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                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 278                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 279                                                  bucket->bu_bhs,
 280                                                  bucket->bu_blocks,
 281                                                  &bucket_xh(bucket)->xh_check);
 282                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 283                 if (rc)
 284                         mlog_errno(rc);
 285         }
 286
 287         if (rc)
 288                 ocfs2_xattr_bucket_relse(bucket);
 289         return rc;
 290 }
 291
 292 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 293                                              struct ocfs2_xattr_bucket *bucket,
 294                                              int type)
 295 {
 296         int i, rc = 0;
 297
 298         for (i = 0; i < bucket->bu_blocks; i++) {
 299                 rc = ocfs2_journal_access(handle, bucket->bu_inode,
 300                                           bucket->bu_bhs[i], type);
 301                 if (rc) {
 302                         mlog_errno(rc);
 303                         break;
 304                 }
 305         }
 306
 307         return rc;
 308 }
 309
 310 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 311                                              struct ocfs2_xattr_bucket *bucket)
 312 {
 313         int i;
 314
 315         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 316         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 317                                    bucket->bu_bhs, bucket->bu_blocks,
 318                                    &bucket_xh(bucket)->xh_check);
 319         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 320
 321         for (i = 0; i < bucket->bu_blocks; i++)
 322                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 323 }
 324
 325 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 326                                          struct ocfs2_xattr_bucket *src)
 327 {
 328         int i;
 329         int blocksize = src->bu_inode->i_sb->s_blocksize;
 330
 331         BUG_ON(dest->bu_blocks != src->bu_blocks);
 332         BUG_ON(dest->bu_inode != src->bu_inode);
 333
 334         for (i = 0; i < src->bu_blocks; i++) {
 335                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 336                        blocksize);
 337         }
 338 }
 339
 340 static int ocfs2_validate_xattr_block(struct super_block *sb,
 341                                       struct buffer_head *bh)
 342 {
 343         int rc;
 344         struct ocfs2_xattr_block *xb =
 345                 (struct ocfs2_xattr_block *)bh->b_data;
 346
 347         mlog(0, "Validating xattr block %llu\n",
 348              (unsigned long long)bh->b_blocknr);
 349
 350         BUG_ON(!buffer_uptodate(bh));
 351
 352         /*
 353          * If the ecc fails, we return the error but otherwise
 354          * leave the filesystem running.  We know any error is
 355          * local to this block.
 356          */
 357         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 358         if (rc)
 359                 return rc;
 360
 361         /*
 362          * Errors after here are fatal
 363          */
 364
 365         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 366                 ocfs2_error(sb,
 367                             "Extended attribute block #%llu has bad "
 368                             "signature %.*s",
 369                             (unsigned long long)bh->b_blocknr, 7,
 370                             xb->xb_signature);
 371                 return -EINVAL;
 372         }
 373
 374         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 375                 ocfs2_error(sb,
 376                             "Extended attribute block #%llu has an "
 377                             "invalid xb_blkno of %llu",
 378                             (unsigned long long)bh->b_blocknr,
 379                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
 380                 return -EINVAL;
 381         }
 382
 383         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 384                 ocfs2_error(sb,
 385                             "Extended attribute block #%llu has an invalid "
 386                             "xb_fs_generation of #%u",
 387                             (unsigned long long)bh->b_blocknr,
 388                             le32_to_cpu(xb->xb_fs_generation));
 389                 return -EINVAL;
 390         }
 391
 392         return 0;
 393 }
 394
 395 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 396                                   struct buffer_head **bh)
 397 {
 398         int rc;
 399         struct buffer_head *tmp = *bh;
 400
 401         rc = ocfs2_read_block(inode, xb_blkno, &tmp,
 402                               ocfs2_validate_xattr_block);
 403
 404         /* If ocfs2_read_block() got us a new bh, pass it up. */
 405         if (!rc && !*bh)
 406                 *bh = tmp;
 407
 408         return rc;
 409 }
 410
 411 static inline const char *ocfs2_xattr_prefix(int name_index)
 412 {
 413         struct xattr_handler *handler = NULL;
 414
 415         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 416                 handler = ocfs2_xattr_handler_map[name_index];
 417
 418         return handler ? handler->prefix : NULL;
 419 }
 420
 421 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 422                                  const char *name,
 423                                  int name_len)
 424 {
 425         /* Get hash value of uuid from super block */
 426         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 427         int i;
 428
 429         /* hash extended attribute name */
 430         for (i = 0; i < name_len; i++) {
 431                 hash = (hash << OCFS2_HASH_SHIFT) ^
 432                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 433                        *name++;
 434         }
 435
 436         return hash;
 437 }
 438
 439 /*
 440  * ocfs2_xattr_hash_entry()
 441  *
 442  * Compute the hash of an extended attribute.
 443  */
 444 static void ocfs2_xattr_hash_entry(struct inode *inode,
 445                                    struct ocfs2_xattr_header *header,
 446                                    struct ocfs2_xattr_entry *entry)
 447 {
 448         u32 hash = 0;
 449         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
 450
 451         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
 452         entry->xe_name_hash = cpu_to_le32(hash);
 453
 454         return;
 455 }
 456
 457 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 458 {
 459         int size = 0;
 460
 461         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
 462                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 463         else
 464                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 465         size += sizeof(struct ocfs2_xattr_entry);
 466
 467         return size;
 468 }
 469
 470 int ocfs2_calc_security_init(struct inode *dir,
 471                              struct ocfs2_security_xattr_info *si,
 472                              int *want_clusters,
 473                              int *xattr_credits,
 474                              struct ocfs2_alloc_context **xattr_ac)
 475 {
 476         int ret = 0;
 477         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 478         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 479                                                  si->value_len);
 480
 481         /*
 482          * The max space of security xattr taken inline is
 483          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 484          * So reserve one metadata block for it is ok.
 485          */
 486         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 487             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 488                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 489                 if (ret) {
 490                         mlog_errno(ret);
 491                         return ret;
 492                 }
 493                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 494         }
 495
 496         /* reserve clusters for xattr value which will be set in B tree*/
 497         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 498                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 499                                                             si->value_len);
 500
 501                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 502                                                            new_clusters);
 503                 *want_clusters += new_clusters;
 504         }
 505         return ret;
 506 }
 507
 508 int ocfs2_calc_xattr_init(struct inode *dir,
 509                           struct buffer_head *dir_bh,
 510                           int mode,
 511                           struct ocfs2_security_xattr_info *si,
 512                           int *want_clusters,
 513                           int *xattr_credits,
 514                           struct ocfs2_alloc_context **xattr_ac)
 515 {
 516         int ret = 0;
 517         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 518         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 519
 520         if (si->enable)
 521                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 522                                                      si->value_len);
 523
 524         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 525                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 526                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 527                                         "", NULL, 0);
 528                 if (acl_len > 0) {
 529                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 530                         if (S_ISDIR(mode))
 531                                 a_size <<= 1;
 532                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 533                         mlog_errno(ret);
 534                         return ret;
 535                 }
 536         }
 537
 538         if (!(s_size + a_size))
 539                 return ret;
 540
 541         /*
 542          * The max space of security xattr taken inline is
 543          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 544          * The max space of acl xattr taken inline is
 545          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 546          * when blocksize = 512, may reserve one more cluser for
 547          * xattr bucket, otherwise reserve one metadata block
 548          * for them is ok.
 549          */
 550         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 551             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 552                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 553                 if (ret) {
 554                         mlog_errno(ret);
 555                         return ret;
 556                 }
 557                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 558         }
 559
 560         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 561             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 562                 *want_clusters += 1;
 563                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 564         }
 565
 566         /*
 567          * reserve credits and clusters for xattrs which has large value
 568          * and have to be set outside
 569          */
 570         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 571                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 572                                                         si->value_len);
 573                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 574                                                            new_clusters);
 575                 *want_clusters += new_clusters;
 576         }
 577         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 578             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 579                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 580                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 581                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 582                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 583                                                            new_clusters);
 584                 *want_clusters += new_clusters;
 585         }
 586
 587         return ret;
 588 }
 589
 590 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 591                                          u32 clusters_to_add,
 592                                          struct ocfs2_xattr_value_buf *vb,
 593                                          struct ocfs2_xattr_set_ctxt *ctxt)
 594 {
 595         int status = 0;
 596         handle_t *handle = ctxt->handle;
 597         enum ocfs2_alloc_restarted why;
 598         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 599         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 600         struct ocfs2_extent_tree et;
 601
 602         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
 603
 604         ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
 605
 606         status = vb->vb_access(handle, inode, vb->vb_bh,
 607                               OCFS2_JOURNAL_ACCESS_WRITE);
 608         if (status < 0) {
 609                 mlog_errno(status);
 610                 goto leave;
 611         }
 612
 613         prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 614         status = ocfs2_add_clusters_in_btree(osb,
 615                                              inode,
 616                                              &logical_start,
 617                                              clusters_to_add,
 618                                              0,
 619                                              &et,
 620                                              handle,
 621                                              ctxt->data_ac,
 622                                              ctxt->meta_ac,
 623                                              &why);
 624         if (status < 0) {
 625                 mlog_errno(status);
 626                 goto leave;
 627         }
 628
 629         status = ocfs2_journal_dirty(handle, vb->vb_bh);
 630         if (status < 0) {
 631                 mlog_errno(status);
 632                 goto leave;
 633         }
 634
 635         clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
 636
 637         /*
 638          * We should have already allocated enough space before the transaction,
 639          * so no need to restart.
 640          */
 641         BUG_ON(why != RESTART_NONE || clusters_to_add);
 642
 643 leave:
 644
 645         return status;
 646 }
 647
 648 static int __ocfs2_remove_xattr_range(struct inode *inode,
 649                                       struct ocfs2_xattr_value_buf *vb,
 650                                       u32 cpos, u32 phys_cpos, u32 len,
 651                                       struct ocfs2_xattr_set_ctxt *ctxt)
 652 {
 653         int ret;
 654         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 655         handle_t *handle = ctxt->handle;
 656         struct ocfs2_extent_tree et;
 657
 658         ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
 659
 660         ret = vb->vb_access(handle, inode, vb->vb_bh,
 661                             OCFS2_JOURNAL_ACCESS_WRITE);
 662         if (ret) {
 663                 mlog_errno(ret);
 664                 goto out;
 665         }
 666
 667         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
 668                                   &ctxt->dealloc);
 669         if (ret) {
 670                 mlog_errno(ret);
 671                 goto out;
 672         }
 673
 674         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 675
 676         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
 677         if (ret) {
 678                 mlog_errno(ret);
 679                 goto out;
 680         }
 681
 682         ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
 683         if (ret)
 684                 mlog_errno(ret);
 685
 686 out:
 687         return ret;
 688 }
 689
 690 static int ocfs2_xattr_shrink_size(struct inode *inode,
 691                                    u32 old_clusters,
 692                                    u32 new_clusters,
 693                                    struct ocfs2_xattr_value_buf *vb,
 694                                    struct ocfs2_xattr_set_ctxt *ctxt)
 695 {
 696         int ret = 0;
 697         u32 trunc_len, cpos, phys_cpos, alloc_size;
 698         u64 block;
 699
 700         if (old_clusters <= new_clusters)
 701                 return 0;
 702
 703         cpos = new_clusters;
 704         trunc_len = old_clusters - new_clusters;
 705         while (trunc_len) {
 706                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 707                                                &alloc_size,
 708                                                &vb->vb_xv->xr_list);
 709                 if (ret) {
 710                         mlog_errno(ret);
 711                         goto out;
 712                 }
 713
 714                 if (alloc_size > trunc_len)
 715                         alloc_size = trunc_len;
 716
 717                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 718                                                  phys_cpos, alloc_size,
 719                                                  ctxt);
 720                 if (ret) {
 721                         mlog_errno(ret);
 722                         goto out;
 723                 }
 724
 725                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 726                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
 727                                                        alloc_size);
 728                 cpos += alloc_size;
 729                 trunc_len -= alloc_size;
 730         }
 731
 732 out:
 733         return ret;
 734 }
 735
 736 static int ocfs2_xattr_value_truncate(struct inode *inode,
 737                                       struct ocfs2_xattr_value_buf *vb,
 738                                       int len,
 739                                       struct ocfs2_xattr_set_ctxt *ctxt)
 740 {
 741         int ret;
 742         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 743         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 744
 745         if (new_clusters == old_clusters)
 746                 return 0;
 747
 748         if (new_clusters > old_clusters)
 749                 ret = ocfs2_xattr_extend_allocation(inode,
 750                                                     new_clusters - old_clusters,
 751                                                     vb, ctxt);
 752         else
 753                 ret = ocfs2_xattr_shrink_size(inode,
 754                                               old_clusters, new_clusters,
 755                                               vb, ctxt);
 756
 757         return ret;
 758 }
 759
 760 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 761                                   size_t *result, const char *prefix,
 762                                   const char *name, int name_len)
 763 {
 764         char *p = buffer + *result;
 765         int prefix_len = strlen(prefix);
 766         int total_len = prefix_len + name_len + 1;
 767
 768         *result += total_len;
 769
 770         /* we are just looking for how big our buffer needs to be */
 771         if (!size)
 772                 return 0;
 773
 774         if (*result > size)
 775                 return -ERANGE;
 776
 777         memcpy(p, prefix, prefix_len);
 778         memcpy(p + prefix_len, name, name_len);
 779         p[prefix_len + name_len] = '\0';
 780
 781         return 0;
 782 }
 783
 784 static int ocfs2_xattr_list_entries(struct inode *inode,
 785                                     struct ocfs2_xattr_header *header,
 786                                     char *buffer, size_t buffer_size)
 787 {
 788         size_t result = 0;
 789         int i, type, ret;
 790         const char *prefix, *name;
 791
 792         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 793                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 794                 type = ocfs2_xattr_get_type(entry);
 795                 prefix = ocfs2_xattr_prefix(type);
 796
 797                 if (prefix) {
 798                         name = (const char *)header +
 799                                 le16_to_cpu(entry->xe_name_offset);
 800
 801                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 802                                                      &result, prefix, name,
 803                                                      entry->xe_name_len);
 804                         if (ret)
 805                                 return ret;
 806                 }
 807         }
 808
 809         return result;
 810 }
 811
 812 static int ocfs2_xattr_ibody_list(struct inode *inode,
 813                                   struct ocfs2_dinode *di,
 814                                   char *buffer,
 815                                   size_t buffer_size)
 816 {
 817         struct ocfs2_xattr_header *header = NULL;
 818         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 819         int ret = 0;
 820
 821         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 822                 return ret;
 823
 824         header = (struct ocfs2_xattr_header *)
 825                  ((void *)di + inode->i_sb->s_blocksize -
 826                  le16_to_cpu(di->i_xattr_inline_size));
 827
 828         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 829
 830         return ret;
 831 }
 832
 833 static int ocfs2_xattr_block_list(struct inode *inode,
 834                                   struct ocfs2_dinode *di,
 835                                   char *buffer,
 836                                   size_t buffer_size)
 837 {
 838         struct buffer_head *blk_bh = NULL;
 839         struct ocfs2_xattr_block *xb;
 840         int ret = 0;
 841
 842         if (!di->i_xattr_loc)
 843                 return ret;
 844
 845         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
 846                                      &blk_bh);
 847         if (ret < 0) {
 848                 mlog_errno(ret);
 849                 return ret;
 850         }
 851
 852         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 853         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 854                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 855                 ret = ocfs2_xattr_list_entries(inode, header,
 856                                                buffer, buffer_size);
 857         } else {
 858                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
 859                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
 860                                                    buffer, buffer_size);
 861         }
 862
 863         brelse(blk_bh);
 864
 865         return ret;
 866 }
 867
 868 ssize_t ocfs2_listxattr(struct dentry *dentry,
 869                         char *buffer,
 870                         size_t size)
 871 {
 872         int ret = 0, i_ret = 0, b_ret = 0;
 873         struct buffer_head *di_bh = NULL;
 874         struct ocfs2_dinode *di = NULL;
 875         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
 876
 877         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
 878                 return -EOPNOTSUPP;
 879
 880         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 881                 return ret;
 882
 883         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
 884         if (ret < 0) {
 885                 mlog_errno(ret);
 886                 return ret;
 887         }
 888
 889         di = (struct ocfs2_dinode *)di_bh->b_data;
 890
 891         down_read(&oi->ip_xattr_sem);
 892         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
 893         if (i_ret < 0)
 894                 b_ret = 0;
 895         else {
 896                 if (buffer) {
 897                         buffer += i_ret;
 898                         size -= i_ret;
 899                 }
 900                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
 901                                                buffer, size);
 902                 if (b_ret < 0)
 903                         i_ret = 0;
 904         }
 905         up_read(&oi->ip_xattr_sem);
 906         ocfs2_inode_unlock(dentry->d_inode, 0);
 907
 908         brelse(di_bh);
 909
 910         return i_ret + b_ret;
 911 }
 912
 913 static int ocfs2_xattr_find_entry(int name_index,
 914                                   const char *name,
 915                                   struct ocfs2_xattr_search *xs)
 916 {
 917         struct ocfs2_xattr_entry *entry;
 918         size_t name_len;
 919         int i, cmp = 1;
 920
 921         if (name == NULL)
 922                 return -EINVAL;
 923
 924         name_len = strlen(name);
 925         entry = xs->here;
 926         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
 927                 cmp = name_index - ocfs2_xattr_get_type(entry);
 928                 if (!cmp)
 929                         cmp = name_len - entry->xe_name_len;
 930                 if (!cmp)
 931                         cmp = memcmp(name, (xs->base +
 932                                      le16_to_cpu(entry->xe_name_offset)),
 933                                      name_len);
 934                 if (cmp == 0)
 935                         break;
 936                 entry += 1;
 937         }
 938         xs->here = entry;
 939
 940         return cmp ? -ENODATA : 0;
 941 }
 942
 943 static int ocfs2_xattr_get_value_outside(struct inode *inode,
 944                                          struct ocfs2_xattr_value_root *xv,
 945                                          void *buffer,
 946                                          size_t len)
 947 {
 948         u32 cpos, p_cluster, num_clusters, bpc, clusters;
 949         u64 blkno;
 950         int i, ret = 0;
 951         size_t cplen, blocksize;
 952         struct buffer_head *bh = NULL;
 953         struct ocfs2_extent_list *el;
 954
 955         el = &xv->xr_list;
 956         clusters = le32_to_cpu(xv->xr_clusters);
 957         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 958         blocksize = inode->i_sb->s_blocksize;
 959
 960         cpos = 0;
 961         while (cpos < clusters) {
 962                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 963                                                &num_clusters, el);
 964                 if (ret) {
 965                         mlog_errno(ret);
 966                         goto out;
 967                 }
 968
 969                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 970                 /* Copy ocfs2_xattr_value */
 971                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 972                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
 973                         if (ret) {
 974                                 mlog_errno(ret);
 975                                 goto out;
 976                         }
 977
 978                         cplen = len >= blocksize ? blocksize : len;
 979                         memcpy(buffer, bh->b_data, cplen);
 980                         len -= cplen;
 981                         buffer += cplen;
 982
 983                         brelse(bh);
 984                         bh = NULL;
 985                         if (len == 0)
 986                                 break;
 987                 }
 988                 cpos += num_clusters;
 989         }
 990 out:
 991         return ret;
 992 }
 993
 994 static int ocfs2_xattr_ibody_get(struct inode *inode,
 995                                  int name_index,
 996                                  const char *name,
 997                                  void *buffer,
 998                                  size_t buffer_size,
 999                                  struct ocfs2_xattr_search *xs)
1000 {
1001         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1002         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1003         struct ocfs2_xattr_value_root *xv;
1004         size_t size;
1005         int ret = 0;
1006
1007         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1008                 return -ENODATA;
1009
1010         xs->end = (void *)di + inode->i_sb->s_blocksize;
1011         xs->header = (struct ocfs2_xattr_header *)
1012                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1013         xs->base = (void *)xs->header;
1014         xs->here = xs->header->xh_entries;
1015
1016         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1017         if (ret)
1018                 return ret;
1019         size = le64_to_cpu(xs->here->xe_value_size);
1020         if (buffer) {
1021                 if (size > buffer_size)
1022                         return -ERANGE;
1023                 if (ocfs2_xattr_is_local(xs->here)) {
1024                         memcpy(buffer, (void *)xs->base +
1025                                le16_to_cpu(xs->here->xe_name_offset) +
1026                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1027                 } else {
1028                         xv = (struct ocfs2_xattr_value_root *)
1029                                 (xs->base + le16_to_cpu(
1030                                  xs->here->xe_name_offset) +
1031                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1032                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1033                                                             buffer, size);
1034                         if (ret < 0) {
1035                                 mlog_errno(ret);
1036                                 return ret;
1037                         }
1038                 }
1039         }
1040
1041         return size;
1042 }
1043
1044 static int ocfs2_xattr_block_get(struct inode *inode,
1045                                  int name_index,
1046                                  const char *name,
1047                                  void *buffer,
1048                                  size_t buffer_size,
1049                                  struct ocfs2_xattr_search *xs)
1050 {
1051         struct ocfs2_xattr_block *xb;
1052         struct ocfs2_xattr_value_root *xv;
1053         size_t size;
1054         int ret = -ENODATA, name_offset, name_len, block_off, i;
1055
1056         xs->bucket = ocfs2_xattr_bucket_new(inode);
1057         if (!xs->bucket) {
1058                 ret = -ENOMEM;
1059                 mlog_errno(ret);
1060                 goto cleanup;
1061         }
1062
1063         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1064         if (ret) {
1065                 mlog_errno(ret);
1066                 goto cleanup;
1067         }
1068
1069         if (xs->not_found) {
1070                 ret = -ENODATA;
1071                 goto cleanup;
1072         }
1073
1074         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1075         size = le64_to_cpu(xs->here->xe_value_size);
1076         if (buffer) {
1077                 ret = -ERANGE;
1078                 if (size > buffer_size)
1079                         goto cleanup;
1080
1081                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1082                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1083                 i = xs->here - xs->header->xh_entries;
1084
1085                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1086                         ret = ocfs2_xattr_bucket_get_name_value(inode,
1087                                                                 bucket_xh(xs->bucket),
1088                                                                 i,
1089                                                                 &block_off,
1090                                                                 &name_offset);
1091                         xs->base = bucket_block(xs->bucket, block_off);
1092                 }
1093                 if (ocfs2_xattr_is_local(xs->here)) {
1094                         memcpy(buffer, (void *)xs->base +
1095                                name_offset + name_len, size);
1096                 } else {
1097                         xv = (struct ocfs2_xattr_value_root *)
1098                                 (xs->base + name_offset + name_len);
1099                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1100                                                             buffer, size);
1101                         if (ret < 0) {
1102                                 mlog_errno(ret);
1103                                 goto cleanup;
1104                         }
1105                 }
1106         }
1107         ret = size;
1108 cleanup:
1109         ocfs2_xattr_bucket_free(xs->bucket);
1110
1111         brelse(xs->xattr_bh);
1112         xs->xattr_bh = NULL;
1113         return ret;
1114 }
1115
1116 int ocfs2_xattr_get_nolock(struct inode *inode,
1117                            struct buffer_head *di_bh,
1118                            int name_index,
1119                            const char *name,
1120                            void *buffer,
1121                            size_t buffer_size)
1122 {
1123         int ret;
1124         struct ocfs2_dinode *di = NULL;
1125         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1126         struct ocfs2_xattr_search xis = {
1127                 .not_found = -ENODATA,
1128         };
1129         struct ocfs2_xattr_search xbs = {
1130                 .not_found = -ENODATA,
1131         };
1132
1133         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1134                 return -EOPNOTSUPP;
1135
1136         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1137                 ret = -ENODATA;
1138
1139         xis.inode_bh = xbs.inode_bh = di_bh;
1140         di = (struct ocfs2_dinode *)di_bh->b_data;
1141
1142         down_read(&oi->ip_xattr_sem);
1143         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1144                                     buffer_size, &xis);
1145         if (ret == -ENODATA && di->i_xattr_loc)
1146                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1147                                             buffer_size, &xbs);
1148         up_read(&oi->ip_xattr_sem);
1149
1150         return ret;
1151 }
1152
1153 /* ocfs2_xattr_get()
1154  *
1155  * Copy an extended attribute into the buffer provided.
