fs/ocfs2/xattr.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * xattr.c
   5  *
   6  * Copyright (C) 2008 Oracle.  All rights reserved.
   7  *
   8  * CREDITS:
   9  * Lots of code in this file is taken from ext3.
  10  *
  11  * This program is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU General Public
  13  * License as published by the Free Software Foundation; either
  14  * version 2 of the License, or (at your option) any later version.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU General Public
  22  * License along with this program; if not, write to the
  23  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  24  * Boston, MA 021110-1307, USA.
  25  */
  26
  27 #include <linux/capability.h>
  28 #include <linux/fs.h>
  29 #include <linux/types.h>
  30 #include <linux/slab.h>
  31 #include <linux/highmem.h>
  32 #include <linux/pagemap.h>
  33 #include <linux/uio.h>
  34 #include <linux/sched.h>
  35 #include <linux/splice.h>
  36 #include <linux/mount.h>
  37 #include <linux/writeback.h>
  38 #include <linux/falloc.h>
  39 #include <linux/sort.h>
  40
  41 #define MLOG_MASK_PREFIX ML_XATTR
  42 #include <cluster/masklog.h>
  43
  44 #include "ocfs2.h"
  45 #include "alloc.h"
  46 #include "dlmglue.h"
  47 #include "file.h"
  48 #include "symlink.h"
  49 #include "sysfile.h"
  50 #include "inode.h"
  51 #include "journal.h"
  52 #include "ocfs2_fs.h"
  53 #include "suballoc.h"
  54 #include "uptodate.h"
  55 #include "buffer_head_io.h"
  56 #include "super.h"
  57 #include "xattr.h"
  58
  59
  60 struct ocfs2_xattr_def_value_root {
  61         struct ocfs2_xattr_value_root   xv;
  62         struct ocfs2_extent_rec         er;
  63 };
  64
  65 struct ocfs2_xattr_bucket {
  66         struct buffer_head *bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  67         struct ocfs2_xattr_header *xh;
  68 };
  69
  70 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  71 #define OCFS2_XATTR_INLINE_SIZE 80
  72
  73 static struct ocfs2_xattr_def_value_root def_xv = {
  74         .xv.xr_list.l_count = cpu_to_le16(1),
  75 };
  76
  77 struct xattr_handler *ocfs2_xattr_handlers[] = {
  78         &ocfs2_xattr_user_handler,
  79         &ocfs2_xattr_trusted_handler,
  80         NULL
  81 };
  82
  83 static struct xattr_handler *ocfs2_xattr_handler_map[] = {
  84         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
  85         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
  86 };
  87
  88 struct ocfs2_xattr_info {
  89         int name_index;
  90         const char *name;
  91         const void *value;
  92         size_t value_len;
  93 };
  94
  95 struct ocfs2_xattr_search {
  96         struct buffer_head *inode_bh;
  97         /*
  98          * xattr_bh point to the block buffer head which has extended attribute
  99          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 100          */
 101         struct buffer_head *xattr_bh;
 102         struct ocfs2_xattr_header *header;
 103         struct ocfs2_xattr_bucket bucket;
 104         void *base;
 105         void *end;
 106         struct ocfs2_xattr_entry *here;
 107         int not_found;
 108 };
 109
 110 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
 111                                              struct ocfs2_xattr_header *xh,
 112                                              int index,
 113                                              int *block_off,
 114                                              int *new_offset);
 115
 116 static int ocfs2_xattr_index_block_find(struct inode *inode,
 117                                         struct buffer_head *root_bh,
 118                                         int name_index,
 119                                         const char *name,
 120                                         struct ocfs2_xattr_search *xs);
 121
 122 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 123                                         struct ocfs2_xattr_tree_root *xt,
 124                                         char *buffer,
 125                                         size_t buffer_size);
 126
 127 static int ocfs2_xattr_create_index_block(struct inode *inode,
 128                                           struct ocfs2_xattr_search *xs);
 129
 130 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 131                                              struct ocfs2_xattr_info *xi,
 132                                              struct ocfs2_xattr_search *xs);
 133
 134 static int ocfs2_delete_xattr_index_block(struct inode *inode,
 135                                           struct buffer_head *xb_bh);
 136
 137 static inline struct xattr_handler *ocfs2_xattr_handler(int name_index)
 138 {
 139         struct xattr_handler *handler = NULL;
 140
 141         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 142                 handler = ocfs2_xattr_handler_map[name_index];
 143
 144         return handler;
 145 }
 146
 147 static inline u32 ocfs2_xattr_name_hash(struct inode *inode,
 148                                         char *prefix,
 149                                         int prefix_len,
 150                                         char *name,
 151                                         int name_len)
 152 {
 153         /* Get hash value of uuid from super block */
 154         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 155         int i;
 156
 157         /* hash extended attribute prefix */
 158         for (i = 0; i < prefix_len; i++) {
 159                 hash = (hash << OCFS2_HASH_SHIFT) ^
 160                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 161                        *prefix++;
 162         }
 163         /* hash extended attribute name */
 164         for (i = 0; i < name_len; i++) {
 165                 hash = (hash << OCFS2_HASH_SHIFT) ^
 166                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 167                        *name++;
 168         }
 169
 170         return hash;
 171 }
 172
 173 /*
 174  * ocfs2_xattr_hash_entry()
 175  *
 176  * Compute the hash of an extended attribute.
 177  */
 178 static void ocfs2_xattr_hash_entry(struct inode *inode,
 179                                    struct ocfs2_xattr_header *header,
 180                                    struct ocfs2_xattr_entry *entry)
 181 {
 182         u32 hash = 0;
 183         struct xattr_handler *handler =
 184                         ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
 185         char *prefix = handler->prefix;
 186         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
 187         int prefix_len = strlen(handler->prefix);
 188
 189         hash = ocfs2_xattr_name_hash(inode, prefix, prefix_len, name,
 190                                      entry->xe_name_len);
 191         entry->xe_name_hash = cpu_to_le32(hash);
 192
 193         return;
 194 }
 195
 196 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 197                                          u32 clusters_to_add,
 198                                          struct buffer_head *xattr_bh,
 199                                          struct ocfs2_xattr_value_root *xv)
 200 {
 201         int status = 0;
 202         int restart_func = 0;
 203         int credits = 0;
 204         handle_t *handle = NULL;
 205         struct ocfs2_alloc_context *data_ac = NULL;
 206         struct ocfs2_alloc_context *meta_ac = NULL;
 207         enum ocfs2_alloc_restarted why;
 208         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 209         u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
 210         struct ocfs2_extent_tree et;
 211
 212         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
 213
 214         ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
 215
 216 restart_all:
 217
 218         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 219                                        &data_ac, &meta_ac);
 220         if (status) {
 221                 mlog_errno(status);
 222                 goto leave;
 223         }
 224
 225         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
 226                                             clusters_to_add);
 227         handle = ocfs2_start_trans(osb, credits);
 228         if (IS_ERR(handle)) {
 229                 status = PTR_ERR(handle);
 230                 handle = NULL;
 231                 mlog_errno(status);
 232                 goto leave;
 233         }
 234
 235 restarted_transaction:
 236         status = ocfs2_journal_access(handle, inode, xattr_bh,
 237                                       OCFS2_JOURNAL_ACCESS_WRITE);
 238         if (status < 0) {
 239                 mlog_errno(status);
 240                 goto leave;
 241         }
 242
 243         prev_clusters = le32_to_cpu(xv->xr_clusters);
 244         status = ocfs2_add_clusters_in_btree(osb,
 245                                              inode,
 246                                              &logical_start,
 247                                              clusters_to_add,
 248                                              0,
 249                                              &et,
 250                                              handle,
 251                                              data_ac,
 252                                              meta_ac,
 253                                              &why);
 254         if ((status < 0) && (status != -EAGAIN)) {
 255                 if (status != -ENOSPC)
 256                         mlog_errno(status);
 257                 goto leave;
 258         }
 259
 260         status = ocfs2_journal_dirty(handle, xattr_bh);
 261         if (status < 0) {
 262                 mlog_errno(status);
 263                 goto leave;
 264         }
 265
 266         clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
 267
 268         if (why != RESTART_NONE && clusters_to_add) {
 269                 if (why == RESTART_META) {
 270                         mlog(0, "restarting function.\n");
 271                         restart_func = 1;
 272                 } else {
 273                         BUG_ON(why != RESTART_TRANS);
 274
 275                         mlog(0, "restarting transaction.\n");
 276                         /* TODO: This can be more intelligent. */
 277                         credits = ocfs2_calc_extend_credits(osb->sb,
 278                                                             et.et_root_el,
 279                                                             clusters_to_add);
 280                         status = ocfs2_extend_trans(handle, credits);
 281                         if (status < 0) {
 282                                 /* handle still has to be committed at
 283                                  * this point. */
 284                                 status = -ENOMEM;
 285                                 mlog_errno(status);
 286                                 goto leave;
 287                         }
 288                         goto restarted_transaction;
 289                 }
 290         }
 291
 292 leave:
 293         if (handle) {
 294                 ocfs2_commit_trans(osb, handle);
 295                 handle = NULL;
 296         }
 297         if (data_ac) {
 298                 ocfs2_free_alloc_context(data_ac);
 299                 data_ac = NULL;
 300         }
 301         if (meta_ac) {
 302                 ocfs2_free_alloc_context(meta_ac);
 303                 meta_ac = NULL;
 304         }
 305         if ((!status) && restart_func) {
 306                 restart_func = 0;
 307                 goto restart_all;
 308         }
 309
 310         return status;
 311 }
 312
 313 static int __ocfs2_remove_xattr_range(struct inode *inode,
 314                                       struct buffer_head *root_bh,
 315                                       struct ocfs2_xattr_value_root *xv,
 316                                       u32 cpos, u32 phys_cpos, u32 len,
 317                                       struct ocfs2_cached_dealloc_ctxt *dealloc)
 318 {
 319         int ret;
 320         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 321         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 322         struct inode *tl_inode = osb->osb_tl_inode;
 323         handle_t *handle;
 324         struct ocfs2_alloc_context *meta_ac = NULL;
 325         struct ocfs2_extent_tree et;
 326
 327         ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
 328
 329         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
 330         if (ret) {
 331                 mlog_errno(ret);
 332                 return ret;
 333         }
 334
 335         mutex_lock(&tl_inode->i_mutex);
 336
 337         if (ocfs2_truncate_log_needs_flush(osb)) {
 338                 ret = __ocfs2_flush_truncate_log(osb);
 339                 if (ret < 0) {
 340                         mlog_errno(ret);
 341                         goto out;
 342                 }
 343         }
 344
 345         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
 346         if (IS_ERR(handle)) {
 347                 ret = PTR_ERR(handle);
 348                 mlog_errno(ret);
 349                 goto out;
 350         }
 351
 352         ret = ocfs2_journal_access(handle, inode, root_bh,
 353                                    OCFS2_JOURNAL_ACCESS_WRITE);
 354         if (ret) {
 355                 mlog_errno(ret);
 356                 goto out_commit;
 357         }
 358
 359         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
 360                                   dealloc);
 361         if (ret) {
 362                 mlog_errno(ret);
 363                 goto out_commit;
 364         }
 365
 366         le32_add_cpu(&xv->xr_clusters, -len);
 367
 368         ret = ocfs2_journal_dirty(handle, root_bh);
 369         if (ret) {
 370                 mlog_errno(ret);
 371                 goto out_commit;
 372         }
 373
 374         ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
 375         if (ret)
 376                 mlog_errno(ret);
 377
 378 out_commit:
 379         ocfs2_commit_trans(osb, handle);
 380 out:
 381         mutex_unlock(&tl_inode->i_mutex);
 382
 383         if (meta_ac)
 384                 ocfs2_free_alloc_context(meta_ac);
 385
 386         return ret;
 387 }
 388
 389 static int ocfs2_xattr_shrink_size(struct inode *inode,
 390                                    u32 old_clusters,
 391                                    u32 new_clusters,
 392                                    struct buffer_head *root_bh,
 393                                    struct ocfs2_xattr_value_root *xv)
 394 {
 395         int ret = 0;
 396         u32 trunc_len, cpos, phys_cpos, alloc_size;
 397         u64 block;
 398         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 399         struct ocfs2_cached_dealloc_ctxt dealloc;
 400
 401         ocfs2_init_dealloc_ctxt(&dealloc);
 402
 403         if (old_clusters <= new_clusters)
 404                 return 0;
 405
 406         cpos = new_clusters;
 407         trunc_len = old_clusters - new_clusters;
 408         while (trunc_len) {
 409                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 410                                                &alloc_size, &xv->xr_list);
 411                 if (ret) {
 412                         mlog_errno(ret);
 413                         goto out;
 414                 }
 415
 416                 if (alloc_size > trunc_len)
 417                         alloc_size = trunc_len;
 418
 419                 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
 420                                                  phys_cpos, alloc_size,
 421                                                  &dealloc);
 422                 if (ret) {
 423                         mlog_errno(ret);
 424                         goto out;
 425                 }
 426
 427                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 428                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
 429                                                        alloc_size);
 430                 cpos += alloc_size;
 431                 trunc_len -= alloc_size;
 432         }
 433
 434 out:
 435         ocfs2_schedule_truncate_log_flush(osb, 1);
 436         ocfs2_run_deallocs(osb, &dealloc);
 437
 438         return ret;
 439 }
 440
 441 static int ocfs2_xattr_value_truncate(struct inode *inode,
 442                                       struct buffer_head *root_bh,
 443                                       struct ocfs2_xattr_value_root *xv,
 444                                       int len)
 445 {
 446         int ret;
 447         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 448         u32 old_clusters = le32_to_cpu(xv->xr_clusters);
 449
 450         if (new_clusters == old_clusters)
 451                 return 0;
 452
 453         if (new_clusters > old_clusters)
 454                 ret = ocfs2_xattr_extend_allocation(inode,
 455                                                     new_clusters - old_clusters,
 456                                                     root_bh, xv);
 457         else
 458                 ret = ocfs2_xattr_shrink_size(inode,
 459                                               old_clusters, new_clusters,
 460                                               root_bh, xv);
 461
 462         return ret;
 463 }
 464
 465 static int ocfs2_xattr_list_entries(struct inode *inode,
 466                                     struct ocfs2_xattr_header *header,
 467                                     char *buffer, size_t buffer_size)
 468 {
 469         size_t rest = buffer_size;
 470         int i;
 471
 472         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 473                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 474                 struct xattr_handler *handler =
 475                         ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
 476
 477                 if (handler) {
 478                         size_t size = handler->list(inode, buffer, rest,
 479                                         ((char *)header +
 480                                         le16_to_cpu(entry->xe_name_offset)),
 481                                         entry->xe_name_len);
 482                         if (buffer) {
 483                                 if (size > rest)
 484                                         return -ERANGE;
 485                                 buffer += size;
 486                         }
 487                         rest -= size;
 488                 }
 489         }
 490
 491         return buffer_size - rest;
 492 }
 493
 494 static int ocfs2_xattr_ibody_list(struct inode *inode,
 495                                   struct ocfs2_dinode *di,
 496                                   char *buffer,
 497                                   size_t buffer_size)
 498 {
 499         struct ocfs2_xattr_header *header = NULL;
 500         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 501         int ret = 0;
 502
 503         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 504                 return ret;
 505
 506         header = (struct ocfs2_xattr_header *)
 507                  ((void *)di + inode->i_sb->s_blocksize -
 508                  le16_to_cpu(di->i_xattr_inline_size));
 509
 510         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 511
 512         return ret;
 513 }
 514
 515 static int ocfs2_xattr_block_list(struct inode *inode,
 516                                   struct ocfs2_dinode *di,
 517                                   char *buffer,
 518                                   size_t buffer_size)
 519 {
 520         struct buffer_head *blk_bh = NULL;
 521         struct ocfs2_xattr_block *xb;
 522         int ret = 0;
 523
 524         if (!di->i_xattr_loc)
 525                 return ret;
 526
 527         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
 528                                le64_to_cpu(di->i_xattr_loc),
 529                                &blk_bh, OCFS2_BH_CACHED, inode);
 530         if (ret < 0) {
 531                 mlog_errno(ret);
 532                 return ret;
 533         }
 534         /*Verify the signature of xattr block*/
 535         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
 536                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
 537                 ret = -EFAULT;
 538                 goto cleanup;
 539         }
 540
 541         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 542
 543         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 544                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 545                 ret = ocfs2_xattr_list_entries(inode, header,
 546                                                buffer, buffer_size);
 547         } else {
 548                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
 549                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
 550                                                    buffer, buffer_size);
 551         }
 552 cleanup:
 553         brelse(blk_bh);
 554
 555         return ret;
 556 }
 557
 558 ssize_t ocfs2_listxattr(struct dentry *dentry,
 559                         char *buffer,
 560                         size_t size)
 561 {
 562         int ret = 0, i_ret = 0, b_ret = 0;
 563         struct buffer_head *di_bh = NULL;
 564         struct ocfs2_dinode *di = NULL;
 565         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
 566
 567         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
 568                 return -EOPNOTSUPP;
 569
 570         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 571                 return ret;
 572
 573         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
 574         if (ret < 0) {
 575                 mlog_errno(ret);
 576                 return ret;
 577         }
 578
 579         di = (struct ocfs2_dinode *)di_bh->b_data;
 580
 581         down_read(&oi->ip_xattr_sem);
 582         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
 583         if (i_ret < 0)
 584                 b_ret = 0;
 585         else {
 586                 if (buffer) {
 587                         buffer += i_ret;
 588                         size -= i_ret;
 589                 }
 590                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
 591                                                buffer, size);
 592                 if (b_ret < 0)
 593                         i_ret = 0;
 594         }
 595         up_read(&oi->ip_xattr_sem);
 596         ocfs2_inode_unlock(dentry->d_inode, 0);
 597
 598         brelse(di_bh);
 599
 600         return i_ret + b_ret;
 601 }
 602
 603 static int ocfs2_xattr_find_entry(int name_index,
 604                                   const char *name,
 605                                   struct ocfs2_xattr_search *xs)
 606 {
 607         struct ocfs2_xattr_entry *entry;
 608         size_t name_len;
 609         int i, cmp = 1;
 610
 611         if (name == NULL)
 612                 return -EINVAL;
 613
 614         name_len = strlen(name);
 615         entry = xs->here;
 616         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
 617                 cmp = name_index - ocfs2_xattr_get_type(entry);
 618                 if (!cmp)
 619                         cmp = name_len - entry->xe_name_len;
 620                 if (!cmp)
 621                         cmp = memcmp(name, (xs->base +
 622                                      le16_to_cpu(entry->xe_name_offset)),
 623                                      name_len);
 624                 if (cmp == 0)
 625                         break;
 626                 entry += 1;
 627         }
 628         xs->here = entry;
