fs/ocfs2/xattr.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * xattr.c
   5  *
   6  * Copyright (C) 2008 Oracle.  All rights reserved.
   7  *
   8  * CREDITS:
   9  * Lots of code in this file is taken from ext3.
  10  *
  11  * This program is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU General Public
  13  * License as published by the Free Software Foundation; either
  14  * version 2 of the License, or (at your option) any later version.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU General Public
  22  * License along with this program; if not, write to the
  23  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  24  * Boston, MA 021110-1307, USA.
  25  */
  26
  27 #include <linux/capability.h>
  28 #include <linux/fs.h>
  29 #include <linux/types.h>
  30 #include <linux/slab.h>
  31 #include <linux/highmem.h>
  32 #include <linux/pagemap.h>
  33 #include <linux/uio.h>
  34 #include <linux/sched.h>
  35 #include <linux/splice.h>
  36 #include <linux/mount.h>
  37 #include <linux/writeback.h>
  38 #include <linux/falloc.h>
  39 #include <linux/sort.h>
  40
  41 #define MLOG_MASK_PREFIX ML_XATTR
  42 #include <cluster/masklog.h>
  43
  44 #include "ocfs2.h"
  45 #include "alloc.h"
  46 #include "dlmglue.h"
  47 #include "file.h"
  48 #include "symlink.h"
  49 #include "sysfile.h"
  50 #include "inode.h"
  51 #include "journal.h"
  52 #include "ocfs2_fs.h"
  53 #include "suballoc.h"
  54 #include "uptodate.h"
  55 #include "buffer_head_io.h"
  56 #include "super.h"
  57 #include "xattr.h"
  58
  59
  60 struct ocfs2_xattr_def_value_root {
  61         struct ocfs2_xattr_value_root   xv;
  62         struct ocfs2_extent_rec         er;
  63 };
  64
  65 struct ocfs2_xattr_bucket {
  66         struct buffer_head *bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  67         struct ocfs2_xattr_header *xh;
  68 };
  69
  70 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  71 #define OCFS2_XATTR_INLINE_SIZE 80
  72
  73 static struct ocfs2_xattr_def_value_root def_xv = {
  74         .xv.xr_list.l_count = cpu_to_le16(1),
  75 };
  76
  77 struct xattr_handler *ocfs2_xattr_handlers[] = {
  78         &ocfs2_xattr_user_handler,
  79         &ocfs2_xattr_trusted_handler,
  80         NULL
  81 };
  82
  83 static struct xattr_handler *ocfs2_xattr_handler_map[] = {
  84         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
  85         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
  86 };
  87
  88 struct ocfs2_xattr_info {
  89         int name_index;
  90         const char *name;
  91         const void *value;
  92         size_t value_len;
  93 };
  94
  95 struct ocfs2_xattr_search {
  96         struct buffer_head *inode_bh;
  97         /*
  98          * xattr_bh point to the block buffer head which has extended attribute
  99          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 100          */
 101         struct buffer_head *xattr_bh;
 102         struct ocfs2_xattr_header *header;
 103         struct ocfs2_xattr_bucket bucket;
 104         void *base;
 105         void *end;
 106         struct ocfs2_xattr_entry *here;
 107         int not_found;
 108 };
 109
 110 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
 111                                              struct ocfs2_xattr_header *xh,
 112                                              int index,
 113                                              int *block_off,
 114                                              int *new_offset);
 115
 116 static int ocfs2_xattr_index_block_find(struct inode *inode,
 117                                         struct buffer_head *root_bh,
 118                                         int name_index,
 119                                         const char *name,
 120                                         struct ocfs2_xattr_search *xs);
 121
 122 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 123                                         struct ocfs2_xattr_tree_root *xt,
 124                                         char *buffer,
 125                                         size_t buffer_size);
 126
 127 static int ocfs2_xattr_create_index_block(struct inode *inode,
 128                                           struct ocfs2_xattr_search *xs);
 129
 130 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 131                                              struct ocfs2_xattr_info *xi,
 132                                              struct ocfs2_xattr_search *xs);
 133
 134 static int ocfs2_delete_xattr_index_block(struct inode *inode,
 135                                           struct buffer_head *xb_bh);
 136
 137 static inline struct xattr_handler *ocfs2_xattr_handler(int name_index)
 138 {
 139         struct xattr_handler *handler = NULL;
 140
 141         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 142                 handler = ocfs2_xattr_handler_map[name_index];
 143
 144         return handler;
 145 }
 146
 147 static inline u32 ocfs2_xattr_name_hash(struct inode *inode,
 148                                         char *prefix,
 149                                         int prefix_len,
 150                                         char *name,
 151                                         int name_len)
 152 {
 153         /* Get hash value of uuid from super block */
 154         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 155         int i;
 156
 157         /* hash extended attribute prefix */
 158         for (i = 0; i < prefix_len; i++) {
 159                 hash = (hash << OCFS2_HASH_SHIFT) ^
 160                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 161                        *prefix++;
 162         }
 163         /* hash extended attribute name */
 164         for (i = 0; i < name_len; i++) {
 165                 hash = (hash << OCFS2_HASH_SHIFT) ^
 166                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 167                        *name++;
 168         }
 169
 170         return hash;
 171 }
 172
 173 /*
 174  * ocfs2_xattr_hash_entry()
 175  *
 176  * Compute the hash of an extended attribute.
 177  */
 178 static void ocfs2_xattr_hash_entry(struct inode *inode,
 179                                    struct ocfs2_xattr_header *header,
 180                                    struct ocfs2_xattr_entry *entry)
 181 {
 182         u32 hash = 0;
 183         struct xattr_handler *handler =
 184                         ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
 185         char *prefix = handler->prefix;
 186         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
 187         int prefix_len = strlen(handler->prefix);
 188
 189         hash = ocfs2_xattr_name_hash(inode, prefix, prefix_len, name,
 190                                      entry->xe_name_len);
 191         entry->xe_name_hash = cpu_to_le32(hash);
 192
 193         return;
 194 }
 195
 196 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 197                                          u32 clusters_to_add,
 198                                          struct buffer_head *xattr_bh,
 199                                          struct ocfs2_xattr_value_root *xv)
 200 {
 201         int status = 0;
 202         int restart_func = 0;
 203         int credits = 0;
 204         handle_t *handle = NULL;
 205         struct ocfs2_alloc_context *data_ac = NULL;
 206         struct ocfs2_alloc_context *meta_ac = NULL;
 207         enum ocfs2_alloc_restarted why;
 208         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 209         struct ocfs2_extent_list *root_el = &xv->xr_list;
 210         u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
 211
 212         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
 213
 214 restart_all:
 215
 216         status = ocfs2_lock_allocators(inode, xattr_bh, root_el,
 217                                        clusters_to_add, 0, &data_ac,
 218                                        &meta_ac, OCFS2_XATTR_VALUE_EXTENT, xv);
 219         if (status) {
 220                 mlog_errno(status);
 221                 goto leave;
 222         }
 223
 224         credits = ocfs2_calc_extend_credits(osb->sb, root_el, clusters_to_add);
 225         handle = ocfs2_start_trans(osb, credits);
 226         if (IS_ERR(handle)) {
 227                 status = PTR_ERR(handle);
 228                 handle = NULL;
 229                 mlog_errno(status);
 230                 goto leave;
 231         }
 232
 233 restarted_transaction:
 234         status = ocfs2_journal_access(handle, inode, xattr_bh,
 235                                       OCFS2_JOURNAL_ACCESS_WRITE);
 236         if (status < 0) {
 237                 mlog_errno(status);
 238                 goto leave;
 239         }
 240
 241         prev_clusters = le32_to_cpu(xv->xr_clusters);
 242         status = ocfs2_add_clusters_in_btree(osb,
 243                                              inode,
 244                                              &logical_start,
 245                                              clusters_to_add,
 246                                              0,
 247                                              xattr_bh,
 248                                              root_el,
 249                                              handle,
 250                                              data_ac,
 251                                              meta_ac,
 252                                              &why,
 253                                              OCFS2_XATTR_VALUE_EXTENT,
 254                                              xv);
 255         if ((status < 0) && (status != -EAGAIN)) {
 256                 if (status != -ENOSPC)
 257                         mlog_errno(status);
 258                 goto leave;
 259         }
 260
 261         status = ocfs2_journal_dirty(handle, xattr_bh);
 262         if (status < 0) {
 263                 mlog_errno(status);
 264                 goto leave;
 265         }
 266
 267         clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
 268
 269         if (why != RESTART_NONE && clusters_to_add) {
 270                 if (why == RESTART_META) {
 271                         mlog(0, "restarting function.\n");
 272                         restart_func = 1;
 273                 } else {
 274                         BUG_ON(why != RESTART_TRANS);
 275
 276                         mlog(0, "restarting transaction.\n");
 277                         /* TODO: This can be more intelligent. */
 278                         credits = ocfs2_calc_extend_credits(osb->sb,
 279                                                             root_el,
 280                                                             clusters_to_add);
 281                         status = ocfs2_extend_trans(handle, credits);
 282                         if (status < 0) {
 283                                 /* handle still has to be committed at
 284                                  * this point. */
 285                                 status = -ENOMEM;
 286                                 mlog_errno(status);
 287                                 goto leave;
 288                         }
 289                         goto restarted_transaction;
 290                 }
 291         }
 292
 293 leave:
 294         if (handle) {
 295                 ocfs2_commit_trans(osb, handle);
 296                 handle = NULL;
 297         }
 298         if (data_ac) {
 299                 ocfs2_free_alloc_context(data_ac);
 300                 data_ac = NULL;
 301         }
 302         if (meta_ac) {
 303                 ocfs2_free_alloc_context(meta_ac);
 304                 meta_ac = NULL;
 305         }
 306         if ((!status) && restart_func) {
 307                 restart_func = 0;
 308                 goto restart_all;
 309         }
 310
 311         return status;
 312 }
 313
 314 static int __ocfs2_remove_xattr_range(struct inode *inode,
 315                                       struct buffer_head *root_bh,
 316                                       struct ocfs2_xattr_value_root *xv,
 317                                       u32 cpos, u32 phys_cpos, u32 len,
 318                                       struct ocfs2_cached_dealloc_ctxt *dealloc)
 319 {
 320         int ret;
 321         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 322         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 323         struct inode *tl_inode = osb->osb_tl_inode;
 324         handle_t *handle;
 325         struct ocfs2_alloc_context *meta_ac = NULL;
 326
 327         ret = ocfs2_lock_allocators(inode, root_bh, &xv->xr_list,
 328                                     0, 1, NULL, &meta_ac,
 329                                     OCFS2_XATTR_VALUE_EXTENT, xv);
 330         if (ret) {
 331                 mlog_errno(ret);
 332                 return ret;
 333         }
 334
 335         mutex_lock(&tl_inode->i_mutex);
 336
 337         if (ocfs2_truncate_log_needs_flush(osb)) {
 338                 ret = __ocfs2_flush_truncate_log(osb);
 339                 if (ret < 0) {
 340                         mlog_errno(ret);
 341                         goto out;
 342                 }
 343         }
 344
 345         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
 346         if (IS_ERR(handle)) {
 347                 ret = PTR_ERR(handle);
 348                 mlog_errno(ret);
 349                 goto out;
 350         }
 351
 352         ret = ocfs2_journal_access(handle, inode, root_bh,
 353                                    OCFS2_JOURNAL_ACCESS_WRITE);
 354         if (ret) {
 355                 mlog_errno(ret);
 356                 goto out_commit;
 357         }
 358
 359         ret = ocfs2_remove_extent(inode, root_bh, cpos, len, handle, meta_ac,
 360                                   dealloc, OCFS2_XATTR_VALUE_EXTENT, xv);
 361         if (ret) {
 362                 mlog_errno(ret);
 363                 goto out_commit;
 364         }
 365
 366         le32_add_cpu(&xv->xr_clusters, -len);
 367
 368         ret = ocfs2_journal_dirty(handle, root_bh);
 369         if (ret) {
 370                 mlog_errno(ret);
 371                 goto out_commit;
 372         }
 373
 374         ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
 375         if (ret)
 376                 mlog_errno(ret);
 377
 378 out_commit:
 379         ocfs2_commit_trans(osb, handle);
 380 out:
 381         mutex_unlock(&tl_inode->i_mutex);
 382
 383         if (meta_ac)
 384                 ocfs2_free_alloc_context(meta_ac);
 385
 386         return ret;
 387 }
 388
 389 static int ocfs2_xattr_shrink_size(struct inode *inode,
 390                                    u32 old_clusters,
 391                                    u32 new_clusters,
 392                                    struct buffer_head *root_bh,
 393                                    struct ocfs2_xattr_value_root *xv)
 394 {
 395         int ret = 0;
 396         u32 trunc_len, cpos, phys_cpos, alloc_size;
 397         u64 block;
 398         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 399         struct ocfs2_cached_dealloc_ctxt dealloc;
 400
 401         ocfs2_init_dealloc_ctxt(&dealloc);
 402
 403         if (old_clusters <= new_clusters)
 404                 return 0;
 405
 406         cpos = new_clusters;
 407         trunc_len = old_clusters - new_clusters;
 408         while (trunc_len) {
 409                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 410                                                &alloc_size, &xv->xr_list);
 411                 if (ret) {
 412                         mlog_errno(ret);
 413                         goto out;
 414                 }
 415
 416                 if (alloc_size > trunc_len)
 417                         alloc_size = trunc_len;
 418
 419                 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
 420                                                  phys_cpos, alloc_size,
 421                                                  &dealloc);
 422                 if (ret) {
 423                         mlog_errno(ret);
 424                         goto out;
 425                 }
 426
 427                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 428                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
 429                                                        alloc_size);
 430                 cpos += alloc_size;
 431                 trunc_len -= alloc_size;
 432         }
 433
 434 out:
 435         ocfs2_schedule_truncate_log_flush(osb, 1);
 436         ocfs2_run_deallocs(osb, &dealloc);
 437
 438         return ret;
 439 }
 440
 441 static int ocfs2_xattr_value_truncate(struct inode *inode,
 442                                       struct buffer_head *root_bh,
 443                                       struct ocfs2_xattr_value_root *xv,
 444                                       int len)
 445 {
 446         int ret;
 447         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 448         u32 old_clusters = le32_to_cpu(xv->xr_clusters);
 449
 450         if (new_clusters == old_clusters)
 451                 return 0;
 452
 453         if (new_clusters > old_clusters)
 454                 ret = ocfs2_xattr_extend_allocation(inode,
 455                                                     new_clusters - old_clusters,
 456                                                     root_bh, xv);
 457         else
 458                 ret = ocfs2_xattr_shrink_size(inode,
 459                                               old_clusters, new_clusters,
 460                                               root_bh, xv);
 461
 462         return ret;
 463 }
 464
 465 static int ocfs2_xattr_list_entries(struct inode *inode,
 466                                     struct ocfs2_xattr_header *header,
 467                                     char *buffer, size_t buffer_size)
 468 {
 469         size_t rest = buffer_size;
 470         int i;
 471
 472         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 473                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 474                 struct xattr_handler *handler =
 475                         ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
 476
 477                 if (handler) {
 478                         size_t size = handler->list(inode, buffer, rest,
 479                                         ((char *)header +
 480                                         le16_to_cpu(entry->xe_name_offset)),
 481                                         entry->xe_name_len);
 482                         if (buffer) {
 483                                 if (size > rest)
 484                                         return -ERANGE;
 485                                 buffer += size;
 486                         }
 487                         rest -= size;
 488                 }
 489         }
 490
 491         return buffer_size - rest;
 492 }
 493
 494 static int ocfs2_xattr_ibody_list(struct inode *inode,
 495                                   struct ocfs2_dinode *di,
 496                                   char *buffer,
 497                                   size_t buffer_size)
 498 {
 499         struct ocfs2_xattr_header *header = NULL;
 500         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 501         int ret = 0;
 502
 503         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 504                 return ret;
 505
 506         header = (struct ocfs2_xattr_header *)
 507                  ((void *)di + inode->i_sb->s_blocksize -
 508                  le16_to_cpu(di->i_xattr_inline_size));
 509
 510         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 511
 512         return ret;
 513 }
 514
 515 static int ocfs2_xattr_block_list(struct inode *inode,
 516                                   struct ocfs2_dinode *di,
 517                                   char *buffer,
 518                                   size_t buffer_size)
 519 {
 520         struct buffer_head *blk_bh = NULL;
 521         struct ocfs2_xattr_block *xb;
 522         int ret = 0;
 523
 524         if (!di->i_xattr_loc)
 525                 return ret;
 526
 527         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
 528                                le64_to_cpu(di->i_xattr_loc),
 529                                &blk_bh, OCFS2_BH_CACHED, inode);
 530         if (ret < 0) {
 531                 mlog_errno(ret);
 532                 return ret;
 533         }
 534         /*Verify the signature of xattr block*/
 535         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
 536                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
 537                 ret = -EFAULT;
 538                 goto cleanup;
 539         }
 540
 541         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 542
 543         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 544                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 545                 ret = ocfs2_xattr_list_entries(inode, header,
 546                                                buffer, buffer_size);
 547         } else {
 548                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
 549                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
 550                                                    buffer, buffer_size);
 551         }
 552 cleanup:
 553         brelse(blk_bh);
 554
 555         return ret;
 556 }
 557
 558 ssize_t ocfs2_listxattr(struct dentry *dentry,
 559                         char *buffer,
 560                         size_t size)
 561 {
 562         int ret = 0, i_ret = 0, b_ret = 0;
 563         struct buffer_head *di_bh = NULL;
 564         struct ocfs2_dinode *di = NULL;
 565         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
 566
 567         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 568                 return ret;
 569
 570         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
 571         if (ret < 0) {
 572                 mlog_errno(ret);
 573                 return ret;
 574         }
 575
 576         di = (struct ocfs2_dinode *)di_bh->b_data;
 577
 578         down_read(&oi->ip_xattr_sem);
 579         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
 580         if (i_ret < 0)
 581                 b_ret = 0;
 582         else {
 583                 if (buffer) {
 584                         buffer += i_ret;
 585                         size -= i_ret;
 586                 }
 587                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
 588                                                buffer, size);
 589                 if (b_ret < 0)
 590                         i_ret = 0;
 591         }
 592         up_read(&oi->ip_xattr_sem);
 593         ocfs2_inode_unlock(dentry->d_inode, 0);
 594
 595         brelse(di_bh);
 596
 597         return i_ret + b_ret;
 598 }
 599
 600 static int ocfs2_xattr_find_entry(int name_index,
 601                                   const char *name,
 602                                   struct ocfs2_xattr_search *xs)
 603 {
 604         struct ocfs2_xattr_entry *entry;
 605         size_t name_len;
 606         int i, cmp = 1;
 607
 608         if (name == NULL)
 609                 return -EINVAL;
 610
 611         name_len = strlen(name);
 612         entry = xs->here;
 613         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
 614                 cmp = name_index - ocfs2_xattr_get_type(entry);
 615                 if (!cmp)
 616                         cmp = name_len - entry->xe_name_len;
 617                 if (!cmp)
 618                         cmp = memcmp(name, (xs->base +
 619                                      le16_to_cpu(entry->xe_name_offset)),
 620                                      name_len);