1156  * Buffer is NULL to compute the size of buffer required.
1157  */
1158 static int ocfs2_xattr_get(struct inode *inode,
1159                            int name_index,
1160                            const char *name,
1161                            void *buffer,
1162                            size_t buffer_size)
1163 {
1164         int ret;
1165         struct buffer_head *di_bh = NULL;
1166
1167         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1168         if (ret < 0) {
1169                 mlog_errno(ret);
1170                 return ret;
1171         }
1172         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1173                                      name, buffer, buffer_size);
1174
1175         ocfs2_inode_unlock(inode, 0);
1176
1177         brelse(di_bh);
1178
1179         return ret;
1180 }
1181
1182 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1183                                            handle_t *handle,
1184                                            struct ocfs2_xattr_value_root *xv,
1185                                            const void *value,
1186                                            int value_len)
1187 {
1188         int ret = 0, i, cp_len;
1189         u16 blocksize = inode->i_sb->s_blocksize;
1190         u32 p_cluster, num_clusters;
1191         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1192         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1193         u64 blkno;
1194         struct buffer_head *bh = NULL;
1195
1196         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1197
1198         while (cpos < clusters) {
1199                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1200                                                &num_clusters, &xv->xr_list);
1201                 if (ret) {
1202                         mlog_errno(ret);
1203                         goto out;
1204                 }
1205
1206                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1207
1208                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1209                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
1210                         if (ret) {
1211                                 mlog_errno(ret);
1212                                 goto out;
1213                         }
1214
1215                         ret = ocfs2_journal_access(handle,
1216                                                    inode,
1217                                                    bh,
1218                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1219                         if (ret < 0) {
1220                                 mlog_errno(ret);
1221                                 goto out;
1222                         }
1223
1224                         cp_len = value_len > blocksize ? blocksize : value_len;
1225                         memcpy(bh->b_data, value, cp_len);
1226                         value_len -= cp_len;
1227                         value += cp_len;
1228                         if (cp_len < blocksize)
1229                                 memset(bh->b_data + cp_len, 0,
1230                                        blocksize - cp_len);
1231
1232                         ret = ocfs2_journal_dirty(handle, bh);
1233                         if (ret < 0) {
1234                                 mlog_errno(ret);
1235                                 goto out;
1236                         }
1237                         brelse(bh);
1238                         bh = NULL;
1239
1240                         /*
1241                          * XXX: do we need to empty all the following
1242                          * blocks in this cluster?
1243                          */
1244                         if (!value_len)
1245                                 break;
1246                 }
1247                 cpos += num_clusters;
1248         }
1249 out:
1250         brelse(bh);
1251
1252         return ret;
1253 }
1254
1255 static int ocfs2_xattr_cleanup(struct inode *inode,
1256                                handle_t *handle,
1257                                struct ocfs2_xattr_info *xi,
1258                                struct ocfs2_xattr_search *xs,
1259                                struct ocfs2_xattr_value_buf *vb,
1260                                size_t offs)
1261 {
1262         int ret = 0;
1263         size_t name_len = strlen(xi->name);
1264         void *val = xs->base + offs;
1265         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1266
1267         ret = vb->vb_access(handle, inode, vb->vb_bh,
1268                             OCFS2_JOURNAL_ACCESS_WRITE);
1269         if (ret) {
1270                 mlog_errno(ret);
1271                 goto out;
1272         }
1273         /* Decrease xattr count */
1274         le16_add_cpu(&xs->header->xh_count, -1);
1275         /* Remove the xattr entry and tree root which has already be set*/
1276         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1277         memset(val, 0, size);
1278
1279         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1280         if (ret < 0)
1281                 mlog_errno(ret);
1282 out:
1283         return ret;
1284 }
1285
1286 static int ocfs2_xattr_update_entry(struct inode *inode,
1287                                     handle_t *handle,
1288                                     struct ocfs2_xattr_info *xi,
1289                                     struct ocfs2_xattr_search *xs,
1290                                     struct ocfs2_xattr_value_buf *vb,
1291                                     size_t offs)
1292 {
1293         int ret;
1294
1295         ret = vb->vb_access(handle, inode, vb->vb_bh,
1296                             OCFS2_JOURNAL_ACCESS_WRITE);
1297         if (ret) {
1298                 mlog_errno(ret);
1299                 goto out;
1300         }
1301
1302         xs->here->xe_name_offset = cpu_to_le16(offs);
1303         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1304         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1305                 ocfs2_xattr_set_local(xs->here, 1);
1306         else
1307                 ocfs2_xattr_set_local(xs->here, 0);
1308         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1309
1310         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1311         if (ret < 0)
1312                 mlog_errno(ret);
1313 out:
1314         return ret;
1315 }
1316
1317 /*
1318  * ocfs2_xattr_set_value_outside()
1319  *
1320  * Set large size value in B tree.
1321  */
1322 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1323                                          struct ocfs2_xattr_info *xi,
1324                                          struct ocfs2_xattr_search *xs,
1325                                          struct ocfs2_xattr_set_ctxt *ctxt,
1326                                          struct ocfs2_xattr_value_buf *vb,
1327                                          size_t offs)
1328 {
1329         size_t name_len = strlen(xi->name);
1330         void *val = xs->base + offs;
1331         struct ocfs2_xattr_value_root *xv = NULL;
1332         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1333         int ret = 0;
1334
1335         memset(val, 0, size);
1336         memcpy(val, xi->name, name_len);
1337         xv = (struct ocfs2_xattr_value_root *)
1338                 (val + OCFS2_XATTR_SIZE(name_len));
1339         xv->xr_clusters = 0;
1340         xv->xr_last_eb_blk = 0;
1341         xv->xr_list.l_tree_depth = 0;
1342         xv->xr_list.l_count = cpu_to_le16(1);
1343         xv->xr_list.l_next_free_rec = 0;
1344         vb->vb_xv = xv;
1345
1346         ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1347         if (ret < 0) {
1348                 mlog_errno(ret);
1349                 return ret;
1350         }
1351         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1352         if (ret < 0) {
1353                 mlog_errno(ret);
1354                 return ret;
1355         }
1356         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
1357                                               xi->value, xi->value_len);
1358         if (ret < 0)
1359                 mlog_errno(ret);
1360
1361         return ret;
1362 }
1363
1364 /*
1365  * ocfs2_xattr_set_entry_local()
1366  *
1367  * Set, replace or remove extended attribute in local.
1368  */
1369 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1370                                         struct ocfs2_xattr_info *xi,
1371                                         struct ocfs2_xattr_search *xs,
1372                                         struct ocfs2_xattr_entry *last,
1373                                         size_t min_offs)
1374 {
1375         size_t name_len = strlen(xi->name);
1376         int i;
1377
1378         if (xi->value && xs->not_found) {
1379                 /* Insert the new xattr entry. */
1380                 le16_add_cpu(&xs->header->xh_count, 1);
1381                 ocfs2_xattr_set_type(last, xi->name_index);
1382                 ocfs2_xattr_set_local(last, 1);
1383                 last->xe_name_len = name_len;
1384         } else {
1385                 void *first_val;
1386                 void *val;
1387                 size_t offs, size;
1388
1389                 first_val = xs->base + min_offs;
1390                 offs = le16_to_cpu(xs->here->xe_name_offset);
1391                 val = xs->base + offs;
1392
1393                 if (le64_to_cpu(xs->here->xe_value_size) >
1394                     OCFS2_XATTR_INLINE_SIZE)
1395                         size = OCFS2_XATTR_SIZE(name_len) +
1396                                 OCFS2_XATTR_ROOT_SIZE;
1397                 else
1398                         size = OCFS2_XATTR_SIZE(name_len) +
1399                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1400
1401                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1402                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1403                         /* The old and the new value have the
1404                            same size. Just replace the value. */
1405                         ocfs2_xattr_set_local(xs->here, 1);
1406                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1407                         /* Clear value bytes. */
1408                         memset(val + OCFS2_XATTR_SIZE(name_len),
1409                                0,
1410                                OCFS2_XATTR_SIZE(xi->value_len));
1411                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1412                                xi->value,
1413                                xi->value_len);
1414                         return;
1415                 }
1416                 /* Remove the old name+value. */
1417                 memmove(first_val + size, first_val, val - first_val);
1418                 memset(first_val, 0, size);
1419                 xs->here->xe_name_hash = 0;
1420                 xs->here->xe_name_offset = 0;
1421                 ocfs2_xattr_set_local(xs->here, 1);
1422                 xs->here->xe_value_size = 0;
1423
1424                 min_offs += size;
1425
1426                 /* Adjust all value offsets. */
1427                 last = xs->header->xh_entries;
1428                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1429                         size_t o = le16_to_cpu(last->xe_name_offset);
1430
1431                         if (o < offs)
1432                                 last->xe_name_offset = cpu_to_le16(o + size);
1433                         last += 1;
1434                 }
1435
1436                 if (!xi->value) {
1437                         /* Remove the old entry. */
1438                         last -= 1;
1439                         memmove(xs->here, xs->here + 1,
1440                                 (void *)last - (void *)xs->here);
1441                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1442                         le16_add_cpu(&xs->header->xh_count, -1);
1443                 }
1444         }
1445         if (xi->value) {
1446                 /* Insert the new name+value. */
1447                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1448                                 OCFS2_XATTR_SIZE(xi->value_len);
1449                 void *val = xs->base + min_offs - size;
1450
1451                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1452                 memset(val, 0, size);
1453                 memcpy(val, xi->name, name_len);
1454                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1455                        xi->value,
1456                        xi->value_len);
1457                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1458                 ocfs2_xattr_set_local(xs->here, 1);
1459                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1460         }
1461
1462         return;
1463 }
1464
1465 /*
1466  * ocfs2_xattr_set_entry()
1467  *
1468  * Set extended attribute entry into inode or block.
1469  *
1470  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1471  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1472  * then set value in B tree with set_value_outside().
1473  */
1474 static int ocfs2_xattr_set_entry(struct inode *inode,
1475                                  struct ocfs2_xattr_info *xi,
1476                                  struct ocfs2_xattr_search *xs,
1477                                  struct ocfs2_xattr_set_ctxt *ctxt,
1478                                  int flag)
1479 {
1480         struct ocfs2_xattr_entry *last;
1481         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1482         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1483         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1484         size_t size_l = 0;
1485         handle_t *handle = ctxt->handle;
1486         int free, i, ret;
1487         struct ocfs2_xattr_info xi_l = {
1488                 .name_index = xi->name_index,
1489                 .name = xi->name,
1490                 .value = xi->value,
1491                 .value_len = xi->value_len,
1492         };
1493         struct ocfs2_xattr_value_buf vb = {
1494                 .vb_bh = xs->xattr_bh,
1495                 .vb_access = ocfs2_journal_access_di,
1496         };
1497
1498         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1499                 BUG_ON(xs->xattr_bh == xs->inode_bh);
1500                 vb.vb_access = ocfs2_journal_access_xb;
1501         } else
1502                 BUG_ON(xs->xattr_bh != xs->inode_bh);
1503
1504         /* Compute min_offs, last and free space. */
1505         last = xs->header->xh_entries;
1506
1507         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1508                 size_t offs = le16_to_cpu(last->xe_name_offset);
1509                 if (offs < min_offs)
1510                         min_offs = offs;
1511                 last += 1;
1512         }
1513
1514         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1515         if (free < 0)
1516                 return -EIO;
1517
1518         if (!xs->not_found) {
1519                 size_t size = 0;
1520                 if (ocfs2_xattr_is_local(xs->here))
1521                         size = OCFS2_XATTR_SIZE(name_len) +
1522                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1523                 else
1524                         size = OCFS2_XATTR_SIZE(name_len) +
1525                                 OCFS2_XATTR_ROOT_SIZE;
1526                 free += (size + sizeof(struct ocfs2_xattr_entry));
1527         }
1528         /* Check free space in inode or block */
1529         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1530                 if (free < sizeof(struct ocfs2_xattr_entry) +
1531                            OCFS2_XATTR_SIZE(name_len) +
1532                            OCFS2_XATTR_ROOT_SIZE) {
1533                         ret = -ENOSPC;
1534                         goto out;
1535                 }
1536                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1537                 xi_l.value = (void *)&def_xv;
1538                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1539         } else if (xi->value) {
1540                 if (free < sizeof(struct ocfs2_xattr_entry) +
1541                            OCFS2_XATTR_SIZE(name_len) +
1542                            OCFS2_XATTR_SIZE(xi->value_len)) {
1543                         ret = -ENOSPC;
1544                         goto out;
1545                 }
1546         }
1547
1548         if (!xs->not_found) {
1549                 /* For existing extended attribute */
1550                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1551                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1552                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1553                 void *val = xs->base + offs;
1554
1555                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1556                         /* Replace existing local xattr with tree root */
1557                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1558                                                             ctxt, &vb, offs);
1559                         if (ret < 0)
1560                                 mlog_errno(ret);
1561                         goto out;
1562                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1563                         /* For existing xattr which has value outside */
1564                         vb.vb_xv = (struct ocfs2_xattr_value_root *)
1565                                 (val + OCFS2_XATTR_SIZE(name_len));
1566
1567                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1568                                 /*
1569                                  * If new value need set outside also,
1570                                  * first truncate old value to new value,
1571                                  * then set new value with set_value_outside().