 629
 630         return cmp ? -ENODATA : 0;
 631 }
 632
 633 static int ocfs2_xattr_get_value_outside(struct inode *inode,
 634                                          struct ocfs2_xattr_value_root *xv,
 635                                          void *buffer,
 636                                          size_t len)
 637 {
 638         u32 cpos, p_cluster, num_clusters, bpc, clusters;
 639         u64 blkno;
 640         int i, ret = 0;
 641         size_t cplen, blocksize;
 642         struct buffer_head *bh = NULL;
 643         struct ocfs2_extent_list *el;
 644
 645         el = &xv->xr_list;
 646         clusters = le32_to_cpu(xv->xr_clusters);
 647         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 648         blocksize = inode->i_sb->s_blocksize;
 649
 650         cpos = 0;
 651         while (cpos < clusters) {
 652                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 653                                                &num_clusters, el);
 654                 if (ret) {
 655                         mlog_errno(ret);
 656                         goto out;
 657                 }
 658
 659                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 660                 /* Copy ocfs2_xattr_value */
 661                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 662                         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
 663                                                &bh, OCFS2_BH_CACHED, inode);
 664                         if (ret) {
 665                                 mlog_errno(ret);
 666                                 goto out;
 667                         }
 668
 669                         cplen = len >= blocksize ? blocksize : len;
 670                         memcpy(buffer, bh->b_data, cplen);
 671                         len -= cplen;
 672                         buffer += cplen;
 673
 674                         brelse(bh);
 675                         bh = NULL;
 676                         if (len == 0)
 677                                 break;
 678                 }
 679                 cpos += num_clusters;
 680         }
 681 out:
 682         return ret;
 683 }
 684
 685 static int ocfs2_xattr_ibody_get(struct inode *inode,
 686                                  int name_index,
 687                                  const char *name,
 688                                  void *buffer,
 689                                  size_t buffer_size,
 690                                  struct ocfs2_xattr_search *xs)
 691 {
 692         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 693         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
 694         struct ocfs2_xattr_value_root *xv;
 695         size_t size;
 696         int ret = 0;
 697
 698         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 699                 return -ENODATA;
 700
 701         xs->end = (void *)di + inode->i_sb->s_blocksize;
 702         xs->header = (struct ocfs2_xattr_header *)
 703                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
 704         xs->base = (void *)xs->header;
 705         xs->here = xs->header->xh_entries;
 706
 707         ret = ocfs2_xattr_find_entry(name_index, name, xs);
 708         if (ret)
 709                 return ret;
 710         size = le64_to_cpu(xs->here->xe_value_size);
 711         if (buffer) {
 712                 if (size > buffer_size)
 713                         return -ERANGE;
 714                 if (ocfs2_xattr_is_local(xs->here)) {
 715                         memcpy(buffer, (void *)xs->base +
 716                                le16_to_cpu(xs->here->xe_name_offset) +
 717                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
 718                 } else {
 719                         xv = (struct ocfs2_xattr_value_root *)
 720                                 (xs->base + le16_to_cpu(
 721                                  xs->here->xe_name_offset) +
 722                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
 723                         ret = ocfs2_xattr_get_value_outside(inode, xv,
 724                                                             buffer, size);
 725                         if (ret < 0) {
 726                                 mlog_errno(ret);
 727                                 return ret;
 728                         }
 729                 }
 730         }
 731
 732         return size;
 733 }
 734
 735 static int ocfs2_xattr_block_get(struct inode *inode,
 736                                  int name_index,
 737                                  const char *name,
 738                                  void *buffer,
 739                                  size_t buffer_size,
 740                                  struct ocfs2_xattr_search *xs)
 741 {
 742         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
 743         struct buffer_head *blk_bh = NULL;
 744         struct ocfs2_xattr_block *xb;
 745         struct ocfs2_xattr_value_root *xv;
 746         size_t size;
 747         int ret = -ENODATA, name_offset, name_len, block_off, i;
 748
 749         if (!di->i_xattr_loc)
 750                 return ret;
 751
 752         memset(&xs->bucket, 0, sizeof(xs->bucket));
 753
 754         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
 755                                le64_to_cpu(di->i_xattr_loc),
 756                                &blk_bh, OCFS2_BH_CACHED, inode);
 757         if (ret < 0) {
 758                 mlog_errno(ret);
 759                 return ret;
 760         }
 761         /*Verify the signature of xattr block*/
 762         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
 763                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
 764                 ret = -EFAULT;
 765                 goto cleanup;
 766         }
 767
 768         xs->xattr_bh = blk_bh;
 769         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 770
 771         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 772                 xs->header = &xb->xb_attrs.xb_header;
 773                 xs->base = (void *)xs->header;
 774                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
 775                 xs->here = xs->header->xh_entries;
 776
 777                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
 778         } else
 779                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
 780                                                    name_index,
 781                                                    name, xs);
 782
 783         if (ret)
 784                 goto cleanup;
 785         size = le64_to_cpu(xs->here->xe_value_size);
 786         if (buffer) {
 787                 ret = -ERANGE;
 788                 if (size > buffer_size)
 789                         goto cleanup;
 790
 791                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
 792                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
 793                 i = xs->here - xs->header->xh_entries;
 794
 795                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
 796                         ret = ocfs2_xattr_bucket_get_name_value(inode,
 797                                                                 xs->bucket.xh,
 798                                                                 i,
 799                                                                 &block_off,
 800                                                                 &name_offset);
 801                         xs->base = xs->bucket.bhs[block_off]->b_data;
 802                 }
 803                 if (ocfs2_xattr_is_local(xs->here)) {
 804                         memcpy(buffer, (void *)xs->base +
 805                                name_offset + name_len, size);
 806                 } else {
 807                         xv = (struct ocfs2_xattr_value_root *)
 808                                 (xs->base + name_offset + name_len);
 809                         ret = ocfs2_xattr_get_value_outside(inode, xv,
 810                                                             buffer, size);
 811                         if (ret < 0) {
 812                                 mlog_errno(ret);
 813                                 goto cleanup;
 814                         }
 815                 }
 816         }
 817         ret = size;
 818 cleanup:
 819         for (i = 0; i < OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET; i++)
 820                 brelse(xs->bucket.bhs[i]);
 821         memset(&xs->bucket, 0, sizeof(xs->bucket));
 822
 823         brelse(blk_bh);
 824         return ret;
 825 }
 826
 827 /* ocfs2_xattr_get()
 828  *
 829  * Copy an extended attribute into the buffer provided.
 830  * Buffer is NULL to compute the size of buffer required.
 831  */
 832 int ocfs2_xattr_get(struct inode *inode,
 833                     int name_index,
 834                     const char *name,
 835                     void *buffer,
 836                     size_t buffer_size)
 837 {
 838         int ret;
 839         struct ocfs2_dinode *di = NULL;
 840         struct buffer_head *di_bh = NULL;
 841         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 842         struct ocfs2_xattr_search xis = {
 843                 .not_found = -ENODATA,
 844         };
 845         struct ocfs2_xattr_search xbs = {
 846                 .not_found = -ENODATA,
 847         };
 848
 849         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
 850                 return -EOPNOTSUPP;
 851
 852         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 853                 ret = -ENODATA;
 854
 855         ret = ocfs2_inode_lock(inode, &di_bh, 0);
 856         if (ret < 0) {
 857                 mlog_errno(ret);
 858                 return ret;
 859         }
 860         xis.inode_bh = xbs.inode_bh = di_bh;
 861         di = (struct ocfs2_dinode *)di_bh->b_data;
 862
 863         down_read(&oi->ip_xattr_sem);
 864         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
 865                                     buffer_size, &xis);
 866         if (ret == -ENODATA)
 867                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
 868                                             buffer_size, &xbs);
 869         up_read(&oi->ip_xattr_sem);
 870         ocfs2_inode_unlock(inode, 0);
 871
 872         brelse(di_bh);
 873
 874         return ret;
 875 }
 876
 877 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
 878                                            struct ocfs2_xattr_value_root *xv,
 879                                            const void *value,
 880                                            int value_len)
 881 {
 882         int ret = 0, i, cp_len, credits;
 883         u16 blocksize = inode->i_sb->s_blocksize;
 884         u32 p_cluster, num_clusters;
 885         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 886         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
 887         u64 blkno;
 888         struct buffer_head *bh = NULL;
 889         handle_t *handle;
 890
 891         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
 892
 893         credits = clusters * bpc;
 894         handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits);
 895         if (IS_ERR(handle)) {
 896                 ret = PTR_ERR(handle);
 897                 mlog_errno(ret);
 898                 goto out;
 899         }
 900
 901         while (cpos < clusters) {
 902                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 903                                                &num_clusters, &xv->xr_list);
 904                 if (ret) {
 905                         mlog_errno(ret);
 906                         goto out_commit;
 907                 }
 908
 909                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 910
 911                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 912                         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
 913                                                &bh, OCFS2_BH_CACHED, inode);
 914                         if (ret) {
 915                                 mlog_errno(ret);
 916                                 goto out_commit;
 917                         }
 918
 919                         ret = ocfs2_journal_access(handle,
 920                                                    inode,
 921                                                    bh,
 922                                                    OCFS2_JOURNAL_ACCESS_WRITE);
 923                         if (ret < 0) {
 924                                 mlog_errno(ret);
 925                                 goto out_commit;
 926                         }
 927
 928                         cp_len = value_len > blocksize ? blocksize : value_len;
 929                         memcpy(bh->b_data, value, cp_len);
 930                         value_len -= cp_len;
 931                         value += cp_len;
 932                         if (cp_len < blocksize)
 933                                 memset(bh->b_data + cp_len, 0,
 934                                        blocksize - cp_len);
 935
 936                         ret = ocfs2_journal_dirty(handle, bh);
 937                         if (ret < 0) {
 938                                 mlog_errno(ret);
 939                                 goto out_commit;
 940                         }
 941                         brelse(bh);
 942                         bh = NULL;
 943
 944                         /*
 945                          * XXX: do we need to empty all the following
 946                          * blocks in this cluster?
 947                          */
 948                         if (!value_len)
 949                                 break;
 950                 }
 951                 cpos += num_clusters;
 952         }
 953 out_commit:
 954         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 955 out:
 956         brelse(bh);
 957
 958         return ret;
 959 }
 960
 961 static int ocfs2_xattr_cleanup(struct inode *inode,
 962                                struct ocfs2_xattr_info *xi,
 963                                struct ocfs2_xattr_search *xs,
 964                                size_t offs)
 965 {
 966         handle_t *handle = NULL;
 967         int ret = 0;
 968         size_t name_len = strlen(xi->name);
 969         void *val = xs->base + offs;
 970         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 971
 972         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
 973                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
 974         if (IS_ERR(handle)) {
 975                 ret = PTR_ERR(handle);
 976                 mlog_errno(ret);
 977                 goto out;
 978         }
 979         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
 980                                    OCFS2_JOURNAL_ACCESS_WRITE);
 981         if (ret) {
 982                 mlog_errno(ret);
 983                 goto out_commit;
 984         }
 985         /* Decrease xattr count */
 986         le16_add_cpu(&xs->header->xh_count, -1);
 987         /* Remove the xattr entry and tree root which has already be set*/
 988         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
 989         memset(val, 0, size);
 990
 991         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
 992         if (ret < 0)
 993                 mlog_errno(ret);
 994 out_commit:
 995         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 996 out:
 997         return ret;
 998 }
 999
1000 static int ocfs2_xattr_update_entry(struct inode *inode,
1001                                     struct ocfs2_xattr_info *xi,
1002                                     struct ocfs2_xattr_search *xs,
1003                                     size_t offs)
1004 {
1005         handle_t *handle = NULL;
1006         int ret = 0;
1007
1008         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1009                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1010         if (IS_ERR(handle)) {
1011                 ret = PTR_ERR(handle);
1012                 mlog_errno(ret);
1013                 goto out;
1014         }
1015         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1016                                    OCFS2_JOURNAL_ACCESS_WRITE);
1017         if (ret) {
1018                 mlog_errno(ret);
1019                 goto out_commit;
1020         }
1021
1022         xs->here->xe_name_offset = cpu_to_le16(offs);
1023         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1024         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1025                 ocfs2_xattr_set_local(xs->here, 1);
1026         else
1027                 ocfs2_xattr_set_local(xs->here, 0);
1028         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1029
1030         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1031         if (ret < 0)
1032                 mlog_errno(ret);
1033 out_commit:
1034         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1035 out:
1036         return ret;
1037 }
1038
1039 /*
1040  * ocfs2_xattr_set_value_outside()
1041  *
1042  * Set large size value in B tree.
1043  */
1044 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1045                                          struct ocfs2_xattr_info *xi,
1046                                          struct ocfs2_xattr_search *xs,
1047                                          size_t offs)
1048 {
1049         size_t name_len = strlen(xi->name);
1050         void *val = xs->base + offs;
1051         struct ocfs2_xattr_value_root *xv = NULL;
1052         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1053         int ret = 0;
1054
1055         memset(val, 0, size);
1056         memcpy(val, xi->name, name_len);
1057         xv = (struct ocfs2_xattr_value_root *)
1058                 (val + OCFS2_XATTR_SIZE(name_len));
1059         xv->xr_clusters = 0;
1060         xv->xr_last_eb_blk = 0;
1061         xv->xr_list.l_tree_depth = 0;
1062         xv->xr_list.l_count = cpu_to_le16(1);
1063         xv->xr_list.l_next_free_rec = 0;
1064
1065         ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1066                                          xi->value_len);
1067         if (ret < 0) {
1068                 mlog_errno(ret);
1069                 return ret;
1070         }
1071         ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value,
1072                                               xi->value_len);
1073         if (ret < 0) {
1074                 mlog_errno(ret);
1075                 return ret;
1076         }
1077         ret = ocfs2_xattr_update_entry(inode, xi, xs, offs);
1078         if (ret < 0)
1079                 mlog_errno(ret);
1080
1081         return ret;
1082 }
1083
1084 /*
1085  * ocfs2_xattr_set_entry_local()
1086  *
1087  * Set, replace or remove extended attribute in local.
1088  */
1089 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1090                                         struct ocfs2_xattr_info *xi,
1091                                         struct ocfs2_xattr_search *xs,
1092                                         struct ocfs2_xattr_entry *last,
1093                                         size_t min_offs)
1094 {
1095         size_t name_len = strlen(xi->name);
1096         int i;
1097
1098         if (xi->value && xs->not_found) {
1099                 /* Insert the new xattr entry. */
1100                 le16_add_cpu(&xs->header->xh_count, 1);
1101                 ocfs2_xattr_set_type(last, xi->name_index);
1102                 ocfs2_xattr_set_local(last, 1);
1103                 last->xe_name_len = name_len;
1104         } else {
1105                 void *first_val;
1106                 void *val;
1107                 size_t offs, size;
1108
1109                 first_val = xs->base + min_offs;
1110                 offs = le16_to_cpu(xs->here->xe_name_offset);
1111                 val = xs->base + offs;
1112
1113                 if (le64_to_cpu(xs->here->xe_value_size) >
1114                     OCFS2_XATTR_INLINE_SIZE)
1115                         size = OCFS2_XATTR_SIZE(name_len) +
1116                                 OCFS2_XATTR_ROOT_SIZE;
1117                 else
1118                         size = OCFS2_XATTR_SIZE(name_len) +
1119                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1120
1121                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1122                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1123                         /* The old and the new value have the
1124                            same size. Just replace the value. */
1125                         ocfs2_xattr_set_local(xs->here, 1);
1126                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1127                         /* Clear value bytes. */
1128                         memset(val + OCFS2_XATTR_SIZE(name_len),
1129                                0,
1130                                OCFS2_XATTR_SIZE(xi->value_len));
1131                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1132                                xi->value,
1133                                xi->value_len);
1134                         return;
1135                 }
1136                 /* Remove the old name+value. */
1137                 memmove(first_val + size, first_val, val - first_val);
1138                 memset(first_val, 0, size);
1139                 xs->here->xe_name_hash = 0;
1140                 xs->here->xe_name_offset = 0;
1141                 ocfs2_xattr_set_local(xs->here, 1);
1142                 xs->here->xe_value_size = 0;
1143
1144                 min_offs += size;
1145
1146                 /* Adjust all value offsets. */
1147                 last = xs->header->xh_entries;
1148                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1149                         size_t o = le16_to_cpu(last->xe_name_offset);
1150
1151                         if (o < offs)
1152                                 last->xe_name_offset = cpu_to_le16(o + size);
1153                         last += 1;
1154                 }
1155
1156                 if (!xi->value) {
1157                         /* Remove the old entry. */
1158                         last -= 1;
1159                         memmove(xs->here, xs->here + 1,
1160                                 (void *)last - (void *)xs->here);
1161                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1162                         le16_add_cpu(&xs->header->xh_count, -1);
1163                 }
1164         }
1165         if (xi->value) {
1166                 /* Insert the new name+value. */
1167                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1168                                 OCFS2_XATTR_SIZE(xi->value_len);
1169                 void *val = xs->base + min_offs - size;
1170
1171                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1172                 memset(val, 0, size);
1173                 memcpy(val, xi->name, name_len);
1174                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1175                        xi->value,
1176                        xi->value_len);
1177                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1178                 ocfs2_xattr_set_local(xs->here, 1);
1179                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1180         }
1181
1182         return;
1183 }
1184
1185 /*
1186  * ocfs2_xattr_set_entry()
1187  *
1188  * Set extended attribute entry into inode or block.
1189  *
1190  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1191  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1192  * then set value in B tree with set_value_outside().
