SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/ocfs2/xattr.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * xattr.c
   5  *
   6  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
   7  *
   8  * CREDITS:
   9  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
  10  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  11  *
  12  * This program is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU General Public
  14  * License version 2 as published by the Free Software Foundation.
  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
  22 #include <linux/capability.h>
  23 #include <linux/fs.h>
  24 #include <linux/types.h>
  25 #include <linux/slab.h>
  26 #include <linux/highmem.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/uio.h>
  29 #include <linux/sched.h>
  30 #include <linux/splice.h>
  31 #include <linux/mount.h>
  32 #include <linux/writeback.h>
  33 #include <linux/falloc.h>
  34 #include <linux/sort.h>
  35 #include <linux/init.h>
  36 #include <linux/module.h>
  37 #include <linux/string.h>
  38 #include <linux/security.h>
  39
  40 #define MLOG_MASK_PREFIX ML_XATTR
  41 #include <cluster/masklog.h>
  42
  43 #include "ocfs2.h"
  44 #include "alloc.h"
  45 #include "blockcheck.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         /* The inode these xattrs are associated with */
  67         struct inode *bu_inode;
  68
  69         /* The actual buffers that make up the bucket */
  70         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  71
  72         /* How many blocks make up one bucket for this filesystem */
  73         int bu_blocks;
  74 };
  75
  76 struct ocfs2_xattr_set_ctxt {
  77         handle_t *handle;
  78         struct ocfs2_alloc_context *meta_ac;
  79         struct ocfs2_alloc_context *data_ac;
  80         struct ocfs2_cached_dealloc_ctxt dealloc;
  81 };
  82
  83 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  84 #define OCFS2_XATTR_INLINE_SIZE 80
  85 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
  86                                          - sizeof(struct ocfs2_xattr_header) \
  87                                          - sizeof(__u32))
  88 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
  89                                          - sizeof(struct ocfs2_xattr_block) \
  90                                          - sizeof(struct ocfs2_xattr_header) \
  91                                          - sizeof(__u32))
  92
  93 static struct ocfs2_xattr_def_value_root def_xv = {
  94         .xv.xr_list.l_count = cpu_to_le16(1),
  95 };
  96
  97 struct xattr_handler *ocfs2_xattr_handlers[] = {
  98         &ocfs2_xattr_user_handler,
  99 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
 100         &ocfs2_xattr_acl_access_handler,
 101         &ocfs2_xattr_acl_default_handler,
 102 #endif
 103         &ocfs2_xattr_trusted_handler,
 104         &ocfs2_xattr_security_handler,
 105         NULL
 106 };
 107
 108 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 109         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
 110 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
 111         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
 112                                         = &ocfs2_xattr_acl_access_handler,
 113         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
 114                                         = &ocfs2_xattr_acl_default_handler,
 115 #endif
 116         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
 117         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 118 };
 119
 120 struct ocfs2_xattr_info {
 121         int name_index;
 122         const char *name;
 123         const void *value;
 124         size_t value_len;
 125 };
 126
 127 struct ocfs2_xattr_search {
 128         struct buffer_head *inode_bh;
 129         /*
 130          * xattr_bh point to the block buffer head which has extended attribute
 131          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 132          */
 133         struct buffer_head *xattr_bh;
 134         struct ocfs2_xattr_header *header;
 135         struct ocfs2_xattr_bucket *bucket;
 136         void *base;
 137         void *end;
 138         struct ocfs2_xattr_entry *here;
 139         int not_found;
 140 };
 141
 142 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
 143                                              struct ocfs2_xattr_header *xh,
 144                                              int index,
 145                                              int *block_off,
 146                                              int *new_offset);
 147
 148 static int ocfs2_xattr_block_find(struct inode *inode,
 149                                   int name_index,
 150                                   const char *name,
 151                                   struct ocfs2_xattr_search *xs);
 152 static int ocfs2_xattr_index_block_find(struct inode *inode,
 153                                         struct buffer_head *root_bh,
 154                                         int name_index,
 155                                         const char *name,
 156                                         struct ocfs2_xattr_search *xs);
 157
 158 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 159                                         struct ocfs2_xattr_tree_root *xt,
 160                                         char *buffer,
 161                                         size_t buffer_size);
 162
 163 static int ocfs2_xattr_create_index_block(struct inode *inode,
 164                                           struct ocfs2_xattr_search *xs,
 165                                           struct ocfs2_xattr_set_ctxt *ctxt);
 166
 167 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 168                                              struct ocfs2_xattr_info *xi,
 169                                              struct ocfs2_xattr_search *xs,
 170                                              struct ocfs2_xattr_set_ctxt *ctxt);
 171
 172 static int ocfs2_delete_xattr_index_block(struct inode *inode,
 173                                           struct buffer_head *xb_bh);
 174 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 175                                   u64 src_blk, u64 last_blk, u64 to_blk,
 176                                   unsigned int start_bucket,
 177                                   u32 *first_hash);
 178
 179 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 180 {
 181         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 182 }
 183
 184 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 185 {
 186         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 187 }
 188
 189 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
 190 {
 191         u16 len = sb->s_blocksize -
 192                  offsetof(struct ocfs2_xattr_header, xh_entries);
 193
 194         return len / sizeof(struct ocfs2_xattr_entry);
 195 }
 196
 197 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 198 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 199 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 200
 201 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 202 {
 203         struct ocfs2_xattr_bucket *bucket;
 204         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 205
 206         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 207
 208         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 209         if (bucket) {
 210                 bucket->bu_inode = inode;
 211                 bucket->bu_blocks = blks;
 212         }
 213
 214         return bucket;
 215 }
 216
 217 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 218 {
 219         int i;
 220
 221         for (i = 0; i < bucket->bu_blocks; i++) {
 222                 brelse(bucket->bu_bhs[i]);
 223                 bucket->bu_bhs[i] = NULL;
 224         }
 225 }
 226
 227 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 228 {
 229         if (bucket) {
 230                 ocfs2_xattr_bucket_relse(bucket);
 231                 bucket->bu_inode = NULL;
 232                 kfree(bucket);
 233         }
 234 }
 235
 236 /*
 237  * A bucket that has never been written to disk doesn't need to be
 238  * read.  We just need the buffer_heads.  Don't call this for
 239  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 240  * them fully.
 241  */
 242 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 243                                    u64 xb_blkno)
 244 {
 245         int i, rc = 0;
 246
 247         for (i = 0; i < bucket->bu_blocks; i++) {
 248                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 249                                               xb_blkno + i);
 250                 if (!bucket->bu_bhs[i]) {
 251                         rc = -EIO;
 252                         mlog_errno(rc);
 253                         break;
 254                 }
 255
 256                 if (!ocfs2_buffer_uptodate(bucket->bu_inode,
 257                                            bucket->bu_bhs[i]))
 258                         ocfs2_set_new_buffer_uptodate(bucket->bu_inode,
 259                                                       bucket->bu_bhs[i]);
 260         }
 261
 262         if (rc)
 263                 ocfs2_xattr_bucket_relse(bucket);
 264         return rc;
 265 }
 266
 267 /* Read the xattr bucket at xb_blkno */
 268 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 269                                    u64 xb_blkno)
 270 {
 271         int rc;
 272
 273         rc = ocfs2_read_blocks(bucket->bu_inode, xb_blkno,
 274                                bucket->bu_blocks, bucket->bu_bhs, 0,
 275                                NULL);
 276         if (!rc) {
 277                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 278                                                  bucket->bu_bhs,
 279                                                  bucket->bu_blocks,
 280                                                  &bucket_xh(bucket)->xh_check);
 281                 if (rc)
 282                         mlog_errno(rc);
 283         }
 284
 285         if (rc)
 286                 ocfs2_xattr_bucket_relse(bucket);
 287         return rc;
 288 }
 289
 290 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 291                                              struct ocfs2_xattr_bucket *bucket,
 292                                              int type)
 293 {
 294         int i, rc = 0;
 295
 296         for (i = 0; i < bucket->bu_blocks; i++) {
 297                 rc = ocfs2_journal_access(handle, bucket->bu_inode,
 298                                           bucket->bu_bhs[i], type);
 299                 if (rc) {
 300                         mlog_errno(rc);
 301                         break;
 302                 }
 303         }
 304
 305         return rc;
 306 }
 307
 308 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 309                                              struct ocfs2_xattr_bucket *bucket)
 310 {
 311         int i;
 312
 313         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 314                                    bucket->bu_bhs, bucket->bu_blocks,
 315                                    &bucket_xh(bucket)->xh_check);
 316
 317         for (i = 0; i < bucket->bu_blocks; i++)
 318                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 319 }
 320
 321 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 322                                          struct ocfs2_xattr_bucket *src)
 323 {
 324         int i;
 325         int blocksize = src->bu_inode->i_sb->s_blocksize;
 326
 327         BUG_ON(dest->bu_blocks != src->bu_blocks);
 328         BUG_ON(dest->bu_inode != src->bu_inode);
 329
 330         for (i = 0; i < src->bu_blocks; i++) {
 331                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 332                        blocksize);
 333         }
 334 }
 335
 336 static int ocfs2_validate_xattr_block(struct super_block *sb,
 337                                       struct buffer_head *bh)
 338 {
 339         int rc;
 340         struct ocfs2_xattr_block *xb =
 341                 (struct ocfs2_xattr_block *)bh->b_data;
 342
 343         mlog(0, "Validating xattr block %llu\n",
 344              (unsigned long long)bh->b_blocknr);
 345
 346         BUG_ON(!buffer_uptodate(bh));
 347
 348         /*
 349          * If the ecc fails, we return the error but otherwise
 350          * leave the filesystem running.  We know any error is
 351          * local to this block.
 352          */
 353         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 354         if (rc)
 355                 return rc;
 356
 357         /*
 358          * Errors after here are fatal
 359          */
 360
 361         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 362                 ocfs2_error(sb,
 363                             "Extended attribute block #%llu has bad "
 364                             "signature %.*s",
 365                             (unsigned long long)bh->b_blocknr, 7,
 366                             xb->xb_signature);
 367                 return -EINVAL;
 368         }
 369
 370         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 371                 ocfs2_error(sb,
 372                             "Extended attribute block #%llu has an "
 373                             "invalid xb_blkno of %llu",
 374                             (unsigned long long)bh->b_blocknr,
 375                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
 376                 return -EINVAL;
 377         }
 378
 379         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 380                 ocfs2_error(sb,
 381                             "Extended attribute block #%llu has an invalid "
 382                             "xb_fs_generation of #%u",
 383                             (unsigned long long)bh->b_blocknr,
 384                             le32_to_cpu(xb->xb_fs_generation));
 385                 return -EINVAL;
 386         }
 387
 388         return 0;
 389 }
 390
 391 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 392                                   struct buffer_head **bh)
 393 {
 394         int rc;
 395         struct buffer_head *tmp = *bh;
 396
 397         rc = ocfs2_read_block(inode, xb_blkno, &tmp,
 398                               ocfs2_validate_xattr_block);
 399
 400         /* If ocfs2_read_block() got us a new bh, pass it up. */
 401         if (!rc && !*bh)
 402                 *bh = tmp;
 403
 404         return rc;
 405 }
 406
 407 static inline const char *ocfs2_xattr_prefix(int name_index)
 408 {
 409         struct xattr_handler *handler = NULL;
 410
 411         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 412                 handler = ocfs2_xattr_handler_map[name_index];
 413
 414         return handler ? handler->prefix : NULL;
 415 }
 416
 417 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 418                                  const char *name,
 419                                  int name_len)
 420 {
 421         /* Get hash value of uuid from super block */
 422         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 423         int i;
 424
 425         /* hash extended attribute name */
 426         for (i = 0; i < name_len; i++) {
 427                 hash = (hash << OCFS2_HASH_SHIFT) ^
 428                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 429                        *name++;
 430         }
 431
 432         return hash;
 433 }
 434
 435 /*
 436  * ocfs2_xattr_hash_entry()
 437  *
 438  * Compute the hash of an extended attribute.
 439  */
 440 static void ocfs2_xattr_hash_entry(struct inode *inode,
 441                                    struct ocfs2_xattr_header *header,
 442                                    struct ocfs2_xattr_entry *entry)
 443 {
 444         u32 hash = 0;
 445         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
 446
 447         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
 448         entry->xe_name_hash = cpu_to_le32(hash);
 449
 450         return;
 451 }
 452
 453 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 454 {
 455         int size = 0;
 456
 457         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
 458                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 459         else
 460                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 461         size += sizeof(struct ocfs2_xattr_entry);
 462
 463         return size;
 464 }
 465
 466 int ocfs2_calc_security_init(struct inode *dir,
 467                              struct ocfs2_security_xattr_info *si,
 468                              int *want_clusters,
 469                              int *xattr_credits,
 470                              struct ocfs2_alloc_context **xattr_ac)
 471 {
 472         int ret = 0;
 473         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 474         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 475                                                  si->value_len);
 476
 477         /*
 478          * The max space of security xattr taken inline is
 479          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 480          * So reserve one metadata block for it is ok.
 481          */
 482         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 483             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 484                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 485                 if (ret) {
 486                         mlog_errno(ret);
 487                         return ret;
 488                 }
 489                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 490         }
 491
 492         /* reserve clusters for xattr value which will be set in B tree*/
 493         if (si->value_len > OCFS2_XATTR_INLINE_SIZE)
 494                 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
 495                                                            si->value_len);
 496         return ret;
 497 }
 498
 499 int ocfs2_calc_xattr_init(struct inode *dir,
 500                           struct buffer_head *dir_bh,
 501                           int mode,
 502                           struct ocfs2_security_xattr_info *si,
 503                           int *want_clusters,
 504                           int *xattr_credits,
 505                           struct ocfs2_alloc_context **xattr_ac)
 506 {
 507         int ret = 0;
 508         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 509         int s_size = 0;
 510         int a_size = 0;
 511         int acl_len = 0;
 512
 513         if (si->enable)
 514                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 515                                                      si->value_len);
 516
 517         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 518                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 519                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 520                                         "", NULL, 0);
 521                 if (acl_len > 0) {
 522                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 523                         if (S_ISDIR(mode))
 524                                 a_size <<= 1;
 525                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 526                         mlog_errno(ret);
 527                         return ret;
 528                 }
 529         }
 530
 531         if (!(s_size + a_size))
 532                 return ret;
 533
 534         /*
 535          * The max space of security xattr taken inline is
 536          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 537          * The max space of acl xattr taken inline is
 538          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 539          * when blocksize = 512, may reserve one more cluser for
 540          * xattr bucket, otherwise reserve one metadata block
 541          * for them is ok.
 542          */
 543         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 544             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 545                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 546                 if (ret) {
 547                         mlog_errno(ret);
 548                         return ret;
 549                 }
 550                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 551         }
 552
 553         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 554             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 555                 *want_clusters += 1;
 556                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 557         }
 558
 559         /* reserve clusters for xattr value which will be set in B tree*/
 560         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE)
 561                 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
 562                                                            si->value_len);
 563         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 564             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 565                 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 566                 if (S_ISDIR(mode))
 567                         *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
 568                                                                    acl_len);
 569         }
 570
 571         return ret;
 572 }
 573
 574 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 575                                          u32 clusters_to_add,
 576                                          struct ocfs2_xattr_value_buf *vb,
 577                                          struct ocfs2_xattr_set_ctxt *ctxt)
 578 {
 579         int status = 0;
 580         handle_t *handle = ctxt->handle;
 581         enum ocfs2_alloc_restarted why;
 582         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 583         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 584         struct ocfs2_extent_tree et;
 585
 586         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
 587
 588         ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
 589
 590         status = vb->vb_access(handle, inode, vb->vb_bh,
 591                               OCFS2_JOURNAL_ACCESS_WRITE);
 592         if (status < 0) {
 593                 mlog_errno(status);
 594                 goto leave;
 595         }
 596
 597         prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 598         status = ocfs2_add_clusters_in_btree(osb,
 599                                              inode,
 600                                              &logical_start,
 601                                              clusters_to_add,
 602                                              0,
 603                                              &et,
 604                                              handle,
 605                                              ctxt->data_ac,
 606                                              ctxt->meta_ac,
 607                                              &why);
 608         if (status < 0) {
 609                 mlog_errno(status);
 610                 goto leave;
 611         }
 612
 613         status = ocfs2_journal_dirty(handle, vb->vb_bh);
 614         if (status < 0) {
 615                 mlog_errno(status);
 616                 goto leave;
 617         }
 618
 619         clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
 620
 621         /*
 622          * We should have already allocated enough space before the transaction,
 623          * so no need to restart.