 621                 if (cmp == 0)
 622                         break;
 623                 entry += 1;
 624         }
 625         xs->here = entry;
 626
 627         return cmp ? -ENODATA : 0;
 628 }
 629
 630 static int ocfs2_xattr_get_value_outside(struct inode *inode,
 631                                          struct ocfs2_xattr_value_root *xv,
 632                                          void *buffer,
 633                                          size_t len)
 634 {
 635         u32 cpos, p_cluster, num_clusters, bpc, clusters;
 636         u64 blkno;
 637         int i, ret = 0;
 638         size_t cplen, blocksize;
 639         struct buffer_head *bh = NULL;
 640         struct ocfs2_extent_list *el;
 641
 642         el = &xv->xr_list;
 643         clusters = le32_to_cpu(xv->xr_clusters);
 644         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 645         blocksize = inode->i_sb->s_blocksize;
 646
 647         cpos = 0;
 648         while (cpos < clusters) {
 649                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 650                                                &num_clusters, el);
 651                 if (ret) {
 652                         mlog_errno(ret);
 653                         goto out;
 654                 }
 655
 656                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 657                 /* Copy ocfs2_xattr_value */
 658                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 659                         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
 660                                                &bh, OCFS2_BH_CACHED, inode);
 661                         if (ret) {
 662                                 mlog_errno(ret);
 663                                 goto out;
 664                         }
 665
 666                         cplen = len >= blocksize ? blocksize : len;
 667                         memcpy(buffer, bh->b_data, cplen);
 668                         len -= cplen;
 669                         buffer += cplen;
 670
 671                         brelse(bh);
 672                         bh = NULL;
 673                         if (len == 0)
 674                                 break;
 675                 }
 676                 cpos += num_clusters;
 677         }
 678 out:
 679         return ret;
 680 }
 681
 682 static int ocfs2_xattr_ibody_get(struct inode *inode,
 683                                  int name_index,
 684                                  const char *name,
 685                                  void *buffer,
 686                                  size_t buffer_size,
 687                                  struct ocfs2_xattr_search *xs)
 688 {
 689         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 690         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
 691         struct ocfs2_xattr_value_root *xv;
 692         size_t size;
 693         int ret = 0;
 694
 695         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 696                 return -ENODATA;
 697
 698         xs->end = (void *)di + inode->i_sb->s_blocksize;
 699         xs->header = (struct ocfs2_xattr_header *)
 700                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
 701         xs->base = (void *)xs->header;
 702         xs->here = xs->header->xh_entries;
 703
 704         ret = ocfs2_xattr_find_entry(name_index, name, xs);
 705         if (ret)
 706                 return ret;
 707         size = le64_to_cpu(xs->here->xe_value_size);
 708         if (buffer) {
 709                 if (size > buffer_size)
 710                         return -ERANGE;
 711                 if (ocfs2_xattr_is_local(xs->here)) {
 712                         memcpy(buffer, (void *)xs->base +
 713                                le16_to_cpu(xs->here->xe_name_offset) +
 714                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
 715                 } else {
 716                         xv = (struct ocfs2_xattr_value_root *)
 717                                 (xs->base + le16_to_cpu(
 718                                  xs->here->xe_name_offset) +
 719                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
 720                         ret = ocfs2_xattr_get_value_outside(inode, xv,
 721                                                             buffer, size);
 722                         if (ret < 0) {
 723                                 mlog_errno(ret);
 724                                 return ret;
 725                         }
 726                 }
 727         }
 728
 729         return size;
 730 }
 731
 732 static int ocfs2_xattr_block_get(struct inode *inode,
 733                                  int name_index,
 734                                  const char *name,
 735                                  void *buffer,
 736                                  size_t buffer_size,
 737                                  struct ocfs2_xattr_search *xs)
 738 {
 739         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
 740         struct buffer_head *blk_bh = NULL;
 741         struct ocfs2_xattr_block *xb;
 742         struct ocfs2_xattr_value_root *xv;
 743         size_t size;
 744         int ret = -ENODATA, name_offset, name_len, block_off, i;
 745
 746         if (!di->i_xattr_loc)
 747                 return ret;
 748
 749         memset(&xs->bucket, 0, sizeof(xs->bucket));
 750
 751         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
 752                                le64_to_cpu(di->i_xattr_loc),
 753                                &blk_bh, OCFS2_BH_CACHED, inode);
 754         if (ret < 0) {
 755                 mlog_errno(ret);
 756                 return ret;
 757         }
 758         /*Verify the signature of xattr block*/
 759         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
 760                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
 761                 ret = -EFAULT;
 762                 goto cleanup;
 763         }
 764
 765         xs->xattr_bh = blk_bh;
 766         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 767
 768         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 769                 xs->header = &xb->xb_attrs.xb_header;
 770                 xs->base = (void *)xs->header;
 771                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
 772                 xs->here = xs->header->xh_entries;
 773
 774                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
 775         } else
 776                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
 777                                                    name_index,
 778                                                    name, xs);
 779
 780         if (ret)
 781                 goto cleanup;
 782         size = le64_to_cpu(xs->here->xe_value_size);
 783         if (buffer) {
 784                 ret = -ERANGE;
 785                 if (size > buffer_size)
 786                         goto cleanup;
 787
 788                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
 789                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
 790                 i = xs->here - xs->header->xh_entries;
 791
 792                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
 793                         ret = ocfs2_xattr_bucket_get_name_value(inode,
 794                                                                 xs->bucket.xh,
 795                                                                 i,
 796                                                                 &block_off,
 797                                                                 &name_offset);
 798                         xs->base = xs->bucket.bhs[block_off]->b_data;
 799                 }
 800                 if (ocfs2_xattr_is_local(xs->here)) {
 801                         memcpy(buffer, (void *)xs->base +
 802                                name_offset + name_len, size);
 803                 } else {
 804                         xv = (struct ocfs2_xattr_value_root *)
 805                                 (xs->base + name_offset + name_len);
 806                         ret = ocfs2_xattr_get_value_outside(inode, xv,
 807                                                             buffer, size);
 808                         if (ret < 0) {
 809                                 mlog_errno(ret);
 810                                 goto cleanup;
 811                         }
 812                 }
 813         }
 814         ret = size;
 815 cleanup:
 816         for (i = 0; i < OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET; i++)
 817                 brelse(xs->bucket.bhs[i]);
 818         memset(&xs->bucket, 0, sizeof(xs->bucket));
 819
 820         brelse(blk_bh);
 821         return ret;
 822 }
 823
 824 /* ocfs2_xattr_get()
 825  *
 826  * Copy an extended attribute into the buffer provided.
 827  * Buffer is NULL to compute the size of buffer required.
 828  */
 829 int ocfs2_xattr_get(struct inode *inode,
 830                     int name_index,
 831                     const char *name,
 832                     void *buffer,
 833                     size_t buffer_size)
 834 {
 835         int ret;
 836         struct ocfs2_dinode *di = NULL;
 837         struct buffer_head *di_bh = NULL;
 838         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 839         struct ocfs2_xattr_search xis = {
 840                 .not_found = -ENODATA,
 841         };
 842         struct ocfs2_xattr_search xbs = {
 843                 .not_found = -ENODATA,
 844         };
 845
 846         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 847                 ret = -ENODATA;
 848
 849         ret = ocfs2_inode_lock(inode, &di_bh, 0);
 850         if (ret < 0) {
 851                 mlog_errno(ret);
 852                 return ret;
 853         }
 854         xis.inode_bh = xbs.inode_bh = di_bh;
 855         di = (struct ocfs2_dinode *)di_bh->b_data;
 856
 857         down_read(&oi->ip_xattr_sem);
 858         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
 859                                     buffer_size, &xis);
 860         if (ret == -ENODATA)
 861                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
 862                                             buffer_size, &xbs);
 863         up_read(&oi->ip_xattr_sem);
 864         ocfs2_inode_unlock(inode, 0);
 865
 866         brelse(di_bh);
 867
 868         return ret;
 869 }
 870
 871 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
 872                                            struct ocfs2_xattr_value_root *xv,
 873                                            const void *value,
 874                                            int value_len)
 875 {
 876         int ret = 0, i, cp_len, credits;
 877         u16 blocksize = inode->i_sb->s_blocksize;
 878         u32 p_cluster, num_clusters;
 879         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 880         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
 881         u64 blkno;
 882         struct buffer_head *bh = NULL;
 883         handle_t *handle;
 884
 885         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
 886
 887         credits = clusters * bpc;
 888         handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits);
 889         if (IS_ERR(handle)) {
 890                 ret = PTR_ERR(handle);
 891                 mlog_errno(ret);
 892                 goto out;
 893         }
 894
 895         while (cpos < clusters) {
 896                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 897                                                &num_clusters, &xv->xr_list);
 898                 if (ret) {
 899                         mlog_errno(ret);
 900                         goto out_commit;
 901                 }
 902
 903                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 904
 905                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 906                         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
 907                                                &bh, OCFS2_BH_CACHED, inode);
 908                         if (ret) {
 909                                 mlog_errno(ret);
 910                                 goto out_commit;
 911                         }
 912
 913                         ret = ocfs2_journal_access(handle,
 914                                                    inode,
 915                                                    bh,
 916                                                    OCFS2_JOURNAL_ACCESS_WRITE);
 917                         if (ret < 0) {
 918                                 mlog_errno(ret);
 919                                 goto out_commit;
 920                         }
 921
 922                         cp_len = value_len > blocksize ? blocksize : value_len;
 923                         memcpy(bh->b_data, value, cp_len);
 924                         value_len -= cp_len;
 925                         value += cp_len;
 926                         if (cp_len < blocksize)
 927                                 memset(bh->b_data + cp_len, 0,
 928                                        blocksize - cp_len);
 929
 930                         ret = ocfs2_journal_dirty(handle, bh);
 931                         if (ret < 0) {
 932                                 mlog_errno(ret);
 933                                 goto out_commit;
 934                         }
 935                         brelse(bh);
 936                         bh = NULL;
 937
 938                         /*
 939                          * XXX: do we need to empty all the following
 940                          * blocks in this cluster?
 941                          */
 942                         if (!value_len)
 943                                 break;
 944                 }
 945                 cpos += num_clusters;
 946         }
 947 out_commit:
 948         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 949 out:
 950         brelse(bh);
 951
 952         return ret;
 953 }
 954
 955 static int ocfs2_xattr_cleanup(struct inode *inode,
 956                                struct ocfs2_xattr_info *xi,
 957                                struct ocfs2_xattr_search *xs,
 958                                size_t offs)
 959 {
 960         handle_t *handle = NULL;
 961         int ret = 0;
 962         size_t name_len = strlen(xi->name);
 963         void *val = xs->base + offs;
 964         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 965
 966         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
 967                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
 968         if (IS_ERR(handle)) {
 969                 ret = PTR_ERR(handle);
 970                 mlog_errno(ret);
 971                 goto out;
 972         }
 973         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
 974                                    OCFS2_JOURNAL_ACCESS_WRITE);
 975         if (ret) {
 976                 mlog_errno(ret);
 977                 goto out_commit;
 978         }
 979         /* Decrease xattr count */
 980         le16_add_cpu(&xs->header->xh_count, -1);
 981         /* Remove the xattr entry and tree root which has already be set*/
 982         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
 983         memset(val, 0, size);
 984
 985         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
 986         if (ret < 0)
 987                 mlog_errno(ret);
 988 out_commit:
 989         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 990 out:
 991         return ret;
 992 }
 993
 994 static int ocfs2_xattr_update_entry(struct inode *inode,
 995                                     struct ocfs2_xattr_info *xi,
 996                                     struct ocfs2_xattr_search *xs,
 997                                     size_t offs)
 998 {
 999         handle_t *handle = NULL;
1000         int ret = 0;
1001
1002         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1003                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1004         if (IS_ERR(handle)) {
1005                 ret = PTR_ERR(handle);
1006                 mlog_errno(ret);
1007                 goto out;
1008         }
1009         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1010                                    OCFS2_JOURNAL_ACCESS_WRITE);
1011         if (ret) {
1012                 mlog_errno(ret);
1013                 goto out_commit;
1014         }
1015
1016         xs->here->xe_name_offset = cpu_to_le16(offs);
1017         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1018         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1019                 ocfs2_xattr_set_local(xs->here, 1);
1020         else
1021                 ocfs2_xattr_set_local(xs->here, 0);
1022         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1023
1024         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1025         if (ret < 0)
1026                 mlog_errno(ret);
1027 out_commit:
1028         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1029 out:
1030         return ret;
1031 }
1032
1033 /*
1034  * ocfs2_xattr_set_value_outside()
1035  *
1036  * Set large size value in B tree.
1037  */
1038 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1039                                          struct ocfs2_xattr_info *xi,
1040                                          struct ocfs2_xattr_search *xs,
1041                                          size_t offs)
1042 {
1043         size_t name_len = strlen(xi->name);
1044         void *val = xs->base + offs;
1045         struct ocfs2_xattr_value_root *xv = NULL;
1046         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1047         int ret = 0;
1048
1049         memset(val, 0, size);
1050         memcpy(val, xi->name, name_len);
1051         xv = (struct ocfs2_xattr_value_root *)
1052                 (val + OCFS2_XATTR_SIZE(name_len));
1053         xv->xr_clusters = 0;
1054         xv->xr_last_eb_blk = 0;
1055         xv->xr_list.l_tree_depth = 0;
1056         xv->xr_list.l_count = cpu_to_le16(1);
1057         xv->xr_list.l_next_free_rec = 0;
1058
1059         ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1060                                          xi->value_len);
1061         if (ret < 0) {
1062                 mlog_errno(ret);
1063                 return ret;
1064         }
1065         ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value,
1066                                               xi->value_len);
1067         if (ret < 0) {
1068                 mlog_errno(ret);
1069                 return ret;
1070         }
1071         ret = ocfs2_xattr_update_entry(inode, xi, xs, offs);
1072         if (ret < 0)
1073                 mlog_errno(ret);
1074
1075         return ret;
1076 }
1077
1078 /*
1079  * ocfs2_xattr_set_entry_local()
1080  *
1081  * Set, replace or remove extended attribute in local.
1082  */
1083 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1084                                         struct ocfs2_xattr_info *xi,
1085                                         struct ocfs2_xattr_search *xs,
1086                                         struct ocfs2_xattr_entry *last,
1087                                         size_t min_offs)
1088 {
1089         size_t name_len = strlen(xi->name);
1090         int i;
1091
1092         if (xi->value && xs->not_found) {
1093                 /* Insert the new xattr entry. */
1094                 le16_add_cpu(&xs->header->xh_count, 1);
1095                 ocfs2_xattr_set_type(last, xi->name_index);
1096                 ocfs2_xattr_set_local(last, 1);
1097                 last->xe_name_len = name_len;
1098         } else {
1099                 void *first_val;
1100                 void *val;
1101                 size_t offs, size;
1102
1103                 first_val = xs->base + min_offs;
1104                 offs = le16_to_cpu(xs->here->xe_name_offset);
1105                 val = xs->base + offs;
1106
1107                 if (le64_to_cpu(xs->here->xe_value_size) >
1108                     OCFS2_XATTR_INLINE_SIZE)
1109                         size = OCFS2_XATTR_SIZE(name_len) +
1110                                 OCFS2_XATTR_ROOT_SIZE;
1111                 else
1112                         size = OCFS2_XATTR_SIZE(name_len) +
1113                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1114
1115                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1116                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1117                         /* The old and the new value have the
1118                            same size. Just replace the value. */
1119                         ocfs2_xattr_set_local(xs->here, 1);
1120                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1121                         /* Clear value bytes. */
1122                         memset(val + OCFS2_XATTR_SIZE(name_len),
1123                                0,
1124                                OCFS2_XATTR_SIZE(xi->value_len));
1125                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1126                                xi->value,
1127                                xi->value_len);
1128                         return;
1129                 }
1130                 /* Remove the old name+value. */
1131                 memmove(first_val + size, first_val, val - first_val);
1132                 memset(first_val, 0, size);
1133                 xs->here->xe_name_hash = 0;
1134                 xs->here->xe_name_offset = 0;
1135                 ocfs2_xattr_set_local(xs->here, 1);
1136                 xs->here->xe_value_size = 0;
1137
1138                 min_offs += size;
1139
1140                 /* Adjust all value offsets. */
1141                 last = xs->header->xh_entries;
1142                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1143                         size_t o = le16_to_cpu(last->xe_name_offset);
1144
1145                         if (o < offs)
1146                                 last->xe_name_offset = cpu_to_le16(o + size);
1147                         last += 1;
1148                 }
1149
1150                 if (!xi->value) {
1151                         /* Remove the old entry. */
1152                         last -= 1;
1153                         memmove(xs->here, xs->here + 1,
1154                                 (void *)last - (void *)xs->here);
1155                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1156                         le16_add_cpu(&xs->header->xh_count, -1);
1157                 }
1158         }
1159         if (xi->value) {
1160                 /* Insert the new name+value. */
1161                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1162                                 OCFS2_XATTR_SIZE(xi->value_len);
1163                 void *val = xs->base + min_offs - size;
1164
1165                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1166                 memset(val, 0, size);
1167                 memcpy(val, xi->name, name_len);
1168                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1169                        xi->value,
1170                        xi->value_len);
1171                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1172                 ocfs2_xattr_set_local(xs->here, 1);
1173                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1174         }
1175
1176         return;
1177 }
1178
1179 /*
1180  * ocfs2_xattr_set_entry()
1181  *
1182  * Set extended attribute entry into inode or block.
1183  *
1184  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1185  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1186  * then set value in B tree with set_value_outside().
