local(xs->here, 1);
1132                 xs->here->xe_value_size = 0;
1133
1134                 min_offs += size;
1135
1136                 /* Adjust all value offsets. */
1137                 last = xs->header->xh_entries;
1138                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1139                         size_t o = le16_to_cpu(last->xe_name_offset);
1140
1141                         if (o < offs)
1142                                 last->xe_name_offset = cpu_to_le16(o + size);
1143                         last += 1;
1144                 }
1145
1146                 if (!xi->value) {
1147                         /* Remove the old entry. */
1148                         last -= 1;
1149                         memmove(xs->here, xs->here + 1,
1150                                 (void *)last - (void *)xs->here);
1151                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1152                         le16_add_cpu(&xs->header->xh_count, -1);
1153                 }
1154         }
1155         if (xi->value) {
1156                 /* Insert the new name+value. */
1157                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1158                                 OCFS2_XATTR_SIZE(xi->value_len);
1159                 void *val = xs->base + min_offs - size;
1160
1161                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1162                 memset(val, 0, size);
1163                 memcpy(val, xi->name, name_len);
1164                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1165                        xi->value,
1166                        xi->value_len);
1167                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1168                 ocfs2_xattr_set_local(xs->here, 1);
1169                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1170         }
1171
1172         return;
1173 }
1174
1175 /*
1176  * ocfs2_xattr_set_entry()
1177  *
1178  * Set extended attribute entry into inode or block.
1179  *
1180  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1181  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1182  * then set value in B tree with set_value_outside().
1183  */
1184 static int ocfs2_xattr_set_entry(struct inode *inode,
1185                                  struct ocfs2_xattr_info *xi,
1186                                  struct ocfs2_xattr_search *xs,
1187                                  int flag)
1188 {
1189         struct ocfs2_xattr_entry *last;
1190         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1191         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1192         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1193         size_t size_l = 0;
1194         handle_t *handle = NULL;
1195         int free, i, ret;
1196         struct ocfs2_xattr_info xi_l = {
1197                 .name_index = xi->name_index,
1198                 .name = xi->name,
1199                 .value = xi->value,
1200                 .value_len = xi->value_len,
1201         };
1202
1203         /* Compute min_offs, last and free space. */
1204         last = xs->header->xh_entries;
1205
1206         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1207                 size_t offs = le16_to_cpu(last->xe_name_offset);
1208                 if (offs < min_offs)
1209                         min_offs = offs;
1210                 last += 1;
1211         }
1212
1213         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1214         if (free < 0)
1215                 return -EFAULT;
1216
1217         if (!xs->not_found) {
1218                 size_t size = 0;
1219                 if (ocfs2_xattr_is_local(xs->here))
1220                         size = OCFS2_XATTR_SIZE(name_len) +
1221                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1222                 else
1223                         size = OCFS2_XATTR_SIZE(name_len) +
1224                                 OCFS2_XATTR_ROOT_SIZE;
1225                 free += (size + sizeof(struct ocfs2_xattr_entry));
1226         }
1227         /* Check free space in inode or block */
1228         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1229                 if (free < sizeof(struct ocfs2_xattr_entry) +
1230                            OCFS2_XATTR_SIZE(name_len) +
1231                            OCFS2_XATTR_ROOT_SIZE) {
1232                         ret = -ENOSPC;
1233                         goto out;
1234                 }
1235                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1236                 xi_l.value = (void *)&def_xv;
1237                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1238         } else if (xi->value) {
1239                 if (free < sizeof(struct ocfs2_xattr_entry) +
1240                            OCFS2_XATTR_SIZE(name_len) +
1241                            OCFS2_XATTR_SIZE(xi->value_len)) {
1242                         ret = -ENOSPC;
1243                         goto out;
1244                 }
1245         }
1246
1247         if (!xs->not_found) {
1248                 /* For existing extended attribute */
1249                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1250                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1251                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1252                 void *val = xs->base + offs;
1253
1254                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1255                         /* Replace existing local xattr with tree root */
1256                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1257                                                             offs);
1258                         if (ret < 0)
1259                                 mlog_errno(ret);
1260                         goto out;
1261                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1262                         /* For existing xattr which has value outside */
1263                         struct ocfs2_xattr_value_root *xv = NULL;
1264                         xv = (struct ocfs2_xattr_value_root *)(val +
1265                                 OCFS2_XATTR_SIZE(name_len));
1266
1267                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1268                                 /*
1269                                  * If new value need set outside also,
1270                                  * first truncate old value to new value,
1271                                  * then set new value with set_value_outside().
1272                                  */
1273                                 ret = ocfs2_xattr_value_truncate(inode,
1274                                                                  xs->xattr_bh,
1275                                                                  xv,
1276                                                                  xi->value_len);
1277                                 if (ret < 0) {
1278                                         mlog_errno(ret);
1279                                         goto out;
1280                                 }
1281
1282                                 ret = __ocfs2_xattr_set_value_outside(inode,
1283                                                                 xv,
1284                                                                 xi->value,
1285                                                                 xi->value_len);
1286                                 if (ret < 0) {
1287                                         mlog_errno(ret);
1288                                         goto out;
1289                                 }
1290
1291                                 ret = ocfs2_xattr_update_entry(inode,
1292                                                                xi,
1293                                                                xs,
1294                                                                offs);
1295                                 if (ret < 0)
1296                                         mlog_errno(ret);
1297                                 goto out;
1298                         } else {
1299                                 /*
1300                                  * If new value need set in local,
1301                                  * just trucate old value to zero.
1302                                  */
1303                                  ret = ocfs2_xattr_value_truncate(inode,
1304                                                                  xs->xattr_bh,
1305                                                                  xv,
1306                                                                  0);
1307                                 if (ret < 0)
1308                                         mlog_errno(ret);
1309                         }
1310                 }
1311         }
1312
1313         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1314                                    OCFS2_INODE_UPDATE_CREDITS);
1315         if (IS_ERR(handle)) {
1316                 ret = PTR_ERR(handle);
1317                 mlog_errno(ret);
1318                 goto out;
1319         }
1320
1321         ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1322                                    OCFS2_JOURNAL_ACCESS_WRITE);
1323         if (ret) {
1324                 mlog_errno(ret);
1325                 goto out_commit;
1326         }
1327
1328         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1329                 /* set extended attribute in external block. */
1330                 ret = ocfs2_extend_trans(handle,
1331                                          OCFS2_INODE_UPDATE_CREDITS +
1332                                          OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1333                 if (ret) {
1334                         mlog_errno(ret);
1335                         goto out_commit;
1336                 }
1337                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1338                                            OCFS2_JOURNAL_ACCESS_WRITE);
1339                 if (ret) {
1340                         mlog_errno(ret);
1341                         goto out_commit;
1342                 }
1343         }
1344
1345         /*
1346          * Set value in local, include set tree root in local.
1347          * This is the first step for value size >INLINE_SIZE.
1348          */
1349         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1350
1351         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1352                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1353                 if (ret < 0) {
1354                         mlog_errno(ret);
1355                         goto out_commit;
1356                 }
1357         }
1358
1359         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1360             (flag & OCFS2_INLINE_XATTR_FL)) {
1361                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1362                 unsigned int xattrsize = osb->s_xattr_inline_size;
1363
1364                 /*
1365                  * Adjust extent record count or inline data size
1366                  * to reserve space for extended attribute.
1367                  */
1368                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1369                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1370                         le16_add_cpu(&idata->id_count, -xattrsize);
1371                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1372                         struct ocfs2_extent_list *el = &di->id2.i_list;
1373                         le16_add_cpu(&el->l_count, -(xattrsize /
1374                                         sizeof(struct ocfs2_extent_rec)));
1375                 }
1376                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1377         }
1378         /* Update xattr flag */
1379         spin_lock(&oi->ip_lock);
1380         oi->ip_dyn_features |= flag;
1381         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1382         spin_unlock(&oi->ip_lock);
1383         /* Update inode ctime */
1384         inode->i_ctime = CURRENT_TIME;
1385         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1386         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1387
1388         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1389         if (ret < 0)
1390                 mlog_errno(ret);
1391
1392 out_commit:
1393         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1394
1395         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1396                 /*
1397                  * Set value outside in B tree.
1398                  * This is the second step for value size > INLINE_SIZE.
1399                  */
1400                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1401                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1402                 if (ret < 0) {
1403                         int ret2;
1404
1405                         mlog_errno(ret);
1406                         /*
1407                          * If set value outside failed, we have to clean
1408                          * the junk tree root we have already set in local.
1409                          */
1410                         ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1411                         if (ret2 < 0)
1412                                 mlog_errno(ret2);
1413                 }
1414         }
1415 out:
1416         return ret;
1417
1418 }
1419
1420 static int ocfs2_remove_value_outside(struct inode*inode,
1421                                       struct buffer_head *bh,
1422                                       struct ocfs2_xattr_header *header)
1423 {
1424         int ret = 0, i;
1425
1426         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1427                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1428
1429                 if (!ocfs2_xattr_is_local(entry)) {
1430                         struct ocfs2_xattr_value_root *xv;
1431                         void *val;
1432
1433                         val = (void *)header +
1434                                 le16_to_cpu(entry->xe_name_offset);
1435                         xv = (struct ocfs2_xattr_value_root *)
1436                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1437                         ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1438                         if (ret < 0) {
1439                                 mlog_errno(ret);
1440                                 return ret;
1441                         }
1442                 }
1443         }
1444
1445         return ret;
1446 }
1447
1448 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1449                                     struct buffer_head *di_bh)
1450 {
1451
1452         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1453         struct ocfs2_xattr_header *header;
1454         int ret;
1455
1456         header = (struct ocfs2_xattr_header *)
1457                  ((void *)di + inode->i_sb->s_blocksize -
1458                  le16_to_cpu(di->i_xattr_inline_size));
1459
1460         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1461
1462         return ret;
1463 }
1464
1465 static int ocfs2_xattr_block_remove(struct inode *inode,
1466                                     struct buffer_head *blk_bh)
1467 {
1468         struct ocfs2_xattr_block *xb;
1469         int ret = 0;
1470
1471         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1472         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1473                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1474                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1475         } else
1476                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1477
1478         return ret;
1479 }
1480
1481 static int ocfs2_xattr_free_block(struct inode *inode,
1482                                   u64 block)
1483 {
1484         struct inode *xb_alloc_inode;
1485         struct buffer_head *xb_alloc_bh = NULL;
1486         struct buffer_head *blk_bh = NULL;
1487         struct ocfs2_xattr_block *xb;
1488         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1489         handle_t *handle;
1490         int ret = 0;
1491         u64 blk, bg_blkno;
1492         u16 bit;
1493
1494         ret = ocfs2_read_block(osb, block, &blk_bh,
1495                                OCFS2_BH_CACHED, inode);
1496         if (ret < 0) {
1497                 mlog_errno(ret);
1498                 goto out;
1499         }
1500
1501         /*Verify the signature of xattr block*/
1502         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1503                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1504                 ret = -EFAULT;
1505                 goto out;
1506         }
1507
1508         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1509         if (ret < 0) {
1510                 mlog_errno(ret);
1511                 goto out;
1512         }
1513
1514         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1515         blk = le64_to_cpu(xb->xb_blkno);
1516         bit = le16_to_cpu(xb->xb_suballoc_bit);
1517         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1518
1519         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1520                                 EXTENT_ALLOC_SYSTEM_INODE,
1521                                 le16_to_cpu(xb->xb_suballoc_slot));
1522         if (!xb_alloc_inode) {
1523                 ret = -ENOMEM;
1524                 mlog_errno(ret);
1525                 goto out;
1526         }
1527         mutex_lock(&xb_alloc_inode->i_mutex);
1528
1529         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1530         if (ret < 0) {
1531                 mlog_errno(ret);
1532                 goto out_mutex;
1533         }
1534
1535         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1536         if (IS_ERR(handle)) {
1537                 ret = PTR_ERR(handle);
1538                 mlog_errno(ret);
1539                 goto out_unlock;
1540         }
1541
1542         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1543                                        bit, bg_blkno, 1);
1544         if (ret < 0)
1545                 mlog_errno(ret);
1546
1547         ocfs2_commit_trans(osb, handle);
1548 out_unlock:
1549         ocfs2_inode_unlock(xb_alloc_inode, 1);
1550         brelse(xb_alloc_bh);
1551 out_mutex:
1552         mutex_unlock(&xb_alloc_inode->i_mutex);
1553         iput(xb_alloc_inode);
1554 out:
1555         brelse(blk_bh);
1556         return ret;
1557 }
1558
1559 /*
1560  * ocfs2_xattr_remove()
1561  *
1562  * Free extended attribute resources associated with this inode.
1563  */
1564 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1565 {
1566         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1567         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1568         handle_t *handle;
1569         int ret;
1570
1571         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1572                 return 0;
1573
1574         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1575                 return 0;
1576
1577         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1578                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1579                 if (ret < 0) {
1580                         mlog_errno(ret);
1581                         goto out;
1582                 }
1583         }
1584
1585         if (di->i_xattr_loc) {
1586                 ret = ocfs2_xattr_free_block(inode,
1587                                              le64_to_cpu(di->i_xattr_loc));
1588                 if (ret < 0) {
1589                         mlog_errno(ret);
1590                         goto out;
1591                 }
1592         }
1593
1594         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1595                                    OCFS2_INODE_UPDATE_CREDITS);
1596         if (IS_ERR(handle)) {
1597                 ret = PTR_ERR(handle);
1598                 mlog_errno(ret);
1599                 goto out;
1600         }
1601         ret = ocfs2_journal_access(handle, inode, di_bh,
1602                                    OCFS2_JOURNAL_ACCESS_WRITE);
1603         if (ret) {
1604                 mlog_errno(ret);
1605                 goto out_commit;
1606         }
1607
1608         di->i_xattr_loc = 0;
1609
1610         spin_lock(&oi->ip_lock);
1611         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1612         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1613         spin_unlock(&oi->ip_lock);
1614
1615         ret = ocfs2_journal_dirty(handle, di_bh);
1616         if (ret < 0)
1617                 mlog_errno(ret);
1618 out_commit:
1619         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1620 out:
1621         return ret;
1622 }
1623
1624 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1625                                         struct ocfs2_dinode *di)
1626 {
1627         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1628         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1629         int free;
1630
1631         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1632                 return 0;
1633
1634         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1635                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1636                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1637         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1638                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1639                         le64_to_cpu(di->i_size);
1640         } else {
1641                 struct ocfs2_extent_list *el = &di->id2.i_list;
1642                 free = (le16_to_cpu(el->l_count) -
1643                         le16_to_cpu(el->l_next_free_rec)) *
1644                         sizeof(struct ocfs2_extent_rec);
1645         }
1646         if (free >= xattrsize)
1647                 return 1;
1648
1649         return 0;
1650 }
1651
1652 /*
1653  * ocfs2_xattr_ibody_find()
1654  *
1655  * Find extended attribute in inode block and
1656  * fill search info into struct ocfs2_xattr_search.
