fs/ocfs2/xattr.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * xattr.c
   5  *
   6  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
   7  *
   8  * CREDITS:
   9  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
  10  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  11  *
  12  * This program is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU General Public
  14  * License version 2 as published by the Free Software Foundation.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * General Public License for more details.
  20  */
  21
  22 #include <linux/capability.h>
  23 #include <linux/fs.h>
  24 #include <linux/types.h>
  25 #include <linux/slab.h>
  26 #include <linux/highmem.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/uio.h>
  29 #include <linux/sched.h>
  30 #include <linux/splice.h>
  31 #include <linux/mount.h>
  32 #include <linux/writeback.h>
  33 #include <linux/falloc.h>
  34 #include <linux/sort.h>
  35 #include <linux/init.h>
  36 #include <linux/module.h>
  37 #include <linux/string.h>
  38 #include <linux/security.h>
  39
  40 #define MLOG_MASK_PREFIX ML_XATTR
  41 #include <cluster/masklog.h>
  42
  43 #include "ocfs2.h"
  44 #include "alloc.h"
  45 #include "blockcheck.h"
  46 #include "dlmglue.h"
  47 #include "file.h"
  48 #include "symlink.h"
  49 #include "sysfile.h"
  50 #include "inode.h"
  51 #include "journal.h"
  52 #include "ocfs2_fs.h"
  53 #include "suballoc.h"
  54 #include "uptodate.h"
  55 #include "buffer_head_io.h"
  56 #include "super.h"
  57 #include "xattr.h"
  58
  59
  60 struct ocfs2_xattr_def_value_root {
  61         struct ocfs2_xattr_value_root   xv;
  62         struct ocfs2_extent_rec         er;
  63 };
  64
  65 struct ocfs2_xattr_bucket {
  66         /* The inode these xattrs are associated with */
  67         struct inode *bu_inode;
  68
  69         /* The actual buffers that make up the bucket */
  70         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  71
  72         /* How many blocks make up one bucket for this filesystem */
  73         int bu_blocks;
  74 };
  75
  76 struct ocfs2_xattr_set_ctxt {
  77         handle_t *handle;
  78         struct ocfs2_alloc_context *meta_ac;
  79         struct ocfs2_alloc_context *data_ac;
  80         struct ocfs2_cached_dealloc_ctxt dealloc;
  81 };
  82
  83 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  84 #define OCFS2_XATTR_INLINE_SIZE 80
  85 #define OCFS2_XATTR_HEADER_GAP  4
  86 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
  87                                          - sizeof(struct ocfs2_xattr_header) \
  88                                          - OCFS2_XATTR_HEADER_GAP)
  89 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
  90                                          - sizeof(struct ocfs2_xattr_block) \
  91                                          - sizeof(struct ocfs2_xattr_header) \
  92                                          - OCFS2_XATTR_HEADER_GAP)
  93
  94 static struct ocfs2_xattr_def_value_root def_xv = {
  95         .xv.xr_list.l_count = cpu_to_le16(1),
  96 };
  97
  98 struct xattr_handler *ocfs2_xattr_handlers[] = {
  99         &ocfs2_xattr_user_handler,
 100 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
 101         &ocfs2_xattr_acl_access_handler,
 102         &ocfs2_xattr_acl_default_handler,
 103 #endif
 104         &ocfs2_xattr_trusted_handler,
 105         &ocfs2_xattr_security_handler,
 106         NULL
 107 };
 108
 109 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 110         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
 111 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
 112         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
 113                                         = &ocfs2_xattr_acl_access_handler,
 114         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
 115                                         = &ocfs2_xattr_acl_default_handler,
 116 #endif
 117         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
 118         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 119 };
 120
 121 struct ocfs2_xattr_info {
 122         int name_index;
 123         const char *name;
 124         const void *value;
 125         size_t value_len;
 126 };
 127
 128 struct ocfs2_xattr_search {
 129         struct buffer_head *inode_bh;
 130         /*
 131          * xattr_bh point to the block buffer head which has extended attribute
 132          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 133          */
 134         struct buffer_head *xattr_bh;
 135         struct ocfs2_xattr_header *header;
 136         struct ocfs2_xattr_bucket *bucket;
 137         void *base;
 138         void *end;
 139         struct ocfs2_xattr_entry *here;
 140         int not_found;
 141 };
 142
 143 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
 144                                              struct ocfs2_xattr_header *xh,
 145                                              int index,
 146                                              int *block_off,
 147                                              int *new_offset);
 148
 149 static int ocfs2_xattr_block_find(struct inode *inode,
 150                                   int name_index,
 151                                   const char *name,
 152                                   struct ocfs2_xattr_search *xs);
 153 static int ocfs2_xattr_index_block_find(struct inode *inode,
 154                                         struct buffer_head *root_bh,
 155                                         int name_index,
 156                                         const char *name,
 157                                         struct ocfs2_xattr_search *xs);
 158
 159 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 160                                         struct ocfs2_xattr_tree_root *xt,
 161                                         char *buffer,
 162                                         size_t buffer_size);
 163
 164 static int ocfs2_xattr_create_index_block(struct inode *inode,
 165                                           struct ocfs2_xattr_search *xs,
 166                                           struct ocfs2_xattr_set_ctxt *ctxt);
 167
 168 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 169                                              struct ocfs2_xattr_info *xi,
 170                                              struct ocfs2_xattr_search *xs,
 171                                              struct ocfs2_xattr_set_ctxt *ctxt);
 172
 173 static int ocfs2_delete_xattr_index_block(struct inode *inode,
 174                                           struct buffer_head *xb_bh);
 175 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 176                                   u64 src_blk, u64 last_blk, u64 to_blk,
 177                                   unsigned int start_bucket,
 178                                   u32 *first_hash);
 179
 180 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 181 {
 182         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 183 }
 184
 185 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 186 {
 187         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 188 }
 189
 190 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
 191 {
 192         u16 len = sb->s_blocksize -
 193                  offsetof(struct ocfs2_xattr_header, xh_entries);
 194
 195         return len / sizeof(struct ocfs2_xattr_entry);
 196 }
 197
 198 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 199 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 200 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 201
 202 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 203 {
 204         struct ocfs2_xattr_bucket *bucket;
 205         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 206
 207         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 208
 209         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 210         if (bucket) {
 211                 bucket->bu_inode = inode;
 212                 bucket->bu_blocks = blks;
 213         }
 214
 215         return bucket;
 216 }
 217
 218 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 219 {
 220         int i;
 221
 222         for (i = 0; i < bucket->bu_blocks; i++) {
 223                 brelse(bucket->bu_bhs[i]);
 224                 bucket->bu_bhs[i] = NULL;
 225         }
 226 }
 227
 228 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 229 {
 230         if (bucket) {
 231                 ocfs2_xattr_bucket_relse(bucket);
 232                 bucket->bu_inode = NULL;
 233                 kfree(bucket);
 234         }
 235 }
 236
 237 /*
 238  * A bucket that has never been written to disk doesn't need to be
 239  * read.  We just need the buffer_heads.  Don't call this for
 240  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 241  * them fully.
 242  */
 243 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 244                                    u64 xb_blkno)
 245 {
 246         int i, rc = 0;
 247
 248         for (i = 0; i < bucket->bu_blocks; i++) {
 249                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 250                                               xb_blkno + i);
 251                 if (!bucket->bu_bhs[i]) {
 252                         rc = -EIO;
 253                         mlog_errno(rc);
 254                         break;
 255                 }
 256
 257                 if (!ocfs2_buffer_uptodate(bucket->bu_inode,
 258                                            bucket->bu_bhs[i]))
 259                         ocfs2_set_new_buffer_uptodate(bucket->bu_inode,
 260                                                       bucket->bu_bhs[i]);
 261         }
 262
 263         if (rc)
 264                 ocfs2_xattr_bucket_relse(bucket);
 265         return rc;
 266 }
 267
 268 /* Read the xattr bucket at xb_blkno */
 269 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 270                                    u64 xb_blkno)
 271 {
 272         int rc;
 273
 274         rc = ocfs2_read_blocks(bucket->bu_inode, xb_blkno,
 275                                bucket->bu_blocks, bucket->bu_bhs, 0,
 276                                NULL);
 277         if (!rc) {
 278                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 279                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 280                                                  bucket->bu_bhs,
 281                                                  bucket->bu_blocks,
 282                                                  &bucket_xh(bucket)->xh_check);
 283                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 284                 if (rc)
 285                         mlog_errno(rc);
 286         }
 287
 288         if (rc)
 289                 ocfs2_xattr_bucket_relse(bucket);
 290         return rc;
 291 }
 292
 293 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 294                                              struct ocfs2_xattr_bucket *bucket,
 295                                              int type)
 296 {
 297         int i, rc = 0;
 298
 299         for (i = 0; i < bucket->bu_blocks; i++) {
 300                 rc = ocfs2_journal_access(handle, bucket->bu_inode,
 301                                           bucket->bu_bhs[i], type);
 302                 if (rc) {
 303                         mlog_errno(rc);
 304                         break;
 305                 }
 306         }
 307
 308         return rc;
 309 }
 310
 311 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 312                                              struct ocfs2_xattr_bucket *bucket)
 313 {
 314         int i;
 315
 316         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 317         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 318                                    bucket->bu_bhs, bucket->bu_blocks,
 319                                    &bucket_xh(bucket)->xh_check);
 320         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 321
 322         for (i = 0; i < bucket->bu_blocks; i++)
 323                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 324 }
 325
 326 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 327                                          struct ocfs2_xattr_bucket *src)
 328 {
 329         int i;
 330         int blocksize = src->bu_inode->i_sb->s_blocksize;
 331
 332         BUG_ON(dest->bu_blocks != src->bu_blocks);
 333         BUG_ON(dest->bu_inode != src->bu_inode);
 334
 335         for (i = 0; i < src->bu_blocks; i++) {
 336                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 337                        blocksize);
 338         }
 339 }
 340
 341 static int ocfs2_validate_xattr_block(struct super_block *sb,
 342                                       struct buffer_head *bh)
 343 {
 344         int rc;
 345         struct ocfs2_xattr_block *xb =
 346                 (struct ocfs2_xattr_block *)bh->b_data;
 347
 348         mlog(0, "Validating xattr block %llu\n",
 349              (unsigned long long)bh->b_blocknr);
 350
 351         BUG_ON(!buffer_uptodate(bh));
 352
 353         /*
 354          * If the ecc fails, we return the error but otherwise
 355          * leave the filesystem running.  We know any error is
 356          * local to this block.
 357          */
 358         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 359         if (rc)
 360                 return rc;
 361
 362         /*
 363          * Errors after here are fatal
 364          */
 365
 366         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 367                 ocfs2_error(sb,
 368                             "Extended attribute block #%llu has bad "
 369                             "signature %.*s",
 370                             (unsigned long long)bh->b_blocknr, 7,
 371                             xb->xb_signature);
 372                 return -EINVAL;
 373         }
 374
 375         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 376                 ocfs2_error(sb,
 377                             "Extended attribute block #%llu has an "
 378                             "invalid xb_blkno of %llu",
 379                             (unsigned long long)bh->b_blocknr,
 380                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
 381                 return -EINVAL;
 382         }
 383
 384         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 385                 ocfs2_error(sb,
 386                             "Extended attribute block #%llu has an invalid "
 387                             "xb_fs_generation of #%u",
 388                             (unsigned long long)bh->b_blocknr,
 389                             le32_to_cpu(xb->xb_fs_generation));
 390                 return -EINVAL;
 391         }
 392
 393         return 0;
 394 }
 395
 396 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 397                                   struct buffer_head **bh)
 398 {
 399         int rc;
 400         struct buffer_head *tmp = *bh;
 401
 402         rc = ocfs2_read_block(inode, xb_blkno, &tmp,
 403                               ocfs2_validate_xattr_block);
 404
 405         /* If ocfs2_read_block() got us a new bh, pass it up. */
 406         if (!rc && !*bh)
 407                 *bh = tmp;
 408
 409         return rc;
 410 }
 411
 412 static inline const char *ocfs2_xattr_prefix(int name_index)
 413 {
 414         struct xattr_handler *handler = NULL;
 415
 416         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 417                 handler = ocfs2_xattr_handler_map[name_index];
 418
 419         return handler ? handler->prefix : NULL;
 420 }
 421
 422 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 423                                  const char *name,
 424                                  int name_len)
 425 {
 426         /* Get hash value of uuid from super block */
 427         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 428         int i;
 429
 430         /* hash extended attribute name */
 431         for (i = 0; i < name_len; i++) {
 432                 hash = (hash << OCFS2_HASH_SHIFT) ^
 433                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 434                        *name++;
 435         }
 436
 437         return hash;
 438 }
 439
 440 /*
 441  * ocfs2_xattr_hash_entry()
 442  *
 443  * Compute the hash of an extended attribute.
 444  */
 445 static void ocfs2_xattr_hash_entry(struct inode *inode,
 446                                    struct ocfs2_xattr_header *header,
 447                                    struct ocfs2_xattr_entry *entry)
 448 {
 449         u32 hash = 0;
 450         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
 451
 452         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
 453         entry->xe_name_hash = cpu_to_le32(hash);
 454
 455         return;
 456 }
 457
 458 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 459 {
 460         int size = 0;
 461
 462         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
 463                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 464         else
 465                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 466         size += sizeof(struct ocfs2_xattr_entry);
 467
 468         return size;
 469 }
 470
 471 int ocfs2_calc_security_init(struct inode *dir,
 472                              struct ocfs2_security_xattr_info *si,
 473                              int *want_clusters,
 474                              int *xattr_credits,
 475                              struct ocfs2_alloc_context **xattr_ac)
 476 {
 477         int ret = 0;
 478         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 479         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 480                                                  si->value_len);
 481
 482         /*
 483          * The max space of security xattr taken inline is
 484          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 485          * So reserve one metadata block for it is ok.
 486          */
 487         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 488             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 489                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 490                 if (ret) {
 491                         mlog_errno(ret);
 492                         return ret;
 493                 }
 494                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 495         }
 496
 497         /* reserve clusters for xattr value which will be set in B tree*/
 498         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 499                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 500                                                             si->value_len);
 501
 502                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 503                                                            new_clusters);
 504                 *want_clusters += new_clusters;
 505         }
 506         return ret;
 507 }
 508
 509 int ocfs2_calc_xattr_init(struct inode *dir,
 510                           struct buffer_head *dir_bh,
 511                           int mode,
 512                           struct ocfs2_security_xattr_info *si,
 513                           int *want_clusters,
 514                           int *xattr_credits,
 515                           int *want_meta)
 516 {
 517         int ret = 0;
 518         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 519         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 520
 521         if (si->enable)
 522                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 523                                                      si->value_len);
 524
 525         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 526                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 527                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 528                                         "", NULL, 0);
 529                 if (acl_len > 0) {
 530                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 531                         if (S_ISDIR(mode))
 532                                 a_size <<= 1;
 533                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 534                         mlog_errno(ret);
 535                         return ret;
 536                 }
 537         }
 538
 539         if (!(s_size + a_size))
 540                 return ret;
 541
 542         /*
 543          * The max space of security xattr taken inline is
 544          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 545          * The max space of acl xattr taken inline is
 546          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 547          * when blocksize = 512, may reserve one more cluser for
 548          * xattr bucket, otherwise reserve one metadata block
 549          * for them is ok.
 550          * If this is a new directory with inline data,
 551          * we choose to reserve the entire inline area for
 552          * directory contents and force an external xattr block.
 553          */
 554         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 555             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 556             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 557                 *want_meta = *want_meta + 1;
 558                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 559         }
 560
 561         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 562             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 563                 *want_clusters += 1;
 564                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 565         }
 566
 567         /*
 568          * reserve credits and clusters for xattrs which has large value
 569          * and have to be set outside
 570          */
 571         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 572                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 573                                                         si->value_len);
 574                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 575                                                            new_clusters);
 576                 *want_clusters += new_clusters;
 577         }
 578         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 579             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 580                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 581                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 582                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 583                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 584                                                            new_clusters);
 585                 *want_clusters += new_clusters;
 586         }
 587
 588         return ret;
 589 }
 590
 591 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 592                                          u32 clusters_to_add,
 593                                          struct ocfs2_xattr_value_buf *vb,
 594                                          struct ocfs2_xattr_set_ctxt *ctxt)
 595 {
 596         int status = 0;
 597         handle_t *handle = ctxt->handle;
 598         enum ocfs2_alloc_restarted why;
 599         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 600         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 601         struct ocfs2_extent_tree et;
 602
 603         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
 604
 605         ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
 606
 607         status = vb->vb_access(handle, inode, vb->vb_bh,
 608                               OCFS2_JOURNAL_ACCESS_WRITE);
 609         if (status < 0) {
 610                 mlog_errno(status);
 611                 goto leave;
 612         }
 613
 614         prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 615         status = ocfs2_add_clusters_in_btree(osb,
 616                                              inode,
 617                                              &logical_start,
 618                                              clusters_to_add,
 619                                              0,
 620                                              &et,
 621                                              handle,
 622                                              ctxt->data_ac,
 623                                              ctxt->meta_ac,
 624                                              &why);
 625         if (status < 0) {
 626                 mlog_errno(status);
 627                 goto leave;
 628         }
 629
 630         status = ocfs2_journal_dirty(handle, vb->vb_bh);
 631         if (status < 0) {
 632                 mlog_errno(status);
 633                 goto leave;
 634         }
 635
 636         clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
 637
 638         /*
 639          * We should have already allocated enough space before the transaction,
 640          * so no need to restart.
