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

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