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
  50 struct ocfs2_cow_context {
  51         struct inode *inode;
  52         struct file *file;
  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 file *file,
  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         ocfs2_set_new_buffer_uptodate(&new_tree->rf_ci, new_bh);
 617
 618         ret = ocfs2_journal_access_rb(handle, &new_tree->rf_ci, new_bh,
 619                                       OCFS2_JOURNAL_ACCESS_CREATE);
 620         if (ret) {
 621                 mlog_errno(ret);
 622                 goto out_commit;
 623         }
 624
 625         /* Initialize ocfs2_refcount_block. */
 626         rb = (struct ocfs2_refcount_block *)new_bh->b_data;
 627         memset(rb, 0, inode->i_sb->s_blocksize);
 628         strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
 629         rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
 630         rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
 631         rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
 632         rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
 633         rb->rf_blkno = cpu_to_le64(first_blkno);
 634         rb->rf_count = cpu_to_le32(1);
 635         rb->rf_records.rl_count =
 636                         cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb));
 637         spin_lock(&osb->osb_lock);
 638         rb->rf_generation = osb->s_next_generation++;
 639         spin_unlock(&osb->osb_lock);
 640
 641         ocfs2_journal_dirty(handle, new_bh);
 642
 643         spin_lock(&oi->ip_lock);
 644         oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
 645         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 646         di->i_refcount_loc = cpu_to_le64(first_blkno);
 647         spin_unlock(&oi->ip_lock);
 648
 649         trace_ocfs2_create_refcount_tree_blkno((unsigned long long)first_blkno);
 650
 651         ocfs2_journal_dirty(handle, di_bh);
 652
 653         /*
 654          * We have to init the tree lock here since it will use
 655          * the generation number to create it.
 656          */
 657         new_tree->rf_generation = le32_to_cpu(rb->rf_generation);
 658         ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno,
 659                                       new_tree->rf_generation);
 660
 661         spin_lock(&osb->osb_lock);
 662         tree = ocfs2_find_refcount_tree(osb, first_blkno);
 663
 664         /*
 665          * We've just created a new refcount tree in this block.  If
 666          * we found a refcount tree on the ocfs2_super, it must be
 667          * one we just deleted.  We free the old tree before
 668          * inserting the new tree.
 669          */
 670         BUG_ON(tree && tree->rf_generation == new_tree->rf_generation);
 671         if (tree)
 672                 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
 673         ocfs2_insert_refcount_tree(osb, new_tree);
 674         spin_unlock(&osb->osb_lock);
 675         new_tree = NULL;
 676         if (tree)
 677                 ocfs2_refcount_tree_put(tree);
 678
 679 out_commit:
 680         ocfs2_commit_trans(osb, handle);
 681
 682 out:
 683         if (new_tree) {
 684                 ocfs2_metadata_cache_exit(&new_tree->rf_ci);
 685                 kfree(new_tree);
 686         }
 687
 688         brelse(new_bh);
 689         if (meta_ac)
 690                 ocfs2_free_alloc_context(meta_ac);
 691
 692         return ret;
 693 }
 694
 695 static int ocfs2_set_refcount_tree(struct inode *inode,
 696                                    struct buffer_head *di_bh,
 697                                    u64 refcount_loc)
 698 {
 699         int ret;
 700         handle_t *handle = NULL;
 701         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 702         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 703         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 704         struct buffer_head *ref_root_bh = NULL;
 705         struct ocfs2_refcount_block *rb;
 706         struct ocfs2_refcount_tree *ref_tree;
 707
 708         BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
 709
 710         ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
 711                                        &ref_tree, &ref_root_bh);
 712         if (ret) {
 713                 mlog_errno(ret);
 714                 return ret;
 715         }
 716
 717         handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS);
 718         if (IS_ERR(handle)) {
 719                 ret = PTR_ERR(handle);
 720                 mlog_errno(ret);
 721                 goto out;
 722         }
 723
 724         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 725                                       OCFS2_JOURNAL_ACCESS_WRITE);
 726         if (ret) {
 727                 mlog_errno(ret);
 728                 goto out_commit;
 729         }
 730
 731         ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh,
 732                                       OCFS2_JOURNAL_ACCESS_WRITE);
 733         if (ret) {
 734                 mlog_errno(ret);
 735                 goto out_commit;
 736         }
 737
 738         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
 739         le32_add_cpu(&rb->rf_count, 1);
 740
 741         ocfs2_journal_dirty(handle, ref_root_bh);
 742
 743         spin_lock(&oi->ip_lock);
 744         oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
 745         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 746         di->i_refcount_loc = cpu_to_le64(refcount_loc);
 747         spin_unlock(&oi->ip_lock);
 748         ocfs2_journal_dirty(handle, di_bh);
 749
 750 out_commit:
 751         ocfs2_commit_trans(osb, handle);
 752 out:
 753         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
 754         brelse(ref_root_bh);
 755
 756         return ret;
 757 }
 758
 759 int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh)
 760 {
 761         int ret, delete_tree = 0;
 762         handle_t *handle = NULL;
 763         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 764         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 765         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 766         struct ocfs2_refcount_block *rb;
 767         struct inode *alloc_inode = NULL;
 768         struct buffer_head *alloc_bh = NULL;
 769         struct buffer_head *blk_bh = NULL;
 770         struct ocfs2_refcount_tree *ref_tree;
 771         int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS;
 772         u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc);
 773         u16 bit = 0;
 774
 775         if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL))
 776                 return 0;
 777
 778         BUG_ON(!ref_blkno);
 779         ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh);
 780         if (ret) {
 781                 mlog_errno(ret);
 782                 return ret;
 783         }
 784
 785         rb = (struct ocfs2_refcount_block *)blk_bh->b_data;
 786
 787         /*
 788          * If we are the last user, we need to free the block.
 789          * So lock the allocator ahead.
 790          */
 791         if (le32_to_cpu(rb->rf_count) == 1) {
 792                 blk = le64_to_cpu(rb->rf_blkno);
 793                 bit = le16_to_cpu(rb->rf_suballoc_bit);
 794                 if (rb->rf_suballoc_loc)
 795                         bg_blkno = le64_to_cpu(rb->rf_suballoc_loc);
 796                 else
 797                         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
 798
 799                 alloc_inode = ocfs2_get_system_file_inode(osb,
 800                                         EXTENT_ALLOC_SYSTEM_INODE,
 801                                         le16_to_cpu(rb->rf_suballoc_slot));
 802                 if (!alloc_inode) {
 803                         ret = -ENOMEM;
 804                         mlog_errno(ret);
 805                         goto out;
 806                 }
 807                 mutex_lock(&alloc_inode->i_mutex);
 808
 809                 ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1);
 810                 if (ret) {
 811                         mlog_errno(ret);
 812                         goto out_mutex;
 813                 }
 814
 815                 credits += OCFS2_SUBALLOC_FREE;
 816         }
 817
 818         handle = ocfs2_start_trans(osb, credits);
 819         if (IS_ERR(handle)) {
 820                 ret = PTR_ERR(handle);
 821                 mlog_errno(ret);
 822                 goto out_unlock;
 823         }
 824
 825         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 826                                       OCFS2_JOURNAL_ACCESS_WRITE);
 827         if (ret) {
 828                 mlog_errno(ret);
 829                 goto out_commit;
 830         }
 831
 832         ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh,
 833                                       OCFS2_JOURNAL_ACCESS_WRITE);
 834         if (ret) {
 835                 mlog_errno(ret);
 836                 goto out_commit;
 837         }
 838
 839         spin_lock(&oi->ip_lock);
 840         oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL;
 841         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 842         di->i_refcount_loc = 0;
 843         spin_unlock(&oi->ip_lock);
 844         ocfs2_journal_dirty(handle, di_bh);
 845
 846         le32_add_cpu(&rb->rf_count , -1);
 847         ocfs2_journal_dirty(handle, blk_bh);
 848
 849         if (!rb->rf_count) {
 850                 delete_tree = 1;
 851                 ocfs2_erase_refcount_tree_from_list(osb, ref_tree);
 852                 ret = ocfs2_free_suballoc_bits(handle, alloc_inode,
 853                                                alloc_bh, bit, bg_blkno, 1);
 854                 if (ret)
 855                         mlog_errno(ret);
 856         }
 857
 858 out_commit:
 859         ocfs2_commit_trans(osb, handle);
 860 out_unlock:
 861         if (alloc_inode) {
 862                 ocfs2_inode_unlock(alloc_inode, 1);
 863                 brelse(alloc_bh);
 864         }
 865 out_mutex:
 866         if (alloc_inode) {
 867                 mutex_unlock(&alloc_inode->i_mutex);
 868                 iput(alloc_inode);
 869         }
 870 out:
 871         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
 872         if (delete_tree)
 873                 ocfs2_refcount_tree_put(ref_tree);
 874         brelse(blk_bh);
 875
 876         return ret;
 877 }
 878
 879 static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info *ci,
 880                                           struct buffer_head *ref_leaf_bh,
 881                                           u64 cpos, unsigned int len,
 882                                           struct ocfs2_refcount_rec *ret_rec,
 883                                           int *index)
 884 {
 885         int i = 0;
 886         struct ocfs2_refcount_block *rb =
 887                 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
 888         struct ocfs2_refcount_rec *rec = NULL;
 889
 890         for (; i < le16_to_cpu(rb->rf_records.rl_used); i++) {
 891                 rec = &rb->rf_records.rl_recs[i];
 892
 893                 if (le64_to_cpu(rec->r_cpos) +
 894                     le32_to_cpu(rec->r_clusters) <= cpos)
 895                         continue;
 896                 else if (le64_to_cpu(rec->r_cpos) > cpos)
 897                         break;
 898
 899                 /* ok, cpos fail in this rec. Just return. */
 900                 if (ret_rec)
 901                         *ret_rec = *rec;
 902                 goto out;
 903         }
 904
 905         if (ret_rec) {
 906                 /* We meet with a hole here, so fake the rec. */
 907                 ret_rec->r_cpos = cpu_to_le64(cpos);
 908                 ret_rec->r_refcount = 0;
 909                 if (i < le16_to_cpu(rb->rf_records.rl_used) &&
 910                     le64_to_cpu(rec->r_cpos) < cpos + len)
 911                         ret_rec->r_clusters =
 912                                 cpu_to_le32(le64_to_cpu(rec->r_cpos) - cpos);
 913                 else
 914                         ret_rec->r_clusters = cpu_to_le32(len);
 915         }
 916
 917 out:
 918         *index = i;
 919 }
 920
 921 /*
 922  * Try to remove refcount tree. The mechanism is:
 923  * 1) Check whether i_clusters == 0, if no, exit.