1193  */
1194 static int ocfs2_xattr_set_entry(struct inode *inode,
1195                                  struct ocfs2_xattr_info *xi,
1196                                  struct ocfs2_xattr_search *xs,
1197                                  int flag)
1198 {
1199         struct ocfs2_xattr_entry *last;
1200         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1201         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1202         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1203         size_t size_l = 0;
1204         handle_t *handle = NULL;
1205         int free, i, ret;
1206         struct ocfs2_xattr_info xi_l = {
1207                 .name_index = xi->name_index,
1208                 .name = xi->name,
1209                 .value = xi->value,
1210                 .value_len = xi->value_len,
1211         };
1212
1213         /* Compute min_offs, last and free space. */
1214         last = xs->header->xh_entries;
1215
1216         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1217                 size_t offs = le16_to_cpu(last->xe_name_offset);
1218                 if (offs < min_offs)
1219                         min_offs = offs;
1220                 last += 1;
1221         }
1222
1223         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1224         if (free < 0)
1225                 return -EFAULT;
1226
1227         if (!xs->not_found) {
1228                 size_t size = 0;
1229                 if (ocfs2_xattr_is_local(xs->here))
1230                         size = OCFS2_XATTR_SIZE(name_len) +
1231                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1232                 else
1233                         size = OCFS2_XATTR_SIZE(name_len) +
1234                                 OCFS2_XATTR_ROOT_SIZE;
1235                 free += (size + sizeof(struct ocfs2_xattr_entry));
1236         }
1237         /* Check free space in inode or block */
1238         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1239                 if (free < sizeof(struct ocfs2_xattr_entry) +
1240                            OCFS2_XATTR_SIZE(name_len) +
1241                            OCFS2_XATTR_ROOT_SIZE) {
1242                         ret = -ENOSPC;
1243                         goto out;
1244                 }
1245                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1246                 xi_l.value = (void *)&def_xv;
1247                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1248         } else if (xi->value) {
1249                 if (free < sizeof(struct ocfs2_xattr_entry) +
1250                            OCFS2_XATTR_SIZE(name_len) +
1251                            OCFS2_XATTR_SIZE(xi->value_len)) {
1252                         ret = -ENOSPC;
1253                         goto out;
1254                 }
1255         }
1256
1257         if (!xs->not_found) {
1258                 /* For existing extended attribute */
1259                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1260                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1261                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1262                 void *val = xs->base + offs;
1263
1264                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1265                         /* Replace existing local xattr with tree root */
1266                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1267                                                             offs);
1268                         if (ret < 0)
1269                                 mlog_errno(ret);
1270                         goto out;
1271                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1272                         /* For existing xattr which has value outside */
1273                         struct ocfs2_xattr_value_root *xv = NULL;
1274                         xv = (struct ocfs2_xattr_value_root *)(val +
1275                                 OCFS2_XATTR_SIZE(name_len));
1276
1277                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1278                                 /*
1279                                  * If new value need set outside also,
1280                                  * first truncate old value to new value,
1281                                  * then set new value with set_value_outside().
1282                                  */
1283                                 ret = ocfs2_xattr_value_truncate(inode,
1284                                                                  xs->xattr_bh,
1285                                                                  xv,
1286                                                                  xi->value_len);
1287                                 if (ret < 0) {
1288                                         mlog_errno(ret);
1289                                         goto out;
1290                                 }
1291
1292                                 ret = __ocfs2_xattr_set_value_outside(inode,
1293                                                                 xv,
1294                                                                 xi->value,
1295                                                                 xi->value_len);
1296                                 if (ret < 0) {
1297                                         mlog_errno(ret);
1298                                         goto out;
1299                                 }
1300
1301                                 ret = ocfs2_xattr_update_entry(inode,
1302                                                                xi,
1303                                                                xs,
1304                                                                offs);
1305                                 if (ret < 0)
1306                                         mlog_errno(ret);
1307                                 goto out;
1308                         } else {
1309                                 /*
1310                                  * If new value need set in local,
1311                                  * just trucate old value to zero.
1312                                  */
1313                                  ret = ocfs2_xattr_value_truncate(inode,
1314                                                                  xs->xattr_bh,
1315                                                                  xv,
1316                                                                  0);
1317                                 if (ret < 0)
1318                                         mlog_errno(ret);
1319                         }
1320                 }
1321         }
1322
1323         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1324                                    OCFS2_INODE_UPDATE_CREDITS);
1325         if (IS_ERR(handle)) {
1326                 ret = PTR_ERR(handle);
1327                 mlog_errno(ret);
1328                 goto out;
1329         }
1330
1331         ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1332                                    OCFS2_JOURNAL_ACCESS_WRITE);
1333         if (ret) {
1334                 mlog_errno(ret);
1335                 goto out_commit;
1336         }
1337
1338         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1339                 /* set extended attribute in external block. */
1340                 ret = ocfs2_extend_trans(handle,
1341                                          OCFS2_INODE_UPDATE_CREDITS +
1342                                          OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1343                 if (ret) {
1344                         mlog_errno(ret);
1345                         goto out_commit;
1346                 }
1347                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1348                                            OCFS2_JOURNAL_ACCESS_WRITE);
1349                 if (ret) {
1350                         mlog_errno(ret);
1351                         goto out_commit;
1352                 }
1353         }
1354
1355         /*
1356          * Set value in local, include set tree root in local.
1357          * This is the first step for value size >INLINE_SIZE.
1358          */
1359         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1360
1361         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1362                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1363                 if (ret < 0) {
1364                         mlog_errno(ret);
1365                         goto out_commit;
1366                 }
1367         }
1368
1369         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1370             (flag & OCFS2_INLINE_XATTR_FL)) {
1371                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1372                 unsigned int xattrsize = osb->s_xattr_inline_size;
1373
1374                 /*
1375                  * Adjust extent record count or inline data size
1376                  * to reserve space for extended attribute.
1377                  */
1378                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1379                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1380                         le16_add_cpu(&idata->id_count, -xattrsize);
1381                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1382                         struct ocfs2_extent_list *el = &di->id2.i_list;
1383                         le16_add_cpu(&el->l_count, -(xattrsize /
1384                                         sizeof(struct ocfs2_extent_rec)));
1385                 }
1386                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1387         }
1388         /* Update xattr flag */
1389         spin_lock(&oi->ip_lock);
1390         oi->ip_dyn_features |= flag;
1391         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1392         spin_unlock(&oi->ip_lock);
1393         /* Update inode ctime */
1394         inode->i_ctime = CURRENT_TIME;
1395         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1396         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1397
1398         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1399         if (ret < 0)
1400                 mlog_errno(ret);
1401
1402 out_commit:
1403         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1404
1405         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1406                 /*
1407                  * Set value outside in B tree.
1408                  * This is the second step for value size > INLINE_SIZE.
1409                  */
1410                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1411                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1412                 if (ret < 0) {
1413                         int ret2;
1414
1415                         mlog_errno(ret);
1416                         /*
1417                          * If set value outside failed, we have to clean
1418                          * the junk tree root we have already set in local.
1419                          */
1420                         ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1421                         if (ret2 < 0)
1422                                 mlog_errno(ret2);
1423                 }
1424         }
1425 out:
1426         return ret;
1427
1428 }
1429
1430 static int ocfs2_xattr_free_block(handle_t *handle,
1431                                   struct ocfs2_super *osb,
1432                                   struct ocfs2_xattr_block *xb)
1433 {
1434         struct inode *xb_alloc_inode;
1435         struct buffer_head *xb_alloc_bh = NULL;
1436         u64 blk = le64_to_cpu(xb->xb_blkno);
1437         u16 bit = le16_to_cpu(xb->xb_suballoc_bit);
1438         u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1439         int ret = 0;
1440
1441         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1442                                 EXTENT_ALLOC_SYSTEM_INODE,
1443                                 le16_to_cpu(xb->xb_suballoc_slot));
1444         if (!xb_alloc_inode) {
1445                 ret = -ENOMEM;
1446                 mlog_errno(ret);
1447                 goto out;
1448         }
1449         mutex_lock(&xb_alloc_inode->i_mutex);
1450
1451         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1452         if (ret < 0) {
1453                 mlog_errno(ret);
1454                 goto out_mutex;
1455         }
1456         ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE);
1457         if (ret < 0) {
1458                 mlog_errno(ret);
1459                 goto out_unlock;
1460         }
1461         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1462                                        bit, bg_blkno, 1);
1463         if (ret < 0)
1464                 mlog_errno(ret);
1465 out_unlock:
1466         ocfs2_inode_unlock(xb_alloc_inode, 1);
1467         brelse(xb_alloc_bh);
1468 out_mutex:
1469         mutex_unlock(&xb_alloc_inode->i_mutex);
1470         iput(xb_alloc_inode);
1471 out:
1472         return ret;
1473 }
1474
1475 static int ocfs2_remove_value_outside(struct inode*inode,
1476                                       struct buffer_head *bh,
1477                                       struct ocfs2_xattr_header *header)
1478 {
1479         int ret = 0, i;
1480
1481         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1482                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1483
1484                 if (!ocfs2_xattr_is_local(entry)) {
1485                         struct ocfs2_xattr_value_root *xv;
1486                         void *val;
1487
1488                         val = (void *)header +
1489                                 le16_to_cpu(entry->xe_name_offset);
1490                         xv = (struct ocfs2_xattr_value_root *)
1491                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1492                         ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1493                         if (ret < 0) {
1494                                 mlog_errno(ret);
1495                                 return ret;
1496                         }
1497                 }
1498         }
1499
1500         return ret;
1501 }
1502
1503 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1504                                     struct buffer_head *di_bh)
1505 {
1506
1507         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1508         struct ocfs2_xattr_header *header;
1509         int ret;
1510
1511         header = (struct ocfs2_xattr_header *)
1512                  ((void *)di + inode->i_sb->s_blocksize -
1513                  le16_to_cpu(di->i_xattr_inline_size));
1514
1515         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1516
1517         return ret;
1518 }
1519
1520 static int ocfs2_xattr_block_remove(struct inode *inode,
1521                                     struct buffer_head *blk_bh)
1522 {
1523         struct ocfs2_xattr_block *xb;
1524         int ret = 0;
1525
1526         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1527         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1528                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1529                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1530         } else
1531                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1532
1533         return ret;
1534 }
1535
1536 /*
1537  * ocfs2_xattr_remove()
1538  *
1539  * Free extended attribute resources associated with this inode.
1540  */
1541 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1542 {
1543         struct ocfs2_xattr_block *xb;
1544         struct buffer_head *blk_bh = NULL;
1545         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1546         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1547         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1548         handle_t *handle;
1549         int ret;
1550
1551         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1552                 return 0;
1553
1554         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1555                 return 0;
1556
1557         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1558                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1559                 if (ret < 0) {
1560                         mlog_errno(ret);
1561                         goto out;
1562                 }
1563         }
1564         if (di->i_xattr_loc) {
1565                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1566                                        le64_to_cpu(di->i_xattr_loc),
1567                                        &blk_bh, OCFS2_BH_CACHED, inode);
1568                 if (ret < 0) {
1569                         mlog_errno(ret);
1570                         return ret;
1571                 }
1572                 /*Verify the signature of xattr block*/
1573                 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1574                            strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1575                         ret = -EFAULT;
1576                         goto out;
1577                 }
1578
1579                 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1580                 if (ret < 0) {
1581                         mlog_errno(ret);
1582                         goto out;
1583                 }
1584         }
1585
1586         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1587                                    OCFS2_INODE_UPDATE_CREDITS);
1588         if (IS_ERR(handle)) {
1589                 ret = PTR_ERR(handle);
1590                 mlog_errno(ret);
1591                 goto out;
1592         }
1593         ret = ocfs2_journal_access(handle, inode, di_bh,
1594                                    OCFS2_JOURNAL_ACCESS_WRITE);
1595         if (ret) {
1596                 mlog_errno(ret);
1597                 goto out_commit;
1598         }
1599
1600         if (di->i_xattr_loc) {
1601                 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1602                 ocfs2_xattr_free_block(handle, osb, xb);
1603                 di->i_xattr_loc = cpu_to_le64(0);
1604         }
1605
1606         spin_lock(&oi->ip_lock);
1607         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1608         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1609         spin_unlock(&oi->ip_lock);
1610
1611         ret = ocfs2_journal_dirty(handle, di_bh);
1612         if (ret < 0)
1613                 mlog_errno(ret);
1614 out_commit:
1615         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1616 out:
1617         brelse(blk_bh);
1618
1619         return ret;
1620 }
1621
1622 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1623                                         struct ocfs2_dinode *di)
1624 {
1625         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1626         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1627         int free;
1628
1629         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1630                 return 0;
1631
1632         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1633                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1634                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1635         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1636                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1637                         le64_to_cpu(di->i_size);
1638         } else {
1639                 struct ocfs2_extent_list *el = &di->id2.i_list;
1640                 free = (le16_to_cpu(el->l_count) -
1641                         le16_to_cpu(el->l_next_free_rec)) *
1642                         sizeof(struct ocfs2_extent_rec);
1643         }
1644         if (free >= xattrsize)
1645                 return 1;
1646
1647         return 0;
1648 }
1649
1650 /*
1651  * ocfs2_xattr_ibody_find()
1652  *
1653  * Find extended attribute in inode block and
1654  * fill search info into struct ocfs2_xattr_search.
1655  */
1656 static int ocfs2_xattr_ibody_find(struct inode *inode,
1657                                   int name_index,
1658                                   const char *name,
1659                                   struct ocfs2_xattr_search *xs)
1660 {
1661         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1662         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1663         int ret;
1664         int has_space = 0;
1665
1666         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1667                 return 0;
1668
1669         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1670                 down_read(&oi->ip_alloc_sem);
1671                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1672                 up_read(&oi->ip_alloc_sem);
1673                 if (!has_space)
1674                         return 0;
1675         }
1676
1677         xs->xattr_bh = xs->inode_bh;
1678         xs->end = (void *)di + inode->i_sb->s_blocksize;
1679         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1680                 xs->header = (struct ocfs2_xattr_header *)
1681                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1682         else
1683                 xs->header = (struct ocfs2_xattr_header *)
1684                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1685         xs->base = (void *)xs->header;
1686         xs->here = xs->header->xh_entries;
1687
1688         /* Find the named attribute. */
1689         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1690                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1691                 if (ret && ret != -ENODATA)
1692                         return ret;
1693                 xs->not_found = ret;
1694         }
1695
1696         return 0;
1697 }
1698
1699 /*
1700  * ocfs2_xattr_ibody_set()
1701  *
1702  * Set, replace or remove an extended attribute into inode block.
1703  *
1704  */
1705 static int ocfs2_xattr_ibody_set(struct inode *inode,
1706                                  struct ocfs2_xattr_info *xi,
1707                                  struct ocfs2_xattr_search *xs)
1708 {
1709         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1710         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1711         int ret;
1712
1713         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1714                 return -ENOSPC;
1715
1716         down_write(&oi->ip_alloc_sem);
1717         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1718                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1719                         ret = -ENOSPC;
1720                         goto out;
1721                 }
1722         }
1723
1724         ret = ocfs2_xattr_set_entry(inode, xi, xs,
1725                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1726 out:
1727         up_write(&oi->ip_alloc_sem);
1728
1729         return ret;
1730 }
1731
1732 /*
1733  * ocfs2_xattr_block_find()
1734  *
1735  * Find extended attribute in external block and
1736  * fill search info into struct ocfs2_xattr_search.
1737  */
1738 static int ocfs2_xattr_block_find(struct inode *inode,
1739                                   int name_index,
1740                                   const char *name,
1741                                   struct ocfs2_xattr_search *xs)
1742 {
1743         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1744         struct buffer_head *blk_bh = NULL;
1745         struct ocfs2_xattr_block *xb;
1746         int ret = 0;
1747
1748         if (!di->i_xattr_loc)
1749                 return ret;
1750
1751         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1752                                le64_to_cpu(di->i_xattr_loc),
1753                                &blk_bh, OCFS2_BH_CACHED, inode);
1754         if (ret < 0) {
1755                 mlog_errno(ret);
1756                 return ret;
1757         }
1758         /*Verify the signature of xattr block*/
1759         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1760                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1761                         ret = -EFAULT;
1762                         goto cleanup;
1763         }
1764
1765         xs->xattr_bh = blk_bh;
1766         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1767
1768         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1769                 xs->header = &xb->xb_attrs.xb_header;
1770                 xs->base = (void *)xs->header;
1771                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1772                 xs->here = xs->header->xh_entries;
1773
1774                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1775         } else
1776                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1777                                                    name_index,
1778                                                    name, xs);
1779
1780         if (ret && ret != -ENODATA) {
1781                 xs->xattr_bh = NULL;
1782                 goto cleanup;
1783         }
1784         xs->not_found = ret;
1785         return 0;
1786 cleanup:
1787         brelse(blk_bh);
1788
1789         return ret;
1790 }
1791
1792 /*
1793  * When all the xattrs are deleted from index btree, the ocfs2_xattr_tree
1794  * will be erased and ocfs2_xattr_block will have its ocfs2_xattr_header
1795  * re-initialized.
1796  */
1797 static int ocfs2_restore_xattr_block(struct inode *inode,
1798                                      struct ocfs2_xattr_search *xs)
1799 {
1800         int ret;
1801         handle_t *handle;
1802         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1803         struct ocfs2_xattr_block *xb =
1804                 (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1805         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
1806         u16 xb_flags = le16_to_cpu(xb->xb_flags);
1807
1808         BUG_ON(!(xb_flags & OCFS2_XATTR_INDEXED) ||
1809                 le16_to_cpu(el->l_next_free_rec) != 0);
1810
1811         handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1812         if (IS_ERR(handle)) {
1813                 ret = PTR_ERR(handle);
1814                 handle = NULL;
1815                 goto out;
1816         }
1817
1818         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1819                                    OCFS2_JOURNAL_ACCESS_WRITE);
1820         if (ret < 0) {
1821                 mlog_errno(ret);
1822                 goto out_commit;
1823         }
1824
1825         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
1826                offsetof(struct ocfs2_xattr_block, xb_attrs));
1827
1828         xb->xb_flags = cpu_to_le16(xb_flags & ~OCFS2_XATTR_INDEXED);
1829
1830         ocfs2_journal_dirty(handle, xs->xattr_bh);
1831
1832 out_commit:
1833         ocfs2_commit_trans(osb, handle);
1834 out:
1835         return ret;
1836 }
1837
1838 /*
1839  * ocfs2_xattr_block_set()
1840  *
1841  * Set, replace or remove an extended attribute into external block.
1842  *
1843  */
1844 static int ocfs2_xattr_block_set(struct inode *inode,
1845                                  struct ocfs2_xattr_info *xi,
1846                                  struct ocfs2_xattr_search *xs)
1847 {
1848         struct buffer_head *new_bh = NULL;
1849         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1850         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
1851         struct ocfs2_alloc_context *meta_ac = NULL;
1852         handle_t *handle = NULL;
1853         struct ocfs2_xattr_block *xblk = NULL;
1854         u16 suballoc_bit_start;
1855         u32 num_got;
1856         u64 first_blkno;
1857         int ret;
1858
1859         if (!xs->xattr_bh) {
1860                 /*
1861                  * Alloc one external block for extended attribute
1862                  * outside of inode.