1572                                  */
1573                                 ret = ocfs2_xattr_value_truncate(inode,
1574                                                                  &vb,
1575                                                                  xi->value_len,
1576                                                                  ctxt);
1577                                 if (ret < 0) {
1578                                         mlog_errno(ret);
1579                                         goto out;
1580                                 }
1581
1582                                 ret = ocfs2_xattr_update_entry(inode,
1583                                                                handle,
1584                                                                xi,
1585                                                                xs,
1586                                                                &vb,
1587                                                                offs);
1588                                 if (ret < 0) {
1589                                         mlog_errno(ret);
1590                                         goto out;
1591                                 }
1592
1593                                 ret = __ocfs2_xattr_set_value_outside(inode,
1594                                                                 handle,
1595                                                                 vb.vb_xv,
1596                                                                 xi->value,
1597                                                                 xi->value_len);
1598                                 if (ret < 0)
1599                                         mlog_errno(ret);
1600                                 goto out;
1601                         } else {
1602                                 /*
1603                                  * If new value need set in local,
1604                                  * just trucate old value to zero.
1605                                  */
1606                                  ret = ocfs2_xattr_value_truncate(inode,
1607                                                                   &vb,
1608                                                                   0,
1609                                                                   ctxt);
1610                                 if (ret < 0)
1611                                         mlog_errno(ret);
1612                         }
1613                 }
1614         }
1615
1616         ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
1617                                       OCFS2_JOURNAL_ACCESS_WRITE);
1618         if (ret) {
1619                 mlog_errno(ret);
1620                 goto out;
1621         }
1622
1623         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1624                 ret = vb.vb_access(handle, inode, vb.vb_bh,
1625                                    OCFS2_JOURNAL_ACCESS_WRITE);
1626                 if (ret) {
1627                         mlog_errno(ret);
1628                         goto out;
1629                 }
1630         }
1631
1632         /*
1633          * Set value in local, include set tree root in local.
1634          * This is the first step for value size >INLINE_SIZE.
1635          */
1636         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1637
1638         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1639                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1640                 if (ret < 0) {
1641                         mlog_errno(ret);
1642                         goto out;
1643                 }
1644         }
1645
1646         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1647             (flag & OCFS2_INLINE_XATTR_FL)) {
1648                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1649                 unsigned int xattrsize = osb->s_xattr_inline_size;
1650
1651                 /*
1652                  * Adjust extent record count or inline data size
1653                  * to reserve space for extended attribute.
1654                  */
1655                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1656                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1657                         le16_add_cpu(&idata->id_count, -xattrsize);
1658                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1659                         struct ocfs2_extent_list *el = &di->id2.i_list;
1660                         le16_add_cpu(&el->l_count, -(xattrsize /
1661                                         sizeof(struct ocfs2_extent_rec)));
1662                 }
1663                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1664         }
1665         /* Update xattr flag */
1666         spin_lock(&oi->ip_lock);
1667         oi->ip_dyn_features |= flag;
1668         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1669         spin_unlock(&oi->ip_lock);
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 ocfs2_xattr_value_buf *vb,
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                         void *val;
1724
1725                         val = (void *)header +
1726                                 le16_to_cpu(entry->xe_name_offset);
1727                         vb->vb_xv = (struct ocfs2_xattr_value_root *)
1728                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1729                         ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1730                         if (ret < 0) {
1731                                 mlog_errno(ret);
1732                                 break;
1733                         }
1734                 }
1735         }
1736
1737         ocfs2_commit_trans(osb, ctxt.handle);
1738         ocfs2_schedule_truncate_log_flush(osb, 1);
1739         ocfs2_run_deallocs(osb, &ctxt.dealloc);
1740 out:
1741         return ret;
1742 }
1743
1744 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1745                                     struct buffer_head *di_bh)
1746 {
1747
1748         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1749         struct ocfs2_xattr_header *header;
1750         int ret;
1751         struct ocfs2_xattr_value_buf vb = {
1752                 .vb_bh = di_bh,
1753                 .vb_access = ocfs2_journal_access_di,
1754         };
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, &vb, 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         struct ocfs2_xattr_value_buf vb = {
1771                 .vb_bh = blk_bh,
1772                 .vb_access = ocfs2_journal_access_xb,
1773         };
1774
1775         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1776         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1777                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1778                 ret = ocfs2_remove_value_outside(inode, &vb, header);
1779         } else
1780                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1781
1782         return ret;
1783 }
1784
1785 static int ocfs2_xattr_free_block(struct inode *inode,
1786                                   u64 block)
1787 {
1788         struct inode *xb_alloc_inode;
1789         struct buffer_head *xb_alloc_bh = NULL;
1790         struct buffer_head *blk_bh = NULL;
1791         struct ocfs2_xattr_block *xb;
1792         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1793         handle_t *handle;
1794         int ret = 0;
1795         u64 blk, bg_blkno;
1796         u16 bit;
1797
1798         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1799         if (ret < 0) {
1800                 mlog_errno(ret);
1801                 goto out;
1802         }
1803
1804         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1805         if (ret < 0) {
1806                 mlog_errno(ret);
1807                 goto out;
1808         }
1809
1810         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1811         blk = le64_to_cpu(xb->xb_blkno);
1812         bit = le16_to_cpu(xb->xb_suballoc_bit);
1813         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1814
1815         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1816                                 EXTENT_ALLOC_SYSTEM_INODE,
1817                                 le16_to_cpu(xb->xb_suballoc_slot));
1818         if (!xb_alloc_inode) {
1819                 ret = -ENOMEM;
1820                 mlog_errno(ret);
1821                 goto out;
1822         }
1823         mutex_lock(&xb_alloc_inode->i_mutex);
1824
1825         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1826         if (ret < 0) {
1827                 mlog_errno(ret);
1828                 goto out_mutex;
1829         }
1830
1831         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1832         if (IS_ERR(handle)) {
1833                 ret = PTR_ERR(handle);
1834                 mlog_errno(ret);
1835                 goto out_unlock;
1836         }
1837
1838         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1839                                        bit, bg_blkno, 1);
1840         if (ret < 0)
1841                 mlog_errno(ret);
1842
1843         ocfs2_commit_trans(osb, handle);
1844 out_unlock:
1845         ocfs2_inode_unlock(xb_alloc_inode, 1);
1846         brelse(xb_alloc_bh);
1847 out_mutex:
1848         mutex_unlock(&xb_alloc_inode->i_mutex);
1849         iput(xb_alloc_inode);
1850 out:
1851         brelse(blk_bh);
1852         return ret;
1853 }
1854
1855 /*
1856  * ocfs2_xattr_remove()
1857  *
1858  * Free extended attribute resources associated with this inode.
1859  */
1860 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1861 {
1862         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1863         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1864         handle_t *handle;
1865         int ret;
1866
1867         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1868                 return 0;
1869
1870         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1871                 return 0;
1872
1873         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1874                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1875                 if (ret < 0) {
1876                         mlog_errno(ret);
1877                         goto out;
1878                 }
1879         }
1880
1881         if (di->i_xattr_loc) {
1882                 ret = ocfs2_xattr_free_block(inode,
1883                                              le64_to_cpu(di->i_xattr_loc));
1884                 if (ret < 0) {
1885                         mlog_errno(ret);
1886                         goto out;
1887                 }
1888         }
1889
1890         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1891                                    OCFS2_INODE_UPDATE_CREDITS);
1892         if (IS_ERR(handle)) {
1893                 ret = PTR_ERR(handle);
1894                 mlog_errno(ret);
1895                 goto out;
1896         }
1897         ret = ocfs2_journal_access_di(handle, inode, di_bh,
1898                                       OCFS2_JOURNAL_ACCESS_WRITE);
1899         if (ret) {
1900                 mlog_errno(ret);
1901                 goto out_commit;
1902         }
1903
1904         di->i_xattr_loc = 0;
1905
1906         spin_lock(&oi->ip_lock);
1907         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1908         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1909         spin_unlock(&oi->ip_lock);
1910
1911         ret = ocfs2_journal_dirty(handle, di_bh);
1912         if (ret < 0)
1913                 mlog_errno(ret);
1914 out_commit:
1915         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1916 out:
1917         return ret;
1918 }
1919
1920 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1921                                         struct ocfs2_dinode *di)
1922 {
1923         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1924         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1925         int free;
1926
1927         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1928                 return 0;
1929
1930         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1931                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1932                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1933         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1934                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1935                         le64_to_cpu(di->i_size);
1936         } else {
1937                 struct ocfs2_extent_list *el = &di->id2.i_list;
1938                 free = (le16_to_cpu(el->l_count) -
1939                         le16_to_cpu(el->l_next_free_rec)) *
1940                         sizeof(struct ocfs2_extent_rec);
1941         }
1942         if (free >= xattrsize)
1943                 return 1;
1944
1945         return 0;
1946 }
1947
1948 /*
1949  * ocfs2_xattr_ibody_find()
1950  *
1951  * Find extended attribute in inode block and
1952  * fill search info into struct ocfs2_xattr_search.
1953  */
1954 static int ocfs2_xattr_ibody_find(struct inode *inode,
1955                                   int name_index,
1956                                   const char *name,
1957                                   struct ocfs2_xattr_search *xs)
1958 {
1959         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1960         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1961         int ret;
1962         int has_space = 0;
1963
1964         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1965                 return 0;
1966
1967         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1968                 down_read(&oi->ip_alloc_sem);
1969                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1970                 up_read(&oi->ip_alloc_sem);
1971                 if (!has_space)
1972                         return 0;
1973         }
1974
1975         xs->xattr_bh = xs->inode_bh;
1976         xs->end = (void *)di + inode->i_sb->s_blocksize;
1977         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1978                 xs->header = (struct ocfs2_xattr_header *)
1979                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1980         else
1981                 xs->header = (struct ocfs2_xattr_header *)
1982                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1983         xs->base = (void *)xs->header;
1984         xs->here = xs->header->xh_entries;
1985
1986         /* Find the named attribute. */
1987         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1988                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1989                 if (ret && ret != -ENODATA)
1990                         return ret;
1991                 xs->not_found = ret;
1992         }
1993
1994         return 0;
1995 }
1996
1997 /*
1998  * ocfs2_xattr_ibody_set()
1999  *
2000  * Set, replace or remove an extended attribute into inode block.
2001  *
2002  */
2003 static int ocfs2_xattr_ibody_set(struct inode *inode,
2004                                  struct ocfs2_xattr_info *xi,
2005                                  struct ocfs2_xattr_search *xs,
2006                                  struct ocfs2_xattr_set_ctxt *ctxt)
2007 {
2008         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2009         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2010         int ret;
2011
2012         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2013                 return -ENOSPC;
2014
2015         down_write(&oi->ip_alloc_sem);
2016         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2017                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2018                         ret = -ENOSPC;
2019                         goto out;
2020                 }
2021         }
2022
2023         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2024                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2025 out:
2026         up_write(&oi->ip_alloc_sem);
2027
2028         return ret;
2029 }
2030
2031 /*
2032  * ocfs2_xattr_block_find()
2033  *
2034  * Find extended attribute in external block and
2035  * fill search info into struct ocfs2_xattr_search.
2036  */
2037 static int ocfs2_xattr_block_find(struct inode *inode,
2038                                   int name_index,
2039                                   const char *name,
2040                                   struct ocfs2_xattr_search *xs)
2041 {
2042         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2043         struct buffer_head *blk_bh = NULL;
2044         struct ocfs2_xattr_block *xb;
2045         int ret = 0;
2046
2047         if (!di->i_xattr_loc)
2048                 return ret;
2049
2050         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2051                                      &blk_bh);
2052         if (ret < 0) {
2053                 mlog_errno(ret);
2054                 return ret;
2055         }
2056
2057         xs->xattr_bh = blk_bh;
2058         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2059
2060         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2061                 xs->header = &xb->xb_attrs.xb_header;
2062                 xs->base = (void *)xs->header;
2063                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2064                 xs->here = xs->header->xh_entries;
2065
2066                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2067         } else
2068                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2069                                                    name_index,
2070                                                    name, xs);
2071
2072         if (ret && ret != -ENODATA) {
2073                 xs->xattr_bh = NULL;
2074                 goto cleanup;
2075         }
2076         xs->not_found = ret;
2077         return 0;
2078 cleanup:
2079         brelse(blk_bh);
2080
2081         return ret;
2082 }
2083
2084 /*
2085  * ocfs2_xattr_block_set()
2086  *
2087  * Set, replace or remove an extended attribute into external block.
2088  *
2089  */
2090 static int ocfs2_xattr_block_set(struct inode *inode,
2091                                  struct ocfs2_xattr_info *xi,
2092                                  struct ocfs2_xattr_search *xs,
2093                                  struct ocfs2_xattr_set_ctxt *ctxt)
2094 {
2095         struct buffer_head *new_bh = NULL;
2096         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2097         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
2098         handle_t *handle = ctxt->handle;
2099         struct ocfs2_xattr_block *xblk = NULL;
2100         u16 suballoc_bit_start;
2101         u32 num_got;
2102         u64 first_blkno;
2103         int ret;
2104
2105         if (!xs->xattr_bh) {
2106                 ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
2107                                               OCFS2_JOURNAL_ACCESS_CREATE);
2108                 if (ret < 0) {
2109                         mlog_errno(ret);
2110                         goto end;
2111                 }
2112
2113                 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2114                                            &suballoc_bit_start, &num_got,
2115                                            &first_blkno);
2116                 if (ret < 0) {
2117                         mlog_errno(ret);
2118                         goto end;
2119                 }
2120
2121                 new_bh = sb_getblk(inode->i_sb, first_blkno);
2122                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2123
2124                 ret = ocfs2_journal_access_xb(handle, inode, new_bh,
2125                                               OCFS2_JOURNAL_ACCESS_CREATE);
2126                 if (ret < 0) {
2127                         mlog_errno(ret);
2128                         goto end;
2129                 }
2130
2131                 /* Initialize ocfs2_xattr_block */
2132                 xs->xattr_bh = new_bh;
2133                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2134                 memset(xblk, 0, inode->i_sb->s_blocksize);
2135                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2136                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2137                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2138                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2139                 xblk->xb_blkno = cpu_to_le64(first_blkno);
2140
2141                 xs->header = &xblk->xb_attrs.xb_header;
2142                 xs->base = (void *)xs->header;
2143                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2144                 xs->here = xs->header->xh_entries;
2145
2146                 ret = ocfs2_journal_dirty(handle, new_bh);
2147                 if (ret < 0) {
2148                         mlog_errno(ret);
2149                         goto end;
2150                 }
2151                 di->i_xattr_loc = cpu_to_le64(first_blkno);
2152                 ocfs2_journal_dirty(handle, xs->inode_bh);
2153         } else
2154                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2155
2156         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2157                 /* Set extended attribute into external block */
2158                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2159                                             OCFS2_HAS_XATTR_FL);
2160                 if (!ret || ret != -ENOSPC)
2161                         goto end;
2162
2163                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2164                 if (ret)
2165                         goto end;
2166         }
2167
2168         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2169
2170 end:
2171
2172         return ret;
2173 }
2174
2175 /* Check whether the new xattr can be inserted into the inode. */
2176 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2177                                        struct ocfs2_xattr_info *xi,
2178                                        struct ocfs2_xattr_search *xs)
2179 {
2180         u64 value_size;
2181         struct ocfs2_xattr_entry *last;
2182         int free, i;
2183         size_t min_offs = xs->end - xs->base;
2184
2185         if (!xs->header)
2186                 return 0;
2187
2188         last = xs->header->xh_entries;
2189
2190         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2191                 size_t offs = le16_to_cpu(last->xe_name_offset);
2192                 if (offs < min_offs)
2193                         min_offs = offs;
2194                 last += 1;
2195         }
2196
2197         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
2198         if (free < 0)
2199                 return 0;
2200
2201         BUG_ON(!xs->not_found);
2202
2203         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2204                 value_size = OCFS2_XATTR_ROOT_SIZE;
2205         else
2206                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2207
2208         if (free >= sizeof(struct ocfs2_xattr_entry) +
2209                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2210                 return 1;
2211
2212         return 0;
2213 }
2214
2215 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2216                                      struct ocfs2_dinode *di,
2217                                      struct ocfs2_xattr_info *xi,
2218                                      struct ocfs2_xattr_search *xis,
2219                                      struct ocfs2_xattr_search *xbs,
2220                                      int *clusters_need,
2221                                      int *meta_need,
2222                                      int *credits_need)
2223 {
2224         int ret = 0, old_in_xb = 0;
2225         int clusters_add = 0, meta_add = 0, credits = 0;
2226         struct buffer_head *bh = NULL;
2227         struct ocfs2_xattr_block *xb = NULL;
2228         struct ocfs2_xattr_entry *xe = NULL;
2229         struct ocfs2_xattr_value_root *xv = NULL;
2230         char *base = NULL;
2231         int name_offset, name_len = 0;
2232         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2233                                                     xi->value_len);
2234         u64 value_size;
2235
2236         /*
2237          * Calculate the clusters we need to write.
2238          * No matter whether we replace an old one or add a new one,
2239          * we need this for writing.
2240          */
2241         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2242                 credits += new_clusters *
2243                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
2244
2245         if (xis->not_found && xbs->not_found) {
2246                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2247
2248                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2249                         clusters_add += new_clusters;
2250                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2251                                                         &def_xv.xv.xr_list,
2252                                                         new_clusters);
2253                 }
2254
2255                 goto meta_guess;
2256         }
2257
2258         if (!xis->not_found) {
2259                 xe = xis->here;
2260                 name_offset = le16_to_cpu(xe->xe_name_offset);
2261                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2262                 base = xis->base;
2263                 credits += OCFS2_INODE_UPDATE_CREDITS;
2264         } else {
2265                 int i, block_off = 0;
2266                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2267                 xe = xbs->here;
2268                 name_offset = le16_to_cpu(xe->xe_name_offset);
2269                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2270                 i = xbs->here - xbs->header->xh_entries;
2271                 old_in_xb = 1;
2272
2273                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2274                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2275                                                         bucket_xh(xbs->bucket),
2276                                                         i, &block_off,
2277                                                         &name_offset);
2278                         base = bucket_block(xbs->bucket, block_off);
2279                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2280                 } else {
2281                         base = xbs->base;
2282                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2283                 }
2284         }
2285
2286         /*
2287          * delete a xattr doesn't need metadata and cluster allocation.
2288          * so just calculate the credits and return.
2289          *
2290          * The credits for removing the value tree will be extended
2291          * by ocfs2_remove_extent itself.