 624          */
 625         BUG_ON(why != RESTART_NONE || clusters_to_add);
 626
 627 leave:
 628
 629         return status;
 630 }
 631
 632 static int __ocfs2_remove_xattr_range(struct inode *inode,
 633                                       struct ocfs2_xattr_value_buf *vb,
 634                                       u32 cpos, u32 phys_cpos, u32 len,
 635                                       struct ocfs2_xattr_set_ctxt *ctxt)
 636 {
 637         int ret;
 638         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 639         handle_t *handle = ctxt->handle;
 640         struct ocfs2_extent_tree et;
 641
 642         ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
 643
 644         ret = vb->vb_access(handle, inode, vb->vb_bh,
 645                             OCFS2_JOURNAL_ACCESS_WRITE);
 646         if (ret) {
 647                 mlog_errno(ret);
 648                 goto out;
 649         }
 650
 651         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
 652                                   &ctxt->dealloc);
 653         if (ret) {
 654                 mlog_errno(ret);
 655                 goto out;
 656         }
 657
 658         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 659
 660         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
 661         if (ret) {
 662                 mlog_errno(ret);
 663                 goto out;
 664         }
 665
 666         ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
 667         if (ret)
 668                 mlog_errno(ret);
 669
 670 out:
 671         return ret;
 672 }
 673
 674 static int ocfs2_xattr_shrink_size(struct inode *inode,
 675                                    u32 old_clusters,
 676                                    u32 new_clusters,
 677                                    struct ocfs2_xattr_value_buf *vb,
 678                                    struct ocfs2_xattr_set_ctxt *ctxt)
 679 {
 680         int ret = 0;
 681         u32 trunc_len, cpos, phys_cpos, alloc_size;
 682         u64 block;
 683
 684         if (old_clusters <= new_clusters)
 685                 return 0;
 686
 687         cpos = new_clusters;
 688         trunc_len = old_clusters - new_clusters;
 689         while (trunc_len) {
 690                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 691                                                &alloc_size,
 692                                                &vb->vb_xv->xr_list);
 693                 if (ret) {
 694                         mlog_errno(ret);
 695                         goto out;
 696                 }
 697
 698                 if (alloc_size > trunc_len)
 699                         alloc_size = trunc_len;
 700
 701                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 702                                                  phys_cpos, alloc_size,
 703                                                  ctxt);
 704                 if (ret) {
 705                         mlog_errno(ret);
 706                         goto out;
 707                 }
 708
 709                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 710                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
 711                                                        alloc_size);
 712                 cpos += alloc_size;
 713                 trunc_len -= alloc_size;
 714         }
 715
 716 out:
 717         return ret;
 718 }
 719
 720 static int ocfs2_xattr_value_truncate(struct inode *inode,
 721                                       struct ocfs2_xattr_value_buf *vb,
 722                                       int len,
 723                                       struct ocfs2_xattr_set_ctxt *ctxt)
 724 {
 725         int ret;
 726         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 727         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 728
 729         if (new_clusters == old_clusters)
 730                 return 0;
 731
 732         if (new_clusters > old_clusters)
 733                 ret = ocfs2_xattr_extend_allocation(inode,
 734                                                     new_clusters - old_clusters,
 735                                                     vb, ctxt);
 736         else
 737                 ret = ocfs2_xattr_shrink_size(inode,
 738                                               old_clusters, new_clusters,
 739                                               vb, ctxt);
 740
 741         return ret;
 742 }
 743
 744 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 745                                   size_t *result, const char *prefix,
 746                                   const char *name, int name_len)
 747 {
 748         char *p = buffer + *result;
 749         int prefix_len = strlen(prefix);
 750         int total_len = prefix_len + name_len + 1;
 751
 752         *result += total_len;
 753
 754         /* we are just looking for how big our buffer needs to be */
 755         if (!size)
 756                 return 0;
 757
 758         if (*result > size)
 759                 return -ERANGE;
 760
 761         memcpy(p, prefix, prefix_len);
 762         memcpy(p + prefix_len, name, name_len);
 763         p[prefix_len + name_len] = '\0';
 764
 765         return 0;
 766 }
 767
 768 static int ocfs2_xattr_list_entries(struct inode *inode,
 769                                     struct ocfs2_xattr_header *header,
 770                                     char *buffer, size_t buffer_size)
 771 {
 772         size_t result = 0;
 773         int i, type, ret;
 774         const char *prefix, *name;
 775
 776         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 777                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 778                 type = ocfs2_xattr_get_type(entry);
 779                 prefix = ocfs2_xattr_prefix(type);
 780
 781                 if (prefix) {
 782                         name = (const char *)header +
 783                                 le16_to_cpu(entry->xe_name_offset);
 784
 785                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 786                                                      &result, prefix, name,
 787                                                      entry->xe_name_len);
 788                         if (ret)
 789                                 return ret;
 790                 }
 791         }
 792
 793         return result;
 794 }
 795
 796 static int ocfs2_xattr_ibody_list(struct inode *inode,
 797                                   struct ocfs2_dinode *di,
 798                                   char *buffer,
 799                                   size_t buffer_size)
 800 {
 801         struct ocfs2_xattr_header *header = NULL;
 802         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 803         int ret = 0;
 804
 805         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 806                 return ret;
 807
 808         header = (struct ocfs2_xattr_header *)
 809                  ((void *)di + inode->i_sb->s_blocksize -
 810                  le16_to_cpu(di->i_xattr_inline_size));
 811
 812         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 813
 814         return ret;
 815 }
 816
 817 static int ocfs2_xattr_block_list(struct inode *inode,
 818                                   struct ocfs2_dinode *di,
 819                                   char *buffer,
 820                                   size_t buffer_size)
 821 {
 822         struct buffer_head *blk_bh = NULL;
 823         struct ocfs2_xattr_block *xb;
 824         int ret = 0;
 825
 826         if (!di->i_xattr_loc)
 827                 return ret;
 828
 829         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
 830                                      &blk_bh);
 831         if (ret < 0) {
 832                 mlog_errno(ret);
 833                 return ret;
 834         }
 835
 836         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 837         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 838                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 839                 ret = ocfs2_xattr_list_entries(inode, header,
 840                                                buffer, buffer_size);
 841         } else {
 842                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
 843                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
 844                                                    buffer, buffer_size);
 845         }
 846
 847         brelse(blk_bh);
 848
 849         return ret;
 850 }
 851
 852 ssize_t ocfs2_listxattr(struct dentry *dentry,
 853                         char *buffer,
 854                         size_t size)
 855 {
 856         int ret = 0, i_ret = 0, b_ret = 0;
 857         struct buffer_head *di_bh = NULL;
 858         struct ocfs2_dinode *di = NULL;
 859         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
 860
 861         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
 862                 return -EOPNOTSUPP;
 863
 864         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 865                 return ret;
 866
 867         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
 868         if (ret < 0) {
 869                 mlog_errno(ret);
 870                 return ret;
 871         }
 872
 873         di = (struct ocfs2_dinode *)di_bh->b_data;
 874
 875         down_read(&oi->ip_xattr_sem);
 876         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
 877         if (i_ret < 0)
 878                 b_ret = 0;
 879         else {
 880                 if (buffer) {
 881                         buffer += i_ret;
 882                         size -= i_ret;
 883                 }
 884                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
 885                                                buffer, size);
 886                 if (b_ret < 0)
 887                         i_ret = 0;
 888         }
 889         up_read(&oi->ip_xattr_sem);
 890         ocfs2_inode_unlock(dentry->d_inode, 0);
 891
 892         brelse(di_bh);
 893
 894         return i_ret + b_ret;
 895 }
 896
 897 static int ocfs2_xattr_find_entry(int name_index,
 898                                   const char *name,
 899                                   struct ocfs2_xattr_search *xs)
 900 {
 901         struct ocfs2_xattr_entry *entry;
 902         size_t name_len;
 903         int i, cmp = 1;
 904
 905         if (name == NULL)
 906                 return -EINVAL;
 907
 908         name_len = strlen(name);
 909         entry = xs->here;
 910         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
 911                 cmp = name_index - ocfs2_xattr_get_type(entry);
 912                 if (!cmp)
 913                         cmp = name_len - entry->xe_name_len;
 914                 if (!cmp)
 915                         cmp = memcmp(name, (xs->base +
 916                                      le16_to_cpu(entry->xe_name_offset)),
 917                                      name_len);
 918                 if (cmp == 0)
 919                         break;
 920                 entry += 1;
 921         }
 922         xs->here = entry;
 923
 924         return cmp ? -ENODATA : 0;
 925 }
 926
 927 static int ocfs2_xattr_get_value_outside(struct inode *inode,
 928                                          struct ocfs2_xattr_value_root *xv,
 929                                          void *buffer,
 930                                          size_t len)
 931 {
 932         u32 cpos, p_cluster, num_clusters, bpc, clusters;
 933         u64 blkno;
 934         int i, ret = 0;
 935         size_t cplen, blocksize;
 936         struct buffer_head *bh = NULL;
 937         struct ocfs2_extent_list *el;
 938
 939         el = &xv->xr_list;
 940         clusters = le32_to_cpu(xv->xr_clusters);
 941         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 942         blocksize = inode->i_sb->s_blocksize;
 943
 944         cpos = 0;
 945         while (cpos < clusters) {
 946                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 947                                                &num_clusters, el);
 948                 if (ret) {
 949                         mlog_errno(ret);
 950                         goto out;
 951                 }
 952
 953                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 954                 /* Copy ocfs2_xattr_value */
 955                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 956                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
 957                         if (ret) {
 958                                 mlog_errno(ret);
 959                                 goto out;
 960                         }
 961
 962                         cplen = len >= blocksize ? blocksize : len;
 963                         memcpy(buffer, bh->b_data, cplen);
 964                         len -= cplen;
 965                         buffer += cplen;
 966
 967                         brelse(bh);
 968                         bh = NULL;
 969                         if (len == 0)
 970                                 break;
 971                 }
 972                 cpos += num_clusters;
 973         }
 974 out:
 975         return ret;
 976 }
 977
 978 static int ocfs2_xattr_ibody_get(struct inode *inode,
 979                                  int name_index,
 980                                  const char *name,
 981                                  void *buffer,
 982                                  size_t buffer_size,
 983                                  struct ocfs2_xattr_search *xs)
 984 {
 985         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 986         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
 987         struct ocfs2_xattr_value_root *xv;
 988         size_t size;
 989         int ret = 0;
 990
 991         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 992                 return -ENODATA;
 993
 994         xs->end = (void *)di + inode->i_sb->s_blocksize;
 995         xs->header = (struct ocfs2_xattr_header *)
 996                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
 997         xs->base = (void *)xs->header;
 998         xs->here = xs->header->xh_entries;
 999
1000         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1001         if (ret)
1002                 return ret;
1003         size = le64_to_cpu(xs->here->xe_value_size);
1004         if (buffer) {
1005                 if (size > buffer_size)
1006                         return -ERANGE;
1007                 if (ocfs2_xattr_is_local(xs->here)) {
1008                         memcpy(buffer, (void *)xs->base +
1009                                le16_to_cpu(xs->here->xe_name_offset) +
1010                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1011                 } else {
1012                         xv = (struct ocfs2_xattr_value_root *)
1013                                 (xs->base + le16_to_cpu(
1014                                  xs->here->xe_name_offset) +
1015                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1016                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1017                                                             buffer, size);
1018                         if (ret < 0) {
1019                                 mlog_errno(ret);
1020                                 return ret;
1021                         }
1022                 }
1023         }
1024
1025         return size;
1026 }
1027
1028 static int ocfs2_xattr_block_get(struct inode *inode,
1029                                  int name_index,
1030                                  const char *name,
1031                                  void *buffer,
1032                                  size_t buffer_size,
1033                                  struct ocfs2_xattr_search *xs)
1034 {
1035         struct ocfs2_xattr_block *xb;
1036         struct ocfs2_xattr_value_root *xv;
1037         size_t size;
1038         int ret = -ENODATA, name_offset, name_len, block_off, i;
1039
1040         xs->bucket = ocfs2_xattr_bucket_new(inode);
1041         if (!xs->bucket) {
1042                 ret = -ENOMEM;
1043                 mlog_errno(ret);
1044                 goto cleanup;
1045         }
1046
1047         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1048         if (ret) {
1049                 mlog_errno(ret);
1050                 goto cleanup;
1051         }
1052
1053         if (xs->not_found) {
1054                 ret = -ENODATA;
1055                 goto cleanup;
1056         }
1057
1058         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1059         size = le64_to_cpu(xs->here->xe_value_size);
1060         if (buffer) {
1061                 ret = -ERANGE;
1062                 if (size > buffer_size)
1063                         goto cleanup;
1064
1065                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1066                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1067                 i = xs->here - xs->header->xh_entries;
1068
1069                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1070                         ret = ocfs2_xattr_bucket_get_name_value(inode,
1071                                                                 bucket_xh(xs->bucket),
1072                                                                 i,
1073                                                                 &block_off,
1074                                                                 &name_offset);
1075                         xs->base = bucket_block(xs->bucket, block_off);
1076                 }
1077                 if (ocfs2_xattr_is_local(xs->here)) {
1078                         memcpy(buffer, (void *)xs->base +
1079                                name_offset + name_len, size);
1080                 } else {
1081                         xv = (struct ocfs2_xattr_value_root *)
1082                                 (xs->base + name_offset + name_len);
1083                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1084                                                             buffer, size);
1085                         if (ret < 0) {
1086                                 mlog_errno(ret);
1087                                 goto cleanup;
1088                         }
1089                 }
1090         }
1091         ret = size;
1092 cleanup:
1093         ocfs2_xattr_bucket_free(xs->bucket);
1094
1095         brelse(xs->xattr_bh);
1096         xs->xattr_bh = NULL;
1097         return ret;
1098 }
1099
1100 int ocfs2_xattr_get_nolock(struct inode *inode,
1101                            struct buffer_head *di_bh,
1102                            int name_index,
1103                            const char *name,
1104                            void *buffer,
1105                            size_t buffer_size)
1106 {
1107         int ret;
1108         struct ocfs2_dinode *di = NULL;
1109         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1110         struct ocfs2_xattr_search xis = {
1111                 .not_found = -ENODATA,
1112         };
1113         struct ocfs2_xattr_search xbs = {
1114                 .not_found = -ENODATA,
1115         };
1116
1117         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1118                 return -EOPNOTSUPP;
1119
1120         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1121                 ret = -ENODATA;
1122
1123         xis.inode_bh = xbs.inode_bh = di_bh;
1124         di = (struct ocfs2_dinode *)di_bh->b_data;
1125
1126         down_read(&oi->ip_xattr_sem);
1127         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1128                                     buffer_size, &xis);
1129         if (ret == -ENODATA && di->i_xattr_loc)
1130                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1131                                             buffer_size, &xbs);
1132         up_read(&oi->ip_xattr_sem);
1133
1134         return ret;
1135 }
1136
1137 /* ocfs2_xattr_get()
1138  *
1139  * Copy an extended attribute into the buffer provided.
1140  * Buffer is NULL to compute the size of buffer required.
1141  */
1142 static int ocfs2_xattr_get(struct inode *inode,
1143                            int name_index,
1144                            const char *name,
1145                            void *buffer,
1146                            size_t buffer_size)
1147 {
1148         int ret;
1149         struct buffer_head *di_bh = NULL;
1150
1151         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1152         if (ret < 0) {
1153                 mlog_errno(ret);
1154                 return ret;
1155         }
1156         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1157                                      name, buffer, buffer_size);
1158
1159         ocfs2_inode_unlock(inode, 0);
1160
1161         brelse(di_bh);
1162
1163         return ret;
1164 }
1165
1166 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1167                                            handle_t *handle,
1168                                            struct ocfs2_xattr_value_root *xv,
1169                                            const void *value,
1170                                            int value_len)
1171 {
1172         int ret = 0, i, cp_len, credits;
1173         u16 blocksize = inode->i_sb->s_blocksize;
1174         u32 p_cluster, num_clusters;
1175         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1176         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1177         u64 blkno;
1178         struct buffer_head *bh = NULL;
1179
1180         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1181
1182         /*
1183          * In __ocfs2_xattr_set_value_outside has already been dirtied,
1184          * so we don't need to worry about whether ocfs2_extend_trans
1185          * will create a new transactio for us or not.
1186          */
1187         credits = clusters * bpc;
1188         ret = ocfs2_extend_trans(handle, credits);
1189         if (ret) {
1190                 mlog_errno(ret);
1191                 goto out;
1192         }
1193
1194         while (cpos < clusters) {
1195                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1196                                                &num_clusters, &xv->xr_list);
1197                 if (ret) {
1198                         mlog_errno(ret);
1199                         goto out;
1200                 }
1201
1202                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1203
1204                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1205                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
1206                         if (ret) {
1207                                 mlog_errno(ret);
1208                                 goto out;
1209                         }
1210
1211                         ret = ocfs2_journal_access(handle,
1212                                                    inode,
1213                                                    bh,
1214                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1215                         if (ret < 0) {
1216                                 mlog_errno(ret);
1217                                 goto out;
1218                         }
1219
1220                         cp_len = value_len > blocksize ? blocksize : value_len;
1221                         memcpy(bh->b_data, value, cp_len);
1222                         value_len -= cp_len;
1223                         value += cp_len;
1224                         if (cp_len < blocksize)
1225                                 memset(bh->b_data + cp_len, 0,
1226                                        blocksize - cp_len);
1227
1228                         ret = ocfs2_journal_dirty(handle, bh);
1229                         if (ret < 0) {
1230                                 mlog_errno(ret);
1231                                 goto out;
1232                         }
1233                         brelse(bh);
1234                         bh = NULL;
1235
1236                         /*
1237                          * XXX: do we need to empty all the following
1238                          * blocks in this cluster?
1239                          */
1240                         if (!value_len)
1241                                 break;
1242                 }
1243                 cpos += num_clusters;
1244         }
1245 out:
1246         brelse(bh);
1247
1248         return ret;
1249 }
1250
1251 static int ocfs2_xattr_cleanup(struct inode *inode,
1252                                handle_t *handle,
1253                                struct ocfs2_xattr_info *xi,
1254                                struct ocfs2_xattr_search *xs,
1255                                struct ocfs2_xattr_value_buf *vb,
1256                                size_t offs)
1257 {
1258         int ret = 0;
1259         size_t name_len = strlen(xi->name);
1260         void *val = xs->base + offs;
1261         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1262
1263         ret = vb->vb_access(handle, inode, vb->vb_bh,
1264                             OCFS2_JOURNAL_ACCESS_WRITE);
1265         if (ret) {
1266                 mlog_errno(ret);
1267                 goto out;
1268         }
1269         /* Decrease xattr count */
1270         le16_add_cpu(&xs->header->xh_count, -1);
1271         /* Remove the xattr entry and tree root which has already be set*/
1272         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1273         memset(val, 0, size);
1274
1275         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1276         if (ret < 0)
1277                 mlog_errno(ret);
1278 out:
1279         return ret;
1280 }
1281
1282 static int ocfs2_xattr_update_entry(struct inode *inode,
1283                                     handle_t *handle,
1284                                     struct ocfs2_xattr_info *xi,
1285                                     struct ocfs2_xattr_search *xs,
1286                                     struct ocfs2_xattr_value_buf *vb,
1287                                     size_t offs)
1288 {
1289         int ret;
1290
1291         ret = vb->vb_access(handle, inode, vb->vb_bh,
1292                             OCFS2_JOURNAL_ACCESS_WRITE);
1293         if (ret) {
1294                 mlog_errno(ret);
1295                 goto out;
1296         }
1297
1298         xs->here->xe_name_offset = cpu_to_le16(offs);
1299         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1300         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1301                 ocfs2_xattr_set_local(xs->here, 1);
1302         else
1303                 ocfs2_xattr_set_local(xs->here, 0);
1304         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1305
1306         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1307         if (ret < 0)
1308                 mlog_errno(ret);
1309 out:
1310         return ret;
1311 }
1312
1313 /*
1314  * ocfs2_xattr_set_value_outside()
1315  *
1316  * Set large size value in B tree.
1317  */
1318 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1319                                          struct ocfs2_xattr_info *xi,
1320                                          struct ocfs2_xattr_search *xs,
1321                                          struct ocfs2_xattr_set_ctxt *ctxt,
1322                                          struct ocfs2_xattr_value_buf *vb,
1323                                          size_t offs)
1324 {
1325         size_t name_len = strlen(xi->name);
1326         void *val = xs->base + offs;
1327         struct ocfs2_xattr_value_root *xv = NULL;
1328         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1329         int ret = 0;
1330
1331         memset(val, 0, size);
1332         memcpy(val, xi->name, name_len);
1333         xv = (struct ocfs2_xattr_value_root *)
1334                 (val + OCFS2_XATTR_SIZE(name_len));
1335         xv->xr_clusters = 0;
1336         xv->xr_last_eb_blk = 0;
1337         xv->xr_list.l_tree_depth = 0;
1338         xv->xr_list.l_count = cpu_to_le16(1);
1339         xv->xr_list.l_next_free_rec = 0;
1340         vb->vb_xv = xv;
1341
1342         ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1343         if (ret < 0) {
1344                 mlog_errno(ret);
1345                 return ret;
1346         }
1347         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1348         if (ret < 0) {
1349                 mlog_errno(ret);
1350                 return ret;
1351         }
1352         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
1353                                               xi->value, xi->value_len);
1354         if (ret < 0)
1355                 mlog_errno(ret);
1356
1357         return ret;
1358 }
1359
1360 /*
1361  * ocfs2_xattr_set_entry_local()
1362  *
1363  * Set, replace or remove extended attribute in local.