 924  * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
 925  * 3) Check whether we have inline xattr stored outside, if yes, exit.
 926  * 4) Remove the tree.
 927  */
 928 int ocfs2_try_remove_refcount_tree(struct inode *inode,
 929                                    struct buffer_head *di_bh)
 930 {
 931         int ret;
 932         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 933         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 934
 935         down_write(&oi->ip_xattr_sem);
 936         down_write(&oi->ip_alloc_sem);
 937
 938         if (oi->ip_clusters)
 939                 goto out;
 940
 941         if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
 942                 goto out;
 943
 944         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
 945             ocfs2_has_inline_xattr_value_outside(inode, di))
 946                 goto out;
 947
 948         ret = ocfs2_remove_refcount_tree(inode, di_bh);
 949         if (ret)
 950                 mlog_errno(ret);
 951 out:
 952         up_write(&oi->ip_alloc_sem);
 953         up_write(&oi->ip_xattr_sem);
 954         return 0;
 955 }
 956
 957 /*
 958  * Find the end range for a leaf refcount block indicated by
 959  * el->l_recs[index].e_blkno.
 960  */
 961 static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info *ci,
 962                                        struct buffer_head *ref_root_bh,
 963                                        struct ocfs2_extent_block *eb,
 964                                        struct ocfs2_extent_list *el,
 965                                        int index,  u32 *cpos_end)
 966 {
 967         int ret, i, subtree_root;
 968         u32 cpos;
 969         u64 blkno;
 970         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
 971         struct ocfs2_path *left_path = NULL, *right_path = NULL;
 972         struct ocfs2_extent_tree et;
 973         struct ocfs2_extent_list *tmp_el;
 974
 975         if (index < le16_to_cpu(el->l_next_free_rec) - 1) {
 976                 /*
 977                  * We have a extent rec after index, so just use the e_cpos
 978                  * of the next extent rec.
 979                  */
 980                 *cpos_end = le32_to_cpu(el->l_recs[index+1].e_cpos);
 981                 return 0;
 982         }
 983
 984         if (!eb || (eb && !eb->h_next_leaf_blk)) {
 985                 /*
 986                  * We are the last extent rec, so any high cpos should
 987                  * be stored in this leaf refcount block.
 988                  */
 989                 *cpos_end = UINT_MAX;
 990                 return 0;
 991         }
 992
 993         /*
 994          * If the extent block isn't the last one, we have to find
 995          * the subtree root between this extent block and the next
 996          * leaf extent block and get the corresponding e_cpos from
 997          * the subroot. Otherwise we may corrupt the b-tree.
 998          */
 999         ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1000
1001         left_path = ocfs2_new_path_from_et(&et);
1002         if (!left_path) {
1003                 ret = -ENOMEM;
1004                 mlog_errno(ret);
1005                 goto out;
1006         }
1007
1008         cpos = le32_to_cpu(eb->h_list.l_recs[index].e_cpos);
1009         ret = ocfs2_find_path(ci, left_path, cpos);
1010         if (ret) {
1011                 mlog_errno(ret);
1012                 goto out;
1013         }
1014
1015         right_path = ocfs2_new_path_from_path(left_path);
1016         if (!right_path) {
1017                 ret = -ENOMEM;
1018                 mlog_errno(ret);
1019                 goto out;
1020         }
1021
1022         ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, &cpos);
1023         if (ret) {
1024                 mlog_errno(ret);
1025                 goto out;
1026         }
1027
1028         ret = ocfs2_find_path(ci, right_path, cpos);
1029         if (ret) {
1030                 mlog_errno(ret);
1031                 goto out;
1032         }
1033
1034         subtree_root = ocfs2_find_subtree_root(&et, left_path,
1035                                                right_path);
1036
1037         tmp_el = left_path->p_node[subtree_root].el;
1038         blkno = left_path->p_node[subtree_root+1].bh->b_blocknr;
1039         for (i = 0; i < le16_to_cpu(tmp_el->l_next_free_rec); i++) {
1040                 if (le64_to_cpu(tmp_el->l_recs[i].e_blkno) == blkno) {
1041                         *cpos_end = le32_to_cpu(tmp_el->l_recs[i+1].e_cpos);
1042                         break;
1043                 }
1044         }
1045
1046         BUG_ON(i == le16_to_cpu(tmp_el->l_next_free_rec));
1047
1048 out:
1049         ocfs2_free_path(left_path);
1050         ocfs2_free_path(right_path);
1051         return ret;
1052 }
1053
1054 /*
1055  * Given a cpos and len, try to find the refcount record which contains cpos.
1056  * 1. If cpos can be found in one refcount record, return the record.
1057  * 2. If cpos can't be found, return a fake record which start from cpos
1058  *    and end at a small value between cpos+len and start of the next record.
1059  *    This fake record has r_refcount = 0.
