fs/ocfs2/xattr.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * xattr.c
   5  *
   6  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
   7  *
   8  * CREDITS:
   9  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
  10  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  11  *
  12  * This program is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU General Public
  14  * License version 2 as published by the Free Software Foundation.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * General Public License for more details.
  20  */
  21
  22 #include <linux/capability.h>
  23 #include <linux/fs.h>
  24 #include <linux/types.h>
  25 #include <linux/slab.h>
  26 #include <linux/highmem.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/uio.h>
  29 #include <linux/sched.h>
  30 #include <linux/splice.h>
  31 #include <linux/mount.h>
  32 #include <linux/writeback.h>
  33 #include <linux/falloc.h>
  34 #include <linux/sort.h>
  35 #include <linux/init.h>
  36 #include <linux/module.h>
  37 #include <linux/string.h>
  38 #include <linux/security.h>
  39
  40 #define MLOG_MASK_PREFIX ML_XATTR
  41 #include <cluster/masklog.h>
  42
  43 #include "ocfs2.h"
  44 #include "alloc.h"
  45 #include "blockcheck.h"
  46 #include "dlmglue.h"
  47 #include "file.h"
  48 #include "symlink.h"
  49 #include "sysfile.h"
  50 #include "inode.h"
  51 #include "journal.h"
  52 #include "ocfs2_fs.h"
  53 #include "suballoc.h"
  54 #include "uptodate.h"
  55 #include "buffer_head_io.h"
  56 #include "super.h"
  57 #include "xattr.h"
  58
  59
  60 struct ocfs2_xattr_def_value_root {
  61         struct ocfs2_xattr_value_root   xv;
  62         struct ocfs2_extent_rec         er;
  63 };
  64
  65 struct ocfs2_xattr_bucket {
  66         /* The inode these xattrs are associated with */
  67         struct inode *bu_inode;
  68
  69         /* The actual buffers that make up the bucket */
  70         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  71
  72         /* How many blocks make up one bucket for this filesystem */
  73         int bu_blocks;
  74 };
  75
  76 struct ocfs2_xattr_set_ctxt {
  77         handle_t *handle;
  78         struct ocfs2_alloc_context *meta_ac;
  79         struct ocfs2_alloc_context *data_ac;
  80         struct ocfs2_cached_dealloc_ctxt dealloc;
  81 };
  82
  83 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  84 #define OCFS2_XATTR_INLINE_SIZE 80
  85 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
  86                                          - sizeof(struct ocfs2_xattr_header) \
  87                                          - sizeof(__u32))
  88 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
  89                                          - sizeof(struct ocfs2_xattr_block) \
  90                                          - sizeof(struct ocfs2_xattr_header) \
  91                                          - sizeof(__u32))
  92
  93 static struct ocfs2_xattr_def_value_root def_xv = {
  94         .xv.xr_list.l_count = cpu_to_le16(1),
  95 };
  96
  97 struct xattr_handler *ocfs2_xattr_handlers[] = {
  98         &ocfs2_xattr_user_handler,
  99 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
 100         &ocfs2_xattr_acl_access_handler,
 101         &ocfs2_xattr_acl_default_handler,
 102 #endif
 103         &ocfs2_xattr_trusted_handler,
 104         &ocfs2_xattr_security_handler,
 105         NULL
 106 };
 107
 108 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 109         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
 110 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
 111         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
 112                                         = &ocfs2_xattr_acl_access_handler,
 113         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
 114                                         = &ocfs2_xattr_acl_default_handler,
 115 #endif
 116         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
 117         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 118 };
 119
 120 struct ocfs2_xattr_info {
 121         int name_index;
 122         const char *name;
 123         const void *value;
 124         size_t value_len;
 125 };
 126
 127 struct ocfs2_xattr_search {
 128         struct buffer_head *inode_bh;
 129         /*
 130          * xattr_bh point to the block buffer head which has extended attribute
 131          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 132          */
 133         struct buffer_head *xattr_bh;
 134         struct ocfs2_xattr_header *header;
 135         struct ocfs2_xattr_bucket *bucket;
 136         void *base;
 137         void *end;
 138         struct ocfs2_xattr_entry *here;
 139         int not_found;
 140 };
 141
 142 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
 143                                              struct ocfs2_xattr_header *xh,
 144                                              int index,
 145                                              int *block_off,
 146                                              int *new_offset);
 147
 148 static int ocfs2_xattr_block_find(struct inode *inode,
 149                                   int name_index,
 150                                   const char *name,
 151                                   struct ocfs2_xattr_search *xs);
 152 static int ocfs2_xattr_index_block_find(struct inode *inode,
 153                                         struct buffer_head *root_bh,
 154                                         int name_index,
 155                                         const char *name,
 156                                         struct ocfs2_xattr_search *xs);
 157
 158 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 159                                         struct ocfs2_xattr_tree_root *xt,
 160                                         char *buffer,
 161                                         size_t buffer_size);
 162
 163 static int ocfs2_xattr_create_index_block(struct inode *inode,
 164                                           struct ocfs2_xattr_search *xs,
 165                                           struct ocfs2_xattr_set_ctxt *ctxt);
 166
 167 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 168                                              struct ocfs2_xattr_info *xi,
 169                                              struct ocfs2_xattr_search *xs,
 170                                              struct ocfs2_xattr_set_ctxt *ctxt);
 171
 172 static int ocfs2_delete_xattr_index_block(struct inode *inode,
 173                                           struct buffer_head *xb_bh);
 174 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 175                                   u64 src_blk, u64 last_blk, u64 to_blk,
 176                                   unsigned int start_bucket,
 177                                   u32 *first_hash);
 178
 179 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 180 {
 181         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 182 }
 183
 184 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 185 {
 186         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 187 }
 188
 189 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
 190 {
 191         u16 len = sb->s_blocksize -
 192                  offsetof(struct ocfs2_xattr_header, xh_entries);
 193
 194         return len / sizeof(struct ocfs2_xattr_entry);
 195 }
 196
 197 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 198 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 199 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 200
 201 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 202 {
 203         struct ocfs2_xattr_bucket *bucket;
 204         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 205
 206         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 207
 208         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 209         if (bucket) {
 210                 bucket->bu_inode = inode;
 211                 bucket->bu_blocks = blks;
 212         }
 213
 214         return bucket;
 215 }
 216
 217 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 218 {
 219         int i;
 220
 221         for (i = 0; i < bucket->bu_blocks; i++) {
 222                 brelse(bucket->bu_bhs[i]);
 223                 bucket->bu_bhs[i] = NULL;
 224         }
 225 }
 226
 227 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 228 {
 229         if (bucket) {
 230                 ocfs2_xattr_bucket_relse(bucket);
 231                 bucket->bu_inode = NULL;
 232                 kfree(bucket);
 233         }
 234 }
 235
 236 /*
 237  * A bucket that has never been written to disk doesn't need to be
 238  * read.  We just need the buffer_heads.  Don't call this for
 239  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 240  * them fully.
 241  */
 242 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 243                                    u64 xb_blkno)
 244 {
 245         int i, rc = 0;
 246
 247         for (i = 0; i < bucket->bu_blocks; i++) {
 248                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 249                                               xb_blkno + i);
 250                 if (!bucket->bu_bhs[i]) {
 251                         rc = -EIO;
 252                         mlog_errno(rc);
 253                         break;
 254                 }
 255
 256                 if (!ocfs2_buffer_uptodate(bucket->bu_inode,
 257                                            bucket->bu_bhs[i]))
 258                         ocfs2_set_new_buffer_uptodate(bucket->bu_inode,
 259                                                       bucket->bu_bhs[i]);
 260         }
 261
 262         if (rc)
 263                 ocfs2_xattr_bucket_relse(bucket);
 264         return rc;
 265 }
 266
 267 /* Read the xattr bucket at xb_blkno */
 268 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 269                                    u64 xb_blkno)
 270 {
 271         int rc;
 272
 273         rc = ocfs2_read_blocks(bucket->bu_inode, xb_blkno,
 274                                bucket->bu_blocks, bucket->bu_bhs, 0,
 275                                NULL);
 276         if (!rc) {
 277                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 278                                                  bucket->bu_bhs,
 279                                                  bucket->bu_blocks,
 280                                                  &bucket_xh(bucket)->xh_check);
 281                 if (rc)
 282                         mlog_errno(rc);
 283         }
 284
 285         if (rc)
 286                 ocfs2_xattr_bucket_relse(bucket);
 287         return rc;
 288 }
 289
 290 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 291                                              struct ocfs2_xattr_bucket *bucket,
 292                                              int type)
 293 {
 294         int i, rc = 0;
 295
 296         for (i = 0; i < bucket->bu_blocks; i++) {
 297                 rc = ocfs2_journal_access(handle, bucket->bu_inode,
 298                                           bucket->bu_bhs[i], type);
 299                 if (rc) {
 300                         mlog_errno(rc);
 301                         break;
 302                 }
 303         }
 304
 305         return rc;
 306 }
 307
 308 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 309                                              struct ocfs2_xattr_bucket *bucket)
 310 {
 311         int i;
 312
 313         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 314                                    bucket->bu_bhs, bucket->bu_blocks,
 315                                    &bucket_xh(bucket)->xh_check);
 316
 317         for (i = 0; i < bucket->bu_blocks; i++)
 318                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 319 }
 320
 321 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 322                                          struct ocfs2_xattr_bucket *src)
 323 {
 324         int i;
 325         int blocksize = src->bu_inode->i_sb->s_blocksize;
 326
 327         BUG_ON(dest->bu_blocks != src->bu_blocks);
 328         BUG_ON(dest->bu_inode != src->bu_inode);
 329
 330         for (i = 0; i < src->bu_blocks; i++) {
 331                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 332                        blocksize);
 333         }
 334 }
 335
 336 static int ocfs2_validate_xattr_block(struct super_block *sb,
 337                                       struct buffer_head *bh)
 338 {
 339         int rc;
 340         struct ocfs2_xattr_block *xb =
 341                 (struct ocfs2_xattr_block *)bh->b_data;
 342
 343         mlog(0, "Validating xattr block %llu\n",
 344              (unsigned long long)bh->b_blocknr);
 345
 346         BUG_ON(!buffer_uptodate(bh));
 347
 348         /*
 349          * If the ecc fails, we return the error but otherwise
 350          * leave the filesystem running.  We know any error is
 351          * local to this block.
 352          */
 353         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 354         if (rc)
 355                 return rc;
 356
 357         /*
 358          * Errors after here are fatal
 359          */
 360
 361         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 362                 ocfs2_error(sb,
 363                             "Extended attribute block #%llu has bad "
 364                             "signature %.*s",
 365                             (unsigned long long)bh->b_blocknr, 7,
 366                             xb->xb_signature);
 367                 return -EINVAL;
 368         }
 369
 370         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 371                 ocfs2_error(sb,
 372                             "Extended attribute block #%llu has an "
 373                             "invalid xb_blkno of %llu",
 374                             (unsigned long long)bh->b_blocknr,
 375                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
 376                 return -EINVAL;
 377         }
 378
 379         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 380                 ocfs2_error(sb,
 381                             "Extended attribute block #%llu has an invalid "
 382                             "xb_fs_generation of #%u",
 383                             (unsigned long long)bh->b_blocknr,
 384                             le32_to_cpu(xb->xb_fs_generation));
 385                 return -EINVAL;
 386         }
 387
 388         return 0;
 389 }
 390
 391 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 392                                   struct buffer_head **bh)
 393 {
 394         int rc;
 395         struct buffer_head *tmp = *bh;
 396
 397         rc = ocfs2_read_block(inode, xb_blkno, &tmp,
 398                               ocfs2_validate_xattr_block);
 399
 400         /* If ocfs2_read_block() got us a new bh, pass it up. */
 401         if (!rc && !*bh)
 402                 *bh = tmp;
 403
 404         return rc;
 405 }
 406
 407 static inline const char *ocfs2_xattr_prefix(int name_index)
 408 {
 409         struct xattr_handler *handler = NULL;
 410
 411         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 412                 handler = ocfs2_xattr_handler_map[name_index];
 413
 414         return handler ? handler->prefix : NULL;
 415 }
 416
 417 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 418                                  const char *name,
 419                                  int name_len)
 420 {
 421         /* Get hash value of uuid from super block */
 422         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 423         int i;
 424
 425         /* hash extended attribute name */
 426         for (i = 0; i < name_len; i++) {
 427                 hash = (hash << OCFS2_HASH_SHIFT) ^
 428                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 429                        *name++;
 430         }
 431
 432         return hash;
 433 }
 434
 435 /*
 436  * ocfs2_xattr_hash_entry()
 437  *
 438  * Compute the hash of an extended attribute.
 439  */
 440 static void ocfs2_xattr_hash_entry(struct inode *inode,
 441                                    struct ocfs2_xattr_header *header,
 442                                    struct ocfs2_xattr_entry *entry)
 443 {
 444         u32 hash = 0;
 445         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
 446
 447         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
 448         entry->xe_name_hash = cpu_to_le32(hash);
 449
 450         return;
 451 }
 452
 453 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 454 {
 455         int size = 0;
 456
 457         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
 458                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 459         else
 460                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 461         size += sizeof(struct ocfs2_xattr_entry);
 462
 463         return size;
 464 }
 465
 466 int ocfs2_calc_security_init(struct inode *dir,
 467                              struct ocfs2_security_xattr_info *si,
 468                              int *want_clusters,
 469                              int *xattr_credits,
 470                              struct ocfs2_alloc_context **xattr_ac)
 471 {
 472         int ret = 0;
 473         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 474         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 475                                                  si->value_len);
 476
 477         /*
 478          * The max space of security xattr taken inline is
 479          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 480          * So reserve one metadata block for it is ok.
 481          */
 482         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 483             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 484                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 485                 if (ret) {
 486                         mlog_errno(ret);
 487                         return ret;
 488                 }
 489                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 490         }
 491
 492         /* reserve clusters for xattr value which will be set in B tree*/
 493         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 494                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 495                                                             si->value_len);
 496
 497                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 498                                                            new_clusters);
 499                 *want_clusters += new_clusters;
 500         }
 501         return ret;
 502 }
 503
 504 int ocfs2_calc_xattr_init(struct inode *dir,
 505                           struct buffer_head *dir_bh,
 506                           int mode,
 507                           struct ocfs2_security_xattr_info *si,
 508                           int *want_clusters,
 509                           int *xattr_credits,
 510                           struct ocfs2_alloc_context **xattr_ac)
 511 {
 512         int ret = 0;
 513         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 514         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 515
 516         if (si->enable)
 517                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 518                                                      si->value_len);
 519
 520         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 521                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 522                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 523                                         "", NULL, 0);
 524                 if (acl_len > 0) {
 525                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 526                         if (S_ISDIR(mode))
 527                                 a_size <<= 1;
 528                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 529                         mlog_errno(ret);
 530                         return ret;
 531                 }
 532         }
 533
 534         if (!(s_size + a_size))
 535                 return ret;
 536
 537         /*
 538          * The max space of security xattr taken inline is
 539          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 540          * The max space of acl xattr taken inline is
 541          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 542          * when blocksize = 512, may reserve one more cluser for
 543          * xattr bucket, otherwise reserve one metadata block
 544          * for them is ok.
 545          */
 546         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 547             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 548                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 549                 if (ret) {
 550                         mlog_errno(ret);
 551                         return ret;
 552                 }
 553                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 554         }
 555
 556         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 557             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 558                 *want_clusters += 1;
 559                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 560         }
 561
 562         /*
 563          * reserve credits and clusters for xattrs which has large value
 564          * and have to be set outside
 565          */
 566         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 567                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 568                                                         si->value_len);
 569                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 570                                                            new_clusters);
 571                 *want_clusters += new_clusters;
 572         }
 573         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 574             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 575                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 576                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 577                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 578                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 579                                                            new_clusters);
 580                 *want_clusters += new_clusters;
 581         }
 582
 583         return ret;
 584 }
 585
 586 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 587                                          u32 clusters_to_add,
 588                                          struct ocfs2_xattr_value_buf *vb,
 589                                          struct ocfs2_xattr_set_ctxt *ctxt)
 590 {
 591         int status = 0;
 592         handle_t *handle = ctxt->handle;
 593         enum ocfs2_alloc_restarted why;
 594         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 595         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 596         struct ocfs2_extent_tree et;
 597
 598         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
 599
 600         ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
 601
 602         status = vb->vb_access(handle, inode, vb->vb_bh,
 603                               OCFS2_JOURNAL_ACCESS_WRITE);
 604         if (status < 0) {
 605                 mlog_errno(status);
 606                 goto leave;
 607         }
 608
 609         prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 610         status = ocfs2_add_clusters_in_btree(osb,
 611                                              inode,
 612                                              &logical_start,
 613                                              clusters_to_add,
 614                                              0,
 615                                              &et,
 616                                              handle,
 617                                              ctxt->data_ac,
 618                                              ctxt->meta_ac,
 619                                              &why);
 620         if (status < 0) {
 621                 mlog_errno(status);
 622                 goto leave;
 623         }
 624
 625         status = ocfs2_journal_dirty(handle, vb->vb_bh);
 626         if (status < 0) {
 627                 mlog_errno(status);
 628                 goto leave;
 629         }
 630
 631         clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
 632
 633         /*
 634          * We should have already allocated enough space before the transaction,
 635          * so no need to restart.