1364  */
1365 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1366                                         struct ocfs2_xattr_info *xi,
1367                                         struct ocfs2_xattr_search *xs,
1368                                         struct ocfs2_xattr_entry *last,
1369                                         size_t min_offs)
1370 {
1371         size_t name_len = strlen(xi->name);
1372         int i;
1373
1374         if (xi->value && xs->not_found) {
1375                 /* Insert the new xattr entry. */
1376                 le16_add_cpu(&xs->header->xh_count, 1);
1377                 ocfs2_xattr_set_type(last, xi->name_index);
1378                 ocfs2_xattr_set_local(last, 1);
1379                 last->xe_name_len = name_len;
1380         } else {
1381                 void *first_val;
1382                 void *val;
1383                 size_t offs, size;
1384
1385                 first_val = xs->base + min_offs;
1386                 offs = le16_to_cpu(xs->here->xe_name_offset);
1387                 val = xs->base + offs;
1388
1389                 if (le64_to_cpu(xs->here->xe_value_size) >
1390                     OCFS2_XATTR_INLINE_SIZE)
1391                         size = OCFS2_XATTR_SIZE(name_len) +
1392                                 OCFS2_XATTR_ROOT_SIZE;
1393                 else
1394                         size = OCFS2_XATTR_SIZE(name_len) +
1395                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1396
1397                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1398                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1399                         /* The old and the new value have the
1400                            same size. Just replace the value. */
1401                         ocfs2_xattr_set_local(xs->here, 1);
1402                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1403                         /* Clear value bytes. */
1404                         memset(val + OCFS2_XATTR_SIZE(name_len),
1405                                0,
1406                                OCFS2_XATTR_SIZE(xi->value_len));
1407                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1408                                xi->value,
1409                                xi->value_len);
1410                         return;
1411                 }
1412                 /* Remove the old name+value. */
1413                 memmove(first_val + size, first_val, val - first_val);
1414                 memset(first_val, 0, size);
1415                 xs->here->xe_name_hash = 0;
1416                 xs->here->xe_name_offset = 0;
1417                 ocfs2_xattr_set_local(xs->here, 1);
1418                 xs->here->xe_value_size = 0;
1419
1420                 min_offs += size;
1421
1422                 /* Adjust all value offsets. */
1423                 last = xs->header->xh_entries;
1424                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1425                         size_t o = le16_to_cpu(last->xe_name_offset);
1426
1427                         if (o < offs)
1428                                 last->xe_name_offset = cpu_to_le16(o + size);
1429                         last += 1;
1430                 }
1431
1432                 if (!xi->value) {
1433                         /* Remove the old entry. */
1434                         last -= 1;
1435                         memmove(xs->here, xs->here + 1,
1436                                 (void *)last - (void *)xs->here);
1437                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1438                         le16_add_cpu(&xs->header->xh_count, -1);
1439                 }
1440         }
1441         if (xi->value) {
1442                 /* Insert the new name+value. */
1443                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1444                                 OCFS2_XATTR_SIZE(xi->value_len);
1445                 void *val = xs->base + min_offs - size;
1446
1447                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1448                 memset(val, 0, size);
1449                 memcpy(val, xi->name, name_len);
1450                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1451                        xi->value,
1452                        xi->value_len);
1453                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1454                 ocfs2_xattr_set_local(xs->here, 1);
1455                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1456         }
1457
1458         return;
1459 }
1460
1461 /*
1462  * ocfs2_xattr_set_entry()
1463  *
1464  * Set extended attribute entry into inode or block.
1465  *
1466  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1467  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1468  * then set value in B tree with set_value_outside().
1469  */
1470 static int ocfs2_xattr_set_entry(struct inode *inode,
1471                                  struct ocfs2_xattr_info *xi,
1472                                  struct ocfs2_xattr_search *xs,
1473                                  struct ocfs2_xattr_set_ctxt *ctxt,
1474                                  int flag)
1475 {
1476         struct ocfs2_xattr_entry *last;
1477         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1478         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1479         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1480         size_t size_l = 0;
1481         handle_t *handle = ctxt->handle;
1482         int free, i, ret;
1483         struct ocfs2_xattr_info xi_l = {
1484                 .name_index = xi->name_index,
1485                 .name = xi->name,
1486                 .value = xi->value,
1487                 .value_len = xi->value_len,
1488         };
1489         struct ocfs2_xattr_value_buf vb = {
1490                 .vb_bh = xs->xattr_bh,
1491                 .vb_access = ocfs2_journal_access_di,
1492         };
1493
1494         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1495                 BUG_ON(xs->xattr_bh == xs->inode_bh);
1496                 vb.vb_access = ocfs2_journal_access_xb;
1497         } else
1498                 BUG_ON(xs->xattr_bh != xs->inode_bh);
1499
1500         /* Compute min_offs, last and free space. */
1501         last = xs->header->xh_entries;
1502
1503         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1504                 size_t offs = le16_to_cpu(last->xe_name_offset);
1505                 if (offs < min_offs)
1506                         min_offs = offs;
1507                 last += 1;
1508         }
1509
1510         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1511         if (free < 0)
1512                 return -EIO;
1513
1514         if (!xs->not_found) {
1515                 size_t size = 0;
1516                 if (ocfs2_xattr_is_local(xs->here))
1517                         size = OCFS2_XATTR_SIZE(name_len) +
1518                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1519                 else
1520                         size = OCFS2_XATTR_SIZE(name_len) +
1521                                 OCFS2_XATTR_ROOT_SIZE;
1522                 free += (size + sizeof(struct ocfs2_xattr_entry));
1523         }
1524         /* Check free space in inode or block */
1525         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1526                 if (free < sizeof(struct ocfs2_xattr_entry) +
1527                            OCFS2_XATTR_SIZE(name_len) +
1528                            OCFS2_XATTR_ROOT_SIZE) {
1529                         ret = -ENOSPC;
1530                         goto out;
1531                 }
1532                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1533                 xi_l.value = (void *)&def_xv;
1534                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1535         } else if (xi->value) {
1536                 if (free < sizeof(struct ocfs2_xattr_entry) +
1537                            OCFS2_XATTR_SIZE(name_len) +
1538                            OCFS2_XATTR_SIZE(xi->value_len)) {
1539                         ret = -ENOSPC;
1540                         goto out;
1541                 }
1542         }
1543
1544         if (!xs->not_found) {
1545                 /* For existing extended attribute */
1546                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1547                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1548                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1549                 void *val = xs->base + offs;
1550
1551                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1552                         /* Replace existing local xattr with tree root */
1553                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1554                                                             ctxt, &vb, offs);
1555                         if (ret < 0)
1556                                 mlog_errno(ret);
1557                         goto out;
1558                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1559                         /* For existing xattr which has value outside */
1560                         vb.vb_xv = (struct ocfs2_xattr_value_root *)
1561                                 (val + OCFS2_XATTR_SIZE(name_len));
1562
1563                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1564                                 /*
1565                                  * If new value need set outside also,
1566                                  * first truncate old value to new value,
1567                                  * then set new value with set_value_outside().
1568                                  */
1569                                 ret = ocfs2_xattr_value_truncate(inode,
1570                                                                  &vb,
1571                                                                  xi->value_len,
1572                                                                  ctxt);
1573                                 if (ret < 0) {
1574                                         mlog_errno(ret);
1575                                         goto out;
1576                                 }
1577
1578                                 ret = ocfs2_xattr_update_entry(inode,
1579                                                                handle,
1580                                                                xi,
1581                                                                xs,
1582                                                                &vb,
1583                                                                offs);
1584                                 if (ret < 0) {
1585                                         mlog_errno(ret);
1586                                         goto out;
1587                                 }
1588
1589                                 ret = __ocfs2_xattr_set_value_outside(inode,
1590                                                                 handle,
1591                                                                 vb.vb_xv,
1592                                                                 xi->value,
1593                                                                 xi->value_len);
1594                                 if (ret < 0)
1595                                         mlog_errno(ret);
1596                                 goto out;
1597                         } else {
1598                                 /*
1599                                  * If new value need set in local,
1600                                  * just trucate old value to zero.
1601                                  */
1602                                  ret = ocfs2_xattr_value_truncate(inode,
1603                                                                   &vb,
1604                                                                   0,
1605                                                                   ctxt);
1606                                 if (ret < 0)
1607                                         mlog_errno(ret);
1608                         }
1609                 }
1610         }
1611
1612         ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
1613                                       OCFS2_JOURNAL_ACCESS_WRITE);
1614         if (ret) {
1615                 mlog_errno(ret);
1616                 goto out;
1617         }
1618
1619         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1620                 ret = vb.vb_access(handle, inode, vb.vb_bh,
1621                                    OCFS2_JOURNAL_ACCESS_WRITE);
1622                 if (ret) {
1623                         mlog_errno(ret);
1624                         goto out;
1625                 }
1626         }
1627
1628         /*
1629          * Set value in local, include set tree root in local.
1630          * This is the first step for value size >INLINE_SIZE.
1631          */
1632         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1633
1634         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1635                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1636                 if (ret < 0) {
1637                         mlog_errno(ret);
1638                         goto out;
1639                 }
1640         }
1641
1642         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1643             (flag & OCFS2_INLINE_XATTR_FL)) {
1644                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1645                 unsigned int xattrsize = osb->s_xattr_inline_size;
1646
1647                 /*
1648                  * Adjust extent record count or inline data size
1649                  * to reserve space for extended attribute.
1650                  */
1651                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1652                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1653                         le16_add_cpu(&idata->id_count, -xattrsize);
1654                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1655                         struct ocfs2_extent_list *el = &di->id2.i_list;
1656                         le16_add_cpu(&el->l_count, -(xattrsize /
1657                                         sizeof(struct ocfs2_extent_rec)));
1658                 }
1659                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1660         }
1661         /* Update xattr flag */
1662         spin_lock(&oi->ip_lock);
1663         oi->ip_dyn_features |= flag;
1664         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1665         spin_unlock(&oi->ip_lock);
1666         /* Update inode ctime */
1667         inode->i_ctime = CURRENT_TIME;
1668         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1669         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1670
1671         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1672         if (ret < 0)
1673                 mlog_errno(ret);
1674
1675         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1676                 /*
1677                  * Set value outside in B tree.
1678                  * This is the second step for value size > INLINE_SIZE.
1679                  */
1680                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1681                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1682                                                     &vb, offs);
1683                 if (ret < 0) {
1684                         int ret2;
1685
1686                         mlog_errno(ret);
1687                         /*
1688                          * If set value outside failed, we have to clean
1689                          * the junk tree root we have already set in local.
1690                          */
1691                         ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1692                                                    xi, xs, &vb, offs);
1693                         if (ret2 < 0)
1694                                 mlog_errno(ret2);
1695                 }
1696         }
1697 out:
1698         return ret;
1699 }
1700
1701 static int ocfs2_remove_value_outside(struct inode*inode,
1702                                       struct buffer_head *bh,
1703                                       struct ocfs2_xattr_header *header)
1704 {
1705         int ret = 0, i;
1706         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1707         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1708
1709         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1710
1711         ctxt.handle = ocfs2_start_trans(osb,
1712                                         ocfs2_remove_extent_credits(osb->sb));
1713         if (IS_ERR(ctxt.handle)) {
1714                 ret = PTR_ERR(ctxt.handle);
1715                 mlog_errno(ret);
1716                 goto out;
1717         }
1718
1719         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1720                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1721
1722                 if (!ocfs2_xattr_is_local(entry)) {
1723                         struct ocfs2_xattr_value_buf vb = {
1724                                 .vb_bh = bh,
1725                                 .vb_access = ocfs2_journal_access,
1726                         };
1727                         void *val;
1728
1729                         val = (void *)header +
1730                                 le16_to_cpu(entry->xe_name_offset);
1731                         vb.vb_xv = (struct ocfs2_xattr_value_root *)
1732                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1733                         ret = ocfs2_xattr_value_truncate(inode, &vb, 0, &ctxt);
1734                         if (ret < 0) {
1735                                 mlog_errno(ret);
1736                                 break;
1737                         }
1738                 }
1739         }
1740
1741         ocfs2_commit_trans(osb, ctxt.handle);
1742         ocfs2_schedule_truncate_log_flush(osb, 1);
1743         ocfs2_run_deallocs(osb, &ctxt.dealloc);
1744 out:
1745         return ret;
1746 }
1747
1748 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1749                                     struct buffer_head *di_bh)
1750 {
1751
1752         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1753         struct ocfs2_xattr_header *header;
1754         int ret;
1755
1756         header = (struct ocfs2_xattr_header *)
1757                  ((void *)di + inode->i_sb->s_blocksize -
1758                  le16_to_cpu(di->i_xattr_inline_size));
1759
1760         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1761
1762         return ret;
1763 }
1764
1765 static int ocfs2_xattr_block_remove(struct inode *inode,
1766                                     struct buffer_head *blk_bh)
1767 {
1768         struct ocfs2_xattr_block *xb;
1769         int ret = 0;
1770
1771         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1772         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1773                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1774                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1775         } else
1776                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1777
1778         return ret;
1779 }
1780
1781 static int ocfs2_xattr_free_block(struct inode *inode,
1782                                   u64 block)
1783 {
1784         struct inode *xb_alloc_inode;
1785         struct buffer_head *xb_alloc_bh = NULL;
1786         struct buffer_head *blk_bh = NULL;
1787         struct ocfs2_xattr_block *xb;
1788         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1789         handle_t *handle;
1790         int ret = 0;
1791         u64 blk, bg_blkno;
1792         u16 bit;
1793
1794         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1795         if (ret < 0) {
1796                 mlog_errno(ret);
1797                 goto out;
1798         }
1799
1800         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1801         if (ret < 0) {
1802                 mlog_errno(ret);
1803                 goto out;
1804         }
1805
1806         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1807         blk = le64_to_cpu(xb->xb_blkno);
1808         bit = le16_to_cpu(xb->xb_suballoc_bit);
1809         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1810
1811         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1812                                 EXTENT_ALLOC_SYSTEM_INODE,
1813                                 le16_to_cpu(xb->xb_suballoc_slot));
1814         if (!xb_alloc_inode) {
1815                 ret = -ENOMEM;
1816                 mlog_errno(ret);
1817                 goto out;
1818         }
1819         mutex_lock(&xb_alloc_inode->i_mutex);
1820
1821         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1822         if (ret < 0) {
1823                 mlog_errno(ret);
1824                 goto out_mutex;
1825         }
1826
1827         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1828         if (IS_ERR(handle)) {
1829                 ret = PTR_ERR(handle);
1830                 mlog_errno(ret);
1831                 goto out_unlock;
1832         }
1833
1834         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1835                                        bit, bg_blkno, 1);
1836         if (ret < 0)
1837                 mlog_errno(ret);
1838
1839         ocfs2_commit_trans(osb, handle);
1840 out_unlock:
1841         ocfs2_inode_unlock(xb_alloc_inode, 1);
1842         brelse(xb_alloc_bh);
1843 out_mutex:
1844         mutex_unlock(&xb_alloc_inode->i_mutex);
1845         iput(xb_alloc_inode);
1846 out:
1847         brelse(blk_bh);
1848         return ret;
1849 }
1850
1851 /*
1852  * ocfs2_xattr_remove()
1853  *
1854  * Free extended attribute resources associated with this inode.
1855  */
1856 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1857 {
1858         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1859         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1860         handle_t *handle;
1861         int ret;
1862
1863         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1864                 return 0;
1865
1866         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1867                 return 0;
1868
1869         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1870                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1871                 if (ret < 0) {
1872                         mlog_errno(ret);
1873                         goto out;
1874                 }
1875         }
1876
1877         if (di->i_xattr_loc) {
1878                 ret = ocfs2_xattr_free_block(inode,
1879                                              le64_to_cpu(di->i_xattr_loc));
1880                 if (ret < 0) {
1881                         mlog_errno(ret);
1882                         goto out;
1883                 }
1884         }
1885
1886         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1887                                    OCFS2_INODE_UPDATE_CREDITS);
1888         if (IS_ERR(handle)) {
1889                 ret = PTR_ERR(handle);
1890                 mlog_errno(ret);
1891                 goto out;
1892         }
1893         ret = ocfs2_journal_access(handle, inode, di_bh,
1894                                    OCFS2_JOURNAL_ACCESS_WRITE);
1895         if (ret) {
1896                 mlog_errno(ret);
1897                 goto out_commit;
1898         }
1899
1900         di->i_xattr_loc = 0;
1901
1902         spin_lock(&oi->ip_lock);
1903         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1904         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1905         spin_unlock(&oi->ip_lock);
1906
1907         ret = ocfs2_journal_dirty(handle, di_bh);
1908         if (ret < 0)
1909                 mlog_errno(ret);
1910 out_commit:
1911         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1912 out:
1913         return ret;
1914 }
1915
1916 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1917                                         struct ocfs2_dinode *di)
1918 {
1919         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1920         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1921         int free;
1922
1923         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1924                 return 0;
1925
1926         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1927                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1928                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1929         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1930                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1931                         le64_to_cpu(di->i_size);
1932         } else {
1933                 struct ocfs2_extent_list *el = &di->id2.i_list;
1934                 free = (le16_to_cpu(el->l_count) -
1935                         le16_to_cpu(el->l_next_free_rec)) *
1936                         sizeof(struct ocfs2_extent_rec);
1937         }
1938         if (free >= xattrsize)
1939                 return 1;
1940
1941         return 0;
1942 }
1943
1944 /*
1945  * ocfs2_xattr_ibody_find()
1946  *
1947  * Find extended attribute in inode block and
1948  * fill search info into struct ocfs2_xattr_search.
1949  */
1950 static int ocfs2_xattr_ibody_find(struct inode *inode,
1951                                   int name_index,
1952                                   const char *name,
1953                                   struct ocfs2_xattr_search *xs)
1954 {
1955         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1956         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1957         int ret;
1958         int has_space = 0;
1959
1960         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1961                 return 0;
1962
1963         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1964                 down_read(&oi->ip_alloc_sem);
1965                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1966                 up_read(&oi->ip_alloc_sem);
1967                 if (!has_space)
1968                         return 0;
1969         }
1970
1971         xs->xattr_bh = xs->inode_bh;
1972         xs->end = (void *)di + inode->i_sb->s_blocksize;
1973         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1974                 xs->header = (struct ocfs2_xattr_header *)
1975                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1976         else
1977                 xs->header = (struct ocfs2_xattr_header *)
1978                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1979         xs->base = (void *)xs->header;
1980         xs->here = xs->header->xh_entries;
1981
1982         /* Find the named attribute. */
1983         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1984                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1985                 if (ret && ret != -ENODATA)
1986                         return ret;
1987                 xs->not_found = ret;
1988         }
1989
1990         return 0;
1991 }
1992
1993 /*
1994  * ocfs2_xattr_ibody_set()
1995  *
1996  * Set, replace or remove an extended attribute into inode block.
1997  *
1998  */
1999 static int ocfs2_xattr_ibody_set(struct inode *inode,
2000                                  struct ocfs2_xattr_info *xi,
2001                                  struct ocfs2_xattr_search *xs,
2002                                  struct ocfs2_xattr_set_ctxt *ctxt)
2003 {
2004         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2005         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2006         int ret;
2007
2008         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2009                 return -ENOSPC;
2010
2011         down_write(&oi->ip_alloc_sem);
2012         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2013                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2014                         ret = -ENOSPC;
2015                         goto out;
2016                 }
2017         }
2018
2019         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2020                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2021 out:
2022         up_write(&oi->ip_alloc_sem);
2023
2024         return ret;
2025 }
2026
2027 /*
2028  * ocfs2_xattr_block_find()
2029  *
2030  * Find extended attribute in external block and
2031  * fill search info into struct ocfs2_xattr_search.
2032  */
2033 static int ocfs2_xattr_block_find(struct inode *inode,
2034                                   int name_index,
2035                                   const char *name,
2036                                   struct ocfs2_xattr_search *xs)
2037 {
2038         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2039         struct buffer_head *blk_bh = NULL;
2040         struct ocfs2_xattr_block *xb;
2041         int ret = 0;
2042
2043         if (!di->i_xattr_loc)
2044                 return ret;
2045
2046         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2047                                      &blk_bh);
2048         if (ret < 0) {
2049                 mlog_errno(ret);
2050                 return ret;
2051         }
2052
2053         xs->xattr_bh = blk_bh;
2054         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2055
2056         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2057                 xs->header = &xb->xb_attrs.xb_header;
2058                 xs->base = (void *)xs->header;
2059                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2060                 xs->here = xs->header->xh_entries;
2061
2062                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2063         } else
2064                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2065                                                    name_index,
2066                                                    name, xs);
2067
2068         if (ret && ret != -ENODATA) {
2069                 xs->xattr_bh = NULL;
2070                 goto cleanup;
2071         }
2072         xs->not_found = ret;
2073         return 0;
2074 cleanup:
2075         brelse(blk_bh);
2076
2077         return ret;
2078 }
2079
2080 /*
2081  * ocfs2_xattr_block_set()
2082  *
2083  * Set, replace or remove an extended attribute into external block.