1187  */
1188 static int ocfs2_xattr_set_entry(struct inode *inode,
1189                                  struct ocfs2_xattr_info *xi,
1190                                  struct ocfs2_xattr_search *xs,
1191                                  int flag)
1192 {
1193         struct ocfs2_xattr_entry *last;
1194         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1195         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1196         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1197         size_t size_l = 0;
1198         handle_t *handle = NULL;
1199         int free, i, ret;
1200         struct ocfs2_xattr_info xi_l = {
1201                 .name_index = xi->name_index,
1202                 .name = xi->name,
1203                 .value = xi->value,
1204                 .value_len = xi->value_len,
1205         };
1206
1207         /* Compute min_offs, last and free space. */
1208         last = xs->header->xh_entries;
1209
1210         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1211                 size_t offs = le16_to_cpu(last->xe_name_offset);
1212                 if (offs < min_offs)
1213                         min_offs = offs;
1214                 last += 1;
1215         }
1216
1217         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1218         if (free < 0)
1219                 return -EFAULT;
1220
1221         if (!xs->not_found) {
1222                 size_t size = 0;
1223                 if (ocfs2_xattr_is_local(xs->here))
1224                         size = OCFS2_XATTR_SIZE(name_len) +
1225                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1226                 else
1227                         size = OCFS2_XATTR_SIZE(name_len) +
1228                                 OCFS2_XATTR_ROOT_SIZE;
1229                 free += (size + sizeof(struct ocfs2_xattr_entry));
1230         }
1231         /* Check free space in inode or block */
1232         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1233                 if (free < sizeof(struct ocfs2_xattr_entry) +
1234                            OCFS2_XATTR_SIZE(name_len) +
1235                            OCFS2_XATTR_ROOT_SIZE) {
1236                         ret = -ENOSPC;
1237                         goto out;
1238                 }
1239                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1240                 xi_l.value = (void *)&def_xv;
1241                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1242         } else if (xi->value) {
1243                 if (free < sizeof(struct ocfs2_xattr_entry) +
1244                            OCFS2_XATTR_SIZE(name_len) +
1245                            OCFS2_XATTR_SIZE(xi->value_len)) {
1246                         ret = -ENOSPC;
1247                         goto out;
1248                 }
1249         }
1250
1251         if (!xs->not_found) {
1252                 /* For existing extended attribute */
1253                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1254                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1255                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1256                 void *val = xs->base + offs;
1257
1258                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1259                         /* Replace existing local xattr with tree root */
1260                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1261                                                             offs);
1262                         if (ret < 0)
1263                                 mlog_errno(ret);
1264                         goto out;
1265                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1266                         /* For existing xattr which has value outside */
1267                         struct ocfs2_xattr_value_root *xv = NULL;
1268                         xv = (struct ocfs2_xattr_value_root *)(val +
1269                                 OCFS2_XATTR_SIZE(name_len));
1270
1271                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1272                                 /*
1273                                  * If new value need set outside also,
1274                                  * first truncate old value to new value,
1275                                  * then set new value with set_value_outside().
1276                                  */
1277                                 ret = ocfs2_xattr_value_truncate(inode,
1278                                                                  xs->xattr_bh,
1279                                                                  xv,
1280                                                                  xi->value_len);
1281                                 if (ret < 0) {
1282                                         mlog_errno(ret);
1283                                         goto out;
1284                                 }
1285
1286                                 ret = __ocfs2_xattr_set_value_outside(inode,
1287                                                                 xv,
1288                                                                 xi->value,
1289                                                                 xi->value_len);
1290                                 if (ret < 0) {
1291                                         mlog_errno(ret);
1292                                         goto out;
1293                                 }
1294
1295                                 ret = ocfs2_xattr_update_entry(inode,
1296                                                                xi,
1297                                                                xs,
1298                                                                offs);
1299                                 if (ret < 0)
1300                                         mlog_errno(ret);
1301                                 goto out;
1302                         } else {
1303                                 /*
1304                                  * If new value need set in local,
1305                                  * just trucate old value to zero.
1306                                  */
1307                                  ret = ocfs2_xattr_value_truncate(inode,
1308                                                                  xs->xattr_bh,
1309                                                                  xv,
1310                                                                  0);
1311                                 if (ret < 0)
1312                                         mlog_errno(ret);
1313                         }
1314                 }
1315         }
1316
1317         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1318                                    OCFS2_INODE_UPDATE_CREDITS);
1319         if (IS_ERR(handle)) {
1320                 ret = PTR_ERR(handle);
1321                 mlog_errno(ret);
1322                 goto out;
1323         }
1324
1325         ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1326                                    OCFS2_JOURNAL_ACCESS_WRITE);
1327         if (ret) {
1328                 mlog_errno(ret);
1329                 goto out_commit;
1330         }
1331
1332         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1333                 /*set extended attribue in external blcok*/
1334                 ret = ocfs2_extend_trans(handle,
1335                                          OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1336                 if (ret) {
1337                         mlog_errno(ret);
1338                         goto out_commit;
1339                 }
1340                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1341                                            OCFS2_JOURNAL_ACCESS_WRITE);
1342                 if (ret) {
1343                         mlog_errno(ret);
1344                         goto out_commit;
1345                 }
1346         }
1347
1348         /*
1349          * Set value in local, include set tree root in local.
1350          * This is the first step for value size >INLINE_SIZE.
1351          */
1352         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1353
1354         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1355                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1356                 if (ret < 0) {
1357                         mlog_errno(ret);
1358                         goto out_commit;
1359                 }
1360         }
1361
1362         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1363             (flag & OCFS2_INLINE_XATTR_FL)) {
1364                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1365                 unsigned int xattrsize = osb->s_xattr_inline_size;
1366
1367                 /*
1368                  * Adjust extent record count or inline data size
1369                  * to reserve space for extended attribute.
1370                  */
1371                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1372                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1373                         le16_add_cpu(&idata->id_count, -xattrsize);
1374                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1375                         struct ocfs2_extent_list *el = &di->id2.i_list;
1376                         le16_add_cpu(&el->l_count, -(xattrsize /
1377                                         sizeof(struct ocfs2_extent_rec)));
1378                 }
1379                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1380         }
1381         /* Update xattr flag */
1382         spin_lock(&oi->ip_lock);
1383         oi->ip_dyn_features |= flag;
1384         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1385         spin_unlock(&oi->ip_lock);
1386         /* Update inode ctime */
1387         inode->i_ctime = CURRENT_TIME;
1388         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1389         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1390
1391         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1392         if (ret < 0)
1393                 mlog_errno(ret);
1394
1395 out_commit:
1396         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1397
1398         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1399                 /*
1400                  * Set value outside in B tree.
1401                  * This is the second step for value size > INLINE_SIZE.
1402                  */
1403                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1404                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1405                 if (ret < 0) {
1406                         int ret2;
1407
1408                         mlog_errno(ret);
1409                         /*
1410                          * If set value outside failed, we have to clean
1411                          * the junk tree root we have already set in local.
1412                          */
1413                         ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1414                         if (ret2 < 0)
1415                                 mlog_errno(ret2);
1416                 }
1417         }
1418 out:
1419         return ret;
1420
1421 }
1422
1423 static int ocfs2_xattr_free_block(handle_t *handle,
1424                                   struct ocfs2_super *osb,
1425                                   struct ocfs2_xattr_block *xb)
1426 {
1427         struct inode *xb_alloc_inode;
1428         struct buffer_head *xb_alloc_bh = NULL;
1429         u64 blk = le64_to_cpu(xb->xb_blkno);
1430         u16 bit = le16_to_cpu(xb->xb_suballoc_bit);
1431         u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1432         int ret = 0;
1433
1434         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1435                                 EXTENT_ALLOC_SYSTEM_INODE,
1436                                 le16_to_cpu(xb->xb_suballoc_slot));
1437         if (!xb_alloc_inode) {
1438                 ret = -ENOMEM;
1439                 mlog_errno(ret);
1440                 goto out;
1441         }
1442         mutex_lock(&xb_alloc_inode->i_mutex);
1443
1444         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1445         if (ret < 0) {
1446                 mlog_errno(ret);
1447                 goto out_mutex;
1448         }
1449         ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE);
1450         if (ret < 0) {
1451                 mlog_errno(ret);
1452                 goto out_unlock;
1453         }
1454         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1455                                        bit, bg_blkno, 1);
1456         if (ret < 0)
1457                 mlog_errno(ret);
1458 out_unlock:
1459         ocfs2_inode_unlock(xb_alloc_inode, 1);
1460         brelse(xb_alloc_bh);
1461 out_mutex:
1462         mutex_unlock(&xb_alloc_inode->i_mutex);
1463         iput(xb_alloc_inode);
1464 out:
1465         return ret;
1466 }
1467
1468 static int ocfs2_remove_value_outside(struct inode*inode,
1469                                       struct buffer_head *bh,
1470                                       struct ocfs2_xattr_header *header)
1471 {
1472         int ret = 0, i;
1473
1474         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1475                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1476
1477                 if (!ocfs2_xattr_is_local(entry)) {
1478                         struct ocfs2_xattr_value_root *xv;
1479                         void *val;
1480
1481                         val = (void *)header +
1482                                 le16_to_cpu(entry->xe_name_offset);
1483                         xv = (struct ocfs2_xattr_value_root *)
1484                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1485                         ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1486                         if (ret < 0) {
1487                                 mlog_errno(ret);
1488                                 return ret;
1489                         }
1490                 }
1491         }
1492
1493         return ret;
1494 }
1495
1496 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1497                                     struct buffer_head *di_bh)
1498 {
1499
1500         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1501         struct ocfs2_xattr_header *header;
1502         int ret;
1503
1504         header = (struct ocfs2_xattr_header *)
1505                  ((void *)di + inode->i_sb->s_blocksize -
1506                  le16_to_cpu(di->i_xattr_inline_size));
1507
1508         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1509
1510         return ret;
1511 }
1512
1513 static int ocfs2_xattr_block_remove(struct inode *inode,
1514                                     struct buffer_head *blk_bh)
1515 {
1516         struct ocfs2_xattr_block *xb;
1517         int ret = 0;
1518
1519         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1520         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1521                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1522                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1523         } else
1524                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1525
1526         return ret;
1527 }
1528
1529 /*
1530  * ocfs2_xattr_remove()
1531  *
1532  * Free extended attribute resources associated with this inode.
1533  */
1534 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1535 {
1536         struct ocfs2_xattr_block *xb;
1537         struct buffer_head *blk_bh = NULL;
1538         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1539         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1540         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1541         handle_t *handle;
1542         int ret;
1543
1544         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1545                 return 0;
1546
1547         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1548                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1549                 if (ret < 0) {
1550                         mlog_errno(ret);
1551                         goto out;
1552                 }
1553         }
1554         if (di->i_xattr_loc) {
1555                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1556                                        le64_to_cpu(di->i_xattr_loc),
1557                                        &blk_bh, OCFS2_BH_CACHED, inode);
1558                 if (ret < 0) {
1559                         mlog_errno(ret);
1560                         return ret;
1561                 }
1562                 /*Verify the signature of xattr block*/
1563                 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1564                            strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1565                         ret = -EFAULT;
1566                         goto out;
1567                 }
1568
1569                 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1570                 if (ret < 0) {
1571                         mlog_errno(ret);
1572                         goto out;
1573                 }
1574         }
1575
1576         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1577                                    OCFS2_INODE_UPDATE_CREDITS);
1578         if (IS_ERR(handle)) {
1579                 ret = PTR_ERR(handle);
1580                 mlog_errno(ret);
1581                 goto out;
1582         }
1583         ret = ocfs2_journal_access(handle, inode, di_bh,
1584                                    OCFS2_JOURNAL_ACCESS_WRITE);
1585         if (ret) {
1586                 mlog_errno(ret);
1587                 goto out_commit;
1588         }
1589
1590         if (di->i_xattr_loc) {
1591                 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1592                 ocfs2_xattr_free_block(handle, osb, xb);
1593                 di->i_xattr_loc = cpu_to_le64(0);
1594         }
1595
1596         spin_lock(&oi->ip_lock);
1597         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1598         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1599         spin_unlock(&oi->ip_lock);
1600
1601         ret = ocfs2_journal_dirty(handle, di_bh);
1602         if (ret < 0)
1603                 mlog_errno(ret);
1604 out_commit:
1605         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1606 out:
1607         brelse(blk_bh);
1608
1609         return ret;
1610 }
1611
1612 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1613                                         struct ocfs2_dinode *di)
1614 {
1615         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1616         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1617         int free;
1618
1619         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1620                 return 0;
1621
1622         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1623                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1624                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1625         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1626                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1627                         le64_to_cpu(di->i_size);
1628         } else {
1629                 struct ocfs2_extent_list *el = &di->id2.i_list;
1630                 free = (le16_to_cpu(el->l_count) -
1631                         le16_to_cpu(el->l_next_free_rec)) *
1632                         sizeof(struct ocfs2_extent_rec);
1633         }
1634         if (free >= xattrsize)
1635                 return 1;
1636
1637         return 0;
1638 }
1639
1640 /*
1641  * ocfs2_xattr_ibody_find()
1642  *
1643  * Find extended attribute in inode block and
1644  * fill search info into struct ocfs2_xattr_search.
1645  */
1646 static int ocfs2_xattr_ibody_find(struct inode *inode,
1647                                   int name_index,
1648                                   const char *name,
1649                                   struct ocfs2_xattr_search *xs)
1650 {
1651         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1652         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1653         int ret;
1654         int has_space = 0;
1655
1656         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1657                 return 0;
1658
1659         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1660                 down_read(&oi->ip_alloc_sem);
1661                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1662                 up_read(&oi->ip_alloc_sem);
1663                 if (!has_space)
1664                         return 0;
1665         }
1666
1667         xs->xattr_bh = xs->inode_bh;
1668         xs->end = (void *)di + inode->i_sb->s_blocksize;
1669         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1670                 xs->header = (struct ocfs2_xattr_header *)
1671                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1672         else
1673                 xs->header = (struct ocfs2_xattr_header *)
1674                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1675         xs->base = (void *)xs->header;
1676         xs->here = xs->header->xh_entries;
1677
1678         /* Find the named attribute. */
1679         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1680                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1681                 if (ret && ret != -ENODATA)
1682                         return ret;
1683                 xs->not_found = ret;
1684         }
1685
1686         return 0;
1687 }
1688
1689 /*
1690  * ocfs2_xattr_ibody_set()
1691  *
1692  * Set, replace or remove an extended attribute into inode block.
1693  *
1694  */
1695 static int ocfs2_xattr_ibody_set(struct inode *inode,
1696                                  struct ocfs2_xattr_info *xi,
1697                                  struct ocfs2_xattr_search *xs)
1698 {
1699         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1700         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1701         int ret;
1702
1703         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1704                 return -ENOSPC;
1705
1706         down_write(&oi->ip_alloc_sem);
1707         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1708                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1709                         ret = -ENOSPC;
1710                         goto out;
1711                 }
1712         }
1713
1714         ret = ocfs2_xattr_set_entry(inode, xi, xs,
1715                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1716 out:
1717         up_write(&oi->ip_alloc_sem);
1718
1719         return ret;
1720 }
1721
1722 /*
1723  * ocfs2_xattr_block_find()
1724  *
1725  * Find extended attribute in external block and
1726  * fill search info into struct ocfs2_xattr_search.
1727  */
1728 static int ocfs2_xattr_block_find(struct inode *inode,
1729                                   int name_index,
1730                                   const char *name,
1731                                   struct ocfs2_xattr_search *xs)
1732 {
1733         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1734         struct buffer_head *blk_bh = NULL;
1735         struct ocfs2_xattr_block *xb;
1736         int ret = 0;
1737
1738         if (!di->i_xattr_loc)
1739                 return ret;
1740
1741         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1742                                le64_to_cpu(di->i_xattr_loc),
1743                                &blk_bh, OCFS2_BH_CACHED, inode);
1744         if (ret < 0) {
1745                 mlog_errno(ret);
1746                 return ret;
1747         }
1748         /*Verify the signature of xattr block*/
1749         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1750                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1751                         ret = -EFAULT;
1752                         goto cleanup;
1753         }
1754
1755         xs->xattr_bh = blk_bh;
1756         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1757
1758         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1759                 xs->header = &xb->xb_attrs.xb_header;
1760                 xs->base = (void *)xs->header;
1761                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1762                 xs->here = xs->header->xh_entries;
1763
1764                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1765         } else
1766                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1767                                                    name_index,
1768                                                    name, xs);
1769
1770         if (ret && ret != -ENODATA) {
1771                 xs->xattr_bh = NULL;
1772                 goto cleanup;
1773         }
1774         xs->not_found = ret;
1775         return 0;
1776 cleanup:
1777         brelse(blk_bh);
1778
1779         return ret;
1780 }
1781
1782 /*
1783  * When all the xattrs are deleted from index btree, the ocfs2_xattr_tree
1784  * will be erased and ocfs2_xattr_block will have its ocfs2_xattr_header
1785  * re-initialized.
1786  */
1787 static int ocfs2_restore_xattr_block(struct inode *inode,
1788                                      struct ocfs2_xattr_search *xs)
1789 {
1790         int ret;
1791         handle_t *handle;
1792         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1793         struct ocfs2_xattr_block *xb =
1794                 (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1795         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
1796         u16 xb_flags = le16_to_cpu(xb->xb_flags);
1797
1798         BUG_ON(!(xb_flags & OCFS2_XATTR_INDEXED) ||
1799                 le16_to_cpu(el->l_next_free_rec) != 0);
1800
1801         handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1802         if (IS_ERR(handle)) {
1803                 ret = PTR_ERR(handle);
1804                 handle = NULL;
1805                 goto out;
1806         }
1807
1808         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1809                                    OCFS2_JOURNAL_ACCESS_WRITE);
1810         if (ret < 0) {
1811                 mlog_errno(ret);
1812                 goto out_commit;
1813         }
1814
1815         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
1816                offsetof(struct ocfs2_xattr_block, xb_attrs));
1817
1818         xb->xb_flags = cpu_to_le16(xb_flags & ~OCFS2_XATTR_INDEXED);
1819
1820         ocfs2_journal_dirty(handle, xs->xattr_bh);
1821
1822 out_commit:
1823         ocfs2_commit_trans(osb, handle);
1824 out:
1825         return ret;
1826 }
1827
1828 /*
1829  * ocfs2_xattr_block_set()
1830  *
1831  * Set, replace or remove an extended attribute into external block.
1832  *
1833  */
1834 static int ocfs2_xattr_block_set(struct inode *inode,
1835                                  struct ocfs2_xattr_info *xi,
1836                                  struct ocfs2_xattr_search *xs)
1837 {
1838         struct buffer_head *new_bh = NULL;
1839         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1840         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
1841         struct ocfs2_alloc_context *meta_ac = NULL;
1842         handle_t *handle = NULL;
1843         struct ocfs2_xattr_block *xblk = NULL;
1844         u16 suballoc_bit_start;
1845         u32 num_got;
1846         u64 first_blkno;
1847         int ret;
1848
1849         if (!xs->xattr_bh) {
1850                 /*
1851                  * Alloc one external block for extended attribute
1852                  * outside of inode.
