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                                     struct ocfs2_xattr_value_buf *vb,
1286                                     size_t offs)
1287 {
1288         int ret;
1289
1290         ret = vb->vb_access(handle, inode, vb->vb_bh,
1291                             OCFS2_JOURNAL_ACCESS_WRITE);
1292         if (ret) {
1293                 mlog_errno(ret);
1294                 goto out;
1295         }
1296
1297         xs->here->xe_name_offset = cpu_to_le16(offs);
1298         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1299         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1300                 ocfs2_xattr_set_local(xs->here, 1);
1301         else
1302                 ocfs2_xattr_set_local(xs->here, 0);
1303         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1304
1305         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1306         if (ret < 0)
1307                 mlog_errno(ret);
1308 out:
1309         return ret;
1310 }
1311
1312 /*
1313  * ocfs2_xattr_set_value_outside()
1314  *
1315  * Set large size value in B tree.
1316  */
1317 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1318                                          struct ocfs2_xattr_info *xi,
1319                                          struct ocfs2_xattr_search *xs,
1320                                          struct ocfs2_xattr_set_ctxt *ctxt,
1321                                          size_t offs)
1322 {
1323         size_t name_len = strlen(xi->name);
1324         void *val = xs->base + offs;
1325         struct ocfs2_xattr_value_root *xv = NULL;
1326         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1327         int ret = 0;
1328         struct ocfs2_xattr_value_buf vb = {
1329                 .vb_bh = xs->xattr_bh,
1330                 .vb_access = ocfs2_journal_access
1331         };
1332
1333         memset(val, 0, size);
1334         memcpy(val, xi->name, name_len);
1335         xv = (struct ocfs2_xattr_value_root *)
1336                 (val + OCFS2_XATTR_SIZE(name_len));
1337         xv->xr_clusters = 0;
1338         xv->xr_last_eb_blk = 0;
1339         xv->xr_list.l_tree_depth = 0;
1340         xv->xr_list.l_count = cpu_to_le16(1);
1341         xv->xr_list.l_next_free_rec = 0;
1342         vb.vb_xv = xv;
1343
1344         ret = ocfs2_xattr_value_truncate(inode, &vb, xi->value_len, ctxt);
1345         if (ret < 0) {
1346                 mlog_errno(ret);
1347                 return ret;
1348         }
1349         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, &vb, offs);
1350         if (ret < 0) {
1351                 mlog_errno(ret);
1352                 return ret;
1353         }
1354         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb.vb_xv,
1355                                               xi->value, xi->value_len);
1356         if (ret < 0)
1357                 mlog_errno(ret);
1358
1359         return ret;
1360 }
1361
1362 /*
1363  * ocfs2_xattr_set_entry_local()
1364  *
1365  * Set, replace or remove extended attribute in local.
1366  */
1367 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1368                                         struct ocfs2_xattr_info *xi,
1369                                         struct ocfs2_xattr_search *xs,
1370                                         struct ocfs2_xattr_entry *last,
1371                                         size_t min_offs)
1372 {
1373         size_t name_len = strlen(xi->name);
1374         int i;
1375
1376         if (xi->value && xs->not_found) {
1377                 /* Insert the new xattr entry. */
1378                 le16_add_cpu(&xs->header->xh_count, 1);
1379                 ocfs2_xattr_set_type(last, xi->name_index);
1380                 ocfs2_xattr_set_local(last, 1);
1381                 last->xe_name_len = name_len;
1382         } else {
1383                 void *first_val;
1384                 void *val;
1385                 size_t offs, size;
1386
1387                 first_val = xs->base + min_offs;
1388                 offs = le16_to_cpu(xs->here->xe_name_offset);
1389                 val = xs->base + offs;
1390
1391                 if (le64_to_cpu(xs->here->xe_value_size) >
1392                     OCFS2_XATTR_INLINE_SIZE)
1393                         size = OCFS2_XATTR_SIZE(name_len) +
1394                                 OCFS2_XATTR_ROOT_SIZE;
1395                 else
1396                         size = OCFS2_XATTR_SIZE(name_len) +
1397                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1398
1399                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1400                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1401                         /* The old and the new value have the
1402                            same size. Just replace the value. */
1403                         ocfs2_xattr_set_local(xs->here, 1);
1404                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1405                         /* Clear value bytes. */
1406                         memset(val + OCFS2_XATTR_SIZE(name_len),
1407                                0,
1408                                OCFS2_XATTR_SIZE(xi->value_len));
1409                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1410                                xi->value,
1411                                xi->value_len);
1412                         return;
1413                 }
1414                 /* Remove the old name+value. */
1415                 memmove(first_val + size, first_val, val - first_val);
1416                 memset(first_val, 0, size);
1417                 xs->here->xe_name_hash = 0;
1418                 xs->here->xe_name_offset = 0;
1419                 ocfs2_xattr_set_local(xs->here, 1);
1420                 xs->here->xe_value_size = 0;
1421
1422                 min_offs += size;
1423
1424                 /* Adjust all value offsets. */
1425                 last = xs->header->xh_entries;
1426                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1427                         size_t o = le16_to_cpu(last->xe_name_offset);
1428
1429                         if (o < offs)
1430                                 last->xe_name_offset = cpu_to_le16(o + size);
1431                         last += 1;
1432                 }
1433
1434                 if (!xi->value) {
1435                         /* Remove the old entry. */
1436                         last -= 1;
1437                         memmove(xs->here, xs->here + 1,
1438                                 (void *)last - (void *)xs->here);
1439                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1440                         le16_add_cpu(&xs->header->xh_count, -1);
1441                 }
1442         }
1443         if (xi->value) {
1444                 /* Insert the new name+value. */
1445                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1446                                 OCFS2_XATTR_SIZE(xi->value_len);
1447                 void *val = xs->base + min_offs - size;
1448
1449                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1450                 memset(val, 0, size);
1451                 memcpy(val, xi->name, name_len);
1452                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1453                        xi->value,
1454                        xi->value_len);
1455                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1456                 ocfs2_xattr_set_local(xs->here, 1);
1457                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1458         }
1459
1460         return;
1461 }
1462
1463 /*
1464  * ocfs2_xattr_set_entry()
1465  *
1466  * Set extended attribute entry into inode or block.
1467  *
1468  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1469  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1470  * then set value in B tree with set_value_outside().
1471  */
1472 static int ocfs2_xattr_set_entry(struct inode *inode,
1473                                  struct ocfs2_xattr_info *xi,
1474                                  struct ocfs2_xattr_search *xs,
1475                                  struct ocfs2_xattr_set_ctxt *ctxt,
1476                                  int flag)
1477 {
1478         struct ocfs2_xattr_entry *last;
1479         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1480         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1481         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1482         size_t size_l = 0;
1483         handle_t *handle = ctxt->handle;
1484         int free, i, ret;
1485         struct ocfs2_xattr_info xi_l = {
1486                 .name_index = xi->name_index,
1487                 .name = xi->name,
1488                 .value = xi->value,
1489                 .value_len = xi->value_len,
1490         };
1491
1492         /* Compute min_offs, last and free space. */
1493         last = xs->header->xh_entries;
1494
1495         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1496                 size_t offs = le16_to_cpu(last->xe_name_offset);
1497                 if (offs < min_offs)
1498                         min_offs = offs;
1499                 last += 1;
1500         }
1501
1502         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1503         if (free < 0)
1504                 return -EIO;
1505
1506         if (!xs->not_found) {
1507                 size_t size = 0;
1508                 if (ocfs2_xattr_is_local(xs->here))
1509                         size = OCFS2_XATTR_SIZE(name_len) +
1510                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1511                 else
1512                         size = OCFS2_XATTR_SIZE(name_len) +
1513                                 OCFS2_XATTR_ROOT_SIZE;
1514                 free += (size + sizeof(struct ocfs2_xattr_entry));
1515         }
1516         /* Check free space in inode or block */
1517         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1518                 if (free < sizeof(struct ocfs2_xattr_entry) +
1519                            OCFS2_XATTR_SIZE(name_len) +
1520                            OCFS2_XATTR_ROOT_SIZE) {
1521                         ret = -ENOSPC;
1522                         goto out;
1523                 }
1524                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1525                 xi_l.value = (void *)&def_xv;
1526                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1527         } else if (xi->value) {
1528                 if (free < sizeof(struct ocfs2_xattr_entry) +
1529                            OCFS2_XATTR_SIZE(name_len) +
1530                            OCFS2_XATTR_SIZE(xi->value_len)) {
1531                         ret = -ENOSPC;
1532                         goto out;
1533                 }
1534         }
1535
1536         if (!xs->not_found) {
1537                 /* For existing extended attribute */
1538                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1539                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1540                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1541                 void *val = xs->base + offs;
1542
1543                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1544                         /* Replace existing local xattr with tree root */
1545                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1546                                                             ctxt, offs);
1547                         if (ret < 0)
1548                                 mlog_errno(ret);
1549                         goto out;
1550                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1551                         /* For existing xattr which has value outside */
1552                         struct ocfs2_xattr_value_buf vb = {
1553                                 .vb_bh = xs->xattr_bh,
1554                                 .vb_xv = (struct ocfs2_xattr_value_root *)
1555                                         (val + OCFS2_XATTR_SIZE(name_len)),
1556                                 .vb_access = ocfs2_journal_access,
1557                         };
1558
1559                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1560                                 /*
1561                                  * If new value need set outside also,
1562                                  * first truncate old value to new value,
1563                                  * then set new value with set_value_outside().
1564                                  */
1565                                 ret = ocfs2_xattr_value_truncate(inode,
1566                                                                  &vb,
1567                                                                  xi->value_len,
1568                                                                  ctxt);
1569                                 if (ret < 0) {
1570                                         mlog_errno(ret);
1571                                         goto out;
1572                                 }
1573
1574                                 ret = ocfs2_xattr_update_entry(inode,
1575                                                                handle,
1576                                                                xi,
1577                                                                xs,
1578                                                                &vb,
1579                                                                offs);
1580                                 if (ret < 0) {
1581                                         mlog_errno(ret);
1582                                         goto out;
1583                                 }
1584
1585                                 ret = __ocfs2_xattr_set_value_outside(inode,
1586                                                                 handle,
1587                                                                 vb.vb_xv,
1588                                                                 xi->value,
1589                                                                 xi->value_len);
1590                                 if (ret < 0)
1591                                         mlog_errno(ret);
1592                                 goto out;
1593                         } else {
1594                                 /*
1595                                  * If new value need set in local,
1596                                  * just trucate old value to zero.
1597                                  */
1598                                  ret = ocfs2_xattr_value_truncate(inode,
1599                                                                   &vb,
1600                                                                   0,
1601                                                                   ctxt);
1602                                 if (ret < 0)
1603                                         mlog_errno(ret);
1604                         }
1605                 }
1606         }
1607
1608         ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1609                                    OCFS2_JOURNAL_ACCESS_WRITE);
1610         if (ret) {
1611                 mlog_errno(ret);
1612                 goto out;
1613         }
1614
1615         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1616                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1617                                            OCFS2_JOURNAL_ACCESS_WRITE);
1618                 if (ret) {
1619                         mlog_errno(ret);
1620                         goto out;
1621                 }
1622         }
1623
1624         /*
1625          * Set value in local, include set tree root in local.
1626          * This is the first step for value size >INLINE_SIZE.
1627          */
1628         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1629
1630         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1631                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1632                 if (ret < 0) {
1633                         mlog_errno(ret);
1634                         goto out;
1635                 }
1636         }
1637
1638         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1639             (flag & OCFS2_INLINE_XATTR_FL)) {
1640                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1641                 unsigned int xattrsize = osb->s_xattr_inline_size;
1642
1643                 /*
1644                  * Adjust extent record count or inline data size
1645                  * to reserve space for extended attribute.
1646                  */
1647                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1648                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1649                         le16_add_cpu(&idata->id_count, -xattrsize);
1650                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1651                         struct ocfs2_extent_list *el = &di->id2.i_list;
1652                         le16_add_cpu(&el->l_count, -(xattrsize /
1653                                         sizeof(struct ocfs2_extent_rec)));
1654                 }
1655                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1656         }
1657         /* Update xattr flag */
1658         spin_lock(&oi->ip_lock);
1659         oi->ip_dyn_features |= flag;
1660         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1661         spin_unlock(&oi->ip_lock);
1662         /* Update inode ctime */
1663         inode->i_ctime = CURRENT_TIME;
1664         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1665         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1666
1667         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1668         if (ret < 0)
1669                 mlog_errno(ret);
1670
1671         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1672                 /*
1673                  * Set value outside in B tree.
1674                  * This is the second step for value size > INLINE_SIZE.
1675                  */
1676                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1677                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt, offs);
1678                 if (ret < 0) {
1679                         int ret2;
1680
1681                         mlog_errno(ret);
1682                         /*
1683                          * If set value outside failed, we have to clean
1684                          * the junk tree root we have already set in local.
1685                          */
1686                         ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1687                                                    xi, xs, offs);
1688                         if (ret2 < 0)
1689                                 mlog_errno(ret2);
1690                 }
1691         }
1692 out:
1693         return ret;
1694 }
1695
1696 static int ocfs2_remove_value_outside(struct inode*inode,
1697                                       struct buffer_head *bh,
1698                                       struct ocfs2_xattr_header *header)
1699 {
1700         int ret = 0, i;
1701         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1702         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1703
1704         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1705
1706         ctxt.handle = ocfs2_start_trans(osb,
1707                                         ocfs2_remove_extent_credits(osb->sb));
1708         if (IS_ERR(ctxt.handle)) {
1709                 ret = PTR_ERR(ctxt.handle);
1710                 mlog_errno(ret);
1711                 goto out;
1712         }
1713
1714         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1715                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1716
1717                 if (!ocfs2_xattr_is_local(entry)) {
1718                         struct ocfs2_xattr_value_buf vb = {
1719                                 .vb_bh = bh,
1720                                 .vb_access = ocfs2_journal_access,
1721                         };
1722                         void *val;
1723
1724                         val = (void *)header +
1725                                 le16_to_cpu(entry->xe_name_offset);
1726                         vb.vb_xv = (struct ocfs2_xattr_value_root *)
1727                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1728                         ret = ocfs2_xattr_value_truncate(inode, &vb, 0, &ctxt);
1729                         if (ret < 0) {
1730                                 mlog_errno(ret);
1731                                 break;
1732                         }
1733                 }
1734         }
1735
1736         ocfs2_commit_trans(osb, ctxt.handle);
1737         ocfs2_schedule_truncate_log_flush(osb, 1);
1738         ocfs2_run_deallocs(osb, &ctxt.dealloc);
1739 out:
1740         return ret;
1741 }
1742
1743 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1744                                     struct buffer_head *di_bh)
1745 {
1746
1747         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1748         struct ocfs2_xattr_header *header;
1749         int ret;
1750
1751         header = (struct ocfs2_xattr_header *)
1752                  ((void *)di + inode->i_sb->s_blocksize -
1753                  le16_to_cpu(di->i_xattr_inline_size));
1754
1755         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1756
1757         return ret;
1758 }
1759
1760 static int ocfs2_xattr_block_remove(struct inode *inode,
1761                                     struct buffer_head *blk_bh)
1762 {
1763         struct ocfs2_xattr_block *xb;
1764         int ret = 0;
1765
1766         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1767         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1768                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1769                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1770         } else
1771                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1772
1773         return ret;
1774 }
1775
1776 static int ocfs2_xattr_free_block(struct inode *inode,
1777                                   u64 block)
1778 {
1779         struct inode *xb_alloc_inode;
1780         struct buffer_head *xb_alloc_bh = NULL;
1781         struct buffer_head *blk_bh = NULL;
1782         struct ocfs2_xattr_block *xb;
1783         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1784         handle_t *handle;
1785         int ret = 0;
1786         u64 blk, bg_blkno;
1787         u16 bit;
1788
1789         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1790         if (ret < 0) {
1791                 mlog_errno(ret);
1792                 goto out;
1793         }
1794
1795         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1796         if (ret < 0) {
1797                 mlog_errno(ret);
1798                 goto out;
1799         }
1800
1801         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1802         blk = le64_to_cpu(xb->xb_blkno);
1803         bit = le16_to_cpu(xb->xb_suballoc_bit);
1804         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1805
1806         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1807                                 EXTENT_ALLOC_SYSTEM_INODE,
1808                                 le16_to_cpu(xb->xb_suballoc_slot));
1809         if (!xb_alloc_inode) {
1810                 ret = -ENOMEM;
1811                 mlog_errno(ret);
1812                 goto out;
1813         }
1814         mutex_lock(&xb_alloc_inode->i_mutex);
1815
1816         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1817         if (ret < 0) {
1818                 mlog_errno(ret);
1819                 goto out_mutex;
1820         }
1821
1822         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1823         if (IS_ERR(handle)) {
1824                 ret = PTR_ERR(handle);
1825                 mlog_errno(ret);
1826                 goto out_unlock;
1827         }
1828
1829         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1830                                        bit, bg_blkno, 1);
1831         if (ret < 0)
1832                 mlog_errno(ret);
1833
1834         ocfs2_commit_trans(osb, handle);
1835 out_unlock:
1836         ocfs2_inode_unlock(xb_alloc_inode, 1);
1837         brelse(xb_alloc_bh);
1838 out_mutex:
1839         mutex_unlock(&xb_alloc_inode->i_mutex);
1840         iput(xb_alloc_inode);
1841 out:
1842         brelse(blk_bh);
1843         return ret;
1844 }
1845
1846 /*
1847  * ocfs2_xattr_remove()
1848  *
1849  * Free extended attribute resources associated with this inode.