1060  */
1061 static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
1062                                   struct buffer_head *ref_root_bh,
1063                                   u64 cpos, unsigned int len,
1064                                   struct ocfs2_refcount_rec *ret_rec,
1065                                   int *index,
1066                                   struct buffer_head **ret_bh)
1067 {
1068         int ret = 0, i, found;
1069         u32 low_cpos, uninitialized_var(cpos_end);
1070         struct ocfs2_extent_list *el;
1071         struct ocfs2_extent_rec *rec = NULL;
1072         struct ocfs2_extent_block *eb = NULL;
1073         struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
1074         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1075         struct ocfs2_refcount_block *rb =
1076                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1077
1078         if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)) {
1079                 ocfs2_find_refcount_rec_in_rl(ci, ref_root_bh, cpos, len,
1080                                               ret_rec, index);
1081                 *ret_bh = ref_root_bh;
1082                 get_bh(ref_root_bh);
1083                 return 0;
1084         }
1085
1086         el = &rb->rf_list;
1087         low_cpos = cpos & OCFS2_32BIT_POS_MASK;
1088
1089         if (el->l_tree_depth) {
1090                 ret = ocfs2_find_leaf(ci, el, low_cpos, &eb_bh);
1091                 if (ret) {
1092                         mlog_errno(ret);
1093                         goto out;
1094                 }
1095
1096                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1097                 el = &eb->h_list;
1098
1099                 if (el->l_tree_depth) {
1100                         ocfs2_error(sb,
1101                         "refcount tree %llu has non zero tree "
1102                         "depth in leaf btree tree block %llu\n",
1103                         (unsigned long long)ocfs2_metadata_cache_owner(ci),
1104                         (unsigned long long)eb_bh->b_blocknr);
1105                         ret = -EROFS;
1106                         goto out;
1107                 }
1108         }
1109
1110         found = 0;
1111         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1112                 rec = &el->l_recs[i];
1113
1114                 if (le32_to_cpu(rec->e_cpos) <= low_cpos) {
1115                         found = 1;
1116                         break;
1117                 }
1118         }
1119
1120         if (found) {
1121                 ret = ocfs2_get_refcount_cpos_end(ci, ref_root_bh,
1122                                                   eb, el, i, &cpos_end);
1123                 if (ret) {
1124                         mlog_errno(ret);
1125                         goto out;
1126                 }
1127
1128                 if (cpos_end < low_cpos + len)
1129                         len = cpos_end - low_cpos;
1130         }
1131
1132         ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
1133                                         &ref_leaf_bh);
1134         if (ret) {
1135                 mlog_errno(ret);
1136                 goto out;
1137         }
1138
1139         ocfs2_find_refcount_rec_in_rl(ci, ref_leaf_bh, cpos, len,
1140                                       ret_rec, index);
1141         *ret_bh = ref_leaf_bh;
1142 out:
1143         brelse(eb_bh);
1144         return ret;
1145 }
1146
1147 enum ocfs2_ref_rec_contig {
1148         REF_CONTIG_NONE = 0,
1149         REF_CONTIG_LEFT,
1150         REF_CONTIG_RIGHT,
1151         REF_CONTIG_LEFTRIGHT,
1152 };
1153
1154 static enum ocfs2_ref_rec_contig
1155         ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block *rb,
1156                                     int index)
1157 {
1158         if ((rb->rf_records.rl_recs[index].r_refcount ==
1159             rb->rf_records.rl_recs[index + 1].r_refcount) &&
1160             (le64_to_cpu(rb->rf_records.rl_recs[index].r_cpos) +
1161             le32_to_cpu(rb->rf_records.rl_recs[index].r_clusters) ==
1162             le64_to_cpu(rb->rf_records.rl_recs[index + 1].r_cpos)))
1163                 return REF_CONTIG_RIGHT;
1164
1165         return REF_CONTIG_NONE;
1166 }
1167
1168 static enum ocfs2_ref_rec_contig
1169         ocfs2_refcount_rec_contig(struct ocfs2_refcount_block *rb,
1170                                   int index)
1171 {
1172         enum ocfs2_ref_rec_contig ret = REF_CONTIG_NONE;
1173
1174         if (index < le16_to_cpu(rb->rf_records.rl_used) - 1)
1175                 ret = ocfs2_refcount_rec_adjacent(rb, index);
1176
1177         if (index > 0) {
1178                 enum ocfs2_ref_rec_contig tmp;
1179
1180                 tmp = ocfs2_refcount_rec_adjacent(rb, index - 1);
1181
1182                 if (tmp == REF_CONTIG_RIGHT) {
1183                         if (ret == REF_CONTIG_RIGHT)
1184                                 ret = REF_CONTIG_LEFTRIGHT;
1185                         else
1186                                 ret = REF_CONTIG_LEFT;
1187                 }
1188         }
1189
1190         return ret;
1191 }
1192
1193 static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block *rb,
1194                                            int index)
1195 {
1196         BUG_ON(rb->rf_records.rl_recs[index].r_refcount !=
1197                rb->rf_records.rl_recs[index+1].r_refcount);
1198
1199         le32_add_cpu(&rb->rf_records.rl_recs[index].r_clusters,
1200                      le32_to_cpu(rb->rf_records.rl_recs[index+1].r_clusters));
1201
1202         if (index < le16_to_cpu(rb->rf_records.rl_used) - 2)
1203                 memmove(&rb->rf_records.rl_recs[index + 1],
1204                         &rb->rf_records.rl_recs[index + 2],
1205                         sizeof(struct ocfs2_refcount_rec) *
1206                         (le16_to_cpu(rb->rf_records.rl_used) - index - 2));
1207
1208         memset(&rb->rf_records.rl_recs[le16_to_cpu(rb->rf_records.rl_used) - 1],
1209                0, sizeof(struct ocfs2_refcount_rec));
1210         le16_add_cpu(&rb->rf_records.rl_used, -1);
1211 }
1212
1213 /*
1214  * Merge the refcount rec if we are contiguous with the adjacent recs.
1215  */
1216 static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
1217                                      int index)
1218 {
1219         enum ocfs2_ref_rec_contig contig =
1220                                 ocfs2_refcount_rec_contig(rb, index);
1221
1222         if (contig == REF_CONTIG_NONE)
1223                 return;
1224
1225         if (contig == REF_CONTIG_LEFT || contig == REF_CONTIG_LEFTRIGHT) {
1226                 BUG_ON(index == 0);
1227                 index--;
1228         }
1229
1230         ocfs2_rotate_refcount_rec_left(rb, index);
1231
1232         if (contig == REF_CONTIG_LEFTRIGHT)
1233                 ocfs2_rotate_refcount_rec_left(rb, index);
1234 }
1235
1236 /*
1237  * Change the refcount indexed by "index" in ref_bh.
1238  * If refcount reaches 0, remove it.
1239  */
1240 static int ocfs2_change_refcount_rec(handle_t *handle,
1241                                      struct ocfs2_caching_info *ci,
1242                                      struct buffer_head *ref_leaf_bh,
1243                                      int index, int merge, int change)
1244 {
1245         int ret;
1246         struct ocfs2_refcount_block *rb =
1247                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1248         struct ocfs2_refcount_list *rl = &rb->rf_records;
1249         struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
1250
1251         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1252                                       OCFS2_JOURNAL_ACCESS_WRITE);
1253         if (ret) {
1254                 mlog_errno(ret);
1255                 goto out;
1256         }
1257
1258         trace_ocfs2_change_refcount_rec(
1259                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1260                 index, le32_to_cpu(rec->r_refcount), change);
1261         le32_add_cpu(&rec->r_refcount, change);
1262
1263         if (!rec->r_refcount) {
1264                 if (index != le16_to_cpu(rl->rl_used) - 1) {
1265                         memmove(rec, rec + 1,
1266                                 (le16_to_cpu(rl->rl_used) - index - 1) *
1267                                 sizeof(struct ocfs2_refcount_rec));
1268                         memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
1269                                0, sizeof(struct ocfs2_refcount_rec));
1270                 }
1271
1272                 le16_add_cpu(&rl->rl_used, -1);
1273         } else if (merge)
1274                 ocfs2_refcount_rec_merge(rb, index);
1275
1276         ocfs2_journal_dirty(handle, ref_leaf_bh);
1277 out:
1278         return ret;
1279 }
1280
1281 static int ocfs2_expand_inline_ref_root(handle_t *handle,
1282                                         struct ocfs2_caching_info *ci,
1283                                         struct buffer_head *ref_root_bh,
1284                                         struct buffer_head **ref_leaf_bh,
1285                                         struct ocfs2_alloc_context *meta_ac)
1286 {
1287         int ret;
1288         u16 suballoc_bit_start;
1289         u32 num_got;
1290         u64 suballoc_loc, blkno;
1291         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1292         struct buffer_head *new_bh = NULL;
1293         struct ocfs2_refcount_block *new_rb;
1294         struct ocfs2_refcount_block *root_rb =
1295                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1296
1297         ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1298                                       OCFS2_JOURNAL_ACCESS_WRITE);
1299         if (ret) {
1300                 mlog_errno(ret);
1301                 goto out;
1302         }
1303
1304         ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1305                                    &suballoc_bit_start, &num_got,
1306                                    &blkno);
1307         if (ret) {
1308                 mlog_errno(ret);
1309                 goto out;
1310         }
1311
1312         new_bh = sb_getblk(sb, blkno);
1313         if (new_bh == NULL) {
1314                 ret = -EIO;
1315                 mlog_errno(ret);
1316                 goto out;
1317         }
1318         ocfs2_set_new_buffer_uptodate(ci, new_bh);
1319
1320         ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1321                                       OCFS2_JOURNAL_ACCESS_CREATE);
1322         if (ret) {
1323                 mlog_errno(ret);
1324                 goto out;
1325         }
1326
1327         /*
1328          * Initialize ocfs2_refcount_block.