1657  */
1658 static int ocfs2_xattr_ibody_find(struct inode *inode,
1659                                   int name_index,
1660                                   const char *name,
1661                                   struct ocfs2_xattr_search *xs)
1662 {
1663         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1664         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1665         int ret;
1666         int has_space = 0;
1667
1668         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1669                 return 0;
1670
1671         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1672                 down_read(&oi->ip_alloc_sem);
1673                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1674                 up_read(&oi->ip_alloc_sem);
1675                 if (!has_space)
1676                         return 0;
1677         }
1678
1679         xs->xattr_bh = xs->inode_bh;
1680         xs->end = (void *)di + inode->i_sb->s_blocksize;
1681         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1682                 xs->header = (struct ocfs2_xattr_header *)
1683                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1684         else
1685                 xs->header = (struct ocfs2_xattr_header *)
1686                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1687         xs->base = (void *)xs->header;
1688         xs->here = xs->header->xh_entries;
1689
1690         /* Find the named attribute. */
1691         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1692                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1693                 if (ret && ret != -ENODATA)
1694                         return ret;
1695                 xs->not_found = ret;
1696         }
1697
1698         return 0;
1699 }
1700
1701 /*
1702  * ocfs2_xattr_ibody_set()
1703  *
1704  * Set, replace or remove an extended attribute into inode block.
1705  *
1706  */
1707 static int ocfs2_xattr_ibody_set(struct inode *inode,
1708                                  struct ocfs2_xattr_info *xi,
1709                                  struct ocfs2_xattr_search *xs)
1710 {
1711         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1712         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1713         int ret;
1714
1715         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1716                 return -ENOSPC;
1717
1718         down_write(&oi->ip_alloc_sem);
1719         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1720                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1721                         ret = -ENOSPC;
1722                         goto out;
1723                 }
1724         }
1725
1726         ret = ocfs2_xattr_set_entry(inode, xi, xs,
1727                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1728 out:
1729         up_write(&oi->ip_alloc_sem);
1730
1731         return ret;
1732 }
1733
1734 /*
1735  * ocfs2_xattr_block_find()
1736  *
1737  * Find extended attribute in external block and
1738  * fill search info into struct ocfs2_xattr_search.
1739  */
1740 static int ocfs2_xattr_block_find(struct inode *inode,
1741                                   int name_index,
1742                                   const char *name,
1743                                   struct ocfs2_xattr_search *xs)
1744 {
1745         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1746         struct buffer_head *blk_bh = NULL;
1747         struct ocfs2_xattr_block *xb;
1748         int ret = 0;
1749
1750         if (!di->i_xattr_loc)
1751                 return ret;
1752
1753         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1754                                le64_to_cpu(di->i_xattr_loc),
1755                                &blk_bh, OCFS2_BH_CACHED, inode);
1756         if (ret < 0) {
1757                 mlog_errno(ret);
1758                 return ret;
1759         }
1760         /*Verify the signature of xattr block*/
1761         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1762                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1763                         ret = -EFAULT;
1764                         goto cleanup;
1765         }
1766
1767         xs->xattr_bh = blk_bh;
1768         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1769
1770         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1771                 xs->header = &xb->xb_attrs.xb_header;
1772                 xs->base = (void *)xs->header;
1773                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1774                 xs->here = xs->header->xh_entries;
1775
1776                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1777         } else
1778                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1779                                                    name_index,
1780                                                    name, xs);
1781
1782         if (ret && ret != -ENODATA) {
1783                 xs->xattr_bh = NULL;
1784                 goto cleanup;
1785         }
1786         xs->not_found = ret;
1787         return 0;
1788 cleanup:
1789         brelse(blk_bh);
1790
1791         return ret;
1792 }
1793
1794 /*
1795  * When all the xattrs are deleted from index btree, the ocfs2_xattr_tree
1796  * will be erased and ocfs2_xattr_block will have its ocfs2_xattr_header
1797  * re-initialized.
1798  */
1799 static int ocfs2_restore_xattr_block(struct inode *inode,
1800                                      struct ocfs2_xattr_search *xs)
1801 {
1802         int ret;
1803         handle_t *handle;
1804         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1805         struct ocfs2_xattr_block *xb =
1806                 (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1807         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
1808         u16 xb_flags = le16_to_cpu(xb->xb_flags);
1809
1810         BUG_ON(!(xb_flags & OCFS2_XATTR_INDEXED) ||
1811                 le16_to_cpu(el->l_next_free_rec) != 0);
1812
1813         handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1814         if (IS_ERR(handle)) {
1815                 ret = PTR_ERR(handle);
1816                 handle = NULL;
1817                 goto out;
1818         }
1819
1820         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1821                                    OCFS2_JOURNAL_ACCESS_WRITE);
1822         if (ret < 0) {
1823                 mlog_errno(ret);
1824                 goto out_commit;
1825         }
1826
1827         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
1828                offsetof(struct ocfs2_xattr_block, xb_attrs));
1829
1830         xb->xb_flags = cpu_to_le16(xb_flags & ~OCFS2_XATTR_INDEXED);
1831
1832         ocfs2_journal_dirty(handle, xs->xattr_bh);
1833
1834 out_commit:
1835         ocfs2_commit_trans(osb, handle);
1836 out:
1837         return ret;
1838 }
1839
1840 /*
1841  * ocfs2_xattr_block_set()
1842  *
1843  * Set, replace or remove an extended attribute into external block.
1844  *
1845  */
1846 static int ocfs2_xattr_block_set(struct inode *inode,
1847                                  struct ocfs2_xattr_info *xi,
1848                                  struct ocfs2_xattr_search *xs)
1849 {
1850         struct buffer_head *new_bh = NULL;
1851         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1852         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
1853         struct ocfs2_alloc_context *meta_ac = NULL;
1854         handle_t *handle = NULL;
1855         struct ocfs2_xattr_block *xblk = NULL;
1856         u16 suballoc_bit_start;
1857         u32 num_got;
1858         u64 first_blkno;
1859         int ret;
1860
1861         if (!xs->xattr_bh) {
1862                 /*
1863                  * Alloc one external block for extended attribute
1864                  * outside of inode.
1865                  */
1866                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1867                 if (ret < 0) {
1868                         mlog_errno(ret);
1869                         goto out;
1870                 }
1871                 handle = ocfs2_start_trans(osb,
1872                                            OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1873                 if (IS_ERR(handle)) {
1874                         ret = PTR_ERR(handle);
1875                         mlog_errno(ret);
1876                         goto out;
1877                 }
1878                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1879                                            OCFS2_JOURNAL_ACCESS_CREATE);
1880                 if (ret < 0) {
1881                         mlog_errno(ret);
1882                         goto out_commit;
1883                 }
1884
1885                 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1886                                            &suballoc_bit_start, &num_got,
1887                                            &first_blkno);
1888                 if (ret < 0) {
1889                         mlog_errno(ret);
1890                         goto out_commit;
1891                 }
1892
1893                 new_bh = sb_getblk(inode->i_sb, first_blkno);
1894                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1895
1896                 ret = ocfs2_journal_access(handle, inode, new_bh,
1897                                            OCFS2_JOURNAL_ACCESS_CREATE);
1898                 if (ret < 0) {
1899                         mlog_errno(ret);
1900                         goto out_commit;
1901                 }
1902
1903                 /* Initialize ocfs2_xattr_block */
1904                 xs->xattr_bh = new_bh;
1905                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1906                 memset(xblk, 0, inode->i_sb->s_blocksize);
1907                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1908                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1909                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1910                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1911                 xblk->xb_blkno = cpu_to_le64(first_blkno);
1912
1913                 xs->header = &xblk->xb_attrs.xb_header;
1914                 xs->base = (void *)xs->header;
1915                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1916                 xs->here = xs->header->xh_entries;
1917
1918
1919                 ret = ocfs2_journal_dirty(handle, new_bh);
1920                 if (ret < 0) {
1921                         mlog_errno(ret);
1922                         goto out_commit;
1923                 }
1924                 di->i_xattr_loc = cpu_to_le64(first_blkno);
1925                 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1926                 if (ret < 0)
1927                         mlog_errno(ret);
1928 out_commit:
1929                 ocfs2_commit_trans(osb, handle);
1930 out:
1931                 if (meta_ac)
1932                         ocfs2_free_alloc_context(meta_ac);
1933                 if (ret < 0)
1934                         return ret;
1935         } else
1936                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1937
1938         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1939                 /* Set extended attribute into external block */
1940                 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
1941                 if (!ret || ret != -ENOSPC)
1942                         goto end;
1943
1944                 ret = ocfs2_xattr_create_index_block(inode, xs);
1945                 if (ret)
1946                         goto end;
1947         }
1948
1949         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
1950         if (!ret && xblk->xb_attrs.xb_root.xt_list.l_next_free_rec == 0)
1951                 ret = ocfs2_restore_xattr_block(inode, xs);
1952
1953 end:
1954
1955         return ret;
1956 }
1957
1958 /*
1959  * ocfs2_xattr_set()
1960  *
1961  * Set, replace or remove an extended attribute for this inode.
1962  * value is NULL to remove an existing extended attribute, else either
1963  * create or replace an extended attribute.
1964  */
1965 int ocfs2_xattr_set(struct inode *inode,
1966                     int name_index,
1967                     const char *name,
1968                     const void *value,
1969                     size_t value_len,
1970                     int flags)
1971 {
1972         struct buffer_head *di_bh = NULL;
1973         struct ocfs2_dinode *di;
1974         int ret;
1975         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
1976
1977         struct ocfs2_xattr_info xi = {
1978                 .name_index = name_index,
1979                 .name = name,
1980                 .value = value,
1981                 .value_len = value_len,
1982         };
1983
1984         struct ocfs2_xattr_search xis = {
1985                 .not_found = -ENODATA,
1986         };
1987
1988         struct ocfs2_xattr_search xbs = {
1989                 .not_found = -ENODATA,
1990         };
1991
1992         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1993                 return -EOPNOTSUPP;
1994
1995         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1996         if (ret < 0) {
1997                 mlog_errno(ret);
1998                 return ret;
1999         }
2000         xis.inode_bh = xbs.inode_bh = di_bh;
2001         di = (struct ocfs2_dinode *)di_bh->b_data;
2002
2003         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2004         /*
2005          * Scan inode and external block to find the same name
2006          * extended attribute and collect search infomation.
2007          */
2008         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2009         if (ret)
2010                 goto cleanup;
2011         if (xis.not_found) {
2012                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2013                 if (ret)
2014                         goto cleanup;
2015         }
2016
2017         if (xis.not_found && xbs.not_found) {
2018                 ret = -ENODATA;
2019                 if (flags & XATTR_REPLACE)
2020                         goto cleanup;
2021                 ret = 0;
2022                 if (!value)
2023                         goto cleanup;
2024         } else {
2025                 ret = -EEXIST;
2026                 if (flags & XATTR_CREATE)
2027                         goto cleanup;
2028         }
2029
2030         if (!value) {
2031                 /* Remove existing extended attribute */
2032                 if (!xis.not_found)
2033                         ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2034                 else if (!xbs.not_found)
2035                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2036         } else {
2037                 /* We always try to set extended attribute into inode first*/
2038                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2039                 if (!ret && !xbs.not_found) {
2040                         /*
2041                          * If succeed and that extended attribute existing in
2042                          * external block, then we will remove it.
2043                          */
2044                         xi.value = NULL;
2045                         xi.value_len = 0;
2046                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2047                 } else if (ret == -ENOSPC) {
2048                         if (di->i_xattr_loc && !xbs.xattr_bh) {
2049                                 ret = ocfs2_xattr_block_find(inode, name_index,
2050                                                              name, &xbs);
2051                                 if (ret)
2052                                         goto cleanup;
2053                         }
2054                         /*
2055                          * If no space in inode, we will set extended attribute
2056                          * into external block.
2057                          */
2058                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2059                         if (ret)
2060                                 goto cleanup;
2061                         if (!xis.not_found) {
2062                                 /*
2063                                  * If succeed and that extended attribute
2064                                  * existing in inode, we will remove it.
2065                                  */
2066                                 xi.value = NULL;
2067                                 xi.value_len = 0;
2068                                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2069                         }
2070                 }
2071         }
2072 cleanup:
2073         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2074         ocfs2_inode_unlock(inode, 1);
2075         brelse(di_bh);
2076         brelse(xbs.xattr_bh);
2077         for (i = 0; i < blk_per_bucket; i++)
2078                 brelse(xbs.bucket.bhs[i]);
2079
2080         return ret;
2081 }
2082
2083 /*
2084  * Find the xattr extent rec which may contains name_hash.
2085  * e_cpos will be the first name hash of the xattr rec.