1850  */
1851 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1852 {
1853         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1854         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1855         handle_t *handle;
1856         int ret;
1857
1858         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1859                 return 0;
1860
1861         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1862                 return 0;
1863
1864         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1865                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1866                 if (ret < 0) {
1867                         mlog_errno(ret);
1868                         goto out;
1869                 }
1870         }
1871
1872         if (di->i_xattr_loc) {
1873                 ret = ocfs2_xattr_free_block(inode,
1874                                              le64_to_cpu(di->i_xattr_loc));
1875                 if (ret < 0) {
1876                         mlog_errno(ret);
1877                         goto out;
1878                 }
1879         }
1880
1881         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1882                                    OCFS2_INODE_UPDATE_CREDITS);
1883         if (IS_ERR(handle)) {
1884                 ret = PTR_ERR(handle);
1885                 mlog_errno(ret);
1886                 goto out;
1887         }
1888         ret = ocfs2_journal_access(handle, inode, di_bh,
1889                                    OCFS2_JOURNAL_ACCESS_WRITE);
1890         if (ret) {
1891                 mlog_errno(ret);
1892                 goto out_commit;
1893         }
1894
1895         di->i_xattr_loc = 0;
1896
1897         spin_lock(&oi->ip_lock);
1898         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1899         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1900         spin_unlock(&oi->ip_lock);
1901
1902         ret = ocfs2_journal_dirty(handle, di_bh);
1903         if (ret < 0)
1904                 mlog_errno(ret);
1905 out_commit:
1906         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1907 out:
1908         return ret;
1909 }
1910
1911 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1912                                         struct ocfs2_dinode *di)
1913 {
1914         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1915         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1916         int free;
1917
1918         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1919                 return 0;
1920
1921         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1922                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1923                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1924         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1925                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1926                         le64_to_cpu(di->i_size);
1927         } else {
1928                 struct ocfs2_extent_list *el = &di->id2.i_list;
1929                 free = (le16_to_cpu(el->l_count) -
1930                         le16_to_cpu(el->l_next_free_rec)) *
1931                         sizeof(struct ocfs2_extent_rec);
1932         }
1933         if (free >= xattrsize)
1934                 return 1;
1935
1936         return 0;
1937 }
1938
1939 /*
1940  * ocfs2_xattr_ibody_find()
1941  *
1942  * Find extended attribute in inode block and
1943  * fill search info into struct ocfs2_xattr_search.
1944  */
1945 static int ocfs2_xattr_ibody_find(struct inode *inode,
1946                                   int name_index,
1947                                   const char *name,
1948                                   struct ocfs2_xattr_search *xs)
1949 {
1950         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1951         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1952         int ret;
1953         int has_space = 0;
1954
1955         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1956                 return 0;
1957
1958         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1959                 down_read(&oi->ip_alloc_sem);
1960                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1961                 up_read(&oi->ip_alloc_sem);
1962                 if (!has_space)
1963                         return 0;
1964         }
1965
1966         xs->xattr_bh = xs->inode_bh;
1967         xs->end = (void *)di + inode->i_sb->s_blocksize;
1968         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1969                 xs->header = (struct ocfs2_xattr_header *)
1970                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1971         else
1972                 xs->header = (struct ocfs2_xattr_header *)
1973                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1974         xs->base = (void *)xs->header;
1975         xs->here = xs->header->xh_entries;
1976
1977         /* Find the named attribute. */
1978         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1979                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1980                 if (ret && ret != -ENODATA)
1981                         return ret;
1982                 xs->not_found = ret;
1983         }
1984
1985         return 0;
1986 }
1987
1988 /*
1989  * ocfs2_xattr_ibody_set()
1990  *
1991  * Set, replace or remove an extended attribute into inode block.
1992  *
1993  */
1994 static int ocfs2_xattr_ibody_set(struct inode *inode,
1995                                  struct ocfs2_xattr_info *xi,
1996                                  struct ocfs2_xattr_search *xs,
1997                                  struct ocfs2_xattr_set_ctxt *ctxt)
1998 {
1999         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2000         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2001         int ret;
2002
2003         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2004                 return -ENOSPC;
2005
2006         down_write(&oi->ip_alloc_sem);
2007         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2008                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2009                         ret = -ENOSPC;
2010                         goto out;
2011                 }
2012         }
2013
2014         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2015                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2016 out:
2017         up_write(&oi->ip_alloc_sem);
2018
2019         return ret;
2020 }
2021
2022 /*
2023  * ocfs2_xattr_block_find()
2024  *
2025  * Find extended attribute in external block and
2026  * fill search info into struct ocfs2_xattr_search.
2027  */
2028 static int ocfs2_xattr_block_find(struct inode *inode,
2029                                   int name_index,
2030                                   const char *name,
2031                                   struct ocfs2_xattr_search *xs)
2032 {
2033         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2034         struct buffer_head *blk_bh = NULL;
2035         struct ocfs2_xattr_block *xb;
2036         int ret = 0;
2037
2038         if (!di->i_xattr_loc)
2039                 return ret;
2040
2041         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2042                                      &blk_bh);
2043         if (ret < 0) {
2044                 mlog_errno(ret);
2045                 return ret;
2046         }
2047
2048         xs->xattr_bh = blk_bh;
2049         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2050
2051         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2052                 xs->header = &xb->xb_attrs.xb_header;
2053                 xs->base = (void *)xs->header;
2054                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2055                 xs->here = xs->header->xh_entries;
2056
2057                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2058         } else
2059                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2060                                                    name_index,
2061                                                    name, xs);
2062
2063         if (ret && ret != -ENODATA) {
2064                 xs->xattr_bh = NULL;
2065                 goto cleanup;
2066         }
2067         xs->not_found = ret;
2068         return 0;
2069 cleanup:
2070         brelse(blk_bh);
2071
2072         return ret;
2073 }
2074
2075 /*
2076  * ocfs2_xattr_block_set()
2077  *
2078  * Set, replace or remove an extended attribute into external block.
2079  *
2080  */
2081 static int ocfs2_xattr_block_set(struct inode *inode,
2082                                  struct ocfs2_xattr_info *xi,
2083                                  struct ocfs2_xattr_search *xs,
2084                                  struct ocfs2_xattr_set_ctxt *ctxt)
2085 {
2086         struct buffer_head *new_bh = NULL;
2087         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2088         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
2089         handle_t *handle = ctxt->handle;
2090         struct ocfs2_xattr_block *xblk = NULL;
2091         u16 suballoc_bit_start;
2092         u32 num_got;
2093         u64 first_blkno;
2094         int ret;
2095
2096         if (!xs->xattr_bh) {
2097                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
2098                                            OCFS2_JOURNAL_ACCESS_CREATE);
2099                 if (ret < 0) {
2100                         mlog_errno(ret);
2101                         goto end;
2102                 }
2103
2104                 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2105                                            &suballoc_bit_start, &num_got,
2106                                            &first_blkno);
2107                 if (ret < 0) {
2108                         mlog_errno(ret);
2109                         goto end;
2110                 }
2111
2112                 new_bh = sb_getblk(inode->i_sb, first_blkno);
2113                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2114
2115                 ret = ocfs2_journal_access(handle, inode, new_bh,
2116                                            OCFS2_JOURNAL_ACCESS_CREATE);
2117                 if (ret < 0) {
2118                         mlog_errno(ret);
2119                         goto end;
2120                 }
2121
2122                 /* Initialize ocfs2_xattr_block */
2123                 xs->xattr_bh = new_bh;
2124                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2125                 memset(xblk, 0, inode->i_sb->s_blocksize);
2126                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2127                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2128                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2129                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2130                 xblk->xb_blkno = cpu_to_le64(first_blkno);
2131
2132                 xs->header = &xblk->xb_attrs.xb_header;
2133                 xs->base = (void *)xs->header;
2134                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2135                 xs->here = xs->header->xh_entries;
2136
2137                 ret = ocfs2_journal_dirty(handle, new_bh);
2138                 if (ret < 0) {
2139                         mlog_errno(ret);
2140                         goto end;
2141                 }
2142                 di->i_xattr_loc = cpu_to_le64(first_blkno);
2143                 ocfs2_journal_dirty(handle, xs->inode_bh);
2144         } else
2145                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2146
2147         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2148                 /* Set extended attribute into external block */
2149                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2150                                             OCFS2_HAS_XATTR_FL);
2151                 if (!ret || ret != -ENOSPC)
2152                         goto end;
2153
2154                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2155                 if (ret)
2156                         goto end;
2157         }
2158
2159         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2160
2161 end:
2162
2163         return ret;
2164 }
2165
2166 /* Check whether the new xattr can be inserted into the inode. */
2167 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2168                                        struct ocfs2_xattr_info *xi,
2169                                        struct ocfs2_xattr_search *xs)
2170 {
2171         u64 value_size;
2172         struct ocfs2_xattr_entry *last;
2173         int free, i;
2174         size_t min_offs = xs->end - xs->base;
2175
2176         if (!xs->header)
2177                 return 0;
2178
2179         last = xs->header->xh_entries;
2180
2181         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2182                 size_t offs = le16_to_cpu(last->xe_name_offset);
2183                 if (offs < min_offs)
2184                         min_offs = offs;
2185                 last += 1;
2186         }
2187
2188         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
2189         if (free < 0)
2190                 return 0;
2191
2192         BUG_ON(!xs->not_found);
2193
2194         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2195                 value_size = OCFS2_XATTR_ROOT_SIZE;
2196         else
2197                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2198
2199         if (free >= sizeof(struct ocfs2_xattr_entry) +
2200                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2201                 return 1;
2202
2203         return 0;
2204 }
2205
2206 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2207                                      struct ocfs2_dinode *di,
2208                                      struct ocfs2_xattr_info *xi,
2209                                      struct ocfs2_xattr_search *xis,
2210                                      struct ocfs2_xattr_search *xbs,
2211                                      int *clusters_need,
2212                                      int *meta_need,
2213                                      int *credits_need)
2214 {
2215         int ret = 0, old_in_xb = 0;
2216         int clusters_add = 0, meta_add = 0, credits = 0;
2217         struct buffer_head *bh = NULL;
2218         struct ocfs2_xattr_block *xb = NULL;
2219         struct ocfs2_xattr_entry *xe = NULL;
2220         struct ocfs2_xattr_value_root *xv = NULL;
2221         char *base = NULL;
2222         int name_offset, name_len = 0;
2223         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2224                                                     xi->value_len);
2225         u64 value_size;
2226
2227         if (xis->not_found && xbs->not_found) {
2228                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2229
2230                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2231                         clusters_add += new_clusters;
2232                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2233                                                         &def_xv.xv.xr_list,
2234                                                         new_clusters);
2235                 }
2236
2237                 goto meta_guess;
2238         }
2239
2240         if (!xis->not_found) {
2241                 xe = xis->here;
2242                 name_offset = le16_to_cpu(xe->xe_name_offset);
2243                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2244                 base = xis->base;
2245                 credits += OCFS2_INODE_UPDATE_CREDITS;
2246         } else {
2247                 int i, block_off = 0;
2248                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2249                 xe = xbs->here;
2250                 name_offset = le16_to_cpu(xe->xe_name_offset);
2251                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2252                 i = xbs->here - xbs->header->xh_entries;
2253                 old_in_xb = 1;
2254
2255                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2256                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2257                                                         bucket_xh(xbs->bucket),
2258                                                         i, &block_off,
2259                                                         &name_offset);
2260                         base = bucket_block(xbs->bucket, block_off);
2261                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2262                 } else {
2263                         base = xbs->base;
2264                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2265                 }
2266         }
2267
2268         /*
2269          * delete a xattr doesn't need metadata and cluster allocation.
2270          * so just calculate the credits and return.
2271          *
2272          * The credits for removing the value tree will be extended
2273          * by ocfs2_remove_extent itself.
2274          */
2275         if (!xi->value) {
2276                 if (!ocfs2_xattr_is_local(xe))
2277                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2278
2279                 goto out;
2280         }
2281
2282         /* do cluster allocation guess first. */
2283         value_size = le64_to_cpu(xe->xe_value_size);
2284
2285         if (old_in_xb) {
2286                 /*
2287                  * In xattr set, we always try to set the xe in inode first,
2288                  * so if it can be inserted into inode successfully, the old
2289                  * one will be removed from the xattr block, and this xattr
2290                  * will be inserted into inode as a new xattr in inode.