 636          */
 637         BUG_ON(why != RESTART_NONE || clusters_to_add);
 638
 639 leave:
 640
 641         return status;
 642 }
 643
 644 static int __ocfs2_remove_xattr_range(struct inode *inode,
 645                                       struct ocfs2_xattr_value_buf *vb,
 646                                       u32 cpos, u32 phys_cpos, u32 len,
 647                                       struct ocfs2_xattr_set_ctxt *ctxt)
 648 {
 649         int ret;
 650         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 651         handle_t *handle = ctxt->handle;
 652         struct ocfs2_extent_tree et;
 653
 654         ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
 655
 656         ret = vb->vb_access(handle, inode, vb->vb_bh,
 657                             OCFS2_JOURNAL_ACCESS_WRITE);
 658         if (ret) {
 659                 mlog_errno(ret);
 660                 goto out;
 661         }
 662
 663         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
 664                                   &ctxt->dealloc);
 665         if (ret) {
 666                 mlog_errno(ret);
 667                 goto out;
 668         }
 669
 670         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 671
 672         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
 673         if (ret) {
 674                 mlog_errno(ret);
 675                 goto out;
 676         }
 677
 678         ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
 679         if (ret)
 680                 mlog_errno(ret);
 681
 682 out:
 683         return ret;
 684 }
 685
 686 static int ocfs2_xattr_shrink_size(struct inode *inode,
 687                                    u32 old_clusters,
 688                                    u32 new_clusters,
 689                                    struct ocfs2_xattr_value_buf *vb,
 690                                    struct ocfs2_xattr_set_ctxt *ctxt)
 691 {
 692         int ret = 0;
 693         u32 trunc_len, cpos, phys_cpos, alloc_size;
 694         u64 block;
 695
 696         if (old_clusters <= new_clusters)
 697                 return 0;
 698
 699         cpos = new_clusters;
 700         trunc_len = old_clusters - new_clusters;
 701         while (trunc_len) {
 702                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 703                                                &alloc_size,
 704                                                &vb->vb_xv->xr_list);
 705                 if (ret) {
 706                         mlog_errno(ret);
 707                         goto out;
 708                 }
 709
 710                 if (alloc_size > trunc_len)
 711                         alloc_size = trunc_len;
 712
 713                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 714                                                  phys_cpos, alloc_size,
 715                                                  ctxt);
 716                 if (ret) {
 717                         mlog_errno(ret);
 718                         goto out;
 719                 }
 720
 721                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 722                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
 723                                                        alloc_size);
 724                 cpos += alloc_size;
 725                 trunc_len -= alloc_size;
 726         }
 727
 728 out:
 729         return ret;
 730 }
 731
 732 static int ocfs2_xattr_value_truncate(struct inode *inode,
 733                                       struct ocfs2_xattr_value_buf *vb,
 734                                       int len,
 735                                       struct ocfs2_xattr_set_ctxt *ctxt)
 736 {
 737         int ret;
 738         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 739         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 740
 741         if (new_clusters == old_clusters)
 742                 return 0;
 743
 744         if (new_clusters > old_clusters)
 745                 ret = ocfs2_xattr_extend_allocation(inode,
 746                                                     new_clusters - old_clusters,
 747                                                     vb, ctxt);
 748         else
 749                 ret = ocfs2_xattr_shrink_size(inode,
 750                                               old_clusters, new_clusters,
 751                                               vb, ctxt);
 752
 753         return ret;
 754 }
 755
 756 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 757                                   size_t *result, const char *prefix,
 758                                   const char *name, int name_len)
 759 {
 760         char *p = buffer + *result;
 761         int prefix_len = strlen(prefix);
 762         int total_len = prefix_len + name_len + 1;
 763
 764         *result += total_len;
 765
 766         /* we are just looking for how big our buffer needs to be */
 767         if (!size)
 768                 return 0;
 769
 770         if (*result > size)
 771                 return -ERANGE;
 772
 773         memcpy(p, prefix, prefix_len);
 774         memcpy(p + prefix_len, name, name_len);
 775         p[prefix_len + name_len] = '\0';
 776
 777         return 0;
 778 }
 779
 780 static int ocfs2_xattr_list_entries(struct inode *inode,
 781                                     struct ocfs2_xattr_header *header,
 782                                     char *buffer, size_t buffer_size)
 783 {
 784         size_t result = 0;
 785         int i, type, ret;
 786         const char *prefix, *name;
 787
 788         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 789                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 790                 type = ocfs2_xattr_get_type(entry);
 791                 prefix = ocfs2_xattr_prefix(type);
 792
 793                 if (prefix) {
 794                         name = (const char *)header +
 795                                 le16_to_cpu(entry->xe_name_offset);
 796
 797                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 798                                                      &result, prefix, name,
 799                                                      entry->xe_name_len);
 800                         if (ret)
 801                                 return ret;
 802                 }
 803         }
 804
 805         return result;
 806 }
 807
 808 static int ocfs2_xattr_ibody_list(struct inode *inode,
 809                                   struct ocfs2_dinode *di,
 810                                   char *buffer,
 811                                   size_t buffer_size)
 812 {
 813         struct ocfs2_xattr_header *header = NULL;
 814         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 815         int ret = 0;
 816
 817         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 818                 return ret;
 819
 820         header = (struct ocfs2_xattr_header *)
 821                  ((void *)di + inode->i_sb->s_blocksize -
 822                  le16_to_cpu(di->i_xattr_inline_size));
 823
 824         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 825
 826         return ret;
 827 }
 828
 829 static int ocfs2_xattr_block_list(struct inode *inode,
 830                                   struct ocfs2_dinode *di,
 831                                   char *buffer,
 832                                   size_t buffer_size)
 833 {
 834         struct buffer_head *blk_bh = NULL;
 835         struct ocfs2_xattr_block *xb;
 836         int ret = 0;
 837
 838         if (!di->i_xattr_loc)
 839                 return ret;
 840
 841         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
 842                                      &blk_bh);
 843         if (ret < 0) {
 844                 mlog_errno(ret);
 845                 return ret;
 846         }
 847
 848         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 849         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 850                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 851                 ret = ocfs2_xattr_list_entries(inode, header,
 852                                                buffer, buffer_size);
 853         } else {
 854                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
 855                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
 856                                                    buffer, buffer_size);
 857         }
 858
 859         brelse(blk_bh);
 860
 861         return ret;
 862 }
 863
 864 ssize_t ocfs2_listxattr(struct dentry *dentry,
 865                         char *buffer,
 866                         size_t size)
 867 {
 868         int ret = 0, i_ret = 0, b_ret = 0;
 869         struct buffer_head *di_bh = NULL;
 870         struct ocfs2_dinode *di = NULL;
 871         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
 872
 873         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
 874                 return -EOPNOTSUPP;
 875
 876         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 877                 return ret;
 878
 879         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
 880         if (ret < 0) {
 881                 mlog_errno(ret);
 882                 return ret;
 883         }
 884
 885         di = (struct ocfs2_dinode *)di_bh->b_data;
 886
 887         down_read(&oi->ip_xattr_sem);
 888         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
 889         if (i_ret < 0)
 890                 b_ret = 0;
 891         else {
 892                 if (buffer) {
 893                         buffer += i_ret;
 894                         size -= i_ret;
 895                 }
 896                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
 897                                                buffer, size);
 898                 if (b_ret < 0)
 899                         i_ret = 0;
 900         }
 901         up_read(&oi->ip_xattr_sem);
 902         ocfs2_inode_unlock(dentry->d_inode, 0);
 903
 904         brelse(di_bh);
 905
 906         return i_ret + b_ret;
 907 }
 908
 909 static int ocfs2_xattr_find_entry(int name_index,
 910                                   const char *name,
 911                                   struct ocfs2_xattr_search *xs)
 912 {
 913         struct ocfs2_xattr_entry *entry;
 914         size_t name_len;
 915         int i, cmp = 1;
 916
 917         if (name == NULL)
 918                 return -EINVAL;
 919
 920         name_len = strlen(name);
 921         entry = xs->here;
 922         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
 923                 cmp = name_index - ocfs2_xattr_get_type(entry);
 924                 if (!cmp)
 925                         cmp = name_len - entry->xe_name_len;
 926                 if (!cmp)
 927                         cmp = memcmp(name, (xs->base +
 928                                      le16_to_cpu(entry->xe_name_offset)),
 929                                      name_len);
 930                 if (cmp == 0)
 931                         break;
 932                 entry += 1;
 933         }
 934         xs->here = entry;
 935
 936         return cmp ? -ENODATA : 0;
 937 }
 938
 939 static int ocfs2_xattr_get_value_outside(struct inode *inode,
 940                                          struct ocfs2_xattr_value_root *xv,
 941                                          void *buffer,
 942                                          size_t len)
 943 {
 944         u32 cpos, p_cluster, num_clusters, bpc, clusters;
 945         u64 blkno;
 946         int i, ret = 0;
 947         size_t cplen, blocksize;
 948         struct buffer_head *bh = NULL;
 949         struct ocfs2_extent_list *el;
 950
 951         el = &xv->xr_list;
 952         clusters = le32_to_cpu(xv->xr_clusters);
 953         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 954         blocksize = inode->i_sb->s_blocksize;
 955
 956         cpos = 0;
 957         while (cpos < clusters) {
 958                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 959                                                &num_clusters, el);
 960                 if (ret) {
 961                         mlog_errno(ret);
 962                         goto out;
 963                 }
 964
 965                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 966                 /* Copy ocfs2_xattr_value */
 967                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 968                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
 969                         if (ret) {
 970                                 mlog_errno(ret);
 971                                 goto out;
 972                         }
 973
 974                         cplen = len >= blocksize ? blocksize : len;
 975                         memcpy(buffer, bh->b_data, cplen);
 976                         len -= cplen;
 977                         buffer += cplen;
 978
 979                         brelse(bh);
 980                         bh = NULL;
 981                         if (len == 0)
 982                                 break;
 983                 }
 984                 cpos += num_clusters;
 985         }
 986 out:
 987         return ret;
 988 }
 989
 990 static int ocfs2_xattr_ibody_get(struct inode *inode,
 991                                  int name_index,
 992                                  const char *name,
 993                                  void *buffer,
 994                                  size_t buffer_size,
 995                                  struct ocfs2_xattr_search *xs)
 996 {
 997         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 998         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
 999         struct ocfs2_xattr_value_root *xv;
1000         size_t size;
1001         int ret = 0;
1002
1003         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1004                 return -ENODATA;
1005
1006         xs->end = (void *)di + inode->i_sb->s_blocksize;
1007         xs->header = (struct ocfs2_xattr_header *)
1008                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1009         xs->base = (void *)xs->header;
1010         xs->here = xs->header->xh_entries;
1011
1012         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1013         if (ret)
1014                 return ret;
1015         size = le64_to_cpu(xs->here->xe_value_size);
1016         if (buffer) {
1017                 if (size > buffer_size)
1018                         return -ERANGE;
1019                 if (ocfs2_xattr_is_local(xs->here)) {
1020                         memcpy(buffer, (void *)xs->base +
1021                                le16_to_cpu(xs->here->xe_name_offset) +
1022                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1023                 } else {
1024                         xv = (struct ocfs2_xattr_value_root *)
1025                                 (xs->base + le16_to_cpu(
1026                                  xs->here->xe_name_offset) +
1027                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1028                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1029                                                             buffer, size);
1030                         if (ret < 0) {
1031                                 mlog_errno(ret);
1032                                 return ret;
1033                         }
1034                 }
1035         }
1036
1037         return size;
1038 }
1039
1040 static int ocfs2_xattr_block_get(struct inode *inode,
1041                                  int name_index,
1042                                  const char *name,
1043                                  void *buffer,
1044                                  size_t buffer_size,
1045                                  struct ocfs2_xattr_search *xs)
1046 {
1047         struct ocfs2_xattr_block *xb;
1048         struct ocfs2_xattr_value_root *xv;
1049         size_t size;
1050         int ret = -ENODATA, name_offset, name_len, block_off, i;
1051
1052         xs->bucket = ocfs2_xattr_bucket_new(inode);
1053         if (!xs->bucket) {
1054                 ret = -ENOMEM;
1055                 mlog_errno(ret);
1056                 goto cleanup;
1057         }
1058
1059         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1060         if (ret) {
1061                 mlog_errno(ret);
1062                 goto cleanup;
1063         }
1064
1065         if (xs->not_found) {
1066                 ret = -ENODATA;
1067                 goto cleanup;
1068         }
1069
1070         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1071         size = le64_to_cpu(xs->here->xe_value_size);
1072         if (buffer) {
1073                 ret = -ERANGE;
1074                 if (size > buffer_size)
1075                         goto cleanup;
1076
1077                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1078                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1079                 i = xs->here - xs->header->xh_entries;
1080
1081                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1082                         ret = ocfs2_xattr_bucket_get_name_value(inode,
1083                                                                 bucket_xh(xs->bucket),
1084                                                                 i,
1085                                                                 &block_off,
1086                                                                 &name_offset);
1087                         xs->base = bucket_block(xs->bucket, block_off);
1088                 }
1089                 if (ocfs2_xattr_is_local(xs->here)) {
1090                         memcpy(buffer, (void *)xs->base +
1091                                name_offset + name_len, size);
1092                 } else {
1093                         xv = (struct ocfs2_xattr_value_root *)
1094                                 (xs->base + name_offset + name_len);
1095                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1096                                                             buffer, size);
1097                         if (ret < 0) {
1098                                 mlog_errno(ret);
1099                                 goto cleanup;
1100                         }
1101                 }
1102         }
1103         ret = size;
1104 cleanup:
1105         ocfs2_xattr_bucket_free(xs->bucket);
1106
1107         brelse(xs->xattr_bh);
1108         xs->xattr_bh = NULL;
1109         return ret;
1110 }
1111
1112 int ocfs2_xattr_get_nolock(struct inode *inode,
1113                            struct buffer_head *di_bh,
1114                            int name_index,
1115                            const char *name,
1116                            void *buffer,
1117                            size_t buffer_size)
1118 {
1119         int ret;
1120         struct ocfs2_dinode *di = NULL;
1121         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1122         struct ocfs2_xattr_search xis = {
1123                 .not_found = -ENODATA,
1124         };
1125         struct ocfs2_xattr_search xbs = {
1126                 .not_found = -ENODATA,
1127         };
1128
1129         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1130                 return -EOPNOTSUPP;
1131
1132         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1133                 ret = -ENODATA;
1134
1135         xis.inode_bh = xbs.inode_bh = di_bh;
1136         di = (struct ocfs2_dinode *)di_bh->b_data;
1137
1138         down_read(&oi->ip_xattr_sem);
1139         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1140                                     buffer_size, &xis);
1141         if (ret == -ENODATA && di->i_xattr_loc)
1142                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1143                                             buffer_size, &xbs);
1144         up_read(&oi->ip_xattr_sem);
1145
1146         return ret;
1147 }
1148
1149 /* ocfs2_xattr_get()
1150  *
1151  * Copy an extended attribute into the buffer provided.
1152  * Buffer is NULL to compute the size of buffer required.
1153  */
1154 static int ocfs2_xattr_get(struct inode *inode,
1155                            int name_index,
1156                            const char *name,
1157                            void *buffer,
1158                            size_t buffer_size)
1159 {
1160         int ret;
1161         struct buffer_head *di_bh = NULL;
1162
1163         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1164         if (ret < 0) {
1165                 mlog_errno(ret);
1166                 return ret;
1167         }
1168         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1169                                      name, buffer, buffer_size);
1170
1171         ocfs2_inode_unlock(inode, 0);
1172
1173         brelse(di_bh);
1174
1175         return ret;
1176 }
1177
1178 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1179                                            handle_t *handle,
1180                                            struct ocfs2_xattr_value_root *xv,
1181                                            const void *value,
1182                                            int value_len)
1183 {
1184         int ret = 0, i, cp_len;
1185         u16 blocksize = inode->i_sb->s_blocksize;
1186         u32 p_cluster, num_clusters;
1187         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1188         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1189         u64 blkno;
1190         struct buffer_head *bh = NULL;
1191
1192         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1193
1194         while (cpos < clusters) {
1195                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1196                                                &num_clusters, &xv->xr_list);
1197                 if (ret) {
1198                         mlog_errno(ret);
1199                         goto out;
1200                 }
1201
1202                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1203
1204                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1205                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
1206                         if (ret) {
1207                                 mlog_errno(ret);
1208                                 goto out;
1209                         }
1210
1211                         ret = ocfs2_journal_access(handle,
1212                                                    inode,
1213                                                    bh,
1214                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1215                         if (ret < 0) {
1216                                 mlog_errno(ret);
1217                                 goto out;
1218                         }
1219
1220                         cp_len = value_len > blocksize ? blocksize : value_len;
1221                         memcpy(bh->b_data, value, cp_len);
1222                         value_len -= cp_len;
1223                         value += cp_len;
1224                         if (cp_len < blocksize)
1225                                 memset(bh->b_data + cp_len, 0,
1226                                        blocksize - cp_len);
1227
1228                         ret = ocfs2_journal_dirty(handle, bh);
1229                         if (ret < 0) {
1230                                 mlog_errno(ret);
1231                                 goto out;
1232                         }
1233                         brelse(bh);
1234                         bh = NULL;
1235
1236                         /*
1237                          * XXX: do we need to empty all the following
1238                          * blocks in this cluster?
1239                          */
1240                         if (!value_len)
1241                                 break;
1242                 }
1243                 cpos += num_clusters;
1244         }
1245 out:
1246         brelse(bh);
1247
1248         return ret;
1249 }
1250
1251 static int ocfs2_xattr_cleanup(struct inode *inode,
1252                                handle_t *handle,
1253                                struct ocfs2_xattr_info *xi,
1254                                struct ocfs2_xattr_search *xs,
1255                                struct ocfs2_xattr_value_buf *vb,
1256                                size_t offs)
1257 {
1258         int ret = 0;
1259         size_t name_len = strlen(xi->name);
1260         void *val = xs->base + offs;
1261         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1262
1263         ret = vb->vb_access(handle, inode, vb->vb_bh,
1264                             OCFS2_JOURNAL_ACCESS_WRITE);
1265         if (ret) {
1266                 mlog_errno(ret);
1267                 goto out;
1268         }
1269         /* Decrease xattr count */
1270         le16_add_cpu(&xs->header->xh_count, -1);
1271         /* Remove the xattr entry and tree root which has already be set*/
1272         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1273         memset(val, 0, size);
1274
1275         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1276         if (ret < 0)
1277                 mlog_errno(ret);
1278 out:
1279         return ret;
1280 }
1281
1282 static int ocfs2_xattr_update_entry(struct inode *inode,
1283                                     handle_t *handle,
1284                                     struct ocfs2_xattr_info *xi,
1285                                     struct ocfs2_xattr_search *xs,
1286                                     struct ocfs2_xattr_value_buf *vb,
1287                                     size_t offs)
1288 {
1289         int ret;
1290
1291         ret = vb->vb_access(handle, inode, vb->vb_bh,
1292                             OCFS2_JOURNAL_ACCESS_WRITE);
1293         if (ret) {
1294                 mlog_errno(ret);
1295                 goto out;
1296         }
1297
1298         xs->here->xe_name_offset = cpu_to_le16(offs);
1299         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1300         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1301                 ocfs2_xattr_set_local(xs->here, 1);
1302         else
1303                 ocfs2_xattr_set_local(xs->here, 0);
1304         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1305
1306         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1307         if (ret < 0)
1308                 mlog_errno(ret);
1309 out:
1310         return ret;
1311 }
1312
1313 /*
1314  * ocfs2_xattr_set_value_outside()
1315  *
1316  * Set large size value in B tree.
1317  */
1318 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1319                                          struct ocfs2_xattr_info *xi,
1320                                          struct ocfs2_xattr_search *xs,
1321                                          struct ocfs2_xattr_set_ctxt *ctxt,
1322                                          struct ocfs2_xattr_value_buf *vb,
1323                                          size_t offs)
1324 {
1325         size_t name_len = strlen(xi->name);
1326         void *val = xs->base + offs;
1327         struct ocfs2_xattr_value_root *xv = NULL;
1328         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1329         int ret = 0;
1330
1331         memset(val, 0, size);
1332         memcpy(val, xi->name, name_len);
1333         xv = (struct ocfs2_xattr_value_root *)
1334                 (val + OCFS2_XATTR_SIZE(name_len));
1335         xv->xr_clusters = 0;
1336         xv->xr_last_eb_blk = 0;
1337         xv->xr_list.l_tree_depth = 0;
1338         xv->xr_list.l_count = cpu_to_le16(1);
1339         xv->xr_list.l_next_free_rec = 0;
1340         vb->vb_xv = xv;
1341
1342         ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1343         if (ret < 0) {
1344                 mlog_errno(ret);
1345                 return ret;
1346         }
1347         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1348         if (ret < 0) {
1349                 mlog_errno(ret);
1350                 return ret;
1351         }
1352         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
1353                                               xi->value, xi->value_len);
1354         if (ret < 0)
1355                 mlog_errno(ret);
1356
1357         return ret;
1358 }
1359
1360 /*
1361  * ocfs2_xattr_set_entry_local()
1362  *
1363  * Set, replace or remove extended attribute in local.