2086  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2087  */
2088 static int ocfs2_xattr_get_rec(struct inode *inode,
2089                                u32 name_hash,
2090                                u64 *p_blkno,
2091                                u32 *e_cpos,
2092                                u32 *num_clusters,
2093                                struct ocfs2_extent_list *el)
2094 {
2095         int ret = 0, i;
2096         struct buffer_head *eb_bh = NULL;
2097         struct ocfs2_extent_block *eb;
2098         struct ocfs2_extent_rec *rec = NULL;
2099         u64 e_blkno = 0;
2100
2101         if (el->l_tree_depth) {
2102                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2103                 if (ret) {
2104                         mlog_errno(ret);
2105                         goto out;
2106                 }
2107
2108                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2109                 el = &eb->h_list;
2110
2111                 if (el->l_tree_depth) {
2112                         ocfs2_error(inode->i_sb,
2113                                     "Inode %lu has non zero tree depth in "
2114                                     "xattr tree block %llu\n", inode->i_ino,
2115                                     (unsigned long long)eb_bh->b_blocknr);
2116                         ret = -EROFS;
2117                         goto out;
2118                 }
2119         }
2120
2121         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2122                 rec = &el->l_recs[i];
2123
2124                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2125                         e_blkno = le64_to_cpu(rec->e_blkno);
2126                         break;
2127                 }
2128         }
2129
2130         if (!e_blkno) {
2131                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2132                             "record (%u, %u, 0) in xattr", inode->i_ino,
2133                             le32_to_cpu(rec->e_cpos),
2134                             ocfs2_rec_clusters(el, rec));
2135                 ret = -EROFS;
2136                 goto out;
2137         }
2138
2139         *p_blkno = le64_to_cpu(rec->e_blkno);
2140         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2141         if (e_cpos)
2142                 *e_cpos = le32_to_cpu(rec->e_cpos);
2143 out:
2144         brelse(eb_bh);
2145         return ret;
2146 }
2147
2148 typedef int (xattr_bucket_func)(struct inode *inode,
2149                                 struct ocfs2_xattr_bucket *bucket,
2150                                 void *para);
2151
2152 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2153                                    struct buffer_head *header_bh,
2154                                    int name_index,
2155                                    const char *name,
2156                                    u32 name_hash,
2157                                    u16 *xe_index,
2158                                    int *found)
2159 {
2160         int i, ret = 0, cmp = 1, block_off, new_offset;
2161         struct ocfs2_xattr_header *xh =
2162                         (struct ocfs2_xattr_header *)header_bh->b_data;
2163         size_t name_len = strlen(name);
2164         struct ocfs2_xattr_entry *xe = NULL;
2165         struct buffer_head *name_bh = NULL;
2166         char *xe_name;
2167
2168         /*
2169          * We don't use binary search in the bucket because there
2170          * may be multiple entries with the same name hash.
2171          */
2172         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2173                 xe = &xh->xh_entries[i];
2174
2175                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2176                         continue;
2177                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2178                         break;
2179
2180                 cmp = name_index - ocfs2_xattr_get_type(xe);
2181                 if (!cmp)
2182                         cmp = name_len - xe->xe_name_len;
2183                 if (cmp)
2184                         continue;
2185
2186                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2187                                                         xh,
2188                                                         i,
2189                                                         &block_off,
2190                                                         &new_offset);
2191                 if (ret) {
2192                         mlog_errno(ret);
2193                         break;
2194                 }
2195
2196                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
2197                                        header_bh->b_blocknr + block_off,
2198                                        &name_bh, OCFS2_BH_CACHED, inode);
2199                 if (ret) {
2200                         mlog_errno(ret);
2201                         break;
2202                 }
2203                 xe_name = name_bh->b_data + new_offset;
2204
2205                 cmp = memcmp(name, xe_name, name_len);
2206                 brelse(name_bh);
2207                 name_bh = NULL;
2208
2209                 if (cmp == 0) {
2210                         *xe_index = i;
2211                         *found = 1;
2212                         ret = 0;
2213                         break;
2214                 }
2215         }
2216
2217         return ret;
2218 }
2219
2220 /*
2221  * Find the specified xattr entry in a series of buckets.
2222  * This series start from p_blkno and last for num_clusters.
2223  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2224  * the num of the valid buckets.
2225  *
2226  * Return the buffer_head this xattr should reside in. And if the xattr's
2227  * hash is in the gap of 2 buckets, return the lower bucket.
2228  */
2229 static int ocfs2_xattr_bucket_find(struct inode *inode,
2230                                    int name_index,
2231                                    const char *name,
2232                                    u32 name_hash,
2233                                    u64 p_blkno,
2234                                    u32 first_hash,
2235                                    u32 num_clusters,
2236                                    struct ocfs2_xattr_search *xs)
2237 {
2238         int ret, found = 0;
2239         struct buffer_head *bh = NULL;
2240         struct buffer_head *lower_bh = NULL;
2241         struct ocfs2_xattr_header *xh = NULL;
2242         struct ocfs2_xattr_entry *xe = NULL;
2243         u16 index = 0;
2244         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2245         int low_bucket = 0, bucket, high_bucket;
2246         u32 last_hash;
2247         u64 blkno;
2248
2249         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno,
2250                                &bh, OCFS2_BH_CACHED, inode);
2251         if (ret) {
2252                 mlog_errno(ret);
2253                 goto out;
2254         }
2255
2256         xh = (struct ocfs2_xattr_header *)bh->b_data;
2257         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2258
2259         while (low_bucket <= high_bucket) {
2260                 brelse(bh);
2261                 bh = NULL;
2262                 bucket = (low_bucket + high_bucket) / 2;
2263
2264                 blkno = p_blkno + bucket * blk_per_bucket;
2265
2266                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
2267                                        &bh, OCFS2_BH_CACHED, inode);
2268                 if (ret) {
2269                         mlog_errno(ret);
2270                         goto out;
2271                 }
2272
2273                 xh = (struct ocfs2_xattr_header *)bh->b_data;
2274                 xe = &xh->xh_entries[0];
2275                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2276                         high_bucket = bucket - 1;
2277                         continue;
2278                 }
2279
2280                 /*
2281                  * Check whether the hash of the last entry in our
2282                  * bucket is larger than the search one. for an empty
2283                  * bucket, the last one is also the first one.
2284                  */
2285                 if (xh->xh_count)
2286                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2287
2288                 last_hash = le32_to_cpu(xe->xe_name_hash);
2289
2290                 /* record lower_bh which may be the insert place. */
2291                 brelse(lower_bh);
2292                 lower_bh = bh;
2293                 bh = NULL;
2294
2295                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2296                         low_bucket = bucket + 1;
2297                         continue;
2298                 }
2299
2300                 /* the searched xattr should reside in this bucket if exists. */
2301                 ret = ocfs2_find_xe_in_bucket(inode, lower_bh,
2302                                               name_index, name, name_hash,
2303                                               &index, &found);
2304                 if (ret) {
2305                         mlog_errno(ret);
2306                         goto out;
2307                 }
2308                 break;
2309         }
2310
2311         /*
2312          * Record the bucket we have found.
2313          * When the xattr's hash value is in the gap of 2 buckets, we will
2314          * always set it to the previous bucket.
2315          */
2316         if (!lower_bh) {
2317                 /*
2318                  * We can't find any bucket whose first name_hash is less
2319                  * than the find name_hash.
2320                  */
2321                 BUG_ON(bh->b_blocknr != p_blkno);
2322                 lower_bh = bh;
2323                 bh = NULL;
2324         }
2325         xs->bucket.bhs[0] = lower_bh;
2326         xs->bucket.xh = (struct ocfs2_xattr_header *)
2327                                         xs->bucket.bhs[0]->b_data;
2328         lower_bh = NULL;
2329
2330         xs->header = xs->bucket.xh;
2331         xs->base = xs->bucket.bhs[0]->b_data;
2332         xs->end = xs->base + inode->i_sb->s_blocksize;
2333
2334         if (found) {
2335                 /*
2336                  * If we have found the xattr enty, read all the blocks in
2337                  * this bucket.
2338                  */
2339                 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2340                                         xs->bucket.bhs[0]->b_blocknr + 1,
2341                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2342                                         OCFS2_BH_CACHED, inode);
2343                 if (ret) {
2344                         mlog_errno(ret);
2345                         goto out;
2346                 }
2347
2348                 xs->here = &xs->header->xh_entries[index];
2349                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2350                      (unsigned long long)xs->bucket.bhs[0]->b_blocknr, index);
2351         } else
2352                 ret = -ENODATA;
2353
2354 out:
2355         brelse(bh);
2356         brelse(lower_bh);
2357         return ret;
2358 }
2359
2360 static int ocfs2_xattr_index_block_find(struct inode *inode,
2361                                         struct buffer_head *root_bh,
2362                                         int name_index,
2363                                         const char *name,
2364                                         struct ocfs2_xattr_search *xs)
2365 {
2366         int ret;
2367         struct ocfs2_xattr_block *xb =
2368                         (struct ocfs2_xattr_block *)root_bh->b_data;
2369         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2370         struct ocfs2_extent_list *el = &xb_root->xt_list;
2371         u64 p_blkno = 0;
2372         u32 first_hash, num_clusters = 0;
2373         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
2374
2375         if (le16_to_cpu(el->l_next_free_rec) == 0)
2376                 return -ENODATA;
2377
2378         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2379              name, name_hash, name_index);
2380
2381         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2382                                   &num_clusters, el);
2383         if (ret) {
2384                 mlog_errno(ret);
2385                 goto out;
2386         }
2387
2388         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2389
2390         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2391              "in the rec is %u\n", num_clusters, p_blkno, first_hash);
2392
2393         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2394                                       p_blkno, first_hash, num_clusters, xs);
2395
2396 out:
2397         return ret;
2398 }
2399
2400 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2401                                        u64 blkno,
2402                                        u32 clusters,
2403                                        xattr_bucket_func *func,
2404                                        void *para)
2405 {
2406         int i, j, ret = 0;
2407         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2408         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2409         u32 num_buckets = clusters * bpc;
2410         struct ocfs2_xattr_bucket bucket;
2411
2412         memset(&bucket, 0, sizeof(bucket));
2413
2414         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2415              clusters, blkno);
2416
2417         for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
2418                 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2419                                         blkno, blk_per_bucket,
2420                                         bucket.bhs, OCFS2_BH_CACHED, inode);
2421                 if (ret) {
2422                         mlog_errno(ret);
2423                         goto out;
2424                 }
2425
2426                 bucket.xh = (struct ocfs2_xattr_header *)bucket.bhs[0]->b_data;
2427                 /*
2428                  * The real bucket num in this series of blocks is stored
2429                  * in the 1st bucket.
2430                  */
2431                 if (i == 0)
2432                         num_buckets = le16_to_cpu(bucket.xh->xh_num_buckets);
2433
2434                 mlog(0, "iterating xattr bucket %llu, first hash %u\n", blkno,
2435                      le32_to_cpu(bucket.xh->xh_entries[0].xe_name_hash));
2436                 if (func) {
2437                         ret = func(inode, &bucket, para);
2438                         if (ret) {
2439                                 mlog_errno(ret);
2440                                 break;
2441                         }
2442                 }
2443
2444                 for (j = 0; j < blk_per_bucket; j++)
2445                         brelse(bucket.bhs[j]);
2446                 memset(&bucket, 0, sizeof(bucket));
2447         }
2448
2449 out:
2450         for (j = 0; j < blk_per_bucket; j++)
2451                 brelse(bucket.bhs[j]);
2452
2453         return ret;
2454 }
2455
2456 struct ocfs2_xattr_tree_list {
2457         char *buffer;
2458         size_t buffer_size;
2459 };
2460
2461 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2462                                              struct ocfs2_xattr_header *xh,
2463                                              int index,
2464                                              int *block_off,
2465                                              int *new_offset)
2466 {
2467         u16 name_offset;
2468
2469         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2470                 return -EINVAL;
2471
2472         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2473
2474         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2475         *new_offset = name_offset % inode->i_sb->s_blocksize;
2476
2477         return 0;
2478 }
2479
2480 static int ocfs2_list_xattr_bucket(struct inode *inode,
2481                                    struct ocfs2_xattr_bucket *bucket,
2482                                    void *para)
2483 {
2484         int ret = 0;
2485         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2486         size_t size;
2487         int i, block_off, new_offset;
2488
2489         for (i = 0 ; i < le16_to_cpu(bucket->xh->xh_count); i++) {
2490                 struct ocfs2_xattr_entry *entry = &bucket->xh->xh_entries[i];
2491                 struct xattr_handler *handler =
2492                         ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
2493
2494                 if (handler) {
2495                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2496                                                                 bucket->xh,
2497                                                                 i,
2498                                                                 &block_off,
2499                                                                 &new_offset);
2500                         if (ret)
2501                                 break;
2502                         size = handler->list(inode, xl->buffer, xl->buffer_size,
2503                                              bucket->bhs[block_off]->b_data +
2504                                              new_offset,
2505                                              entry->xe_name_len);
2506                         if (xl->buffer) {
2507                                 if (size > xl->buffer_size)
2508                                         return -ERANGE;
2509                                 xl->buffer += size;
2510                         }
2511                         xl->buffer_size -= size;
2512                 }
2513         }
2514
2515         return ret;
2516 }
2517
2518 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2519                                              struct ocfs2_xattr_tree_root *xt,
2520                                              char *buffer,
2521                                              size_t buffer_size)
2522 {
2523         struct ocfs2_extent_list *el = &xt->xt_list;
2524         int ret = 0;
2525         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2526         u64 p_blkno = 0;
2527         struct ocfs2_xattr_tree_list xl = {
2528                 .buffer = buffer,
2529                 .buffer_size = buffer_size,
2530         };
2531
2532         if (le16_to_cpu(el->l_next_free_rec) == 0)
2533                 return 0;
2534
2535         while (name_hash > 0) {
2536                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2537                                           &e_cpos, &num_clusters, el);
2538                 if (ret) {
2539                         mlog_errno(ret);
2540                         goto out;
2541                 }
2542
2543                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2544                                                   ocfs2_list_xattr_bucket,
2545                                                   &xl);
2546                 if (ret) {
2547                         mlog_errno(ret);
2548                         goto out;
2549                 }
2550
2551                 if (e_cpos == 0)
2552                         break;
2553
2554                 name_hash = e_cpos - 1;
2555         }
2556
2557         ret = buffer_size - xl.buffer_size;
2558 out:
2559         return ret;
2560 }
2561
2562 static int cmp_xe(const void *a, const void *b)
2563 {
2564         const struct ocfs2_xattr_entry *l = a, *r = b;
2565         u32 l_hash = le32_to_cpu(l->xe_name_hash);
2566         u32 r_hash = le32_to_cpu(r->xe_name_hash);
2567
2568         if (l_hash > r_hash)
2569                 return 1;
2570         if (l_hash < r_hash)
2571                 return -1;
2572         return 0;
2573 }
2574
2575 static void swap_xe(void *a, void *b, int size)
2576 {
2577         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2578
2579         tmp = *l;
2580         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2581         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2582 }
2583
2584 /*
2585  * When the ocfs2_xattr_block is filled up, new bucket will be created
2586  * and all the xattr entries will be moved to the new bucket.