2291                  */
2292                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2293                         clusters_add += new_clusters;
2294                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
2295                                     OCFS2_INODE_UPDATE_CREDITS;
2296                         if (!ocfs2_xattr_is_local(xe))
2297                                 credits += ocfs2_calc_extend_credits(
2298                                                         inode->i_sb,
2299                                                         &def_xv.xv.xr_list,
2300                                                         new_clusters);
2301                         goto out;
2302                 }
2303         }
2304
2305         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2306                 /* the new values will be stored outside. */
2307                 u32 old_clusters = 0;
2308
2309                 if (!ocfs2_xattr_is_local(xe)) {
2310                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2311                                                                  value_size);
2312                         xv = (struct ocfs2_xattr_value_root *)
2313                              (base + name_offset + name_len);
2314                         value_size = OCFS2_XATTR_ROOT_SIZE;
2315                 } else
2316                         xv = &def_xv.xv;
2317
2318                 if (old_clusters >= new_clusters) {
2319                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2320                         goto out;
2321                 } else {
2322                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2323                         clusters_add += new_clusters - old_clusters;
2324                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2325                                                              &xv->xr_list,
2326                                                              new_clusters -
2327                                                              old_clusters);
2328                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2329                                 goto out;
2330                 }
2331         } else {
2332                 /*
2333                  * Now the new value will be stored inside. So if the new
2334                  * value is smaller than the size of value root or the old
2335                  * value, we don't need any allocation, otherwise we have
2336                  * to guess metadata allocation.
2337                  */
2338                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2339                     (!ocfs2_xattr_is_local(xe) &&
2340                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2341                         goto out;
2342         }
2343
2344 meta_guess:
2345         /* calculate metadata allocation. */
2346         if (di->i_xattr_loc) {
2347                 if (!xbs->xattr_bh) {
2348                         ret = ocfs2_read_xattr_block(inode,
2349                                                      le64_to_cpu(di->i_xattr_loc),
2350                                                      &bh);
2351                         if (ret) {
2352                                 mlog_errno(ret);
2353                                 goto out;
2354                         }
2355
2356                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2357                 } else
2358                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2359
2360                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2361                         struct ocfs2_extent_list *el =
2362                                  &xb->xb_attrs.xb_root.xt_list;
2363                         meta_add += ocfs2_extend_meta_needed(el);
2364                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2365                                                              el, 1);
2366                 }
2367
2368                 /*
2369                  * This cluster will be used either for new bucket or for
2370                  * new xattr block.
2371                  * If the cluster size is the same as the bucket size, one
2372                  * more is needed since we may need to extend the bucket
2373                  * also.
2374                  */
2375                 clusters_add += 1;
2376                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2377                 if (OCFS2_XATTR_BUCKET_SIZE ==
2378                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2379                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2380                         clusters_add += 1;
2381                 }
2382         } else {
2383                 meta_add += 1;
2384                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2385         }
2386 out:
2387         if (clusters_need)
2388                 *clusters_need = clusters_add;
2389         if (meta_need)
2390                 *meta_need = meta_add;
2391         if (credits_need)
2392                 *credits_need = credits;
2393         brelse(bh);
2394         return ret;
2395 }
2396
2397 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2398                                      struct ocfs2_dinode *di,
2399                                      struct ocfs2_xattr_info *xi,
2400                                      struct ocfs2_xattr_search *xis,
2401                                      struct ocfs2_xattr_search *xbs,
2402                                      struct ocfs2_xattr_set_ctxt *ctxt,
2403                                      int *credits)
2404 {
2405         int clusters_add, meta_add, ret;
2406         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2407
2408         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2409
2410         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2411
2412         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2413                                         &clusters_add, &meta_add, credits);
2414         if (ret) {
2415                 mlog_errno(ret);
2416                 return ret;
2417         }
2418
2419         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2420              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2421
2422         if (meta_add) {
2423                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2424                                                         &ctxt->meta_ac);
2425                 if (ret) {
2426                         mlog_errno(ret);
2427                         goto out;
2428                 }
2429         }
2430
2431         if (clusters_add) {
2432                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2433                 if (ret)
2434                         mlog_errno(ret);
2435         }
2436 out:
2437         if (ret) {
2438                 if (ctxt->meta_ac) {
2439                         ocfs2_free_alloc_context(ctxt->meta_ac);
2440                         ctxt->meta_ac = NULL;
2441                 }
2442
2443                 /*
2444                  * We cannot have an error and a non null ctxt->data_ac.
2445                  */
2446         }
2447
2448         return ret;
2449 }
2450
2451 static int __ocfs2_xattr_set_handle(struct inode *inode,
2452                                     struct ocfs2_dinode *di,
2453                                     struct ocfs2_xattr_info *xi,
2454                                     struct ocfs2_xattr_search *xis,
2455                                     struct ocfs2_xattr_search *xbs,
2456                                     struct ocfs2_xattr_set_ctxt *ctxt)
2457 {
2458         int ret = 0, credits, old_found;
2459
2460         if (!xi->value) {
2461                 /* Remove existing extended attribute */
2462                 if (!xis->not_found)
2463                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2464                 else if (!xbs->not_found)
2465                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2466         } else {
2467                 /* We always try to set extended attribute into inode first*/
2468                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2469                 if (!ret && !xbs->not_found) {
2470                         /*
2471                          * If succeed and that extended attribute existing in
2472                          * external block, then we will remove it.
2473                          */
2474                         xi->value = NULL;
2475                         xi->value_len = 0;
2476
2477                         old_found = xis->not_found;
2478                         xis->not_found = -ENODATA;
2479                         ret = ocfs2_calc_xattr_set_need(inode,
2480                                                         di,
2481                                                         xi,
2482                                                         xis,
2483                                                         xbs,
2484                                                         NULL,
2485                                                         NULL,
2486                                                         &credits);
2487                         xis->not_found = old_found;
2488                         if (ret) {
2489                                 mlog_errno(ret);
2490                                 goto out;
2491                         }
2492
2493                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2494                                         ctxt->handle->h_buffer_credits);
2495                         if (ret) {
2496                                 mlog_errno(ret);
2497                                 goto out;
2498                         }
2499                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2500                 } else if (ret == -ENOSPC) {
2501                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2502                                 ret = ocfs2_xattr_block_find(inode,
2503                                                              xi->name_index,
2504                                                              xi->name, xbs);
2505                                 if (ret)
2506                                         goto out;
2507
2508                                 old_found = xis->not_found;
2509                                 xis->not_found = -ENODATA;
2510                                 ret = ocfs2_calc_xattr_set_need(inode,
2511                                                                 di,
2512                                                                 xi,
2513                                                                 xis,
2514                                                                 xbs,
2515                                                                 NULL,
2516                                                                 NULL,
2517                                                                 &credits);
2518                                 xis->not_found = old_found;
2519                                 if (ret) {
2520                                         mlog_errno(ret);
2521                                         goto out;
2522                                 }
2523
2524                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2525                                         ctxt->handle->h_buffer_credits);
2526                                 if (ret) {
2527                                         mlog_errno(ret);
2528                                         goto out;
2529                                 }
2530                         }
2531                         /*
2532                          * If no space in inode, we will set extended attribute
2533                          * into external block.
2534                          */
2535                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2536                         if (ret)
2537                                 goto out;
2538                         if (!xis->not_found) {
2539                                 /*
2540                                  * If succeed and that extended attribute
2541                                  * existing in inode, we will remove it.
2542                                  */
2543                                 xi->value = NULL;
2544                                 xi->value_len = 0;
2545                                 xbs->not_found = -ENODATA;
2546                                 ret = ocfs2_calc_xattr_set_need(inode,
2547                                                                 di,
2548                                                                 xi,
2549                                                                 xis,
2550                                                                 xbs,
2551                                                                 NULL,
2552                                                                 NULL,
2553                                                                 &credits);
2554                                 if (ret) {
2555                                         mlog_errno(ret);
2556                                         goto out;
2557                                 }
2558
2559                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2560                                                 ctxt->handle->h_buffer_credits);
2561                                 if (ret) {
2562                                         mlog_errno(ret);
2563                                         goto out;
2564                                 }
2565                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2566                                                             xis, ctxt);
2567                         }
2568                 }
2569         }
2570
2571 out:
2572         return ret;
2573 }
2574
2575 /*
2576  * This function only called duing creating inode
2577  * for init security/acl xattrs of the new inode.
2578  * The xattrs could be put into ibody or extent block,
2579  * xattr bucket would not be use in this case.
2580  * transanction credits also be reserved in here.
2581  */
2582 int ocfs2_xattr_set_handle(handle_t *handle,
2583                            struct inode *inode,
2584                            struct buffer_head *di_bh,
2585                            int name_index,
2586                            const char *name,
2587                            const void *value,
2588                            size_t value_len,
2589                            int flags,
2590                            struct ocfs2_alloc_context *meta_ac,
2591                            struct ocfs2_alloc_context *data_ac)
2592 {
2593         struct ocfs2_dinode *di;
2594         int ret;
2595
2596         struct ocfs2_xattr_info xi = {
2597                 .name_index = name_index,
2598                 .name = name,
2599                 .value = value,
2600                 .value_len = value_len,
2601         };
2602
2603         struct ocfs2_xattr_search xis = {
2604                 .not_found = -ENODATA,
2605         };
2606
2607         struct ocfs2_xattr_search xbs = {
2608                 .not_found = -ENODATA,
2609         };
2610
2611         struct ocfs2_xattr_set_ctxt ctxt = {
2612                 .handle = handle,
2613                 .meta_ac = meta_ac,
2614                 .data_ac = data_ac,
2615         };
2616
2617         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2618                 return -EOPNOTSUPP;
2619
2620         xis.inode_bh = xbs.inode_bh = di_bh;
2621         di = (struct ocfs2_dinode *)di_bh->b_data;
2622
2623         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2624
2625         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2626         if (ret)
2627                 goto cleanup;
2628         if (xis.not_found) {
2629                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2630                 if (ret)
2631                         goto cleanup;
2632         }
2633
2634         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2635
2636 cleanup:
2637         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2638         brelse(xbs.xattr_bh);
2639
2640         return ret;
2641 }
2642
2643 /*
2644  * ocfs2_xattr_set()
2645  *
2646  * Set, replace or remove an extended attribute for this inode.
2647  * value is NULL to remove an existing extended attribute, else either
2648  * create or replace an extended attribute.
2649  */
2650 int ocfs2_xattr_set(struct inode *inode,
2651                     int name_index,
2652                     const char *name,
2653                     const void *value,
2654                     size_t value_len,
2655                     int flags)
2656 {
2657         struct buffer_head *di_bh = NULL;
2658         struct ocfs2_dinode *di;
2659         int ret, credits;
2660         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2661         struct inode *tl_inode = osb->osb_tl_inode;
2662         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2663
2664         struct ocfs2_xattr_info xi = {
2665                 .name_index = name_index,
2666                 .name = name,
2667                 .value = value,
2668                 .value_len = value_len,
2669         };
2670
2671         struct ocfs2_xattr_search xis = {
2672                 .not_found = -ENODATA,
2673         };
2674
2675         struct ocfs2_xattr_search xbs = {
2676                 .not_found = -ENODATA,
2677         };
2678
2679         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2680                 return -EOPNOTSUPP;
2681
2682         /*
2683          * Only xbs will be used on indexed trees.  xis doesn't need a
2684          * bucket.
2685          */
2686         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2687         if (!xbs.bucket) {
2688                 mlog_errno(-ENOMEM);
2689                 return -ENOMEM;
2690         }
2691
2692         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2693         if (ret < 0) {
2694                 mlog_errno(ret);
2695                 goto cleanup_nolock;
2696         }
2697         xis.inode_bh = xbs.inode_bh = di_bh;
2698         di = (struct ocfs2_dinode *)di_bh->b_data;
2699
2700         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2701         /*
2702          * Scan inode and external block to find the same name
2703          * extended attribute and collect search infomation.
2704          */
2705         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2706         if (ret)
2707                 goto cleanup;
2708         if (xis.not_found) {
2709                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2710                 if (ret)
2711                         goto cleanup;
2712         }
2713
2714         if (xis.not_found && xbs.not_found) {
2715                 ret = -ENODATA;
2716                 if (flags & XATTR_REPLACE)
2717                         goto cleanup;
2718                 ret = 0;
2719                 if (!value)
2720                         goto cleanup;
2721         } else {
2722                 ret = -EEXIST;
2723                 if (flags & XATTR_CREATE)
2724                         goto cleanup;
2725         }
2726
2727
2728         mutex_lock(&tl_inode->i_mutex);
2729
2730         if (ocfs2_truncate_log_needs_flush(osb)) {
2731                 ret = __ocfs2_flush_truncate_log(osb);
2732                 if (ret < 0) {
2733                         mutex_unlock(&tl_inode->i_mutex);
2734                         mlog_errno(ret);
2735                         goto cleanup;
2736                 }
2737         }
2738         mutex_unlock(&tl_inode->i_mutex);
2739
2740         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2741                                         &xbs, &ctxt, &credits);
2742         if (ret) {
2743                 mlog_errno(ret);
2744                 goto cleanup;
2745         }
2746
2747         ctxt.handle = ocfs2_start_trans(osb, credits);
2748         if (IS_ERR(ctxt.handle)) {
2749                 ret = PTR_ERR(ctxt.handle);
2750                 mlog_errno(ret);
2751                 goto cleanup;
2752         }
2753
2754         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2755
2756         ocfs2_commit_trans(osb, ctxt.handle);
2757
2758         if (ctxt.data_ac)
2759                 ocfs2_free_alloc_context(ctxt.data_ac);
2760         if (ctxt.meta_ac)
2761                 ocfs2_free_alloc_context(ctxt.meta_ac);
2762         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2763                 ocfs2_schedule_truncate_log_flush(osb, 1);
2764         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2765 cleanup:
2766         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2767         ocfs2_inode_unlock(inode, 1);
2768 cleanup_nolock:
2769         brelse(di_bh);
2770         brelse(xbs.xattr_bh);
2771         ocfs2_xattr_bucket_free(xbs.bucket);
2772
2773         return ret;
2774 }
2775
2776 /*
2777  * Find the xattr extent rec which may contains name_hash.
2778  * e_cpos will be the first name hash of the xattr rec.
2779  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2780  */
2781 static int ocfs2_xattr_get_rec(struct inode *inode,
2782                                u32 name_hash,
2783                                u64 *p_blkno,
2784                                u32 *e_cpos,
2785                                u32 *num_clusters,
2786                                struct ocfs2_extent_list *el)
2787 {
2788         int ret = 0, i;
2789         struct buffer_head *eb_bh = NULL;
2790         struct ocfs2_extent_block *eb;
2791         struct ocfs2_extent_rec *rec = NULL;
2792         u64 e_blkno = 0;
2793
2794         if (el->l_tree_depth) {
2795                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2796                 if (ret) {
2797                         mlog_errno(ret);
2798                         goto out;
2799                 }
2800
2801                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2802                 el = &eb->h_list;
2803
2804                 if (el->l_tree_depth) {
2805                         ocfs2_error(inode->i_sb,
2806                                     "Inode %lu has non zero tree depth in "
2807                                     "xattr tree block %llu\n", inode->i_ino,
2808                                     (unsigned long long)eb_bh->b_blocknr);
2809                         ret = -EROFS;
2810                         goto out;
2811                 }
2812         }
2813
2814         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2815                 rec = &el->l_recs[i];
2816
2817                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2818                         e_blkno = le64_to_cpu(rec->e_blkno);
2819                         break;
2820                 }
2821         }
2822
2823         if (!e_blkno) {
2824                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2825                             "record (%u, %u, 0) in xattr", inode->i_ino,
2826                             le32_to_cpu(rec->e_cpos),
2827                             ocfs2_rec_clusters(el, rec));
2828                 ret = -EROFS;
2829                 goto out;
2830         }
2831
2832         *p_blkno = le64_to_cpu(rec->e_blkno);
2833         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2834         if (e_cpos)
2835                 *e_cpos = le32_to_cpu(rec->e_cpos);
2836 out:
2837         brelse(eb_bh);
2838         return ret;
2839 }
2840
2841 typedef int (xattr_bucket_func)(struct inode *inode,
2842                                 struct ocfs2_xattr_bucket *bucket,
2843                                 void *para);
2844
2845 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2846                                    struct ocfs2_xattr_bucket *bucket,
2847                                    int name_index,
2848                                    const char *name,
2849                                    u32 name_hash,
2850                                    u16 *xe_index,
2851                                    int *found)
2852 {
2853         int i, ret = 0, cmp = 1, block_off, new_offset;
2854         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2855         size_t name_len = strlen(name);
2856         struct ocfs2_xattr_entry *xe = NULL;
2857         char *xe_name;
2858
2859         /*
2860          * We don't use binary search in the bucket because there
2861          * may be multiple entries with the same name hash.