1364  */
1365 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1366                                         struct ocfs2_xattr_info *xi,
1367                                         struct ocfs2_xattr_search *xs,
1368                                         struct ocfs2_xattr_entry *last,
1369                                         size_t min_offs)
1370 {
1371         size_t name_len = strlen(xi->name);
1372         int i;
1373
1374         if (xi->value && xs->not_found) {
1375                 /* Insert the new xattr entry. */
1376                 le16_add_cpu(&xs->header->xh_count, 1);
1377                 ocfs2_xattr_set_type(last, xi->name_index);
1378                 ocfs2_xattr_set_local(last, 1);
1379                 last->xe_name_len = name_len;
1380         } else {
1381                 void *first_val;
1382                 void *val;
1383                 size_t offs, size;
1384
1385                 first_val = xs->base + min_offs;
1386                 offs = le16_to_cpu(xs->here->xe_name_offset);
1387                 val = xs->base + offs;
1388
1389                 if (le64_to_cpu(xs->here->xe_value_size) >
1390                     OCFS2_XATTR_INLINE_SIZE)
1391                         size = OCFS2_XATTR_SIZE(name_len) +
1392                                 OCFS2_XATTR_ROOT_SIZE;
1393                 else
1394                         size = OCFS2_XATTR_SIZE(name_len) +
1395                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1396
1397                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1398                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1399                         /* The old and the new value have the
1400                            same size. Just replace the value. */
1401                         ocfs2_xattr_set_local(xs->here, 1);
1402                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1403                         /* Clear value bytes. */
1404                         memset(val + OCFS2_XATTR_SIZE(name_len),
1405                                0,
1406                                OCFS2_XATTR_SIZE(xi->value_len));
1407                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1408                                xi->value,
1409                                xi->value_len);
1410                         return;
1411                 }
1412                 /* Remove the old name+value. */
1413                 memmove(first_val + size, first_val, val - first_val);
1414                 memset(first_val, 0, size);
1415                 xs->here->xe_name_hash = 0;
1416                 xs->here->xe_name_offset = 0;
1417                 ocfs2_xattr_set_local(xs->here, 1);
1418                 xs->here->xe_value_size = 0;
1419
1420                 min_offs += size;
1421
1422                 /* Adjust all value offsets. */
1423                 last = xs->header->xh_entries;
1424                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1425                         size_t o = le16_to_cpu(last->xe_name_offset);
1426
1427                         if (o < offs)
1428                                 last->xe_name_offset = cpu_to_le16(o + size);
1429                         last += 1;
1430                 }
1431
1432                 if (!xi->value) {
1433                         /* Remove the old entry. */
1434                         last -= 1;
1435                         memmove(xs->here, xs->here + 1,
1436                                 (void *)last - (void *)xs->here);
1437                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1438                         le16_add_cpu(&xs->header->xh_count, -1);
1439                 }
1440         }
1441         if (xi->value) {
1442                 /* Insert the new name+value. */
1443                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1444                                 OCFS2_XATTR_SIZE(xi->value_len);
1445                 void *val = xs->base + min_offs - size;
1446
1447                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1448                 memset(val, 0, size);
1449                 memcpy(val, xi->name, name_len);
1450                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1451                        xi->value,
1452                        xi->value_len);
1453                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1454                 ocfs2_xattr_set_local(xs->here, 1);
1455                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1456         }
1457
1458         return;
1459 }
1460
1461 /*
1462  * ocfs2_xattr_set_entry()
1463  *
1464  * Set extended attribute entry into inode or block.
1465  *
1466  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1467  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1468  * then set value in B tree with set_value_outside().
1469  */
1470 static int ocfs2_xattr_set_entry(struct inode *inode,
1471                                  struct ocfs2_xattr_info *xi,
1472                                  struct ocfs2_xattr_search *xs,
1473                                  struct ocfs2_xattr_set_ctxt *ctxt,
1474                                  int flag)
1475 {
1476         struct ocfs2_xattr_entry *last;
1477         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1478         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1479         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1480         size_t size_l = 0;
1481         handle_t *handle = ctxt->handle;
1482         int free, i, ret;
1483         struct ocfs2_xattr_info xi_l = {
1484                 .name_index = xi->name_index,
1485                 .name = xi->name,
1486                 .value = xi->value,
1487                 .value_len = xi->value_len,
1488         };
1489         struct ocfs2_xattr_value_buf vb = {
1490                 .vb_bh = xs->xattr_bh,
1491                 .vb_access = ocfs2_journal_access_di,
1492         };
1493
1494         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1495                 BUG_ON(xs->xattr_bh == xs->inode_bh);
1496                 vb.vb_access = ocfs2_journal_access_xb;
1497         } else
1498                 BUG_ON(xs->xattr_bh != xs->inode_bh);
1499
1500         /* Compute min_offs, last and free space. */
1501         last = xs->header->xh_entries;
1502
1503         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1504                 size_t offs = le16_to_cpu(last->xe_name_offset);
1505                 if (offs < min_offs)
1506                         min_offs = offs;
1507                 last += 1;
1508         }
1509
1510         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1511         if (free < 0)
1512                 return -EIO;
1513
1514         if (!xs->not_found) {
1515                 size_t size = 0;
1516                 if (ocfs2_xattr_is_local(xs->here))
1517                         size = OCFS2_XATTR_SIZE(name_len) +
1518                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1519                 else
1520                         size = OCFS2_XATTR_SIZE(name_len) +
1521                                 OCFS2_XATTR_ROOT_SIZE;
1522                 free += (size + sizeof(struct ocfs2_xattr_entry));
1523         }
1524         /* Check free space in inode or block */
1525         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1526                 if (free < sizeof(struct ocfs2_xattr_entry) +
1527                            OCFS2_XATTR_SIZE(name_len) +
1528                            OCFS2_XATTR_ROOT_SIZE) {
1529                         ret = -ENOSPC;
1530                         goto out;
1531                 }
1532                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1533                 xi_l.value = (void *)&def_xv;
1534                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1535         } else if (xi->value) {
1536                 if (free < sizeof(struct ocfs2_xattr_entry) +
1537                            OCFS2_XATTR_SIZE(name_len) +
1538                            OCFS2_XATTR_SIZE(xi->value_len)) {
1539                         ret = -ENOSPC;
1540                         goto out;
1541                 }
1542         }
1543
1544         if (!xs->not_found) {
1545                 /* For existing extended attribute */
1546                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1547                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1548                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1549                 void *val = xs->base + offs;
1550
1551                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1552                         /* Replace existing local xattr with tree root */
1553                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1554                                                             ctxt, &vb, offs);
1555                         if (ret < 0)
1556                                 mlog_errno(ret);
1557                         goto out;
1558                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1559                         /* For existing xattr which has value outside */
1560                         vb.vb_xv = (struct ocfs2_xattr_value_root *)
1561                                 (val + OCFS2_XATTR_SIZE(name_len));
1562
1563                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1564                                 /*
1565                                  * If new value need set outside also,
1566                                  * first truncate old value to new value,
1567                                  * then set new value with set_value_outside().
1568                                  */
1569                                 ret = ocfs2_xattr_value_truncate(inode,
1570                                                                  &vb,
1571                                                                  xi->value_len,
1572                                                                  ctxt);
1573                                 if (ret < 0) {
1574                                         mlog_errno(ret);
1575                                         goto out;
1576                                 }
1577
1578                                 ret = ocfs2_xattr_update_entry(inode,
1579                                                                handle,
1580                                                                xi,
1581                                                                xs,
1582                                                                &vb,
1583                                                                offs);
1584                                 if (ret < 0) {
1585                                         mlog_errno(ret);
1586                                         goto out;
1587                                 }
1588
1589                                 ret = __ocfs2_xattr_set_value_outside(inode,
1590                                                                 handle,
1591                                                                 vb.vb_xv,
1592                                                                 xi->value,
1593                                                                 xi->value_len);
1594                                 if (ret < 0)
1595                                         mlog_errno(ret);
1596                                 goto out;
1597                         } else {
1598                                 /*
1599                                  * If new value need set in local,
1600                                  * just trucate old value to zero.
1601                                  */
1602                                  ret = ocfs2_xattr_value_truncate(inode,
1603                                                                   &vb,
1604                                                                   0,
1605                                                                   ctxt);
1606                                 if (ret < 0)
1607                                         mlog_errno(ret);
1608                         }
1609                 }
1610         }
1611
1612         ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
1613                                       OCFS2_JOURNAL_ACCESS_WRITE);
1614         if (ret) {
1615                 mlog_errno(ret);
1616                 goto out;
1617         }
1618
1619         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1620                 ret = vb.vb_access(handle, inode, vb.vb_bh,
1621                                    OCFS2_JOURNAL_ACCESS_WRITE);
1622                 if (ret) {
1623                         mlog_errno(ret);
1624                         goto out;
1625                 }
1626         }
1627
1628         /*
1629          * Set value in local, include set tree root in local.
1630          * This is the first step for value size >INLINE_SIZE.
1631          */
1632         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1633
1634         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1635                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1636                 if (ret < 0) {
1637                         mlog_errno(ret);
1638                         goto out;
1639                 }
1640         }
1641
1642         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1643             (flag & OCFS2_INLINE_XATTR_FL)) {
1644                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1645                 unsigned int xattrsize = osb->s_xattr_inline_size;
1646
1647                 /*
1648                  * Adjust extent record count or inline data size
1649                  * to reserve space for extended attribute.
1650                  */
1651                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1652                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1653                         le16_add_cpu(&idata->id_count, -xattrsize);
1654                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1655                         struct ocfs2_extent_list *el = &di->id2.i_list;
1656                         le16_add_cpu(&el->l_count, -(xattrsize /
1657                                         sizeof(struct ocfs2_extent_rec)));
1658                 }
1659                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1660         }
1661         /* Update xattr flag */
1662         spin_lock(&oi->ip_lock);
1663         oi->ip_dyn_features |= flag;
1664         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1665         spin_unlock(&oi->ip_lock);
1666
1667         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1668         if (ret < 0)
1669                 mlog_errno(ret);
1670
1671         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1672                 /*
1673                  * Set value outside in B tree.
1674                  * This is the second step for value size > INLINE_SIZE.
1675                  */
1676                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1677                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1678                                                     &vb, offs);
1679                 if (ret < 0) {
1680                         int ret2;
1681
1682                         mlog_errno(ret);
1683                         /*
1684                          * If set value outside failed, we have to clean
1685                          * the junk tree root we have already set in local.
1686                          */
1687                         ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1688                                                    xi, xs, &vb, offs);
1689                         if (ret2 < 0)
1690                                 mlog_errno(ret2);
1691                 }
1692         }
1693 out:
1694         return ret;
1695 }
1696
1697 static int ocfs2_remove_value_outside(struct inode*inode,
1698                                       struct ocfs2_xattr_value_buf *vb,
1699                                       struct ocfs2_xattr_header *header)
1700 {
1701         int ret = 0, i;
1702         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1703         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1704
1705         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1706
1707         ctxt.handle = ocfs2_start_trans(osb,
1708                                         ocfs2_remove_extent_credits(osb->sb));
1709         if (IS_ERR(ctxt.handle)) {
1710                 ret = PTR_ERR(ctxt.handle);
1711                 mlog_errno(ret);
1712                 goto out;
1713         }
1714
1715         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1716                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1717
1718                 if (!ocfs2_xattr_is_local(entry)) {
1719                         void *val;
1720
1721                         val = (void *)header +
1722                                 le16_to_cpu(entry->xe_name_offset);
1723                         vb->vb_xv = (struct ocfs2_xattr_value_root *)
1724                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1725                         ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1726                         if (ret < 0) {
1727                                 mlog_errno(ret);
1728                                 break;
1729                         }
1730                 }
1731         }
1732
1733         ocfs2_commit_trans(osb, ctxt.handle);
1734         ocfs2_schedule_truncate_log_flush(osb, 1);
1735         ocfs2_run_deallocs(osb, &ctxt.dealloc);
1736 out:
1737         return ret;
1738 }
1739
1740 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1741                                     struct buffer_head *di_bh)
1742 {
1743
1744         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1745         struct ocfs2_xattr_header *header;
1746         int ret;
1747         struct ocfs2_xattr_value_buf vb = {
1748                 .vb_bh = di_bh,
1749                 .vb_access = ocfs2_journal_access_di,
1750         };
1751
1752         header = (struct ocfs2_xattr_header *)
1753                  ((void *)di + inode->i_sb->s_blocksize -
1754                  le16_to_cpu(di->i_xattr_inline_size));
1755
1756         ret = ocfs2_remove_value_outside(inode, &vb, header);
1757
1758         return ret;
1759 }
1760
1761 static int ocfs2_xattr_block_remove(struct inode *inode,
1762                                     struct buffer_head *blk_bh)
1763 {
1764         struct ocfs2_xattr_block *xb;
1765         int ret = 0;
1766         struct ocfs2_xattr_value_buf vb = {
1767                 .vb_bh = blk_bh,
1768                 .vb_access = ocfs2_journal_access_xb,
1769         };
1770
1771         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1772         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1773                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1774                 ret = ocfs2_remove_value_outside(inode, &vb, header);
1775         } else
1776                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1777
1778         return ret;
1779 }
1780
1781 static int ocfs2_xattr_free_block(struct inode *inode,
1782                                   u64 block)
1783 {
1784         struct inode *xb_alloc_inode;
1785         struct buffer_head *xb_alloc_bh = NULL;
1786         struct buffer_head *blk_bh = NULL;
1787         struct ocfs2_xattr_block *xb;
1788         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1789         handle_t *handle;
1790         int ret = 0;
1791         u64 blk, bg_blkno;
1792         u16 bit;
1793
1794         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1795         if (ret < 0) {
1796                 mlog_errno(ret);
1797                 goto out;
1798         }
1799
1800         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1801         if (ret < 0) {
1802                 mlog_errno(ret);
1803                 goto out;
1804         }
1805
1806         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1807         blk = le64_to_cpu(xb->xb_blkno);
1808         bit = le16_to_cpu(xb->xb_suballoc_bit);
1809         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1810
1811         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1812                                 EXTENT_ALLOC_SYSTEM_INODE,
1813                                 le16_to_cpu(xb->xb_suballoc_slot));
1814         if (!xb_alloc_inode) {
1815                 ret = -ENOMEM;
1816                 mlog_errno(ret);
1817                 goto out;
1818         }
1819         mutex_lock(&xb_alloc_inode->i_mutex);
1820
1821         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1822         if (ret < 0) {
1823                 mlog_errno(ret);
1824                 goto out_mutex;
1825         }
1826
1827         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1828         if (IS_ERR(handle)) {
1829                 ret = PTR_ERR(handle);
1830                 mlog_errno(ret);
1831                 goto out_unlock;
1832         }
1833
1834         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1835                                        bit, bg_blkno, 1);
1836         if (ret < 0)
1837                 mlog_errno(ret);
1838
1839         ocfs2_commit_trans(osb, handle);
1840 out_unlock:
1841         ocfs2_inode_unlock(xb_alloc_inode, 1);
1842         brelse(xb_alloc_bh);
1843 out_mutex:
1844         mutex_unlock(&xb_alloc_inode->i_mutex);
1845         iput(xb_alloc_inode);
1846 out:
1847         brelse(blk_bh);
1848         return ret;
1849 }
1850
1851 /*
1852  * ocfs2_xattr_remove()
1853  *
1854  * Free extended attribute resources associated with this inode.
1855  */
1856 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1857 {
1858         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1859         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1860         handle_t *handle;
1861         int ret;
1862
1863         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1864                 return 0;
1865
1866         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1867                 return 0;
1868
1869         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1870                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1871                 if (ret < 0) {
1872                         mlog_errno(ret);
1873                         goto out;
1874                 }
1875         }
1876
1877         if (di->i_xattr_loc) {
1878                 ret = ocfs2_xattr_free_block(inode,
1879                                              le64_to_cpu(di->i_xattr_loc));
1880                 if (ret < 0) {
1881                         mlog_errno(ret);
1882                         goto out;
1883                 }
1884         }
1885
1886         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1887                                    OCFS2_INODE_UPDATE_CREDITS);
1888         if (IS_ERR(handle)) {
1889                 ret = PTR_ERR(handle);
1890                 mlog_errno(ret);
1891                 goto out;
1892         }
1893         ret = ocfs2_journal_access_di(handle, inode, di_bh,
1894                                       OCFS2_JOURNAL_ACCESS_WRITE);
1895         if (ret) {
1896                 mlog_errno(ret);
1897                 goto out_commit;
1898         }
1899
1900         di->i_xattr_loc = 0;
1901
1902         spin_lock(&oi->ip_lock);
1903         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1904         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1905         spin_unlock(&oi->ip_lock);
1906
1907         ret = ocfs2_journal_dirty(handle, di_bh);
1908         if (ret < 0)
1909                 mlog_errno(ret);
1910 out_commit:
1911         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1912 out:
1913         return ret;
1914 }
1915
1916 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1917                                         struct ocfs2_dinode *di)
1918 {
1919         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1920         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1921         int free;
1922
1923         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1924                 return 0;
1925
1926         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1927                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1928                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1929         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1930                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1931                         le64_to_cpu(di->i_size);
1932         } else {
1933                 struct ocfs2_extent_list *el = &di->id2.i_list;
1934                 free = (le16_to_cpu(el->l_count) -
1935                         le16_to_cpu(el->l_next_free_rec)) *
1936                         sizeof(struct ocfs2_extent_rec);
1937         }
1938         if (free >= xattrsize)
1939                 return 1;
1940
1941         return 0;
1942 }
1943
1944 /*
1945  * ocfs2_xattr_ibody_find()
1946  *
1947  * Find extended attribute in inode block and
1948  * fill search info into struct ocfs2_xattr_search.
1949  */
1950 static int ocfs2_xattr_ibody_find(struct inode *inode,
1951                                   int name_index,
1952                                   const char *name,
1953                                   struct ocfs2_xattr_search *xs)
1954 {
1955         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1956         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1957         int ret;
1958         int has_space = 0;
1959
1960         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1961                 return 0;
1962
1963         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1964                 down_read(&oi->ip_alloc_sem);
1965                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1966                 up_read(&oi->ip_alloc_sem);
1967                 if (!has_space)
1968                         return 0;
1969         }
1970
1971         xs->xattr_bh = xs->inode_bh;
1972         xs->end = (void *)di + inode->i_sb->s_blocksize;
1973         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1974                 xs->header = (struct ocfs2_xattr_header *)
1975                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1976         else
1977                 xs->header = (struct ocfs2_xattr_header *)
1978                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1979         xs->base = (void *)xs->header;
1980         xs->here = xs->header->xh_entries;
1981
1982         /* Find the named attribute. */
1983         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1984                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1985                 if (ret && ret != -ENODATA)
1986                         return ret;
1987                 xs->not_found = ret;
1988         }
1989
1990         return 0;
1991 }
1992
1993 /*
1994  * ocfs2_xattr_ibody_set()
1995  *
1996  * Set, replace or remove an extended attribute into inode block.
1997  *
1998  */
1999 static int ocfs2_xattr_ibody_set(struct inode *inode,
2000                                  struct ocfs2_xattr_info *xi,
2001                                  struct ocfs2_xattr_search *xs,
2002                                  struct ocfs2_xattr_set_ctxt *ctxt)
2003 {
2004         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2005         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2006         int ret;
2007
2008         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2009                 return -ENOSPC;
2010
2011         down_write(&oi->ip_alloc_sem);
2012         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2013                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2014                         ret = -ENOSPC;
2015                         goto out;
2016                 }
2017         }
2018
2019         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2020                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2021 out:
2022         up_write(&oi->ip_alloc_sem);
2023
2024         return ret;
2025 }
2026
2027 /*
2028  * ocfs2_xattr_block_find()
2029  *
2030  * Find extended attribute in external block and
2031  * fill search info into struct ocfs2_xattr_search.
2032  */
2033 static int ocfs2_xattr_block_find(struct inode *inode,
2034                                   int name_index,
2035                                   const char *name,
2036                                   struct ocfs2_xattr_search *xs)
2037 {
2038         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2039         struct buffer_head *blk_bh = NULL;
2040         struct ocfs2_xattr_block *xb;
2041         int ret = 0;
2042
2043         if (!di->i_xattr_loc)
2044                 return ret;
2045
2046         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2047                                      &blk_bh);
2048         if (ret < 0) {
2049                 mlog_errno(ret);
2050                 return ret;
2051         }
2052
2053         xs->xattr_bh = blk_bh;
2054         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2055
2056         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2057                 xs->header = &xb->xb_attrs.xb_header;
2058                 xs->base = (void *)xs->header;
2059                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2060                 xs->here = xs->header->xh_entries;
2061
2062                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2063         } else
2064                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2065                                                    name_index,
2066                                                    name, xs);
2067
2068         if (ret && ret != -ENODATA) {
2069                 xs->xattr_bh = NULL;
2070                 goto cleanup;
2071         }
2072         xs->not_found = ret;
2073         return 0;
2074 cleanup:
2075         brelse(blk_bh);
2076
2077         return ret;
2078 }
2079
2080 /*
2081  * ocfs2_xattr_block_set()
2082  *
2083  * Set, replace or remove an extended attribute into external block.