2084  *
2085  */
2086 static int ocfs2_xattr_block_set(struct inode *inode,
2087                                  struct ocfs2_xattr_info *xi,
2088                                  struct ocfs2_xattr_search *xs,
2089                                  struct ocfs2_xattr_set_ctxt *ctxt)
2090 {
2091         struct buffer_head *new_bh = NULL;
2092         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2093         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
2094         handle_t *handle = ctxt->handle;
2095         struct ocfs2_xattr_block *xblk = NULL;
2096         u16 suballoc_bit_start;
2097         u32 num_got;
2098         u64 first_blkno;
2099         int ret;
2100
2101         if (!xs->xattr_bh) {
2102                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
2103                                            OCFS2_JOURNAL_ACCESS_CREATE);
2104                 if (ret < 0) {
2105                         mlog_errno(ret);
2106                         goto end;
2107                 }
2108
2109                 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2110                                            &suballoc_bit_start, &num_got,
2111                                            &first_blkno);
2112                 if (ret < 0) {
2113                         mlog_errno(ret);
2114                         goto end;
2115                 }
2116
2117                 new_bh = sb_getblk(inode->i_sb, first_blkno);
2118                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2119
2120                 ret = ocfs2_journal_access(handle, inode, new_bh,
2121                                            OCFS2_JOURNAL_ACCESS_CREATE);
2122                 if (ret < 0) {
2123                         mlog_errno(ret);
2124                         goto end;
2125                 }
2126
2127                 /* Initialize ocfs2_xattr_block */
2128                 xs->xattr_bh = new_bh;
2129                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2130                 memset(xblk, 0, inode->i_sb->s_blocksize);
2131                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2132                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2133                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2134                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2135                 xblk->xb_blkno = cpu_to_le64(first_blkno);
2136
2137                 xs->header = &xblk->xb_attrs.xb_header;
2138                 xs->base = (void *)xs->header;
2139                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2140                 xs->here = xs->header->xh_entries;
2141
2142                 ret = ocfs2_journal_dirty(handle, new_bh);
2143                 if (ret < 0) {
2144                         mlog_errno(ret);
2145                         goto end;
2146                 }
2147                 di->i_xattr_loc = cpu_to_le64(first_blkno);
2148                 ocfs2_journal_dirty(handle, xs->inode_bh);
2149         } else
2150                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2151
2152         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2153                 /* Set extended attribute into external block */
2154                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2155                                             OCFS2_HAS_XATTR_FL);
2156                 if (!ret || ret != -ENOSPC)
2157                         goto end;
2158
2159                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2160                 if (ret)
2161                         goto end;
2162         }
2163
2164         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2165
2166 end:
2167
2168         return ret;
2169 }
2170
2171 /* Check whether the new xattr can be inserted into the inode. */
2172 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2173                                        struct ocfs2_xattr_info *xi,
2174                                        struct ocfs2_xattr_search *xs)
2175 {
2176         u64 value_size;
2177         struct ocfs2_xattr_entry *last;
2178         int free, i;
2179         size_t min_offs = xs->end - xs->base;
2180
2181         if (!xs->header)
2182                 return 0;
2183
2184         last = xs->header->xh_entries;
2185
2186         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2187                 size_t offs = le16_to_cpu(last->xe_name_offset);
2188                 if (offs < min_offs)
2189                         min_offs = offs;
2190                 last += 1;
2191         }
2192
2193         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
2194         if (free < 0)
2195                 return 0;
2196
2197         BUG_ON(!xs->not_found);
2198
2199         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2200                 value_size = OCFS2_XATTR_ROOT_SIZE;
2201         else
2202                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2203
2204         if (free >= sizeof(struct ocfs2_xattr_entry) +
2205                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2206                 return 1;
2207
2208         return 0;
2209 }
2210
2211 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2212                                      struct ocfs2_dinode *di,
2213                                      struct ocfs2_xattr_info *xi,
2214                                      struct ocfs2_xattr_search *xis,
2215                                      struct ocfs2_xattr_search *xbs,
2216                                      int *clusters_need,
2217                                      int *meta_need,
2218                                      int *credits_need)
2219 {
2220         int ret = 0, old_in_xb = 0;
2221         int clusters_add = 0, meta_add = 0, credits = 0;
2222         struct buffer_head *bh = NULL;
2223         struct ocfs2_xattr_block *xb = NULL;
2224         struct ocfs2_xattr_entry *xe = NULL;
2225         struct ocfs2_xattr_value_root *xv = NULL;
2226         char *base = NULL;
2227         int name_offset, name_len = 0;
2228         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2229                                                     xi->value_len);
2230         u64 value_size;
2231
2232         if (xis->not_found && xbs->not_found) {
2233                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2234
2235                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2236                         clusters_add += new_clusters;
2237                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2238                                                         &def_xv.xv.xr_list,
2239                                                         new_clusters);
2240                 }
2241
2242                 goto meta_guess;
2243         }
2244
2245         if (!xis->not_found) {
2246                 xe = xis->here;
2247                 name_offset = le16_to_cpu(xe->xe_name_offset);
2248                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2249                 base = xis->base;
2250                 credits += OCFS2_INODE_UPDATE_CREDITS;
2251         } else {
2252                 int i, block_off = 0;
2253                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2254                 xe = xbs->here;
2255                 name_offset = le16_to_cpu(xe->xe_name_offset);
2256                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2257                 i = xbs->here - xbs->header->xh_entries;
2258                 old_in_xb = 1;
2259
2260                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2261                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2262                                                         bucket_xh(xbs->bucket),
2263                                                         i, &block_off,
2264                                                         &name_offset);
2265                         base = bucket_block(xbs->bucket, block_off);
2266                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2267                 } else {
2268                         base = xbs->base;
2269                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2270                 }
2271         }
2272
2273         /*
2274          * delete a xattr doesn't need metadata and cluster allocation.
2275          * so just calculate the credits and return.
2276          *
2277          * The credits for removing the value tree will be extended
2278          * by ocfs2_remove_extent itself.
2279          */
2280         if (!xi->value) {
2281                 if (!ocfs2_xattr_is_local(xe))
2282                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2283
2284                 goto out;
2285         }
2286
2287         /* do cluster allocation guess first. */
2288         value_size = le64_to_cpu(xe->xe_value_size);
2289
2290         if (old_in_xb) {
2291                 /*
2292                  * In xattr set, we always try to set the xe in inode first,
2293                  * so if it can be inserted into inode successfully, the old
2294                  * one will be removed from the xattr block, and this xattr
2295                  * will be inserted into inode as a new xattr in inode.
2296                  */
2297                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2298                         clusters_add += new_clusters;
2299                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
2300                                     OCFS2_INODE_UPDATE_CREDITS;
2301                         if (!ocfs2_xattr_is_local(xe))
2302                                 credits += ocfs2_calc_extend_credits(
2303                                                         inode->i_sb,
2304                                                         &def_xv.xv.xr_list,
2305                                                         new_clusters);
2306                         goto out;
2307                 }
2308         }
2309
2310         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2311                 /* the new values will be stored outside. */
2312                 u32 old_clusters = 0;
2313
2314                 if (!ocfs2_xattr_is_local(xe)) {
2315                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2316                                                                  value_size);
2317                         xv = (struct ocfs2_xattr_value_root *)
2318                              (base + name_offset + name_len);
2319                         value_size = OCFS2_XATTR_ROOT_SIZE;
2320                 } else
2321                         xv = &def_xv.xv;
2322
2323                 if (old_clusters >= new_clusters) {
2324                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2325                         goto out;
2326                 } else {
2327                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2328                         clusters_add += new_clusters - old_clusters;
2329                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2330                                                              &xv->xr_list,
2331                                                              new_clusters -
2332                                                              old_clusters);
2333                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2334                                 goto out;
2335                 }
2336         } else {
2337                 /*
2338                  * Now the new value will be stored inside. So if the new
2339                  * value is smaller than the size of value root or the old
2340                  * value, we don't need any allocation, otherwise we have
2341                  * to guess metadata allocation.
2342                  */
2343                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2344                     (!ocfs2_xattr_is_local(xe) &&
2345                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2346                         goto out;
2347         }
2348
2349 meta_guess:
2350         /* calculate metadata allocation. */
2351         if (di->i_xattr_loc) {
2352                 if (!xbs->xattr_bh) {
2353                         ret = ocfs2_read_xattr_block(inode,
2354                                                      le64_to_cpu(di->i_xattr_loc),
2355                                                      &bh);
2356                         if (ret) {
2357                                 mlog_errno(ret);
2358                                 goto out;
2359                         }
2360
2361                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2362                 } else
2363                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2364
2365                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2366                         struct ocfs2_extent_list *el =
2367                                  &xb->xb_attrs.xb_root.xt_list;
2368                         meta_add += ocfs2_extend_meta_needed(el);
2369                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2370                                                              el, 1);
2371                 }
2372
2373                 /*
2374                  * This cluster will be used either for new bucket or for
2375                  * new xattr block.
2376                  * If the cluster size is the same as the bucket size, one
2377                  * more is needed since we may need to extend the bucket
2378                  * also.
2379                  */
2380                 clusters_add += 1;
2381                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2382                 if (OCFS2_XATTR_BUCKET_SIZE ==
2383                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2384                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2385                         clusters_add += 1;
2386                 }
2387         } else {
2388                 meta_add += 1;
2389                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2390         }
2391 out:
2392         if (clusters_need)
2393                 *clusters_need = clusters_add;
2394         if (meta_need)
2395                 *meta_need = meta_add;
2396         if (credits_need)
2397                 *credits_need = credits;
2398         brelse(bh);
2399         return ret;
2400 }
2401
2402 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2403                                      struct ocfs2_dinode *di,
2404                                      struct ocfs2_xattr_info *xi,
2405                                      struct ocfs2_xattr_search *xis,
2406                                      struct ocfs2_xattr_search *xbs,
2407                                      struct ocfs2_xattr_set_ctxt *ctxt,
2408                                      int *credits)
2409 {
2410         int clusters_add, meta_add, ret;
2411         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2412
2413         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2414
2415         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2416
2417         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2418                                         &clusters_add, &meta_add, credits);
2419         if (ret) {
2420                 mlog_errno(ret);
2421                 return ret;
2422         }
2423
2424         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2425              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2426
2427         if (meta_add) {
2428                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2429                                                         &ctxt->meta_ac);
2430                 if (ret) {
2431                         mlog_errno(ret);
2432                         goto out;
2433                 }
2434         }
2435
2436         if (clusters_add) {
2437                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2438                 if (ret)
2439                         mlog_errno(ret);
2440         }
2441 out:
2442         if (ret) {
2443                 if (ctxt->meta_ac) {
2444                         ocfs2_free_alloc_context(ctxt->meta_ac);
2445                         ctxt->meta_ac = NULL;
2446                 }
2447
2448                 /*
2449                  * We cannot have an error and a non null ctxt->data_ac.
2450                  */
2451         }
2452
2453         return ret;
2454 }
2455
2456 static int __ocfs2_xattr_set_handle(struct inode *inode,
2457                                     struct ocfs2_dinode *di,
2458                                     struct ocfs2_xattr_info *xi,
2459                                     struct ocfs2_xattr_search *xis,
2460                                     struct ocfs2_xattr_search *xbs,
2461                                     struct ocfs2_xattr_set_ctxt *ctxt)
2462 {
2463         int ret = 0, credits, old_found;
2464
2465         if (!xi->value) {
2466                 /* Remove existing extended attribute */
2467                 if (!xis->not_found)
2468                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2469                 else if (!xbs->not_found)
2470                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2471         } else {
2472                 /* We always try to set extended attribute into inode first*/
2473                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2474                 if (!ret && !xbs->not_found) {
2475                         /*
2476                          * If succeed and that extended attribute existing in
2477                          * external block, then we will remove it.
2478                          */
2479                         xi->value = NULL;
2480                         xi->value_len = 0;
2481
2482                         old_found = xis->not_found;
2483                         xis->not_found = -ENODATA;
2484                         ret = ocfs2_calc_xattr_set_need(inode,
2485                                                         di,
2486                                                         xi,
2487                                                         xis,
2488                                                         xbs,
2489                                                         NULL,
2490                                                         NULL,
2491                                                         &credits);
2492                         xis->not_found = old_found;
2493                         if (ret) {
2494                                 mlog_errno(ret);
2495                                 goto out;
2496                         }
2497
2498                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2499                                         ctxt->handle->h_buffer_credits);
2500                         if (ret) {
2501                                 mlog_errno(ret);
2502                                 goto out;
2503                         }
2504                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2505                 } else if (ret == -ENOSPC) {
2506                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2507                                 ret = ocfs2_xattr_block_find(inode,
2508                                                              xi->name_index,
2509                                                              xi->name, xbs);
2510                                 if (ret)
2511                                         goto out;
2512
2513                                 old_found = xis->not_found;
2514                                 xis->not_found = -ENODATA;
2515                                 ret = ocfs2_calc_xattr_set_need(inode,
2516                                                                 di,
2517                                                                 xi,
2518                                                                 xis,
2519                                                                 xbs,
2520                                                                 NULL,
2521                                                                 NULL,
2522                                                                 &credits);
2523                                 xis->not_found = old_found;
2524                                 if (ret) {
2525                                         mlog_errno(ret);
2526                                         goto out;
2527                                 }
2528
2529                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2530                                         ctxt->handle->h_buffer_credits);
2531                                 if (ret) {
2532                                         mlog_errno(ret);
2533                                         goto out;
2534                                 }
2535                         }
2536                         /*
2537                          * If no space in inode, we will set extended attribute
2538                          * into external block.
2539                          */
2540                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2541                         if (ret)
2542                                 goto out;
2543                         if (!xis->not_found) {
2544                                 /*
2545                                  * If succeed and that extended attribute
2546                                  * existing in inode, we will remove it.
2547                                  */
2548                                 xi->value = NULL;
2549                                 xi->value_len = 0;
2550                                 xbs->not_found = -ENODATA;
2551                                 ret = ocfs2_calc_xattr_set_need(inode,
2552                                                                 di,
2553                                                                 xi,
2554                                                                 xis,
2555                                                                 xbs,
2556                                                                 NULL,
2557                                                                 NULL,
2558                                                                 &credits);
2559                                 if (ret) {
2560                                         mlog_errno(ret);
2561                                         goto out;
2562                                 }
2563
2564                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2565                                                 ctxt->handle->h_buffer_credits);
2566                                 if (ret) {
2567                                         mlog_errno(ret);
2568                                         goto out;
2569                                 }
2570                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2571                                                             xis, ctxt);
2572                         }
2573                 }
2574         }
2575
2576 out:
2577         return ret;
2578 }
2579
2580 /*
2581  * This function only called duing creating inode
2582  * for init security/acl xattrs of the new inode.
2583  * The xattrs could be put into ibody or extent block,
2584  * xattr bucket would not be use in this case.
2585  * transanction credits also be reserved in here.
2586  */
2587 int ocfs2_xattr_set_handle(handle_t *handle,
2588                            struct inode *inode,
2589                            struct buffer_head *di_bh,
2590                            int name_index,
2591                            const char *name,
2592                            const void *value,
2593                            size_t value_len,
2594                            int flags,
2595                            struct ocfs2_alloc_context *meta_ac,
2596                            struct ocfs2_alloc_context *data_ac)
2597 {
2598         struct ocfs2_dinode *di;
2599         int ret;
2600
2601         struct ocfs2_xattr_info xi = {
2602                 .name_index = name_index,
2603                 .name = name,
2604                 .value = value,
2605                 .value_len = value_len,
2606         };
2607
2608         struct ocfs2_xattr_search xis = {
2609                 .not_found = -ENODATA,
2610         };
2611
2612         struct ocfs2_xattr_search xbs = {
2613                 .not_found = -ENODATA,
2614         };
2615
2616         struct ocfs2_xattr_set_ctxt ctxt = {
2617                 .handle = handle,
2618                 .meta_ac = meta_ac,
2619                 .data_ac = data_ac,
2620         };
2621
2622         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2623                 return -EOPNOTSUPP;
2624
2625         xis.inode_bh = xbs.inode_bh = di_bh;
2626         di = (struct ocfs2_dinode *)di_bh->b_data;
2627
2628         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2629
2630         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2631         if (ret)
2632                 goto cleanup;
2633         if (xis.not_found) {
2634                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2635                 if (ret)
2636                         goto cleanup;
2637         }
2638
2639         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2640
2641 cleanup:
2642         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2643         brelse(xbs.xattr_bh);
2644
2645         return ret;
2646 }
2647
2648 /*
2649  * ocfs2_xattr_set()
2650  *
2651  * Set, replace or remove an extended attribute for this inode.
2652  * value is NULL to remove an existing extended attribute, else either
2653  * create or replace an extended attribute.
2654  */
2655 int ocfs2_xattr_set(struct inode *inode,
2656                     int name_index,
2657                     const char *name,
2658                     const void *value,
2659                     size_t value_len,
2660                     int flags)
2661 {
2662         struct buffer_head *di_bh = NULL;
2663         struct ocfs2_dinode *di;
2664         int ret, credits;
2665         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2666         struct inode *tl_inode = osb->osb_tl_inode;
2667         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2668
2669         struct ocfs2_xattr_info xi = {
2670                 .name_index = name_index,
2671                 .name = name,
2672                 .value = value,
2673                 .value_len = value_len,
2674         };
2675
2676         struct ocfs2_xattr_search xis = {
2677                 .not_found = -ENODATA,
2678         };
2679
2680         struct ocfs2_xattr_search xbs = {
2681                 .not_found = -ENODATA,
2682         };
2683
2684         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2685                 return -EOPNOTSUPP;
2686
2687         /*
2688          * Only xbs will be used on indexed trees.  xis doesn't need a
2689          * bucket.
2690          */
2691         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2692         if (!xbs.bucket) {
2693                 mlog_errno(-ENOMEM);
2694                 return -ENOMEM;
2695         }
2696
2697         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2698         if (ret < 0) {
2699                 mlog_errno(ret);
2700                 goto cleanup_nolock;
2701         }
2702         xis.inode_bh = xbs.inode_bh = di_bh;
2703         di = (struct ocfs2_dinode *)di_bh->b_data;
2704
2705         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2706         /*
2707          * Scan inode and external block to find the same name
2708          * extended attribute and collect search infomation.
2709          */
2710         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2711         if (ret)
2712                 goto cleanup;
2713         if (xis.not_found) {
2714                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2715                 if (ret)
2716                         goto cleanup;
2717         }
2718
2719         if (xis.not_found && xbs.not_found) {
2720                 ret = -ENODATA;
2721                 if (flags & XATTR_REPLACE)
2722                         goto cleanup;
2723                 ret = 0;
2724                 if (!value)
2725                         goto cleanup;
2726         } else {
2727                 ret = -EEXIST;
2728                 if (flags & XATTR_CREATE)
2729                         goto cleanup;
2730         }
2731
2732
2733         mutex_lock(&tl_inode->i_mutex);
2734
2735         if (ocfs2_truncate_log_needs_flush(osb)) {
2736                 ret = __ocfs2_flush_truncate_log(osb);
2737                 if (ret < 0) {
2738                         mutex_unlock(&tl_inode->i_mutex);
2739                         mlog_errno(ret);
2740                         goto cleanup;
2741                 }
2742         }
2743         mutex_unlock(&tl_inode->i_mutex);
2744
2745         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2746                                         &xbs, &ctxt, &credits);
2747         if (ret) {
2748                 mlog_errno(ret);
2749                 goto cleanup;
2750         }
2751
2752         ctxt.handle = ocfs2_start_trans(osb, credits);
2753         if (IS_ERR(ctxt.handle)) {
2754                 ret = PTR_ERR(ctxt.handle);
2755                 mlog_errno(ret);
2756                 goto cleanup;
2757         }
2758
2759         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2760
2761         ocfs2_commit_trans(osb, ctxt.handle);
2762
2763         if (ctxt.data_ac)
2764                 ocfs2_free_alloc_context(ctxt.data_ac);
2765         if (ctxt.meta_ac)
2766                 ocfs2_free_alloc_context(ctxt.meta_ac);
2767         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2768                 ocfs2_schedule_truncate_log_flush(osb, 1);
2769         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2770 cleanup:
2771         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2772         ocfs2_inode_unlock(inode, 1);
2773 cleanup_nolock:
2774         brelse(di_bh);
2775         brelse(xbs.xattr_bh);
2776         ocfs2_xattr_bucket_free(xbs.bucket);
2777
2778         return ret;
2779 }
2780
2781 /*
2782  * Find the xattr extent rec which may contains name_hash.
