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_HEADER_GAP  4
  86 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
  87                                          - sizeof(struct ocfs2_xattr_header) \
  88                                          - OCFS2_XATTR_HEADER_GAP)
  89 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
  90                                          - sizeof(struct ocfs2_xattr_block) \
  91                                          - sizeof(struct ocfs2_xattr_header) \
  92                                          - OCFS2_XATTR_HEADER_GAP)
  93
  94 static struct ocfs2_xattr_def_value_root def_xv = {
  95         .xv.xr_list.l_count = cpu_to_le16(1),
  96 };
  97
  98 struct xattr_handler *ocfs2_xattr_handlers[] = {
  99         &ocfs2_xattr_user_handler,
 100 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
 101         &ocfs2_xattr_acl_access_handler,
 102         &ocfs2_xattr_acl_default_handler,
 103 #endif
 104         &ocfs2_xattr_trusted_handler,
 105         &ocfs2_xattr_security_handler,
 106         NULL
 107 };
 108
 109 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 110         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
 111 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
 112         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
 113                                         = &ocfs2_xattr_acl_access_handler,
 114         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
 115                                         = &ocfs2_xattr_acl_default_handler,
 116 #endif
 117         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
 118         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 119 };
 120
 121 struct ocfs2_xattr_info {
 122         int name_index;
 123         const char *name;
 124         const void *value;
 125         size_t value_len;
 126 };
 127
 128 struct ocfs2_xattr_search {
 129         struct buffer_head *inode_bh;
 130         /*
 131          * xattr_bh point to the block buffer head which has extended attribute
 132          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 133          */
 134         struct buffer_head *xattr_bh;
 135         struct ocfs2_xattr_header *header;
 136         struct ocfs2_xattr_bucket *bucket;
 137         void *base;
 138         void *end;
 139         struct ocfs2_xattr_entry *here;
 140         int not_found;
 141 };
 142
 143 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
 144                                              struct ocfs2_xattr_header *xh,
 145                                              int index,
 146                                              int *block_off,
 147                                              int *new_offset);
 148
 149 static int ocfs2_xattr_block_find(struct inode *inode,
 150                                   int name_index,
 151                                   const char *name,
 152                                   struct ocfs2_xattr_search *xs);
 153 static int ocfs2_xattr_index_block_find(struct inode *inode,
 154                                         struct buffer_head *root_bh,
 155                                         int name_index,
 156                                         const char *name,
 157                                         struct ocfs2_xattr_search *xs);
 158
 159 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 160                                         struct ocfs2_xattr_tree_root *xt,
 161                                         char *buffer,
 162                                         size_t buffer_size);
 163
 164 static int ocfs2_xattr_create_index_block(struct inode *inode,
 165                                           struct ocfs2_xattr_search *xs,
 166                                           struct ocfs2_xattr_set_ctxt *ctxt);
 167
 168 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 169                                              struct ocfs2_xattr_info *xi,
 170                                              struct ocfs2_xattr_search *xs,
 171                                              struct ocfs2_xattr_set_ctxt *ctxt);
 172
 173 static int ocfs2_delete_xattr_index_block(struct inode *inode,
 174                                           struct buffer_head *xb_bh);
 175 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 176                                   u64 src_blk, u64 last_blk, u64 to_blk,
 177                                   unsigned int start_bucket,
 178                                   u32 *first_hash);
 179
 180 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 181 {
 182         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 183 }
 184
 185 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 186 {
 187         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 188 }
 189
 190 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
 191 {
 192         u16 len = sb->s_blocksize -
 193                  offsetof(struct ocfs2_xattr_header, xh_entries);
 194
 195         return len / sizeof(struct ocfs2_xattr_entry);
 196 }
 197
 198 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 199 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 200 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 201
 202 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 203 {
 204         struct ocfs2_xattr_bucket *bucket;
 205         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 206
 207         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 208
 209         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 210         if (bucket) {
 211                 bucket->bu_inode = inode;
 212                 bucket->bu_blocks = blks;
 213         }
 214
 215         return bucket;
 216 }
 217
 218 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 219 {
 220         int i;
 221
 222         for (i = 0; i < bucket->bu_blocks; i++) {
 223                 brelse(bucket->bu_bhs[i]);
 224                 bucket->bu_bhs[i] = NULL;
 225         }
 226 }
 227
 228 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 229 {
 230         if (bucket) {
 231                 ocfs2_xattr_bucket_relse(bucket);
 232                 bucket->bu_inode = NULL;
 233                 kfree(bucket);
 234         }
 235 }
 236
 237 /*
 238  * A bucket that has never been written to disk doesn't need to be
 239  * read.  We just need the buffer_heads.  Don't call this for
 240  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 241  * them fully.
 242  */
 243 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 244                                    u64 xb_blkno)
 245 {
 246         int i, rc = 0;
 247
 248         for (i = 0; i < bucket->bu_blocks; i++) {
 249                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 250                                               xb_blkno + i);
 251                 if (!bucket->bu_bhs[i]) {
 252                         rc = -EIO;
 253                         mlog_errno(rc);
 254                         break;
 255                 }
 256
 257                 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 258                                            bucket->bu_bhs[i]))
 259                         ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 260                                                       bucket->bu_bhs[i]);
 261         }
 262
 263         if (rc)
 264                 ocfs2_xattr_bucket_relse(bucket);
 265         return rc;
 266 }
 267
 268 /* Read the xattr bucket at xb_blkno */
 269 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 270                                    u64 xb_blkno)
 271 {
 272         int rc;
 273
 274         rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
 275                                bucket->bu_blocks, bucket->bu_bhs, 0,
 276                                NULL);
 277         if (!rc) {
 278                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 279                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 280                                                  bucket->bu_bhs,
 281                                                  bucket->bu_blocks,
 282                                                  &bucket_xh(bucket)->xh_check);
 283                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 284                 if (rc)
 285                         mlog_errno(rc);
 286         }
 287
 288         if (rc)
 289                 ocfs2_xattr_bucket_relse(bucket);
 290         return rc;
 291 }
 292
 293 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 294                                              struct ocfs2_xattr_bucket *bucket,
 295                                              int type)
 296 {
 297         int i, rc = 0;
 298
 299         for (i = 0; i < bucket->bu_blocks; i++) {
 300                 rc = ocfs2_journal_access(handle,
 301                                           INODE_CACHE(bucket->bu_inode),
 302                                           bucket->bu_bhs[i], type);
 303                 if (rc) {
 304                         mlog_errno(rc);
 305                         break;
 306                 }
 307         }
 308
 309         return rc;
 310 }
 311
 312 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 313                                              struct ocfs2_xattr_bucket *bucket)
 314 {
 315         int i;
 316
 317         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 318         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 319                                    bucket->bu_bhs, bucket->bu_blocks,
 320                                    &bucket_xh(bucket)->xh_check);
 321         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 322
 323         for (i = 0; i < bucket->bu_blocks; i++)
 324                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 325 }
 326
 327 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 328                                          struct ocfs2_xattr_bucket *src)
 329 {
 330         int i;
 331         int blocksize = src->bu_inode->i_sb->s_blocksize;
 332
 333         BUG_ON(dest->bu_blocks != src->bu_blocks);
 334         BUG_ON(dest->bu_inode != src->bu_inode);
 335
 336         for (i = 0; i < src->bu_blocks; i++) {
 337                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 338                        blocksize);
 339         }
 340 }
 341
 342 static int ocfs2_validate_xattr_block(struct super_block *sb,
 343                                       struct buffer_head *bh)
 344 {
 345         int rc;
 346         struct ocfs2_xattr_block *xb =
 347                 (struct ocfs2_xattr_block *)bh->b_data;
 348
 349         mlog(0, "Validating xattr block %llu\n",
 350              (unsigned long long)bh->b_blocknr);
 351
 352         BUG_ON(!buffer_uptodate(bh));
 353
 354         /*
 355          * If the ecc fails, we return the error but otherwise
 356          * leave the filesystem running.  We know any error is
 357          * local to this block.
 358          */
 359         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 360         if (rc)
 361                 return rc;
 362
 363         /*
 364          * Errors after here are fatal
 365          */
 366
 367         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 368                 ocfs2_error(sb,
 369                             "Extended attribute block #%llu has bad "
 370                             "signature %.*s",
 371                             (unsigned long long)bh->b_blocknr, 7,
 372                             xb->xb_signature);
 373                 return -EINVAL;
 374         }
 375
 376         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 377                 ocfs2_error(sb,
 378                             "Extended attribute block #%llu has an "
 379                             "invalid xb_blkno of %llu",
 380                             (unsigned long long)bh->b_blocknr,
 381                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
 382                 return -EINVAL;
 383         }
 384
 385         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 386                 ocfs2_error(sb,
 387                             "Extended attribute block #%llu has an invalid "
 388                             "xb_fs_generation of #%u",
 389                             (unsigned long long)bh->b_blocknr,
 390                             le32_to_cpu(xb->xb_fs_generation));
 391                 return -EINVAL;
 392         }
 393
 394         return 0;
 395 }
 396
 397 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 398                                   struct buffer_head **bh)
 399 {
 400         int rc;
 401         struct buffer_head *tmp = *bh;
 402
 403         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
 404                               ocfs2_validate_xattr_block);
 405
 406         /* If ocfs2_read_block() got us a new bh, pass it up. */
 407         if (!rc && !*bh)
 408                 *bh = tmp;
 409
 410         return rc;
 411 }
 412
 413 static inline const char *ocfs2_xattr_prefix(int name_index)
 414 {
 415         struct xattr_handler *handler = NULL;
 416
 417         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 418                 handler = ocfs2_xattr_handler_map[name_index];
 419
 420         return handler ? handler->prefix : NULL;
 421 }
 422
 423 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 424                                  const char *name,
 425                                  int name_len)
 426 {
 427         /* Get hash value of uuid from super block */
 428         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 429         int i;
 430
 431         /* hash extended attribute name */
 432         for (i = 0; i < name_len; i++) {
 433                 hash = (hash << OCFS2_HASH_SHIFT) ^
 434                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 435                        *name++;
 436         }
 437
 438         return hash;
 439 }
 440
 441 /*
 442  * ocfs2_xattr_hash_entry()
 443  *
 444  * Compute the hash of an extended attribute.
 445  */
 446 static void ocfs2_xattr_hash_entry(struct inode *inode,
 447                                    struct ocfs2_xattr_header *header,
 448                                    struct ocfs2_xattr_entry *entry)
 449 {
 450         u32 hash = 0;
 451         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
 452
 453         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
 454         entry->xe_name_hash = cpu_to_le32(hash);
 455
 456         return;
 457 }
 458
 459 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 460 {
 461         int size = 0;
 462
 463         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
 464                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 465         else
 466                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 467         size += sizeof(struct ocfs2_xattr_entry);
 468
 469         return size;
 470 }
 471
 472 int ocfs2_calc_security_init(struct inode *dir,
 473                              struct ocfs2_security_xattr_info *si,
 474                              int *want_clusters,
 475                              int *xattr_credits,
 476                              struct ocfs2_alloc_context **xattr_ac)
 477 {
 478         int ret = 0;
 479         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 480         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 481                                                  si->value_len);
 482
 483         /*
 484          * The max space of security xattr taken inline is
 485          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 486          * So reserve one metadata block for it is ok.
 487          */
 488         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 489             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 490                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 491                 if (ret) {
 492                         mlog_errno(ret);
 493                         return ret;
 494                 }
 495                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 496         }
 497
 498         /* reserve clusters for xattr value which will be set in B tree*/
 499         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 500                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 501                                                             si->value_len);
 502
 503                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 504                                                            new_clusters);
 505                 *want_clusters += new_clusters;
 506         }
 507         return ret;
 508 }
 509
 510 int ocfs2_calc_xattr_init(struct inode *dir,
 511                           struct buffer_head *dir_bh,
 512                           int mode,
 513                           struct ocfs2_security_xattr_info *si,
 514                           int *want_clusters,
 515                           int *xattr_credits,
 516                           int *want_meta)
 517 {
 518         int ret = 0;
 519         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 520         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 521
 522         if (si->enable)
 523                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 524                                                      si->value_len);
 525
 526         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 527                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 528                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 529                                         "", NULL, 0);
 530                 if (acl_len > 0) {
 531                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 532                         if (S_ISDIR(mode))
 533                                 a_size <<= 1;
 534                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 535                         mlog_errno(ret);
 536                         return ret;
 537                 }
 538         }
 539
 540         if (!(s_size + a_size))
 541                 return ret;
 542
 543         /*
 544          * The max space of security xattr taken inline is
 545          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 546          * The max space of acl xattr taken inline is
 547          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 548          * when blocksize = 512, may reserve one more cluser for
 549          * xattr bucket, otherwise reserve one metadata block
 550          * for them is ok.
 551          * If this is a new directory with inline data,
 552          * we choose to reserve the entire inline area for
 553          * directory contents and force an external xattr block.
 554          */
 555         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 556             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 557             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 558                 *want_meta = *want_meta + 1;
 559                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 560         }
 561
 562         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 563             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 564                 *want_clusters += 1;
 565                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 566         }
 567
 568         /*
 569          * reserve credits and clusters for xattrs which has large value
 570          * and have to be set outside
 571          */
 572         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 573                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 574                                                         si->value_len);
 575                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 576                                                            new_clusters);
 577                 *want_clusters += new_clusters;
 578         }
 579         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 580             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 581                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 582                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 583                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 584                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 585                                                            new_clusters);
 586                 *want_clusters += new_clusters;
 587         }
 588
 589         return ret;
 590 }
 591
 592 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 593                                          u32 clusters_to_add,
 594                                          struct ocfs2_xattr_value_buf *vb,
 595                                          struct ocfs2_xattr_set_ctxt *ctxt)
 596 {
 597         int status = 0;
 598         handle_t *handle = ctxt->handle;
 599         enum ocfs2_alloc_restarted why;
 600         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 601         struct ocfs2_extent_tree et;
 602
 603         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
 604
 605         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 606
 607         status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 608                               OCFS2_JOURNAL_ACCESS_WRITE);
 609         if (status < 0) {
 610                 mlog_errno(status);
 611                 goto leave;
 612         }
 613
 614         prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 615         status = ocfs2_add_clusters_in_btree(handle,
 616                                              &et,
 617                                              &logical_start,
 618                                              clusters_to_add,
 619                                              0,
 620                                              ctxt->data_ac,
 621                                              ctxt->meta_ac,
 622                                              &why);
 623         if (status < 0) {
 624                 mlog_errno(status);
 625                 goto leave;
 626         }
 627
 628         status = ocfs2_journal_dirty(handle, vb->vb_bh);
 629         if (status < 0) {
 630                 mlog_errno(status);
 631                 goto leave;
 632         }
 633
 634         clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
 635
 636         /*
 637          * We should have already allocated enough space before the transaction,
 638          * so no need to restart.