 641          */
 642         BUG_ON(why != RESTART_NONE || clusters_to_add);
 643
 644 leave:
 645
 646         return status;
 647 }
 648
 649 static int __ocfs2_remove_xattr_range(struct inode *inode,
 650                                       struct ocfs2_xattr_value_buf *vb,
 651                                       u32 cpos, u32 phys_cpos, u32 len,
 652                                       struct ocfs2_xattr_set_ctxt *ctxt)
 653 {
 654         int ret;
 655         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 656         handle_t *handle = ctxt->handle;
 657         struct ocfs2_extent_tree et;
 658
 659         ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
 660
 661         ret = vb->vb_access(handle, inode, vb->vb_bh,
 662                             OCFS2_JOURNAL_ACCESS_WRITE);
 663         if (ret) {
 664                 mlog_errno(ret);
 665                 goto out;
 666         }
 667
 668         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
 669                                   &ctxt->dealloc);
 670         if (ret) {
 671                 mlog_errno(ret);
 672                 goto out;
 673         }
 674
 675         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 676
 677         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
 678         if (ret) {
 679                 mlog_errno(ret);
 680                 goto out;
 681         }
 682
 683         ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
 684         if (ret)
 685                 mlog_errno(ret);
 686
 687 out:
 688         return ret;
 689 }
 690
 691 static int ocfs2_xattr_shrink_size(struct inode *inode,
 692                                    u32 old_clusters,
 693                                    u32 new_clusters,
 694                                    struct ocfs2_xattr_value_buf *vb,
 695                                    struct ocfs2_xattr_set_ctxt *ctxt)
 696 {
 697         int ret = 0;
 698         u32 trunc_len, cpos, phys_cpos, alloc_size;
 699         u64 block;
 700
 701         if (old_clusters <= new_clusters)
 702                 return 0;
 703
 704         cpos = new_clusters;
 705         trunc_len = old_clusters - new_clusters;
 706         while (trunc_len) {
 707                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 708                                                &alloc_size,
 709                                                &vb->vb_xv->xr_list);
 710                 if (ret) {
 711                         mlog_errno(ret);
 712                         goto out;
 713                 }
 714
 715                 if (alloc_size > trunc_len)
 716                         alloc_size = trunc_len;
 717
 718                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 719                                                  phys_cpos, alloc_size,
 720                                                  ctxt);
 721                 if (ret) {
 722                         mlog_errno(ret);
 723                         goto out;
 724                 }
 725
 726                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 727                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
 728                                                        alloc_size);
 729                 cpos += alloc_size;
 730                 trunc_len -= alloc_size;
 731         }
 732
 733 out:
 734         return ret;
 735 }
 736
 737 static int ocfs2_xattr_value_truncate(struct inode *inode,
 738                                       struct ocfs2_xattr_value_buf *vb,
 739                                       int len,
 740                                       struct ocfs2_xattr_set_ctxt *ctxt)
 741 {
 742         int ret;
 743         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 744         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 745
 746         if (new_clusters == old_clusters)
 747                 return 0;
 748
 749         if (new_clusters > old_clusters)
 750                 ret = ocfs2_xattr_extend_allocation(inode,
 751                                                     new_clusters - old_clusters,
 752                                                     vb, ctxt);
 753         else
 754                 ret = ocfs2_xattr_shrink_size(inode,
 755                                               old_clusters, new_clusters,
 756                                               vb, ctxt);
 757
 758         return ret;
 759 }
 760
 761 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 762                                   size_t *result, const char *prefix,
 763                                   const char *name, int name_len)
 764 {
 765         char *p = buffer + *result;
 766         int prefix_len = strlen(prefix);
 767         int total_len = prefix_len + name_len + 1;
 768
 769         *result += total_len;
 770
 771         /* we are just looking for how big our buffer needs to be */
 772         if (!size)
 773                 return 0;
 774
 775         if (*result > size)
 776                 return -ERANGE;
 777
 778         memcpy(p, prefix, prefix_len);
 779         memcpy(p + prefix_len, name, name_len);
 780         p[prefix_len + name_len] = '\0';
 781
 782         return 0;
 783 }
 784
 785 static int ocfs2_xattr_list_entries(struct inode *inode,
 786                                     struct ocfs2_xattr_header *header,
 787                                     char *buffer, size_t buffer_size)
 788 {
 789         size_t result = 0;
 790         int i, type, ret;
 791         const char *prefix, *name;
 792
 793         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 794                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 795                 type = ocfs2_xattr_get_type(entry);
 796                 prefix = ocfs2_xattr_prefix(type);
 797
 798                 if (prefix) {
 799                         name = (const char *)header +
 800                                 le16_to_cpu(entry->xe_name_offset);
 801
 802                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 803                                                      &result, prefix, name,
 804                                                      entry->xe_name_len);
 805                         if (ret)
 806                                 return ret;
 807                 }
 808         }
 809
 810         return result;
 811 }
 812
 813 static int ocfs2_xattr_ibody_list(struct inode *inode,
 814                                   struct ocfs2_dinode *di,
 815                                   char *buffer,
 816                                   size_t buffer_size)
 817 {
 818         struct ocfs2_xattr_header *header = NULL;
 819         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 820         int ret = 0;
 821
 822         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 823                 return ret;
 824
 825         header = (struct ocfs2_xattr_header *)
 826                  ((void *)di + inode->i_sb->s_blocksize -
 827                  le16_to_cpu(di->i_xattr_inline_size));
 828
 829         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 830
 831         return ret;
 832 }
 833
 834 static int ocfs2_xattr_block_list(struct inode *inode,
 835                                   struct ocfs2_dinode *di,
 836                                   char *buffer,
 837                                   size_t buffer_size)
 838 {
 839         struct buffer_head *blk_bh = NULL;
 840         struct ocfs2_xattr_block *xb;
 841         int ret = 0;
 842
 843         if (!di->i_xattr_loc)
 844                 return ret;
 845
 846         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
 847                                      &blk_bh);
 848         if (ret < 0) {
 849                 mlog_errno(ret);
 850                 return ret;
 851         }
 852
 853         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 854         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 855                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 856                 ret = ocfs2_xattr_list_entries(inode, header,
 857                                                buffer, buffer_size);
 858         } else {
 859                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
 860                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
 861                                                    buffer, buffer_size);
 862         }
 863
 864         brelse(blk_bh);
 865
 866         return ret;
 867 }
 868
 869 ssize_t ocfs2_listxattr(struct dentry *dentry,
 870                         char *buffer,
 871                         size_t size)
 872 {
 873         int ret = 0, i_ret = 0, b_ret = 0;
 874         struct buffer_head *di_bh = NULL;
 875         struct ocfs2_dinode *di = NULL;
 876         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
 877
 878         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
 879                 return -EOPNOTSUPP;
 880
 881         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 882                 return ret;
 883
 884         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
 885         if (ret < 0) {
 886                 mlog_errno(ret);
 887                 return ret;
 888         }
 889
 890         di = (struct ocfs2_dinode *)di_bh->b_data;
 891
 892         down_read(&oi->ip_xattr_sem);
 893         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
 894         if (i_ret < 0)
 895                 b_ret = 0;
 896         else {
 897                 if (buffer) {
 898                         buffer += i_ret;
 899                         size -= i_ret;
 900                 }
 901                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
 902                                                buffer, size);
 903                 if (b_ret < 0)
 904                         i_ret = 0;
 905         }
 906         up_read(&oi->ip_xattr_sem);
 907         ocfs2_inode_unlock(dentry->d_inode, 0);
 908
 909         brelse(di_bh);
 910
 911         return i_ret + b_ret;
 912 }
 913
 914 static int ocfs2_xattr_find_entry(int name_index,
 915                                   const char *name,
 916                                   struct ocfs2_xattr_search *xs)
 917 {
 918         struct ocfs2_xattr_entry *entry;
 919         size_t name_len;
 920         int i, cmp = 1;
 921
 922         if (name == NULL)
 923                 return -EINVAL;
 924
 925         name_len = strlen(name);
 926         entry = xs->here;
 927         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
 928                 cmp = name_index - ocfs2_xattr_get_type(entry);
 929                 if (!cmp)
 930                         cmp = name_len - entry->xe_name_len;
 931                 if (!cmp)
 932                         cmp = memcmp(name, (xs->base +
 933                                      le16_to_cpu(entry->xe_name_offset)),
 934                                      name_len);
 935                 if (cmp == 0)
 936                         break;
 937                 entry += 1;
 938         }
 939         xs->here = entry;
 940
 941         return cmp ? -ENODATA : 0;
 942 }
 943
 944 static int ocfs2_xattr_get_value_outside(struct inode *inode,
 945                                          struct ocfs2_xattr_value_root *xv,
 946                                          void *buffer,
 947                                          size_t len)
 948 {
 949         u32 cpos, p_cluster, num_clusters, bpc, clusters;
 950         u64 blkno;
 951         int i, ret = 0;
 952         size_t cplen, blocksize;
 953         struct buffer_head *bh = NULL;
 954         struct ocfs2_extent_list *el;
 955
 956         el = &xv->xr_list;
 957         clusters = le32_to_cpu(xv->xr_clusters);
 958         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 959         blocksize = inode->i_sb->s_blocksize;
 960
 961         cpos = 0;
 962         while (cpos < clusters) {
 963                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 964                                                &num_clusters, el);
 965                 if (ret) {
 966                         mlog_errno(ret);
 967                         goto out;
 968                 }
 969
 970                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 971                 /* Copy ocfs2_xattr_value */
 972                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 973                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
 974                         if (ret) {
 975                                 mlog_errno(ret);
 976                                 goto out;
 977                         }
 978
 979                         cplen = len >= blocksize ? blocksize : len;
 980                         memcpy(buffer, bh->b_data, cplen);
 981                         len -= cplen;
 982                         buffer += cplen;
 983
 984                         brelse(bh);
 985                         bh = NULL;
 986                         if (len == 0)
 987                                 break;
 988                 }
 989                 cpos += num_clusters;
 990         }
 991 out:
 992         return ret;
 993 }
 994
 995 static int ocfs2_xattr_ibody_get(struct inode *inode,
 996                                  int name_index,
 997                                  const char *name,
 998                                  void *buffer,
 999                                  size_t buffer_size,
1000                                  struct ocfs2_xattr_search *xs)
1001 {
1002         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1003         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1004         struct ocfs2_xattr_value_root *xv;
1005         size_t size;
1006         int ret = 0;
1007
1008         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1009                 return -ENODATA;
1010
1011         xs->end = (void *)di + inode->i_sb->s_blocksize;
1012         xs->header = (struct ocfs2_xattr_header *)
1013                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1014         xs->base = (void *)xs->header;
1015         xs->here = xs->header->xh_entries;
1016
1017         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1018         if (ret)
1019                 return ret;
1020         size = le64_to_cpu(xs->here->xe_value_size);
1021         if (buffer) {
1022                 if (size > buffer_size)
1023                         return -ERANGE;
1024                 if (ocfs2_xattr_is_local(xs->here)) {
1025                         memcpy(buffer, (void *)xs->base +
1026                                le16_to_cpu(xs->here->xe_name_offset) +
1027                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1028                 } else {
1029                         xv = (struct ocfs2_xattr_value_root *)
1030                                 (xs->base + le16_to_cpu(
1031                                  xs->here->xe_name_offset) +
1032                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1033                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1034                                                             buffer, size);
1035                         if (ret < 0) {
1036                                 mlog_errno(ret);
1037                                 return ret;
1038                         }
1039                 }
1040         }
1041
1042         return size;
1043 }
1044
1045 static int ocfs2_xattr_block_get(struct inode *inode,
1046                                  int name_index,
1047                                  const char *name,
1048                                  void *buffer,
1049                                  size_t buffer_size,
1050                                  struct ocfs2_xattr_search *xs)
1051 {
1052         struct ocfs2_xattr_block *xb;
1053         struct ocfs2_xattr_value_root *xv;
1054         size_t size;
1055         int ret = -ENODATA, name_offset, name_len, i;
1056         int uninitialized_var(block_off);
1057
1058         xs->bucket = ocfs2_xattr_bucket_new(inode);
1059         if (!xs->bucket) {
1060                 ret = -ENOMEM;
1061                 mlog_errno(ret);
1062                 goto cleanup;
1063         }
1064
1065         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1066         if (ret) {
1067                 mlog_errno(ret);
1068                 goto cleanup;
1069         }
1070
1071         if (xs->not_found) {
1072                 ret = -ENODATA;
1073                 goto cleanup;
1074         }
1075
1076         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1077         size = le64_to_cpu(xs->here->xe_value_size);
1078         if (buffer) {
1079                 ret = -ERANGE;
1080                 if (size > buffer_size)
1081                         goto cleanup;
1082
1083                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1084                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1085                 i = xs->here - xs->header->xh_entries;
1086
1087                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1088                         ret = ocfs2_xattr_bucket_get_name_value(inode,
1089                                                                 bucket_xh(xs->bucket),
1090                                                                 i,
1091                                                                 &block_off,
1092                                                                 &name_offset);
1093                         xs->base = bucket_block(xs->bucket, block_off);
1094                 }
1095                 if (ocfs2_xattr_is_local(xs->here)) {
1096                         memcpy(buffer, (void *)xs->base +
1097                                name_offset + name_len, size);
1098                 } else {
1099                         xv = (struct ocfs2_xattr_value_root *)
1100                                 (xs->base + name_offset + name_len);
1101                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1102                                                             buffer, size);
1103                         if (ret < 0) {
1104                                 mlog_errno(ret);
1105                                 goto cleanup;
1106                         }
1107                 }
1108         }
1109         ret = size;
1110 cleanup:
1111         ocfs2_xattr_bucket_free(xs->bucket);
1112
1113         brelse(xs->xattr_bh);
1114         xs->xattr_bh = NULL;
1115         return ret;
1116 }
1117
1118 int ocfs2_xattr_get_nolock(struct inode *inode,
1119                            struct buffer_head *di_bh,
1120                            int name_index,
1121                            const char *name,
1122                            void *buffer,
1123                            size_t buffer_size)
1124 {
1125         int ret;
1126         struct ocfs2_dinode *di = NULL;
1127         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1128         struct ocfs2_xattr_search xis = {
1129                 .not_found = -ENODATA,
1130         };
1131         struct ocfs2_xattr_search xbs = {
1132                 .not_found = -ENODATA,
1133         };
1134
1135         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1136                 return -EOPNOTSUPP;
1137
1138         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1139                 ret = -ENODATA;
1140
1141         xis.inode_bh = xbs.inode_bh = di_bh;
1142         di = (struct ocfs2_dinode *)di_bh->b_data;
1143
1144         down_read(&oi->ip_xattr_sem);
1145         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1146                                     buffer_size, &xis);
1147         if (ret == -ENODATA && di->i_xattr_loc)
1148                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1149                                             buffer_size, &xbs);
1150         up_read(&oi->ip_xattr_sem);
1151
1152         return ret;
1153 }
1154
1155 /* ocfs2_xattr_get()
1156  *
1157  * Copy an extended attribute into the buffer provided.
1158  * Buffer is NULL to compute the size of buffer required.
1159  */
1160 static int ocfs2_xattr_get(struct inode *inode,
1161                            int name_index,
1162                            const char *name,
1163                            void *buffer,
1164                            size_t buffer_size)
1165 {
1166         int ret;
1167         struct buffer_head *di_bh = NULL;
1168
1169         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1170         if (ret < 0) {
1171                 mlog_errno(ret);
1172                 return ret;
1173         }
1174         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1175                                      name, buffer, buffer_size);
1176
1177         ocfs2_inode_unlock(inode, 0);
1178
1179         brelse(di_bh);
1180
1181         return ret;
1182 }
1183
1184 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1185                                            handle_t *handle,
1186                                            struct ocfs2_xattr_value_root *xv,
1187                                            const void *value,
1188                                            int value_len)
1189 {
1190         int ret = 0, i, cp_len;
1191         u16 blocksize = inode->i_sb->s_blocksize;
1192         u32 p_cluster, num_clusters;
1193         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1194         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1195         u64 blkno;
1196         struct buffer_head *bh = NULL;
1197
1198         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1199
1200         while (cpos < clusters) {
1201                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1202                                                &num_clusters, &xv->xr_list);
1203                 if (ret) {
1204                         mlog_errno(ret);
1205                         goto out;
1206                 }
1207
1208                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1209
1210                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1211                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
1212                         if (ret) {
1213                                 mlog_errno(ret);
1214                                 goto out;
1215                         }
1216
1217                         ret = ocfs2_journal_access(handle,
1218                                                    inode,
1219                                                    bh,
1220                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1221                         if (ret < 0) {
1222                                 mlog_errno(ret);
1223                                 goto out;
1224                         }
1225
1226                         cp_len = value_len > blocksize ? blocksize : value_len;
1227                         memcpy(bh->b_data, value, cp_len);
1228                         value_len -= cp_len;
1229                         value += cp_len;
1230                         if (cp_len < blocksize)
1231                                 memset(bh->b_data + cp_len, 0,
1232                                        blocksize - cp_len);
1233
1234                         ret = ocfs2_journal_dirty(handle, bh);
1235                         if (ret < 0) {
1236                                 mlog_errno(ret);
1237                                 goto out;
1238                         }
1239                         brelse(bh);
1240                         bh = NULL;
1241
1242                         /*
1243                          * XXX: do we need to empty all the following
1244                          * blocks in this cluster?
1245                          */
1246                         if (!value_len)
1247                                 break;
1248                 }
1249                 cpos += num_clusters;
1250         }
1251 out:
1252         brelse(bh);
1253
1254         return ret;
1255 }
1256
1257 static int ocfs2_xattr_cleanup(struct inode *inode,
1258                                handle_t *handle,
1259                                struct ocfs2_xattr_info *xi,
1260                                struct ocfs2_xattr_search *xs,
1261                                struct ocfs2_xattr_value_buf *vb,
1262                                size_t offs)
1263 {
1264         int ret = 0;
1265         size_t name_len = strlen(xi->name);
1266         void *val = xs->base + offs;
1267         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1268
1269         ret = vb->vb_access(handle, inode, vb->vb_bh,
1270                             OCFS2_JOURNAL_ACCESS_WRITE);
1271         if (ret) {
1272                 mlog_errno(ret);
1273                 goto out;
1274         }
1275         /* Decrease xattr count */
1276         le16_add_cpu(&xs->header->xh_count, -1);
1277         /* Remove the xattr entry and tree root which has already be set*/
1278         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1279         memset(val, 0, size);
1280
1281         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1282         if (ret < 0)
1283                 mlog_errno(ret);
1284 out:
1285         return ret;
1286 }
1287
1288 static int ocfs2_xattr_update_entry(struct inode *inode,
1289                                     handle_t *handle,
1290                                     struct ocfs2_xattr_info *xi,
1291                                     struct ocfs2_xattr_search *xs,
1292                                     struct ocfs2_xattr_value_buf *vb,
1293                                     size_t offs)
1294 {
1295         int ret;
1296
1297         ret = vb->vb_access(handle, inode, vb->vb_bh,
1298                             OCFS2_JOURNAL_ACCESS_WRITE);
1299         if (ret) {
1300                 mlog_errno(ret);
1301                 goto out;
1302         }
1303
1304         xs->here->xe_name_offset = cpu_to_le16(offs);
1305         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1306         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1307                 ocfs2_xattr_set_local(xs->here, 1);
1308         else
1309                 ocfs2_xattr_set_local(xs->here, 0);
1310         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1311
1312         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1313         if (ret < 0)
1314                 mlog_errno(ret);
1315 out:
1316         return ret;
1317 }
1318
1319 /*
1320  * ocfs2_xattr_set_value_outside()
1321  *
1322  * Set large size value in B tree.
1323  */
1324 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1325                                          struct ocfs2_xattr_info *xi,
1326                                          struct ocfs2_xattr_search *xs,
1327                                          struct ocfs2_xattr_set_ctxt *ctxt,
1328                                          struct ocfs2_xattr_value_buf *vb,
1329                                          size_t offs)
1330 {
1331         size_t name_len = strlen(xi->name);
1332         void *val = xs->base + offs;
1333         struct ocfs2_xattr_value_root *xv = NULL;
1334         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1335         int ret = 0;
1336
1337         memset(val, 0, size);
1338         memcpy(val, xi->name, name_len);
1339         xv = (struct ocfs2_xattr_value_root *)
1340                 (val + OCFS2_XATTR_SIZE(name_len));
1341         xv->xr_clusters = 0;
1342         xv->xr_last_eb_blk = 0;
1343         xv->xr_list.l_tree_depth = 0;
1344         xv->xr_list.l_count = cpu_to_le16(1);
1345         xv->xr_list.l_next_free_rec = 0;
1346         vb->vb_xv = xv;
1347
1348         ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1349         if (ret < 0) {
1350                 mlog_errno(ret);
1351                 return ret;
1352         }
1353         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1354         if (ret < 0) {
1355                 mlog_errno(ret);
1356                 return ret;
1357         }
1358         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
1359                                               xi->value, xi->value_len);
1360         if (ret < 0)
1361                 mlog_errno(ret);
1362
1363         return ret;
1364 }
1365
1366 /*
1367  * ocfs2_xattr_set_entry_local()
1368  *
1369  * Set, replace or remove extended attribute in local.
