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

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