fs/ocfs2/refcounttree.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * refcounttree.c
   5  *
   6  * Copyright (C) 2009 Oracle.  All rights reserved.
   7  *
   8  * This program is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU General Public
  10  * License version 2 as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License for more details.
  16  */
  17
  18 #include <linux/sort.h>
  19 #include <cluster/masklog.h>
  20 #include "ocfs2.h"
  21 #include "inode.h"
  22 #include "alloc.h"
  23 #include "suballoc.h"
  24 #include "journal.h"
  25 #include "uptodate.h"
  26 #include "super.h"
  27 #include "buffer_head_io.h"
  28 #include "blockcheck.h"
  29 #include "refcounttree.h"
  30 #include "sysfile.h"
  31 #include "dlmglue.h"
  32 #include "extent_map.h"
  33 #include "aops.h"
  34 #include "xattr.h"
  35 #include "namei.h"
  36 #include "ocfs2_trace.h"
  37
  38 #include <linux/bio.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/slab.h>
  41 #include <linux/writeback.h>
  42 #include <linux/pagevec.h>
  43 #include <linux/swap.h>
  44 #include <linux/security.h>
  45 #include <linux/fsnotify.h>
  46 #include <linux/quotaops.h>
  47 #include <linux/namei.h>
  48 #include <linux/mount.h>
  49 #include <linux/posix_acl.h>
  50
  51 struct ocfs2_cow_context {
  52         struct inode *inode;
  53         u32 cow_start;
  54         u32 cow_len;
  55         struct ocfs2_extent_tree data_et;
  56         struct ocfs2_refcount_tree *ref_tree;
  57         struct buffer_head *ref_root_bh;
  58         struct ocfs2_alloc_context *meta_ac;
  59         struct ocfs2_alloc_context *data_ac;
  60         struct ocfs2_cached_dealloc_ctxt dealloc;
  61         void *cow_object;
  62         struct ocfs2_post_refcount *post_refcount;
  63         int extra_credits;
  64         int (*get_clusters)(struct ocfs2_cow_context *context,
  65                             u32 v_cluster, u32 *p_cluster,
  66                             u32 *num_clusters,
  67                             unsigned int *extent_flags);
  68         int (*cow_duplicate_clusters)(handle_t *handle,
  69                                       struct inode *inode,
  70                                       u32 cpos, u32 old_cluster,
  71                                       u32 new_cluster, u32 new_len);
  72 };
  73
  74 static inline struct ocfs2_refcount_tree *
  75 cache_info_to_refcount(struct ocfs2_caching_info *ci)
  76 {
  77         return container_of(ci, struct ocfs2_refcount_tree, rf_ci);
  78 }
  79
  80 static int ocfs2_validate_refcount_block(struct super_block *sb,
  81                                          struct buffer_head *bh)
  82 {
  83         int rc;
  84         struct ocfs2_refcount_block *rb =
  85                 (struct ocfs2_refcount_block *)bh->b_data;
  86
  87         trace_ocfs2_validate_refcount_block((unsigned long long)bh->b_blocknr);
  88
  89         BUG_ON(!buffer_uptodate(bh));
  90
  91         /*
  92          * If the ecc fails, we return the error but otherwise
  93          * leave the filesystem running.  We know any error is
  94          * local to this block.
  95          */
  96         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &rb->rf_check);
  97         if (rc) {
  98                 mlog(ML_ERROR, "Checksum failed for refcount block %llu\n",
  99                      (unsigned long long)bh->b_blocknr);
 100                 return rc;
 101         }
 102
 103
 104         if (!OCFS2_IS_VALID_REFCOUNT_BLOCK(rb)) {
 105                 ocfs2_error(sb,
 106                             "Refcount block #%llu has bad signature %.*s",
 107                             (unsigned long long)bh->b_blocknr, 7,
 108                             rb->rf_signature);
 109                 return -EINVAL;
 110         }
 111
 112         if (le64_to_cpu(rb->rf_blkno) != bh->b_blocknr) {
 113                 ocfs2_error(sb,
 114                             "Refcount block #%llu has an invalid rf_blkno "
 115                             "of %llu",
 116                             (unsigned long long)bh->b_blocknr,
 117                             (unsigned long long)le64_to_cpu(rb->rf_blkno));
 118                 return -EINVAL;
 119         }
 120
 121         if (le32_to_cpu(rb->rf_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 122                 ocfs2_error(sb,
 123                             "Refcount block #%llu has an invalid "
 124                             "rf_fs_generation of #%u",
 125                             (unsigned long long)bh->b_blocknr,
 126                             le32_to_cpu(rb->rf_fs_generation));
 127                 return -EINVAL;
 128         }
 129
 130         return 0;
 131 }
 132
 133 static int ocfs2_read_refcount_block(struct ocfs2_caching_info *ci,
 134                                      u64 rb_blkno,
 135                                      struct buffer_head **bh)
 136 {
 137         int rc;
 138         struct buffer_head *tmp = *bh;
 139
 140         rc = ocfs2_read_block(ci, rb_blkno, &tmp,
 141                               ocfs2_validate_refcount_block);
 142
 143         /* If ocfs2_read_block() got us a new bh, pass it up. */
 144         if (!rc && !*bh)
 145                 *bh = tmp;
 146
 147         return rc;
 148 }
 149
 150 static u64 ocfs2_refcount_cache_owner(struct ocfs2_caching_info *ci)
 151 {
 152         struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 153
 154         return rf->rf_blkno;
 155 }
 156
 157 static struct super_block *
 158 ocfs2_refcount_cache_get_super(struct ocfs2_caching_info *ci)
 159 {
 160         struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 161
 162         return rf->rf_sb;
 163 }
 164
 165 static void ocfs2_refcount_cache_lock(struct ocfs2_caching_info *ci)
 166 {
 167         struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 168
 169         spin_lock(&rf->rf_lock);
 170 }
 171
 172 static void ocfs2_refcount_cache_unlock(struct ocfs2_caching_info *ci)
 173 {
 174         struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 175
 176         spin_unlock(&rf->rf_lock);
 177 }
 178
 179 static void ocfs2_refcount_cache_io_lock(struct ocfs2_caching_info *ci)
 180 {
 181         struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 182
 183         mutex_lock(&rf->rf_io_mutex);
 184 }
 185
 186 static void ocfs2_refcount_cache_io_unlock(struct ocfs2_caching_info *ci)
 187 {
 188         struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 189
 190         mutex_unlock(&rf->rf_io_mutex);
 191 }
 192
 193 static const struct ocfs2_caching_operations ocfs2_refcount_caching_ops = {
 194         .co_owner               = ocfs2_refcount_cache_owner,
 195         .co_get_super           = ocfs2_refcount_cache_get_super,
 196         .co_cache_lock          = ocfs2_refcount_cache_lock,
 197         .co_cache_unlock        = ocfs2_refcount_cache_unlock,
 198         .co_io_lock             = ocfs2_refcount_cache_io_lock,
 199         .co_io_unlock           = ocfs2_refcount_cache_io_unlock,
 200 };
 201
 202 static struct ocfs2_refcount_tree *
 203 ocfs2_find_refcount_tree(struct ocfs2_super *osb, u64 blkno)
 204 {
 205         struct rb_node *n = osb->osb_rf_lock_tree.rb_node;
 206         struct ocfs2_refcount_tree *tree = NULL;
 207
 208         while (n) {
 209                 tree = rb_entry(n, struct ocfs2_refcount_tree, rf_node);
 210
 211                 if (blkno < tree->rf_blkno)
 212                         n = n->rb_left;
 213                 else if (blkno > tree->rf_blkno)
 214                         n = n->rb_right;
 215                 else
 216                         return tree;
 217         }
 218
 219         return NULL;
 220 }
 221
 222 /* osb_lock is already locked. */
 223 static void ocfs2_insert_refcount_tree(struct ocfs2_super *osb,
 224                                        struct ocfs2_refcount_tree *new)
 225 {
 226         u64 rf_blkno = new->rf_blkno;
 227         struct rb_node *parent = NULL;
 228         struct rb_node **p = &osb->osb_rf_lock_tree.rb_node;
 229         struct ocfs2_refcount_tree *tmp;
 230
 231         while (*p) {
 232                 parent = *p;
 233
 234                 tmp = rb_entry(parent, struct ocfs2_refcount_tree,
 235                                rf_node);
 236
 237                 if (rf_blkno < tmp->rf_blkno)
 238                         p = &(*p)->rb_left;
 239                 else if (rf_blkno > tmp->rf_blkno)
 240                         p = &(*p)->rb_right;
 241                 else {
 242                         /* This should never happen! */
 243                         mlog(ML_ERROR, "Duplicate refcount block %llu found!\n",
 244                              (unsigned long long)rf_blkno);
 245                         BUG();
 246                 }
 247         }
 248
 249         rb_link_node(&new->rf_node, parent, p);
 250         rb_insert_color(&new->rf_node, &osb->osb_rf_lock_tree);
 251 }
 252
 253 static void ocfs2_free_refcount_tree(struct ocfs2_refcount_tree *tree)
 254 {
 255         ocfs2_metadata_cache_exit(&tree->rf_ci);
 256         ocfs2_simple_drop_lockres(OCFS2_SB(tree->rf_sb), &tree->rf_lockres);
 257         ocfs2_lock_res_free(&tree->rf_lockres);
 258         kfree(tree);
 259 }
 260
 261 static inline void
 262 ocfs2_erase_refcount_tree_from_list_no_lock(struct ocfs2_super *osb,
 263                                         struct ocfs2_refcount_tree *tree)
 264 {
 265         rb_erase(&tree->rf_node, &osb->osb_rf_lock_tree);
 266         if (osb->osb_ref_tree_lru && osb->osb_ref_tree_lru == tree)
 267                 osb->osb_ref_tree_lru = NULL;
 268 }
 269
 270 static void ocfs2_erase_refcount_tree_from_list(struct ocfs2_super *osb,
 271                                         struct ocfs2_refcount_tree *tree)
 272 {
 273         spin_lock(&osb->osb_lock);
 274         ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
 275         spin_unlock(&osb->osb_lock);
 276 }
 277
 278 static void ocfs2_kref_remove_refcount_tree(struct kref *kref)
 279 {
 280         struct ocfs2_refcount_tree *tree =
 281                 container_of(kref, struct ocfs2_refcount_tree, rf_getcnt);
 282
 283         ocfs2_free_refcount_tree(tree);
 284 }
 285
 286 static inline void
 287 ocfs2_refcount_tree_get(struct ocfs2_refcount_tree *tree)
 288 {
 289         kref_get(&tree->rf_getcnt);
 290 }
 291
 292 static inline void
 293 ocfs2_refcount_tree_put(struct ocfs2_refcount_tree *tree)
 294 {
 295         kref_put(&tree->rf_getcnt, ocfs2_kref_remove_refcount_tree);
 296 }
 297
 298 static inline void ocfs2_init_refcount_tree_ci(struct ocfs2_refcount_tree *new,
 299                                                struct super_block *sb)
 300 {
 301         ocfs2_metadata_cache_init(&new->rf_ci, &ocfs2_refcount_caching_ops);
 302         mutex_init(&new->rf_io_mutex);
 303         new->rf_sb = sb;
 304         spin_lock_init(&new->rf_lock);
 305 }
 306
 307 static inline void ocfs2_init_refcount_tree_lock(struct ocfs2_super *osb,
 308                                         struct ocfs2_refcount_tree *new,
 309                                         u64 rf_blkno, u32 generation)
 310 {
 311         init_rwsem(&new->rf_sem);
 312         ocfs2_refcount_lock_res_init(&new->rf_lockres, osb,
 313                                      rf_blkno, generation);
 314 }
 315
 316 static struct ocfs2_refcount_tree*
 317 ocfs2_allocate_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno)
 318 {
 319         struct ocfs2_refcount_tree *new;
 320
 321         new = kzalloc(sizeof(struct ocfs2_refcount_tree), GFP_NOFS);
 322         if (!new)
 323                 return NULL;
 324
 325         new->rf_blkno = rf_blkno;
 326         kref_init(&new->rf_getcnt);
 327         ocfs2_init_refcount_tree_ci(new, osb->sb);
 328
 329         return new;
 330 }
 331
 332 static int ocfs2_get_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno,
 333                                    struct ocfs2_refcount_tree **ret_tree)
 334 {
 335         int ret = 0;
 336         struct ocfs2_refcount_tree *tree, *new = NULL;
 337         struct buffer_head *ref_root_bh = NULL;
 338         struct ocfs2_refcount_block *ref_rb;
 339
 340         spin_lock(&osb->osb_lock);
 341         if (osb->osb_ref_tree_lru &&
 342             osb->osb_ref_tree_lru->rf_blkno == rf_blkno)
 343                 tree = osb->osb_ref_tree_lru;
 344         else
 345                 tree = ocfs2_find_refcount_tree(osb, rf_blkno);
 346         if (tree)
 347                 goto out;
 348
 349         spin_unlock(&osb->osb_lock);
 350
 351         new = ocfs2_allocate_refcount_tree(osb, rf_blkno);
 352         if (!new) {
 353                 ret = -ENOMEM;
 354                 mlog_errno(ret);
 355                 return ret;
 356         }
 357         /*
 358          * We need the generation to create the refcount tree lock and since
 359          * it isn't changed during the tree modification, we are safe here to
 360          * read without protection.
 361          * We also have to purge the cache after we create the lock since the
 362          * refcount block may have the stale data. It can only be trusted when
 363          * we hold the refcount lock.
 364          */
 365         ret = ocfs2_read_refcount_block(&new->rf_ci, rf_blkno, &ref_root_bh);
 366         if (ret) {
 367                 mlog_errno(ret);
 368                 ocfs2_metadata_cache_exit(&new->rf_ci);
 369                 kfree(new);
 370                 return ret;
 371         }
 372
 373         ref_rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
 374         new->rf_generation = le32_to_cpu(ref_rb->rf_generation);
 375         ocfs2_init_refcount_tree_lock(osb, new, rf_blkno,
 376                                       new->rf_generation);
 377         ocfs2_metadata_cache_purge(&new->rf_ci);
 378
 379         spin_lock(&osb->osb_lock);
 380         tree = ocfs2_find_refcount_tree(osb, rf_blkno);
 381         if (tree)
 382                 goto out;
 383
 384         ocfs2_insert_refcount_tree(osb, new);
 385
 386         tree = new;
 387         new = NULL;
 388
 389 out:
 390         *ret_tree = tree;
 391
 392         osb->osb_ref_tree_lru = tree;
 393
 394         spin_unlock(&osb->osb_lock);
 395
 396         if (new)
 397                 ocfs2_free_refcount_tree(new);
 398
 399         brelse(ref_root_bh);
 400         return ret;
 401 }
 402
 403 static int ocfs2_get_refcount_block(struct inode *inode, u64 *ref_blkno)
 404 {
 405         int ret;
 406         struct buffer_head *di_bh = NULL;
 407         struct ocfs2_dinode *di;
 408
 409         ret = ocfs2_read_inode_block(inode, &di_bh);
 410         if (ret) {
 411                 mlog_errno(ret);
 412                 goto out;
 413         }
 414
 415         BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
 416
 417         di = (struct ocfs2_dinode *)di_bh->b_data;
 418         *ref_blkno = le64_to_cpu(di->i_refcount_loc);
 419         brelse(di_bh);
 420 out:
 421         return ret;
 422 }
 423
 424 static int __ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
 425                                       struct ocfs2_refcount_tree *tree, int rw)
 426 {
 427         int ret;
 428
 429         ret = ocfs2_refcount_lock(tree, rw);
 430         if (ret) {
 431                 mlog_errno(ret);
 432                 goto out;
 433         }
 434
 435         if (rw)
 436                 down_write(&tree->rf_sem);
 437         else
 438                 down_read(&tree->rf_sem);
 439
 440 out:
 441         return ret;
 442 }
 443
 444 /*
 445  * Lock the refcount tree pointed by ref_blkno and return the tree.
 446  * In most case, we lock the tree and read the refcount block.
 447  * So read it here if the caller really needs it.
 448  *
 449  * If the tree has been re-created by other node, it will free the
 450  * old one and re-create it.
 451  */
 452 int ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
 453                              u64 ref_blkno, int rw,
 454                              struct ocfs2_refcount_tree **ret_tree,
 455                              struct buffer_head **ref_bh)
 456 {
 457         int ret, delete_tree = 0;
 458         struct ocfs2_refcount_tree *tree = NULL;
 459         struct buffer_head *ref_root_bh = NULL;
 460         struct ocfs2_refcount_block *rb;
 461
 462 again:
 463         ret = ocfs2_get_refcount_tree(osb, ref_blkno, &tree);
 464         if (ret) {
 465                 mlog_errno(ret);
 466                 return ret;
 467         }
 468
 469         ocfs2_refcount_tree_get(tree);
 470
 471         ret = __ocfs2_lock_refcount_tree(osb, tree, rw);
 472         if (ret) {
 473                 mlog_errno(ret);
 474                 ocfs2_refcount_tree_put(tree);
 475                 goto out;
 476         }
 477
 478         ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
 479                                         &ref_root_bh);
 480         if (ret) {
 481                 mlog_errno(ret);
 482                 ocfs2_unlock_refcount_tree(osb, tree, rw);
 483                 ocfs2_refcount_tree_put(tree);
 484                 goto out;
 485         }
 486
 487         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
 488         /*
 489          * If the refcount block has been freed and re-created, we may need
 490          * to recreate the refcount tree also.
 491          *
 492          * Here we just remove the tree from the rb-tree, and the last
 493          * kref holder will unlock and delete this refcount_tree.
 494          * Then we goto "again" and ocfs2_get_refcount_tree will create
 495          * the new refcount tree for us.
 496          */
 497         if (tree->rf_generation != le32_to_cpu(rb->rf_generation)) {
 498                 if (!tree->rf_removed) {
 499                         ocfs2_erase_refcount_tree_from_list(osb, tree);
 500                         tree->rf_removed = 1;
 501                         delete_tree = 1;
 502                 }
 503
 504                 ocfs2_unlock_refcount_tree(osb, tree, rw);
 505                 /*
 506                  * We get an extra reference when we create the refcount
 507                  * tree, so another put will destroy it.