1329          * It should contain the same information as the old root.
1330          * so just memcpy it and change the corresponding field.
1331          */
1332         memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
1333
1334         new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1335         new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1336         new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1337         new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1338         new_rb->rf_blkno = cpu_to_le64(blkno);
1339         new_rb->rf_cpos = cpu_to_le32(0);
1340         new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1341         new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1342         ocfs2_journal_dirty(handle, new_bh);
1343
1344         /* Now change the root. */
1345         memset(&root_rb->rf_list, 0, sb->s_blocksize -
1346                offsetof(struct ocfs2_refcount_block, rf_list));
1347         root_rb->rf_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_rb(sb));
1348         root_rb->rf_clusters = cpu_to_le32(1);
1349         root_rb->rf_list.l_next_free_rec = cpu_to_le16(1);
1350         root_rb->rf_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
1351         root_rb->rf_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
1352         root_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_TREE_FL);
1353
1354         ocfs2_journal_dirty(handle, ref_root_bh);
1355
1356         trace_ocfs2_expand_inline_ref_root((unsigned long long)blkno,
1357                 le16_to_cpu(new_rb->rf_records.rl_used));
1358
1359         *ref_leaf_bh = new_bh;
1360         new_bh = NULL;
1361 out:
1362         brelse(new_bh);
1363         return ret;
1364 }
1365
1366 static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec *prev,
1367                                            struct ocfs2_refcount_rec *next)
1368 {
1369         if (ocfs2_get_ref_rec_low_cpos(prev) + le32_to_cpu(prev->r_clusters) <=
1370                 ocfs2_get_ref_rec_low_cpos(next))
1371                 return 1;
1372
1373         return 0;
1374 }
1375
1376 static int cmp_refcount_rec_by_low_cpos(const void *a, const void *b)
1377 {
1378         const struct ocfs2_refcount_rec *l = a, *r = b;
1379         u32 l_cpos = ocfs2_get_ref_rec_low_cpos(l);
1380         u32 r_cpos = ocfs2_get_ref_rec_low_cpos(r);
1381
1382         if (l_cpos > r_cpos)
1383                 return 1;
1384         if (l_cpos < r_cpos)
1385                 return -1;
1386         return 0;
1387 }
1388
1389 static int cmp_refcount_rec_by_cpos(const void *a, const void *b)
1390 {
1391         const struct ocfs2_refcount_rec *l = a, *r = b;
1392         u64 l_cpos = le64_to_cpu(l->r_cpos);
1393         u64 r_cpos = le64_to_cpu(r->r_cpos);
1394
1395         if (l_cpos > r_cpos)
1396                 return 1;
1397         if (l_cpos < r_cpos)
1398                 return -1;
1399         return 0;
1400 }
1401
1402 static void swap_refcount_rec(void *a, void *b, int size)
1403 {
1404         struct ocfs2_refcount_rec *l = a, *r = b, tmp;
1405
1406         tmp = *(struct ocfs2_refcount_rec *)l;
1407         *(struct ocfs2_refcount_rec *)l =
1408                         *(struct ocfs2_refcount_rec *)r;
1409         *(struct ocfs2_refcount_rec *)r = tmp;
1410 }
1411
1412 /*
1413  * The refcount cpos are ordered by their 64bit cpos,
1414  * But we will use the low 32 bit to be the e_cpos in the b-tree.
1415  * So we need to make sure that this pos isn't intersected with others.
1416  *
1417  * Note: The refcount block is already sorted by their low 32 bit cpos,
1418  *       So just try the middle pos first, and we will exit when we find
1419  *       the good position.
1420  */
1421 static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
1422                                          u32 *split_pos, int *split_index)
1423 {
1424         int num_used = le16_to_cpu(rl->rl_used);
1425         int delta, middle = num_used / 2;
1426
1427         for (delta = 0; delta < middle; delta++) {
1428                 /* Let's check delta earlier than middle */
1429                 if (ocfs2_refcount_rec_no_intersect(
1430                                         &rl->rl_recs[middle - delta - 1],
1431                                         &rl->rl_recs[middle - delta])) {
1432                         *split_index = middle - delta;
1433                         break;
1434                 }
1435
1436                 /* For even counts, don't walk off the end */
1437                 if ((middle + delta + 1) == num_used)
1438                         continue;
1439
1440                 /* Now try delta past middle */
1441                 if (ocfs2_refcount_rec_no_intersect(
1442                                         &rl->rl_recs[middle + delta],
1443                                         &rl->rl_recs[middle + delta + 1])) {
1444                         *split_index = middle + delta + 1;
1445                         break;
1446                 }
1447         }
1448
1449         if (delta >= middle)
1450                 return -ENOSPC;
1451
1452         *split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
1453         return 0;
1454 }
1455
1456 static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
1457                                             struct buffer_head *new_bh,
1458                                             u32 *split_cpos)
1459 {
1460         int split_index = 0, num_moved, ret;
1461         u32 cpos = 0;
1462         struct ocfs2_refcount_block *rb =
1463                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1464         struct ocfs2_refcount_list *rl = &rb->rf_records;
1465         struct ocfs2_refcount_block *new_rb =
1466                         (struct ocfs2_refcount_block *)new_bh->b_data;
1467         struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
1468
1469         trace_ocfs2_divide_leaf_refcount_block(
1470                 (unsigned long long)ref_leaf_bh->b_blocknr,
1471                 le16_to_cpu(rl->rl_count), le16_to_cpu(rl->rl_used));
1472
1473         /*
1474          * XXX: Improvement later.
1475          * If we know all the high 32 bit cpos is the same, no need to sort.
1476          *
1477          * In order to make the whole process safe, we do:
1478          * 1. sort the entries by their low 32 bit cpos first so that we can
1479          *    find the split cpos easily.
1480          * 2. call ocfs2_insert_extent to insert the new refcount block.
1481          * 3. move the refcount rec to the new block.
1482          * 4. sort the entries by their 64 bit cpos.
1483          * 5. dirty the new_rb and rb.
1484          */
1485         sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1486              sizeof(struct ocfs2_refcount_rec),
1487              cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
1488
1489         ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
1490         if (ret) {
1491                 mlog_errno(ret);
1492                 return ret;
1493         }
1494
1495         new_rb->rf_cpos = cpu_to_le32(cpos);
1496
1497         /* move refcount records starting from split_index to the new block. */
1498         num_moved = le16_to_cpu(rl->rl_used) - split_index;
1499         memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
1500                num_moved * sizeof(struct ocfs2_refcount_rec));
1501
1502         /*ok, remove the entries we just moved over to the other block. */
1503         memset(&rl->rl_recs[split_index], 0,
1504                num_moved * sizeof(struct ocfs2_refcount_rec));
1505
1506         /* change old and new rl_used accordingly. */
1507         le16_add_cpu(&rl->rl_used, -num_moved);
1508         new_rl->rl_used = cpu_to_le16(num_moved);
1509
1510         sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1511              sizeof(struct ocfs2_refcount_rec),
1512              cmp_refcount_rec_by_cpos, swap_refcount_rec);
1513
1514         sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
1515              sizeof(struct ocfs2_refcount_rec),
1516              cmp_refcount_rec_by_cpos, swap_refcount_rec);
1517
1518         *split_cpos = cpos;
1519         return 0;
1520 }
1521
1522 static int ocfs2_new_leaf_refcount_block(handle_t *handle,
1523                                          struct ocfs2_caching_info *ci,
1524                                          struct buffer_head *ref_root_bh,
1525                                          struct buffer_head *ref_leaf_bh,
1526                                          struct ocfs2_alloc_context *meta_ac)
1527 {
1528         int ret;
1529         u16 suballoc_bit_start;
1530         u32 num_got, new_cpos;
1531         u64 suballoc_loc, blkno;
1532         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1533         struct ocfs2_refcount_block *root_rb =
1534                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1535         struct buffer_head *new_bh = NULL;
1536         struct ocfs2_refcount_block *new_rb;
1537         struct ocfs2_extent_tree ref_et;
1538
1539         BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
1540
1541         ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1542                                       OCFS2_JOURNAL_ACCESS_WRITE);
1543         if (ret) {
1544                 mlog_errno(ret);
1545                 goto out;
1546         }
1547
1548         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1549                                       OCFS2_JOURNAL_ACCESS_WRITE);
1550         if (ret) {
1551                 mlog_errno(ret);
1552                 goto out;
1553         }
1554
1555         ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1556                                    &suballoc_bit_start, &num_got,
1557                                    &blkno);
1558         if (ret) {
1559                 mlog_errno(ret);
1560                 goto out;
1561         }
1562
1563         new_bh = sb_getblk(sb, blkno);
1564         if (new_bh == NULL) {
1565                 ret = -EIO;
1566                 mlog_errno(ret);
1567                 goto out;
1568         }
1569         ocfs2_set_new_buffer_uptodate(ci, new_bh);
1570
1571         ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1572                                       OCFS2_JOURNAL_ACCESS_CREATE);
1573         if (ret) {
1574                 mlog_errno(ret);
1575                 goto out;
1576         }
1577
1578         /* Initialize ocfs2_refcount_block. */
1579         new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1580         memset(new_rb, 0, sb->s_blocksize);
1581         strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
1582         new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1583         new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1584         new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1585         new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1586         new_rb->rf_blkno = cpu_to_le64(blkno);
1587         new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1588         new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1589         new_rb->rf_records.rl_count =
1590                                 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
1591         new_rb->rf_generation = root_rb->rf_generation;
1592
1593         ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
1594         if (ret) {
1595                 mlog_errno(ret);
1596                 goto out;
1597         }
1598
1599         ocfs2_journal_dirty(handle, ref_leaf_bh);
1600         ocfs2_journal_dirty(handle, new_bh);
1601
1602         ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
1603
1604         trace_ocfs2_new_leaf_refcount_block(
1605                         (unsigned long long)new_bh->b_blocknr, new_cpos);
1606
1607         /* Insert the new leaf block with the specific offset cpos. */
1608         ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
1609                                   1, 0, meta_ac);
1610         if (ret)
1611                 mlog_errno(ret);
1612
1613 out:
1614         brelse(new_bh);
1615         return ret;
1616 }
1617
1618 static int ocfs2_expand_refcount_tree(handle_t *handle,
1619                                       struct ocfs2_caching_info *ci,
1620                                       struct buffer_head *ref_root_bh,
1621                                       struct buffer_head *ref_leaf_bh,
1622                                       struct ocfs2_alloc_context *meta_ac)
1623 {
1624         int ret;
1625         struct buffer_head *expand_bh = NULL;
1626
1627         if (ref_root_bh == ref_leaf_bh) {
1628                 /*
1629                  * the old root bh hasn't been expanded to a b-tree,
1630                  * so expand it first.