2292          */
2293         if (!xi->value) {
2294                 if (!ocfs2_xattr_is_local(xe))
2295                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2296
2297                 goto out;
2298         }
2299
2300         /* do cluster allocation guess first. */
2301         value_size = le64_to_cpu(xe->xe_value_size);
2302
2303         if (old_in_xb) {
2304                 /*
2305                  * In xattr set, we always try to set the xe in inode first,
2306                  * so if it can be inserted into inode successfully, the old
2307                  * one will be removed from the xattr block, and this xattr
2308                  * will be inserted into inode as a new xattr in inode.
2309                  */
2310                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2311                         clusters_add += new_clusters;
2312                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
2313                                     OCFS2_INODE_UPDATE_CREDITS;
2314                         if (!ocfs2_xattr_is_local(xe))
2315                                 credits += ocfs2_calc_extend_credits(
2316                                                         inode->i_sb,
2317                                                         &def_xv.xv.xr_list,
2318                                                         new_clusters);
2319                         goto out;
2320                 }
2321         }
2322
2323         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2324                 /* the new values will be stored outside. */
2325                 u32 old_clusters = 0;
2326
2327                 if (!ocfs2_xattr_is_local(xe)) {
2328                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2329                                                                  value_size);
2330                         xv = (struct ocfs2_xattr_value_root *)
2331                              (base + name_offset + name_len);
2332                         value_size = OCFS2_XATTR_ROOT_SIZE;
2333                 } else
2334                         xv = &def_xv.xv;
2335
2336                 if (old_clusters >= new_clusters) {
2337                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2338                         goto out;
2339                 } else {
2340                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2341                         clusters_add += new_clusters - old_clusters;
2342                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2343                                                              &xv->xr_list,
2344                                                              new_clusters -
2345                                                              old_clusters);
2346                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2347                                 goto out;
2348                 }
2349         } else {
2350                 /*
2351                  * Now the new value will be stored inside. So if the new
2352                  * value is smaller than the size of value root or the old
2353                  * value, we don't need any allocation, otherwise we have
2354                  * to guess metadata allocation.
2355                  */
2356                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2357                     (!ocfs2_xattr_is_local(xe) &&
2358                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2359                         goto out;
2360         }
2361
2362 meta_guess:
2363         /* calculate metadata allocation. */
2364         if (di->i_xattr_loc) {
2365                 if (!xbs->xattr_bh) {
2366                         ret = ocfs2_read_xattr_block(inode,
2367                                                      le64_to_cpu(di->i_xattr_loc),
2368                                                      &bh);
2369                         if (ret) {
2370                                 mlog_errno(ret);
2371                                 goto out;
2372                         }
2373
2374                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2375                 } else
2376                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2377
2378                 /*
2379                  * If there is already an xattr tree, good, we can calculate
2380                  * like other b-trees. Otherwise we may have the chance of
2381                  * create a tree, the credit calculation is borrowed from
2382                  * ocfs2_calc_extend_credits with root_el = NULL. And the
2383                  * new tree will be cluster based, so no meta is needed.
2384                  */
2385                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2386                         struct ocfs2_extent_list *el =
2387                                  &xb->xb_attrs.xb_root.xt_list;
2388                         meta_add += ocfs2_extend_meta_needed(el);
2389                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2390                                                              el, 1);
2391                 } else
2392                         credits += OCFS2_SUBALLOC_ALLOC + 1;
2393
2394                 /*
2395                  * This cluster will be used either for new bucket or for
2396                  * new xattr block.
2397                  * If the cluster size is the same as the bucket size, one
2398                  * more is needed since we may need to extend the bucket
2399                  * also.
2400                  */
2401                 clusters_add += 1;
2402                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2403                 if (OCFS2_XATTR_BUCKET_SIZE ==
2404                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2405                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2406                         clusters_add += 1;
2407                 }
2408         } else {
2409                 meta_add += 1;
2410                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2411         }
2412 out:
2413         if (clusters_need)
2414                 *clusters_need = clusters_add;
2415         if (meta_need)
2416                 *meta_need = meta_add;
2417         if (credits_need)
2418                 *credits_need = credits;
2419         brelse(bh);
2420         return ret;
2421 }
2422
2423 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2424                                      struct ocfs2_dinode *di,
2425                                      struct ocfs2_xattr_info *xi,
2426                                      struct ocfs2_xattr_search *xis,
2427                                      struct ocfs2_xattr_search *xbs,
2428                                      struct ocfs2_xattr_set_ctxt *ctxt,
2429                                      int *credits)
2430 {
2431         int clusters_add, meta_add, ret;
2432         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2433
2434         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2435
2436         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2437
2438         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2439                                         &clusters_add, &meta_add, credits);
2440         if (ret) {
2441                 mlog_errno(ret);
2442                 return ret;
2443         }
2444
2445         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2446              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2447
2448         if (meta_add) {
2449                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2450                                                         &ctxt->meta_ac);
2451                 if (ret) {
2452                         mlog_errno(ret);
2453                         goto out;
2454                 }
2455         }
2456
2457         if (clusters_add) {
2458                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2459                 if (ret)
2460                         mlog_errno(ret);
2461         }
2462 out:
2463         if (ret) {
2464                 if (ctxt->meta_ac) {
2465                         ocfs2_free_alloc_context(ctxt->meta_ac);
2466                         ctxt->meta_ac = NULL;
2467                 }
2468
2469                 /*
2470                  * We cannot have an error and a non null ctxt->data_ac.
2471                  */
2472         }
2473
2474         return ret;
2475 }
2476
2477 static int __ocfs2_xattr_set_handle(struct inode *inode,
2478                                     struct ocfs2_dinode *di,
2479                                     struct ocfs2_xattr_info *xi,
2480                                     struct ocfs2_xattr_search *xis,
2481                                     struct ocfs2_xattr_search *xbs,
2482                                     struct ocfs2_xattr_set_ctxt *ctxt)
2483 {
2484         int ret = 0, credits, old_found;
2485
2486         if (!xi->value) {
2487                 /* Remove existing extended attribute */
2488                 if (!xis->not_found)
2489                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2490                 else if (!xbs->not_found)
2491                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2492         } else {
2493                 /* We always try to set extended attribute into inode first*/
2494                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2495                 if (!ret && !xbs->not_found) {
2496                         /*
2497                          * If succeed and that extended attribute existing in
2498                          * external block, then we will remove it.
2499                          */
2500                         xi->value = NULL;
2501                         xi->value_len = 0;
2502
2503                         old_found = xis->not_found;
2504                         xis->not_found = -ENODATA;
2505                         ret = ocfs2_calc_xattr_set_need(inode,
2506                                                         di,
2507                                                         xi,
2508                                                         xis,
2509                                                         xbs,
2510                                                         NULL,
2511                                                         NULL,
2512                                                         &credits);
2513                         xis->not_found = old_found;
2514                         if (ret) {
2515                                 mlog_errno(ret);
2516                                 goto out;
2517                         }
2518
2519                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2520                                         ctxt->handle->h_buffer_credits);
2521                         if (ret) {
2522                                 mlog_errno(ret);
2523                                 goto out;
2524                         }
2525                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2526                 } else if (ret == -ENOSPC) {
2527                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2528                                 ret = ocfs2_xattr_block_find(inode,
2529                                                              xi->name_index,
2530                                                              xi->name, xbs);
2531                                 if (ret)
2532                                         goto out;
2533
2534                                 old_found = xis->not_found;
2535                                 xis->not_found = -ENODATA;
2536                                 ret = ocfs2_calc_xattr_set_need(inode,
2537                                                                 di,
2538                                                                 xi,
2539                                                                 xis,
2540                                                                 xbs,
2541                                                                 NULL,
2542                                                                 NULL,
2543                                                                 &credits);
2544                                 xis->not_found = old_found;
2545                                 if (ret) {
2546                                         mlog_errno(ret);
2547                                         goto out;
2548                                 }
2549
2550                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2551                                         ctxt->handle->h_buffer_credits);
2552                                 if (ret) {
2553                                         mlog_errno(ret);
2554                                         goto out;
2555                                 }
2556                         }
2557                         /*
2558                          * If no space in inode, we will set extended attribute
2559                          * into external block.
2560                          */
2561                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2562                         if (ret)
2563                                 goto out;
2564                         if (!xis->not_found) {
2565                                 /*
2566                                  * If succeed and that extended attribute
2567                                  * existing in inode, we will remove it.
2568                                  */
2569                                 xi->value = NULL;
2570                                 xi->value_len = 0;
2571                                 xbs->not_found = -ENODATA;
2572                                 ret = ocfs2_calc_xattr_set_need(inode,
2573                                                                 di,
2574                                                                 xi,
2575                                                                 xis,
2576                                                                 xbs,
2577                                                                 NULL,
2578                                                                 NULL,
2579                                                                 &credits);
2580                                 if (ret) {
2581                                         mlog_errno(ret);
2582                                         goto out;
2583                                 }
2584
2585                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2586                                                 ctxt->handle->h_buffer_credits);
2587                                 if (ret) {
2588                                         mlog_errno(ret);
2589                                         goto out;
2590                                 }
2591                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2592                                                             xis, ctxt);
2593                         }
2594                 }
2595         }
2596
2597         if (!ret) {
2598                 /* Update inode ctime. */
2599                 ret = ocfs2_journal_access_di(ctxt->handle, inode,
2600                                               xis->inode_bh,
2601                                               OCFS2_JOURNAL_ACCESS_WRITE);
2602                 if (ret) {
2603                         mlog_errno(ret);
2604                         goto out;
2605                 }
2606
2607                 inode->i_ctime = CURRENT_TIME;
2608                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2609                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2610                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2611         }
2612 out:
2613         return ret;
2614 }
2615
2616 /*
2617  * This function only called duing creating inode
2618  * for init security/acl xattrs of the new inode.
2619  * All transanction credits have been reserved in mknod.
2620  */
2621 int ocfs2_xattr_set_handle(handle_t *handle,
2622                            struct inode *inode,
2623                            struct buffer_head *di_bh,
2624                            int name_index,
2625                            const char *name,
2626                            const void *value,
2627                            size_t value_len,
2628                            int flags,
2629                            struct ocfs2_alloc_context *meta_ac,
2630                            struct ocfs2_alloc_context *data_ac)
2631 {
2632         struct ocfs2_dinode *di;
2633         int ret;
2634
2635         struct ocfs2_xattr_info xi = {
2636                 .name_index = name_index,
2637                 .name = name,
2638                 .value = value,
2639                 .value_len = value_len,
2640         };
2641
2642         struct ocfs2_xattr_search xis = {
2643                 .not_found = -ENODATA,
2644         };
2645
2646         struct ocfs2_xattr_search xbs = {
2647                 .not_found = -ENODATA,
2648         };
2649
2650         struct ocfs2_xattr_set_ctxt ctxt = {
2651                 .handle = handle,
2652                 .meta_ac = meta_ac,
2653                 .data_ac = data_ac,
2654         };
2655
2656         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2657                 return -EOPNOTSUPP;
2658
2659         /*
2660          * In extreme situation, may need xattr bucket when
2661          * block size is too small. And we have already reserved
2662          * the credits for bucket in mknod.
2663          */
2664         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
2665                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2666                 if (!xbs.bucket) {
2667                         mlog_errno(-ENOMEM);
2668                         return -ENOMEM;
2669                 }
2670         }
2671
2672         xis.inode_bh = xbs.inode_bh = di_bh;
2673         di = (struct ocfs2_dinode *)di_bh->b_data;
2674
2675         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2676
2677         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2678         if (ret)
2679                 goto cleanup;
2680         if (xis.not_found) {
2681                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2682                 if (ret)
2683                         goto cleanup;
2684         }
2685
2686         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2687
2688 cleanup:
2689         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2690         brelse(xbs.xattr_bh);
2691         ocfs2_xattr_bucket_free(xbs.bucket);
2692
2693         return ret;
2694 }
2695
2696 /*
2697  * ocfs2_xattr_set()
2698  *
2699  * Set, replace or remove an extended attribute for this inode.
2700  * value is NULL to remove an existing extended attribute, else either
2701  * create or replace an extended attribute.
2702  */
2703 int ocfs2_xattr_set(struct inode *inode,
2704                     int name_index,
2705                     const char *name,
2706                     const void *value,
2707                     size_t value_len,
2708                     int flags)
2709 {
2710         struct buffer_head *di_bh = NULL;
2711         struct ocfs2_dinode *di;
2712         int ret, credits;
2713         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2714         struct inode *tl_inode = osb->osb_tl_inode;
2715         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2716
2717         struct ocfs2_xattr_info xi = {
2718                 .name_index = name_index,
2719                 .name = name,
2720                 .value = value,
2721                 .value_len = value_len,
2722         };
2723
2724         struct ocfs2_xattr_search xis = {
2725                 .not_found = -ENODATA,
2726         };
2727
2728         struct ocfs2_xattr_search xbs = {
2729                 .not_found = -ENODATA,
2730         };
2731
2732         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2733                 return -EOPNOTSUPP;
2734
2735         /*
2736          * Only xbs will be used on indexed trees.  xis doesn't need a
2737          * bucket.
2738          */
2739         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2740         if (!xbs.bucket) {
2741                 mlog_errno(-ENOMEM);
2742                 return -ENOMEM;
2743         }
2744
2745         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2746         if (ret < 0) {
2747                 mlog_errno(ret);
2748                 goto cleanup_nolock;
2749         }
2750         xis.inode_bh = xbs.inode_bh = di_bh;
2751         di = (struct ocfs2_dinode *)di_bh->b_data;
2752
2753         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2754         /*
2755          * Scan inode and external block to find the same name
2756          * extended attribute and collect search infomation.
2757          */
2758         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2759         if (ret)
2760                 goto cleanup;
2761         if (xis.not_found) {
2762                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2763                 if (ret)
2764                         goto cleanup;
2765         }
2766
2767         if (xis.not_found && xbs.not_found) {
2768                 ret = -ENODATA;
2769                 if (flags & XATTR_REPLACE)
2770                         goto cleanup;
2771                 ret = 0;
2772                 if (!value)
2773                         goto cleanup;
2774         } else {
2775                 ret = -EEXIST;
2776                 if (flags & XATTR_CREATE)
2777                         goto cleanup;
2778         }
2779
2780
2781         mutex_lock(&tl_inode->i_mutex);
2782
2783         if (ocfs2_truncate_log_needs_flush(osb)) {
2784                 ret = __ocfs2_flush_truncate_log(osb);
2785                 if (ret < 0) {
2786                         mutex_unlock(&tl_inode->i_mutex);
2787                         mlog_errno(ret);
2788                         goto cleanup;
2789                 }
2790         }
2791         mutex_unlock(&tl_inode->i_mutex);
2792
2793         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2794                                         &xbs, &ctxt, &credits);
2795         if (ret) {
2796                 mlog_errno(ret);
2797                 goto cleanup;
2798         }
2799
2800         /* we need to update inode's ctime field, so add credit for it. */
2801         credits += OCFS2_INODE_UPDATE_CREDITS;
2802         ctxt.handle = ocfs2_start_trans(osb, credits);
2803         if (IS_ERR(ctxt.handle)) {
2804                 ret = PTR_ERR(ctxt.handle);
2805                 mlog_errno(ret);
2806                 goto cleanup;
2807         }
2808
2809         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2810
2811         ocfs2_commit_trans(osb, ctxt.handle);
2812
2813         if (ctxt.data_ac)
2814                 ocfs2_free_alloc_context(ctxt.data_ac);
2815         if (ctxt.meta_ac)
2816                 ocfs2_free_alloc_context(ctxt.meta_ac);
2817         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2818                 ocfs2_schedule_truncate_log_flush(osb, 1);
2819         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2820 cleanup:
2821         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2822         ocfs2_inode_unlock(inode, 1);
2823 cleanup_nolock:
2824         brelse(di_bh);
2825         brelse(xbs.xattr_bh);
2826         ocfs2_xattr_bucket_free(xbs.bucket);
2827
2828         return ret;
2829 }
2830
2831 /*
2832  * Find the xattr extent rec which may contains name_hash.
2833  * e_cpos will be the first name hash of the xattr rec.
2834  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2835  */
2836 static int ocfs2_xattr_get_rec(struct inode *inode,
2837                                u32 name_hash,
2838                                u64 *p_blkno,
2839                                u32 *e_cpos,
2840                                u32 *num_clusters,
2841                                struct ocfs2_extent_list *el)
2842 {
2843         int ret = 0, i;
2844         struct buffer_head *eb_bh = NULL;
2845         struct ocfs2_extent_block *eb;
2846         struct ocfs2_extent_rec *rec = NULL;
2847         u64 e_blkno = 0;
2848
2849         if (el->l_tree_depth) {
2850                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2851                 if (ret) {
2852                         mlog_errno(ret);
2853                         goto out;
2854                 }
2855
2856                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2857                 el = &eb->h_list;
2858
2859                 if (el->l_tree_depth) {
2860                         ocfs2_error(inode->i_sb,
2861                                     "Inode %lu has non zero tree depth in "
2862                                     "xattr tree block %llu\n", inode->i_ino,
2863                                     (unsigned long long)eb_bh->b_blocknr);
2864                         ret = -EROFS;
2865                         goto out;
2866                 }
2867         }
2868
2869         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2870                 rec = &el->l_recs[i];
2871
2872                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2873                         e_blkno = le64_to_cpu(rec->e_blkno);
2874                         break;
2875                 }
2876         }
2877
2878         if (!e_blkno) {
2879                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2880                             "record (%u, %u, 0) in xattr", inode->i_ino,
2881                             le32_to_cpu(rec->e_cpos),
2882                             ocfs2_rec_clusters(el, rec));
2883                 ret = -EROFS;
2884                 goto out;
2885         }
2886
2887         *p_blkno = le64_to_cpu(rec->e_blkno);
2888         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2889         if (e_cpos)
2890                 *e_cpos = le32_to_cpu(rec->e_cpos);
2891 out:
2892         brelse(eb_bh);
2893         return ret;
2894 }
2895
2896 typedef int (xattr_bucket_func)(struct inode *inode,
2897                                 struct ocfs2_xattr_bucket *bucket,
2898                                 void *para);
2899
2900 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2901                                    struct ocfs2_xattr_bucket *bucket,
2902                                    int name_index,
2903                                    const char *name,
2904                                    u32 name_hash,
2905                                    u16 *xe_index,
2906                                    int *found)
2907 {
2908         int i, ret = 0, cmp = 1, block_off, new_offset;
2909         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2910         size_t name_len = strlen(name);
2911         struct ocfs2_xattr_entry *xe = NULL;
2912         char *xe_name;
2913
2914         /*
2915          * We don't use binary search in the bucket because there
2916          * may be multiple entries with the same name hash.