 508                  */
 509                 if (delete_tree)
 510                         ocfs2_refcount_tree_put(tree);
 511                 brelse(ref_root_bh);
 512                 ref_root_bh = NULL;
 513                 goto again;
 514         }
 515
 516         *ret_tree = tree;
 517         if (ref_bh) {
 518                 *ref_bh = ref_root_bh;
 519                 ref_root_bh = NULL;
 520         }
 521 out:
 522         brelse(ref_root_bh);
 523         return ret;
 524 }
 525
 526 void ocfs2_unlock_refcount_tree(struct ocfs2_super *osb,
 527                                 struct ocfs2_refcount_tree *tree, int rw)
 528 {
 529         if (rw)
 530                 up_write(&tree->rf_sem);
 531         else
 532                 up_read(&tree->rf_sem);
 533
 534         ocfs2_refcount_unlock(tree, rw);
 535         ocfs2_refcount_tree_put(tree);
 536 }
 537
 538 void ocfs2_purge_refcount_trees(struct ocfs2_super *osb)
 539 {
 540         struct rb_node *node;
 541         struct ocfs2_refcount_tree *tree;
 542         struct rb_root *root = &osb->osb_rf_lock_tree;
 543
 544         while ((node = rb_last(root)) != NULL) {
 545                 tree = rb_entry(node, struct ocfs2_refcount_tree, rf_node);
 546
 547                 trace_ocfs2_purge_refcount_trees(
 548                                 (unsigned long long) tree->rf_blkno);
 549
 550                 rb_erase(&tree->rf_node, root);
 551                 ocfs2_free_refcount_tree(tree);
 552         }
 553 }
 554
 555 /*
 556  * Create a refcount tree for an inode.
 557  * We take for granted that the inode is already locked.
 558  */
 559 static int ocfs2_create_refcount_tree(struct inode *inode,
 560                                       struct buffer_head *di_bh)
 561 {
 562         int ret;
 563         handle_t *handle = NULL;
 564         struct ocfs2_alloc_context *meta_ac = NULL;
 565         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 566         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 567         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 568         struct buffer_head *new_bh = NULL;
 569         struct ocfs2_refcount_block *rb;
 570         struct ocfs2_refcount_tree *new_tree = NULL, *tree = NULL;
 571         u16 suballoc_bit_start;
 572         u32 num_got;
 573         u64 suballoc_loc, first_blkno;
 574
 575         BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
 576
 577         trace_ocfs2_create_refcount_tree(
 578                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
 579
 580         ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
 581         if (ret) {
 582                 mlog_errno(ret);
 583                 goto out;
 584         }
 585
 586         handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_CREATE_CREDITS);
 587         if (IS_ERR(handle)) {
 588                 ret = PTR_ERR(handle);
 589                 mlog_errno(ret);
 590                 goto out;
 591         }
 592
 593         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 594                                       OCFS2_JOURNAL_ACCESS_WRITE);
 595         if (ret) {
 596                 mlog_errno(ret);
 597                 goto out_commit;
 598         }
 599
 600         ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
 601                                    &suballoc_bit_start, &num_got,
 602                                    &first_blkno);
 603         if (ret) {
 604                 mlog_errno(ret);
 605                 goto out_commit;
 606         }
 607
 608         new_tree = ocfs2_allocate_refcount_tree(osb, first_blkno);
 609         if (!new_tree) {
 610                 ret = -ENOMEM;
 611                 mlog_errno(ret);
 612                 goto out_commit;
 613         }
 614
 615         new_bh = sb_getblk(inode->i_sb, first_blkno);
 616         if (!new_bh) {
 617                 ret = -ENOMEM;
 618                 mlog_errno(ret);
 619                 goto out_commit;
 620         }
 621         ocfs2_set_new_buffer_uptodate(&new_tree->rf_ci, new_bh);
 622
 623         ret = ocfs2_journal_access_rb(handle, &new_tree->rf_ci, new_bh,
 624                                       OCFS2_JOURNAL_ACCESS_CREATE);
 625         if (ret) {
 626                 mlog_errno(ret);
 627                 goto out_commit;
 628         }
 629
 630         /* Initialize ocfs2_refcount_block. */
 631         rb = (struct ocfs2_refcount_block *)new_bh->b_data;
 632         memset(rb, 0, inode->i_sb->s_blocksize);
 633         strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
 634         rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
 635         rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
 636         rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
 637         rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
 638         rb->rf_blkno = cpu_to_le64(first_blkno);
 639         rb->rf_count = cpu_to_le32(1);
 640         rb->rf_records.rl_count =
 641                         cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb));
 642         spin_lock(&osb->osb_lock);
 643         rb->rf_generation = osb->s_next_generation++;
 644         spin_unlock(&osb->osb_lock);
 645
 646         ocfs2_journal_dirty(handle, new_bh);
 647
 648         spin_lock(&oi->ip_lock);
 649         oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
 650         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 651         di->i_refcount_loc = cpu_to_le64(first_blkno);
 652         spin_unlock(&oi->ip_lock);
 653
 654         trace_ocfs2_create_refcount_tree_blkno((unsigned long long)first_blkno);
 655
 656         ocfs2_journal_dirty(handle, di_bh);
 657
 658         /*
 659          * We have to init the tree lock here since it will use
 660          * the generation number to create it.
 661          */
 662         new_tree->rf_generation = le32_to_cpu(rb->rf_generation);
 663         ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno,
 664                                       new_tree->rf_generation);
 665
 666         spin_lock(&osb->osb_lock);
 667         tree = ocfs2_find_refcount_tree(osb, first_blkno);
 668
 669         /*
 670          * We've just created a new refcount tree in this block.  If
 671          * we found a refcount tree on the ocfs2_super, it must be
 672          * one we just deleted.  We free the old tree before
 673          * inserting the new tree.
 674          */
 675         BUG_ON(tree && tree->rf_generation == new_tree->rf_generation);
 676         if (tree)
 677                 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
 678         ocfs2_insert_refcount_tree(osb, new_tree);
 679         spin_unlock(&osb->osb_lock);
 680         new_tree = NULL;
 681         if (tree)
 682                 ocfs2_refcount_tree_put(tree);
 683
 684 out_commit:
 685         ocfs2_commit_trans(osb, handle);
 686
 687 out:
 688         if (new_tree) {
 689                 ocfs2_metadata_cache_exit(&new_tree->rf_ci);
 690                 kfree(new_tree);
 691         }
 692
 693         brelse(new_bh);
 694         if (meta_ac)
 695                 ocfs2_free_alloc_context(meta_ac);
 696
 697         return ret;
 698 }
 699
 700 static int ocfs2_set_refcount_tree(struct inode *inode,
 701                                    struct buffer_head *di_bh,
 702                                    u64 refcount_loc)
 703 {
 704         int ret;
 705         handle_t *handle = NULL;
 706         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 707         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 708         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 709         struct buffer_head *ref_root_bh = NULL;
 710         struct ocfs2_refcount_block *rb;
 711         struct ocfs2_refcount_tree *ref_tree;
 712
 713         BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
 714
 715         ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
 716                                        &ref_tree, &ref_root_bh);
 717         if (ret) {
 718                 mlog_errno(ret);
 719                 return ret;
 720         }
 721
 722         handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS);
 723         if (IS_ERR(handle)) {
 724                 ret = PTR_ERR(handle);
 725                 mlog_errno(ret);
 726                 goto out;
 727         }
 728
 729         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 730                                       OCFS2_JOURNAL_ACCESS_WRITE);
 731         if (ret) {
 732                 mlog_errno(ret);
 733                 goto out_commit;
 734         }
 735
 736         ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh,
 737                                       OCFS2_JOURNAL_ACCESS_WRITE);
 738         if (ret) {
 739                 mlog_errno(ret);
 740                 goto out_commit;
 741         }
 742
 743         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
 744         le32_add_cpu(&rb->rf_count, 1);
 745
 746         ocfs2_journal_dirty(handle, ref_root_bh);
 747
 748         spin_lock(&oi->ip_lock);
 749         oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
 750         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 751         di->i_refcount_loc = cpu_to_le64(refcount_loc);
 752         spin_unlock(&oi->ip_lock);
 753         ocfs2_journal_dirty(handle, di_bh);
 754
 755 out_commit:
 756         ocfs2_commit_trans(osb, handle);
 757 out:
 758         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
 759         brelse(ref_root_bh);
 760
 761         return ret;
 762 }
 763
 764 int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh)
 765 {
 766         int ret, delete_tree = 0;
 767         handle_t *handle = NULL;
 768         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 769         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 770         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 771         struct ocfs2_refcount_block *rb;
 772         struct inode *alloc_inode = NULL;
 773         struct buffer_head *alloc_bh = NULL;
 774         struct buffer_head *blk_bh = NULL;
 775         struct ocfs2_refcount_tree *ref_tree;
 776         int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS;
 777         u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc);
 778         u16 bit = 0;
 779
 780         if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL))
 781                 return 0;
 782
 783         BUG_ON(!ref_blkno);
 784         ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh);
 785         if (ret) {
 786                 mlog_errno(ret);
 787                 return ret;
 788         }
 789
 790         rb = (struct ocfs2_refcount_block *)blk_bh->b_data;
 791
 792         /*
 793          * If we are the last user, we need to free the block.
 794          * So lock the allocator ahead.
 795          */
 796         if (le32_to_cpu(rb->rf_count) == 1) {
 797                 blk = le64_to_cpu(rb->rf_blkno);
 798                 bit = le16_to_cpu(rb->rf_suballoc_bit);
 799                 if (rb->rf_suballoc_loc)
 800                         bg_blkno = le64_to_cpu(rb->rf_suballoc_loc);
 801                 else
 802                         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
 803
 804                 alloc_inode = ocfs2_get_system_file_inode(osb,
 805                                         EXTENT_ALLOC_SYSTEM_INODE,
 806                                         le16_to_cpu(rb->rf_suballoc_slot));
 807                 if (!alloc_inode) {
 808                         ret = -ENOMEM;
 809                         mlog_errno(ret);
 810                         goto out;
 811                 }
 812                 mutex_lock(&alloc_inode->i_mutex);
 813
 814                 ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1);
 815                 if (ret) {
 816                         mlog_errno(ret);
 817                         goto out_mutex;
 818                 }
 819
 820                 credits += OCFS2_SUBALLOC_FREE;
 821         }
 822
 823         handle = ocfs2_start_trans(osb, credits);
 824         if (IS_ERR(handle)) {
 825                 ret = PTR_ERR(handle);
 826                 mlog_errno(ret);
 827                 goto out_unlock;
 828         }
 829
 830         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 831                                       OCFS2_JOURNAL_ACCESS_WRITE);
 832         if (ret) {
 833                 mlog_errno(ret);
 834                 goto out_commit;
 835         }
 836
 837         ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh,
 838                                       OCFS2_JOURNAL_ACCESS_WRITE);
 839         if (ret) {
 840                 mlog_errno(ret);
 841                 goto out_commit;
 842         }
 843
 844         spin_lock(&oi->ip_lock);
 845         oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL;
 846         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 847         di->i_refcount_loc = 0;
 848         spin_unlock(&oi->ip_lock);
 849         ocfs2_journal_dirty(handle, di_bh);
 850
 851         le32_add_cpu(&rb->rf_count , -1);
 852         ocfs2_journal_dirty(handle, blk_bh);
 853
 854         if (!rb->rf_count) {
 855                 delete_tree = 1;
 856                 ocfs2_erase_refcount_tree_from_list(osb, ref_tree);
 857                 ret = ocfs2_free_suballoc_bits(handle, alloc_inode,
 858                                                alloc_bh, bit, bg_blkno, 1);
 859                 if (ret)
 860                         mlog_errno(ret);
 861         }
 862
 863 out_commit:
 864         ocfs2_commit_trans(osb, handle);
 865 out_unlock:
 866         if (alloc_inode) {
 867                 ocfs2_inode_unlock(alloc_inode, 1);
 868                 brelse(alloc_bh);
 869         }
 870 out_mutex:
 871         if (alloc_inode) {
 872                 mutex_unlock(&alloc_inode->i_mutex);
 873                 iput(alloc_inode);
 874         }
 875 out:
 876         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
 877         if (delete_tree)
 878                 ocfs2_refcount_tree_put(ref_tree);
 879         brelse(blk_bh);
 880
 881         return ret;
 882 }
 883
 884 static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info *ci,
 885                                           struct buffer_head *ref_leaf_bh,
 886                                           u64 cpos, unsigned int len,
 887                                           struct ocfs2_refcount_rec *ret_rec,
 888                                           int *index)
 889 {
 890         int i = 0;
 891         struct ocfs2_refcount_block *rb =
 892                 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
 893         struct ocfs2_refcount_rec *rec = NULL;
 894
 895         for (; i < le16_to_cpu(rb->rf_records.rl_used); i++) {
 896                 rec = &rb->rf_records.rl_recs[i];
 897
 898                 if (le64_to_cpu(rec->r_cpos) +
 899                     le32_to_cpu(rec->r_clusters) <= cpos)
 900                         continue;
 901                 else if (le64_to_cpu(rec->r_cpos) > cpos)
 902                         break;
 903
 904                 /* ok, cpos fail in this rec. Just return. */
 905                 if (ret_rec)
 906                         *ret_rec = *rec;
 907                 goto out;
 908         }
 909
 910         if (ret_rec) {
 911                 /* We meet with a hole here, so fake the rec. */
 912                 ret_rec->r_cpos = cpu_to_le64(cpos);
 913                 ret_rec->r_refcount = 0;
 914                 if (i < le16_to_cpu(rb->rf_records.rl_used) &&
 915                     le64_to_cpu(rec->r_cpos) < cpos + len)
 916                         ret_rec->r_clusters =
 917                                 cpu_to_le32(le64_to_cpu(rec->r_cpos) - cpos);
 918                 else
 919                         ret_rec->r_clusters = cpu_to_le32(len);
 920         }
 921
 922 out:
 923         *index = i;
 924 }
 925
 926 /*
 927  * Try to remove refcount tree. The mechanism is:
 928  * 1) Check whether i_clusters == 0, if no, exit.
 929  * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
 930  * 3) Check whether we have inline xattr stored outside, if yes, exit.
 931  * 4) Remove the tree.
 932  */
 933 int ocfs2_try_remove_refcount_tree(struct inode *inode,
 934                                    struct buffer_head *di_bh)
 935 {
 936         int ret;
 937         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 938         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 939
 940         down_write(&oi->ip_xattr_sem);
 941         down_write(&oi->ip_alloc_sem);
 942
 943         if (oi->ip_clusters)
 944                 goto out;
 945
 946         if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
 947                 goto out;
 948
 949         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
 950             ocfs2_has_inline_xattr_value_outside(inode, di))
 951                 goto out;
 952
 953         ret = ocfs2_remove_refcount_tree(inode, di_bh);
 954         if (ret)
 955                 mlog_errno(ret);
 956 out:
 957         up_write(&oi->ip_alloc_sem);
 958         up_write(&oi->ip_xattr_sem);
 959         return 0;
 960 }
 961
 962 /*
 963  * Find the end range for a leaf refcount block indicated by
 964  * el->l_recs[index].e_blkno.
 965  */
 966 static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info *ci,
 967                                        struct buffer_head *ref_root_bh,
 968                                        struct ocfs2_extent_block *eb,
 969                                        struct ocfs2_extent_list *el,
 970                                        int index,  u32 *cpos_end)
 971 {
 972         int ret, i, subtree_root;
 973         u32 cpos;
 974         u64 blkno;
 975         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
 976         struct ocfs2_path *left_path = NULL, *right_path = NULL;
 977         struct ocfs2_extent_tree et;
 978         struct ocfs2_extent_list *tmp_el;
 979
 980         if (index < le16_to_cpu(el->l_next_free_rec) - 1) {
 981                 /*
 982                  * We have a extent rec after index, so just use the e_cpos
 983                  * of the next extent rec.
 984                  */
 985                 *cpos_end = le32_to_cpu(el->l_recs[index+1].e_cpos);
 986                 return 0;
 987         }
 988
 989         if (!eb || (eb && !eb->h_next_leaf_blk)) {
 990                 /*
 991                  * We are the last extent rec, so any high cpos should
 992                  * be stored in this leaf refcount block.
 993                  */
 994                 *cpos_end = UINT_MAX;
 995                 return 0;
 996         }
 997
 998         /*
 999          * If the extent block isn't the last one, we have to find
1000          * the subtree root between this extent block and the next
1001          * leaf extent block and get the corresponding e_cpos from
1002          * the subroot. Otherwise we may corrupt the b-tree.
1003          */
1004         ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1005
1006         left_path = ocfs2_new_path_from_et(&et);
1007         if (!left_path) {
1008                 ret = -ENOMEM;
1009                 mlog_errno(ret);
1010                 goto out;
1011         }
1012
1013         cpos = le32_to_cpu(eb->h_list.l_recs[index].e_cpos);
1014         ret = ocfs2_find_path(ci, left_path, cpos);
1015         if (ret) {
1016                 mlog_errno(ret);
1017                 goto out;
1018         }
1019
1020         right_path = ocfs2_new_path_from_path(left_path);
1021         if (!right_path) {
1022                 ret = -ENOMEM;
1023                 mlog_errno(ret);
1024                 goto out;
1025         }
1026
1027         ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, &cpos);
1028         if (ret) {
1029                 mlog_errno(ret);
1030                 goto out;
1031         }
1032
1033         ret = ocfs2_find_path(ci, right_path, cpos);
1034         if (ret) {
1035                 mlog_errno(ret);
1036                 goto out;
1037         }
1038
1039         subtree_root = ocfs2_find_subtree_root(&et, left_path,
1040                                                right_path);
1041
1042         tmp_el = left_path->p_node[subtree_root].el;
1043         blkno = left_path->p_node[subtree_root+1].bh->b_blocknr;
1044         for (i = 0; i < le16_to_cpu(tmp_el->l_next_free_rec); i++) {
1045                 if (le64_to_cpu(tmp_el->l_recs[i].e_blkno) == blkno) {
1046                         *cpos_end = le32_to_cpu(tmp_el->l_recs[i+1].e_cpos);
1047                         break;
1048                 }
1049         }
1050
1051         BUG_ON(i == le16_to_cpu(tmp_el->l_next_free_rec));
1052
1053 out:
1054         ocfs2_free_path(left_path);
1055         ocfs2_free_path(right_path);
1056         return ret;
1057 }
1058
1059 /*
1060  * Given a cpos and len, try to find the refcount record which contains cpos.
1061  * 1. If cpos can be found in one refcount record, return the record.
1062  * 2. If cpos can't be found, return a fake record which start from cpos
1063  *    and end at a small value between cpos+len and start of the next record.
1064  *    This fake record has r_refcount = 0.