2862          */
2863         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2864                 xe = &xh->xh_entries[i];
2865
2866                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2867                         continue;
2868                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2869                         break;
2870
2871                 cmp = name_index - ocfs2_xattr_get_type(xe);
2872                 if (!cmp)
2873                         cmp = name_len - xe->xe_name_len;
2874                 if (cmp)
2875                         continue;
2876
2877                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2878                                                         xh,
2879                                                         i,
2880                                                         &block_off,
2881                                                         &new_offset);
2882                 if (ret) {
2883                         mlog_errno(ret);
2884                         break;
2885                 }
2886
2887
2888                 xe_name = bucket_block(bucket, block_off) + new_offset;
2889                 if (!memcmp(name, xe_name, name_len)) {
2890                         *xe_index = i;
2891                         *found = 1;
2892                         ret = 0;
2893                         break;
2894                 }
2895         }
2896
2897         return ret;
2898 }
2899
2900 /*
2901  * Find the specified xattr entry in a series of buckets.
2902  * This series start from p_blkno and last for num_clusters.
2903  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2904  * the num of the valid buckets.
2905  *
2906  * Return the buffer_head this xattr should reside in. And if the xattr's
2907  * hash is in the gap of 2 buckets, return the lower bucket.
2908  */
2909 static int ocfs2_xattr_bucket_find(struct inode *inode,
2910                                    int name_index,
2911                                    const char *name,
2912                                    u32 name_hash,
2913                                    u64 p_blkno,
2914                                    u32 first_hash,
2915                                    u32 num_clusters,
2916                                    struct ocfs2_xattr_search *xs)
2917 {
2918         int ret, found = 0;
2919         struct ocfs2_xattr_header *xh = NULL;
2920         struct ocfs2_xattr_entry *xe = NULL;
2921         u16 index = 0;
2922         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2923         int low_bucket = 0, bucket, high_bucket;
2924         struct ocfs2_xattr_bucket *search;
2925         u32 last_hash;
2926         u64 blkno, lower_blkno = 0;
2927
2928         search = ocfs2_xattr_bucket_new(inode);
2929         if (!search) {
2930                 ret = -ENOMEM;
2931                 mlog_errno(ret);
2932                 goto out;
2933         }
2934
2935         ret = ocfs2_read_xattr_bucket(search, p_blkno);
2936         if (ret) {
2937                 mlog_errno(ret);
2938                 goto out;
2939         }
2940
2941         xh = bucket_xh(search);
2942         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2943         while (low_bucket <= high_bucket) {
2944                 ocfs2_xattr_bucket_relse(search);
2945
2946                 bucket = (low_bucket + high_bucket) / 2;
2947                 blkno = p_blkno + bucket * blk_per_bucket;
2948                 ret = ocfs2_read_xattr_bucket(search, blkno);
2949                 if (ret) {
2950                         mlog_errno(ret);
2951                         goto out;
2952                 }
2953
2954                 xh = bucket_xh(search);
2955                 xe = &xh->xh_entries[0];
2956                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2957                         high_bucket = bucket - 1;
2958                         continue;
2959                 }
2960
2961                 /*
2962                  * Check whether the hash of the last entry in our
2963                  * bucket is larger than the search one. for an empty
2964                  * bucket, the last one is also the first one.
2965                  */
2966                 if (xh->xh_count)
2967                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2968
2969                 last_hash = le32_to_cpu(xe->xe_name_hash);
2970
2971                 /* record lower_blkno which may be the insert place. */
2972                 lower_blkno = blkno;
2973
2974                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2975                         low_bucket = bucket + 1;
2976                         continue;
2977                 }
2978
2979                 /* the searched xattr should reside in this bucket if exists. */
2980                 ret = ocfs2_find_xe_in_bucket(inode, search,
2981                                               name_index, name, name_hash,
2982                                               &index, &found);
2983                 if (ret) {
2984                         mlog_errno(ret);
2985                         goto out;
2986                 }
2987                 break;
2988         }
2989
2990         /*
2991          * Record the bucket we have found.
2992          * When the xattr's hash value is in the gap of 2 buckets, we will
2993          * always set it to the previous bucket.
2994          */
2995         if (!lower_blkno)
2996                 lower_blkno = p_blkno;
2997
2998         /* This should be in cache - we just read it during the search */
2999         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3000         if (ret) {
3001                 mlog_errno(ret);
3002                 goto out;
3003         }
3004
3005         xs->header = bucket_xh(xs->bucket);
3006         xs->base = bucket_block(xs->bucket, 0);
3007         xs->end = xs->base + inode->i_sb->s_blocksize;
3008
3009         if (found) {
3010                 xs->here = &xs->header->xh_entries[index];
3011                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3012                      (unsigned long long)bucket_blkno(xs->bucket), index);
3013         } else
3014                 ret = -ENODATA;
3015
3016 out:
3017         ocfs2_xattr_bucket_free(search);
3018         return ret;
3019 }
3020
3021 static int ocfs2_xattr_index_block_find(struct inode *inode,
3022                                         struct buffer_head *root_bh,
3023                                         int name_index,
3024                                         const char *name,
3025                                         struct ocfs2_xattr_search *xs)
3026 {
3027         int ret;
3028         struct ocfs2_xattr_block *xb =
3029                         (struct ocfs2_xattr_block *)root_bh->b_data;
3030         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3031         struct ocfs2_extent_list *el = &xb_root->xt_list;
3032         u64 p_blkno = 0;
3033         u32 first_hash, num_clusters = 0;
3034         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3035
3036         if (le16_to_cpu(el->l_next_free_rec) == 0)
3037                 return -ENODATA;
3038
3039         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3040              name, name_hash, name_index);
3041
3042         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3043                                   &num_clusters, el);
3044         if (ret) {
3045                 mlog_errno(ret);
3046                 goto out;
3047         }
3048
3049         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3050
3051         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3052              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3053              first_hash);
3054
3055         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3056                                       p_blkno, first_hash, num_clusters, xs);
3057
3058 out:
3059         return ret;
3060 }
3061
3062 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3063                                        u64 blkno,
3064                                        u32 clusters,
3065                                        xattr_bucket_func *func,
3066                                        void *para)
3067 {
3068         int i, ret = 0;
3069         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3070         u32 num_buckets = clusters * bpc;
3071         struct ocfs2_xattr_bucket *bucket;
3072
3073         bucket = ocfs2_xattr_bucket_new(inode);
3074         if (!bucket) {
3075                 mlog_errno(-ENOMEM);
3076                 return -ENOMEM;
3077         }
3078
3079         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3080              clusters, (unsigned long long)blkno);
3081
3082         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3083                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3084                 if (ret) {
3085                         mlog_errno(ret);
3086                         break;
3087                 }
3088
3089                 /*
3090                  * The real bucket num in this series of blocks is stored
3091                  * in the 1st bucket.
3092                  */
3093                 if (i == 0)
3094                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3095
3096                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3097                      (unsigned long long)blkno,
3098                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3099                 if (func) {
3100                         ret = func(inode, bucket, para);
3101                         if (ret)
3102                                 mlog_errno(ret);
3103                         /* Fall through to bucket_relse() */
3104                 }
3105
3106                 ocfs2_xattr_bucket_relse(bucket);
3107                 if (ret)
3108                         break;
3109         }
3110
3111         ocfs2_xattr_bucket_free(bucket);
3112         return ret;
3113 }
3114
3115 struct ocfs2_xattr_tree_list {
3116         char *buffer;
3117         size_t buffer_size;
3118         size_t result;
3119 };
3120
3121 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3122                                              struct ocfs2_xattr_header *xh,
3123                                              int index,
3124                                              int *block_off,
3125                                              int *new_offset)
3126 {
3127         u16 name_offset;
3128
3129         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3130                 return -EINVAL;
3131
3132         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3133
3134         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3135         *new_offset = name_offset % inode->i_sb->s_blocksize;
3136
3137         return 0;
3138 }
3139
3140 static int ocfs2_list_xattr_bucket(struct inode *inode,
3141                                    struct ocfs2_xattr_bucket *bucket,
3142                                    void *para)
3143 {
3144         int ret = 0, type;
3145         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3146         int i, block_off, new_offset;
3147         const char *prefix, *name;
3148
3149         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3150                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3151                 type = ocfs2_xattr_get_type(entry);
3152                 prefix = ocfs2_xattr_prefix(type);
3153
3154                 if (prefix) {
3155                         ret = ocfs2_xattr_bucket_get_name_value(inode,
3156                                                                 bucket_xh(bucket),
3157                                                                 i,
3158                                                                 &block_off,
3159                                                                 &new_offset);
3160                         if (ret)
3161                                 break;
3162
3163                         name = (const char *)bucket_block(bucket, block_off) +
3164                                 new_offset;
3165                         ret = ocfs2_xattr_list_entry(xl->buffer,
3166                                                      xl->buffer_size,
3167                                                      &xl->result,
3168                                                      prefix, name,
3169                                                      entry->xe_name_len);
3170                         if (ret)
3171                                 break;
3172                 }
3173         }
3174
3175         return ret;
3176 }
3177
3178 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3179                                              struct ocfs2_xattr_tree_root *xt,
3180                                              char *buffer,
3181                                              size_t buffer_size)
3182 {
3183         struct ocfs2_extent_list *el = &xt->xt_list;
3184         int ret = 0;
3185         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3186         u64 p_blkno = 0;
3187         struct ocfs2_xattr_tree_list xl = {
3188                 .buffer = buffer,
3189                 .buffer_size = buffer_size,
3190                 .result = 0,
3191         };
3192
3193         if (le16_to_cpu(el->l_next_free_rec) == 0)
3194                 return 0;
3195
3196         while (name_hash > 0) {
3197                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3198                                           &e_cpos, &num_clusters, el);
3199                 if (ret) {
3200                         mlog_errno(ret);
3201                         goto out;
3202                 }
3203
3204                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3205                                                   ocfs2_list_xattr_bucket,
3206                                                   &xl);
3207                 if (ret) {
3208                         mlog_errno(ret);
3209                         goto out;
3210                 }
3211
3212                 if (e_cpos == 0)
3213                         break;
3214
3215                 name_hash = e_cpos - 1;
3216         }
3217
3218         ret = xl.result;
3219 out:
3220         return ret;
3221 }
3222
3223 static int cmp_xe(const void *a, const void *b)
3224 {
3225         const struct ocfs2_xattr_entry *l = a, *r = b;
3226         u32 l_hash = le32_to_cpu(l->xe_name_hash);
3227         u32 r_hash = le32_to_cpu(r->xe_name_hash);
3228
3229         if (l_hash > r_hash)
3230                 return 1;
3231         if (l_hash < r_hash)
3232                 return -1;
3233         return 0;
3234 }
3235
3236 static void swap_xe(void *a, void *b, int size)
3237 {
3238         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3239
3240         tmp = *l;
3241         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3242         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3243 }
3244
3245 /*
3246  * When the ocfs2_xattr_block is filled up, new bucket will be created
3247  * and all the xattr entries will be moved to the new bucket.
3248  * The header goes at the start of the bucket, and the names+values are
3249  * filled from the end.  This is why *target starts as the last buffer.
3250  * Note: we need to sort the entries since they are not saved in order
3251  * in the ocfs2_xattr_block.
3252  */
3253 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3254                                            struct buffer_head *xb_bh,
3255                                            struct ocfs2_xattr_bucket *bucket)
3256 {
3257         int i, blocksize = inode->i_sb->s_blocksize;
3258         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3259         u16 offset, size, off_change;
3260         struct ocfs2_xattr_entry *xe;
3261         struct ocfs2_xattr_block *xb =
3262                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
3263         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3264         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3265         u16 count = le16_to_cpu(xb_xh->xh_count);
3266         char *src = xb_bh->b_data;
3267         char *target = bucket_block(bucket, blks - 1);
3268
3269         mlog(0, "cp xattr from block %llu to bucket %llu\n",
3270              (unsigned long long)xb_bh->b_blocknr,
3271              (unsigned long long)bucket_blkno(bucket));
3272
3273         for (i = 0; i < blks; i++)
3274                 memset(bucket_block(bucket, i), 0, blocksize);
3275
3276         /*
3277          * Since the xe_name_offset is based on ocfs2_xattr_header,
3278          * there is a offset change corresponding to the change of
3279          * ocfs2_xattr_header's position.
3280          */
3281         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3282         xe = &xb_xh->xh_entries[count - 1];
3283         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3284         size = blocksize - offset;
3285
3286         /* copy all the names and values. */
3287         memcpy(target + offset, src + offset, size);
3288
3289         /* Init new header now. */
3290         xh->xh_count = xb_xh->xh_count;
3291         xh->xh_num_buckets = cpu_to_le16(1);
3292         xh->xh_name_value_len = cpu_to_le16(size);
3293         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3294
3295         /* copy all the entries. */
3296         target = bucket_block(bucket, 0);
3297         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3298         size = count * sizeof(struct ocfs2_xattr_entry);
3299         memcpy(target + offset, (char *)xb_xh + offset, size);
3300
3301         /* Change the xe offset for all the xe because of the move. */
3302         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3303                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3304         for (i = 0; i < count; i++)
3305                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3306
3307         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3308              offset, size, off_change);
3309
3310         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3311              cmp_xe, swap_xe);
3312 }
3313
3314 /*
3315  * After we move xattr from block to index btree, we have to
3316  * update ocfs2_xattr_search to the new xe and base.
3317  *
3318  * When the entry is in xattr block, xattr_bh indicates the storage place.
3319  * While if the entry is in index b-tree, "bucket" indicates the
3320  * real place of the xattr.
3321  */
3322 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3323                                             struct ocfs2_xattr_search *xs,
3324                                             struct buffer_head *old_bh)
3325 {
3326         char *buf = old_bh->b_data;
3327         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3328         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3329         int i;
3330
3331         xs->header = bucket_xh(xs->bucket);
3332         xs->base = bucket_block(xs->bucket, 0);
3333         xs->end = xs->base + inode->i_sb->s_blocksize;
3334
3335         if (xs->not_found)
3336                 return;
3337
3338         i = xs->here - old_xh->xh_entries;
3339         xs->here = &xs->header->xh_entries[i];
3340 }
3341
3342 static int ocfs2_xattr_create_index_block(struct inode *inode,
3343                                           struct ocfs2_xattr_search *xs,
3344                                           struct ocfs2_xattr_set_ctxt *ctxt)
3345 {
3346         int ret;
3347         u32 bit_off, len;
3348         u64 blkno;
3349         handle_t *handle = ctxt->handle;
3350         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3351         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3352         struct buffer_head *xb_bh = xs->xattr_bh;
3353         struct ocfs2_xattr_block *xb =
3354                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3355         struct ocfs2_xattr_tree_root *xr;
3356         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3357
3358         mlog(0, "create xattr index block for %llu\n",
3359              (unsigned long long)xb_bh->b_blocknr);
3360
3361         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3362         BUG_ON(!xs->bucket);
3363
3364         /*
3365          * XXX:
3366          * We can use this lock for now, and maybe move to a dedicated mutex
3367          * if performance becomes a problem later.