2587  * Note: we need to sort the entries since they are not saved in order
2588  * in the ocfs2_xattr_block.
2589  */
2590 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2591                                            struct buffer_head *xb_bh,
2592                                            struct buffer_head *xh_bh,
2593                                            struct buffer_head *data_bh)
2594 {
2595         int i, blocksize = inode->i_sb->s_blocksize;
2596         u16 offset, size, off_change;
2597         struct ocfs2_xattr_entry *xe;
2598         struct ocfs2_xattr_block *xb =
2599                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
2600         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2601         struct ocfs2_xattr_header *xh =
2602                                 (struct ocfs2_xattr_header *)xh_bh->b_data;
2603         u16 count = le16_to_cpu(xb_xh->xh_count);
2604         char *target = xh_bh->b_data, *src = xb_bh->b_data;
2605
2606         mlog(0, "cp xattr from block %llu to bucket %llu\n",
2607              (unsigned long long)xb_bh->b_blocknr,
2608              (unsigned long long)xh_bh->b_blocknr);
2609
2610         memset(xh_bh->b_data, 0, blocksize);
2611         if (data_bh)
2612                 memset(data_bh->b_data, 0, blocksize);
2613         /*
2614          * Since the xe_name_offset is based on ocfs2_xattr_header,
2615          * there is a offset change corresponding to the change of
2616          * ocfs2_xattr_header's position.
2617          */
2618         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2619         xe = &xb_xh->xh_entries[count - 1];
2620         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2621         size = blocksize - offset;
2622
2623         /* copy all the names and values. */
2624         if (data_bh)
2625                 target = data_bh->b_data;
2626         memcpy(target + offset, src + offset, size);
2627
2628         /* Init new header now. */
2629         xh->xh_count = xb_xh->xh_count;
2630         xh->xh_num_buckets = cpu_to_le16(1);
2631         xh->xh_name_value_len = cpu_to_le16(size);
2632         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2633
2634         /* copy all the entries. */
2635         target = xh_bh->b_data;
2636         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2637         size = count * sizeof(struct ocfs2_xattr_entry);
2638         memcpy(target + offset, (char *)xb_xh + offset, size);
2639
2640         /* Change the xe offset for all the xe because of the move. */
2641         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2642                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2643         for (i = 0; i < count; i++)
2644                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2645
2646         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2647              offset, size, off_change);
2648
2649         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2650              cmp_xe, swap_xe);
2651 }
2652
2653 /*
2654  * After we move xattr from block to index btree, we have to
2655  * update ocfs2_xattr_search to the new xe and base.
2656  *
2657  * When the entry is in xattr block, xattr_bh indicates the storage place.
2658  * While if the entry is in index b-tree, "bucket" indicates the
2659  * real place of the xattr.
2660  */
2661 static int ocfs2_xattr_update_xattr_search(struct inode *inode,
2662                                            struct ocfs2_xattr_search *xs,
2663                                            struct buffer_head *old_bh,
2664                                            struct buffer_head *new_bh)
2665 {
2666         int ret = 0;
2667         char *buf = old_bh->b_data;
2668         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2669         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2670         int i, blocksize = inode->i_sb->s_blocksize;
2671         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2672
2673         xs->bucket.bhs[0] = new_bh;
2674         get_bh(new_bh);
2675         xs->bucket.xh = (struct ocfs2_xattr_header *)xs->bucket.bhs[0]->b_data;
2676         xs->header = xs->bucket.xh;
2677
2678         xs->base = new_bh->b_data;
2679         xs->end = xs->base + inode->i_sb->s_blocksize;
2680
2681         if (!xs->not_found) {
2682                 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) {
2683                         ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2684                                         xs->bucket.bhs[0]->b_blocknr + 1,
2685                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2686                                         OCFS2_BH_CACHED, inode);
2687                         if (ret) {
2688                                 mlog_errno(ret);
2689                                 return ret;
2690                         }
2691
2692                         i = xs->here - old_xh->xh_entries;
2693                         xs->here = &xs->header->xh_entries[i];
2694                 }
2695         }
2696
2697         return ret;
2698 }
2699
2700 static int ocfs2_xattr_create_index_block(struct inode *inode,
2701                                           struct ocfs2_xattr_search *xs)
2702 {
2703         int ret, credits = OCFS2_SUBALLOC_ALLOC;
2704         u32 bit_off, len;
2705         u64 blkno;
2706         handle_t *handle;
2707         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2708         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2709         struct ocfs2_alloc_context *data_ac;
2710         struct buffer_head *xh_bh = NULL, *data_bh = NULL;
2711         struct buffer_head *xb_bh = xs->xattr_bh;
2712         struct ocfs2_xattr_block *xb =
2713                         (struct ocfs2_xattr_block *)xb_bh->b_data;
2714         struct ocfs2_xattr_tree_root *xr;
2715         u16 xb_flags = le16_to_cpu(xb->xb_flags);
2716         u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2717
2718         mlog(0, "create xattr index block for %llu\n",
2719              (unsigned long long)xb_bh->b_blocknr);
2720
2721         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2722
2723         ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2724         if (ret) {
2725                 mlog_errno(ret);
2726                 goto out;
2727         }
2728
2729         /*
2730          * XXX:
2731          * We can use this lock for now, and maybe move to a dedicated mutex
2732          * if performance becomes a problem later.
2733          */
2734         down_write(&oi->ip_alloc_sem);
2735
2736         /*
2737          * 3 more credits, one for xattr block update, one for the 1st block
2738          * of the new xattr bucket and one for the value/data.
2739          */
2740         credits += 3;
2741         handle = ocfs2_start_trans(osb, credits);
2742         if (IS_ERR(handle)) {
2743                 ret = PTR_ERR(handle);
2744                 mlog_errno(ret);
2745                 goto out_sem;
2746         }
2747
2748         ret = ocfs2_journal_access(handle, inode, xb_bh,
2749                                    OCFS2_JOURNAL_ACCESS_WRITE);
2750         if (ret) {
2751                 mlog_errno(ret);
2752                 goto out_commit;
2753         }
2754
2755         ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2756         if (ret) {
2757                 mlog_errno(ret);
2758                 goto out_commit;
2759         }
2760
2761         /*
2762          * The bucket may spread in many blocks, and
2763          * we will only touch the 1st block and the last block
2764          * in the whole bucket(one for entry and one for data).
2765          */
2766         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2767
2768         mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno);
2769
2770         xh_bh = sb_getblk(inode->i_sb, blkno);
2771         if (!xh_bh) {
2772                 ret = -EIO;
2773                 mlog_errno(ret);
2774                 goto out_commit;
2775         }
2776
2777         ocfs2_set_new_buffer_uptodate(inode, xh_bh);
2778
2779         ret = ocfs2_journal_access(handle, inode, xh_bh,
2780                                    OCFS2_JOURNAL_ACCESS_CREATE);
2781         if (ret) {
2782                 mlog_errno(ret);
2783                 goto out_commit;
2784         }
2785
2786         if (bpb > 1) {
2787                 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1);
2788                 if (!data_bh) {
2789                         ret = -EIO;
2790                         mlog_errno(ret);
2791                         goto out_commit;
2792                 }
2793
2794                 ocfs2_set_new_buffer_uptodate(inode, data_bh);
2795
2796                 ret = ocfs2_journal_access(handle, inode, data_bh,
2797                                            OCFS2_JOURNAL_ACCESS_CREATE);
2798                 if (ret) {
2799                         mlog_errno(ret);
2800                         goto out_commit;
2801                 }
2802         }
2803
2804         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh);
2805
2806         ocfs2_journal_dirty(handle, xh_bh);
2807         if (data_bh)
2808                 ocfs2_journal_dirty(handle, data_bh);
2809
2810         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
2811
2812         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2813         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2814                offsetof(struct ocfs2_xattr_block, xb_attrs));
2815
2816         xr = &xb->xb_attrs.xb_root;
2817         xr->xt_clusters = cpu_to_le32(1);
2818         xr->xt_last_eb_blk = 0;
2819         xr->xt_list.l_tree_depth = 0;
2820         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2821         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2822
2823         xr->xt_list.l_recs[0].e_cpos = 0;
2824         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2825         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2826
2827         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2828
2829         ret = ocfs2_journal_dirty(handle, xb_bh);
2830         if (ret) {
2831                 mlog_errno(ret);
2832                 goto out_commit;
2833         }
2834
2835 out_commit:
2836         ocfs2_commit_trans(osb, handle);
2837
2838 out_sem:
2839         up_write(&oi->ip_alloc_sem);
2840
2841 out:
2842         if (data_ac)
2843                 ocfs2_free_alloc_context(data_ac);
2844
2845         brelse(xh_bh);
2846         brelse(data_bh);
2847
2848         return ret;
2849 }
2850
2851 static int cmp_xe_offset(const void *a, const void *b)
2852 {
2853         const struct ocfs2_xattr_entry *l = a, *r = b;
2854         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2855         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2856
2857         if (l_name_offset < r_name_offset)
2858                 return 1;
2859         if (l_name_offset > r_name_offset)
2860                 return -1;
2861         return 0;
2862 }
2863
2864 /*
2865  * defrag a xattr bucket if we find that the bucket has some
2866  * holes beteen name/value pairs.
2867  * We will move all the name/value pairs to the end of the bucket
2868  * so that we can spare some space for insertion.
2869  */
2870 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2871                                      struct ocfs2_xattr_bucket *bucket)
2872 {
2873         int ret, i;
2874         size_t end, offset, len, value_len;
2875         struct ocfs2_xattr_header *xh;
2876         char *entries, *buf, *bucket_buf = NULL;
2877         u64 blkno = bucket->bhs[0]->b_blocknr;
2878         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2879         u16 xh_free_start;
2880         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2881         size_t blocksize = inode->i_sb->s_blocksize;
2882         handle_t *handle;
2883         struct buffer_head **bhs;
2884         struct ocfs2_xattr_entry *xe;
2885
2886         bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
2887                         GFP_NOFS);
2888         if (!bhs)
2889                 return -ENOMEM;
2890
2891         ret = ocfs2_read_blocks(osb, blkno, blk_per_bucket, bhs,
2892                                 OCFS2_BH_CACHED, inode);
2893         if (ret)
2894                 goto out;
2895
2896         /*
2897          * In order to make the operation more efficient and generic,
2898          * we copy all the blocks into a contiguous memory and do the
2899          * defragment there, so if anything is error, we will not touch
2900          * the real block.
2901          */
2902         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2903         if (!bucket_buf) {
2904                 ret = -EIO;
2905                 goto out;
2906         }
2907
2908         buf = bucket_buf;
2909         for (i = 0; i < blk_per_bucket; i++, buf += blocksize)
2910                 memcpy(buf, bhs[i]->b_data, blocksize);
2911
2912         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket);
2913         if (IS_ERR(handle)) {
2914                 ret = PTR_ERR(handle);
2915                 handle = NULL;
2916                 mlog_errno(ret);
2917                 goto out;
2918         }
2919
2920         for (i = 0; i < blk_per_bucket; i++) {
2921                 ret = ocfs2_journal_access(handle, inode, bhs[i],
2922                                            OCFS2_JOURNAL_ACCESS_WRITE);
2923                 if (ret < 0) {
2924                         mlog_errno(ret);
2925                         goto commit;
2926                 }
2927         }
2928
2929         xh = (struct ocfs2_xattr_header *)bucket_buf;
2930         entries = (char *)xh->xh_entries;
2931         xh_free_start = le16_to_cpu(xh->xh_free_start);
2932
2933         mlog(0, "adjust xattr bucket in %llu, count = %u, "
2934              "xh_free_start = %u, xh_name_value_len = %u.\n",
2935              blkno, le16_to_cpu(xh->xh_count), xh_free_start,
2936              le16_to_cpu(xh->xh_name_value_len));
2937
2938         /*
2939          * sort all the entries by their offset.
2940          * the largest will be the first, so that we can
2941          * move them to the end one by one.
2942          */
2943         sort(entries, le16_to_cpu(xh->xh_count),
2944              sizeof(struct ocfs2_xattr_entry),
2945              cmp_xe_offset, swap_xe);
2946
2947         /* Move all name/values to the end of the bucket. */
2948         xe = xh->xh_entries;
2949         end = OCFS2_XATTR_BUCKET_SIZE;
2950         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
2951                 offset = le16_to_cpu(xe->xe_name_offset);
2952                 if (ocfs2_xattr_is_local(xe))
2953                         value_len = OCFS2_XATTR_SIZE(
2954                                         le64_to_cpu(xe->xe_value_size));
2955                 else
2956                         value_len = OCFS2_XATTR_ROOT_SIZE;
2957                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
2958
2959                 /*
2960                  * We must make sure that the name/value pair
2961                  * exist in the same block. So adjust end to
2962                  * the previous block end if needed.