2783  * e_cpos will be the first name hash of the xattr rec.
2784  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2785  */
2786 static int ocfs2_xattr_get_rec(struct inode *inode,
2787                                u32 name_hash,
2788                                u64 *p_blkno,
2789                                u32 *e_cpos,
2790                                u32 *num_clusters,
2791                                struct ocfs2_extent_list *el)
2792 {
2793         int ret = 0, i;
2794         struct buffer_head *eb_bh = NULL;
2795         struct ocfs2_extent_block *eb;
2796         struct ocfs2_extent_rec *rec = NULL;
2797         u64 e_blkno = 0;
2798
2799         if (el->l_tree_depth) {
2800                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2801                 if (ret) {
2802                         mlog_errno(ret);
2803                         goto out;
2804                 }
2805
2806                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2807                 el = &eb->h_list;
2808
2809                 if (el->l_tree_depth) {
2810                         ocfs2_error(inode->i_sb,
2811                                     "Inode %lu has non zero tree depth in "
2812                                     "xattr tree block %llu\n", inode->i_ino,
2813                                     (unsigned long long)eb_bh->b_blocknr);
2814                         ret = -EROFS;
2815                         goto out;
2816                 }
2817         }
2818
2819         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2820                 rec = &el->l_recs[i];
2821
2822                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2823                         e_blkno = le64_to_cpu(rec->e_blkno);
2824                         break;
2825                 }
2826         }
2827
2828         if (!e_blkno) {
2829                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2830                             "record (%u, %u, 0) in xattr", inode->i_ino,
2831                             le32_to_cpu(rec->e_cpos),
2832                             ocfs2_rec_clusters(el, rec));
2833                 ret = -EROFS;
2834                 goto out;
2835         }
2836
2837         *p_blkno = le64_to_cpu(rec->e_blkno);
2838         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2839         if (e_cpos)
2840                 *e_cpos = le32_to_cpu(rec->e_cpos);
2841 out:
2842         brelse(eb_bh);
2843         return ret;
2844 }
2845
2846 typedef int (xattr_bucket_func)(struct inode *inode,
2847                                 struct ocfs2_xattr_bucket *bucket,
2848                                 void *para);
2849
2850 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2851                                    struct ocfs2_xattr_bucket *bucket,
2852                                    int name_index,
2853                                    const char *name,
2854                                    u32 name_hash,
2855                                    u16 *xe_index,
2856                                    int *found)
2857 {
2858         int i, ret = 0, cmp = 1, block_off, new_offset;
2859         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2860         size_t name_len = strlen(name);
2861         struct ocfs2_xattr_entry *xe = NULL;
2862         char *xe_name;
2863
2864         /*
2865          * We don't use binary search in the bucket because there
2866          * may be multiple entries with the same name hash.
2867          */
2868         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2869                 xe = &xh->xh_entries[i];
2870
2871                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2872                         continue;
2873                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2874                         break;
2875
2876                 cmp = name_index - ocfs2_xattr_get_type(xe);
2877                 if (!cmp)
2878                         cmp = name_len - xe->xe_name_len;
2879                 if (cmp)
2880                         continue;
2881
2882                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2883                                                         xh,
2884                                                         i,
2885                                                         &block_off,
2886                                                         &new_offset);
2887                 if (ret) {
2888                         mlog_errno(ret);
2889                         break;
2890                 }
2891
2892
2893                 xe_name = bucket_block(bucket, block_off) + new_offset;
2894                 if (!memcmp(name, xe_name, name_len)) {
2895                         *xe_index = i;
2896                         *found = 1;
2897                         ret = 0;
2898                         break;
2899                 }
2900         }
2901
2902         return ret;
2903 }
2904
2905 /*
2906  * Find the specified xattr entry in a series of buckets.
2907  * This series start from p_blkno and last for num_clusters.
2908  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2909  * the num of the valid buckets.
2910  *
2911  * Return the buffer_head this xattr should reside in. And if the xattr's
2912  * hash is in the gap of 2 buckets, return the lower bucket.
2913  */
2914 static int ocfs2_xattr_bucket_find(struct inode *inode,
2915                                    int name_index,
2916                                    const char *name,
2917                                    u32 name_hash,
2918                                    u64 p_blkno,
2919                                    u32 first_hash,
2920                                    u32 num_clusters,
2921                                    struct ocfs2_xattr_search *xs)
2922 {
2923         int ret, found = 0;
2924         struct ocfs2_xattr_header *xh = NULL;
2925         struct ocfs2_xattr_entry *xe = NULL;
2926         u16 index = 0;
2927         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2928         int low_bucket = 0, bucket, high_bucket;
2929         struct ocfs2_xattr_bucket *search;
2930         u32 last_hash;
2931         u64 blkno, lower_blkno = 0;
2932
2933         search = ocfs2_xattr_bucket_new(inode);
2934         if (!search) {
2935                 ret = -ENOMEM;
2936                 mlog_errno(ret);
2937                 goto out;
2938         }
2939
2940         ret = ocfs2_read_xattr_bucket(search, p_blkno);
2941         if (ret) {
2942                 mlog_errno(ret);
2943                 goto out;
2944         }
2945
2946         xh = bucket_xh(search);
2947         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2948         while (low_bucket <= high_bucket) {
2949                 ocfs2_xattr_bucket_relse(search);
2950
2951                 bucket = (low_bucket + high_bucket) / 2;
2952                 blkno = p_blkno + bucket * blk_per_bucket;
2953                 ret = ocfs2_read_xattr_bucket(search, blkno);
2954                 if (ret) {
2955                         mlog_errno(ret);
2956                         goto out;
2957                 }
2958
2959                 xh = bucket_xh(search);
2960                 xe = &xh->xh_entries[0];
2961                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2962                         high_bucket = bucket - 1;
2963                         continue;
2964                 }
2965
2966                 /*
2967                  * Check whether the hash of the last entry in our
2968                  * bucket is larger than the search one. for an empty
2969                  * bucket, the last one is also the first one.
2970                  */
2971                 if (xh->xh_count)
2972                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2973
2974                 last_hash = le32_to_cpu(xe->xe_name_hash);
2975
2976                 /* record lower_blkno which may be the insert place. */
2977                 lower_blkno = blkno;
2978
2979                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2980                         low_bucket = bucket + 1;
2981                         continue;
2982                 }
2983
2984                 /* the searched xattr should reside in this bucket if exists. */
2985                 ret = ocfs2_find_xe_in_bucket(inode, search,
2986                                               name_index, name, name_hash,
2987                                               &index, &found);
2988                 if (ret) {
2989                         mlog_errno(ret);
2990                         goto out;
2991                 }
2992                 break;
2993         }
2994
2995         /*
2996          * Record the bucket we have found.
2997          * When the xattr's hash value is in the gap of 2 buckets, we will
2998          * always set it to the previous bucket.
2999          */
3000         if (!lower_blkno)
3001                 lower_blkno = p_blkno;
3002
3003         /* This should be in cache - we just read it during the search */
3004         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3005         if (ret) {
3006                 mlog_errno(ret);
3007                 goto out;
3008         }
3009
3010         xs->header = bucket_xh(xs->bucket);
3011         xs->base = bucket_block(xs->bucket, 0);
3012         xs->end = xs->base + inode->i_sb->s_blocksize;
3013
3014         if (found) {
3015                 xs->here = &xs->header->xh_entries[index];
3016                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3017                      (unsigned long long)bucket_blkno(xs->bucket), index);
3018         } else
3019                 ret = -ENODATA;
3020
3021 out:
3022         ocfs2_xattr_bucket_free(search);
3023         return ret;
3024 }
3025
3026 static int ocfs2_xattr_index_block_find(struct inode *inode,
3027                                         struct buffer_head *root_bh,
3028                                         int name_index,
3029                                         const char *name,
3030                                         struct ocfs2_xattr_search *xs)
3031 {
3032         int ret;
3033         struct ocfs2_xattr_block *xb =
3034                         (struct ocfs2_xattr_block *)root_bh->b_data;
3035         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3036         struct ocfs2_extent_list *el = &xb_root->xt_list;
3037         u64 p_blkno = 0;
3038         u32 first_hash, num_clusters = 0;
3039         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3040
3041         if (le16_to_cpu(el->l_next_free_rec) == 0)
3042                 return -ENODATA;
3043
3044         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3045              name, name_hash, name_index);
3046
3047         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3048                                   &num_clusters, el);
3049         if (ret) {
3050                 mlog_errno(ret);
3051                 goto out;
3052         }
3053
3054         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3055
3056         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3057              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3058              first_hash);
3059
3060         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3061                                       p_blkno, first_hash, num_clusters, xs);
3062
3063 out:
3064         return ret;
3065 }
3066
3067 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3068                                        u64 blkno,
3069                                        u32 clusters,
3070                                        xattr_bucket_func *func,
3071                                        void *para)
3072 {
3073         int i, ret = 0;
3074         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3075         u32 num_buckets = clusters * bpc;
3076         struct ocfs2_xattr_bucket *bucket;
3077
3078         bucket = ocfs2_xattr_bucket_new(inode);
3079         if (!bucket) {
3080                 mlog_errno(-ENOMEM);
3081                 return -ENOMEM;
3082         }
3083
3084         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3085              clusters, (unsigned long long)blkno);
3086
3087         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3088                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3089                 if (ret) {
3090                         mlog_errno(ret);
3091                         break;
3092                 }
3093
3094                 /*
3095                  * The real bucket num in this series of blocks is stored
3096                  * in the 1st bucket.
3097                  */
3098                 if (i == 0)
3099                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3100
3101                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3102                      (unsigned long long)blkno,
3103                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3104                 if (func) {
3105                         ret = func(inode, bucket, para);
3106                         if (ret)
3107                                 mlog_errno(ret);
3108                         /* Fall through to bucket_relse() */
3109                 }
3110
3111                 ocfs2_xattr_bucket_relse(bucket);
3112                 if (ret)
3113                         break;
3114         }
3115
3116         ocfs2_xattr_bucket_free(bucket);
3117         return ret;
3118 }
3119
3120 struct ocfs2_xattr_tree_list {
3121         char *buffer;
3122         size_t buffer_size;
3123         size_t result;
3124 };
3125
3126 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3127                                              struct ocfs2_xattr_header *xh,
3128                                              int index,
3129                                              int *block_off,
3130                                              int *new_offset)
3131 {
3132         u16 name_offset;
3133
3134         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3135                 return -EINVAL;
3136
3137         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3138
3139         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3140         *new_offset = name_offset % inode->i_sb->s_blocksize;
3141
3142         return 0;
3143 }
3144
3145 static int ocfs2_list_xattr_bucket(struct inode *inode,
3146                                    struct ocfs2_xattr_bucket *bucket,
3147                                    void *para)
3148 {
3149         int ret = 0, type;
3150         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3151         int i, block_off, new_offset;
3152         const char *prefix, *name;
3153
3154         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3155                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3156                 type = ocfs2_xattr_get_type(entry);
3157                 prefix = ocfs2_xattr_prefix(type);
3158
3159                 if (prefix) {
3160                         ret = ocfs2_xattr_bucket_get_name_value(inode,
3161                                                                 bucket_xh(bucket),
3162                                                                 i,
3163                                                                 &block_off,
3164                                                                 &new_offset);
3165                         if (ret)
3166                                 break;
3167
3168                         name = (const char *)bucket_block(bucket, block_off) +
3169                                 new_offset;
3170                         ret = ocfs2_xattr_list_entry(xl->buffer,
3171                                                      xl->buffer_size,
3172                                                      &xl->result,
3173                                                      prefix, name,
3174                                                      entry->xe_name_len);
3175                         if (ret)
3176                                 break;
3177                 }
3178         }
3179
3180         return ret;
3181 }
3182
3183 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3184                                              struct ocfs2_xattr_tree_root *xt,
3185                                              char *buffer,
3186                                              size_t buffer_size)
3187 {
3188         struct ocfs2_extent_list *el = &xt->xt_list;
3189         int ret = 0;
3190         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3191         u64 p_blkno = 0;
3192         struct ocfs2_xattr_tree_list xl = {
3193                 .buffer = buffer,
3194                 .buffer_size = buffer_size,
3195                 .result = 0,
3196         };
3197
3198         if (le16_to_cpu(el->l_next_free_rec) == 0)
3199                 return 0;
3200
3201         while (name_hash > 0) {
3202                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3203                                           &e_cpos, &num_clusters, el);
3204                 if (ret) {
3205                         mlog_errno(ret);
3206                         goto out;
3207                 }
3208
3209                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3210                                                   ocfs2_list_xattr_bucket,
3211                                                   &xl);
3212                 if (ret) {
3213                         mlog_errno(ret);
3214                         goto out;
3215                 }
3216
3217                 if (e_cpos == 0)
3218                         break;
3219
3220                 name_hash = e_cpos - 1;
3221         }
3222
3223         ret = xl.result;
3224 out:
3225         return ret;
3226 }
3227
3228 static int cmp_xe(const void *a, const void *b)
3229 {
3230         const struct ocfs2_xattr_entry *l = a, *r = b;
3231         u32 l_hash = le32_to_cpu(l->xe_name_hash);
3232         u32 r_hash = le32_to_cpu(r->xe_name_hash);
3233
3234         if (l_hash > r_hash)
3235                 return 1;
3236         if (l_hash < r_hash)
3237                 return -1;
3238         return 0;
3239 }
3240
3241 static void swap_xe(void *a, void *b, int size)
3242 {
3243         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3244
3245         tmp = *l;
3246         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3247         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3248 }
3249
3250 /*
3251  * When the ocfs2_xattr_block is filled up, new bucket will be created
3252  * and all the xattr entries will be moved to the new bucket.
3253  * The header goes at the start of the bucket, and the names+values are
3254  * filled from the end.  This is why *target starts as the last buffer.
3255  * Note: we need to sort the entries since they are not saved in order
3256  * in the ocfs2_xattr_block.
3257  */
3258 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3259                                            struct buffer_head *xb_bh,
3260                                            struct ocfs2_xattr_bucket *bucket)
3261 {
3262         int i, blocksize = inode->i_sb->s_blocksize;
3263         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3264         u16 offset, size, off_change;
3265         struct ocfs2_xattr_entry *xe;
3266         struct ocfs2_xattr_block *xb =
3267                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
3268         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3269         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3270         u16 count = le16_to_cpu(xb_xh->xh_count);
3271         char *src = xb_bh->b_data;
3272         char *target = bucket_block(bucket, blks - 1);
3273
3274         mlog(0, "cp xattr from block %llu to bucket %llu\n",
3275              (unsigned long long)xb_bh->b_blocknr,
3276              (unsigned long long)bucket_blkno(bucket));
3277
3278         for (i = 0; i < blks; i++)
3279                 memset(bucket_block(bucket, i), 0, blocksize);
3280
3281         /*
3282          * Since the xe_name_offset is based on ocfs2_xattr_header,
3283          * there is a offset change corresponding to the change of
3284          * ocfs2_xattr_header's position.
3285          */
3286         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3287         xe = &xb_xh->xh_entries[count - 1];
3288         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3289         size = blocksize - offset;
3290
3291         /* copy all the names and values. */
3292         memcpy(target + offset, src + offset, size);
3293
3294         /* Init new header now. */
3295         xh->xh_count = xb_xh->xh_count;
3296         xh->xh_num_buckets = cpu_to_le16(1);
3297         xh->xh_name_value_len = cpu_to_le16(size);
3298         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3299
3300         /* copy all the entries. */
3301         target = bucket_block(bucket, 0);
3302         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3303         size = count * sizeof(struct ocfs2_xattr_entry);
3304         memcpy(target + offset, (char *)xb_xh + offset, size);
3305
3306         /* Change the xe offset for all the xe because of the move. */
3307         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3308                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3309         for (i = 0; i < count; i++)
3310                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3311
3312         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3313              offset, size, off_change);
3314
3315         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3316              cmp_xe, swap_xe);
3317 }
3318
3319 /*
3320  * After we move xattr from block to index btree, we have to
3321  * update ocfs2_xattr_search to the new xe and base.
3322  *
3323  * When the entry is in xattr block, xattr_bh indicates the storage place.
3324  * While if the entry is in index b-tree, "bucket" indicates the
3325  * real place of the xattr.
3326  */
3327 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3328                                             struct ocfs2_xattr_search *xs,
3329                                             struct buffer_head *old_bh)
3330 {
3331         char *buf = old_bh->b_data;
3332         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3333         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3334         int i;
3335
3336         xs->header = bucket_xh(xs->bucket);
3337         xs->base = bucket_block(xs->bucket, 0);
3338         xs->end = xs->base + inode->i_sb->s_blocksize;
3339
3340         if (xs->not_found)
3341                 return;
3342
3343         i = xs->here - old_xh->xh_entries;
3344         xs->here = &xs->header->xh_entries[i];
3345 }
3346
3347 static int ocfs2_xattr_create_index_block(struct inode *inode,
3348                                           struct ocfs2_xattr_search *xs,
3349                                           struct ocfs2_xattr_set_ctxt *ctxt)
3350 {
3351         int ret;
3352         u32 bit_off, len;
3353         u64 blkno;
3354         handle_t *handle = ctxt->handle;
3355         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3356         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3357         struct buffer_head *xb_bh = xs->xattr_bh;
3358         struct ocfs2_xattr_block *xb =
3359                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3360         struct ocfs2_xattr_tree_root *xr;
3361         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3362
3363         mlog(0, "create xattr index block for %llu\n",
3364              (unsigned long long)xb_bh->b_blocknr);
3365
3366         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3367         BUG_ON(!xs->bucket);
3368
3369         /*
3370          * XXX:
3371          * We can use this lock for now, and maybe move to a dedicated mutex
3372          * if performance becomes a problem later.
3373          */
3374         down_write(&oi->ip_alloc_sem);
3375
3376         ret = ocfs2_journal_access(handle, inode, xb_bh,
3377                                    OCFS2_JOURNAL_ACCESS_WRITE);
3378         if (ret) {
3379                 mlog_errno(ret);
3380                 goto out;
3381         }
3382
3383         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3384                                      1, 1, &bit_off, &len);
3385         if (ret) {
3386                 mlog_errno(ret);
3387                 goto out;
3388         }
3389
3390         /*
3391          * The bucket may spread in many blocks, and
3392          * we will only touch the 1st block and the last block
3393          * in the whole bucket(one for entry and one for data).