1065  */
1066 static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
1067                                   struct buffer_head *ref_root_bh,
1068                                   u64 cpos, unsigned int len,
1069                                   struct ocfs2_refcount_rec *ret_rec,
1070                                   int *index,
1071                                   struct buffer_head **ret_bh)
1072 {
1073         int ret = 0, i, found;
1074         u32 low_cpos, uninitialized_var(cpos_end);
1075         struct ocfs2_extent_list *el;
1076         struct ocfs2_extent_rec *rec = NULL;
1077         struct ocfs2_extent_block *eb = NULL;
1078         struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
1079         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1080         struct ocfs2_refcount_block *rb =
1081                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1082
1083         if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)) {
1084                 ocfs2_find_refcount_rec_in_rl(ci, ref_root_bh, cpos, len,
1085                                               ret_rec, index);
1086                 *ret_bh = ref_root_bh;
1087                 get_bh(ref_root_bh);
1088                 return 0;
1089         }
1090
1091         el = &rb->rf_list;
1092         low_cpos = cpos & OCFS2_32BIT_POS_MASK;
1093
1094         if (el->l_tree_depth) {
1095                 ret = ocfs2_find_leaf(ci, el, low_cpos, &eb_bh);
1096                 if (ret) {
1097                         mlog_errno(ret);
1098                         goto out;
1099                 }
1100
1101                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1102                 el = &eb->h_list;
1103
1104                 if (el->l_tree_depth) {
1105                         ocfs2_error(sb,
1106                         "refcount tree %llu has non zero tree "
1107                         "depth in leaf btree tree block %llu\n",
1108                         (unsigned long long)ocfs2_metadata_cache_owner(ci),
1109                         (unsigned long long)eb_bh->b_blocknr);
1110                         ret = -EROFS;
1111                         goto out;
1112                 }
1113         }
1114
1115         found = 0;
1116         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1117                 rec = &el->l_recs[i];
1118
1119                 if (le32_to_cpu(rec->e_cpos) <= low_cpos) {
1120                         found = 1;
1121                         break;
1122                 }
1123         }
1124
1125         if (found) {
1126                 ret = ocfs2_get_refcount_cpos_end(ci, ref_root_bh,
1127                                                   eb, el, i, &cpos_end);
1128                 if (ret) {
1129                         mlog_errno(ret);
1130                         goto out;
1131                 }
1132
1133                 if (cpos_end < low_cpos + len)
1134                         len = cpos_end - low_cpos;
1135         }
1136
1137         ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
1138                                         &ref_leaf_bh);
1139         if (ret) {
1140                 mlog_errno(ret);
1141                 goto out;
1142         }
1143
1144         ocfs2_find_refcount_rec_in_rl(ci, ref_leaf_bh, cpos, len,
1145                                       ret_rec, index);
1146         *ret_bh = ref_leaf_bh;
1147 out:
1148         brelse(eb_bh);
1149         return ret;
1150 }
1151
1152 enum ocfs2_ref_rec_contig {
1153         REF_CONTIG_NONE = 0,
1154         REF_CONTIG_LEFT,
1155         REF_CONTIG_RIGHT,
1156         REF_CONTIG_LEFTRIGHT,
1157 };
1158
1159 static enum ocfs2_ref_rec_contig
1160         ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block *rb,
1161                                     int index)
1162 {
1163         if ((rb->rf_records.rl_recs[index].r_refcount ==
1164             rb->rf_records.rl_recs[index + 1].r_refcount) &&
1165             (le64_to_cpu(rb->rf_records.rl_recs[index].r_cpos) +
1166             le32_to_cpu(rb->rf_records.rl_recs[index].r_clusters) ==
1167             le64_to_cpu(rb->rf_records.rl_recs[index + 1].r_cpos)))
1168                 return REF_CONTIG_RIGHT;
1169
1170         return REF_CONTIG_NONE;
1171 }
1172
1173 static enum ocfs2_ref_rec_contig
1174         ocfs2_refcount_rec_contig(struct ocfs2_refcount_block *rb,
1175                                   int index)
1176 {
1177         enum ocfs2_ref_rec_contig ret = REF_CONTIG_NONE;
1178
1179         if (index < le16_to_cpu(rb->rf_records.rl_used) - 1)
1180                 ret = ocfs2_refcount_rec_adjacent(rb, index);
1181
1182         if (index > 0) {
1183                 enum ocfs2_ref_rec_contig tmp;
1184
1185                 tmp = ocfs2_refcount_rec_adjacent(rb, index - 1);
1186
1187                 if (tmp == REF_CONTIG_RIGHT) {
1188                         if (ret == REF_CONTIG_RIGHT)
1189                                 ret = REF_CONTIG_LEFTRIGHT;
1190                         else
1191                                 ret = REF_CONTIG_LEFT;
1192                 }
1193         }
1194
1195         return ret;
1196 }
1197
1198 static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block *rb,
1199                                            int index)
1200 {
1201         BUG_ON(rb->rf_records.rl_recs[index].r_refcount !=
1202                rb->rf_records.rl_recs[index+1].r_refcount);
1203
1204         le32_add_cpu(&rb->rf_records.rl_recs[index].r_clusters,
1205                      le32_to_cpu(rb->rf_records.rl_recs[index+1].r_clusters));
1206
1207         if (index < le16_to_cpu(rb->rf_records.rl_used) - 2)
1208                 memmove(&rb->rf_records.rl_recs[index + 1],
1209                         &rb->rf_records.rl_recs[index + 2],
1210                         sizeof(struct ocfs2_refcount_rec) *
1211                         (le16_to_cpu(rb->rf_records.rl_used) - index - 2));
1212
1213         memset(&rb->rf_records.rl_recs[le16_to_cpu(rb->rf_records.rl_used) - 1],
1214                0, sizeof(struct ocfs2_refcount_rec));
1215         le16_add_cpu(&rb->rf_records.rl_used, -1);
1216 }
1217
1218 /*
1219  * Merge the refcount rec if we are contiguous with the adjacent recs.
1220  */
1221 static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
1222                                      int index)
1223 {
1224         enum ocfs2_ref_rec_contig contig =
1225                                 ocfs2_refcount_rec_contig(rb, index);
1226
1227         if (contig == REF_CONTIG_NONE)
1228                 return;
1229
1230         if (contig == REF_CONTIG_LEFT || contig == REF_CONTIG_LEFTRIGHT) {
1231                 BUG_ON(index == 0);
1232                 index--;
1233         }
1234
1235         ocfs2_rotate_refcount_rec_left(rb, index);
1236
1237         if (contig == REF_CONTIG_LEFTRIGHT)
1238                 ocfs2_rotate_refcount_rec_left(rb, index);
1239 }
1240
1241 /*
1242  * Change the refcount indexed by "index" in ref_bh.
1243  * If refcount reaches 0, remove it.
1244  */
1245 static int ocfs2_change_refcount_rec(handle_t *handle,
1246                                      struct ocfs2_caching_info *ci,
1247                                      struct buffer_head *ref_leaf_bh,
1248                                      int index, int merge, int change)
1249 {
1250         int ret;
1251         struct ocfs2_refcount_block *rb =
1252                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1253         struct ocfs2_refcount_list *rl = &rb->rf_records;
1254         struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
1255
1256         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1257                                       OCFS2_JOURNAL_ACCESS_WRITE);
1258         if (ret) {
1259                 mlog_errno(ret);
1260                 goto out;
1261         }
1262
1263         trace_ocfs2_change_refcount_rec(
1264                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1265                 index, le32_to_cpu(rec->r_refcount), change);
1266         le32_add_cpu(&rec->r_refcount, change);
1267
1268         if (!rec->r_refcount) {
1269                 if (index != le16_to_cpu(rl->rl_used) - 1) {
1270                         memmove(rec, rec + 1,
1271                                 (le16_to_cpu(rl->rl_used) - index - 1) *
1272                                 sizeof(struct ocfs2_refcount_rec));
1273                         memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
1274                                0, sizeof(struct ocfs2_refcount_rec));
1275                 }
1276
1277                 le16_add_cpu(&rl->rl_used, -1);
1278         } else if (merge)
1279                 ocfs2_refcount_rec_merge(rb, index);
1280
1281         ocfs2_journal_dirty(handle, ref_leaf_bh);
1282 out:
1283         return ret;
1284 }
1285
1286 static int ocfs2_expand_inline_ref_root(handle_t *handle,
1287                                         struct ocfs2_caching_info *ci,
1288                                         struct buffer_head *ref_root_bh,
1289                                         struct buffer_head **ref_leaf_bh,
1290                                         struct ocfs2_alloc_context *meta_ac)
1291 {
1292         int ret;
1293         u16 suballoc_bit_start;
1294         u32 num_got;
1295         u64 suballoc_loc, blkno;
1296         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1297         struct buffer_head *new_bh = NULL;
1298         struct ocfs2_refcount_block *new_rb;
1299         struct ocfs2_refcount_block *root_rb =
1300                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1301
1302         ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1303                                       OCFS2_JOURNAL_ACCESS_WRITE);
1304         if (ret) {
1305                 mlog_errno(ret);
1306                 goto out;
1307         }
1308
1309         ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1310                                    &suballoc_bit_start, &num_got,
1311                                    &blkno);
1312         if (ret) {
1313                 mlog_errno(ret);
1314                 goto out;
1315         }
1316
1317         new_bh = sb_getblk(sb, blkno);
1318         if (new_bh == NULL) {
1319                 ret = -ENOMEM;
1320                 mlog_errno(ret);
1321                 goto out;
1322         }
1323         ocfs2_set_new_buffer_uptodate(ci, new_bh);
1324
1325         ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1326                                       OCFS2_JOURNAL_ACCESS_CREATE);
1327         if (ret) {
1328                 mlog_errno(ret);
1329                 goto out;
1330         }
1331
1332         /*
1333          * Initialize ocfs2_refcount_block.
1334          * It should contain the same information as the old root.
1335          * so just memcpy it and change the corresponding field.
1336          */
1337         memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
1338
1339         new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1340         new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1341         new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1342         new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1343         new_rb->rf_blkno = cpu_to_le64(blkno);
1344         new_rb->rf_cpos = cpu_to_le32(0);
1345         new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1346         new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1347         ocfs2_journal_dirty(handle, new_bh);
1348
1349         /* Now change the root. */
1350         memset(&root_rb->rf_list, 0, sb->s_blocksize -
1351                offsetof(struct ocfs2_refcount_block, rf_list));
1352         root_rb->rf_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_rb(sb));
1353         root_rb->rf_clusters = cpu_to_le32(1);
1354         root_rb->rf_list.l_next_free_rec = cpu_to_le16(1);
1355         root_rb->rf_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
1356         root_rb->rf_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
1357         root_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_TREE_FL);
1358
1359         ocfs2_journal_dirty(handle, ref_root_bh);
1360
1361         trace_ocfs2_expand_inline_ref_root((unsigned long long)blkno,
1362                 le16_to_cpu(new_rb->rf_records.rl_used));
1363
1364         *ref_leaf_bh = new_bh;
1365         new_bh = NULL;
1366 out:
1367         brelse(new_bh);
1368         return ret;
1369 }
1370
1371 static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec *prev,
1372                                            struct ocfs2_refcount_rec *next)
1373 {
1374         if (ocfs2_get_ref_rec_low_cpos(prev) + le32_to_cpu(prev->r_clusters) <=
1375                 ocfs2_get_ref_rec_low_cpos(next))
1376                 return 1;
1377
1378         return 0;
1379 }
1380
1381 static int cmp_refcount_rec_by_low_cpos(const void *a, const void *b)
1382 {
1383         const struct ocfs2_refcount_rec *l = a, *r = b;
1384         u32 l_cpos = ocfs2_get_ref_rec_low_cpos(l);
1385         u32 r_cpos = ocfs2_get_ref_rec_low_cpos(r);
1386
1387         if (l_cpos > r_cpos)
1388                 return 1;
1389         if (l_cpos < r_cpos)
1390                 return -1;
1391         return 0;
1392 }
1393
1394 static int cmp_refcount_rec_by_cpos(const void *a, const void *b)
1395 {
1396         const struct ocfs2_refcount_rec *l = a, *r = b;
1397         u64 l_cpos = le64_to_cpu(l->r_cpos);
1398         u64 r_cpos = le64_to_cpu(r->r_cpos);
1399
1400         if (l_cpos > r_cpos)
1401                 return 1;
1402         if (l_cpos < r_cpos)
1403                 return -1;
1404         return 0;
1405 }
1406
1407 static void swap_refcount_rec(void *a, void *b, int size)
1408 {
1409         struct ocfs2_refcount_rec *l = a, *r = b;
1410
1411         swap(*l, *r);
1412 }
1413
1414 /*
1415  * The refcount cpos are ordered by their 64bit cpos,
1416  * But we will use the low 32 bit to be the e_cpos in the b-tree.
1417  * So we need to make sure that this pos isn't intersected with others.
1418  *
1419  * Note: The refcount block is already sorted by their low 32 bit cpos,
1420  *       So just try the middle pos first, and we will exit when we find
1421  *       the good position.
1422  */
1423 static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
1424                                          u32 *split_pos, int *split_index)
1425 {
1426         int num_used = le16_to_cpu(rl->rl_used);
1427         int delta, middle = num_used / 2;
1428
1429         for (delta = 0; delta < middle; delta++) {
1430                 /* Let's check delta earlier than middle */
1431                 if (ocfs2_refcount_rec_no_intersect(
1432                                         &rl->rl_recs[middle - delta - 1],
1433                                         &rl->rl_recs[middle - delta])) {
1434                         *split_index = middle - delta;
1435                         break;
1436                 }
1437
1438                 /* For even counts, don't walk off the end */
1439                 if ((middle + delta + 1) == num_used)
1440                         continue;
1441
1442                 /* Now try delta past middle */
1443                 if (ocfs2_refcount_rec_no_intersect(
1444                                         &rl->rl_recs[middle + delta],
1445                                         &rl->rl_recs[middle + delta + 1])) {
1446                         *split_index = middle + delta + 1;
1447                         break;
1448                 }
1449         }
1450
1451         if (delta >= middle)
1452                 return -ENOSPC;
1453
1454         *split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
1455         return 0;
1456 }
1457
1458 static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
1459                                             struct buffer_head *new_bh,
1460                                             u32 *split_cpos)
1461 {
1462         int split_index = 0, num_moved, ret;
1463         u32 cpos = 0;
1464         struct ocfs2_refcount_block *rb =
1465                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1466         struct ocfs2_refcount_list *rl = &rb->rf_records;
1467         struct ocfs2_refcount_block *new_rb =
1468                         (struct ocfs2_refcount_block *)new_bh->b_data;
1469         struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
1470
1471         trace_ocfs2_divide_leaf_refcount_block(
1472                 (unsigned long long)ref_leaf_bh->b_blocknr,
1473                 le16_to_cpu(rl->rl_count), le16_to_cpu(rl->rl_used));
1474
1475         /*
1476          * XXX: Improvement later.
1477          * If we know all the high 32 bit cpos is the same, no need to sort.
1478          *
1479          * In order to make the whole process safe, we do:
1480          * 1. sort the entries by their low 32 bit cpos first so that we can
1481          *    find the split cpos easily.
1482          * 2. call ocfs2_insert_extent to insert the new refcount block.
1483          * 3. move the refcount rec to the new block.
1484          * 4. sort the entries by their 64 bit cpos.
1485          * 5. dirty the new_rb and rb.
1486          */
1487         sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1488              sizeof(struct ocfs2_refcount_rec),
1489              cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
1490
1491         ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
1492         if (ret) {
1493                 mlog_errno(ret);
1494                 return ret;
1495         }
1496
1497         new_rb->rf_cpos = cpu_to_le32(cpos);
1498
1499         /* move refcount records starting from split_index to the new block. */
1500         num_moved = le16_to_cpu(rl->rl_used) - split_index;
1501         memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
1502                num_moved * sizeof(struct ocfs2_refcount_rec));
1503
1504         /*ok, remove the entries we just moved over to the other block. */
1505         memset(&rl->rl_recs[split_index], 0,
1506                num_moved * sizeof(struct ocfs2_refcount_rec));
1507
1508         /* change old and new rl_used accordingly. */
1509         le16_add_cpu(&rl->rl_used, -num_moved);
1510         new_rl->rl_used = cpu_to_le16(num_moved);
1511
1512         sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1513              sizeof(struct ocfs2_refcount_rec),
1514              cmp_refcount_rec_by_cpos, swap_refcount_rec);
1515
1516         sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
1517              sizeof(struct ocfs2_refcount_rec),
1518              cmp_refcount_rec_by_cpos, swap_refcount_rec);
1519
1520         *split_cpos = cpos;
1521         return 0;
1522 }
1523
1524 static int ocfs2_new_leaf_refcount_block(handle_t *handle,
1525                                          struct ocfs2_caching_info *ci,
1526                                          struct buffer_head *ref_root_bh,
1527                                          struct buffer_head *ref_leaf_bh,
1528                                          struct ocfs2_alloc_context *meta_ac)
1529 {
1530         int ret;
1531         u16 suballoc_bit_start;
1532         u32 num_got, new_cpos;
1533         u64 suballoc_loc, blkno;
1534         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1535         struct ocfs2_refcount_block *root_rb =
1536                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1537         struct buffer_head *new_bh = NULL;
1538         struct ocfs2_refcount_block *new_rb;
1539         struct ocfs2_extent_tree ref_et;
1540
1541         BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
1542
1543         ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1544                                       OCFS2_JOURNAL_ACCESS_WRITE);
1545         if (ret) {
1546                 mlog_errno(ret);
1547                 goto out;
1548         }
1549
1550         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1551                                       OCFS2_JOURNAL_ACCESS_WRITE);
1552         if (ret) {
1553                 mlog_errno(ret);
1554                 goto out;
1555         }
1556
1557         ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1558                                    &suballoc_bit_start, &num_got,
1559                                    &blkno);
1560         if (ret) {
1561                 mlog_errno(ret);
1562                 goto out;
1563         }
1564
1565         new_bh = sb_getblk(sb, blkno);
1566         if (new_bh == NULL) {
1567                 ret = -ENOMEM;
1568                 mlog_errno(ret);
1569                 goto out;
1570         }
1571         ocfs2_set_new_buffer_uptodate(ci, new_bh);
1572
1573         ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1574                                       OCFS2_JOURNAL_ACCESS_CREATE);
1575         if (ret) {
1576                 mlog_errno(ret);
1577                 goto out;
1578         }
1579
1580         /* Initialize ocfs2_refcount_block. */
1581         new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1582         memset(new_rb, 0, sb->s_blocksize);
1583         strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
1584         new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1585         new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1586         new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1587         new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1588         new_rb->rf_blkno = cpu_to_le64(blkno);
1589         new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1590         new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1591         new_rb->rf_records.rl_count =
1592                                 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
1593         new_rb->rf_generation = root_rb->rf_generation;
1594
1595         ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
1596         if (ret) {
1597                 mlog_errno(ret);
1598                 goto out;
1599         }
1600
1601         ocfs2_journal_dirty(handle, ref_leaf_bh);
1602         ocfs2_journal_dirty(handle, new_bh);
1603
1604         ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
1605
1606         trace_ocfs2_new_leaf_refcount_block(
1607                         (unsigned long long)new_bh->b_blocknr, new_cpos);
1608
1609         /* Insert the new leaf block with the specific offset cpos. */
1610         ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
1611                                   1, 0, meta_ac);
1612         if (ret)
1613                 mlog_errno(ret);
1614
1615 out:
1616         brelse(new_bh);
1617         return ret;
1618 }
1619
1620 static int ocfs2_expand_refcount_tree(handle_t *handle,
1621                                       struct ocfs2_caching_info *ci,
1622                                       struct buffer_head *ref_root_bh,
1623                                       struct buffer_head *ref_leaf_bh,
1624                                       struct ocfs2_alloc_context *meta_ac)
1625 {
1626         int ret;
1627         struct buffer_head *expand_bh = NULL;
1628
1629         if (ref_root_bh == ref_leaf_bh) {
1630                 /*
1631                  * the old root bh hasn't been expanded to a b-tree,
1632                  * so expand it first.