2084  *
2085  */
2086 static int ocfs2_xattr_block_set(struct inode *inode,
2087                                  struct ocfs2_xattr_info *xi,
2088                                  struct ocfs2_xattr_search *xs,
2089                                  struct ocfs2_xattr_set_ctxt *ctxt)
2090 {
2091         struct buffer_head *new_bh = NULL;
2092         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2093         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
2094         handle_t *handle = ctxt->handle;
2095         struct ocfs2_xattr_block *xblk = NULL;
2096         u16 suballoc_bit_start;
2097         u32 num_got;
2098         u64 first_blkno;
2099         int ret;
2100
2101         if (!xs->xattr_bh) {
2102                 ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
2103                                               OCFS2_JOURNAL_ACCESS_CREATE);
2104                 if (ret < 0) {
2105                         mlog_errno(ret);
2106                         goto end;
2107                 }
2108
2109                 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2110                                            &suballoc_bit_start, &num_got,
2111                                            &first_blkno);
2112                 if (ret < 0) {
2113                         mlog_errno(ret);
2114                         goto end;
2115                 }
2116
2117                 new_bh = sb_getblk(inode->i_sb, first_blkno);
2118                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2119
2120                 ret = ocfs2_journal_access_xb(handle, inode, new_bh,
2121                                               OCFS2_JOURNAL_ACCESS_CREATE);
2122                 if (ret < 0) {
2123                         mlog_errno(ret);
2124                         goto end;
2125                 }
2126
2127                 /* Initialize ocfs2_xattr_block */
2128                 xs->xattr_bh = new_bh;
2129                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2130                 memset(xblk, 0, inode->i_sb->s_blocksize);
2131                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2132                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2133                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2134                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2135                 xblk->xb_blkno = cpu_to_le64(first_blkno);
2136
2137                 xs->header = &xblk->xb_attrs.xb_header;
2138                 xs->base = (void *)xs->header;
2139                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2140                 xs->here = xs->header->xh_entries;
2141
2142                 ret = ocfs2_journal_dirty(handle, new_bh);
2143                 if (ret < 0) {
2144                         mlog_errno(ret);
2145                         goto end;
2146                 }
2147                 di->i_xattr_loc = cpu_to_le64(first_blkno);
2148                 ocfs2_journal_dirty(handle, xs->inode_bh);
2149         } else
2150                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2151
2152         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2153                 /* Set extended attribute into external block */
2154                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2155                                             OCFS2_HAS_XATTR_FL);
2156                 if (!ret || ret != -ENOSPC)
2157                         goto end;
2158
2159                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2160                 if (ret)
2161                         goto end;
2162         }
2163
2164         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2165
2166 end:
2167
2168         return ret;
2169 }
2170
2171 /* Check whether the new xattr can be inserted into the inode. */
2172 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2173                                        struct ocfs2_xattr_info *xi,
2174                                        struct ocfs2_xattr_search *xs)
2175 {
2176         u64 value_size;
2177         struct ocfs2_xattr_entry *last;
2178         int free, i;
2179         size_t min_offs = xs->end - xs->base;
2180
2181         if (!xs->header)
2182                 return 0;
2183
2184         last = xs->header->xh_entries;
2185
2186         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2187                 size_t offs = le16_to_cpu(last->xe_name_offset);
2188                 if (offs < min_offs)
2189                         min_offs = offs;
2190                 last += 1;
2191         }
2192
2193         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
2194         if (free < 0)
2195                 return 0;
2196
2197         BUG_ON(!xs->not_found);
2198
2199         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2200                 value_size = OCFS2_XATTR_ROOT_SIZE;
2201         else
2202                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2203
2204         if (free >= sizeof(struct ocfs2_xattr_entry) +
2205                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2206                 return 1;
2207
2208         return 0;
2209 }
2210
2211 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2212                                      struct ocfs2_dinode *di,
2213                                      struct ocfs2_xattr_info *xi,
2214                                      struct ocfs2_xattr_search *xis,
2215                                      struct ocfs2_xattr_search *xbs,
2216                                      int *clusters_need,
2217                                      int *meta_need,
2218                                      int *credits_need)
2219 {
2220         int ret = 0, old_in_xb = 0;
2221         int clusters_add = 0, meta_add = 0, credits = 0;
2222         struct buffer_head *bh = NULL;
2223         struct ocfs2_xattr_block *xb = NULL;
2224         struct ocfs2_xattr_entry *xe = NULL;
2225         struct ocfs2_xattr_value_root *xv = NULL;
2226         char *base = NULL;
2227         int name_offset, name_len = 0;
2228         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2229                                                     xi->value_len);
2230         u64 value_size;
2231
2232         /*
2233          * Calculate the clusters we need to write.
2234          * No matter whether we replace an old one or add a new one,
2235          * we need this for writing.
2236          */
2237         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2238                 credits += new_clusters *
2239                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
2240
2241         if (xis->not_found && xbs->not_found) {
2242                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2243
2244                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2245                         clusters_add += new_clusters;
2246                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2247                                                         &def_xv.xv.xr_list,
2248                                                         new_clusters);
2249                 }
2250
2251                 goto meta_guess;
2252         }
2253
2254         if (!xis->not_found) {
2255                 xe = xis->here;
2256                 name_offset = le16_to_cpu(xe->xe_name_offset);
2257                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2258                 base = xis->base;
2259                 credits += OCFS2_INODE_UPDATE_CREDITS;
2260         } else {
2261                 int i, block_off = 0;
2262                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2263                 xe = xbs->here;
2264                 name_offset = le16_to_cpu(xe->xe_name_offset);
2265                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2266                 i = xbs->here - xbs->header->xh_entries;
2267                 old_in_xb = 1;
2268
2269                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2270                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2271                                                         bucket_xh(xbs->bucket),
2272                                                         i, &block_off,
2273                                                         &name_offset);
2274                         base = bucket_block(xbs->bucket, block_off);
2275                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2276                 } else {
2277                         base = xbs->base;
2278                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2279                 }
2280         }
2281
2282         /*
2283          * delete a xattr doesn't need metadata and cluster allocation.
2284          * so just calculate the credits and return.
2285          *
2286          * The credits for removing the value tree will be extended
2287          * by ocfs2_remove_extent itself.
2288          */
2289         if (!xi->value) {
2290                 if (!ocfs2_xattr_is_local(xe))
2291                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2292
2293                 goto out;
2294         }
2295
2296         /* do cluster allocation guess first. */
2297         value_size = le64_to_cpu(xe->xe_value_size);
2298
2299         if (old_in_xb) {
2300                 /*
2301                  * In xattr set, we always try to set the xe in inode first,
2302                  * so if it can be inserted into inode successfully, the old
2303                  * one will be removed from the xattr block, and this xattr
2304                  * will be inserted into inode as a new xattr in inode.
2305                  */
2306                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2307                         clusters_add += new_clusters;
2308                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
2309                                     OCFS2_INODE_UPDATE_CREDITS;
2310                         if (!ocfs2_xattr_is_local(xe))
2311                                 credits += ocfs2_calc_extend_credits(
2312                                                         inode->i_sb,
2313                                                         &def_xv.xv.xr_list,
2314                                                         new_clusters);
2315                         goto out;
2316                 }
2317         }
2318
2319         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2320                 /* the new values will be stored outside. */
2321                 u32 old_clusters = 0;
2322
2323                 if (!ocfs2_xattr_is_local(xe)) {
2324                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2325                                                                  value_size);
2326                         xv = (struct ocfs2_xattr_value_root *)
2327                              (base + name_offset + name_len);
2328                         value_size = OCFS2_XATTR_ROOT_SIZE;
2329                 } else
2330                         xv = &def_xv.xv;
2331
2332                 if (old_clusters >= new_clusters) {
2333                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2334                         goto out;
2335                 } else {
2336                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2337                         clusters_add += new_clusters - old_clusters;
2338                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2339                                                              &xv->xr_list,
2340                                                              new_clusters -
2341                                                              old_clusters);
2342                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2343                                 goto out;
2344                 }
2345         } else {
2346                 /*
2347                  * Now the new value will be stored inside. So if the new
2348                  * value is smaller than the size of value root or the old
2349                  * value, we don't need any allocation, otherwise we have
2350                  * to guess metadata allocation.
2351                  */
2352                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2353                     (!ocfs2_xattr_is_local(xe) &&
2354                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2355                         goto out;
2356         }
2357
2358 meta_guess:
2359         /* calculate metadata allocation. */
2360         if (di->i_xattr_loc) {
2361                 if (!xbs->xattr_bh) {
2362                         ret = ocfs2_read_xattr_block(inode,
2363                                                      le64_to_cpu(di->i_xattr_loc),
2364                                                      &bh);
2365                         if (ret) {
2366                                 mlog_errno(ret);
2367                                 goto out;
2368                         }
2369
2370                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2371                 } else
2372                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2373
2374                 /*
2375                  * If there is already an xattr tree, good, we can calculate
2376                  * like other b-trees. Otherwise we may have the chance of
2377                  * create a tree, the credit calculation is borrowed from
2378                  * ocfs2_calc_extend_credits with root_el = NULL. And the
2379                  * new tree will be cluster based, so no meta is needed.
2380                  */
2381                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2382                         struct ocfs2_extent_list *el =
2383                                  &xb->xb_attrs.xb_root.xt_list;
2384                         meta_add += ocfs2_extend_meta_needed(el);
2385                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2386                                                              el, 1);
2387                 } else
2388                         credits += OCFS2_SUBALLOC_ALLOC + 1;
2389
2390                 /*
2391                  * This cluster will be used either for new bucket or for
2392                  * new xattr block.
2393                  * If the cluster size is the same as the bucket size, one
2394                  * more is needed since we may need to extend the bucket
2395                  * also.
2396                  */
2397                 clusters_add += 1;
2398                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2399                 if (OCFS2_XATTR_BUCKET_SIZE ==
2400                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2401                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2402                         clusters_add += 1;
2403                 }
2404         } else {
2405                 meta_add += 1;
2406                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2407         }
2408 out:
2409         if (clusters_need)
2410                 *clusters_need = clusters_add;
2411         if (meta_need)
2412                 *meta_need = meta_add;
2413         if (credits_need)
2414                 *credits_need = credits;
2415         brelse(bh);
2416         return ret;
2417 }
2418
2419 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2420                                      struct ocfs2_dinode *di,
2421                                      struct ocfs2_xattr_info *xi,
2422                                      struct ocfs2_xattr_search *xis,
2423                                      struct ocfs2_xattr_search *xbs,
2424                                      struct ocfs2_xattr_set_ctxt *ctxt,
2425                                      int *credits)
2426 {
2427         int clusters_add, meta_add, ret;
2428         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2429
2430         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2431
2432         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2433
2434         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2435                                         &clusters_add, &meta_add, credits);
2436         if (ret) {
2437                 mlog_errno(ret);
2438                 return ret;
2439         }
2440
2441         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2442              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2443
2444         if (meta_add) {
2445                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2446                                                         &ctxt->meta_ac);
2447                 if (ret) {
2448                         mlog_errno(ret);
2449                         goto out;
2450                 }
2451         }
2452
2453         if (clusters_add) {
2454                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2455                 if (ret)
2456                         mlog_errno(ret);
2457         }
2458 out:
2459         if (ret) {
2460                 if (ctxt->meta_ac) {
2461                         ocfs2_free_alloc_context(ctxt->meta_ac);
2462                         ctxt->meta_ac = NULL;
2463                 }
2464
2465                 /*
2466                  * We cannot have an error and a non null ctxt->data_ac.
2467                  */
2468         }
2469
2470         return ret;
2471 }
2472
2473 static int __ocfs2_xattr_set_handle(struct inode *inode,
2474                                     struct ocfs2_dinode *di,
2475                                     struct ocfs2_xattr_info *xi,
2476                                     struct ocfs2_xattr_search *xis,
2477                                     struct ocfs2_xattr_search *xbs,
2478                                     struct ocfs2_xattr_set_ctxt *ctxt)
2479 {
2480         int ret = 0, credits, old_found;
2481
2482         if (!xi->value) {
2483                 /* Remove existing extended attribute */
2484                 if (!xis->not_found)
2485                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2486                 else if (!xbs->not_found)
2487                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2488         } else {
2489                 /* We always try to set extended attribute into inode first*/
2490                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2491                 if (!ret && !xbs->not_found) {
2492                         /*
2493                          * If succeed and that extended attribute existing in
2494                          * external block, then we will remove it.
2495                          */
2496                         xi->value = NULL;
2497                         xi->value_len = 0;
2498
2499                         old_found = xis->not_found;
2500                         xis->not_found = -ENODATA;
2501                         ret = ocfs2_calc_xattr_set_need(inode,
2502                                                         di,
2503                                                         xi,
2504                                                         xis,
2505                                                         xbs,
2506                                                         NULL,
2507                                                         NULL,
2508                                                         &credits);
2509                         xis->not_found = old_found;
2510                         if (ret) {
2511                                 mlog_errno(ret);
2512                                 goto out;
2513                         }
2514
2515                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2516                                         ctxt->handle->h_buffer_credits);
2517                         if (ret) {
2518                                 mlog_errno(ret);
2519                                 goto out;
2520                         }
2521                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2522                 } else if (ret == -ENOSPC) {
2523                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2524                                 ret = ocfs2_xattr_block_find(inode,
2525                                                              xi->name_index,
2526                                                              xi->name, xbs);
2527                                 if (ret)
2528                                         goto out;
2529
2530                                 old_found = xis->not_found;
2531                                 xis->not_found = -ENODATA;
2532                                 ret = ocfs2_calc_xattr_set_need(inode,
2533                                                                 di,
2534                                                                 xi,
2535                                                                 xis,
2536                                                                 xbs,
2537                                                                 NULL,
2538                                                                 NULL,
2539                                                                 &credits);
2540                                 xis->not_found = old_found;
2541                                 if (ret) {
2542                                         mlog_errno(ret);
2543                                         goto out;
2544                                 }
2545
2546                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2547                                         ctxt->handle->h_buffer_credits);
2548                                 if (ret) {
2549                                         mlog_errno(ret);
2550                                         goto out;
2551                                 }
2552                         }
2553                         /*
2554                          * If no space in inode, we will set extended attribute
2555                          * into external block.
2556                          */
2557                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2558                         if (ret)
2559                                 goto out;
2560                         if (!xis->not_found) {
2561                                 /*
2562                                  * If succeed and that extended attribute
2563                                  * existing in inode, we will remove it.
2564                                  */
2565                                 xi->value = NULL;
2566                                 xi->value_len = 0;
2567                                 xbs->not_found = -ENODATA;
2568                                 ret = ocfs2_calc_xattr_set_need(inode,
2569                                                                 di,
2570                                                                 xi,
2571                                                                 xis,
2572                                                                 xbs,
2573                                                                 NULL,
2574                                                                 NULL,
2575                                                                 &credits);
2576                                 if (ret) {
2577                                         mlog_errno(ret);
2578                                         goto out;
2579                                 }
2580
2581                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2582                                                 ctxt->handle->h_buffer_credits);
2583                                 if (ret) {
2584                                         mlog_errno(ret);
2585                                         goto out;
2586                                 }
2587                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2588                                                             xis, ctxt);
2589                         }
2590                 }
2591         }
2592
2593         if (!ret) {
2594                 /* Update inode ctime. */
2595                 ret = ocfs2_journal_access(ctxt->handle, inode, xis->inode_bh,
2596                                            OCFS2_JOURNAL_ACCESS_WRITE);
2597                 if (ret) {
2598                         mlog_errno(ret);
2599                         goto out;
2600                 }
2601
2602                 inode->i_ctime = CURRENT_TIME;
2603                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2604                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2605                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2606         }
2607 out:
2608         return ret;
2609 }
2610
2611 /*
2612  * This function only called duing creating inode
2613  * for init security/acl xattrs of the new inode.
2614  * All transanction credits have been reserved in mknod.
2615  */
2616 int ocfs2_xattr_set_handle(handle_t *handle,
2617                            struct inode *inode,
2618                            struct buffer_head *di_bh,
2619                            int name_index,
2620                            const char *name,
2621                            const void *value,
2622                            size_t value_len,
2623                            int flags,
2624                            struct ocfs2_alloc_context *meta_ac,
2625                            struct ocfs2_alloc_context *data_ac)
2626 {
2627         struct ocfs2_dinode *di;
2628         int ret;
2629
2630         struct ocfs2_xattr_info xi = {
2631                 .name_index = name_index,
2632                 .name = name,
2633                 .value = value,
2634                 .value_len = value_len,
2635         };
2636
2637         struct ocfs2_xattr_search xis = {
2638                 .not_found = -ENODATA,
2639         };
2640
2641         struct ocfs2_xattr_search xbs = {
2642                 .not_found = -ENODATA,
2643         };
2644
2645         struct ocfs2_xattr_set_ctxt ctxt = {
2646                 .handle = handle,
2647                 .meta_ac = meta_ac,
2648                 .data_ac = data_ac,
2649         };
2650
2651         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2652                 return -EOPNOTSUPP;
2653
2654         /*
2655          * In extreme situation, may need xattr bucket when
2656          * block size is too small. And we have already reserved
2657          * the credits for bucket in mknod.
2658          */
2659         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
2660                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2661                 if (!xbs.bucket) {
2662                         mlog_errno(-ENOMEM);
2663                         return -ENOMEM;
2664                 }
2665         }
2666
2667         xis.inode_bh = xbs.inode_bh = di_bh;
2668         di = (struct ocfs2_dinode *)di_bh->b_data;
2669
2670         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2671
2672         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2673         if (ret)
2674                 goto cleanup;
2675         if (xis.not_found) {
2676                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2677                 if (ret)
2678                         goto cleanup;
2679         }
2680
2681         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2682
2683 cleanup:
2684         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2685         brelse(xbs.xattr_bh);
2686         ocfs2_xattr_bucket_free(xbs.bucket);
2687
2688         return ret;
2689 }
2690
2691 /*
2692  * ocfs2_xattr_set()
2693  *
2694  * Set, replace or remove an extended attribute for this inode.
2695  * value is NULL to remove an existing extended attribute, else either
2696  * create or replace an extended attribute.
2697  */
2698 int ocfs2_xattr_set(struct inode *inode,
2699                     int name_index,
2700                     const char *name,
2701                     const void *value,
2702                     size_t value_len,
2703                     int flags)
2704 {
2705         struct buffer_head *di_bh = NULL;
2706         struct ocfs2_dinode *di;
2707         int ret, credits;
2708         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2709         struct inode *tl_inode = osb->osb_tl_inode;
2710         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2711
2712         struct ocfs2_xattr_info xi = {
2713                 .name_index = name_index,
2714                 .name = name,
2715                 .value = value,
2716                 .value_len = value_len,
2717         };
2718
2719         struct ocfs2_xattr_search xis = {
2720                 .not_found = -ENODATA,
2721         };
2722
2723         struct ocfs2_xattr_search xbs = {
2724                 .not_found = -ENODATA,
2725         };
2726
2727         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2728                 return -EOPNOTSUPP;
2729
2730         /*
2731          * Only xbs will be used on indexed trees.  xis doesn't need a
2732          * bucket.
2733          */
2734         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2735         if (!xbs.bucket) {
2736                 mlog_errno(-ENOMEM);
2737                 return -ENOMEM;
2738         }
2739
2740         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2741         if (ret < 0) {
2742                 mlog_errno(ret);
2743                 goto cleanup_nolock;
2744         }
2745         xis.inode_bh = xbs.inode_bh = di_bh;
2746         di = (struct ocfs2_dinode *)di_bh->b_data;
2747
2748         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2749         /*
2750          * Scan inode and external block to find the same name
2751          * extended attribute and collect search infomation.