3394          */
3395         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3396
3397         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3398              (unsigned long long)blkno);
3399
3400         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3401         if (ret) {
3402                 mlog_errno(ret);
3403                 goto out;
3404         }
3405
3406         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3407                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3408         if (ret) {
3409                 mlog_errno(ret);
3410                 goto out;
3411         }
3412
3413         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3414         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3415
3416         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3417
3418         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3419         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3420                offsetof(struct ocfs2_xattr_block, xb_attrs));
3421
3422         xr = &xb->xb_attrs.xb_root;
3423         xr->xt_clusters = cpu_to_le32(1);
3424         xr->xt_last_eb_blk = 0;
3425         xr->xt_list.l_tree_depth = 0;
3426         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3427         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3428
3429         xr->xt_list.l_recs[0].e_cpos = 0;
3430         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3431         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3432
3433         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3434
3435         ocfs2_journal_dirty(handle, xb_bh);
3436
3437 out:
3438         up_write(&oi->ip_alloc_sem);
3439
3440         return ret;
3441 }
3442
3443 static int cmp_xe_offset(const void *a, const void *b)
3444 {
3445         const struct ocfs2_xattr_entry *l = a, *r = b;
3446         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3447         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3448
3449         if (l_name_offset < r_name_offset)
3450                 return 1;
3451         if (l_name_offset > r_name_offset)
3452                 return -1;
3453         return 0;
3454 }
3455
3456 /*
3457  * defrag a xattr bucket if we find that the bucket has some
3458  * holes beteen name/value pairs.
3459  * We will move all the name/value pairs to the end of the bucket
3460  * so that we can spare some space for insertion.
3461  */
3462 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3463                                      handle_t *handle,
3464                                      struct ocfs2_xattr_bucket *bucket)
3465 {
3466         int ret, i;
3467         size_t end, offset, len, value_len;
3468         struct ocfs2_xattr_header *xh;
3469         char *entries, *buf, *bucket_buf = NULL;
3470         u64 blkno = bucket_blkno(bucket);
3471         u16 xh_free_start;
3472         size_t blocksize = inode->i_sb->s_blocksize;
3473         struct ocfs2_xattr_entry *xe;
3474
3475         /*
3476          * In order to make the operation more efficient and generic,
3477          * we copy all the blocks into a contiguous memory and do the
3478          * defragment there, so if anything is error, we will not touch
3479          * the real block.
3480          */
3481         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3482         if (!bucket_buf) {
3483                 ret = -EIO;
3484                 goto out;
3485         }
3486
3487         buf = bucket_buf;
3488         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3489                 memcpy(buf, bucket_block(bucket, i), blocksize);
3490
3491         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3492                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3493         if (ret < 0) {
3494                 mlog_errno(ret);
3495                 goto out;
3496         }
3497
3498         xh = (struct ocfs2_xattr_header *)bucket_buf;
3499         entries = (char *)xh->xh_entries;
3500         xh_free_start = le16_to_cpu(xh->xh_free_start);
3501
3502         mlog(0, "adjust xattr bucket in %llu, count = %u, "
3503              "xh_free_start = %u, xh_name_value_len = %u.\n",
3504              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3505              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3506
3507         /*
3508          * sort all the entries by their offset.
3509          * the largest will be the first, so that we can
3510          * move them to the end one by one.
3511          */
3512         sort(entries, le16_to_cpu(xh->xh_count),
3513              sizeof(struct ocfs2_xattr_entry),
3514              cmp_xe_offset, swap_xe);
3515
3516         /* Move all name/values to the end of the bucket. */
3517         xe = xh->xh_entries;
3518         end = OCFS2_XATTR_BUCKET_SIZE;
3519         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3520                 offset = le16_to_cpu(xe->xe_name_offset);
3521                 if (ocfs2_xattr_is_local(xe))
3522                         value_len = OCFS2_XATTR_SIZE(
3523                                         le64_to_cpu(xe->xe_value_size));
3524                 else
3525                         value_len = OCFS2_XATTR_ROOT_SIZE;
3526                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3527
3528                 /*
3529                  * We must make sure that the name/value pair
3530                  * exist in the same block. So adjust end to
3531                  * the previous block end if needed.
3532                  */
3533                 if (((end - len) / blocksize !=
3534                         (end - 1) / blocksize))
3535                         end = end - end % blocksize;
3536
3537                 if (end > offset + len) {
3538                         memmove(bucket_buf + end - len,
3539                                 bucket_buf + offset, len);
3540                         xe->xe_name_offset = cpu_to_le16(end - len);
3541                 }
3542
3543                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3544                                 "bucket %llu\n", (unsigned long long)blkno);
3545
3546                 end -= len;
3547         }
3548
3549         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3550                         "bucket %llu\n", (unsigned long long)blkno);
3551
3552         if (xh_free_start == end)
3553                 goto out;
3554
3555         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3556         xh->xh_free_start = cpu_to_le16(end);
3557
3558         /* sort the entries by their name_hash. */
3559         sort(entries, le16_to_cpu(xh->xh_count),
3560              sizeof(struct ocfs2_xattr_entry),
3561              cmp_xe, swap_xe);
3562
3563         buf = bucket_buf;
3564         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3565                 memcpy(bucket_block(bucket, i), buf, blocksize);
3566         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3567
3568 out:
3569         kfree(bucket_buf);
3570         return ret;
3571 }
3572
3573 /*
3574  * prev_blkno points to the start of an existing extent.  new_blkno
3575  * points to a newly allocated extent.  Because we know each of our
3576  * clusters contains more than bucket, we can easily split one cluster
3577  * at a bucket boundary.  So we take the last cluster of the existing
3578  * extent and split it down the middle.  We move the last half of the
3579  * buckets in the last cluster of the existing extent over to the new
3580  * extent.
3581  *
3582  * first_bh is the buffer at prev_blkno so we can update the existing
3583  * extent's bucket count.  header_bh is the bucket were we were hoping
3584  * to insert our xattr.  If the bucket move places the target in the new
3585  * extent, we'll update first_bh and header_bh after modifying the old
3586  * extent.
3587  *
3588  * first_hash will be set as the 1st xe's name_hash in the new extent.
3589  */
3590 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3591                                                handle_t *handle,
3592                                                struct ocfs2_xattr_bucket *first,
3593                                                struct ocfs2_xattr_bucket *target,
3594                                                u64 new_blkno,
3595                                                u32 num_clusters,
3596                                                u32 *first_hash)
3597 {
3598         int ret;
3599         struct super_block *sb = inode->i_sb;
3600         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3601         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3602         int to_move = num_buckets / 2;
3603         u64 src_blkno;
3604         u64 last_cluster_blkno = bucket_blkno(first) +
3605                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3606
3607         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3608         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3609
3610         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3611              (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3612
3613         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3614                                      last_cluster_blkno, new_blkno,
3615                                      to_move, first_hash);
3616         if (ret) {
3617                 mlog_errno(ret);
3618                 goto out;
3619         }
3620
3621         /* This is the first bucket that got moved */
3622         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3623
3624         /*
3625          * If the target bucket was part of the moved buckets, we need to
3626          * update first and target.
3627          */
3628         if (bucket_blkno(target) >= src_blkno) {
3629                 /* Find the block for the new target bucket */
3630                 src_blkno = new_blkno +
3631                         (bucket_blkno(target) - src_blkno);
3632
3633                 ocfs2_xattr_bucket_relse(first);
3634                 ocfs2_xattr_bucket_relse(target);
3635
3636                 /*
3637                  * These shouldn't fail - the buffers are in the
3638                  * journal from ocfs2_cp_xattr_bucket().
3639                  */
3640                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3641                 if (ret) {
3642                         mlog_errno(ret);
3643                         goto out;
3644                 }
3645                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3646                 if (ret)
3647                         mlog_errno(ret);
3648
3649         }
3650
3651 out:
3652         return ret;
3653 }
3654
3655 /*
3656  * Find the suitable pos when we divide a bucket into 2.
3657  * We have to make sure the xattrs with the same hash value exist
3658  * in the same bucket.
3659  *
3660  * If this ocfs2_xattr_header covers more than one hash value, find a
3661  * place where the hash value changes.  Try to find the most even split.
3662  * The most common case is that all entries have different hash values,
3663  * and the first check we make will find a place to split.
3664  */
3665 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3666 {
3667         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3668         int count = le16_to_cpu(xh->xh_count);
3669         int delta, middle = count / 2;
3670
3671         /*
3672          * We start at the middle.  Each step gets farther away in both
3673          * directions.  We therefore hit the change in hash value
3674          * nearest to the middle.  Note that this loop does not execute for
3675          * count < 2.
3676          */
3677         for (delta = 0; delta < middle; delta++) {
3678                 /* Let's check delta earlier than middle */
3679                 if (cmp_xe(&entries[middle - delta - 1],
3680                            &entries[middle - delta]))
3681                         return middle - delta;
3682
3683                 /* For even counts, don't walk off the end */
3684                 if ((middle + delta + 1) == count)
3685                         continue;
3686
3687                 /* Now try delta past middle */
3688                 if (cmp_xe(&entries[middle + delta],
3689                            &entries[middle + delta + 1]))
3690                         return middle + delta + 1;
3691         }
3692
3693         /* Every entry had the same hash */
3694         return count;
3695 }
3696
3697 /*
3698  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3699  * first_hash will record the 1st hash of the new bucket.
3700  *
3701  * Normally half of the xattrs will be moved.  But we have to make
3702  * sure that the xattrs with the same hash value are stored in the
3703  * same bucket. If all the xattrs in this bucket have the same hash
3704  * value, the new bucket will be initialized as an empty one and the
3705  * first_hash will be initialized as (hash_value+1).
3706  */
3707 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3708                                     handle_t *handle,
3709                                     u64 blk,
3710                                     u64 new_blk,
3711                                     u32 *first_hash,
3712                                     int new_bucket_head)
3713 {
3714         int ret, i;
3715         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3716         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3717         struct ocfs2_xattr_header *xh;
3718         struct ocfs2_xattr_entry *xe;
3719         int blocksize = inode->i_sb->s_blocksize;
3720
3721         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3722              (unsigned long long)blk, (unsigned long long)new_blk);
3723
3724         s_bucket = ocfs2_xattr_bucket_new(inode);
3725         t_bucket = ocfs2_xattr_bucket_new(inode);
3726         if (!s_bucket || !t_bucket) {
3727                 ret = -ENOMEM;
3728                 mlog_errno(ret);
3729                 goto out;
3730         }
3731
3732         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3733         if (ret) {
3734                 mlog_errno(ret);
3735                 goto out;
3736         }
3737
3738         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3739                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3740         if (ret) {
3741                 mlog_errno(ret);
3742                 goto out;
3743         }
3744
3745         /*
3746          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3747          * there's no need to read it.
3748          */
3749         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3750         if (ret) {
3751                 mlog_errno(ret);
3752                 goto out;
3753         }
3754
3755         /*
3756          * Hey, if we're overwriting t_bucket, what difference does
3757          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
3758          * same part of ocfs2_cp_xattr_bucket().
3759          */
3760         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3761                                                 new_bucket_head ?
3762                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3763                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3764         if (ret) {
3765                 mlog_errno(ret);
3766                 goto out;
3767         }
3768
3769         xh = bucket_xh(s_bucket);
3770         count = le16_to_cpu(xh->xh_count);
3771         start = ocfs2_xattr_find_divide_pos(xh);
3772
3773         if (start == count) {
3774                 xe = &xh->xh_entries[start-1];
3775
3776                 /*
3777                  * initialized a new empty bucket here.
3778                  * The hash value is set as one larger than
3779                  * that of the last entry in the previous bucket.
3780                  */
3781                 for (i = 0; i < t_bucket->bu_blocks; i++)
3782                         memset(bucket_block(t_bucket, i), 0, blocksize);
3783
3784                 xh = bucket_xh(t_bucket);
3785                 xh->xh_free_start = cpu_to_le16(blocksize);
3786                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3787                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3788
3789                 goto set_num_buckets;
3790         }
3791
3792         /* copy the whole bucket to the new first. */
3793         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3794
3795         /* update the new bucket. */
3796         xh = bucket_xh(t_bucket);
3797
3798         /*
3799          * Calculate the total name/value len and xh_free_start for
3800          * the old bucket first.
3801          */
3802         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3803         name_value_len = 0;
3804         for (i = 0; i < start; i++) {
3805                 xe = &xh->xh_entries[i];
3806                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3807                 if (ocfs2_xattr_is_local(xe))
3808                         xe_len +=
3809                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3810                 else
3811                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3812                 name_value_len += xe_len;
3813                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3814                         name_offset = le16_to_cpu(xe->xe_name_offset);
3815         }
3816
3817         /*
3818          * Now begin the modification to the new bucket.
3819          *
3820          * In the new bucket, We just move the xattr entry to the beginning
3821          * and don't touch the name/value. So there will be some holes in the
3822          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3823          * called.
3824          */
3825         xe = &xh->xh_entries[start];
3826         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3827         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3828              (int)((char *)xe - (char *)xh),
3829              (int)((char *)xh->xh_entries - (char *)xh));
3830         memmove((char *)xh->xh_entries, (char *)xe, len);
3831         xe = &xh->xh_entries[count - start];
3832         len = sizeof(struct ocfs2_xattr_entry) * start;
3833         memset((char *)xe, 0, len);
3834
3835         le16_add_cpu(&xh->xh_count, -start);
3836         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3837
3838         /* Calculate xh_free_start for the new bucket. */
3839         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3840         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3841                 xe = &xh->xh_entries[i];
3842                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3843                 if (ocfs2_xattr_is_local(xe))
3844                         xe_len +=
3845                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3846                 else
3847                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3848                 if (le16_to_cpu(xe->xe_name_offset) <
3849                     le16_to_cpu(xh->xh_free_start))
3850                         xh->xh_free_start = xe->xe_name_offset;
3851         }
3852
3853 set_num_buckets:
3854         /* set xh->xh_num_buckets for the new xh. */
3855         if (new_bucket_head)
3856                 xh->xh_num_buckets = cpu_to_le16(1);
3857         else
3858                 xh->xh_num_buckets = 0;
3859
3860         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3861
3862         /* store the first_hash of the new bucket. */
3863         if (first_hash)
3864                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3865
3866         /*
3867          * Now only update the 1st block of the old bucket.  If we
3868          * just added a new empty bucket, there is no need to modify
3869          * it.
3870          */
3871         if (start == count)
3872                 goto out;
3873
3874         xh = bucket_xh(s_bucket);
3875         memset(&xh->xh_entries[start], 0,
3876                sizeof(struct ocfs2_xattr_entry) * (count - start));
3877         xh->xh_count = cpu_to_le16(start);
3878         xh->xh_free_start = cpu_to_le16(name_offset);
3879         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3880
3881         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3882
3883 out:
3884         ocfs2_xattr_bucket_free(s_bucket);
3885         ocfs2_xattr_bucket_free(t_bucket);
3886
3887         return ret;
3888 }
3889
3890 /*
3891  * Copy xattr from one bucket to another bucket.
3892  *
3893  * The caller must make sure that the journal transaction
3894  * has enough space for journaling.
3895  */
3896 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3897                                  handle_t *handle,
3898                                  u64 s_blkno,
3899                                  u64 t_blkno,
3900                                  int t_is_new)
3901 {
3902         int ret;
3903         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3904
3905         BUG_ON(s_blkno == t_blkno);
3906
3907         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3908              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3909              t_is_new);
3910
3911         s_bucket = ocfs2_xattr_bucket_new(inode);
3912         t_bucket = ocfs2_xattr_bucket_new(inode);
3913         if (!s_bucket || !t_bucket) {
3914                 ret = -ENOMEM;
3915                 mlog_errno(ret);
3916                 goto out;
3917         }
3918
3919         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3920         if (ret)
3921                 goto out;
3922
3923         /*
3924          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3925          * there's no need to read it.
3926          */
3927         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3928         if (ret)
3929                 goto out;
3930
3931         /*
3932          * Hey, if we're overwriting t_bucket, what difference does
3933          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
3934          * cluster to fill, we came here from
3935          * ocfs2_mv_xattr_buckets(), and it is really new -
3936          * ACCESS_CREATE is required.  But we also might have moved data
3937          * out of t_bucket before extending back into it.
3938          * ocfs2_add_new_xattr_bucket() can do this - its call to
3939          * ocfs2_add_new_xattr_cluster() may have created a new extent
3940          * and copied out the end of the old extent.  Then it re-extends
3941          * the old extent back to create space for new xattrs.  That's
3942          * how we get here, and the bucket isn't really new.
3943          */
3944         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3945                                                 t_is_new ?
3946                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3947                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3948         if (ret)
3949                 goto out;
3950
3951         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3952         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3953
3954 out:
3955         ocfs2_xattr_bucket_free(t_bucket);
3956         ocfs2_xattr_bucket_free(s_bucket);
3957
3958         return ret;
3959 }
3960
3961 /*
3962  * src_blk points to the start of an existing extent.  last_blk points to
3963  * last cluster in that extent.  to_blk points to a newly allocated
3964  * extent.  We copy the buckets from the cluster at last_blk to the new
3965  * extent.  If start_bucket is non-zero, we skip that many buckets before
3966  * we start copying.  The new extent's xh_num_buckets gets set to the
3967  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
3968  * by the same amount.
3969  */
3970 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
3971                                   u64 src_blk, u64 last_blk, u64 to_blk,
3972                                   unsigned int start_bucket,
3973                                   u32 *first_hash)
3974 {
3975         int i, ret, credits;
3976         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3977         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3978         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3979         struct ocfs2_xattr_bucket *old_first, *new_first;
3980
3981         mlog(0, "mv xattrs from cluster %llu to %llu\n",
3982              (unsigned long long)last_blk, (unsigned long long)to_blk);
3983
3984         BUG_ON(start_bucket >= num_buckets);
3985         if (start_bucket) {
3986                 num_buckets -= start_bucket;
3987                 last_blk += (start_bucket * blks_per_bucket);
3988         }
3989
3990         /* The first bucket of the original extent */
3991         old_first = ocfs2_xattr_bucket_new(inode);
3992         /* The first bucket of the new extent */
3993         new_first = ocfs2_xattr_bucket_new(inode);
3994         if (!old_first || !new_first) {
3995                 ret = -ENOMEM;
3996                 mlog_errno(ret);
3997                 goto out;
3998         }
3999
4000         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4001         if (ret) {
4002                 mlog_errno(ret);
4003                 goto out;
4004         }
4005
4006         /*
4007          * We need to update the first bucket of the old extent and all
4008          * the buckets going to the new extent.
4009          */
4010         credits = ((num_buckets + 1) * blks_per_bucket) +
4011                 handle->h_buffer_credits;
4012         ret = ocfs2_extend_trans(handle, credits);
4013         if (ret) {
4014                 mlog_errno(ret);
4015                 goto out;
4016         }
4017
4018         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4019                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4020         if (ret) {
4021                 mlog_errno(ret);
4022                 goto out;
4023         }
4024
4025         for (i = 0; i < num_buckets; i++) {
4026                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4027                                             last_blk + (i * blks_per_bucket),
4028                                             to_blk + (i * blks_per_bucket),
4029                                             1);
4030                 if (ret) {
4031                         mlog_errno(ret);
4032                         goto out;
4033                 }
4034         }
4035
4036         /*
4037          * Get the new bucket ready before we dirty anything
4038          * (This actually shouldn't fail, because we already dirtied
4039          * it once in ocfs2_cp_xattr_bucket()).
4040          */
4041         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4042         if (ret) {
4043                 mlog_errno(ret);
4044                 goto out;
4045         }
4046         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4047                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4048         if (ret) {
4049                 mlog_errno(ret);
4050                 goto out;
4051         }
4052
4053         /* Now update the headers */
4054         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4055         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4056
4057         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4058         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4059
4060         if (first_hash)
4061                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4062
4063 out:
4064         ocfs2_xattr_bucket_free(new_first);
4065         ocfs2_xattr_bucket_free(old_first);
4066         return ret;
4067 }
4068
4069 /*
4070  * Move some xattrs in this cluster to the new cluster.
4071  * This function should only be called when bucket size == cluster size.