2917          */
2918         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2919                 xe = &xh->xh_entries[i];
2920
2921                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2922                         continue;
2923                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2924                         break;
2925
2926                 cmp = name_index - ocfs2_xattr_get_type(xe);
2927                 if (!cmp)
2928                         cmp = name_len - xe->xe_name_len;
2929                 if (cmp)
2930                         continue;
2931
2932                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2933                                                         xh,
2934                                                         i,
2935                                                         &block_off,
2936                                                         &new_offset);
2937                 if (ret) {
2938                         mlog_errno(ret);
2939                         break;
2940                 }
2941
2942
2943                 xe_name = bucket_block(bucket, block_off) + new_offset;
2944                 if (!memcmp(name, xe_name, name_len)) {
2945                         *xe_index = i;
2946                         *found = 1;
2947                         ret = 0;
2948                         break;
2949                 }
2950         }
2951
2952         return ret;
2953 }
2954
2955 /*
2956  * Find the specified xattr entry in a series of buckets.
2957  * This series start from p_blkno and last for num_clusters.
2958  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2959  * the num of the valid buckets.
2960  *
2961  * Return the buffer_head this xattr should reside in. And if the xattr's
2962  * hash is in the gap of 2 buckets, return the lower bucket.
2963  */
2964 static int ocfs2_xattr_bucket_find(struct inode *inode,
2965                                    int name_index,
2966                                    const char *name,
2967                                    u32 name_hash,
2968                                    u64 p_blkno,
2969                                    u32 first_hash,
2970                                    u32 num_clusters,
2971                                    struct ocfs2_xattr_search *xs)
2972 {
2973         int ret, found = 0;
2974         struct ocfs2_xattr_header *xh = NULL;
2975         struct ocfs2_xattr_entry *xe = NULL;
2976         u16 index = 0;
2977         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2978         int low_bucket = 0, bucket, high_bucket;
2979         struct ocfs2_xattr_bucket *search;
2980         u32 last_hash;
2981         u64 blkno, lower_blkno = 0;
2982
2983         search = ocfs2_xattr_bucket_new(inode);
2984         if (!search) {
2985                 ret = -ENOMEM;
2986                 mlog_errno(ret);
2987                 goto out;
2988         }
2989
2990         ret = ocfs2_read_xattr_bucket(search, p_blkno);
2991         if (ret) {
2992                 mlog_errno(ret);
2993                 goto out;
2994         }
2995
2996         xh = bucket_xh(search);
2997         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2998         while (low_bucket <= high_bucket) {
2999                 ocfs2_xattr_bucket_relse(search);
3000
3001                 bucket = (low_bucket + high_bucket) / 2;
3002                 blkno = p_blkno + bucket * blk_per_bucket;
3003                 ret = ocfs2_read_xattr_bucket(search, blkno);
3004                 if (ret) {
3005                         mlog_errno(ret);
3006                         goto out;
3007                 }
3008
3009                 xh = bucket_xh(search);
3010                 xe = &xh->xh_entries[0];
3011                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3012                         high_bucket = bucket - 1;
3013                         continue;
3014                 }
3015
3016                 /*
3017                  * Check whether the hash of the last entry in our
3018                  * bucket is larger than the search one. for an empty
3019                  * bucket, the last one is also the first one.
3020                  */
3021                 if (xh->xh_count)
3022                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3023
3024                 last_hash = le32_to_cpu(xe->xe_name_hash);
3025
3026                 /* record lower_blkno which may be the insert place. */
3027                 lower_blkno = blkno;
3028
3029                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3030                         low_bucket = bucket + 1;
3031                         continue;
3032                 }
3033
3034                 /* the searched xattr should reside in this bucket if exists. */
3035                 ret = ocfs2_find_xe_in_bucket(inode, search,
3036                                               name_index, name, name_hash,
3037                                               &index, &found);
3038                 if (ret) {
3039                         mlog_errno(ret);
3040                         goto out;
3041                 }
3042                 break;
3043         }
3044
3045         /*
3046          * Record the bucket we have found.
3047          * When the xattr's hash value is in the gap of 2 buckets, we will
3048          * always set it to the previous bucket.
3049          */
3050         if (!lower_blkno)
3051                 lower_blkno = p_blkno;
3052
3053         /* This should be in cache - we just read it during the search */
3054         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3055         if (ret) {
3056                 mlog_errno(ret);
3057                 goto out;
3058         }
3059
3060         xs->header = bucket_xh(xs->bucket);
3061         xs->base = bucket_block(xs->bucket, 0);
3062         xs->end = xs->base + inode->i_sb->s_blocksize;
3063
3064         if (found) {
3065                 xs->here = &xs->header->xh_entries[index];
3066                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3067                      (unsigned long long)bucket_blkno(xs->bucket), index);
3068         } else
3069                 ret = -ENODATA;
3070
3071 out:
3072         ocfs2_xattr_bucket_free(search);
3073         return ret;
3074 }
3075
3076 static int ocfs2_xattr_index_block_find(struct inode *inode,
3077                                         struct buffer_head *root_bh,
3078                                         int name_index,
3079                                         const char *name,
3080                                         struct ocfs2_xattr_search *xs)
3081 {
3082         int ret;
3083         struct ocfs2_xattr_block *xb =
3084                         (struct ocfs2_xattr_block *)root_bh->b_data;
3085         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3086         struct ocfs2_extent_list *el = &xb_root->xt_list;
3087         u64 p_blkno = 0;
3088         u32 first_hash, num_clusters = 0;
3089         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3090
3091         if (le16_to_cpu(el->l_next_free_rec) == 0)
3092                 return -ENODATA;
3093
3094         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3095              name, name_hash, name_index);
3096
3097         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3098                                   &num_clusters, el);
3099         if (ret) {
3100                 mlog_errno(ret);
3101                 goto out;
3102         }
3103
3104         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3105
3106         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3107              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3108              first_hash);
3109
3110         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3111                                       p_blkno, first_hash, num_clusters, xs);
3112
3113 out:
3114         return ret;
3115 }
3116
3117 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3118                                        u64 blkno,
3119                                        u32 clusters,
3120                                        xattr_bucket_func *func,
3121                                        void *para)
3122 {
3123         int i, ret = 0;
3124         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3125         u32 num_buckets = clusters * bpc;
3126         struct ocfs2_xattr_bucket *bucket;
3127
3128         bucket = ocfs2_xattr_bucket_new(inode);
3129         if (!bucket) {
3130                 mlog_errno(-ENOMEM);
3131                 return -ENOMEM;
3132         }
3133
3134         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3135              clusters, (unsigned long long)blkno);
3136
3137         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3138                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3139                 if (ret) {
3140                         mlog_errno(ret);
3141                         break;
3142                 }
3143
3144                 /*
3145                  * The real bucket num in this series of blocks is stored
3146                  * in the 1st bucket.
3147                  */
3148                 if (i == 0)
3149                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3150
3151                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3152                      (unsigned long long)blkno,
3153                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3154                 if (func) {
3155                         ret = func(inode, bucket, para);
3156                         if (ret)
3157                                 mlog_errno(ret);
3158                         /* Fall through to bucket_relse() */
3159                 }
3160
3161                 ocfs2_xattr_bucket_relse(bucket);
3162                 if (ret)
3163                         break;
3164         }
3165
3166         ocfs2_xattr_bucket_free(bucket);
3167         return ret;
3168 }
3169
3170 struct ocfs2_xattr_tree_list {
3171         char *buffer;
3172         size_t buffer_size;
3173         size_t result;
3174 };
3175
3176 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3177                                              struct ocfs2_xattr_header *xh,
3178                                              int index,
3179                                              int *block_off,
3180                                              int *new_offset)
3181 {
3182         u16 name_offset;
3183
3184         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3185                 return -EINVAL;
3186
3187         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3188
3189         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3190         *new_offset = name_offset % inode->i_sb->s_blocksize;
3191
3192         return 0;
3193 }
3194
3195 static int ocfs2_list_xattr_bucket(struct inode *inode,
3196                                    struct ocfs2_xattr_bucket *bucket,
3197                                    void *para)
3198 {
3199         int ret = 0, type;
3200         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3201         int i, block_off, new_offset;
3202         const char *prefix, *name;
3203
3204         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3205                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3206                 type = ocfs2_xattr_get_type(entry);
3207                 prefix = ocfs2_xattr_prefix(type);
3208
3209                 if (prefix) {
3210                         ret = ocfs2_xattr_bucket_get_name_value(inode,
3211                                                                 bucket_xh(bucket),
3212                                                                 i,
3213                                                                 &block_off,
3214                                                                 &new_offset);
3215                         if (ret)
3216                                 break;
3217
3218                         name = (const char *)bucket_block(bucket, block_off) +
3219                                 new_offset;
3220                         ret = ocfs2_xattr_list_entry(xl->buffer,
3221                                                      xl->buffer_size,
3222                                                      &xl->result,
3223                                                      prefix, name,
3224                                                      entry->xe_name_len);
3225                         if (ret)
3226                                 break;
3227                 }
3228         }
3229
3230         return ret;
3231 }
3232
3233 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3234                                              struct ocfs2_xattr_tree_root *xt,
3235                                              char *buffer,
3236                                              size_t buffer_size)
3237 {
3238         struct ocfs2_extent_list *el = &xt->xt_list;
3239         int ret = 0;
3240         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3241         u64 p_blkno = 0;
3242         struct ocfs2_xattr_tree_list xl = {
3243                 .buffer = buffer,
3244                 .buffer_size = buffer_size,
3245                 .result = 0,
3246         };
3247
3248         if (le16_to_cpu(el->l_next_free_rec) == 0)
3249                 return 0;
3250
3251         while (name_hash > 0) {
3252                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3253                                           &e_cpos, &num_clusters, el);
3254                 if (ret) {
3255                         mlog_errno(ret);
3256                         goto out;
3257                 }
3258
3259                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3260                                                   ocfs2_list_xattr_bucket,
3261                                                   &xl);
3262                 if (ret) {
3263                         mlog_errno(ret);
3264                         goto out;
3265                 }
3266
3267                 if (e_cpos == 0)
3268                         break;
3269
3270                 name_hash = e_cpos - 1;
3271         }
3272
3273         ret = xl.result;
3274 out:
3275         return ret;
3276 }
3277
3278 static int cmp_xe(const void *a, const void *b)
3279 {
3280         const struct ocfs2_xattr_entry *l = a, *r = b;
3281         u32 l_hash = le32_to_cpu(l->xe_name_hash);
3282         u32 r_hash = le32_to_cpu(r->xe_name_hash);
3283
3284         if (l_hash > r_hash)
3285                 return 1;
3286         if (l_hash < r_hash)
3287                 return -1;
3288         return 0;
3289 }
3290
3291 static void swap_xe(void *a, void *b, int size)
3292 {
3293         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3294
3295         tmp = *l;
3296         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3297         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3298 }
3299
3300 /*
3301  * When the ocfs2_xattr_block is filled up, new bucket will be created
3302  * and all the xattr entries will be moved to the new bucket.
3303  * The header goes at the start of the bucket, and the names+values are
3304  * filled from the end.  This is why *target starts as the last buffer.
3305  * Note: we need to sort the entries since they are not saved in order
3306  * in the ocfs2_xattr_block.
3307  */
3308 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3309                                            struct buffer_head *xb_bh,
3310                                            struct ocfs2_xattr_bucket *bucket)
3311 {
3312         int i, blocksize = inode->i_sb->s_blocksize;
3313         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3314         u16 offset, size, off_change;
3315         struct ocfs2_xattr_entry *xe;
3316         struct ocfs2_xattr_block *xb =
3317                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
3318         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3319         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3320         u16 count = le16_to_cpu(xb_xh->xh_count);
3321         char *src = xb_bh->b_data;
3322         char *target = bucket_block(bucket, blks - 1);
3323
3324         mlog(0, "cp xattr from block %llu to bucket %llu\n",
3325              (unsigned long long)xb_bh->b_blocknr,
3326              (unsigned long long)bucket_blkno(bucket));
3327
3328         for (i = 0; i < blks; i++)
3329                 memset(bucket_block(bucket, i), 0, blocksize);
3330
3331         /*
3332          * Since the xe_name_offset is based on ocfs2_xattr_header,
3333          * there is a offset change corresponding to the change of
3334          * ocfs2_xattr_header's position.
3335          */
3336         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3337         xe = &xb_xh->xh_entries[count - 1];
3338         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3339         size = blocksize - offset;
3340
3341         /* copy all the names and values. */
3342         memcpy(target + offset, src + offset, size);
3343
3344         /* Init new header now. */
3345         xh->xh_count = xb_xh->xh_count;
3346         xh->xh_num_buckets = cpu_to_le16(1);
3347         xh->xh_name_value_len = cpu_to_le16(size);
3348         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3349
3350         /* copy all the entries. */
3351         target = bucket_block(bucket, 0);
3352         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3353         size = count * sizeof(struct ocfs2_xattr_entry);
3354         memcpy(target + offset, (char *)xb_xh + offset, size);
3355
3356         /* Change the xe offset for all the xe because of the move. */
3357         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3358                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3359         for (i = 0; i < count; i++)
3360                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3361
3362         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3363              offset, size, off_change);
3364
3365         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3366              cmp_xe, swap_xe);
3367 }
3368
3369 /*
3370  * After we move xattr from block to index btree, we have to
3371  * update ocfs2_xattr_search to the new xe and base.
3372  *
3373  * When the entry is in xattr block, xattr_bh indicates the storage place.
3374  * While if the entry is in index b-tree, "bucket" indicates the
3375  * real place of the xattr.
3376  */
3377 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3378                                             struct ocfs2_xattr_search *xs,
3379                                             struct buffer_head *old_bh)
3380 {
3381         char *buf = old_bh->b_data;
3382         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3383         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3384         int i;
3385
3386         xs->header = bucket_xh(xs->bucket);
3387         xs->base = bucket_block(xs->bucket, 0);
3388         xs->end = xs->base + inode->i_sb->s_blocksize;
3389
3390         if (xs->not_found)
3391                 return;
3392
3393         i = xs->here - old_xh->xh_entries;
3394         xs->here = &xs->header->xh_entries[i];
3395 }
3396
3397 static int ocfs2_xattr_create_index_block(struct inode *inode,
3398                                           struct ocfs2_xattr_search *xs,
3399                                           struct ocfs2_xattr_set_ctxt *ctxt)
3400 {
3401         int ret;
3402         u32 bit_off, len;
3403         u64 blkno;
3404         handle_t *handle = ctxt->handle;
3405         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3406         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3407         struct buffer_head *xb_bh = xs->xattr_bh;
3408         struct ocfs2_xattr_block *xb =
3409                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3410         struct ocfs2_xattr_tree_root *xr;
3411         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3412
3413         mlog(0, "create xattr index block for %llu\n",
3414              (unsigned long long)xb_bh->b_blocknr);
3415
3416         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3417         BUG_ON(!xs->bucket);
3418
3419         /*
3420          * XXX:
3421          * We can use this lock for now, and maybe move to a dedicated mutex
3422          * if performance becomes a problem later.
3423          */
3424         down_write(&oi->ip_alloc_sem);
3425
3426         ret = ocfs2_journal_access_xb(handle, inode, xb_bh,
3427                                       OCFS2_JOURNAL_ACCESS_WRITE);
3428         if (ret) {
3429                 mlog_errno(ret);
3430                 goto out;
3431         }
3432
3433         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3434                                      1, 1, &bit_off, &len);
3435         if (ret) {
3436                 mlog_errno(ret);
3437                 goto out;
3438         }
3439
3440         /*
3441          * The bucket may spread in many blocks, and
3442          * we will only touch the 1st block and the last block
3443          * in the whole bucket(one for entry and one for data).
3444          */
3445         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3446
3447         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3448              (unsigned long long)blkno);
3449
3450         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3451         if (ret) {
3452                 mlog_errno(ret);
3453                 goto out;
3454         }
3455
3456         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3457                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3458         if (ret) {
3459                 mlog_errno(ret);
3460                 goto out;
3461         }
3462
3463         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3464         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3465
3466         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3467
3468         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3469         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3470                offsetof(struct ocfs2_xattr_block, xb_attrs));
3471
3472         xr = &xb->xb_attrs.xb_root;
3473         xr->xt_clusters = cpu_to_le32(1);
3474         xr->xt_last_eb_blk = 0;
3475         xr->xt_list.l_tree_depth = 0;
3476         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3477         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3478
3479         xr->xt_list.l_recs[0].e_cpos = 0;
3480         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3481         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3482
3483         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3484
3485         ocfs2_journal_dirty(handle, xb_bh);
3486
3487 out:
3488         up_write(&oi->ip_alloc_sem);
3489
3490         return ret;
3491 }
3492
3493 static int cmp_xe_offset(const void *a, const void *b)
3494 {
3495         const struct ocfs2_xattr_entry *l = a, *r = b;
3496         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3497         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3498
3499         if (l_name_offset < r_name_offset)
3500                 return 1;
3501         if (l_name_offset > r_name_offset)
3502                 return -1;
3503         return 0;
3504 }
3505
3506 /*
3507  * defrag a xattr bucket if we find that the bucket has some
3508  * holes beteen name/value pairs.
3509  * We will move all the name/value pairs to the end of the bucket
3510  * so that we can spare some space for insertion.
3511  */
3512 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3513                                      handle_t *handle,
3514                                      struct ocfs2_xattr_bucket *bucket)
3515 {
3516         int ret, i;
3517         size_t end, offset, len, value_len;
3518         struct ocfs2_xattr_header *xh;
3519         char *entries, *buf, *bucket_buf = NULL;
3520         u64 blkno = bucket_blkno(bucket);
3521         u16 xh_free_start;
3522         size_t blocksize = inode->i_sb->s_blocksize;
3523         struct ocfs2_xattr_entry *xe;
3524
3525         /*
3526          * In order to make the operation more efficient and generic,
3527          * we copy all the blocks into a contiguous memory and do the
3528          * defragment there, so if anything is error, we will not touch
3529          * the real block.
3530          */
3531         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3532         if (!bucket_buf) {
3533                 ret = -EIO;
3534                 goto out;
3535         }
3536
3537         buf = bucket_buf;
3538         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3539                 memcpy(buf, bucket_block(bucket, i), blocksize);
3540
3541         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3542                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3543         if (ret < 0) {
3544                 mlog_errno(ret);
3545                 goto out;
3546         }
3547
3548         xh = (struct ocfs2_xattr_header *)bucket_buf;
3549         entries = (char *)xh->xh_entries;
3550         xh_free_start = le16_to_cpu(xh->xh_free_start);
3551
3552         mlog(0, "adjust xattr bucket in %llu, count = %u, "
3553              "xh_free_start = %u, xh_name_value_len = %u.\n",
3554              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3555              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3556
3557         /*
3558          * sort all the entries by their offset.
3559          * the largest will be the first, so that we can
3560          * move them to the end one by one.