 639          */
 640         BUG_ON(why != RESTART_NONE || clusters_to_add);
 641
 642 leave:
 643
 644         return status;
 645 }
 646
 647 static int __ocfs2_remove_xattr_range(struct inode *inode,
 648                                       struct ocfs2_xattr_value_buf *vb,
 649                                       u32 cpos, u32 phys_cpos, u32 len,
 650                                       struct ocfs2_xattr_set_ctxt *ctxt)
 651 {
 652         int ret;
 653         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 654         handle_t *handle = ctxt->handle;
 655         struct ocfs2_extent_tree et;
 656
 657         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 658
 659         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 660                             OCFS2_JOURNAL_ACCESS_WRITE);
 661         if (ret) {
 662                 mlog_errno(ret);
 663                 goto out;
 664         }
 665
 666         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 667                                   &ctxt->dealloc);
 668         if (ret) {
 669                 mlog_errno(ret);
 670                 goto out;
 671         }
 672
 673         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 674
 675         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
 676         if (ret) {
 677                 mlog_errno(ret);
 678                 goto out;
 679         }
 680
 681         ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
 682         if (ret)
 683                 mlog_errno(ret);
 684
 685 out:
 686         return ret;
 687 }
 688
 689 static int ocfs2_xattr_shrink_size(struct inode *inode,
 690                                    u32 old_clusters,
 691                                    u32 new_clusters,
 692                                    struct ocfs2_xattr_value_buf *vb,
 693                                    struct ocfs2_xattr_set_ctxt *ctxt)
 694 {
 695         int ret = 0;
 696         u32 trunc_len, cpos, phys_cpos, alloc_size;
 697         u64 block;
 698
 699         if (old_clusters <= new_clusters)
 700                 return 0;
 701
 702         cpos = new_clusters;
 703         trunc_len = old_clusters - new_clusters;
 704         while (trunc_len) {
 705                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 706                                                &alloc_size,
 707                                                &vb->vb_xv->xr_list, NULL);
 708                 if (ret) {
 709                         mlog_errno(ret);
 710                         goto out;
 711                 }
 712
 713                 if (alloc_size > trunc_len)
 714                         alloc_size = trunc_len;
 715
 716                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 717                                                  phys_cpos, alloc_size,
 718                                                  ctxt);
 719                 if (ret) {
 720                         mlog_errno(ret);
 721                         goto out;
 722                 }
 723
 724                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 725                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 726                                                        block, alloc_size);
 727                 cpos += alloc_size;
 728                 trunc_len -= alloc_size;
 729         }
 730
 731 out:
 732         return ret;
 733 }
 734
 735 static int ocfs2_xattr_value_truncate(struct inode *inode,
 736                                       struct ocfs2_xattr_value_buf *vb,
 737                                       int len,
 738                                       struct ocfs2_xattr_set_ctxt *ctxt)
 739 {
 740         int ret;
 741         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 742         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 743
 744         if (new_clusters == old_clusters)
 745                 return 0;
 746
 747         if (new_clusters > old_clusters)
 748                 ret = ocfs2_xattr_extend_allocation(inode,
 749                                                     new_clusters - old_clusters,
 750                                                     vb, ctxt);
 751         else
 752                 ret = ocfs2_xattr_shrink_size(inode,
 753                                               old_clusters, new_clusters,
 754                                               vb, ctxt);
 755
 756         return ret;
 757 }
 758
 759 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 760                                   size_t *result, const char *prefix,
 761                                   const char *name, int name_len)
 762 {
 763         char *p = buffer + *result;
 764         int prefix_len = strlen(prefix);
 765         int total_len = prefix_len + name_len + 1;
 766
 767         *result += total_len;
 768
 769         /* we are just looking for how big our buffer needs to be */
 770         if (!size)
 771                 return 0;
 772
 773         if (*result > size)
 774                 return -ERANGE;
 775
 776         memcpy(p, prefix, prefix_len);
 777         memcpy(p + prefix_len, name, name_len);
 778         p[prefix_len + name_len] = '\0';
 779
 780         return 0;
 781 }
 782
 783 static int ocfs2_xattr_list_entries(struct inode *inode,
 784                                     struct ocfs2_xattr_header *header,
 785                                     char *buffer, size_t buffer_size)
 786 {
 787         size_t result = 0;
 788         int i, type, ret;
 789         const char *prefix, *name;
 790
 791         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 792                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 793                 type = ocfs2_xattr_get_type(entry);
 794                 prefix = ocfs2_xattr_prefix(type);
 795
 796                 if (prefix) {
 797                         name = (const char *)header +
 798                                 le16_to_cpu(entry->xe_name_offset);
 799
 800                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 801                                                      &result, prefix, name,
 802                                                      entry->xe_name_len);
 803                         if (ret)
 804                                 return ret;
 805                 }
 806         }
 807
 808         return result;
 809 }
 810
 811 static int ocfs2_xattr_ibody_list(struct inode *inode,
 812                                   struct ocfs2_dinode *di,
 813                                   char *buffer,
 814                                   size_t buffer_size)
 815 {
 816         struct ocfs2_xattr_header *header = NULL;
 817         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 818         int ret = 0;
 819
 820         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 821                 return ret;
 822
 823         header = (struct ocfs2_xattr_header *)
 824                  ((void *)di + inode->i_sb->s_blocksize -
 825                  le16_to_cpu(di->i_xattr_inline_size));
 826
 827         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 828
 829         return ret;
 830 }
 831
 832 static int ocfs2_xattr_block_list(struct inode *inode,
 833                                   struct ocfs2_dinode *di,
 834                                   char *buffer,
 835                                   size_t buffer_size)
 836 {
 837         struct buffer_head *blk_bh = NULL;
 838         struct ocfs2_xattr_block *xb;
 839         int ret = 0;
 840
 841         if (!di->i_xattr_loc)
 842                 return ret;
 843
 844         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
 845                                      &blk_bh);
 846         if (ret < 0) {
 847                 mlog_errno(ret);
 848                 return ret;
 849         }
 850
 851         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 852         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 853                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 854                 ret = ocfs2_xattr_list_entries(inode, header,
 855                                                buffer, buffer_size);
 856         } else {
 857                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
 858                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
 859                                                    buffer, buffer_size);
 860         }
 861
 862         brelse(blk_bh);
 863
 864         return ret;
 865 }
 866
 867 ssize_t ocfs2_listxattr(struct dentry *dentry,
 868                         char *buffer,
 869                         size_t size)
 870 {
 871         int ret = 0, i_ret = 0, b_ret = 0;
 872         struct buffer_head *di_bh = NULL;
 873         struct ocfs2_dinode *di = NULL;
 874         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
 875
 876         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
 877                 return -EOPNOTSUPP;
 878
 879         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 880                 return ret;
 881
 882         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
 883         if (ret < 0) {
 884                 mlog_errno(ret);
 885                 return ret;
 886         }
 887
 888         di = (struct ocfs2_dinode *)di_bh->b_data;
 889
 890         down_read(&oi->ip_xattr_sem);
 891         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
 892         if (i_ret < 0)
 893                 b_ret = 0;
 894         else {
 895                 if (buffer) {
 896                         buffer += i_ret;
 897                         size -= i_ret;
 898                 }
 899                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
 900                                                buffer, size);
 901                 if (b_ret < 0)
 902                         i_ret = 0;
 903         }
 904         up_read(&oi->ip_xattr_sem);
 905         ocfs2_inode_unlock(dentry->d_inode, 0);
 906
 907         brelse(di_bh);
 908
 909         return i_ret + b_ret;
 910 }
 911
 912 static int ocfs2_xattr_find_entry(int name_index,
 913                                   const char *name,
 914                                   struct ocfs2_xattr_search *xs)
 915 {
 916         struct ocfs2_xattr_entry *entry;
 917         size_t name_len;
 918         int i, cmp = 1;
 919
 920         if (name == NULL)
 921                 return -EINVAL;
 922
 923         name_len = strlen(name);
 924         entry = xs->here;
 925         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
 926                 cmp = name_index - ocfs2_xattr_get_type(entry);
 927                 if (!cmp)
 928                         cmp = name_len - entry->xe_name_len;
 929                 if (!cmp)
 930                         cmp = memcmp(name, (xs->base +
 931                                      le16_to_cpu(entry->xe_name_offset)),
 932                                      name_len);
 933                 if (cmp == 0)
 934                         break;
 935                 entry += 1;
 936         }
 937         xs->here = entry;
 938
 939         return cmp ? -ENODATA : 0;
 940 }
 941
 942 static int ocfs2_xattr_get_value_outside(struct inode *inode,
 943                                          struct ocfs2_xattr_value_root *xv,
 944                                          void *buffer,
 945                                          size_t len)
 946 {
 947         u32 cpos, p_cluster, num_clusters, bpc, clusters;
 948         u64 blkno;
 949         int i, ret = 0;
 950         size_t cplen, blocksize;
 951         struct buffer_head *bh = NULL;
 952         struct ocfs2_extent_list *el;
 953
 954         el = &xv->xr_list;
 955         clusters = le32_to_cpu(xv->xr_clusters);
 956         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 957         blocksize = inode->i_sb->s_blocksize;
 958
 959         cpos = 0;
 960         while (cpos < clusters) {
 961                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 962                                                &num_clusters, el, NULL);
 963                 if (ret) {
 964                         mlog_errno(ret);
 965                         goto out;
 966                 }
 967
 968                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 969                 /* Copy ocfs2_xattr_value */
 970                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 971                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
 972                                                &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, i;
1055         int uninitialized_var(block_off);
1056
1057         xs->bucket = ocfs2_xattr_bucket_new(inode);
1058         if (!xs->bucket) {
1059                 ret = -ENOMEM;
1060                 mlog_errno(ret);
1061                 goto cleanup;
1062         }
1063
1064         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1065         if (ret) {
1066                 mlog_errno(ret);
1067                 goto cleanup;
1068         }
1069
1070         if (xs->not_found) {
1071                 ret = -ENODATA;
1072                 goto cleanup;
1073         }
1074
1075         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1076         size = le64_to_cpu(xs->here->xe_value_size);
1077         if (buffer) {
1078                 ret = -ERANGE;
1079                 if (size > buffer_size)
1080                         goto cleanup;
1081
1082                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1083                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1084                 i = xs->here - xs->header->xh_entries;
1085
1086                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1087                         ret = ocfs2_xattr_bucket_get_name_value(inode,
1088                                                                 bucket_xh(xs->bucket),
1089                                                                 i,
1090                                                                 &block_off,
1091                                                                 &name_offset);
1092                         xs->base = bucket_block(xs->bucket, block_off);
1093                 }
1094                 if (ocfs2_xattr_is_local(xs->here)) {
1095                         memcpy(buffer, (void *)xs->base +
1096                                name_offset + name_len, size);
1097                 } else {
1098                         xv = (struct ocfs2_xattr_value_root *)
1099                                 (xs->base + name_offset + name_len);
1100                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1101                                                             buffer, size);
1102                         if (ret < 0) {
1103                                 mlog_errno(ret);
1104                                 goto cleanup;
1105                         }
1106                 }
1107         }
1108         ret = size;
1109 cleanup:
1110         ocfs2_xattr_bucket_free(xs->bucket);
1111
1112         brelse(xs->xattr_bh);
1113         xs->xattr_bh = NULL;
1114         return ret;
1115 }
1116
1117 int ocfs2_xattr_get_nolock(struct inode *inode,
1118                            struct buffer_head *di_bh,
1119                            int name_index,
1120                            const char *name,
1121                            void *buffer,
1122                            size_t buffer_size)
1123 {
1124         int ret;
1125         struct ocfs2_dinode *di = NULL;
1126         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1127         struct ocfs2_xattr_search xis = {
1128                 .not_found = -ENODATA,
1129         };
1130         struct ocfs2_xattr_search xbs = {
1131                 .not_found = -ENODATA,
1132         };
1133
1134         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1135                 return -EOPNOTSUPP;
1136
1137         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1138                 ret = -ENODATA;
1139
1140         xis.inode_bh = xbs.inode_bh = di_bh;
1141         di = (struct ocfs2_dinode *)di_bh->b_data;
1142
1143         down_read(&oi->ip_xattr_sem);
1144         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1145                                     buffer_size, &xis);
1146         if (ret == -ENODATA && di->i_xattr_loc)
1147                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1148                                             buffer_size, &xbs);
1149         up_read(&oi->ip_xattr_sem);
1150
1151         return ret;
1152 }
1153
1154 /* ocfs2_xattr_get()
1155  *
1156  * Copy an extended attribute into the buffer provided.
1157  * Buffer is NULL to compute the size of buffer required.
1158  */
1159 static int ocfs2_xattr_get(struct inode *inode,
1160                            int name_index,
1161                            const char *name,
1162                            void *buffer,
1163                            size_t buffer_size)
1164 {
1165         int ret;
1166         struct buffer_head *di_bh = NULL;
1167
1168         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1169         if (ret < 0) {
1170                 mlog_errno(ret);
1171                 return ret;
1172         }
1173         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1174                                      name, buffer, buffer_size);
1175
1176         ocfs2_inode_unlock(inode, 0);
1177
1178         brelse(di_bh);
1179
1180         return ret;
1181 }
1182
1183 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1184                                            handle_t *handle,
1185                                            struct ocfs2_xattr_value_root *xv,
1186                                            const void *value,
1187                                            int value_len)
1188 {
1189         int ret = 0, i, cp_len;
1190         u16 blocksize = inode->i_sb->s_blocksize;
1191         u32 p_cluster, num_clusters;
1192         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1193         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1194         u64 blkno;
1195         struct buffer_head *bh = NULL;
1196
1197         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1198
1199         while (cpos < clusters) {
1200                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1201                                                &num_clusters, &xv->xr_list,
1202                                                NULL);
1203                 if (ret) {
1204                         mlog_errno(ret);
1205                         goto out;
1206                 }
1207
1208                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1209
1210                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1211                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1212                                                &bh, NULL);
1213                         if (ret) {
1214                                 mlog_errno(ret);
1215                                 goto out;
1216                         }
1217
1218                         ret = ocfs2_journal_access(handle,
1219                                                    INODE_CACHE(inode),
1220                                                    bh,
1221                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1222                         if (ret < 0) {
1223                                 mlog_errno(ret);
1224                                 goto out;
1225                         }
1226
1227                         cp_len = value_len > blocksize ? blocksize : value_len;
1228                         memcpy(bh->b_data, value, cp_len);
1229                         value_len -= cp_len;
1230                         value += cp_len;
1231                         if (cp_len < blocksize)
1232                                 memset(bh->b_data + cp_len, 0,
1233                                        blocksize - cp_len);
1234
1235                         ret = ocfs2_journal_dirty(handle, bh);
1236                         if (ret < 0) {
1237                                 mlog_errno(ret);
1238                                 goto out;
1239                         }
1240                         brelse(bh);
1241                         bh = NULL;
1242
1243                         /*
1244                          * XXX: do we need to empty all the following
1245                          * blocks in this cluster?
1246                          */
1247                         if (!value_len)
1248                                 break;
1249                 }
1250                 cpos += num_clusters;
1251         }
1252 out:
1253         brelse(bh);
1254
1255         return ret;
1256 }
1257
1258 static int ocfs2_xattr_cleanup(struct inode *inode,
1259                                handle_t *handle,
1260                                struct ocfs2_xattr_info *xi,
1261                                struct ocfs2_xattr_search *xs,
1262                                struct ocfs2_xattr_value_buf *vb,
1263                                size_t offs)
1264 {
1265         int ret = 0;
1266         size_t name_len = strlen(xi->name);
1267         void *val = xs->base + offs;
1268         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1269
1270         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1271                             OCFS2_JOURNAL_ACCESS_WRITE);
1272         if (ret) {
1273                 mlog_errno(ret);
1274                 goto out;
1275         }
1276         /* Decrease xattr count */
1277         le16_add_cpu(&xs->header->xh_count, -1);
1278         /* Remove the xattr entry and tree root which has already be set*/
1279         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1280         memset(val, 0, size);
1281
1282         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1283         if (ret < 0)
1284                 mlog_errno(ret);
1285 out:
1286         return ret;
1287 }
1288
1289 static int ocfs2_xattr_update_entry(struct inode *inode,
1290                                     handle_t *handle,
1291                                     struct ocfs2_xattr_info *xi,
1292                                     struct ocfs2_xattr_search *xs,
1293                                     struct ocfs2_xattr_value_buf *vb,
1294                                     size_t offs)
1295 {
1296         int ret;
1297
1298         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1299                             OCFS2_JOURNAL_ACCESS_WRITE);
1300         if (ret) {
1301                 mlog_errno(ret);
1302                 goto out;
1303         }
1304
1305         xs->here->xe_name_offset = cpu_to_le16(offs);
1306         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1307         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1308                 ocfs2_xattr_set_local(xs->here, 1);
1309         else
1310                 ocfs2_xattr_set_local(xs->here, 0);
1311         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1312
1313         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1314         if (ret < 0)
1315                 mlog_errno(ret);
1316 out:
1317         return ret;
1318 }
1319
1320 /*
1321  * ocfs2_xattr_set_value_outside()
1322  *
1323  * Set large size value in B tree.
1324  */
1325 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1326                                          struct ocfs2_xattr_info *xi,
1327                                          struct ocfs2_xattr_search *xs,
1328                                          struct ocfs2_xattr_set_ctxt *ctxt,
1329                                          struct ocfs2_xattr_value_buf *vb,
1330                                          size_t offs)
1331 {
1332         size_t name_len = strlen(xi->name);
1333         void *val = xs->base + offs;
1334         struct ocfs2_xattr_value_root *xv = NULL;
1335         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1336         int ret = 0;
1337
1338         memset(val, 0, size);
1339         memcpy(val, xi->name, name_len);
1340         xv = (struct ocfs2_xattr_value_root *)
1341                 (val + OCFS2_XATTR_SIZE(name_len));
1342         xv->xr_clusters = 0;
1343         xv->xr_last_eb_blk = 0;
1344         xv->xr_list.l_tree_depth = 0;
1345         xv->xr_list.l_count = cpu_to_le16(1);
1346         xv->xr_list.l_next_free_rec = 0;
1347         vb->vb_xv = xv;
1348
1349         ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1350         if (ret < 0) {
1351                 mlog_errno(ret);
1352                 return ret;
1353         }
1354         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1355         if (ret < 0) {
1356                 mlog_errno(ret);
1357                 return ret;
1358         }
1359         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
1360                                               xi->value, xi->value_len);
1361         if (ret < 0)
1362                 mlog_errno(ret);
1363
1364         return ret;
1365 }
1366
1367 /*
1368  * ocfs2_xattr_set_entry_local()
1369  *
1370  * Set, replace or remove extended attribute in local.
1371  */
1372 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1373                                         struct ocfs2_xattr_info *xi,
1374                                         struct ocfs2_xattr_search *xs,
1375                                         struct ocfs2_xattr_entry *last,
1376                                         size_t min_offs)
1377 {
1378         size_t name_len = strlen(xi->name);
1379         int i;
1380
1381         if (xi->value && xs->not_found) {
1382                 /* Insert the new xattr entry. */
1383                 le16_add_cpu(&xs->header->xh_count, 1);
1384                 ocfs2_xattr_set_type(last, xi->name_index);
1385                 ocfs2_xattr_set_local(last, 1);
1386                 last->xe_name_len = name_len;
1387         } else {
1388                 void *first_val;
1389                 void *val;
1390                 size_t offs, size;
1391
1392                 first_val = xs->base + min_offs;
1393                 offs = le16_to_cpu(xs->here->xe_name_offset);
1394                 val = xs->base + offs;
1395
1396                 if (le64_to_cpu(xs->here->xe_value_size) >
1397                     OCFS2_XATTR_INLINE_SIZE)
1398                         size = OCFS2_XATTR_SIZE(name_len) +
1399                                 OCFS2_XATTR_ROOT_SIZE;
1400                 else
1401                         size = OCFS2_XATTR_SIZE(name_len) +
1402                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1403
1404                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1405                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1406                         /* The old and the new value have the
1407                            same size. Just replace the value. */
1408                         ocfs2_xattr_set_local(xs->here, 1);
1409                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1410                         /* Clear value bytes. */
1411                         memset(val + OCFS2_XATTR_SIZE(name_len),
1412                                0,
1413                                OCFS2_XATTR_SIZE(xi->value_len));
1414                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1415                                xi->value,
1416                                xi->value_len);
1417                         return;
1418                 }
1419                 /* Remove the old name+value. */
1420                 memmove(first_val + size, first_val, val - first_val);
1421                 memset(first_val, 0, size);
1422                 xs->here->xe_name_hash = 0;
1423                 xs->here->xe_name_offset = 0;
1424                 ocfs2_xattr_set_local(xs->here, 1);
1425                 xs->here->xe_value_size = 0;
1426
1427                 min_offs += size;
1428
1429                 /* Adjust all value offsets. */
1430                 last = xs->header->xh_entries;
1431                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1432                         size_t o = le16_to_cpu(last->xe_name_offset);
1433
1434                         if (o < offs)
1435                                 last->xe_name_offset = cpu_to_le16(o + size);
1436                         last += 1;
1437                 }
1438
1439                 if (!xi->value) {
1440                         /* Remove the old entry. */
1441                         last -= 1;
1442                         memmove(xs->here, xs->here + 1,
1443                                 (void *)last - (void *)xs->here);
1444                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1445                         le16_add_cpu(&xs->header->xh_count, -1);
1446                 }
1447         }
1448         if (xi->value) {
1449                 /* Insert the new name+value. */
1450                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1451                                 OCFS2_XATTR_SIZE(xi->value_len);
1452                 void *val = xs->base + min_offs - size;
1453
1454                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1455                 memset(val, 0, size);
1456                 memcpy(val, xi->name, name_len);
1457                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1458                        xi->value,
1459                        xi->value_len);
1460                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1461                 ocfs2_xattr_set_local(xs->here, 1);
1462                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1463         }
1464
1465         return;
1466 }
1467
1468 /*
1469  * ocfs2_xattr_set_entry()
1470  *
1471  * Set extended attribute entry into inode or block.
1472  *
1473  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1474  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1475  * then set value in B tree with set_value_outside().
1476  */
1477 static int ocfs2_xattr_set_entry(struct inode *inode,
1478                                  struct ocfs2_xattr_info *xi,
1479                                  struct ocfs2_xattr_search *xs,
1480                                  struct ocfs2_xattr_set_ctxt *ctxt,
1481                                  int flag)
1482 {
1483         struct ocfs2_xattr_entry *last;
1484         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1485         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1486         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1487         size_t size_l = 0;
1488         handle_t *handle = ctxt->handle;
1489         int free, i, ret;
1490         struct ocfs2_xattr_info xi_l = {
1491                 .name_index = xi->name_index,
1492                 .name = xi->name,
1493                 .value = xi->value,
1494                 .value_len = xi->value_len,
1495         };
1496         struct ocfs2_xattr_value_buf vb = {
1497                 .vb_bh = xs->xattr_bh,
1498                 .vb_access = ocfs2_journal_access_di,
1499         };
1500
1501         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1502                 BUG_ON(xs->xattr_bh == xs->inode_bh);
1503                 vb.vb_access = ocfs2_journal_access_xb;
1504         } else
1505                 BUG_ON(xs->xattr_bh != xs->inode_bh);
1506
1507         /* Compute min_offs, last and free space. */
1508         last = xs->header->xh_entries;
1509
1510         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1511                 size_t offs = le16_to_cpu(last->xe_name_offset);
1512                 if (offs < min_offs)
1513                         min_offs = offs;
1514                 last += 1;
1515         }
1516
1517         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
1518         if (free < 0)
1519                 return -EIO;
1520
1521         if (!xs->not_found) {
1522                 size_t size = 0;
1523                 if (ocfs2_xattr_is_local(xs->here))
1524                         size = OCFS2_XATTR_SIZE(name_len) +
1525                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1526                 else
1527                         size = OCFS2_XATTR_SIZE(name_len) +
1528                                 OCFS2_XATTR_ROOT_SIZE;
1529                 free += (size + sizeof(struct ocfs2_xattr_entry));
1530         }
1531         /* Check free space in inode or block */
1532         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1533                 if (free < sizeof(struct ocfs2_xattr_entry) +
1534                            OCFS2_XATTR_SIZE(name_len) +
1535                            OCFS2_XATTR_ROOT_SIZE) {
1536                         ret = -ENOSPC;
1537                         goto out;
1538                 }
1539                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1540                 xi_l.value = (void *)&def_xv;
1541                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1542         } else if (xi->value) {
1543                 if (free < sizeof(struct ocfs2_xattr_entry) +
1544                            OCFS2_XATTR_SIZE(name_len) +
1545                            OCFS2_XATTR_SIZE(xi->value_len)) {
1546                         ret = -ENOSPC;
1547                         goto out;
1548                 }
1549         }
1550
1551         if (!xs->not_found) {
1552                 /* For existing extended attribute */
1553                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1554                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1555                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1556                 void *val = xs->base + offs;
1557
1558                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1559                         /* Replace existing local xattr with tree root */
1560                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1561                                                             ctxt, &vb, offs);
1562                         if (ret < 0)
1563                                 mlog_errno(ret);
1564                         goto out;
1565                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1566                         /* For existing xattr which has value outside */
1567                         vb.vb_xv = (struct ocfs2_xattr_value_root *)
1568                                 (val + OCFS2_XATTR_SIZE(name_len));
1569
1570                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1571                                 /*
1572                                  * If new value need set outside also,
1573                                  * first truncate old value to new value,
1574                                  * then set new value with set_value_outside().