1853                  */
1854                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1855                 if (ret < 0) {
1856                         mlog_errno(ret);
1857                         goto out;
1858                 }
1859                 handle = ocfs2_start_trans(osb,
1860                                            OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1861                 if (IS_ERR(handle)) {
1862                         ret = PTR_ERR(handle);
1863                         mlog_errno(ret);
1864                         goto out;
1865                 }
1866                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1867                                            OCFS2_JOURNAL_ACCESS_CREATE);
1868                 if (ret < 0) {
1869                         mlog_errno(ret);
1870                         goto out_commit;
1871                 }
1872
1873                 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1874                                            &suballoc_bit_start, &num_got,
1875                                            &first_blkno);
1876                 if (ret < 0) {
1877                         mlog_errno(ret);
1878                         goto out_commit;
1879                 }
1880
1881                 new_bh = sb_getblk(inode->i_sb, first_blkno);
1882                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1883
1884                 ret = ocfs2_journal_access(handle, inode, new_bh,
1885                                            OCFS2_JOURNAL_ACCESS_CREATE);
1886                 if (ret < 0) {
1887                         mlog_errno(ret);
1888                         goto out_commit;
1889                 }
1890
1891                 /* Initialize ocfs2_xattr_block */
1892                 xs->xattr_bh = new_bh;
1893                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1894                 memset(xblk, 0, inode->i_sb->s_blocksize);
1895                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1896                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1897                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1898                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1899                 xblk->xb_blkno = cpu_to_le64(first_blkno);
1900
1901                 xs->header = &xblk->xb_attrs.xb_header;
1902                 xs->base = (void *)xs->header;
1903                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1904                 xs->here = xs->header->xh_entries;
1905
1906
1907                 ret = ocfs2_journal_dirty(handle, new_bh);
1908                 if (ret < 0) {
1909                         mlog_errno(ret);
1910                         goto out_commit;
1911                 }
1912                 di->i_xattr_loc = cpu_to_le64(first_blkno);
1913                 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1914                 if (ret < 0)
1915                         mlog_errno(ret);
1916 out_commit:
1917                 ocfs2_commit_trans(osb, handle);
1918 out:
1919                 if (meta_ac)
1920                         ocfs2_free_alloc_context(meta_ac);
1921                 if (ret < 0)
1922                         return ret;
1923         } else
1924                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1925
1926         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1927                 /* Set extended attribute into external block */
1928                 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
1929                 if (!ret || ret != -ENOSPC)
1930                         goto end;
1931
1932                 ret = ocfs2_xattr_create_index_block(inode, xs);
1933                 if (ret)
1934                         goto end;
1935         }
1936
1937         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
1938         if (!ret && xblk->xb_attrs.xb_root.xt_list.l_next_free_rec == 0)
1939                 ret = ocfs2_restore_xattr_block(inode, xs);
1940
1941 end:
1942
1943         return ret;
1944 }
1945
1946 /*
1947  * ocfs2_xattr_set()
1948  *
1949  * Set, replace or remove an extended attribute for this inode.
1950  * value is NULL to remove an existing extended attribute, else either
1951  * create or replace an extended attribute.
1952  */
1953 int ocfs2_xattr_set(struct inode *inode,
1954                     int name_index,
1955                     const char *name,
1956                     const void *value,
1957                     size_t value_len,
1958                     int flags)
1959 {
1960         struct buffer_head *di_bh = NULL;
1961         struct ocfs2_dinode *di;
1962         int ret;
1963         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
1964
1965         struct ocfs2_xattr_info xi = {
1966                 .name_index = name_index,
1967                 .name = name,
1968                 .value = value,
1969                 .value_len = value_len,
1970         };
1971
1972         struct ocfs2_xattr_search xis = {
1973                 .not_found = -ENODATA,
1974         };
1975
1976         struct ocfs2_xattr_search xbs = {
1977                 .not_found = -ENODATA,
1978         };
1979
1980         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1981         if (ret < 0) {
1982                 mlog_errno(ret);
1983                 return ret;
1984         }
1985         xis.inode_bh = xbs.inode_bh = di_bh;
1986         di = (struct ocfs2_dinode *)di_bh->b_data;
1987
1988         down_write(&OCFS2_I(inode)->ip_xattr_sem);
1989         /*
1990          * Scan inode and external block to find the same name
1991          * extended attribute and collect search infomation.
1992          */
1993         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
1994         if (ret)
1995                 goto cleanup;
1996         if (xis.not_found) {
1997                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
1998                 if (ret)
1999                         goto cleanup;
2000         }
2001
2002         if (xis.not_found && xbs.not_found) {
2003                 ret = -ENODATA;
2004                 if (flags & XATTR_REPLACE)
2005                         goto cleanup;
2006                 ret = 0;
2007                 if (!value)
2008                         goto cleanup;
2009         } else {
2010                 ret = -EEXIST;
2011                 if (flags & XATTR_CREATE)
2012                         goto cleanup;
2013         }
2014
2015         if (!value) {
2016                 /* Remove existing extended attribute */
2017                 if (!xis.not_found)
2018                         ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2019                 else if (!xbs.not_found)
2020                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2021         } else {
2022                 /* We always try to set extended attribute into inode first*/
2023                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2024                 if (!ret && !xbs.not_found) {
2025                         /*
2026                          * If succeed and that extended attribute existing in
2027                          * external block, then we will remove it.
2028                          */
2029                         xi.value = NULL;
2030                         xi.value_len = 0;
2031                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2032                 } else if (ret == -ENOSPC) {
2033                         if (di->i_xattr_loc && !xbs.xattr_bh) {
2034                                 ret = ocfs2_xattr_block_find(inode, name_index,
2035                                                              name, &xbs);
2036                                 if (ret)
2037                                         goto cleanup;
2038                         }
2039                         /*
2040                          * If no space in inode, we will set extended attribute
2041                          * into external block.
2042                          */
2043                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2044                         if (ret)
2045                                 goto cleanup;
2046                         if (!xis.not_found) {
2047                                 /*
2048                                  * If succeed and that extended attribute
2049                                  * existing in inode, we will remove it.
2050                                  */
2051                                 xi.value = NULL;
2052                                 xi.value_len = 0;
2053                                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2054                         }
2055                 }
2056         }
2057 cleanup:
2058         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2059         ocfs2_inode_unlock(inode, 1);
2060         brelse(di_bh);
2061         brelse(xbs.xattr_bh);
2062         for (i = 0; i < blk_per_bucket; i++)
2063                 brelse(xbs.bucket.bhs[i]);
2064
2065         return ret;
2066 }
2067
2068 static inline u32 ocfs2_xattr_hash_by_name(struct inode *inode,
2069                                            int name_index,
2070                                            const char *suffix_name)
2071 {
2072         struct xattr_handler *handler = ocfs2_xattr_handler(name_index);
2073         char *prefix = handler->prefix;
2074         int prefix_len = strlen(handler->prefix);
2075
2076         return ocfs2_xattr_name_hash(inode, prefix, prefix_len,
2077                                      (char *)suffix_name, strlen(suffix_name));
2078 }
2079
2080 /*
2081  * Find the xattr extent rec which may contains name_hash.
2082  * e_cpos will be the first name hash of the xattr rec.
2083  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2084  */
2085 static int ocfs2_xattr_get_rec(struct inode *inode,
2086                                u32 name_hash,
2087                                u64 *p_blkno,
2088                                u32 *e_cpos,
2089                                u32 *num_clusters,
2090                                struct ocfs2_extent_list *el)
2091 {
2092         int ret = 0, i;
2093         struct buffer_head *eb_bh = NULL;
2094         struct ocfs2_extent_block *eb;
2095         struct ocfs2_extent_rec *rec = NULL;
2096         u64 e_blkno = 0;
2097
2098         if (el->l_tree_depth) {
2099                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2100                 if (ret) {
2101                         mlog_errno(ret);
2102                         goto out;
2103                 }
2104
2105                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2106                 el = &eb->h_list;
2107
2108                 if (el->l_tree_depth) {
2109                         ocfs2_error(inode->i_sb,
2110                                     "Inode %lu has non zero tree depth in "
2111                                     "xattr tree block %llu\n", inode->i_ino,
2112                                     (unsigned long long)eb_bh->b_blocknr);
2113                         ret = -EROFS;
2114                         goto out;
2115                 }
2116         }
2117
2118         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2119                 rec = &el->l_recs[i];
2120
2121                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2122                         e_blkno = le64_to_cpu(rec->e_blkno);
2123                         break;
2124                 }
2125         }
2126
2127         if (!e_blkno) {
2128                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2129                             "record (%u, %u, 0) in xattr", inode->i_ino,
2130                             le32_to_cpu(rec->e_cpos),
2131                             ocfs2_rec_clusters(el, rec));
2132                 ret = -EROFS;
2133                 goto out;
2134         }
2135
2136         *p_blkno = le64_to_cpu(rec->e_blkno);
2137         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2138         if (e_cpos)
2139                 *e_cpos = le32_to_cpu(rec->e_cpos);
2140 out:
2141         brelse(eb_bh);
2142         return ret;
2143 }
2144
2145 typedef int (xattr_bucket_func)(struct inode *inode,
2146                                 struct ocfs2_xattr_bucket *bucket,
2147                                 void *para);
2148
2149 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2150                                    struct buffer_head *header_bh,
2151                                    int name_index,
2152                                    const char *name,
2153                                    u32 name_hash,
2154                                    u16 *xe_index,
2155                                    int *found)
2156 {
2157         int i, ret = 0, cmp = 1, block_off, new_offset;
2158         struct ocfs2_xattr_header *xh =
2159                         (struct ocfs2_xattr_header *)header_bh->b_data;
2160         size_t name_len = strlen(name);
2161         struct ocfs2_xattr_entry *xe = NULL;
2162         struct buffer_head *name_bh = NULL;
2163         char *xe_name;
2164
2165         /*
2166          * We don't use binary search in the bucket because there
2167          * may be multiple entries with the same name hash.
2168          */
2169         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2170                 xe = &xh->xh_entries[i];
2171
2172                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2173                         continue;
2174                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2175                         break;
2176
2177                 cmp = name_index - ocfs2_xattr_get_type(xe);
2178                 if (!cmp)
2179                         cmp = name_len - xe->xe_name_len;
2180                 if (cmp)
2181                         continue;
2182
2183                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2184                                                         xh,
2185                                                         i,
2186                                                         &block_off,
2187                                                         &new_offset);
2188                 if (ret) {
2189                         mlog_errno(ret);
2190                         break;
2191                 }
2192
2193                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
2194                                        header_bh->b_blocknr + block_off,
2195                                        &name_bh, OCFS2_BH_CACHED, inode);
2196                 if (ret) {
2197                         mlog_errno(ret);
2198                         break;
2199                 }
2200                 xe_name = name_bh->b_data + new_offset;
2201
2202                 cmp = memcmp(name, xe_name, name_len);
2203                 brelse(name_bh);
2204                 name_bh = NULL;
2205
2206                 if (cmp == 0) {
2207                         *xe_index = i;
2208                         *found = 1;
2209                         ret = 0;
2210                         break;
2211                 }
2212         }
2213
2214         return ret;
2215 }
2216
2217 /*
2218  * Find the specified xattr entry in a series of buckets.
2219  * This series start from p_blkno and last for num_clusters.
2220  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2221  * the num of the valid buckets.
2222  *
2223  * Return the buffer_head this xattr should reside in. And if the xattr's
2224  * hash is in the gap of 2 buckets, return the lower bucket.
2225  */
2226 static int ocfs2_xattr_bucket_find(struct inode *inode,
2227                                    int name_index,
2228                                    const char *name,
2229                                    u32 name_hash,
2230                                    u64 p_blkno,
2231                                    u32 first_hash,
2232                                    u32 num_clusters,
2233                                    struct ocfs2_xattr_search *xs)
2234 {
2235         int ret, found = 0;
2236         struct buffer_head *bh = NULL;
2237         struct buffer_head *lower_bh = NULL;
2238         struct ocfs2_xattr_header *xh = NULL;
2239         struct ocfs2_xattr_entry *xe = NULL;
2240         u16 index = 0;
2241         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2242         int low_bucket = 0, bucket, high_bucket;
2243         u32 last_hash;
2244         u64 blkno;
2245
2246         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno,
2247                                &bh, OCFS2_BH_CACHED, inode);
2248         if (ret) {
2249                 mlog_errno(ret);
2250                 goto out;
2251         }
2252
2253         xh = (struct ocfs2_xattr_header *)bh->b_data;
2254         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2255
2256         while (low_bucket <= high_bucket) {
2257                 brelse(bh);
2258                 bh = NULL;
2259                 bucket = (low_bucket + high_bucket) / 2;
2260
2261                 blkno = p_blkno + bucket * blk_per_bucket;
2262
2263                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
2264                                        &bh, OCFS2_BH_CACHED, inode);
2265                 if (ret) {
2266                         mlog_errno(ret);
2267                         goto out;
2268                 }
2269
2270                 xh = (struct ocfs2_xattr_header *)bh->b_data;
2271                 xe = &xh->xh_entries[0];
2272                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2273                         high_bucket = bucket - 1;
2274                         continue;
2275                 }
2276
2277                 /*
2278                  * Check whether the hash of the last entry in our
2279                  * bucket is larger than the search one.
2280                  */
2281                 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2282                 last_hash = le32_to_cpu(xe->xe_name_hash);
2283
2284                 /* record lower_bh which may be the insert place. */
2285                 brelse(lower_bh);
2286                 lower_bh = bh;
2287                 bh = NULL;
2288
2289                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2290                         low_bucket = bucket + 1;
2291                         continue;
2292                 }
2293
2294                 /* the searched xattr should reside in this bucket if exists. */
2295                 ret = ocfs2_find_xe_in_bucket(inode, lower_bh,
2296                                               name_index, name, name_hash,
2297                                               &index, &found);
2298                 if (ret) {
2299                         mlog_errno(ret);
2300                         goto out;
2301                 }
2302                 break;
2303         }
2304
2305         /*
2306          * Record the bucket we have found.
2307          * When the xattr's hash value is in the gap of 2 buckets, we will
2308          * always set it to the previous bucket.
2309          */
2310         if (!lower_bh) {
2311                 /*
2312                  * We can't find any bucket whose first name_hash is less
2313                  * than the find name_hash.
2314                  */
2315                 BUG_ON(bh->b_blocknr != p_blkno);
2316                 lower_bh = bh;
2317                 bh = NULL;
2318         }
2319         xs->bucket.bhs[0] = lower_bh;
2320         xs->bucket.xh = (struct ocfs2_xattr_header *)
2321                                         xs->bucket.bhs[0]->b_data;
2322         lower_bh = NULL;
2323
2324         xs->header = xs->bucket.xh;
2325         xs->base = xs->bucket.bhs[0]->b_data;
2326         xs->end = xs->base + inode->i_sb->s_blocksize;
2327
2328         if (found) {
2329                 /*
2330                  * If we have found the xattr enty, read all the blocks in
2331                  * this bucket.
2332                  */
2333                 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2334                                         xs->bucket.bhs[0]->b_blocknr + 1,
2335                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2336                                         OCFS2_BH_CACHED, inode);
2337                 if (ret) {
2338                         mlog_errno(ret);
2339                         goto out;
2340                 }
2341
2342                 xs->here = &xs->header->xh_entries[index];
2343                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2344                      (unsigned long long)xs->bucket.bhs[0]->b_blocknr, index);
2345         } else
2346                 ret = -ENODATA;
2347
2348 out:
2349         brelse(bh);
2350         brelse(lower_bh);
2351         return ret;
2352 }
2353
2354 static int ocfs2_xattr_index_block_find(struct inode *inode,
2355                                         struct buffer_head *root_bh,
2356                                         int name_index,
2357                                         const char *name,
2358                                         struct ocfs2_xattr_search *xs)
2359 {
2360         int ret;
2361         struct ocfs2_xattr_block *xb =
2362                         (struct ocfs2_xattr_block *)root_bh->b_data;
2363         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2364         struct ocfs2_extent_list *el = &xb_root->xt_list;
2365         u64 p_blkno = 0;
2366         u32 first_hash, num_clusters = 0;
2367         u32 name_hash = ocfs2_xattr_hash_by_name(inode, name_index, name);
2368
2369         if (le16_to_cpu(el->l_next_free_rec) == 0)
2370                 return -ENODATA;
2371
2372         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2373              name, name_hash, name_index);
2374
2375         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2376                                   &num_clusters, el);
2377         if (ret) {
2378                 mlog_errno(ret);
2379                 goto out;
2380         }
2381
2382         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2383
2384         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2385              "in the rec is %u\n", num_clusters, p_blkno, first_hash);
2386
2387         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2388                                       p_blkno, first_hash, num_clusters, xs);
2389
2390 out:
2391         return ret;
2392 }
2393
2394 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2395                                        u64 blkno,
2396                                        u32 clusters,
2397                                        xattr_bucket_func *func,
2398                                        void *para)
2399 {
2400         int i, j, ret = 0;
2401         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2402         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2403         u32 num_buckets = clusters * bpc;
2404         struct ocfs2_xattr_bucket bucket;
2405
2406         memset(&bucket, 0, sizeof(bucket));
2407
2408         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2409              clusters, blkno);
2410
2411         for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
2412                 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2413                                         blkno, blk_per_bucket,
2414                                         bucket.bhs, OCFS2_BH_CACHED, inode);
2415                 if (ret) {
2416                         mlog_errno(ret);
2417                         goto out;
2418                 }
2419
2420                 bucket.xh = (struct ocfs2_xattr_header *)bucket.bhs[0]->b_data;
2421                 /*
2422                  * The real bucket num in this series of blocks is stored
2423                  * in the 1st bucket.