3561          */
3562         sort(entries, le16_to_cpu(xh->xh_count),
3563              sizeof(struct ocfs2_xattr_entry),
3564              cmp_xe_offset, swap_xe);
3565
3566         /* Move all name/values to the end of the bucket. */
3567         xe = xh->xh_entries;
3568         end = OCFS2_XATTR_BUCKET_SIZE;
3569         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3570                 offset = le16_to_cpu(xe->xe_name_offset);
3571                 if (ocfs2_xattr_is_local(xe))
3572                         value_len = OCFS2_XATTR_SIZE(
3573                                         le64_to_cpu(xe->xe_value_size));
3574                 else
3575                         value_len = OCFS2_XATTR_ROOT_SIZE;
3576                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3577
3578                 /*
3579                  * We must make sure that the name/value pair
3580                  * exist in the same block. So adjust end to
3581                  * the previous block end if needed.
3582                  */
3583                 if (((end - len) / blocksize !=
3584                         (end - 1) / blocksize))
3585                         end = end - end % blocksize;
3586
3587                 if (end > offset + len) {
3588                         memmove(bucket_buf + end - len,
3589                                 bucket_buf + offset, len);
3590                         xe->xe_name_offset = cpu_to_le16(end - len);
3591                 }
3592
3593                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3594                                 "bucket %llu\n", (unsigned long long)blkno);
3595
3596                 end -= len;
3597         }
3598
3599         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3600                         "bucket %llu\n", (unsigned long long)blkno);
3601
3602         if (xh_free_start == end)
3603                 goto out;
3604
3605         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3606         xh->xh_free_start = cpu_to_le16(end);
3607
3608         /* sort the entries by their name_hash. */
3609         sort(entries, le16_to_cpu(xh->xh_count),
3610              sizeof(struct ocfs2_xattr_entry),
3611              cmp_xe, swap_xe);
3612
3613         buf = bucket_buf;
3614         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3615                 memcpy(bucket_block(bucket, i), buf, blocksize);
3616         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3617
3618 out:
3619         kfree(bucket_buf);
3620         return ret;
3621 }
3622
3623 /*
3624  * prev_blkno points to the start of an existing extent.  new_blkno
3625  * points to a newly allocated extent.  Because we know each of our
3626  * clusters contains more than bucket, we can easily split one cluster
3627  * at a bucket boundary.  So we take the last cluster of the existing
3628  * extent and split it down the middle.  We move the last half of the
3629  * buckets in the last cluster of the existing extent over to the new
3630  * extent.
3631  *
3632  * first_bh is the buffer at prev_blkno so we can update the existing
3633  * extent's bucket count.  header_bh is the bucket were we were hoping
3634  * to insert our xattr.  If the bucket move places the target in the new
3635  * extent, we'll update first_bh and header_bh after modifying the old
3636  * extent.
3637  *
3638  * first_hash will be set as the 1st xe's name_hash in the new extent.
3639  */
3640 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3641                                                handle_t *handle,
3642                                                struct ocfs2_xattr_bucket *first,
3643                                                struct ocfs2_xattr_bucket *target,
3644                                                u64 new_blkno,
3645                                                u32 num_clusters,
3646                                                u32 *first_hash)
3647 {
3648         int ret;
3649         struct super_block *sb = inode->i_sb;
3650         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3651         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3652         int to_move = num_buckets / 2;
3653         u64 src_blkno;
3654         u64 last_cluster_blkno = bucket_blkno(first) +
3655                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3656
3657         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3658         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3659
3660         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3661              (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3662
3663         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3664                                      last_cluster_blkno, new_blkno,
3665                                      to_move, first_hash);
3666         if (ret) {
3667                 mlog_errno(ret);
3668                 goto out;
3669         }
3670
3671         /* This is the first bucket that got moved */
3672         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3673
3674         /*
3675          * If the target bucket was part of the moved buckets, we need to
3676          * update first and target.
3677          */
3678         if (bucket_blkno(target) >= src_blkno) {
3679                 /* Find the block for the new target bucket */
3680                 src_blkno = new_blkno +
3681                         (bucket_blkno(target) - src_blkno);
3682
3683                 ocfs2_xattr_bucket_relse(first);
3684                 ocfs2_xattr_bucket_relse(target);
3685
3686                 /*
3687                  * These shouldn't fail - the buffers are in the
3688                  * journal from ocfs2_cp_xattr_bucket().
3689                  */
3690                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3691                 if (ret) {
3692                         mlog_errno(ret);
3693                         goto out;
3694                 }
3695                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3696                 if (ret)
3697                         mlog_errno(ret);
3698
3699         }
3700
3701 out:
3702         return ret;
3703 }
3704
3705 /*
3706  * Find the suitable pos when we divide a bucket into 2.
3707  * We have to make sure the xattrs with the same hash value exist
3708  * in the same bucket.
3709  *
3710  * If this ocfs2_xattr_header covers more than one hash value, find a
3711  * place where the hash value changes.  Try to find the most even split.
3712  * The most common case is that all entries have different hash values,
3713  * and the first check we make will find a place to split.
3714  */
3715 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3716 {
3717         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3718         int count = le16_to_cpu(xh->xh_count);
3719         int delta, middle = count / 2;
3720
3721         /*
3722          * We start at the middle.  Each step gets farther away in both
3723          * directions.  We therefore hit the change in hash value
3724          * nearest to the middle.  Note that this loop does not execute for
3725          * count < 2.
3726          */
3727         for (delta = 0; delta < middle; delta++) {
3728                 /* Let's check delta earlier than middle */
3729                 if (cmp_xe(&entries[middle - delta - 1],
3730                            &entries[middle - delta]))
3731                         return middle - delta;
3732
3733                 /* For even counts, don't walk off the end */
3734                 if ((middle + delta + 1) == count)
3735                         continue;
3736
3737                 /* Now try delta past middle */
3738                 if (cmp_xe(&entries[middle + delta],
3739                            &entries[middle + delta + 1]))
3740                         return middle + delta + 1;
3741         }
3742
3743         /* Every entry had the same hash */
3744         return count;
3745 }
3746
3747 /*
3748  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3749  * first_hash will record the 1st hash of the new bucket.
3750  *
3751  * Normally half of the xattrs will be moved.  But we have to make
3752  * sure that the xattrs with the same hash value are stored in the
3753  * same bucket. If all the xattrs in this bucket have the same hash
3754  * value, the new bucket will be initialized as an empty one and the
3755  * first_hash will be initialized as (hash_value+1).
3756  */
3757 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3758                                     handle_t *handle,
3759                                     u64 blk,
3760                                     u64 new_blk,
3761                                     u32 *first_hash,
3762                                     int new_bucket_head)
3763 {
3764         int ret, i;
3765         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3766         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3767         struct ocfs2_xattr_header *xh;
3768         struct ocfs2_xattr_entry *xe;
3769         int blocksize = inode->i_sb->s_blocksize;
3770
3771         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3772              (unsigned long long)blk, (unsigned long long)new_blk);
3773
3774         s_bucket = ocfs2_xattr_bucket_new(inode);
3775         t_bucket = ocfs2_xattr_bucket_new(inode);
3776         if (!s_bucket || !t_bucket) {
3777                 ret = -ENOMEM;
3778                 mlog_errno(ret);
3779                 goto out;
3780         }
3781
3782         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3783         if (ret) {
3784                 mlog_errno(ret);
3785                 goto out;
3786         }
3787
3788         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3789                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3790         if (ret) {
3791                 mlog_errno(ret);
3792                 goto out;
3793         }
3794
3795         /*
3796          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3797          * there's no need to read it.
3798          */
3799         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3800         if (ret) {
3801                 mlog_errno(ret);
3802                 goto out;
3803         }
3804
3805         /*
3806          * Hey, if we're overwriting t_bucket, what difference does
3807          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
3808          * same part of ocfs2_cp_xattr_bucket().
3809          */
3810         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3811                                                 new_bucket_head ?
3812                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3813                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3814         if (ret) {
3815                 mlog_errno(ret);
3816                 goto out;
3817         }
3818
3819         xh = bucket_xh(s_bucket);
3820         count = le16_to_cpu(xh->xh_count);
3821         start = ocfs2_xattr_find_divide_pos(xh);
3822
3823         if (start == count) {
3824                 xe = &xh->xh_entries[start-1];
3825
3826                 /*
3827                  * initialized a new empty bucket here.
3828                  * The hash value is set as one larger than
3829                  * that of the last entry in the previous bucket.
3830                  */
3831                 for (i = 0; i < t_bucket->bu_blocks; i++)
3832                         memset(bucket_block(t_bucket, i), 0, blocksize);
3833
3834                 xh = bucket_xh(t_bucket);
3835                 xh->xh_free_start = cpu_to_le16(blocksize);
3836                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3837                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3838
3839                 goto set_num_buckets;
3840         }
3841
3842         /* copy the whole bucket to the new first. */
3843         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3844
3845         /* update the new bucket. */
3846         xh = bucket_xh(t_bucket);
3847
3848         /*
3849          * Calculate the total name/value len and xh_free_start for
3850          * the old bucket first.
3851          */
3852         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3853         name_value_len = 0;
3854         for (i = 0; i < start; i++) {
3855                 xe = &xh->xh_entries[i];
3856                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3857                 if (ocfs2_xattr_is_local(xe))
3858                         xe_len +=
3859                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3860                 else
3861                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3862                 name_value_len += xe_len;
3863                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3864                         name_offset = le16_to_cpu(xe->xe_name_offset);
3865         }
3866
3867         /*
3868          * Now begin the modification to the new bucket.
3869          *
3870          * In the new bucket, We just move the xattr entry to the beginning
3871          * and don't touch the name/value. So there will be some holes in the
3872          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3873          * called.
3874          */
3875         xe = &xh->xh_entries[start];
3876         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3877         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3878              (int)((char *)xe - (char *)xh),
3879              (int)((char *)xh->xh_entries - (char *)xh));
3880         memmove((char *)xh->xh_entries, (char *)xe, len);
3881         xe = &xh->xh_entries[count - start];
3882         len = sizeof(struct ocfs2_xattr_entry) * start;
3883         memset((char *)xe, 0, len);
3884
3885         le16_add_cpu(&xh->xh_count, -start);
3886         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3887
3888         /* Calculate xh_free_start for the new bucket. */
3889         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3890         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3891                 xe = &xh->xh_entries[i];
3892                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3893                 if (ocfs2_xattr_is_local(xe))
3894                         xe_len +=
3895                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3896                 else
3897                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3898                 if (le16_to_cpu(xe->xe_name_offset) <
3899                     le16_to_cpu(xh->xh_free_start))
3900                         xh->xh_free_start = xe->xe_name_offset;
3901         }
3902
3903 set_num_buckets:
3904         /* set xh->xh_num_buckets for the new xh. */
3905         if (new_bucket_head)
3906                 xh->xh_num_buckets = cpu_to_le16(1);
3907         else
3908                 xh->xh_num_buckets = 0;
3909
3910         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3911
3912         /* store the first_hash of the new bucket. */
3913         if (first_hash)
3914                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3915
3916         /*
3917          * Now only update the 1st block of the old bucket.  If we
3918          * just added a new empty bucket, there is no need to modify
3919          * it.
3920          */
3921         if (start == count)
3922                 goto out;
3923
3924         xh = bucket_xh(s_bucket);
3925         memset(&xh->xh_entries[start], 0,
3926                sizeof(struct ocfs2_xattr_entry) * (count - start));
3927         xh->xh_count = cpu_to_le16(start);
3928         xh->xh_free_start = cpu_to_le16(name_offset);
3929         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3930
3931         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3932
3933 out:
3934         ocfs2_xattr_bucket_free(s_bucket);
3935         ocfs2_xattr_bucket_free(t_bucket);
3936
3937         return ret;
3938 }
3939
3940 /*
3941  * Copy xattr from one bucket to another bucket.
3942  *
3943  * The caller must make sure that the journal transaction
3944  * has enough space for journaling.
3945  */
3946 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3947                                  handle_t *handle,
3948                                  u64 s_blkno,
3949                                  u64 t_blkno,
3950                                  int t_is_new)
3951 {
3952         int ret;
3953         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3954
3955         BUG_ON(s_blkno == t_blkno);
3956
3957         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3958              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3959              t_is_new);
3960
3961         s_bucket = ocfs2_xattr_bucket_new(inode);
3962         t_bucket = ocfs2_xattr_bucket_new(inode);
3963         if (!s_bucket || !t_bucket) {
3964                 ret = -ENOMEM;
3965                 mlog_errno(ret);
3966                 goto out;
3967         }
3968
3969         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3970         if (ret)
3971                 goto out;
3972
3973         /*
3974          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3975          * there's no need to read it.
3976          */
3977         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3978         if (ret)
3979                 goto out;
3980
3981         /*
3982          * Hey, if we're overwriting t_bucket, what difference does
3983          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
3984          * cluster to fill, we came here from
3985          * ocfs2_mv_xattr_buckets(), and it is really new -
3986          * ACCESS_CREATE is required.  But we also might have moved data
3987          * out of t_bucket before extending back into it.
3988          * ocfs2_add_new_xattr_bucket() can do this - its call to
3989          * ocfs2_add_new_xattr_cluster() may have created a new extent
3990          * and copied out the end of the old extent.  Then it re-extends
3991          * the old extent back to create space for new xattrs.  That's
3992          * how we get here, and the bucket isn't really new.
3993          */
3994         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3995                                                 t_is_new ?
3996                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3997                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3998         if (ret)
3999                 goto out;
4000
4001         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4002         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4003
4004 out:
4005         ocfs2_xattr_bucket_free(t_bucket);
4006         ocfs2_xattr_bucket_free(s_bucket);
4007
4008         return ret;
4009 }
4010
4011 /*
4012  * src_blk points to the start of an existing extent.  last_blk points to
4013  * last cluster in that extent.  to_blk points to a newly allocated
4014  * extent.  We copy the buckets from the cluster at last_blk to the new
4015  * extent.  If start_bucket is non-zero, we skip that many buckets before
4016  * we start copying.  The new extent's xh_num_buckets gets set to the
4017  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4018  * by the same amount.
4019  */
4020 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4021                                   u64 src_blk, u64 last_blk, u64 to_blk,
4022                                   unsigned int start_bucket,
4023                                   u32 *first_hash)
4024 {
4025         int i, ret, credits;
4026         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4027         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4028         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4029         struct ocfs2_xattr_bucket *old_first, *new_first;
4030
4031         mlog(0, "mv xattrs from cluster %llu to %llu\n",
4032              (unsigned long long)last_blk, (unsigned long long)to_blk);
4033
4034         BUG_ON(start_bucket >= num_buckets);
4035         if (start_bucket) {
4036                 num_buckets -= start_bucket;
4037                 last_blk += (start_bucket * blks_per_bucket);
4038         }
4039
4040         /* The first bucket of the original extent */
4041         old_first = ocfs2_xattr_bucket_new(inode);
4042         /* The first bucket of the new extent */
4043         new_first = ocfs2_xattr_bucket_new(inode);
4044         if (!old_first || !new_first) {
4045                 ret = -ENOMEM;
4046                 mlog_errno(ret);
4047                 goto out;
4048         }
4049
4050         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4051         if (ret) {
4052                 mlog_errno(ret);
4053                 goto out;
4054         }
4055
4056         /*
4057          * We need to update the first bucket of the old extent and all
4058          * the buckets going to the new extent.
4059          */
4060         credits = ((num_buckets + 1) * blks_per_bucket) +
4061                 handle->h_buffer_credits;
4062         ret = ocfs2_extend_trans(handle, credits);
4063         if (ret) {
4064                 mlog_errno(ret);
4065                 goto out;
4066         }
4067
4068         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4069                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4070         if (ret) {
4071                 mlog_errno(ret);
4072                 goto out;
4073         }
4074
4075         for (i = 0; i < num_buckets; i++) {
4076                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4077                                             last_blk + (i * blks_per_bucket),
4078                                             to_blk + (i * blks_per_bucket),
4079                                             1);
4080                 if (ret) {
4081                         mlog_errno(ret);
4082                         goto out;
4083                 }
4084         }
4085
4086         /*
4087          * Get the new bucket ready before we dirty anything
4088          * (This actually shouldn't fail, because we already dirtied
4089          * it once in ocfs2_cp_xattr_bucket()).
4090          */
4091         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4092         if (ret) {
4093                 mlog_errno(ret);
4094                 goto out;
4095         }
4096         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4097                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4098         if (ret) {
4099                 mlog_errno(ret);
4100                 goto out;
4101         }
4102
4103         /* Now update the headers */
4104         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4105         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4106
4107         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4108         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4109
4110         if (first_hash)
4111                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4112
4113 out:
4114         ocfs2_xattr_bucket_free(new_first);
4115         ocfs2_xattr_bucket_free(old_first);
4116         return ret;
4117 }
4118
4119 /*
4120  * Move some xattrs in this cluster to the new cluster.
4121  * This function should only be called when bucket size == cluster size.
4122  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4123  */
4124 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4125                                       handle_t *handle,
4126                                       u64 prev_blk,
4127                                       u64 new_blk,
4128                                       u32 *first_hash)
4129 {
4130         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4131         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4132
4133         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4134
4135         ret = ocfs2_extend_trans(handle, credits);
4136         if (ret) {
4137                 mlog_errno(ret);
4138                 return ret;
4139         }
4140
4141         /* Move half of the xattr in start_blk to the next bucket. */
4142         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4143                                           new_blk, first_hash, 1);
4144 }
4145
4146 /*
4147  * Move some xattrs from the old cluster to the new one since they are not
4148  * contiguous in ocfs2 xattr tree.
4149  *
4150  * new_blk starts a new separate cluster, and we will move some xattrs from
4151  * prev_blk to it. v_start will be set as the first name hash value in this
4152  * new cluster so that it can be used as e_cpos during tree insertion and
4153  * don't collide with our original b-tree operations. first_bh and header_bh
4154  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4155  * to extend the insert bucket.
4156  *
4157  * The problem is how much xattr should we move to the new one and when should
4158  * we update first_bh and header_bh?
4159  * 1. If cluster size > bucket size, that means the previous cluster has more
4160  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4161  *    update the first_bh and header_bh if the insert bucket has been moved
4162  *    to the new cluster.
4163  * 2. If cluster_size == bucket_size:
4164  *    a) If the previous extent rec has more than one cluster and the insert
4165  *       place isn't in the last cluster, copy the entire last cluster to the
4166  *       new one. This time, we don't need to upate the first_bh and header_bh
4167  *       since they will not be moved into the new cluster.
4168  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4169  *       the new one. And we set the extend flag to zero if the insert place is
4170  *       moved into the new allocated cluster since no extend is needed.