1575                                  */
1576                                 ret = ocfs2_xattr_value_truncate(inode,
1577                                                                  &vb,
1578                                                                  xi->value_len,
1579                                                                  ctxt);
1580                                 if (ret < 0) {
1581                                         mlog_errno(ret);
1582                                         goto out;
1583                                 }
1584
1585                                 ret = ocfs2_xattr_update_entry(inode,
1586                                                                handle,
1587                                                                xi,
1588                                                                xs,
1589                                                                &vb,
1590                                                                offs);
1591                                 if (ret < 0) {
1592                                         mlog_errno(ret);
1593                                         goto out;
1594                                 }
1595
1596                                 ret = __ocfs2_xattr_set_value_outside(inode,
1597                                                                 handle,
1598                                                                 vb.vb_xv,
1599                                                                 xi->value,
1600                                                                 xi->value_len);
1601                                 if (ret < 0)
1602                                         mlog_errno(ret);
1603                                 goto out;
1604                         } else {
1605                                 /*
1606                                  * If new value need set in local,
1607                                  * just trucate old value to zero.
1608                                  */
1609                                  ret = ocfs2_xattr_value_truncate(inode,
1610                                                                   &vb,
1611                                                                   0,
1612                                                                   ctxt);
1613                                 if (ret < 0)
1614                                         mlog_errno(ret);
1615                         }
1616                 }
1617         }
1618
1619         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), xs->inode_bh,
1620                                       OCFS2_JOURNAL_ACCESS_WRITE);
1621         if (ret) {
1622                 mlog_errno(ret);
1623                 goto out;
1624         }
1625
1626         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1627                 ret = vb.vb_access(handle, INODE_CACHE(inode), vb.vb_bh,
1628                                    OCFS2_JOURNAL_ACCESS_WRITE);
1629                 if (ret) {
1630                         mlog_errno(ret);
1631                         goto out;
1632                 }
1633         }
1634
1635         /*
1636          * Set value in local, include set tree root in local.
1637          * This is the first step for value size >INLINE_SIZE.
1638          */
1639         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1640
1641         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1642                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1643                 if (ret < 0) {
1644                         mlog_errno(ret);
1645                         goto out;
1646                 }
1647         }
1648
1649         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1650             (flag & OCFS2_INLINE_XATTR_FL)) {
1651                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1652                 unsigned int xattrsize = osb->s_xattr_inline_size;
1653
1654                 /*
1655                  * Adjust extent record count or inline data size
1656                  * to reserve space for extended attribute.
1657                  */
1658                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1659                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1660                         le16_add_cpu(&idata->id_count, -xattrsize);
1661                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1662                         struct ocfs2_extent_list *el = &di->id2.i_list;
1663                         le16_add_cpu(&el->l_count, -(xattrsize /
1664                                         sizeof(struct ocfs2_extent_rec)));
1665                 }
1666                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1667         }
1668         /* Update xattr flag */
1669         spin_lock(&oi->ip_lock);
1670         oi->ip_dyn_features |= flag;
1671         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1672         spin_unlock(&oi->ip_lock);
1673
1674         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1675         if (ret < 0)
1676                 mlog_errno(ret);
1677
1678         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1679                 /*
1680                  * Set value outside in B tree.
1681                  * This is the second step for value size > INLINE_SIZE.
1682                  */
1683                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1684                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1685                                                     &vb, offs);
1686                 if (ret < 0) {
1687                         int ret2;
1688
1689                         mlog_errno(ret);
1690                         /*
1691                          * If set value outside failed, we have to clean
1692                          * the junk tree root we have already set in local.
1693                          */
1694                         ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1695                                                    xi, xs, &vb, offs);
1696                         if (ret2 < 0)
1697                                 mlog_errno(ret2);
1698                 }
1699         }
1700 out:
1701         return ret;
1702 }
1703
1704 static int ocfs2_remove_value_outside(struct inode*inode,
1705                                       struct ocfs2_xattr_value_buf *vb,
1706                                       struct ocfs2_xattr_header *header)
1707 {
1708         int ret = 0, i;
1709         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1710         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1711
1712         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1713
1714         ctxt.handle = ocfs2_start_trans(osb,
1715                                         ocfs2_remove_extent_credits(osb->sb));
1716         if (IS_ERR(ctxt.handle)) {
1717                 ret = PTR_ERR(ctxt.handle);
1718                 mlog_errno(ret);
1719                 goto out;
1720         }
1721
1722         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1723                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1724
1725                 if (!ocfs2_xattr_is_local(entry)) {
1726                         void *val;
1727
1728                         val = (void *)header +
1729                                 le16_to_cpu(entry->xe_name_offset);
1730                         vb->vb_xv = (struct ocfs2_xattr_value_root *)
1731                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1732                         ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1733                         if (ret < 0) {
1734                                 mlog_errno(ret);
1735                                 break;
1736                         }
1737                 }
1738         }
1739
1740         ocfs2_commit_trans(osb, ctxt.handle);
1741         ocfs2_schedule_truncate_log_flush(osb, 1);
1742         ocfs2_run_deallocs(osb, &ctxt.dealloc);
1743 out:
1744         return ret;
1745 }
1746
1747 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1748                                     struct buffer_head *di_bh)
1749 {
1750
1751         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1752         struct ocfs2_xattr_header *header;
1753         int ret;
1754         struct ocfs2_xattr_value_buf vb = {
1755                 .vb_bh = di_bh,
1756                 .vb_access = ocfs2_journal_access_di,
1757         };
1758
1759         header = (struct ocfs2_xattr_header *)
1760                  ((void *)di + inode->i_sb->s_blocksize -
1761                  le16_to_cpu(di->i_xattr_inline_size));
1762
1763         ret = ocfs2_remove_value_outside(inode, &vb, header);
1764
1765         return ret;
1766 }
1767
1768 static int ocfs2_xattr_block_remove(struct inode *inode,
1769                                     struct buffer_head *blk_bh)
1770 {
1771         struct ocfs2_xattr_block *xb;
1772         int ret = 0;
1773         struct ocfs2_xattr_value_buf vb = {
1774                 .vb_bh = blk_bh,
1775                 .vb_access = ocfs2_journal_access_xb,
1776         };
1777
1778         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1779         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1780                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1781                 ret = ocfs2_remove_value_outside(inode, &vb, header);
1782         } else
1783                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1784
1785         return ret;
1786 }
1787
1788 static int ocfs2_xattr_free_block(struct inode *inode,
1789                                   u64 block)
1790 {
1791         struct inode *xb_alloc_inode;
1792         struct buffer_head *xb_alloc_bh = NULL;
1793         struct buffer_head *blk_bh = NULL;
1794         struct ocfs2_xattr_block *xb;
1795         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1796         handle_t *handle;
1797         int ret = 0;
1798         u64 blk, bg_blkno;
1799         u16 bit;
1800
1801         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1802         if (ret < 0) {
1803                 mlog_errno(ret);
1804                 goto out;
1805         }
1806
1807         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1808         if (ret < 0) {
1809                 mlog_errno(ret);
1810                 goto out;
1811         }
1812
1813         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1814         blk = le64_to_cpu(xb->xb_blkno);
1815         bit = le16_to_cpu(xb->xb_suballoc_bit);
1816         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1817
1818         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1819                                 EXTENT_ALLOC_SYSTEM_INODE,
1820                                 le16_to_cpu(xb->xb_suballoc_slot));
1821         if (!xb_alloc_inode) {
1822                 ret = -ENOMEM;
1823                 mlog_errno(ret);
1824                 goto out;
1825         }
1826         mutex_lock(&xb_alloc_inode->i_mutex);
1827
1828         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1829         if (ret < 0) {
1830                 mlog_errno(ret);
1831                 goto out_mutex;
1832         }
1833
1834         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1835         if (IS_ERR(handle)) {
1836                 ret = PTR_ERR(handle);
1837                 mlog_errno(ret);
1838                 goto out_unlock;
1839         }
1840
1841         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1842                                        bit, bg_blkno, 1);
1843         if (ret < 0)
1844                 mlog_errno(ret);
1845
1846         ocfs2_commit_trans(osb, handle);
1847 out_unlock:
1848         ocfs2_inode_unlock(xb_alloc_inode, 1);
1849         brelse(xb_alloc_bh);
1850 out_mutex:
1851         mutex_unlock(&xb_alloc_inode->i_mutex);
1852         iput(xb_alloc_inode);
1853 out:
1854         brelse(blk_bh);
1855         return ret;
1856 }
1857
1858 /*
1859  * ocfs2_xattr_remove()
1860  *
1861  * Free extended attribute resources associated with this inode.
1862  */
1863 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1864 {
1865         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1866         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1867         handle_t *handle;
1868         int ret;
1869
1870         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1871                 return 0;
1872
1873         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1874                 return 0;
1875
1876         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1877                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1878                 if (ret < 0) {
1879                         mlog_errno(ret);
1880                         goto out;
1881                 }
1882         }
1883
1884         if (di->i_xattr_loc) {
1885                 ret = ocfs2_xattr_free_block(inode,
1886                                              le64_to_cpu(di->i_xattr_loc));
1887                 if (ret < 0) {
1888                         mlog_errno(ret);
1889                         goto out;
1890                 }
1891         }
1892
1893         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1894                                    OCFS2_INODE_UPDATE_CREDITS);
1895         if (IS_ERR(handle)) {
1896                 ret = PTR_ERR(handle);
1897                 mlog_errno(ret);
1898                 goto out;
1899         }
1900         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1901                                       OCFS2_JOURNAL_ACCESS_WRITE);
1902         if (ret) {
1903                 mlog_errno(ret);
1904                 goto out_commit;
1905         }
1906
1907         di->i_xattr_loc = 0;
1908
1909         spin_lock(&oi->ip_lock);
1910         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1911         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1912         spin_unlock(&oi->ip_lock);
1913
1914         ret = ocfs2_journal_dirty(handle, di_bh);
1915         if (ret < 0)
1916                 mlog_errno(ret);
1917 out_commit:
1918         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1919 out:
1920         return ret;
1921 }
1922
1923 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1924                                         struct ocfs2_dinode *di)
1925 {
1926         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1927         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1928         int free;
1929
1930         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1931                 return 0;
1932
1933         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1934                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1935                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1936         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1937                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1938                         le64_to_cpu(di->i_size);
1939         } else {
1940                 struct ocfs2_extent_list *el = &di->id2.i_list;
1941                 free = (le16_to_cpu(el->l_count) -
1942                         le16_to_cpu(el->l_next_free_rec)) *
1943                         sizeof(struct ocfs2_extent_rec);
1944         }
1945         if (free >= xattrsize)
1946                 return 1;
1947
1948         return 0;
1949 }
1950
1951 /*
1952  * ocfs2_xattr_ibody_find()
1953  *
1954  * Find extended attribute in inode block and
1955  * fill search info into struct ocfs2_xattr_search.
1956  */
1957 static int ocfs2_xattr_ibody_find(struct inode *inode,
1958                                   int name_index,
1959                                   const char *name,
1960                                   struct ocfs2_xattr_search *xs)
1961 {
1962         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1963         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1964         int ret;
1965         int has_space = 0;
1966
1967         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1968                 return 0;
1969
1970         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1971                 down_read(&oi->ip_alloc_sem);
1972                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1973                 up_read(&oi->ip_alloc_sem);
1974                 if (!has_space)
1975                         return 0;
1976         }
1977
1978         xs->xattr_bh = xs->inode_bh;
1979         xs->end = (void *)di + inode->i_sb->s_blocksize;
1980         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1981                 xs->header = (struct ocfs2_xattr_header *)
1982                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1983         else
1984                 xs->header = (struct ocfs2_xattr_header *)
1985                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1986         xs->base = (void *)xs->header;
1987         xs->here = xs->header->xh_entries;
1988
1989         /* Find the named attribute. */
1990         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1991                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1992                 if (ret && ret != -ENODATA)
1993                         return ret;
1994                 xs->not_found = ret;
1995         }
1996
1997         return 0;
1998 }
1999
2000 /*
2001  * ocfs2_xattr_ibody_set()
2002  *
2003  * Set, replace or remove an extended attribute into inode block.
2004  *
2005  */
2006 static int ocfs2_xattr_ibody_set(struct inode *inode,
2007                                  struct ocfs2_xattr_info *xi,
2008                                  struct ocfs2_xattr_search *xs,
2009                                  struct ocfs2_xattr_set_ctxt *ctxt)
2010 {
2011         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2012         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2013         int ret;
2014
2015         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2016                 return -ENOSPC;
2017
2018         down_write(&oi->ip_alloc_sem);
2019         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2020                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2021                         ret = -ENOSPC;
2022                         goto out;
2023                 }
2024         }
2025
2026         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2027                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2028 out:
2029         up_write(&oi->ip_alloc_sem);
2030
2031         return ret;
2032 }
2033
2034 /*
2035  * ocfs2_xattr_block_find()
2036  *
2037  * Find extended attribute in external block and
2038  * fill search info into struct ocfs2_xattr_search.
2039  */
2040 static int ocfs2_xattr_block_find(struct inode *inode,
2041                                   int name_index,
2042                                   const char *name,
2043                                   struct ocfs2_xattr_search *xs)
2044 {
2045         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2046         struct buffer_head *blk_bh = NULL;
2047         struct ocfs2_xattr_block *xb;
2048         int ret = 0;
2049
2050         if (!di->i_xattr_loc)
2051                 return ret;
2052
2053         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2054                                      &blk_bh);
2055         if (ret < 0) {
2056                 mlog_errno(ret);
2057                 return ret;
2058         }
2059
2060         xs->xattr_bh = blk_bh;
2061         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2062
2063         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2064                 xs->header = &xb->xb_attrs.xb_header;
2065                 xs->base = (void *)xs->header;
2066                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2067                 xs->here = xs->header->xh_entries;
2068
2069                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2070         } else
2071                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2072                                                    name_index,
2073                                                    name, xs);
2074
2075         if (ret && ret != -ENODATA) {
2076                 xs->xattr_bh = NULL;
2077                 goto cleanup;
2078         }
2079         xs->not_found = ret;
2080         return 0;
2081 cleanup:
2082         brelse(blk_bh);
2083
2084         return ret;
2085 }
2086
2087 /*
2088  * ocfs2_xattr_block_set()
2089  *
2090  * Set, replace or remove an extended attribute into external block.
2091  *
2092  */
2093 static int ocfs2_xattr_block_set(struct inode *inode,
2094                                  struct ocfs2_xattr_info *xi,
2095                                  struct ocfs2_xattr_search *xs,
2096                                  struct ocfs2_xattr_set_ctxt *ctxt)
2097 {
2098         struct buffer_head *new_bh = NULL;
2099         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2100         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
2101         handle_t *handle = ctxt->handle;
2102         struct ocfs2_xattr_block *xblk = NULL;
2103         u16 suballoc_bit_start;
2104         u32 num_got;
2105         u64 first_blkno;
2106         int ret;
2107
2108         if (!xs->xattr_bh) {
2109                 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2110                                               xs->inode_bh,
2111                                               OCFS2_JOURNAL_ACCESS_CREATE);
2112                 if (ret < 0) {
2113                         mlog_errno(ret);
2114                         goto end;
2115                 }
2116
2117                 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2118                                            &suballoc_bit_start, &num_got,
2119                                            &first_blkno);
2120                 if (ret < 0) {
2121                         mlog_errno(ret);
2122                         goto end;
2123                 }
2124
2125                 new_bh = sb_getblk(inode->i_sb, first_blkno);
2126                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2127
2128                 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode),
2129                                               new_bh,
2130                                               OCFS2_JOURNAL_ACCESS_CREATE);
2131                 if (ret < 0) {
2132                         mlog_errno(ret);
2133                         goto end;
2134                 }
2135
2136                 /* Initialize ocfs2_xattr_block */
2137                 xs->xattr_bh = new_bh;
2138                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2139                 memset(xblk, 0, inode->i_sb->s_blocksize);
2140                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2141                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2142                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2143                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2144                 xblk->xb_blkno = cpu_to_le64(first_blkno);
2145
2146                 xs->header = &xblk->xb_attrs.xb_header;
2147                 xs->base = (void *)xs->header;
2148                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2149                 xs->here = xs->header->xh_entries;
2150
2151                 ret = ocfs2_journal_dirty(handle, new_bh);
2152                 if (ret < 0) {
2153                         mlog_errno(ret);
2154                         goto end;
2155                 }
2156                 di->i_xattr_loc = cpu_to_le64(first_blkno);
2157                 ocfs2_journal_dirty(handle, xs->inode_bh);
2158         } else
2159                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2160
2161         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2162                 /* Set extended attribute into external block */
2163                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2164                                             OCFS2_HAS_XATTR_FL);
2165                 if (!ret || ret != -ENOSPC)
2166                         goto end;
2167
2168                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2169                 if (ret)
2170                         goto end;
2171         }
2172
2173         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2174
2175 end:
2176
2177         return ret;
2178 }
2179
2180 /* Check whether the new xattr can be inserted into the inode. */
2181 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2182                                        struct ocfs2_xattr_info *xi,
2183                                        struct ocfs2_xattr_search *xs)
2184 {
2185         u64 value_size;
2186         struct ocfs2_xattr_entry *last;
2187         int free, i;
2188         size_t min_offs = xs->end - xs->base;
2189
2190         if (!xs->header)
2191                 return 0;
2192
2193         last = xs->header->xh_entries;
2194
2195         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2196                 size_t offs = le16_to_cpu(last->xe_name_offset);
2197                 if (offs < min_offs)
2198                         min_offs = offs;
2199                 last += 1;
2200         }
2201
2202         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2203         if (free < 0)
2204                 return 0;
2205
2206         BUG_ON(!xs->not_found);
2207
2208         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2209                 value_size = OCFS2_XATTR_ROOT_SIZE;
2210         else
2211                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2212
2213         if (free >= sizeof(struct ocfs2_xattr_entry) +
2214                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2215                 return 1;
2216
2217         return 0;
2218 }
2219
2220 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2221                                      struct ocfs2_dinode *di,
2222                                      struct ocfs2_xattr_info *xi,
2223                                      struct ocfs2_xattr_search *xis,
2224                                      struct ocfs2_xattr_search *xbs,
2225                                      int *clusters_need,
2226                                      int *meta_need,
2227                                      int *credits_need)
2228 {
2229         int ret = 0, old_in_xb = 0;
2230         int clusters_add = 0, meta_add = 0, credits = 0;
2231         struct buffer_head *bh = NULL;
2232         struct ocfs2_xattr_block *xb = NULL;
2233         struct ocfs2_xattr_entry *xe = NULL;
2234         struct ocfs2_xattr_value_root *xv = NULL;
2235         char *base = NULL;
2236         int name_offset, name_len = 0;
2237         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2238                                                     xi->value_len);
2239         u64 value_size;
2240
2241         /*
2242          * Calculate the clusters we need to write.