1631                  */
1632                 ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
1633                                                    &expand_bh, meta_ac);
1634                 if (ret) {
1635                         mlog_errno(ret);
1636                         goto out;
1637                 }
1638         } else {
1639                 expand_bh = ref_leaf_bh;
1640                 get_bh(expand_bh);
1641         }
1642
1643
1644         /* Now add a new refcount block into the tree.*/
1645         ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
1646                                             expand_bh, meta_ac);
1647         if (ret)
1648                 mlog_errno(ret);
1649 out:
1650         brelse(expand_bh);
1651         return ret;
1652 }
1653
1654 /*
1655  * Adjust the extent rec in b-tree representing ref_leaf_bh.
1656  *
1657  * Only called when we have inserted a new refcount rec at index 0
1658  * which means ocfs2_extent_rec.e_cpos may need some change.
1659  */
1660 static int ocfs2_adjust_refcount_rec(handle_t *handle,
1661                                      struct ocfs2_caching_info *ci,
1662                                      struct buffer_head *ref_root_bh,
1663                                      struct buffer_head *ref_leaf_bh,
1664                                      struct ocfs2_refcount_rec *rec)
1665 {
1666         int ret = 0, i;
1667         u32 new_cpos, old_cpos;
1668         struct ocfs2_path *path = NULL;
1669         struct ocfs2_extent_tree et;
1670         struct ocfs2_refcount_block *rb =
1671                 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1672         struct ocfs2_extent_list *el;
1673
1674         if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
1675                 goto out;
1676
1677         rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1678         old_cpos = le32_to_cpu(rb->rf_cpos);
1679         new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
1680         if (old_cpos <= new_cpos)
1681                 goto out;
1682
1683         ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1684
1685         path = ocfs2_new_path_from_et(&et);
1686         if (!path) {
1687                 ret = -ENOMEM;
1688                 mlog_errno(ret);
1689                 goto out;
1690         }
1691
1692         ret = ocfs2_find_path(ci, path, old_cpos);
1693         if (ret) {
1694                 mlog_errno(ret);
1695                 goto out;
1696         }
1697
1698         /*
1699          * 2 more credits, one for the leaf refcount block, one for
1700          * the extent block contains the extent rec.
1701          */
1702         ret = ocfs2_extend_trans(handle, 2);
1703         if (ret < 0) {
1704                 mlog_errno(ret);
1705                 goto out;
1706         }
1707
1708         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1709                                       OCFS2_JOURNAL_ACCESS_WRITE);
1710         if (ret < 0) {
1711                 mlog_errno(ret);
1712                 goto out;
1713         }
1714
1715         ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
1716                                       OCFS2_JOURNAL_ACCESS_WRITE);
1717         if (ret < 0) {
1718                 mlog_errno(ret);
1719                 goto out;
1720         }
1721
1722         /* change the leaf extent block first. */
1723         el = path_leaf_el(path);
1724
1725         for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
1726                 if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
1727                         break;
1728
1729         BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
1730
1731         el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1732
1733         /* change the r_cpos in the leaf block. */
1734         rb->rf_cpos = cpu_to_le32(new_cpos);
1735
1736         ocfs2_journal_dirty(handle, path_leaf_bh(path));
1737         ocfs2_journal_dirty(handle, ref_leaf_bh);
1738
1739 out:
1740         ocfs2_free_path(path);
1741         return ret;
1742 }
1743
1744 static int ocfs2_insert_refcount_rec(handle_t *handle,
1745                                      struct ocfs2_caching_info *ci,
1746                                      struct buffer_head *ref_root_bh,
1747                                      struct buffer_head *ref_leaf_bh,
1748                                      struct ocfs2_refcount_rec *rec,
1749                                      int index, int merge,
1750                                      struct ocfs2_alloc_context *meta_ac)
1751 {
1752         int ret;
1753         struct ocfs2_refcount_block *rb =
1754                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1755         struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1756         struct buffer_head *new_bh = NULL;
1757
1758         BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1759
1760         if (rf_list->rl_used == rf_list->rl_count) {
1761                 u64 cpos = le64_to_cpu(rec->r_cpos);
1762                 u32 len = le32_to_cpu(rec->r_clusters);
1763
1764                 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1765                                                  ref_leaf_bh, meta_ac);
1766                 if (ret) {
1767                         mlog_errno(ret);
1768                         goto out;
1769                 }
1770
1771                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1772                                              cpos, len, NULL, &index,
1773                                              &new_bh);
1774                 if (ret) {
1775                         mlog_errno(ret);
1776                         goto out;
1777                 }
1778
1779                 ref_leaf_bh = new_bh;
1780                 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1781                 rf_list = &rb->rf_records;
1782         }
1783
1784         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1785                                       OCFS2_JOURNAL_ACCESS_WRITE);
1786         if (ret) {
1787                 mlog_errno(ret);
1788                 goto out;
1789         }
1790
1791         if (index < le16_to_cpu(rf_list->rl_used))
1792                 memmove(&rf_list->rl_recs[index + 1],
1793                         &rf_list->rl_recs[index],
1794                         (le16_to_cpu(rf_list->rl_used) - index) *
1795                          sizeof(struct ocfs2_refcount_rec));
1796
1797         trace_ocfs2_insert_refcount_rec(
1798                 (unsigned long long)ref_leaf_bh->b_blocknr, index,
1799                 (unsigned long long)le64_to_cpu(rec->r_cpos),
1800                 le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount));
1801
1802         rf_list->rl_recs[index] = *rec;
1803
1804         le16_add_cpu(&rf_list->rl_used, 1);
1805
1806         if (merge)
1807                 ocfs2_refcount_rec_merge(rb, index);
1808
1809         ocfs2_journal_dirty(handle, ref_leaf_bh);
1810
1811         if (index == 0) {
1812                 ret = ocfs2_adjust_refcount_rec(handle, ci,
1813                                                 ref_root_bh,
1814                                                 ref_leaf_bh, rec);
1815                 if (ret)
1816                         mlog_errno(ret);
1817         }
1818 out:
1819         brelse(new_bh);
1820         return ret;
1821 }
1822
1823 /*
1824  * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1825  * This is much simple than our b-tree code.