2752          */
2753         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2754         if (ret)
2755                 goto cleanup;
2756         if (xis.not_found) {
2757                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2758                 if (ret)
2759                         goto cleanup;
2760         }
2761
2762         if (xis.not_found && xbs.not_found) {
2763                 ret = -ENODATA;
2764                 if (flags & XATTR_REPLACE)
2765                         goto cleanup;
2766                 ret = 0;
2767                 if (!value)
2768                         goto cleanup;
2769         } else {
2770                 ret = -EEXIST;
2771                 if (flags & XATTR_CREATE)
2772                         goto cleanup;
2773         }
2774
2775
2776         mutex_lock(&tl_inode->i_mutex);
2777
2778         if (ocfs2_truncate_log_needs_flush(osb)) {
2779                 ret = __ocfs2_flush_truncate_log(osb);
2780                 if (ret < 0) {
2781                         mutex_unlock(&tl_inode->i_mutex);
2782                         mlog_errno(ret);
2783                         goto cleanup;
2784                 }
2785         }
2786         mutex_unlock(&tl_inode->i_mutex);
2787
2788         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2789                                         &xbs, &ctxt, &credits);
2790         if (ret) {
2791                 mlog_errno(ret);
2792                 goto cleanup;
2793         }
2794
2795         /* we need to update inode's ctime field, so add credit for it. */
2796         credits += OCFS2_INODE_UPDATE_CREDITS;
2797         ctxt.handle = ocfs2_start_trans(osb, credits);
2798         if (IS_ERR(ctxt.handle)) {
2799                 ret = PTR_ERR(ctxt.handle);
2800                 mlog_errno(ret);
2801                 goto cleanup;
2802         }
2803
2804         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2805
2806         ocfs2_commit_trans(osb, ctxt.handle);
2807
2808         if (ctxt.data_ac)
2809                 ocfs2_free_alloc_context(ctxt.data_ac);
2810         if (ctxt.meta_ac)
2811                 ocfs2_free_alloc_context(ctxt.meta_ac);
2812         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2813                 ocfs2_schedule_truncate_log_flush(osb, 1);
2814         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2815 cleanup:
2816         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2817         ocfs2_inode_unlock(inode, 1);
2818 cleanup_nolock:
2819         brelse(di_bh);
2820         brelse(xbs.xattr_bh);
2821         ocfs2_xattr_bucket_free(xbs.bucket);
2822
2823         return ret;
2824 }
2825
2826 /*
2827  * Find the xattr extent rec which may contains name_hash.
2828  * e_cpos will be the first name hash of the xattr rec.
2829  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2830  */
2831 static int ocfs2_xattr_get_rec(struct inode *inode,
2832                                u32 name_hash,
2833                                u64 *p_blkno,
2834                                u32 *e_cpos,
2835                                u32 *num_clusters,
2836                                struct ocfs2_extent_list *el)
2837 {
2838         int ret = 0, i;
2839         struct buffer_head *eb_bh = NULL;
2840         struct ocfs2_extent_block *eb;
2841         struct ocfs2_extent_rec *rec = NULL;
2842         u64 e_blkno = 0;
2843
2844         if (el->l_tree_depth) {
2845                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2846                 if (ret) {
2847                         mlog_errno(ret);
2848                         goto out;
2849                 }
2850
2851                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2852                 el = &eb->h_list;
2853
2854                 if (el->l_tree_depth) {
2855                         ocfs2_error(inode->i_sb,
2856                                     "Inode %lu has non zero tree depth in "
2857                                     "xattr tree block %llu\n", inode->i_ino,
2858                                     (unsigned long long)eb_bh->b_blocknr);
2859                         ret = -EROFS;
2860                         goto out;
2861                 }
2862         }
2863
2864         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2865                 rec = &el->l_recs[i];
2866
2867                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2868                         e_blkno = le64_to_cpu(rec->e_blkno);
2869                         break;
2870                 }
2871         }
2872
2873         if (!e_blkno) {
2874                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2875                             "record (%u, %u, 0) in xattr", inode->i_ino,
2876                             le32_to_cpu(rec->e_cpos),
2877                             ocfs2_rec_clusters(el, rec));
2878                 ret = -EROFS;
2879                 goto out;
2880         }
2881
2882         *p_blkno = le64_to_cpu(rec->e_blkno);
2883         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2884         if (e_cpos)
2885                 *e_cpos = le32_to_cpu(rec->e_cpos);
2886 out:
2887         brelse(eb_bh);
2888         return ret;
2889 }
2890
2891 typedef int (xattr_bucket_func)(struct inode *inode,
2892                                 struct ocfs2_xattr_bucket *bucket,
2893                                 void *para);
2894
2895 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2896                                    struct ocfs2_xattr_bucket *bucket,
2897                                    int name_index,
2898                                    const char *name,
2899                                    u32 name_hash,
2900                                    u16 *xe_index,
2901                                    int *found)
2902 {
2903         int i, ret = 0, cmp = 1, block_off, new_offset;
2904         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2905         size_t name_len = strlen(name);
2906         struct ocfs2_xattr_entry *xe = NULL;
2907         char *xe_name;
2908
2909         /*
2910          * We don't use binary search in the bucket because there
2911          * may be multiple entries with the same name hash.
2912          */
2913         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2914                 xe = &xh->xh_entries[i];
2915
2916                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2917                         continue;
2918                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2919                         break;
2920
2921                 cmp = name_index - ocfs2_xattr_get_type(xe);
2922                 if (!cmp)
2923                         cmp = name_len - xe->xe_name_len;
2924                 if (cmp)
2925                         continue;
2926
2927                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2928                                                         xh,
2929                                                         i,
2930                                                         &block_off,
2931                                                         &new_offset);
2932                 if (ret) {
2933                         mlog_errno(ret);
2934                         break;
2935                 }
2936
2937
2938                 xe_name = bucket_block(bucket, block_off) + new_offset;
2939                 if (!memcmp(name, xe_name, name_len)) {
2940                         *xe_index = i;
2941                         *found = 1;
2942                         ret = 0;
2943                         break;
2944                 }
2945         }
2946
2947         return ret;
2948 }
2949
2950 /*
2951  * Find the specified xattr entry in a series of buckets.
2952  * This series start from p_blkno and last for num_clusters.
2953  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2954  * the num of the valid buckets.
2955  *
2956  * Return the buffer_head this xattr should reside in. And if the xattr's
2957  * hash is in the gap of 2 buckets, return the lower bucket.
2958  */
2959 static int ocfs2_xattr_bucket_find(struct inode *inode,
2960                                    int name_index,
2961                                    const char *name,
2962                                    u32 name_hash,
2963                                    u64 p_blkno,
2964                                    u32 first_hash,
2965                                    u32 num_clusters,
2966                                    struct ocfs2_xattr_search *xs)
2967 {
2968         int ret, found = 0;
2969         struct ocfs2_xattr_header *xh = NULL;
2970         struct ocfs2_xattr_entry *xe = NULL;
2971         u16 index = 0;
2972         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2973         int low_bucket = 0, bucket, high_bucket;
2974         struct ocfs2_xattr_bucket *search;
2975         u32 last_hash;
2976         u64 blkno, lower_blkno = 0;
2977
2978         search = ocfs2_xattr_bucket_new(inode);
2979         if (!search) {
2980                 ret = -ENOMEM;
2981                 mlog_errno(ret);
2982                 goto out;
2983         }
2984
2985         ret = ocfs2_read_xattr_bucket(search, p_blkno);
2986         if (ret) {
2987                 mlog_errno(ret);
2988                 goto out;
2989         }
2990
2991         xh = bucket_xh(search);
2992         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2993         while (low_bucket <= high_bucket) {
2994                 ocfs2_xattr_bucket_relse(search);
2995
2996                 bucket = (low_bucket + high_bucket) / 2;
2997                 blkno = p_blkno + bucket * blk_per_bucket;
2998                 ret = ocfs2_read_xattr_bucket(search, blkno);
2999                 if (ret) {
3000                         mlog_errno(ret);
3001                         goto out;
3002                 }
3003
3004                 xh = bucket_xh(search);
3005                 xe = &xh->xh_entries[0];
3006                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3007                         high_bucket = bucket - 1;
3008                         continue;
3009                 }
3010
3011                 /*
3012                  * Check whether the hash of the last entry in our
3013                  * bucket is larger than the search one. for an empty
3014                  * bucket, the last one is also the first one.
3015                  */
3016                 if (xh->xh_count)
3017                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3018
3019                 last_hash = le32_to_cpu(xe->xe_name_hash);
3020
3021                 /* record lower_blkno which may be the insert place. */
3022                 lower_blkno = blkno;
3023
3024                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3025                         low_bucket = bucket + 1;
3026                         continue;
3027                 }
3028
3029                 /* the searched xattr should reside in this bucket if exists. */
3030                 ret = ocfs2_find_xe_in_bucket(inode, search,
3031                                               name_index, name, name_hash,
3032                                               &index, &found);
3033                 if (ret) {
3034                         mlog_errno(ret);
3035                         goto out;
3036                 }
3037                 break;
3038         }
3039
3040         /*
3041          * Record the bucket we have found.
3042          * When the xattr's hash value is in the gap of 2 buckets, we will
3043          * always set it to the previous bucket.
3044          */
3045         if (!lower_blkno)
3046                 lower_blkno = p_blkno;
3047
3048         /* This should be in cache - we just read it during the search */
3049         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3050         if (ret) {
3051                 mlog_errno(ret);
3052                 goto out;
3053         }
3054
3055         xs->header = bucket_xh(xs->bucket);
3056         xs->base = bucket_block(xs->bucket, 0);
3057         xs->end = xs->base + inode->i_sb->s_blocksize;
3058
3059         if (found) {
3060                 xs->here = &xs->header->xh_entries[index];
3061                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3062                      (unsigned long long)bucket_blkno(xs->bucket), index);
3063         } else
3064                 ret = -ENODATA;
3065
3066 out:
3067         ocfs2_xattr_bucket_free(search);
3068         return ret;
3069 }
3070
3071 static int ocfs2_xattr_index_block_find(struct inode *inode,
3072                                         struct buffer_head *root_bh,
3073                                         int name_index,
3074                                         const char *name,
3075                                         struct ocfs2_xattr_search *xs)
3076 {
3077         int ret;
3078         struct ocfs2_xattr_block *xb =
3079                         (struct ocfs2_xattr_block *)root_bh->b_data;
3080         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3081         struct ocfs2_extent_list *el = &xb_root->xt_list;
3082         u64 p_blkno = 0;
3083         u32 first_hash, num_clusters = 0;
3084         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3085
3086         if (le16_to_cpu(el->l_next_free_rec) == 0)
3087                 return -ENODATA;
3088
3089         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3090              name, name_hash, name_index);
3091
3092         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3093                                   &num_clusters, el);
3094         if (ret) {
3095                 mlog_errno(ret);
3096                 goto out;
3097         }
3098
3099         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3100
3101         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3102              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3103              first_hash);
3104
3105         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3106                                       p_blkno, first_hash, num_clusters, xs);
3107
3108 out:
3109         return ret;
3110 }
3111
3112 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3113                                        u64 blkno,
3114                                        u32 clusters,
3115                                        xattr_bucket_func *func,
3116                                        void *para)
3117 {
3118         int i, ret = 0;
3119         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3120         u32 num_buckets = clusters * bpc;
3121         struct ocfs2_xattr_bucket *bucket;
3122
3123         bucket = ocfs2_xattr_bucket_new(inode);
3124         if (!bucket) {
3125                 mlog_errno(-ENOMEM);
3126                 return -ENOMEM;
3127         }
3128
3129         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3130              clusters, (unsigned long long)blkno);
3131
3132         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3133                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3134                 if (ret) {
3135                         mlog_errno(ret);
3136                         break;
3137                 }
3138
3139                 /*
3140                  * The real bucket num in this series of blocks is stored
3141                  * in the 1st bucket.
3142                  */
3143                 if (i == 0)
3144                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3145
3146                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3147                      (unsigned long long)blkno,
3148                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3149                 if (func) {
3150                         ret = func(inode, bucket, para);
3151                         if (ret)
3152                                 mlog_errno(ret);
3153                         /* Fall through to bucket_relse() */
3154                 }
3155
3156                 ocfs2_xattr_bucket_relse(bucket);
3157                 if (ret)
3158                         break;
3159         }
3160
3161         ocfs2_xattr_bucket_free(bucket);
3162         return ret;
3163 }
3164
3165 struct ocfs2_xattr_tree_list {
3166         char *buffer;
3167         size_t buffer_size;
3168         size_t result;
3169 };
3170
3171 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3172                                              struct ocfs2_xattr_header *xh,
3173                                              int index,
3174                                              int *block_off,
3175                                              int *new_offset)
3176 {
3177         u16 name_offset;
3178
3179         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3180                 return -EINVAL;
3181
3182         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3183
3184         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3185         *new_offset = name_offset % inode->i_sb->s_blocksize;
3186
3187         return 0;
3188 }
3189
3190 static int ocfs2_list_xattr_bucket(struct inode *inode,
3191                                    struct ocfs2_xattr_bucket *bucket,
3192                                    void *para)
3193 {
3194         int ret = 0, type;
3195         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3196         int i, block_off, new_offset;
3197         const char *prefix, *name;
3198
3199         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3200                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3201                 type = ocfs2_xattr_get_type(entry);
3202                 prefix = ocfs2_xattr_prefix(type);
3203
3204                 if (prefix) {
3205                         ret = ocfs2_xattr_bucket_get_name_value(inode,
3206                                                                 bucket_xh(bucket),
3207                                                                 i,
3208                                                                 &block_off,
3209                                                                 &new_offset);
3210                         if (ret)
3211                                 break;
3212
3213                         name = (const char *)bucket_block(bucket, block_off) +
3214                                 new_offset;
3215                         ret = ocfs2_xattr_list_entry(xl->buffer,
3216                                                      xl->buffer_size,
3217                                                      &xl->result,
3218                                                      prefix, name,
3219                                                      entry->xe_name_len);
3220                         if (ret)
3221                                 break;
3222                 }
3223         }
3224
3225         return ret;
3226 }
3227
3228 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3229                                              struct ocfs2_xattr_tree_root *xt,
3230                                              char *buffer,
3231                                              size_t buffer_size)
3232 {
3233         struct ocfs2_extent_list *el = &xt->xt_list;
3234         int ret = 0;
3235         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3236         u64 p_blkno = 0;
3237         struct ocfs2_xattr_tree_list xl = {
3238                 .buffer = buffer,
3239                 .buffer_size = buffer_size,
3240                 .result = 0,
3241         };
3242
3243         if (le16_to_cpu(el->l_next_free_rec) == 0)
3244                 return 0;
3245
3246         while (name_hash > 0) {
3247                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3248                                           &e_cpos, &num_clusters, el);
3249                 if (ret) {
3250                         mlog_errno(ret);
3251                         goto out;
3252                 }
3253
3254                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3255                                                   ocfs2_list_xattr_bucket,
3256                                                   &xl);
3257                 if (ret) {
3258                         mlog_errno(ret);
3259                         goto out;
3260                 }
3261
3262                 if (e_cpos == 0)
3263                         break;
3264
3265                 name_hash = e_cpos - 1;
3266         }
3267
3268         ret = xl.result;
3269 out:
3270         return ret;
3271 }
3272
3273 static int cmp_xe(const void *a, const void *b)
3274 {
3275         const struct ocfs2_xattr_entry *l = a, *r = b;
3276         u32 l_hash = le32_to_cpu(l->xe_name_hash);
3277         u32 r_hash = le32_to_cpu(r->xe_name_hash);
3278
3279         if (l_hash > r_hash)
3280                 return 1;
3281         if (l_hash < r_hash)
3282                 return -1;
3283         return 0;
3284 }
3285
3286 static void swap_xe(void *a, void *b, int size)
3287 {
3288         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3289
3290         tmp = *l;
3291         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3292         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3293 }
3294
3295 /*
3296  * When the ocfs2_xattr_block is filled up, new bucket will be created
3297  * and all the xattr entries will be moved to the new bucket.
3298  * The header goes at the start of the bucket, and the names+values are
3299  * filled from the end.  This is why *target starts as the last buffer.
3300  * Note: we need to sort the entries since they are not saved in order
3301  * in the ocfs2_xattr_block.
3302  */
3303 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3304                                            struct buffer_head *xb_bh,
3305                                            struct ocfs2_xattr_bucket *bucket)
3306 {
3307         int i, blocksize = inode->i_sb->s_blocksize;
3308         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3309         u16 offset, size, off_change;
3310         struct ocfs2_xattr_entry *xe;
3311         struct ocfs2_xattr_block *xb =
3312                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
3313         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3314         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3315         u16 count = le16_to_cpu(xb_xh->xh_count);
3316         char *src = xb_bh->b_data;
3317         char *target = bucket_block(bucket, blks - 1);
3318
3319         mlog(0, "cp xattr from block %llu to bucket %llu\n",
3320              (unsigned long long)xb_bh->b_blocknr,
3321              (unsigned long long)bucket_blkno(bucket));
3322
3323         for (i = 0; i < blks; i++)
3324                 memset(bucket_block(bucket, i), 0, blocksize);
3325
3326         /*
3327          * Since the xe_name_offset is based on ocfs2_xattr_header,
3328          * there is a offset change corresponding to the change of
3329          * ocfs2_xattr_header's position.
3330          */
3331         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3332         xe = &xb_xh->xh_entries[count - 1];
3333         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3334         size = blocksize - offset;
3335
3336         /* copy all the names and values. */
3337         memcpy(target + offset, src + offset, size);
3338
3339         /* Init new header now. */
3340         xh->xh_count = xb_xh->xh_count;
3341         xh->xh_num_buckets = cpu_to_le16(1);
3342         xh->xh_name_value_len = cpu_to_le16(size);
3343         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3344
3345         /* copy all the entries. */
3346         target = bucket_block(bucket, 0);
3347         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3348         size = count * sizeof(struct ocfs2_xattr_entry);
3349         memcpy(target + offset, (char *)xb_xh + offset, size);
3350
3351         /* Change the xe offset for all the xe because of the move. */
3352         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3353                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3354         for (i = 0; i < count; i++)
3355                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3356
3357         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3358              offset, size, off_change);
3359
3360         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3361              cmp_xe, swap_xe);
3362 }
3363
3364 /*
3365  * After we move xattr from block to index btree, we have to
3366  * update ocfs2_xattr_search to the new xe and base.
3367  *
3368  * When the entry is in xattr block, xattr_bh indicates the storage place.
3369  * While if the entry is in index b-tree, "bucket" indicates the
3370  * real place of the xattr.
3371  */
3372 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3373                                             struct ocfs2_xattr_search *xs,
3374                                             struct buffer_head *old_bh)
3375 {
3376         char *buf = old_bh->b_data;
3377         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3378         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3379         int i;
3380
3381         xs->header = bucket_xh(xs->bucket);
3382         xs->base = bucket_block(xs->bucket, 0);
3383         xs->end = xs->base + inode->i_sb->s_blocksize;
3384
3385         if (xs->not_found)
3386                 return;
3387
3388         i = xs->here - old_xh->xh_entries;
3389         xs->here = &xs->header->xh_entries[i];
3390 }
3391
3392 static int ocfs2_xattr_create_index_block(struct inode *inode,
3393                                           struct ocfs2_xattr_search *xs,
3394                                           struct ocfs2_xattr_set_ctxt *ctxt)
3395 {
3396         int ret;
3397         u32 bit_off, len;
3398         u64 blkno;
3399         handle_t *handle = ctxt->handle;
3400         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3401         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3402         struct buffer_head *xb_bh = xs->xattr_bh;
3403         struct ocfs2_xattr_block *xb =
3404                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3405         struct ocfs2_xattr_tree_root *xr;
3406         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3407
3408         mlog(0, "create xattr index block for %llu\n",
3409              (unsigned long long)xb_bh->b_blocknr);
3410
3411         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3412         BUG_ON(!xs->bucket);
3413
3414         /*
3415          * XXX:
3416          * We can use this lock for now, and maybe move to a dedicated mutex
3417          * if performance becomes a problem later.
3418          */
3419         down_write(&oi->ip_alloc_sem);
3420
3421         ret = ocfs2_journal_access_xb(handle, inode, xb_bh,
3422                                       OCFS2_JOURNAL_ACCESS_WRITE);
3423         if (ret) {
3424                 mlog_errno(ret);
3425                 goto out;
3426         }
3427
3428         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3429                                      1, 1, &bit_off, &len);
3430         if (ret) {
3431                 mlog_errno(ret);
3432                 goto out;
3433         }
3434
3435         /*
3436          * The bucket may spread in many blocks, and
3437          * we will only touch the 1st block and the last block
3438          * in the whole bucket(one for entry and one for data).