2963                  */
2964                 if (((end - len) / blocksize !=
2965                         (end - 1) / blocksize))
2966                         end = end - end % blocksize;
2967
2968                 if (end > offset + len) {
2969                         memmove(bucket_buf + end - len,
2970                                 bucket_buf + offset, len);
2971                         xe->xe_name_offset = cpu_to_le16(end - len);
2972                 }
2973
2974                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
2975                                 "bucket %llu\n", (unsigned long long)blkno);
2976
2977                 end -= len;
2978         }
2979
2980         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
2981                         "bucket %llu\n", (unsigned long long)blkno);
2982
2983         if (xh_free_start == end)
2984                 goto commit;
2985
2986         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
2987         xh->xh_free_start = cpu_to_le16(end);
2988
2989         /* sort the entries by their name_hash. */
2990         sort(entries, le16_to_cpu(xh->xh_count),
2991              sizeof(struct ocfs2_xattr_entry),
2992              cmp_xe, swap_xe);
2993
2994         buf = bucket_buf;
2995         for (i = 0; i < blk_per_bucket; i++, buf += blocksize) {
2996                 memcpy(bhs[i]->b_data, buf, blocksize);
2997                 ocfs2_journal_dirty(handle, bhs[i]);
2998         }
2999
3000 commit:
3001         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3002 out:
3003
3004         if (bhs) {
3005                 for (i = 0; i < blk_per_bucket; i++)
3006                         brelse(bhs[i]);
3007         }
3008         kfree(bhs);
3009
3010         kfree(bucket_buf);
3011         return ret;
3012 }
3013
3014 /*
3015  * Move half nums of the xattr bucket in the previous cluster to this new
3016  * cluster. We only touch the last cluster of the previous extend record.
3017  *
3018  * first_bh is the first buffer_head of a series of bucket in the same
3019  * extent rec and header_bh is the header of one bucket in this cluster.
3020  * They will be updated if we move the data header_bh contains to the new
3021  * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3022  */
3023 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3024                                                handle_t *handle,
3025                                                struct buffer_head **first_bh,
3026                                                struct buffer_head **header_bh,
3027                                                u64 new_blkno,
3028                                                u64 prev_blkno,
3029                                                u32 num_clusters,
3030                                                u32 *first_hash)
3031 {
3032         int i, ret, credits;
3033         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3034         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3035         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3036         int blocksize = inode->i_sb->s_blocksize;
3037         struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3038         struct ocfs2_xattr_header *new_xh;
3039         struct ocfs2_xattr_header *xh =
3040                         (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3041
3042         BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3043         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3044
3045         prev_bh = *first_bh;
3046         get_bh(prev_bh);
3047         xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3048
3049         prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3050
3051         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3052              prev_blkno, new_blkno);
3053
3054         /*
3055          * We need to update the 1st half of the new cluster and
3056          * 1 more for the update of the 1st bucket of the previous
3057          * extent record.
3058          */
3059         credits = bpc / 2 + 1;
3060         ret = ocfs2_extend_trans(handle, credits);
3061         if (ret) {
3062                 mlog_errno(ret);
3063                 goto out;
3064         }
3065
3066         ret = ocfs2_journal_access(handle, inode, prev_bh,
3067                                    OCFS2_JOURNAL_ACCESS_WRITE);
3068         if (ret) {
3069                 mlog_errno(ret);
3070                 goto out;
3071         }
3072
3073         for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3074                 old_bh = new_bh = NULL;
3075                 new_bh = sb_getblk(inode->i_sb, new_blkno);
3076                 if (!new_bh) {
3077                         ret = -EIO;
3078                         mlog_errno(ret);
3079                         goto out;
3080                 }
3081
3082                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3083
3084                 ret = ocfs2_journal_access(handle, inode, new_bh,
3085                                            OCFS2_JOURNAL_ACCESS_CREATE);
3086                 if (ret < 0) {
3087                         mlog_errno(ret);
3088                         brelse(new_bh);
3089                         goto out;
3090                 }
3091
3092                 ret = ocfs2_read_block(osb, prev_blkno,
3093                                        &old_bh, OCFS2_BH_CACHED, inode);
3094                 if (ret < 0) {
3095                         mlog_errno(ret);
3096                         brelse(new_bh);
3097                         goto out;
3098                 }
3099
3100                 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3101
3102                 if (i == 0) {
3103                         new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3104                         new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3105
3106                         if (first_hash)
3107                                 *first_hash = le32_to_cpu(
3108                                         new_xh->xh_entries[0].xe_name_hash);
3109                         new_first_bh = new_bh;
3110                         get_bh(new_first_bh);
3111                 }
3112
3113                 ocfs2_journal_dirty(handle, new_bh);
3114
3115                 if (*header_bh == old_bh) {
3116                         brelse(*header_bh);
3117                         *header_bh = new_bh;
3118                         get_bh(*header_bh);
3119
3120                         brelse(*first_bh);
3121                         *first_bh = new_first_bh;
3122                         get_bh(*first_bh);
3123                 }
3124                 brelse(new_bh);
3125                 brelse(old_bh);
3126         }
3127
3128         le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3129
3130         ocfs2_journal_dirty(handle, prev_bh);
3131 out:
3132         brelse(prev_bh);
3133         brelse(new_first_bh);
3134         return ret;
3135 }
3136
3137 static int ocfs2_read_xattr_bucket(struct inode *inode,
3138                                    u64 blkno,
3139                                    struct buffer_head **bhs,
3140                                    int new)
3141 {
3142         int ret = 0;
3143         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3144
3145         if (!new)
3146                 return ocfs2_read_blocks(OCFS2_SB(inode->i_sb), blkno,
3147                                          blk_per_bucket, bhs,
3148                                          OCFS2_BH_CACHED, inode);
3149
3150         for (i = 0; i < blk_per_bucket; i++) {
3151                 bhs[i] = sb_getblk(inode->i_sb, blkno + i);
3152                 if (bhs[i] == NULL) {
3153                         ret = -EIO;
3154                         mlog_errno(ret);
3155                         break;
3156                 }
3157                 ocfs2_set_new_buffer_uptodate(inode, bhs[i]);
3158         }
3159
3160         return ret;
3161 }
3162
3163 /*
3164  * Move half num of the xattrs in old bucket(blk) to new bucket(new_blk).
3165  * first_hash will record the 1st hash of the new bucket.
3166  */
3167 static int ocfs2_half_xattr_bucket(struct inode *inode,
3168                                    handle_t *handle,
3169                                    u64 blk,
3170                                    u64 new_blk,
3171                                    u32 *first_hash,
3172                                    int new_bucket_head)
3173 {
3174         int ret, i;
3175         u16 count, start, len, name_value_len, xe_len, name_offset;
3176         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3177         struct buffer_head **s_bhs, **t_bhs = NULL;
3178         struct ocfs2_xattr_header *xh;
3179         struct ocfs2_xattr_entry *xe;
3180         int blocksize = inode->i_sb->s_blocksize;
3181
3182         mlog(0, "move half of xattrs from bucket %llu to %llu\n",
3183              blk, new_blk);
3184
3185         s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3186         if (!s_bhs)
3187                 return -ENOMEM;
3188
3189         ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0);
3190         if (ret) {
3191                 mlog_errno(ret);
3192                 goto out;
3193         }
3194
3195         ret = ocfs2_journal_access(handle, inode, s_bhs[0],
3196                                    OCFS2_JOURNAL_ACCESS_WRITE);
3197         if (ret) {
3198                 mlog_errno(ret);
3199                 goto out;
3200         }
3201
3202         t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3203         if (!t_bhs) {
3204                 ret = -ENOMEM;
3205                 goto out;
3206         }
3207
3208         ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head);
3209         if (ret) {
3210                 mlog_errno(ret);
3211                 goto out;
3212         }
3213
3214         for (i = 0; i < blk_per_bucket; i++) {
3215                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3216                                            OCFS2_JOURNAL_ACCESS_CREATE);
3217                 if (ret) {
3218                         mlog_errno(ret);
3219                         goto out;
3220                 }
3221         }
3222
3223         /* copy the whole bucket to the new first. */
3224         for (i = 0; i < blk_per_bucket; i++)
3225                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3226
3227         /* update the new bucket. */
3228         xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3229         count = le16_to_cpu(xh->xh_count);
3230         start = count / 2;
3231
3232         /*
3233          * Calculate the total name/value len and xh_free_start for
3234          * the old bucket first.
3235          */
3236         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3237         name_value_len = 0;
3238         for (i = 0; i < start; i++) {
3239                 xe = &xh->xh_entries[i];
3240                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3241                 if (ocfs2_xattr_is_local(xe))
3242                         xe_len +=
3243                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3244                 else
3245                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3246                 name_value_len += xe_len;
3247                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3248                         name_offset = le16_to_cpu(xe->xe_name_offset);
3249         }
3250
3251         /*
3252          * Now begin the modification to the new bucket.
3253          *
3254          * In the new bucket, We just move the xattr entry to the beginning
3255          * and don't touch the name/value. So there will be some holes in the
3256          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3257          * called.
3258          */
3259         xe = &xh->xh_entries[start];
3260         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3261         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3262              (int)((char *)xe - (char *)xh),
3263              (int)((char *)xh->xh_entries - (char *)xh));
3264         memmove((char *)xh->xh_entries, (char *)xe, len);
3265         xe = &xh->xh_entries[count - start];
3266         len = sizeof(struct ocfs2_xattr_entry) * start;
3267         memset((char *)xe, 0, len);
3268
3269         le16_add_cpu(&xh->xh_count, -start);
3270         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3271
3272         /* Calculate xh_free_start for the new bucket. */
3273         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3274         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3275                 xe = &xh->xh_entries[i];
3276                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3277                 if (ocfs2_xattr_is_local(xe))
3278                         xe_len +=
3279                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3280                 else
3281                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3282                 if (le16_to_cpu(xe->xe_name_offset) <
3283                     le16_to_cpu(xh->xh_free_start))
3284                         xh->xh_free_start = xe->xe_name_offset;
3285         }
3286
3287         /* set xh->xh_num_buckets for the new xh. */
3288         if (new_bucket_head)
3289                 xh->xh_num_buckets = cpu_to_le16(1);
3290         else
3291                 xh->xh_num_buckets = 0;
3292
3293         for (i = 0; i < blk_per_bucket; i++) {
3294                 ocfs2_journal_dirty(handle, t_bhs[i]);
3295                 if (ret)
3296                         mlog_errno(ret);
3297         }
3298
3299         /* store the first_hash of the new bucket. */
3300         if (first_hash)
3301                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3302
3303         /*
3304          * Now only update the 1st block of the old bucket.
3305          * Please note that the entry has been sorted already above.
3306          */
3307         xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3308         memset(&xh->xh_entries[start], 0,
3309                sizeof(struct ocfs2_xattr_entry) * (count - start));
3310         xh->xh_count = cpu_to_le16(start);
3311         xh->xh_free_start = cpu_to_le16(name_offset);
3312         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3313
3314         ocfs2_journal_dirty(handle, s_bhs[0]);
3315         if (ret)
3316                 mlog_errno(ret);
3317
3318 out:
3319         if (s_bhs) {
3320                 for (i = 0; i < blk_per_bucket; i++)
3321                         brelse(s_bhs[i]);
3322         }
3323         kfree(s_bhs);
3324
3325         if (t_bhs) {
3326                 for (i = 0; i < blk_per_bucket; i++)
3327                         brelse(t_bhs[i]);
3328         }
3329         kfree(t_bhs);
3330
3331         return ret;
3332 }
3333
3334 /*
3335  * Copy xattr from one bucket to another bucket.
3336  *
3337  * The caller must make sure that the journal transaction
3338  * has enough space for journaling.
3339  */
3340 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3341                                  handle_t *handle,
3342                                  u64 s_blkno,
3343                                  u64 t_blkno,
3344                                  int t_is_new)
3345 {
3346         int ret, i;
3347         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3348         int blocksize = inode->i_sb->s_blocksize;
3349         struct buffer_head **s_bhs, **t_bhs = NULL;
3350
3351         BUG_ON(s_blkno == t_blkno);
3352
3353         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3354              s_blkno, t_blkno, t_is_new);
3355
3356         s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3357                         GFP_NOFS);
3358         if (!s_bhs)
3359                 return -ENOMEM;
3360
3361         ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0);
3362         if (ret)
3363                 goto out;
3364
3365         t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3366                         GFP_NOFS);
3367         if (!t_bhs) {
3368                 ret = -ENOMEM;
3369                 goto out;
3370         }
3371
3372         ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new);
3373         if (ret)
3374                 goto out;
3375
3376         for (i = 0; i < blk_per_bucket; i++) {
3377                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3378                                            OCFS2_JOURNAL_ACCESS_WRITE);
3379                 if (ret)
3380                         goto out;
3381         }
3382
3383         for (i = 0; i < blk_per_bucket; i++) {
3384                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3385                 ocfs2_journal_dirty(handle, t_bhs[i]);
3386         }
3387
3388 out:
3389         if (s_bhs) {
3390                 for (i = 0; i < blk_per_bucket; i++)
3391                         brelse(s_bhs[i]);
3392         }
3393         kfree(s_bhs);
3394
3395         if (t_bhs) {
3396                 for (i = 0; i < blk_per_bucket; i++)
3397                         brelse(t_bhs[i]);
3398         }
3399         kfree(t_bhs);
3400
3401         return ret;
3402 }
3403
3404 /*
3405  * Copy one xattr cluster from src_blk to to_blk.
3406  * The to_blk will become the first bucket header of the cluster, so its
3407  * xh_num_buckets will be initialized as the bucket num in the cluster.
3408  */
3409 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3410                                   handle_t *handle,
3411                                   struct buffer_head *first_bh,
3412                                   u64 src_blk,
3413                                   u64 to_blk,
3414                                   u32 *first_hash)
3415 {
3416         int i, ret, credits;
3417         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3418         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3419         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3420         struct buffer_head *bh = NULL;
3421         struct ocfs2_xattr_header *xh;
3422         u64 to_blk_start = to_blk;
3423
3424         mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk);
3425
3426         /*
3427          * We need to update the new cluster and 1 more for the update of
3428          * the 1st bucket of the previous extent rec.