4072  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4073  */
4074 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4075                                       handle_t *handle,
4076                                       u64 prev_blk,
4077                                       u64 new_blk,
4078                                       u32 *first_hash)
4079 {
4080         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4081         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4082
4083         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4084
4085         ret = ocfs2_extend_trans(handle, credits);
4086         if (ret) {
4087                 mlog_errno(ret);
4088                 return ret;
4089         }
4090
4091         /* Move half of the xattr in start_blk to the next bucket. */
4092         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4093                                           new_blk, first_hash, 1);
4094 }
4095
4096 /*
4097  * Move some xattrs from the old cluster to the new one since they are not
4098  * contiguous in ocfs2 xattr tree.
4099  *
4100  * new_blk starts a new separate cluster, and we will move some xattrs from
4101  * prev_blk to it. v_start will be set as the first name hash value in this
4102  * new cluster so that it can be used as e_cpos during tree insertion and
4103  * don't collide with our original b-tree operations. first_bh and header_bh
4104  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4105  * to extend the insert bucket.
4106  *
4107  * The problem is how much xattr should we move to the new one and when should
4108  * we update first_bh and header_bh?
4109  * 1. If cluster size > bucket size, that means the previous cluster has more
4110  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4111  *    update the first_bh and header_bh if the insert bucket has been moved
4112  *    to the new cluster.
4113  * 2. If cluster_size == bucket_size:
4114  *    a) If the previous extent rec has more than one cluster and the insert
4115  *       place isn't in the last cluster, copy the entire last cluster to the
4116  *       new one. This time, we don't need to upate the first_bh and header_bh
4117  *       since they will not be moved into the new cluster.
4118  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4119  *       the new one. And we set the extend flag to zero if the insert place is
4120  *       moved into the new allocated cluster since no extend is needed.
4121  */
4122 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4123                                             handle_t *handle,
4124                                             struct ocfs2_xattr_bucket *first,
4125                                             struct ocfs2_xattr_bucket *target,
4126                                             u64 new_blk,
4127                                             u32 prev_clusters,
4128                                             u32 *v_start,
4129                                             int *extend)
4130 {
4131         int ret;
4132
4133         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4134              (unsigned long long)bucket_blkno(first), prev_clusters,
4135              (unsigned long long)new_blk);
4136
4137         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4138                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4139                                                           handle,
4140                                                           first, target,
4141                                                           new_blk,
4142                                                           prev_clusters,
4143                                                           v_start);
4144                 if (ret)
4145                         mlog_errno(ret);
4146         } else {
4147                 /* The start of the last cluster in the first extent */
4148                 u64 last_blk = bucket_blkno(first) +
4149                         ((prev_clusters - 1) *
4150                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
4151
4152                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4153                         ret = ocfs2_mv_xattr_buckets(inode, handle,
4154                                                      bucket_blkno(first),
4155                                                      last_blk, new_blk, 0,
4156                                                      v_start);
4157                         if (ret)
4158                                 mlog_errno(ret);
4159                 } else {
4160                         ret = ocfs2_divide_xattr_cluster(inode, handle,
4161                                                          last_blk, new_blk,
4162                                                          v_start);
4163                         if (ret)
4164                                 mlog_errno(ret);
4165
4166                         if ((bucket_blkno(target) == last_blk) && extend)
4167                                 *extend = 0;
4168                 }
4169         }
4170
4171         return ret;
4172 }
4173
4174 /*
4175  * Add a new cluster for xattr storage.
4176  *
4177  * If the new cluster is contiguous with the previous one, it will be
4178  * appended to the same extent record, and num_clusters will be updated.
4179  * If not, we will insert a new extent for it and move some xattrs in
4180  * the last cluster into the new allocated one.
4181  * We also need to limit the maximum size of a btree leaf, otherwise we'll
4182  * lose the benefits of hashing because we'll have to search large leaves.
4183  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4184  * if it's bigger).
4185  *
4186  * first_bh is the first block of the previous extent rec and header_bh
4187  * indicates the bucket we will insert the new xattrs. They will be updated
4188  * when the header_bh is moved into the new cluster.
4189  */
4190 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4191                                        struct buffer_head *root_bh,
4192                                        struct ocfs2_xattr_bucket *first,
4193                                        struct ocfs2_xattr_bucket *target,
4194                                        u32 *num_clusters,
4195                                        u32 prev_cpos,
4196                                        int *extend,
4197                                        struct ocfs2_xattr_set_ctxt *ctxt)
4198 {
4199         int ret;
4200         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4201         u32 prev_clusters = *num_clusters;
4202         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4203         u64 block;
4204         handle_t *handle = ctxt->handle;
4205         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4206         struct ocfs2_extent_tree et;
4207
4208         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4209              "previous xattr blkno = %llu\n",
4210              (unsigned long long)OCFS2_I(inode)->ip_blkno,
4211              prev_cpos, (unsigned long long)bucket_blkno(first));
4212
4213         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4214
4215         ret = ocfs2_journal_access(handle, inode, root_bh,
4216                                    OCFS2_JOURNAL_ACCESS_WRITE);
4217         if (ret < 0) {
4218                 mlog_errno(ret);
4219                 goto leave;
4220         }
4221
4222         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4223                                      clusters_to_add, &bit_off, &num_bits);
4224         if (ret < 0) {
4225                 if (ret != -ENOSPC)
4226                         mlog_errno(ret);
4227                 goto leave;
4228         }
4229
4230         BUG_ON(num_bits > clusters_to_add);
4231
4232         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4233         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4234              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4235
4236         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4237             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4238              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4239                 /*
4240                  * If this cluster is contiguous with the old one and
4241                  * adding this new cluster, we don't surpass the limit of
4242                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4243                  * initialized and used like other buckets in the previous
4244                  * cluster.
4245                  * So add it as a contiguous one. The caller will handle
4246                  * its init process.
4247                  */
4248                 v_start = prev_cpos + prev_clusters;
4249                 *num_clusters = prev_clusters + num_bits;
4250                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4251                      num_bits);
4252         } else {
4253                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4254                                                        handle,
4255                                                        first,
4256                                                        target,
4257                                                        block,
4258                                                        prev_clusters,
4259                                                        &v_start,
4260                                                        extend);
4261                 if (ret) {
4262                         mlog_errno(ret);
4263                         goto leave;
4264                 }
4265         }
4266
4267         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4268              num_bits, (unsigned long long)block, v_start);
4269         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
4270                                   num_bits, 0, ctxt->meta_ac);
4271         if (ret < 0) {
4272                 mlog_errno(ret);
4273                 goto leave;
4274         }
4275
4276         ret = ocfs2_journal_dirty(handle, root_bh);
4277         if (ret < 0)
4278                 mlog_errno(ret);
4279
4280 leave:
4281         return ret;
4282 }
4283
4284 /*
4285  * We are given an extent.  'first' is the bucket at the very front of
4286  * the extent.  The extent has space for an additional bucket past
4287  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
4288  * of the target bucket.  We wish to shift every bucket past the target
4289  * down one, filling in that additional space.  When we get back to the
4290  * target, we split the target between itself and the now-empty bucket
4291  * at target+1 (aka, target_blkno + blks_per_bucket).
4292  */
4293 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4294                                      handle_t *handle,
4295                                      struct ocfs2_xattr_bucket *first,
4296                                      u64 target_blk,
4297                                      u32 num_clusters)
4298 {
4299         int ret, credits;
4300         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4301         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4302         u64 end_blk;
4303         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4304
4305         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4306              "from %llu, len = %u\n", (unsigned long long)target_blk,
4307              (unsigned long long)bucket_blkno(first), num_clusters);
4308
4309         /* The extent must have room for an additional bucket */
4310         BUG_ON(new_bucket >=
4311                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4312
4313         /* end_blk points to the last existing bucket */
4314         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4315
4316         /*
4317          * end_blk is the start of the last existing bucket.
4318          * Thus, (end_blk - target_blk) covers the target bucket and
4319          * every bucket after it up to, but not including, the last
4320          * existing bucket.  Then we add the last existing bucket, the
4321          * new bucket, and the first bucket (3 * blk_per_bucket).
4322          */
4323         credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4324                   handle->h_buffer_credits;
4325         ret = ocfs2_extend_trans(handle, credits);
4326         if (ret) {
4327                 mlog_errno(ret);
4328                 goto out;
4329         }
4330
4331         ret = ocfs2_xattr_bucket_journal_access(handle, first,
4332                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4333         if (ret) {
4334                 mlog_errno(ret);
4335                 goto out;
4336         }
4337
4338         while (end_blk != target_blk) {
4339                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4340                                             end_blk + blk_per_bucket, 0);
4341                 if (ret)
4342                         goto out;
4343                 end_blk -= blk_per_bucket;
4344         }
4345
4346         /* Move half of the xattr in target_blkno to the next bucket. */
4347         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4348                                         target_blk + blk_per_bucket, NULL, 0);
4349
4350         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4351         ocfs2_xattr_bucket_journal_dirty(handle, first);
4352
4353 out:
4354         return ret;
4355 }
4356
4357 /*
4358  * Add new xattr bucket in an extent record and adjust the buckets
4359  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
4360  * bucket we want to insert into.
4361  *
4362  * In the easy case, we will move all the buckets after target down by
4363  * one. Half of target's xattrs will be moved to the next bucket.
4364  *
4365  * If current cluster is full, we'll allocate a new one.  This may not
4366  * be contiguous.  The underlying calls will make sure that there is
4367  * space for the insert, shifting buckets around if necessary.
4368  * 'target' may be moved by those calls.
4369  */
4370 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4371                                       struct buffer_head *xb_bh,
4372                                       struct ocfs2_xattr_bucket *target,
4373                                       struct ocfs2_xattr_set_ctxt *ctxt)
4374 {
4375         struct ocfs2_xattr_block *xb =
4376                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4377         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4378         struct ocfs2_extent_list *el = &xb_root->xt_list;
4379         u32 name_hash =
4380                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4381         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4382         int ret, num_buckets, extend = 1;
4383         u64 p_blkno;
4384         u32 e_cpos, num_clusters;
4385         /* The bucket at the front of the extent */
4386         struct ocfs2_xattr_bucket *first;
4387
4388         mlog(0, "Add new xattr bucket starting from %llu\n",
4389              (unsigned long long)bucket_blkno(target));
4390
4391         /* The first bucket of the original extent */
4392         first = ocfs2_xattr_bucket_new(inode);
4393         if (!first) {
4394                 ret = -ENOMEM;
4395                 mlog_errno(ret);
4396                 goto out;
4397         }
4398
4399         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4400                                   &num_clusters, el);
4401         if (ret) {
4402                 mlog_errno(ret);
4403                 goto out;
4404         }
4405
4406         ret = ocfs2_read_xattr_bucket(first, p_blkno);
4407         if (ret) {
4408                 mlog_errno(ret);
4409                 goto out;
4410         }
4411
4412         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4413         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4414                 /*
4415                  * This can move first+target if the target bucket moves
4416                  * to the new extent.
4417                  */
4418                 ret = ocfs2_add_new_xattr_cluster(inode,
4419                                                   xb_bh,
4420                                                   first,
4421                                                   target,
4422                                                   &num_clusters,
4423                                                   e_cpos,
4424                                                   &extend,
4425                                                   ctxt);
4426                 if (ret) {
4427                         mlog_errno(ret);
4428                         goto out;
4429                 }
4430         }
4431
4432         if (extend) {
4433                 ret = ocfs2_extend_xattr_bucket(inode,
4434                                                 ctxt->handle,
4435                                                 first,
4436                                                 bucket_blkno(target),
4437                                                 num_clusters);
4438                 if (ret)
4439                         mlog_errno(ret);
4440         }
4441
4442 out:
4443         ocfs2_xattr_bucket_free(first);
4444
4445         return ret;
4446 }
4447
4448 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4449                                         struct ocfs2_xattr_bucket *bucket,
4450                                         int offs)
4451 {
4452         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4453
4454         offs = offs % inode->i_sb->s_blocksize;
4455         return bucket_block(bucket, block_off) + offs;
4456 }
4457
4458 /*
4459  * Handle the normal xattr set, including replace, delete and new.
4460  *
4461  * Note: "local" indicates the real data's locality. So we can't
4462  * just its bucket locality by its length.
4463  */
4464 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4465                                          struct ocfs2_xattr_info *xi,
4466                                          struct ocfs2_xattr_search *xs,
4467                                          u32 name_hash,
4468                                          int local)
4469 {
4470         struct ocfs2_xattr_entry *last, *xe;
4471         int name_len = strlen(xi->name);
4472         struct ocfs2_xattr_header *xh = xs->header;
4473         u16 count = le16_to_cpu(xh->xh_count), start;
4474         size_t blocksize = inode->i_sb->s_blocksize;
4475         char *val;
4476         size_t offs, size, new_size;
4477
4478         last = &xh->xh_entries[count];
4479         if (!xs->not_found) {
4480                 xe = xs->here;
4481                 offs = le16_to_cpu(xe->xe_name_offset);
4482                 if (ocfs2_xattr_is_local(xe))
4483                         size = OCFS2_XATTR_SIZE(name_len) +
4484                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4485                 else
4486                         size = OCFS2_XATTR_SIZE(name_len) +
4487                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4488
4489                 /*
4490                  * If the new value will be stored outside, xi->value has been
4491                  * initalized as an empty ocfs2_xattr_value_root, and the same
4492                  * goes with xi->value_len, so we can set new_size safely here.
4493                  * See ocfs2_xattr_set_in_bucket.
4494                  */
4495                 new_size = OCFS2_XATTR_SIZE(name_len) +
4496                            OCFS2_XATTR_SIZE(xi->value_len);
4497
4498                 le16_add_cpu(&xh->xh_name_value_len, -size);
4499                 if (xi->value) {
4500                         if (new_size > size)
4501                                 goto set_new_name_value;
4502
4503                         /* Now replace the old value with new one. */
4504                         if (local)
4505                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4506                         else
4507                                 xe->xe_value_size = 0;
4508
4509                         val = ocfs2_xattr_bucket_get_val(inode,
4510                                                          xs->bucket, offs);
4511                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4512                                size - OCFS2_XATTR_SIZE(name_len));
4513                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4514                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4515                                        xi->value, xi->value_len);
4516
4517                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4518                         ocfs2_xattr_set_local(xe, local);
4519                         return;
4520                 } else {
4521                         /*
4522                          * Remove the old entry if there is more than one.
4523                          * We don't remove the last entry so that we can
4524                          * use it to indicate the hash value of the empty
4525                          * bucket.
4526                          */
4527                         last -= 1;
4528                         le16_add_cpu(&xh->xh_count, -1);
4529                         if (xh->xh_count) {
4530                                 memmove(xe, xe + 1,
4531                                         (void *)last - (void *)xe);
4532                                 memset(last, 0,
4533                                        sizeof(struct ocfs2_xattr_entry));
4534                         } else
4535                                 xh->xh_free_start =
4536                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4537
4538                         return;
4539                 }
4540         } else {
4541                 /* find a new entry for insert. */
4542                 int low = 0, high = count - 1, tmp;
4543                 struct ocfs2_xattr_entry *tmp_xe;
4544
4545                 while (low <= high && count) {
4546                         tmp = (low + high) / 2;
4547                         tmp_xe = &xh->xh_entries[tmp];
4548
4549                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4550                                 low = tmp + 1;
4551                         else if (name_hash <
4552                                  le32_to_cpu(tmp_xe->xe_name_hash))
4553                                 high = tmp - 1;
4554                         else {
4555                                 low = tmp;
4556                                 break;
4557                         }
4558                 }
4559
4560                 xe = &xh->xh_entries[low];
4561                 if (low != count)
4562                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4563
4564                 le16_add_cpu(&xh->xh_count, 1);
4565                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4566                 xe->xe_name_hash = cpu_to_le32(name_hash);
4567                 xe->xe_name_len = name_len;
4568                 ocfs2_xattr_set_type(xe, xi->name_index);
4569         }
4570
4571 set_new_name_value:
4572         /* Insert the new name+value. */
4573         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4574
4575         /*
4576          * We must make sure that the name/value pair
4577          * exists in the same block.
4578          */
4579         offs = le16_to_cpu(xh->xh_free_start);
4580         start = offs - size;
4581
4582         if (start >> inode->i_sb->s_blocksize_bits !=
4583             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4584                 offs = offs - offs % blocksize;
4585                 xh->xh_free_start = cpu_to_le16(offs);
4586         }
4587
4588         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4589         xe->xe_name_offset = cpu_to_le16(offs - size);
4590
4591         memset(val, 0, size);
4592         memcpy(val, xi->name, name_len);
4593         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4594
4595         xe->xe_value_size = cpu_to_le64(xi->value_len);
4596         ocfs2_xattr_set_local(xe, local);
4597         xs->here = xe;
4598         le16_add_cpu(&xh->xh_free_start, -size);
4599         le16_add_cpu(&xh->xh_name_value_len, size);
4600
4601         return;
4602 }
4603
4604 /*
4605  * Set the xattr entry in the specified bucket.
4606  * The bucket is indicated by xs->bucket and it should have the enough
4607  * space for the xattr insertion.
4608  */
4609 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4610                                            handle_t *handle,
4611                                            struct ocfs2_xattr_info *xi,
4612                                            struct ocfs2_xattr_search *xs,
4613                                            u32 name_hash,
4614                                            int local)
4615 {
4616         int ret;
4617         u64 blkno;
4618
4619         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4620              (unsigned long)xi->value_len, xi->name_index,
4621              (unsigned long long)bucket_blkno(xs->bucket));
4622
4623         if (!xs->bucket->bu_bhs[1]) {
4624                 blkno = bucket_blkno(xs->bucket);
4625                 ocfs2_xattr_bucket_relse(xs->bucket);
4626                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4627                 if (ret) {
4628                         mlog_errno(ret);
4629                         goto out;
4630                 }
4631         }
4632
4633         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4634                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4635         if (ret < 0) {
4636                 mlog_errno(ret);
4637                 goto out;
4638         }
4639
4640         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4641         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4642
4643 out:
4644         return ret;
4645 }
4646
4647 /*
4648  * Truncate the specified xe_off entry in xattr bucket.
4649  * bucket is indicated by header_bh and len is the new length.
4650  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4651  *
4652  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4653  */
4654 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4655                                              struct ocfs2_xattr_bucket *bucket,
4656                                              int xe_off,
4657                                              int len,
4658                                              struct ocfs2_xattr_set_ctxt *ctxt)
4659 {
4660         int ret, offset;
4661         u64 value_blk;
4662         struct ocfs2_xattr_entry *xe;
4663         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4664         size_t blocksize = inode->i_sb->s_blocksize;
4665         struct ocfs2_xattr_value_buf vb = {
4666                 .vb_access = ocfs2_journal_access,
4667         };
4668
4669         xe = &xh->xh_entries[xe_off];
4670
4671         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4672
4673         offset = le16_to_cpu(xe->xe_name_offset) +
4674                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4675
4676         value_blk = offset / blocksize;
4677
4678         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4679         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4680
4681         vb.vb_bh = bucket->bu_bhs[value_blk];
4682         BUG_ON(!vb.vb_bh);
4683
4684         vb.vb_xv = (struct ocfs2_xattr_value_root *)
4685                 (vb.vb_bh->b_data + offset % blocksize);
4686
4687         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4688                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4689         if (ret) {
4690                 mlog_errno(ret);
4691                 goto out;
4692         }
4693
4694         /*
4695          * From here on out we have to dirty the bucket.  The generic
4696          * value calls only modify one of the bucket's bhs, but we need
4697          * to send the bucket at once.  So if they error, they *could* have
4698          * modified something.  We have to assume they did, and dirty
4699          * the whole bucket.  This leaves us in a consistent state.