1370  */
1371 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1372                                         struct ocfs2_xattr_info *xi,
1373                                         struct ocfs2_xattr_search *xs,
1374                                         struct ocfs2_xattr_entry *last,
1375                                         size_t min_offs)
1376 {
1377         size_t name_len = strlen(xi->name);
1378         int i;
1379
1380         if (xi->value && xs->not_found) {
1381                 /* Insert the new xattr entry. */
1382                 le16_add_cpu(&xs->header->xh_count, 1);
1383                 ocfs2_xattr_set_type(last, xi->name_index);
1384                 ocfs2_xattr_set_local(last, 1);
1385                 last->xe_name_len = name_len;
1386         } else {
1387                 void *first_val;
1388                 void *val;
1389                 size_t offs, size;
1390
1391                 first_val = xs->base + min_offs;
1392                 offs = le16_to_cpu(xs->here->xe_name_offset);
1393                 val = xs->base + offs;
1394
1395                 if (le64_to_cpu(xs->here->xe_value_size) >
1396                     OCFS2_XATTR_INLINE_SIZE)
1397                         size = OCFS2_XATTR_SIZE(name_len) +
1398                                 OCFS2_XATTR_ROOT_SIZE;
1399                 else
1400                         size = OCFS2_XATTR_SIZE(name_len) +
1401                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1402
1403                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1404                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1405                         /* The old and the new value have the
1406                            same size. Just replace the value. */
1407                         ocfs2_xattr_set_local(xs->here, 1);
1408                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1409                         /* Clear value bytes. */
1410                         memset(val + OCFS2_XATTR_SIZE(name_len),
1411                                0,
1412                                OCFS2_XATTR_SIZE(xi->value_len));
1413                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1414                                xi->value,
1415                                xi->value_len);
1416                         return;
1417                 }
1418                 /* Remove the old name+value. */
1419                 memmove(first_val + size, first_val, val - first_val);
1420                 memset(first_val, 0, size);
1421                 xs->here->xe_name_hash = 0;
1422                 xs->here->xe_name_offset = 0;
1423                 ocfs2_xattr_set_local(xs->here, 1);
1424                 xs->here->xe_value_size = 0;
1425
1426                 min_offs += size;
1427
1428                 /* Adjust all value offsets. */
1429                 last = xs->header->xh_entries;
1430                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1431                         size_t o = le16_to_cpu(last->xe_name_offset);
1432
1433                         if (o < offs)
1434                                 last->xe_name_offset = cpu_to_le16(o + size);
1435                         last += 1;
1436                 }
1437
1438                 if (!xi->value) {
1439                         /* Remove the old entry. */
1440                         last -= 1;
1441                         memmove(xs->here, xs->here + 1,
1442                                 (void *)last - (void *)xs->here);
1443                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1444                         le16_add_cpu(&xs->header->xh_count, -1);
1445                 }
1446         }
1447         if (xi->value) {
1448                 /* Insert the new name+value. */
1449                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1450                                 OCFS2_XATTR_SIZE(xi->value_len);
1451                 void *val = xs->base + min_offs - size;
1452
1453                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1454                 memset(val, 0, size);
1455                 memcpy(val, xi->name, name_len);
1456                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1457                        xi->value,
1458                        xi->value_len);
1459                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1460                 ocfs2_xattr_set_local(xs->here, 1);
1461                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1462         }
1463
1464         return;
1465 }
1466
1467 /*
1468  * ocfs2_xattr_set_entry()
1469  *
1470  * Set extended attribute entry into inode or block.
1471  *
1472  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1473  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1474  * then set value in B tree with set_value_outside().
1475  */
1476 static int ocfs2_xattr_set_entry(struct inode *inode,
1477                                  struct ocfs2_xattr_info *xi,
1478                                  struct ocfs2_xattr_search *xs,
1479                                  struct ocfs2_xattr_set_ctxt *ctxt,
1480                                  int flag)
1481 {
1482         struct ocfs2_xattr_entry *last;
1483         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1484         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1485         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1486         size_t size_l = 0;
1487         handle_t *handle = ctxt->handle;
1488         int free, i, ret;
1489         struct ocfs2_xattr_info xi_l = {
1490                 .name_index = xi->name_index,
1491                 .name = xi->name,
1492                 .value = xi->value,
1493                 .value_len = xi->value_len,
1494         };
1495         struct ocfs2_xattr_value_buf vb = {
1496                 .vb_bh = xs->xattr_bh,
1497                 .vb_access = ocfs2_journal_access_di,
1498         };
1499
1500         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1501                 BUG_ON(xs->xattr_bh == xs->inode_bh);
1502                 vb.vb_access = ocfs2_journal_access_xb;
1503         } else
1504                 BUG_ON(xs->xattr_bh != xs->inode_bh);
1505
1506         /* Compute min_offs, last and free space. */
1507         last = xs->header->xh_entries;
1508
1509         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1510                 size_t offs = le16_to_cpu(last->xe_name_offset);
1511                 if (offs < min_offs)
1512                         min_offs = offs;
1513                 last += 1;
1514         }
1515
1516         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
1517         if (free < 0)
1518                 return -EIO;
1519
1520         if (!xs->not_found) {
1521                 size_t size = 0;
1522                 if (ocfs2_xattr_is_local(xs->here))
1523                         size = OCFS2_XATTR_SIZE(name_len) +
1524                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1525                 else
1526                         size = OCFS2_XATTR_SIZE(name_len) +
1527                                 OCFS2_XATTR_ROOT_SIZE;
1528                 free += (size + sizeof(struct ocfs2_xattr_entry));
1529         }
1530         /* Check free space in inode or block */
1531         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1532                 if (free < sizeof(struct ocfs2_xattr_entry) +
1533                            OCFS2_XATTR_SIZE(name_len) +
1534                            OCFS2_XATTR_ROOT_SIZE) {
1535                         ret = -ENOSPC;
1536                         goto out;
1537                 }
1538                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1539                 xi_l.value = (void *)&def_xv;
1540                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1541         } else if (xi->value) {
1542                 if (free < sizeof(struct ocfs2_xattr_entry) +
1543                            OCFS2_XATTR_SIZE(name_len) +
1544                            OCFS2_XATTR_SIZE(xi->value_len)) {
1545                         ret = -ENOSPC;
1546                         goto out;
1547                 }
1548         }
1549
1550         if (!xs->not_found) {
1551                 /* For existing extended attribute */
1552                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1553                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1554                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1555                 void *val = xs->base + offs;
1556
1557                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1558                         /* Replace existing local xattr with tree root */
1559                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1560                                                             ctxt, &vb, offs);
1561                         if (ret < 0)
1562                                 mlog_errno(ret);
1563                         goto out;
1564                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1565                         /* For existing xattr which has value outside */
1566                         vb.vb_xv = (struct ocfs2_xattr_value_root *)
1567                                 (val + OCFS2_XATTR_SIZE(name_len));
1568
1569                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1570                                 /*
1571                                  * If new value need set outside also,
1572                                  * first truncate old value to new value,
1573                                  * then set new value with set_value_outside().
1574                                  */
1575                                 ret = ocfs2_xattr_value_truncate(inode,
1576                                                                  &vb,
1577                                                                  xi->value_len,
1578                                                                  ctxt);
1579                                 if (ret < 0) {
1580                                         mlog_errno(ret);
1581                                         goto out;
1582                                 }
1583
1584                                 ret = ocfs2_xattr_update_entry(inode,
1585                                                                handle,
1586                                                                xi,
1587                                                                xs,
1588                                                                &vb,
1589                                                                offs);
1590                                 if (ret < 0) {
1591                                         mlog_errno(ret);
1592                                         goto out;
1593                                 }
1594
1595                                 ret = __ocfs2_xattr_set_value_outside(inode,
1596                                                                 handle,
1597                                                                 vb.vb_xv,
1598                                                                 xi->value,
1599                                                                 xi->value_len);
1600                                 if (ret < 0)
1601                                         mlog_errno(ret);
1602                                 goto out;
1603                         } else {
1604                                 /*
1605                                  * If new value need set in local,
1606                                  * just trucate old value to zero.
1607                                  */
1608                                  ret = ocfs2_xattr_value_truncate(inode,
1609                                                                   &vb,
1610                                                                   0,
1611                                                                   ctxt);
1612                                 if (ret < 0)
1613                                         mlog_errno(ret);
1614                         }
1615                 }
1616         }
1617
1618         ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
1619                                       OCFS2_JOURNAL_ACCESS_WRITE);
1620         if (ret) {
1621                 mlog_errno(ret);
1622                 goto out;
1623         }
1624
1625         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1626                 ret = vb.vb_access(handle, inode, vb.vb_bh,
1627                                    OCFS2_JOURNAL_ACCESS_WRITE);
1628                 if (ret) {
1629                         mlog_errno(ret);
1630                         goto out;
1631                 }
1632         }
1633
1634         /*
1635          * Set value in local, include set tree root in local.
1636          * This is the first step for value size >INLINE_SIZE.
1637          */
1638         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1639
1640         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1641                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1642                 if (ret < 0) {
1643                         mlog_errno(ret);
1644                         goto out;
1645                 }
1646         }
1647
1648         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1649             (flag & OCFS2_INLINE_XATTR_FL)) {
1650                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1651                 unsigned int xattrsize = osb->s_xattr_inline_size;
1652
1653                 /*
1654                  * Adjust extent record count or inline data size
1655                  * to reserve space for extended attribute.
1656                  */
1657                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1658                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1659                         le16_add_cpu(&idata->id_count, -xattrsize);
1660                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1661                         struct ocfs2_extent_list *el = &di->id2.i_list;
1662                         le16_add_cpu(&el->l_count, -(xattrsize /
1663                                         sizeof(struct ocfs2_extent_rec)));
1664                 }
1665                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1666         }
1667         /* Update xattr flag */
1668         spin_lock(&oi->ip_lock);
1669         oi->ip_dyn_features |= flag;
1670         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1671         spin_unlock(&oi->ip_lock);
1672
1673         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1674         if (ret < 0)
1675                 mlog_errno(ret);
1676
1677         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1678                 /*
1679                  * Set value outside in B tree.
1680                  * This is the second step for value size > INLINE_SIZE.
1681                  */
1682                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1683                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1684                                                     &vb, offs);
1685                 if (ret < 0) {
1686                         int ret2;
1687
1688                         mlog_errno(ret);
1689                         /*
1690                          * If set value outside failed, we have to clean
1691                          * the junk tree root we have already set in local.
1692                          */
1693                         ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1694                                                    xi, xs, &vb, offs);
1695                         if (ret2 < 0)
1696                                 mlog_errno(ret2);
1697                 }
1698         }
1699 out:
1700         return ret;
1701 }
1702
1703 static int ocfs2_remove_value_outside(struct inode*inode,
1704                                       struct ocfs2_xattr_value_buf *vb,
1705                                       struct ocfs2_xattr_header *header)
1706 {
1707         int ret = 0, i;
1708         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1709         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1710
1711         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1712
1713         ctxt.handle = ocfs2_start_trans(osb,
1714                                         ocfs2_remove_extent_credits(osb->sb));
1715         if (IS_ERR(ctxt.handle)) {
1716                 ret = PTR_ERR(ctxt.handle);
1717                 mlog_errno(ret);
1718                 goto out;
1719         }
1720
1721         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1722                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1723
1724                 if (!ocfs2_xattr_is_local(entry)) {
1725                         void *val;
1726
1727                         val = (void *)header +
1728                                 le16_to_cpu(entry->xe_name_offset);
1729                         vb->vb_xv = (struct ocfs2_xattr_value_root *)
1730                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1731                         ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1732                         if (ret < 0) {
1733                                 mlog_errno(ret);
1734                                 break;
1735                         }
1736                 }
1737         }
1738
1739         ocfs2_commit_trans(osb, ctxt.handle);
1740         ocfs2_schedule_truncate_log_flush(osb, 1);
1741         ocfs2_run_deallocs(osb, &ctxt.dealloc);
1742 out:
1743         return ret;
1744 }
1745
1746 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1747                                     struct buffer_head *di_bh)
1748 {
1749
1750         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1751         struct ocfs2_xattr_header *header;
1752         int ret;
1753         struct ocfs2_xattr_value_buf vb = {
1754                 .vb_bh = di_bh,
1755                 .vb_access = ocfs2_journal_access_di,
1756         };
1757
1758         header = (struct ocfs2_xattr_header *)
1759                  ((void *)di + inode->i_sb->s_blocksize -
1760                  le16_to_cpu(di->i_xattr_inline_size));
1761
1762         ret = ocfs2_remove_value_outside(inode, &vb, header);
1763
1764         return ret;
1765 }
1766
1767 static int ocfs2_xattr_block_remove(struct inode *inode,
1768                                     struct buffer_head *blk_bh)
1769 {
1770         struct ocfs2_xattr_block *xb;
1771         int ret = 0;
1772         struct ocfs2_xattr_value_buf vb = {
1773                 .vb_bh = blk_bh,
1774                 .vb_access = ocfs2_journal_access_xb,
1775         };
1776
1777         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1778         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1779                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1780                 ret = ocfs2_remove_value_outside(inode, &vb, header);
1781         } else
1782                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1783
1784         return ret;
1785 }
1786
1787 static int ocfs2_xattr_free_block(struct inode *inode,
1788                                   u64 block)
1789 {
1790         struct inode *xb_alloc_inode;
1791         struct buffer_head *xb_alloc_bh = NULL;
1792         struct buffer_head *blk_bh = NULL;
1793         struct ocfs2_xattr_block *xb;
1794         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1795         handle_t *handle;
1796         int ret = 0;
1797         u64 blk, bg_blkno;
1798         u16 bit;
1799
1800         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1801         if (ret < 0) {
1802                 mlog_errno(ret);
1803                 goto out;
1804         }
1805
1806         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1807         if (ret < 0) {
1808                 mlog_errno(ret);
1809                 goto out;
1810         }
1811
1812         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1813         blk = le64_to_cpu(xb->xb_blkno);
1814         bit = le16_to_cpu(xb->xb_suballoc_bit);
1815         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1816
1817         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1818                                 EXTENT_ALLOC_SYSTEM_INODE,
1819                                 le16_to_cpu(xb->xb_suballoc_slot));
1820         if (!xb_alloc_inode) {
1821                 ret = -ENOMEM;
1822                 mlog_errno(ret);
1823                 goto out;
1824         }
1825         mutex_lock(&xb_alloc_inode->i_mutex);
1826
1827         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1828         if (ret < 0) {
1829                 mlog_errno(ret);
1830                 goto out_mutex;
1831         }
1832
1833         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1834         if (IS_ERR(handle)) {
1835                 ret = PTR_ERR(handle);
1836                 mlog_errno(ret);
1837                 goto out_unlock;
1838         }
1839
1840         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1841                                        bit, bg_blkno, 1);
1842         if (ret < 0)
1843                 mlog_errno(ret);
1844
1845         ocfs2_commit_trans(osb, handle);
1846 out_unlock:
1847         ocfs2_inode_unlock(xb_alloc_inode, 1);
1848         brelse(xb_alloc_bh);
1849 out_mutex:
1850         mutex_unlock(&xb_alloc_inode->i_mutex);
1851         iput(xb_alloc_inode);
1852 out:
1853         brelse(blk_bh);
1854         return ret;
1855 }
1856
1857 /*
1858  * ocfs2_xattr_remove()
1859  *
1860  * Free extended attribute resources associated with this inode.
1861  */
1862 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1863 {
1864         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1865         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1866         handle_t *handle;
1867         int ret;
1868
1869         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1870                 return 0;
1871
1872         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1873                 return 0;
1874
1875         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1876                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1877                 if (ret < 0) {
1878                         mlog_errno(ret);
1879                         goto out;
1880                 }
1881         }
1882
1883         if (di->i_xattr_loc) {
1884                 ret = ocfs2_xattr_free_block(inode,
1885                                              le64_to_cpu(di->i_xattr_loc));
1886                 if (ret < 0) {
1887                         mlog_errno(ret);
1888                         goto out;
1889                 }
1890         }
1891
1892         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1893                                    OCFS2_INODE_UPDATE_CREDITS);
1894         if (IS_ERR(handle)) {
1895                 ret = PTR_ERR(handle);
1896                 mlog_errno(ret);
1897                 goto out;
1898         }
1899         ret = ocfs2_journal_access_di(handle, inode, di_bh,
1900                                       OCFS2_JOURNAL_ACCESS_WRITE);
1901         if (ret) {
1902                 mlog_errno(ret);
1903                 goto out_commit;
1904         }
1905
1906         di->i_xattr_loc = 0;
1907
1908         spin_lock(&oi->ip_lock);
1909         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1910         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1911         spin_unlock(&oi->ip_lock);
1912
1913         ret = ocfs2_journal_dirty(handle, di_bh);
1914         if (ret < 0)
1915                 mlog_errno(ret);
1916 out_commit:
1917         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1918 out:
1919         return ret;
1920 }
1921
1922 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1923                                         struct ocfs2_dinode *di)
1924 {
1925         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1926         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1927         int free;
1928
1929         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1930                 return 0;
1931
1932         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1933                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1934                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1935         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1936                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1937                         le64_to_cpu(di->i_size);
1938         } else {
1939                 struct ocfs2_extent_list *el = &di->id2.i_list;
1940                 free = (le16_to_cpu(el->l_count) -
1941                         le16_to_cpu(el->l_next_free_rec)) *
1942                         sizeof(struct ocfs2_extent_rec);
1943         }
1944         if (free >= xattrsize)
1945                 return 1;
1946
1947         return 0;
1948 }
1949
1950 /*
1951  * ocfs2_xattr_ibody_find()
1952  *
1953  * Find extended attribute in inode block and
1954  * fill search info into struct ocfs2_xattr_search.
1955  */
1956 static int ocfs2_xattr_ibody_find(struct inode *inode,
1957                                   int name_index,
1958                                   const char *name,
1959                                   struct ocfs2_xattr_search *xs)
1960 {
1961         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1962         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1963         int ret;
1964         int has_space = 0;
1965
1966         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1967                 return 0;
1968
1969         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1970                 down_read(&oi->ip_alloc_sem);
1971                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1972                 up_read(&oi->ip_alloc_sem);
1973                 if (!has_space)
1974                         return 0;
1975         }
1976
1977         xs->xattr_bh = xs->inode_bh;
1978         xs->end = (void *)di + inode->i_sb->s_blocksize;
1979         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1980                 xs->header = (struct ocfs2_xattr_header *)
1981                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1982         else
1983                 xs->header = (struct ocfs2_xattr_header *)
1984                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1985         xs->base = (void *)xs->header;
1986         xs->here = xs->header->xh_entries;
1987
1988         /* Find the named attribute. */
1989         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1990                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1991                 if (ret && ret != -ENODATA)
1992                         return ret;
1993                 xs->not_found = ret;
1994         }
1995
1996         return 0;
1997 }
1998
1999 /*
2000  * ocfs2_xattr_ibody_set()
2001  *
2002  * Set, replace or remove an extended attribute into inode block.
2003  *
2004  */
2005 static int ocfs2_xattr_ibody_set(struct inode *inode,
2006                                  struct ocfs2_xattr_info *xi,
2007                                  struct ocfs2_xattr_search *xs,
2008                                  struct ocfs2_xattr_set_ctxt *ctxt)
2009 {
2010         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2011         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2012         int ret;
2013
2014         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2015                 return -ENOSPC;
2016
2017         down_write(&oi->ip_alloc_sem);
2018         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2019                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2020                         ret = -ENOSPC;
2021                         goto out;
2022                 }
2023         }
2024
2025         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2026                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2027 out:
2028         up_write(&oi->ip_alloc_sem);
2029
2030         return ret;
2031 }
2032
2033 /*
2034  * ocfs2_xattr_block_find()
2035  *
2036  * Find extended attribute in external block and
2037  * fill search info into struct ocfs2_xattr_search.
2038  */
2039 static int ocfs2_xattr_block_find(struct inode *inode,
2040                                   int name_index,
2041                                   const char *name,
2042                                   struct ocfs2_xattr_search *xs)
2043 {
2044         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2045         struct buffer_head *blk_bh = NULL;
2046         struct ocfs2_xattr_block *xb;
2047         int ret = 0;
2048
2049         if (!di->i_xattr_loc)
2050                 return ret;
2051
2052         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2053                                      &blk_bh);
2054         if (ret < 0) {
2055                 mlog_errno(ret);
2056                 return ret;
2057         }
2058
2059         xs->xattr_bh = blk_bh;
2060         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2061
2062         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2063                 xs->header = &xb->xb_attrs.xb_header;
2064                 xs->base = (void *)xs->header;
2065                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2066                 xs->here = xs->header->xh_entries;
2067
2068                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2069         } else
2070                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2071                                                    name_index,
2072                                                    name, xs);
2073
2074         if (ret && ret != -ENODATA) {
2075                 xs->xattr_bh = NULL;
2076                 goto cleanup;
2077         }
2078         xs->not_found = ret;
2079         return 0;
2080 cleanup:
2081         brelse(blk_bh);
2082
2083         return ret;
2084 }
2085
2086 /*
2087  * ocfs2_xattr_block_set()
2088  *
2089  * Set, replace or remove an extended attribute into external block.