1633                  */
1634                 ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
1635                                                    &expand_bh, meta_ac);
1636                 if (ret) {
1637                         mlog_errno(ret);
1638                         goto out;
1639                 }
1640         } else {
1641                 expand_bh = ref_leaf_bh;
1642                 get_bh(expand_bh);
1643         }
1644
1645
1646         /* Now add a new refcount block into the tree.*/
1647         ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
1648                                             expand_bh, meta_ac);
1649         if (ret)
1650                 mlog_errno(ret);
1651 out:
1652         brelse(expand_bh);
1653         return ret;
1654 }
1655
1656 /*
1657  * Adjust the extent rec in b-tree representing ref_leaf_bh.
1658  *
1659  * Only called when we have inserted a new refcount rec at index 0
1660  * which means ocfs2_extent_rec.e_cpos may need some change.
1661  */
1662 static int ocfs2_adjust_refcount_rec(handle_t *handle,
1663                                      struct ocfs2_caching_info *ci,
1664                                      struct buffer_head *ref_root_bh,
1665                                      struct buffer_head *ref_leaf_bh,
1666                                      struct ocfs2_refcount_rec *rec)
1667 {
1668         int ret = 0, i;
1669         u32 new_cpos, old_cpos;
1670         struct ocfs2_path *path = NULL;
1671         struct ocfs2_extent_tree et;
1672         struct ocfs2_refcount_block *rb =
1673                 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1674         struct ocfs2_extent_list *el;
1675
1676         if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
1677                 goto out;
1678
1679         rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1680         old_cpos = le32_to_cpu(rb->rf_cpos);
1681         new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
1682         if (old_cpos <= new_cpos)
1683                 goto out;
1684
1685         ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1686
1687         path = ocfs2_new_path_from_et(&et);
1688         if (!path) {
1689                 ret = -ENOMEM;
1690                 mlog_errno(ret);
1691                 goto out;
1692         }
1693
1694         ret = ocfs2_find_path(ci, path, old_cpos);
1695         if (ret) {
1696                 mlog_errno(ret);
1697                 goto out;
1698         }
1699
1700         /*
1701          * 2 more credits, one for the leaf refcount block, one for
1702          * the extent block contains the extent rec.
1703          */
1704         ret = ocfs2_extend_trans(handle, 2);
1705         if (ret < 0) {
1706                 mlog_errno(ret);
1707                 goto out;
1708         }
1709
1710         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1711                                       OCFS2_JOURNAL_ACCESS_WRITE);
1712         if (ret < 0) {
1713                 mlog_errno(ret);
1714                 goto out;
1715         }
1716
1717         ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
1718                                       OCFS2_JOURNAL_ACCESS_WRITE);
1719         if (ret < 0) {
1720                 mlog_errno(ret);
1721                 goto out;
1722         }
1723
1724         /* change the leaf extent block first. */
1725         el = path_leaf_el(path);
1726
1727         for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
1728                 if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
1729                         break;
1730
1731         BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
1732
1733         el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1734
1735         /* change the r_cpos in the leaf block. */
1736         rb->rf_cpos = cpu_to_le32(new_cpos);
1737
1738         ocfs2_journal_dirty(handle, path_leaf_bh(path));
1739         ocfs2_journal_dirty(handle, ref_leaf_bh);
1740
1741 out:
1742         ocfs2_free_path(path);
1743         return ret;
1744 }
1745
1746 static int ocfs2_insert_refcount_rec(handle_t *handle,
1747                                      struct ocfs2_caching_info *ci,
1748                                      struct buffer_head *ref_root_bh,
1749                                      struct buffer_head *ref_leaf_bh,
1750                                      struct ocfs2_refcount_rec *rec,
1751                                      int index, int merge,
1752                                      struct ocfs2_alloc_context *meta_ac)
1753 {
1754         int ret;
1755         struct ocfs2_refcount_block *rb =
1756                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1757         struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1758         struct buffer_head *new_bh = NULL;
1759
1760         BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1761
1762         if (rf_list->rl_used == rf_list->rl_count) {
1763                 u64 cpos = le64_to_cpu(rec->r_cpos);
1764                 u32 len = le32_to_cpu(rec->r_clusters);
1765
1766                 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1767                                                  ref_leaf_bh, meta_ac);
1768                 if (ret) {
1769                         mlog_errno(ret);
1770                         goto out;
1771                 }
1772
1773                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1774                                              cpos, len, NULL, &index,
1775                                              &new_bh);
1776                 if (ret) {
1777                         mlog_errno(ret);
1778                         goto out;
1779                 }
1780
1781                 ref_leaf_bh = new_bh;
1782                 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1783                 rf_list = &rb->rf_records;
1784         }
1785
1786         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1787                                       OCFS2_JOURNAL_ACCESS_WRITE);
1788         if (ret) {
1789                 mlog_errno(ret);
1790                 goto out;
1791         }
1792
1793         if (index < le16_to_cpu(rf_list->rl_used))
1794                 memmove(&rf_list->rl_recs[index + 1],
1795                         &rf_list->rl_recs[index],
1796                         (le16_to_cpu(rf_list->rl_used) - index) *
1797                          sizeof(struct ocfs2_refcount_rec));
1798
1799         trace_ocfs2_insert_refcount_rec(
1800                 (unsigned long long)ref_leaf_bh->b_blocknr, index,
1801                 (unsigned long long)le64_to_cpu(rec->r_cpos),
1802                 le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount));
1803
1804         rf_list->rl_recs[index] = *rec;
1805
1806         le16_add_cpu(&rf_list->rl_used, 1);
1807
1808         if (merge)
1809                 ocfs2_refcount_rec_merge(rb, index);
1810
1811         ocfs2_journal_dirty(handle, ref_leaf_bh);
1812
1813         if (index == 0) {
1814                 ret = ocfs2_adjust_refcount_rec(handle, ci,
1815                                                 ref_root_bh,
1816                                                 ref_leaf_bh, rec);
1817                 if (ret)
1818                         mlog_errno(ret);
1819         }
1820 out:
1821         brelse(new_bh);
1822         return ret;
1823 }
1824
1825 /*
1826  * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1827  * This is much simple than our b-tree code.
1828  * split_rec is the new refcount rec we want to insert.
1829  * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1830  * increase refcount or decrease a refcount to non-zero).
1831  * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1832  * rec( in case we decrease a refcount to zero).
1833  */
1834 static int ocfs2_split_refcount_rec(handle_t *handle,
1835                                     struct ocfs2_caching_info *ci,
1836                                     struct buffer_head *ref_root_bh,
1837                                     struct buffer_head *ref_leaf_bh,
1838                                     struct ocfs2_refcount_rec *split_rec,
1839                                     int index, int merge,
1840                                     struct ocfs2_alloc_context *meta_ac,
1841                                     struct ocfs2_cached_dealloc_ctxt *dealloc)
1842 {
1843         int ret, recs_need;
1844         u32 len;
1845         struct ocfs2_refcount_block *rb =
1846                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1847         struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1848         struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
1849         struct ocfs2_refcount_rec *tail_rec = NULL;
1850         struct buffer_head *new_bh = NULL;
1851
1852         BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1853
1854         trace_ocfs2_split_refcount_rec(le64_to_cpu(orig_rec->r_cpos),
1855                 le32_to_cpu(orig_rec->r_clusters),
1856                 le32_to_cpu(orig_rec->r_refcount),
1857                 le64_to_cpu(split_rec->r_cpos),
1858                 le32_to_cpu(split_rec->r_clusters),
1859                 le32_to_cpu(split_rec->r_refcount));
1860
1861         /*
1862          * If we just need to split the header or tail clusters,
1863          * no more recs are needed, just split is OK.
1864          * Otherwise we at least need one new recs.
1865          */
1866         if (!split_rec->r_refcount &&
1867             (split_rec->r_cpos == orig_rec->r_cpos ||
1868              le64_to_cpu(split_rec->r_cpos) +
1869              le32_to_cpu(split_rec->r_clusters) ==
1870              le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1871                 recs_need = 0;
1872         else
1873                 recs_need = 1;
1874
1875         /*
1876          * We need one more rec if we split in the middle and the new rec have
1877          * some refcount in it.
1878          */
1879         if (split_rec->r_refcount &&
1880             (split_rec->r_cpos != orig_rec->r_cpos &&
1881              le64_to_cpu(split_rec->r_cpos) +
1882              le32_to_cpu(split_rec->r_clusters) !=
1883              le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1884                 recs_need++;
1885
1886         /* If the leaf block don't have enough record, expand it. */
1887         if (le16_to_cpu(rf_list->rl_used) + recs_need >
1888                                          le16_to_cpu(rf_list->rl_count)) {
1889                 struct ocfs2_refcount_rec tmp_rec;
1890                 u64 cpos = le64_to_cpu(orig_rec->r_cpos);
1891                 len = le32_to_cpu(orig_rec->r_clusters);
1892                 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1893                                                  ref_leaf_bh, meta_ac);
1894                 if (ret) {
1895                         mlog_errno(ret);
1896                         goto out;
1897                 }
1898
1899                 /*
1900                  * We have to re-get it since now cpos may be moved to
1901                  * another leaf block.
1902                  */
1903                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1904                                              cpos, len, &tmp_rec, &index,
1905                                              &new_bh);
1906                 if (ret) {
1907                         mlog_errno(ret);
1908                         goto out;
1909                 }
1910
1911                 ref_leaf_bh = new_bh;
1912                 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1913                 rf_list = &rb->rf_records;
1914                 orig_rec = &rf_list->rl_recs[index];
1915         }
1916
1917         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1918                                       OCFS2_JOURNAL_ACCESS_WRITE);
1919         if (ret) {
1920                 mlog_errno(ret);
1921                 goto out;
1922         }
1923
1924         /*
1925          * We have calculated out how many new records we need and store
1926          * in recs_need, so spare enough space first by moving the records
1927          * after "index" to the end.
1928          */
1929         if (index != le16_to_cpu(rf_list->rl_used) - 1)
1930                 memmove(&rf_list->rl_recs[index + 1 + recs_need],
1931                         &rf_list->rl_recs[index + 1],
1932                         (le16_to_cpu(rf_list->rl_used) - index - 1) *
1933                          sizeof(struct ocfs2_refcount_rec));
1934
1935         len = (le64_to_cpu(orig_rec->r_cpos) +
1936               le32_to_cpu(orig_rec->r_clusters)) -
1937               (le64_to_cpu(split_rec->r_cpos) +
1938               le32_to_cpu(split_rec->r_clusters));
1939
1940         /*
1941          * If we have "len", the we will split in the tail and move it
1942          * to the end of the space we have just spared.
1943          */
1944         if (len) {
1945                 tail_rec = &rf_list->rl_recs[index + recs_need];
1946
1947                 memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
1948                 le64_add_cpu(&tail_rec->r_cpos,
1949                              le32_to_cpu(tail_rec->r_clusters) - len);
1950                 tail_rec->r_clusters = cpu_to_le32(len);
1951         }
1952
1953         /*
1954          * If the split pos isn't the same as the original one, we need to
1955          * split in the head.
1956          *
1957          * Note: We have the chance that split_rec.r_refcount = 0,
1958          * recs_need = 0 and len > 0, which means we just cut the head from
1959          * the orig_rec and in that case we have done some modification in
1960          * orig_rec above, so the check for r_cpos is faked.
1961          */
1962         if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
1963                 len = le64_to_cpu(split_rec->r_cpos) -
1964                       le64_to_cpu(orig_rec->r_cpos);
1965                 orig_rec->r_clusters = cpu_to_le32(len);
1966                 index++;
1967         }
1968
1969         le16_add_cpu(&rf_list->rl_used, recs_need);
1970
1971         if (split_rec->r_refcount) {
1972                 rf_list->rl_recs[index] = *split_rec;
1973                 trace_ocfs2_split_refcount_rec_insert(
1974                         (unsigned long long)ref_leaf_bh->b_blocknr, index,
1975                         (unsigned long long)le64_to_cpu(split_rec->r_cpos),
1976                         le32_to_cpu(split_rec->r_clusters),
1977                         le32_to_cpu(split_rec->r_refcount));
1978
1979                 if (merge)
1980                         ocfs2_refcount_rec_merge(rb, index);
1981         }
1982
1983         ocfs2_journal_dirty(handle, ref_leaf_bh);
1984
1985 out:
1986         brelse(new_bh);
1987         return ret;
1988 }
1989
1990 static int __ocfs2_increase_refcount(handle_t *handle,
1991                                      struct ocfs2_caching_info *ci,
1992                                      struct buffer_head *ref_root_bh,
1993                                      u64 cpos, u32 len, int merge,
1994                                      struct ocfs2_alloc_context *meta_ac,
1995                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
1996 {
1997         int ret = 0, index;
1998         struct buffer_head *ref_leaf_bh = NULL;
1999         struct ocfs2_refcount_rec rec;
2000         unsigned int set_len = 0;
2001
2002         trace_ocfs2_increase_refcount_begin(
2003              (unsigned long long)ocfs2_metadata_cache_owner(ci),
2004              (unsigned long long)cpos, len);
2005
2006         while (len) {
2007                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2008                                              cpos, len, &rec, &index,
2009                                              &ref_leaf_bh);
2010                 if (ret) {
2011                         mlog_errno(ret);
2012                         goto out;
2013                 }
2014
2015                 set_len = le32_to_cpu(rec.r_clusters);
2016
2017                 /*
2018                  * Here we may meet with 3 situations:
2019                  *
2020                  * 1. If we find an already existing record, and the length
2021                  *    is the same, cool, we just need to increase the r_refcount
2022                  *    and it is OK.
2023                  * 2. If we find a hole, just insert it with r_refcount = 1.
2024                  * 3. If we are in the middle of one extent record, split
2025                  *    it.
2026                  */
2027                 if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
2028                     set_len <= len) {
2029                         trace_ocfs2_increase_refcount_change(
2030                                 (unsigned long long)cpos, set_len,
2031                                 le32_to_cpu(rec.r_refcount));
2032                         ret = ocfs2_change_refcount_rec(handle, ci,
2033                                                         ref_leaf_bh, index,
2034                                                         merge, 1);
2035                         if (ret) {
2036                                 mlog_errno(ret);
2037                                 goto out;
2038                         }
2039                 } else if (!rec.r_refcount) {
2040                         rec.r_refcount = cpu_to_le32(1);
2041
2042                         trace_ocfs2_increase_refcount_insert(
2043                              (unsigned long long)le64_to_cpu(rec.r_cpos),
2044                              set_len);
2045                         ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
2046                                                         ref_leaf_bh,
2047                                                         &rec, index,
2048                                                         merge, meta_ac);
2049                         if (ret) {
2050                                 mlog_errno(ret);
2051                                 goto out;
2052                         }
2053                 } else  {
2054                         set_len = min((u64)(cpos + len),
2055                                       le64_to_cpu(rec.r_cpos) + set_len) - cpos;
2056                         rec.r_cpos = cpu_to_le64(cpos);
2057                         rec.r_clusters = cpu_to_le32(set_len);
2058                         le32_add_cpu(&rec.r_refcount, 1);
2059
2060                         trace_ocfs2_increase_refcount_split(
2061                              (unsigned long long)le64_to_cpu(rec.r_cpos),
2062                              set_len, le32_to_cpu(rec.r_refcount));
2063                         ret = ocfs2_split_refcount_rec(handle, ci,
2064                                                        ref_root_bh, ref_leaf_bh,
2065                                                        &rec, index, merge,
2066                                                        meta_ac, dealloc);
2067                         if (ret) {
2068                                 mlog_errno(ret);
2069                                 goto out;
2070                         }
2071                 }
2072
2073                 cpos += set_len;
2074                 len -= set_len;
2075                 brelse(ref_leaf_bh);
2076                 ref_leaf_bh = NULL;
2077         }
2078
2079 out:
2080         brelse(ref_leaf_bh);
2081         return ret;
2082 }
2083
2084 static int ocfs2_remove_refcount_extent(handle_t *handle,
2085                                 struct ocfs2_caching_info *ci,
2086                                 struct buffer_head *ref_root_bh,
2087                                 struct buffer_head *ref_leaf_bh,
2088                                 struct ocfs2_alloc_context *meta_ac,
2089                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
2090 {
2091         int ret;
2092         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2093         struct ocfs2_refcount_block *rb =
2094                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2095         struct ocfs2_extent_tree et;
2096
2097         BUG_ON(rb->rf_records.rl_used);
2098
2099         trace_ocfs2_remove_refcount_extent(
2100                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2101                 (unsigned long long)ref_leaf_bh->b_blocknr,
2102                 le32_to_cpu(rb->rf_cpos));
2103
2104         ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2105         ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
2106                                   1, meta_ac, dealloc);
2107         if (ret) {
2108                 mlog_errno(ret);
2109                 goto out;
2110         }
2111
2112         ocfs2_remove_from_cache(ci, ref_leaf_bh);
2113
2114         /*
2115          * add the freed block to the dealloc so that it will be freed
2116          * when we run dealloc.
2117          */
2118         ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
2119                                         le16_to_cpu(rb->rf_suballoc_slot),
2120                                         le64_to_cpu(rb->rf_suballoc_loc),
2121                                         le64_to_cpu(rb->rf_blkno),
2122                                         le16_to_cpu(rb->rf_suballoc_bit));
2123         if (ret) {
2124                 mlog_errno(ret);
2125                 goto out;
2126         }
2127
2128         ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
2129                                       OCFS2_JOURNAL_ACCESS_WRITE);
2130         if (ret) {
2131                 mlog_errno(ret);
2132                 goto out;
2133         }
2134
2135         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2136
2137         le32_add_cpu(&rb->rf_clusters, -1);
2138
2139         /*
2140          * check whether we need to restore the root refcount block if
2141          * there is no leaf extent block at atll.