1826  * split_rec is the new refcount rec we want to insert.
1827  * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1828  * increase refcount or decrease a refcount to non-zero).
1829  * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1830  * rec( in case we decrease a refcount to zero).
1831  */
1832 static int ocfs2_split_refcount_rec(handle_t *handle,
1833                                     struct ocfs2_caching_info *ci,
1834                                     struct buffer_head *ref_root_bh,
1835                                     struct buffer_head *ref_leaf_bh,
1836                                     struct ocfs2_refcount_rec *split_rec,
1837                                     int index, int merge,
1838                                     struct ocfs2_alloc_context *meta_ac,
1839                                     struct ocfs2_cached_dealloc_ctxt *dealloc)
1840 {
1841         int ret, recs_need;
1842         u32 len;
1843         struct ocfs2_refcount_block *rb =
1844                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1845         struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1846         struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
1847         struct ocfs2_refcount_rec *tail_rec = NULL;
1848         struct buffer_head *new_bh = NULL;
1849
1850         BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1851
1852         trace_ocfs2_split_refcount_rec(le64_to_cpu(orig_rec->r_cpos),
1853                 le32_to_cpu(orig_rec->r_clusters),
1854                 le32_to_cpu(orig_rec->r_refcount),
1855                 le64_to_cpu(split_rec->r_cpos),
1856                 le32_to_cpu(split_rec->r_clusters),
1857                 le32_to_cpu(split_rec->r_refcount));
1858
1859         /*
1860          * If we just need to split the header or tail clusters,
1861          * no more recs are needed, just split is OK.
1862          * Otherwise we at least need one new recs.
1863          */
1864         if (!split_rec->r_refcount &&
1865             (split_rec->r_cpos == orig_rec->r_cpos ||
1866              le64_to_cpu(split_rec->r_cpos) +
1867              le32_to_cpu(split_rec->r_clusters) ==
1868              le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1869                 recs_need = 0;
1870         else
1871                 recs_need = 1;
1872
1873         /*
1874          * We need one more rec if we split in the middle and the new rec have
1875          * some refcount in it.
1876          */
1877         if (split_rec->r_refcount &&
1878             (split_rec->r_cpos != orig_rec->r_cpos &&
1879              le64_to_cpu(split_rec->r_cpos) +
1880              le32_to_cpu(split_rec->r_clusters) !=
1881              le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1882                 recs_need++;
1883
1884         /* If the leaf block don't have enough record, expand it. */
1885         if (le16_to_cpu(rf_list->rl_used) + recs_need >
1886                                          le16_to_cpu(rf_list->rl_count)) {
1887                 struct ocfs2_refcount_rec tmp_rec;
1888                 u64 cpos = le64_to_cpu(orig_rec->r_cpos);
1889                 len = le32_to_cpu(orig_rec->r_clusters);
1890                 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1891                                                  ref_leaf_bh, meta_ac);
1892                 if (ret) {
1893                         mlog_errno(ret);
1894                         goto out;
1895                 }
1896
1897                 /*
1898                  * We have to re-get it since now cpos may be moved to
1899                  * another leaf block.
1900                  */
1901                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1902                                              cpos, len, &tmp_rec, &index,
1903                                              &new_bh);
1904                 if (ret) {
1905                         mlog_errno(ret);
1906                         goto out;
1907                 }
1908
1909                 ref_leaf_bh = new_bh;
1910                 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1911                 rf_list = &rb->rf_records;
1912                 orig_rec = &rf_list->rl_recs[index];
1913         }
1914
1915         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1916                                       OCFS2_JOURNAL_ACCESS_WRITE);
1917         if (ret) {
1918                 mlog_errno(ret);
1919                 goto out;
1920         }
1921
1922         /*
1923          * We have calculated out how many new records we need and store
1924          * in recs_need, so spare enough space first by moving the records
1925          * after "index" to the end.
1926          */
1927         if (index != le16_to_cpu(rf_list->rl_used) - 1)
1928                 memmove(&rf_list->rl_recs[index + 1 + recs_need],
1929                         &rf_list->rl_recs[index + 1],
1930                         (le16_to_cpu(rf_list->rl_used) - index - 1) *
1931                          sizeof(struct ocfs2_refcount_rec));
1932
1933         len = (le64_to_cpu(orig_rec->r_cpos) +
1934               le32_to_cpu(orig_rec->r_clusters)) -
1935               (le64_to_cpu(split_rec->r_cpos) +
1936               le32_to_cpu(split_rec->r_clusters));
1937
1938         /*
1939          * If we have "len", the we will split in the tail and move it
1940          * to the end of the space we have just spared.
1941          */
1942         if (len) {
1943                 tail_rec = &rf_list->rl_recs[index + recs_need];
1944
1945                 memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
1946                 le64_add_cpu(&tail_rec->r_cpos,
1947                              le32_to_cpu(tail_rec->r_clusters) - len);
1948                 tail_rec->r_clusters = cpu_to_le32(len);
1949         }
1950
1951         /*
1952          * If the split pos isn't the same as the original one, we need to
1953          * split in the head.
1954          *
1955          * Note: We have the chance that split_rec.r_refcount = 0,
1956          * recs_need = 0 and len > 0, which means we just cut the head from
1957          * the orig_rec and in that case we have done some modification in
1958          * orig_rec above, so the check for r_cpos is faked.
1959          */
1960         if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
1961                 len = le64_to_cpu(split_rec->r_cpos) -
1962                       le64_to_cpu(orig_rec->r_cpos);
1963                 orig_rec->r_clusters = cpu_to_le32(len);
1964                 index++;
1965         }
1966
1967         le16_add_cpu(&rf_list->rl_used, recs_need);
1968
1969         if (split_rec->r_refcount) {
1970                 rf_list->rl_recs[index] = *split_rec;
1971                 trace_ocfs2_split_refcount_rec_insert(
1972                         (unsigned long long)ref_leaf_bh->b_blocknr, index,
1973                         (unsigned long long)le64_to_cpu(split_rec->r_cpos),
1974                         le32_to_cpu(split_rec->r_clusters),
1975                         le32_to_cpu(split_rec->r_refcount));
1976
1977                 if (merge)
1978                         ocfs2_refcount_rec_merge(rb, index);
1979         }
1980
1981         ocfs2_journal_dirty(handle, ref_leaf_bh);
1982
1983 out:
1984         brelse(new_bh);
1985         return ret;
1986 }
1987
1988 static int __ocfs2_increase_refcount(handle_t *handle,
1989                                      struct ocfs2_caching_info *ci,
1990                                      struct buffer_head *ref_root_bh,
1991                                      u64 cpos, u32 len, int merge,
1992                                      struct ocfs2_alloc_context *meta_ac,
1993                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
1994 {
1995         int ret = 0, index;
1996         struct buffer_head *ref_leaf_bh = NULL;
1997         struct ocfs2_refcount_rec rec;
1998         unsigned int set_len = 0;
1999
2000         trace_ocfs2_increase_refcount_begin(
2001              (unsigned long long)ocfs2_metadata_cache_owner(ci),
2002              (unsigned long long)cpos, len);
2003
2004         while (len) {
2005                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2006                                              cpos, len, &rec, &index,
2007                                              &ref_leaf_bh);
2008                 if (ret) {
2009                         mlog_errno(ret);
2010                         goto out;
2011                 }
2012
2013                 set_len = le32_to_cpu(rec.r_clusters);
2014
2015                 /*
2016                  * Here we may meet with 3 situations:
2017                  *
2018                  * 1. If we find an already existing record, and the length
2019                  *    is the same, cool, we just need to increase the r_refcount
2020                  *    and it is OK.
2021                  * 2. If we find a hole, just insert it with r_refcount = 1.
2022                  * 3. If we are in the middle of one extent record, split
2023                  *    it.