1863                  */
1864                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1865                 if (ret < 0) {
1866                         mlog_errno(ret);
1867                         goto out;
1868                 }
1869                 handle = ocfs2_start_trans(osb,
1870                                            OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1871                 if (IS_ERR(handle)) {
1872                         ret = PTR_ERR(handle);
1873                         mlog_errno(ret);
1874                         goto out;
1875                 }
1876                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1877                                            OCFS2_JOURNAL_ACCESS_CREATE);
1878                 if (ret < 0) {
1879                         mlog_errno(ret);
1880                         goto out_commit;
1881                 }
1882
1883                 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1884                                            &suballoc_bit_start, &num_got,
1885                                            &first_blkno);
1886                 if (ret < 0) {
1887                         mlog_errno(ret);
1888                         goto out_commit;
1889                 }
1890
1891                 new_bh = sb_getblk(inode->i_sb, first_blkno);
1892                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1893
1894                 ret = ocfs2_journal_access(handle, inode, new_bh,
1895                                            OCFS2_JOURNAL_ACCESS_CREATE);
1896                 if (ret < 0) {
1897                         mlog_errno(ret);
1898                         goto out_commit;
1899                 }
1900
1901                 /* Initialize ocfs2_xattr_block */
1902                 xs->xattr_bh = new_bh;
1903                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1904                 memset(xblk, 0, inode->i_sb->s_blocksize);
1905                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1906                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1907                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1908                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1909                 xblk->xb_blkno = cpu_to_le64(first_blkno);
1910
1911                 xs->header = &xblk->xb_attrs.xb_header;
1912                 xs->base = (void *)xs->header;
1913                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1914                 xs->here = xs->header->xh_entries;
1915
1916
1917                 ret = ocfs2_journal_dirty(handle, new_bh);
1918                 if (ret < 0) {
1919                         mlog_errno(ret);
1920                         goto out_commit;
1921                 }
1922                 di->i_xattr_loc = cpu_to_le64(first_blkno);
1923                 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1924                 if (ret < 0)
1925                         mlog_errno(ret);
1926 out_commit:
1927                 ocfs2_commit_trans(osb, handle);
1928 out:
1929                 if (meta_ac)
1930                         ocfs2_free_alloc_context(meta_ac);
1931                 if (ret < 0)
1932                         return ret;
1933         } else
1934                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1935
1936         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1937                 /* Set extended attribute into external block */
1938                 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
1939                 if (!ret || ret != -ENOSPC)
1940                         goto end;
1941
1942                 ret = ocfs2_xattr_create_index_block(inode, xs);
1943                 if (ret)
1944                         goto end;
1945         }
1946
1947         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
1948         if (!ret && xblk->xb_attrs.xb_root.xt_list.l_next_free_rec == 0)
1949                 ret = ocfs2_restore_xattr_block(inode, xs);
1950
1951 end:
1952
1953         return ret;
1954 }
1955
1956 /*
1957  * ocfs2_xattr_set()
1958  *
1959  * Set, replace or remove an extended attribute for this inode.
1960  * value is NULL to remove an existing extended attribute, else either
1961  * create or replace an extended attribute.
1962  */
1963 int ocfs2_xattr_set(struct inode *inode,
1964                     int name_index,
1965                     const char *name,
1966                     const void *value,
1967                     size_t value_len,
1968                     int flags)
1969 {
1970         struct buffer_head *di_bh = NULL;
1971         struct ocfs2_dinode *di;
1972         int ret;
1973         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
1974
1975         struct ocfs2_xattr_info xi = {
1976                 .name_index = name_index,
1977                 .name = name,
1978                 .value = value,
1979                 .value_len = value_len,
1980         };
1981
1982         struct ocfs2_xattr_search xis = {
1983                 .not_found = -ENODATA,
1984         };
1985
1986         struct ocfs2_xattr_search xbs = {
1987                 .not_found = -ENODATA,
1988         };
1989
1990         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1991                 return -EOPNOTSUPP;
1992
1993         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1994         if (ret < 0) {
1995                 mlog_errno(ret);
1996                 return ret;
1997         }
1998         xis.inode_bh = xbs.inode_bh = di_bh;
1999         di = (struct ocfs2_dinode *)di_bh->b_data;
2000
2001         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2002         /*
2003          * Scan inode and external block to find the same name
2004          * extended attribute and collect search infomation.
2005          */
2006         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2007         if (ret)
2008                 goto cleanup;
2009         if (xis.not_found) {
2010                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2011                 if (ret)
2012                         goto cleanup;
2013         }
2014
2015         if (xis.not_found && xbs.not_found) {
2016                 ret = -ENODATA;
2017                 if (flags & XATTR_REPLACE)
2018                         goto cleanup;
2019                 ret = 0;
2020                 if (!value)
2021                         goto cleanup;
2022         } else {
2023                 ret = -EEXIST;
2024                 if (flags & XATTR_CREATE)
2025                         goto cleanup;
2026         }
2027
2028         if (!value) {
2029                 /* Remove existing extended attribute */
2030                 if (!xis.not_found)
2031                         ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2032                 else if (!xbs.not_found)
2033                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2034         } else {
2035                 /* We always try to set extended attribute into inode first*/
2036                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2037                 if (!ret && !xbs.not_found) {
2038                         /*
2039                          * If succeed and that extended attribute existing in
2040                          * external block, then we will remove it.
2041                          */
2042                         xi.value = NULL;
2043                         xi.value_len = 0;
2044                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2045                 } else if (ret == -ENOSPC) {
2046                         if (di->i_xattr_loc && !xbs.xattr_bh) {
2047                                 ret = ocfs2_xattr_block_find(inode, name_index,
2048                                                              name, &xbs);
2049                                 if (ret)
2050                                         goto cleanup;
2051                         }
2052                         /*
2053                          * If no space in inode, we will set extended attribute
2054                          * into external block.
2055                          */
2056                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2057                         if (ret)
2058                                 goto cleanup;
2059                         if (!xis.not_found) {
2060                                 /*
2061                                  * If succeed and that extended attribute
2062                                  * existing in inode, we will remove it.
2063                                  */
2064                                 xi.value = NULL;
2065                                 xi.value_len = 0;
2066                                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2067                         }
2068                 }
2069         }
2070 cleanup:
2071         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2072         ocfs2_inode_unlock(inode, 1);
2073         brelse(di_bh);
2074         brelse(xbs.xattr_bh);
2075         for (i = 0; i < blk_per_bucket; i++)
2076                 brelse(xbs.bucket.bhs[i]);
2077
2078         return ret;
2079 }
2080
2081 static inline u32 ocfs2_xattr_hash_by_name(struct inode *inode,
2082                                            int name_index,
2083                                            const char *suffix_name)
2084 {
2085         struct xattr_handler *handler = ocfs2_xattr_handler(name_index);
2086         char *prefix = handler->prefix;
2087         int prefix_len = strlen(handler->prefix);
2088
2089         return ocfs2_xattr_name_hash(inode, prefix, prefix_len,
2090                                      (char *)suffix_name, strlen(suffix_name));
2091 }
2092
2093 /*
2094  * Find the xattr extent rec which may contains name_hash.
2095  * e_cpos will be the first name hash of the xattr rec.
2096  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2097  */
2098 static int ocfs2_xattr_get_rec(struct inode *inode,
2099                                u32 name_hash,
2100                                u64 *p_blkno,
2101                                u32 *e_cpos,
2102                                u32 *num_clusters,
2103                                struct ocfs2_extent_list *el)
2104 {
2105         int ret = 0, i;
2106         struct buffer_head *eb_bh = NULL;
2107         struct ocfs2_extent_block *eb;
2108         struct ocfs2_extent_rec *rec = NULL;
2109         u64 e_blkno = 0;
2110
2111         if (el->l_tree_depth) {
2112                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2113                 if (ret) {
2114                         mlog_errno(ret);
2115                         goto out;
2116                 }
2117
2118                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2119                 el = &eb->h_list;
2120
2121                 if (el->l_tree_depth) {
2122                         ocfs2_error(inode->i_sb,
2123                                     "Inode %lu has non zero tree depth in "
2124                                     "xattr tree block %llu\n", inode->i_ino,
2125                                     (unsigned long long)eb_bh->b_blocknr);
2126                         ret = -EROFS;
2127                         goto out;
2128                 }
2129         }
2130
2131         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2132                 rec = &el->l_recs[i];
2133
2134                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2135                         e_blkno = le64_to_cpu(rec->e_blkno);
2136                         break;
2137                 }
2138         }
2139
2140         if (!e_blkno) {
2141                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2142                             "record (%u, %u, 0) in xattr", inode->i_ino,
2143                             le32_to_cpu(rec->e_cpos),
2144                             ocfs2_rec_clusters(el, rec));
2145                 ret = -EROFS;
2146                 goto out;
2147         }
2148
2149         *p_blkno = le64_to_cpu(rec->e_blkno);
2150         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2151         if (e_cpos)
2152                 *e_cpos = le32_to_cpu(rec->e_cpos);
2153 out:
2154         brelse(eb_bh);
2155         return ret;
2156 }
2157
2158 typedef int (xattr_bucket_func)(struct inode *inode,
2159                                 struct ocfs2_xattr_bucket *bucket,
2160                                 void *para);
2161
2162 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2163                                    struct buffer_head *header_bh,
2164                                    int name_index,
2165                                    const char *name,
2166                                    u32 name_hash,
2167                                    u16 *xe_index,
2168                                    int *found)
2169 {
2170         int i, ret = 0, cmp = 1, block_off, new_offset;
2171         struct ocfs2_xattr_header *xh =
2172                         (struct ocfs2_xattr_header *)header_bh->b_data;
2173         size_t name_len = strlen(name);
2174         struct ocfs2_xattr_entry *xe = NULL;
2175         struct buffer_head *name_bh = NULL;
2176         char *xe_name;
2177
2178         /*
2179          * We don't use binary search in the bucket because there
2180          * may be multiple entries with the same name hash.
2181          */
2182         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2183                 xe = &xh->xh_entries[i];
2184
2185                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2186                         continue;
2187                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2188                         break;
2189
2190                 cmp = name_index - ocfs2_xattr_get_type(xe);
2191                 if (!cmp)
2192                         cmp = name_len - xe->xe_name_len;
2193                 if (cmp)
2194                         continue;
2195
2196                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2197                                                         xh,
2198                                                         i,
2199                                                         &block_off,
2200                                                         &new_offset);
2201                 if (ret) {
2202                         mlog_errno(ret);
2203                         break;
2204                 }
2205
2206                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
2207                                        header_bh->b_blocknr + block_off,
2208                                        &name_bh, OCFS2_BH_CACHED, inode);
2209                 if (ret) {
2210                         mlog_errno(ret);
2211                         break;
2212                 }
2213                 xe_name = name_bh->b_data + new_offset;
2214
2215                 cmp = memcmp(name, xe_name, name_len);
2216                 brelse(name_bh);
2217                 name_bh = NULL;
2218
2219                 if (cmp == 0) {
2220                         *xe_index = i;
2221                         *found = 1;
2222                         ret = 0;
2223                         break;
2224                 }
2225         }
2226
2227         return ret;
2228 }
2229
2230 /*
2231  * Find the specified xattr entry in a series of buckets.
2232  * This series start from p_blkno and last for num_clusters.
2233  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2234  * the num of the valid buckets.
2235  *
2236  * Return the buffer_head this xattr should reside in. And if the xattr's
2237  * hash is in the gap of 2 buckets, return the lower bucket.
2238  */
2239 static int ocfs2_xattr_bucket_find(struct inode *inode,
2240                                    int name_index,
2241                                    const char *name,
2242                                    u32 name_hash,
2243                                    u64 p_blkno,
2244                                    u32 first_hash,
2245                                    u32 num_clusters,
2246                                    struct ocfs2_xattr_search *xs)
2247 {
2248         int ret, found = 0;
2249         struct buffer_head *bh = NULL;
2250         struct buffer_head *lower_bh = NULL;
2251         struct ocfs2_xattr_header *xh = NULL;
2252         struct ocfs2_xattr_entry *xe = NULL;
2253         u16 index = 0;
2254         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2255         int low_bucket = 0, bucket, high_bucket;
2256         u32 last_hash;
2257         u64 blkno;
2258
2259         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno,
2260                                &bh, OCFS2_BH_CACHED, inode);
2261         if (ret) {
2262                 mlog_errno(ret);
2263                 goto out;
2264         }
2265
2266         xh = (struct ocfs2_xattr_header *)bh->b_data;
2267         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2268
2269         while (low_bucket <= high_bucket) {
2270                 brelse(bh);
2271                 bh = NULL;
2272                 bucket = (low_bucket + high_bucket) / 2;
2273
2274                 blkno = p_blkno + bucket * blk_per_bucket;
2275
2276                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
2277                                        &bh, OCFS2_BH_CACHED, inode);
2278                 if (ret) {
2279                         mlog_errno(ret);
2280                         goto out;
2281                 }
2282
2283                 xh = (struct ocfs2_xattr_header *)bh->b_data;
2284                 xe = &xh->xh_entries[0];
2285                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2286                         high_bucket = bucket - 1;
2287                         continue;
2288                 }
2289
2290                 /*
2291                  * Check whether the hash of the last entry in our
2292                  * bucket is larger than the search one.
2293                  */
2294                 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2295                 last_hash = le32_to_cpu(xe->xe_name_hash);
2296
2297                 /* record lower_bh which may be the insert place. */
2298                 brelse(lower_bh);
2299                 lower_bh = bh;
2300                 bh = NULL;
2301
2302                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2303                         low_bucket = bucket + 1;
2304                         continue;
2305                 }
2306
2307                 /* the searched xattr should reside in this bucket if exists. */
2308                 ret = ocfs2_find_xe_in_bucket(inode, lower_bh,
2309                                               name_index, name, name_hash,
2310                                               &index, &found);
2311                 if (ret) {
2312                         mlog_errno(ret);
2313                         goto out;
2314                 }
2315                 break;
2316         }
2317
2318         /*
2319          * Record the bucket we have found.
2320          * When the xattr's hash value is in the gap of 2 buckets, we will
2321          * always set it to the previous bucket.
2322          */
2323         if (!lower_bh) {
2324                 /*
2325                  * We can't find any bucket whose first name_hash is less
2326                  * than the find name_hash.
2327                  */
2328                 BUG_ON(bh->b_blocknr != p_blkno);
2329                 lower_bh = bh;
2330                 bh = NULL;
2331         }
2332         xs->bucket.bhs[0] = lower_bh;
2333         xs->bucket.xh = (struct ocfs2_xattr_header *)
2334                                         xs->bucket.bhs[0]->b_data;
2335         lower_bh = NULL;
2336
2337         xs->header = xs->bucket.xh;
2338         xs->base = xs->bucket.bhs[0]->b_data;
2339         xs->end = xs->base + inode->i_sb->s_blocksize;
2340
2341         if (found) {
2342                 /*
2343                  * If we have found the xattr enty, read all the blocks in
2344                  * this bucket.
2345                  */
2346                 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2347                                         xs->bucket.bhs[0]->b_blocknr + 1,
2348                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2349                                         OCFS2_BH_CACHED, inode);
2350                 if (ret) {
2351                         mlog_errno(ret);
2352                         goto out;
2353                 }
2354
2355                 xs->here = &xs->header->xh_entries[index];
2356                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2357                      (unsigned long long)xs->bucket.bhs[0]->b_blocknr, index);
2358         } else
2359                 ret = -ENODATA;
2360
2361 out:
2362         brelse(bh);
2363         brelse(lower_bh);
2364         return ret;
2365 }
2366
2367 static int ocfs2_xattr_index_block_find(struct inode *inode,
2368                                         struct buffer_head *root_bh,
2369                                         int name_index,
2370                                         const char *name,
2371                                         struct ocfs2_xattr_search *xs)
2372 {
2373         int ret;
2374         struct ocfs2_xattr_block *xb =
2375                         (struct ocfs2_xattr_block *)root_bh->b_data;
2376         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2377         struct ocfs2_extent_list *el = &xb_root->xt_list;
2378         u64 p_blkno = 0;
2379         u32 first_hash, num_clusters = 0;
2380         u32 name_hash = ocfs2_xattr_hash_by_name(inode, name_index, name);
2381
2382         if (le16_to_cpu(el->l_next_free_rec) == 0)
2383                 return -ENODATA;
2384
2385         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2386              name, name_hash, name_index);
2387
2388         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2389                                   &num_clusters, el);
2390         if (ret) {
2391                 mlog_errno(ret);
2392                 goto out;
2393         }
2394
2395         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2396
2397         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2398              "in the rec is %u\n", num_clusters, p_blkno, first_hash);
2399
2400         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2401                                       p_blkno, first_hash, num_clusters, xs);
2402
2403 out:
2404         return ret;
2405 }
2406
2407 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2408                                        u64 blkno,
2409                                        u32 clusters,
2410                                        xattr_bucket_func *func,
2411                                        void *para)
2412 {
2413         int i, j, ret = 0;
2414         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2415         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2416         u32 num_buckets = clusters * bpc;
2417         struct ocfs2_xattr_bucket bucket;
2418
2419         memset(&bucket, 0, sizeof(bucket));
2420
2421         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2422              clusters, blkno);
2423
2424         for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
2425                 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2426                                         blkno, blk_per_bucket,
2427                                         bucket.bhs, OCFS2_BH_CACHED, inode);
2428                 if (ret) {
2429                         mlog_errno(ret);
2430                         goto out;
2431                 }
2432
2433                 bucket.xh = (struct ocfs2_xattr_header *)bucket.bhs[0]->b_data;
2434                 /*
2435                  * The real bucket num in this series of blocks is stored
2436                  * in the 1st bucket.