3368          */
3369         down_write(&oi->ip_alloc_sem);
3370
3371         ret = ocfs2_journal_access(handle, inode, xb_bh,
3372                                    OCFS2_JOURNAL_ACCESS_WRITE);
3373         if (ret) {
3374                 mlog_errno(ret);
3375                 goto out;
3376         }
3377
3378         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3379                                      1, 1, &bit_off, &len);
3380         if (ret) {
3381                 mlog_errno(ret);
3382                 goto out;
3383         }
3384
3385         /*
3386          * The bucket may spread in many blocks, and
3387          * we will only touch the 1st block and the last block
3388          * in the whole bucket(one for entry and one for data).
3389          */
3390         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3391
3392         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3393              (unsigned long long)blkno);
3394
3395         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3396         if (ret) {
3397                 mlog_errno(ret);
3398                 goto out;
3399         }
3400
3401         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3402                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3403         if (ret) {
3404                 mlog_errno(ret);
3405                 goto out;
3406         }
3407
3408         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3409         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3410
3411         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3412
3413         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3414         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3415                offsetof(struct ocfs2_xattr_block, xb_attrs));
3416
3417         xr = &xb->xb_attrs.xb_root;
3418         xr->xt_clusters = cpu_to_le32(1);
3419         xr->xt_last_eb_blk = 0;
3420         xr->xt_list.l_tree_depth = 0;
3421         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3422         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3423
3424         xr->xt_list.l_recs[0].e_cpos = 0;
3425         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3426         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3427
3428         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3429
3430         ocfs2_journal_dirty(handle, xb_bh);
3431
3432 out:
3433         up_write(&oi->ip_alloc_sem);
3434
3435         return ret;
3436 }
3437
3438 static int cmp_xe_offset(const void *a, const void *b)
3439 {
3440         const struct ocfs2_xattr_entry *l = a, *r = b;
3441         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3442         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3443
3444         if (l_name_offset < r_name_offset)
3445                 return 1;
3446         if (l_name_offset > r_name_offset)
3447                 return -1;
3448         return 0;
3449 }
3450
3451 /*
3452  * defrag a xattr bucket if we find that the bucket has some
3453  * holes beteen name/value pairs.
3454  * We will move all the name/value pairs to the end of the bucket
3455  * so that we can spare some space for insertion.
3456  */
3457 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3458                                      handle_t *handle,
3459                                      struct ocfs2_xattr_bucket *bucket)
3460 {
3461         int ret, i;
3462         size_t end, offset, len, value_len;
3463         struct ocfs2_xattr_header *xh;
3464         char *entries, *buf, *bucket_buf = NULL;
3465         u64 blkno = bucket_blkno(bucket);
3466         u16 xh_free_start;
3467         size_t blocksize = inode->i_sb->s_blocksize;
3468         struct ocfs2_xattr_entry *xe;
3469
3470         /*
3471          * In order to make the operation more efficient and generic,
3472          * we copy all the blocks into a contiguous memory and do the
3473          * defragment there, so if anything is error, we will not touch
3474          * the real block.
3475          */
3476         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3477         if (!bucket_buf) {
3478                 ret = -EIO;
3479                 goto out;
3480         }
3481
3482         buf = bucket_buf;
3483         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3484                 memcpy(buf, bucket_block(bucket, i), blocksize);
3485
3486         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3487                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3488         if (ret < 0) {
3489                 mlog_errno(ret);
3490                 goto out;
3491         }
3492
3493         xh = (struct ocfs2_xattr_header *)bucket_buf;
3494         entries = (char *)xh->xh_entries;
3495         xh_free_start = le16_to_cpu(xh->xh_free_start);
3496
3497         mlog(0, "adjust xattr bucket in %llu, count = %u, "
3498              "xh_free_start = %u, xh_name_value_len = %u.\n",
3499              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3500              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3501
3502         /*
3503          * sort all the entries by their offset.
3504          * the largest will be the first, so that we can
3505          * move them to the end one by one.
3506          */
3507         sort(entries, le16_to_cpu(xh->xh_count),
3508              sizeof(struct ocfs2_xattr_entry),
3509              cmp_xe_offset, swap_xe);
3510
3511         /* Move all name/values to the end of the bucket. */
3512         xe = xh->xh_entries;
3513         end = OCFS2_XATTR_BUCKET_SIZE;
3514         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3515                 offset = le16_to_cpu(xe->xe_name_offset);
3516                 if (ocfs2_xattr_is_local(xe))
3517                         value_len = OCFS2_XATTR_SIZE(
3518                                         le64_to_cpu(xe->xe_value_size));
3519                 else
3520                         value_len = OCFS2_XATTR_ROOT_SIZE;
3521                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3522
3523                 /*
3524                  * We must make sure that the name/value pair
3525                  * exist in the same block. So adjust end to
3526                  * the previous block end if needed.
3527                  */
3528                 if (((end - len) / blocksize !=
3529                         (end - 1) / blocksize))
3530                         end = end - end % blocksize;
3531
3532                 if (end > offset + len) {
3533                         memmove(bucket_buf + end - len,
3534                                 bucket_buf + offset, len);
3535                         xe->xe_name_offset = cpu_to_le16(end - len);
3536                 }
3537
3538                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3539                                 "bucket %llu\n", (unsigned long long)blkno);
3540
3541                 end -= len;
3542         }
3543
3544         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3545                         "bucket %llu\n", (unsigned long long)blkno);
3546
3547         if (xh_free_start == end)
3548                 goto out;
3549
3550         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3551         xh->xh_free_start = cpu_to_le16(end);
3552
3553         /* sort the entries by their name_hash. */
3554         sort(entries, le16_to_cpu(xh->xh_count),
3555              sizeof(struct ocfs2_xattr_entry),
3556              cmp_xe, swap_xe);
3557
3558         buf = bucket_buf;
3559         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3560                 memcpy(bucket_block(bucket, i), buf, blocksize);
3561         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3562
3563 out:
3564         kfree(bucket_buf);
3565         return ret;
3566 }
3567
3568 /*
3569  * prev_blkno points to the start of an existing extent.  new_blkno
3570  * points to a newly allocated extent.  Because we know each of our
3571  * clusters contains more than bucket, we can easily split one cluster
3572  * at a bucket boundary.  So we take the last cluster of the existing
3573  * extent and split it down the middle.  We move the last half of the
3574  * buckets in the last cluster of the existing extent over to the new
3575  * extent.
3576  *
3577  * first_bh is the buffer at prev_blkno so we can update the existing
3578  * extent's bucket count.  header_bh is the bucket were we were hoping
3579  * to insert our xattr.  If the bucket move places the target in the new
3580  * extent, we'll update first_bh and header_bh after modifying the old
3581  * extent.
3582  *
3583  * first_hash will be set as the 1st xe's name_hash in the new extent.
3584  */
3585 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3586                                                handle_t *handle,
3587                                                struct ocfs2_xattr_bucket *first,
3588                                                struct ocfs2_xattr_bucket *target,
3589                                                u64 new_blkno,
3590                                                u32 num_clusters,
3591                                                u32 *first_hash)
3592 {
3593         int ret;
3594         struct super_block *sb = inode->i_sb;
3595         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3596         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3597         int to_move = num_buckets / 2;
3598         u64 src_blkno;
3599         u64 last_cluster_blkno = bucket_blkno(first) +
3600                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3601
3602         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3603         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3604
3605         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3606              (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3607
3608         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3609                                      last_cluster_blkno, new_blkno,
3610                                      to_move, first_hash);
3611         if (ret) {
3612                 mlog_errno(ret);
3613                 goto out;
3614         }
3615
3616         /* This is the first bucket that got moved */
3617         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3618
3619         /*
3620          * If the target bucket was part of the moved buckets, we need to
3621          * update first and target.
3622          */
3623         if (bucket_blkno(target) >= src_blkno) {
3624                 /* Find the block for the new target bucket */
3625                 src_blkno = new_blkno +
3626                         (bucket_blkno(target) - src_blkno);
3627
3628                 ocfs2_xattr_bucket_relse(first);
3629                 ocfs2_xattr_bucket_relse(target);
3630
3631                 /*
3632                  * These shouldn't fail - the buffers are in the
3633                  * journal from ocfs2_cp_xattr_bucket().
3634                  */
3635                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3636                 if (ret) {
3637                         mlog_errno(ret);
3638                         goto out;
3639                 }
3640                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3641                 if (ret)
3642                         mlog_errno(ret);
3643
3644         }
3645
3646 out:
3647         return ret;
3648 }
3649
3650 /*
3651  * Find the suitable pos when we divide a bucket into 2.
3652  * We have to make sure the xattrs with the same hash value exist
3653  * in the same bucket.
3654  *
3655  * If this ocfs2_xattr_header covers more than one hash value, find a
3656  * place where the hash value changes.  Try to find the most even split.
3657  * The most common case is that all entries have different hash values,
3658  * and the first check we make will find a place to split.
3659  */
3660 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3661 {
3662         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3663         int count = le16_to_cpu(xh->xh_count);
3664         int delta, middle = count / 2;
3665
3666         /*
3667          * We start at the middle.  Each step gets farther away in both
3668          * directions.  We therefore hit the change in hash value
3669          * nearest to the middle.  Note that this loop does not execute for
3670          * count < 2.
3671          */
3672         for (delta = 0; delta < middle; delta++) {
3673                 /* Let's check delta earlier than middle */
3674                 if (cmp_xe(&entries[middle - delta - 1],
3675                            &entries[middle - delta]))
3676                         return middle - delta;
3677
3678                 /* For even counts, don't walk off the end */
3679                 if ((middle + delta + 1) == count)
3680                         continue;
3681
3682                 /* Now try delta past middle */
3683                 if (cmp_xe(&entries[middle + delta],
3684                            &entries[middle + delta + 1]))
3685                         return middle + delta + 1;
3686         }
3687
3688         /* Every entry had the same hash */
3689         return count;
3690 }
3691
3692 /*
3693  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3694  * first_hash will record the 1st hash of the new bucket.
3695  *
3696  * Normally half of the xattrs will be moved.  But we have to make
3697  * sure that the xattrs with the same hash value are stored in the
3698  * same bucket. If all the xattrs in this bucket have the same hash
3699  * value, the new bucket will be initialized as an empty one and the
3700  * first_hash will be initialized as (hash_value+1).
3701  */
3702 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3703                                     handle_t *handle,
3704                                     u64 blk,
3705                                     u64 new_blk,
3706                                     u32 *first_hash,
3707                                     int new_bucket_head)
3708 {
3709         int ret, i;
3710         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3711         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3712         struct ocfs2_xattr_header *xh;
3713         struct ocfs2_xattr_entry *xe;
3714         int blocksize = inode->i_sb->s_blocksize;
3715
3716         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3717              (unsigned long long)blk, (unsigned long long)new_blk);
3718
3719         s_bucket = ocfs2_xattr_bucket_new(inode);
3720         t_bucket = ocfs2_xattr_bucket_new(inode);
3721         if (!s_bucket || !t_bucket) {
3722                 ret = -ENOMEM;
3723                 mlog_errno(ret);
3724                 goto out;
3725         }
3726
3727         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3728         if (ret) {
3729                 mlog_errno(ret);
3730                 goto out;
3731         }
3732
3733         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3734                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3735         if (ret) {
3736                 mlog_errno(ret);
3737                 goto out;
3738         }
3739
3740         /*
3741          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3742          * there's no need to read it.
3743          */
3744         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3745         if (ret) {
3746                 mlog_errno(ret);
3747                 goto out;
3748         }
3749
3750         /*
3751          * Hey, if we're overwriting t_bucket, what difference does
3752          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
3753          * same part of ocfs2_cp_xattr_bucket().
3754          */
3755         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3756                                                 new_bucket_head ?
3757                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3758                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3759         if (ret) {
3760                 mlog_errno(ret);
3761                 goto out;
3762         }
3763
3764         xh = bucket_xh(s_bucket);
3765         count = le16_to_cpu(xh->xh_count);
3766         start = ocfs2_xattr_find_divide_pos(xh);
3767
3768         if (start == count) {
3769                 xe = &xh->xh_entries[start-1];
3770
3771                 /*
3772                  * initialized a new empty bucket here.
3773                  * The hash value is set as one larger than
3774                  * that of the last entry in the previous bucket.
3775                  */
3776                 for (i = 0; i < t_bucket->bu_blocks; i++)
3777                         memset(bucket_block(t_bucket, i), 0, blocksize);
3778
3779                 xh = bucket_xh(t_bucket);
3780                 xh->xh_free_start = cpu_to_le16(blocksize);
3781                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3782                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3783
3784                 goto set_num_buckets;
3785         }
3786
3787         /* copy the whole bucket to the new first. */
3788         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3789
3790         /* update the new bucket. */
3791         xh = bucket_xh(t_bucket);
3792
3793         /*
3794          * Calculate the total name/value len and xh_free_start for
3795          * the old bucket first.
3796          */
3797         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3798         name_value_len = 0;
3799         for (i = 0; i < start; i++) {
3800                 xe = &xh->xh_entries[i];
3801                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3802                 if (ocfs2_xattr_is_local(xe))
3803                         xe_len +=
3804                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3805                 else
3806                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3807                 name_value_len += xe_len;
3808                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3809                         name_offset = le16_to_cpu(xe->xe_name_offset);
3810         }
3811
3812         /*
3813          * Now begin the modification to the new bucket.
3814          *
3815          * In the new bucket, We just move the xattr entry to the beginning
3816          * and don't touch the name/value. So there will be some holes in the
3817          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3818          * called.
3819          */
3820         xe = &xh->xh_entries[start];
3821         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3822         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3823              (int)((char *)xe - (char *)xh),
3824              (int)((char *)xh->xh_entries - (char *)xh));
3825         memmove((char *)xh->xh_entries, (char *)xe, len);
3826         xe = &xh->xh_entries[count - start];
3827         len = sizeof(struct ocfs2_xattr_entry) * start;
3828         memset((char *)xe, 0, len);
3829
3830         le16_add_cpu(&xh->xh_count, -start);
3831         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3832
3833         /* Calculate xh_free_start for the new bucket. */
3834         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3835         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3836                 xe = &xh->xh_entries[i];
3837                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3838                 if (ocfs2_xattr_is_local(xe))
3839                         xe_len +=
3840                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3841                 else
3842                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3843                 if (le16_to_cpu(xe->xe_name_offset) <
3844                     le16_to_cpu(xh->xh_free_start))
3845                         xh->xh_free_start = xe->xe_name_offset;
3846         }
3847
3848 set_num_buckets:
3849         /* set xh->xh_num_buckets for the new xh. */
3850         if (new_bucket_head)
3851                 xh->xh_num_buckets = cpu_to_le16(1);
3852         else
3853                 xh->xh_num_buckets = 0;
3854
3855         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3856
3857         /* store the first_hash of the new bucket. */
3858         if (first_hash)
3859                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3860
3861         /*
3862          * Now only update the 1st block of the old bucket.  If we
3863          * just added a new empty bucket, there is no need to modify
3864          * it.