2090  *
2091  */
2092 static int ocfs2_xattr_block_set(struct inode *inode,
2093                                  struct ocfs2_xattr_info *xi,
2094                                  struct ocfs2_xattr_search *xs,
2095                                  struct ocfs2_xattr_set_ctxt *ctxt)
2096 {
2097         struct buffer_head *new_bh = NULL;
2098         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2099         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
2100         handle_t *handle = ctxt->handle;
2101         struct ocfs2_xattr_block *xblk = NULL;
2102         u16 suballoc_bit_start;
2103         u32 num_got;
2104         u64 first_blkno;
2105         int ret;
2106
2107         if (!xs->xattr_bh) {
2108                 ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
2109                                               OCFS2_JOURNAL_ACCESS_CREATE);
2110                 if (ret < 0) {
2111                         mlog_errno(ret);
2112                         goto end;
2113                 }
2114
2115                 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2116                                            &suballoc_bit_start, &num_got,
2117                                            &first_blkno);
2118                 if (ret < 0) {
2119                         mlog_errno(ret);
2120                         goto end;
2121                 }
2122
2123                 new_bh = sb_getblk(inode->i_sb, first_blkno);
2124                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2125
2126                 ret = ocfs2_journal_access_xb(handle, inode, new_bh,
2127                                               OCFS2_JOURNAL_ACCESS_CREATE);
2128                 if (ret < 0) {
2129                         mlog_errno(ret);
2130                         goto end;
2131                 }
2132
2133                 /* Initialize ocfs2_xattr_block */
2134                 xs->xattr_bh = new_bh;
2135                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2136                 memset(xblk, 0, inode->i_sb->s_blocksize);
2137                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2138                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2139                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2140                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2141                 xblk->xb_blkno = cpu_to_le64(first_blkno);
2142
2143                 xs->header = &xblk->xb_attrs.xb_header;
2144                 xs->base = (void *)xs->header;
2145                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2146                 xs->here = xs->header->xh_entries;
2147
2148                 ret = ocfs2_journal_dirty(handle, new_bh);
2149                 if (ret < 0) {
2150                         mlog_errno(ret);
2151                         goto end;
2152                 }
2153                 di->i_xattr_loc = cpu_to_le64(first_blkno);
2154                 ocfs2_journal_dirty(handle, xs->inode_bh);
2155         } else
2156                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2157
2158         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2159                 /* Set extended attribute into external block */
2160                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2161                                             OCFS2_HAS_XATTR_FL);
2162                 if (!ret || ret != -ENOSPC)
2163                         goto end;
2164
2165                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2166                 if (ret)
2167                         goto end;
2168         }
2169
2170         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2171
2172 end:
2173
2174         return ret;
2175 }
2176
2177 /* Check whether the new xattr can be inserted into the inode. */
2178 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2179                                        struct ocfs2_xattr_info *xi,
2180                                        struct ocfs2_xattr_search *xs)
2181 {
2182         u64 value_size;
2183         struct ocfs2_xattr_entry *last;
2184         int free, i;
2185         size_t min_offs = xs->end - xs->base;
2186
2187         if (!xs->header)
2188                 return 0;
2189
2190         last = xs->header->xh_entries;
2191
2192         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2193                 size_t offs = le16_to_cpu(last->xe_name_offset);
2194                 if (offs < min_offs)
2195                         min_offs = offs;
2196                 last += 1;
2197         }
2198
2199         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2200         if (free < 0)
2201                 return 0;
2202
2203         BUG_ON(!xs->not_found);
2204
2205         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2206                 value_size = OCFS2_XATTR_ROOT_SIZE;
2207         else
2208                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2209
2210         if (free >= sizeof(struct ocfs2_xattr_entry) +
2211                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2212                 return 1;
2213
2214         return 0;
2215 }
2216
2217 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2218                                      struct ocfs2_dinode *di,
2219                                      struct ocfs2_xattr_info *xi,
2220                                      struct ocfs2_xattr_search *xis,
2221                                      struct ocfs2_xattr_search *xbs,
2222                                      int *clusters_need,
2223                                      int *meta_need,
2224                                      int *credits_need)
2225 {
2226         int ret = 0, old_in_xb = 0;
2227         int clusters_add = 0, meta_add = 0, credits = 0;
2228         struct buffer_head *bh = NULL;
2229         struct ocfs2_xattr_block *xb = NULL;
2230         struct ocfs2_xattr_entry *xe = NULL;
2231         struct ocfs2_xattr_value_root *xv = NULL;
2232         char *base = NULL;
2233         int name_offset, name_len = 0;
2234         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2235                                                     xi->value_len);
2236         u64 value_size;
2237
2238         /*
2239          * Calculate the clusters we need to write.
2240          * No matter whether we replace an old one or add a new one,
2241          * we need this for writing.
2242          */
2243         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2244                 credits += new_clusters *
2245                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
2246
2247         if (xis->not_found && xbs->not_found) {
2248                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2249
2250                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2251                         clusters_add += new_clusters;
2252                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2253                                                         &def_xv.xv.xr_list,
2254                                                         new_clusters);
2255                 }
2256
2257                 goto meta_guess;
2258         }
2259
2260         if (!xis->not_found) {
2261                 xe = xis->here;
2262                 name_offset = le16_to_cpu(xe->xe_name_offset);
2263                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2264                 base = xis->base;
2265                 credits += OCFS2_INODE_UPDATE_CREDITS;
2266         } else {
2267                 int i, block_off = 0;
2268                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2269                 xe = xbs->here;
2270                 name_offset = le16_to_cpu(xe->xe_name_offset);
2271                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2272                 i = xbs->here - xbs->header->xh_entries;
2273                 old_in_xb = 1;
2274
2275                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2276                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2277                                                         bucket_xh(xbs->bucket),
2278                                                         i, &block_off,
2279                                                         &name_offset);
2280                         base = bucket_block(xbs->bucket, block_off);
2281                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2282                 } else {
2283                         base = xbs->base;
2284                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2285                 }
2286         }
2287
2288         /*
2289          * delete a xattr doesn't need metadata and cluster allocation.
2290          * so just calculate the credits and return.
2291          *
2292          * The credits for removing the value tree will be extended
2293          * by ocfs2_remove_extent itself.
2294          */
2295         if (!xi->value) {
2296                 if (!ocfs2_xattr_is_local(xe))
2297                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2298
2299                 goto out;
2300         }
2301
2302         /* do cluster allocation guess first. */
2303         value_size = le64_to_cpu(xe->xe_value_size);
2304
2305         if (old_in_xb) {
2306                 /*
2307                  * In xattr set, we always try to set the xe in inode first,
2308                  * so if it can be inserted into inode successfully, the old
2309                  * one will be removed from the xattr block, and this xattr
2310                  * will be inserted into inode as a new xattr in inode.
2311                  */
2312                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2313                         clusters_add += new_clusters;
2314                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
2315                                     OCFS2_INODE_UPDATE_CREDITS;
2316                         if (!ocfs2_xattr_is_local(xe))
2317                                 credits += ocfs2_calc_extend_credits(
2318                                                         inode->i_sb,
2319                                                         &def_xv.xv.xr_list,
2320                                                         new_clusters);
2321                         goto out;
2322                 }
2323         }
2324
2325         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2326                 /* the new values will be stored outside. */
2327                 u32 old_clusters = 0;
2328
2329                 if (!ocfs2_xattr_is_local(xe)) {
2330                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2331                                                                  value_size);
2332                         xv = (struct ocfs2_xattr_value_root *)
2333                              (base + name_offset + name_len);
2334                         value_size = OCFS2_XATTR_ROOT_SIZE;
2335                 } else
2336                         xv = &def_xv.xv;
2337
2338                 if (old_clusters >= new_clusters) {
2339                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2340                         goto out;
2341                 } else {
2342                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2343                         clusters_add += new_clusters - old_clusters;
2344                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2345                                                              &xv->xr_list,
2346                                                              new_clusters -
2347                                                              old_clusters);
2348                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2349                                 goto out;
2350                 }
2351         } else {
2352                 /*
2353                  * Now the new value will be stored inside. So if the new
2354                  * value is smaller than the size of value root or the old
2355                  * value, we don't need any allocation, otherwise we have
2356                  * to guess metadata allocation.
2357                  */
2358                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2359                     (!ocfs2_xattr_is_local(xe) &&
2360                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2361                         goto out;
2362         }
2363
2364 meta_guess:
2365         /* calculate metadata allocation. */
2366         if (di->i_xattr_loc) {
2367                 if (!xbs->xattr_bh) {
2368                         ret = ocfs2_read_xattr_block(inode,
2369                                                      le64_to_cpu(di->i_xattr_loc),
2370                                                      &bh);
2371                         if (ret) {
2372                                 mlog_errno(ret);
2373                                 goto out;
2374                         }
2375
2376                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2377                 } else
2378                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2379
2380                 /*
2381                  * If there is already an xattr tree, good, we can calculate
2382                  * like other b-trees. Otherwise we may have the chance of
2383                  * create a tree, the credit calculation is borrowed from
2384                  * ocfs2_calc_extend_credits with root_el = NULL. And the
2385                  * new tree will be cluster based, so no meta is needed.
2386                  */
2387                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2388                         struct ocfs2_extent_list *el =
2389                                  &xb->xb_attrs.xb_root.xt_list;
2390                         meta_add += ocfs2_extend_meta_needed(el);
2391                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2392                                                              el, 1);
2393                 } else
2394                         credits += OCFS2_SUBALLOC_ALLOC + 1;
2395
2396                 /*
2397                  * This cluster will be used either for new bucket or for
2398                  * new xattr block.
2399                  * If the cluster size is the same as the bucket size, one
2400                  * more is needed since we may need to extend the bucket
2401                  * also.
2402                  */
2403                 clusters_add += 1;
2404                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2405                 if (OCFS2_XATTR_BUCKET_SIZE ==
2406                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2407                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2408                         clusters_add += 1;
2409                 }
2410         } else {
2411                 meta_add += 1;
2412                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2413         }
2414 out:
2415         if (clusters_need)
2416                 *clusters_need = clusters_add;
2417         if (meta_need)
2418                 *meta_need = meta_add;
2419         if (credits_need)
2420                 *credits_need = credits;
2421         brelse(bh);
2422         return ret;
2423 }
2424
2425 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2426                                      struct ocfs2_dinode *di,
2427                                      struct ocfs2_xattr_info *xi,
2428                                      struct ocfs2_xattr_search *xis,
2429                                      struct ocfs2_xattr_search *xbs,
2430                                      struct ocfs2_xattr_set_ctxt *ctxt,
2431                                      int *credits)
2432 {
2433         int clusters_add, meta_add, ret;
2434         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2435
2436         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2437
2438         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2439
2440         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2441                                         &clusters_add, &meta_add, credits);
2442         if (ret) {
2443                 mlog_errno(ret);
2444                 return ret;
2445         }
2446
2447         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2448              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2449
2450         if (meta_add) {
2451                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2452                                                         &ctxt->meta_ac);
2453                 if (ret) {
2454                         mlog_errno(ret);
2455                         goto out;
2456                 }
2457         }
2458
2459         if (clusters_add) {
2460                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2461                 if (ret)
2462                         mlog_errno(ret);
2463         }
2464 out:
2465         if (ret) {
2466                 if (ctxt->meta_ac) {
2467                         ocfs2_free_alloc_context(ctxt->meta_ac);
2468                         ctxt->meta_ac = NULL;
2469                 }
2470
2471                 /*
2472                  * We cannot have an error and a non null ctxt->data_ac.
2473                  */
2474         }
2475
2476         return ret;
2477 }
2478
2479 static int __ocfs2_xattr_set_handle(struct inode *inode,
2480                                     struct ocfs2_dinode *di,
2481                                     struct ocfs2_xattr_info *xi,
2482                                     struct ocfs2_xattr_search *xis,
2483                                     struct ocfs2_xattr_search *xbs,
2484                                     struct ocfs2_xattr_set_ctxt *ctxt)
2485 {
2486         int ret = 0, credits, old_found;
2487
2488         if (!xi->value) {
2489                 /* Remove existing extended attribute */
2490                 if (!xis->not_found)
2491                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2492                 else if (!xbs->not_found)
2493                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2494         } else {
2495                 /* We always try to set extended attribute into inode first*/
2496                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2497                 if (!ret && !xbs->not_found) {
2498                         /*
2499                          * If succeed and that extended attribute existing in
2500                          * external block, then we will remove it.
2501                          */
2502                         xi->value = NULL;
2503                         xi->value_len = 0;
2504
2505                         old_found = xis->not_found;
2506                         xis->not_found = -ENODATA;
2507                         ret = ocfs2_calc_xattr_set_need(inode,
2508                                                         di,
2509                                                         xi,
2510                                                         xis,
2511                                                         xbs,
2512                                                         NULL,
2513                                                         NULL,
2514                                                         &credits);
2515                         xis->not_found = old_found;
2516                         if (ret) {
2517                                 mlog_errno(ret);
2518                                 goto out;
2519                         }
2520
2521                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2522                                         ctxt->handle->h_buffer_credits);
2523                         if (ret) {
2524                                 mlog_errno(ret);
2525                                 goto out;
2526                         }
2527                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2528                 } else if (ret == -ENOSPC) {
2529                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2530                                 ret = ocfs2_xattr_block_find(inode,
2531                                                              xi->name_index,
2532                                                              xi->name, xbs);
2533                                 if (ret)
2534                                         goto out;
2535
2536                                 old_found = xis->not_found;
2537                                 xis->not_found = -ENODATA;
2538                                 ret = ocfs2_calc_xattr_set_need(inode,
2539                                                                 di,
2540                                                                 xi,
2541                                                                 xis,
2542                                                                 xbs,
2543                                                                 NULL,
2544                                                                 NULL,
2545                                                                 &credits);
2546                                 xis->not_found = old_found;
2547                                 if (ret) {
2548                                         mlog_errno(ret);
2549                                         goto out;
2550                                 }
2551
2552                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2553                                         ctxt->handle->h_buffer_credits);
2554                                 if (ret) {
2555                                         mlog_errno(ret);
2556                                         goto out;
2557                                 }
2558                         }
2559                         /*
2560                          * If no space in inode, we will set extended attribute
2561                          * into external block.
2562                          */
2563                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2564                         if (ret)
2565                                 goto out;
2566                         if (!xis->not_found) {
2567                                 /*
2568                                  * If succeed and that extended attribute
2569                                  * existing in inode, we will remove it.
2570                                  */
2571                                 xi->value = NULL;
2572                                 xi->value_len = 0;
2573                                 xbs->not_found = -ENODATA;
2574                                 ret = ocfs2_calc_xattr_set_need(inode,
2575                                                                 di,
2576                                                                 xi,
2577                                                                 xis,
2578                                                                 xbs,
2579                                                                 NULL,
2580                                                                 NULL,
2581                                                                 &credits);
2582                                 if (ret) {
2583                                         mlog_errno(ret);
2584                                         goto out;
2585                                 }
2586
2587                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2588                                                 ctxt->handle->h_buffer_credits);
2589                                 if (ret) {
2590                                         mlog_errno(ret);
2591                                         goto out;
2592                                 }
2593                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2594                                                             xis, ctxt);
2595                         }
2596                 }
2597         }
2598
2599         if (!ret) {
2600                 /* Update inode ctime. */
2601                 ret = ocfs2_journal_access_di(ctxt->handle, inode,
2602                                               xis->inode_bh,
2603                                               OCFS2_JOURNAL_ACCESS_WRITE);
2604                 if (ret) {
2605                         mlog_errno(ret);
2606                         goto out;
2607                 }
2608
2609                 inode->i_ctime = CURRENT_TIME;
2610                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2611                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2612                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2613         }
2614 out:
2615         return ret;
2616 }
2617
2618 /*
2619  * This function only called duing creating inode
2620  * for init security/acl xattrs of the new inode.
2621  * All transanction credits have been reserved in mknod.
2622  */
2623 int ocfs2_xattr_set_handle(handle_t *handle,
2624                            struct inode *inode,
2625                            struct buffer_head *di_bh,
2626                            int name_index,
2627                            const char *name,
2628                            const void *value,
2629                            size_t value_len,
2630                            int flags,
2631                            struct ocfs2_alloc_context *meta_ac,
2632                            struct ocfs2_alloc_context *data_ac)
2633 {
2634         struct ocfs2_dinode *di;
2635         int ret;
2636
2637         struct ocfs2_xattr_info xi = {
2638                 .name_index = name_index,
2639                 .name = name,
2640                 .value = value,
2641                 .value_len = value_len,
2642         };
2643
2644         struct ocfs2_xattr_search xis = {
2645                 .not_found = -ENODATA,
2646         };
2647
2648         struct ocfs2_xattr_search xbs = {
2649                 .not_found = -ENODATA,
2650         };
2651
2652         struct ocfs2_xattr_set_ctxt ctxt = {
2653                 .handle = handle,
2654                 .meta_ac = meta_ac,
2655                 .data_ac = data_ac,
2656         };
2657
2658         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2659                 return -EOPNOTSUPP;
2660
2661         /*
2662          * In extreme situation, may need xattr bucket when
2663          * block size is too small. And we have already reserved
2664          * the credits for bucket in mknod.
2665          */
2666         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
2667                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2668                 if (!xbs.bucket) {
2669                         mlog_errno(-ENOMEM);
2670                         return -ENOMEM;
2671                 }
2672         }
2673
2674         xis.inode_bh = xbs.inode_bh = di_bh;
2675         di = (struct ocfs2_dinode *)di_bh->b_data;
2676
2677         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2678
2679         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2680         if (ret)
2681                 goto cleanup;
2682         if (xis.not_found) {
2683                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2684                 if (ret)
2685                         goto cleanup;
2686         }
2687
2688         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2689
2690 cleanup:
2691         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2692         brelse(xbs.xattr_bh);
2693         ocfs2_xattr_bucket_free(xbs.bucket);
2694
2695         return ret;
2696 }
2697
2698 /*
2699  * ocfs2_xattr_set()
2700  *
2701  * Set, replace or remove an extended attribute for this inode.
2702  * value is NULL to remove an existing extended attribute, else either
2703  * create or replace an extended attribute.
2704  */
2705 int ocfs2_xattr_set(struct inode *inode,
2706                     int name_index,
2707                     const char *name,
2708                     const void *value,
2709                     size_t value_len,
2710                     int flags)
2711 {
2712         struct buffer_head *di_bh = NULL;
2713         struct ocfs2_dinode *di;
2714         int ret, credits;
2715         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2716         struct inode *tl_inode = osb->osb_tl_inode;
2717         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2718
2719         struct ocfs2_xattr_info xi = {
2720                 .name_index = name_index,
2721                 .name = name,
2722                 .value = value,
2723                 .value_len = value_len,
2724         };
2725
2726         struct ocfs2_xattr_search xis = {
2727                 .not_found = -ENODATA,
2728         };
2729
2730         struct ocfs2_xattr_search xbs = {
2731                 .not_found = -ENODATA,
2732         };
2733
2734         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2735                 return -EOPNOTSUPP;
2736
2737         /*
2738          * Only xbs will be used on indexed trees.  xis doesn't need a
2739          * bucket.
2740          */
2741         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2742         if (!xbs.bucket) {
2743                 mlog_errno(-ENOMEM);
2744                 return -ENOMEM;
2745         }
2746
2747         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2748         if (ret < 0) {
2749                 mlog_errno(ret);
2750                 goto cleanup_nolock;
2751         }
2752         xis.inode_bh = xbs.inode_bh = di_bh;
2753         di = (struct ocfs2_dinode *)di_bh->b_data;
2754
2755         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2756         /*
2757          * Scan inode and external block to find the same name
2758          * extended attribute and collect search infomation.