3429          */
3430         credits = bpc + 1;
3431         ret = ocfs2_extend_trans(handle, credits);
3432         if (ret) {
3433                 mlog_errno(ret);
3434                 goto out;
3435         }
3436
3437         ret = ocfs2_journal_access(handle, inode, first_bh,
3438                                    OCFS2_JOURNAL_ACCESS_WRITE);
3439         if (ret) {
3440                 mlog_errno(ret);
3441                 goto out;
3442         }
3443
3444         for (i = 0; i < num_buckets; i++) {
3445                 ret = ocfs2_cp_xattr_bucket(inode, handle,
3446                                             src_blk, to_blk, 1);
3447                 if (ret) {
3448                         mlog_errno(ret);
3449                         goto out;
3450                 }
3451
3452                 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3453                 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3454         }
3455
3456         /* update the old bucket header. */
3457         xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3458         le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3459
3460         ocfs2_journal_dirty(handle, first_bh);
3461
3462         /* update the new bucket header. */
3463         ret = ocfs2_read_block(osb, to_blk_start, &bh, OCFS2_BH_CACHED, inode);
3464         if (ret < 0) {
3465                 mlog_errno(ret);
3466                 goto out;
3467         }
3468
3469         ret = ocfs2_journal_access(handle, inode, bh,
3470                                    OCFS2_JOURNAL_ACCESS_WRITE);
3471         if (ret) {
3472                 mlog_errno(ret);
3473                 goto out;
3474         }
3475
3476         xh = (struct ocfs2_xattr_header *)bh->b_data;
3477         xh->xh_num_buckets = cpu_to_le16(num_buckets);
3478
3479         ocfs2_journal_dirty(handle, bh);
3480
3481         if (first_hash)
3482                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3483 out:
3484         brelse(bh);
3485         return ret;
3486 }
3487
3488 /*
3489  * Move half of the xattrs in this cluster to the new cluster.
3490  * This function should only be called when bucket size == cluster size.
3491  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3492  */
3493 static int ocfs2_half_xattr_cluster(struct inode *inode,
3494                                     handle_t *handle,
3495                                     u64 prev_blk,
3496                                     u64 new_blk,
3497                                     u32 *first_hash)
3498 {
3499         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3500         int ret, credits = 2 * blk_per_bucket;
3501
3502         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3503
3504         ret = ocfs2_extend_trans(handle, credits);
3505         if (ret) {
3506                 mlog_errno(ret);
3507                 return ret;
3508         }
3509
3510         /* Move half of the xattr in start_blk to the next bucket. */
3511         return  ocfs2_half_xattr_bucket(inode, handle, prev_blk,
3512                                         new_blk, first_hash, 1);
3513 }
3514
3515 /*
3516  * Move some xattrs from the old cluster to the new one since they are not
3517  * contiguous in ocfs2 xattr tree.
3518  *
3519  * new_blk starts a new separate cluster, and we will move some xattrs from
3520  * prev_blk to it. v_start will be set as the first name hash value in this
3521  * new cluster so that it can be used as e_cpos during tree insertion and
3522  * don't collide with our original b-tree operations. first_bh and header_bh
3523  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3524  * to extend the insert bucket.
3525  *
3526  * The problem is how much xattr should we move to the new one and when should
3527  * we update first_bh and header_bh?
3528  * 1. If cluster size > bucket size, that means the previous cluster has more
3529  *    than 1 bucket, so just move half nums of bucket into the new cluster and
3530  *    update the first_bh and header_bh if the insert bucket has been moved
3531  *    to the new cluster.
3532  * 2. If cluster_size == bucket_size:
3533  *    a) If the previous extent rec has more than one cluster and the insert
3534  *       place isn't in the last cluster, copy the entire last cluster to the
3535  *       new one. This time, we don't need to upate the first_bh and header_bh
3536  *       since they will not be moved into the new cluster.
3537  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
3538  *       the new one. And we set the extend flag to zero if the insert place is
3539  *       moved into the new allocated cluster since no extend is needed.
3540  */
3541 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3542                                             handle_t *handle,
3543                                             struct buffer_head **first_bh,
3544                                             struct buffer_head **header_bh,
3545                                             u64 new_blk,
3546                                             u64 prev_blk,
3547                                             u32 prev_clusters,
3548                                             u32 *v_start,
3549                                             int *extend)
3550 {
3551         int ret = 0;
3552         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3553
3554         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3555              prev_blk, prev_clusters, new_blk);
3556
3557         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3558                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3559                                                           handle,
3560                                                           first_bh,
3561                                                           header_bh,
3562                                                           new_blk,
3563                                                           prev_blk,
3564                                                           prev_clusters,
3565                                                           v_start);
3566         else {
3567                 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3568
3569                 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3570                         ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3571                                                      last_blk, new_blk,
3572                                                      v_start);
3573                 else {
3574                         ret = ocfs2_half_xattr_cluster(inode, handle,
3575                                                        last_blk, new_blk,
3576                                                        v_start);
3577
3578                         if ((*header_bh)->b_blocknr == last_blk && extend)
3579                                 *extend = 0;
3580                 }
3581         }
3582
3583         return ret;
3584 }
3585
3586 /*
3587  * Add a new cluster for xattr storage.
3588  *
3589  * If the new cluster is contiguous with the previous one, it will be
3590  * appended to the same extent record, and num_clusters will be updated.
3591  * If not, we will insert a new extent for it and move some xattrs in
3592  * the last cluster into the new allocated one.
3593  * We also need to limit the maximum size of a btree leaf, otherwise we'll
3594  * lose the benefits of hashing because we'll have to search large leaves.
3595  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3596  * if it's bigger).
3597  *
3598  * first_bh is the first block of the previous extent rec and header_bh
3599  * indicates the bucket we will insert the new xattrs. They will be updated
3600  * when the header_bh is moved into the new cluster.
3601  */
3602 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3603                                        struct buffer_head *root_bh,
3604                                        struct buffer_head **first_bh,
3605                                        struct buffer_head **header_bh,
3606                                        u32 *num_clusters,
3607                                        u32 prev_cpos,
3608                                        u64 prev_blkno,
3609                                        int *extend)
3610 {
3611         int ret, credits;
3612         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3613         u32 prev_clusters = *num_clusters;
3614         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3615         u64 block;
3616         handle_t *handle = NULL;
3617         struct ocfs2_alloc_context *data_ac = NULL;
3618         struct ocfs2_alloc_context *meta_ac = NULL;
3619         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3620         struct ocfs2_extent_tree et;
3621
3622         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3623              "previous xattr blkno = %llu\n",
3624              (unsigned long long)OCFS2_I(inode)->ip_blkno,
3625              prev_cpos, prev_blkno);
3626
3627         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
3628
3629         ret = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
3630                                     &data_ac, &meta_ac);
3631         if (ret) {
3632                 mlog_errno(ret);
3633                 goto leave;
3634         }
3635
3636         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
3637                                             clusters_to_add);
3638         handle = ocfs2_start_trans(osb, credits);
3639         if (IS_ERR(handle)) {
3640                 ret = PTR_ERR(handle);
3641                 handle = NULL;
3642                 mlog_errno(ret);
3643                 goto leave;
3644         }
3645
3646         ret = ocfs2_journal_access(handle, inode, root_bh,
3647                                    OCFS2_JOURNAL_ACCESS_WRITE);
3648         if (ret < 0) {
3649                 mlog_errno(ret);
3650                 goto leave;
3651         }
3652
3653         ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3654                                      clusters_to_add, &bit_off, &num_bits);
3655         if (ret < 0) {
3656                 if (ret != -ENOSPC)
3657                         mlog_errno(ret);
3658                 goto leave;
3659         }
3660
3661         BUG_ON(num_bits > clusters_to_add);
3662
3663         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3664         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3665              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3666
3667         if (prev_blkno + prev_clusters * bpc == block &&
3668             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3669              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3670                 /*
3671                  * If this cluster is contiguous with the old one and
3672                  * adding this new cluster, we don't surpass the limit of
3673                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3674                  * initialized and used like other buckets in the previous
3675                  * cluster.
3676                  * So add it as a contiguous one. The caller will handle
3677                  * its init process.
3678                  */
3679                 v_start = prev_cpos + prev_clusters;
3680                 *num_clusters = prev_clusters + num_bits;
3681                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3682                      num_bits);
3683         } else {
3684                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3685                                                        handle,
3686                                                        first_bh,
3687                                                        header_bh,
3688                                                        block,
3689                                                        prev_blkno,
3690                                                        prev_clusters,
3691                                                        &v_start,
3692                                                        extend);
3693                 if (ret) {
3694                         mlog_errno(ret);
3695                         goto leave;
3696                 }
3697         }
3698
3699         if (handle->h_buffer_credits < credits) {
3700                 /*
3701                  * The journal has been restarted before, and don't
3702                  * have enough space for the insertion, so extend it
3703                  * here.
3704                  */
3705                 ret = ocfs2_extend_trans(handle, credits);
3706                 if (ret) {
3707                         mlog_errno(ret);
3708                         goto leave;
3709                 }
3710         }
3711         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3712              num_bits, block, v_start);
3713         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
3714                                   num_bits, 0, meta_ac);
3715         if (ret < 0) {
3716                 mlog_errno(ret);
3717                 goto leave;
3718         }
3719
3720         ret = ocfs2_journal_dirty(handle, root_bh);
3721         if (ret < 0) {
3722                 mlog_errno(ret);
3723                 goto leave;
3724         }
3725
3726 leave:
3727         if (handle)
3728                 ocfs2_commit_trans(osb, handle);
3729         if (data_ac)
3730                 ocfs2_free_alloc_context(data_ac);
3731         if (meta_ac)
3732                 ocfs2_free_alloc_context(meta_ac);
3733
3734         return ret;
3735 }
3736
3737 /*
3738  * Extend a new xattr bucket and move xattrs to the end one by one until
3739  * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3740  */
3741 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3742                                      struct buffer_head *first_bh,
3743                                      struct buffer_head *start_bh,
3744                                      u32 num_clusters)
3745 {
3746         int ret, credits;
3747         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3748         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3749         u64 start_blk = start_bh->b_blocknr, end_blk;
3750         u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3751         handle_t *handle;
3752         struct ocfs2_xattr_header *first_xh =
3753                                 (struct ocfs2_xattr_header *)first_bh->b_data;
3754         u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3755
3756         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3757              "from %llu, len = %u\n", start_blk,
3758              (unsigned long long)first_bh->b_blocknr, num_clusters);
3759
3760         BUG_ON(bucket >= num_buckets);
3761
3762         end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3763
3764         /*
3765          * We will touch all the buckets after the start_bh(include it).
3766          * Add one more bucket and modify the first_bh.
3767          */
3768         credits = end_blk - start_blk + 2 * blk_per_bucket + 1;
3769         handle = ocfs2_start_trans(osb, credits);
3770         if (IS_ERR(handle)) {
3771                 ret = PTR_ERR(handle);
3772                 handle = NULL;
3773                 mlog_errno(ret);
3774                 goto out;
3775         }
3776
3777         ret = ocfs2_journal_access(handle, inode, first_bh,
3778                                    OCFS2_JOURNAL_ACCESS_WRITE);
3779         if (ret) {
3780                 mlog_errno(ret);
3781                 goto commit;
3782         }
3783
3784         while (end_blk != start_blk) {
3785                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3786                                             end_blk + blk_per_bucket, 0);
3787                 if (ret)
3788                         goto commit;
3789                 end_blk -= blk_per_bucket;
3790         }
3791
3792         /* Move half of the xattr in start_blk to the next bucket. */
3793         ret = ocfs2_half_xattr_bucket(inode, handle, start_blk,
3794                                       start_blk + blk_per_bucket, NULL, 0);
3795
3796         le16_add_cpu(&first_xh->xh_num_buckets, 1);
3797         ocfs2_journal_dirty(handle, first_bh);
3798
3799 commit:
3800         ocfs2_commit_trans(osb, handle);
3801 out:
3802         return ret;
3803 }
3804
3805 /*
3806  * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3807  * xb_bh is the ocfs2_xattr_block.
3808  * We will move all the buckets starting from header_bh to the next place. As
3809  * for this one, half num of its xattrs will be moved to the next one.
3810  *
3811  * We will allocate a new cluster if current cluster is full and adjust
3812  * header_bh and first_bh if the insert place is moved to the new cluster.
3813  */
3814 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3815                                       struct buffer_head *xb_bh,
3816                                       struct buffer_head *header_bh)
3817 {
3818         struct ocfs2_xattr_header *first_xh = NULL;
3819         struct buffer_head *first_bh = NULL;
3820         struct ocfs2_xattr_block *xb =
3821                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3822         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3823         struct ocfs2_extent_list *el = &xb_root->xt_list;
3824         struct ocfs2_xattr_header *xh =
3825                         (struct ocfs2_xattr_header *)header_bh->b_data;
3826         u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3827         struct super_block *sb = inode->i_sb;
3828         struct ocfs2_super *osb = OCFS2_SB(sb);
3829         int ret, num_buckets, extend = 1;
3830         u64 p_blkno;
3831         u32 e_cpos, num_clusters;
3832
3833         mlog(0, "Add new xattr bucket starting form %llu\n",
3834              (unsigned long long)header_bh->b_blocknr);
3835
3836         /*
3837          * Add refrence for header_bh here because it may be
3838          * changed in ocfs2_add_new_xattr_cluster and we need
3839          * to free it in the end.
3840          */
3841         get_bh(header_bh);
3842
3843         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3844                                   &num_clusters, el);
3845         if (ret) {
3846                 mlog_errno(ret);
3847                 goto out;
3848         }
3849
3850         ret = ocfs2_read_block(osb, p_blkno,
3851                                &first_bh, OCFS2_BH_CACHED, inode);
3852         if (ret) {
3853                 mlog_errno(ret);
3854                 goto out;
3855         }
3856
3857         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3858         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3859
3860         if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3861                 ret = ocfs2_add_new_xattr_cluster(inode,
3862                                                   xb_bh,
3863                                                   &first_bh,
3864                                                   &header_bh,
3865                                                   &num_clusters,
3866                                                   e_cpos,
3867                                                   p_blkno,
3868                                                   &extend);
3869                 if (ret) {
3870                         mlog_errno(ret);
3871                         goto out;
3872                 }
3873         }
3874
3875         if (extend)
3876                 ret = ocfs2_extend_xattr_bucket(inode,
3877                                                 first_bh,
3878                                                 header_bh,
3879                                                 num_clusters);
3880         if (ret)
3881                 mlog_errno(ret);
3882 out:
3883         brelse(first_bh);
3884         brelse(header_bh);
3885         return ret;
3886 }
3887
3888 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3889                                         struct ocfs2_xattr_bucket *bucket,
3890                                         int offs)
3891 {
3892         int block_off = offs >> inode->i_sb->s_blocksize_bits;
3893
3894         offs = offs % inode->i_sb->s_blocksize;
3895         return bucket->bhs[block_off]->b_data + offs;
3896 }
3897
3898 /*
3899  * Handle the normal xattr set, including replace, delete and new.