2424                  */
2425                 if (i == 0)
2426                         num_buckets = le16_to_cpu(bucket.xh->xh_num_buckets);
2427
2428                 mlog(0, "iterating xattr bucket %llu\n", blkno);
2429                 if (func) {
2430                         ret = func(inode, &bucket, para);
2431                         if (ret) {
2432                                 mlog_errno(ret);
2433                                 break;
2434                         }
2435                 }
2436
2437                 for (j = 0; j < blk_per_bucket; j++)
2438                         brelse(bucket.bhs[j]);
2439                 memset(&bucket, 0, sizeof(bucket));
2440         }
2441
2442 out:
2443         for (j = 0; j < blk_per_bucket; j++)
2444                 brelse(bucket.bhs[j]);
2445
2446         return ret;
2447 }
2448
2449 struct ocfs2_xattr_tree_list {
2450         char *buffer;
2451         size_t buffer_size;
2452 };
2453
2454 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2455                                              struct ocfs2_xattr_header *xh,
2456                                              int index,
2457                                              int *block_off,
2458                                              int *new_offset)
2459 {
2460         u16 name_offset;
2461
2462         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2463                 return -EINVAL;
2464
2465         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2466
2467         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2468         *new_offset = name_offset % inode->i_sb->s_blocksize;
2469
2470         return 0;
2471 }
2472
2473 static int ocfs2_list_xattr_bucket(struct inode *inode,
2474                                    struct ocfs2_xattr_bucket *bucket,
2475                                    void *para)
2476 {
2477         int ret = 0;
2478         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2479         size_t size;
2480         int i, block_off, new_offset;
2481
2482         for (i = 0 ; i < le16_to_cpu(bucket->xh->xh_count); i++) {
2483                 struct ocfs2_xattr_entry *entry = &bucket->xh->xh_entries[i];
2484                 struct xattr_handler *handler =
2485                         ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
2486
2487                 if (handler) {
2488                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2489                                                                 bucket->xh,
2490                                                                 i,
2491                                                                 &block_off,
2492                                                                 &new_offset);
2493                         if (ret)
2494                                 break;
2495                         size = handler->list(inode, xl->buffer, xl->buffer_size,
2496                                              bucket->bhs[block_off]->b_data +
2497                                              new_offset,
2498                                              entry->xe_name_len);
2499                         if (xl->buffer) {
2500                                 if (size > xl->buffer_size)
2501                                         return -ERANGE;
2502                                 xl->buffer += size;
2503                         }
2504                         xl->buffer_size -= size;
2505                 }
2506         }
2507
2508         return ret;
2509 }
2510
2511 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2512                                              struct ocfs2_xattr_tree_root *xt,
2513                                              char *buffer,
2514                                              size_t buffer_size)
2515 {
2516         struct ocfs2_extent_list *el = &xt->xt_list;
2517         int ret = 0;
2518         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2519         u64 p_blkno = 0;
2520         struct ocfs2_xattr_tree_list xl = {
2521                 .buffer = buffer,
2522                 .buffer_size = buffer_size,
2523         };
2524
2525         if (le16_to_cpu(el->l_next_free_rec) == 0)
2526                 return 0;
2527
2528         while (name_hash > 0) {
2529                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2530                                           &e_cpos, &num_clusters, el);
2531                 if (ret) {
2532                         mlog_errno(ret);
2533                         goto out;
2534                 }
2535
2536                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2537                                                   ocfs2_list_xattr_bucket,
2538                                                   &xl);
2539                 if (ret) {
2540                         mlog_errno(ret);
2541                         goto out;
2542                 }
2543
2544                 if (e_cpos == 0)
2545                         break;
2546
2547                 name_hash = e_cpos - 1;
2548         }
2549
2550         ret = buffer_size - xl.buffer_size;
2551 out:
2552         return ret;
2553 }
2554
2555 static int cmp_xe(const void *a, const void *b)
2556 {
2557         const struct ocfs2_xattr_entry *l = a, *r = b;
2558         u32 l_hash = le32_to_cpu(l->xe_name_hash);
2559         u32 r_hash = le32_to_cpu(r->xe_name_hash);
2560
2561         if (l_hash > r_hash)
2562                 return 1;
2563         if (l_hash < r_hash)
2564                 return -1;
2565         return 0;
2566 }
2567
2568 static void swap_xe(void *a, void *b, int size)
2569 {
2570         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2571
2572         tmp = *l;
2573         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2574         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2575 }
2576
2577 /*
2578  * When the ocfs2_xattr_block is filled up, new bucket will be created
2579  * and all the xattr entries will be moved to the new bucket.
2580  * Note: we need to sort the entries since they are not saved in order
2581  * in the ocfs2_xattr_block.
2582  */
2583 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2584                                            struct buffer_head *xb_bh,
2585                                            struct buffer_head *xh_bh,
2586                                            struct buffer_head *data_bh)
2587 {
2588         int i, blocksize = inode->i_sb->s_blocksize;
2589         u16 offset, size, off_change;
2590         struct ocfs2_xattr_entry *xe;
2591         struct ocfs2_xattr_block *xb =
2592                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
2593         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2594         struct ocfs2_xattr_header *xh =
2595                                 (struct ocfs2_xattr_header *)xh_bh->b_data;
2596         u16 count = le16_to_cpu(xb_xh->xh_count);
2597         char *target = xh_bh->b_data, *src = xb_bh->b_data;
2598
2599         mlog(0, "cp xattr from block %llu to bucket %llu\n",
2600              (unsigned long long)xb_bh->b_blocknr,
2601              (unsigned long long)xh_bh->b_blocknr);
2602
2603         memset(xh_bh->b_data, 0, blocksize);
2604         if (data_bh)
2605                 memset(data_bh->b_data, 0, blocksize);
2606         /*
2607          * Since the xe_name_offset is based on ocfs2_xattr_header,
2608          * there is a offset change corresponding to the change of
2609          * ocfs2_xattr_header's position.
2610          */
2611         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2612         xe = &xb_xh->xh_entries[count - 1];
2613         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2614         size = blocksize - offset;
2615
2616         /* copy all the names and values. */
2617         if (data_bh)
2618                 target = data_bh->b_data;
2619         memcpy(target + offset, src + offset, size);
2620
2621         /* Init new header now. */
2622         xh->xh_count = xb_xh->xh_count;
2623         xh->xh_num_buckets = cpu_to_le16(1);
2624         xh->xh_name_value_len = cpu_to_le16(size);
2625         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2626
2627         /* copy all the entries. */
2628         target = xh_bh->b_data;
2629         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2630         size = count * sizeof(struct ocfs2_xattr_entry);
2631         memcpy(target + offset, (char *)xb_xh + offset, size);
2632
2633         /* Change the xe offset for all the xe because of the move. */
2634         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2635                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2636         for (i = 0; i < count; i++)
2637                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2638
2639         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2640              offset, size, off_change);
2641
2642         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2643              cmp_xe, swap_xe);
2644 }
2645
2646 /*
2647  * After we move xattr from block to index btree, we have to
2648  * update ocfs2_xattr_search to the new xe and base.
2649  *
2650  * When the entry is in xattr block, xattr_bh indicates the storage place.
2651  * While if the entry is in index b-tree, "bucket" indicates the
2652  * real place of the xattr.
2653  */
2654 static int ocfs2_xattr_update_xattr_search(struct inode *inode,
2655                                            struct ocfs2_xattr_search *xs,
2656                                            struct buffer_head *old_bh,
2657                                            struct buffer_head *new_bh)
2658 {
2659         int ret = 0;
2660         char *buf = old_bh->b_data;
2661         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2662         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2663         int i, blocksize = inode->i_sb->s_blocksize;
2664         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2665
2666         xs->bucket.bhs[0] = new_bh;
2667         get_bh(new_bh);
2668         xs->bucket.xh = (struct ocfs2_xattr_header *)xs->bucket.bhs[0]->b_data;
2669         xs->header = xs->bucket.xh;
2670
2671         xs->base = new_bh->b_data;
2672         xs->end = xs->base + inode->i_sb->s_blocksize;
2673
2674         if (!xs->not_found) {
2675                 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) {
2676                         ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2677                                         xs->bucket.bhs[0]->b_blocknr + 1,
2678                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2679                                         OCFS2_BH_CACHED, inode);
2680                         if (ret) {
2681                                 mlog_errno(ret);
2682                                 return ret;
2683                         }
2684
2685                         i = xs->here - old_xh->xh_entries;
2686                         xs->here = &xs->header->xh_entries[i];
2687                 }
2688         }
2689
2690         return ret;
2691 }
2692
2693 static int ocfs2_xattr_create_index_block(struct inode *inode,
2694                                           struct ocfs2_xattr_search *xs)
2695 {
2696         int ret, credits = OCFS2_SUBALLOC_ALLOC;
2697         u32 bit_off, len;
2698         u64 blkno;
2699         handle_t *handle;
2700         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2701         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2702         struct ocfs2_alloc_context *data_ac;
2703         struct buffer_head *xh_bh = NULL, *data_bh = NULL;
2704         struct buffer_head *xb_bh = xs->xattr_bh;
2705         struct ocfs2_xattr_block *xb =
2706                         (struct ocfs2_xattr_block *)xb_bh->b_data;
2707         struct ocfs2_xattr_tree_root *xr;
2708         u16 xb_flags = le16_to_cpu(xb->xb_flags);
2709         u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2710
2711         mlog(0, "create xattr index block for %llu\n",
2712              (unsigned long long)xb_bh->b_blocknr);
2713
2714         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2715
2716         ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2717         if (ret) {
2718                 mlog_errno(ret);
2719                 goto out;
2720         }
2721
2722         /*
2723          * XXX:
2724          * We can use this lock for now, and maybe move to a dedicated mutex
2725          * if performance becomes a problem later.
2726          */
2727         down_write(&oi->ip_alloc_sem);
2728
2729         /*
2730          * 3 more credits, one for xattr block update, one for the 1st block
2731          * of the new xattr bucket and one for the value/data.
2732          */
2733         credits += 3;
2734         handle = ocfs2_start_trans(osb, credits);
2735         if (IS_ERR(handle)) {
2736                 ret = PTR_ERR(handle);
2737                 mlog_errno(ret);
2738                 goto out_sem;
2739         }
2740
2741         ret = ocfs2_journal_access(handle, inode, xb_bh,
2742                                    OCFS2_JOURNAL_ACCESS_WRITE);
2743         if (ret) {
2744                 mlog_errno(ret);
2745                 goto out_commit;
2746         }
2747
2748         ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2749         if (ret) {
2750                 mlog_errno(ret);
2751                 goto out_commit;
2752         }
2753
2754         /*
2755          * The bucket may spread in many blocks, and
2756          * we will only touch the 1st block and the last block
2757          * in the whole bucket(one for entry and one for data).
2758          */
2759         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2760
2761         mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno);
2762
2763         xh_bh = sb_getblk(inode->i_sb, blkno);
2764         if (!xh_bh) {
2765                 ret = -EIO;
2766                 mlog_errno(ret);
2767                 goto out_commit;
2768         }
2769
2770         ocfs2_set_new_buffer_uptodate(inode, xh_bh);
2771
2772         ret = ocfs2_journal_access(handle, inode, xh_bh,
2773                                    OCFS2_JOURNAL_ACCESS_CREATE);
2774         if (ret) {
2775                 mlog_errno(ret);
2776                 goto out_commit;
2777         }
2778
2779         if (bpb > 1) {
2780                 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1);
2781                 if (!data_bh) {
2782                         ret = -EIO;
2783                         mlog_errno(ret);
2784                         goto out_commit;
2785                 }
2786
2787                 ocfs2_set_new_buffer_uptodate(inode, data_bh);
2788
2789                 ret = ocfs2_journal_access(handle, inode, data_bh,
2790                                            OCFS2_JOURNAL_ACCESS_CREATE);
2791                 if (ret) {
2792                         mlog_errno(ret);
2793                         goto out_commit;
2794                 }
2795         }
2796
2797         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh);
2798
2799         ocfs2_journal_dirty(handle, xh_bh);
2800         if (data_bh)
2801                 ocfs2_journal_dirty(handle, data_bh);
2802
2803         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
2804
2805         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2806         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2807                offsetof(struct ocfs2_xattr_block, xb_attrs));
2808
2809         xr = &xb->xb_attrs.xb_root;
2810         xr->xt_clusters = cpu_to_le32(1);
2811         xr->xt_last_eb_blk = 0;
2812         xr->xt_list.l_tree_depth = 0;
2813         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2814         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2815
2816         xr->xt_list.l_recs[0].e_cpos = 0;
2817         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2818         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2819
2820         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2821
2822         ret = ocfs2_journal_dirty(handle, xb_bh);
2823         if (ret) {
2824                 mlog_errno(ret);
2825                 goto out_commit;
2826         }
2827
2828 out_commit:
2829         ocfs2_commit_trans(osb, handle);
2830
2831 out_sem:
2832         up_write(&oi->ip_alloc_sem);
2833
2834 out:
2835         if (data_ac)
2836                 ocfs2_free_alloc_context(data_ac);
2837
2838         brelse(xh_bh);
2839         brelse(data_bh);
2840
2841         return ret;
2842 }
2843
2844 static int cmp_xe_offset(const void *a, const void *b)
2845 {
2846         const struct ocfs2_xattr_entry *l = a, *r = b;
2847         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2848         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2849
2850         if (l_name_offset < r_name_offset)
2851                 return 1;
2852         if (l_name_offset > r_name_offset)
2853                 return -1;
2854         return 0;
2855 }
2856
2857 /*
2858  * defrag a xattr bucket if we find that the bucket has some
2859  * holes beteen name/value pairs.
2860  * We will move all the name/value pairs to the end of the bucket
2861  * so that we can spare some space for insertion.
2862  */
2863 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2864                                      struct ocfs2_xattr_bucket *bucket)
2865 {
2866         int ret, i;
2867         size_t end, offset, len, value_len;
2868         struct ocfs2_xattr_header *xh;
2869         char *entries, *buf, *bucket_buf = NULL;
2870         u64 blkno = bucket->bhs[0]->b_blocknr;
2871         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2872         u16 xh_free_start;
2873         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2874         size_t blocksize = inode->i_sb->s_blocksize;
2875         handle_t *handle;
2876         struct buffer_head **bhs;
2877         struct ocfs2_xattr_entry *xe;
2878
2879         bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
2880                         GFP_NOFS);
2881         if (!bhs)
2882                 return -ENOMEM;
2883
2884         ret = ocfs2_read_blocks(osb, blkno, blk_per_bucket, bhs,
2885                                 OCFS2_BH_CACHED, inode);
2886         if (ret)
2887                 goto out;
2888
2889         /*
2890          * In order to make the operation more efficient and generic,
2891          * we copy all the blocks into a contiguous memory and do the
2892          * defragment there, so if anything is error, we will not touch
2893          * the real block.
2894          */
2895         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2896         if (!bucket_buf) {
2897                 ret = -EIO;
2898                 goto out;
2899         }
2900
2901         buf = bucket_buf;
2902         for (i = 0; i < blk_per_bucket; i++, buf += blocksize)
2903                 memcpy(buf, bhs[i]->b_data, blocksize);
2904
2905         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket);
2906         if (IS_ERR(handle)) {
2907                 ret = PTR_ERR(handle);
2908                 handle = NULL;
2909                 mlog_errno(ret);
2910                 goto out;
2911         }
2912
2913         for (i = 0; i < blk_per_bucket; i++) {
2914                 ret = ocfs2_journal_access(handle, inode, bhs[i],
2915                                            OCFS2_JOURNAL_ACCESS_WRITE);
2916                 if (ret < 0) {
2917                         mlog_errno(ret);
2918                         goto commit;
2919                 }
2920         }
2921
2922         xh = (struct ocfs2_xattr_header *)bucket_buf;
2923         entries = (char *)xh->xh_entries;
2924         xh_free_start = le16_to_cpu(xh->xh_free_start);
2925
2926         mlog(0, "adjust xattr bucket in %llu, count = %u, "
2927              "xh_free_start = %u, xh_name_value_len = %u.\n",
2928              blkno, le16_to_cpu(xh->xh_count), xh_free_start,
2929              le16_to_cpu(xh->xh_name_value_len));
2930
2931         /*
2932          * sort all the entries by their offset.
2933          * the largest will be the first, so that we can
2934          * move them to the end one by one.
2935          */
2936         sort(entries, le16_to_cpu(xh->xh_count),
2937              sizeof(struct ocfs2_xattr_entry),
2938              cmp_xe_offset, swap_xe);
2939
2940         /* Move all name/values to the end of the bucket. */
2941         xe = xh->xh_entries;
2942         end = OCFS2_XATTR_BUCKET_SIZE;
2943         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
2944                 offset = le16_to_cpu(xe->xe_name_offset);
2945                 if (ocfs2_xattr_is_local(xe))
2946                         value_len = OCFS2_XATTR_SIZE(
2947                                         le64_to_cpu(xe->xe_value_size));
2948                 else
2949                         value_len = OCFS2_XATTR_ROOT_SIZE;
2950                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
2951
2952                 /*
2953                  * We must make sure that the name/value pair
2954                  * exist in the same block. So adjust end to
2955                  * the previous block end if needed.
2956                  */
2957                 if (((end - len) / blocksize !=
2958                         (end - 1) / blocksize))
2959                         end = end - end % blocksize;
2960
2961                 if (end > offset + len) {
2962                         memmove(bucket_buf + end - len,
2963                                 bucket_buf + offset, len);
2964                         xe->xe_name_offset = cpu_to_le16(end - len);
2965                 }
2966
2967                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
2968                                 "bucket %llu\n", (unsigned long long)blkno);
2969
2970                 end -= len;
2971         }
2972
2973         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
2974                         "bucket %llu\n", (unsigned long long)blkno);
2975
2976         if (xh_free_start == end)
2977                 goto commit;
2978
2979         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
2980         xh->xh_free_start = cpu_to_le16(end);
2981
2982         /* sort the entries by their name_hash. */
2983         sort(entries, le16_to_cpu(xh->xh_count),
2984              sizeof(struct ocfs2_xattr_entry),
2985              cmp_xe, swap_xe);
2986
2987         buf = bucket_buf;
2988         for (i = 0; i < blk_per_bucket; i++, buf += blocksize) {
2989                 memcpy(bhs[i]->b_data, buf, blocksize);
2990                 ocfs2_journal_dirty(handle, bhs[i]);
2991         }
2992
2993 commit:
2994         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2995 out:
2996
2997         if (bhs) {
2998                 for (i = 0; i < blk_per_bucket; i++)
2999                         brelse(bhs[i]);
3000         }
3001         kfree(bhs);
3002
3003         kfree(bucket_buf);
3004         return ret;
3005 }
3006
3007 /*
3008  * Move half nums of the xattr bucket in the previous cluster to this new
3009  * cluster. We only touch the last cluster of the previous extend record.
3010  *
3011  * first_bh is the first buffer_head of a series of bucket in the same
3012  * extent rec and header_bh is the header of one bucket in this cluster.
3013  * They will be updated if we move the data header_bh contains to the new
3014  * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3015  */
3016 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3017                                                handle_t *handle,
3018                                                struct buffer_head **first_bh,
3019                                                struct buffer_head **header_bh,
3020                                                u64 new_blkno,
3021                                                u64 prev_blkno,
3022                                                u32 num_clusters,
3023                                                u32 *first_hash)
3024 {
3025         int i, ret, credits;
3026         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3027         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3028         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3029         int blocksize = inode->i_sb->s_blocksize;
3030         struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3031         struct ocfs2_xattr_header *new_xh;
3032         struct ocfs2_xattr_header *xh =
3033                         (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3034
3035         BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3036         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3037
3038         prev_bh = *first_bh;
3039         get_bh(prev_bh);
3040         xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3041
3042         prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3043
3044         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3045              prev_blkno, new_blkno);
3046
3047         /*
3048          * We need to update the 1st half of the new cluster and
3049          * 1 more for the update of the 1st bucket of the previous
3050          * extent record.