2437                  */
2438                 if (i == 0)
2439                         num_buckets = le16_to_cpu(bucket.xh->xh_num_buckets);
2440
2441                 mlog(0, "iterating xattr bucket %llu\n", blkno);
2442                 if (func) {
2443                         ret = func(inode, &bucket, para);
2444                         if (ret) {
2445                                 mlog_errno(ret);
2446                                 break;
2447                         }
2448                 }
2449
2450                 for (j = 0; j < blk_per_bucket; j++)
2451                         brelse(bucket.bhs[j]);
2452                 memset(&bucket, 0, sizeof(bucket));
2453         }
2454
2455 out:
2456         for (j = 0; j < blk_per_bucket; j++)
2457                 brelse(bucket.bhs[j]);
2458
2459         return ret;
2460 }
2461
2462 struct ocfs2_xattr_tree_list {
2463         char *buffer;
2464         size_t buffer_size;
2465 };
2466
2467 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2468                                              struct ocfs2_xattr_header *xh,
2469                                              int index,
2470                                              int *block_off,
2471                                              int *new_offset)
2472 {
2473         u16 name_offset;
2474
2475         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2476                 return -EINVAL;
2477
2478         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2479
2480         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2481         *new_offset = name_offset % inode->i_sb->s_blocksize;
2482
2483         return 0;
2484 }
2485
2486 static int ocfs2_list_xattr_bucket(struct inode *inode,
2487                                    struct ocfs2_xattr_bucket *bucket,
2488                                    void *para)
2489 {
2490         int ret = 0;
2491         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2492         size_t size;
2493         int i, block_off, new_offset;
2494
2495         for (i = 0 ; i < le16_to_cpu(bucket->xh->xh_count); i++) {
2496                 struct ocfs2_xattr_entry *entry = &bucket->xh->xh_entries[i];
2497                 struct xattr_handler *handler =
2498                         ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
2499
2500                 if (handler) {
2501                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2502                                                                 bucket->xh,
2503                                                                 i,
2504                                                                 &block_off,
2505                                                                 &new_offset);
2506                         if (ret)
2507                                 break;
2508                         size = handler->list(inode, xl->buffer, xl->buffer_size,
2509                                              bucket->bhs[block_off]->b_data +
2510                                              new_offset,
2511                                              entry->xe_name_len);
2512                         if (xl->buffer) {
2513                                 if (size > xl->buffer_size)
2514                                         return -ERANGE;
2515                                 xl->buffer += size;
2516                         }
2517                         xl->buffer_size -= size;
2518                 }
2519         }
2520
2521         return ret;
2522 }
2523
2524 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2525                                              struct ocfs2_xattr_tree_root *xt,
2526                                              char *buffer,
2527                                              size_t buffer_size)
2528 {
2529         struct ocfs2_extent_list *el = &xt->xt_list;
2530         int ret = 0;
2531         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2532         u64 p_blkno = 0;
2533         struct ocfs2_xattr_tree_list xl = {
2534                 .buffer = buffer,
2535                 .buffer_size = buffer_size,
2536         };
2537
2538         if (le16_to_cpu(el->l_next_free_rec) == 0)
2539                 return 0;
2540
2541         while (name_hash > 0) {
2542                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2543                                           &e_cpos, &num_clusters, el);
2544                 if (ret) {
2545                         mlog_errno(ret);
2546                         goto out;
2547                 }
2548
2549                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2550                                                   ocfs2_list_xattr_bucket,
2551                                                   &xl);
2552                 if (ret) {
2553                         mlog_errno(ret);
2554                         goto out;
2555                 }
2556
2557                 if (e_cpos == 0)
2558                         break;
2559
2560                 name_hash = e_cpos - 1;
2561         }
2562
2563         ret = buffer_size - xl.buffer_size;
2564 out:
2565         return ret;
2566 }
2567
2568 static int cmp_xe(const void *a, const void *b)
2569 {
2570         const struct ocfs2_xattr_entry *l = a, *r = b;
2571         u32 l_hash = le32_to_cpu(l->xe_name_hash);
2572         u32 r_hash = le32_to_cpu(r->xe_name_hash);
2573
2574         if (l_hash > r_hash)
2575                 return 1;
2576         if (l_hash < r_hash)
2577                 return -1;
2578         return 0;
2579 }
2580
2581 static void swap_xe(void *a, void *b, int size)
2582 {
2583         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2584
2585         tmp = *l;
2586         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2587         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2588 }
2589
2590 /*
2591  * When the ocfs2_xattr_block is filled up, new bucket will be created
2592  * and all the xattr entries will be moved to the new bucket.
2593  * Note: we need to sort the entries since they are not saved in order
2594  * in the ocfs2_xattr_block.
2595  */
2596 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2597                                            struct buffer_head *xb_bh,
2598                                            struct buffer_head *xh_bh,
2599                                            struct buffer_head *data_bh)
2600 {
2601         int i, blocksize = inode->i_sb->s_blocksize;
2602         u16 offset, size, off_change;
2603         struct ocfs2_xattr_entry *xe;
2604         struct ocfs2_xattr_block *xb =
2605                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
2606         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2607         struct ocfs2_xattr_header *xh =
2608                                 (struct ocfs2_xattr_header *)xh_bh->b_data;
2609         u16 count = le16_to_cpu(xb_xh->xh_count);
2610         char *target = xh_bh->b_data, *src = xb_bh->b_data;
2611
2612         mlog(0, "cp xattr from block %llu to bucket %llu\n",
2613              (unsigned long long)xb_bh->b_blocknr,
2614              (unsigned long long)xh_bh->b_blocknr);
2615
2616         memset(xh_bh->b_data, 0, blocksize);
2617         if (data_bh)
2618                 memset(data_bh->b_data, 0, blocksize);
2619         /*
2620          * Since the xe_name_offset is based on ocfs2_xattr_header,
2621          * there is a offset change corresponding to the change of
2622          * ocfs2_xattr_header's position.
2623          */
2624         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2625         xe = &xb_xh->xh_entries[count - 1];
2626         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2627         size = blocksize - offset;
2628
2629         /* copy all the names and values. */
2630         if (data_bh)
2631                 target = data_bh->b_data;
2632         memcpy(target + offset, src + offset, size);
2633
2634         /* Init new header now. */
2635         xh->xh_count = xb_xh->xh_count;
2636         xh->xh_num_buckets = cpu_to_le16(1);
2637         xh->xh_name_value_len = cpu_to_le16(size);
2638         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2639
2640         /* copy all the entries. */
2641         target = xh_bh->b_data;
2642         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2643         size = count * sizeof(struct ocfs2_xattr_entry);
2644         memcpy(target + offset, (char *)xb_xh + offset, size);
2645
2646         /* Change the xe offset for all the xe because of the move. */
2647         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2648                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2649         for (i = 0; i < count; i++)
2650                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2651
2652         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2653              offset, size, off_change);
2654
2655         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2656              cmp_xe, swap_xe);
2657 }
2658
2659 /*
2660  * After we move xattr from block to index btree, we have to
2661  * update ocfs2_xattr_search to the new xe and base.
2662  *
2663  * When the entry is in xattr block, xattr_bh indicates the storage place.
2664  * While if the entry is in index b-tree, "bucket" indicates the
2665  * real place of the xattr.
2666  */
2667 static int ocfs2_xattr_update_xattr_search(struct inode *inode,
2668                                            struct ocfs2_xattr_search *xs,
2669                                            struct buffer_head *old_bh,
2670                                            struct buffer_head *new_bh)
2671 {
2672         int ret = 0;
2673         char *buf = old_bh->b_data;
2674         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2675         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2676         int i, blocksize = inode->i_sb->s_blocksize;
2677         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2678
2679         xs->bucket.bhs[0] = new_bh;
2680         get_bh(new_bh);
2681         xs->bucket.xh = (struct ocfs2_xattr_header *)xs->bucket.bhs[0]->b_data;
2682         xs->header = xs->bucket.xh;
2683
2684         xs->base = new_bh->b_data;
2685         xs->end = xs->base + inode->i_sb->s_blocksize;
2686
2687         if (!xs->not_found) {
2688                 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) {
2689                         ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2690                                         xs->bucket.bhs[0]->b_blocknr + 1,
2691                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2692                                         OCFS2_BH_CACHED, inode);
2693                         if (ret) {
2694                                 mlog_errno(ret);
2695                                 return ret;
2696                         }
2697
2698                         i = xs->here - old_xh->xh_entries;
2699                         xs->here = &xs->header->xh_entries[i];
2700                 }
2701         }
2702
2703         return ret;
2704 }
2705
2706 static int ocfs2_xattr_create_index_block(struct inode *inode,
2707                                           struct ocfs2_xattr_search *xs)
2708 {
2709         int ret, credits = OCFS2_SUBALLOC_ALLOC;
2710         u32 bit_off, len;
2711         u64 blkno;
2712         handle_t *handle;
2713         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2714         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2715         struct ocfs2_alloc_context *data_ac;
2716         struct buffer_head *xh_bh = NULL, *data_bh = NULL;
2717         struct buffer_head *xb_bh = xs->xattr_bh;
2718         struct ocfs2_xattr_block *xb =
2719                         (struct ocfs2_xattr_block *)xb_bh->b_data;
2720         struct ocfs2_xattr_tree_root *xr;
2721         u16 xb_flags = le16_to_cpu(xb->xb_flags);
2722         u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2723
2724         mlog(0, "create xattr index block for %llu\n",
2725              (unsigned long long)xb_bh->b_blocknr);
2726
2727         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2728
2729         ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2730         if (ret) {
2731                 mlog_errno(ret);
2732                 goto out;
2733         }
2734
2735         /*
2736          * XXX:
2737          * We can use this lock for now, and maybe move to a dedicated mutex
2738          * if performance becomes a problem later.
2739          */
2740         down_write(&oi->ip_alloc_sem);
2741
2742         /*
2743          * 3 more credits, one for xattr block update, one for the 1st block
2744          * of the new xattr bucket and one for the value/data.
2745          */
2746         credits += 3;
2747         handle = ocfs2_start_trans(osb, credits);
2748         if (IS_ERR(handle)) {
2749                 ret = PTR_ERR(handle);
2750                 mlog_errno(ret);
2751                 goto out_sem;
2752         }
2753
2754         ret = ocfs2_journal_access(handle, inode, xb_bh,
2755                                    OCFS2_JOURNAL_ACCESS_WRITE);
2756         if (ret) {
2757                 mlog_errno(ret);
2758                 goto out_commit;
2759         }
2760
2761         ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2762         if (ret) {
2763                 mlog_errno(ret);
2764                 goto out_commit;
2765         }
2766
2767         /*
2768          * The bucket may spread in many blocks, and
2769          * we will only touch the 1st block and the last block
2770          * in the whole bucket(one for entry and one for data).
2771          */
2772         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2773
2774         mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno);
2775
2776         xh_bh = sb_getblk(inode->i_sb, blkno);
2777         if (!xh_bh) {
2778                 ret = -EIO;
2779                 mlog_errno(ret);
2780                 goto out_commit;
2781         }
2782
2783         ocfs2_set_new_buffer_uptodate(inode, xh_bh);
2784
2785         ret = ocfs2_journal_access(handle, inode, xh_bh,
2786                                    OCFS2_JOURNAL_ACCESS_CREATE);
2787         if (ret) {
2788                 mlog_errno(ret);
2789                 goto out_commit;
2790         }
2791
2792         if (bpb > 1) {
2793                 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1);
2794                 if (!data_bh) {
2795                         ret = -EIO;
2796                         mlog_errno(ret);
2797                         goto out_commit;
2798                 }
2799
2800                 ocfs2_set_new_buffer_uptodate(inode, data_bh);
2801
2802                 ret = ocfs2_journal_access(handle, inode, data_bh,
2803                                            OCFS2_JOURNAL_ACCESS_CREATE);
2804                 if (ret) {
2805                         mlog_errno(ret);
2806                         goto out_commit;
2807                 }
2808         }
2809
2810         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh);
2811
2812         ocfs2_journal_dirty(handle, xh_bh);
2813         if (data_bh)
2814                 ocfs2_journal_dirty(handle, data_bh);
2815
2816         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
2817
2818         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2819         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2820                offsetof(struct ocfs2_xattr_block, xb_attrs));
2821
2822         xr = &xb->xb_attrs.xb_root;
2823         xr->xt_clusters = cpu_to_le32(1);
2824         xr->xt_last_eb_blk = 0;
2825         xr->xt_list.l_tree_depth = 0;
2826         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2827         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2828
2829         xr->xt_list.l_recs[0].e_cpos = 0;
2830         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2831         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2832
2833         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2834
2835         ret = ocfs2_journal_dirty(handle, xb_bh);
2836         if (ret) {
2837                 mlog_errno(ret);
2838                 goto out_commit;
2839         }
2840
2841 out_commit:
2842         ocfs2_commit_trans(osb, handle);
2843
2844 out_sem:
2845         up_write(&oi->ip_alloc_sem);
2846
2847 out:
2848         if (data_ac)
2849                 ocfs2_free_alloc_context(data_ac);
2850
2851         brelse(xh_bh);
2852         brelse(data_bh);
2853
2854         return ret;
2855 }
2856
2857 static int cmp_xe_offset(const void *a, const void *b)
2858 {
2859         const struct ocfs2_xattr_entry *l = a, *r = b;
2860         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2861         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2862
2863         if (l_name_offset < r_name_offset)
2864                 return 1;
2865         if (l_name_offset > r_name_offset)
2866                 return -1;
2867         return 0;
2868 }
2869
2870 /*
2871  * defrag a xattr bucket if we find that the bucket has some
2872  * holes beteen name/value pairs.
2873  * We will move all the name/value pairs to the end of the bucket
2874  * so that we can spare some space for insertion.
2875  */
2876 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2877                                      struct ocfs2_xattr_bucket *bucket)
2878 {
2879         int ret, i;
2880         size_t end, offset, len, value_len;
2881         struct ocfs2_xattr_header *xh;
2882         char *entries, *buf, *bucket_buf = NULL;
2883         u64 blkno = bucket->bhs[0]->b_blocknr;
2884         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2885         u16 xh_free_start;
2886         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2887         size_t blocksize = inode->i_sb->s_blocksize;
2888         handle_t *handle;
2889         struct buffer_head **bhs;
2890         struct ocfs2_xattr_entry *xe;
2891
2892         bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
2893                         GFP_NOFS);
2894         if (!bhs)
2895                 return -ENOMEM;
2896
2897         ret = ocfs2_read_blocks(osb, blkno, blk_per_bucket, bhs,
2898                                 OCFS2_BH_CACHED, inode);
2899         if (ret)
2900                 goto out;
2901
2902         /*
2903          * In order to make the operation more efficient and generic,
2904          * we copy all the blocks into a contiguous memory and do the
2905          * defragment there, so if anything is error, we will not touch
2906          * the real block.
2907          */
2908         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2909         if (!bucket_buf) {
2910                 ret = -EIO;
2911                 goto out;
2912         }
2913
2914         buf = bucket_buf;
2915         for (i = 0; i < blk_per_bucket; i++, buf += blocksize)
2916                 memcpy(buf, bhs[i]->b_data, blocksize);
2917
2918         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket);
2919         if (IS_ERR(handle)) {
2920                 ret = PTR_ERR(handle);
2921                 handle = NULL;
2922                 mlog_errno(ret);
2923                 goto out;
2924         }
2925
2926         for (i = 0; i < blk_per_bucket; i++) {
2927                 ret = ocfs2_journal_access(handle, inode, bhs[i],
2928                                            OCFS2_JOURNAL_ACCESS_WRITE);
2929                 if (ret < 0) {
2930                         mlog_errno(ret);
2931                         goto commit;
2932                 }
2933         }
2934
2935         xh = (struct ocfs2_xattr_header *)bucket_buf;
2936         entries = (char *)xh->xh_entries;
2937         xh_free_start = le16_to_cpu(xh->xh_free_start);
2938
2939         mlog(0, "adjust xattr bucket in %llu, count = %u, "
2940              "xh_free_start = %u, xh_name_value_len = %u.\n",
2941              blkno, le16_to_cpu(xh->xh_count), xh_free_start,
2942              le16_to_cpu(xh->xh_name_value_len));
2943
2944         /*
2945          * sort all the entries by their offset.
2946          * the largest will be the first, so that we can
2947          * move them to the end one by one.
2948          */
2949         sort(entries, le16_to_cpu(xh->xh_count),
2950              sizeof(struct ocfs2_xattr_entry),
2951              cmp_xe_offset, swap_xe);
2952
2953         /* Move all name/values to the end of the bucket. */
2954         xe = xh->xh_entries;
2955         end = OCFS2_XATTR_BUCKET_SIZE;
2956         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
2957                 offset = le16_to_cpu(xe->xe_name_offset);
2958                 if (ocfs2_xattr_is_local(xe))
2959                         value_len = OCFS2_XATTR_SIZE(
2960                                         le64_to_cpu(xe->xe_value_size));
2961                 else
2962                         value_len = OCFS2_XATTR_ROOT_SIZE;
2963                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
2964
2965                 /*
2966                  * We must make sure that the name/value pair
2967                  * exist in the same block. So adjust end to
2968                  * the previous block end if needed.
2969                  */
2970                 if (((end - len) / blocksize !=
2971                         (end - 1) / blocksize))
2972                         end = end - end % blocksize;
2973
2974                 if (end > offset + len) {
2975                         memmove(bucket_buf + end - len,
2976                                 bucket_buf + offset, len);
2977                         xe->xe_name_offset = cpu_to_le16(end - len);
2978                 }
2979
2980                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
2981                                 "bucket %llu\n", (unsigned long long)blkno);
2982
2983                 end -= len;
2984         }
2985
2986         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
2987                         "bucket %llu\n", (unsigned long long)blkno);
2988
2989         if (xh_free_start == end)
2990                 goto commit;
2991
2992         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
2993         xh->xh_free_start = cpu_to_le16(end);
2994
2995         /* sort the entries by their name_hash. */
2996         sort(entries, le16_to_cpu(xh->xh_count),
2997              sizeof(struct ocfs2_xattr_entry),
2998              cmp_xe, swap_xe);
2999
3000         buf = bucket_buf;
3001         for (i = 0; i < blk_per_bucket; i++, buf += blocksize) {
3002                 memcpy(bhs[i]->b_data, buf, blocksize);
3003                 ocfs2_journal_dirty(handle, bhs[i]);
3004         }
3005
3006 commit:
3007         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3008 out:
3009
3010         if (bhs) {
3011                 for (i = 0; i < blk_per_bucket; i++)
3012                         brelse(bhs[i]);
3013         }
3014         kfree(bhs);
3015
3016         kfree(bucket_buf);
3017         return ret;
3018 }
3019
3020 /*
3021  * Move half nums of the xattr bucket in the previous cluster to this new
3022  * cluster. We only touch the last cluster of the previous extend record.
3023  *
3024  * first_bh is the first buffer_head of a series of bucket in the same
3025  * extent rec and header_bh is the header of one bucket in this cluster.
3026  * They will be updated if we move the data header_bh contains to the new
3027  * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3028  */
3029 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3030                                                handle_t *handle,
3031                                                struct buffer_head **first_bh,
3032                                                struct buffer_head **header_bh,
3033                                                u64 new_blkno,
3034                                                u64 prev_blkno,
3035                                                u32 num_clusters,
3036                                                u32 *first_hash)
3037 {
3038         int i, ret, credits;
3039         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3040         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3041         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3042         int blocksize = inode->i_sb->s_blocksize;
3043         struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3044         struct ocfs2_xattr_header *new_xh;
3045         struct ocfs2_xattr_header *xh =
3046                         (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3047
3048         BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3049         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3050
3051         prev_bh = *first_bh;
3052         get_bh(prev_bh);
3053         xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3054
3055         prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3056
3057         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3058              prev_blkno, new_blkno);
3059
3060         /*
3061          * We need to update the 1st half of the new cluster and
3062          * 1 more for the update of the 1st bucket of the previous
3063          * extent record.