4700          */
4701         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4702              xe_off, (unsigned long long)bucket_blkno(bucket), len);
4703         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4704         if (ret) {
4705                 mlog_errno(ret);
4706                 goto out_dirty;
4707         }
4708
4709         xe->xe_value_size = cpu_to_le64(len);
4710
4711 out_dirty:
4712         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4713
4714 out:
4715         return ret;
4716 }
4717
4718 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4719                                         struct ocfs2_xattr_search *xs,
4720                                         int len,
4721                                         struct ocfs2_xattr_set_ctxt *ctxt)
4722 {
4723         int ret, offset;
4724         struct ocfs2_xattr_entry *xe = xs->here;
4725         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4726
4727         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4728
4729         offset = xe - xh->xh_entries;
4730         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4731                                                 offset, len, ctxt);
4732         if (ret)
4733                 mlog_errno(ret);
4734
4735         return ret;
4736 }
4737
4738 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4739                                                 handle_t *handle,
4740                                                 struct ocfs2_xattr_search *xs,
4741                                                 char *val,
4742                                                 int value_len)
4743 {
4744         int offset;
4745         struct ocfs2_xattr_value_root *xv;
4746         struct ocfs2_xattr_entry *xe = xs->here;
4747
4748         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4749
4750         offset = le16_to_cpu(xe->xe_name_offset) +
4751                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4752
4753         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4754
4755         return __ocfs2_xattr_set_value_outside(inode, handle,
4756                                                xv, val, value_len);
4757 }
4758
4759 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4760                                   struct buffer_head *root_bh,
4761                                   u64 blkno,
4762                                   u32 cpos,
4763                                   u32 len)
4764 {
4765         int ret;
4766         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4767         struct inode *tl_inode = osb->osb_tl_inode;
4768         handle_t *handle;
4769         struct ocfs2_xattr_block *xb =
4770                         (struct ocfs2_xattr_block *)root_bh->b_data;
4771         struct ocfs2_alloc_context *meta_ac = NULL;
4772         struct ocfs2_cached_dealloc_ctxt dealloc;
4773         struct ocfs2_extent_tree et;
4774
4775         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4776
4777         ocfs2_init_dealloc_ctxt(&dealloc);
4778
4779         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4780              cpos, len, (unsigned long long)blkno);
4781
4782         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4783
4784         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4785         if (ret) {
4786                 mlog_errno(ret);
4787                 return ret;
4788         }
4789
4790         mutex_lock(&tl_inode->i_mutex);
4791
4792         if (ocfs2_truncate_log_needs_flush(osb)) {
4793                 ret = __ocfs2_flush_truncate_log(osb);
4794                 if (ret < 0) {
4795                         mlog_errno(ret);
4796                         goto out;
4797                 }
4798         }
4799
4800         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4801         if (IS_ERR(handle)) {
4802                 ret = -ENOMEM;
4803                 mlog_errno(ret);
4804                 goto out;
4805         }
4806
4807         ret = ocfs2_journal_access(handle, inode, root_bh,
4808                                    OCFS2_JOURNAL_ACCESS_WRITE);
4809         if (ret) {
4810                 mlog_errno(ret);
4811                 goto out_commit;
4812         }
4813
4814         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4815                                   &dealloc);
4816         if (ret) {
4817                 mlog_errno(ret);
4818                 goto out_commit;
4819         }
4820
4821         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4822
4823         ret = ocfs2_journal_dirty(handle, root_bh);
4824         if (ret) {
4825                 mlog_errno(ret);
4826                 goto out_commit;
4827         }
4828
4829         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4830         if (ret)
4831                 mlog_errno(ret);
4832
4833 out_commit:
4834         ocfs2_commit_trans(osb, handle);
4835 out:
4836         ocfs2_schedule_truncate_log_flush(osb, 1);
4837
4838         mutex_unlock(&tl_inode->i_mutex);
4839
4840         if (meta_ac)
4841                 ocfs2_free_alloc_context(meta_ac);
4842
4843         ocfs2_run_deallocs(osb, &dealloc);
4844
4845         return ret;
4846 }
4847
4848 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4849                                          handle_t *handle,
4850                                          struct ocfs2_xattr_search *xs)
4851 {
4852         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4853         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4854                                                 le16_to_cpu(xh->xh_count) - 1];
4855         int ret = 0;
4856
4857         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4858                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4859         if (ret) {
4860                 mlog_errno(ret);
4861                 return;
4862         }
4863
4864         /* Remove the old entry. */
4865         memmove(xs->here, xs->here + 1,
4866                 (void *)last - (void *)xs->here);
4867         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4868         le16_add_cpu(&xh->xh_count, -1);
4869
4870         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4871 }
4872
4873 /*
4874  * Set the xattr name/value in the bucket specified in xs.
4875  *
4876  * As the new value in xi may be stored in the bucket or in an outside cluster,
4877  * we divide the whole process into 3 steps:
4878  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4879  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4880  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4881  * 4. If the clusters for the new outside value can't be allocated, we need
4882  *    to free the xattr we allocated in set.
4883  */
4884 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4885                                      struct ocfs2_xattr_info *xi,
4886                                      struct ocfs2_xattr_search *xs,
4887                                      struct ocfs2_xattr_set_ctxt *ctxt)
4888 {
4889         int ret, local = 1;
4890         size_t value_len;
4891         char *val = (char *)xi->value;
4892         struct ocfs2_xattr_entry *xe = xs->here;
4893         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4894                                               strlen(xi->name));
4895
4896         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4897                 /*
4898                  * We need to truncate the xattr storage first.
4899                  *
4900                  * If both the old and new value are stored to
4901                  * outside block, we only need to truncate
4902                  * the storage and then set the value outside.
4903                  *
4904                  * If the new value should be stored within block,
4905                  * we should free all the outside block first and
4906                  * the modification to the xattr block will be done
4907                  * by following steps.
4908                  */
4909                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4910                         value_len = xi->value_len;
4911                 else
4912                         value_len = 0;
4913
4914                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4915                                                            value_len,
4916                                                            ctxt);
4917                 if (ret)
4918                         goto out;
4919
4920                 if (value_len)
4921                         goto set_value_outside;
4922         }
4923
4924         value_len = xi->value_len;
4925         /* So we have to handle the inside block change now. */
4926         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4927                 /*
4928                  * If the new value will be stored outside of block,
4929                  * initalize a new empty value root and insert it first.
4930                  */
4931                 local = 0;
4932                 xi->value = &def_xv;
4933                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4934         }
4935
4936         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
4937                                               name_hash, local);
4938         if (ret) {
4939                 mlog_errno(ret);
4940                 goto out;
4941         }
4942
4943         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4944                 goto out;
4945
4946         /* allocate the space now for the outside block storage. */
4947         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4948                                                    value_len, ctxt);
4949         if (ret) {
4950                 mlog_errno(ret);
4951
4952                 if (xs->not_found) {
4953                         /*
4954                          * We can't allocate enough clusters for outside
4955                          * storage and we have allocated xattr already,
4956                          * so need to remove it.
4957                          */
4958                         ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
4959                 }
4960                 goto out;
4961         }
4962
4963 set_value_outside:
4964         ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
4965                                                    xs, val, value_len);
4966 out:
4967         return ret;
4968 }
4969
4970 /*
4971  * check whether the xattr bucket is filled up with the same hash value.
4972  * If we want to insert the xattr with the same hash, return -ENOSPC.
4973  * If we want to insert a xattr with different hash value, go ahead
4974  * and ocfs2_divide_xattr_bucket will handle this.
4975  */
4976 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4977                                               struct ocfs2_xattr_bucket *bucket,
4978                                               const char *name)
4979 {
4980         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4981         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
4982
4983         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
4984                 return 0;
4985
4986         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4987             xh->xh_entries[0].xe_name_hash) {
4988                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4989                      "hash = %u\n",
4990                      (unsigned long long)bucket_blkno(bucket),
4991                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4992                 return -ENOSPC;
4993         }
4994
4995         return 0;
4996 }
4997
4998 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4999                                              struct ocfs2_xattr_info *xi,
5000                                              struct ocfs2_xattr_search *xs,
5001                                              struct ocfs2_xattr_set_ctxt *ctxt)
5002 {
5003         struct ocfs2_xattr_header *xh;
5004         struct ocfs2_xattr_entry *xe;
5005         u16 count, header_size, xh_free_start;
5006         int free, max_free, need, old;
5007         size_t value_size = 0, name_len = strlen(xi->name);
5008         size_t blocksize = inode->i_sb->s_blocksize;
5009         int ret, allocation = 0;
5010
5011         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5012
5013 try_again:
5014         xh = xs->header;
5015         count = le16_to_cpu(xh->xh_count);
5016         xh_free_start = le16_to_cpu(xh->xh_free_start);
5017         header_size = sizeof(struct ocfs2_xattr_header) +
5018                         count * sizeof(struct ocfs2_xattr_entry);
5019         max_free = OCFS2_XATTR_BUCKET_SIZE -
5020                 le16_to_cpu(xh->xh_name_value_len) - header_size;
5021
5022         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5023                         "of %u which exceed block size\n",
5024                         (unsigned long long)bucket_blkno(xs->bucket),
5025                         header_size);
5026
5027         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5028                 value_size = OCFS2_XATTR_ROOT_SIZE;
5029         else if (xi->value)
5030                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5031
5032         if (xs->not_found)
5033                 need = sizeof(struct ocfs2_xattr_entry) +
5034                         OCFS2_XATTR_SIZE(name_len) + value_size;
5035         else {
5036                 need = value_size + OCFS2_XATTR_SIZE(name_len);
5037
5038                 /*
5039                  * We only replace the old value if the new length is smaller
5040                  * than the old one. Otherwise we will allocate new space in the
5041                  * bucket to store it.
5042                  */
5043                 xe = xs->here;
5044                 if (ocfs2_xattr_is_local(xe))
5045                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5046                 else
5047                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5048
5049                 if (old >= value_size)
5050                         need = 0;
5051         }
5052
5053         free = xh_free_start - header_size;
5054         /*
5055          * We need to make sure the new name/value pair
5056          * can exist in the same block.
5057          */
5058         if (xh_free_start % blocksize < need)
5059                 free -= xh_free_start % blocksize;
5060
5061         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5062              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5063              " %u\n", xs->not_found,
5064              (unsigned long long)bucket_blkno(xs->bucket),
5065              free, need, max_free, le16_to_cpu(xh->xh_free_start),
5066              le16_to_cpu(xh->xh_name_value_len));
5067
5068         if (free < need ||
5069             (xs->not_found &&
5070              count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5071                 if (need <= max_free &&
5072                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5073                         /*
5074                          * We can create the space by defragment. Since only the
5075                          * name/value will be moved, the xe shouldn't be changed
5076                          * in xs.
5077                          */
5078                         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5079                                                         xs->bucket);
5080                         if (ret) {
5081                                 mlog_errno(ret);
5082                                 goto out;
5083                         }
5084
5085                         xh_free_start = le16_to_cpu(xh->xh_free_start);
5086                         free = xh_free_start - header_size;
5087                         if (xh_free_start % blocksize < need)
5088                                 free -= xh_free_start % blocksize;
5089
5090                         if (free >= need)
5091                                 goto xattr_set;
5092
5093                         mlog(0, "Can't get enough space for xattr insert by "
5094                              "defragment. Need %u bytes, but we have %d, so "
5095                              "allocate new bucket for it.\n", need, free);
5096                 }
5097
5098                 /*
5099                  * We have to add new buckets or clusters and one
5100                  * allocation should leave us enough space for insert.
5101                  */
5102                 BUG_ON(allocation);
5103
5104                 /*
5105                  * We do not allow for overlapping ranges between buckets. And
5106                  * the maximum number of collisions we will allow for then is
5107                  * one bucket's worth, so check it here whether we need to
5108                  * add a new bucket for the insert.
5109                  */
5110                 ret = ocfs2_check_xattr_bucket_collision(inode,
5111                                                          xs->bucket,
5112                                                          xi->name);
5113                 if (ret) {
5114                         mlog_errno(ret);
5115                         goto out;
5116                 }
5117
5118                 ret = ocfs2_add_new_xattr_bucket(inode,
5119                                                  xs->xattr_bh,
5120                                                  xs->bucket,
5121                                                  ctxt);
5122                 if (ret) {
5123                         mlog_errno(ret);
5124                         goto out;
5125                 }
5126
5127                 /*
5128                  * ocfs2_add_new_xattr_bucket() will have updated
5129                  * xs->bucket if it moved, but it will not have updated
5130                  * any of the other search fields.  Thus, we drop it and
5131                  * re-search.  Everything should be cached, so it'll be
5132                  * quick.
5133                  */
5134                 ocfs2_xattr_bucket_relse(xs->bucket);
5135                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5136                                                    xi->name_index,
5137                                                    xi->name, xs);
5138                 if (ret && ret != -ENODATA)
5139                         goto out;
5140                 xs->not_found = ret;
5141                 allocation = 1;
5142                 goto try_again;
5143         }
5144
5145 xattr_set:
5146         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5147 out:
5148         mlog_exit(ret);
5149         return ret;
5150 }
5151
5152 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5153                                         struct ocfs2_xattr_bucket *bucket,
5154                                         void *para)
5155 {
5156         int ret = 0;
5157         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5158         u16 i;
5159         struct ocfs2_xattr_entry *xe;
5160         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5161         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5162         int credits = ocfs2_remove_extent_credits(osb->sb) +
5163                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5164
5165
5166         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5167
5168         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5169                 xe = &xh->xh_entries[i];
5170                 if (ocfs2_xattr_is_local(xe))
5171                         continue;
5172
5173                 ctxt.handle = ocfs2_start_trans(osb, credits);
5174                 if (IS_ERR(ctxt.handle)) {
5175                         ret = PTR_ERR(ctxt.handle);
5176                         mlog_errno(ret);
5177                         break;
5178                 }
5179
5180                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5181                                                         i, 0, &ctxt);
5182
5183                 ocfs2_commit_trans(osb, ctxt.handle);
5184                 if (ret) {
5185                         mlog_errno(ret);
5186                         break;
5187                 }
5188         }
5189
5190         ocfs2_schedule_truncate_log_flush(osb, 1);
5191         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5192         return ret;
5193 }
5194
5195 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5196                                           struct buffer_head *xb_bh)
5197 {
5198         struct ocfs2_xattr_block *xb =
5199                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5200         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5201         int ret = 0;
5202         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5203         u64 p_blkno;
5204
5205         if (le16_to_cpu(el->l_next_free_rec) == 0)
5206                 return 0;
5207
5208         while (name_hash > 0) {
5209                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5210                                           &e_cpos, &num_clusters, el);
5211                 if (ret) {
5212                         mlog_errno(ret);
5213                         goto out;
5214                 }
5215
5216                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5217                                                   ocfs2_delete_xattr_in_bucket,
5218                                                   NULL);
5219                 if (ret) {
5220                         mlog_errno(ret);
5221                         goto out;
5222                 }
5223
5224                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5225                                              p_blkno, e_cpos, num_clusters);
5226                 if (ret) {
5227                         mlog_errno(ret);
5228                         break;
5229                 }
5230
5231                 if (e_cpos == 0)
5232                         break;
5233
5234                 name_hash = e_cpos - 1;
5235         }
5236
5237 out:
5238         return ret;
5239 }
5240
5241 /*
5242  * 'security' attributes support
5243  */
5244 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5245                                         size_t list_size, const char *name,
5246                                         size_t name_len)
5247 {
5248         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5249         const size_t total_len = prefix_len + name_len + 1;
5250
5251         if (list && total_len <= list_size) {
5252                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5253                 memcpy(list + prefix_len, name, name_len);
5254                 list[prefix_len + name_len] = '\0';
5255         }
5256         return total_len;
5257 }
5258
5259 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5260                                     void *buffer, size_t size)
5261 {
5262         if (strcmp(name, "") == 0)
5263                 return -EINVAL;
5264         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5265                                buffer, size);
5266 }
5267
5268 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5269                                     const void *value, size_t size, int flags)
5270 {
5271         if (strcmp(name, "") == 0)
5272                 return -EINVAL;
5273
5274         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5275                                size, flags);
5276 }
5277
5278 int ocfs2_init_security_get(struct inode *inode,
5279                             struct inode *dir,
5280                             struct ocfs2_security_xattr_info *si)
5281 {
5282         return security_inode_init_security(inode, dir, &si->name, &si->value,
5283                                             &si->value_len);
5284 }
5285
5286 int ocfs2_init_security_set(handle_t *handle,
5287                             struct inode *inode,
5288                             struct buffer_head *di_bh,
5289                             struct ocfs2_security_xattr_info *si,
5290                             struct ocfs2_alloc_context *xattr_ac,
5291                             struct ocfs2_alloc_context *data_ac)
5292 {
5293         return ocfs2_xattr_set_handle(handle, inode, di_bh,
5294                                      OCFS2_XATTR_INDEX_SECURITY,
5295                                      si->name, si->value, si->value_len, 0,
5296                                      xattr_ac, data_ac);
5297 }
5298
5299 struct xattr_handler ocfs2_xattr_security_handler = {
5300         .prefix = XATTR_SECURITY_PREFIX,
5301         .list   = ocfs2_xattr_security_list,
5302         .get    = ocfs2_xattr_security_get,
5303         .set    = ocfs2_xattr_security_set,
5304 };
5305
5306 /*
5307  * 'trusted' attributes support
5308  */
5309 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5310                                        size_t list_size, const char *name,
5311                                        size_t name_len)
5312 {
5313         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5314         const size_t total_len = prefix_len + name_len + 1;
5315
5316         if (list && total_len <= list_size) {
5317                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5318                 memcpy(list + prefix_len, name, name_len);
5319                 list[prefix_len + name_len] = '\0';
5320         }
5321         return total_len;
5322 }
5323
5324 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5325                                    void *buffer, size_t size)
5326 {
5327         if (strcmp(name, "") == 0)
5328                 return -EINVAL;
5329         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5330                                buffer, size);
5331 }
5332
5333 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5334                                    const void *value, size_t size, int flags)
5335 {
5336         if (strcmp(name, "") == 0)
5337                 return -EINVAL;
5338
5339         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5340                                size, flags);
5341 }
5342
5343 struct xattr_handler ocfs2_xattr_trusted_handler = {
5344         .prefix = XATTR_TRUSTED_PREFIX,
5345         .list   = ocfs2_xattr_trusted_list,
5346         .get    = ocfs2_xattr_trusted_get,
5347         .set    = ocfs2_xattr_trusted_set,
5348 };
5349
5350 /*
5351  * 'user' attributes support
5352  */
5353 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5354                                     size_t list_size, const char *name,
5355                                     size_t name_len)
5356 {
5357         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5358         const size_t total_len = prefix_len + name_len + 1;
5359         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5360
5361         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5362                 return 0;
5363
5364         if (list && total_len <= list_size) {
5365                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5366                 memcpy(list + prefix_len, name, name_len);
5367                 list[prefix_len + name_len] = '\0';
5368         }
5369         return total_len;
5370 }
5371
5372 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5373                                 void *buffer, size_t size)
5374 {
5375         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5376
5377         if (strcmp(name, "") == 0)
5378                 return -EINVAL;
5379         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5380                 return -EOPNOTSUPP;
5381         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5382                                buffer, size);
5383 }
5384
5385 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5386                                 const void *value, size_t size, int flags)
5387 {
5388         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5389
5390         if (strcmp(name, "") == 0)
5391                 return -EINVAL;
5392         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5393                 return -EOPNOTSUPP;
5394
5395         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5396                                size, flags);
5397 }
5398
5399 struct xattr_handler ocfs2_xattr_user_handler = {
5400         .prefix = XATTR_USER_PREFIX,
5401         .list   = ocfs2_xattr_user_list,
5402         .get    = ocfs2_xattr_user_get,
5403         .set    = ocfs2_xattr_user_set,
5404 };