3439          */
3440         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3441
3442         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3443              (unsigned long long)blkno);
3444
3445         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3446         if (ret) {
3447                 mlog_errno(ret);
3448                 goto out;
3449         }
3450
3451         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3452                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3453         if (ret) {
3454                 mlog_errno(ret);
3455                 goto out;
3456         }
3457
3458         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3459         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3460
3461         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3462
3463         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3464         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3465                offsetof(struct ocfs2_xattr_block, xb_attrs));
3466
3467         xr = &xb->xb_attrs.xb_root;
3468         xr->xt_clusters = cpu_to_le32(1);
3469         xr->xt_last_eb_blk = 0;
3470         xr->xt_list.l_tree_depth = 0;
3471         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3472         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3473
3474         xr->xt_list.l_recs[0].e_cpos = 0;
3475         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3476         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3477
3478         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3479
3480         ocfs2_journal_dirty(handle, xb_bh);
3481
3482 out:
3483         up_write(&oi->ip_alloc_sem);
3484
3485         return ret;
3486 }
3487
3488 static int cmp_xe_offset(const void *a, const void *b)
3489 {
3490         const struct ocfs2_xattr_entry *l = a, *r = b;
3491         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3492         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3493
3494         if (l_name_offset < r_name_offset)
3495                 return 1;
3496         if (l_name_offset > r_name_offset)
3497                 return -1;
3498         return 0;
3499 }
3500
3501 /*
3502  * defrag a xattr bucket if we find that the bucket has some
3503  * holes beteen name/value pairs.
3504  * We will move all the name/value pairs to the end of the bucket
3505  * so that we can spare some space for insertion.
3506  */
3507 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3508                                      handle_t *handle,
3509                                      struct ocfs2_xattr_bucket *bucket)
3510 {
3511         int ret, i;
3512         size_t end, offset, len, value_len;
3513         struct ocfs2_xattr_header *xh;
3514         char *entries, *buf, *bucket_buf = NULL;
3515         u64 blkno = bucket_blkno(bucket);
3516         u16 xh_free_start;
3517         size_t blocksize = inode->i_sb->s_blocksize;
3518         struct ocfs2_xattr_entry *xe;
3519
3520         /*
3521          * In order to make the operation more efficient and generic,
3522          * we copy all the blocks into a contiguous memory and do the
3523          * defragment there, so if anything is error, we will not touch
3524          * the real block.
3525          */
3526         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3527         if (!bucket_buf) {
3528                 ret = -EIO;
3529                 goto out;
3530         }
3531
3532         buf = bucket_buf;
3533         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3534                 memcpy(buf, bucket_block(bucket, i), blocksize);
3535
3536         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3537                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3538         if (ret < 0) {
3539                 mlog_errno(ret);
3540                 goto out;
3541         }
3542
3543         xh = (struct ocfs2_xattr_header *)bucket_buf;
3544         entries = (char *)xh->xh_entries;
3545         xh_free_start = le16_to_cpu(xh->xh_free_start);
3546
3547         mlog(0, "adjust xattr bucket in %llu, count = %u, "
3548              "xh_free_start = %u, xh_name_value_len = %u.\n",
3549              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3550              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3551
3552         /*
3553          * sort all the entries by their offset.
3554          * the largest will be the first, so that we can
3555          * move them to the end one by one.
3556          */
3557         sort(entries, le16_to_cpu(xh->xh_count),
3558              sizeof(struct ocfs2_xattr_entry),
3559              cmp_xe_offset, swap_xe);
3560
3561         /* Move all name/values to the end of the bucket. */
3562         xe = xh->xh_entries;
3563         end = OCFS2_XATTR_BUCKET_SIZE;
3564         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3565                 offset = le16_to_cpu(xe->xe_name_offset);
3566                 if (ocfs2_xattr_is_local(xe))
3567                         value_len = OCFS2_XATTR_SIZE(
3568                                         le64_to_cpu(xe->xe_value_size));
3569                 else
3570                         value_len = OCFS2_XATTR_ROOT_SIZE;
3571                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3572
3573                 /*
3574                  * We must make sure that the name/value pair
3575                  * exist in the same block. So adjust end to
3576                  * the previous block end if needed.
3577                  */
3578                 if (((end - len) / blocksize !=
3579                         (end - 1) / blocksize))
3580                         end = end - end % blocksize;
3581
3582                 if (end > offset + len) {
3583                         memmove(bucket_buf + end - len,
3584                                 bucket_buf + offset, len);
3585                         xe->xe_name_offset = cpu_to_le16(end - len);
3586                 }
3587
3588                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3589                                 "bucket %llu\n", (unsigned long long)blkno);
3590
3591                 end -= len;
3592         }
3593
3594         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3595                         "bucket %llu\n", (unsigned long long)blkno);
3596
3597         if (xh_free_start == end)
3598                 goto out;
3599
3600         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3601         xh->xh_free_start = cpu_to_le16(end);
3602
3603         /* sort the entries by their name_hash. */
3604         sort(entries, le16_to_cpu(xh->xh_count),
3605              sizeof(struct ocfs2_xattr_entry),
3606              cmp_xe, swap_xe);
3607
3608         buf = bucket_buf;
3609         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3610                 memcpy(bucket_block(bucket, i), buf, blocksize);
3611         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3612
3613 out:
3614         kfree(bucket_buf);
3615         return ret;
3616 }
3617
3618 /*
3619  * prev_blkno points to the start of an existing extent.  new_blkno
3620  * points to a newly allocated extent.  Because we know each of our
3621  * clusters contains more than bucket, we can easily split one cluster
3622  * at a bucket boundary.  So we take the last cluster of the existing
3623  * extent and split it down the middle.  We move the last half of the
3624  * buckets in the last cluster of the existing extent over to the new
3625  * extent.
3626  *
3627  * first_bh is the buffer at prev_blkno so we can update the existing
3628  * extent's bucket count.  header_bh is the bucket were we were hoping
3629  * to insert our xattr.  If the bucket move places the target in the new
3630  * extent, we'll update first_bh and header_bh after modifying the old
3631  * extent.
3632  *
3633  * first_hash will be set as the 1st xe's name_hash in the new extent.
3634  */
3635 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3636                                                handle_t *handle,
3637                                                struct ocfs2_xattr_bucket *first,
3638                                                struct ocfs2_xattr_bucket *target,
3639                                                u64 new_blkno,
3640                                                u32 num_clusters,
3641                                                u32 *first_hash)
3642 {
3643         int ret;
3644         struct super_block *sb = inode->i_sb;
3645         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3646         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3647         int to_move = num_buckets / 2;
3648         u64 src_blkno;
3649         u64 last_cluster_blkno = bucket_blkno(first) +
3650                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3651
3652         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3653         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3654
3655         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3656              (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3657
3658         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3659                                      last_cluster_blkno, new_blkno,
3660                                      to_move, first_hash);
3661         if (ret) {
3662                 mlog_errno(ret);
3663                 goto out;
3664         }
3665
3666         /* This is the first bucket that got moved */
3667         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3668
3669         /*
3670          * If the target bucket was part of the moved buckets, we need to
3671          * update first and target.
3672          */
3673         if (bucket_blkno(target) >= src_blkno) {
3674                 /* Find the block for the new target bucket */
3675                 src_blkno = new_blkno +
3676                         (bucket_blkno(target) - src_blkno);
3677
3678                 ocfs2_xattr_bucket_relse(first);
3679                 ocfs2_xattr_bucket_relse(target);
3680
3681                 /*
3682                  * These shouldn't fail - the buffers are in the
3683                  * journal from ocfs2_cp_xattr_bucket().
3684                  */
3685                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3686                 if (ret) {
3687                         mlog_errno(ret);
3688                         goto out;
3689                 }
3690                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3691                 if (ret)
3692                         mlog_errno(ret);
3693
3694         }
3695
3696 out:
3697         return ret;
3698 }
3699
3700 /*
3701  * Find the suitable pos when we divide a bucket into 2.
3702  * We have to make sure the xattrs with the same hash value exist
3703  * in the same bucket.
3704  *
3705  * If this ocfs2_xattr_header covers more than one hash value, find a
3706  * place where the hash value changes.  Try to find the most even split.
3707  * The most common case is that all entries have different hash values,
3708  * and the first check we make will find a place to split.
3709  */
3710 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3711 {
3712         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3713         int count = le16_to_cpu(xh->xh_count);
3714         int delta, middle = count / 2;
3715
3716         /*
3717          * We start at the middle.  Each step gets farther away in both
3718          * directions.  We therefore hit the change in hash value
3719          * nearest to the middle.  Note that this loop does not execute for
3720          * count < 2.
3721          */
3722         for (delta = 0; delta < middle; delta++) {
3723                 /* Let's check delta earlier than middle */
3724                 if (cmp_xe(&entries[middle - delta - 1],
3725                            &entries[middle - delta]))
3726                         return middle - delta;
3727
3728                 /* For even counts, don't walk off the end */
3729                 if ((middle + delta + 1) == count)
3730                         continue;
3731
3732                 /* Now try delta past middle */
3733                 if (cmp_xe(&entries[middle + delta],
3734                            &entries[middle + delta + 1]))
3735                         return middle + delta + 1;
3736         }
3737
3738         /* Every entry had the same hash */
3739         return count;
3740 }
3741
3742 /*
3743  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3744  * first_hash will record the 1st hash of the new bucket.
3745  *
3746  * Normally half of the xattrs will be moved.  But we have to make
3747  * sure that the xattrs with the same hash value are stored in the
3748  * same bucket. If all the xattrs in this bucket have the same hash
3749  * value, the new bucket will be initialized as an empty one and the
3750  * first_hash will be initialized as (hash_value+1).
3751  */
3752 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3753                                     handle_t *handle,
3754                                     u64 blk,
3755                                     u64 new_blk,
3756                                     u32 *first_hash,
3757                                     int new_bucket_head)
3758 {
3759         int ret, i;
3760         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3761         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3762         struct ocfs2_xattr_header *xh;
3763         struct ocfs2_xattr_entry *xe;
3764         int blocksize = inode->i_sb->s_blocksize;
3765
3766         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3767              (unsigned long long)blk, (unsigned long long)new_blk);
3768
3769         s_bucket = ocfs2_xattr_bucket_new(inode);
3770         t_bucket = ocfs2_xattr_bucket_new(inode);
3771         if (!s_bucket || !t_bucket) {
3772                 ret = -ENOMEM;
3773                 mlog_errno(ret);
3774                 goto out;
3775         }
3776
3777         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3778         if (ret) {
3779                 mlog_errno(ret);
3780                 goto out;
3781         }
3782
3783         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3784                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3785         if (ret) {
3786                 mlog_errno(ret);
3787                 goto out;
3788         }
3789
3790         /*
3791          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3792          * there's no need to read it.
3793          */
3794         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3795         if (ret) {
3796                 mlog_errno(ret);
3797                 goto out;
3798         }
3799
3800         /*
3801          * Hey, if we're overwriting t_bucket, what difference does
3802          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
3803          * same part of ocfs2_cp_xattr_bucket().
3804          */
3805         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3806                                                 new_bucket_head ?
3807                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3808                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3809         if (ret) {
3810                 mlog_errno(ret);
3811                 goto out;
3812         }
3813
3814         xh = bucket_xh(s_bucket);
3815         count = le16_to_cpu(xh->xh_count);
3816         start = ocfs2_xattr_find_divide_pos(xh);
3817
3818         if (start == count) {
3819                 xe = &xh->xh_entries[start-1];
3820
3821                 /*
3822                  * initialized a new empty bucket here.
3823                  * The hash value is set as one larger than
3824                  * that of the last entry in the previous bucket.
3825                  */
3826                 for (i = 0; i < t_bucket->bu_blocks; i++)
3827                         memset(bucket_block(t_bucket, i), 0, blocksize);
3828
3829                 xh = bucket_xh(t_bucket);
3830                 xh->xh_free_start = cpu_to_le16(blocksize);
3831                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3832                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3833
3834                 goto set_num_buckets;
3835         }
3836
3837         /* copy the whole bucket to the new first. */
3838         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3839
3840         /* update the new bucket. */
3841         xh = bucket_xh(t_bucket);
3842
3843         /*
3844          * Calculate the total name/value len and xh_free_start for
3845          * the old bucket first.
3846          */
3847         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3848         name_value_len = 0;
3849         for (i = 0; i < start; i++) {
3850                 xe = &xh->xh_entries[i];
3851                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3852                 if (ocfs2_xattr_is_local(xe))
3853                         xe_len +=
3854                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3855                 else
3856                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3857                 name_value_len += xe_len;
3858                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3859                         name_offset = le16_to_cpu(xe->xe_name_offset);
3860         }
3861
3862         /*
3863          * Now begin the modification to the new bucket.
3864          *
3865          * In the new bucket, We just move the xattr entry to the beginning
3866          * and don't touch the name/value. So there will be some holes in the
3867          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3868          * called.
3869          */
3870         xe = &xh->xh_entries[start];
3871         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3872         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3873              (int)((char *)xe - (char *)xh),
3874              (int)((char *)xh->xh_entries - (char *)xh));
3875         memmove((char *)xh->xh_entries, (char *)xe, len);
3876         xe = &xh->xh_entries[count - start];
3877         len = sizeof(struct ocfs2_xattr_entry) * start;
3878         memset((char *)xe, 0, len);
3879
3880         le16_add_cpu(&xh->xh_count, -start);
3881         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3882
3883         /* Calculate xh_free_start for the new bucket. */
3884         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3885         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3886                 xe = &xh->xh_entries[i];
3887                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3888                 if (ocfs2_xattr_is_local(xe))
3889                         xe_len +=
3890                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3891                 else
3892                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3893                 if (le16_to_cpu(xe->xe_name_offset) <
3894                     le16_to_cpu(xh->xh_free_start))
3895                         xh->xh_free_start = xe->xe_name_offset;
3896         }
3897
3898 set_num_buckets:
3899         /* set xh->xh_num_buckets for the new xh. */
3900         if (new_bucket_head)
3901                 xh->xh_num_buckets = cpu_to_le16(1);
3902         else
3903                 xh->xh_num_buckets = 0;
3904
3905         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3906
3907         /* store the first_hash of the new bucket. */
3908         if (first_hash)
3909                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3910
3911         /*
3912          * Now only update the 1st block of the old bucket.  If we
3913          * just added a new empty bucket, there is no need to modify
3914          * it.
3915          */
3916         if (start == count)
3917                 goto out;
3918
3919         xh = bucket_xh(s_bucket);
3920         memset(&xh->xh_entries[start], 0,
3921                sizeof(struct ocfs2_xattr_entry) * (count - start));
3922         xh->xh_count = cpu_to_le16(start);
3923         xh->xh_free_start = cpu_to_le16(name_offset);
3924         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3925
3926         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3927
3928 out:
3929         ocfs2_xattr_bucket_free(s_bucket);
3930         ocfs2_xattr_bucket_free(t_bucket);
3931
3932         return ret;
3933 }
3934
3935 /*
3936  * Copy xattr from one bucket to another bucket.
3937  *
3938  * The caller must make sure that the journal transaction
3939  * has enough space for journaling.
3940  */
3941 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3942                                  handle_t *handle,
3943                                  u64 s_blkno,
3944                                  u64 t_blkno,
3945                                  int t_is_new)
3946 {
3947         int ret;
3948         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3949
3950         BUG_ON(s_blkno == t_blkno);
3951
3952         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3953              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3954              t_is_new);
3955
3956         s_bucket = ocfs2_xattr_bucket_new(inode);
3957         t_bucket = ocfs2_xattr_bucket_new(inode);
3958         if (!s_bucket || !t_bucket) {
3959                 ret = -ENOMEM;
3960                 mlog_errno(ret);
3961                 goto out;
3962         }
3963
3964         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3965         if (ret)
3966                 goto out;
3967
3968         /*
3969          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3970          * there's no need to read it.
3971          */
3972         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3973         if (ret)
3974                 goto out;
3975
3976         /*
3977          * Hey, if we're overwriting t_bucket, what difference does
3978          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
3979          * cluster to fill, we came here from
3980          * ocfs2_mv_xattr_buckets(), and it is really new -
3981          * ACCESS_CREATE is required.  But we also might have moved data
3982          * out of t_bucket before extending back into it.
3983          * ocfs2_add_new_xattr_bucket() can do this - its call to
3984          * ocfs2_add_new_xattr_cluster() may have created a new extent
3985          * and copied out the end of the old extent.  Then it re-extends
3986          * the old extent back to create space for new xattrs.  That's
3987          * how we get here, and the bucket isn't really new.
3988          */
3989         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3990                                                 t_is_new ?
3991                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3992                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3993         if (ret)
3994                 goto out;
3995
3996         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3997         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3998
3999 out:
4000         ocfs2_xattr_bucket_free(t_bucket);
4001         ocfs2_xattr_bucket_free(s_bucket);
4002
4003         return ret;
4004 }
4005
4006 /*
4007  * src_blk points to the start of an existing extent.  last_blk points to
4008  * last cluster in that extent.  to_blk points to a newly allocated
4009  * extent.  We copy the buckets from the cluster at last_blk to the new
4010  * extent.  If start_bucket is non-zero, we skip that many buckets before
4011  * we start copying.  The new extent's xh_num_buckets gets set to the
4012  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4013  * by the same amount.
4014  */
4015 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4016                                   u64 src_blk, u64 last_blk, u64 to_blk,
4017                                   unsigned int start_bucket,
4018                                   u32 *first_hash)
4019 {
4020         int i, ret, credits;
4021         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4022         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4023         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4024         struct ocfs2_xattr_bucket *old_first, *new_first;
4025
4026         mlog(0, "mv xattrs from cluster %llu to %llu\n",
4027              (unsigned long long)last_blk, (unsigned long long)to_blk);
4028
4029         BUG_ON(start_bucket >= num_buckets);
4030         if (start_bucket) {
4031                 num_buckets -= start_bucket;
4032                 last_blk += (start_bucket * blks_per_bucket);
4033         }
4034
4035         /* The first bucket of the original extent */
4036         old_first = ocfs2_xattr_bucket_new(inode);
4037         /* The first bucket of the new extent */
4038         new_first = ocfs2_xattr_bucket_new(inode);
4039         if (!old_first || !new_first) {
4040                 ret = -ENOMEM;
4041                 mlog_errno(ret);
4042                 goto out;
4043         }
4044
4045         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4046         if (ret) {
4047                 mlog_errno(ret);
4048                 goto out;
4049         }
4050
4051         /*
4052          * We need to update the first bucket of the old extent and all
4053          * the buckets going to the new extent.
4054          */
4055         credits = ((num_buckets + 1) * blks_per_bucket) +
4056                 handle->h_buffer_credits;
4057         ret = ocfs2_extend_trans(handle, credits);
4058         if (ret) {
4059                 mlog_errno(ret);
4060                 goto out;
4061         }
4062
4063         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4064                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4065         if (ret) {
4066                 mlog_errno(ret);
4067                 goto out;
4068         }
4069
4070         for (i = 0; i < num_buckets; i++) {
4071                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4072                                             last_blk + (i * blks_per_bucket),
4073                                             to_blk + (i * blks_per_bucket),
4074                                             1);
4075                 if (ret) {
4076                         mlog_errno(ret);
4077                         goto out;
4078                 }
4079         }
4080
4081         /*
4082          * Get the new bucket ready before we dirty anything
4083          * (This actually shouldn't fail, because we already dirtied
4084          * it once in ocfs2_cp_xattr_bucket()).