3064          */
3065         credits = bpc / 2 + 1;
3066         ret = ocfs2_extend_trans(handle, credits);
3067         if (ret) {
3068                 mlog_errno(ret);
3069                 goto out;
3070         }
3071
3072         ret = ocfs2_journal_access(handle, inode, prev_bh,
3073                                    OCFS2_JOURNAL_ACCESS_WRITE);
3074         if (ret) {
3075                 mlog_errno(ret);
3076                 goto out;
3077         }
3078
3079         for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3080                 old_bh = new_bh = NULL;
3081                 new_bh = sb_getblk(inode->i_sb, new_blkno);
3082                 if (!new_bh) {
3083                         ret = -EIO;
3084                         mlog_errno(ret);
3085                         goto out;
3086                 }
3087
3088                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3089
3090                 ret = ocfs2_journal_access(handle, inode, new_bh,
3091                                            OCFS2_JOURNAL_ACCESS_CREATE);
3092                 if (ret < 0) {
3093                         mlog_errno(ret);
3094                         brelse(new_bh);
3095                         goto out;
3096                 }
3097
3098                 ret = ocfs2_read_block(osb, prev_blkno,
3099                                        &old_bh, OCFS2_BH_CACHED, inode);
3100                 if (ret < 0) {
3101                         mlog_errno(ret);
3102                         brelse(new_bh);
3103                         goto out;
3104                 }
3105
3106                 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3107
3108                 if (i == 0) {
3109                         new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3110                         new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3111
3112                         if (first_hash)
3113                                 *first_hash = le32_to_cpu(
3114                                         new_xh->xh_entries[0].xe_name_hash);
3115                         new_first_bh = new_bh;
3116                         get_bh(new_first_bh);
3117                 }
3118
3119                 ocfs2_journal_dirty(handle, new_bh);
3120
3121                 if (*header_bh == old_bh) {
3122                         brelse(*header_bh);
3123                         *header_bh = new_bh;
3124                         get_bh(*header_bh);
3125
3126                         brelse(*first_bh);
3127                         *first_bh = new_first_bh;
3128                         get_bh(*first_bh);
3129                 }
3130                 brelse(new_bh);
3131                 brelse(old_bh);
3132         }
3133
3134         le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3135
3136         ocfs2_journal_dirty(handle, prev_bh);
3137 out:
3138         brelse(prev_bh);
3139         brelse(new_first_bh);
3140         return ret;
3141 }
3142
3143 static int ocfs2_read_xattr_bucket(struct inode *inode,
3144                                    u64 blkno,
3145                                    struct buffer_head **bhs,
3146                                    int new)
3147 {
3148         int ret = 0;
3149         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3150
3151         if (!new)
3152                 return ocfs2_read_blocks(OCFS2_SB(inode->i_sb), blkno,
3153                                          blk_per_bucket, bhs,
3154                                          OCFS2_BH_CACHED, inode);
3155
3156         for (i = 0; i < blk_per_bucket; i++) {
3157                 bhs[i] = sb_getblk(inode->i_sb, blkno + i);
3158                 if (bhs[i] == NULL) {
3159                         ret = -EIO;
3160                         mlog_errno(ret);
3161                         break;
3162                 }
3163                 ocfs2_set_new_buffer_uptodate(inode, bhs[i]);
3164         }
3165
3166         return ret;
3167 }
3168
3169 /*
3170  * Move half num of the xattrs in old bucket(blk) to new bucket(new_blk).
3171  * first_hash will record the 1st hash of the new bucket.
3172  */
3173 static int ocfs2_half_xattr_bucket(struct inode *inode,
3174                                    handle_t *handle,
3175                                    u64 blk,
3176                                    u64 new_blk,
3177                                    u32 *first_hash,
3178                                    int new_bucket_head)
3179 {
3180         int ret, i;
3181         u16 count, start, len, name_value_len, xe_len, name_offset;
3182         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3183         struct buffer_head **s_bhs, **t_bhs = NULL;
3184         struct ocfs2_xattr_header *xh;
3185         struct ocfs2_xattr_entry *xe;
3186         int blocksize = inode->i_sb->s_blocksize;
3187
3188         mlog(0, "move half of xattrs from bucket %llu to %llu\n",
3189              blk, new_blk);
3190
3191         s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3192         if (!s_bhs)
3193                 return -ENOMEM;
3194
3195         ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0);
3196         if (ret) {
3197                 mlog_errno(ret);
3198                 goto out;
3199         }
3200
3201         ret = ocfs2_journal_access(handle, inode, s_bhs[0],
3202                                    OCFS2_JOURNAL_ACCESS_WRITE);
3203         if (ret) {
3204                 mlog_errno(ret);
3205                 goto out;
3206         }
3207
3208         t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3209         if (!t_bhs) {
3210                 ret = -ENOMEM;
3211                 goto out;
3212         }
3213
3214         ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head);
3215         if (ret) {
3216                 mlog_errno(ret);
3217                 goto out;
3218         }
3219
3220         for (i = 0; i < blk_per_bucket; i++) {
3221                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3222                                            OCFS2_JOURNAL_ACCESS_CREATE);
3223                 if (ret) {
3224                         mlog_errno(ret);
3225                         goto out;
3226                 }
3227         }
3228
3229         /* copy the whole bucket to the new first. */
3230         for (i = 0; i < blk_per_bucket; i++)
3231                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3232
3233         /* update the new bucket. */
3234         xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3235         count = le16_to_cpu(xh->xh_count);
3236         start = count / 2;
3237
3238         /*
3239          * Calculate the total name/value len and xh_free_start for
3240          * the old bucket first.
3241          */
3242         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3243         name_value_len = 0;
3244         for (i = 0; i < start; i++) {
3245                 xe = &xh->xh_entries[i];
3246                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3247                 if (ocfs2_xattr_is_local(xe))
3248                         xe_len +=
3249                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3250                 else
3251                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3252                 name_value_len += xe_len;
3253                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3254                         name_offset = le16_to_cpu(xe->xe_name_offset);
3255         }
3256
3257         /*
3258          * Now begin the modification to the new bucket.
3259          *
3260          * In the new bucket, We just move the xattr entry to the beginning
3261          * and don't touch the name/value. So there will be some holes in the
3262          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3263          * called.
3264          */
3265         xe = &xh->xh_entries[start];
3266         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3267         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3268              (int)((char *)xe - (char *)xh),
3269              (int)((char *)xh->xh_entries - (char *)xh));
3270         memmove((char *)xh->xh_entries, (char *)xe, len);
3271         xe = &xh->xh_entries[count - start];
3272         len = sizeof(struct ocfs2_xattr_entry) * start;
3273         memset((char *)xe, 0, len);
3274
3275         le16_add_cpu(&xh->xh_count, -start);
3276         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3277
3278         /* Calculate xh_free_start for the new bucket. */
3279         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3280         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3281                 xe = &xh->xh_entries[i];
3282                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3283                 if (ocfs2_xattr_is_local(xe))
3284                         xe_len +=
3285                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3286                 else
3287                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3288                 if (le16_to_cpu(xe->xe_name_offset) <
3289                     le16_to_cpu(xh->xh_free_start))
3290                         xh->xh_free_start = xe->xe_name_offset;
3291         }
3292
3293         /* set xh->xh_num_buckets for the new xh. */
3294         if (new_bucket_head)
3295                 xh->xh_num_buckets = cpu_to_le16(1);
3296         else
3297                 xh->xh_num_buckets = 0;
3298
3299         for (i = 0; i < blk_per_bucket; i++) {
3300                 ocfs2_journal_dirty(handle, t_bhs[i]);
3301                 if (ret)
3302                         mlog_errno(ret);
3303         }
3304
3305         /* store the first_hash of the new bucket. */
3306         if (first_hash)
3307                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3308
3309         /*
3310          * Now only update the 1st block of the old bucket.
3311          * Please note that the entry has been sorted already above.
3312          */
3313         xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3314         memset(&xh->xh_entries[start], 0,
3315                sizeof(struct ocfs2_xattr_entry) * (count - start));
3316         xh->xh_count = cpu_to_le16(start);
3317         xh->xh_free_start = cpu_to_le16(name_offset);
3318         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3319
3320         ocfs2_journal_dirty(handle, s_bhs[0]);
3321         if (ret)
3322                 mlog_errno(ret);
3323
3324 out:
3325         if (s_bhs) {
3326                 for (i = 0; i < blk_per_bucket; i++)
3327                         brelse(s_bhs[i]);
3328         }
3329         kfree(s_bhs);
3330
3331         if (t_bhs) {
3332                 for (i = 0; i < blk_per_bucket; i++)
3333                         brelse(t_bhs[i]);
3334         }
3335         kfree(t_bhs);
3336
3337         return ret;
3338 }
3339
3340 /*
3341  * Copy xattr from one bucket to another bucket.
3342  *
3343  * The caller must make sure that the journal transaction
3344  * has enough space for journaling.
3345  */
3346 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3347                                  handle_t *handle,
3348                                  u64 s_blkno,
3349                                  u64 t_blkno,
3350                                  int t_is_new)
3351 {
3352         int ret, i;
3353         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3354         int blocksize = inode->i_sb->s_blocksize;
3355         struct buffer_head **s_bhs, **t_bhs = NULL;
3356
3357         BUG_ON(s_blkno == t_blkno);
3358
3359         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3360              s_blkno, t_blkno, t_is_new);
3361
3362         s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3363                         GFP_NOFS);
3364         if (!s_bhs)
3365                 return -ENOMEM;
3366
3367         ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0);
3368         if (ret)
3369                 goto out;
3370
3371         t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3372                         GFP_NOFS);
3373         if (!t_bhs) {
3374                 ret = -ENOMEM;
3375                 goto out;
3376         }
3377
3378         ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new);
3379         if (ret)
3380                 goto out;
3381
3382         for (i = 0; i < blk_per_bucket; i++) {
3383                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3384                                            OCFS2_JOURNAL_ACCESS_WRITE);
3385                 if (ret)
3386                         goto out;
3387         }
3388
3389         for (i = 0; i < blk_per_bucket; i++) {
3390                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3391                 ocfs2_journal_dirty(handle, t_bhs[i]);
3392         }
3393
3394 out:
3395         if (s_bhs) {
3396                 for (i = 0; i < blk_per_bucket; i++)
3397                         brelse(s_bhs[i]);
3398         }
3399         kfree(s_bhs);
3400
3401         if (t_bhs) {
3402                 for (i = 0; i < blk_per_bucket; i++)
3403                         brelse(t_bhs[i]);
3404         }
3405         kfree(t_bhs);
3406
3407         return ret;
3408 }
3409
3410 /*
3411  * Copy one xattr cluster from src_blk to to_blk.
3412  * The to_blk will become the first bucket header of the cluster, so its
3413  * xh_num_buckets will be initialized as the bucket num in the cluster.
3414  */
3415 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3416                                   handle_t *handle,
3417                                   struct buffer_head *first_bh,
3418                                   u64 src_blk,
3419                                   u64 to_blk,
3420                                   u32 *first_hash)
3421 {
3422         int i, ret, credits;
3423         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3424         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3425         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3426         struct buffer_head *bh = NULL;
3427         struct ocfs2_xattr_header *xh;
3428         u64 to_blk_start = to_blk;
3429
3430         mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk);
3431
3432         /*
3433          * We need to update the new cluster and 1 more for the update of
3434          * the 1st bucket of the previous extent rec.
3435          */
3436         credits = bpc + 1;
3437         ret = ocfs2_extend_trans(handle, credits);
3438         if (ret) {
3439                 mlog_errno(ret);
3440                 goto out;
3441         }
3442
3443         ret = ocfs2_journal_access(handle, inode, first_bh,
3444                                    OCFS2_JOURNAL_ACCESS_WRITE);
3445         if (ret) {
3446                 mlog_errno(ret);
3447                 goto out;
3448         }
3449
3450         for (i = 0; i < num_buckets; i++) {
3451                 ret = ocfs2_cp_xattr_bucket(inode, handle,
3452                                             src_blk, to_blk, 1);
3453                 if (ret) {
3454                         mlog_errno(ret);
3455                         goto out;
3456                 }
3457
3458                 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3459                 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3460         }
3461
3462         /* update the old bucket header. */
3463         xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3464         le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3465
3466         ocfs2_journal_dirty(handle, first_bh);
3467
3468         /* update the new bucket header. */
3469         ret = ocfs2_read_block(osb, to_blk_start, &bh, OCFS2_BH_CACHED, inode);
3470         if (ret < 0) {
3471                 mlog_errno(ret);
3472                 goto out;
3473         }
3474
3475         ret = ocfs2_journal_access(handle, inode, bh,
3476                                    OCFS2_JOURNAL_ACCESS_WRITE);
3477         if (ret) {
3478                 mlog_errno(ret);
3479                 goto out;
3480         }
3481
3482         xh = (struct ocfs2_xattr_header *)bh->b_data;
3483         xh->xh_num_buckets = cpu_to_le16(num_buckets);
3484
3485         ocfs2_journal_dirty(handle, bh);
3486
3487         if (first_hash)
3488                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3489 out:
3490         brelse(bh);
3491         return ret;
3492 }
3493
3494 /*
3495  * Move half of the xattrs in this cluster to the new cluster.
3496  * This function should only be called when bucket size == cluster size.
3497  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3498  */
3499 static int ocfs2_half_xattr_cluster(struct inode *inode,
3500                                     handle_t *handle,
3501                                     u64 prev_blk,
3502                                     u64 new_blk,
3503                                     u32 *first_hash)
3504 {
3505         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3506         int ret, credits = 2 * blk_per_bucket;
3507
3508         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3509
3510         ret = ocfs2_extend_trans(handle, credits);
3511         if (ret) {
3512                 mlog_errno(ret);
3513                 return ret;
3514         }
3515
3516         /* Move half of the xattr in start_blk to the next bucket. */
3517         return  ocfs2_half_xattr_bucket(inode, handle, prev_blk,
3518                                         new_blk, first_hash, 1);
3519 }
3520
3521 /*
3522  * Move some xattrs from the old cluster to the new one since they are not
3523  * contiguous in ocfs2 xattr tree.
3524  *
3525  * new_blk starts a new separate cluster, and we will move some xattrs from
3526  * prev_blk to it. v_start will be set as the first name hash value in this
3527  * new cluster so that it can be used as e_cpos during tree insertion and
3528  * don't collide with our original b-tree operations. first_bh and header_bh
3529  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3530  * to extend the insert bucket.
3531  *
3532  * The problem is how much xattr should we move to the new one and when should
3533  * we update first_bh and header_bh?
3534  * 1. If cluster size > bucket size, that means the previous cluster has more
3535  *    than 1 bucket, so just move half nums of bucket into the new cluster and
3536  *    update the first_bh and header_bh if the insert bucket has been moved
3537  *    to the new cluster.
3538  * 2. If cluster_size == bucket_size:
3539  *    a) If the previous extent rec has more than one cluster and the insert
3540  *       place isn't in the last cluster, copy the entire last cluster to the
3541  *       new one. This time, we don't need to upate the first_bh and header_bh
3542  *       since they will not be moved into the new cluster.
3543  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
3544  *       the new one. And we set the extend flag to zero if the insert place is
3545  *       moved into the new allocated cluster since no extend is needed.
3546  */
3547 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3548                                             handle_t *handle,
3549                                             struct buffer_head **first_bh,
3550                                             struct buffer_head **header_bh,
3551                                             u64 new_blk,
3552                                             u64 prev_blk,
3553                                             u32 prev_clusters,
3554                                             u32 *v_start,
3555                                             int *extend)
3556 {
3557         int ret = 0;
3558         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3559
3560         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3561              prev_blk, prev_clusters, new_blk);
3562
3563         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3564                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3565                                                           handle,
3566                                                           first_bh,
3567                                                           header_bh,
3568                                                           new_blk,
3569                                                           prev_blk,
3570                                                           prev_clusters,
3571                                                           v_start);
3572         else {
3573                 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3574
3575                 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3576                         ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3577                                                      last_blk, new_blk,
3578                                                      v_start);
3579                 else {
3580                         ret = ocfs2_half_xattr_cluster(inode, handle,
3581                                                        last_blk, new_blk,
3582                                                        v_start);
3583
3584                         if ((*header_bh)->b_blocknr == last_blk && extend)
3585                                 *extend = 0;
3586                 }
3587         }
3588
3589         return ret;
3590 }
3591
3592 /*
3593  * Add a new cluster for xattr storage.
3594  *
3595  * If the new cluster is contiguous with the previous one, it will be
3596  * appended to the same extent record, and num_clusters will be updated.
3597  * If not, we will insert a new extent for it and move some xattrs in
3598  * the last cluster into the new allocated one.
3599  * We also need to limit the maximum size of a btree leaf, otherwise we'll
3600  * lose the benefits of hashing because we'll have to search large leaves.
3601  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3602  * if it's bigger).
3603  *
3604  * first_bh is the first block of the previous extent rec and header_bh
3605  * indicates the bucket we will insert the new xattrs. They will be updated
3606  * when the header_bh is moved into the new cluster.