2142          */
2143         if (!rb->rf_list.l_next_free_rec) {
2144                 BUG_ON(rb->rf_clusters);
2145
2146                 trace_ocfs2_restore_refcount_block(
2147                      (unsigned long long)ref_root_bh->b_blocknr);
2148
2149                 rb->rf_flags = 0;
2150                 rb->rf_parent = 0;
2151                 rb->rf_cpos = 0;
2152                 memset(&rb->rf_records, 0, sb->s_blocksize -
2153                        offsetof(struct ocfs2_refcount_block, rf_records));
2154                 rb->rf_records.rl_count =
2155                                 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
2156         }
2157
2158         ocfs2_journal_dirty(handle, ref_root_bh);
2159
2160 out:
2161         return ret;
2162 }
2163
2164 int ocfs2_increase_refcount(handle_t *handle,
2165                             struct ocfs2_caching_info *ci,
2166                             struct buffer_head *ref_root_bh,
2167                             u64 cpos, u32 len,
2168                             struct ocfs2_alloc_context *meta_ac,
2169                             struct ocfs2_cached_dealloc_ctxt *dealloc)
2170 {
2171         return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
2172                                          cpos, len, 1,
2173                                          meta_ac, dealloc);
2174 }
2175
2176 static int ocfs2_decrease_refcount_rec(handle_t *handle,
2177                                 struct ocfs2_caching_info *ci,
2178                                 struct buffer_head *ref_root_bh,
2179                                 struct buffer_head *ref_leaf_bh,
2180                                 int index, u64 cpos, unsigned int len,
2181                                 struct ocfs2_alloc_context *meta_ac,
2182                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
2183 {
2184         int ret;
2185         struct ocfs2_refcount_block *rb =
2186                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2187         struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
2188
2189         BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
2190         BUG_ON(cpos + len >
2191                le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
2192
2193         trace_ocfs2_decrease_refcount_rec(
2194                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2195                 (unsigned long long)cpos, len);
2196
2197         if (cpos == le64_to_cpu(rec->r_cpos) &&
2198             len == le32_to_cpu(rec->r_clusters))
2199                 ret = ocfs2_change_refcount_rec(handle, ci,
2200                                                 ref_leaf_bh, index, 1, -1);
2201         else {
2202                 struct ocfs2_refcount_rec split = *rec;
2203                 split.r_cpos = cpu_to_le64(cpos);
2204                 split.r_clusters = cpu_to_le32(len);
2205
2206                 le32_add_cpu(&split.r_refcount, -1);
2207
2208                 ret = ocfs2_split_refcount_rec(handle, ci,
2209                                                ref_root_bh, ref_leaf_bh,
2210                                                &split, index, 1,
2211                                                meta_ac, dealloc);
2212         }
2213
2214         if (ret) {
2215                 mlog_errno(ret);
2216                 goto out;
2217         }
2218
2219         /* Remove the leaf refcount block if it contains no refcount record. */
2220         if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
2221                 ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
2222                                                    ref_leaf_bh, meta_ac,
2223                                                    dealloc);
2224                 if (ret)
2225                         mlog_errno(ret);
2226         }
2227
2228 out:
2229         return ret;
2230 }
2231
2232 static int __ocfs2_decrease_refcount(handle_t *handle,
2233                                      struct ocfs2_caching_info *ci,
2234                                      struct buffer_head *ref_root_bh,
2235                                      u64 cpos, u32 len,
2236                                      struct ocfs2_alloc_context *meta_ac,
2237                                      struct ocfs2_cached_dealloc_ctxt *dealloc,
2238                                      int delete)
2239 {
2240         int ret = 0, index = 0;
2241         struct ocfs2_refcount_rec rec;
2242         unsigned int r_count = 0, r_len;
2243         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2244         struct buffer_head *ref_leaf_bh = NULL;
2245
2246         trace_ocfs2_decrease_refcount(
2247                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2248                 (unsigned long long)cpos, len, delete);
2249
2250         while (len) {
2251                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2252                                              cpos, len, &rec, &index,
2253                                              &ref_leaf_bh);
2254                 if (ret) {
2255                         mlog_errno(ret);
2256                         goto out;
2257                 }
2258
2259                 r_count = le32_to_cpu(rec.r_refcount);
2260                 BUG_ON(r_count == 0);
2261                 if (!delete)
2262                         BUG_ON(r_count > 1);
2263
2264                 r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
2265                               le32_to_cpu(rec.r_clusters)) - cpos;
2266
2267                 ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
2268                                                   ref_leaf_bh, index,
2269                                                   cpos, r_len,
2270                                                   meta_ac, dealloc);
2271                 if (ret) {
2272                         mlog_errno(ret);
2273                         goto out;
2274                 }
2275
2276                 if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
2277                         ret = ocfs2_cache_cluster_dealloc(dealloc,
2278                                           ocfs2_clusters_to_blocks(sb, cpos),
2279                                                           r_len);
2280                         if (ret) {
2281                                 mlog_errno(ret);
2282                                 goto out;
2283                         }
2284                 }
2285
2286                 cpos += r_len;
2287                 len -= r_len;
2288                 brelse(ref_leaf_bh);
2289                 ref_leaf_bh = NULL;
2290         }
2291
2292 out:
2293         brelse(ref_leaf_bh);
2294         return ret;
2295 }
2296
2297 /* Caller must hold refcount tree lock. */
2298 int ocfs2_decrease_refcount(struct inode *inode,
2299                             handle_t *handle, u32 cpos, u32 len,
2300                             struct ocfs2_alloc_context *meta_ac,
2301                             struct ocfs2_cached_dealloc_ctxt *dealloc,
2302                             int delete)
2303 {
2304         int ret;
2305         u64 ref_blkno;
2306         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2307         struct buffer_head *ref_root_bh = NULL;
2308         struct ocfs2_refcount_tree *tree;
2309
2310         BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2311
2312         ret = ocfs2_get_refcount_block(inode, &ref_blkno);
2313         if (ret) {
2314                 mlog_errno(ret);
2315                 goto out;
2316         }
2317
2318         ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
2319         if (ret) {
2320                 mlog_errno(ret);
2321                 goto out;
2322         }
2323
2324         ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
2325                                         &ref_root_bh);
2326         if (ret) {
2327                 mlog_errno(ret);
2328                 goto out;
2329         }
2330
2331         ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
2332                                         cpos, len, meta_ac, dealloc, delete);
2333         if (ret)
2334                 mlog_errno(ret);
2335 out:
2336         brelse(ref_root_bh);
2337         return ret;
2338 }
2339
2340 /*
2341  * Mark the already-existing extent at cpos as refcounted for len clusters.
2342  * This adds the refcount extent flag.
2343  *
2344  * If the existing extent is larger than the request, initiate a
2345  * split. An attempt will be made at merging with adjacent extents.
2346  *
2347  * The caller is responsible for passing down meta_ac if we'll need it.
2348  */
2349 static int ocfs2_mark_extent_refcounted(struct inode *inode,
2350                                 struct ocfs2_extent_tree *et,
2351                                 handle_t *handle, u32 cpos,
2352                                 u32 len, u32 phys,
2353                                 struct ocfs2_alloc_context *meta_ac,
2354                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
2355 {
2356         int ret;
2357
2358         trace_ocfs2_mark_extent_refcounted(OCFS2_I(inode)->ip_blkno,
2359                                            cpos, len, phys);
2360
2361         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2362                 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2363                             "tree, but the feature bit is not set in the "
2364                             "super block.", inode->i_ino);
2365                 ret = -EROFS;
2366                 goto out;
2367         }
2368
2369         ret = ocfs2_change_extent_flag(handle, et, cpos,
2370                                        len, phys, meta_ac, dealloc,
2371                                        OCFS2_EXT_REFCOUNTED, 0);
2372         if (ret)
2373                 mlog_errno(ret);
2374
2375 out:
2376         return ret;
2377 }
2378
2379 /*
2380  * Given some contiguous physical clusters, calculate what we need
2381  * for modifying their refcount.
2382  */
2383 static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
2384                                             struct ocfs2_caching_info *ci,
2385                                             struct buffer_head *ref_root_bh,
2386                                             u64 start_cpos,
2387                                             u32 clusters,
2388                                             int *meta_add,
2389                                             int *credits)
2390 {
2391         int ret = 0, index, ref_blocks = 0, recs_add = 0;
2392         u64 cpos = start_cpos;
2393         struct ocfs2_refcount_block *rb;
2394         struct ocfs2_refcount_rec rec;
2395         struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
2396         u32 len;
2397
2398         while (clusters) {
2399                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2400                                              cpos, clusters, &rec,
2401                                              &index, &ref_leaf_bh);
2402                 if (ret) {
2403                         mlog_errno(ret);
2404                         goto out;
2405                 }
2406
2407                 if (ref_leaf_bh != prev_bh) {
2408                         /*
2409                          * Now we encounter a new leaf block, so calculate
2410                          * whether we need to extend the old leaf.
2411                          */
2412                         if (prev_bh) {
2413                                 rb = (struct ocfs2_refcount_block *)
2414                                                         prev_bh->b_data;
2415
2416                                 if (le16_to_cpu(rb->rf_records.rl_used) +
2417                                     recs_add >
2418                                     le16_to_cpu(rb->rf_records.rl_count))
2419                                         ref_blocks++;
2420                         }
2421
2422                         recs_add = 0;
2423                         *credits += 1;
2424                         brelse(prev_bh);
2425                         prev_bh = ref_leaf_bh;
2426                         get_bh(prev_bh);
2427                 }
2428
2429                 trace_ocfs2_calc_refcount_meta_credits_iterate(
2430                                 recs_add, (unsigned long long)cpos, clusters,
2431                                 (unsigned long long)le64_to_cpu(rec.r_cpos),
2432                                 le32_to_cpu(rec.r_clusters),
2433                                 le32_to_cpu(rec.r_refcount), index);
2434
2435                 len = min((u64)cpos + clusters, le64_to_cpu(rec.r_cpos) +
2436                           le32_to_cpu(rec.r_clusters)) - cpos;
2437                 /*
2438                  * We record all the records which will be inserted to the
2439                  * same refcount block, so that we can tell exactly whether
2440                  * we need a new refcount block or not.
2441                  *
2442                  * If we will insert a new one, this is easy and only happens
2443                  * during adding refcounted flag to the extent, so we don't
2444                  * have a chance of spliting. We just need one record.
2445                  *
2446                  * If the refcount rec already exists, that would be a little
2447                  * complicated. we may have to:
2448                  * 1) split at the beginning if the start pos isn't aligned.
2449                  *    we need 1 more record in this case.
2450                  * 2) split int the end if the end pos isn't aligned.
2451                  *    we need 1 more record in this case.
2452                  * 3) split in the middle because of file system fragmentation.
2453                  *    we need 2 more records in this case(we can't detect this
2454                  *    beforehand, so always think of the worst case).
2455                  */
2456                 if (rec.r_refcount) {
2457                         recs_add += 2;
2458                         /* Check whether we need a split at the beginning. */
2459                         if (cpos == start_cpos &&
2460                             cpos != le64_to_cpu(rec.r_cpos))
2461                                 recs_add++;
2462
2463                         /* Check whether we need a split in the end. */
2464                         if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
2465                             le32_to_cpu(rec.r_clusters))
2466                                 recs_add++;
2467                 } else
2468                         recs_add++;
2469
2470                 brelse(ref_leaf_bh);
2471                 ref_leaf_bh = NULL;
2472                 clusters -= len;
2473                 cpos += len;
2474         }
2475
2476         if (prev_bh) {
2477                 rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
2478
2479                 if (le16_to_cpu(rb->rf_records.rl_used) + recs_add >
2480                     le16_to_cpu(rb->rf_records.rl_count))
2481                         ref_blocks++;
2482
2483                 *credits += 1;
2484         }
2485
2486         if (!ref_blocks)
2487                 goto out;
2488
2489         *meta_add += ref_blocks;
2490         *credits += ref_blocks;
2491
2492         /*
2493          * So we may need ref_blocks to insert into the tree.
2494          * That also means we need to change the b-tree and add that number
2495          * of records since we never merge them.
2496          * We need one more block for expansion since the new created leaf
2497          * block is also full and needs split.
2498          */
2499         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2500         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
2501                 struct ocfs2_extent_tree et;
2502
2503                 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2504                 *meta_add += ocfs2_extend_meta_needed(et.et_root_el);
2505                 *credits += ocfs2_calc_extend_credits(sb,
2506                                                       et.et_root_el);
2507         } else {
2508                 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
2509                 *meta_add += 1;
2510         }
2511
2512 out:
2513
2514         trace_ocfs2_calc_refcount_meta_credits(
2515                 (unsigned long long)start_cpos, clusters,
2516                 *meta_add, *credits);
2517         brelse(ref_leaf_bh);
2518         brelse(prev_bh);
2519         return ret;
2520 }
2521
2522 /*
2523  * For refcount tree, we will decrease some contiguous clusters
2524  * refcount count, so just go through it to see how many blocks
2525  * we gonna touch and whether we need to create new blocks.
2526  *
2527  * Normally the refcount blocks store these refcount should be
2528  * contiguous also, so that we can get the number easily.
2529  * We will at most add split 2 refcount records and 2 more
2530  * refcount blocks, so just check it in a rough way.
2531  *
2532  * Caller must hold refcount tree lock.
2533  */
2534 int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
2535                                           u64 refcount_loc,
2536                                           u64 phys_blkno,
2537                                           u32 clusters,
2538                                           int *credits,
2539                                           int *ref_blocks)
2540 {
2541         int ret;
2542         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2543         struct buffer_head *ref_root_bh = NULL;
2544         struct ocfs2_refcount_tree *tree;
2545         u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
2546
2547         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2548                 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2549                             "tree, but the feature bit is not set in the "
2550                             "super block.", inode->i_ino);
2551                 ret = -EROFS;
2552                 goto out;
2553         }
2554
2555         BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2556
2557         ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
2558                                       refcount_loc, &tree);
2559         if (ret) {
2560                 mlog_errno(ret);
2561                 goto out;
2562         }
2563
2564         ret = ocfs2_read_refcount_block(&tree->rf_ci, refcount_loc,
2565                                         &ref_root_bh);
2566         if (ret) {
2567                 mlog_errno(ret);
2568                 goto out;
2569         }
2570
2571         ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
2572                                                &tree->rf_ci,
2573                                                ref_root_bh,
2574                                                start_cpos, clusters,
2575                                                ref_blocks, credits);
2576         if (ret) {
2577                 mlog_errno(ret);
2578                 goto out;
2579         }
2580
2581         trace_ocfs2_prepare_refcount_change_for_del(*ref_blocks, *credits);
2582
2583 out:
2584         brelse(ref_root_bh);
2585         return ret;
2586 }
2587
2588 #define MAX_CONTIG_BYTES        1048576
2589
2590 static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
2591 {
2592         return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
2593 }
2594
2595 static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
2596 {
2597         return ~(ocfs2_cow_contig_clusters(sb) - 1);
2598 }
2599
2600 /*
2601  * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2602  * find an offset (start + (n * contig_clusters)) that is closest to cpos
2603  * while still being less than or equal to it.
2604  *
2605  * The goal is to break the extent at a multiple of contig_clusters.
2606  */
2607 static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
2608                                                  unsigned int start,
2609                                                  unsigned int cpos)
2610 {
2611         BUG_ON(start > cpos);
2612
2613         return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
2614 }
2615
2616 /*
2617  * Given a cluster count of len, pad it out so that it is a multiple
2618  * of contig_clusters.
2619  */
2620 static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
2621                                                   unsigned int len)
2622 {
2623         unsigned int padded =
2624                 (len + (ocfs2_cow_contig_clusters(sb) - 1)) &
2625                 ocfs2_cow_contig_mask(sb);
2626
2627         /* Did we wrap? */
2628         if (padded < len)
2629                 padded = UINT_MAX;
2630
2631         return padded;
2632 }
2633
2634 /*
2635  * Calculate out the start and number of virtual clusters we need to to CoW.
2636  *
2637  * cpos is vitual start cluster position we want to do CoW in a
2638  * file and write_len is the cluster length.
2639  * max_cpos is the place where we want to stop CoW intentionally.
2640  *
2641  * Normal we will start CoW from the beginning of extent record cotaining cpos.
2642  * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2643  * get good I/O from the resulting extent tree.
2644  */
2645 static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
2646                                            struct ocfs2_extent_list *el,
2647                                            u32 cpos,
2648                                            u32 write_len,
2649                                            u32 max_cpos,
2650                                            u32 *cow_start,
2651                                            u32 *cow_len)
2652 {
2653         int ret = 0;
2654         int tree_height = le16_to_cpu(el->l_tree_depth), i;
2655         struct buffer_head *eb_bh = NULL;
2656         struct ocfs2_extent_block *eb = NULL;
2657         struct ocfs2_extent_rec *rec;
2658         unsigned int want_clusters, rec_end = 0;
2659         int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
2660         int leaf_clusters;
2661
2662         BUG_ON(cpos + write_len > max_cpos);
2663
2664         if (tree_height > 0) {
2665                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
2666                 if (ret) {
2667                         mlog_errno(ret);
2668                         goto out;
2669                 }
2670
2671                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2672                 el = &eb->h_list;
2673
2674                 if (el->l_tree_depth) {
2675                         ocfs2_error(inode->i_sb,
2676                                     "Inode %lu has non zero tree depth in "
2677                                     "leaf block %llu\n", inode->i_ino,
2678                                     (unsigned long long)eb_bh->b_blocknr);
2679                         ret = -EROFS;
2680                         goto out;
2681                 }
2682         }
2683
2684         *cow_len = 0;
2685         for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
2686                 rec = &el->l_recs[i];
2687
2688                 if (ocfs2_is_empty_extent(rec)) {
2689                         mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
2690                                         "index %d\n", inode->i_ino, i);
2691                         continue;
2692                 }
2693
2694                 if (le32_to_cpu(rec->e_cpos) +
2695                     le16_to_cpu(rec->e_leaf_clusters) <= cpos)
2696                         continue;
2697
2698                 if (*cow_len == 0) {
2699                         /*
2700                          * We should find a refcounted record in the
2701                          * first pass.
2702                          */
2703                         BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
2704                         *cow_start = le32_to_cpu(rec->e_cpos);
2705                 }
2706
2707                 /*
2708                  * If we encounter a hole, a non-refcounted record or
2709                  * pass the max_cpos, stop the search.