2243          * No matter whether we replace an old one or add a new one,
2244          * we need this for writing.
2245          */
2246         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2247                 credits += new_clusters *
2248                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
2249
2250         if (xis->not_found && xbs->not_found) {
2251                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2252
2253                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2254                         clusters_add += new_clusters;
2255                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2256                                                         &def_xv.xv.xr_list,
2257                                                         new_clusters);
2258                 }
2259
2260                 goto meta_guess;
2261         }
2262
2263         if (!xis->not_found) {
2264                 xe = xis->here;
2265                 name_offset = le16_to_cpu(xe->xe_name_offset);
2266                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2267                 base = xis->base;
2268                 credits += OCFS2_INODE_UPDATE_CREDITS;
2269         } else {
2270                 int i, block_off = 0;
2271                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2272                 xe = xbs->here;
2273                 name_offset = le16_to_cpu(xe->xe_name_offset);
2274                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2275                 i = xbs->here - xbs->header->xh_entries;
2276                 old_in_xb = 1;
2277
2278                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2279                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2280                                                         bucket_xh(xbs->bucket),
2281                                                         i, &block_off,
2282                                                         &name_offset);
2283                         base = bucket_block(xbs->bucket, block_off);
2284                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2285                 } else {
2286                         base = xbs->base;
2287                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2288                 }
2289         }
2290
2291         /*
2292          * delete a xattr doesn't need metadata and cluster allocation.
2293          * so just calculate the credits and return.
2294          *
2295          * The credits for removing the value tree will be extended
2296          * by ocfs2_remove_extent itself.
2297          */
2298         if (!xi->value) {
2299                 if (!ocfs2_xattr_is_local(xe))
2300                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2301
2302                 goto out;
2303         }
2304
2305         /* do cluster allocation guess first. */
2306         value_size = le64_to_cpu(xe->xe_value_size);
2307
2308         if (old_in_xb) {
2309                 /*
2310                  * In xattr set, we always try to set the xe in inode first,
2311                  * so if it can be inserted into inode successfully, the old
2312                  * one will be removed from the xattr block, and this xattr
2313                  * will be inserted into inode as a new xattr in inode.
2314                  */
2315                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2316                         clusters_add += new_clusters;
2317                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
2318                                     OCFS2_INODE_UPDATE_CREDITS;
2319                         if (!ocfs2_xattr_is_local(xe))
2320                                 credits += ocfs2_calc_extend_credits(
2321                                                         inode->i_sb,
2322                                                         &def_xv.xv.xr_list,
2323                                                         new_clusters);
2324                         goto out;
2325                 }
2326         }
2327
2328         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2329                 /* the new values will be stored outside. */
2330                 u32 old_clusters = 0;
2331
2332                 if (!ocfs2_xattr_is_local(xe)) {
2333                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2334                                                                  value_size);
2335                         xv = (struct ocfs2_xattr_value_root *)
2336                              (base + name_offset + name_len);
2337                         value_size = OCFS2_XATTR_ROOT_SIZE;
2338                 } else
2339                         xv = &def_xv.xv;
2340
2341                 if (old_clusters >= new_clusters) {
2342                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2343                         goto out;
2344                 } else {
2345                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2346                         clusters_add += new_clusters - old_clusters;
2347                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2348                                                              &xv->xr_list,
2349                                                              new_clusters -
2350                                                              old_clusters);
2351                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2352                                 goto out;
2353                 }
2354         } else {
2355                 /*
2356                  * Now the new value will be stored inside. So if the new
2357                  * value is smaller than the size of value root or the old
2358                  * value, we don't need any allocation, otherwise we have
2359                  * to guess metadata allocation.
2360                  */
2361                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2362                     (!ocfs2_xattr_is_local(xe) &&
2363                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2364                         goto out;
2365         }
2366
2367 meta_guess:
2368         /* calculate metadata allocation. */
2369         if (di->i_xattr_loc) {
2370                 if (!xbs->xattr_bh) {
2371                         ret = ocfs2_read_xattr_block(inode,
2372                                                      le64_to_cpu(di->i_xattr_loc),
2373                                                      &bh);
2374                         if (ret) {
2375                                 mlog_errno(ret);
2376                                 goto out;
2377                         }
2378
2379                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2380                 } else
2381                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2382
2383                 /*
2384                  * If there is already an xattr tree, good, we can calculate
2385                  * like other b-trees. Otherwise we may have the chance of
2386                  * create a tree, the credit calculation is borrowed from
2387                  * ocfs2_calc_extend_credits with root_el = NULL. And the
2388                  * new tree will be cluster based, so no meta is needed.
2389                  */
2390                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2391                         struct ocfs2_extent_list *el =
2392                                  &xb->xb_attrs.xb_root.xt_list;
2393                         meta_add += ocfs2_extend_meta_needed(el);
2394                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2395                                                              el, 1);
2396                 } else
2397                         credits += OCFS2_SUBALLOC_ALLOC + 1;
2398
2399                 /*
2400                  * This cluster will be used either for new bucket or for
2401                  * new xattr block.
2402                  * If the cluster size is the same as the bucket size, one
2403                  * more is needed since we may need to extend the bucket
2404                  * also.
2405                  */
2406                 clusters_add += 1;
2407                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2408                 if (OCFS2_XATTR_BUCKET_SIZE ==
2409                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2410                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2411                         clusters_add += 1;
2412                 }
2413         } else {
2414                 meta_add += 1;
2415                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2416         }
2417 out:
2418         if (clusters_need)
2419                 *clusters_need = clusters_add;
2420         if (meta_need)
2421                 *meta_need = meta_add;
2422         if (credits_need)
2423                 *credits_need = credits;
2424         brelse(bh);
2425         return ret;
2426 }
2427
2428 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2429                                      struct ocfs2_dinode *di,
2430                                      struct ocfs2_xattr_info *xi,
2431                                      struct ocfs2_xattr_search *xis,
2432                                      struct ocfs2_xattr_search *xbs,
2433                                      struct ocfs2_xattr_set_ctxt *ctxt,
2434                                      int *credits)
2435 {
2436         int clusters_add, meta_add, ret;
2437         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2438
2439         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2440
2441         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2442
2443         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2444                                         &clusters_add, &meta_add, credits);
2445         if (ret) {
2446                 mlog_errno(ret);
2447                 return ret;
2448         }
2449
2450         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2451              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2452
2453         if (meta_add) {
2454                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2455                                                         &ctxt->meta_ac);
2456                 if (ret) {
2457                         mlog_errno(ret);
2458                         goto out;
2459                 }
2460         }
2461
2462         if (clusters_add) {
2463                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2464                 if (ret)
2465                         mlog_errno(ret);
2466         }
2467 out:
2468         if (ret) {
2469                 if (ctxt->meta_ac) {
2470                         ocfs2_free_alloc_context(ctxt->meta_ac);
2471                         ctxt->meta_ac = NULL;
2472                 }
2473
2474                 /*
2475                  * We cannot have an error and a non null ctxt->data_ac.
2476                  */
2477         }
2478
2479         return ret;
2480 }
2481
2482 static int __ocfs2_xattr_set_handle(struct inode *inode,
2483                                     struct ocfs2_dinode *di,
2484                                     struct ocfs2_xattr_info *xi,
2485                                     struct ocfs2_xattr_search *xis,
2486                                     struct ocfs2_xattr_search *xbs,
2487                                     struct ocfs2_xattr_set_ctxt *ctxt)
2488 {
2489         int ret = 0, credits, old_found;
2490
2491         if (!xi->value) {
2492                 /* Remove existing extended attribute */
2493                 if (!xis->not_found)
2494                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2495                 else if (!xbs->not_found)
2496                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2497         } else {
2498                 /* We always try to set extended attribute into inode first*/
2499                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2500                 if (!ret && !xbs->not_found) {
2501                         /*
2502                          * If succeed and that extended attribute existing in
2503                          * external block, then we will remove it.
2504                          */
2505                         xi->value = NULL;
2506                         xi->value_len = 0;
2507
2508                         old_found = xis->not_found;
2509                         xis->not_found = -ENODATA;
2510                         ret = ocfs2_calc_xattr_set_need(inode,
2511                                                         di,
2512                                                         xi,
2513                                                         xis,
2514                                                         xbs,
2515                                                         NULL,
2516                                                         NULL,
2517                                                         &credits);
2518                         xis->not_found = old_found;
2519                         if (ret) {
2520                                 mlog_errno(ret);
2521                                 goto out;
2522                         }
2523
2524                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2525                                         ctxt->handle->h_buffer_credits);
2526                         if (ret) {
2527                                 mlog_errno(ret);
2528                                 goto out;
2529                         }
2530                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2531                 } else if (ret == -ENOSPC) {
2532                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2533                                 ret = ocfs2_xattr_block_find(inode,
2534                                                              xi->name_index,
2535                                                              xi->name, xbs);
2536                                 if (ret)
2537                                         goto out;
2538
2539                                 old_found = xis->not_found;
2540                                 xis->not_found = -ENODATA;
2541                                 ret = ocfs2_calc_xattr_set_need(inode,
2542                                                                 di,
2543                                                                 xi,
2544                                                                 xis,
2545                                                                 xbs,
2546                                                                 NULL,
2547                                                                 NULL,
2548                                                                 &credits);
2549                                 xis->not_found = old_found;
2550                                 if (ret) {
2551                                         mlog_errno(ret);
2552                                         goto out;
2553                                 }
2554
2555                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2556                                         ctxt->handle->h_buffer_credits);
2557                                 if (ret) {
2558                                         mlog_errno(ret);
2559                                         goto out;
2560                                 }
2561                         }
2562                         /*
2563                          * If no space in inode, we will set extended attribute
2564                          * into external block.
2565                          */
2566                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2567                         if (ret)
2568                                 goto out;
2569                         if (!xis->not_found) {
2570                                 /*
2571                                  * If succeed and that extended attribute
2572                                  * existing in inode, we will remove it.
2573                                  */
2574                                 xi->value = NULL;
2575                                 xi->value_len = 0;
2576                                 xbs->not_found = -ENODATA;
2577                                 ret = ocfs2_calc_xattr_set_need(inode,
2578                                                                 di,
2579                                                                 xi,
2580                                                                 xis,
2581                                                                 xbs,
2582                                                                 NULL,
2583                                                                 NULL,
2584                                                                 &credits);
2585                                 if (ret) {
2586                                         mlog_errno(ret);
2587                                         goto out;
2588                                 }
2589
2590                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2591                                                 ctxt->handle->h_buffer_credits);
2592                                 if (ret) {
2593                                         mlog_errno(ret);
2594                                         goto out;
2595                                 }
2596                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2597                                                             xis, ctxt);
2598                         }
2599                 }
2600         }
2601
2602         if (!ret) {
2603                 /* Update inode ctime. */
2604                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2605                                               xis->inode_bh,
2606                                               OCFS2_JOURNAL_ACCESS_WRITE);
2607                 if (ret) {
2608                         mlog_errno(ret);
2609                         goto out;
2610                 }
2611
2612                 inode->i_ctime = CURRENT_TIME;
2613                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2614                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2615                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2616         }
2617 out:
2618         return ret;
2619 }
2620
2621 /*
2622  * This function only called duing creating inode
2623  * for init security/acl xattrs of the new inode.
2624  * All transanction credits have been reserved in mknod.
2625  */
2626 int ocfs2_xattr_set_handle(handle_t *handle,
2627                            struct inode *inode,
2628                            struct buffer_head *di_bh,
2629                            int name_index,
2630                            const char *name,
2631                            const void *value,
2632                            size_t value_len,
2633                            int flags,
2634                            struct ocfs2_alloc_context *meta_ac,
2635                            struct ocfs2_alloc_context *data_ac)
2636 {
2637         struct ocfs2_dinode *di;
2638         int ret;
2639
2640         struct ocfs2_xattr_info xi = {
2641                 .name_index = name_index,
2642                 .name = name,
2643                 .value = value,
2644                 .value_len = value_len,
2645         };
2646
2647         struct ocfs2_xattr_search xis = {
2648                 .not_found = -ENODATA,
2649         };
2650
2651         struct ocfs2_xattr_search xbs = {
2652                 .not_found = -ENODATA,
2653         };
2654
2655         struct ocfs2_xattr_set_ctxt ctxt = {
2656                 .handle = handle,
2657                 .meta_ac = meta_ac,
2658                 .data_ac = data_ac,
2659         };
2660
2661         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2662                 return -EOPNOTSUPP;
2663
2664         /*
2665          * In extreme situation, may need xattr bucket when
2666          * block size is too small. And we have already reserved
2667          * the credits for bucket in mknod.
2668          */
2669         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
2670                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2671                 if (!xbs.bucket) {
2672                         mlog_errno(-ENOMEM);
2673                         return -ENOMEM;
2674                 }
2675         }
2676
2677         xis.inode_bh = xbs.inode_bh = di_bh;
2678         di = (struct ocfs2_dinode *)di_bh->b_data;
2679
2680         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2681
2682         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2683         if (ret)
2684                 goto cleanup;
2685         if (xis.not_found) {
2686                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2687                 if (ret)
2688                         goto cleanup;
2689         }
2690
2691         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2692
2693 cleanup:
2694         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2695         brelse(xbs.xattr_bh);
2696         ocfs2_xattr_bucket_free(xbs.bucket);
2697
2698         return ret;
2699 }
2700
2701 /*
2702  * ocfs2_xattr_set()
2703  *
2704  * Set, replace or remove an extended attribute for this inode.
2705  * value is NULL to remove an existing extended attribute, else either
2706  * create or replace an extended attribute.
2707  */
2708 int ocfs2_xattr_set(struct inode *inode,
2709                     int name_index,
2710                     const char *name,
2711                     const void *value,
2712                     size_t value_len,
2713                     int flags)
2714 {
2715         struct buffer_head *di_bh = NULL;
2716         struct ocfs2_dinode *di;
2717         int ret, credits;
2718         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2719         struct inode *tl_inode = osb->osb_tl_inode;
2720         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2721
2722         struct ocfs2_xattr_info xi = {
2723                 .name_index = name_index,
2724                 .name = name,
2725                 .value = value,
2726                 .value_len = value_len,
2727         };
2728
2729         struct ocfs2_xattr_search xis = {
2730                 .not_found = -ENODATA,
2731         };
2732
2733         struct ocfs2_xattr_search xbs = {
2734                 .not_found = -ENODATA,
2735         };
2736
2737         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2738                 return -EOPNOTSUPP;
2739
2740         /*
2741          * Only xbs will be used on indexed trees.  xis doesn't need a
2742          * bucket.
2743          */
2744         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2745         if (!xbs.bucket) {
2746                 mlog_errno(-ENOMEM);
2747                 return -ENOMEM;
2748         }
2749
2750         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2751         if (ret < 0) {
2752                 mlog_errno(ret);
2753                 goto cleanup_nolock;
2754         }
2755         xis.inode_bh = xbs.inode_bh = di_bh;
2756         di = (struct ocfs2_dinode *)di_bh->b_data;
2757
2758         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2759         /*
2760          * Scan inode and external block to find the same name
2761          * extended attribute and collect search infomation.
2762          */
2763         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2764         if (ret)
2765                 goto cleanup;
2766         if (xis.not_found) {
2767                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2768                 if (ret)
2769                         goto cleanup;
2770         }
2771
2772         if (xis.not_found && xbs.not_found) {
2773                 ret = -ENODATA;
2774                 if (flags & XATTR_REPLACE)
2775                         goto cleanup;
2776                 ret = 0;
2777                 if (!value)
2778                         goto cleanup;
2779         } else {
2780                 ret = -EEXIST;
2781                 if (flags & XATTR_CREATE)
2782                         goto cleanup;
2783         }
2784
2785
2786         mutex_lock(&tl_inode->i_mutex);
2787
2788         if (ocfs2_truncate_log_needs_flush(osb)) {
2789                 ret = __ocfs2_flush_truncate_log(osb);
2790                 if (ret < 0) {
2791                         mutex_unlock(&tl_inode->i_mutex);
2792                         mlog_errno(ret);
2793                         goto cleanup;
2794                 }
2795         }
2796         mutex_unlock(&tl_inode->i_mutex);
2797
2798         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2799                                         &xbs, &ctxt, &credits);
2800         if (ret) {
2801                 mlog_errno(ret);
2802                 goto cleanup;
2803         }
2804
2805         /* we need to update inode's ctime field, so add credit for it. */
2806         credits += OCFS2_INODE_UPDATE_CREDITS;
2807         ctxt.handle = ocfs2_start_trans(osb, credits);
2808         if (IS_ERR(ctxt.handle)) {
2809                 ret = PTR_ERR(ctxt.handle);
2810                 mlog_errno(ret);
2811                 goto cleanup;
2812         }
2813
2814         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2815
2816         ocfs2_commit_trans(osb, ctxt.handle);
2817
2818         if (ctxt.data_ac)
2819                 ocfs2_free_alloc_context(ctxt.data_ac);
2820         if (ctxt.meta_ac)
2821                 ocfs2_free_alloc_context(ctxt.meta_ac);
2822         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2823                 ocfs2_schedule_truncate_log_flush(osb, 1);
2824         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2825 cleanup:
2826         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2827         ocfs2_inode_unlock(inode, 1);
2828 cleanup_nolock:
2829         brelse(di_bh);
2830         brelse(xbs.xattr_bh);
2831         ocfs2_xattr_bucket_free(xbs.bucket);
2832
2833         return ret;
2834 }
2835
2836 /*
2837  * Find the xattr extent rec which may contains name_hash.
2838  * e_cpos will be the first name hash of the xattr rec.