3607  */
3608 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3609                                        struct buffer_head *root_bh,
3610                                        struct buffer_head **first_bh,
3611                                        struct buffer_head **header_bh,
3612                                        u32 *num_clusters,
3613                                        u32 prev_cpos,
3614                                        u64 prev_blkno,
3615                                        int *extend)
3616 {
3617         int ret, credits;
3618         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3619         u32 prev_clusters = *num_clusters;
3620         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3621         u64 block;
3622         handle_t *handle = NULL;
3623         struct ocfs2_alloc_context *data_ac = NULL;
3624         struct ocfs2_alloc_context *meta_ac = NULL;
3625         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3626         struct ocfs2_extent_tree et;
3627
3628         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3629              "previous xattr blkno = %llu\n",
3630              (unsigned long long)OCFS2_I(inode)->ip_blkno,
3631              prev_cpos, prev_blkno);
3632
3633         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
3634
3635         ret = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
3636                                     &data_ac, &meta_ac);
3637         if (ret) {
3638                 mlog_errno(ret);
3639                 goto leave;
3640         }
3641
3642         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
3643                                             clusters_to_add);
3644         handle = ocfs2_start_trans(osb, credits);
3645         if (IS_ERR(handle)) {
3646                 ret = PTR_ERR(handle);
3647                 handle = NULL;
3648                 mlog_errno(ret);
3649                 goto leave;
3650         }
3651
3652         ret = ocfs2_journal_access(handle, inode, root_bh,
3653                                    OCFS2_JOURNAL_ACCESS_WRITE);
3654         if (ret < 0) {
3655                 mlog_errno(ret);
3656                 goto leave;
3657         }
3658
3659         ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3660                                      clusters_to_add, &bit_off, &num_bits);
3661         if (ret < 0) {
3662                 if (ret != -ENOSPC)
3663                         mlog_errno(ret);
3664                 goto leave;
3665         }
3666
3667         BUG_ON(num_bits > clusters_to_add);
3668
3669         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3670         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3671              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3672
3673         if (prev_blkno + prev_clusters * bpc == block &&
3674             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3675              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3676                 /*
3677                  * If this cluster is contiguous with the old one and
3678                  * adding this new cluster, we don't surpass the limit of
3679                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3680                  * initialized and used like other buckets in the previous
3681                  * cluster.
3682                  * So add it as a contiguous one. The caller will handle
3683                  * its init process.
3684                  */
3685                 v_start = prev_cpos + prev_clusters;
3686                 *num_clusters = prev_clusters + num_bits;
3687                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3688                      num_bits);
3689         } else {
3690                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3691                                                        handle,
3692                                                        first_bh,
3693                                                        header_bh,
3694                                                        block,
3695                                                        prev_blkno,
3696                                                        prev_clusters,
3697                                                        &v_start,
3698                                                        extend);
3699                 if (ret) {
3700                         mlog_errno(ret);
3701                         goto leave;
3702                 }
3703         }
3704
3705         if (handle->h_buffer_credits < credits) {
3706                 /*
3707                  * The journal has been restarted before, and don't
3708                  * have enough space for the insertion, so extend it
3709                  * here.
3710                  */
3711                 ret = ocfs2_extend_trans(handle, credits);
3712                 if (ret) {
3713                         mlog_errno(ret);
3714                         goto leave;
3715                 }
3716         }
3717         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3718              num_bits, block, v_start);
3719         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
3720                                   num_bits, 0, meta_ac);
3721         if (ret < 0) {
3722                 mlog_errno(ret);
3723                 goto leave;
3724         }
3725
3726         ret = ocfs2_journal_dirty(handle, root_bh);
3727         if (ret < 0) {
3728                 mlog_errno(ret);
3729                 goto leave;
3730         }
3731
3732 leave:
3733         if (handle)
3734                 ocfs2_commit_trans(osb, handle);
3735         if (data_ac)
3736                 ocfs2_free_alloc_context(data_ac);
3737         if (meta_ac)
3738                 ocfs2_free_alloc_context(meta_ac);
3739
3740         return ret;
3741 }
3742
3743 /*
3744  * Extend a new xattr bucket and move xattrs to the end one by one until
3745  * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3746  */
3747 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3748                                      struct buffer_head *first_bh,
3749                                      struct buffer_head *start_bh,
3750                                      u32 num_clusters)
3751 {
3752         int ret, credits;
3753         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3754         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3755         u64 start_blk = start_bh->b_blocknr, end_blk;
3756         u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3757         handle_t *handle;
3758         struct ocfs2_xattr_header *first_xh =
3759                                 (struct ocfs2_xattr_header *)first_bh->b_data;
3760         u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3761
3762         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3763              "from %llu, len = %u\n", start_blk,
3764              (unsigned long long)first_bh->b_blocknr, num_clusters);
3765
3766         BUG_ON(bucket >= num_buckets);
3767
3768         end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3769
3770         /*
3771          * We will touch all the buckets after the start_bh(include it).
3772          * Add one more bucket and modify the first_bh.
3773          */
3774         credits = end_blk - start_blk + 2 * blk_per_bucket + 1;
3775         handle = ocfs2_start_trans(osb, credits);
3776         if (IS_ERR(handle)) {
3777                 ret = PTR_ERR(handle);
3778                 handle = NULL;
3779                 mlog_errno(ret);
3780                 goto out;
3781         }
3782
3783         ret = ocfs2_journal_access(handle, inode, first_bh,
3784                                    OCFS2_JOURNAL_ACCESS_WRITE);
3785         if (ret) {
3786                 mlog_errno(ret);
3787                 goto commit;
3788         }
3789
3790         while (end_blk != start_blk) {
3791                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3792                                             end_blk + blk_per_bucket, 0);
3793                 if (ret)
3794                         goto commit;
3795                 end_blk -= blk_per_bucket;
3796         }
3797
3798         /* Move half of the xattr in start_blk to the next bucket. */
3799         ret = ocfs2_half_xattr_bucket(inode, handle, start_blk,
3800                                       start_blk + blk_per_bucket, NULL, 0);
3801
3802         le16_add_cpu(&first_xh->xh_num_buckets, 1);
3803         ocfs2_journal_dirty(handle, first_bh);
3804
3805 commit:
3806         ocfs2_commit_trans(osb, handle);
3807 out:
3808         return ret;
3809 }
3810
3811 /*
3812  * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3813  * xb_bh is the ocfs2_xattr_block.
3814  * We will move all the buckets starting from header_bh to the next place. As
3815  * for this one, half num of its xattrs will be moved to the next one.
3816  *
3817  * We will allocate a new cluster if current cluster is full and adjust
3818  * header_bh and first_bh if the insert place is moved to the new cluster.
3819  */
3820 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3821                                       struct buffer_head *xb_bh,
3822                                       struct buffer_head *header_bh)
3823 {
3824         struct ocfs2_xattr_header *first_xh = NULL;
3825         struct buffer_head *first_bh = NULL;
3826         struct ocfs2_xattr_block *xb =
3827                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3828         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3829         struct ocfs2_extent_list *el = &xb_root->xt_list;
3830         struct ocfs2_xattr_header *xh =
3831                         (struct ocfs2_xattr_header *)header_bh->b_data;
3832         u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3833         struct super_block *sb = inode->i_sb;
3834         struct ocfs2_super *osb = OCFS2_SB(sb);
3835         int ret, num_buckets, extend = 1;
3836         u64 p_blkno;
3837         u32 e_cpos, num_clusters;
3838
3839         mlog(0, "Add new xattr bucket starting form %llu\n",
3840              (unsigned long long)header_bh->b_blocknr);
3841
3842         /*
3843          * Add refrence for header_bh here because it may be
3844          * changed in ocfs2_add_new_xattr_cluster and we need
3845          * to free it in the end.
3846          */
3847         get_bh(header_bh);
3848
3849         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3850                                   &num_clusters, el);
3851         if (ret) {
3852                 mlog_errno(ret);
3853                 goto out;
3854         }
3855
3856         ret = ocfs2_read_block(osb, p_blkno,
3857                                &first_bh, OCFS2_BH_CACHED, inode);
3858         if (ret) {
3859                 mlog_errno(ret);
3860                 goto out;
3861         }
3862
3863         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3864         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3865
3866         if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3867                 ret = ocfs2_add_new_xattr_cluster(inode,
3868                                                   xb_bh,
3869                                                   &first_bh,
3870                                                   &header_bh,
3871                                                   &num_clusters,
3872                                                   e_cpos,
3873                                                   p_blkno,
3874                                                   &extend);
3875                 if (ret) {
3876                         mlog_errno(ret);
3877                         goto out;
3878                 }
3879         }
3880
3881         if (extend)
3882                 ret = ocfs2_extend_xattr_bucket(inode,
3883                                                 first_bh,
3884                                                 header_bh,
3885                                                 num_clusters);
3886         if (ret)
3887                 mlog_errno(ret);
3888 out:
3889         brelse(first_bh);
3890         brelse(header_bh);
3891         return ret;
3892 }
3893
3894 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3895                                         struct ocfs2_xattr_bucket *bucket,
3896                                         int offs)
3897 {
3898         int block_off = offs >> inode->i_sb->s_blocksize_bits;
3899
3900         offs = offs % inode->i_sb->s_blocksize;
3901         return bucket->bhs[block_off]->b_data + offs;
3902 }
3903
3904 /*
3905  * Handle the normal xattr set, including replace, delete and new.
3906  * When the bucket is empty, "is_empty" is set and the caller can
3907  * free this bucket.
3908  *
3909  * Note: "local" indicates the real data's locality. So we can't
3910  * just its bucket locality by its length.
3911  */
3912 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3913                                          struct ocfs2_xattr_info *xi,
3914                                          struct ocfs2_xattr_search *xs,
3915                                          u32 name_hash,
3916                                          int local,
3917                                          int *is_empty)
3918 {
3919         struct ocfs2_xattr_entry *last, *xe;
3920         int name_len = strlen(xi->name);
3921         struct ocfs2_xattr_header *xh = xs->header;
3922         u16 count = le16_to_cpu(xh->xh_count), start;
3923         size_t blocksize = inode->i_sb->s_blocksize;
3924         char *val;
3925         size_t offs, size, new_size;
3926
3927         last = &xh->xh_entries[count];
3928         if (!xs->not_found) {
3929                 xe = xs->here;
3930                 offs = le16_to_cpu(xe->xe_name_offset);
3931                 if (ocfs2_xattr_is_local(xe))
3932                         size = OCFS2_XATTR_SIZE(name_len) +
3933                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3934                 else
3935                         size = OCFS2_XATTR_SIZE(name_len) +
3936                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3937
3938                 /*
3939                  * If the new value will be stored outside, xi->value has been
3940                  * initalized as an empty ocfs2_xattr_value_root, and the same
3941                  * goes with xi->value_len, so we can set new_size safely here.
3942                  * See ocfs2_xattr_set_in_bucket.
3943                  */
3944                 new_size = OCFS2_XATTR_SIZE(name_len) +
3945                            OCFS2_XATTR_SIZE(xi->value_len);
3946
3947                 le16_add_cpu(&xh->xh_name_value_len, -size);
3948                 if (xi->value) {
3949                         if (new_size > size)
3950                                 goto set_new_name_value;
3951
3952                         /* Now replace the old value with new one. */
3953                         if (local)
3954                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
3955                         else
3956                                 xe->xe_value_size = 0;
3957
3958                         val = ocfs2_xattr_bucket_get_val(inode,
3959                                                          &xs->bucket, offs);
3960                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
3961                                size - OCFS2_XATTR_SIZE(name_len));
3962                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
3963                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
3964                                        xi->value, xi->value_len);
3965
3966                         le16_add_cpu(&xh->xh_name_value_len, new_size);
3967                         ocfs2_xattr_set_local(xe, local);
3968                         return;
3969                 } else {
3970                         /* Remove the old entry. */
3971                         last -= 1;
3972                         memmove(xe, xe + 1,
3973                                 (void *)last - (void *)xe);
3974                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
3975                         le16_add_cpu(&xh->xh_count, -1);
3976                         if (xh->xh_count == 0 && is_empty)
3977                                 *is_empty = 1;
3978                         return;
3979                 }
3980         } else {
3981                 /* find a new entry for insert. */
3982                 int low = 0, high = count - 1, tmp;
3983                 struct ocfs2_xattr_entry *tmp_xe;
3984
3985                 while (low <= high) {
3986                         tmp = (low + high) / 2;
3987                         tmp_xe = &xh->xh_entries[tmp];
3988
3989                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
3990                                 low = tmp + 1;
3991                         else if (name_hash <
3992                                  le32_to_cpu(tmp_xe->xe_name_hash))
3993                                 high = tmp - 1;
3994                         else
3995                                 break;
3996                 }
3997
3998                 xe = &xh->xh_entries[low];
3999                 if (low != count)
4000                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4001
4002                 le16_add_cpu(&xh->xh_count, 1);
4003                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4004                 xe->xe_name_hash = cpu_to_le32(name_hash);
4005                 xe->xe_name_len = name_len;
4006                 ocfs2_xattr_set_type(xe, xi->name_index);
4007         }
4008
4009 set_new_name_value:
4010         /* Insert the new name+value. */
4011         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4012
4013         /*
4014          * We must make sure that the name/value pair
4015          * exists in the same block.
4016          */
4017         offs = le16_to_cpu(xh->xh_free_start);
4018         start = offs - size;
4019
4020         if (start >> inode->i_sb->s_blocksize_bits !=
4021             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4022                 offs = offs - offs % blocksize;
4023                 xh->xh_free_start = cpu_to_le16(offs);
4024         }
4025
4026         val = ocfs2_xattr_bucket_get_val(inode,
4027                                          &xs->bucket, offs - size);
4028         xe->xe_name_offset = cpu_to_le16(offs - size);
4029
4030         memset(val, 0, size);
4031         memcpy(val, xi->name, name_len);
4032         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4033
4034         xe->xe_value_size = cpu_to_le64(xi->value_len);
4035         ocfs2_xattr_set_local(xe, local);
4036         xs->here = xe;
4037         le16_add_cpu(&xh->xh_free_start, -size);
4038         le16_add_cpu(&xh->xh_name_value_len, size);
4039
4040         return;
4041 }
4042
4043 static int ocfs2_xattr_bucket_handle_journal(struct inode *inode,
4044                                              handle_t *handle,
4045                                              struct ocfs2_xattr_search *xs,
4046                                              struct buffer_head **bhs,
4047                                              u16 bh_num)
4048 {
4049         int ret = 0, off, block_off;
4050         struct ocfs2_xattr_entry *xe = xs->here;
4051
4052         /*
4053          * First calculate all the blocks we should journal_access
4054          * and journal_dirty. The first block should always be touched.
4055          */
4056         ret = ocfs2_journal_dirty(handle, bhs[0]);
4057         if (ret)
4058                 mlog_errno(ret);
4059
4060         /* calc the data. */
4061         off = le16_to_cpu(xe->xe_name_offset);
4062         block_off = off >> inode->i_sb->s_blocksize_bits;
4063         ret = ocfs2_journal_dirty(handle, bhs[block_off]);
4064         if (ret)
4065                 mlog_errno(ret);
4066
4067         return ret;
4068 }
4069
4070 /*
4071  * Set the xattr entry in the specified bucket.
4072  * The bucket is indicated by xs->bucket and it should have the enough
4073  * space for the xattr insertion.
4074  */
4075 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4076                                            struct ocfs2_xattr_info *xi,
4077                                            struct ocfs2_xattr_search *xs,
4078                                            u32 name_hash,
4079                                            int local,
4080                                            int *bucket_empty)
4081 {
4082         int i, ret;
4083         handle_t *handle = NULL;
4084         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4085         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4086
4087         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4088              (unsigned long)xi->value_len, xi->name_index,
4089              (unsigned long long)xs->bucket.bhs[0]->b_blocknr);
4090
4091         if (!xs->bucket.bhs[1]) {
4092                 ret = ocfs2_read_blocks(osb,
4093                                         xs->bucket.bhs[0]->b_blocknr + 1,
4094                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
4095                                         OCFS2_BH_CACHED, inode);
4096                 if (ret) {
4097                         mlog_errno(ret);
4098                         goto out;
4099                 }
4100         }
4101
4102         handle = ocfs2_start_trans(osb, blk_per_bucket);
4103         if (IS_ERR(handle)) {
4104                 ret = PTR_ERR(handle);
4105                 handle = NULL;
4106                 mlog_errno(ret);
4107                 goto out;
4108         }
4109
4110         for (i = 0; i < blk_per_bucket; i++) {
4111                 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[i],
4112                                            OCFS2_JOURNAL_ACCESS_WRITE);
4113                 if (ret < 0) {
4114                         mlog_errno(ret);
4115                         goto out;
4116                 }
4117         }
4118
4119         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash,
4120                                      local, bucket_empty);
4121
4122         /*Only dirty the blocks we have touched in set xattr. */
4123         ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs,
4124                                                 xs->bucket.bhs, blk_per_bucket);
4125         if (ret)
4126                 mlog_errno(ret);
4127 out:
4128         ocfs2_commit_trans(osb, handle);
4129
4130         return ret;
4131 }
4132
4133 static int ocfs2_xattr_value_update_size(struct inode *inode,
4134                                          struct buffer_head *xe_bh,
4135                                          struct ocfs2_xattr_entry *xe,
4136                                          u64 new_size)
4137 {
4138         int ret;
4139         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4140         handle_t *handle = NULL;
4141
4142         handle = ocfs2_start_trans(osb, 1);
4143         if (handle == NULL) {
4144                 ret = -ENOMEM;
4145                 mlog_errno(ret);
4146                 goto out;
4147         }
4148
4149         ret = ocfs2_journal_access(handle, inode, xe_bh,
4150                                    OCFS2_JOURNAL_ACCESS_WRITE);
4151         if (ret < 0) {
4152                 mlog_errno(ret);
4153                 goto out_commit;
4154         }
4155
4156         xe->xe_value_size = cpu_to_le64(new_size);
4157
4158         ret = ocfs2_journal_dirty(handle, xe_bh);
4159         if (ret < 0)
4160                 mlog_errno(ret);
4161
4162 out_commit:
4163         ocfs2_commit_trans(osb, handle);
4164 out:
4165         return ret;
4166 }
4167
4168 /*
4169  * Truncate the specified xe_off entry in xattr bucket.
4170  * bucket is indicated by header_bh and len is the new length.