2710                  */
2711                 if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
2712                     (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
2713                     (max_cpos <= le32_to_cpu(rec->e_cpos)))
2714                         break;
2715
2716                 leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
2717                 rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
2718                 if (rec_end > max_cpos) {
2719                         rec_end = max_cpos;
2720                         leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
2721                 }
2722
2723                 /*
2724                  * How many clusters do we actually need from
2725                  * this extent?  First we see how many we actually
2726                  * need to complete the write.  If that's smaller
2727                  * than contig_clusters, we try for contig_clusters.
2728                  */
2729                 if (!*cow_len)
2730                         want_clusters = write_len;
2731                 else
2732                         want_clusters = (cpos + write_len) -
2733                                 (*cow_start + *cow_len);
2734                 if (want_clusters < contig_clusters)
2735                         want_clusters = contig_clusters;
2736
2737                 /*
2738                  * If the write does not cover the whole extent, we
2739                  * need to calculate how we're going to split the extent.
2740                  * We try to do it on contig_clusters boundaries.
2741                  *
2742                  * Any extent smaller than contig_clusters will be
2743                  * CoWed in its entirety.
2744                  */
2745                 if (leaf_clusters <= contig_clusters)
2746                         *cow_len += leaf_clusters;
2747                 else if (*cow_len || (*cow_start == cpos)) {
2748                         /*
2749                          * This extent needs to be CoW'd from its
2750                          * beginning, so all we have to do is compute
2751                          * how many clusters to grab.  We align
2752                          * want_clusters to the edge of contig_clusters
2753                          * to get better I/O.
2754                          */
2755                         want_clusters = ocfs2_cow_align_length(inode->i_sb,
2756                                                                want_clusters);
2757
2758                         if (leaf_clusters < want_clusters)
2759                                 *cow_len += leaf_clusters;
2760                         else
2761                                 *cow_len += want_clusters;
2762                 } else if ((*cow_start + contig_clusters) >=
2763                            (cpos + write_len)) {
2764                         /*
2765                          * Breaking off contig_clusters at the front
2766                          * of the extent will cover our write.  That's
2767                          * easy.
2768                          */
2769                         *cow_len = contig_clusters;
2770                 } else if ((rec_end - cpos) <= contig_clusters) {
2771                         /*
2772                          * Breaking off contig_clusters at the tail of
2773                          * this extent will cover cpos.
2774                          */
2775                         *cow_start = rec_end - contig_clusters;
2776                         *cow_len = contig_clusters;
2777                 } else if ((rec_end - cpos) <= want_clusters) {
2778                         /*
2779                          * While we can't fit the entire write in this
2780                          * extent, we know that the write goes from cpos
2781                          * to the end of the extent.  Break that off.
2782                          * We try to break it at some multiple of
2783                          * contig_clusters from the front of the extent.
2784                          * Failing that (ie, cpos is within
2785                          * contig_clusters of the front), we'll CoW the
2786                          * entire extent.
2787                          */
2788                         *cow_start = ocfs2_cow_align_start(inode->i_sb,
2789                                                            *cow_start, cpos);
2790                         *cow_len = rec_end - *cow_start;
2791                 } else {
2792                         /*
2793                          * Ok, the entire write lives in the middle of
2794                          * this extent.  Let's try to slice the extent up
2795                          * nicely.  Optimally, our CoW region starts at
2796                          * m*contig_clusters from the beginning of the
2797                          * extent and goes for n*contig_clusters,
2798                          * covering the entire write.
2799                          */
2800                         *cow_start = ocfs2_cow_align_start(inode->i_sb,
2801                                                            *cow_start, cpos);
2802
2803                         want_clusters = (cpos + write_len) - *cow_start;
2804                         want_clusters = ocfs2_cow_align_length(inode->i_sb,
2805                                                                want_clusters);
2806                         if (*cow_start + want_clusters <= rec_end)
2807                                 *cow_len = want_clusters;
2808                         else
2809                                 *cow_len = rec_end - *cow_start;
2810                 }
2811
2812                 /* Have we covered our entire write yet? */
2813                 if ((*cow_start + *cow_len) >= (cpos + write_len))
2814                         break;
2815
2816                 /*
2817                  * If we reach the end of the extent block and don't get enough
2818                  * clusters, continue with the next extent block if possible.
2819                  */
2820                 if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
2821                     eb && eb->h_next_leaf_blk) {
2822                         brelse(eb_bh);
2823                         eb_bh = NULL;
2824
2825                         ret = ocfs2_read_extent_block(INODE_CACHE(inode),
2826                                                le64_to_cpu(eb->h_next_leaf_blk),
2827                                                &eb_bh);
2828                         if (ret) {
2829                                 mlog_errno(ret);
2830                                 goto out;
2831                         }
2832
2833                         eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2834                         el = &eb->h_list;
2835                         i = -1;
2836                 }
2837         }
2838
2839 out:
2840         brelse(eb_bh);
2841         return ret;
2842 }
2843
2844 /*
2845  * Prepare meta_ac, data_ac and calculate credits when we want to add some
2846  * num_clusters in data_tree "et" and change the refcount for the old
2847  * clusters(starting form p_cluster) in the refcount tree.
2848  *
2849  * Note:
2850  * 1. since we may split the old tree, so we at most will need num_clusters + 2
2851  *    more new leaf records.
2852  * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2853  *    just give data_ac = NULL.
2854  */
2855 static int ocfs2_lock_refcount_allocators(struct super_block *sb,
2856                                         u32 p_cluster, u32 num_clusters,
2857                                         struct ocfs2_extent_tree *et,
2858                                         struct ocfs2_caching_info *ref_ci,
2859                                         struct buffer_head *ref_root_bh,
2860                                         struct ocfs2_alloc_context **meta_ac,
2861                                         struct ocfs2_alloc_context **data_ac,
2862                                         int *credits)
2863 {
2864         int ret = 0, meta_add = 0;
2865         int num_free_extents = ocfs2_num_free_extents(OCFS2_SB(sb), et);
2866
2867         if (num_free_extents < 0) {
2868                 ret = num_free_extents;
2869                 mlog_errno(ret);
2870                 goto out;
2871         }
2872
2873         if (num_free_extents < num_clusters + 2)
2874                 meta_add =
2875                         ocfs2_extend_meta_needed(et->et_root_el);
2876
2877         *credits += ocfs2_calc_extend_credits(sb, et->et_root_el);
2878
2879         ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
2880                                                p_cluster, num_clusters,
2881                                                &meta_add, credits);
2882         if (ret) {
2883                 mlog_errno(ret);
2884                 goto out;
2885         }
2886
2887         trace_ocfs2_lock_refcount_allocators(meta_add, *credits);
2888         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
2889                                                 meta_ac);
2890         if (ret) {
2891                 mlog_errno(ret);
2892                 goto out;
2893         }
2894
2895         if (data_ac) {
2896                 ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
2897                                              data_ac);
2898                 if (ret)
2899                         mlog_errno(ret);
2900         }
2901
2902 out:
2903         if (ret) {
2904                 if (*meta_ac) {
2905                         ocfs2_free_alloc_context(*meta_ac);
2906                         *meta_ac = NULL;
2907                 }
2908         }
2909
2910         return ret;
2911 }
2912
2913 static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
2914 {
2915         BUG_ON(buffer_dirty(bh));
2916
2917         clear_buffer_mapped(bh);
2918
2919         return 0;
2920 }
2921
2922 int ocfs2_duplicate_clusters_by_page(handle_t *handle,
2923                                      struct inode *inode,
2924                                      u32 cpos, u32 old_cluster,
2925                                      u32 new_cluster, u32 new_len)
2926 {
2927         int ret = 0, partial;
2928         struct super_block *sb = inode->i_sb;
2929         u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
2930         struct page *page;
2931         pgoff_t page_index;
2932         unsigned int from, to, readahead_pages;
2933         loff_t offset, end, map_end;
2934         struct address_space *mapping = inode->i_mapping;
2935
2936         trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
2937                                                new_cluster, new_len);
2938
2939         readahead_pages =
2940                 (ocfs2_cow_contig_clusters(sb) <<
2941                  OCFS2_SB(sb)->s_clustersize_bits) >> PAGE_CACHE_SHIFT;
2942         offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
2943         end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
2944         /*
2945          * We only duplicate pages until we reach the page contains i_size - 1.
2946          * So trim 'end' to i_size.
2947          */
2948         if (end > i_size_read(inode))
2949                 end = i_size_read(inode);
2950
2951         while (offset < end) {
2952                 page_index = offset >> PAGE_CACHE_SHIFT;
2953                 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
2954                 if (map_end > end)
2955                         map_end = end;
2956
2957                 /* from, to is the offset within the page. */
2958                 from = offset & (PAGE_CACHE_SIZE - 1);
2959                 to = PAGE_CACHE_SIZE;
2960                 if (map_end & (PAGE_CACHE_SIZE - 1))
2961                         to = map_end & (PAGE_CACHE_SIZE - 1);
2962
2963                 page = find_or_create_page(mapping, page_index, GFP_NOFS);
2964                 if (!page) {
2965                         ret = -ENOMEM;
2966                         mlog_errno(ret);
2967                         break;
2968                 }
2969
2970                 /*
2971                  * In case PAGE_CACHE_SIZE <= CLUSTER_SIZE, This page
2972                  * can't be dirtied before we CoW it out.
2973                  */
2974                 if (PAGE_CACHE_SIZE <= OCFS2_SB(sb)->s_clustersize)
2975                         BUG_ON(PageDirty(page));
2976
2977                 if (!PageUptodate(page)) {
2978                         ret = block_read_full_page(page, ocfs2_get_block);
2979                         if (ret) {
2980                                 mlog_errno(ret);
2981                                 goto unlock;
2982                         }
2983                         lock_page(page);
2984                 }
2985
2986                 if (page_has_buffers(page)) {
2987                         ret = walk_page_buffers(handle, page_buffers(page),
2988                                                 from, to, &partial,
2989                                                 ocfs2_clear_cow_buffer);
2990                         if (ret) {
2991                                 mlog_errno(ret);
2992                                 goto unlock;
2993                         }
2994                 }
2995
2996                 ocfs2_map_and_dirty_page(inode,
2997                                          handle, from, to,
2998                                          page, 0, &new_block);
2999                 mark_page_accessed(page);
3000 unlock:
3001                 unlock_page(page);
3002                 page_cache_release(page);
3003                 page = NULL;
3004                 offset = map_end;
3005                 if (ret)
3006                         break;
3007         }
3008
3009         return ret;
3010 }
3011
3012 int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
3013                                     struct inode *inode,
3014                                     u32 cpos, u32 old_cluster,
3015                                     u32 new_cluster, u32 new_len)
3016 {
3017         int ret = 0;
3018         struct super_block *sb = inode->i_sb;
3019         struct ocfs2_caching_info *ci = INODE_CACHE(inode);
3020         int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
3021         u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
3022         u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
3023         struct ocfs2_super *osb = OCFS2_SB(sb);
3024         struct buffer_head *old_bh = NULL;
3025         struct buffer_head *new_bh = NULL;
3026
3027         trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
3028                                                new_cluster, new_len);
3029
3030         for (i = 0; i < blocks; i++, old_block++, new_block++) {
3031                 new_bh = sb_getblk(osb->sb, new_block);
3032                 if (new_bh == NULL) {
3033                         ret = -ENOMEM;
3034                         mlog_errno(ret);
3035                         break;
3036                 }
3037
3038                 ocfs2_set_new_buffer_uptodate(ci, new_bh);
3039
3040                 ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
3041                 if (ret) {
3042                         mlog_errno(ret);
3043                         break;
3044                 }
3045
3046                 ret = ocfs2_journal_access(handle, ci, new_bh,
3047                                            OCFS2_JOURNAL_ACCESS_CREATE);
3048                 if (ret) {
3049                         mlog_errno(ret);
3050                         break;
3051                 }
3052
3053                 memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
3054                 ocfs2_journal_dirty(handle, new_bh);
3055
3056                 brelse(new_bh);
3057                 brelse(old_bh);
3058                 new_bh = NULL;
3059                 old_bh = NULL;
3060         }
3061
3062         brelse(new_bh);
3063         brelse(old_bh);
3064         return ret;
3065 }
3066
3067 static int ocfs2_clear_ext_refcount(handle_t *handle,
3068                                     struct ocfs2_extent_tree *et,
3069                                     u32 cpos, u32 p_cluster, u32 len,
3070                                     unsigned int ext_flags,
3071                                     struct ocfs2_alloc_context *meta_ac,
3072                                     struct ocfs2_cached_dealloc_ctxt *dealloc)
3073 {
3074         int ret, index;
3075         struct ocfs2_extent_rec replace_rec;
3076         struct ocfs2_path *path = NULL;
3077         struct ocfs2_extent_list *el;
3078         struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
3079         u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
3080
3081         trace_ocfs2_clear_ext_refcount((unsigned long long)ino,
3082                                        cpos, len, p_cluster, ext_flags);
3083
3084         memset(&replace_rec, 0, sizeof(replace_rec));
3085         replace_rec.e_cpos = cpu_to_le32(cpos);
3086         replace_rec.e_leaf_clusters = cpu_to_le16(len);
3087         replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
3088                                                                    p_cluster));
3089         replace_rec.e_flags = ext_flags;
3090         replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
3091
3092         path = ocfs2_new_path_from_et(et);
3093         if (!path) {
3094                 ret = -ENOMEM;
3095                 mlog_errno(ret);
3096                 goto out;
3097         }
3098
3099         ret = ocfs2_find_path(et->et_ci, path, cpos);
3100         if (ret) {
3101                 mlog_errno(ret);
3102                 goto out;
3103         }
3104
3105         el = path_leaf_el(path);
3106
3107         index = ocfs2_search_extent_list(el, cpos);
3108         if (index == -1) {
3109                 ocfs2_error(sb,
3110                             "Inode %llu has an extent at cpos %u which can no "
3111                             "longer be found.\n",
3112                             (unsigned long long)ino, cpos);
3113                 ret = -EROFS;
3114                 goto out;
3115         }
3116
3117         ret = ocfs2_split_extent(handle, et, path, index,
3118                                  &replace_rec, meta_ac, dealloc);
3119         if (ret)
3120                 mlog_errno(ret);
3121
3122 out:
3123         ocfs2_free_path(path);
3124         return ret;
3125 }
3126
3127 static int ocfs2_replace_clusters(handle_t *handle,
3128                                   struct ocfs2_cow_context *context,
3129                                   u32 cpos, u32 old,
3130                                   u32 new, u32 len,
3131                                   unsigned int ext_flags)
3132 {
3133         int ret;
3134         struct ocfs2_caching_info *ci = context->data_et.et_ci;
3135         u64 ino = ocfs2_metadata_cache_owner(ci);
3136
3137         trace_ocfs2_replace_clusters((unsigned long long)ino,
3138                                      cpos, old, new, len, ext_flags);
3139
3140         /*If the old clusters is unwritten, no need to duplicate. */
3141         if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
3142                 ret = context->cow_duplicate_clusters(handle, context->inode,
3143                                                       cpos, old, new, len);
3144                 if (ret) {
3145                         mlog_errno(ret);
3146                         goto out;
3147                 }
3148         }
3149
3150         ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
3151                                        cpos, new, len, ext_flags,
3152                                        context->meta_ac, &context->dealloc);
3153         if (ret)
3154                 mlog_errno(ret);
3155 out:
3156         return ret;
3157 }
3158
3159 int ocfs2_cow_sync_writeback(struct super_block *sb,
3160                              struct inode *inode,
3161                              u32 cpos, u32 num_clusters)
3162 {
3163         int ret = 0;
3164         loff_t offset, end, map_end;
3165         pgoff_t page_index;
3166         struct page *page;
3167
3168         if (ocfs2_should_order_data(inode))
3169                 return 0;
3170
3171         offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
3172         end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
3173
3174         ret = filemap_fdatawrite_range(inode->i_mapping,
3175                                        offset, end - 1);
3176         if (ret < 0) {
3177                 mlog_errno(ret);
3178                 return ret;
3179         }
3180
3181         while (offset < end) {
3182                 page_index = offset >> PAGE_CACHE_SHIFT;
3183                 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
3184                 if (map_end > end)
3185                         map_end = end;
3186
3187                 page = find_or_create_page(inode->i_mapping,
3188                                            page_index, GFP_NOFS);
3189                 BUG_ON(!page);
3190
3191                 wait_on_page_writeback(page);
3192                 if (PageError(page)) {
3193                         ret = -EIO;
3194                         mlog_errno(ret);
3195                 } else
3196                         mark_page_accessed(page);
3197
3198                 unlock_page(page);
3199                 page_cache_release(page);
3200                 page = NULL;
3201                 offset = map_end;
3202                 if (ret)
3203                         break;
3204         }
3205
3206         return ret;
3207 }
3208
3209 static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
3210                                  u32 v_cluster, u32 *p_cluster,
3211                                  u32 *num_clusters,
3212                                  unsigned int *extent_flags)
3213 {
3214         return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
3215                                   num_clusters, extent_flags);
3216 }
3217
3218 static int ocfs2_make_clusters_writable(struct super_block *sb,
3219                                         struct ocfs2_cow_context *context,
3220                                         u32 cpos, u32 p_cluster,
3221                                         u32 num_clusters, unsigned int e_flags)
3222 {
3223         int ret, delete, index, credits =  0;
3224         u32 new_bit, new_len, orig_num_clusters;
3225         unsigned int set_len;
3226         struct ocfs2_super *osb = OCFS2_SB(sb);
3227         handle_t *handle;
3228         struct buffer_head *ref_leaf_bh = NULL;
3229         struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
3230         struct ocfs2_refcount_rec rec;
3231
3232         trace_ocfs2_make_clusters_writable(cpos, p_cluster,
3233                                            num_clusters, e_flags);
3234
3235         ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
3236                                              &context->data_et,
3237                                              ref_ci,
3238                                              context->ref_root_bh,
3239                                              &context->meta_ac,
3240                                              &context->data_ac, &credits);
3241         if (ret) {
3242                 mlog_errno(ret);
3243                 return ret;
3244         }
3245
3246         if (context->post_refcount)
3247                 credits += context->post_refcount->credits;
3248
3249         credits += context->extra_credits;
3250         handle = ocfs2_start_trans(osb, credits);
3251         if (IS_ERR(handle)) {
3252                 ret = PTR_ERR(handle);
3253                 mlog_errno(ret);
3254                 goto out;
3255         }
3256
3257         orig_num_clusters = num_clusters;
3258
3259         while (num_clusters) {
3260                 ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
3261                                              p_cluster, num_clusters,
3262                                              &rec, &index, &ref_leaf_bh);
3263                 if (ret) {
3264                         mlog_errno(ret);
3265                         goto out_commit;
3266                 }
3267
3268                 BUG_ON(!rec.r_refcount);
3269                 set_len = min((u64)p_cluster + num_clusters,
3270                               le64_to_cpu(rec.r_cpos) +
3271                               le32_to_cpu(rec.r_clusters)) - p_cluster;
3272
3273                 /*
3274                  * There are many different situation here.