3051          */
3052         credits = bpc / 2 + 1;
3053         ret = ocfs2_extend_trans(handle, credits);
3054         if (ret) {
3055                 mlog_errno(ret);
3056                 goto out;
3057         }
3058
3059         ret = ocfs2_journal_access(handle, inode, prev_bh,
3060                                    OCFS2_JOURNAL_ACCESS_WRITE);
3061         if (ret) {
3062                 mlog_errno(ret);
3063                 goto out;
3064         }
3065
3066         for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3067                 old_bh = new_bh = NULL;
3068                 new_bh = sb_getblk(inode->i_sb, new_blkno);
3069                 if (!new_bh) {
3070                         ret = -EIO;
3071                         mlog_errno(ret);
3072                         goto out;
3073                 }
3074
3075                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3076
3077                 ret = ocfs2_journal_access(handle, inode, new_bh,
3078                                            OCFS2_JOURNAL_ACCESS_CREATE);
3079                 if (ret < 0) {
3080                         mlog_errno(ret);
3081                         brelse(new_bh);
3082                         goto out;
3083                 }
3084
3085                 ret = ocfs2_read_block(osb, prev_blkno,
3086                                        &old_bh, OCFS2_BH_CACHED, inode);
3087                 if (ret < 0) {
3088                         mlog_errno(ret);
3089                         brelse(new_bh);
3090                         goto out;
3091                 }
3092
3093                 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3094
3095                 if (i == 0) {
3096                         new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3097                         new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3098
3099                         if (first_hash)
3100                                 *first_hash = le32_to_cpu(
3101                                         new_xh->xh_entries[0].xe_name_hash);
3102                         new_first_bh = new_bh;
3103                         get_bh(new_first_bh);
3104                 }
3105
3106                 ocfs2_journal_dirty(handle, new_bh);
3107
3108                 if (*header_bh == old_bh) {
3109                         brelse(*header_bh);
3110                         *header_bh = new_bh;
3111                         get_bh(*header_bh);
3112
3113                         brelse(*first_bh);
3114                         *first_bh = new_first_bh;
3115                         get_bh(*first_bh);
3116                 }
3117                 brelse(new_bh);
3118                 brelse(old_bh);
3119         }
3120
3121         le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3122
3123         ocfs2_journal_dirty(handle, prev_bh);
3124 out:
3125         brelse(prev_bh);
3126         brelse(new_first_bh);
3127         return ret;
3128 }
3129
3130 static int ocfs2_read_xattr_bucket(struct inode *inode,
3131                                    u64 blkno,
3132                                    struct buffer_head **bhs,
3133                                    int new)
3134 {
3135         int ret = 0;
3136         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3137
3138         if (!new)
3139                 return ocfs2_read_blocks(OCFS2_SB(inode->i_sb), blkno,
3140                                          blk_per_bucket, bhs,
3141                                          OCFS2_BH_CACHED, inode);
3142
3143         for (i = 0; i < blk_per_bucket; i++) {
3144                 bhs[i] = sb_getblk(inode->i_sb, blkno + i);
3145                 if (bhs[i] == NULL) {
3146                         ret = -EIO;
3147                         mlog_errno(ret);
3148                         break;
3149                 }
3150                 ocfs2_set_new_buffer_uptodate(inode, bhs[i]);
3151         }
3152
3153         return ret;
3154 }
3155
3156 /*
3157  * Move half num of the xattrs in old bucket(blk) to new bucket(new_blk).
3158  * first_hash will record the 1st hash of the new bucket.
3159  */
3160 static int ocfs2_half_xattr_bucket(struct inode *inode,
3161                                    handle_t *handle,
3162                                    u64 blk,
3163                                    u64 new_blk,
3164                                    u32 *first_hash,
3165                                    int new_bucket_head)
3166 {
3167         int ret, i;
3168         u16 count, start, len, name_value_len, xe_len, name_offset;
3169         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3170         struct buffer_head **s_bhs, **t_bhs = NULL;
3171         struct ocfs2_xattr_header *xh;
3172         struct ocfs2_xattr_entry *xe;
3173         int blocksize = inode->i_sb->s_blocksize;
3174
3175         mlog(0, "move half of xattrs from bucket %llu to %llu\n",
3176              blk, new_blk);
3177
3178         s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3179         if (!s_bhs)
3180                 return -ENOMEM;
3181
3182         ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0);
3183         if (ret) {
3184                 mlog_errno(ret);
3185                 goto out;
3186         }
3187
3188         ret = ocfs2_journal_access(handle, inode, s_bhs[0],
3189                                    OCFS2_JOURNAL_ACCESS_WRITE);
3190         if (ret) {
3191                 mlog_errno(ret);
3192                 goto out;
3193         }
3194
3195         t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3196         if (!t_bhs) {
3197                 ret = -ENOMEM;
3198                 goto out;
3199         }
3200
3201         ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head);
3202         if (ret) {
3203                 mlog_errno(ret);
3204                 goto out;
3205         }
3206
3207         for (i = 0; i < blk_per_bucket; i++) {
3208                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3209                                            OCFS2_JOURNAL_ACCESS_CREATE);
3210                 if (ret) {
3211                         mlog_errno(ret);
3212                         goto out;
3213                 }
3214         }
3215
3216         /* copy the whole bucket to the new first. */
3217         for (i = 0; i < blk_per_bucket; i++)
3218                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3219
3220         /* update the new bucket. */
3221         xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3222         count = le16_to_cpu(xh->xh_count);
3223         start = count / 2;
3224
3225         /*
3226          * Calculate the total name/value len and xh_free_start for
3227          * the old bucket first.
3228          */
3229         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3230         name_value_len = 0;
3231         for (i = 0; i < start; i++) {
3232                 xe = &xh->xh_entries[i];
3233                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3234                 if (ocfs2_xattr_is_local(xe))
3235                         xe_len +=
3236                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3237                 else
3238                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3239                 name_value_len += xe_len;
3240                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3241                         name_offset = le16_to_cpu(xe->xe_name_offset);
3242         }
3243
3244         /*
3245          * Now begin the modification to the new bucket.
3246          *
3247          * In the new bucket, We just move the xattr entry to the beginning
3248          * and don't touch the name/value. So there will be some holes in the
3249          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3250          * called.
3251          */
3252         xe = &xh->xh_entries[start];
3253         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3254         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3255              (char *)xe - (char *)xh, (char *)xh->xh_entries - (char *)xh);
3256         memmove((char *)xh->xh_entries, (char *)xe, len);
3257         xe = &xh->xh_entries[count - start];
3258         len = sizeof(struct ocfs2_xattr_entry) * start;
3259         memset((char *)xe, 0, len);
3260
3261         le16_add_cpu(&xh->xh_count, -start);
3262         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3263
3264         /* Calculate xh_free_start for the new bucket. */
3265         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3266         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3267                 xe = &xh->xh_entries[i];
3268                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3269                 if (ocfs2_xattr_is_local(xe))
3270                         xe_len +=
3271                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3272                 else
3273                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3274                 if (le16_to_cpu(xe->xe_name_offset) <
3275                     le16_to_cpu(xh->xh_free_start))
3276                         xh->xh_free_start = xe->xe_name_offset;
3277         }
3278
3279         /* set xh->xh_num_buckets for the new xh. */
3280         if (new_bucket_head)
3281                 xh->xh_num_buckets = cpu_to_le16(1);
3282         else
3283                 xh->xh_num_buckets = 0;
3284
3285         for (i = 0; i < blk_per_bucket; i++) {
3286                 ocfs2_journal_dirty(handle, t_bhs[i]);
3287                 if (ret)
3288                         mlog_errno(ret);
3289         }
3290
3291         /* store the first_hash of the new bucket. */
3292         if (first_hash)
3293                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3294
3295         /*
3296          * Now only update the 1st block of the old bucket.
3297          * Please note that the entry has been sorted already above.
3298          */
3299         xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3300         memset(&xh->xh_entries[start], 0,
3301                sizeof(struct ocfs2_xattr_entry) * (count - start));
3302         xh->xh_count = cpu_to_le16(start);
3303         xh->xh_free_start = cpu_to_le16(name_offset);
3304         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3305
3306         ocfs2_journal_dirty(handle, s_bhs[0]);
3307         if (ret)
3308                 mlog_errno(ret);
3309
3310 out:
3311         if (s_bhs) {
3312                 for (i = 0; i < blk_per_bucket; i++)
3313                         brelse(s_bhs[i]);
3314         }
3315         kfree(s_bhs);
3316
3317         if (t_bhs) {
3318                 for (i = 0; i < blk_per_bucket; i++)
3319                         brelse(t_bhs[i]);
3320         }
3321         kfree(t_bhs);
3322
3323         return ret;
3324 }
3325
3326 /*
3327  * Copy xattr from one bucket to another bucket.
3328  *
3329  * The caller must make sure that the journal transaction
3330  * has enough space for journaling.
3331  */
3332 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3333                                  handle_t *handle,
3334                                  u64 s_blkno,
3335                                  u64 t_blkno,
3336                                  int t_is_new)
3337 {
3338         int ret, i;
3339         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3340         int blocksize = inode->i_sb->s_blocksize;
3341         struct buffer_head **s_bhs, **t_bhs = NULL;
3342
3343         BUG_ON(s_blkno == t_blkno);
3344
3345         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3346              s_blkno, t_blkno, t_is_new);
3347
3348         s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3349                         GFP_NOFS);
3350         if (!s_bhs)
3351                 return -ENOMEM;
3352
3353         ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0);
3354         if (ret)
3355                 goto out;
3356
3357         t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3358                         GFP_NOFS);
3359         if (!t_bhs) {
3360                 ret = -ENOMEM;
3361                 goto out;
3362         }
3363
3364         ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new);
3365         if (ret)
3366                 goto out;
3367
3368         for (i = 0; i < blk_per_bucket; i++) {
3369                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3370                                            OCFS2_JOURNAL_ACCESS_WRITE);
3371                 if (ret)
3372                         goto out;
3373         }
3374
3375         for (i = 0; i < blk_per_bucket; i++) {
3376                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3377                 ocfs2_journal_dirty(handle, t_bhs[i]);
3378         }
3379
3380 out:
3381         if (s_bhs) {
3382                 for (i = 0; i < blk_per_bucket; i++)
3383                         brelse(s_bhs[i]);
3384         }
3385         kfree(s_bhs);
3386
3387         if (t_bhs) {
3388                 for (i = 0; i < blk_per_bucket; i++)
3389                         brelse(t_bhs[i]);
3390         }
3391         kfree(t_bhs);
3392
3393         return ret;
3394 }
3395
3396 /*
3397  * Copy one xattr cluster from src_blk to to_blk.
3398  * The to_blk will become the first bucket header of the cluster, so its
3399  * xh_num_buckets will be initialized as the bucket num in the cluster.
3400  */
3401 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3402                                   handle_t *handle,
3403                                   struct buffer_head *first_bh,
3404                                   u64 src_blk,
3405                                   u64 to_blk,
3406                                   u32 *first_hash)
3407 {
3408         int i, ret, credits;
3409         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3410         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3411         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3412         struct buffer_head *bh = NULL;
3413         struct ocfs2_xattr_header *xh;
3414         u64 to_blk_start = to_blk;
3415
3416         mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk);
3417
3418         /*
3419          * We need to update the new cluster and 1 more for the update of
3420          * the 1st bucket of the previous extent rec.
3421          */
3422         credits = bpc + 1;
3423         ret = ocfs2_extend_trans(handle, credits);
3424         if (ret) {
3425                 mlog_errno(ret);
3426                 goto out;
3427         }
3428
3429         ret = ocfs2_journal_access(handle, inode, first_bh,
3430                                    OCFS2_JOURNAL_ACCESS_WRITE);
3431         if (ret) {
3432                 mlog_errno(ret);
3433                 goto out;
3434         }
3435
3436         for (i = 0; i < num_buckets; i++) {
3437                 ret = ocfs2_cp_xattr_bucket(inode, handle,
3438                                             src_blk, to_blk, 1);
3439                 if (ret) {
3440                         mlog_errno(ret);
3441                         goto out;
3442                 }
3443
3444                 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3445                 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3446         }
3447
3448         /* update the old bucket header. */
3449         xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3450         le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3451
3452         ocfs2_journal_dirty(handle, first_bh);
3453
3454         /* update the new bucket header. */
3455         ret = ocfs2_read_block(osb, to_blk_start, &bh, OCFS2_BH_CACHED, inode);
3456         if (ret < 0) {
3457                 mlog_errno(ret);
3458                 goto out;
3459         }
3460
3461         ret = ocfs2_journal_access(handle, inode, bh,
3462                                    OCFS2_JOURNAL_ACCESS_WRITE);
3463         if (ret) {
3464                 mlog_errno(ret);
3465                 goto out;
3466         }
3467
3468         xh = (struct ocfs2_xattr_header *)bh->b_data;
3469         xh->xh_num_buckets = cpu_to_le16(num_buckets);
3470
3471         ocfs2_journal_dirty(handle, bh);
3472
3473         if (first_hash)
3474                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3475 out:
3476         brelse(bh);
3477         return ret;
3478 }
3479
3480 /*
3481  * Move half of the xattrs in this cluster to the new cluster.
3482  * This function should only be called when bucket size == cluster size.
3483  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3484  */
3485 static int ocfs2_half_xattr_cluster(struct inode *inode,
3486                                     handle_t *handle,
3487                                     u64 prev_blk,
3488                                     u64 new_blk,
3489                                     u32 *first_hash)
3490 {
3491         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3492         int ret, credits = 2 * blk_per_bucket;
3493
3494         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3495
3496         ret = ocfs2_extend_trans(handle, credits);
3497         if (ret) {
3498                 mlog_errno(ret);
3499                 return ret;
3500         }
3501
3502         /* Move half of the xattr in start_blk to the next bucket. */
3503         return  ocfs2_half_xattr_bucket(inode, handle, prev_blk,
3504                                         new_blk, first_hash, 1);
3505 }
3506
3507 /*
3508  * Move some xattrs from the old cluster to the new one since they are not
3509  * contiguous in ocfs2 xattr tree.
3510  *
3511  * new_blk starts a new separate cluster, and we will move some xattrs from
3512  * prev_blk to it. v_start will be set as the first name hash value in this
3513  * new cluster so that it can be used as e_cpos during tree insertion and
3514  * don't collide with our original b-tree operations. first_bh and header_bh
3515  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3516  * to extend the insert bucket.
3517  *
3518  * The problem is how much xattr should we move to the new one and when should
3519  * we update first_bh and header_bh?
3520  * 1. If cluster size > bucket size, that means the previous cluster has more
3521  *    than 1 bucket, so just move half nums of bucket into the new cluster and
3522  *    update the first_bh and header_bh if the insert bucket has been moved
3523  *    to the new cluster.
3524  * 2. If cluster_size == bucket_size:
3525  *    a) If the previous extent rec has more than one cluster and the insert
3526  *       place isn't in the last cluster, copy the entire last cluster to the
3527  *       new one. This time, we don't need to upate the first_bh and header_bh
3528  *       since they will not be moved into the new cluster.
3529  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
3530  *       the new one. And we set the extend flag to zero if the insert place is
3531  *       moved into the new allocated cluster since no extend is needed.
3532  */
3533 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3534                                             handle_t *handle,
3535                                             struct buffer_head **first_bh,
3536                                             struct buffer_head **header_bh,
3537                                             u64 new_blk,
3538                                             u64 prev_blk,
3539                                             u32 prev_clusters,
3540                                             u32 *v_start,
3541                                             int *extend)
3542 {
3543         int ret = 0;
3544         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3545
3546         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3547              prev_blk, prev_clusters, new_blk);
3548
3549         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3550                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3551                                                           handle,
3552                                                           first_bh,
3553                                                           header_bh,
3554                                                           new_blk,
3555                                                           prev_blk,
3556                                                           prev_clusters,
3557                                                           v_start);
3558         else {
3559                 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3560
3561                 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3562                         ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3563                                                      last_blk, new_blk,
3564                                                      v_start);
3565                 else {
3566                         ret = ocfs2_half_xattr_cluster(inode, handle,
3567                                                        last_blk, new_blk,
3568                                                        v_start);
3569
3570                         if ((*header_bh)->b_blocknr == last_blk && extend)
3571                                 *extend = 0;
3572                 }
3573         }
3574
3575         return ret;
3576 }
3577
3578 /*
3579  * Add a new cluster for xattr storage.
3580  *
3581  * If the new cluster is contiguous with the previous one, it will be
3582  * appended to the same extent record, and num_clusters will be updated.
3583  * If not, we will insert a new extent for it and move some xattrs in
3584  * the last cluster into the new allocated one.
3585  * We also need to limit the maximum size of a btree leaf, otherwise we'll
3586  * lose the benefits of hashing because we'll have to search large leaves.
3587  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3588  * if it's bigger).
3589  *
3590  * first_bh is the first block of the previous extent rec and header_bh
3591  * indicates the bucket we will insert the new xattrs. They will be updated
3592  * when the header_bh is moved into the new cluster.
3593  */
3594 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3595                                        struct buffer_head *root_bh,
3596                                        struct buffer_head **first_bh,
3597                                        struct buffer_head **header_bh,
3598                                        u32 *num_clusters,
3599                                        u32 prev_cpos,
3600                                        u64 prev_blkno,
3601                                        int *extend)
3602 {
3603         int ret, credits;
3604         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3605         u32 prev_clusters = *num_clusters;
3606         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3607         u64 block;
3608         handle_t *handle = NULL;
3609         struct ocfs2_alloc_context *data_ac = NULL;
3610         struct ocfs2_alloc_context *meta_ac = NULL;
3611         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3612         struct ocfs2_xattr_block *xb =
3613                         (struct ocfs2_xattr_block *)root_bh->b_data;
3614         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3615         struct ocfs2_extent_list *root_el = &xb_root->xt_list;
3616         enum ocfs2_extent_tree_type type = OCFS2_XATTR_TREE_EXTENT;
3617
3618         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3619              "previous xattr blkno = %llu\n",
3620              (unsigned long long)OCFS2_I(inode)->ip_blkno,
3621              prev_cpos, prev_blkno);
3622
3623         ret = ocfs2_lock_allocators(inode, root_bh, root_el,
3624                                     clusters_to_add, 0, &data_ac,
3625                                     &meta_ac, type, NULL);
3626         if (ret) {
3627                 mlog_errno(ret);
3628                 goto leave;
3629         }
3630
3631         credits = ocfs2_calc_extend_credits(osb->sb, root_el, clusters_to_add);
3632         handle = ocfs2_start_trans(osb, credits);
3633         if (IS_ERR(handle)) {
3634                 ret = PTR_ERR(handle);
3635                 handle = NULL;
3636                 mlog_errno(ret);
3637                 goto leave;
3638         }
3639
3640         ret = ocfs2_journal_access(handle, inode, root_bh,
3641                                    OCFS2_JOURNAL_ACCESS_WRITE);
3642         if (ret < 0) {
3643                 mlog_errno(ret);
3644                 goto leave;
3645         }
3646
3647         ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3648                                      clusters_to_add, &bit_off, &num_bits);
3649         if (ret < 0) {
3650                 if (ret != -ENOSPC)
3651                         mlog_errno(ret);
3652                 goto leave;
3653         }
3654
3655         BUG_ON(num_bits > clusters_to_add);
3656
3657         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3658         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3659              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3660
3661         if (prev_blkno + prev_clusters * bpc == block &&
3662             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3663              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3664                 /*
3665                  * If this cluster is contiguous with the old one and
3666                  * adding this new cluster, we don't surpass the limit of
3667                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3668                  * initialized and used like other buckets in the previous
3669                  * cluster.