3865          */
3866         if (start == count)
3867                 goto out;
3868
3869         xh = bucket_xh(s_bucket);
3870         memset(&xh->xh_entries[start], 0,
3871                sizeof(struct ocfs2_xattr_entry) * (count - start));
3872         xh->xh_count = cpu_to_le16(start);
3873         xh->xh_free_start = cpu_to_le16(name_offset);
3874         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3875
3876         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3877
3878 out:
3879         ocfs2_xattr_bucket_free(s_bucket);
3880         ocfs2_xattr_bucket_free(t_bucket);
3881
3882         return ret;
3883 }
3884
3885 /*
3886  * Copy xattr from one bucket to another bucket.
3887  *
3888  * The caller must make sure that the journal transaction
3889  * has enough space for journaling.
3890  */
3891 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3892                                  handle_t *handle,
3893                                  u64 s_blkno,
3894                                  u64 t_blkno,
3895                                  int t_is_new)
3896 {
3897         int ret;
3898         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3899
3900         BUG_ON(s_blkno == t_blkno);
3901
3902         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3903              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3904              t_is_new);
3905
3906         s_bucket = ocfs2_xattr_bucket_new(inode);
3907         t_bucket = ocfs2_xattr_bucket_new(inode);
3908         if (!s_bucket || !t_bucket) {
3909                 ret = -ENOMEM;
3910                 mlog_errno(ret);
3911                 goto out;
3912         }
3913
3914         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3915         if (ret)
3916                 goto out;
3917
3918         /*
3919          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3920          * there's no need to read it.
3921          */
3922         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3923         if (ret)
3924                 goto out;
3925
3926         /*
3927          * Hey, if we're overwriting t_bucket, what difference does
3928          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
3929          * cluster to fill, we came here from
3930          * ocfs2_mv_xattr_buckets(), and it is really new -
3931          * ACCESS_CREATE is required.  But we also might have moved data
3932          * out of t_bucket before extending back into it.
3933          * ocfs2_add_new_xattr_bucket() can do this - its call to
3934          * ocfs2_add_new_xattr_cluster() may have created a new extent
3935          * and copied out the end of the old extent.  Then it re-extends
3936          * the old extent back to create space for new xattrs.  That's
3937          * how we get here, and the bucket isn't really new.
3938          */
3939         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3940                                                 t_is_new ?
3941                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3942                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3943         if (ret)
3944                 goto out;
3945
3946         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3947         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3948
3949 out:
3950         ocfs2_xattr_bucket_free(t_bucket);
3951         ocfs2_xattr_bucket_free(s_bucket);
3952
3953         return ret;
3954 }
3955
3956 /*
3957  * src_blk points to the start of an existing extent.  last_blk points to
3958  * last cluster in that extent.  to_blk points to a newly allocated
3959  * extent.  We copy the buckets from the cluster at last_blk to the new
3960  * extent.  If start_bucket is non-zero, we skip that many buckets before
3961  * we start copying.  The new extent's xh_num_buckets gets set to the
3962  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
3963  * by the same amount.
3964  */
3965 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
3966                                   u64 src_blk, u64 last_blk, u64 to_blk,
3967                                   unsigned int start_bucket,
3968                                   u32 *first_hash)
3969 {
3970         int i, ret, credits;
3971         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3972         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3973         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3974         struct ocfs2_xattr_bucket *old_first, *new_first;
3975
3976         mlog(0, "mv xattrs from cluster %llu to %llu\n",
3977              (unsigned long long)last_blk, (unsigned long long)to_blk);
3978
3979         BUG_ON(start_bucket >= num_buckets);
3980         if (start_bucket) {
3981                 num_buckets -= start_bucket;
3982                 last_blk += (start_bucket * blks_per_bucket);
3983         }
3984
3985         /* The first bucket of the original extent */
3986         old_first = ocfs2_xattr_bucket_new(inode);
3987         /* The first bucket of the new extent */
3988         new_first = ocfs2_xattr_bucket_new(inode);
3989         if (!old_first || !new_first) {
3990                 ret = -ENOMEM;
3991                 mlog_errno(ret);
3992                 goto out;
3993         }
3994
3995         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
3996         if (ret) {
3997                 mlog_errno(ret);
3998                 goto out;
3999         }
4000
4001         /*
4002          * We need to update the first bucket of the old extent and all
4003          * the buckets going to the new extent.
4004          */
4005         credits = ((num_buckets + 1) * blks_per_bucket) +
4006                 handle->h_buffer_credits;
4007         ret = ocfs2_extend_trans(handle, credits);
4008         if (ret) {
4009                 mlog_errno(ret);
4010                 goto out;
4011         }
4012
4013         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4014                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4015         if (ret) {
4016                 mlog_errno(ret);
4017                 goto out;
4018         }
4019
4020         for (i = 0; i < num_buckets; i++) {
4021                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4022                                             last_blk + (i * blks_per_bucket),
4023                                             to_blk + (i * blks_per_bucket),
4024                                             1);
4025                 if (ret) {
4026                         mlog_errno(ret);
4027                         goto out;
4028                 }
4029         }
4030
4031         /*
4032          * Get the new bucket ready before we dirty anything
4033          * (This actually shouldn't fail, because we already dirtied
4034          * it once in ocfs2_cp_xattr_bucket()).
4035          */
4036         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4037         if (ret) {
4038                 mlog_errno(ret);
4039                 goto out;
4040         }
4041         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4042                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4043         if (ret) {
4044                 mlog_errno(ret);
4045                 goto out;
4046         }
4047
4048         /* Now update the headers */
4049         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4050         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4051
4052         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4053         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4054
4055         if (first_hash)
4056                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4057
4058 out:
4059         ocfs2_xattr_bucket_free(new_first);
4060         ocfs2_xattr_bucket_free(old_first);
4061         return ret;
4062 }
4063
4064 /*
4065  * Move some xattrs in this cluster to the new cluster.
4066  * This function should only be called when bucket size == cluster size.
4067  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4068  */
4069 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4070                                       handle_t *handle,
4071                                       u64 prev_blk,
4072                                       u64 new_blk,
4073                                       u32 *first_hash)
4074 {
4075         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4076         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4077
4078         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4079
4080         ret = ocfs2_extend_trans(handle, credits);
4081         if (ret) {
4082                 mlog_errno(ret);
4083                 return ret;
4084         }
4085
4086         /* Move half of the xattr in start_blk to the next bucket. */
4087         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4088                                           new_blk, first_hash, 1);
4089 }
4090
4091 /*
4092  * Move some xattrs from the old cluster to the new one since they are not
4093  * contiguous in ocfs2 xattr tree.
4094  *
4095  * new_blk starts a new separate cluster, and we will move some xattrs from
4096  * prev_blk to it. v_start will be set as the first name hash value in this
4097  * new cluster so that it can be used as e_cpos during tree insertion and
4098  * don't collide with our original b-tree operations. first_bh and header_bh
4099  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4100  * to extend the insert bucket.
4101  *
4102  * The problem is how much xattr should we move to the new one and when should
4103  * we update first_bh and header_bh?
4104  * 1. If cluster size > bucket size, that means the previous cluster has more
4105  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4106  *    update the first_bh and header_bh if the insert bucket has been moved
4107  *    to the new cluster.
4108  * 2. If cluster_size == bucket_size:
4109  *    a) If the previous extent rec has more than one cluster and the insert
4110  *       place isn't in the last cluster, copy the entire last cluster to the
4111  *       new one. This time, we don't need to upate the first_bh and header_bh
4112  *       since they will not be moved into the new cluster.
4113  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4114  *       the new one. And we set the extend flag to zero if the insert place is
4115  *       moved into the new allocated cluster since no extend is needed.
4116  */
4117 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4118                                             handle_t *handle,
4119                                             struct ocfs2_xattr_bucket *first,
4120                                             struct ocfs2_xattr_bucket *target,
4121                                             u64 new_blk,
4122                                             u32 prev_clusters,
4123                                             u32 *v_start,
4124                                             int *extend)
4125 {
4126         int ret;
4127
4128         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4129              (unsigned long long)bucket_blkno(first), prev_clusters,
4130              (unsigned long long)new_blk);
4131
4132         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4133                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4134                                                           handle,
4135                                                           first, target,
4136                                                           new_blk,
4137                                                           prev_clusters,
4138                                                           v_start);
4139                 if (ret)
4140                         mlog_errno(ret);
4141         } else {
4142                 /* The start of the last cluster in the first extent */
4143                 u64 last_blk = bucket_blkno(first) +
4144                         ((prev_clusters - 1) *
4145                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
4146
4147                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4148                         ret = ocfs2_mv_xattr_buckets(inode, handle,
4149                                                      bucket_blkno(first),
4150                                                      last_blk, new_blk, 0,
4151                                                      v_start);
4152                         if (ret)
4153                                 mlog_errno(ret);
4154                 } else {
4155                         ret = ocfs2_divide_xattr_cluster(inode, handle,
4156                                                          last_blk, new_blk,
4157                                                          v_start);
4158                         if (ret)
4159                                 mlog_errno(ret);
4160
4161                         if ((bucket_blkno(target) == last_blk) && extend)
4162                                 *extend = 0;
4163                 }
4164         }
4165
4166         return ret;
4167 }
4168
4169 /*
4170  * Add a new cluster for xattr storage.
4171  *
4172  * If the new cluster is contiguous with the previous one, it will be
4173  * appended to the same extent record, and num_clusters will be updated.
4174  * If not, we will insert a new extent for it and move some xattrs in
4175  * the last cluster into the new allocated one.
4176  * We also need to limit the maximum size of a btree leaf, otherwise we'll
4177  * lose the benefits of hashing because we'll have to search large leaves.
4178  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4179  * if it's bigger).
4180  *
4181  * first_bh is the first block of the previous extent rec and header_bh
4182  * indicates the bucket we will insert the new xattrs. They will be updated
4183  * when the header_bh is moved into the new cluster.
4184  */
4185 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4186                                        struct buffer_head *root_bh,
4187                                        struct ocfs2_xattr_bucket *first,
4188                                        struct ocfs2_xattr_bucket *target,
4189                                        u32 *num_clusters,
4190                                        u32 prev_cpos,
4191                                        int *extend,
4192                                        struct ocfs2_xattr_set_ctxt *ctxt)
4193 {
4194         int ret;
4195         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4196         u32 prev_clusters = *num_clusters;
4197         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4198         u64 block;
4199         handle_t *handle = ctxt->handle;
4200         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4201         struct ocfs2_extent_tree et;
4202
4203         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4204              "previous xattr blkno = %llu\n",
4205              (unsigned long long)OCFS2_I(inode)->ip_blkno,
4206              prev_cpos, (unsigned long long)bucket_blkno(first));
4207
4208         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4209
4210         ret = ocfs2_journal_access(handle, inode, root_bh,
4211                                    OCFS2_JOURNAL_ACCESS_WRITE);
4212         if (ret < 0) {
4213                 mlog_errno(ret);
4214                 goto leave;
4215         }
4216
4217         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4218                                      clusters_to_add, &bit_off, &num_bits);
4219         if (ret < 0) {
4220                 if (ret != -ENOSPC)
4221                         mlog_errno(ret);
4222                 goto leave;
4223         }
4224
4225         BUG_ON(num_bits > clusters_to_add);
4226
4227         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4228         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4229              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4230
4231         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4232             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4233              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4234                 /*
4235                  * If this cluster is contiguous with the old one and
4236                  * adding this new cluster, we don't surpass the limit of
4237                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4238                  * initialized and used like other buckets in the previous
4239                  * cluster.
4240                  * So add it as a contiguous one. The caller will handle
4241                  * its init process.
4242                  */
4243                 v_start = prev_cpos + prev_clusters;
4244                 *num_clusters = prev_clusters + num_bits;
4245                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4246                      num_bits);
4247         } else {
4248                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4249                                                        handle,
4250                                                        first,
4251                                                        target,
4252                                                        block,
4253                                                        prev_clusters,
4254                                                        &v_start,
4255                                                        extend);
4256                 if (ret) {
4257                         mlog_errno(ret);
4258                         goto leave;
4259                 }
4260         }
4261
4262         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4263              num_bits, (unsigned long long)block, v_start);
4264         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
4265                                   num_bits, 0, ctxt->meta_ac);
4266         if (ret < 0) {
4267                 mlog_errno(ret);
4268                 goto leave;
4269         }
4270
4271         ret = ocfs2_journal_dirty(handle, root_bh);
4272         if (ret < 0)
4273                 mlog_errno(ret);
4274
4275 leave:
4276         return ret;
4277 }
4278
4279 /*
4280  * We are given an extent.  'first' is the bucket at the very front of
4281  * the extent.  The extent has space for an additional bucket past
4282  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
4283  * of the target bucket.  We wish to shift every bucket past the target
4284  * down one, filling in that additional space.  When we get back to the
4285  * target, we split the target between itself and the now-empty bucket
4286  * at target+1 (aka, target_blkno + blks_per_bucket).
4287  */
4288 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4289                                      handle_t *handle,
4290                                      struct ocfs2_xattr_bucket *first,
4291                                      u64 target_blk,
4292                                      u32 num_clusters)
4293 {
4294         int ret, credits;
4295         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4296         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4297         u64 end_blk;
4298         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4299
4300         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4301              "from %llu, len = %u\n", (unsigned long long)target_blk,
4302              (unsigned long long)bucket_blkno(first), num_clusters);
4303
4304         /* The extent must have room for an additional bucket */
4305         BUG_ON(new_bucket >=
4306                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4307
4308         /* end_blk points to the last existing bucket */
4309         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4310
4311         /*
4312          * end_blk is the start of the last existing bucket.
4313          * Thus, (end_blk - target_blk) covers the target bucket and
4314          * every bucket after it up to, but not including, the last
4315          * existing bucket.  Then we add the last existing bucket, the
4316          * new bucket, and the first bucket (3 * blk_per_bucket).
4317          */
4318         credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4319                   handle->h_buffer_credits;
4320         ret = ocfs2_extend_trans(handle, credits);
4321         if (ret) {
4322                 mlog_errno(ret);
4323                 goto out;
4324         }
4325
4326         ret = ocfs2_xattr_bucket_journal_access(handle, first,
4327                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4328         if (ret) {
4329                 mlog_errno(ret);
4330                 goto out;
4331         }
4332
4333         while (end_blk != target_blk) {
4334                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4335                                             end_blk + blk_per_bucket, 0);
4336                 if (ret)
4337                         goto out;
4338                 end_blk -= blk_per_bucket;
4339         }
4340
4341         /* Move half of the xattr in target_blkno to the next bucket. */
4342         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4343                                         target_blk + blk_per_bucket, NULL, 0);
4344
4345         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4346         ocfs2_xattr_bucket_journal_dirty(handle, first);
4347
4348 out:
4349         return ret;
4350 }
4351
4352 /*
4353  * Add new xattr bucket in an extent record and adjust the buckets
4354  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
4355  * bucket we want to insert into.