2839  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2840  */
2841 static int ocfs2_xattr_get_rec(struct inode *inode,
2842                                u32 name_hash,
2843                                u64 *p_blkno,
2844                                u32 *e_cpos,
2845                                u32 *num_clusters,
2846                                struct ocfs2_extent_list *el)
2847 {
2848         int ret = 0, i;
2849         struct buffer_head *eb_bh = NULL;
2850         struct ocfs2_extent_block *eb;
2851         struct ocfs2_extent_rec *rec = NULL;
2852         u64 e_blkno = 0;
2853
2854         if (el->l_tree_depth) {
2855                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
2856                                       &eb_bh);
2857                 if (ret) {
2858                         mlog_errno(ret);
2859                         goto out;
2860                 }
2861
2862                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2863                 el = &eb->h_list;
2864
2865                 if (el->l_tree_depth) {
2866                         ocfs2_error(inode->i_sb,
2867                                     "Inode %lu has non zero tree depth in "
2868                                     "xattr tree block %llu\n", inode->i_ino,
2869                                     (unsigned long long)eb_bh->b_blocknr);
2870                         ret = -EROFS;
2871                         goto out;
2872                 }
2873         }
2874
2875         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2876                 rec = &el->l_recs[i];
2877
2878                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2879                         e_blkno = le64_to_cpu(rec->e_blkno);
2880                         break;
2881                 }
2882         }
2883
2884         if (!e_blkno) {
2885                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2886                             "record (%u, %u, 0) in xattr", inode->i_ino,
2887                             le32_to_cpu(rec->e_cpos),
2888                             ocfs2_rec_clusters(el, rec));
2889                 ret = -EROFS;
2890                 goto out;
2891         }
2892
2893         *p_blkno = le64_to_cpu(rec->e_blkno);
2894         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2895         if (e_cpos)
2896                 *e_cpos = le32_to_cpu(rec->e_cpos);
2897 out:
2898         brelse(eb_bh);
2899         return ret;
2900 }
2901
2902 typedef int (xattr_bucket_func)(struct inode *inode,
2903                                 struct ocfs2_xattr_bucket *bucket,
2904                                 void *para);
2905
2906 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2907                                    struct ocfs2_xattr_bucket *bucket,
2908                                    int name_index,
2909                                    const char *name,
2910                                    u32 name_hash,
2911                                    u16 *xe_index,
2912                                    int *found)
2913 {
2914         int i, ret = 0, cmp = 1, block_off, new_offset;
2915         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2916         size_t name_len = strlen(name);
2917         struct ocfs2_xattr_entry *xe = NULL;
2918         char *xe_name;
2919
2920         /*
2921          * We don't use binary search in the bucket because there
2922          * may be multiple entries with the same name hash.
2923          */
2924         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2925                 xe = &xh->xh_entries[i];
2926
2927                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2928                         continue;
2929                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2930                         break;
2931
2932                 cmp = name_index - ocfs2_xattr_get_type(xe);
2933                 if (!cmp)
2934                         cmp = name_len - xe->xe_name_len;
2935                 if (cmp)
2936                         continue;
2937
2938                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2939                                                         xh,
2940                                                         i,
2941                                                         &block_off,
2942                                                         &new_offset);
2943                 if (ret) {
2944                         mlog_errno(ret);
2945                         break;
2946                 }
2947
2948
2949                 xe_name = bucket_block(bucket, block_off) + new_offset;
2950                 if (!memcmp(name, xe_name, name_len)) {
2951                         *xe_index = i;
2952                         *found = 1;
2953                         ret = 0;
2954                         break;
2955                 }
2956         }
2957
2958         return ret;
2959 }
2960
2961 /*
2962  * Find the specified xattr entry in a series of buckets.
2963  * This series start from p_blkno and last for num_clusters.
2964  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2965  * the num of the valid buckets.
2966  *
2967  * Return the buffer_head this xattr should reside in. And if the xattr's
2968  * hash is in the gap of 2 buckets, return the lower bucket.
2969  */
2970 static int ocfs2_xattr_bucket_find(struct inode *inode,
2971                                    int name_index,
2972                                    const char *name,
2973                                    u32 name_hash,
2974                                    u64 p_blkno,
2975                                    u32 first_hash,
2976                                    u32 num_clusters,
2977                                    struct ocfs2_xattr_search *xs)
2978 {
2979         int ret, found = 0;
2980         struct ocfs2_xattr_header *xh = NULL;
2981         struct ocfs2_xattr_entry *xe = NULL;
2982         u16 index = 0;
2983         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2984         int low_bucket = 0, bucket, high_bucket;
2985         struct ocfs2_xattr_bucket *search;
2986         u32 last_hash;
2987         u64 blkno, lower_blkno = 0;
2988
2989         search = ocfs2_xattr_bucket_new(inode);
2990         if (!search) {
2991                 ret = -ENOMEM;
2992                 mlog_errno(ret);
2993                 goto out;
2994         }
2995
2996         ret = ocfs2_read_xattr_bucket(search, p_blkno);
2997         if (ret) {
2998                 mlog_errno(ret);
2999                 goto out;
3000         }
3001
3002         xh = bucket_xh(search);
3003         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3004         while (low_bucket <= high_bucket) {
3005                 ocfs2_xattr_bucket_relse(search);
3006
3007                 bucket = (low_bucket + high_bucket) / 2;
3008                 blkno = p_blkno + bucket * blk_per_bucket;
3009                 ret = ocfs2_read_xattr_bucket(search, blkno);
3010                 if (ret) {
3011                         mlog_errno(ret);
3012                         goto out;
3013                 }
3014
3015                 xh = bucket_xh(search);
3016                 xe = &xh->xh_entries[0];
3017                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3018                         high_bucket = bucket - 1;
3019                         continue;
3020                 }
3021
3022                 /*
3023                  * Check whether the hash of the last entry in our
3024                  * bucket is larger than the search one. for an empty
3025                  * bucket, the last one is also the first one.
3026                  */
3027                 if (xh->xh_count)
3028                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3029
3030                 last_hash = le32_to_cpu(xe->xe_name_hash);
3031
3032                 /* record lower_blkno which may be the insert place. */
3033                 lower_blkno = blkno;
3034
3035                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3036                         low_bucket = bucket + 1;
3037                         continue;
3038                 }
3039
3040                 /* the searched xattr should reside in this bucket if exists. */
3041                 ret = ocfs2_find_xe_in_bucket(inode, search,
3042                                               name_index, name, name_hash,
3043                                               &index, &found);
3044                 if (ret) {
3045                         mlog_errno(ret);
3046                         goto out;
3047                 }
3048                 break;
3049         }
3050
3051         /*
3052          * Record the bucket we have found.
3053          * When the xattr's hash value is in the gap of 2 buckets, we will
3054          * always set it to the previous bucket.
3055          */
3056         if (!lower_blkno)
3057                 lower_blkno = p_blkno;
3058
3059         /* This should be in cache - we just read it during the search */
3060         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3061         if (ret) {
3062                 mlog_errno(ret);
3063                 goto out;
3064         }
3065
3066         xs->header = bucket_xh(xs->bucket);
3067         xs->base = bucket_block(xs->bucket, 0);
3068         xs->end = xs->base + inode->i_sb->s_blocksize;
3069
3070         if (found) {
3071                 xs->here = &xs->header->xh_entries[index];
3072                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3073                      (unsigned long long)bucket_blkno(xs->bucket), index);
3074         } else
3075                 ret = -ENODATA;
3076
3077 out:
3078         ocfs2_xattr_bucket_free(search);
3079         return ret;
3080 }
3081
3082 static int ocfs2_xattr_index_block_find(struct inode *inode,
3083                                         struct buffer_head *root_bh,
3084                                         int name_index,
3085                                         const char *name,
3086                                         struct ocfs2_xattr_search *xs)
3087 {
3088         int ret;
3089         struct ocfs2_xattr_block *xb =
3090                         (struct ocfs2_xattr_block *)root_bh->b_data;
3091         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3092         struct ocfs2_extent_list *el = &xb_root->xt_list;
3093         u64 p_blkno = 0;
3094         u32 first_hash, num_clusters = 0;
3095         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3096
3097         if (le16_to_cpu(el->l_next_free_rec) == 0)
3098                 return -ENODATA;
3099
3100         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3101              name, name_hash, name_index);
3102
3103         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3104                                   &num_clusters, el);
3105         if (ret) {
3106                 mlog_errno(ret);
3107                 goto out;
3108         }
3109
3110         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3111
3112         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3113              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3114              first_hash);
3115
3116         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3117                                       p_blkno, first_hash, num_clusters, xs);
3118
3119 out:
3120         return ret;
3121 }
3122
3123 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3124                                        u64 blkno,
3125                                        u32 clusters,
3126                                        xattr_bucket_func *func,
3127                                        void *para)
3128 {
3129         int i, ret = 0;
3130         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3131         u32 num_buckets = clusters * bpc;
3132         struct ocfs2_xattr_bucket *bucket;
3133
3134         bucket = ocfs2_xattr_bucket_new(inode);
3135         if (!bucket) {
3136                 mlog_errno(-ENOMEM);
3137                 return -ENOMEM;
3138         }
3139
3140         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3141              clusters, (unsigned long long)blkno);
3142
3143         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3144                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3145                 if (ret) {
3146                         mlog_errno(ret);
3147                         break;
3148                 }
3149
3150                 /*
3151                  * The real bucket num in this series of blocks is stored
3152                  * in the 1st bucket.
3153                  */
3154                 if (i == 0)
3155                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3156
3157                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3158                      (unsigned long long)blkno,
3159                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3160                 if (func) {
3161                         ret = func(inode, bucket, para);
3162                         if (ret && ret != -ERANGE)
3163                                 mlog_errno(ret);
3164                         /* Fall through to bucket_relse() */
3165                 }
3166
3167                 ocfs2_xattr_bucket_relse(bucket);
3168                 if (ret)
3169                         break;
3170         }
3171
3172         ocfs2_xattr_bucket_free(bucket);
3173         return ret;
3174 }
3175
3176 struct ocfs2_xattr_tree_list {
3177         char *buffer;
3178         size_t buffer_size;
3179         size_t result;
3180 };
3181
3182 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3183                                              struct ocfs2_xattr_header *xh,
3184                                              int index,
3185                                              int *block_off,
3186                                              int *new_offset)
3187 {
3188         u16 name_offset;
3189
3190         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3191                 return -EINVAL;
3192
3193         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3194
3195         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3196         *new_offset = name_offset % inode->i_sb->s_blocksize;
3197
3198         return 0;
3199 }
3200
3201 static int ocfs2_list_xattr_bucket(struct inode *inode,
3202                                    struct ocfs2_xattr_bucket *bucket,
3203                                    void *para)
3204 {
3205         int ret = 0, type;
3206         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3207         int i, block_off, new_offset;
3208         const char *prefix, *name;
3209
3210         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3211                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3212                 type = ocfs2_xattr_get_type(entry);
3213                 prefix = ocfs2_xattr_prefix(type);
3214
3215                 if (prefix) {
3216                         ret = ocfs2_xattr_bucket_get_name_value(inode,
3217                                                                 bucket_xh(bucket),
3218                                                                 i,
3219                                                                 &block_off,
3220                                                                 &new_offset);
3221                         if (ret)
3222                                 break;
3223
3224                         name = (const char *)bucket_block(bucket, block_off) +
3225                                 new_offset;
3226                         ret = ocfs2_xattr_list_entry(xl->buffer,
3227                                                      xl->buffer_size,
3228                                                      &xl->result,
3229                                                      prefix, name,
3230                                                      entry->xe_name_len);
3231                         if (ret)
3232                                 break;
3233                 }
3234         }
3235
3236         return ret;
3237 }
3238
3239 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3240                                              struct ocfs2_xattr_tree_root *xt,
3241                                              char *buffer,
3242                                              size_t buffer_size)
3243 {
3244         struct ocfs2_extent_list *el = &xt->xt_list;
3245         int ret = 0;
3246         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3247         u64 p_blkno = 0;
3248         struct ocfs2_xattr_tree_list xl = {
3249                 .buffer = buffer,
3250                 .buffer_size = buffer_size,
3251                 .result = 0,
3252         };
3253
3254         if (le16_to_cpu(el->l_next_free_rec) == 0)
3255                 return 0;
3256
3257         while (name_hash > 0) {
3258                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3259                                           &e_cpos, &num_clusters, el);
3260                 if (ret) {
3261                         mlog_errno(ret);
3262                         goto out;
3263                 }
3264
3265                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3266                                                   ocfs2_list_xattr_bucket,
3267                                                   &xl);
3268                 if (ret) {
3269                         if (ret != -ERANGE)
3270                                 mlog_errno(ret);
3271                         goto out;
3272                 }
3273
3274                 if (e_cpos == 0)
3275                         break;
3276
3277                 name_hash = e_cpos - 1;
3278         }
3279
3280         ret = xl.result;
3281 out:
3282         return ret;
3283 }
3284
3285 static int cmp_xe(const void *a, const void *b)
3286 {
3287         const struct ocfs2_xattr_entry *l = a, *r = b;
3288         u32 l_hash = le32_to_cpu(l->xe_name_hash);
3289         u32 r_hash = le32_to_cpu(r->xe_name_hash);
3290
3291         if (l_hash > r_hash)
3292                 return 1;
3293         if (l_hash < r_hash)
3294                 return -1;
3295         return 0;
3296 }
3297
3298 static void swap_xe(void *a, void *b, int size)
3299 {
3300         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3301
3302         tmp = *l;
3303         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3304         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3305 }
3306
3307 /*
3308  * When the ocfs2_xattr_block is filled up, new bucket will be created
3309  * and all the xattr entries will be moved to the new bucket.
3310  * The header goes at the start of the bucket, and the names+values are
3311  * filled from the end.  This is why *target starts as the last buffer.
3312  * Note: we need to sort the entries since they are not saved in order
3313  * in the ocfs2_xattr_block.
3314  */
3315 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3316                                            struct buffer_head *xb_bh,
3317                                            struct ocfs2_xattr_bucket *bucket)
3318 {
3319         int i, blocksize = inode->i_sb->s_blocksize;
3320         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3321         u16 offset, size, off_change;
3322         struct ocfs2_xattr_entry *xe;
3323         struct ocfs2_xattr_block *xb =
3324                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
3325         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3326         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3327         u16 count = le16_to_cpu(xb_xh->xh_count);
3328         char *src = xb_bh->b_data;
3329         char *target = bucket_block(bucket, blks - 1);
3330
3331         mlog(0, "cp xattr from block %llu to bucket %llu\n",
3332              (unsigned long long)xb_bh->b_blocknr,
3333              (unsigned long long)bucket_blkno(bucket));
3334
3335         for (i = 0; i < blks; i++)
3336                 memset(bucket_block(bucket, i), 0, blocksize);
3337
3338         /*
3339          * Since the xe_name_offset is based on ocfs2_xattr_header,
3340          * there is a offset change corresponding to the change of
3341          * ocfs2_xattr_header's position.
3342          */
3343         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3344         xe = &xb_xh->xh_entries[count - 1];
3345         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3346         size = blocksize - offset;
3347
3348         /* copy all the names and values. */
3349         memcpy(target + offset, src + offset, size);
3350
3351         /* Init new header now. */
3352         xh->xh_count = xb_xh->xh_count;
3353         xh->xh_num_buckets = cpu_to_le16(1);
3354         xh->xh_name_value_len = cpu_to_le16(size);
3355         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3356
3357         /* copy all the entries. */
3358         target = bucket_block(bucket, 0);
3359         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3360         size = count * sizeof(struct ocfs2_xattr_entry);
3361         memcpy(target + offset, (char *)xb_xh + offset, size);
3362
3363         /* Change the xe offset for all the xe because of the move. */
3364         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3365                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3366         for (i = 0; i < count; i++)
3367                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3368
3369         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3370              offset, size, off_change);
3371
3372         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3373              cmp_xe, swap_xe);
3374 }
3375
3376 /*
3377  * After we move xattr from block to index btree, we have to
3378  * update ocfs2_xattr_search to the new xe and base.
3379  *
3380  * When the entry is in xattr block, xattr_bh indicates the storage place.
3381  * While if the entry is in index b-tree, "bucket" indicates the
3382  * real place of the xattr.
3383  */
3384 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3385                                             struct ocfs2_xattr_search *xs,
3386                                             struct buffer_head *old_bh)
3387 {
3388         char *buf = old_bh->b_data;
3389         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3390         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3391         int i;
3392
3393         xs->header = bucket_xh(xs->bucket);
3394         xs->base = bucket_block(xs->bucket, 0);
3395         xs->end = xs->base + inode->i_sb->s_blocksize;
3396
3397         if (xs->not_found)
3398                 return;
3399
3400         i = xs->here - old_xh->xh_entries;
3401         xs->here = &xs->header->xh_entries[i];
3402 }
3403
3404 static int ocfs2_xattr_create_index_block(struct inode *inode,
3405                                           struct ocfs2_xattr_search *xs,
3406                                           struct ocfs2_xattr_set_ctxt *ctxt)
3407 {
3408         int ret;
3409         u32 bit_off, len;
3410         u64 blkno;
3411         handle_t *handle = ctxt->handle;
3412         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3413         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3414         struct buffer_head *xb_bh = xs->xattr_bh;
3415         struct ocfs2_xattr_block *xb =
3416                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3417         struct ocfs2_xattr_tree_root *xr;
3418         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3419
3420         mlog(0, "create xattr index block for %llu\n",
3421              (unsigned long long)xb_bh->b_blocknr);
3422
3423         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3424         BUG_ON(!xs->bucket);
3425
3426         /*
3427          * XXX:
3428          * We can use this lock for now, and maybe move to a dedicated mutex
3429          * if performance becomes a problem later.
3430          */
3431         down_write(&oi->ip_alloc_sem);
3432
3433         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
3434                                       OCFS2_JOURNAL_ACCESS_WRITE);
3435         if (ret) {
3436                 mlog_errno(ret);
3437                 goto out;
3438         }
3439
3440         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3441                                      1, 1, &bit_off, &len);
3442         if (ret) {
3443                 mlog_errno(ret);
3444                 goto out;
3445         }
3446
3447         /*
3448          * The bucket may spread in many blocks, and
3449          * we will only touch the 1st block and the last block
3450          * in the whole bucket(one for entry and one for data).
3451          */
3452         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3453
3454         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3455              (unsigned long long)blkno);
3456
3457         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3458         if (ret) {
3459                 mlog_errno(ret);
3460                 goto out;
3461         }
3462
3463         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3464                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3465         if (ret) {
3466                 mlog_errno(ret);
3467                 goto out;
3468         }
3469
3470         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3471         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3472
3473         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3474
3475         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3476         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3477                offsetof(struct ocfs2_xattr_block, xb_attrs));
3478
3479         xr = &xb->xb_attrs.xb_root;
3480         xr->xt_clusters = cpu_to_le32(1);
3481         xr->xt_last_eb_blk = 0;
3482         xr->xt_list.l_tree_depth = 0;
3483         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3484         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3485
3486         xr->xt_list.l_recs[0].e_cpos = 0;
3487         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3488         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3489
3490         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3491
3492         ocfs2_journal_dirty(handle, xb_bh);
3493
3494 out:
3495         up_write(&oi->ip_alloc_sem);
3496
3497         return ret;
3498 }
3499
3500 static int cmp_xe_offset(const void *a, const void *b)
3501 {
3502         const struct ocfs2_xattr_entry *l = a, *r = b;
3503         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3504         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3505
3506         if (l_name_offset < r_name_offset)
3507                 return 1;
3508         if (l_name_offset > r_name_offset)
3509                 return -1;
3510         return 0;
3511 }
3512
3513 /*
3514  * defrag a xattr bucket if we find that the bucket has some
3515  * holes beteen name/value pairs.
3516  * We will move all the name/value pairs to the end of the bucket
3517  * so that we can spare some space for insertion.
3518  */
3519 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3520                                      handle_t *handle,
3521                                      struct ocfs2_xattr_bucket *bucket)
3522 {
3523         int ret, i;
3524         size_t end, offset, len, value_len;
3525         struct ocfs2_xattr_header *xh;
3526         char *entries, *buf, *bucket_buf = NULL;
3527         u64 blkno = bucket_blkno(bucket);
3528         u16 xh_free_start;
3529         size_t blocksize = inode->i_sb->s_blocksize;
3530         struct ocfs2_xattr_entry *xe;
3531
3532         /*
3533          * In order to make the operation more efficient and generic,
3534          * we copy all the blocks into a contiguous memory and do the
3535          * defragment there, so if anything is error, we will not touch
3536          * the real block.