2759          */
2760         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2761         if (ret)
2762                 goto cleanup;
2763         if (xis.not_found) {
2764                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2765                 if (ret)
2766                         goto cleanup;
2767         }
2768
2769         if (xis.not_found && xbs.not_found) {
2770                 ret = -ENODATA;
2771                 if (flags & XATTR_REPLACE)
2772                         goto cleanup;
2773                 ret = 0;
2774                 if (!value)
2775                         goto cleanup;
2776         } else {
2777                 ret = -EEXIST;
2778                 if (flags & XATTR_CREATE)
2779                         goto cleanup;
2780         }
2781
2782
2783         mutex_lock(&tl_inode->i_mutex);
2784
2785         if (ocfs2_truncate_log_needs_flush(osb)) {
2786                 ret = __ocfs2_flush_truncate_log(osb);
2787                 if (ret < 0) {
2788                         mutex_unlock(&tl_inode->i_mutex);
2789                         mlog_errno(ret);
2790                         goto cleanup;
2791                 }
2792         }
2793         mutex_unlock(&tl_inode->i_mutex);
2794
2795         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2796                                         &xbs, &ctxt, &credits);
2797         if (ret) {
2798                 mlog_errno(ret);
2799                 goto cleanup;
2800         }
2801
2802         /* we need to update inode's ctime field, so add credit for it. */
2803         credits += OCFS2_INODE_UPDATE_CREDITS;
2804         ctxt.handle = ocfs2_start_trans(osb, credits);
2805         if (IS_ERR(ctxt.handle)) {
2806                 ret = PTR_ERR(ctxt.handle);
2807                 mlog_errno(ret);
2808                 goto cleanup;
2809         }
2810
2811         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2812
2813         ocfs2_commit_trans(osb, ctxt.handle);
2814
2815         if (ctxt.data_ac)
2816                 ocfs2_free_alloc_context(ctxt.data_ac);
2817         if (ctxt.meta_ac)
2818                 ocfs2_free_alloc_context(ctxt.meta_ac);
2819         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2820                 ocfs2_schedule_truncate_log_flush(osb, 1);
2821         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2822 cleanup:
2823         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2824         ocfs2_inode_unlock(inode, 1);
2825 cleanup_nolock:
2826         brelse(di_bh);
2827         brelse(xbs.xattr_bh);
2828         ocfs2_xattr_bucket_free(xbs.bucket);
2829
2830         return ret;
2831 }
2832
2833 /*
2834  * Find the xattr extent rec which may contains name_hash.
2835  * e_cpos will be the first name hash of the xattr rec.
2836  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2837  */
2838 static int ocfs2_xattr_get_rec(struct inode *inode,
2839                                u32 name_hash,
2840                                u64 *p_blkno,
2841                                u32 *e_cpos,
2842                                u32 *num_clusters,
2843                                struct ocfs2_extent_list *el)
2844 {
2845         int ret = 0, i;
2846         struct buffer_head *eb_bh = NULL;
2847         struct ocfs2_extent_block *eb;
2848         struct ocfs2_extent_rec *rec = NULL;
2849         u64 e_blkno = 0;
2850
2851         if (el->l_tree_depth) {
2852                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2853                 if (ret) {
2854                         mlog_errno(ret);
2855                         goto out;
2856                 }
2857
2858                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2859                 el = &eb->h_list;
2860
2861                 if (el->l_tree_depth) {
2862                         ocfs2_error(inode->i_sb,
2863                                     "Inode %lu has non zero tree depth in "
2864                                     "xattr tree block %llu\n", inode->i_ino,
2865                                     (unsigned long long)eb_bh->b_blocknr);
2866                         ret = -EROFS;
2867                         goto out;
2868                 }
2869         }
2870
2871         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2872                 rec = &el->l_recs[i];
2873
2874                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2875                         e_blkno = le64_to_cpu(rec->e_blkno);
2876                         break;
2877                 }
2878         }
2879
2880         if (!e_blkno) {
2881                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2882                             "record (%u, %u, 0) in xattr", inode->i_ino,
2883                             le32_to_cpu(rec->e_cpos),
2884                             ocfs2_rec_clusters(el, rec));
2885                 ret = -EROFS;
2886                 goto out;
2887         }
2888
2889         *p_blkno = le64_to_cpu(rec->e_blkno);
2890         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2891         if (e_cpos)
2892                 *e_cpos = le32_to_cpu(rec->e_cpos);
2893 out:
2894         brelse(eb_bh);
2895         return ret;
2896 }
2897
2898 typedef int (xattr_bucket_func)(struct inode *inode,
2899                                 struct ocfs2_xattr_bucket *bucket,
2900                                 void *para);
2901
2902 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2903                                    struct ocfs2_xattr_bucket *bucket,
2904                                    int name_index,
2905                                    const char *name,
2906                                    u32 name_hash,
2907                                    u16 *xe_index,
2908                                    int *found)
2909 {
2910         int i, ret = 0, cmp = 1, block_off, new_offset;
2911         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2912         size_t name_len = strlen(name);
2913         struct ocfs2_xattr_entry *xe = NULL;
2914         char *xe_name;
2915
2916         /*
2917          * We don't use binary search in the bucket because there
2918          * may be multiple entries with the same name hash.
2919          */
2920         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2921                 xe = &xh->xh_entries[i];
2922
2923                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2924                         continue;
2925                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2926                         break;
2927
2928                 cmp = name_index - ocfs2_xattr_get_type(xe);
2929                 if (!cmp)
2930                         cmp = name_len - xe->xe_name_len;
2931                 if (cmp)
2932                         continue;
2933
2934                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2935                                                         xh,
2936                                                         i,
2937                                                         &block_off,
2938                                                         &new_offset);
2939                 if (ret) {
2940                         mlog_errno(ret);
2941                         break;
2942                 }
2943
2944
2945                 xe_name = bucket_block(bucket, block_off) + new_offset;
2946                 if (!memcmp(name, xe_name, name_len)) {
2947                         *xe_index = i;
2948                         *found = 1;
2949                         ret = 0;
2950                         break;
2951                 }
2952         }
2953
2954         return ret;
2955 }
2956
2957 /*
2958  * Find the specified xattr entry in a series of buckets.
2959  * This series start from p_blkno and last for num_clusters.
2960  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2961  * the num of the valid buckets.
2962  *
2963  * Return the buffer_head this xattr should reside in. And if the xattr's
2964  * hash is in the gap of 2 buckets, return the lower bucket.
2965  */
2966 static int ocfs2_xattr_bucket_find(struct inode *inode,
2967                                    int name_index,
2968                                    const char *name,
2969                                    u32 name_hash,
2970                                    u64 p_blkno,
2971                                    u32 first_hash,
2972                                    u32 num_clusters,
2973                                    struct ocfs2_xattr_search *xs)
2974 {
2975         int ret, found = 0;
2976         struct ocfs2_xattr_header *xh = NULL;
2977         struct ocfs2_xattr_entry *xe = NULL;
2978         u16 index = 0;
2979         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2980         int low_bucket = 0, bucket, high_bucket;
2981         struct ocfs2_xattr_bucket *search;
2982         u32 last_hash;
2983         u64 blkno, lower_blkno = 0;
2984
2985         search = ocfs2_xattr_bucket_new(inode);
2986         if (!search) {
2987                 ret = -ENOMEM;
2988                 mlog_errno(ret);
2989                 goto out;
2990         }
2991
2992         ret = ocfs2_read_xattr_bucket(search, p_blkno);
2993         if (ret) {
2994                 mlog_errno(ret);
2995                 goto out;
2996         }
2997
2998         xh = bucket_xh(search);
2999         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3000         while (low_bucket <= high_bucket) {
3001                 ocfs2_xattr_bucket_relse(search);
3002
3003                 bucket = (low_bucket + high_bucket) / 2;
3004                 blkno = p_blkno + bucket * blk_per_bucket;
3005                 ret = ocfs2_read_xattr_bucket(search, blkno);
3006                 if (ret) {
3007                         mlog_errno(ret);
3008                         goto out;
3009                 }
3010
3011                 xh = bucket_xh(search);
3012                 xe = &xh->xh_entries[0];
3013                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3014                         high_bucket = bucket - 1;
3015                         continue;
3016                 }
3017
3018                 /*
3019                  * Check whether the hash of the last entry in our
3020                  * bucket is larger than the search one. for an empty
3021                  * bucket, the last one is also the first one.
3022                  */
3023                 if (xh->xh_count)
3024                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3025
3026                 last_hash = le32_to_cpu(xe->xe_name_hash);
3027
3028                 /* record lower_blkno which may be the insert place. */
3029                 lower_blkno = blkno;
3030
3031                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3032                         low_bucket = bucket + 1;
3033                         continue;
3034                 }
3035
3036                 /* the searched xattr should reside in this bucket if exists. */
3037                 ret = ocfs2_find_xe_in_bucket(inode, search,
3038                                               name_index, name, name_hash,
3039                                               &index, &found);
3040                 if (ret) {
3041                         mlog_errno(ret);
3042                         goto out;
3043                 }
3044                 break;
3045         }
3046
3047         /*
3048          * Record the bucket we have found.
3049          * When the xattr's hash value is in the gap of 2 buckets, we will
3050          * always set it to the previous bucket.
3051          */
3052         if (!lower_blkno)
3053                 lower_blkno = p_blkno;
3054
3055         /* This should be in cache - we just read it during the search */
3056         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3057         if (ret) {
3058                 mlog_errno(ret);
3059                 goto out;
3060         }
3061
3062         xs->header = bucket_xh(xs->bucket);
3063         xs->base = bucket_block(xs->bucket, 0);
3064         xs->end = xs->base + inode->i_sb->s_blocksize;
3065
3066         if (found) {
3067                 xs->here = &xs->header->xh_entries[index];
3068                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3069                      (unsigned long long)bucket_blkno(xs->bucket), index);
3070         } else
3071                 ret = -ENODATA;
3072
3073 out:
3074         ocfs2_xattr_bucket_free(search);
3075         return ret;
3076 }
3077
3078 static int ocfs2_xattr_index_block_find(struct inode *inode,
3079                                         struct buffer_head *root_bh,
3080                                         int name_index,
3081                                         const char *name,
3082                                         struct ocfs2_xattr_search *xs)
3083 {
3084         int ret;
3085         struct ocfs2_xattr_block *xb =
3086                         (struct ocfs2_xattr_block *)root_bh->b_data;
3087         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3088         struct ocfs2_extent_list *el = &xb_root->xt_list;
3089         u64 p_blkno = 0;
3090         u32 first_hash, num_clusters = 0;
3091         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3092
3093         if (le16_to_cpu(el->l_next_free_rec) == 0)
3094                 return -ENODATA;
3095
3096         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3097              name, name_hash, name_index);
3098
3099         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3100                                   &num_clusters, el);
3101         if (ret) {
3102                 mlog_errno(ret);
3103                 goto out;
3104         }
3105
3106         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3107
3108         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3109              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3110              first_hash);
3111
3112         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3113                                       p_blkno, first_hash, num_clusters, xs);
3114
3115 out:
3116         return ret;
3117 }
3118
3119 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3120                                        u64 blkno,
3121                                        u32 clusters,
3122                                        xattr_bucket_func *func,
3123                                        void *para)
3124 {
3125         int i, ret = 0;
3126         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3127         u32 num_buckets = clusters * bpc;
3128         struct ocfs2_xattr_bucket *bucket;
3129
3130         bucket = ocfs2_xattr_bucket_new(inode);
3131         if (!bucket) {
3132                 mlog_errno(-ENOMEM);
3133                 return -ENOMEM;
3134         }
3135
3136         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3137              clusters, (unsigned long long)blkno);
3138
3139         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3140                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3141                 if (ret) {
3142                         mlog_errno(ret);
3143                         break;
3144                 }
3145
3146                 /*
3147                  * The real bucket num in this series of blocks is stored
3148                  * in the 1st bucket.
3149                  */
3150                 if (i == 0)
3151                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3152
3153                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3154                      (unsigned long long)blkno,
3155                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3156                 if (func) {
3157                         ret = func(inode, bucket, para);
3158                         if (ret && ret != -ERANGE)
3159                                 mlog_errno(ret);
3160                         /* Fall through to bucket_relse() */
3161                 }
3162
3163                 ocfs2_xattr_bucket_relse(bucket);
3164                 if (ret)
3165                         break;
3166         }
3167
3168         ocfs2_xattr_bucket_free(bucket);
3169         return ret;
3170 }
3171
3172 struct ocfs2_xattr_tree_list {
3173         char *buffer;
3174         size_t buffer_size;
3175         size_t result;
3176 };
3177
3178 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3179                                              struct ocfs2_xattr_header *xh,
3180                                              int index,
3181                                              int *block_off,
3182                                              int *new_offset)
3183 {
3184         u16 name_offset;
3185
3186         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3187                 return -EINVAL;
3188
3189         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3190
3191         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3192         *new_offset = name_offset % inode->i_sb->s_blocksize;
3193
3194         return 0;
3195 }
3196
3197 static int ocfs2_list_xattr_bucket(struct inode *inode,
3198                                    struct ocfs2_xattr_bucket *bucket,
3199                                    void *para)
3200 {
3201         int ret = 0, type;
3202         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3203         int i, block_off, new_offset;
3204         const char *prefix, *name;
3205
3206         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3207                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3208                 type = ocfs2_xattr_get_type(entry);
3209                 prefix = ocfs2_xattr_prefix(type);
3210
3211                 if (prefix) {
3212                         ret = ocfs2_xattr_bucket_get_name_value(inode,
3213                                                                 bucket_xh(bucket),
3214                                                                 i,
3215                                                                 &block_off,
3216                                                                 &new_offset);
3217                         if (ret)
3218                                 break;
3219
3220                         name = (const char *)bucket_block(bucket, block_off) +
3221                                 new_offset;
3222                         ret = ocfs2_xattr_list_entry(xl->buffer,
3223                                                      xl->buffer_size,
3224                                                      &xl->result,
3225                                                      prefix, name,
3226                                                      entry->xe_name_len);
3227                         if (ret)
3228                                 break;
3229                 }
3230         }
3231
3232         return ret;
3233 }
3234
3235 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3236                                              struct ocfs2_xattr_tree_root *xt,
3237                                              char *buffer,
3238                                              size_t buffer_size)
3239 {
3240         struct ocfs2_extent_list *el = &xt->xt_list;
3241         int ret = 0;
3242         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3243         u64 p_blkno = 0;
3244         struct ocfs2_xattr_tree_list xl = {
3245                 .buffer = buffer,
3246                 .buffer_size = buffer_size,
3247                 .result = 0,
3248         };
3249
3250         if (le16_to_cpu(el->l_next_free_rec) == 0)
3251                 return 0;
3252
3253         while (name_hash > 0) {
3254                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3255                                           &e_cpos, &num_clusters, el);
3256                 if (ret) {
3257                         mlog_errno(ret);
3258                         goto out;
3259                 }
3260
3261                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3262                                                   ocfs2_list_xattr_bucket,
3263                                                   &xl);
3264                 if (ret) {
3265                         if (ret != -ERANGE)
3266                                 mlog_errno(ret);
3267                         goto out;
3268                 }
3269
3270                 if (e_cpos == 0)
3271                         break;
3272
3273                 name_hash = e_cpos - 1;
3274         }
3275
3276         ret = xl.result;
3277 out:
3278         return ret;
3279 }
3280
3281 static int cmp_xe(const void *a, const void *b)
3282 {
3283         const struct ocfs2_xattr_entry *l = a, *r = b;
3284         u32 l_hash = le32_to_cpu(l->xe_name_hash);
3285         u32 r_hash = le32_to_cpu(r->xe_name_hash);
3286
3287         if (l_hash > r_hash)
3288                 return 1;
3289         if (l_hash < r_hash)
3290                 return -1;
3291         return 0;
3292 }
3293
3294 static void swap_xe(void *a, void *b, int size)
3295 {
3296         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3297
3298         tmp = *l;
3299         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3300         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3301 }
3302
3303 /*
3304  * When the ocfs2_xattr_block is filled up, new bucket will be created
3305  * and all the xattr entries will be moved to the new bucket.
3306  * The header goes at the start of the bucket, and the names+values are
3307  * filled from the end.  This is why *target starts as the last buffer.
3308  * Note: we need to sort the entries since they are not saved in order
3309  * in the ocfs2_xattr_block.
3310  */
3311 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3312                                            struct buffer_head *xb_bh,
3313                                            struct ocfs2_xattr_bucket *bucket)
3314 {
3315         int i, blocksize = inode->i_sb->s_blocksize;
3316         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3317         u16 offset, size, off_change;
3318         struct ocfs2_xattr_entry *xe;
3319         struct ocfs2_xattr_block *xb =
3320                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
3321         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3322         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3323         u16 count = le16_to_cpu(xb_xh->xh_count);
3324         char *src = xb_bh->b_data;
3325         char *target = bucket_block(bucket, blks - 1);
3326
3327         mlog(0, "cp xattr from block %llu to bucket %llu\n",
3328              (unsigned long long)xb_bh->b_blocknr,
3329              (unsigned long long)bucket_blkno(bucket));
3330
3331         for (i = 0; i < blks; i++)
3332                 memset(bucket_block(bucket, i), 0, blocksize);
3333
3334         /*
3335          * Since the xe_name_offset is based on ocfs2_xattr_header,
3336          * there is a offset change corresponding to the change of
3337          * ocfs2_xattr_header's position.
3338          */
3339         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3340         xe = &xb_xh->xh_entries[count - 1];
3341         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3342         size = blocksize - offset;
3343
3344         /* copy all the names and values. */
3345         memcpy(target + offset, src + offset, size);
3346
3347         /* Init new header now. */
3348         xh->xh_count = xb_xh->xh_count;
3349         xh->xh_num_buckets = cpu_to_le16(1);
3350         xh->xh_name_value_len = cpu_to_le16(size);
3351         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3352
3353         /* copy all the entries. */
3354         target = bucket_block(bucket, 0);
3355         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3356         size = count * sizeof(struct ocfs2_xattr_entry);
3357         memcpy(target + offset, (char *)xb_xh + offset, size);
3358
3359         /* Change the xe offset for all the xe because of the move. */
3360         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3361                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3362         for (i = 0; i < count; i++)
3363                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3364
3365         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3366              offset, size, off_change);
3367
3368         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3369              cmp_xe, swap_xe);
3370 }
3371
3372 /*
3373  * After we move xattr from block to index btree, we have to
3374  * update ocfs2_xattr_search to the new xe and base.
3375  *
3376  * When the entry is in xattr block, xattr_bh indicates the storage place.
3377  * While if the entry is in index b-tree, "bucket" indicates the
3378  * real place of the xattr.
3379  */
3380 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3381                                             struct ocfs2_xattr_search *xs,
3382                                             struct buffer_head *old_bh)
3383 {
3384         char *buf = old_bh->b_data;
3385         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3386         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3387         int i;
3388
3389         xs->header = bucket_xh(xs->bucket);
3390         xs->base = bucket_block(xs->bucket, 0);
3391         xs->end = xs->base + inode->i_sb->s_blocksize;
3392
3393         if (xs->not_found)
3394                 return;
3395
3396         i = xs->here - old_xh->xh_entries;
3397         xs->here = &xs->header->xh_entries[i];
3398 }
3399
3400 static int ocfs2_xattr_create_index_block(struct inode *inode,
3401                                           struct ocfs2_xattr_search *xs,
3402                                           struct ocfs2_xattr_set_ctxt *ctxt)
3403 {
3404         int ret;
3405         u32 bit_off, len;
3406         u64 blkno;
3407         handle_t *handle = ctxt->handle;
3408         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3409         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3410         struct buffer_head *xb_bh = xs->xattr_bh;
3411         struct ocfs2_xattr_block *xb =
3412                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3413         struct ocfs2_xattr_tree_root *xr;
3414         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3415
3416         mlog(0, "create xattr index block for %llu\n",
3417              (unsigned long long)xb_bh->b_blocknr);
3418
3419         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3420         BUG_ON(!xs->bucket);
3421
3422         /*
3423          * XXX:
3424          * We can use this lock for now, and maybe move to a dedicated mutex
3425          * if performance becomes a problem later.
3426          */
3427         down_write(&oi->ip_alloc_sem);
3428
3429         ret = ocfs2_journal_access_xb(handle, inode, xb_bh,
3430                                       OCFS2_JOURNAL_ACCESS_WRITE);
3431         if (ret) {
3432                 mlog_errno(ret);
3433                 goto out;
3434         }
3435
3436         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3437                                      1, 1, &bit_off, &len);
3438         if (ret) {
3439                 mlog_errno(ret);
3440                 goto out;
3441         }
3442
3443         /*
3444          * The bucket may spread in many blocks, and
3445          * we will only touch the 1st block and the last block
3446          * in the whole bucket(one for entry and one for data).