3670                  * So add it as a contiguous one. The caller will handle
3671                  * its init process.
3672                  */
3673                 v_start = prev_cpos + prev_clusters;
3674                 *num_clusters = prev_clusters + num_bits;
3675                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3676                      num_bits);
3677         } else {
3678                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3679                                                        handle,
3680                                                        first_bh,
3681                                                        header_bh,
3682                                                        block,
3683                                                        prev_blkno,
3684                                                        prev_clusters,
3685                                                        &v_start,
3686                                                        extend);
3687                 if (ret) {
3688                         mlog_errno(ret);
3689                         goto leave;
3690                 }
3691         }
3692
3693         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3694              num_bits, block, v_start);
3695         ret = ocfs2_xattr_tree_insert_extent(osb, handle, inode, root_bh,
3696                                              v_start, block, num_bits,
3697                                              0, meta_ac);
3698         if (ret < 0) {
3699                 mlog_errno(ret);
3700                 goto leave;
3701         }
3702
3703         ret = ocfs2_journal_dirty(handle, root_bh);
3704         if (ret < 0) {
3705                 mlog_errno(ret);
3706                 goto leave;
3707         }
3708
3709 leave:
3710         if (handle)
3711                 ocfs2_commit_trans(osb, handle);
3712         if (data_ac)
3713                 ocfs2_free_alloc_context(data_ac);
3714         if (meta_ac)
3715                 ocfs2_free_alloc_context(meta_ac);
3716
3717         return ret;
3718 }
3719
3720 /*
3721  * Extend a new xattr bucket and move xattrs to the end one by one until
3722  * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3723  */
3724 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3725                                      struct buffer_head *first_bh,
3726                                      struct buffer_head *start_bh,
3727                                      u32 num_clusters)
3728 {
3729         int ret, credits;
3730         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3731         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3732         u64 start_blk = start_bh->b_blocknr, end_blk;
3733         u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3734         handle_t *handle;
3735         struct ocfs2_xattr_header *first_xh =
3736                                 (struct ocfs2_xattr_header *)first_bh->b_data;
3737         u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3738
3739         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3740              "from %llu, len = %u\n", start_blk,
3741              (unsigned long long)first_bh->b_blocknr, num_clusters);
3742
3743         BUG_ON(bucket >= num_buckets);
3744
3745         end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3746
3747         /*
3748          * We will touch all the buckets after the start_bh(include it).
3749          * Add one more bucket and modify the first_bh.
3750          */
3751         credits = end_blk - start_blk + 2 * blk_per_bucket + 1;
3752         handle = ocfs2_start_trans(osb, credits);
3753         if (IS_ERR(handle)) {
3754                 ret = PTR_ERR(handle);
3755                 handle = NULL;
3756                 mlog_errno(ret);
3757                 goto out;
3758         }
3759
3760         ret = ocfs2_journal_access(handle, inode, first_bh,
3761                                    OCFS2_JOURNAL_ACCESS_WRITE);
3762         if (ret) {
3763                 mlog_errno(ret);
3764                 goto commit;
3765         }
3766
3767         while (end_blk != start_blk) {
3768                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3769                                             end_blk + blk_per_bucket, 0);
3770                 if (ret)
3771                         goto commit;
3772                 end_blk -= blk_per_bucket;
3773         }
3774
3775         /* Move half of the xattr in start_blk to the next bucket. */
3776         ret = ocfs2_half_xattr_bucket(inode, handle, start_blk,
3777                                       start_blk + blk_per_bucket, NULL, 0);
3778
3779         le16_add_cpu(&first_xh->xh_num_buckets, 1);
3780         ocfs2_journal_dirty(handle, first_bh);
3781
3782 commit:
3783         ocfs2_commit_trans(osb, handle);
3784 out:
3785         return ret;
3786 }
3787
3788 /*
3789  * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3790  * xb_bh is the ocfs2_xattr_block.
3791  * We will move all the buckets starting from header_bh to the next place. As
3792  * for this one, half num of its xattrs will be moved to the next one.
3793  *
3794  * We will allocate a new cluster if current cluster is full and adjust
3795  * header_bh and first_bh if the insert place is moved to the new cluster.
3796  */
3797 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3798                                       struct buffer_head *xb_bh,
3799                                       struct buffer_head *header_bh)
3800 {
3801         struct ocfs2_xattr_header *first_xh = NULL;
3802         struct buffer_head *first_bh = NULL;
3803         struct ocfs2_xattr_block *xb =
3804                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3805         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3806         struct ocfs2_extent_list *el = &xb_root->xt_list;
3807         struct ocfs2_xattr_header *xh =
3808                         (struct ocfs2_xattr_header *)header_bh->b_data;
3809         u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3810         struct super_block *sb = inode->i_sb;
3811         struct ocfs2_super *osb = OCFS2_SB(sb);
3812         int ret, num_buckets, extend = 1;
3813         u64 p_blkno;
3814         u32 e_cpos, num_clusters;
3815
3816         mlog(0, "Add new xattr bucket starting form %llu\n",
3817              (unsigned long long)header_bh->b_blocknr);
3818
3819         /*
3820          * Add refrence for header_bh here because it may be
3821          * changed in ocfs2_add_new_xattr_cluster and we need
3822          * to free it in the end.
3823          */
3824         get_bh(header_bh);
3825
3826         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3827                                   &num_clusters, el);
3828         if (ret) {
3829                 mlog_errno(ret);
3830                 goto out;
3831         }
3832
3833         ret = ocfs2_read_block(osb, p_blkno,
3834                                &first_bh, OCFS2_BH_CACHED, inode);
3835         if (ret) {
3836                 mlog_errno(ret);
3837                 goto out;
3838         }
3839
3840         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3841         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3842
3843         if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3844                 ret = ocfs2_add_new_xattr_cluster(inode,
3845                                                   xb_bh,
3846                                                   &first_bh,
3847                                                   &header_bh,
3848                                                   &num_clusters,
3849                                                   e_cpos,
3850                                                   p_blkno,
3851                                                   &extend);
3852                 if (ret) {
3853                         mlog_errno(ret);
3854                         goto out;
3855                 }
3856         }
3857
3858         if (extend)
3859                 ret = ocfs2_extend_xattr_bucket(inode,
3860                                                 first_bh,
3861                                                 header_bh,
3862                                                 num_clusters);
3863         if (ret)
3864                 mlog_errno(ret);
3865 out:
3866         brelse(first_bh);
3867         brelse(header_bh);
3868         return ret;
3869 }
3870
3871 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3872                                         struct ocfs2_xattr_bucket *bucket,
3873                                         int offs)
3874 {
3875         int block_off = offs >> inode->i_sb->s_blocksize_bits;
3876
3877         offs = offs % inode->i_sb->s_blocksize;
3878         return bucket->bhs[block_off]->b_data + offs;
3879 }
3880
3881 /*
3882  * Handle the normal xattr set, including replace, delete and new.
3883  * When the bucket is empty, "is_empty" is set and the caller can
3884  * free this bucket.
3885  *
3886  * Note: "local" indicates the real data's locality. So we can't
3887  * just its bucket locality by its length.
3888  */
3889 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3890                                          struct ocfs2_xattr_info *xi,
3891                                          struct ocfs2_xattr_search *xs,
3892                                          u32 name_hash,
3893                                          int local,
3894                                          int *is_empty)
3895 {
3896         struct ocfs2_xattr_entry *last, *xe;
3897         int name_len = strlen(xi->name);
3898         struct ocfs2_xattr_header *xh = xs->header;
3899         u16 count = le16_to_cpu(xh->xh_count), start;
3900         size_t blocksize = inode->i_sb->s_blocksize;
3901         char *val;
3902         size_t offs, size, new_size;
3903
3904         last = &xh->xh_entries[count];
3905         if (!xs->not_found) {
3906                 xe = xs->here;
3907                 offs = le16_to_cpu(xe->xe_name_offset);
3908                 if (ocfs2_xattr_is_local(xe))
3909                         size = OCFS2_XATTR_SIZE(name_len) +
3910                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3911                 else
3912                         size = OCFS2_XATTR_SIZE(name_len) +
3913                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3914
3915                 /*
3916                  * If the new value will be stored outside, xi->value has been
3917                  * initalized as an empty ocfs2_xattr_value_root, and the same
3918                  * goes with xi->value_len, so we can set new_size safely here.
3919                  * See ocfs2_xattr_set_in_bucket.
3920                  */
3921                 new_size = OCFS2_XATTR_SIZE(name_len) +
3922                            OCFS2_XATTR_SIZE(xi->value_len);
3923
3924                 le16_add_cpu(&xh->xh_name_value_len, -size);
3925                 if (xi->value) {
3926                         if (new_size > size)
3927                                 goto set_new_name_value;
3928
3929                         /* Now replace the old value with new one. */
3930                         if (local)
3931                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
3932                         else
3933                                 xe->xe_value_size = 0;
3934
3935                         val = ocfs2_xattr_bucket_get_val(inode,
3936                                                          &xs->bucket, offs);
3937                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
3938                                size - OCFS2_XATTR_SIZE(name_len));
3939                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
3940                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
3941                                        xi->value, xi->value_len);
3942
3943                         le16_add_cpu(&xh->xh_name_value_len, new_size);
3944                         ocfs2_xattr_set_local(xe, local);
3945                         return;
3946                 } else {
3947                         /* Remove the old entry. */
3948                         last -= 1;
3949                         memmove(xe, xe + 1,
3950                                 (void *)last - (void *)xe);
3951                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
3952                         le16_add_cpu(&xh->xh_count, -1);
3953                         if (xh->xh_count == 0 && is_empty)
3954                                 *is_empty = 1;
3955                         return;
3956                 }
3957         } else {
3958                 /* find a new entry for insert. */
3959                 int low = 0, high = count - 1, tmp;
3960                 struct ocfs2_xattr_entry *tmp_xe;
3961
3962                 while (low <= high) {
3963                         tmp = (low + high) / 2;
3964                         tmp_xe = &xh->xh_entries[tmp];
3965
3966                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
3967                                 low = tmp + 1;
3968                         else if (name_hash <
3969                                  le32_to_cpu(tmp_xe->xe_name_hash))
3970                                 high = tmp - 1;
3971                         else
3972                                 break;
3973                 }
3974
3975                 xe = &xh->xh_entries[low];
3976                 if (low != count)
3977                         memmove(xe + 1, xe, (void *)last - (void *)xe);
3978
3979                 le16_add_cpu(&xh->xh_count, 1);
3980                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
3981                 xe->xe_name_hash = cpu_to_le32(name_hash);
3982                 xe->xe_name_len = name_len;
3983                 ocfs2_xattr_set_type(xe, xi->name_index);
3984         }
3985
3986 set_new_name_value:
3987         /* Insert the new name+value. */
3988         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
3989
3990         /*
3991          * We must make sure that the name/value pair
3992          * exists in the same block.
3993          */
3994         offs = le16_to_cpu(xh->xh_free_start);
3995         start = offs - size;
3996
3997         if (start >> inode->i_sb->s_blocksize_bits !=
3998             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
3999                 offs = offs - offs % blocksize;
4000                 xh->xh_free_start = cpu_to_le16(offs);
4001         }
4002
4003         val = ocfs2_xattr_bucket_get_val(inode,
4004                                          &xs->bucket, offs - size);
4005         xe->xe_name_offset = cpu_to_le16(offs - size);
4006
4007         memset(val, 0, size);
4008         memcpy(val, xi->name, name_len);
4009         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4010
4011         xe->xe_value_size = cpu_to_le64(xi->value_len);
4012         ocfs2_xattr_set_local(xe, local);
4013         xs->here = xe;
4014         le16_add_cpu(&xh->xh_free_start, -size);
4015         le16_add_cpu(&xh->xh_name_value_len, size);
4016
4017         return;
4018 }
4019
4020 static int ocfs2_xattr_bucket_handle_journal(struct inode *inode,
4021                                              handle_t *handle,
4022                                              struct ocfs2_xattr_search *xs,
4023                                              struct buffer_head **bhs,
4024                                              u16 bh_num)
4025 {
4026         int ret = 0, off, block_off;
4027         struct ocfs2_xattr_entry *xe = xs->here;
4028
4029         /*
4030          * First calculate all the blocks we should journal_access
4031          * and journal_dirty. The first block should always be touched.
4032          */
4033         ret = ocfs2_journal_dirty(handle, bhs[0]);
4034         if (ret)
4035                 mlog_errno(ret);
4036
4037         /* calc the data. */
4038         off = le16_to_cpu(xe->xe_name_offset);
4039         block_off = off >> inode->i_sb->s_blocksize_bits;
4040         ret = ocfs2_journal_dirty(handle, bhs[block_off]);
4041         if (ret)
4042                 mlog_errno(ret);
4043
4044         return ret;
4045 }
4046
4047 /*
4048  * Set the xattr entry in the specified bucket.
4049  * The bucket is indicated by xs->bucket and it should have the enough
4050  * space for the xattr insertion.
4051  */
4052 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4053                                            struct ocfs2_xattr_info *xi,
4054                                            struct ocfs2_xattr_search *xs,
4055                                            u32 name_hash,
4056                                            int local,
4057                                            int *bucket_empty)
4058 {
4059         int i, ret;
4060         handle_t *handle = NULL;
4061         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4062         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4063
4064         mlog(0, "Set xattr entry len = %d index = %d in bucket %llu\n",
4065              xi->value_len, xi->name_index,
4066              (unsigned long long)xs->bucket.bhs[0]->b_blocknr);
4067
4068         if (!xs->bucket.bhs[1]) {
4069                 ret = ocfs2_read_blocks(osb,
4070                                         xs->bucket.bhs[0]->b_blocknr + 1,
4071                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
4072                                         OCFS2_BH_CACHED, inode);
4073                 if (ret) {
4074                         mlog_errno(ret);
4075                         goto out;
4076                 }
4077         }
4078
4079         handle = ocfs2_start_trans(osb, blk_per_bucket);
4080         if (IS_ERR(handle)) {
4081                 ret = PTR_ERR(handle);
4082                 handle = NULL;
4083                 mlog_errno(ret);
4084                 goto out;
4085         }
4086
4087         for (i = 0; i < blk_per_bucket; i++) {
4088                 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[i],
4089                                            OCFS2_JOURNAL_ACCESS_WRITE);
4090                 if (ret < 0) {
4091                         mlog_errno(ret);
4092                         goto out;
4093                 }
4094         }
4095
4096         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash,
4097                                      local, bucket_empty);
4098
4099         /*Only dirty the blocks we have touched in set xattr. */
4100         ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs,
4101                                                 xs->bucket.bhs, blk_per_bucket);
4102         if (ret)
4103                 mlog_errno(ret);
4104 out:
4105         ocfs2_commit_trans(osb, handle);
4106
4107         return ret;
4108 }
4109
4110 static int ocfs2_xattr_value_update_size(struct inode *inode,
4111                                          struct buffer_head *xe_bh,
4112                                          struct ocfs2_xattr_entry *xe,
4113                                          u64 new_size)
4114 {
4115         int ret;
4116         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4117         handle_t *handle = NULL;
4118
4119         handle = ocfs2_start_trans(osb, 1);
4120         if (handle == NULL) {
4121                 ret = -ENOMEM;
4122                 mlog_errno(ret);
4123                 goto out;
4124         }
4125
4126         ret = ocfs2_journal_access(handle, inode, xe_bh,
4127                                    OCFS2_JOURNAL_ACCESS_WRITE);
4128         if (ret < 0) {
4129                 mlog_errno(ret);
4130                 goto out_commit;
4131         }
4132
4133         xe->xe_value_size = cpu_to_le64(new_size);
4134
4135         ret = ocfs2_journal_dirty(handle, xe_bh);
4136         if (ret < 0)
4137                 mlog_errno(ret);
4138
4139 out_commit:
4140         ocfs2_commit_trans(osb, handle);
4141 out:
4142         return ret;
4143 }
4144
4145 /*
4146  * Truncate the specified xe_off entry in xattr bucket.
4147  * bucket is indicated by header_bh and len is the new length.
4148  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4149  *
4150  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4151  */
4152 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4153                                              struct buffer_head *header_bh,
4154                                              int xe_off,
4155                                              int len)
4156 {
4157         int ret, offset;
4158         u64 value_blk;
4159         struct buffer_head *value_bh = NULL;
4160         struct ocfs2_xattr_value_root *xv;
4161         struct ocfs2_xattr_entry *xe;
4162         struct ocfs2_xattr_header *xh =
4163                         (struct ocfs2_xattr_header *)header_bh->b_data;
4164         size_t blocksize = inode->i_sb->s_blocksize;
4165
4166         xe = &xh->xh_entries[xe_off];
4167
4168         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4169
4170         offset = le16_to_cpu(xe->xe_name_offset) +
4171                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4172
4173         value_blk = offset / blocksize;
4174
4175         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4176         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4177         value_blk += header_bh->b_blocknr;
4178
4179         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), value_blk,
4180                                &value_bh, OCFS2_BH_CACHED, inode);
4181         if (ret) {
4182                 mlog_errno(ret);
4183                 goto out;
4184         }
4185
4186         xv = (struct ocfs2_xattr_value_root *)
4187                 (value_bh->b_data + offset % blocksize);
4188
4189         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4190              xe_off, (unsigned long long)header_bh->b_blocknr, len);
4191         ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4192         if (ret) {
4193                 mlog_errno(ret);
4194                 goto out;
4195         }
4196
4197         ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4198         if (ret) {
4199                 mlog_errno(ret);
4200                 goto out;
4201         }
4202
4203 out:
4204         brelse(value_bh);
4205         return ret;
4206 }
4207
4208 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4209                                                 struct ocfs2_xattr_search *xs,
4210                                                 int len)
4211 {
4212         int ret, offset;
4213         struct ocfs2_xattr_entry *xe = xs->here;
4214         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4215
4216         BUG_ON(!xs->bucket.bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4217
4218         offset = xe - xh->xh_entries;
4219         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bhs[0],
4220                                                 offset, len);
4221         if (ret)
4222                 mlog_errno(ret);
4223
4224         return ret;
4225 }
4226
4227 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4228                                                 struct ocfs2_xattr_search *xs,
4229                                                 char *val,
4230                                                 int value_len)
4231 {
4232         int offset;
4233         struct ocfs2_xattr_value_root *xv;
4234         struct ocfs2_xattr_entry *xe = xs->here;
4235
4236         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4237
4238         offset = le16_to_cpu(xe->xe_name_offset) +
4239                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4240
4241         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4242
4243         return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4244 }
4245
4246 /*
4247  * Remove the xattr bucket pointed by bucket_bh.