3537          */
3538         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3539         if (!bucket_buf) {
3540                 ret = -EIO;
3541                 goto out;
3542         }
3543
3544         buf = bucket_buf;
3545         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3546                 memcpy(buf, bucket_block(bucket, i), blocksize);
3547
3548         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3549                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3550         if (ret < 0) {
3551                 mlog_errno(ret);
3552                 goto out;
3553         }
3554
3555         xh = (struct ocfs2_xattr_header *)bucket_buf;
3556         entries = (char *)xh->xh_entries;
3557         xh_free_start = le16_to_cpu(xh->xh_free_start);
3558
3559         mlog(0, "adjust xattr bucket in %llu, count = %u, "
3560              "xh_free_start = %u, xh_name_value_len = %u.\n",
3561              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3562              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3563
3564         /*
3565          * sort all the entries by their offset.
3566          * the largest will be the first, so that we can
3567          * move them to the end one by one.
3568          */
3569         sort(entries, le16_to_cpu(xh->xh_count),
3570              sizeof(struct ocfs2_xattr_entry),
3571              cmp_xe_offset, swap_xe);
3572
3573         /* Move all name/values to the end of the bucket. */
3574         xe = xh->xh_entries;
3575         end = OCFS2_XATTR_BUCKET_SIZE;
3576         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3577                 offset = le16_to_cpu(xe->xe_name_offset);
3578                 if (ocfs2_xattr_is_local(xe))
3579                         value_len = OCFS2_XATTR_SIZE(
3580                                         le64_to_cpu(xe->xe_value_size));
3581                 else
3582                         value_len = OCFS2_XATTR_ROOT_SIZE;
3583                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3584
3585                 /*
3586                  * We must make sure that the name/value pair
3587                  * exist in the same block. So adjust end to
3588                  * the previous block end if needed.
3589                  */
3590                 if (((end - len) / blocksize !=
3591                         (end - 1) / blocksize))
3592                         end = end - end % blocksize;
3593
3594                 if (end > offset + len) {
3595                         memmove(bucket_buf + end - len,
3596                                 bucket_buf + offset, len);
3597                         xe->xe_name_offset = cpu_to_le16(end - len);
3598                 }
3599
3600                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3601                                 "bucket %llu\n", (unsigned long long)blkno);
3602
3603                 end -= len;
3604         }
3605
3606         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3607                         "bucket %llu\n", (unsigned long long)blkno);
3608
3609         if (xh_free_start == end)
3610                 goto out;
3611
3612         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3613         xh->xh_free_start = cpu_to_le16(end);
3614
3615         /* sort the entries by their name_hash. */
3616         sort(entries, le16_to_cpu(xh->xh_count),
3617              sizeof(struct ocfs2_xattr_entry),
3618              cmp_xe, swap_xe);
3619
3620         buf = bucket_buf;
3621         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3622                 memcpy(bucket_block(bucket, i), buf, blocksize);
3623         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3624
3625 out:
3626         kfree(bucket_buf);
3627         return ret;
3628 }
3629
3630 /*
3631  * prev_blkno points to the start of an existing extent.  new_blkno
3632  * points to a newly allocated extent.  Because we know each of our
3633  * clusters contains more than bucket, we can easily split one cluster
3634  * at a bucket boundary.  So we take the last cluster of the existing
3635  * extent and split it down the middle.  We move the last half of the
3636  * buckets in the last cluster of the existing extent over to the new
3637  * extent.
3638  *
3639  * first_bh is the buffer at prev_blkno so we can update the existing
3640  * extent's bucket count.  header_bh is the bucket were we were hoping
3641  * to insert our xattr.  If the bucket move places the target in the new
3642  * extent, we'll update first_bh and header_bh after modifying the old
3643  * extent.
3644  *
3645  * first_hash will be set as the 1st xe's name_hash in the new extent.
3646  */
3647 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3648                                                handle_t *handle,
3649                                                struct ocfs2_xattr_bucket *first,
3650                                                struct ocfs2_xattr_bucket *target,
3651                                                u64 new_blkno,
3652                                                u32 num_clusters,
3653                                                u32 *first_hash)
3654 {
3655         int ret;
3656         struct super_block *sb = inode->i_sb;
3657         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3658         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3659         int to_move = num_buckets / 2;
3660         u64 src_blkno;
3661         u64 last_cluster_blkno = bucket_blkno(first) +
3662                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3663
3664         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3665         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3666
3667         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3668              (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3669
3670         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3671                                      last_cluster_blkno, new_blkno,
3672                                      to_move, first_hash);
3673         if (ret) {
3674                 mlog_errno(ret);
3675                 goto out;
3676         }
3677
3678         /* This is the first bucket that got moved */
3679         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3680
3681         /*
3682          * If the target bucket was part of the moved buckets, we need to
3683          * update first and target.
3684          */
3685         if (bucket_blkno(target) >= src_blkno) {
3686                 /* Find the block for the new target bucket */
3687                 src_blkno = new_blkno +
3688                         (bucket_blkno(target) - src_blkno);
3689
3690                 ocfs2_xattr_bucket_relse(first);
3691                 ocfs2_xattr_bucket_relse(target);
3692
3693                 /*
3694                  * These shouldn't fail - the buffers are in the
3695                  * journal from ocfs2_cp_xattr_bucket().
3696                  */
3697                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3698                 if (ret) {
3699                         mlog_errno(ret);
3700                         goto out;
3701                 }
3702                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3703                 if (ret)
3704                         mlog_errno(ret);
3705
3706         }
3707
3708 out:
3709         return ret;
3710 }
3711
3712 /*
3713  * Find the suitable pos when we divide a bucket into 2.
3714  * We have to make sure the xattrs with the same hash value exist
3715  * in the same bucket.
3716  *
3717  * If this ocfs2_xattr_header covers more than one hash value, find a
3718  * place where the hash value changes.  Try to find the most even split.
3719  * The most common case is that all entries have different hash values,
3720  * and the first check we make will find a place to split.
3721  */
3722 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3723 {
3724         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3725         int count = le16_to_cpu(xh->xh_count);
3726         int delta, middle = count / 2;
3727
3728         /*
3729          * We start at the middle.  Each step gets farther away in both
3730          * directions.  We therefore hit the change in hash value
3731          * nearest to the middle.  Note that this loop does not execute for
3732          * count < 2.
3733          */
3734         for (delta = 0; delta < middle; delta++) {
3735                 /* Let's check delta earlier than middle */
3736                 if (cmp_xe(&entries[middle - delta - 1],
3737                            &entries[middle - delta]))
3738                         return middle - delta;
3739
3740                 /* For even counts, don't walk off the end */
3741                 if ((middle + delta + 1) == count)
3742                         continue;
3743
3744                 /* Now try delta past middle */
3745                 if (cmp_xe(&entries[middle + delta],
3746                            &entries[middle + delta + 1]))
3747                         return middle + delta + 1;
3748         }
3749
3750         /* Every entry had the same hash */
3751         return count;
3752 }
3753
3754 /*
3755  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3756  * first_hash will record the 1st hash of the new bucket.
3757  *
3758  * Normally half of the xattrs will be moved.  But we have to make
3759  * sure that the xattrs with the same hash value are stored in the
3760  * same bucket. If all the xattrs in this bucket have the same hash
3761  * value, the new bucket will be initialized as an empty one and the
3762  * first_hash will be initialized as (hash_value+1).
3763  */
3764 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3765                                     handle_t *handle,
3766                                     u64 blk,
3767                                     u64 new_blk,
3768                                     u32 *first_hash,
3769                                     int new_bucket_head)
3770 {
3771         int ret, i;
3772         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3773         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3774         struct ocfs2_xattr_header *xh;
3775         struct ocfs2_xattr_entry *xe;
3776         int blocksize = inode->i_sb->s_blocksize;
3777
3778         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3779              (unsigned long long)blk, (unsigned long long)new_blk);
3780
3781         s_bucket = ocfs2_xattr_bucket_new(inode);
3782         t_bucket = ocfs2_xattr_bucket_new(inode);
3783         if (!s_bucket || !t_bucket) {
3784                 ret = -ENOMEM;
3785                 mlog_errno(ret);
3786                 goto out;
3787         }
3788
3789         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3790         if (ret) {
3791                 mlog_errno(ret);
3792                 goto out;
3793         }
3794
3795         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3796                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3797         if (ret) {
3798                 mlog_errno(ret);
3799                 goto out;
3800         }
3801
3802         /*
3803          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3804          * there's no need to read it.
3805          */
3806         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3807         if (ret) {
3808                 mlog_errno(ret);
3809                 goto out;
3810         }
3811
3812         /*
3813          * Hey, if we're overwriting t_bucket, what difference does
3814          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
3815          * same part of ocfs2_cp_xattr_bucket().
3816          */
3817         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3818                                                 new_bucket_head ?
3819                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3820                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3821         if (ret) {
3822                 mlog_errno(ret);
3823                 goto out;
3824         }
3825
3826         xh = bucket_xh(s_bucket);
3827         count = le16_to_cpu(xh->xh_count);
3828         start = ocfs2_xattr_find_divide_pos(xh);
3829
3830         if (start == count) {
3831                 xe = &xh->xh_entries[start-1];
3832
3833                 /*
3834                  * initialized a new empty bucket here.
3835                  * The hash value is set as one larger than
3836                  * that of the last entry in the previous bucket.
3837                  */
3838                 for (i = 0; i < t_bucket->bu_blocks; i++)
3839                         memset(bucket_block(t_bucket, i), 0, blocksize);
3840
3841                 xh = bucket_xh(t_bucket);
3842                 xh->xh_free_start = cpu_to_le16(blocksize);
3843                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3844                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3845
3846                 goto set_num_buckets;
3847         }
3848
3849         /* copy the whole bucket to the new first. */
3850         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3851
3852         /* update the new bucket. */
3853         xh = bucket_xh(t_bucket);
3854
3855         /*
3856          * Calculate the total name/value len and xh_free_start for
3857          * the old bucket first.
3858          */
3859         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3860         name_value_len = 0;
3861         for (i = 0; i < start; i++) {
3862                 xe = &xh->xh_entries[i];
3863                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3864                 if (ocfs2_xattr_is_local(xe))
3865                         xe_len +=
3866                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3867                 else
3868                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3869                 name_value_len += xe_len;
3870                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3871                         name_offset = le16_to_cpu(xe->xe_name_offset);
3872         }
3873
3874         /*
3875          * Now begin the modification to the new bucket.
3876          *
3877          * In the new bucket, We just move the xattr entry to the beginning
3878          * and don't touch the name/value. So there will be some holes in the
3879          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3880          * called.
3881          */
3882         xe = &xh->xh_entries[start];
3883         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3884         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3885              (int)((char *)xe - (char *)xh),
3886              (int)((char *)xh->xh_entries - (char *)xh));
3887         memmove((char *)xh->xh_entries, (char *)xe, len);
3888         xe = &xh->xh_entries[count - start];
3889         len = sizeof(struct ocfs2_xattr_entry) * start;
3890         memset((char *)xe, 0, len);
3891
3892         le16_add_cpu(&xh->xh_count, -start);
3893         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3894
3895         /* Calculate xh_free_start for the new bucket. */
3896         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3897         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3898                 xe = &xh->xh_entries[i];
3899                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3900                 if (ocfs2_xattr_is_local(xe))
3901                         xe_len +=
3902                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3903                 else
3904                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3905                 if (le16_to_cpu(xe->xe_name_offset) <
3906                     le16_to_cpu(xh->xh_free_start))
3907                         xh->xh_free_start = xe->xe_name_offset;
3908         }
3909
3910 set_num_buckets:
3911         /* set xh->xh_num_buckets for the new xh. */
3912         if (new_bucket_head)
3913                 xh->xh_num_buckets = cpu_to_le16(1);
3914         else
3915                 xh->xh_num_buckets = 0;
3916
3917         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3918
3919         /* store the first_hash of the new bucket. */
3920         if (first_hash)
3921                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3922
3923         /*
3924          * Now only update the 1st block of the old bucket.  If we
3925          * just added a new empty bucket, there is no need to modify
3926          * it.
3927          */
3928         if (start == count)
3929                 goto out;
3930
3931         xh = bucket_xh(s_bucket);
3932         memset(&xh->xh_entries[start], 0,
3933                sizeof(struct ocfs2_xattr_entry) * (count - start));
3934         xh->xh_count = cpu_to_le16(start);
3935         xh->xh_free_start = cpu_to_le16(name_offset);
3936         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3937
3938         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3939
3940 out:
3941         ocfs2_xattr_bucket_free(s_bucket);
3942         ocfs2_xattr_bucket_free(t_bucket);
3943
3944         return ret;
3945 }
3946
3947 /*
3948  * Copy xattr from one bucket to another bucket.
3949  *
3950  * The caller must make sure that the journal transaction
3951  * has enough space for journaling.
3952  */
3953 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3954                                  handle_t *handle,
3955                                  u64 s_blkno,
3956                                  u64 t_blkno,
3957                                  int t_is_new)
3958 {
3959         int ret;
3960         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3961
3962         BUG_ON(s_blkno == t_blkno);
3963
3964         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3965              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3966              t_is_new);
3967
3968         s_bucket = ocfs2_xattr_bucket_new(inode);
3969         t_bucket = ocfs2_xattr_bucket_new(inode);
3970         if (!s_bucket || !t_bucket) {
3971                 ret = -ENOMEM;
3972                 mlog_errno(ret);
3973                 goto out;
3974         }
3975
3976         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3977         if (ret)
3978                 goto out;
3979
3980         /*
3981          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3982          * there's no need to read it.
3983          */
3984         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3985         if (ret)
3986                 goto out;
3987
3988         /*
3989          * Hey, if we're overwriting t_bucket, what difference does
3990          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
3991          * cluster to fill, we came here from
3992          * ocfs2_mv_xattr_buckets(), and it is really new -
3993          * ACCESS_CREATE is required.  But we also might have moved data
3994          * out of t_bucket before extending back into it.
3995          * ocfs2_add_new_xattr_bucket() can do this - its call to
3996          * ocfs2_add_new_xattr_cluster() may have created a new extent
3997          * and copied out the end of the old extent.  Then it re-extends
3998          * the old extent back to create space for new xattrs.  That's
3999          * how we get here, and the bucket isn't really new.
4000          */
4001         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4002                                                 t_is_new ?
4003                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4004                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4005         if (ret)
4006                 goto out;
4007
4008         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4009         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4010
4011 out:
4012         ocfs2_xattr_bucket_free(t_bucket);
4013         ocfs2_xattr_bucket_free(s_bucket);
4014
4015         return ret;
4016 }
4017
4018 /*
4019  * src_blk points to the start of an existing extent.  last_blk points to
4020  * last cluster in that extent.  to_blk points to a newly allocated
4021  * extent.  We copy the buckets from the cluster at last_blk to the new
4022  * extent.  If start_bucket is non-zero, we skip that many buckets before
4023  * we start copying.  The new extent's xh_num_buckets gets set to the
4024  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4025  * by the same amount.
4026  */
4027 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4028                                   u64 src_blk, u64 last_blk, u64 to_blk,
4029                                   unsigned int start_bucket,
4030                                   u32 *first_hash)
4031 {
4032         int i, ret, credits;
4033         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4034         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4035         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4036         struct ocfs2_xattr_bucket *old_first, *new_first;
4037
4038         mlog(0, "mv xattrs from cluster %llu to %llu\n",
4039              (unsigned long long)last_blk, (unsigned long long)to_blk);
4040
4041         BUG_ON(start_bucket >= num_buckets);
4042         if (start_bucket) {
4043                 num_buckets -= start_bucket;
4044                 last_blk += (start_bucket * blks_per_bucket);
4045         }
4046
4047         /* The first bucket of the original extent */
4048         old_first = ocfs2_xattr_bucket_new(inode);
4049         /* The first bucket of the new extent */
4050         new_first = ocfs2_xattr_bucket_new(inode);
4051         if (!old_first || !new_first) {
4052                 ret = -ENOMEM;
4053                 mlog_errno(ret);
4054                 goto out;
4055         }
4056
4057         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4058         if (ret) {
4059                 mlog_errno(ret);
4060                 goto out;
4061         }
4062
4063         /*
4064          * We need to update the first bucket of the old extent and all
4065          * the buckets going to the new extent.
4066          */
4067         credits = ((num_buckets + 1) * blks_per_bucket) +
4068                 handle->h_buffer_credits;
4069         ret = ocfs2_extend_trans(handle, credits);
4070         if (ret) {
4071                 mlog_errno(ret);
4072                 goto out;
4073         }
4074
4075         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4076                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4077         if (ret) {
4078                 mlog_errno(ret);
4079                 goto out;
4080         }
4081
4082         for (i = 0; i < num_buckets; i++) {
4083                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4084                                             last_blk + (i * blks_per_bucket),
4085                                             to_blk + (i * blks_per_bucket),
4086                                             1);
4087                 if (ret) {
4088                         mlog_errno(ret);
4089                         goto out;
4090                 }
4091         }
4092
4093         /*
4094          * Get the new bucket ready before we dirty anything
4095          * (This actually shouldn't fail, because we already dirtied
4096          * it once in ocfs2_cp_xattr_bucket()).
4097          */
4098         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4099         if (ret) {
4100                 mlog_errno(ret);
4101                 goto out;
4102         }
4103         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4104                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4105         if (ret) {
4106                 mlog_errno(ret);
4107                 goto out;
4108         }
4109
4110         /* Now update the headers */
4111         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4112         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4113
4114         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4115         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4116
4117         if (first_hash)
4118                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4119
4120 out:
4121         ocfs2_xattr_bucket_free(new_first);
4122         ocfs2_xattr_bucket_free(old_first);
4123         return ret;
4124 }
4125
4126 /*
4127  * Move some xattrs in this cluster to the new cluster.
4128  * This function should only be called when bucket size == cluster size.
4129  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4130  */
4131 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4132                                       handle_t *handle,
4133                                       u64 prev_blk,
4134                                       u64 new_blk,
4135                                       u32 *first_hash)
4136 {
4137         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4138         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4139
4140         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4141
4142         ret = ocfs2_extend_trans(handle, credits);
4143         if (ret) {
4144                 mlog_errno(ret);
4145                 return ret;
4146         }
4147
4148         /* Move half of the xattr in start_blk to the next bucket. */
4149         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4150                                           new_blk, first_hash, 1);
4151 }
4152
4153 /*
4154  * Move some xattrs from the old cluster to the new one since they are not
4155  * contiguous in ocfs2 xattr tree.
4156  *
4157  * new_blk starts a new separate cluster, and we will move some xattrs from
4158  * prev_blk to it. v_start will be set as the first name hash value in this
4159  * new cluster so that it can be used as e_cpos during tree insertion and
4160  * don't collide with our original b-tree operations. first_bh and header_bh
4161  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4162  * to extend the insert bucket.
4163  *
4164  * The problem is how much xattr should we move to the new one and when should
4165  * we update first_bh and header_bh?
4166  * 1. If cluster size > bucket size, that means the previous cluster has more
4167  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4168  *    update the first_bh and header_bh if the insert bucket has been moved
4169  *    to the new cluster.