3447          */
3448         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3449
3450         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3451              (unsigned long long)blkno);
3452
3453         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3454         if (ret) {
3455                 mlog_errno(ret);
3456                 goto out;
3457         }
3458
3459         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3460                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3461         if (ret) {
3462                 mlog_errno(ret);
3463                 goto out;
3464         }
3465
3466         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3467         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3468
3469         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3470
3471         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3472         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3473                offsetof(struct ocfs2_xattr_block, xb_attrs));
3474
3475         xr = &xb->xb_attrs.xb_root;
3476         xr->xt_clusters = cpu_to_le32(1);
3477         xr->xt_last_eb_blk = 0;
3478         xr->xt_list.l_tree_depth = 0;
3479         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3480         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3481
3482         xr->xt_list.l_recs[0].e_cpos = 0;
3483         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3484         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3485
3486         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3487
3488         ocfs2_journal_dirty(handle, xb_bh);
3489
3490 out:
3491         up_write(&oi->ip_alloc_sem);
3492
3493         return ret;
3494 }
3495
3496 static int cmp_xe_offset(const void *a, const void *b)
3497 {
3498         const struct ocfs2_xattr_entry *l = a, *r = b;
3499         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3500         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3501
3502         if (l_name_offset < r_name_offset)
3503                 return 1;
3504         if (l_name_offset > r_name_offset)
3505                 return -1;
3506         return 0;
3507 }
3508
3509 /*
3510  * defrag a xattr bucket if we find that the bucket has some
3511  * holes beteen name/value pairs.
3512  * We will move all the name/value pairs to the end of the bucket
3513  * so that we can spare some space for insertion.
3514  */
3515 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3516                                      handle_t *handle,
3517                                      struct ocfs2_xattr_bucket *bucket)
3518 {
3519         int ret, i;
3520         size_t end, offset, len, value_len;
3521         struct ocfs2_xattr_header *xh;
3522         char *entries, *buf, *bucket_buf = NULL;
3523         u64 blkno = bucket_blkno(bucket);
3524         u16 xh_free_start;
3525         size_t blocksize = inode->i_sb->s_blocksize;
3526         struct ocfs2_xattr_entry *xe;
3527
3528         /*
3529          * In order to make the operation more efficient and generic,
3530          * we copy all the blocks into a contiguous memory and do the
3531          * defragment there, so if anything is error, we will not touch
3532          * the real block.
3533          */
3534         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3535         if (!bucket_buf) {
3536                 ret = -EIO;
3537                 goto out;
3538         }
3539
3540         buf = bucket_buf;
3541         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3542                 memcpy(buf, bucket_block(bucket, i), blocksize);
3543
3544         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3545                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3546         if (ret < 0) {
3547                 mlog_errno(ret);
3548                 goto out;
3549         }
3550
3551         xh = (struct ocfs2_xattr_header *)bucket_buf;
3552         entries = (char *)xh->xh_entries;
3553         xh_free_start = le16_to_cpu(xh->xh_free_start);
3554
3555         mlog(0, "adjust xattr bucket in %llu, count = %u, "
3556              "xh_free_start = %u, xh_name_value_len = %u.\n",
3557              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3558              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3559
3560         /*
3561          * sort all the entries by their offset.
3562          * the largest will be the first, so that we can
3563          * move them to the end one by one.
3564          */
3565         sort(entries, le16_to_cpu(xh->xh_count),
3566              sizeof(struct ocfs2_xattr_entry),
3567              cmp_xe_offset, swap_xe);
3568
3569         /* Move all name/values to the end of the bucket. */
3570         xe = xh->xh_entries;
3571         end = OCFS2_XATTR_BUCKET_SIZE;
3572         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3573                 offset = le16_to_cpu(xe->xe_name_offset);
3574                 if (ocfs2_xattr_is_local(xe))
3575                         value_len = OCFS2_XATTR_SIZE(
3576                                         le64_to_cpu(xe->xe_value_size));
3577                 else
3578                         value_len = OCFS2_XATTR_ROOT_SIZE;
3579                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3580
3581                 /*
3582                  * We must make sure that the name/value pair
3583                  * exist in the same block. So adjust end to
3584                  * the previous block end if needed.
3585                  */
3586                 if (((end - len) / blocksize !=
3587                         (end - 1) / blocksize))
3588                         end = end - end % blocksize;
3589
3590                 if (end > offset + len) {
3591                         memmove(bucket_buf + end - len,
3592                                 bucket_buf + offset, len);
3593                         xe->xe_name_offset = cpu_to_le16(end - len);
3594                 }
3595
3596                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3597                                 "bucket %llu\n", (unsigned long long)blkno);
3598
3599                 end -= len;
3600         }
3601
3602         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3603                         "bucket %llu\n", (unsigned long long)blkno);
3604
3605         if (xh_free_start == end)
3606                 goto out;
3607
3608         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3609         xh->xh_free_start = cpu_to_le16(end);
3610
3611         /* sort the entries by their name_hash. */
3612         sort(entries, le16_to_cpu(xh->xh_count),
3613              sizeof(struct ocfs2_xattr_entry),
3614              cmp_xe, swap_xe);
3615
3616         buf = bucket_buf;
3617         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3618                 memcpy(bucket_block(bucket, i), buf, blocksize);
3619         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3620
3621 out:
3622         kfree(bucket_buf);
3623         return ret;
3624 }
3625
3626 /*
3627  * prev_blkno points to the start of an existing extent.  new_blkno
3628  * points to a newly allocated extent.  Because we know each of our
3629  * clusters contains more than bucket, we can easily split one cluster
3630  * at a bucket boundary.  So we take the last cluster of the existing
3631  * extent and split it down the middle.  We move the last half of the
3632  * buckets in the last cluster of the existing extent over to the new
3633  * extent.
3634  *
3635  * first_bh is the buffer at prev_blkno so we can update the existing
3636  * extent's bucket count.  header_bh is the bucket were we were hoping
3637  * to insert our xattr.  If the bucket move places the target in the new
3638  * extent, we'll update first_bh and header_bh after modifying the old
3639  * extent.
3640  *
3641  * first_hash will be set as the 1st xe's name_hash in the new extent.
3642  */
3643 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3644                                                handle_t *handle,
3645                                                struct ocfs2_xattr_bucket *first,
3646                                                struct ocfs2_xattr_bucket *target,
3647                                                u64 new_blkno,
3648                                                u32 num_clusters,
3649                                                u32 *first_hash)
3650 {
3651         int ret;
3652         struct super_block *sb = inode->i_sb;
3653         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3654         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3655         int to_move = num_buckets / 2;
3656         u64 src_blkno;
3657         u64 last_cluster_blkno = bucket_blkno(first) +
3658                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3659
3660         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3661         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3662
3663         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3664              (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3665
3666         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3667                                      last_cluster_blkno, new_blkno,
3668                                      to_move, first_hash);
3669         if (ret) {
3670                 mlog_errno(ret);
3671                 goto out;
3672         }
3673
3674         /* This is the first bucket that got moved */
3675         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3676
3677         /*
3678          * If the target bucket was part of the moved buckets, we need to
3679          * update first and target.
3680          */
3681         if (bucket_blkno(target) >= src_blkno) {
3682                 /* Find the block for the new target bucket */
3683                 src_blkno = new_blkno +
3684                         (bucket_blkno(target) - src_blkno);
3685
3686                 ocfs2_xattr_bucket_relse(first);
3687                 ocfs2_xattr_bucket_relse(target);
3688
3689                 /*
3690                  * These shouldn't fail - the buffers are in the
3691                  * journal from ocfs2_cp_xattr_bucket().
3692                  */
3693                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3694                 if (ret) {
3695                         mlog_errno(ret);
3696                         goto out;
3697                 }
3698                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3699                 if (ret)
3700                         mlog_errno(ret);
3701
3702         }
3703
3704 out:
3705         return ret;
3706 }
3707
3708 /*
3709  * Find the suitable pos when we divide a bucket into 2.
3710  * We have to make sure the xattrs with the same hash value exist
3711  * in the same bucket.
3712  *
3713  * If this ocfs2_xattr_header covers more than one hash value, find a
3714  * place where the hash value changes.  Try to find the most even split.
3715  * The most common case is that all entries have different hash values,
3716  * and the first check we make will find a place to split.
3717  */
3718 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3719 {
3720         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3721         int count = le16_to_cpu(xh->xh_count);
3722         int delta, middle = count / 2;
3723
3724         /*
3725          * We start at the middle.  Each step gets farther away in both
3726          * directions.  We therefore hit the change in hash value
3727          * nearest to the middle.  Note that this loop does not execute for
3728          * count < 2.
3729          */
3730         for (delta = 0; delta < middle; delta++) {
3731                 /* Let's check delta earlier than middle */
3732                 if (cmp_xe(&entries[middle - delta - 1],
3733                            &entries[middle - delta]))
3734                         return middle - delta;
3735
3736                 /* For even counts, don't walk off the end */
3737                 if ((middle + delta + 1) == count)
3738                         continue;
3739
3740                 /* Now try delta past middle */
3741                 if (cmp_xe(&entries[middle + delta],
3742                            &entries[middle + delta + 1]))
3743                         return middle + delta + 1;
3744         }
3745
3746         /* Every entry had the same hash */
3747         return count;
3748 }
3749
3750 /*
3751  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3752  * first_hash will record the 1st hash of the new bucket.
3753  *
3754  * Normally half of the xattrs will be moved.  But we have to make
3755  * sure that the xattrs with the same hash value are stored in the
3756  * same bucket. If all the xattrs in this bucket have the same hash
3757  * value, the new bucket will be initialized as an empty one and the
3758  * first_hash will be initialized as (hash_value+1).
3759  */
3760 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3761                                     handle_t *handle,
3762                                     u64 blk,
3763                                     u64 new_blk,
3764                                     u32 *first_hash,
3765                                     int new_bucket_head)
3766 {
3767         int ret, i;
3768         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3769         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3770         struct ocfs2_xattr_header *xh;
3771         struct ocfs2_xattr_entry *xe;
3772         int blocksize = inode->i_sb->s_blocksize;
3773
3774         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3775              (unsigned long long)blk, (unsigned long long)new_blk);
3776
3777         s_bucket = ocfs2_xattr_bucket_new(inode);
3778         t_bucket = ocfs2_xattr_bucket_new(inode);
3779         if (!s_bucket || !t_bucket) {
3780                 ret = -ENOMEM;
3781                 mlog_errno(ret);
3782                 goto out;
3783         }
3784
3785         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3786         if (ret) {
3787                 mlog_errno(ret);
3788                 goto out;
3789         }
3790
3791         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3792                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3793         if (ret) {
3794                 mlog_errno(ret);
3795                 goto out;
3796         }
3797
3798         /*
3799          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3800          * there's no need to read it.
3801          */
3802         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3803         if (ret) {
3804                 mlog_errno(ret);
3805                 goto out;
3806         }
3807
3808         /*
3809          * Hey, if we're overwriting t_bucket, what difference does
3810          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
3811          * same part of ocfs2_cp_xattr_bucket().
3812          */
3813         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3814                                                 new_bucket_head ?
3815                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3816                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3817         if (ret) {
3818                 mlog_errno(ret);
3819                 goto out;
3820         }
3821
3822         xh = bucket_xh(s_bucket);
3823         count = le16_to_cpu(xh->xh_count);
3824         start = ocfs2_xattr_find_divide_pos(xh);
3825
3826         if (start == count) {
3827                 xe = &xh->xh_entries[start-1];
3828
3829                 /*
3830                  * initialized a new empty bucket here.
3831                  * The hash value is set as one larger than
3832                  * that of the last entry in the previous bucket.
3833                  */
3834                 for (i = 0; i < t_bucket->bu_blocks; i++)
3835                         memset(bucket_block(t_bucket, i), 0, blocksize);
3836
3837                 xh = bucket_xh(t_bucket);
3838                 xh->xh_free_start = cpu_to_le16(blocksize);
3839                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3840                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3841
3842                 goto set_num_buckets;
3843         }
3844
3845         /* copy the whole bucket to the new first. */
3846         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3847
3848         /* update the new bucket. */
3849         xh = bucket_xh(t_bucket);
3850
3851         /*
3852          * Calculate the total name/value len and xh_free_start for
3853          * the old bucket first.
3854          */
3855         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3856         name_value_len = 0;
3857         for (i = 0; i < start; i++) {
3858                 xe = &xh->xh_entries[i];
3859                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3860                 if (ocfs2_xattr_is_local(xe))
3861                         xe_len +=
3862                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3863                 else
3864                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3865                 name_value_len += xe_len;
3866                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3867                         name_offset = le16_to_cpu(xe->xe_name_offset);
3868         }
3869
3870         /*
3871          * Now begin the modification to the new bucket.
3872          *
3873          * In the new bucket, We just move the xattr entry to the beginning
3874          * and don't touch the name/value. So there will be some holes in the
3875          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3876          * called.
3877          */
3878         xe = &xh->xh_entries[start];
3879         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3880         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3881              (int)((char *)xe - (char *)xh),
3882              (int)((char *)xh->xh_entries - (char *)xh));
3883         memmove((char *)xh->xh_entries, (char *)xe, len);
3884         xe = &xh->xh_entries[count - start];
3885         len = sizeof(struct ocfs2_xattr_entry) * start;
3886         memset((char *)xe, 0, len);
3887
3888         le16_add_cpu(&xh->xh_count, -start);
3889         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3890
3891         /* Calculate xh_free_start for the new bucket. */
3892         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3893         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3894                 xe = &xh->xh_entries[i];
3895                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3896                 if (ocfs2_xattr_is_local(xe))
3897                         xe_len +=
3898                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3899                 else
3900                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3901                 if (le16_to_cpu(xe->xe_name_offset) <
3902                     le16_to_cpu(xh->xh_free_start))
3903                         xh->xh_free_start = xe->xe_name_offset;
3904         }
3905
3906 set_num_buckets:
3907         /* set xh->xh_num_buckets for the new xh. */
3908         if (new_bucket_head)
3909                 xh->xh_num_buckets = cpu_to_le16(1);
3910         else
3911                 xh->xh_num_buckets = 0;
3912
3913         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3914
3915         /* store the first_hash of the new bucket. */
3916         if (first_hash)
3917                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3918
3919         /*
3920          * Now only update the 1st block of the old bucket.  If we
3921          * just added a new empty bucket, there is no need to modify
3922          * it.
3923          */
3924         if (start == count)
3925                 goto out;
3926
3927         xh = bucket_xh(s_bucket);
3928         memset(&xh->xh_entries[start], 0,
3929                sizeof(struct ocfs2_xattr_entry) * (count - start));
3930         xh->xh_count = cpu_to_le16(start);
3931         xh->xh_free_start = cpu_to_le16(name_offset);
3932         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3933
3934         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3935
3936 out:
3937         ocfs2_xattr_bucket_free(s_bucket);
3938         ocfs2_xattr_bucket_free(t_bucket);
3939
3940         return ret;
3941 }
3942
3943 /*
3944  * Copy xattr from one bucket to another bucket.
3945  *
3946  * The caller must make sure that the journal transaction
3947  * has enough space for journaling.
3948  */
3949 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3950                                  handle_t *handle,
3951                                  u64 s_blkno,
3952                                  u64 t_blkno,
3953                                  int t_is_new)
3954 {
3955         int ret;
3956         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3957
3958         BUG_ON(s_blkno == t_blkno);
3959
3960         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3961              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3962              t_is_new);
3963
3964         s_bucket = ocfs2_xattr_bucket_new(inode);
3965         t_bucket = ocfs2_xattr_bucket_new(inode);
3966         if (!s_bucket || !t_bucket) {
3967                 ret = -ENOMEM;
3968                 mlog_errno(ret);
3969                 goto out;
3970         }
3971
3972         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3973         if (ret)
3974                 goto out;
3975
3976         /*
3977          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3978          * there's no need to read it.
3979          */
3980         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3981         if (ret)
3982                 goto out;
3983
3984         /*
3985          * Hey, if we're overwriting t_bucket, what difference does
3986          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
3987          * cluster to fill, we came here from
3988          * ocfs2_mv_xattr_buckets(), and it is really new -
3989          * ACCESS_CREATE is required.  But we also might have moved data
3990          * out of t_bucket before extending back into it.
3991          * ocfs2_add_new_xattr_bucket() can do this - its call to
3992          * ocfs2_add_new_xattr_cluster() may have created a new extent
3993          * and copied out the end of the old extent.  Then it re-extends
3994          * the old extent back to create space for new xattrs.  That's
3995          * how we get here, and the bucket isn't really new.
3996          */
3997         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3998                                                 t_is_new ?
3999                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4000                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4001         if (ret)
4002                 goto out;
4003
4004         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4005         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4006
4007 out:
4008         ocfs2_xattr_bucket_free(t_bucket);
4009         ocfs2_xattr_bucket_free(s_bucket);
4010
4011         return ret;
4012 }
4013
4014 /*
4015  * src_blk points to the start of an existing extent.  last_blk points to
4016  * last cluster in that extent.  to_blk points to a newly allocated
4017  * extent.  We copy the buckets from the cluster at last_blk to the new
4018  * extent.  If start_bucket is non-zero, we skip that many buckets before
4019  * we start copying.  The new extent's xh_num_buckets gets set to the
4020  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4021  * by the same amount.
4022  */
4023 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4024                                   u64 src_blk, u64 last_blk, u64 to_blk,
4025                                   unsigned int start_bucket,
4026                                   u32 *first_hash)
4027 {
4028         int i, ret, credits;
4029         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4030         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4031         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4032         struct ocfs2_xattr_bucket *old_first, *new_first;
4033
4034         mlog(0, "mv xattrs from cluster %llu to %llu\n",
4035              (unsigned long long)last_blk, (unsigned long long)to_blk);
4036
4037         BUG_ON(start_bucket >= num_buckets);
4038         if (start_bucket) {
4039                 num_buckets -= start_bucket;
4040                 last_blk += (start_bucket * blks_per_bucket);
4041         }
4042
4043         /* The first bucket of the original extent */
4044         old_first = ocfs2_xattr_bucket_new(inode);
4045         /* The first bucket of the new extent */
4046         new_first = ocfs2_xattr_bucket_new(inode);
4047         if (!old_first || !new_first) {
4048                 ret = -ENOMEM;
4049                 mlog_errno(ret);
4050                 goto out;
4051         }
4052
4053         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4054         if (ret) {
4055                 mlog_errno(ret);
4056                 goto out;
4057         }
4058
4059         /*
4060          * We need to update the first bucket of the old extent and all
4061          * the buckets going to the new extent.
4062          */
4063         credits = ((num_buckets + 1) * blks_per_bucket) +
4064                 handle->h_buffer_credits;
4065         ret = ocfs2_extend_trans(handle, credits);
4066         if (ret) {
4067                 mlog_errno(ret);
4068                 goto out;
4069         }
4070
4071         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4072                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4073         if (ret) {
4074                 mlog_errno(ret);
4075                 goto out;
4076         }
4077
4078         for (i = 0; i < num_buckets; i++) {
4079                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4080                                             last_blk + (i * blks_per_bucket),
4081                                             to_blk + (i * blks_per_bucket),
4082                                             1);
4083                 if (ret) {
4084                         mlog_errno(ret);
4085                         goto out;
4086                 }
4087         }
4088
4089         /*
4090          * Get the new bucket ready before we dirty anything
4091          * (This actually shouldn't fail, because we already dirtied
4092          * it once in ocfs2_cp_xattr_bucket()).
4093          */
4094         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4095         if (ret) {
4096                 mlog_errno(ret);
4097                 goto out;
4098         }
4099         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4100                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4101         if (ret) {
4102                 mlog_errno(ret);
4103                 goto out;
4104         }
4105
4106         /* Now update the headers */
4107         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4108         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4109
4110         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4111         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4112
4113         if (first_hash)
4114                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4115
4116 out:
4117         ocfs2_xattr_bucket_free(new_first);
4118         ocfs2_xattr_bucket_free(old_first);
4119         return ret;
4120 }
4121
4122 /*
4123  * Move some xattrs in this cluster to the new cluster.
4124  * This function should only be called when bucket size == cluster size.