4356  *
4357  * In the easy case, we will move all the buckets after target down by
4358  * one. Half of target's xattrs will be moved to the next bucket.
4359  *
4360  * If current cluster is full, we'll allocate a new one.  This may not
4361  * be contiguous.  The underlying calls will make sure that there is
4362  * space for the insert, shifting buckets around if necessary.
4363  * 'target' may be moved by those calls.
4364  */
4365 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4366                                       struct buffer_head *xb_bh,
4367                                       struct ocfs2_xattr_bucket *target,
4368                                       struct ocfs2_xattr_set_ctxt *ctxt)
4369 {
4370         struct ocfs2_xattr_block *xb =
4371                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4372         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4373         struct ocfs2_extent_list *el = &xb_root->xt_list;
4374         u32 name_hash =
4375                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4376         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4377         int ret, num_buckets, extend = 1;
4378         u64 p_blkno;
4379         u32 e_cpos, num_clusters;
4380         /* The bucket at the front of the extent */
4381         struct ocfs2_xattr_bucket *first;
4382
4383         mlog(0, "Add new xattr bucket starting from %llu\n",
4384              (unsigned long long)bucket_blkno(target));
4385
4386         /* The first bucket of the original extent */
4387         first = ocfs2_xattr_bucket_new(inode);
4388         if (!first) {
4389                 ret = -ENOMEM;
4390                 mlog_errno(ret);
4391                 goto out;
4392         }
4393
4394         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4395                                   &num_clusters, el);
4396         if (ret) {
4397                 mlog_errno(ret);
4398                 goto out;
4399         }
4400
4401         ret = ocfs2_read_xattr_bucket(first, p_blkno);
4402         if (ret) {
4403                 mlog_errno(ret);
4404                 goto out;
4405         }
4406
4407         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4408         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4409                 /*
4410                  * This can move first+target if the target bucket moves
4411                  * to the new extent.
4412                  */
4413                 ret = ocfs2_add_new_xattr_cluster(inode,
4414                                                   xb_bh,
4415                                                   first,
4416                                                   target,
4417                                                   &num_clusters,
4418                                                   e_cpos,
4419                                                   &extend,
4420                                                   ctxt);
4421                 if (ret) {
4422                         mlog_errno(ret);
4423                         goto out;
4424                 }
4425         }
4426
4427         if (extend) {
4428                 ret = ocfs2_extend_xattr_bucket(inode,
4429                                                 ctxt->handle,
4430                                                 first,
4431                                                 bucket_blkno(target),
4432                                                 num_clusters);
4433                 if (ret)
4434                         mlog_errno(ret);
4435         }
4436
4437 out:
4438         ocfs2_xattr_bucket_free(first);
4439
4440         return ret;
4441 }
4442
4443 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4444                                         struct ocfs2_xattr_bucket *bucket,
4445                                         int offs)
4446 {
4447         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4448
4449         offs = offs % inode->i_sb->s_blocksize;
4450         return bucket_block(bucket, block_off) + offs;
4451 }
4452
4453 /*
4454  * Handle the normal xattr set, including replace, delete and new.
4455  *
4456  * Note: "local" indicates the real data's locality. So we can't
4457  * just its bucket locality by its length.
4458  */
4459 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4460                                          struct ocfs2_xattr_info *xi,
4461                                          struct ocfs2_xattr_search *xs,
4462                                          u32 name_hash,
4463                                          int local)
4464 {
4465         struct ocfs2_xattr_entry *last, *xe;
4466         int name_len = strlen(xi->name);
4467         struct ocfs2_xattr_header *xh = xs->header;
4468         u16 count = le16_to_cpu(xh->xh_count), start;
4469         size_t blocksize = inode->i_sb->s_blocksize;
4470         char *val;
4471         size_t offs, size, new_size;
4472
4473         last = &xh->xh_entries[count];
4474         if (!xs->not_found) {
4475                 xe = xs->here;
4476                 offs = le16_to_cpu(xe->xe_name_offset);
4477                 if (ocfs2_xattr_is_local(xe))
4478                         size = OCFS2_XATTR_SIZE(name_len) +
4479                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4480                 else
4481                         size = OCFS2_XATTR_SIZE(name_len) +
4482                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4483
4484                 /*
4485                  * If the new value will be stored outside, xi->value has been
4486                  * initalized as an empty ocfs2_xattr_value_root, and the same
4487                  * goes with xi->value_len, so we can set new_size safely here.
4488                  * See ocfs2_xattr_set_in_bucket.
4489                  */
4490                 new_size = OCFS2_XATTR_SIZE(name_len) +
4491                            OCFS2_XATTR_SIZE(xi->value_len);
4492
4493                 le16_add_cpu(&xh->xh_name_value_len, -size);
4494                 if (xi->value) {
4495                         if (new_size > size)
4496                                 goto set_new_name_value;
4497
4498                         /* Now replace the old value with new one. */
4499                         if (local)
4500                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4501                         else
4502                                 xe->xe_value_size = 0;
4503
4504                         val = ocfs2_xattr_bucket_get_val(inode,
4505                                                          xs->bucket, offs);
4506                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4507                                size - OCFS2_XATTR_SIZE(name_len));
4508                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4509                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4510                                        xi->value, xi->value_len);
4511
4512                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4513                         ocfs2_xattr_set_local(xe, local);
4514                         return;
4515                 } else {
4516                         /*
4517                          * Remove the old entry if there is more than one.
4518                          * We don't remove the last entry so that we can
4519                          * use it to indicate the hash value of the empty
4520                          * bucket.
4521                          */
4522                         last -= 1;
4523                         le16_add_cpu(&xh->xh_count, -1);
4524                         if (xh->xh_count) {
4525                                 memmove(xe, xe + 1,
4526                                         (void *)last - (void *)xe);
4527                                 memset(last, 0,
4528                                        sizeof(struct ocfs2_xattr_entry));
4529                         } else
4530                                 xh->xh_free_start =
4531                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4532
4533                         return;
4534                 }
4535         } else {
4536                 /* find a new entry for insert. */
4537                 int low = 0, high = count - 1, tmp;
4538                 struct ocfs2_xattr_entry *tmp_xe;
4539
4540                 while (low <= high && count) {
4541                         tmp = (low + high) / 2;
4542                         tmp_xe = &xh->xh_entries[tmp];
4543
4544                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4545                                 low = tmp + 1;
4546                         else if (name_hash <
4547                                  le32_to_cpu(tmp_xe->xe_name_hash))
4548                                 high = tmp - 1;
4549                         else {
4550                                 low = tmp;
4551                                 break;
4552                         }
4553                 }
4554
4555                 xe = &xh->xh_entries[low];
4556                 if (low != count)
4557                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4558
4559                 le16_add_cpu(&xh->xh_count, 1);
4560                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4561                 xe->xe_name_hash = cpu_to_le32(name_hash);
4562                 xe->xe_name_len = name_len;
4563                 ocfs2_xattr_set_type(xe, xi->name_index);
4564         }
4565
4566 set_new_name_value:
4567         /* Insert the new name+value. */
4568         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4569
4570         /*
4571          * We must make sure that the name/value pair
4572          * exists in the same block.
4573          */
4574         offs = le16_to_cpu(xh->xh_free_start);
4575         start = offs - size;
4576
4577         if (start >> inode->i_sb->s_blocksize_bits !=
4578             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4579                 offs = offs - offs % blocksize;
4580                 xh->xh_free_start = cpu_to_le16(offs);
4581         }
4582
4583         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4584         xe->xe_name_offset = cpu_to_le16(offs - size);
4585
4586         memset(val, 0, size);
4587         memcpy(val, xi->name, name_len);
4588         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4589
4590         xe->xe_value_size = cpu_to_le64(xi->value_len);
4591         ocfs2_xattr_set_local(xe, local);
4592         xs->here = xe;
4593         le16_add_cpu(&xh->xh_free_start, -size);
4594         le16_add_cpu(&xh->xh_name_value_len, size);
4595
4596         return;
4597 }
4598
4599 /*
4600  * Set the xattr entry in the specified bucket.
4601  * The bucket is indicated by xs->bucket and it should have the enough
4602  * space for the xattr insertion.
4603  */
4604 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4605                                            handle_t *handle,
4606                                            struct ocfs2_xattr_info *xi,
4607                                            struct ocfs2_xattr_search *xs,
4608                                            u32 name_hash,
4609                                            int local)
4610 {
4611         int ret;
4612         u64 blkno;
4613
4614         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4615              (unsigned long)xi->value_len, xi->name_index,
4616              (unsigned long long)bucket_blkno(xs->bucket));
4617
4618         if (!xs->bucket->bu_bhs[1]) {
4619                 blkno = bucket_blkno(xs->bucket);
4620                 ocfs2_xattr_bucket_relse(xs->bucket);
4621                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4622                 if (ret) {
4623                         mlog_errno(ret);
4624                         goto out;
4625                 }
4626         }
4627
4628         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4629                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4630         if (ret < 0) {
4631                 mlog_errno(ret);
4632                 goto out;
4633         }
4634
4635         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4636         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4637
4638 out:
4639         return ret;
4640 }
4641
4642 /*
4643  * Truncate the specified xe_off entry in xattr bucket.
4644  * bucket is indicated by header_bh and len is the new length.
4645  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4646  *
4647  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4648  */
4649 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4650                                              struct ocfs2_xattr_bucket *bucket,
4651                                              int xe_off,
4652                                              int len,
4653                                              struct ocfs2_xattr_set_ctxt *ctxt)
4654 {
4655         int ret, offset;
4656         u64 value_blk;
4657         struct ocfs2_xattr_entry *xe;
4658         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4659         size_t blocksize = inode->i_sb->s_blocksize;
4660         struct ocfs2_xattr_value_buf vb = {
4661                 .vb_access = ocfs2_journal_access,
4662         };
4663
4664         xe = &xh->xh_entries[xe_off];
4665
4666         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4667
4668         offset = le16_to_cpu(xe->xe_name_offset) +
4669                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4670
4671         value_blk = offset / blocksize;
4672
4673         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4674         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4675
4676         vb.vb_bh = bucket->bu_bhs[value_blk];
4677         BUG_ON(!vb.vb_bh);
4678
4679         vb.vb_xv = (struct ocfs2_xattr_value_root *)
4680                 (vb.vb_bh->b_data + offset % blocksize);
4681
4682         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4683                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4684         if (ret) {
4685                 mlog_errno(ret);
4686                 goto out;
4687         }
4688
4689         /*
4690          * From here on out we have to dirty the bucket.  The generic
4691          * value calls only modify one of the bucket's bhs, but we need
4692          * to send the bucket at once.  So if they error, they *could* have
4693          * modified something.  We have to assume they did, and dirty
4694          * the whole bucket.  This leaves us in a consistent state.
4695          */
4696         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4697              xe_off, (unsigned long long)bucket_blkno(bucket), len);
4698         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4699         if (ret) {
4700                 mlog_errno(ret);
4701                 goto out_dirty;
4702         }
4703
4704         xe->xe_value_size = cpu_to_le64(len);
4705
4706 out_dirty:
4707         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4708
4709 out:
4710         return ret;
4711 }
4712
4713 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4714                                         struct ocfs2_xattr_search *xs,
4715                                         int len,
4716                                         struct ocfs2_xattr_set_ctxt *ctxt)
4717 {
4718         int ret, offset;
4719         struct ocfs2_xattr_entry *xe = xs->here;
4720         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4721
4722         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4723
4724         offset = xe - xh->xh_entries;
4725         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4726                                                 offset, len, ctxt);
4727         if (ret)
4728                 mlog_errno(ret);
4729
4730         return ret;
4731 }
4732
4733 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4734                                                 handle_t *handle,
4735                                                 struct ocfs2_xattr_search *xs,
4736                                                 char *val,
4737                                                 int value_len)
4738 {
4739         int offset;
4740         struct ocfs2_xattr_value_root *xv;
4741         struct ocfs2_xattr_entry *xe = xs->here;
4742
4743         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4744
4745         offset = le16_to_cpu(xe->xe_name_offset) +
4746                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4747
4748         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4749
4750         return __ocfs2_xattr_set_value_outside(inode, handle,
4751                                                xv, val, value_len);
4752 }
4753
4754 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4755                                   struct buffer_head *root_bh,
4756                                   u64 blkno,
4757                                   u32 cpos,
4758                                   u32 len)
4759 {
4760         int ret;
4761         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4762         struct inode *tl_inode = osb->osb_tl_inode;
4763         handle_t *handle;
4764         struct ocfs2_xattr_block *xb =
4765                         (struct ocfs2_xattr_block *)root_bh->b_data;
4766         struct ocfs2_alloc_context *meta_ac = NULL;
4767         struct ocfs2_cached_dealloc_ctxt dealloc;
4768         struct ocfs2_extent_tree et;
4769
4770         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4771
4772         ocfs2_init_dealloc_ctxt(&dealloc);
4773
4774         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4775              cpos, len, (unsigned long long)blkno);
4776
4777         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4778
4779         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4780         if (ret) {
4781                 mlog_errno(ret);
4782                 return ret;
4783         }
4784
4785         mutex_lock(&tl_inode->i_mutex);
4786
4787         if (ocfs2_truncate_log_needs_flush(osb)) {
4788                 ret = __ocfs2_flush_truncate_log(osb);
4789                 if (ret < 0) {
4790                         mlog_errno(ret);
4791                         goto out;
4792                 }
4793         }
4794
4795         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4796         if (IS_ERR(handle)) {
4797                 ret = -ENOMEM;
4798                 mlog_errno(ret);
4799                 goto out;
4800         }
4801
4802         ret = ocfs2_journal_access(handle, inode, root_bh,
4803                                    OCFS2_JOURNAL_ACCESS_WRITE);
4804         if (ret) {
4805                 mlog_errno(ret);
4806                 goto out_commit;
4807         }
4808
4809         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4810                                   &dealloc);
4811         if (ret) {
4812                 mlog_errno(ret);
4813                 goto out_commit;
4814         }
4815
4816         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4817
4818         ret = ocfs2_journal_dirty(handle, root_bh);
4819         if (ret) {
4820                 mlog_errno(ret);
4821                 goto out_commit;
4822         }
4823
4824         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4825         if (ret)
4826                 mlog_errno(ret);
4827
4828 out_commit:
4829         ocfs2_commit_trans(osb, handle);
4830 out:
4831         ocfs2_schedule_truncate_log_flush(osb, 1);
4832
4833         mutex_unlock(&tl_inode->i_mutex);
4834
4835         if (meta_ac)
4836                 ocfs2_free_alloc_context(meta_ac);
4837
4838         ocfs2_run_deallocs(osb, &dealloc);
4839
4840         return ret;
4841 }
4842
4843 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4844                                          handle_t *handle,
4845                                          struct ocfs2_xattr_search *xs)
4846 {
4847         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4848         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4849                                                 le16_to_cpu(xh->xh_count) - 1];
4850         int ret = 0;
4851
4852         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4853                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4854         if (ret) {
4855                 mlog_errno(ret);
4856                 return;
4857         }
4858
4859         /* Remove the old entry. */
4860         memmove(xs->here, xs->here + 1,
4861                 (void *)last - (void *)xs->here);
4862         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4863         le16_add_cpu(&xh->xh_count, -1);
4864
4865         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4866 }
4867
4868 /*
4869  * Set the xattr name/value in the bucket specified in xs.