4170  * 2. If cluster_size == bucket_size:
4171  *    a) If the previous extent rec has more than one cluster and the insert
4172  *       place isn't in the last cluster, copy the entire last cluster to the
4173  *       new one. This time, we don't need to upate the first_bh and header_bh
4174  *       since they will not be moved into the new cluster.
4175  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4176  *       the new one. And we set the extend flag to zero if the insert place is
4177  *       moved into the new allocated cluster since no extend is needed.
4178  */
4179 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4180                                             handle_t *handle,
4181                                             struct ocfs2_xattr_bucket *first,
4182                                             struct ocfs2_xattr_bucket *target,
4183                                             u64 new_blk,
4184                                             u32 prev_clusters,
4185                                             u32 *v_start,
4186                                             int *extend)
4187 {
4188         int ret;
4189
4190         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4191              (unsigned long long)bucket_blkno(first), prev_clusters,
4192              (unsigned long long)new_blk);
4193
4194         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4195                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4196                                                           handle,
4197                                                           first, target,
4198                                                           new_blk,
4199                                                           prev_clusters,
4200                                                           v_start);
4201                 if (ret)
4202                         mlog_errno(ret);
4203         } else {
4204                 /* The start of the last cluster in the first extent */
4205                 u64 last_blk = bucket_blkno(first) +
4206                         ((prev_clusters - 1) *
4207                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
4208
4209                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4210                         ret = ocfs2_mv_xattr_buckets(inode, handle,
4211                                                      bucket_blkno(first),
4212                                                      last_blk, new_blk, 0,
4213                                                      v_start);
4214                         if (ret)
4215                                 mlog_errno(ret);
4216                 } else {
4217                         ret = ocfs2_divide_xattr_cluster(inode, handle,
4218                                                          last_blk, new_blk,
4219                                                          v_start);
4220                         if (ret)
4221                                 mlog_errno(ret);
4222
4223                         if ((bucket_blkno(target) == last_blk) && extend)
4224                                 *extend = 0;
4225                 }
4226         }
4227
4228         return ret;
4229 }
4230
4231 /*
4232  * Add a new cluster for xattr storage.
4233  *
4234  * If the new cluster is contiguous with the previous one, it will be
4235  * appended to the same extent record, and num_clusters will be updated.
4236  * If not, we will insert a new extent for it and move some xattrs in
4237  * the last cluster into the new allocated one.
4238  * We also need to limit the maximum size of a btree leaf, otherwise we'll
4239  * lose the benefits of hashing because we'll have to search large leaves.
4240  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4241  * if it's bigger).
4242  *
4243  * first_bh is the first block of the previous extent rec and header_bh
4244  * indicates the bucket we will insert the new xattrs. They will be updated
4245  * when the header_bh is moved into the new cluster.
4246  */
4247 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4248                                        struct buffer_head *root_bh,
4249                                        struct ocfs2_xattr_bucket *first,
4250                                        struct ocfs2_xattr_bucket *target,
4251                                        u32 *num_clusters,
4252                                        u32 prev_cpos,
4253                                        int *extend,
4254                                        struct ocfs2_xattr_set_ctxt *ctxt)
4255 {
4256         int ret;
4257         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4258         u32 prev_clusters = *num_clusters;
4259         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4260         u64 block;
4261         handle_t *handle = ctxt->handle;
4262         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4263         struct ocfs2_extent_tree et;
4264
4265         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4266              "previous xattr blkno = %llu\n",
4267              (unsigned long long)OCFS2_I(inode)->ip_blkno,
4268              prev_cpos, (unsigned long long)bucket_blkno(first));
4269
4270         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4271
4272         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4273                                       OCFS2_JOURNAL_ACCESS_WRITE);
4274         if (ret < 0) {
4275                 mlog_errno(ret);
4276                 goto leave;
4277         }
4278
4279         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4280                                      clusters_to_add, &bit_off, &num_bits);
4281         if (ret < 0) {
4282                 if (ret != -ENOSPC)
4283                         mlog_errno(ret);
4284                 goto leave;
4285         }
4286
4287         BUG_ON(num_bits > clusters_to_add);
4288
4289         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4290         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4291              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4292
4293         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4294             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4295              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4296                 /*
4297                  * If this cluster is contiguous with the old one and
4298                  * adding this new cluster, we don't surpass the limit of
4299                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4300                  * initialized and used like other buckets in the previous
4301                  * cluster.
4302                  * So add it as a contiguous one. The caller will handle
4303                  * its init process.
4304                  */
4305                 v_start = prev_cpos + prev_clusters;
4306                 *num_clusters = prev_clusters + num_bits;
4307                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4308                      num_bits);
4309         } else {
4310                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4311                                                        handle,
4312                                                        first,
4313                                                        target,
4314                                                        block,
4315                                                        prev_clusters,
4316                                                        &v_start,
4317                                                        extend);
4318                 if (ret) {
4319                         mlog_errno(ret);
4320                         goto leave;
4321                 }
4322         }
4323
4324         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4325              num_bits, (unsigned long long)block, v_start);
4326         ret = ocfs2_insert_extent(handle, &et, v_start, block,
4327                                   num_bits, 0, ctxt->meta_ac);
4328         if (ret < 0) {
4329                 mlog_errno(ret);
4330                 goto leave;
4331         }
4332
4333         ret = ocfs2_journal_dirty(handle, root_bh);
4334         if (ret < 0)
4335                 mlog_errno(ret);
4336
4337 leave:
4338         return ret;
4339 }
4340
4341 /*
4342  * We are given an extent.  'first' is the bucket at the very front of
4343  * the extent.  The extent has space for an additional bucket past
4344  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
4345  * of the target bucket.  We wish to shift every bucket past the target
4346  * down one, filling in that additional space.  When we get back to the
4347  * target, we split the target between itself and the now-empty bucket
4348  * at target+1 (aka, target_blkno + blks_per_bucket).
4349  */
4350 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4351                                      handle_t *handle,
4352                                      struct ocfs2_xattr_bucket *first,
4353                                      u64 target_blk,
4354                                      u32 num_clusters)
4355 {
4356         int ret, credits;
4357         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4358         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4359         u64 end_blk;
4360         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4361
4362         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4363              "from %llu, len = %u\n", (unsigned long long)target_blk,
4364              (unsigned long long)bucket_blkno(first), num_clusters);
4365
4366         /* The extent must have room for an additional bucket */
4367         BUG_ON(new_bucket >=
4368                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4369
4370         /* end_blk points to the last existing bucket */
4371         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4372
4373         /*
4374          * end_blk is the start of the last existing bucket.
4375          * Thus, (end_blk - target_blk) covers the target bucket and
4376          * every bucket after it up to, but not including, the last
4377          * existing bucket.  Then we add the last existing bucket, the
4378          * new bucket, and the first bucket (3 * blk_per_bucket).
4379          */
4380         credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4381                   handle->h_buffer_credits;
4382         ret = ocfs2_extend_trans(handle, credits);
4383         if (ret) {
4384                 mlog_errno(ret);
4385                 goto out;
4386         }
4387
4388         ret = ocfs2_xattr_bucket_journal_access(handle, first,
4389                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4390         if (ret) {
4391                 mlog_errno(ret);
4392                 goto out;
4393         }
4394
4395         while (end_blk != target_blk) {
4396                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4397                                             end_blk + blk_per_bucket, 0);
4398                 if (ret)
4399                         goto out;
4400                 end_blk -= blk_per_bucket;
4401         }
4402
4403         /* Move half of the xattr in target_blkno to the next bucket. */
4404         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4405                                         target_blk + blk_per_bucket, NULL, 0);
4406
4407         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4408         ocfs2_xattr_bucket_journal_dirty(handle, first);
4409
4410 out:
4411         return ret;
4412 }
4413
4414 /*
4415  * Add new xattr bucket in an extent record and adjust the buckets
4416  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
4417  * bucket we want to insert into.
4418  *
4419  * In the easy case, we will move all the buckets after target down by
4420  * one. Half of target's xattrs will be moved to the next bucket.
4421  *
4422  * If current cluster is full, we'll allocate a new one.  This may not
4423  * be contiguous.  The underlying calls will make sure that there is
4424  * space for the insert, shifting buckets around if necessary.
4425  * 'target' may be moved by those calls.
4426  */
4427 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4428                                       struct buffer_head *xb_bh,
4429                                       struct ocfs2_xattr_bucket *target,
4430                                       struct ocfs2_xattr_set_ctxt *ctxt)
4431 {
4432         struct ocfs2_xattr_block *xb =
4433                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4434         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4435         struct ocfs2_extent_list *el = &xb_root->xt_list;
4436         u32 name_hash =
4437                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4438         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4439         int ret, num_buckets, extend = 1;
4440         u64 p_blkno;
4441         u32 e_cpos, num_clusters;
4442         /* The bucket at the front of the extent */
4443         struct ocfs2_xattr_bucket *first;
4444
4445         mlog(0, "Add new xattr bucket starting from %llu\n",
4446              (unsigned long long)bucket_blkno(target));
4447
4448         /* The first bucket of the original extent */
4449         first = ocfs2_xattr_bucket_new(inode);
4450         if (!first) {
4451                 ret = -ENOMEM;
4452                 mlog_errno(ret);
4453                 goto out;
4454         }
4455
4456         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4457                                   &num_clusters, el);
4458         if (ret) {
4459                 mlog_errno(ret);
4460                 goto out;
4461         }
4462
4463         ret = ocfs2_read_xattr_bucket(first, p_blkno);
4464         if (ret) {
4465                 mlog_errno(ret);
4466                 goto out;
4467         }
4468
4469         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4470         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4471                 /*
4472                  * This can move first+target if the target bucket moves
4473                  * to the new extent.
4474                  */
4475                 ret = ocfs2_add_new_xattr_cluster(inode,
4476                                                   xb_bh,
4477                                                   first,
4478                                                   target,
4479                                                   &num_clusters,
4480                                                   e_cpos,
4481                                                   &extend,
4482                                                   ctxt);
4483                 if (ret) {
4484                         mlog_errno(ret);
4485                         goto out;
4486                 }
4487         }
4488
4489         if (extend) {
4490                 ret = ocfs2_extend_xattr_bucket(inode,
4491                                                 ctxt->handle,
4492                                                 first,
4493                                                 bucket_blkno(target),
4494                                                 num_clusters);
4495                 if (ret)
4496                         mlog_errno(ret);
4497         }
4498
4499 out:
4500         ocfs2_xattr_bucket_free(first);
4501
4502         return ret;
4503 }
4504
4505 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4506                                         struct ocfs2_xattr_bucket *bucket,
4507                                         int offs)
4508 {
4509         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4510
4511         offs = offs % inode->i_sb->s_blocksize;
4512         return bucket_block(bucket, block_off) + offs;
4513 }
4514
4515 /*
4516  * Handle the normal xattr set, including replace, delete and new.
4517  *
4518  * Note: "local" indicates the real data's locality. So we can't
4519  * just its bucket locality by its length.
4520  */
4521 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4522                                          struct ocfs2_xattr_info *xi,
4523                                          struct ocfs2_xattr_search *xs,
4524                                          u32 name_hash,
4525                                          int local)
4526 {
4527         struct ocfs2_xattr_entry *last, *xe;
4528         int name_len = strlen(xi->name);
4529         struct ocfs2_xattr_header *xh = xs->header;
4530         u16 count = le16_to_cpu(xh->xh_count), start;
4531         size_t blocksize = inode->i_sb->s_blocksize;
4532         char *val;
4533         size_t offs, size, new_size;
4534
4535         last = &xh->xh_entries[count];
4536         if (!xs->not_found) {
4537                 xe = xs->here;
4538                 offs = le16_to_cpu(xe->xe_name_offset);
4539                 if (ocfs2_xattr_is_local(xe))
4540                         size = OCFS2_XATTR_SIZE(name_len) +
4541                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4542                 else
4543                         size = OCFS2_XATTR_SIZE(name_len) +
4544                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4545
4546                 /*
4547                  * If the new value will be stored outside, xi->value has been
4548                  * initalized as an empty ocfs2_xattr_value_root, and the same
4549                  * goes with xi->value_len, so we can set new_size safely here.
4550                  * See ocfs2_xattr_set_in_bucket.
4551                  */
4552                 new_size = OCFS2_XATTR_SIZE(name_len) +
4553                            OCFS2_XATTR_SIZE(xi->value_len);
4554
4555                 le16_add_cpu(&xh->xh_name_value_len, -size);
4556                 if (xi->value) {
4557                         if (new_size > size)
4558                                 goto set_new_name_value;
4559
4560                         /* Now replace the old value with new one. */
4561                         if (local)
4562                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4563                         else
4564                                 xe->xe_value_size = 0;
4565
4566                         val = ocfs2_xattr_bucket_get_val(inode,
4567                                                          xs->bucket, offs);
4568                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4569                                size - OCFS2_XATTR_SIZE(name_len));
4570                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4571                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4572                                        xi->value, xi->value_len);
4573
4574                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4575                         ocfs2_xattr_set_local(xe, local);
4576                         return;
4577                 } else {
4578                         /*
4579                          * Remove the old entry if there is more than one.
4580                          * We don't remove the last entry so that we can
4581                          * use it to indicate the hash value of the empty
4582                          * bucket.
4583                          */
4584                         last -= 1;
4585                         le16_add_cpu(&xh->xh_count, -1);
4586                         if (xh->xh_count) {
4587                                 memmove(xe, xe + 1,
4588                                         (void *)last - (void *)xe);
4589                                 memset(last, 0,
4590                                        sizeof(struct ocfs2_xattr_entry));
4591                         } else
4592                                 xh->xh_free_start =
4593                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4594
4595                         return;
4596                 }
4597         } else {
4598                 /* find a new entry for insert. */
4599                 int low = 0, high = count - 1, tmp;
4600                 struct ocfs2_xattr_entry *tmp_xe;
4601
4602                 while (low <= high && count) {
4603                         tmp = (low + high) / 2;
4604                         tmp_xe = &xh->xh_entries[tmp];
4605
4606                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4607                                 low = tmp + 1;
4608                         else if (name_hash <
4609                                  le32_to_cpu(tmp_xe->xe_name_hash))
4610                                 high = tmp - 1;
4611                         else {
4612                                 low = tmp;
4613                                 break;
4614                         }
4615                 }
4616
4617                 xe = &xh->xh_entries[low];
4618                 if (low != count)
4619                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4620
4621                 le16_add_cpu(&xh->xh_count, 1);
4622                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4623                 xe->xe_name_hash = cpu_to_le32(name_hash);
4624                 xe->xe_name_len = name_len;
4625                 ocfs2_xattr_set_type(xe, xi->name_index);
4626         }
4627
4628 set_new_name_value:
4629         /* Insert the new name+value. */
4630         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4631
4632         /*
4633          * We must make sure that the name/value pair
4634          * exists in the same block.
4635          */
4636         offs = le16_to_cpu(xh->xh_free_start);
4637         start = offs - size;
4638
4639         if (start >> inode->i_sb->s_blocksize_bits !=
4640             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4641                 offs = offs - offs % blocksize;
4642                 xh->xh_free_start = cpu_to_le16(offs);
4643         }
4644
4645         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4646         xe->xe_name_offset = cpu_to_le16(offs - size);
4647
4648         memset(val, 0, size);
4649         memcpy(val, xi->name, name_len);
4650         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4651
4652         xe->xe_value_size = cpu_to_le64(xi->value_len);
4653         ocfs2_xattr_set_local(xe, local);
4654         xs->here = xe;
4655         le16_add_cpu(&xh->xh_free_start, -size);
4656         le16_add_cpu(&xh->xh_name_value_len, size);
4657
4658         return;
4659 }
4660
4661 /*
4662  * Set the xattr entry in the specified bucket.
4663  * The bucket is indicated by xs->bucket and it should have the enough
4664  * space for the xattr insertion.
4665  */
4666 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4667                                            handle_t *handle,
4668                                            struct ocfs2_xattr_info *xi,
4669                                            struct ocfs2_xattr_search *xs,
4670                                            u32 name_hash,
4671                                            int local)
4672 {
4673         int ret;
4674         u64 blkno;
4675
4676         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4677              (unsigned long)xi->value_len, xi->name_index,
4678              (unsigned long long)bucket_blkno(xs->bucket));
4679
4680         if (!xs->bucket->bu_bhs[1]) {
4681                 blkno = bucket_blkno(xs->bucket);
4682                 ocfs2_xattr_bucket_relse(xs->bucket);
4683                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4684                 if (ret) {
4685                         mlog_errno(ret);
4686                         goto out;
4687                 }
4688         }
4689
4690         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4691                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4692         if (ret < 0) {
4693                 mlog_errno(ret);
4694                 goto out;
4695         }
4696
4697         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4698         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4699
4700 out:
4701         return ret;
4702 }
4703
4704 /*
4705  * Truncate the specified xe_off entry in xattr bucket.
4706  * bucket is indicated by header_bh and len is the new length.
4707  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4708  *
4709  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4710  */
4711 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4712                                              struct ocfs2_xattr_bucket *bucket,
4713                                              int xe_off,
4714                                              int len,
4715                                              struct ocfs2_xattr_set_ctxt *ctxt)
4716 {
4717         int ret, offset;
4718         u64 value_blk;
4719         struct ocfs2_xattr_entry *xe;
4720         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4721         size_t blocksize = inode->i_sb->s_blocksize;
4722         struct ocfs2_xattr_value_buf vb = {
4723                 .vb_access = ocfs2_journal_access,
4724         };
4725
4726         xe = &xh->xh_entries[xe_off];
4727
4728         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4729
4730         offset = le16_to_cpu(xe->xe_name_offset) +
4731                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4732
4733         value_blk = offset / blocksize;
4734
4735         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4736         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4737
4738         vb.vb_bh = bucket->bu_bhs[value_blk];
4739         BUG_ON(!vb.vb_bh);
4740
4741         vb.vb_xv = (struct ocfs2_xattr_value_root *)
4742                 (vb.vb_bh->b_data + offset % blocksize);
4743
4744         /*
4745          * From here on out we have to dirty the bucket.  The generic
4746          * value calls only modify one of the bucket's bhs, but we need
4747          * to send the bucket at once.  So if they error, they *could* have
4748          * modified something.  We have to assume they did, and dirty
4749          * the whole bucket.  This leaves us in a consistent state.