4171  */
4172 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4173                                             handle_t *handle,
4174                                             struct ocfs2_xattr_bucket *first,
4175                                             struct ocfs2_xattr_bucket *target,
4176                                             u64 new_blk,
4177                                             u32 prev_clusters,
4178                                             u32 *v_start,
4179                                             int *extend)
4180 {
4181         int ret;
4182
4183         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4184              (unsigned long long)bucket_blkno(first), prev_clusters,
4185              (unsigned long long)new_blk);
4186
4187         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4188                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4189                                                           handle,
4190                                                           first, target,
4191                                                           new_blk,
4192                                                           prev_clusters,
4193                                                           v_start);
4194                 if (ret)
4195                         mlog_errno(ret);
4196         } else {
4197                 /* The start of the last cluster in the first extent */
4198                 u64 last_blk = bucket_blkno(first) +
4199                         ((prev_clusters - 1) *
4200                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
4201
4202                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4203                         ret = ocfs2_mv_xattr_buckets(inode, handle,
4204                                                      bucket_blkno(first),
4205                                                      last_blk, new_blk, 0,
4206                                                      v_start);
4207                         if (ret)
4208                                 mlog_errno(ret);
4209                 } else {
4210                         ret = ocfs2_divide_xattr_cluster(inode, handle,
4211                                                          last_blk, new_blk,
4212                                                          v_start);
4213                         if (ret)
4214                                 mlog_errno(ret);
4215
4216                         if ((bucket_blkno(target) == last_blk) && extend)
4217                                 *extend = 0;
4218                 }
4219         }
4220
4221         return ret;
4222 }
4223
4224 /*
4225  * Add a new cluster for xattr storage.
4226  *
4227  * If the new cluster is contiguous with the previous one, it will be
4228  * appended to the same extent record, and num_clusters will be updated.
4229  * If not, we will insert a new extent for it and move some xattrs in
4230  * the last cluster into the new allocated one.
4231  * We also need to limit the maximum size of a btree leaf, otherwise we'll
4232  * lose the benefits of hashing because we'll have to search large leaves.
4233  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4234  * if it's bigger).
4235  *
4236  * first_bh is the first block of the previous extent rec and header_bh
4237  * indicates the bucket we will insert the new xattrs. They will be updated
4238  * when the header_bh is moved into the new cluster.
4239  */
4240 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4241                                        struct buffer_head *root_bh,
4242                                        struct ocfs2_xattr_bucket *first,
4243                                        struct ocfs2_xattr_bucket *target,
4244                                        u32 *num_clusters,
4245                                        u32 prev_cpos,
4246                                        int *extend,
4247                                        struct ocfs2_xattr_set_ctxt *ctxt)
4248 {
4249         int ret;
4250         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4251         u32 prev_clusters = *num_clusters;
4252         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4253         u64 block;
4254         handle_t *handle = ctxt->handle;
4255         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4256         struct ocfs2_extent_tree et;
4257
4258         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4259              "previous xattr blkno = %llu\n",
4260              (unsigned long long)OCFS2_I(inode)->ip_blkno,
4261              prev_cpos, (unsigned long long)bucket_blkno(first));
4262
4263         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4264
4265         ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4266                                       OCFS2_JOURNAL_ACCESS_WRITE);
4267         if (ret < 0) {
4268                 mlog_errno(ret);
4269                 goto leave;
4270         }
4271
4272         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4273                                      clusters_to_add, &bit_off, &num_bits);
4274         if (ret < 0) {
4275                 if (ret != -ENOSPC)
4276                         mlog_errno(ret);
4277                 goto leave;
4278         }
4279
4280         BUG_ON(num_bits > clusters_to_add);
4281
4282         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4283         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4284              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4285
4286         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4287             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4288              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4289                 /*
4290                  * If this cluster is contiguous with the old one and
4291                  * adding this new cluster, we don't surpass the limit of
4292                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4293                  * initialized and used like other buckets in the previous
4294                  * cluster.
4295                  * So add it as a contiguous one. The caller will handle
4296                  * its init process.
4297                  */
4298                 v_start = prev_cpos + prev_clusters;
4299                 *num_clusters = prev_clusters + num_bits;
4300                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4301                      num_bits);
4302         } else {
4303                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4304                                                        handle,
4305                                                        first,
4306                                                        target,
4307                                                        block,
4308                                                        prev_clusters,
4309                                                        &v_start,
4310                                                        extend);
4311                 if (ret) {
4312                         mlog_errno(ret);
4313                         goto leave;
4314                 }
4315         }
4316
4317         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4318              num_bits, (unsigned long long)block, v_start);
4319         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
4320                                   num_bits, 0, ctxt->meta_ac);
4321         if (ret < 0) {
4322                 mlog_errno(ret);
4323                 goto leave;
4324         }
4325
4326         ret = ocfs2_journal_dirty(handle, root_bh);
4327         if (ret < 0)
4328                 mlog_errno(ret);
4329
4330 leave:
4331         return ret;
4332 }
4333
4334 /*
4335  * We are given an extent.  'first' is the bucket at the very front of
4336  * the extent.  The extent has space for an additional bucket past
4337  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
4338  * of the target bucket.  We wish to shift every bucket past the target
4339  * down one, filling in that additional space.  When we get back to the
4340  * target, we split the target between itself and the now-empty bucket
4341  * at target+1 (aka, target_blkno + blks_per_bucket).
4342  */
4343 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4344                                      handle_t *handle,
4345                                      struct ocfs2_xattr_bucket *first,
4346                                      u64 target_blk,
4347                                      u32 num_clusters)
4348 {
4349         int ret, credits;
4350         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4351         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4352         u64 end_blk;
4353         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4354
4355         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4356              "from %llu, len = %u\n", (unsigned long long)target_blk,
4357              (unsigned long long)bucket_blkno(first), num_clusters);
4358
4359         /* The extent must have room for an additional bucket */
4360         BUG_ON(new_bucket >=
4361                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4362
4363         /* end_blk points to the last existing bucket */
4364         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4365
4366         /*
4367          * end_blk is the start of the last existing bucket.
4368          * Thus, (end_blk - target_blk) covers the target bucket and
4369          * every bucket after it up to, but not including, the last
4370          * existing bucket.  Then we add the last existing bucket, the
4371          * new bucket, and the first bucket (3 * blk_per_bucket).
4372          */
4373         credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4374                   handle->h_buffer_credits;
4375         ret = ocfs2_extend_trans(handle, credits);
4376         if (ret) {
4377                 mlog_errno(ret);
4378                 goto out;
4379         }
4380
4381         ret = ocfs2_xattr_bucket_journal_access(handle, first,
4382                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4383         if (ret) {
4384                 mlog_errno(ret);
4385                 goto out;
4386         }
4387
4388         while (end_blk != target_blk) {
4389                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4390                                             end_blk + blk_per_bucket, 0);
4391                 if (ret)
4392                         goto out;
4393                 end_blk -= blk_per_bucket;
4394         }
4395
4396         /* Move half of the xattr in target_blkno to the next bucket. */
4397         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4398                                         target_blk + blk_per_bucket, NULL, 0);
4399
4400         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4401         ocfs2_xattr_bucket_journal_dirty(handle, first);
4402
4403 out:
4404         return ret;
4405 }
4406
4407 /*
4408  * Add new xattr bucket in an extent record and adjust the buckets
4409  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
4410  * bucket we want to insert into.
4411  *
4412  * In the easy case, we will move all the buckets after target down by
4413  * one. Half of target's xattrs will be moved to the next bucket.
4414  *
4415  * If current cluster is full, we'll allocate a new one.  This may not
4416  * be contiguous.  The underlying calls will make sure that there is
4417  * space for the insert, shifting buckets around if necessary.
4418  * 'target' may be moved by those calls.
4419  */
4420 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4421                                       struct buffer_head *xb_bh,
4422                                       struct ocfs2_xattr_bucket *target,
4423                                       struct ocfs2_xattr_set_ctxt *ctxt)
4424 {
4425         struct ocfs2_xattr_block *xb =
4426                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4427         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4428         struct ocfs2_extent_list *el = &xb_root->xt_list;
4429         u32 name_hash =
4430                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4431         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4432         int ret, num_buckets, extend = 1;
4433         u64 p_blkno;
4434         u32 e_cpos, num_clusters;
4435         /* The bucket at the front of the extent */
4436         struct ocfs2_xattr_bucket *first;
4437
4438         mlog(0, "Add new xattr bucket starting from %llu\n",
4439              (unsigned long long)bucket_blkno(target));
4440
4441         /* The first bucket of the original extent */
4442         first = ocfs2_xattr_bucket_new(inode);
4443         if (!first) {
4444                 ret = -ENOMEM;
4445                 mlog_errno(ret);
4446                 goto out;
4447         }
4448
4449         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4450                                   &num_clusters, el);
4451         if (ret) {
4452                 mlog_errno(ret);
4453                 goto out;
4454         }
4455
4456         ret = ocfs2_read_xattr_bucket(first, p_blkno);
4457         if (ret) {
4458                 mlog_errno(ret);
4459                 goto out;
4460         }
4461
4462         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4463         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4464                 /*
4465                  * This can move first+target if the target bucket moves
4466                  * to the new extent.
4467                  */
4468                 ret = ocfs2_add_new_xattr_cluster(inode,
4469                                                   xb_bh,
4470                                                   first,
4471                                                   target,
4472                                                   &num_clusters,
4473                                                   e_cpos,
4474                                                   &extend,
4475                                                   ctxt);
4476                 if (ret) {
4477                         mlog_errno(ret);
4478                         goto out;
4479                 }
4480         }
4481
4482         if (extend) {
4483                 ret = ocfs2_extend_xattr_bucket(inode,
4484                                                 ctxt->handle,
4485                                                 first,
4486                                                 bucket_blkno(target),
4487                                                 num_clusters);
4488                 if (ret)
4489                         mlog_errno(ret);
4490         }
4491
4492 out:
4493         ocfs2_xattr_bucket_free(first);
4494
4495         return ret;
4496 }
4497
4498 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4499                                         struct ocfs2_xattr_bucket *bucket,
4500                                         int offs)
4501 {
4502         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4503
4504         offs = offs % inode->i_sb->s_blocksize;
4505         return bucket_block(bucket, block_off) + offs;
4506 }
4507
4508 /*
4509  * Handle the normal xattr set, including replace, delete and new.
4510  *
4511  * Note: "local" indicates the real data's locality. So we can't
4512  * just its bucket locality by its length.
4513  */
4514 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4515                                          struct ocfs2_xattr_info *xi,
4516                                          struct ocfs2_xattr_search *xs,
4517                                          u32 name_hash,
4518                                          int local)
4519 {
4520         struct ocfs2_xattr_entry *last, *xe;
4521         int name_len = strlen(xi->name);
4522         struct ocfs2_xattr_header *xh = xs->header;
4523         u16 count = le16_to_cpu(xh->xh_count), start;
4524         size_t blocksize = inode->i_sb->s_blocksize;
4525         char *val;
4526         size_t offs, size, new_size;
4527
4528         last = &xh->xh_entries[count];
4529         if (!xs->not_found) {
4530                 xe = xs->here;
4531                 offs = le16_to_cpu(xe->xe_name_offset);
4532                 if (ocfs2_xattr_is_local(xe))
4533                         size = OCFS2_XATTR_SIZE(name_len) +
4534                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4535                 else
4536                         size = OCFS2_XATTR_SIZE(name_len) +
4537                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4538
4539                 /*
4540                  * If the new value will be stored outside, xi->value has been
4541                  * initalized as an empty ocfs2_xattr_value_root, and the same
4542                  * goes with xi->value_len, so we can set new_size safely here.
4543                  * See ocfs2_xattr_set_in_bucket.
4544                  */
4545                 new_size = OCFS2_XATTR_SIZE(name_len) +
4546                            OCFS2_XATTR_SIZE(xi->value_len);
4547
4548                 le16_add_cpu(&xh->xh_name_value_len, -size);
4549                 if (xi->value) {
4550                         if (new_size > size)
4551                                 goto set_new_name_value;
4552
4553                         /* Now replace the old value with new one. */
4554                         if (local)
4555                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4556                         else
4557                                 xe->xe_value_size = 0;
4558
4559                         val = ocfs2_xattr_bucket_get_val(inode,
4560                                                          xs->bucket, offs);
4561                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4562                                size - OCFS2_XATTR_SIZE(name_len));
4563                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4564                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4565                                        xi->value, xi->value_len);
4566
4567                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4568                         ocfs2_xattr_set_local(xe, local);
4569                         return;
4570                 } else {
4571                         /*
4572                          * Remove the old entry if there is more than one.
4573                          * We don't remove the last entry so that we can
4574                          * use it to indicate the hash value of the empty
4575                          * bucket.
4576                          */
4577                         last -= 1;
4578                         le16_add_cpu(&xh->xh_count, -1);
4579                         if (xh->xh_count) {
4580                                 memmove(xe, xe + 1,
4581                                         (void *)last - (void *)xe);
4582                                 memset(last, 0,
4583                                        sizeof(struct ocfs2_xattr_entry));
4584                         } else
4585                                 xh->xh_free_start =
4586                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4587
4588                         return;
4589                 }
4590         } else {
4591                 /* find a new entry for insert. */
4592                 int low = 0, high = count - 1, tmp;
4593                 struct ocfs2_xattr_entry *tmp_xe;
4594
4595                 while (low <= high && count) {
4596                         tmp = (low + high) / 2;
4597                         tmp_xe = &xh->xh_entries[tmp];
4598
4599                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4600                                 low = tmp + 1;
4601                         else if (name_hash <
4602                                  le32_to_cpu(tmp_xe->xe_name_hash))
4603                                 high = tmp - 1;
4604                         else {
4605                                 low = tmp;
4606                                 break;
4607                         }
4608                 }
4609
4610                 xe = &xh->xh_entries[low];
4611                 if (low != count)
4612                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4613
4614                 le16_add_cpu(&xh->xh_count, 1);
4615                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4616                 xe->xe_name_hash = cpu_to_le32(name_hash);
4617                 xe->xe_name_len = name_len;
4618                 ocfs2_xattr_set_type(xe, xi->name_index);
4619         }
4620
4621 set_new_name_value:
4622         /* Insert the new name+value. */
4623         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4624
4625         /*
4626          * We must make sure that the name/value pair
4627          * exists in the same block.
4628          */
4629         offs = le16_to_cpu(xh->xh_free_start);
4630         start = offs - size;
4631
4632         if (start >> inode->i_sb->s_blocksize_bits !=
4633             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4634                 offs = offs - offs % blocksize;
4635                 xh->xh_free_start = cpu_to_le16(offs);
4636         }
4637
4638         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4639         xe->xe_name_offset = cpu_to_le16(offs - size);
4640
4641         memset(val, 0, size);
4642         memcpy(val, xi->name, name_len);
4643         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4644
4645         xe->xe_value_size = cpu_to_le64(xi->value_len);
4646         ocfs2_xattr_set_local(xe, local);
4647         xs->here = xe;
4648         le16_add_cpu(&xh->xh_free_start, -size);
4649         le16_add_cpu(&xh->xh_name_value_len, size);
4650
4651         return;
4652 }
4653
4654 /*
4655  * Set the xattr entry in the specified bucket.
4656  * The bucket is indicated by xs->bucket and it should have the enough
4657  * space for the xattr insertion.
4658  */
4659 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4660                                            handle_t *handle,
4661                                            struct ocfs2_xattr_info *xi,
4662                                            struct ocfs2_xattr_search *xs,
4663                                            u32 name_hash,
4664                                            int local)
4665 {
4666         int ret;
4667         u64 blkno;
4668
4669         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4670              (unsigned long)xi->value_len, xi->name_index,
4671              (unsigned long long)bucket_blkno(xs->bucket));
4672
4673         if (!xs->bucket->bu_bhs[1]) {
4674                 blkno = bucket_blkno(xs->bucket);
4675                 ocfs2_xattr_bucket_relse(xs->bucket);
4676                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4677                 if (ret) {
4678                         mlog_errno(ret);
4679                         goto out;
4680                 }
4681         }
4682
4683         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4684                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4685         if (ret < 0) {
4686                 mlog_errno(ret);
4687                 goto out;
4688         }
4689
4690         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4691         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4692
4693 out:
4694         return ret;
4695 }
4696
4697 /*
4698  * Truncate the specified xe_off entry in xattr bucket.
4699  * bucket is indicated by header_bh and len is the new length.
4700  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4701  *
4702  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4703  */
4704 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4705                                              struct ocfs2_xattr_bucket *bucket,
4706                                              int xe_off,
4707                                              int len,
4708                                              struct ocfs2_xattr_set_ctxt *ctxt)
4709 {
4710         int ret, offset;
4711         u64 value_blk;
4712         struct ocfs2_xattr_entry *xe;
4713         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4714         size_t blocksize = inode->i_sb->s_blocksize;
4715         struct ocfs2_xattr_value_buf vb = {
4716                 .vb_access = ocfs2_journal_access,
4717         };
4718
4719         xe = &xh->xh_entries[xe_off];
4720
4721         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4722
4723         offset = le16_to_cpu(xe->xe_name_offset) +
4724                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4725
4726         value_blk = offset / blocksize;
4727
4728         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4729         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4730
4731         vb.vb_bh = bucket->bu_bhs[value_blk];
4732         BUG_ON(!vb.vb_bh);
4733
4734         vb.vb_xv = (struct ocfs2_xattr_value_root *)
4735                 (vb.vb_bh->b_data + offset % blocksize);
4736
4737         /*
4738          * From here on out we have to dirty the bucket.  The generic
4739          * value calls only modify one of the bucket's bhs, but we need
4740          * to send the bucket at once.  So if they error, they *could* have
4741          * modified something.  We have to assume they did, and dirty
4742          * the whole bucket.  This leaves us in a consistent state.