4870  *
4871  * As the new value in xi may be stored in the bucket or in an outside cluster,
4872  * we divide the whole process into 3 steps:
4873  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4874  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4875  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4876  * 4. If the clusters for the new outside value can't be allocated, we need
4877  *    to free the xattr we allocated in set.
4878  */
4879 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4880                                      struct ocfs2_xattr_info *xi,
4881                                      struct ocfs2_xattr_search *xs,
4882                                      struct ocfs2_xattr_set_ctxt *ctxt)
4883 {
4884         int ret, local = 1;
4885         size_t value_len;
4886         char *val = (char *)xi->value;
4887         struct ocfs2_xattr_entry *xe = xs->here;
4888         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4889                                               strlen(xi->name));
4890
4891         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4892                 /*
4893                  * We need to truncate the xattr storage first.
4894                  *
4895                  * If both the old and new value are stored to
4896                  * outside block, we only need to truncate
4897                  * the storage and then set the value outside.
4898                  *
4899                  * If the new value should be stored within block,
4900                  * we should free all the outside block first and
4901                  * the modification to the xattr block will be done
4902                  * by following steps.
4903                  */
4904                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4905                         value_len = xi->value_len;
4906                 else
4907                         value_len = 0;
4908
4909                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4910                                                            value_len,
4911                                                            ctxt);
4912                 if (ret)
4913                         goto out;
4914
4915                 if (value_len)
4916                         goto set_value_outside;
4917         }
4918
4919         value_len = xi->value_len;
4920         /* So we have to handle the inside block change now. */
4921         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4922                 /*
4923                  * If the new value will be stored outside of block,
4924                  * initalize a new empty value root and insert it first.
4925                  */
4926                 local = 0;
4927                 xi->value = &def_xv;
4928                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4929         }
4930
4931         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
4932                                               name_hash, local);
4933         if (ret) {
4934                 mlog_errno(ret);
4935                 goto out;
4936         }
4937
4938         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4939                 goto out;
4940
4941         /* allocate the space now for the outside block storage. */
4942         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4943                                                    value_len, ctxt);
4944         if (ret) {
4945                 mlog_errno(ret);
4946
4947                 if (xs->not_found) {
4948                         /*
4949                          * We can't allocate enough clusters for outside
4950                          * storage and we have allocated xattr already,
4951                          * so need to remove it.
4952                          */
4953                         ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
4954                 }
4955                 goto out;
4956         }
4957
4958 set_value_outside:
4959         ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
4960                                                    xs, val, value_len);
4961 out:
4962         return ret;
4963 }
4964
4965 /*
4966  * check whether the xattr bucket is filled up with the same hash value.
4967  * If we want to insert the xattr with the same hash, return -ENOSPC.
4968  * If we want to insert a xattr with different hash value, go ahead
4969  * and ocfs2_divide_xattr_bucket will handle this.
4970  */
4971 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4972                                               struct ocfs2_xattr_bucket *bucket,
4973                                               const char *name)
4974 {
4975         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4976         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
4977
4978         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
4979                 return 0;
4980
4981         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4982             xh->xh_entries[0].xe_name_hash) {
4983                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4984                      "hash = %u\n",
4985                      (unsigned long long)bucket_blkno(bucket),
4986                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4987                 return -ENOSPC;
4988         }
4989
4990         return 0;
4991 }
4992
4993 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4994                                              struct ocfs2_xattr_info *xi,
4995                                              struct ocfs2_xattr_search *xs,
4996                                              struct ocfs2_xattr_set_ctxt *ctxt)
4997 {
4998         struct ocfs2_xattr_header *xh;
4999         struct ocfs2_xattr_entry *xe;
5000         u16 count, header_size, xh_free_start;
5001         int free, max_free, need, old;
5002         size_t value_size = 0, name_len = strlen(xi->name);
5003         size_t blocksize = inode->i_sb->s_blocksize;
5004         int ret, allocation = 0;
5005
5006         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5007
5008 try_again:
5009         xh = xs->header;
5010         count = le16_to_cpu(xh->xh_count);
5011         xh_free_start = le16_to_cpu(xh->xh_free_start);
5012         header_size = sizeof(struct ocfs2_xattr_header) +
5013                         count * sizeof(struct ocfs2_xattr_entry);
5014         max_free = OCFS2_XATTR_BUCKET_SIZE -
5015                 le16_to_cpu(xh->xh_name_value_len) - header_size;
5016
5017         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5018                         "of %u which exceed block size\n",
5019                         (unsigned long long)bucket_blkno(xs->bucket),
5020                         header_size);
5021
5022         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5023                 value_size = OCFS2_XATTR_ROOT_SIZE;
5024         else if (xi->value)
5025                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5026
5027         if (xs->not_found)
5028                 need = sizeof(struct ocfs2_xattr_entry) +
5029                         OCFS2_XATTR_SIZE(name_len) + value_size;
5030         else {
5031                 need = value_size + OCFS2_XATTR_SIZE(name_len);
5032
5033                 /*
5034                  * We only replace the old value if the new length is smaller
5035                  * than the old one. Otherwise we will allocate new space in the
5036                  * bucket to store it.
5037                  */
5038                 xe = xs->here;
5039                 if (ocfs2_xattr_is_local(xe))
5040                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5041                 else
5042                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5043
5044                 if (old >= value_size)
5045                         need = 0;
5046         }
5047
5048         free = xh_free_start - header_size;
5049         /*
5050          * We need to make sure the new name/value pair
5051          * can exist in the same block.
5052          */
5053         if (xh_free_start % blocksize < need)
5054                 free -= xh_free_start % blocksize;
5055
5056         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5057              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5058              " %u\n", xs->not_found,
5059              (unsigned long long)bucket_blkno(xs->bucket),
5060              free, need, max_free, le16_to_cpu(xh->xh_free_start),
5061              le16_to_cpu(xh->xh_name_value_len));
5062
5063         if (free < need ||
5064             (xs->not_found &&
5065              count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5066                 if (need <= max_free &&
5067                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5068                         /*
5069                          * We can create the space by defragment. Since only the
5070                          * name/value will be moved, the xe shouldn't be changed
5071                          * in xs.
5072                          */
5073                         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5074                                                         xs->bucket);
5075                         if (ret) {
5076                                 mlog_errno(ret);
5077                                 goto out;
5078                         }
5079
5080                         xh_free_start = le16_to_cpu(xh->xh_free_start);
5081                         free = xh_free_start - header_size;
5082                         if (xh_free_start % blocksize < need)
5083                                 free -= xh_free_start % blocksize;
5084
5085                         if (free >= need)
5086                                 goto xattr_set;
5087
5088                         mlog(0, "Can't get enough space for xattr insert by "
5089                              "defragment. Need %u bytes, but we have %d, so "
5090                              "allocate new bucket for it.\n", need, free);
5091                 }
5092
5093                 /*
5094                  * We have to add new buckets or clusters and one
5095                  * allocation should leave us enough space for insert.
5096                  */
5097                 BUG_ON(allocation);
5098
5099                 /*
5100                  * We do not allow for overlapping ranges between buckets. And
5101                  * the maximum number of collisions we will allow for then is
5102                  * one bucket's worth, so check it here whether we need to
5103                  * add a new bucket for the insert.
5104                  */
5105                 ret = ocfs2_check_xattr_bucket_collision(inode,
5106                                                          xs->bucket,
5107                                                          xi->name);
5108                 if (ret) {
5109                         mlog_errno(ret);
5110                         goto out;
5111                 }
5112
5113                 ret = ocfs2_add_new_xattr_bucket(inode,
5114                                                  xs->xattr_bh,
5115                                                  xs->bucket,
5116                                                  ctxt);
5117                 if (ret) {
5118                         mlog_errno(ret);
5119                         goto out;
5120                 }
5121
5122                 /*
5123                  * ocfs2_add_new_xattr_bucket() will have updated
5124                  * xs->bucket if it moved, but it will not have updated
5125                  * any of the other search fields.  Thus, we drop it and
5126                  * re-search.  Everything should be cached, so it'll be
5127                  * quick.
5128                  */
5129                 ocfs2_xattr_bucket_relse(xs->bucket);
5130                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5131                                                    xi->name_index,
5132                                                    xi->name, xs);
5133                 if (ret && ret != -ENODATA)
5134                         goto out;
5135                 xs->not_found = ret;
5136                 allocation = 1;
5137                 goto try_again;
5138         }
5139
5140 xattr_set:
5141         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5142 out:
5143         mlog_exit(ret);
5144         return ret;
5145 }
5146
5147 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5148                                         struct ocfs2_xattr_bucket *bucket,
5149                                         void *para)
5150 {
5151         int ret = 0;
5152         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5153         u16 i;
5154         struct ocfs2_xattr_entry *xe;
5155         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5156         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5157         int credits = ocfs2_remove_extent_credits(osb->sb) +
5158                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5159
5160
5161         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5162
5163         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5164                 xe = &xh->xh_entries[i];
5165                 if (ocfs2_xattr_is_local(xe))
5166                         continue;
5167
5168                 ctxt.handle = ocfs2_start_trans(osb, credits);
5169                 if (IS_ERR(ctxt.handle)) {
5170                         ret = PTR_ERR(ctxt.handle);
5171                         mlog_errno(ret);
5172                         break;
5173                 }
5174
5175                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5176                                                         i, 0, &ctxt);
5177
5178                 ocfs2_commit_trans(osb, ctxt.handle);
5179                 if (ret) {
5180                         mlog_errno(ret);
5181                         break;
5182                 }
5183         }
5184
5185         ocfs2_schedule_truncate_log_flush(osb, 1);
5186         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5187         return ret;
5188 }
5189
5190 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5191                                           struct buffer_head *xb_bh)
5192 {
5193         struct ocfs2_xattr_block *xb =
5194                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5195         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5196         int ret = 0;
5197         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5198         u64 p_blkno;
5199
5200         if (le16_to_cpu(el->l_next_free_rec) == 0)
5201                 return 0;
5202
5203         while (name_hash > 0) {
5204                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5205                                           &e_cpos, &num_clusters, el);
5206                 if (ret) {
5207                         mlog_errno(ret);
5208                         goto out;
5209                 }
5210
5211                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5212                                                   ocfs2_delete_xattr_in_bucket,
5213                                                   NULL);
5214                 if (ret) {
5215                         mlog_errno(ret);
5216                         goto out;
5217                 }
5218
5219                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5220                                              p_blkno, e_cpos, num_clusters);
5221                 if (ret) {
5222                         mlog_errno(ret);
5223                         break;
5224                 }
5225
5226                 if (e_cpos == 0)
5227                         break;
5228
5229                 name_hash = e_cpos - 1;
5230         }
5231
5232 out:
5233         return ret;
5234 }
5235
5236 /*
5237  * 'security' attributes support
5238  */
5239 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5240                                         size_t list_size, const char *name,
5241                                         size_t name_len)
5242 {
5243         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5244         const size_t total_len = prefix_len + name_len + 1;
5245
5246         if (list && total_len <= list_size) {
5247                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5248                 memcpy(list + prefix_len, name, name_len);
5249                 list[prefix_len + name_len] = '\0';
5250         }
5251         return total_len;
5252 }
5253
5254 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5255                                     void *buffer, size_t size)
5256 {
5257         if (strcmp(name, "") == 0)
5258                 return -EINVAL;
5259         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5260                                buffer, size);
5261 }
5262
5263 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5264                                     const void *value, size_t size, int flags)
5265 {
5266         if (strcmp(name, "") == 0)
5267                 return -EINVAL;
5268
5269         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5270                                size, flags);
5271 }
5272
5273 int ocfs2_init_security_get(struct inode *inode,
5274                             struct inode *dir,
5275                             struct ocfs2_security_xattr_info *si)
5276 {
5277         return security_inode_init_security(inode, dir, &si->name, &si->value,
5278                                             &si->value_len);
5279 }
5280
5281 int ocfs2_init_security_set(handle_t *handle,
5282                             struct inode *inode,
5283                             struct buffer_head *di_bh,
5284                             struct ocfs2_security_xattr_info *si,
5285                             struct ocfs2_alloc_context *xattr_ac,
5286                             struct ocfs2_alloc_context *data_ac)
5287 {
5288         return ocfs2_xattr_set_handle(handle, inode, di_bh,
5289                                      OCFS2_XATTR_INDEX_SECURITY,
5290                                      si->name, si->value, si->value_len, 0,
5291                                      xattr_ac, data_ac);
5292 }
5293
5294 struct xattr_handler ocfs2_xattr_security_handler = {
5295         .prefix = XATTR_SECURITY_PREFIX,
5296         .list   = ocfs2_xattr_security_list,
5297         .get    = ocfs2_xattr_security_get,
5298         .set    = ocfs2_xattr_security_set,
5299 };
5300
5301 /*
5302  * 'trusted' attributes support
5303  */
5304 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5305                                        size_t list_size, const char *name,
5306                                        size_t name_len)
5307 {
5308         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5309         const size_t total_len = prefix_len + name_len + 1;
5310
5311         if (list && total_len <= list_size) {
5312                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5313                 memcpy(list + prefix_len, name, name_len);
5314                 list[prefix_len + name_len] = '\0';
5315         }
5316         return total_len;
5317 }
5318
5319 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5320                                    void *buffer, size_t size)
5321 {
5322         if (strcmp(name, "") == 0)
5323                 return -EINVAL;
5324         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5325                                buffer, size);
5326 }
5327
5328 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5329                                    const void *value, size_t size, int flags)
5330 {
5331         if (strcmp(name, "") == 0)
5332                 return -EINVAL;
5333
5334         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5335                                size, flags);
5336 }
5337
5338 struct xattr_handler ocfs2_xattr_trusted_handler = {
5339         .prefix = XATTR_TRUSTED_PREFIX,
5340         .list   = ocfs2_xattr_trusted_list,
5341         .get    = ocfs2_xattr_trusted_get,
5342         .set    = ocfs2_xattr_trusted_set,
5343 };
5344
5345 /*
5346  * 'user' attributes support
5347  */
5348 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5349                                     size_t list_size, const char *name,
5350                                     size_t name_len)
5351 {
5352         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5353         const size_t total_len = prefix_len + name_len + 1;
5354         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5355
5356         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5357                 return 0;
5358
5359         if (list && total_len <= list_size) {
5360                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5361                 memcpy(list + prefix_len, name, name_len);
5362                 list[prefix_len + name_len] = '\0';
5363         }
5364         return total_len;
5365 }
5366
5367 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5368                                 void *buffer, size_t size)
5369 {
5370         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5371
5372         if (strcmp(name, "") == 0)
5373                 return -EINVAL;
5374         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5375                 return -EOPNOTSUPP;
5376         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5377                                buffer, size);
5378 }
5379
5380 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5381                                 const void *value, size_t size, int flags)
5382 {
5383         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5384
5385         if (strcmp(name, "") == 0)
5386                 return -EINVAL;
5387         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5388                 return -EOPNOTSUPP;
5389
5390         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5391                                size, flags);
5392 }
5393
5394 struct xattr_handler ocfs2_xattr_user_handler = {
5395         .prefix = XATTR_USER_PREFIX,
5396         .list   = ocfs2_xattr_user_list,
5397         .get    = ocfs2_xattr_user_get,
5398         .set    = ocfs2_xattr_user_set,
5399 };