4750          */
4751         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4752              xe_off, (unsigned long long)bucket_blkno(bucket), len);
4753         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4754         if (ret) {
4755                 mlog_errno(ret);
4756                 goto out;
4757         }
4758
4759         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4760                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4761         if (ret) {
4762                 mlog_errno(ret);
4763                 goto out;
4764         }
4765
4766         xe->xe_value_size = cpu_to_le64(len);
4767
4768         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4769
4770 out:
4771         return ret;
4772 }
4773
4774 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4775                                         struct ocfs2_xattr_search *xs,
4776                                         int len,
4777                                         struct ocfs2_xattr_set_ctxt *ctxt)
4778 {
4779         int ret, offset;
4780         struct ocfs2_xattr_entry *xe = xs->here;
4781         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4782
4783         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4784
4785         offset = xe - xh->xh_entries;
4786         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4787                                                 offset, len, ctxt);
4788         if (ret)
4789                 mlog_errno(ret);
4790
4791         return ret;
4792 }
4793
4794 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4795                                                 handle_t *handle,
4796                                                 struct ocfs2_xattr_search *xs,
4797                                                 char *val,
4798                                                 int value_len)
4799 {
4800         int ret, offset, block_off;
4801         struct ocfs2_xattr_value_root *xv;
4802         struct ocfs2_xattr_entry *xe = xs->here;
4803         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4804         void *base;
4805
4806         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4807
4808         ret = ocfs2_xattr_bucket_get_name_value(inode, xh,
4809                                                 xe - xh->xh_entries,
4810                                                 &block_off,
4811                                                 &offset);
4812         if (ret) {
4813                 mlog_errno(ret);
4814                 goto out;
4815         }
4816
4817         base = bucket_block(xs->bucket, block_off);
4818         xv = (struct ocfs2_xattr_value_root *)(base + offset +
4819                  OCFS2_XATTR_SIZE(xe->xe_name_len));
4820
4821         ret = __ocfs2_xattr_set_value_outside(inode, handle,
4822                                               xv, val, value_len);
4823         if (ret)
4824                 mlog_errno(ret);
4825 out:
4826         return ret;
4827 }
4828
4829 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4830                                   struct buffer_head *root_bh,
4831                                   u64 blkno,
4832                                   u32 cpos,
4833                                   u32 len)
4834 {
4835         int ret;
4836         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4837         struct inode *tl_inode = osb->osb_tl_inode;
4838         handle_t *handle;
4839         struct ocfs2_xattr_block *xb =
4840                         (struct ocfs2_xattr_block *)root_bh->b_data;
4841         struct ocfs2_alloc_context *meta_ac = NULL;
4842         struct ocfs2_cached_dealloc_ctxt dealloc;
4843         struct ocfs2_extent_tree et;
4844
4845         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4846
4847         ocfs2_init_dealloc_ctxt(&dealloc);
4848
4849         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4850              cpos, len, (unsigned long long)blkno);
4851
4852         ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
4853                                                len);
4854
4855         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4856         if (ret) {
4857                 mlog_errno(ret);
4858                 return ret;
4859         }
4860
4861         mutex_lock(&tl_inode->i_mutex);
4862
4863         if (ocfs2_truncate_log_needs_flush(osb)) {
4864                 ret = __ocfs2_flush_truncate_log(osb);
4865                 if (ret < 0) {
4866                         mlog_errno(ret);
4867                         goto out;
4868                 }
4869         }
4870
4871         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4872         if (IS_ERR(handle)) {
4873                 ret = -ENOMEM;
4874                 mlog_errno(ret);
4875                 goto out;
4876         }
4877
4878         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4879                                       OCFS2_JOURNAL_ACCESS_WRITE);
4880         if (ret) {
4881                 mlog_errno(ret);
4882                 goto out_commit;
4883         }
4884
4885         ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
4886                                   &dealloc);
4887         if (ret) {
4888                 mlog_errno(ret);
4889                 goto out_commit;
4890         }
4891
4892         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4893
4894         ret = ocfs2_journal_dirty(handle, root_bh);
4895         if (ret) {
4896                 mlog_errno(ret);
4897                 goto out_commit;
4898         }
4899
4900         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4901         if (ret)
4902                 mlog_errno(ret);
4903
4904 out_commit:
4905         ocfs2_commit_trans(osb, handle);
4906 out:
4907         ocfs2_schedule_truncate_log_flush(osb, 1);
4908
4909         mutex_unlock(&tl_inode->i_mutex);
4910
4911         if (meta_ac)
4912                 ocfs2_free_alloc_context(meta_ac);
4913
4914         ocfs2_run_deallocs(osb, &dealloc);
4915
4916         return ret;
4917 }
4918
4919 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4920                                          handle_t *handle,
4921                                          struct ocfs2_xattr_search *xs)
4922 {
4923         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4924         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4925                                                 le16_to_cpu(xh->xh_count) - 1];
4926         int ret = 0;
4927
4928         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4929                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4930         if (ret) {
4931                 mlog_errno(ret);
4932                 return;
4933         }
4934
4935         /* Remove the old entry. */
4936         memmove(xs->here, xs->here + 1,
4937                 (void *)last - (void *)xs->here);
4938         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4939         le16_add_cpu(&xh->xh_count, -1);
4940
4941         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4942 }
4943
4944 /*
4945  * Set the xattr name/value in the bucket specified in xs.
4946  *
4947  * As the new value in xi may be stored in the bucket or in an outside cluster,
4948  * we divide the whole process into 3 steps:
4949  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4950  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4951  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4952  * 4. If the clusters for the new outside value can't be allocated, we need
4953  *    to free the xattr we allocated in set.
4954  */
4955 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4956                                      struct ocfs2_xattr_info *xi,
4957                                      struct ocfs2_xattr_search *xs,
4958                                      struct ocfs2_xattr_set_ctxt *ctxt)
4959 {
4960         int ret, local = 1;
4961         size_t value_len;
4962         char *val = (char *)xi->value;
4963         struct ocfs2_xattr_entry *xe = xs->here;
4964         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4965                                               strlen(xi->name));
4966
4967         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4968                 /*
4969                  * We need to truncate the xattr storage first.
4970                  *
4971                  * If both the old and new value are stored to
4972                  * outside block, we only need to truncate
4973                  * the storage and then set the value outside.
4974                  *
4975                  * If the new value should be stored within block,
4976                  * we should free all the outside block first and
4977                  * the modification to the xattr block will be done
4978                  * by following steps.
4979                  */
4980                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4981                         value_len = xi->value_len;
4982                 else
4983                         value_len = 0;
4984
4985                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4986                                                            value_len,
4987                                                            ctxt);
4988                 if (ret)
4989                         goto out;
4990
4991                 if (value_len)
4992                         goto set_value_outside;
4993         }
4994
4995         value_len = xi->value_len;
4996         /* So we have to handle the inside block change now. */
4997         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4998                 /*
4999                  * If the new value will be stored outside of block,
5000                  * initalize a new empty value root and insert it first.
5001                  */
5002                 local = 0;
5003                 xi->value = &def_xv;
5004                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
5005         }
5006
5007         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
5008                                               name_hash, local);
5009         if (ret) {
5010                 mlog_errno(ret);
5011                 goto out;
5012         }
5013
5014         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
5015                 goto out;
5016
5017         /* allocate the space now for the outside block storage. */
5018         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5019                                                    value_len, ctxt);
5020         if (ret) {
5021                 mlog_errno(ret);
5022
5023                 if (xs->not_found) {
5024                         /*
5025                          * We can't allocate enough clusters for outside
5026                          * storage and we have allocated xattr already,
5027                          * so need to remove it.
5028                          */
5029                         ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
5030                 }
5031                 goto out;
5032         }
5033
5034 set_value_outside:
5035         ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5036                                                    xs, val, value_len);
5037 out:
5038         return ret;
5039 }
5040
5041 /*
5042  * check whether the xattr bucket is filled up with the same hash value.
5043  * If we want to insert the xattr with the same hash, return -ENOSPC.
5044  * If we want to insert a xattr with different hash value, go ahead
5045  * and ocfs2_divide_xattr_bucket will handle this.
5046  */
5047 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5048                                               struct ocfs2_xattr_bucket *bucket,
5049                                               const char *name)
5050 {
5051         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5052         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5053
5054         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5055                 return 0;
5056
5057         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5058             xh->xh_entries[0].xe_name_hash) {
5059                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5060                      "hash = %u\n",
5061                      (unsigned long long)bucket_blkno(bucket),
5062                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5063                 return -ENOSPC;
5064         }
5065
5066         return 0;
5067 }
5068
5069 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5070                                              struct ocfs2_xattr_info *xi,
5071                                              struct ocfs2_xattr_search *xs,
5072                                              struct ocfs2_xattr_set_ctxt *ctxt)
5073 {
5074         struct ocfs2_xattr_header *xh;
5075         struct ocfs2_xattr_entry *xe;
5076         u16 count, header_size, xh_free_start;
5077         int free, max_free, need, old;
5078         size_t value_size = 0, name_len = strlen(xi->name);
5079         size_t blocksize = inode->i_sb->s_blocksize;
5080         int ret, allocation = 0;
5081
5082         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5083
5084 try_again:
5085         xh = xs->header;
5086         count = le16_to_cpu(xh->xh_count);
5087         xh_free_start = le16_to_cpu(xh->xh_free_start);
5088         header_size = sizeof(struct ocfs2_xattr_header) +
5089                         count * sizeof(struct ocfs2_xattr_entry);
5090         max_free = OCFS2_XATTR_BUCKET_SIZE - header_size -
5091                 le16_to_cpu(xh->xh_name_value_len) - OCFS2_XATTR_HEADER_GAP;
5092
5093         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5094                         "of %u which exceed block size\n",
5095                         (unsigned long long)bucket_blkno(xs->bucket),
5096                         header_size);
5097
5098         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5099                 value_size = OCFS2_XATTR_ROOT_SIZE;
5100         else if (xi->value)
5101                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5102
5103         if (xs->not_found)
5104                 need = sizeof(struct ocfs2_xattr_entry) +
5105                         OCFS2_XATTR_SIZE(name_len) + value_size;
5106         else {
5107                 need = value_size + OCFS2_XATTR_SIZE(name_len);
5108
5109                 /*
5110                  * We only replace the old value if the new length is smaller
5111                  * than the old one. Otherwise we will allocate new space in the
5112                  * bucket to store it.
5113                  */
5114                 xe = xs->here;
5115                 if (ocfs2_xattr_is_local(xe))
5116                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5117                 else
5118                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5119
5120                 if (old >= value_size)
5121                         need = 0;
5122         }
5123
5124         free = xh_free_start - header_size - OCFS2_XATTR_HEADER_GAP;
5125         /*
5126          * We need to make sure the new name/value pair
5127          * can exist in the same block.
5128          */
5129         if (xh_free_start % blocksize < need)
5130                 free -= xh_free_start % blocksize;
5131
5132         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5133              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5134              " %u\n", xs->not_found,
5135              (unsigned long long)bucket_blkno(xs->bucket),
5136              free, need, max_free, le16_to_cpu(xh->xh_free_start),
5137              le16_to_cpu(xh->xh_name_value_len));
5138
5139         if (free < need ||
5140             (xs->not_found &&
5141              count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5142                 if (need <= max_free &&
5143                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5144                         /*
5145                          * We can create the space by defragment. Since only the
5146                          * name/value will be moved, the xe shouldn't be changed
5147                          * in xs.
5148                          */
5149                         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5150                                                         xs->bucket);
5151                         if (ret) {
5152                                 mlog_errno(ret);
5153                                 goto out;
5154                         }
5155
5156                         xh_free_start = le16_to_cpu(xh->xh_free_start);
5157                         free = xh_free_start - header_size
5158                                 - OCFS2_XATTR_HEADER_GAP;
5159                         if (xh_free_start % blocksize < need)
5160                                 free -= xh_free_start % blocksize;
5161
5162                         if (free >= need)
5163                                 goto xattr_set;
5164
5165                         mlog(0, "Can't get enough space for xattr insert by "
5166                              "defragment. Need %u bytes, but we have %d, so "
5167                              "allocate new bucket for it.\n", need, free);
5168                 }
5169
5170                 /*
5171                  * We have to add new buckets or clusters and one
5172                  * allocation should leave us enough space for insert.
5173                  */
5174                 BUG_ON(allocation);
5175
5176                 /*
5177                  * We do not allow for overlapping ranges between buckets. And
5178                  * the maximum number of collisions we will allow for then is
5179                  * one bucket's worth, so check it here whether we need to
5180                  * add a new bucket for the insert.
5181                  */
5182                 ret = ocfs2_check_xattr_bucket_collision(inode,
5183                                                          xs->bucket,
5184                                                          xi->name);
5185                 if (ret) {
5186                         mlog_errno(ret);
5187                         goto out;
5188                 }
5189
5190                 ret = ocfs2_add_new_xattr_bucket(inode,
5191                                                  xs->xattr_bh,
5192                                                  xs->bucket,
5193                                                  ctxt);
5194                 if (ret) {
5195                         mlog_errno(ret);
5196                         goto out;
5197                 }
5198
5199                 /*
5200                  * ocfs2_add_new_xattr_bucket() will have updated
5201                  * xs->bucket if it moved, but it will not have updated
5202                  * any of the other search fields.  Thus, we drop it and
5203                  * re-search.  Everything should be cached, so it'll be
5204                  * quick.
5205                  */
5206                 ocfs2_xattr_bucket_relse(xs->bucket);
5207                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5208                                                    xi->name_index,
5209                                                    xi->name, xs);
5210                 if (ret && ret != -ENODATA)
5211                         goto out;
5212                 xs->not_found = ret;
5213                 allocation = 1;
5214                 goto try_again;
5215         }
5216
5217 xattr_set:
5218         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5219 out:
5220         mlog_exit(ret);
5221         return ret;
5222 }
5223
5224 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5225                                         struct ocfs2_xattr_bucket *bucket,
5226                                         void *para)
5227 {
5228         int ret = 0;
5229         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5230         u16 i;
5231         struct ocfs2_xattr_entry *xe;
5232         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5233         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5234         int credits = ocfs2_remove_extent_credits(osb->sb) +
5235                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5236
5237
5238         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5239
5240         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5241                 xe = &xh->xh_entries[i];
5242                 if (ocfs2_xattr_is_local(xe))
5243                         continue;
5244
5245                 ctxt.handle = ocfs2_start_trans(osb, credits);
5246                 if (IS_ERR(ctxt.handle)) {
5247                         ret = PTR_ERR(ctxt.handle);
5248                         mlog_errno(ret);
5249                         break;
5250                 }
5251
5252                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5253                                                         i, 0, &ctxt);
5254
5255                 ocfs2_commit_trans(osb, ctxt.handle);
5256                 if (ret) {
5257                         mlog_errno(ret);
5258                         break;
5259                 }
5260         }
5261
5262         ocfs2_schedule_truncate_log_flush(osb, 1);
5263         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5264         return ret;
5265 }
5266
5267 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5268                                           struct buffer_head *xb_bh)
5269 {
5270         struct ocfs2_xattr_block *xb =
5271                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5272         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5273         int ret = 0;
5274         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5275         u64 p_blkno;
5276
5277         if (le16_to_cpu(el->l_next_free_rec) == 0)
5278                 return 0;
5279
5280         while (name_hash > 0) {
5281                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5282                                           &e_cpos, &num_clusters, el);
5283                 if (ret) {
5284                         mlog_errno(ret);
5285                         goto out;
5286                 }
5287
5288                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5289                                                   ocfs2_delete_xattr_in_bucket,
5290                                                   NULL);
5291                 if (ret) {
5292                         mlog_errno(ret);
5293                         goto out;
5294                 }
5295
5296                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5297                                              p_blkno, e_cpos, num_clusters);
5298                 if (ret) {
5299                         mlog_errno(ret);
5300                         break;
5301                 }
5302
5303                 if (e_cpos == 0)
5304                         break;
5305
5306                 name_hash = e_cpos - 1;
5307         }
5308
5309 out:
5310         return ret;
5311 }
5312
5313 /*
5314  * 'security' attributes support
5315  */
5316 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5317                                         size_t list_size, const char *name,
5318                                         size_t name_len)
5319 {
5320         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5321         const size_t total_len = prefix_len + name_len + 1;
5322
5323         if (list && total_len <= list_size) {
5324                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5325                 memcpy(list + prefix_len, name, name_len);
5326                 list[prefix_len + name_len] = '\0';
5327         }
5328         return total_len;
5329 }
5330
5331 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5332                                     void *buffer, size_t size)
5333 {
5334         if (strcmp(name, "") == 0)
5335                 return -EINVAL;
5336         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5337                                buffer, size);
5338 }
5339
5340 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5341                                     const void *value, size_t size, int flags)
5342 {
5343         if (strcmp(name, "") == 0)
5344                 return -EINVAL;
5345
5346         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5347                                size, flags);
5348 }
5349
5350 int ocfs2_init_security_get(struct inode *inode,
5351                             struct inode *dir,
5352                             struct ocfs2_security_xattr_info *si)
5353 {
5354         /* check whether ocfs2 support feature xattr */
5355         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
5356                 return -EOPNOTSUPP;
5357         return security_inode_init_security(inode, dir, &si->name, &si->value,
5358                                             &si->value_len);
5359 }
5360
5361 int ocfs2_init_security_set(handle_t *handle,
5362                             struct inode *inode,
5363                             struct buffer_head *di_bh,
5364                             struct ocfs2_security_xattr_info *si,
5365                             struct ocfs2_alloc_context *xattr_ac,
5366                             struct ocfs2_alloc_context *data_ac)
5367 {
5368         return ocfs2_xattr_set_handle(handle, inode, di_bh,
5369                                      OCFS2_XATTR_INDEX_SECURITY,
5370                                      si->name, si->value, si->value_len, 0,
5371                                      xattr_ac, data_ac);
5372 }
5373
5374 struct xattr_handler ocfs2_xattr_security_handler = {
5375         .prefix = XATTR_SECURITY_PREFIX,
5376         .list   = ocfs2_xattr_security_list,
5377         .get    = ocfs2_xattr_security_get,
5378         .set    = ocfs2_xattr_security_set,
5379 };
5380
5381 /*
5382  * 'trusted' attributes support
5383  */
5384 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5385                                        size_t list_size, const char *name,
5386                                        size_t name_len)
5387 {
5388         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5389         const size_t total_len = prefix_len + name_len + 1;
5390
5391         if (list && total_len <= list_size) {
5392                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5393                 memcpy(list + prefix_len, name, name_len);
5394                 list[prefix_len + name_len] = '\0';
5395         }
5396         return total_len;
5397 }
5398
5399 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5400                                    void *buffer, size_t size)
5401 {
5402         if (strcmp(name, "") == 0)
5403                 return -EINVAL;
5404         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5405                                buffer, size);
5406 }
5407
5408 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5409                                    const void *value, size_t size, int flags)
5410 {
5411         if (strcmp(name, "") == 0)
5412                 return -EINVAL;
5413
5414         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5415                                size, flags);
5416 }
5417
5418 struct xattr_handler ocfs2_xattr_trusted_handler = {
5419         .prefix = XATTR_TRUSTED_PREFIX,
5420         .list   = ocfs2_xattr_trusted_list,
5421         .get    = ocfs2_xattr_trusted_get,
5422         .set    = ocfs2_xattr_trusted_set,
5423 };
5424
5425 /*
5426  * 'user' attributes support
5427  */
5428 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5429                                     size_t list_size, const char *name,
5430                                     size_t name_len)
5431 {
5432         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5433         const size_t total_len = prefix_len + name_len + 1;
5434         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5435
5436         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5437                 return 0;
5438
5439         if (list && total_len <= list_size) {
5440                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5441                 memcpy(list + prefix_len, name, name_len);
5442                 list[prefix_len + name_len] = '\0';
5443         }
5444         return total_len;
5445 }
5446
5447 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5448                                 void *buffer, size_t size)
5449 {
5450         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5451
5452         if (strcmp(name, "") == 0)
5453                 return -EINVAL;
5454         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5455                 return -EOPNOTSUPP;
5456         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5457                                buffer, size);
5458 }
5459
5460 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5461                                 const void *value, size_t size, int flags)
5462 {
5463         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5464
5465         if (strcmp(name, "") == 0)
5466                 return -EINVAL;
5467         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5468                 return -EOPNOTSUPP;
5469
5470         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5471                                size, flags);
5472 }
5473
5474 struct xattr_handler ocfs2_xattr_user_handler = {
5475         .prefix = XATTR_USER_PREFIX,
5476         .list   = ocfs2_xattr_user_list,
5477         .get    = ocfs2_xattr_user_get,
5478         .set    = ocfs2_xattr_user_set,
5479 };