4125  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4126  */
4127 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4128                                       handle_t *handle,
4129                                       u64 prev_blk,
4130                                       u64 new_blk,
4131                                       u32 *first_hash)
4132 {
4133         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4134         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4135
4136         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4137
4138         ret = ocfs2_extend_trans(handle, credits);
4139         if (ret) {
4140                 mlog_errno(ret);
4141                 return ret;
4142         }
4143
4144         /* Move half of the xattr in start_blk to the next bucket. */
4145         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4146                                           new_blk, first_hash, 1);
4147 }
4148
4149 /*
4150  * Move some xattrs from the old cluster to the new one since they are not
4151  * contiguous in ocfs2 xattr tree.
4152  *
4153  * new_blk starts a new separate cluster, and we will move some xattrs from
4154  * prev_blk to it. v_start will be set as the first name hash value in this
4155  * new cluster so that it can be used as e_cpos during tree insertion and
4156  * don't collide with our original b-tree operations. first_bh and header_bh
4157  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4158  * to extend the insert bucket.
4159  *
4160  * The problem is how much xattr should we move to the new one and when should
4161  * we update first_bh and header_bh?
4162  * 1. If cluster size > bucket size, that means the previous cluster has more
4163  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4164  *    update the first_bh and header_bh if the insert bucket has been moved
4165  *    to the new cluster.
4166  * 2. If cluster_size == bucket_size:
4167  *    a) If the previous extent rec has more than one cluster and the insert
4168  *       place isn't in the last cluster, copy the entire last cluster to the
4169  *       new one. This time, we don't need to upate the first_bh and header_bh
4170  *       since they will not be moved into the new cluster.
4171  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4172  *       the new one. And we set the extend flag to zero if the insert place is
4173  *       moved into the new allocated cluster since no extend is needed.
4174  */
4175 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4176                                             handle_t *handle,
4177                                             struct ocfs2_xattr_bucket *first,
4178                                             struct ocfs2_xattr_bucket *target,
4179                                             u64 new_blk,
4180                                             u32 prev_clusters,
4181                                             u32 *v_start,
4182                                             int *extend)
4183 {
4184         int ret;
4185
4186         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4187              (unsigned long long)bucket_blkno(first), prev_clusters,
4188              (unsigned long long)new_blk);
4189
4190         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4191                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4192                                                           handle,
4193                                                           first, target,
4194                                                           new_blk,
4195                                                           prev_clusters,
4196                                                           v_start);
4197                 if (ret)
4198                         mlog_errno(ret);
4199         } else {
4200                 /* The start of the last cluster in the first extent */
4201                 u64 last_blk = bucket_blkno(first) +
4202                         ((prev_clusters - 1) *
4203                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
4204
4205                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4206                         ret = ocfs2_mv_xattr_buckets(inode, handle,
4207                                                      bucket_blkno(first),
4208                                                      last_blk, new_blk, 0,
4209                                                      v_start);
4210                         if (ret)
4211                                 mlog_errno(ret);
4212                 } else {
4213                         ret = ocfs2_divide_xattr_cluster(inode, handle,
4214                                                          last_blk, new_blk,
4215                                                          v_start);
4216                         if (ret)
4217                                 mlog_errno(ret);
4218
4219                         if ((bucket_blkno(target) == last_blk) && extend)
4220                                 *extend = 0;
4221                 }
4222         }
4223
4224         return ret;
4225 }
4226
4227 /*
4228  * Add a new cluster for xattr storage.
4229  *
4230  * If the new cluster is contiguous with the previous one, it will be
4231  * appended to the same extent record, and num_clusters will be updated.
4232  * If not, we will insert a new extent for it and move some xattrs in
4233  * the last cluster into the new allocated one.
4234  * We also need to limit the maximum size of a btree leaf, otherwise we'll
4235  * lose the benefits of hashing because we'll have to search large leaves.
4236  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4237  * if it's bigger).
4238  *
4239  * first_bh is the first block of the previous extent rec and header_bh
4240  * indicates the bucket we will insert the new xattrs. They will be updated
4241  * when the header_bh is moved into the new cluster.
4242  */
4243 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4244                                        struct buffer_head *root_bh,
4245                                        struct ocfs2_xattr_bucket *first,
4246                                        struct ocfs2_xattr_bucket *target,
4247                                        u32 *num_clusters,
4248                                        u32 prev_cpos,
4249                                        int *extend,
4250                                        struct ocfs2_xattr_set_ctxt *ctxt)
4251 {
4252         int ret;
4253         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4254         u32 prev_clusters = *num_clusters;
4255         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4256         u64 block;
4257         handle_t *handle = ctxt->handle;
4258         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4259         struct ocfs2_extent_tree et;
4260
4261         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4262              "previous xattr blkno = %llu\n",
4263              (unsigned long long)OCFS2_I(inode)->ip_blkno,
4264              prev_cpos, (unsigned long long)bucket_blkno(first));
4265
4266         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4267
4268         ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4269                                       OCFS2_JOURNAL_ACCESS_WRITE);
4270         if (ret < 0) {
4271                 mlog_errno(ret);
4272                 goto leave;
4273         }
4274
4275         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4276                                      clusters_to_add, &bit_off, &num_bits);
4277         if (ret < 0) {
4278                 if (ret != -ENOSPC)
4279                         mlog_errno(ret);
4280                 goto leave;
4281         }
4282
4283         BUG_ON(num_bits > clusters_to_add);
4284
4285         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4286         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4287              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4288
4289         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4290             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4291              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4292                 /*
4293                  * If this cluster is contiguous with the old one and
4294                  * adding this new cluster, we don't surpass the limit of
4295                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4296                  * initialized and used like other buckets in the previous
4297                  * cluster.
4298                  * So add it as a contiguous one. The caller will handle
4299                  * its init process.
4300                  */
4301                 v_start = prev_cpos + prev_clusters;
4302                 *num_clusters = prev_clusters + num_bits;
4303                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4304                      num_bits);
4305         } else {
4306                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4307                                                        handle,
4308                                                        first,
4309                                                        target,
4310                                                        block,
4311                                                        prev_clusters,
4312                                                        &v_start,
4313                                                        extend);
4314                 if (ret) {
4315                         mlog_errno(ret);
4316                         goto leave;
4317                 }
4318         }
4319
4320         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4321              num_bits, (unsigned long long)block, v_start);
4322         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
4323                                   num_bits, 0, ctxt->meta_ac);
4324         if (ret < 0) {
4325                 mlog_errno(ret);
4326                 goto leave;
4327         }
4328
4329         ret = ocfs2_journal_dirty(handle, root_bh);
4330         if (ret < 0)
4331                 mlog_errno(ret);
4332
4333 leave:
4334         return ret;
4335 }
4336
4337 /*
4338  * We are given an extent.  'first' is the bucket at the very front of
4339  * the extent.  The extent has space for an additional bucket past
4340  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
4341  * of the target bucket.  We wish to shift every bucket past the target
4342  * down one, filling in that additional space.  When we get back to the
4343  * target, we split the target between itself and the now-empty bucket
4344  * at target+1 (aka, target_blkno + blks_per_bucket).
4345  */
4346 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4347                                      handle_t *handle,
4348                                      struct ocfs2_xattr_bucket *first,
4349                                      u64 target_blk,
4350                                      u32 num_clusters)
4351 {
4352         int ret, credits;
4353         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4354         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4355         u64 end_blk;
4356         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4357
4358         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4359              "from %llu, len = %u\n", (unsigned long long)target_blk,
4360              (unsigned long long)bucket_blkno(first), num_clusters);
4361
4362         /* The extent must have room for an additional bucket */
4363         BUG_ON(new_bucket >=
4364                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4365
4366         /* end_blk points to the last existing bucket */
4367         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4368
4369         /*
4370          * end_blk is the start of the last existing bucket.
4371          * Thus, (end_blk - target_blk) covers the target bucket and
4372          * every bucket after it up to, but not including, the last
4373          * existing bucket.  Then we add the last existing bucket, the
4374          * new bucket, and the first bucket (3 * blk_per_bucket).
4375          */
4376         credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4377                   handle->h_buffer_credits;
4378         ret = ocfs2_extend_trans(handle, credits);
4379         if (ret) {
4380                 mlog_errno(ret);
4381                 goto out;
4382         }
4383
4384         ret = ocfs2_xattr_bucket_journal_access(handle, first,
4385                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4386         if (ret) {
4387                 mlog_errno(ret);
4388                 goto out;
4389         }
4390
4391         while (end_blk != target_blk) {
4392                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4393                                             end_blk + blk_per_bucket, 0);
4394                 if (ret)
4395                         goto out;
4396                 end_blk -= blk_per_bucket;
4397         }
4398
4399         /* Move half of the xattr in target_blkno to the next bucket. */
4400         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4401                                         target_blk + blk_per_bucket, NULL, 0);
4402
4403         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4404         ocfs2_xattr_bucket_journal_dirty(handle, first);
4405
4406 out:
4407         return ret;
4408 }
4409
4410 /*
4411  * Add new xattr bucket in an extent record and adjust the buckets
4412  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
4413  * bucket we want to insert into.
4414  *
4415  * In the easy case, we will move all the buckets after target down by
4416  * one. Half of target's xattrs will be moved to the next bucket.
4417  *
4418  * If current cluster is full, we'll allocate a new one.  This may not
4419  * be contiguous.  The underlying calls will make sure that there is
4420  * space for the insert, shifting buckets around if necessary.
4421  * 'target' may be moved by those calls.
4422  */
4423 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4424                                       struct buffer_head *xb_bh,
4425                                       struct ocfs2_xattr_bucket *target,
4426                                       struct ocfs2_xattr_set_ctxt *ctxt)
4427 {
4428         struct ocfs2_xattr_block *xb =
4429                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4430         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4431         struct ocfs2_extent_list *el = &xb_root->xt_list;
4432         u32 name_hash =
4433                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4434         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4435         int ret, num_buckets, extend = 1;
4436         u64 p_blkno;
4437         u32 e_cpos, num_clusters;
4438         /* The bucket at the front of the extent */
4439         struct ocfs2_xattr_bucket *first;
4440
4441         mlog(0, "Add new xattr bucket starting from %llu\n",
4442              (unsigned long long)bucket_blkno(target));
4443
4444         /* The first bucket of the original extent */
4445         first = ocfs2_xattr_bucket_new(inode);
4446         if (!first) {
4447                 ret = -ENOMEM;
4448                 mlog_errno(ret);
4449                 goto out;
4450         }
4451
4452         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4453                                   &num_clusters, el);
4454         if (ret) {
4455                 mlog_errno(ret);
4456                 goto out;
4457         }
4458
4459         ret = ocfs2_read_xattr_bucket(first, p_blkno);
4460         if (ret) {
4461                 mlog_errno(ret);
4462                 goto out;
4463         }
4464
4465         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4466         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4467                 /*
4468                  * This can move first+target if the target bucket moves
4469                  * to the new extent.
4470                  */
4471                 ret = ocfs2_add_new_xattr_cluster(inode,
4472                                                   xb_bh,
4473                                                   first,
4474                                                   target,
4475                                                   &num_clusters,
4476                                                   e_cpos,
4477                                                   &extend,
4478                                                   ctxt);
4479                 if (ret) {
4480                         mlog_errno(ret);
4481                         goto out;
4482                 }
4483         }
4484
4485         if (extend) {
4486                 ret = ocfs2_extend_xattr_bucket(inode,
4487                                                 ctxt->handle,
4488                                                 first,
4489                                                 bucket_blkno(target),
4490                                                 num_clusters);
4491                 if (ret)
4492                         mlog_errno(ret);
4493         }
4494
4495 out:
4496         ocfs2_xattr_bucket_free(first);
4497
4498         return ret;
4499 }
4500
4501 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4502                                         struct ocfs2_xattr_bucket *bucket,
4503                                         int offs)
4504 {
4505         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4506
4507         offs = offs % inode->i_sb->s_blocksize;
4508         return bucket_block(bucket, block_off) + offs;
4509 }
4510
4511 /*
4512  * Handle the normal xattr set, including replace, delete and new.
4513  *
4514  * Note: "local" indicates the real data's locality. So we can't
4515  * just its bucket locality by its length.
4516  */
4517 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4518                                          struct ocfs2_xattr_info *xi,
4519                                          struct ocfs2_xattr_search *xs,
4520                                          u32 name_hash,
4521                                          int local)
4522 {
4523         struct ocfs2_xattr_entry *last, *xe;
4524         int name_len = strlen(xi->name);
4525         struct ocfs2_xattr_header *xh = xs->header;
4526         u16 count = le16_to_cpu(xh->xh_count), start;
4527         size_t blocksize = inode->i_sb->s_blocksize;
4528         char *val;
4529         size_t offs, size, new_size;
4530
4531         last = &xh->xh_entries[count];
4532         if (!xs->not_found) {
4533                 xe = xs->here;
4534                 offs = le16_to_cpu(xe->xe_name_offset);
4535                 if (ocfs2_xattr_is_local(xe))
4536                         size = OCFS2_XATTR_SIZE(name_len) +
4537                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4538                 else
4539                         size = OCFS2_XATTR_SIZE(name_len) +
4540                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4541
4542                 /*
4543                  * If the new value will be stored outside, xi->value has been
4544                  * initalized as an empty ocfs2_xattr_value_root, and the same
4545                  * goes with xi->value_len, so we can set new_size safely here.
4546                  * See ocfs2_xattr_set_in_bucket.
4547                  */
4548                 new_size = OCFS2_XATTR_SIZE(name_len) +
4549                            OCFS2_XATTR_SIZE(xi->value_len);
4550
4551                 le16_add_cpu(&xh->xh_name_value_len, -size);
4552                 if (xi->value) {
4553                         if (new_size > size)
4554                                 goto set_new_name_value;
4555
4556                         /* Now replace the old value with new one. */
4557                         if (local)
4558                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4559                         else
4560                                 xe->xe_value_size = 0;
4561
4562                         val = ocfs2_xattr_bucket_get_val(inode,
4563                                                          xs->bucket, offs);
4564                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4565                                size - OCFS2_XATTR_SIZE(name_len));
4566                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4567                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4568                                        xi->value, xi->value_len);
4569
4570                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4571                         ocfs2_xattr_set_local(xe, local);
4572                         return;
4573                 } else {
4574                         /*
4575                          * Remove the old entry if there is more than one.
4576                          * We don't remove the last entry so that we can
4577                          * use it to indicate the hash value of the empty
4578                          * bucket.
4579                          */
4580                         last -= 1;
4581                         le16_add_cpu(&xh->xh_count, -1);
4582                         if (xh->xh_count) {
4583                                 memmove(xe, xe + 1,
4584                                         (void *)last - (void *)xe);
4585                                 memset(last, 0,
4586                                        sizeof(struct ocfs2_xattr_entry));
4587                         } else
4588                                 xh->xh_free_start =
4589                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4590
4591                         return;
4592                 }
4593         } else {
4594                 /* find a new entry for insert. */
4595                 int low = 0, high = count - 1, tmp;
4596                 struct ocfs2_xattr_entry *tmp_xe;
4597
4598                 while (low <= high && count) {
4599                         tmp = (low + high) / 2;
4600                         tmp_xe = &xh->xh_entries[tmp];
4601
4602                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4603                                 low = tmp + 1;
4604                         else if (name_hash <
4605                                  le32_to_cpu(tmp_xe->xe_name_hash))
4606                                 high = tmp - 1;
4607                         else {
4608                                 low = tmp;
4609                                 break;
4610                         }
4611                 }
4612
4613                 xe = &xh->xh_entries[low];
4614                 if (low != count)
4615                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4616
4617                 le16_add_cpu(&xh->xh_count, 1);
4618                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4619                 xe->xe_name_hash = cpu_to_le32(name_hash);
4620                 xe->xe_name_len = name_len;
4621                 ocfs2_xattr_set_type(xe, xi->name_index);
4622         }
4623
4624 set_new_name_value:
4625         /* Insert the new name+value. */
4626         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4627
4628         /*
4629          * We must make sure that the name/value pair
4630          * exists in the same block.
4631          */
4632         offs = le16_to_cpu(xh->xh_free_start);
4633         start = offs - size;
4634
4635         if (start >> inode->i_sb->s_blocksize_bits !=
4636             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4637                 offs = offs - offs % blocksize;
4638                 xh->xh_free_start = cpu_to_le16(offs);
4639         }
4640
4641         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4642         xe->xe_name_offset = cpu_to_le16(offs - size);
4643
4644         memset(val, 0, size);
4645         memcpy(val, xi->name, name_len);
4646         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4647
4648         xe->xe_value_size = cpu_to_le64(xi->value_len);
4649         ocfs2_xattr_set_local(xe, local);
4650         xs->here = xe;
4651         le16_add_cpu(&xh->xh_free_start, -size);
4652         le16_add_cpu(&xh->xh_name_value_len, size);
4653
4654         return;
4655 }
4656
4657 /*
4658  * Set the xattr entry in the specified bucket.
4659  * The bucket is indicated by xs->bucket and it should have the enough
4660  * space for the xattr insertion.
4661  */
4662 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4663                                            handle_t *handle,
4664                                            struct ocfs2_xattr_info *xi,
4665                                            struct ocfs2_xattr_search *xs,
4666                                            u32 name_hash,
4667                                            int local)
4668 {
4669         int ret;
4670         u64 blkno;
4671
4672         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4673              (unsigned long)xi->value_len, xi->name_index,
4674              (unsigned long long)bucket_blkno(xs->bucket));
4675
4676         if (!xs->bucket->bu_bhs[1]) {
4677                 blkno = bucket_blkno(xs->bucket);
4678                 ocfs2_xattr_bucket_relse(xs->bucket);
4679                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4680                 if (ret) {
4681                         mlog_errno(ret);
4682                         goto out;
4683                 }
4684         }
4685
4686         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4687                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4688         if (ret < 0) {
4689                 mlog_errno(ret);
4690                 goto out;
4691         }
4692
4693         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4694         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4695
4696 out:
4697         return ret;
4698 }
4699
4700 /*
4701  * Truncate the specified xe_off entry in xattr bucket.
4702  * bucket is indicated by header_bh and len is the new length.
4703  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4704  *
4705  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4706  */
4707 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4708                                              struct ocfs2_xattr_bucket *bucket,
4709                                              int xe_off,
4710                                              int len,
4711                                              struct ocfs2_xattr_set_ctxt *ctxt)
4712 {
4713         int ret, offset;
4714         u64 value_blk;
4715         struct ocfs2_xattr_entry *xe;
4716         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4717         size_t blocksize = inode->i_sb->s_blocksize;
4718         struct ocfs2_xattr_value_buf vb = {
4719                 .vb_access = ocfs2_journal_access,
4720         };
4721
4722         xe = &xh->xh_entries[xe_off];
4723
4724         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4725
4726         offset = le16_to_cpu(xe->xe_name_offset) +
4727                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4728
4729         value_blk = offset / blocksize;
4730
4731         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4732         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4733
4734         vb.vb_bh = bucket->bu_bhs[value_blk];
4735         BUG_ON(!vb.vb_bh);
4736
4737         vb.vb_xv = (struct ocfs2_xattr_value_root *)
4738                 (vb.vb_bh->b_data + offset % blocksize);
4739
4740         /*
4741          * From here on out we have to dirty the bucket.  The generic
4742          * value calls only modify one of the bucket's bhs, but we need
4743          * to send the bucket at once.  So if they error, they *could* have
4744          * modified something.  We have to assume they did, and dirty
4745          * the whole bucket.  This leaves us in a consistent state.