4743          */
4744         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4745              xe_off, (unsigned long long)bucket_blkno(bucket), len);
4746         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4747         if (ret) {
4748                 mlog_errno(ret);
4749                 goto out;
4750         }
4751
4752         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4753                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4754         if (ret) {
4755                 mlog_errno(ret);
4756                 goto out;
4757         }
4758
4759         xe->xe_value_size = cpu_to_le64(len);
4760
4761         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4762
4763 out:
4764         return ret;
4765 }
4766
4767 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4768                                         struct ocfs2_xattr_search *xs,
4769                                         int len,
4770                                         struct ocfs2_xattr_set_ctxt *ctxt)
4771 {
4772         int ret, offset;
4773         struct ocfs2_xattr_entry *xe = xs->here;
4774         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4775
4776         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4777
4778         offset = xe - xh->xh_entries;
4779         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4780                                                 offset, len, ctxt);
4781         if (ret)
4782                 mlog_errno(ret);
4783
4784         return ret;
4785 }
4786
4787 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4788                                                 handle_t *handle,
4789                                                 struct ocfs2_xattr_search *xs,
4790                                                 char *val,
4791                                                 int value_len)
4792 {
4793         int offset;
4794         struct ocfs2_xattr_value_root *xv;
4795         struct ocfs2_xattr_entry *xe = xs->here;
4796
4797         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4798
4799         offset = le16_to_cpu(xe->xe_name_offset) +
4800                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4801
4802         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4803
4804         return __ocfs2_xattr_set_value_outside(inode, handle,
4805                                                xv, val, value_len);
4806 }
4807
4808 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4809                                   struct buffer_head *root_bh,
4810                                   u64 blkno,
4811                                   u32 cpos,
4812                                   u32 len)
4813 {
4814         int ret;
4815         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4816         struct inode *tl_inode = osb->osb_tl_inode;
4817         handle_t *handle;
4818         struct ocfs2_xattr_block *xb =
4819                         (struct ocfs2_xattr_block *)root_bh->b_data;
4820         struct ocfs2_alloc_context *meta_ac = NULL;
4821         struct ocfs2_cached_dealloc_ctxt dealloc;
4822         struct ocfs2_extent_tree et;
4823
4824         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4825
4826         ocfs2_init_dealloc_ctxt(&dealloc);
4827
4828         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4829              cpos, len, (unsigned long long)blkno);
4830
4831         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4832
4833         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4834         if (ret) {
4835                 mlog_errno(ret);
4836                 return ret;
4837         }
4838
4839         mutex_lock(&tl_inode->i_mutex);
4840
4841         if (ocfs2_truncate_log_needs_flush(osb)) {
4842                 ret = __ocfs2_flush_truncate_log(osb);
4843                 if (ret < 0) {
4844                         mlog_errno(ret);
4845                         goto out;
4846                 }
4847         }
4848
4849         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4850         if (IS_ERR(handle)) {
4851                 ret = -ENOMEM;
4852                 mlog_errno(ret);
4853                 goto out;
4854         }
4855
4856         ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4857                                       OCFS2_JOURNAL_ACCESS_WRITE);
4858         if (ret) {
4859                 mlog_errno(ret);
4860                 goto out_commit;
4861         }
4862
4863         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4864                                   &dealloc);
4865         if (ret) {
4866                 mlog_errno(ret);
4867                 goto out_commit;
4868         }
4869
4870         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4871
4872         ret = ocfs2_journal_dirty(handle, root_bh);
4873         if (ret) {
4874                 mlog_errno(ret);
4875                 goto out_commit;
4876         }
4877
4878         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4879         if (ret)
4880                 mlog_errno(ret);
4881
4882 out_commit:
4883         ocfs2_commit_trans(osb, handle);
4884 out:
4885         ocfs2_schedule_truncate_log_flush(osb, 1);
4886
4887         mutex_unlock(&tl_inode->i_mutex);
4888
4889         if (meta_ac)
4890                 ocfs2_free_alloc_context(meta_ac);
4891
4892         ocfs2_run_deallocs(osb, &dealloc);
4893
4894         return ret;
4895 }
4896
4897 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4898                                          handle_t *handle,
4899                                          struct ocfs2_xattr_search *xs)
4900 {
4901         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4902         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4903                                                 le16_to_cpu(xh->xh_count) - 1];
4904         int ret = 0;
4905
4906         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4907                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4908         if (ret) {
4909                 mlog_errno(ret);
4910                 return;
4911         }
4912
4913         /* Remove the old entry. */
4914         memmove(xs->here, xs->here + 1,
4915                 (void *)last - (void *)xs->here);
4916         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4917         le16_add_cpu(&xh->xh_count, -1);
4918
4919         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4920 }
4921
4922 /*
4923  * Set the xattr name/value in the bucket specified in xs.
4924  *
4925  * As the new value in xi may be stored in the bucket or in an outside cluster,
4926  * we divide the whole process into 3 steps:
4927  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4928  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4929  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4930  * 4. If the clusters for the new outside value can't be allocated, we need
4931  *    to free the xattr we allocated in set.
4932  */
4933 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4934                                      struct ocfs2_xattr_info *xi,
4935                                      struct ocfs2_xattr_search *xs,
4936                                      struct ocfs2_xattr_set_ctxt *ctxt)
4937 {
4938         int ret, local = 1;
4939         size_t value_len;
4940         char *val = (char *)xi->value;
4941         struct ocfs2_xattr_entry *xe = xs->here;
4942         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4943                                               strlen(xi->name));
4944
4945         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4946                 /*
4947                  * We need to truncate the xattr storage first.
4948                  *
4949                  * If both the old and new value are stored to
4950                  * outside block, we only need to truncate
4951                  * the storage and then set the value outside.
4952                  *
4953                  * If the new value should be stored within block,
4954                  * we should free all the outside block first and
4955                  * the modification to the xattr block will be done
4956                  * by following steps.
4957                  */
4958                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4959                         value_len = xi->value_len;
4960                 else
4961                         value_len = 0;
4962
4963                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4964                                                            value_len,
4965                                                            ctxt);
4966                 if (ret)
4967                         goto out;
4968
4969                 if (value_len)
4970                         goto set_value_outside;
4971         }
4972
4973         value_len = xi->value_len;
4974         /* So we have to handle the inside block change now. */
4975         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4976                 /*
4977                  * If the new value will be stored outside of block,
4978                  * initalize a new empty value root and insert it first.
4979                  */
4980                 local = 0;
4981                 xi->value = &def_xv;
4982                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4983         }
4984
4985         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
4986                                               name_hash, local);
4987         if (ret) {
4988                 mlog_errno(ret);
4989                 goto out;
4990         }
4991
4992         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4993                 goto out;
4994
4995         /* allocate the space now for the outside block storage. */
4996         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4997                                                    value_len, ctxt);
4998         if (ret) {
4999                 mlog_errno(ret);
5000
5001                 if (xs->not_found) {
5002                         /*
5003                          * We can't allocate enough clusters for outside
5004                          * storage and we have allocated xattr already,
5005                          * so need to remove it.
5006                          */
5007                         ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
5008                 }
5009                 goto out;
5010         }
5011
5012 set_value_outside:
5013         ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5014                                                    xs, val, value_len);
5015 out:
5016         return ret;
5017 }
5018
5019 /*
5020  * check whether the xattr bucket is filled up with the same hash value.
5021  * If we want to insert the xattr with the same hash, return -ENOSPC.
5022  * If we want to insert a xattr with different hash value, go ahead
5023  * and ocfs2_divide_xattr_bucket will handle this.
5024  */
5025 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5026                                               struct ocfs2_xattr_bucket *bucket,
5027                                               const char *name)
5028 {
5029         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5030         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5031
5032         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5033                 return 0;
5034
5035         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5036             xh->xh_entries[0].xe_name_hash) {
5037                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5038                      "hash = %u\n",
5039                      (unsigned long long)bucket_blkno(bucket),
5040                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5041                 return -ENOSPC;
5042         }
5043
5044         return 0;
5045 }
5046
5047 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5048                                              struct ocfs2_xattr_info *xi,
5049                                              struct ocfs2_xattr_search *xs,
5050                                              struct ocfs2_xattr_set_ctxt *ctxt)
5051 {
5052         struct ocfs2_xattr_header *xh;
5053         struct ocfs2_xattr_entry *xe;
5054         u16 count, header_size, xh_free_start;
5055         int free, max_free, need, old;
5056         size_t value_size = 0, name_len = strlen(xi->name);
5057         size_t blocksize = inode->i_sb->s_blocksize;
5058         int ret, allocation = 0;
5059
5060         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5061
5062 try_again:
5063         xh = xs->header;
5064         count = le16_to_cpu(xh->xh_count);
5065         xh_free_start = le16_to_cpu(xh->xh_free_start);
5066         header_size = sizeof(struct ocfs2_xattr_header) +
5067                         count * sizeof(struct ocfs2_xattr_entry);
5068         max_free = OCFS2_XATTR_BUCKET_SIZE -
5069                 le16_to_cpu(xh->xh_name_value_len) - header_size;
5070
5071         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5072                         "of %u which exceed block size\n",
5073                         (unsigned long long)bucket_blkno(xs->bucket),
5074                         header_size);
5075
5076         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5077                 value_size = OCFS2_XATTR_ROOT_SIZE;
5078         else if (xi->value)
5079                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5080
5081         if (xs->not_found)
5082                 need = sizeof(struct ocfs2_xattr_entry) +
5083                         OCFS2_XATTR_SIZE(name_len) + value_size;
5084         else {
5085                 need = value_size + OCFS2_XATTR_SIZE(name_len);
5086
5087                 /*
5088                  * We only replace the old value if the new length is smaller
5089                  * than the old one. Otherwise we will allocate new space in the
5090                  * bucket to store it.
5091                  */
5092                 xe = xs->here;
5093                 if (ocfs2_xattr_is_local(xe))
5094                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5095                 else
5096                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5097
5098                 if (old >= value_size)
5099                         need = 0;
5100         }
5101
5102         free = xh_free_start - header_size;
5103         /*
5104          * We need to make sure the new name/value pair
5105          * can exist in the same block.
5106          */
5107         if (xh_free_start % blocksize < need)
5108                 free -= xh_free_start % blocksize;
5109
5110         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5111              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5112              " %u\n", xs->not_found,
5113              (unsigned long long)bucket_blkno(xs->bucket),
5114              free, need, max_free, le16_to_cpu(xh->xh_free_start),
5115              le16_to_cpu(xh->xh_name_value_len));
5116
5117         if (free < need ||
5118             (xs->not_found &&
5119              count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5120                 if (need <= max_free &&
5121                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5122                         /*
5123                          * We can create the space by defragment. Since only the
5124                          * name/value will be moved, the xe shouldn't be changed
5125                          * in xs.
5126                          */
5127                         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5128                                                         xs->bucket);
5129                         if (ret) {
5130                                 mlog_errno(ret);
5131                                 goto out;
5132                         }
5133
5134                         xh_free_start = le16_to_cpu(xh->xh_free_start);
5135                         free = xh_free_start - header_size;
5136                         if (xh_free_start % blocksize < need)
5137                                 free -= xh_free_start % blocksize;
5138
5139                         if (free >= need)
5140                                 goto xattr_set;
5141
5142                         mlog(0, "Can't get enough space for xattr insert by "
5143                              "defragment. Need %u bytes, but we have %d, so "
5144                              "allocate new bucket for it.\n", need, free);
5145                 }
5146
5147                 /*
5148                  * We have to add new buckets or clusters and one
5149                  * allocation should leave us enough space for insert.
5150                  */
5151                 BUG_ON(allocation);
5152
5153                 /*
5154                  * We do not allow for overlapping ranges between buckets. And
5155                  * the maximum number of collisions we will allow for then is
5156                  * one bucket's worth, so check it here whether we need to
5157                  * add a new bucket for the insert.
5158                  */
5159                 ret = ocfs2_check_xattr_bucket_collision(inode,
5160                                                          xs->bucket,
5161                                                          xi->name);
5162                 if (ret) {
5163                         mlog_errno(ret);
5164                         goto out;
5165                 }
5166
5167                 ret = ocfs2_add_new_xattr_bucket(inode,
5168                                                  xs->xattr_bh,
5169                                                  xs->bucket,
5170                                                  ctxt);
5171                 if (ret) {
5172                         mlog_errno(ret);
5173                         goto out;
5174                 }
5175
5176                 /*
5177                  * ocfs2_add_new_xattr_bucket() will have updated
5178                  * xs->bucket if it moved, but it will not have updated
5179                  * any of the other search fields.  Thus, we drop it and
5180                  * re-search.  Everything should be cached, so it'll be
5181                  * quick.
5182                  */
5183                 ocfs2_xattr_bucket_relse(xs->bucket);
5184                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5185                                                    xi->name_index,
5186                                                    xi->name, xs);
5187                 if (ret && ret != -ENODATA)
5188                         goto out;
5189                 xs->not_found = ret;
5190                 allocation = 1;
5191                 goto try_again;
5192         }
5193
5194 xattr_set:
5195         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5196 out:
5197         mlog_exit(ret);
5198         return ret;
5199 }
5200
5201 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5202                                         struct ocfs2_xattr_bucket *bucket,
5203                                         void *para)
5204 {
5205         int ret = 0;
5206         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5207         u16 i;
5208         struct ocfs2_xattr_entry *xe;
5209         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5210         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5211         int credits = ocfs2_remove_extent_credits(osb->sb) +
5212                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5213
5214
5215         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5216
5217         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5218                 xe = &xh->xh_entries[i];
5219                 if (ocfs2_xattr_is_local(xe))
5220                         continue;
5221
5222                 ctxt.handle = ocfs2_start_trans(osb, credits);
5223                 if (IS_ERR(ctxt.handle)) {
5224                         ret = PTR_ERR(ctxt.handle);
5225                         mlog_errno(ret);
5226                         break;
5227                 }
5228
5229                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5230                                                         i, 0, &ctxt);
5231
5232                 ocfs2_commit_trans(osb, ctxt.handle);
5233                 if (ret) {
5234                         mlog_errno(ret);
5235                         break;
5236                 }
5237         }
5238
5239         ocfs2_schedule_truncate_log_flush(osb, 1);
5240         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5241         return ret;
5242 }
5243
5244 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5245                                           struct buffer_head *xb_bh)
5246 {
5247         struct ocfs2_xattr_block *xb =
5248                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5249         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5250         int ret = 0;
5251         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5252         u64 p_blkno;
5253
5254         if (le16_to_cpu(el->l_next_free_rec) == 0)
5255                 return 0;
5256
5257         while (name_hash > 0) {
5258                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5259                                           &e_cpos, &num_clusters, el);
5260                 if (ret) {
5261                         mlog_errno(ret);
5262                         goto out;
5263                 }
5264
5265                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5266                                                   ocfs2_delete_xattr_in_bucket,
5267                                                   NULL);
5268                 if (ret) {
5269                         mlog_errno(ret);
5270                         goto out;
5271                 }
5272
5273                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5274                                              p_blkno, e_cpos, num_clusters);
5275                 if (ret) {
5276                         mlog_errno(ret);
5277                         break;
5278                 }
5279
5280                 if (e_cpos == 0)
5281                         break;
5282
5283                 name_hash = e_cpos - 1;
5284         }
5285
5286 out:
5287         return ret;
5288 }
5289
5290 /*
5291  * 'security' attributes support
5292  */
5293 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5294                                         size_t list_size, const char *name,
5295                                         size_t name_len)
5296 {
5297         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5298         const size_t total_len = prefix_len + name_len + 1;
5299
5300         if (list && total_len <= list_size) {
5301                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5302                 memcpy(list + prefix_len, name, name_len);
5303                 list[prefix_len + name_len] = '\0';
5304         }
5305         return total_len;
5306 }
5307
5308 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5309                                     void *buffer, size_t size)
5310 {
5311         if (strcmp(name, "") == 0)
5312                 return -EINVAL;
5313         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5314                                buffer, size);
5315 }
5316
5317 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5318                                     const void *value, size_t size, int flags)
5319 {
5320         if (strcmp(name, "") == 0)
5321                 return -EINVAL;
5322
5323         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5324                                size, flags);
5325 }
5326
5327 int ocfs2_init_security_get(struct inode *inode,
5328                             struct inode *dir,
5329                             struct ocfs2_security_xattr_info *si)
5330 {
5331         /* check whether ocfs2 support feature xattr */
5332         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
5333                 return -EOPNOTSUPP;
5334         return security_inode_init_security(inode, dir, &si->name, &si->value,
5335                                             &si->value_len);
5336 }
5337
5338 int ocfs2_init_security_set(handle_t *handle,
5339                             struct inode *inode,
5340                             struct buffer_head *di_bh,
5341                             struct ocfs2_security_xattr_info *si,
5342                             struct ocfs2_alloc_context *xattr_ac,
5343                             struct ocfs2_alloc_context *data_ac)
5344 {
5345         return ocfs2_xattr_set_handle(handle, inode, di_bh,
5346                                      OCFS2_XATTR_INDEX_SECURITY,
5347                                      si->name, si->value, si->value_len, 0,
5348                                      xattr_ac, data_ac);
5349 }
5350
5351 struct xattr_handler ocfs2_xattr_security_handler = {
5352         .prefix = XATTR_SECURITY_PREFIX,
5353         .list   = ocfs2_xattr_security_list,
5354         .get    = ocfs2_xattr_security_get,
5355         .set    = ocfs2_xattr_security_set,
5356 };
5357
5358 /*
5359  * 'trusted' attributes support
5360  */
5361 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5362                                        size_t list_size, const char *name,
5363                                        size_t name_len)
5364 {
5365         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5366         const size_t total_len = prefix_len + name_len + 1;
5367
5368         if (list && total_len <= list_size) {
5369                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5370                 memcpy(list + prefix_len, name, name_len);
5371                 list[prefix_len + name_len] = '\0';
5372         }
5373         return total_len;
5374 }
5375
5376 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5377                                    void *buffer, size_t size)
5378 {
5379         if (strcmp(name, "") == 0)
5380                 return -EINVAL;
5381         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5382                                buffer, size);
5383 }
5384
5385 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5386                                    const void *value, size_t size, int flags)
5387 {
5388         if (strcmp(name, "") == 0)
5389                 return -EINVAL;
5390
5391         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5392                                size, flags);
5393 }
5394
5395 struct xattr_handler ocfs2_xattr_trusted_handler = {
5396         .prefix = XATTR_TRUSTED_PREFIX,
5397         .list   = ocfs2_xattr_trusted_list,
5398         .get    = ocfs2_xattr_trusted_get,
5399         .set    = ocfs2_xattr_trusted_set,
5400 };
5401
5402 /*
5403  * 'user' attributes support
5404  */
5405 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5406                                     size_t list_size, const char *name,
5407                                     size_t name_len)
5408 {
5409         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5410         const size_t total_len = prefix_len + name_len + 1;
5411         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5412
5413         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5414                 return 0;
5415
5416         if (list && total_len <= list_size) {
5417                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5418                 memcpy(list + prefix_len, name, name_len);
5419                 list[prefix_len + name_len] = '\0';
5420         }
5421         return total_len;
5422 }
5423
5424 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5425                                 void *buffer, size_t size)
5426 {
5427         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5428
5429         if (strcmp(name, "") == 0)
5430                 return -EINVAL;
5431         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5432                 return -EOPNOTSUPP;
5433         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5434                                buffer, size);
5435 }
5436
5437 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5438                                 const void *value, size_t size, int flags)
5439 {
5440         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5441
5442         if (strcmp(name, "") == 0)
5443                 return -EINVAL;
5444         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5445                 return -EOPNOTSUPP;
5446
5447         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5448                                size, flags);
5449 }
5450
5451 struct xattr_handler ocfs2_xattr_user_handler = {
5452         .prefix = XATTR_USER_PREFIX,
5453         .list   = ocfs2_xattr_user_list,
5454         .get    = ocfs2_xattr_user_get,
5455         .set    = ocfs2_xattr_user_set,
5456 };