2024                  */
2025                 if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
2026                     set_len <= len) {
2027                         trace_ocfs2_increase_refcount_change(
2028                                 (unsigned long long)cpos, set_len,
2029                                 le32_to_cpu(rec.r_refcount));
2030                         ret = ocfs2_change_refcount_rec(handle, ci,
2031                                                         ref_leaf_bh, index,
2032                                                         merge, 1);
2033                         if (ret) {
2034                                 mlog_errno(ret);
2035                                 goto out;
2036                         }
2037                 } else if (!rec.r_refcount) {
2038                         rec.r_refcount = cpu_to_le32(1);
2039
2040                         trace_ocfs2_increase_refcount_insert(
2041                              (unsigned long long)le64_to_cpu(rec.r_cpos),
2042                              set_len);
2043                         ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
2044                                                         ref_leaf_bh,
2045                                                         &rec, index,
2046                                                         merge, meta_ac);
2047                         if (ret) {
2048                                 mlog_errno(ret);
2049                                 goto out;
2050                         }
2051                 } else  {
2052                         set_len = min((u64)(cpos + len),
2053                                       le64_to_cpu(rec.r_cpos) + set_len) - cpos;
2054                         rec.r_cpos = cpu_to_le64(cpos);
2055                         rec.r_clusters = cpu_to_le32(set_len);
2056                         le32_add_cpu(&rec.r_refcount, 1);
2057
2058                         trace_ocfs2_increase_refcount_split(
2059                              (unsigned long long)le64_to_cpu(rec.r_cpos),
2060                              set_len, le32_to_cpu(rec.r_refcount));
2061                         ret = ocfs2_split_refcount_rec(handle, ci,
2062                                                        ref_root_bh, ref_leaf_bh,
2063                                                        &rec, index, merge,
2064                                                        meta_ac, dealloc);
2065                         if (ret) {
2066                                 mlog_errno(ret);
2067                                 goto out;
2068                         }
2069                 }
2070
2071                 cpos += set_len;
2072                 len -= set_len;
2073                 brelse(ref_leaf_bh);
2074                 ref_leaf_bh = NULL;
2075         }
2076
2077 out:
2078         brelse(ref_leaf_bh);
2079         return ret;
2080 }
2081
2082 static int ocfs2_remove_refcount_extent(handle_t *handle,
2083                                 struct ocfs2_caching_info *ci,
2084                                 struct buffer_head *ref_root_bh,
2085                                 struct buffer_head *ref_leaf_bh,
2086                                 struct ocfs2_alloc_context *meta_ac,
2087                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
2088 {
2089         int ret;
2090         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2091         struct ocfs2_refcount_block *rb =
2092                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2093         struct ocfs2_extent_tree et;
2094
2095         BUG_ON(rb->rf_records.rl_used);
2096
2097         trace_ocfs2_remove_refcount_extent(
2098                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2099                 (unsigned long long)ref_leaf_bh->b_blocknr,
2100                 le32_to_cpu(rb->rf_cpos));
2101
2102         ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2103         ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
2104                                   1, meta_ac, dealloc);
2105         if (ret) {
2106                 mlog_errno(ret);
2107                 goto out;
2108         }
2109
2110         ocfs2_remove_from_cache(ci, ref_leaf_bh);
2111
2112         /*
2113          * add the freed block to the dealloc so that it will be freed
2114          * when we run dealloc.
2115          */
2116         ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
2117                                         le16_to_cpu(rb->rf_suballoc_slot),
2118                                         le64_to_cpu(rb->rf_suballoc_loc),
2119                                         le64_to_cpu(rb->rf_blkno),
2120                                         le16_to_cpu(rb->rf_suballoc_bit));
2121         if (ret) {
2122                 mlog_errno(ret);
2123                 goto out;
2124         }
2125
2126         ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
2127                                       OCFS2_JOURNAL_ACCESS_WRITE);
2128         if (ret) {
2129                 mlog_errno(ret);
2130                 goto out;
2131         }
2132
2133         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2134
2135         le32_add_cpu(&rb->rf_clusters, -1);
2136
2137         /*
2138          * check whether we need to restore the root refcount block if
2139          * there is no leaf extent block at atll.
2140          */
2141         if (!rb->rf_list.l_next_free_rec) {
2142                 BUG_ON(rb->rf_clusters);
2143
2144                 trace_ocfs2_restore_refcount_block(
2145                      (unsigned long long)ref_root_bh->b_blocknr);
2146
2147                 rb->rf_flags = 0;
2148                 rb->rf_parent = 0;
2149                 rb->rf_cpos = 0;
2150                 memset(&rb->rf_records, 0, sb->s_blocksize -
2151                        offsetof(struct ocfs2_refcount_block, rf_records));
2152                 rb->rf_records.rl_count =
2153                                 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
2154         }
2155
2156         ocfs2_journal_dirty(handle, ref_root_bh);
2157
2158 out:
2159         return ret;
2160 }
2161
2162 int ocfs2_increase_refcount(handle_t *handle,
2163                             struct ocfs2_caching_info *ci,
2164                             struct buffer_head *ref_root_bh,
2165                             u64 cpos, u32 len,
2166                             struct ocfs2_alloc_context *meta_ac,
2167                             struct ocfs2_cached_dealloc_ctxt *dealloc)
2168 {
2169         return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
2170                                          cpos, len, 1,
2171                                          meta_ac, dealloc);
2172 }
2173
2174 static int ocfs2_decrease_refcount_rec(handle_t *handle,
2175                                 struct ocfs2_caching_info *ci,
2176                                 struct buffer_head *ref_root_bh,
2177                                 struct buffer_head *ref_leaf_bh,
2178                                 int index, u64 cpos, unsigned int len,
2179                                 struct ocfs2_alloc_context *meta_ac,
2180                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
2181 {
2182         int ret;
2183         struct ocfs2_refcount_block *rb =
2184                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2185         struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
2186
2187         BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
2188         BUG_ON(cpos + len >
2189                le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
2190
2191         trace_ocfs2_decrease_refcount_rec(
2192                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2193                 (unsigned long long)cpos, len);
2194
2195         if (cpos == le64_to_cpu(rec->r_cpos) &&
2196             len == le32_to_cpu(rec->r_clusters))
2197                 ret = ocfs2_change_refcount_rec(handle, ci,
2198                                                 ref_leaf_bh, index, 1, -1);
2199         else {
2200                 struct ocfs2_refcount_rec split = *rec;
2201                 split.r_cpos = cpu_to_le64(cpos);
2202                 split.r_clusters = cpu_to_le32(len);
2203
2204                 le32_add_cpu(&split.r_refcount, -1);
2205
2206                 ret = ocfs2_split_refcount_rec(handle, ci,
2207                                                ref_root_bh, ref_leaf_bh,
2208                                                &split, index, 1,
2209                                                meta_ac, dealloc);
2210         }
2211
2212         if (ret) {
2213                 mlog_errno(ret);
2214                 goto out;
2215         }
2216
2217         /* Remove the leaf refcount block if it contains no refcount record. */
2218         if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
2219                 ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
2220                                                    ref_leaf_bh, meta_ac,
2221                                                    dealloc);
2222                 if (ret)
2223                         mlog_errno(ret);
2224         }
2225
2226 out:
2227         return ret;
2228 }
2229
2230 static int __ocfs2_decrease_refcount(handle_t *handle,
2231                                      struct ocfs2_caching_info *ci,
2232                                      struct buffer_head *ref_root_bh,
2233                                      u64 cpos, u32 len,
2234                                      struct ocfs2_alloc_context *meta_ac,
2235                                      struct ocfs2_cached_dealloc_ctxt *dealloc,
2236                                      int delete)
2237 {
2238         int ret = 0, index = 0;
2239         struct ocfs2_refcount_rec rec;
2240         unsigned int r_count = 0, r_len;
2241         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2242         struct buffer_head *ref_leaf_bh = NULL;
2243
2244         trace_ocfs2_decrease_refcount(
2245                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2246                 (unsigned long long)cpos, len, delete);
2247
2248         while (len) {
2249                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2250                                              cpos, len, &rec, &index,
2251                                              &ref_leaf_bh);
2252                 if (ret) {
2253                         mlog_errno(ret);
2254                         goto out;
2255                 }
2256
2257                 r_count = le32_to_cpu(rec.r_refcount);
2258                 BUG_ON(r_count == 0);
2259                 if (!delete)
2260                         BUG_ON(r_count > 1);
2261
2262                 r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
2263                               le32_to_cpu(rec.r_clusters)) - cpos;
2264
2265                 ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
2266                                                   ref_leaf_bh, index,
2267                                                   cpos, r_len,
2268                                                   meta_ac, dealloc);
2269                 if (ret) {
2270                         mlog_errno(ret);
2271                         goto out;
2272                 }
2273
2274                 if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
2275                         ret = ocfs2_cache_cluster_dealloc(dealloc,
2276                                           ocfs2_clusters_to_blocks(sb, cpos),
2277                                                           r_len);
2278                         if (ret) {
2279                                 mlog_errno(ret);
2280                                 goto out;
2281                         }
2282                 }
2283
2284                 cpos += r_len;
2285                 len -= r_len;
2286                 brelse(ref_leaf_bh);
2287                 ref_leaf_bh = NULL;
2288         }
2289
2290 out:
2291         brelse(ref_leaf_bh);
2292         return ret;
2293 }
2294
2295 /* Caller must hold refcount tree lock. */
2296 int ocfs2_decrease_refcount(struct inode *inode,
2297                             handle_t *handle, u32 cpos, u32 len,
2298                             struct ocfs2_alloc_context *meta_ac,
2299                             struct ocfs2_cached_dealloc_ctxt *dealloc,
2300                             int delete)
2301 {
2302         int ret;
2303         u64 ref_blkno;
2304         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2305         struct buffer_head *ref_root_bh = NULL;
2306         struct ocfs2_refcount_tree *tree;
2307
2308         BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2309
2310         ret = ocfs2_get_refcount_block(inode, &ref_blkno);
2311         if (ret) {
2312                 mlog_errno(ret);
2313                 goto out;
2314         }
2315
2316         ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
2317         if (ret) {
2318                 mlog_errno(ret);
2319                 goto out;
2320         }
2321
2322         ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
2323                                         &ref_root_bh);
2324         if (ret) {
2325                 mlog_errno(ret);
2326                 goto out;
2327         }
2328
2329         ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
2330                                         cpos, len, meta_ac, dealloc, delete);
2331         if (ret)
2332                 mlog_errno(ret);
2333 out:
2334         brelse(ref_root_bh);
2335         return ret;
2336 }
2337
2338 /*
2339  * Mark the already-existing extent at cpos as refcounted for len clusters.