3900  *
3901  * Note: "local" indicates the real data's locality. So we can't
3902  * just its bucket locality by its length.
3903  */
3904 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3905                                          struct ocfs2_xattr_info *xi,
3906                                          struct ocfs2_xattr_search *xs,
3907                                          u32 name_hash,
3908                                          int local)
3909 {
3910         struct ocfs2_xattr_entry *last, *xe;
3911         int name_len = strlen(xi->name);
3912         struct ocfs2_xattr_header *xh = xs->header;
3913         u16 count = le16_to_cpu(xh->xh_count), start;
3914         size_t blocksize = inode->i_sb->s_blocksize;
3915         char *val;
3916         size_t offs, size, new_size;
3917
3918         last = &xh->xh_entries[count];
3919         if (!xs->not_found) {
3920                 xe = xs->here;
3921                 offs = le16_to_cpu(xe->xe_name_offset);
3922                 if (ocfs2_xattr_is_local(xe))
3923                         size = OCFS2_XATTR_SIZE(name_len) +
3924                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3925                 else
3926                         size = OCFS2_XATTR_SIZE(name_len) +
3927                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3928
3929                 /*
3930                  * If the new value will be stored outside, xi->value has been
3931                  * initalized as an empty ocfs2_xattr_value_root, and the same
3932                  * goes with xi->value_len, so we can set new_size safely here.
3933                  * See ocfs2_xattr_set_in_bucket.
3934                  */
3935                 new_size = OCFS2_XATTR_SIZE(name_len) +
3936                            OCFS2_XATTR_SIZE(xi->value_len);
3937
3938                 le16_add_cpu(&xh->xh_name_value_len, -size);
3939                 if (xi->value) {
3940                         if (new_size > size)
3941                                 goto set_new_name_value;
3942
3943                         /* Now replace the old value with new one. */
3944                         if (local)
3945                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
3946                         else
3947                                 xe->xe_value_size = 0;
3948
3949                         val = ocfs2_xattr_bucket_get_val(inode,
3950                                                          &xs->bucket, offs);
3951                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
3952                                size - OCFS2_XATTR_SIZE(name_len));
3953                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
3954                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
3955                                        xi->value, xi->value_len);
3956
3957                         le16_add_cpu(&xh->xh_name_value_len, new_size);
3958                         ocfs2_xattr_set_local(xe, local);
3959                         return;
3960                 } else {
3961                         /*
3962                          * Remove the old entry if there is more than one.
3963                          * We don't remove the last entry so that we can
3964                          * use it to indicate the hash value of the empty
3965                          * bucket.
3966                          */
3967                         last -= 1;
3968                         le16_add_cpu(&xh->xh_count, -1);
3969                         if (xh->xh_count) {
3970                                 memmove(xe, xe + 1,
3971                                         (void *)last - (void *)xe);
3972                                 memset(last, 0,
3973                                        sizeof(struct ocfs2_xattr_entry));
3974                         } else
3975                                 xh->xh_free_start =
3976                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3977
3978                         return;
3979                 }
3980         } else {
3981                 /* find a new entry for insert. */
3982                 int low = 0, high = count - 1, tmp;
3983                 struct ocfs2_xattr_entry *tmp_xe;
3984
3985                 while (low <= high && count) {
3986                         tmp = (low + high) / 2;
3987                         tmp_xe = &xh->xh_entries[tmp];
3988
3989                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
3990                                 low = tmp + 1;
3991                         else if (name_hash <
3992                                  le32_to_cpu(tmp_xe->xe_name_hash))
3993                                 high = tmp - 1;
3994                         else {
3995                                 low = tmp;
3996                                 break;
3997                         }
3998                 }
3999
4000                 xe = &xh->xh_entries[low];
4001                 if (low != count)
4002                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4003
4004                 le16_add_cpu(&xh->xh_count, 1);
4005                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4006                 xe->xe_name_hash = cpu_to_le32(name_hash);
4007                 xe->xe_name_len = name_len;
4008                 ocfs2_xattr_set_type(xe, xi->name_index);
4009         }
4010
4011 set_new_name_value:
4012         /* Insert the new name+value. */
4013         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4014
4015         /*
4016          * We must make sure that the name/value pair
4017          * exists in the same block.
4018          */
4019         offs = le16_to_cpu(xh->xh_free_start);
4020         start = offs - size;
4021
4022         if (start >> inode->i_sb->s_blocksize_bits !=
4023             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4024                 offs = offs - offs % blocksize;
4025                 xh->xh_free_start = cpu_to_le16(offs);
4026         }
4027
4028         val = ocfs2_xattr_bucket_get_val(inode,
4029                                          &xs->bucket, offs - size);
4030         xe->xe_name_offset = cpu_to_le16(offs - size);
4031
4032         memset(val, 0, size);
4033         memcpy(val, xi->name, name_len);
4034         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4035
4036         xe->xe_value_size = cpu_to_le64(xi->value_len);
4037         ocfs2_xattr_set_local(xe, local);
4038         xs->here = xe;
4039         le16_add_cpu(&xh->xh_free_start, -size);
4040         le16_add_cpu(&xh->xh_name_value_len, size);
4041
4042         return;
4043 }
4044
4045 static int ocfs2_xattr_bucket_handle_journal(struct inode *inode,
4046                                              handle_t *handle,
4047                                              struct ocfs2_xattr_search *xs,
4048                                              struct buffer_head **bhs,
4049                                              u16 bh_num)
4050 {
4051         int ret = 0, off, block_off;
4052         struct ocfs2_xattr_entry *xe = xs->here;
4053
4054         /*
4055          * First calculate all the blocks we should journal_access
4056          * and journal_dirty. The first block should always be touched.
4057          */
4058         ret = ocfs2_journal_dirty(handle, bhs[0]);
4059         if (ret)
4060                 mlog_errno(ret);
4061
4062         /* calc the data. */
4063         off = le16_to_cpu(xe->xe_name_offset);
4064         block_off = off >> inode->i_sb->s_blocksize_bits;
4065         ret = ocfs2_journal_dirty(handle, bhs[block_off]);
4066         if (ret)
4067                 mlog_errno(ret);
4068
4069         return ret;
4070 }
4071
4072 /*
4073  * Set the xattr entry in the specified bucket.
4074  * The bucket is indicated by xs->bucket and it should have the enough
4075  * space for the xattr insertion.
4076  */
4077 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4078                                            struct ocfs2_xattr_info *xi,
4079                                            struct ocfs2_xattr_search *xs,
4080                                            u32 name_hash,
4081                                            int local)
4082 {
4083         int i, ret;
4084         handle_t *handle = NULL;
4085         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4086         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4087
4088         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4089              (unsigned long)xi->value_len, xi->name_index,
4090              (unsigned long long)xs->bucket.bhs[0]->b_blocknr);
4091
4092         if (!xs->bucket.bhs[1]) {
4093                 ret = ocfs2_read_blocks(osb,
4094                                         xs->bucket.bhs[0]->b_blocknr + 1,
4095                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
4096                                         OCFS2_BH_CACHED, inode);
4097                 if (ret) {
4098                         mlog_errno(ret);
4099                         goto out;
4100                 }
4101         }
4102
4103         handle = ocfs2_start_trans(osb, blk_per_bucket);
4104         if (IS_ERR(handle)) {
4105                 ret = PTR_ERR(handle);
4106                 handle = NULL;
4107                 mlog_errno(ret);
4108                 goto out;
4109         }
4110
4111         for (i = 0; i < blk_per_bucket; i++) {
4112                 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[i],
4113                                            OCFS2_JOURNAL_ACCESS_WRITE);
4114                 if (ret < 0) {
4115                         mlog_errno(ret);
4116                         goto out;
4117                 }
4118         }
4119
4120         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4121
4122         /*Only dirty the blocks we have touched in set xattr. */
4123         ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs,
4124                                                 xs->bucket.bhs, blk_per_bucket);
4125         if (ret)
4126                 mlog_errno(ret);
4127 out:
4128         ocfs2_commit_trans(osb, handle);
4129
4130         return ret;
4131 }
4132
4133 static int ocfs2_xattr_value_update_size(struct inode *inode,
4134                                          struct buffer_head *xe_bh,
4135                                          struct ocfs2_xattr_entry *xe,
4136                                          u64 new_size)
4137 {
4138         int ret;
4139         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4140         handle_t *handle = NULL;
4141
4142         handle = ocfs2_start_trans(osb, 1);
4143         if (handle == NULL) {
4144                 ret = -ENOMEM;
4145                 mlog_errno(ret);
4146                 goto out;
4147         }
4148
4149         ret = ocfs2_journal_access(handle, inode, xe_bh,
4150                                    OCFS2_JOURNAL_ACCESS_WRITE);
4151         if (ret < 0) {
4152                 mlog_errno(ret);
4153                 goto out_commit;
4154         }
4155
4156         xe->xe_value_size = cpu_to_le64(new_size);
4157
4158         ret = ocfs2_journal_dirty(handle, xe_bh);
4159         if (ret < 0)
4160                 mlog_errno(ret);
4161
4162 out_commit:
4163         ocfs2_commit_trans(osb, handle);
4164 out:
4165         return ret;
4166 }
4167
4168 /*
4169  * Truncate the specified xe_off entry in xattr bucket.
4170  * bucket is indicated by header_bh and len is the new length.
4171  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4172  *
4173  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4174  */
4175 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4176                                              struct buffer_head *header_bh,
4177                                              int xe_off,
4178                                              int len)
4179 {
4180         int ret, offset;
4181         u64 value_blk;
4182         struct buffer_head *value_bh = NULL;
4183         struct ocfs2_xattr_value_root *xv;
4184         struct ocfs2_xattr_entry *xe;
4185         struct ocfs2_xattr_header *xh =
4186                         (struct ocfs2_xattr_header *)header_bh->b_data;
4187         size_t blocksize = inode->i_sb->s_blocksize;
4188
4189         xe = &xh->xh_entries[xe_off];
4190
4191         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4192
4193         offset = le16_to_cpu(xe->xe_name_offset) +
4194                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4195
4196         value_blk = offset / blocksize;
4197
4198         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4199         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4200         value_blk += header_bh->b_blocknr;
4201
4202         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), value_blk,
4203                                &value_bh, OCFS2_BH_CACHED, inode);
4204         if (ret) {
4205                 mlog_errno(ret);
4206                 goto out;
4207         }
4208
4209         xv = (struct ocfs2_xattr_value_root *)
4210                 (value_bh->b_data + offset % blocksize);
4211
4212         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4213              xe_off, (unsigned long long)header_bh->b_blocknr, len);
4214         ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4215         if (ret) {
4216                 mlog_errno(ret);
4217                 goto out;
4218         }
4219
4220         ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4221         if (ret) {
4222                 mlog_errno(ret);
4223                 goto out;
4224         }
4225
4226 out:
4227         brelse(value_bh);
4228         return ret;
4229 }
4230
4231 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4232                                                 struct ocfs2_xattr_search *xs,
4233                                                 int len)
4234 {
4235         int ret, offset;
4236         struct ocfs2_xattr_entry *xe = xs->here;
4237         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4238
4239         BUG_ON(!xs->bucket.bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4240
4241         offset = xe - xh->xh_entries;
4242         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bhs[0],
4243                                                 offset, len);
4244         if (ret)
4245                 mlog_errno(ret);
4246
4247         return ret;
4248 }
4249
4250 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4251                                                 struct ocfs2_xattr_search *xs,
4252                                                 char *val,
4253                                                 int value_len)
4254 {
4255         int offset;
4256         struct ocfs2_xattr_value_root *xv;
4257         struct ocfs2_xattr_entry *xe = xs->here;
4258
4259         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4260
4261         offset = le16_to_cpu(xe->xe_name_offset) +
4262                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4263
4264         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4265
4266         return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4267 }
4268
4269 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4270                                   struct buffer_head *root_bh,
4271                                   u64 blkno,
4272                                   u32 cpos,
4273                                   u32 len)
4274 {
4275         int ret;
4276         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4277         struct inode *tl_inode = osb->osb_tl_inode;
4278         handle_t *handle;
4279         struct ocfs2_xattr_block *xb =
4280                         (struct ocfs2_xattr_block *)root_bh->b_data;
4281         struct ocfs2_alloc_context *meta_ac = NULL;
4282         struct ocfs2_cached_dealloc_ctxt dealloc;
4283         struct ocfs2_extent_tree et;
4284
4285         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4286
4287         ocfs2_init_dealloc_ctxt(&dealloc);
4288
4289         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4290              cpos, len, (unsigned long long)blkno);
4291
4292         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4293
4294         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4295         if (ret) {
4296                 mlog_errno(ret);
4297                 return ret;
4298         }
4299
4300         mutex_lock(&tl_inode->i_mutex);
4301
4302         if (ocfs2_truncate_log_needs_flush(osb)) {
4303                 ret = __ocfs2_flush_truncate_log(osb);
4304                 if (ret < 0) {
4305                         mlog_errno(ret);
4306                         goto out;
4307                 }
4308         }
4309
4310         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4311         if (handle == NULL) {
4312                 ret = -ENOMEM;
4313                 mlog_errno(ret);
4314                 goto out;
4315         }
4316
4317         ret = ocfs2_journal_access(handle, inode, root_bh,
4318                                    OCFS2_JOURNAL_ACCESS_WRITE);
4319         if (ret) {
4320                 mlog_errno(ret);
4321                 goto out_commit;
4322         }
4323
4324         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4325                                   &dealloc);
4326         if (ret) {
4327                 mlog_errno(ret);
4328                 goto out_commit;
4329         }
4330
4331         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4332
4333         ret = ocfs2_journal_dirty(handle, root_bh);
4334         if (ret) {
4335                 mlog_errno(ret);
4336                 goto out_commit;
4337         }
4338
4339         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4340         if (ret)
4341                 mlog_errno(ret);
4342
4343 out_commit:
4344         ocfs2_commit_trans(osb, handle);
4345 out:
4346         ocfs2_schedule_truncate_log_flush(osb, 1);
4347
4348         mutex_unlock(&tl_inode->i_mutex);
4349
4350         if (meta_ac)
4351                 ocfs2_free_alloc_context(meta_ac);
4352
4353         ocfs2_run_deallocs(osb, &dealloc);
4354
4355         return ret;
4356 }
4357
4358 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4359                                          struct ocfs2_xattr_search *xs)
4360 {
4361         handle_t *handle = NULL;
4362         struct ocfs2_xattr_header *xh = xs->bucket.xh;
4363         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4364                                                 le16_to_cpu(xh->xh_count) - 1];
4365         int ret = 0;
4366
4367         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1);
4368         if (IS_ERR(handle)) {
4369                 ret = PTR_ERR(handle);
4370                 mlog_errno(ret);
4371                 return;
4372         }
4373
4374         ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[0],
4375                                    OCFS2_JOURNAL_ACCESS_WRITE);
4376         if (ret) {
4377                 mlog_errno(ret);
4378                 goto out_commit;
4379         }
4380
4381         /* Remove the old entry. */
4382         memmove(xs->here, xs->here + 1,
4383                 (void *)last - (void *)xs->here);
4384         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4385         le16_add_cpu(&xh->xh_count, -1);
4386
4387         ret = ocfs2_journal_dirty(handle, xs->bucket.bhs[0]);
4388         if (ret < 0)
4389                 mlog_errno(ret);
4390 out_commit:
4391         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4392 }
4393
4394 /*
4395  * Set the xattr name/value in the bucket specified in xs.