4248  * All the buckets after it in the same xattr extent rec will be
4249  * move forward one by one.
4250  */
4251 static int ocfs2_rm_xattr_bucket(struct inode *inode,
4252                                  struct buffer_head *first_bh,
4253                                  struct ocfs2_xattr_bucket *bucket)
4254 {
4255         int ret = 0, credits;
4256         struct ocfs2_xattr_header *xh =
4257                                 (struct ocfs2_xattr_header *)first_bh->b_data;
4258         u16 bucket_num = le16_to_cpu(xh->xh_num_buckets);
4259         u64 end, start = bucket->bhs[0]->b_blocknr;
4260         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4261         handle_t *handle;
4262         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4263
4264         end = first_bh->b_blocknr + (bucket_num - 1) * blk_per_bucket;
4265
4266         mlog(0, "rm xattr bucket %llu\n", start);
4267         /*
4268          * We need to update the first xattr_header and all the buckets starting
4269          * from start in this xattr rec.
4270          *
4271          * XXX: Should we empty the old last bucket here?
4272          */
4273         credits = 1 + end - start;
4274         handle = ocfs2_start_trans(osb, credits);
4275         if (IS_ERR(handle)) {
4276                 ret = PTR_ERR(handle);
4277                 mlog_errno(ret);
4278                 return ret;
4279         }
4280
4281         ret = ocfs2_journal_access(handle, inode, first_bh,
4282                                    OCFS2_JOURNAL_ACCESS_WRITE);
4283         if (ret) {
4284                 mlog_errno(ret);
4285                 goto out_commit;
4286         }
4287
4288
4289         while (start < end) {
4290                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4291                                             start + blk_per_bucket,
4292                                             start, 0);
4293                 if (ret) {
4294                         mlog_errno(ret);
4295                         goto out_commit;
4296                 }
4297                 start += blk_per_bucket;
4298         }
4299
4300         /* update the first_bh. */
4301         xh->xh_num_buckets = cpu_to_le16(bucket_num - 1);
4302         ocfs2_journal_dirty(handle, first_bh);
4303
4304 out_commit:
4305         ocfs2_commit_trans(osb, handle);
4306         return ret;
4307 }
4308
4309 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4310                                   struct buffer_head *root_bh,
4311                                   u64 blkno,
4312                                   u32 cpos,
4313                                   u32 len)
4314 {
4315         int ret;
4316         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4317         struct inode *tl_inode = osb->osb_tl_inode;
4318         handle_t *handle;
4319         struct ocfs2_xattr_block *xb =
4320                         (struct ocfs2_xattr_block *)root_bh->b_data;
4321         struct ocfs2_extent_list *root_el = &xb->xb_attrs.xb_root.xt_list;
4322         struct ocfs2_alloc_context *meta_ac = NULL;
4323         struct ocfs2_cached_dealloc_ctxt dealloc;
4324
4325         ocfs2_init_dealloc_ctxt(&dealloc);
4326
4327         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4328              cpos, len, (unsigned long long)blkno);
4329
4330         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4331
4332         ret = ocfs2_lock_allocators(inode, root_bh, root_el,
4333                                     0, 1, NULL, &meta_ac,
4334                                     OCFS2_XATTR_TREE_EXTENT, NULL);
4335         if (ret) {
4336                 mlog_errno(ret);
4337                 return ret;
4338         }
4339
4340         mutex_lock(&tl_inode->i_mutex);
4341
4342         if (ocfs2_truncate_log_needs_flush(osb)) {
4343                 ret = __ocfs2_flush_truncate_log(osb);
4344                 if (ret < 0) {
4345                         mlog_errno(ret);
4346                         goto out;
4347                 }
4348         }
4349
4350         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4351         if (handle == NULL) {
4352                 ret = -ENOMEM;
4353                 mlog_errno(ret);
4354                 goto out;
4355         }
4356
4357         ret = ocfs2_journal_access(handle, inode, root_bh,
4358                                    OCFS2_JOURNAL_ACCESS_WRITE);
4359         if (ret) {
4360                 mlog_errno(ret);
4361                 goto out_commit;
4362         }
4363
4364         ret = ocfs2_remove_extent(inode, root_bh, cpos, len, handle, meta_ac,
4365                                   &dealloc, OCFS2_XATTR_TREE_EXTENT, NULL);
4366         if (ret) {
4367                 mlog_errno(ret);
4368                 goto out_commit;
4369         }
4370
4371         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4372
4373         ret = ocfs2_journal_dirty(handle, root_bh);
4374         if (ret) {
4375                 mlog_errno(ret);
4376                 goto out_commit;
4377         }
4378
4379         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4380         if (ret)
4381                 mlog_errno(ret);
4382
4383 out_commit:
4384         ocfs2_commit_trans(osb, handle);
4385 out:
4386         ocfs2_schedule_truncate_log_flush(osb, 1);
4387
4388         mutex_unlock(&tl_inode->i_mutex);
4389
4390         if (meta_ac)
4391                 ocfs2_free_alloc_context(meta_ac);
4392
4393         ocfs2_run_deallocs(osb, &dealloc);
4394
4395         return ret;
4396 }
4397
4398 /*
4399  * Free the xattr bucket indicated by xs->bucket and if all the buckets
4400  * in the clusters is free, free the clusters also.
4401  */
4402 static int ocfs2_xattr_bucket_shrink(struct inode *inode,
4403                                      struct ocfs2_xattr_info *xi,
4404                                      struct ocfs2_xattr_search *xs,
4405                                      u32 name_hash)
4406 {
4407         int ret;
4408         u32 e_cpos, num_clusters;
4409         u64 p_blkno;
4410         struct buffer_head *first_bh = NULL;
4411         struct ocfs2_xattr_header *first_xh;
4412         struct ocfs2_xattr_block *xb =
4413                         (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
4414
4415         BUG_ON(xs->header->xh_count != 0);
4416
4417         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4418                                   &e_cpos, &num_clusters,
4419                                   &xb->xb_attrs.xb_root.xt_list);
4420         if (ret) {
4421                 mlog_errno(ret);
4422                 return ret;
4423         }
4424
4425         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno,
4426                                &first_bh, OCFS2_BH_CACHED, inode);
4427         if (ret) {
4428                 mlog_errno(ret);
4429                 return ret;
4430         }
4431
4432         ret = ocfs2_rm_xattr_bucket(inode, first_bh, &xs->bucket);
4433         if (ret) {
4434                 mlog_errno(ret);
4435                 goto out;
4436         }
4437
4438         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
4439         if (first_xh->xh_num_buckets == 0)
4440                 ret = ocfs2_rm_xattr_cluster(inode, xs->xattr_bh,
4441                                              p_blkno, e_cpos,
4442                                              num_clusters);
4443
4444 out:
4445         brelse(first_bh);
4446         return ret;
4447 }
4448
4449 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4450                                          struct ocfs2_xattr_search *xs)
4451 {
4452         handle_t *handle = NULL;
4453         struct ocfs2_xattr_header *xh = xs->bucket.xh;
4454         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4455                                                 le16_to_cpu(xh->xh_count) - 1];
4456         int ret = 0;
4457
4458         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1);
4459         if (IS_ERR(handle)) {
4460                 ret = PTR_ERR(handle);
4461                 mlog_errno(ret);
4462                 return;
4463         }
4464
4465         ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[0],
4466                                    OCFS2_JOURNAL_ACCESS_WRITE);
4467         if (ret) {
4468                 mlog_errno(ret);
4469                 goto out_commit;
4470         }
4471
4472         /* Remove the old entry. */
4473         memmove(xs->here, xs->here + 1,
4474                 (void *)last - (void *)xs->here);
4475         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4476         le16_add_cpu(&xh->xh_count, -1);
4477
4478         ret = ocfs2_journal_dirty(handle, xs->bucket.bhs[0]);
4479         if (ret < 0)
4480                 mlog_errno(ret);
4481 out_commit:
4482         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4483 }
4484
4485 /*
4486  * Set the xattr name/value in the bucket specified in xs.
4487  *
4488  * As the new value in xi may be stored in the bucket or in an outside cluster,
4489  * we divide the whole process into 3 steps:
4490  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4491  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4492  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4493  * 4. If the clusters for the new outside value can't be allocated, we need
4494  *    to free the xattr we allocated in set.
4495  */
4496 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4497                                      struct ocfs2_xattr_info *xi,
4498                                      struct ocfs2_xattr_search *xs)
4499 {
4500         int ret, local = 1, bucket_empty = 0;
4501         size_t value_len;
4502         char *val = (char *)xi->value;
4503         struct ocfs2_xattr_entry *xe = xs->here;
4504         u32 name_hash = ocfs2_xattr_hash_by_name(inode,
4505                                                  xi->name_index, xi->name);
4506
4507         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4508                 /*
4509                  * We need to truncate the xattr storage first.
4510                  *
4511                  * If both the old and new value are stored to
4512                  * outside block, we only need to truncate
4513                  * the storage and then set the value outside.
4514                  *
4515                  * If the new value should be stored within block,
4516                  * we should free all the outside block first and
4517                  * the modification to the xattr block will be done
4518                  * by following steps.
4519                  */
4520                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4521                         value_len = xi->value_len;
4522                 else
4523                         value_len = 0;
4524
4525                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4526                                                            value_len);
4527                 if (ret)
4528                         goto out;
4529
4530                 if (value_len)
4531                         goto set_value_outside;
4532         }
4533
4534         value_len = xi->value_len;
4535         /* So we have to handle the inside block change now. */
4536         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4537                 /*
4538                  * If the new value will be stored outside of block,
4539                  * initalize a new empty value root and insert it first.
4540                  */
4541                 local = 0;
4542                 xi->value = &def_xv;
4543                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4544         }
4545
4546         ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash,
4547                                               local, &bucket_empty);
4548         if (ret) {
4549                 mlog_errno(ret);
4550                 goto out;
4551         }
4552
4553         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4554                 /* allocate the space now for the outside block storage. */
4555                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4556                                                            value_len);
4557                 if (ret) {
4558                         mlog_errno(ret);
4559
4560                         if (xs->not_found) {
4561                                 /*
4562                                  * We can't allocate enough clusters for outside
4563                                  * storage and we have allocated xattr already,
4564                                  * so need to remove it.
4565                                  */
4566                                 ocfs2_xattr_bucket_remove_xs(inode, xs);
4567                         }
4568                         goto out;
4569                 }
4570         } else {
4571                 if (bucket_empty)
4572                         ret = ocfs2_xattr_bucket_shrink(inode, xi,
4573                                                         xs, name_hash);
4574                 goto out;
4575         }
4576
4577 set_value_outside:
4578         ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4579 out:
4580         return ret;
4581 }
4582
4583 /* check whether the xattr bucket is filled up with the same hash value. */
4584 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4585                                               struct ocfs2_xattr_bucket *bucket)
4586 {
4587         struct ocfs2_xattr_header *xh = bucket->xh;
4588
4589         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4590             xh->xh_entries[0].xe_name_hash) {
4591                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4592                      "hash = %u\n",
4593                      (unsigned long long)bucket->bhs[0]->b_blocknr,
4594                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4595                 return -ENOSPC;
4596         }
4597
4598         return 0;
4599 }
4600
4601 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4602                                              struct ocfs2_xattr_info *xi,
4603                                              struct ocfs2_xattr_search *xs)
4604 {
4605         struct ocfs2_xattr_header *xh;
4606         struct ocfs2_xattr_entry *xe;
4607         u16 count, header_size, xh_free_start;
4608         int i, free, max_free, need, old;
4609         size_t value_size = 0, name_len = strlen(xi->name);
4610         size_t blocksize = inode->i_sb->s_blocksize;
4611         int ret, allocation = 0;
4612         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4613
4614         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4615
4616 try_again:
4617         xh = xs->header;
4618         count = le16_to_cpu(xh->xh_count);
4619         xh_free_start = le16_to_cpu(xh->xh_free_start);
4620         header_size = sizeof(struct ocfs2_xattr_header) +
4621                         count * sizeof(struct ocfs2_xattr_entry);
4622         max_free = OCFS2_XATTR_BUCKET_SIZE -
4623                 le16_to_cpu(xh->xh_name_value_len) - header_size;
4624
4625         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4626                         "of %u which exceed block size\n",
4627                         (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4628                         header_size);
4629
4630         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4631                 value_size = OCFS2_XATTR_ROOT_SIZE;
4632         else if (xi->value)
4633                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4634
4635         if (xs->not_found)
4636                 need = sizeof(struct ocfs2_xattr_entry) +
4637                         OCFS2_XATTR_SIZE(name_len) + value_size;
4638         else {
4639                 need = value_size + OCFS2_XATTR_SIZE(name_len);
4640
4641                 /*
4642                  * We only replace the old value if the new length is smaller
4643                  * than the old one. Otherwise we will allocate new space in the
4644                  * bucket to store it.
4645                  */
4646                 xe = xs->here;
4647                 if (ocfs2_xattr_is_local(xe))
4648                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4649                 else
4650                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4651
4652                 if (old >= value_size)
4653                         need = 0;
4654         }
4655
4656         free = xh_free_start - header_size;
4657         /*
4658          * We need to make sure the new name/value pair
4659          * can exist in the same block.
4660          */
4661         if (xh_free_start % blocksize < need)
4662                 free -= xh_free_start % blocksize;
4663
4664         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4665              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4666              " %u\n", xs->not_found,
4667              (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4668              free, need, max_free, le16_to_cpu(xh->xh_free_start),
4669              le16_to_cpu(xh->xh_name_value_len));
4670
4671         if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4672                 if (need <= max_free &&
4673                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4674                         /*
4675                          * We can create the space by defragment. Since only the
4676                          * name/value will be moved, the xe shouldn't be changed
4677                          * in xs.
4678                          */
4679                         ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket);
4680                         if (ret) {
4681                                 mlog_errno(ret);
4682                                 goto out;
4683                         }
4684
4685                         xh_free_start = le16_to_cpu(xh->xh_free_start);
4686                         free = xh_free_start - header_size;
4687                         if (xh_free_start % blocksize < need)
4688                                 free -= xh_free_start % blocksize;
4689
4690                         if (free >= need)
4691                                 goto xattr_set;
4692
4693                         mlog(0, "Can't get enough space for xattr insert by "
4694                              "defragment. Need %u bytes, but we have %d, so "
4695                              "allocate new bucket for it.\n", need, free);
4696                 }
4697
4698                 /*
4699                  * We have to add new buckets or clusters and one
4700                  * allocation should leave us enough space for insert.
4701                  */
4702                 BUG_ON(allocation);
4703
4704                 /*
4705                  * We do not allow for overlapping ranges between buckets. And
4706                  * the maximum number of collisions we will allow for then is
4707                  * one bucket's worth, so check it here whether we need to
4708                  * add a new bucket for the insert.
4709                  */
4710                 ret = ocfs2_check_xattr_bucket_collision(inode, &xs->bucket);
4711                 if (ret) {
4712                         mlog_errno(ret);
4713                         goto out;
4714                 }
4715
4716                 ret = ocfs2_add_new_xattr_bucket(inode,
4717                                                  xs->xattr_bh,
4718                                                  xs->bucket.bhs[0]);
4719                 if (ret) {
4720                         mlog_errno(ret);
4721                         goto out;
4722                 }
4723
4724                 for (i = 0; i < blk_per_bucket; i++)
4725                         brelse(xs->bucket.bhs[i]);
4726
4727                 memset(&xs->bucket, 0, sizeof(xs->bucket));
4728
4729                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4730                                                    xi->name_index,
4731                                                    xi->name, xs);
4732                 if (ret && ret != -ENODATA)
4733                         goto out;
4734                 xs->not_found = ret;
4735                 allocation = 1;
4736                 goto try_again;
4737         }
4738
4739 xattr_set:
4740         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4741 out:
4742         mlog_exit(ret);
4743         return ret;
4744 }
4745
4746 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4747                                         struct ocfs2_xattr_bucket *bucket,
4748                                         void *para)
4749 {
4750         int ret = 0;
4751         struct ocfs2_xattr_header *xh = bucket->xh;
4752         u16 i;
4753         struct ocfs2_xattr_entry *xe;
4754
4755         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4756                 xe = &xh->xh_entries[i];
4757                 if (ocfs2_xattr_is_local(xe))
4758                         continue;
4759
4760                 ret = ocfs2_xattr_bucket_value_truncate(inode,
4761                                                         bucket->bhs[0],
4762                                                         i, 0);
4763                 if (ret) {
4764                         mlog_errno(ret);
4765                         break;
4766                 }
4767         }
4768
4769         return ret;
4770 }
4771
4772 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4773                                           struct buffer_head *xb_bh)
4774 {
4775         struct ocfs2_xattr_block *xb =
4776                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4777         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4778         int ret = 0;
4779         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4780         u64 p_blkno;
4781
4782         if (le16_to_cpu(el->l_next_free_rec) == 0)
4783                 return 0;
4784
4785         while (name_hash > 0) {
4786                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4787                                           &e_cpos, &num_clusters, el);
4788                 if (ret) {
4789                         mlog_errno(ret);
4790                         goto out;
4791                 }
4792
4793                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4794                                                   ocfs2_delete_xattr_in_bucket,
4795                                                   NULL);
4796                 if (ret) {
4797                         mlog_errno(ret);
4798                         goto out;
4799                 }
4800
4801                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4802                                              p_blkno, e_cpos, num_clusters);
4803                 if (ret) {
4804                         mlog_errno(ret);
4805                         break;
4806                 }
4807
4808                 if (e_cpos == 0)
4809                         break;
4810
4811                 name_hash = e_cpos - 1;
4812         }
4813
4814 out:
4815         return ret;
4816 }