4171  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4172  *
4173  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4174  */
4175 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4176                                              struct buffer_head *header_bh,
4177                                              int xe_off,
4178                                              int len)
4179 {
4180         int ret, offset;
4181         u64 value_blk;
4182         struct buffer_head *value_bh = NULL;
4183         struct ocfs2_xattr_value_root *xv;
4184         struct ocfs2_xattr_entry *xe;
4185         struct ocfs2_xattr_header *xh =
4186                         (struct ocfs2_xattr_header *)header_bh->b_data;
4187         size_t blocksize = inode->i_sb->s_blocksize;
4188
4189         xe = &xh->xh_entries[xe_off];
4190
4191         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4192
4193         offset = le16_to_cpu(xe->xe_name_offset) +
4194                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4195
4196         value_blk = offset / blocksize;
4197
4198         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4199         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4200         value_blk += header_bh->b_blocknr;
4201
4202         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), value_blk,
4203                                &value_bh, OCFS2_BH_CACHED, inode);
4204         if (ret) {
4205                 mlog_errno(ret);
4206                 goto out;
4207         }
4208
4209         xv = (struct ocfs2_xattr_value_root *)
4210                 (value_bh->b_data + offset % blocksize);
4211
4212         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4213              xe_off, (unsigned long long)header_bh->b_blocknr, len);
4214         ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4215         if (ret) {
4216                 mlog_errno(ret);
4217                 goto out;
4218         }
4219
4220         ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4221         if (ret) {
4222                 mlog_errno(ret);
4223                 goto out;
4224         }
4225
4226 out:
4227         brelse(value_bh);
4228         return ret;
4229 }
4230
4231 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4232                                                 struct ocfs2_xattr_search *xs,
4233                                                 int len)
4234 {
4235         int ret, offset;
4236         struct ocfs2_xattr_entry *xe = xs->here;
4237         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4238
4239         BUG_ON(!xs->bucket.bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4240
4241         offset = xe - xh->xh_entries;
4242         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bhs[0],
4243                                                 offset, len);
4244         if (ret)
4245                 mlog_errno(ret);
4246
4247         return ret;
4248 }
4249
4250 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4251                                                 struct ocfs2_xattr_search *xs,
4252                                                 char *val,
4253                                                 int value_len)
4254 {
4255         int offset;
4256         struct ocfs2_xattr_value_root *xv;
4257         struct ocfs2_xattr_entry *xe = xs->here;
4258
4259         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4260
4261         offset = le16_to_cpu(xe->xe_name_offset) +
4262                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4263
4264         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4265
4266         return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4267 }
4268
4269 /*
4270  * Remove the xattr bucket pointed by bucket_bh.
4271  * All the buckets after it in the same xattr extent rec will be
4272  * move forward one by one.
4273  */
4274 static int ocfs2_rm_xattr_bucket(struct inode *inode,
4275                                  struct buffer_head *first_bh,
4276                                  struct ocfs2_xattr_bucket *bucket)
4277 {
4278         int ret = 0, credits;
4279         struct ocfs2_xattr_header *xh =
4280                                 (struct ocfs2_xattr_header *)first_bh->b_data;
4281         u16 bucket_num = le16_to_cpu(xh->xh_num_buckets);
4282         u64 end, start = bucket->bhs[0]->b_blocknr;
4283         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4284         handle_t *handle;
4285         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4286
4287         end = first_bh->b_blocknr + (bucket_num - 1) * blk_per_bucket;
4288
4289         mlog(0, "rm xattr bucket %llu\n", start);
4290         /*
4291          * We need to update the first xattr_header and all the buckets starting
4292          * from start in this xattr rec.
4293          *
4294          * XXX: Should we empty the old last bucket here?
4295          */
4296         credits = 1 + end - start;
4297         handle = ocfs2_start_trans(osb, credits);
4298         if (IS_ERR(handle)) {
4299                 ret = PTR_ERR(handle);
4300                 mlog_errno(ret);
4301                 return ret;
4302         }
4303
4304         ret = ocfs2_journal_access(handle, inode, first_bh,
4305                                    OCFS2_JOURNAL_ACCESS_WRITE);
4306         if (ret) {
4307                 mlog_errno(ret);
4308                 goto out_commit;
4309         }
4310
4311
4312         while (start < end) {
4313                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4314                                             start + blk_per_bucket,
4315                                             start, 0);
4316                 if (ret) {
4317                         mlog_errno(ret);
4318                         goto out_commit;
4319                 }
4320                 start += blk_per_bucket;
4321         }
4322
4323         /* update the first_bh. */
4324         xh->xh_num_buckets = cpu_to_le16(bucket_num - 1);
4325         ocfs2_journal_dirty(handle, first_bh);
4326
4327 out_commit:
4328         ocfs2_commit_trans(osb, handle);
4329         return ret;
4330 }
4331
4332 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4333                                   struct buffer_head *root_bh,
4334                                   u64 blkno,
4335                                   u32 cpos,
4336                                   u32 len)
4337 {
4338         int ret;
4339         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4340         struct inode *tl_inode = osb->osb_tl_inode;
4341         handle_t *handle;
4342         struct ocfs2_xattr_block *xb =
4343                         (struct ocfs2_xattr_block *)root_bh->b_data;
4344         struct ocfs2_alloc_context *meta_ac = NULL;
4345         struct ocfs2_cached_dealloc_ctxt dealloc;
4346         struct ocfs2_extent_tree et;
4347
4348         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4349
4350         ocfs2_init_dealloc_ctxt(&dealloc);
4351
4352         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4353              cpos, len, (unsigned long long)blkno);
4354
4355         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4356
4357         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4358         if (ret) {
4359                 mlog_errno(ret);
4360                 return ret;
4361         }
4362
4363         mutex_lock(&tl_inode->i_mutex);
4364
4365         if (ocfs2_truncate_log_needs_flush(osb)) {
4366                 ret = __ocfs2_flush_truncate_log(osb);
4367                 if (ret < 0) {
4368                         mlog_errno(ret);
4369                         goto out;
4370                 }
4371         }
4372
4373         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4374         if (handle == NULL) {
4375                 ret = -ENOMEM;
4376                 mlog_errno(ret);
4377                 goto out;
4378         }
4379
4380         ret = ocfs2_journal_access(handle, inode, root_bh,
4381                                    OCFS2_JOURNAL_ACCESS_WRITE);
4382         if (ret) {
4383                 mlog_errno(ret);
4384                 goto out_commit;
4385         }
4386
4387         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4388                                   &dealloc);
4389         if (ret) {
4390                 mlog_errno(ret);
4391                 goto out_commit;
4392         }
4393
4394         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4395
4396         ret = ocfs2_journal_dirty(handle, root_bh);
4397         if (ret) {
4398                 mlog_errno(ret);
4399                 goto out_commit;
4400         }
4401
4402         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4403         if (ret)
4404                 mlog_errno(ret);
4405
4406 out_commit:
4407         ocfs2_commit_trans(osb, handle);
4408 out:
4409         ocfs2_schedule_truncate_log_flush(osb, 1);
4410
4411         mutex_unlock(&tl_inode->i_mutex);
4412
4413         if (meta_ac)
4414                 ocfs2_free_alloc_context(meta_ac);
4415
4416         ocfs2_run_deallocs(osb, &dealloc);
4417
4418         return ret;
4419 }
4420
4421 /*
4422  * Free the xattr bucket indicated by xs->bucket and if all the buckets
4423  * in the clusters is free, free the clusters also.
4424  */
4425 static int ocfs2_xattr_bucket_shrink(struct inode *inode,
4426                                      struct ocfs2_xattr_info *xi,
4427                                      struct ocfs2_xattr_search *xs,
4428                                      u32 name_hash)
4429 {
4430         int ret;
4431         u32 e_cpos, num_clusters;
4432         u64 p_blkno;
4433         struct buffer_head *first_bh = NULL;
4434         struct ocfs2_xattr_header *first_xh;
4435         struct ocfs2_xattr_block *xb =
4436                         (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
4437
4438         BUG_ON(xs->header->xh_count != 0);
4439
4440         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4441                                   &e_cpos, &num_clusters,
4442                                   &xb->xb_attrs.xb_root.xt_list);
4443         if (ret) {
4444                 mlog_errno(ret);
4445                 return ret;
4446         }
4447
4448         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno,
4449                                &first_bh, OCFS2_BH_CACHED, inode);
4450         if (ret) {
4451                 mlog_errno(ret);
4452                 return ret;
4453         }
4454
4455         ret = ocfs2_rm_xattr_bucket(inode, first_bh, &xs->bucket);
4456         if (ret) {
4457                 mlog_errno(ret);
4458                 goto out;
4459         }
4460
4461         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
4462         if (first_xh->xh_num_buckets == 0)
4463                 ret = ocfs2_rm_xattr_cluster(inode, xs->xattr_bh,
4464                                              p_blkno, e_cpos,
4465                                              num_clusters);
4466
4467 out:
4468         brelse(first_bh);
4469         return ret;
4470 }
4471
4472 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4473                                          struct ocfs2_xattr_search *xs)
4474 {
4475         handle_t *handle = NULL;
4476         struct ocfs2_xattr_header *xh = xs->bucket.xh;
4477         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4478                                                 le16_to_cpu(xh->xh_count) - 1];
4479         int ret = 0;
4480
4481         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1);
4482         if (IS_ERR(handle)) {
4483                 ret = PTR_ERR(handle);
4484                 mlog_errno(ret);
4485                 return;
4486         }
4487
4488         ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[0],
4489                                    OCFS2_JOURNAL_ACCESS_WRITE);
4490         if (ret) {
4491                 mlog_errno(ret);
4492                 goto out_commit;
4493         }
4494
4495         /* Remove the old entry. */
4496         memmove(xs->here, xs->here + 1,
4497                 (void *)last - (void *)xs->here);
4498         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4499         le16_add_cpu(&xh->xh_count, -1);
4500
4501         ret = ocfs2_journal_dirty(handle, xs->bucket.bhs[0]);
4502         if (ret < 0)
4503                 mlog_errno(ret);
4504 out_commit:
4505         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4506 }
4507
4508 /*
4509  * Set the xattr name/value in the bucket specified in xs.
4510  *
4511  * As the new value in xi may be stored in the bucket or in an outside cluster,
4512  * we divide the whole process into 3 steps:
4513  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4514  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4515  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4516  * 4. If the clusters for the new outside value can't be allocated, we need
4517  *    to free the xattr we allocated in set.
4518  */
4519 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4520                                      struct ocfs2_xattr_info *xi,
4521                                      struct ocfs2_xattr_search *xs)
4522 {
4523         int ret, local = 1, bucket_empty = 0;
4524         size_t value_len;
4525         char *val = (char *)xi->value;
4526         struct ocfs2_xattr_entry *xe = xs->here;
4527         u32 name_hash = ocfs2_xattr_hash_by_name(inode,
4528                                                  xi->name_index, xi->name);
4529
4530         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4531                 /*
4532                  * We need to truncate the xattr storage first.
4533                  *
4534                  * If both the old and new value are stored to
4535                  * outside block, we only need to truncate
4536                  * the storage and then set the value outside.
4537                  *
4538                  * If the new value should be stored within block,
4539                  * we should free all the outside block first and
4540                  * the modification to the xattr block will be done
4541                  * by following steps.
4542                  */
4543                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4544                         value_len = xi->value_len;
4545                 else
4546                         value_len = 0;
4547
4548                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4549                                                            value_len);
4550                 if (ret)
4551                         goto out;
4552
4553                 if (value_len)
4554                         goto set_value_outside;
4555         }
4556
4557         value_len = xi->value_len;
4558         /* So we have to handle the inside block change now. */
4559         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4560                 /*
4561                  * If the new value will be stored outside of block,
4562                  * initalize a new empty value root and insert it first.
4563                  */
4564                 local = 0;
4565                 xi->value = &def_xv;
4566                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4567         }
4568
4569         ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash,
4570                                               local, &bucket_empty);
4571         if (ret) {
4572                 mlog_errno(ret);
4573                 goto out;
4574         }
4575
4576         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4577                 /* allocate the space now for the outside block storage. */
4578                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4579                                                            value_len);
4580                 if (ret) {
4581                         mlog_errno(ret);
4582
4583                         if (xs->not_found) {
4584                                 /*
4585                                  * We can't allocate enough clusters for outside
4586                                  * storage and we have allocated xattr already,
4587                                  * so need to remove it.
4588                                  */
4589                                 ocfs2_xattr_bucket_remove_xs(inode, xs);
4590                         }
4591                         goto out;
4592                 }
4593         } else {
4594                 if (bucket_empty)
4595                         ret = ocfs2_xattr_bucket_shrink(inode, xi,
4596                                                         xs, name_hash);
4597                 goto out;
4598         }
4599
4600 set_value_outside:
4601         ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4602 out:
4603         return ret;
4604 }
4605
4606 /* check whether the xattr bucket is filled up with the same hash value. */
4607 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4608                                               struct ocfs2_xattr_bucket *bucket)
4609 {
4610         struct ocfs2_xattr_header *xh = bucket->xh;
4611
4612         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4613             xh->xh_entries[0].xe_name_hash) {
4614                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4615                      "hash = %u\n",
4616                      (unsigned long long)bucket->bhs[0]->b_blocknr,
4617                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4618                 return -ENOSPC;
4619         }
4620
4621         return 0;
4622 }
4623
4624 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4625                                              struct ocfs2_xattr_info *xi,
4626                                              struct ocfs2_xattr_search *xs)
4627 {
4628         struct ocfs2_xattr_header *xh;
4629         struct ocfs2_xattr_entry *xe;
4630         u16 count, header_size, xh_free_start;
4631         int i, free, max_free, need, old;
4632         size_t value_size = 0, name_len = strlen(xi->name);
4633         size_t blocksize = inode->i_sb->s_blocksize;
4634         int ret, allocation = 0;
4635         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4636
4637         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4638
4639 try_again:
4640         xh = xs->header;
4641         count = le16_to_cpu(xh->xh_count);
4642         xh_free_start = le16_to_cpu(xh->xh_free_start);
4643         header_size = sizeof(struct ocfs2_xattr_header) +
4644                         count * sizeof(struct ocfs2_xattr_entry);
4645         max_free = OCFS2_XATTR_BUCKET_SIZE -
4646                 le16_to_cpu(xh->xh_name_value_len) - header_size;
4647
4648         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4649                         "of %u which exceed block size\n",
4650                         (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4651                         header_size);
4652
4653         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4654                 value_size = OCFS2_XATTR_ROOT_SIZE;
4655         else if (xi->value)
4656                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4657
4658         if (xs->not_found)
4659                 need = sizeof(struct ocfs2_xattr_entry) +
4660                         OCFS2_XATTR_SIZE(name_len) + value_size;
4661         else {
4662                 need = value_size + OCFS2_XATTR_SIZE(name_len);
4663
4664                 /*
4665                  * We only replace the old value if the new length is smaller
4666                  * than the old one. Otherwise we will allocate new space in the
4667                  * bucket to store it.
4668                  */
4669                 xe = xs->here;
4670                 if (ocfs2_xattr_is_local(xe))
4671                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4672                 else
4673                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4674
4675                 if (old >= value_size)
4676                         need = 0;
4677         }
4678
4679         free = xh_free_start - header_size;
4680         /*
4681          * We need to make sure the new name/value pair
4682          * can exist in the same block.
4683          */
4684         if (xh_free_start % blocksize < need)
4685                 free -= xh_free_start % blocksize;
4686
4687         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4688              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4689              " %u\n", xs->not_found,
4690              (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4691              free, need, max_free, le16_to_cpu(xh->xh_free_start),
4692              le16_to_cpu(xh->xh_name_value_len));
4693
4694         if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4695                 if (need <= max_free &&
4696                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4697                         /*
4698                          * We can create the space by defragment. Since only the
4699                          * name/value will be moved, the xe shouldn't be changed
4700                          * in xs.
4701                          */
4702                         ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket);
4703                         if (ret) {
4704                                 mlog_errno(ret);
4705                                 goto out;
4706                         }
4707
4708                         xh_free_start = le16_to_cpu(xh->xh_free_start);
4709                         free = xh_free_start - header_size;
4710                         if (xh_free_start % blocksize < need)
4711                                 free -= xh_free_start % blocksize;
4712
4713                         if (free >= need)
4714                                 goto xattr_set;
4715
4716                         mlog(0, "Can't get enough space for xattr insert by "
4717                              "defragment. Need %u bytes, but we have %d, so "
4718                              "allocate new bucket for it.\n", need, free);
4719                 }
4720
4721                 /*
4722                  * We have to add new buckets or clusters and one
4723                  * allocation should leave us enough space for insert.
4724                  */
4725                 BUG_ON(allocation);
4726
4727                 /*
4728                  * We do not allow for overlapping ranges between buckets. And
4729                  * the maximum number of collisions we will allow for then is
4730                  * one bucket's worth, so check it here whether we need to
4731                  * add a new bucket for the insert.
4732                  */
4733                 ret = ocfs2_check_xattr_bucket_collision(inode, &xs->bucket);
4734                 if (ret) {
4735                         mlog_errno(ret);
4736                         goto out;
4737                 }
4738
4739                 ret = ocfs2_add_new_xattr_bucket(inode,
4740                                                  xs->xattr_bh,
4741                                                  xs->bucket.bhs[0]);
4742                 if (ret) {
4743                         mlog_errno(ret);
4744                         goto out;
4745                 }
4746
4747                 for (i = 0; i < blk_per_bucket; i++)
4748                         brelse(xs->bucket.bhs[i]);
4749
4750                 memset(&xs->bucket, 0, sizeof(xs->bucket));
4751
4752                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4753                                                    xi->name_index,
4754                                                    xi->name, xs);
4755                 if (ret && ret != -ENODATA)
4756                         goto out;
4757                 xs->not_found = ret;
4758                 allocation = 1;
4759                 goto try_again;
4760         }
4761
4762 xattr_set:
4763         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4764 out:
4765         mlog_exit(ret);
4766         return ret;
4767 }
4768
4769 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4770                                         struct ocfs2_xattr_bucket *bucket,
4771                                         void *para)
4772 {
4773         int ret = 0;
4774         struct ocfs2_xattr_header *xh = bucket->xh;
4775         u16 i;
4776         struct ocfs2_xattr_entry *xe;
4777
4778         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4779                 xe = &xh->xh_entries[i];
4780                 if (ocfs2_xattr_is_local(xe))
4781                         continue;
4782
4783                 ret = ocfs2_xattr_bucket_value_truncate(inode,
4784                                                         bucket->bhs[0],
4785                                                         i, 0);
4786                 if (ret) {
4787                         mlog_errno(ret);
4788                         break;
4789                 }
4790         }
4791
4792         return ret;
4793 }
4794
4795 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4796                                           struct buffer_head *xb_bh)
4797 {
4798         struct ocfs2_xattr_block *xb =
4799                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4800         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4801         int ret = 0;
4802         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4803         u64 p_blkno;
4804
4805         if (le16_to_cpu(el->l_next_free_rec) == 0)
4806                 return 0;
4807
4808         while (name_hash > 0) {
4809                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4810                                           &e_cpos, &num_clusters, el);
4811                 if (ret) {
4812                         mlog_errno(ret);
4813                         goto out;
4814                 }
4815
4816                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4817                                                   ocfs2_delete_xattr_in_bucket,
4818                                                   NULL);
4819                 if (ret) {
4820                         mlog_errno(ret);
4821                         goto out;
4822                 }
4823
4824                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4825                                              p_blkno, e_cpos, num_clusters);
4826                 if (ret) {
4827                         mlog_errno(ret);
4828                         break;
4829                 }
4830
4831                 if (e_cpos == 0)
4832                         break;
4833
4834                 name_hash = e_cpos - 1;
4835         }
4836
4837 out:
4838         return ret;
4839 }