4746          */
4747         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4748              xe_off, (unsigned long long)bucket_blkno(bucket), len);
4749         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4750         if (ret) {
4751                 mlog_errno(ret);
4752                 goto out;
4753         }
4754
4755         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4756                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4757         if (ret) {
4758                 mlog_errno(ret);
4759                 goto out;
4760         }
4761
4762         xe->xe_value_size = cpu_to_le64(len);
4763
4764         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4765
4766 out:
4767         return ret;
4768 }
4769
4770 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4771                                         struct ocfs2_xattr_search *xs,
4772                                         int len,
4773                                         struct ocfs2_xattr_set_ctxt *ctxt)
4774 {
4775         int ret, offset;
4776         struct ocfs2_xattr_entry *xe = xs->here;
4777         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4778
4779         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4780
4781         offset = xe - xh->xh_entries;
4782         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4783                                                 offset, len, ctxt);
4784         if (ret)
4785                 mlog_errno(ret);
4786
4787         return ret;
4788 }
4789
4790 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4791                                                 handle_t *handle,
4792                                                 struct ocfs2_xattr_search *xs,
4793                                                 char *val,
4794                                                 int value_len)
4795 {
4796         int ret, offset, block_off;
4797         struct ocfs2_xattr_value_root *xv;
4798         struct ocfs2_xattr_entry *xe = xs->here;
4799         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4800         void *base;
4801
4802         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4803
4804         ret = ocfs2_xattr_bucket_get_name_value(inode, xh,
4805                                                 xe - xh->xh_entries,
4806                                                 &block_off,
4807                                                 &offset);
4808         if (ret) {
4809                 mlog_errno(ret);
4810                 goto out;
4811         }
4812
4813         base = bucket_block(xs->bucket, block_off);
4814         xv = (struct ocfs2_xattr_value_root *)(base + offset +
4815                  OCFS2_XATTR_SIZE(xe->xe_name_len));
4816
4817         ret = __ocfs2_xattr_set_value_outside(inode, handle,
4818                                               xv, val, value_len);
4819         if (ret)
4820                 mlog_errno(ret);
4821 out:
4822         return ret;
4823 }
4824
4825 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4826                                   struct buffer_head *root_bh,
4827                                   u64 blkno,
4828                                   u32 cpos,
4829                                   u32 len)
4830 {
4831         int ret;
4832         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4833         struct inode *tl_inode = osb->osb_tl_inode;
4834         handle_t *handle;
4835         struct ocfs2_xattr_block *xb =
4836                         (struct ocfs2_xattr_block *)root_bh->b_data;
4837         struct ocfs2_alloc_context *meta_ac = NULL;
4838         struct ocfs2_cached_dealloc_ctxt dealloc;
4839         struct ocfs2_extent_tree et;
4840
4841         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4842
4843         ocfs2_init_dealloc_ctxt(&dealloc);
4844
4845         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4846              cpos, len, (unsigned long long)blkno);
4847
4848         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4849
4850         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4851         if (ret) {
4852                 mlog_errno(ret);
4853                 return ret;
4854         }
4855
4856         mutex_lock(&tl_inode->i_mutex);
4857
4858         if (ocfs2_truncate_log_needs_flush(osb)) {
4859                 ret = __ocfs2_flush_truncate_log(osb);
4860                 if (ret < 0) {
4861                         mlog_errno(ret);
4862                         goto out;
4863                 }
4864         }
4865
4866         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4867         if (IS_ERR(handle)) {
4868                 ret = -ENOMEM;
4869                 mlog_errno(ret);
4870                 goto out;
4871         }
4872
4873         ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4874                                       OCFS2_JOURNAL_ACCESS_WRITE);
4875         if (ret) {
4876                 mlog_errno(ret);
4877                 goto out_commit;
4878         }
4879
4880         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4881                                   &dealloc);
4882         if (ret) {
4883                 mlog_errno(ret);
4884                 goto out_commit;
4885         }
4886
4887         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4888
4889         ret = ocfs2_journal_dirty(handle, root_bh);
4890         if (ret) {
4891                 mlog_errno(ret);
4892                 goto out_commit;
4893         }
4894
4895         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4896         if (ret)
4897                 mlog_errno(ret);
4898
4899 out_commit:
4900         ocfs2_commit_trans(osb, handle);
4901 out:
4902         ocfs2_schedule_truncate_log_flush(osb, 1);
4903
4904         mutex_unlock(&tl_inode->i_mutex);
4905
4906         if (meta_ac)
4907                 ocfs2_free_alloc_context(meta_ac);
4908
4909         ocfs2_run_deallocs(osb, &dealloc);
4910
4911         return ret;
4912 }
4913
4914 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4915                                          handle_t *handle,
4916                                          struct ocfs2_xattr_search *xs)
4917 {
4918         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4919         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4920                                                 le16_to_cpu(xh->xh_count) - 1];
4921         int ret = 0;
4922
4923         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4924                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4925         if (ret) {
4926                 mlog_errno(ret);
4927                 return;
4928         }
4929
4930         /* Remove the old entry. */
4931         memmove(xs->here, xs->here + 1,
4932                 (void *)last - (void *)xs->here);
4933         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4934         le16_add_cpu(&xh->xh_count, -1);
4935
4936         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4937 }
4938
4939 /*
4940  * Set the xattr name/value in the bucket specified in xs.
4941  *
4942  * As the new value in xi may be stored in the bucket or in an outside cluster,
4943  * we divide the whole process into 3 steps:
4944  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4945  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4946  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4947  * 4. If the clusters for the new outside value can't be allocated, we need
4948  *    to free the xattr we allocated in set.
4949  */
4950 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4951                                      struct ocfs2_xattr_info *xi,
4952                                      struct ocfs2_xattr_search *xs,
4953                                      struct ocfs2_xattr_set_ctxt *ctxt)
4954 {
4955         int ret, local = 1;
4956         size_t value_len;
4957         char *val = (char *)xi->value;
4958         struct ocfs2_xattr_entry *xe = xs->here;
4959         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4960                                               strlen(xi->name));
4961
4962         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4963                 /*
4964                  * We need to truncate the xattr storage first.
4965                  *
4966                  * If both the old and new value are stored to
4967                  * outside block, we only need to truncate
4968                  * the storage and then set the value outside.
4969                  *
4970                  * If the new value should be stored within block,
4971                  * we should free all the outside block first and
4972                  * the modification to the xattr block will be done
4973                  * by following steps.
4974                  */
4975                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4976                         value_len = xi->value_len;
4977                 else
4978                         value_len = 0;
4979
4980                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4981                                                            value_len,
4982                                                            ctxt);
4983                 if (ret)
4984                         goto out;
4985
4986                 if (value_len)
4987                         goto set_value_outside;
4988         }
4989
4990         value_len = xi->value_len;
4991         /* So we have to handle the inside block change now. */
4992         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4993                 /*
4994                  * If the new value will be stored outside of block,
4995                  * initalize a new empty value root and insert it first.
4996                  */
4997                 local = 0;
4998                 xi->value = &def_xv;
4999                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
5000         }
5001
5002         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
5003                                               name_hash, local);
5004         if (ret) {
5005                 mlog_errno(ret);
5006                 goto out;
5007         }
5008
5009         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
5010                 goto out;
5011
5012         /* allocate the space now for the outside block storage. */
5013         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5014                                                    value_len, ctxt);
5015         if (ret) {
5016                 mlog_errno(ret);
5017
5018                 if (xs->not_found) {
5019                         /*
5020                          * We can't allocate enough clusters for outside
5021                          * storage and we have allocated xattr already,
5022                          * so need to remove it.
5023                          */
5024                         ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
5025                 }
5026                 goto out;
5027         }
5028
5029 set_value_outside:
5030         ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5031                                                    xs, val, value_len);
5032 out:
5033         return ret;
5034 }
5035
5036 /*
5037  * check whether the xattr bucket is filled up with the same hash value.
5038  * If we want to insert the xattr with the same hash, return -ENOSPC.
5039  * If we want to insert a xattr with different hash value, go ahead
5040  * and ocfs2_divide_xattr_bucket will handle this.
5041  */
5042 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5043                                               struct ocfs2_xattr_bucket *bucket,
5044                                               const char *name)
5045 {
5046         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5047         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5048
5049         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5050                 return 0;
5051
5052         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5053             xh->xh_entries[0].xe_name_hash) {
5054                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5055                      "hash = %u\n",
5056                      (unsigned long long)bucket_blkno(bucket),
5057                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5058                 return -ENOSPC;
5059         }
5060
5061         return 0;
5062 }
5063
5064 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5065                                              struct ocfs2_xattr_info *xi,
5066                                              struct ocfs2_xattr_search *xs,
5067                                              struct ocfs2_xattr_set_ctxt *ctxt)
5068 {
5069         struct ocfs2_xattr_header *xh;
5070         struct ocfs2_xattr_entry *xe;
5071         u16 count, header_size, xh_free_start;
5072         int free, max_free, need, old;
5073         size_t value_size = 0, name_len = strlen(xi->name);
5074         size_t blocksize = inode->i_sb->s_blocksize;
5075         int ret, allocation = 0;
5076
5077         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5078
5079 try_again:
5080         xh = xs->header;
5081         count = le16_to_cpu(xh->xh_count);
5082         xh_free_start = le16_to_cpu(xh->xh_free_start);
5083         header_size = sizeof(struct ocfs2_xattr_header) +
5084                         count * sizeof(struct ocfs2_xattr_entry);
5085         max_free = OCFS2_XATTR_BUCKET_SIZE - header_size -
5086                 le16_to_cpu(xh->xh_name_value_len) - OCFS2_XATTR_HEADER_GAP;
5087
5088         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5089                         "of %u which exceed block size\n",
5090                         (unsigned long long)bucket_blkno(xs->bucket),
5091                         header_size);
5092
5093         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5094                 value_size = OCFS2_XATTR_ROOT_SIZE;
5095         else if (xi->value)
5096                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5097
5098         if (xs->not_found)
5099                 need = sizeof(struct ocfs2_xattr_entry) +
5100                         OCFS2_XATTR_SIZE(name_len) + value_size;
5101         else {
5102                 need = value_size + OCFS2_XATTR_SIZE(name_len);
5103
5104                 /*
5105                  * We only replace the old value if the new length is smaller
5106                  * than the old one. Otherwise we will allocate new space in the
5107                  * bucket to store it.
5108                  */
5109                 xe = xs->here;
5110                 if (ocfs2_xattr_is_local(xe))
5111                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5112                 else
5113                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5114
5115                 if (old >= value_size)
5116                         need = 0;
5117         }
5118
5119         free = xh_free_start - header_size - OCFS2_XATTR_HEADER_GAP;
5120         /*
5121          * We need to make sure the new name/value pair
5122          * can exist in the same block.
5123          */
5124         if (xh_free_start % blocksize < need)
5125                 free -= xh_free_start % blocksize;
5126
5127         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5128              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5129              " %u\n", xs->not_found,
5130              (unsigned long long)bucket_blkno(xs->bucket),
5131              free, need, max_free, le16_to_cpu(xh->xh_free_start),
5132              le16_to_cpu(xh->xh_name_value_len));
5133
5134         if (free < need ||
5135             (xs->not_found &&
5136              count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5137                 if (need <= max_free &&
5138                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5139                         /*
5140                          * We can create the space by defragment. Since only the
5141                          * name/value will be moved, the xe shouldn't be changed
5142                          * in xs.
5143                          */
5144                         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5145                                                         xs->bucket);
5146                         if (ret) {
5147                                 mlog_errno(ret);
5148                                 goto out;
5149                         }
5150
5151                         xh_free_start = le16_to_cpu(xh->xh_free_start);
5152                         free = xh_free_start - header_size
5153                                 - OCFS2_XATTR_HEADER_GAP;
5154                         if (xh_free_start % blocksize < need)
5155                                 free -= xh_free_start % blocksize;
5156
5157                         if (free >= need)
5158                                 goto xattr_set;
5159
5160                         mlog(0, "Can't get enough space for xattr insert by "
5161                              "defragment. Need %u bytes, but we have %d, so "
5162                              "allocate new bucket for it.\n", need, free);
5163                 }
5164
5165                 /*
5166                  * We have to add new buckets or clusters and one
5167                  * allocation should leave us enough space for insert.
5168                  */
5169                 BUG_ON(allocation);
5170
5171                 /*
5172                  * We do not allow for overlapping ranges between buckets. And
5173                  * the maximum number of collisions we will allow for then is
5174                  * one bucket's worth, so check it here whether we need to
5175                  * add a new bucket for the insert.
5176                  */
5177                 ret = ocfs2_check_xattr_bucket_collision(inode,
5178                                                          xs->bucket,
5179                                                          xi->name);
5180                 if (ret) {
5181                         mlog_errno(ret);
5182                         goto out;
5183                 }
5184
5185                 ret = ocfs2_add_new_xattr_bucket(inode,
5186                                                  xs->xattr_bh,
5187                                                  xs->bucket,
5188                                                  ctxt);
5189                 if (ret) {
5190                         mlog_errno(ret);
5191                         goto out;
5192                 }
5193
5194                 /*
5195                  * ocfs2_add_new_xattr_bucket() will have updated
5196                  * xs->bucket if it moved, but it will not have updated
5197                  * any of the other search fields.  Thus, we drop it and
5198                  * re-search.  Everything should be cached, so it'll be
5199                  * quick.
5200                  */
5201                 ocfs2_xattr_bucket_relse(xs->bucket);
5202                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5203                                                    xi->name_index,
5204                                                    xi->name, xs);
5205                 if (ret && ret != -ENODATA)
5206                         goto out;
5207                 xs->not_found = ret;
5208                 allocation = 1;
5209                 goto try_again;
5210         }
5211
5212 xattr_set:
5213         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5214 out:
5215         mlog_exit(ret);
5216         return ret;
5217 }
5218
5219 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5220                                         struct ocfs2_xattr_bucket *bucket,
5221                                         void *para)
5222 {
5223         int ret = 0;
5224         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5225         u16 i;
5226         struct ocfs2_xattr_entry *xe;
5227         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5228         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5229         int credits = ocfs2_remove_extent_credits(osb->sb) +
5230                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5231
5232
5233         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5234
5235         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5236                 xe = &xh->xh_entries[i];
5237                 if (ocfs2_xattr_is_local(xe))
5238                         continue;
5239
5240                 ctxt.handle = ocfs2_start_trans(osb, credits);
5241                 if (IS_ERR(ctxt.handle)) {
5242                         ret = PTR_ERR(ctxt.handle);
5243                         mlog_errno(ret);
5244                         break;
5245                 }
5246
5247                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5248                                                         i, 0, &ctxt);
5249
5250                 ocfs2_commit_trans(osb, ctxt.handle);
5251                 if (ret) {
5252                         mlog_errno(ret);
5253                         break;
5254                 }
5255         }
5256
5257         ocfs2_schedule_truncate_log_flush(osb, 1);
5258         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5259         return ret;
5260 }
5261
5262 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5263                                           struct buffer_head *xb_bh)
5264 {
5265         struct ocfs2_xattr_block *xb =
5266                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5267         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5268         int ret = 0;
5269         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5270         u64 p_blkno;
5271
5272         if (le16_to_cpu(el->l_next_free_rec) == 0)
5273                 return 0;
5274
5275         while (name_hash > 0) {
5276                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5277                                           &e_cpos, &num_clusters, el);
5278                 if (ret) {
5279                         mlog_errno(ret);
5280                         goto out;
5281                 }
5282
5283                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5284                                                   ocfs2_delete_xattr_in_bucket,
5285                                                   NULL);
5286                 if (ret) {
5287                         mlog_errno(ret);
5288                         goto out;
5289                 }
5290
5291                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5292                                              p_blkno, e_cpos, num_clusters);
5293                 if (ret) {
5294                         mlog_errno(ret);
5295                         break;
5296                 }
5297
5298                 if (e_cpos == 0)
5299                         break;
5300
5301                 name_hash = e_cpos - 1;
5302         }
5303
5304 out:
5305         return ret;
5306 }
5307
5308 /*
5309  * 'security' attributes support
5310  */
5311 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5312                                         size_t list_size, const char *name,
5313                                         size_t name_len)
5314 {
5315         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5316         const size_t total_len = prefix_len + name_len + 1;
5317
5318         if (list && total_len <= list_size) {
5319                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5320                 memcpy(list + prefix_len, name, name_len);
5321                 list[prefix_len + name_len] = '\0';
5322         }
5323         return total_len;
5324 }
5325
5326 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5327                                     void *buffer, size_t size)
5328 {
5329         if (strcmp(name, "") == 0)
5330                 return -EINVAL;
5331         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5332                                buffer, size);
5333 }
5334
5335 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5336                                     const void *value, size_t size, int flags)
5337 {
5338         if (strcmp(name, "") == 0)
5339                 return -EINVAL;
5340
5341         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5342                                size, flags);
5343 }
5344
5345 int ocfs2_init_security_get(struct inode *inode,
5346                             struct inode *dir,
5347                             struct ocfs2_security_xattr_info *si)
5348 {
5349         /* check whether ocfs2 support feature xattr */
5350         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
5351                 return -EOPNOTSUPP;
5352         return security_inode_init_security(inode, dir, &si->name, &si->value,
5353                                             &si->value_len);
5354 }
5355
5356 int ocfs2_init_security_set(handle_t *handle,
5357                             struct inode *inode,
5358                             struct buffer_head *di_bh,
5359                             struct ocfs2_security_xattr_info *si,
5360                             struct ocfs2_alloc_context *xattr_ac,
5361                             struct ocfs2_alloc_context *data_ac)
5362 {
5363         return ocfs2_xattr_set_handle(handle, inode, di_bh,
5364                                      OCFS2_XATTR_INDEX_SECURITY,
5365                                      si->name, si->value, si->value_len, 0,
5366                                      xattr_ac, data_ac);
5367 }
5368
5369 struct xattr_handler ocfs2_xattr_security_handler = {
5370         .prefix = XATTR_SECURITY_PREFIX,
5371         .list   = ocfs2_xattr_security_list,
5372         .get    = ocfs2_xattr_security_get,
5373         .set    = ocfs2_xattr_security_set,
5374 };
5375
5376 /*
5377  * 'trusted' attributes support
5378  */
5379 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5380                                        size_t list_size, const char *name,
5381                                        size_t name_len)
5382 {
5383         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5384         const size_t total_len = prefix_len + name_len + 1;
5385
5386         if (list && total_len <= list_size) {
5387                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5388                 memcpy(list + prefix_len, name, name_len);
5389                 list[prefix_len + name_len] = '\0';
5390         }
5391         return total_len;
5392 }
5393
5394 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5395                                    void *buffer, size_t size)
5396 {
5397         if (strcmp(name, "") == 0)
5398                 return -EINVAL;
5399         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5400                                buffer, size);
5401 }
5402
5403 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5404                                    const void *value, size_t size, int flags)
5405 {
5406         if (strcmp(name, "") == 0)
5407                 return -EINVAL;
5408
5409         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5410                                size, flags);
5411 }
5412
5413 struct xattr_handler ocfs2_xattr_trusted_handler = {
5414         .prefix = XATTR_TRUSTED_PREFIX,
5415         .list   = ocfs2_xattr_trusted_list,
5416         .get    = ocfs2_xattr_trusted_get,
5417         .set    = ocfs2_xattr_trusted_set,
5418 };
5419
5420 /*
5421  * 'user' attributes support
5422  */
5423 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5424                                     size_t list_size, const char *name,
5425                                     size_t name_len)
5426 {
5427         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5428         const size_t total_len = prefix_len + name_len + 1;
5429         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5430
5431         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5432                 return 0;
5433
5434         if (list && total_len <= list_size) {
5435                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5436                 memcpy(list + prefix_len, name, name_len);
5437                 list[prefix_len + name_len] = '\0';
5438         }
5439         return total_len;
5440 }
5441
5442 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5443                                 void *buffer, size_t size)
5444 {
5445         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5446
5447         if (strcmp(name, "") == 0)
5448                 return -EINVAL;
5449         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5450                 return -EOPNOTSUPP;
5451         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5452                                buffer, size);
5453 }
5454
5455 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5456                                 const void *value, size_t size, int flags)
5457 {
5458         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5459
5460         if (strcmp(name, "") == 0)
5461                 return -EINVAL;
5462         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5463                 return -EOPNOTSUPP;
5464
5465         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5466                                size, flags);
5467 }
5468
5469 struct xattr_handler ocfs2_xattr_user_handler = {
5470         .prefix = XATTR_USER_PREFIX,
5471         .list   = ocfs2_xattr_user_list,
5472         .get    = ocfs2_xattr_user_get,
5473         .set    = ocfs2_xattr_user_set,
5474 };