4085          */
4086         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4087         if (ret) {
4088                 mlog_errno(ret);
4089                 goto out;
4090         }
4091         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4092                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4093         if (ret) {
4094                 mlog_errno(ret);
4095                 goto out;
4096         }
4097
4098         /* Now update the headers */
4099         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4100         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4101
4102         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4103         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4104
4105         if (first_hash)
4106                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4107
4108 out:
4109         ocfs2_xattr_bucket_free(new_first);
4110         ocfs2_xattr_bucket_free(old_first);
4111         return ret;
4112 }
4113
4114 /*
4115  * Move some xattrs in this cluster to the new cluster.
4116  * This function should only be called when bucket size == cluster size.
4117  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4118  */
4119 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4120                                       handle_t *handle,
4121                                       u64 prev_blk,
4122                                       u64 new_blk,
4123                                       u32 *first_hash)
4124 {
4125         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4126         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4127
4128         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4129
4130         ret = ocfs2_extend_trans(handle, credits);
4131         if (ret) {
4132                 mlog_errno(ret);
4133                 return ret;
4134         }
4135
4136         /* Move half of the xattr in start_blk to the next bucket. */
4137         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4138                                           new_blk, first_hash, 1);
4139 }
4140
4141 /*
4142  * Move some xattrs from the old cluster to the new one since they are not
4143  * contiguous in ocfs2 xattr tree.
4144  *
4145  * new_blk starts a new separate cluster, and we will move some xattrs from
4146  * prev_blk to it. v_start will be set as the first name hash value in this
4147  * new cluster so that it can be used as e_cpos during tree insertion and
4148  * don't collide with our original b-tree operations. first_bh and header_bh
4149  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4150  * to extend the insert bucket.
4151  *
4152  * The problem is how much xattr should we move to the new one and when should
4153  * we update first_bh and header_bh?
4154  * 1. If cluster size > bucket size, that means the previous cluster has more
4155  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4156  *    update the first_bh and header_bh if the insert bucket has been moved
4157  *    to the new cluster.
4158  * 2. If cluster_size == bucket_size:
4159  *    a) If the previous extent rec has more than one cluster and the insert
4160  *       place isn't in the last cluster, copy the entire last cluster to the
4161  *       new one. This time, we don't need to upate the first_bh and header_bh
4162  *       since they will not be moved into the new cluster.
4163  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4164  *       the new one. And we set the extend flag to zero if the insert place is
4165  *       moved into the new allocated cluster since no extend is needed.
4166  */
4167 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4168                                             handle_t *handle,
4169                                             struct ocfs2_xattr_bucket *first,
4170                                             struct ocfs2_xattr_bucket *target,
4171                                             u64 new_blk,
4172                                             u32 prev_clusters,
4173                                             u32 *v_start,
4174                                             int *extend)
4175 {
4176         int ret;
4177
4178         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4179              (unsigned long long)bucket_blkno(first), prev_clusters,
4180              (unsigned long long)new_blk);
4181
4182         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4183                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4184                                                           handle,
4185                                                           first, target,
4186                                                           new_blk,
4187                                                           prev_clusters,
4188                                                           v_start);
4189                 if (ret)
4190                         mlog_errno(ret);
4191         } else {
4192                 /* The start of the last cluster in the first extent */
4193                 u64 last_blk = bucket_blkno(first) +
4194                         ((prev_clusters - 1) *
4195                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
4196
4197                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4198                         ret = ocfs2_mv_xattr_buckets(inode, handle,
4199                                                      bucket_blkno(first),
4200                                                      last_blk, new_blk, 0,
4201                                                      v_start);
4202                         if (ret)
4203                                 mlog_errno(ret);
4204                 } else {
4205                         ret = ocfs2_divide_xattr_cluster(inode, handle,
4206                                                          last_blk, new_blk,
4207                                                          v_start);
4208                         if (ret)
4209                                 mlog_errno(ret);
4210
4211                         if ((bucket_blkno(target) == last_blk) && extend)
4212                                 *extend = 0;
4213                 }
4214         }
4215
4216         return ret;
4217 }
4218
4219 /*
4220  * Add a new cluster for xattr storage.
4221  *
4222  * If the new cluster is contiguous with the previous one, it will be
4223  * appended to the same extent record, and num_clusters will be updated.
4224  * If not, we will insert a new extent for it and move some xattrs in
4225  * the last cluster into the new allocated one.
4226  * We also need to limit the maximum size of a btree leaf, otherwise we'll
4227  * lose the benefits of hashing because we'll have to search large leaves.
4228  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4229  * if it's bigger).
4230  *
4231  * first_bh is the first block of the previous extent rec and header_bh
4232  * indicates the bucket we will insert the new xattrs. They will be updated
4233  * when the header_bh is moved into the new cluster.
4234  */
4235 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4236                                        struct buffer_head *root_bh,
4237                                        struct ocfs2_xattr_bucket *first,
4238                                        struct ocfs2_xattr_bucket *target,
4239                                        u32 *num_clusters,
4240                                        u32 prev_cpos,
4241                                        int *extend,
4242                                        struct ocfs2_xattr_set_ctxt *ctxt)
4243 {
4244         int ret;
4245         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4246         u32 prev_clusters = *num_clusters;
4247         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4248         u64 block;
4249         handle_t *handle = ctxt->handle;
4250         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4251         struct ocfs2_extent_tree et;
4252
4253         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4254              "previous xattr blkno = %llu\n",
4255              (unsigned long long)OCFS2_I(inode)->ip_blkno,
4256              prev_cpos, (unsigned long long)bucket_blkno(first));
4257
4258         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4259
4260         ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4261                                       OCFS2_JOURNAL_ACCESS_WRITE);
4262         if (ret < 0) {
4263                 mlog_errno(ret);
4264                 goto leave;
4265         }
4266
4267         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4268                                      clusters_to_add, &bit_off, &num_bits);
4269         if (ret < 0) {
4270                 if (ret != -ENOSPC)
4271                         mlog_errno(ret);
4272                 goto leave;
4273         }
4274
4275         BUG_ON(num_bits > clusters_to_add);
4276
4277         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4278         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4279              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4280
4281         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4282             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4283              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4284                 /*
4285                  * If this cluster is contiguous with the old one and
4286                  * adding this new cluster, we don't surpass the limit of
4287                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4288                  * initialized and used like other buckets in the previous
4289                  * cluster.
4290                  * So add it as a contiguous one. The caller will handle
4291                  * its init process.
4292                  */
4293                 v_start = prev_cpos + prev_clusters;
4294                 *num_clusters = prev_clusters + num_bits;
4295                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4296                      num_bits);
4297         } else {
4298                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4299                                                        handle,
4300                                                        first,
4301                                                        target,
4302                                                        block,
4303                                                        prev_clusters,
4304                                                        &v_start,
4305                                                        extend);
4306                 if (ret) {
4307                         mlog_errno(ret);
4308                         goto leave;
4309                 }
4310         }
4311
4312         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4313              num_bits, (unsigned long long)block, v_start);
4314         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
4315                                   num_bits, 0, ctxt->meta_ac);
4316         if (ret < 0) {
4317                 mlog_errno(ret);
4318                 goto leave;
4319         }
4320
4321         ret = ocfs2_journal_dirty(handle, root_bh);
4322         if (ret < 0)
4323                 mlog_errno(ret);
4324
4325 leave:
4326         return ret;
4327 }
4328
4329 /*
4330  * We are given an extent.  'first' is the bucket at the very front of
4331  * the extent.  The extent has space for an additional bucket past
4332  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
4333  * of the target bucket.  We wish to shift every bucket past the target
4334  * down one, filling in that additional space.  When we get back to the
4335  * target, we split the target between itself and the now-empty bucket
4336  * at target+1 (aka, target_blkno + blks_per_bucket).
4337  */
4338 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4339                                      handle_t *handle,
4340                                      struct ocfs2_xattr_bucket *first,
4341                                      u64 target_blk,
4342                                      u32 num_clusters)
4343 {
4344         int ret, credits;
4345         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4346         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4347         u64 end_blk;
4348         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4349
4350         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4351              "from %llu, len = %u\n", (unsigned long long)target_blk,
4352              (unsigned long long)bucket_blkno(first), num_clusters);
4353
4354         /* The extent must have room for an additional bucket */
4355         BUG_ON(new_bucket >=
4356                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4357
4358         /* end_blk points to the last existing bucket */
4359         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4360
4361         /*
4362          * end_blk is the start of the last existing bucket.
4363          * Thus, (end_blk - target_blk) covers the target bucket and
4364          * every bucket after it up to, but not including, the last
4365          * existing bucket.  Then we add the last existing bucket, the
4366          * new bucket, and the first bucket (3 * blk_per_bucket).
4367          */
4368         credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4369                   handle->h_buffer_credits;
4370         ret = ocfs2_extend_trans(handle, credits);
4371         if (ret) {
4372                 mlog_errno(ret);
4373                 goto out;
4374         }
4375
4376         ret = ocfs2_xattr_bucket_journal_access(handle, first,
4377                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4378         if (ret) {
4379                 mlog_errno(ret);
4380                 goto out;
4381         }
4382
4383         while (end_blk != target_blk) {
4384                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4385                                             end_blk + blk_per_bucket, 0);
4386                 if (ret)
4387                         goto out;
4388                 end_blk -= blk_per_bucket;
4389         }
4390
4391         /* Move half of the xattr in target_blkno to the next bucket. */
4392         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4393                                         target_blk + blk_per_bucket, NULL, 0);
4394
4395         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4396         ocfs2_xattr_bucket_journal_dirty(handle, first);
4397
4398 out:
4399         return ret;
4400 }
4401
4402 /*
4403  * Add new xattr bucket in an extent record and adjust the buckets
4404  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
4405  * bucket we want to insert into.
4406  *
4407  * In the easy case, we will move all the buckets after target down by
4408  * one. Half of target's xattrs will be moved to the next bucket.
4409  *
4410  * If current cluster is full, we'll allocate a new one.  This may not
4411  * be contiguous.  The underlying calls will make sure that there is
4412  * space for the insert, shifting buckets around if necessary.
4413  * 'target' may be moved by those calls.
4414  */
4415 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4416                                       struct buffer_head *xb_bh,
4417                                       struct ocfs2_xattr_bucket *target,
4418                                       struct ocfs2_xattr_set_ctxt *ctxt)
4419 {
4420         struct ocfs2_xattr_block *xb =
4421                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4422         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4423         struct ocfs2_extent_list *el = &xb_root->xt_list;
4424         u32 name_hash =
4425                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4426         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4427         int ret, num_buckets, extend = 1;
4428         u64 p_blkno;
4429         u32 e_cpos, num_clusters;
4430         /* The bucket at the front of the extent */
4431         struct ocfs2_xattr_bucket *first;
4432
4433         mlog(0, "Add new xattr bucket starting from %llu\n",
4434              (unsigned long long)bucket_blkno(target));
4435
4436         /* The first bucket of the original extent */
4437         first = ocfs2_xattr_bucket_new(inode);
4438         if (!first) {
4439                 ret = -ENOMEM;
4440                 mlog_errno(ret);
4441                 goto out;
4442         }
4443
4444         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4445                                   &num_clusters, el);
4446         if (ret) {
4447                 mlog_errno(ret);
4448                 goto out;
4449         }
4450
4451         ret = ocfs2_read_xattr_bucket(first, p_blkno);
4452         if (ret) {
4453                 mlog_errno(ret);
4454                 goto out;
4455         }
4456
4457         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4458         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4459                 /*
4460                  * This can move first+target if the target bucket moves
4461                  * to the new extent.
4462                  */
4463                 ret = ocfs2_add_new_xattr_cluster(inode,
4464                                                   xb_bh,
4465                                                   first,
4466                                                   target,
4467                                                   &num_clusters,
4468                                                   e_cpos,
4469                                                   &extend,
4470                                                   ctxt);
4471                 if (ret) {
4472                         mlog_errno(ret);
4473                         goto out;
4474                 }
4475         }
4476
4477         if (extend) {
4478                 ret = ocfs2_extend_xattr_bucket(inode,
4479                                                 ctxt->handle,
4480                                                 first,
4481                                                 bucket_blkno(target),
4482                                                 num_clusters);
4483                 if (ret)
4484                         mlog_errno(ret);
4485         }
4486
4487 out:
4488         ocfs2_xattr_bucket_free(first);
4489
4490         return ret;
4491 }
4492
4493 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4494                                         struct ocfs2_xattr_bucket *bucket,
4495                                         int offs)
4496 {
4497         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4498
4499         offs = offs % inode->i_sb->s_blocksize;
4500         return bucket_block(bucket, block_off) + offs;
4501 }
4502
4503 /*
4504  * Handle the normal xattr set, including replace, delete and new.
4505  *
4506  * Note: "local" indicates the real data's locality. So we can't
4507  * just its bucket locality by its length.
4508  */
4509 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4510                                          struct ocfs2_xattr_info *xi,
4511                                          struct ocfs2_xattr_search *xs,
4512                                          u32 name_hash,
4513                                          int local)
4514 {
4515         struct ocfs2_xattr_entry *last, *xe;
4516         int name_len = strlen(xi->name);
4517         struct ocfs2_xattr_header *xh = xs->header;
4518         u16 count = le16_to_cpu(xh->xh_count), start;
4519         size_t blocksize = inode->i_sb->s_blocksize;
4520         char *val;
4521         size_t offs, size, new_size;
4522
4523         last = &xh->xh_entries[count];
4524         if (!xs->not_found) {
4525                 xe = xs->here;
4526                 offs = le16_to_cpu(xe->xe_name_offset);
4527                 if (ocfs2_xattr_is_local(xe))
4528                         size = OCFS2_XATTR_SIZE(name_len) +
4529                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4530                 else
4531                         size = OCFS2_XATTR_SIZE(name_len) +
4532                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4533
4534                 /*
4535                  * If the new value will be stored outside, xi->value has been
4536                  * initalized as an empty ocfs2_xattr_value_root, and the same
4537                  * goes with xi->value_len, so we can set new_size safely here.
4538                  * See ocfs2_xattr_set_in_bucket.
4539                  */
4540                 new_size = OCFS2_XATTR_SIZE(name_len) +
4541                            OCFS2_XATTR_SIZE(xi->value_len);
4542
4543                 le16_add_cpu(&xh->xh_name_value_len, -size);
4544                 if (xi->value) {
4545                         if (new_size > size)
4546                                 goto set_new_name_value;
4547
4548                         /* Now replace the old value with new one. */
4549                         if (local)
4550                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4551                         else
4552                                 xe->xe_value_size = 0;
4553
4554                         val = ocfs2_xattr_bucket_get_val(inode,
4555                                                          xs->bucket, offs);
4556                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4557                                size - OCFS2_XATTR_SIZE(name_len));
4558                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4559                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4560                                        xi->value, xi->value_len);
4561
4562                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4563                         ocfs2_xattr_set_local(xe, local);
4564                         return;
4565                 } else {
4566                         /*
4567                          * Remove the old entry if there is more than one.
4568                          * We don't remove the last entry so that we can
4569                          * use it to indicate the hash value of the empty
4570                          * bucket.
4571                          */
4572                         last -= 1;
4573                         le16_add_cpu(&xh->xh_count, -1);
4574                         if (xh->xh_count) {
4575                                 memmove(xe, xe + 1,
4576                                         (void *)last - (void *)xe);
4577                                 memset(last, 0,
4578                                        sizeof(struct ocfs2_xattr_entry));
4579                         } else
4580                                 xh->xh_free_start =
4581                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4582
4583                         return;
4584                 }
4585         } else {
4586                 /* find a new entry for insert. */
4587                 int low = 0, high = count - 1, tmp;
4588                 struct ocfs2_xattr_entry *tmp_xe;
4589
4590                 while (low <= high && count) {
4591                         tmp = (low + high) / 2;
4592                         tmp_xe = &xh->xh_entries[tmp];
4593
4594                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4595                                 low = tmp + 1;
4596                         else if (name_hash <
4597                                  le32_to_cpu(tmp_xe->xe_name_hash))
4598                                 high = tmp - 1;
4599                         else {
4600                                 low = tmp;
4601                                 break;
4602                         }
4603                 }
4604
4605                 xe = &xh->xh_entries[low];
4606                 if (low != count)
4607                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4608
4609                 le16_add_cpu(&xh->xh_count, 1);
4610                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4611                 xe->xe_name_hash = cpu_to_le32(name_hash);
4612                 xe->xe_name_len = name_len;
4613                 ocfs2_xattr_set_type(xe, xi->name_index);
4614         }
4615
4616 set_new_name_value:
4617         /* Insert the new name+value. */
4618         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4619
4620         /*
4621          * We must make sure that the name/value pair
4622          * exists in the same block.
4623          */
4624         offs = le16_to_cpu(xh->xh_free_start);
4625         start = offs - size;
4626
4627         if (start >> inode->i_sb->s_blocksize_bits !=
4628             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4629                 offs = offs - offs % blocksize;
4630                 xh->xh_free_start = cpu_to_le16(offs);
4631         }
4632
4633         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4634         xe->xe_name_offset = cpu_to_le16(offs - size);
4635
4636         memset(val, 0, size);
4637         memcpy(val, xi->name, name_len);
4638         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4639
4640         xe->xe_value_size = cpu_to_le64(xi->value_len);
4641         ocfs2_xattr_set_local(xe, local);
4642         xs->here = xe;
4643         le16_add_cpu(&xh->xh_free_start, -size);
4644         le16_add_cpu(&xh->xh_name_value_len, size);
4645
4646         return;
4647 }
4648
4649 /*
4650  * Set the xattr entry in the specified bucket.
4651  * The bucket is indicated by xs->bucket and it should have the enough
4652  * space for the xattr insertion.
4653  */
4654 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4655                                            handle_t *handle,
4656                                            struct ocfs2_xattr_info *xi,
4657                                            struct ocfs2_xattr_search *xs,
4658                                            u32 name_hash,
4659                                            int local)
4660 {
4661         int ret;
4662         u64 blkno;
4663
4664         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4665              (unsigned long)xi->value_len, xi->name_index,
4666              (unsigned long long)bucket_blkno(xs->bucket));
4667
4668         if (!xs->bucket->bu_bhs[1]) {
4669                 blkno = bucket_blkno(xs->bucket);
4670                 ocfs2_xattr_bucket_relse(xs->bucket);
4671                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4672                 if (ret) {
4673                         mlog_errno(ret);
4674                         goto out;
4675                 }
4676         }
4677
4678         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4679                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4680         if (ret < 0) {
4681                 mlog_errno(ret);
4682                 goto out;
4683         }
4684
4685         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4686         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4687
4688 out:
4689         return ret;
4690 }
4691
4692 /*
4693  * Truncate the specified xe_off entry in xattr bucket.
4694  * bucket is indicated by header_bh and len is the new length.
4695  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4696  *
4697  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4698  */
4699 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4700                                              struct ocfs2_xattr_bucket *bucket,
4701                                              int xe_off,
4702                                              int len,
4703                                              struct ocfs2_xattr_set_ctxt *ctxt)
4704 {
4705         int ret, offset;
4706         u64 value_blk;
4707         struct ocfs2_xattr_entry *xe;
4708         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4709         size_t blocksize = inode->i_sb->s_blocksize;
4710         struct ocfs2_xattr_value_buf vb = {
4711                 .vb_access = ocfs2_journal_access,
4712         };
4713
4714         xe = &xh->xh_entries[xe_off];
4715
4716         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4717
4718         offset = le16_to_cpu(xe->xe_name_offset) +
4719                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4720
4721         value_blk = offset / blocksize;
4722
4723         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4724         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4725
4726         vb.vb_bh = bucket->bu_bhs[value_blk];
4727         BUG_ON(!vb.vb_bh);
4728
4729         vb.vb_xv = (struct ocfs2_xattr_value_root *)
4730                 (vb.vb_bh->b_data + offset % blocksize);
4731
4732         /*
4733          * From here on out we have to dirty the bucket.  The generic
4734          * value calls only modify one of the bucket's bhs, but we need
4735          * to send the bucket at once.  So if they error, they *could* have
4736          * modified something.  We have to assume they did, and dirty
4737          * the whole bucket.  This leaves us in a consistent state.