4396  *
4397  * As the new value in xi may be stored in the bucket or in an outside cluster,
4398  * we divide the whole process into 3 steps:
4399  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4400  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4401  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4402  * 4. If the clusters for the new outside value can't be allocated, we need
4403  *    to free the xattr we allocated in set.
4404  */
4405 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4406                                      struct ocfs2_xattr_info *xi,
4407                                      struct ocfs2_xattr_search *xs)
4408 {
4409         int ret, local = 1;
4410         size_t value_len;
4411         char *val = (char *)xi->value;
4412         struct ocfs2_xattr_entry *xe = xs->here;
4413         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4414                                               strlen(xi->name));
4415
4416         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4417                 /*
4418                  * We need to truncate the xattr storage first.
4419                  *
4420                  * If both the old and new value are stored to
4421                  * outside block, we only need to truncate
4422                  * the storage and then set the value outside.
4423                  *
4424                  * If the new value should be stored within block,
4425                  * we should free all the outside block first and
4426                  * the modification to the xattr block will be done
4427                  * by following steps.
4428                  */
4429                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4430                         value_len = xi->value_len;
4431                 else
4432                         value_len = 0;
4433
4434                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4435                                                            value_len);
4436                 if (ret)
4437                         goto out;
4438
4439                 if (value_len)
4440                         goto set_value_outside;
4441         }
4442
4443         value_len = xi->value_len;
4444         /* So we have to handle the inside block change now. */
4445         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4446                 /*
4447                  * If the new value will be stored outside of block,
4448                  * initalize a new empty value root and insert it first.
4449                  */
4450                 local = 0;
4451                 xi->value = &def_xv;
4452                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4453         }
4454
4455         ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash, local);
4456         if (ret) {
4457                 mlog_errno(ret);
4458                 goto out;
4459         }
4460
4461         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4462                 goto out;
4463
4464         /* allocate the space now for the outside block storage. */
4465         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4466                                                    value_len);
4467         if (ret) {
4468                 mlog_errno(ret);
4469
4470                 if (xs->not_found) {
4471                         /*
4472                          * We can't allocate enough clusters for outside
4473                          * storage and we have allocated xattr already,
4474                          * so need to remove it.
4475                          */
4476                         ocfs2_xattr_bucket_remove_xs(inode, xs);
4477                 }
4478                 goto out;
4479         }
4480
4481 set_value_outside:
4482         ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4483 out:
4484         return ret;
4485 }
4486
4487 /* check whether the xattr bucket is filled up with the same hash value. */
4488 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4489                                               struct ocfs2_xattr_bucket *bucket)
4490 {
4491         struct ocfs2_xattr_header *xh = bucket->xh;
4492
4493         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4494             xh->xh_entries[0].xe_name_hash) {
4495                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4496                      "hash = %u\n",
4497                      (unsigned long long)bucket->bhs[0]->b_blocknr,
4498                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4499                 return -ENOSPC;
4500         }
4501
4502         return 0;
4503 }
4504
4505 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4506                                              struct ocfs2_xattr_info *xi,
4507                                              struct ocfs2_xattr_search *xs)
4508 {
4509         struct ocfs2_xattr_header *xh;
4510         struct ocfs2_xattr_entry *xe;
4511         u16 count, header_size, xh_free_start;
4512         int i, free, max_free, need, old;
4513         size_t value_size = 0, name_len = strlen(xi->name);
4514         size_t blocksize = inode->i_sb->s_blocksize;
4515         int ret, allocation = 0;
4516         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4517
4518         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4519
4520 try_again:
4521         xh = xs->header;
4522         count = le16_to_cpu(xh->xh_count);
4523         xh_free_start = le16_to_cpu(xh->xh_free_start);
4524         header_size = sizeof(struct ocfs2_xattr_header) +
4525                         count * sizeof(struct ocfs2_xattr_entry);
4526         max_free = OCFS2_XATTR_BUCKET_SIZE -
4527                 le16_to_cpu(xh->xh_name_value_len) - header_size;
4528
4529         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4530                         "of %u which exceed block size\n",
4531                         (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4532                         header_size);
4533
4534         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4535                 value_size = OCFS2_XATTR_ROOT_SIZE;
4536         else if (xi->value)
4537                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4538
4539         if (xs->not_found)
4540                 need = sizeof(struct ocfs2_xattr_entry) +
4541                         OCFS2_XATTR_SIZE(name_len) + value_size;
4542         else {
4543                 need = value_size + OCFS2_XATTR_SIZE(name_len);
4544
4545                 /*
4546                  * We only replace the old value if the new length is smaller
4547                  * than the old one. Otherwise we will allocate new space in the
4548                  * bucket to store it.
4549                  */
4550                 xe = xs->here;
4551                 if (ocfs2_xattr_is_local(xe))
4552                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4553                 else
4554                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4555
4556                 if (old >= value_size)
4557                         need = 0;
4558         }
4559
4560         free = xh_free_start - header_size;
4561         /*
4562          * We need to make sure the new name/value pair
4563          * can exist in the same block.
4564          */
4565         if (xh_free_start % blocksize < need)
4566                 free -= xh_free_start % blocksize;
4567
4568         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4569              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4570              " %u\n", xs->not_found,
4571              (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4572              free, need, max_free, le16_to_cpu(xh->xh_free_start),
4573              le16_to_cpu(xh->xh_name_value_len));
4574
4575         if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4576                 if (need <= max_free &&
4577                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4578                         /*
4579                          * We can create the space by defragment. Since only the
4580                          * name/value will be moved, the xe shouldn't be changed
4581                          * in xs.
4582                          */
4583                         ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket);
4584                         if (ret) {
4585                                 mlog_errno(ret);
4586                                 goto out;
4587                         }
4588
4589                         xh_free_start = le16_to_cpu(xh->xh_free_start);
4590                         free = xh_free_start - header_size;
4591                         if (xh_free_start % blocksize < need)
4592                                 free -= xh_free_start % blocksize;
4593
4594                         if (free >= need)
4595                                 goto xattr_set;
4596
4597                         mlog(0, "Can't get enough space for xattr insert by "
4598                              "defragment. Need %u bytes, but we have %d, so "
4599                              "allocate new bucket for it.\n", need, free);
4600                 }
4601
4602                 /*
4603                  * We have to add new buckets or clusters and one
4604                  * allocation should leave us enough space for insert.
4605                  */
4606                 BUG_ON(allocation);
4607
4608                 /*
4609                  * We do not allow for overlapping ranges between buckets. And
4610                  * the maximum number of collisions we will allow for then is
4611                  * one bucket's worth, so check it here whether we need to
4612                  * add a new bucket for the insert.
4613                  */
4614                 ret = ocfs2_check_xattr_bucket_collision(inode, &xs->bucket);
4615                 if (ret) {
4616                         mlog_errno(ret);
4617                         goto out;
4618                 }
4619
4620                 ret = ocfs2_add_new_xattr_bucket(inode,
4621                                                  xs->xattr_bh,
4622                                                  xs->bucket.bhs[0]);
4623                 if (ret) {
4624                         mlog_errno(ret);
4625                         goto out;
4626                 }
4627
4628                 for (i = 0; i < blk_per_bucket; i++)
4629                         brelse(xs->bucket.bhs[i]);
4630
4631                 memset(&xs->bucket, 0, sizeof(xs->bucket));
4632
4633                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4634                                                    xi->name_index,
4635                                                    xi->name, xs);
4636                 if (ret && ret != -ENODATA)
4637                         goto out;
4638                 xs->not_found = ret;
4639                 allocation = 1;
4640                 goto try_again;
4641         }
4642
4643 xattr_set:
4644         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4645 out:
4646         mlog_exit(ret);
4647         return ret;
4648 }
4649
4650 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4651                                         struct ocfs2_xattr_bucket *bucket,
4652                                         void *para)
4653 {
4654         int ret = 0;
4655         struct ocfs2_xattr_header *xh = bucket->xh;
4656         u16 i;
4657         struct ocfs2_xattr_entry *xe;
4658
4659         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4660                 xe = &xh->xh_entries[i];
4661                 if (ocfs2_xattr_is_local(xe))
4662                         continue;
4663
4664                 ret = ocfs2_xattr_bucket_value_truncate(inode,
4665                                                         bucket->bhs[0],
4666                                                         i, 0);
4667                 if (ret) {
4668                         mlog_errno(ret);
4669                         break;
4670                 }
4671         }
4672
4673         return ret;
4674 }
4675
4676 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4677                                           struct buffer_head *xb_bh)
4678 {
4679         struct ocfs2_xattr_block *xb =
4680                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4681         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4682         int ret = 0;
4683         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4684         u64 p_blkno;
4685
4686         if (le16_to_cpu(el->l_next_free_rec) == 0)
4687                 return 0;
4688
4689         while (name_hash > 0) {
4690                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4691                                           &e_cpos, &num_clusters, el);
4692                 if (ret) {
4693                         mlog_errno(ret);
4694                         goto out;
4695                 }
4696
4697                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4698                                                   ocfs2_delete_xattr_in_bucket,
4699                                                   NULL);
4700                 if (ret) {
4701                         mlog_errno(ret);
4702                         goto out;
4703                 }
4704
4705                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4706                                              p_blkno, e_cpos, num_clusters);
4707                 if (ret) {
4708                         mlog_errno(ret);
4709                         break;
4710                 }
4711
4712                 if (e_cpos == 0)
4713                         break;
4714
4715                 name_hash = e_cpos - 1;
4716         }
4717
4718 out:
4719         return ret;
4720 }
4721
4722 /*
4723  * 'trusted' attributes support
4724  */
4725
4726 #define XATTR_TRUSTED_PREFIX "trusted."
4727
4728 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
4729                                        size_t list_size, const char *name,
4730                                        size_t name_len)
4731 {
4732         const size_t prefix_len = sizeof(XATTR_TRUSTED_PREFIX) - 1;
4733         const size_t total_len = prefix_len + name_len + 1;
4734
4735         if (list && total_len <= list_size) {
4736                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
4737                 memcpy(list + prefix_len, name, name_len);
4738                 list[prefix_len + name_len] = '\0';
4739         }
4740         return total_len;
4741 }
4742
4743 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
4744                                    void *buffer, size_t size)
4745 {
4746         if (strcmp(name, "") == 0)
4747                 return -EINVAL;
4748         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
4749                                buffer, size);
4750 }
4751
4752 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
4753                                    const void *value, size_t size, int flags)
4754 {
4755         if (strcmp(name, "") == 0)
4756                 return -EINVAL;
4757
4758         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
4759                                size, flags);
4760 }
4761
4762 struct xattr_handler ocfs2_xattr_trusted_handler = {
4763         .prefix = XATTR_TRUSTED_PREFIX,
4764         .list   = ocfs2_xattr_trusted_list,
4765         .get    = ocfs2_xattr_trusted_get,
4766         .set    = ocfs2_xattr_trusted_set,
4767 };
4768
4769
4770 /*
4771  * 'user' attributes support
4772  */
4773
4774 #define XATTR_USER_PREFIX "user."
4775
4776 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
4777                                     size_t list_size, const char *name,
4778                                     size_t name_len)
4779 {
4780         const size_t prefix_len = sizeof(XATTR_USER_PREFIX) - 1;
4781         const size_t total_len = prefix_len + name_len + 1;
4782         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4783
4784         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4785                 return 0;
4786
4787         if (list && total_len <= list_size) {
4788                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
4789                 memcpy(list + prefix_len, name, name_len);
4790                 list[prefix_len + name_len] = '\0';
4791         }
4792         return total_len;
4793 }
4794
4795 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
4796                                 void *buffer, size_t size)
4797 {
4798         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4799
4800         if (strcmp(name, "") == 0)
4801                 return -EINVAL;
4802         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4803                 return -EOPNOTSUPP;
4804         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
4805                                buffer, size);
4806 }
4807
4808 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
4809                                 const void *value, size_t size, int flags)
4810 {
4811         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4812
4813         if (strcmp(name, "") == 0)
4814                 return -EINVAL;
4815         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4816                 return -EOPNOTSUPP;
4817
4818         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
4819                                size, flags);
4820 }
4821
4822 struct xattr_handler ocfs2_xattr_user_handler = {
4823         .prefix = XATTR_USER_PREFIX,
4824         .list   = ocfs2_xattr_user_list,
4825         .get    = ocfs2_xattr_user_get,
4826         .set    = ocfs2_xattr_user_set,
4827 };