3275                  * 1. If refcount == 1, remove the flag and don't COW.
3276                  * 2. If refcount > 1, allocate clusters.
3277                  *    Here we may not allocate r_len once at a time, so continue
3278                  *    until we reach num_clusters.
3279                  */
3280                 if (le32_to_cpu(rec.r_refcount) == 1) {
3281                         delete = 0;
3282                         ret = ocfs2_clear_ext_refcount(handle,
3283                                                        &context->data_et,
3284                                                        cpos, p_cluster,
3285                                                        set_len, e_flags,
3286                                                        context->meta_ac,
3287                                                        &context->dealloc);
3288                         if (ret) {
3289                                 mlog_errno(ret);
3290                                 goto out_commit;
3291                         }
3292                 } else {
3293                         delete = 1;
3294
3295                         ret = __ocfs2_claim_clusters(handle,
3296                                                      context->data_ac,
3297                                                      1, set_len,
3298                                                      &new_bit, &new_len);
3299                         if (ret) {
3300                                 mlog_errno(ret);
3301                                 goto out_commit;
3302                         }
3303
3304                         ret = ocfs2_replace_clusters(handle, context,
3305                                                      cpos, p_cluster, new_bit,
3306                                                      new_len, e_flags);
3307                         if (ret) {
3308                                 mlog_errno(ret);
3309                                 goto out_commit;
3310                         }
3311                         set_len = new_len;
3312                 }
3313
3314                 ret = __ocfs2_decrease_refcount(handle, ref_ci,
3315                                                 context->ref_root_bh,
3316                                                 p_cluster, set_len,
3317                                                 context->meta_ac,
3318                                                 &context->dealloc, delete);
3319                 if (ret) {
3320                         mlog_errno(ret);
3321                         goto out_commit;
3322                 }
3323
3324                 cpos += set_len;
3325                 p_cluster += set_len;
3326                 num_clusters -= set_len;
3327                 brelse(ref_leaf_bh);
3328                 ref_leaf_bh = NULL;
3329         }
3330
3331         /* handle any post_cow action. */
3332         if (context->post_refcount && context->post_refcount->func) {
3333                 ret = context->post_refcount->func(context->inode, handle,
3334                                                 context->post_refcount->para);
3335                 if (ret) {
3336                         mlog_errno(ret);
3337                         goto out_commit;
3338                 }
3339         }
3340
3341         /*
3342          * Here we should write the new page out first if we are
3343          * in write-back mode.
3344          */
3345         if (context->get_clusters == ocfs2_di_get_clusters) {
3346                 ret = ocfs2_cow_sync_writeback(sb, context->inode, cpos,
3347                                                orig_num_clusters);
3348                 if (ret)
3349                         mlog_errno(ret);
3350         }
3351
3352 out_commit:
3353         ocfs2_commit_trans(osb, handle);
3354
3355 out:
3356         if (context->data_ac) {
3357                 ocfs2_free_alloc_context(context->data_ac);
3358                 context->data_ac = NULL;
3359         }
3360         if (context->meta_ac) {
3361                 ocfs2_free_alloc_context(context->meta_ac);
3362                 context->meta_ac = NULL;
3363         }
3364         brelse(ref_leaf_bh);
3365
3366         return ret;
3367 }
3368
3369 static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
3370 {
3371         int ret = 0;
3372         struct inode *inode = context->inode;
3373         u32 cow_start = context->cow_start, cow_len = context->cow_len;
3374         u32 p_cluster, num_clusters;
3375         unsigned int ext_flags;
3376         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3377
3378         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
3379                 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
3380                             "tree, but the feature bit is not set in the "
3381                             "super block.", inode->i_ino);
3382                 return -EROFS;
3383         }
3384
3385         ocfs2_init_dealloc_ctxt(&context->dealloc);
3386
3387         while (cow_len) {
3388                 ret = context->get_clusters(context, cow_start, &p_cluster,
3389                                             &num_clusters, &ext_flags);
3390                 if (ret) {
3391                         mlog_errno(ret);
3392                         break;
3393                 }
3394
3395                 BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
3396
3397                 if (cow_len < num_clusters)
3398                         num_clusters = cow_len;
3399
3400                 ret = ocfs2_make_clusters_writable(inode->i_sb, context,
3401                                                    cow_start, p_cluster,
3402                                                    num_clusters, ext_flags);
3403                 if (ret) {
3404                         mlog_errno(ret);
3405                         break;
3406                 }
3407
3408                 cow_len -= num_clusters;
3409                 cow_start += num_clusters;
3410         }
3411
3412         if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
3413                 ocfs2_schedule_truncate_log_flush(osb, 1);
3414                 ocfs2_run_deallocs(osb, &context->dealloc);
3415         }
3416
3417         return ret;
3418 }
3419
3420 /*
3421  * Starting at cpos, try to CoW write_len clusters.  Don't CoW
3422  * past max_cpos.  This will stop when it runs into a hole or an
3423  * unrefcounted extent.
3424  */
3425 static int ocfs2_refcount_cow_hunk(struct inode *inode,
3426                                    struct buffer_head *di_bh,
3427                                    u32 cpos, u32 write_len, u32 max_cpos)
3428 {
3429         int ret;
3430         u32 cow_start = 0, cow_len = 0;
3431         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3432         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3433         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3434         struct buffer_head *ref_root_bh = NULL;
3435         struct ocfs2_refcount_tree *ref_tree;
3436         struct ocfs2_cow_context *context = NULL;
3437
3438         BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3439
3440         ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
3441                                               cpos, write_len, max_cpos,
3442                                               &cow_start, &cow_len);
3443         if (ret) {
3444                 mlog_errno(ret);
3445                 goto out;
3446         }
3447
3448         trace_ocfs2_refcount_cow_hunk(OCFS2_I(inode)->ip_blkno,
3449                                       cpos, write_len, max_cpos,
3450                                       cow_start, cow_len);
3451
3452         BUG_ON(cow_len == 0);
3453
3454         context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3455         if (!context) {
3456                 ret = -ENOMEM;
3457                 mlog_errno(ret);
3458                 goto out;
3459         }
3460
3461         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
3462                                        1, &ref_tree, &ref_root_bh);
3463         if (ret) {
3464                 mlog_errno(ret);
3465                 goto out;
3466         }
3467
3468         context->inode = inode;
3469         context->cow_start = cow_start;
3470         context->cow_len = cow_len;
3471         context->ref_tree = ref_tree;
3472         context->ref_root_bh = ref_root_bh;
3473         context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
3474         context->get_clusters = ocfs2_di_get_clusters;
3475
3476         ocfs2_init_dinode_extent_tree(&context->data_et,
3477                                       INODE_CACHE(inode), di_bh);
3478
3479         ret = ocfs2_replace_cow(context);
3480         if (ret)
3481                 mlog_errno(ret);
3482
3483         /*
3484          * truncate the extent map here since no matter whether we meet with
3485          * any error during the action, we shouldn't trust cached extent map
3486          * any more.
3487          */
3488         ocfs2_extent_map_trunc(inode, cow_start);
3489
3490         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3491         brelse(ref_root_bh);
3492 out:
3493         kfree(context);
3494         return ret;
3495 }
3496
3497 /*
3498  * CoW any and all clusters between cpos and cpos+write_len.
3499  * Don't CoW past max_cpos.  If this returns successfully, all
3500  * clusters between cpos and cpos+write_len are safe to modify.
3501  */
3502 int ocfs2_refcount_cow(struct inode *inode,
3503                        struct buffer_head *di_bh,
3504                        u32 cpos, u32 write_len, u32 max_cpos)
3505 {
3506         int ret = 0;
3507         u32 p_cluster, num_clusters;
3508         unsigned int ext_flags;
3509
3510         while (write_len) {
3511                 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3512                                          &num_clusters, &ext_flags);
3513                 if (ret) {
3514                         mlog_errno(ret);
3515                         break;
3516                 }
3517
3518                 if (write_len < num_clusters)
3519                         num_clusters = write_len;
3520
3521                 if (ext_flags & OCFS2_EXT_REFCOUNTED) {
3522                         ret = ocfs2_refcount_cow_hunk(inode, di_bh, cpos,
3523                                                       num_clusters, max_cpos);
3524                         if (ret) {
3525                                 mlog_errno(ret);
3526                                 break;
3527                         }
3528                 }
3529
3530                 write_len -= num_clusters;
3531                 cpos += num_clusters;
3532         }
3533
3534         return ret;
3535 }
3536
3537 static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
3538                                           u32 v_cluster, u32 *p_cluster,
3539                                           u32 *num_clusters,
3540                                           unsigned int *extent_flags)
3541 {
3542         struct inode *inode = context->inode;
3543         struct ocfs2_xattr_value_root *xv = context->cow_object;
3544
3545         return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
3546                                         num_clusters, &xv->xr_list,
3547                                         extent_flags);
3548 }
3549
3550 /*
3551  * Given a xattr value root, calculate the most meta/credits we need for
3552  * refcount tree change if we truncate it to 0.
3553  */
3554 int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
3555                                        struct ocfs2_caching_info *ref_ci,
3556                                        struct buffer_head *ref_root_bh,
3557                                        struct ocfs2_xattr_value_root *xv,
3558                                        int *meta_add, int *credits)
3559 {
3560         int ret = 0, index, ref_blocks = 0;
3561         u32 p_cluster, num_clusters;
3562         u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
3563         struct ocfs2_refcount_block *rb;
3564         struct ocfs2_refcount_rec rec;
3565         struct buffer_head *ref_leaf_bh = NULL;
3566
3567         while (cpos < clusters) {
3568                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
3569                                                &num_clusters, &xv->xr_list,
3570                                                NULL);
3571                 if (ret) {
3572                         mlog_errno(ret);
3573                         goto out;
3574                 }
3575
3576                 cpos += num_clusters;
3577
3578                 while (num_clusters) {
3579                         ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
3580                                                      p_cluster, num_clusters,
3581                                                      &rec, &index,
3582                                                      &ref_leaf_bh);
3583                         if (ret) {
3584                                 mlog_errno(ret);
3585                                 goto out;
3586                         }
3587
3588                         BUG_ON(!rec.r_refcount);
3589
3590                         rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
3591
3592                         /*
3593                          * We really don't know whether the other clusters is in
3594                          * this refcount block or not, so just take the worst
3595                          * case that all the clusters are in this block and each
3596                          * one will split a refcount rec, so totally we need
3597                          * clusters * 2 new refcount rec.
3598                          */
3599                         if (le16_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
3600                             le16_to_cpu(rb->rf_records.rl_count))
3601                                 ref_blocks++;
3602
3603                         *credits += 1;
3604                         brelse(ref_leaf_bh);
3605                         ref_leaf_bh = NULL;
3606
3607                         if (num_clusters <= le32_to_cpu(rec.r_clusters))
3608                                 break;
3609                         else
3610                                 num_clusters -= le32_to_cpu(rec.r_clusters);
3611                         p_cluster += num_clusters;
3612                 }
3613         }
3614
3615         *meta_add += ref_blocks;
3616         if (!ref_blocks)
3617                 goto out;
3618
3619         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
3620         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
3621                 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
3622         else {
3623                 struct ocfs2_extent_tree et;
3624
3625                 ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
3626                 *credits += ocfs2_calc_extend_credits(inode->i_sb,
3627                                                       et.et_root_el);
3628         }
3629
3630 out:
3631         brelse(ref_leaf_bh);
3632         return ret;
3633 }
3634
3635 /*
3636  * Do CoW for xattr.
3637  */
3638 int ocfs2_refcount_cow_xattr(struct inode *inode,
3639                              struct ocfs2_dinode *di,
3640                              struct ocfs2_xattr_value_buf *vb,
3641                              struct ocfs2_refcount_tree *ref_tree,
3642                              struct buffer_head *ref_root_bh,
3643                              u32 cpos, u32 write_len,
3644                              struct ocfs2_post_refcount *post)
3645 {
3646         int ret;
3647         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
3648         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3649         struct ocfs2_cow_context *context = NULL;
3650         u32 cow_start, cow_len;
3651
3652         BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3653
3654         ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
3655                                               cpos, write_len, UINT_MAX,
3656                                               &cow_start, &cow_len);
3657         if (ret) {
3658                 mlog_errno(ret);
3659                 goto out;
3660         }
3661
3662         BUG_ON(cow_len == 0);
3663
3664         context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3665         if (!context) {
3666                 ret = -ENOMEM;
3667                 mlog_errno(ret);
3668                 goto out;
3669         }
3670
3671         context->inode = inode;
3672         context->cow_start = cow_start;
3673         context->cow_len = cow_len;
3674         context->ref_tree = ref_tree;
3675         context->ref_root_bh = ref_root_bh;
3676         context->cow_object = xv;
3677
3678         context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
3679         /* We need the extra credits for duplicate_clusters by jbd. */
3680         context->extra_credits =
3681                 ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
3682         context->get_clusters = ocfs2_xattr_value_get_clusters;
3683         context->post_refcount = post;
3684
3685         ocfs2_init_xattr_value_extent_tree(&context->data_et,
3686                                            INODE_CACHE(inode), vb);
3687
3688         ret = ocfs2_replace_cow(context);
3689         if (ret)
3690                 mlog_errno(ret);
3691
3692 out:
3693         kfree(context);
3694         return ret;
3695 }
3696
3697 /*
3698  * Insert a new extent into refcount tree and mark a extent rec
3699  * as refcounted in the dinode tree.
3700  */
3701 int ocfs2_add_refcount_flag(struct inode *inode,
3702                             struct ocfs2_extent_tree *data_et,
3703                             struct ocfs2_caching_info *ref_ci,
3704                             struct buffer_head *ref_root_bh,
3705                             u32 cpos, u32 p_cluster, u32 num_clusters,
3706                             struct ocfs2_cached_dealloc_ctxt *dealloc,
3707                             struct ocfs2_post_refcount *post)
3708 {
3709         int ret;
3710         handle_t *handle;
3711         int credits = 1, ref_blocks = 0;
3712         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3713         struct ocfs2_alloc_context *meta_ac = NULL;
3714
3715         ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
3716                                                ref_ci, ref_root_bh,
3717                                                p_cluster, num_clusters,
3718                                                &ref_blocks, &credits);
3719         if (ret) {
3720                 mlog_errno(ret);
3721                 goto out;
3722         }
3723
3724         trace_ocfs2_add_refcount_flag(ref_blocks, credits);
3725
3726         if (ref_blocks) {
3727                 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
3728                                                         ref_blocks, &meta_ac);
3729                 if (ret) {
3730                         mlog_errno(ret);
3731                         goto out;
3732                 }
3733         }
3734
3735         if (post)
3736                 credits += post->credits;
3737
3738         handle = ocfs2_start_trans(osb, credits);
3739         if (IS_ERR(handle)) {
3740                 ret = PTR_ERR(handle);
3741                 mlog_errno(ret);
3742                 goto out;
3743         }
3744
3745         ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
3746                                            cpos, num_clusters, p_cluster,
3747                                            meta_ac, dealloc);
3748         if (ret) {
3749                 mlog_errno(ret);
3750                 goto out_commit;
3751         }
3752
3753         ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3754                                         p_cluster, num_clusters, 0,
3755                                         meta_ac, dealloc);
3756         if (ret) {
3757                 mlog_errno(ret);
3758                 goto out_commit;
3759         }
3760
3761         if (post && post->func) {
3762                 ret = post->func(inode, handle, post->para);
3763                 if (ret)
3764                         mlog_errno(ret);
3765         }
3766
3767 out_commit:
3768         ocfs2_commit_trans(osb, handle);
3769 out:
3770         if (meta_ac)
3771                 ocfs2_free_alloc_context(meta_ac);
3772         return ret;
3773 }
3774
3775 static int ocfs2_change_ctime(struct inode *inode,
3776                               struct buffer_head *di_bh)
3777 {
3778         int ret;
3779         handle_t *handle;
3780         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3781
3782         handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
3783                                    OCFS2_INODE_UPDATE_CREDITS);
3784         if (IS_ERR(handle)) {
3785                 ret = PTR_ERR(handle);
3786                 mlog_errno(ret);
3787                 goto out;
3788         }
3789
3790         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
3791                                       OCFS2_JOURNAL_ACCESS_WRITE);
3792         if (ret) {
3793                 mlog_errno(ret);
3794                 goto out_commit;
3795         }
3796
3797         inode->i_ctime = CURRENT_TIME;
3798         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3799         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3800
3801         ocfs2_journal_dirty(handle, di_bh);
3802
3803 out_commit:
3804         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3805 out:
3806         return ret;
3807 }
3808
3809 static int ocfs2_attach_refcount_tree(struct inode *inode,
3810                                       struct buffer_head *di_bh)
3811 {
3812         int ret, data_changed = 0;
3813         struct buffer_head *ref_root_bh = NULL;
3814         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3815         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3816         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3817         struct ocfs2_refcount_tree *ref_tree;
3818         unsigned int ext_flags;
3819         loff_t size;
3820         u32 cpos, num_clusters, clusters, p_cluster;
3821         struct ocfs2_cached_dealloc_ctxt dealloc;
3822         struct ocfs2_extent_tree di_et;
3823
3824         ocfs2_init_dealloc_ctxt(&dealloc);
3825
3826         if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) {
3827                 ret = ocfs2_create_refcount_tree(inode, di_bh);
3828                 if (ret) {
3829                         mlog_errno(ret);
3830                         goto out;
3831                 }
3832         }
3833
3834         BUG_ON(!di->i_refcount_loc);
3835         ret = ocfs2_lock_refcount_tree(osb,
3836                                        le64_to_cpu(di->i_refcount_loc), 1,
3837                                        &ref_tree, &ref_root_bh);
3838         if (ret) {
3839                 mlog_errno(ret);
3840                 goto out;
3841         }
3842
3843         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
3844                 goto attach_xattr;
3845
3846         ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
3847
3848         size = i_size_read(inode);
3849         clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
3850
3851         cpos = 0;
3852         while (cpos < clusters) {
3853                 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3854                                          &num_clusters, &ext_flags);
3855                 if (ret) {
3856                         mlog_errno(ret);
3857                         goto unlock;
3858                 }
3859                 if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
3860                         ret = ocfs2_add_refcount_flag(inode, &di_et,
3861                                                       &ref_tree->rf_ci,
3862                                                       ref_root_bh, cpos,
3863                                                       p_cluster, num_clusters,
3864                                                       &dealloc, NULL);
3865                         if (ret) {
3866                                 mlog_errno(ret);
3867                                 goto unlock;
3868                         }
3869
3870                         data_changed = 1;
3871                 }
3872                 cpos += num_clusters;
3873         }
3874
3875 attach_xattr:
3876         if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
3877                 ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
3878                                                        &ref_tree->rf_ci,
3879                                                        ref_root_bh,
3880                                                        &dealloc);
3881                 if (ret) {
3882                         mlog_errno(ret);
3883                         goto unlock;
3884                 }
3885         }
3886
3887         if (data_changed) {
3888                 ret = ocfs2_change_ctime(inode, di_bh);
3889                 if (ret)
3890                         mlog_errno(ret);
3891         }
3892
3893 unlock:
3894         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3895         brelse(ref_root_bh);
3896
3897         if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
3898                 ocfs2_schedule_truncate_log_flush(osb, 1);
3899                 ocfs2_run_deallocs(osb, &dealloc);
3900         }
3901 out:
3902         /*
3903          * Empty the extent map so that we may get the right extent
3904          * record from the disk.