4738          */
4739         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4740              xe_off, (unsigned long long)bucket_blkno(bucket), len);
4741         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4742         if (ret) {
4743                 mlog_errno(ret);
4744                 goto out;
4745         }
4746
4747         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4748                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4749         if (ret) {
4750                 mlog_errno(ret);
4751                 goto out;
4752         }
4753
4754         xe->xe_value_size = cpu_to_le64(len);
4755
4756         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4757
4758 out:
4759         return ret;
4760 }
4761
4762 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4763                                         struct ocfs2_xattr_search *xs,
4764                                         int len,
4765                                         struct ocfs2_xattr_set_ctxt *ctxt)
4766 {
4767         int ret, offset;
4768         struct ocfs2_xattr_entry *xe = xs->here;
4769         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4770
4771         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4772
4773         offset = xe - xh->xh_entries;
4774         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4775                                                 offset, len, ctxt);
4776         if (ret)
4777                 mlog_errno(ret);
4778
4779         return ret;
4780 }
4781
4782 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4783                                                 handle_t *handle,
4784                                                 struct ocfs2_xattr_search *xs,
4785                                                 char *val,
4786                                                 int value_len)
4787 {
4788         int offset;
4789         struct ocfs2_xattr_value_root *xv;
4790         struct ocfs2_xattr_entry *xe = xs->here;
4791
4792         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4793
4794         offset = le16_to_cpu(xe->xe_name_offset) +
4795                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4796
4797         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4798
4799         return __ocfs2_xattr_set_value_outside(inode, handle,
4800                                                xv, val, value_len);
4801 }
4802
4803 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4804                                   struct buffer_head *root_bh,
4805                                   u64 blkno,
4806                                   u32 cpos,
4807                                   u32 len)
4808 {
4809         int ret;
4810         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4811         struct inode *tl_inode = osb->osb_tl_inode;
4812         handle_t *handle;
4813         struct ocfs2_xattr_block *xb =
4814                         (struct ocfs2_xattr_block *)root_bh->b_data;
4815         struct ocfs2_alloc_context *meta_ac = NULL;
4816         struct ocfs2_cached_dealloc_ctxt dealloc;
4817         struct ocfs2_extent_tree et;
4818
4819         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4820
4821         ocfs2_init_dealloc_ctxt(&dealloc);
4822
4823         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4824              cpos, len, (unsigned long long)blkno);
4825
4826         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4827
4828         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4829         if (ret) {
4830                 mlog_errno(ret);
4831                 return ret;
4832         }
4833
4834         mutex_lock(&tl_inode->i_mutex);
4835
4836         if (ocfs2_truncate_log_needs_flush(osb)) {
4837                 ret = __ocfs2_flush_truncate_log(osb);
4838                 if (ret < 0) {
4839                         mlog_errno(ret);
4840                         goto out;
4841                 }
4842         }
4843
4844         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4845         if (IS_ERR(handle)) {
4846                 ret = -ENOMEM;
4847                 mlog_errno(ret);
4848                 goto out;
4849         }
4850
4851         ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4852                                       OCFS2_JOURNAL_ACCESS_WRITE);
4853         if (ret) {
4854                 mlog_errno(ret);
4855                 goto out_commit;
4856         }
4857
4858         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4859                                   &dealloc);
4860         if (ret) {
4861                 mlog_errno(ret);
4862                 goto out_commit;
4863         }
4864
4865         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4866
4867         ret = ocfs2_journal_dirty(handle, root_bh);
4868         if (ret) {
4869                 mlog_errno(ret);
4870                 goto out_commit;
4871         }
4872
4873         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4874         if (ret)
4875                 mlog_errno(ret);
4876
4877 out_commit:
4878         ocfs2_commit_trans(osb, handle);
4879 out:
4880         ocfs2_schedule_truncate_log_flush(osb, 1);
4881
4882         mutex_unlock(&tl_inode->i_mutex);
4883
4884         if (meta_ac)
4885                 ocfs2_free_alloc_context(meta_ac);
4886
4887         ocfs2_run_deallocs(osb, &dealloc);
4888
4889         return ret;
4890 }
4891
4892 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4893                                          handle_t *handle,
4894                                          struct ocfs2_xattr_search *xs)
4895 {
4896         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4897         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4898                                                 le16_to_cpu(xh->xh_count) - 1];
4899         int ret = 0;
4900
4901         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4902                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4903         if (ret) {
4904                 mlog_errno(ret);
4905                 return;
4906         }
4907
4908         /* Remove the old entry. */
4909         memmove(xs->here, xs->here + 1,
4910                 (void *)last - (void *)xs->here);
4911         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4912         le16_add_cpu(&xh->xh_count, -1);
4913
4914         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4915 }
4916
4917 /*
4918  * Set the xattr name/value in the bucket specified in xs.
4919  *
4920  * As the new value in xi may be stored in the bucket or in an outside cluster,
4921  * we divide the whole process into 3 steps:
4922  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4923  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4924  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4925  * 4. If the clusters for the new outside value can't be allocated, we need
4926  *    to free the xattr we allocated in set.
4927  */
4928 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4929                                      struct ocfs2_xattr_info *xi,
4930                                      struct ocfs2_xattr_search *xs,
4931                                      struct ocfs2_xattr_set_ctxt *ctxt)
4932 {
4933         int ret, local = 1;
4934         size_t value_len;
4935         char *val = (char *)xi->value;
4936         struct ocfs2_xattr_entry *xe = xs->here;
4937         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4938                                               strlen(xi->name));
4939
4940         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4941                 /*
4942                  * We need to truncate the xattr storage first.
4943                  *
4944                  * If both the old and new value are stored to
4945                  * outside block, we only need to truncate
4946                  * the storage and then set the value outside.
4947                  *
4948                  * If the new value should be stored within block,
4949                  * we should free all the outside block first and
4950                  * the modification to the xattr block will be done
4951                  * by following steps.
4952                  */
4953                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4954                         value_len = xi->value_len;
4955                 else
4956                         value_len = 0;
4957
4958                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4959                                                            value_len,
4960                                                            ctxt);
4961                 if (ret)
4962                         goto out;
4963
4964                 if (value_len)
4965                         goto set_value_outside;
4966         }
4967
4968         value_len = xi->value_len;
4969         /* So we have to handle the inside block change now. */
4970         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4971                 /*
4972                  * If the new value will be stored outside of block,
4973                  * initalize a new empty value root and insert it first.
4974                  */
4975                 local = 0;
4976                 xi->value = &def_xv;
4977                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4978         }
4979
4980         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
4981                                               name_hash, local);
4982         if (ret) {
4983                 mlog_errno(ret);
4984                 goto out;
4985         }
4986
4987         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4988                 goto out;
4989
4990         /* allocate the space now for the outside block storage. */
4991         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4992                                                    value_len, ctxt);
4993         if (ret) {
4994                 mlog_errno(ret);
4995
4996                 if (xs->not_found) {
4997                         /*
4998                          * We can't allocate enough clusters for outside
4999                          * storage and we have allocated xattr already,
5000                          * so need to remove it.
5001                          */
5002                         ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
5003                 }
5004                 goto out;
5005         }
5006
5007 set_value_outside:
5008         ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5009                                                    xs, val, value_len);
5010 out:
5011         return ret;
5012 }
5013
5014 /*
5015  * check whether the xattr bucket is filled up with the same hash value.
5016  * If we want to insert the xattr with the same hash, return -ENOSPC.
5017  * If we want to insert a xattr with different hash value, go ahead
5018  * and ocfs2_divide_xattr_bucket will handle this.
5019  */
5020 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5021                                               struct ocfs2_xattr_bucket *bucket,
5022                                               const char *name)
5023 {
5024         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5025         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5026
5027         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5028                 return 0;
5029
5030         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5031             xh->xh_entries[0].xe_name_hash) {
5032                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5033                      "hash = %u\n",
5034                      (unsigned long long)bucket_blkno(bucket),
5035                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5036                 return -ENOSPC;
5037         }
5038
5039         return 0;
5040 }
5041
5042 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5043                                              struct ocfs2_xattr_info *xi,
5044                                              struct ocfs2_xattr_search *xs,
5045                                              struct ocfs2_xattr_set_ctxt *ctxt)
5046 {
5047         struct ocfs2_xattr_header *xh;
5048         struct ocfs2_xattr_entry *xe;
5049         u16 count, header_size, xh_free_start;
5050         int free, max_free, need, old;
5051         size_t value_size = 0, name_len = strlen(xi->name);
5052         size_t blocksize = inode->i_sb->s_blocksize;
5053         int ret, allocation = 0;
5054
5055         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5056
5057 try_again:
5058         xh = xs->header;
5059         count = le16_to_cpu(xh->xh_count);
5060         xh_free_start = le16_to_cpu(xh->xh_free_start);
5061         header_size = sizeof(struct ocfs2_xattr_header) +
5062                         count * sizeof(struct ocfs2_xattr_entry);
5063         max_free = OCFS2_XATTR_BUCKET_SIZE -
5064                 le16_to_cpu(xh->xh_name_value_len) - header_size;
5065
5066         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5067                         "of %u which exceed block size\n",
5068                         (unsigned long long)bucket_blkno(xs->bucket),
5069                         header_size);
5070
5071         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5072                 value_size = OCFS2_XATTR_ROOT_SIZE;
5073         else if (xi->value)
5074                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5075
5076         if (xs->not_found)
5077                 need = sizeof(struct ocfs2_xattr_entry) +
5078                         OCFS2_XATTR_SIZE(name_len) + value_size;
5079         else {
5080                 need = value_size + OCFS2_XATTR_SIZE(name_len);
5081
5082                 /*
5083                  * We only replace the old value if the new length is smaller
5084                  * than the old one. Otherwise we will allocate new space in the
5085                  * bucket to store it.
5086                  */
5087                 xe = xs->here;
5088                 if (ocfs2_xattr_is_local(xe))
5089                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5090                 else
5091                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5092
5093                 if (old >= value_size)
5094                         need = 0;
5095         }
5096
5097         free = xh_free_start - header_size;
5098         /*
5099          * We need to make sure the new name/value pair
5100          * can exist in the same block.
5101          */
5102         if (xh_free_start % blocksize < need)
5103                 free -= xh_free_start % blocksize;
5104
5105         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5106              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5107              " %u\n", xs->not_found,
5108              (unsigned long long)bucket_blkno(xs->bucket),
5109              free, need, max_free, le16_to_cpu(xh->xh_free_start),
5110              le16_to_cpu(xh->xh_name_value_len));
5111
5112         if (free < need ||
5113             (xs->not_found &&
5114              count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5115                 if (need <= max_free &&
5116                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5117                         /*
5118                          * We can create the space by defragment. Since only the
5119                          * name/value will be moved, the xe shouldn't be changed
5120                          * in xs.
5121                          */
5122                         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5123                                                         xs->bucket);
5124                         if (ret) {
5125                                 mlog_errno(ret);
5126                                 goto out;
5127                         }
5128
5129                         xh_free_start = le16_to_cpu(xh->xh_free_start);
5130                         free = xh_free_start - header_size;
5131                         if (xh_free_start % blocksize < need)
5132                                 free -= xh_free_start % blocksize;
5133
5134                         if (free >= need)
5135                                 goto xattr_set;
5136
5137                         mlog(0, "Can't get enough space for xattr insert by "
5138                              "defragment. Need %u bytes, but we have %d, so "
5139                              "allocate new bucket for it.\n", need, free);
5140                 }
5141
5142                 /*
5143                  * We have to add new buckets or clusters and one
5144                  * allocation should leave us enough space for insert.
5145                  */
5146                 BUG_ON(allocation);
5147
5148                 /*
5149                  * We do not allow for overlapping ranges between buckets. And
5150                  * the maximum number of collisions we will allow for then is
5151                  * one bucket's worth, so check it here whether we need to
5152                  * add a new bucket for the insert.
5153                  */
5154                 ret = ocfs2_check_xattr_bucket_collision(inode,
5155                                                          xs->bucket,
5156                                                          xi->name);
5157                 if (ret) {
5158                         mlog_errno(ret);
5159                         goto out;
5160                 }
5161
5162                 ret = ocfs2_add_new_xattr_bucket(inode,
5163                                                  xs->xattr_bh,
5164                                                  xs->bucket,
5165                                                  ctxt);
5166                 if (ret) {
5167                         mlog_errno(ret);
5168                         goto out;
5169                 }
5170
5171                 /*
5172                  * ocfs2_add_new_xattr_bucket() will have updated
5173                  * xs->bucket if it moved, but it will not have updated
5174                  * any of the other search fields.  Thus, we drop it and
5175                  * re-search.  Everything should be cached, so it'll be
5176                  * quick.
5177                  */
5178                 ocfs2_xattr_bucket_relse(xs->bucket);
5179                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5180                                                    xi->name_index,
5181                                                    xi->name, xs);
5182                 if (ret && ret != -ENODATA)
5183                         goto out;
5184                 xs->not_found = ret;
5185                 allocation = 1;
5186                 goto try_again;
5187         }
5188
5189 xattr_set:
5190         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5191 out:
5192         mlog_exit(ret);
5193         return ret;
5194 }
5195
5196 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5197                                         struct ocfs2_xattr_bucket *bucket,
5198                                         void *para)
5199 {
5200         int ret = 0;
5201         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5202         u16 i;
5203         struct ocfs2_xattr_entry *xe;
5204         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5205         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5206         int credits = ocfs2_remove_extent_credits(osb->sb) +
5207                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5208
5209
5210         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5211
5212         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5213                 xe = &xh->xh_entries[i];
5214                 if (ocfs2_xattr_is_local(xe))
5215                         continue;
5216
5217                 ctxt.handle = ocfs2_start_trans(osb, credits);
5218                 if (IS_ERR(ctxt.handle)) {
5219                         ret = PTR_ERR(ctxt.handle);
5220                         mlog_errno(ret);
5221                         break;
5222                 }
5223
5224                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5225                                                         i, 0, &ctxt);
5226
5227                 ocfs2_commit_trans(osb, ctxt.handle);
5228                 if (ret) {
5229                         mlog_errno(ret);
5230                         break;
5231                 }
5232         }
5233
5234         ocfs2_schedule_truncate_log_flush(osb, 1);
5235         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5236         return ret;
5237 }
5238
5239 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5240                                           struct buffer_head *xb_bh)
5241 {
5242         struct ocfs2_xattr_block *xb =
5243                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5244         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5245         int ret = 0;
5246         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5247         u64 p_blkno;
5248
5249         if (le16_to_cpu(el->l_next_free_rec) == 0)
5250                 return 0;
5251
5252         while (name_hash > 0) {
5253                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5254                                           &e_cpos, &num_clusters, el);
5255                 if (ret) {
5256                         mlog_errno(ret);
5257                         goto out;
5258                 }
5259
5260                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5261                                                   ocfs2_delete_xattr_in_bucket,
5262                                                   NULL);
5263                 if (ret) {
5264                         mlog_errno(ret);
5265                         goto out;
5266                 }
5267
5268                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5269                                              p_blkno, e_cpos, num_clusters);
5270                 if (ret) {
5271                         mlog_errno(ret);
5272                         break;
5273                 }
5274
5275                 if (e_cpos == 0)
5276                         break;
5277
5278                 name_hash = e_cpos - 1;
5279         }
5280
5281 out:
5282         return ret;
5283 }
5284
5285 /*
5286  * 'security' attributes support
5287  */
5288 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5289                                         size_t list_size, const char *name,
5290                                         size_t name_len)
5291 {
5292         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5293         const size_t total_len = prefix_len + name_len + 1;
5294
5295         if (list && total_len <= list_size) {
5296                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5297                 memcpy(list + prefix_len, name, name_len);
5298                 list[prefix_len + name_len] = '\0';
5299         }
5300         return total_len;
5301 }
5302
5303 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5304                                     void *buffer, size_t size)
5305 {
5306         if (strcmp(name, "") == 0)
5307                 return -EINVAL;
5308         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5309                                buffer, size);
5310 }
5311
5312 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5313                                     const void *value, size_t size, int flags)
5314 {
5315         if (strcmp(name, "") == 0)
5316                 return -EINVAL;
5317
5318         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5319                                size, flags);
5320 }
5321
5322 int ocfs2_init_security_get(struct inode *inode,
5323                             struct inode *dir,
5324                             struct ocfs2_security_xattr_info *si)
5325 {
5326         /* check whether ocfs2 support feature xattr */
5327         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
5328                 return -EOPNOTSUPP;
5329         return security_inode_init_security(inode, dir, &si->name, &si->value,
5330                                             &si->value_len);
5331 }
5332
5333 int ocfs2_init_security_set(handle_t *handle,
5334                             struct inode *inode,
5335                             struct buffer_head *di_bh,
5336                             struct ocfs2_security_xattr_info *si,
5337                             struct ocfs2_alloc_context *xattr_ac,
5338                             struct ocfs2_alloc_context *data_ac)
5339 {
5340         return ocfs2_xattr_set_handle(handle, inode, di_bh,
5341                                      OCFS2_XATTR_INDEX_SECURITY,
5342                                      si->name, si->value, si->value_len, 0,
5343                                      xattr_ac, data_ac);
5344 }
5345
5346 struct xattr_handler ocfs2_xattr_security_handler = {
5347         .prefix = XATTR_SECURITY_PREFIX,
5348         .list   = ocfs2_xattr_security_list,
5349         .get    = ocfs2_xattr_security_get,
5350         .set    = ocfs2_xattr_security_set,
5351 };
5352
5353 /*
5354  * 'trusted' attributes support
5355  */
5356 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5357                                        size_t list_size, const char *name,
5358                                        size_t name_len)
5359 {
5360         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5361         const size_t total_len = prefix_len + name_len + 1;
5362
5363         if (list && total_len <= list_size) {
5364                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5365                 memcpy(list + prefix_len, name, name_len);
5366                 list[prefix_len + name_len] = '\0';
5367         }
5368         return total_len;
5369 }
5370
5371 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5372                                    void *buffer, size_t size)
5373 {
5374         if (strcmp(name, "") == 0)
5375                 return -EINVAL;
5376         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5377                                buffer, size);
5378 }
5379
5380 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5381                                    const void *value, size_t size, int flags)
5382 {
5383         if (strcmp(name, "") == 0)
5384                 return -EINVAL;
5385
5386         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5387                                size, flags);
5388 }
5389
5390 struct xattr_handler ocfs2_xattr_trusted_handler = {
5391         .prefix = XATTR_TRUSTED_PREFIX,
5392         .list   = ocfs2_xattr_trusted_list,
5393         .get    = ocfs2_xattr_trusted_get,
5394         .set    = ocfs2_xattr_trusted_set,
5395 };
5396
5397 /*
5398  * 'user' attributes support
5399  */
5400 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5401                                     size_t list_size, const char *name,
5402                                     size_t name_len)
5403 {
5404         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5405         const size_t total_len = prefix_len + name_len + 1;
5406         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5407
5408         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5409                 return 0;
5410
5411         if (list && total_len <= list_size) {
5412                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5413                 memcpy(list + prefix_len, name, name_len);
5414                 list[prefix_len + name_len] = '\0';
5415         }
5416         return total_len;
5417 }
5418
5419 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5420                                 void *buffer, size_t size)
5421 {
5422         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5423
5424         if (strcmp(name, "") == 0)
5425                 return -EINVAL;
5426         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5427                 return -EOPNOTSUPP;
5428         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5429                                buffer, size);
5430 }
5431
5432 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5433                                 const void *value, size_t size, int flags)
5434 {
5435         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5436
5437         if (strcmp(name, "") == 0)
5438                 return -EINVAL;
5439         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5440                 return -EOPNOTSUPP;
5441
5442         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5443                                size, flags);
5444 }
5445
5446 struct xattr_handler ocfs2_xattr_user_handler = {
5447         .prefix = XATTR_USER_PREFIX,
5448         .list   = ocfs2_xattr_user_list,
5449         .get    = ocfs2_xattr_user_get,
5450         .set    = ocfs2_xattr_user_set,
5451 };