3905          */
3906         ocfs2_extent_map_trunc(inode, 0);
3907
3908         return ret;
3909 }
3910
3911 static int ocfs2_add_refcounted_extent(struct inode *inode,
3912                                    struct ocfs2_extent_tree *et,
3913                                    struct ocfs2_caching_info *ref_ci,
3914                                    struct buffer_head *ref_root_bh,
3915                                    u32 cpos, u32 p_cluster, u32 num_clusters,
3916                                    unsigned int ext_flags,
3917                                    struct ocfs2_cached_dealloc_ctxt *dealloc)
3918 {
3919         int ret;
3920         handle_t *handle;
3921         int credits = 0;
3922         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3923         struct ocfs2_alloc_context *meta_ac = NULL;
3924
3925         ret = ocfs2_lock_refcount_allocators(inode->i_sb,
3926                                              p_cluster, num_clusters,
3927                                              et, ref_ci,
3928                                              ref_root_bh, &meta_ac,
3929                                              NULL, &credits);
3930         if (ret) {
3931                 mlog_errno(ret);
3932                 goto out;
3933         }
3934
3935         handle = ocfs2_start_trans(osb, credits);
3936         if (IS_ERR(handle)) {
3937                 ret = PTR_ERR(handle);
3938                 mlog_errno(ret);
3939                 goto out;
3940         }
3941
3942         ret = ocfs2_insert_extent(handle, et, cpos,
3943                         ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
3944                         num_clusters, ext_flags, meta_ac);
3945         if (ret) {
3946                 mlog_errno(ret);
3947                 goto out_commit;
3948         }
3949
3950         ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3951                                       p_cluster, num_clusters,
3952                                       meta_ac, dealloc);
3953         if (ret)
3954                 mlog_errno(ret);
3955
3956 out_commit:
3957         ocfs2_commit_trans(osb, handle);
3958 out:
3959         if (meta_ac)
3960                 ocfs2_free_alloc_context(meta_ac);
3961         return ret;
3962 }
3963
3964 static int ocfs2_duplicate_inline_data(struct inode *s_inode,
3965                                        struct buffer_head *s_bh,
3966                                        struct inode *t_inode,
3967                                        struct buffer_head *t_bh)
3968 {
3969         int ret;
3970         handle_t *handle;
3971         struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
3972         struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
3973         struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
3974
3975         BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
3976
3977         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
3978         if (IS_ERR(handle)) {
3979                 ret = PTR_ERR(handle);
3980                 mlog_errno(ret);
3981                 goto out;
3982         }
3983
3984         ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
3985                                       OCFS2_JOURNAL_ACCESS_WRITE);
3986         if (ret) {
3987                 mlog_errno(ret);
3988                 goto out_commit;
3989         }
3990
3991         t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
3992         memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
3993                le16_to_cpu(s_di->id2.i_data.id_count));
3994         spin_lock(&OCFS2_I(t_inode)->ip_lock);
3995         OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
3996         t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
3997         spin_unlock(&OCFS2_I(t_inode)->ip_lock);
3998
3999         ocfs2_journal_dirty(handle, t_bh);
4000
4001 out_commit:
4002         ocfs2_commit_trans(osb, handle);
4003 out:
4004         return ret;
4005 }
4006
4007 static int ocfs2_duplicate_extent_list(struct inode *s_inode,
4008                                 struct inode *t_inode,
4009                                 struct buffer_head *t_bh,
4010                                 struct ocfs2_caching_info *ref_ci,
4011                                 struct buffer_head *ref_root_bh,
4012                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
4013 {
4014         int ret = 0;
4015         u32 p_cluster, num_clusters, clusters, cpos;
4016         loff_t size;
4017         unsigned int ext_flags;
4018         struct ocfs2_extent_tree et;
4019
4020         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
4021
4022         size = i_size_read(s_inode);
4023         clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
4024
4025         cpos = 0;
4026         while (cpos < clusters) {
4027                 ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
4028                                          &num_clusters, &ext_flags);
4029                 if (ret) {
4030                         mlog_errno(ret);
4031                         goto out;
4032                 }
4033                 if (p_cluster) {
4034                         ret = ocfs2_add_refcounted_extent(t_inode, &et,
4035                                                           ref_ci, ref_root_bh,
4036                                                           cpos, p_cluster,
4037                                                           num_clusters,
4038                                                           ext_flags,
4039                                                           dealloc);
4040                         if (ret) {
4041                                 mlog_errno(ret);
4042                                 goto out;
4043                         }
4044                 }
4045
4046                 cpos += num_clusters;
4047         }
4048
4049 out:
4050         return ret;
4051 }
4052
4053 /*
4054  * change the new file's attributes to the src.
4055  *
4056  * reflink creates a snapshot of a file, that means the attributes
4057  * must be identical except for three exceptions - nlink, ino, and ctime.
4058  */
4059 static int ocfs2_complete_reflink(struct inode *s_inode,
4060                                   struct buffer_head *s_bh,
4061                                   struct inode *t_inode,
4062                                   struct buffer_head *t_bh,
4063                                   bool preserve)
4064 {
4065         int ret;
4066         handle_t *handle;
4067         struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4068         struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
4069         loff_t size = i_size_read(s_inode);
4070
4071         handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
4072                                    OCFS2_INODE_UPDATE_CREDITS);
4073         if (IS_ERR(handle)) {
4074                 ret = PTR_ERR(handle);
4075                 mlog_errno(ret);
4076                 return ret;
4077         }
4078
4079         ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4080                                       OCFS2_JOURNAL_ACCESS_WRITE);
4081         if (ret) {
4082                 mlog_errno(ret);
4083                 goto out_commit;
4084         }
4085
4086         spin_lock(&OCFS2_I(t_inode)->ip_lock);
4087         OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
4088         OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
4089         OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
4090         spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4091         i_size_write(t_inode, size);
4092         t_inode->i_blocks = s_inode->i_blocks;
4093
4094         di->i_xattr_inline_size = s_di->i_xattr_inline_size;
4095         di->i_clusters = s_di->i_clusters;
4096         di->i_size = s_di->i_size;
4097         di->i_dyn_features = s_di->i_dyn_features;
4098         di->i_attr = s_di->i_attr;
4099
4100         if (preserve) {
4101                 t_inode->i_uid = s_inode->i_uid;
4102                 t_inode->i_gid = s_inode->i_gid;
4103                 t_inode->i_mode = s_inode->i_mode;
4104                 di->i_uid = s_di->i_uid;
4105                 di->i_gid = s_di->i_gid;
4106                 di->i_mode = s_di->i_mode;
4107
4108                 /*
4109                  * update time.
4110                  * we want mtime to appear identical to the source and
4111                  * update ctime.
4112                  */
4113                 t_inode->i_ctime = CURRENT_TIME;
4114
4115                 di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
4116                 di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
4117
4118                 t_inode->i_mtime = s_inode->i_mtime;
4119                 di->i_mtime = s_di->i_mtime;
4120                 di->i_mtime_nsec = s_di->i_mtime_nsec;
4121         }
4122
4123         ocfs2_journal_dirty(handle, t_bh);
4124
4125 out_commit:
4126         ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
4127         return ret;
4128 }
4129
4130 static int ocfs2_create_reflink_node(struct inode *s_inode,
4131                                      struct buffer_head *s_bh,
4132                                      struct inode *t_inode,
4133                                      struct buffer_head *t_bh,
4134                                      bool preserve)
4135 {
4136         int ret;
4137         struct buffer_head *ref_root_bh = NULL;
4138         struct ocfs2_cached_dealloc_ctxt dealloc;
4139         struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4140         struct ocfs2_refcount_block *rb;
4141         struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
4142         struct ocfs2_refcount_tree *ref_tree;
4143
4144         ocfs2_init_dealloc_ctxt(&dealloc);
4145
4146         ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4147                                       le64_to_cpu(di->i_refcount_loc));
4148         if (ret) {
4149                 mlog_errno(ret);
4150                 goto out;
4151         }
4152
4153         if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4154                 ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
4155                                                   t_inode, t_bh);
4156                 if (ret)
4157                         mlog_errno(ret);
4158                 goto out;
4159         }
4160
4161         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
4162                                        1, &ref_tree, &ref_root_bh);
4163         if (ret) {
4164                 mlog_errno(ret);
4165                 goto out;
4166         }
4167         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
4168
4169         ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
4170                                           &ref_tree->rf_ci, ref_root_bh,
4171                                           &dealloc);
4172         if (ret) {
4173                 mlog_errno(ret);
4174                 goto out_unlock_refcount;
4175         }
4176
4177 out_unlock_refcount:
4178         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4179         brelse(ref_root_bh);
4180 out:
4181         if (ocfs2_dealloc_has_cluster(&dealloc)) {
4182                 ocfs2_schedule_truncate_log_flush(osb, 1);
4183                 ocfs2_run_deallocs(osb, &dealloc);
4184         }
4185
4186         return ret;
4187 }
4188
4189 static int __ocfs2_reflink(struct dentry *old_dentry,
4190                            struct buffer_head *old_bh,
4191                            struct inode *new_inode,
4192                            bool preserve)
4193 {
4194         int ret;
4195         struct inode *inode = d_inode(old_dentry);
4196         struct buffer_head *new_bh = NULL;
4197
4198         if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
4199                 ret = -EINVAL;
4200                 mlog_errno(ret);
4201                 goto out;
4202         }
4203
4204         ret = filemap_fdatawrite(inode->i_mapping);
4205         if (ret) {
4206                 mlog_errno(ret);
4207                 goto out;
4208         }
4209
4210         ret = ocfs2_attach_refcount_tree(inode, old_bh);
4211         if (ret) {
4212                 mlog_errno(ret);
4213                 goto out;
4214         }
4215
4216         mutex_lock_nested(&new_inode->i_mutex, I_MUTEX_CHILD);
4217         ret = ocfs2_inode_lock_nested(new_inode, &new_bh, 1,
4218                                       OI_LS_REFLINK_TARGET);
4219         if (ret) {
4220                 mlog_errno(ret);
4221                 goto out_unlock;
4222         }
4223
4224         ret = ocfs2_create_reflink_node(inode, old_bh,
4225                                         new_inode, new_bh, preserve);
4226         if (ret) {
4227                 mlog_errno(ret);
4228                 goto inode_unlock;
4229         }
4230
4231         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
4232                 ret = ocfs2_reflink_xattrs(inode, old_bh,
4233                                            new_inode, new_bh,
4234                                            preserve);
4235                 if (ret) {
4236                         mlog_errno(ret);
4237                         goto inode_unlock;
4238                 }
4239         }
4240
4241         ret = ocfs2_complete_reflink(inode, old_bh,
4242                                      new_inode, new_bh, preserve);
4243         if (ret)
4244                 mlog_errno(ret);
4245
4246 inode_unlock:
4247         ocfs2_inode_unlock(new_inode, 1);
4248         brelse(new_bh);
4249 out_unlock:
4250         mutex_unlock(&new_inode->i_mutex);
4251 out:
4252         if (!ret) {
4253                 ret = filemap_fdatawait(inode->i_mapping);
4254                 if (ret)
4255                         mlog_errno(ret);
4256         }
4257         return ret;
4258 }
4259
4260 static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
4261                          struct dentry *new_dentry, bool preserve)
4262 {
4263         int error;
4264         struct inode *inode = d_inode(old_dentry);
4265         struct buffer_head *old_bh = NULL;
4266         struct inode *new_orphan_inode = NULL;
4267         struct posix_acl *default_acl, *acl;
4268         umode_t mode;
4269
4270         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4271                 return -EOPNOTSUPP;
4272
4273         mode = inode->i_mode;
4274         error = posix_acl_create(dir, &mode, &default_acl, &acl);
4275         if (error) {
4276                 mlog_errno(error);
4277                 return error;
4278         }
4279
4280         error = ocfs2_create_inode_in_orphan(dir, mode,
4281                                              &new_orphan_inode);
4282         if (error) {
4283                 mlog_errno(error);
4284                 goto out;
4285         }
4286
4287         error = ocfs2_rw_lock(inode, 1);
4288         if (error) {
4289                 mlog_errno(error);
4290                 goto out;
4291         }
4292
4293         error = ocfs2_inode_lock(inode, &old_bh, 1);
4294         if (error) {
4295                 mlog_errno(error);
4296                 ocfs2_rw_unlock(inode, 1);
4297                 goto out;
4298         }
4299
4300         down_write(&OCFS2_I(inode)->ip_xattr_sem);
4301         down_write(&OCFS2_I(inode)->ip_alloc_sem);
4302         error = __ocfs2_reflink(old_dentry, old_bh,
4303                                 new_orphan_inode, preserve);
4304         up_write(&OCFS2_I(inode)->ip_alloc_sem);
4305         up_write(&OCFS2_I(inode)->ip_xattr_sem);
4306
4307         ocfs2_inode_unlock(inode, 1);
4308         ocfs2_rw_unlock(inode, 1);
4309         brelse(old_bh);
4310
4311         if (error) {
4312                 mlog_errno(error);
4313                 goto out;
4314         }
4315
4316         /* If the security isn't preserved, we need to re-initialize them. */
4317         if (!preserve) {
4318                 error = ocfs2_init_security_and_acl(dir, new_orphan_inode,
4319                                                     &new_dentry->d_name,
4320                                                     default_acl, acl);
4321                 if (error)
4322                         mlog_errno(error);
4323         }
4324 out:
4325         if (default_acl)
4326                 posix_acl_release(default_acl);
4327         if (acl)
4328                 posix_acl_release(acl);
4329         if (!error) {
4330                 error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
4331                                                        new_dentry);
4332                 if (error)
4333                         mlog_errno(error);
4334         }
4335
4336         if (new_orphan_inode) {
4337                 /*
4338                  * We need to open_unlock the inode no matter whether we
4339                  * succeed or not, so that other nodes can delete it later.
4340                  */
4341                 ocfs2_open_unlock(new_orphan_inode);
4342                 if (error)
4343                         iput(new_orphan_inode);
4344         }
4345
4346         return error;
4347 }
4348
4349 /*
4350  * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
4351  * sys_reflink().  This will go away when vfs_reflink() exists in
4352  * fs/namei.c.
4353  */
4354
4355 /* copied from may_create in VFS. */
4356 static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
4357 {
4358         if (d_really_is_positive(child))
4359                 return -EEXIST;
4360         if (IS_DEADDIR(dir))
4361                 return -ENOENT;
4362         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
4363 }
4364
4365 /**
4366  * ocfs2_vfs_reflink - Create a reference-counted link
4367  *
4368  * @old_dentry:        source dentry + inode
4369  * @dir:       directory to create the target
4370  * @new_dentry:        target dentry
4371  * @preserve:  if true, preserve all file attributes
4372  */
4373 static int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
4374                              struct dentry *new_dentry, bool preserve)
4375 {
4376         struct inode *inode = d_inode(old_dentry);
4377         int error;
4378
4379         if (!inode)
4380                 return -ENOENT;
4381
4382         error = ocfs2_may_create(dir, new_dentry);
4383         if (error)
4384                 return error;
4385
4386         if (dir->i_sb != inode->i_sb)
4387                 return -EXDEV;
4388
4389         /*
4390          * A reflink to an append-only or immutable file cannot be created.
4391          */
4392         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4393                 return -EPERM;
4394
4395         /* Only regular files can be reflinked. */
4396         if (!S_ISREG(inode->i_mode))
4397                 return -EPERM;
4398
4399         /*
4400          * If the caller wants to preserve ownership, they require the
4401          * rights to do so.
4402          */
4403         if (preserve) {
4404                 if (!uid_eq(current_fsuid(), inode->i_uid) && !capable(CAP_CHOWN))
4405                         return -EPERM;
4406                 if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
4407                         return -EPERM;
4408         }
4409
4410         /*
4411          * If the caller is modifying any aspect of the attributes, they
4412          * are not creating a snapshot.  They need read permission on the
4413          * file.
4414          */
4415         if (!preserve) {
4416                 error = inode_permission(inode, MAY_READ);
4417                 if (error)
4418                         return error;
4419         }
4420
4421         mutex_lock(&inode->i_mutex);
4422         dquot_initialize(dir);
4423         error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
4424         mutex_unlock(&inode->i_mutex);
4425         if (!error)
4426                 fsnotify_create(dir, new_dentry);
4427         return error;
4428 }
4429 /*
4430  * Most codes are copied from sys_linkat.
4431  */
4432 int ocfs2_reflink_ioctl(struct inode *inode,
4433                         const char __user *oldname,
4434                         const char __user *newname,
4435                         bool preserve)
4436 {
4437         struct dentry *new_dentry;
4438         struct path old_path, new_path;
4439         int error;
4440
4441         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4442                 return -EOPNOTSUPP;
4443
4444         error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
4445         if (error) {
4446                 mlog_errno(error);
4447                 return error;
4448         }
4449
4450         new_dentry = user_path_create(AT_FDCWD, newname, &new_path, 0);
4451         error = PTR_ERR(new_dentry);
4452         if (IS_ERR(new_dentry)) {
4453                 mlog_errno(error);
4454                 goto out;
4455         }
4456
4457         error = -EXDEV;
4458         if (old_path.mnt != new_path.mnt) {
4459                 mlog_errno(error);
4460                 goto out_dput;
4461         }
4462
4463         error = ocfs2_vfs_reflink(old_path.dentry,
4464                                   d_inode(new_path.dentry),
4465                                   new_dentry, preserve);
4466 out_dput:
4467         done_path_create(&new_path, new_dentry);
4468 out:
4469         path_put(&old_path);
4470
4471         return error;
4472 }