2340  * This adds the refcount extent flag.
2341  *
2342  * If the existing extent is larger than the request, initiate a
2343  * split. An attempt will be made at merging with adjacent extents.
2344  *
2345  * The caller is responsible for passing down meta_ac if we'll need it.
2346  */
2347 static int ocfs2_mark_extent_refcounted(struct inode *inode,
2348                                 struct ocfs2_extent_tree *et,
2349                                 handle_t *handle, u32 cpos,
2350                                 u32 len, u32 phys,
2351                                 struct ocfs2_alloc_context *meta_ac,
2352                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
2353 {
2354         int ret;
2355
2356         trace_ocfs2_mark_extent_refcounted(OCFS2_I(inode)->ip_blkno,
2357                                            cpos, len, phys);
2358
2359         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2360                 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2361                             "tree, but the feature bit is not set in the "
2362                             "super block.", inode->i_ino);
2363                 ret = -EROFS;
2364                 goto out;
2365         }
2366
2367         ret = ocfs2_change_extent_flag(handle, et, cpos,
2368                                        len, phys, meta_ac, dealloc,
2369                                        OCFS2_EXT_REFCOUNTED, 0);
2370         if (ret)
2371                 mlog_errno(ret);
2372
2373 out:
2374         return ret;
2375 }
2376
2377 /*
2378  * Given some contiguous physical clusters, calculate what we need
2379  * for modifying their refcount.
2380  */
2381 static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
2382                                             struct ocfs2_caching_info *ci,
2383                                             struct buffer_head *ref_root_bh,
2384                                             u64 start_cpos,
2385                                             u32 clusters,
2386                                             int *meta_add,
2387                                             int *credits)
2388 {
2389         int ret = 0, index, ref_blocks = 0, recs_add = 0;
2390         u64 cpos = start_cpos;
2391         struct ocfs2_refcount_block *rb;
2392         struct ocfs2_refcount_rec rec;
2393         struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
2394         u32 len;
2395
2396         while (clusters) {
2397                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2398                                              cpos, clusters, &rec,
2399                                              &index, &ref_leaf_bh);
2400                 if (ret) {
2401                         mlog_errno(ret);
2402                         goto out;
2403                 }
2404
2405                 if (ref_leaf_bh != prev_bh) {
2406                         /*
2407                          * Now we encounter a new leaf block, so calculate
2408                          * whether we need to extend the old leaf.
2409                          */
2410                         if (prev_bh) {
2411                                 rb = (struct ocfs2_refcount_block *)
2412                                                         prev_bh->b_data;
2413
2414                                 if (le16_to_cpu(rb->rf_records.rl_used) +
2415                                     recs_add >
2416                                     le16_to_cpu(rb->rf_records.rl_count))
2417                                         ref_blocks++;
2418                         }
2419
2420                         recs_add = 0;
2421                         *credits += 1;
2422                         brelse(prev_bh);
2423                         prev_bh = ref_leaf_bh;
2424                         get_bh(prev_bh);
2425                 }
2426
2427                 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
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                                                       ref_blocks);
2508         } else {
2509                 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
2510                 *meta_add += 1;
2511         }
2512
2513 out:
2514
2515         trace_ocfs2_calc_refcount_meta_credits(
2516                 (unsigned long long)start_cpos, clusters,
2517                 *meta_add, *credits);
2518         brelse(ref_leaf_bh);
2519         brelse(prev_bh);
2520         return ret;
2521 }
2522
2523 /*
2524  * For refcount tree, we will decrease some contiguous clusters
2525  * refcount count, so just go through it to see how many blocks
2526  * we gonna touch and whether we need to create new blocks.
2527  *
2528  * Normally the refcount blocks store these refcount should be
2529  * contiguous also, so that we can get the number easily.
2530  * We will at most add split 2 refcount records and 2 more
2531  * refcount blocks, so just check it in a rough way.
2532  *
2533  * Caller must hold refcount tree lock.
2534  */
2535 int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
2536                                           u64 refcount_loc,
2537                                           u64 phys_blkno,
2538                                           u32 clusters,
2539                                           int *credits,
2540                                           int *ref_blocks)
2541 {
2542         int ret;
2543         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2544         struct buffer_head *ref_root_bh = NULL;
2545         struct ocfs2_refcount_tree *tree;
2546         u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
2547
2548         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2549                 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2550                             "tree, but the feature bit is not set in the "
2551                             "super block.", inode->i_ino);
2552                 ret = -EROFS;
2553                 goto out;
2554         }
2555
2556         BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2557
2558         ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
2559                                       refcount_loc, &tree);
2560         if (ret) {
2561                 mlog_errno(ret);
2562                 goto out;
2563         }
2564
2565         ret = ocfs2_read_refcount_block(&tree->rf_ci, refcount_loc,
2566                                         &ref_root_bh);
2567         if (ret) {
2568                 mlog_errno(ret);
2569                 goto out;
2570         }
2571
2572         ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
2573                                                &tree->rf_ci,
2574                                                ref_root_bh,
2575                                                start_cpos, clusters,
2576                                                ref_blocks, credits);
2577         if (ret) {
2578                 mlog_errno(ret);
2579                 goto out;
2580         }
2581
2582         trace_ocfs2_prepare_refcount_change_for_del(*ref_blocks, *credits);
2583
2584 out:
2585         brelse(ref_root_bh);
2586         return ret;
2587 }
2588
2589 #define MAX_CONTIG_BYTES        1048576
2590
2591 static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
2592 {
2593         return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
2594 }
2595
2596 static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
2597 {
2598         return ~(ocfs2_cow_contig_clusters(sb) - 1);
2599 }
2600
2601 /*
2602  * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2603  * find an offset (start + (n * contig_clusters)) that is closest to cpos
2604  * while still being less than or equal to it.
2605  *
2606  * The goal is to break the extent at a multiple of contig_clusters.
2607  */
2608 static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
2609                                                  unsigned int start,
2610                                                  unsigned int cpos)
2611 {
2612         BUG_ON(start > cpos);
2613
2614         return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
2615 }
2616
2617 /*
2618  * Given a cluster count of len, pad it out so that it is a multiple
2619  * of contig_clusters.
2620  */
2621 static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
2622                                                   unsigned int len)
2623 {
2624         unsigned int padded =
2625                 (len + (ocfs2_cow_contig_clusters(sb) - 1)) &
2626                 ocfs2_cow_contig_mask(sb);
2627
2628         /* Did we wrap? */
2629         if (padded < len)
2630                 padded = UINT_MAX;
2631
2632         return padded;
2633 }
2634
2635 /*
2636  * Calculate out the start and number of virtual clusters we need to to CoW.
2637  *
2638  * cpos is vitual start cluster position we want to do CoW in a
2639  * file and write_len is the cluster length.
2640  * max_cpos is the place where we want to stop CoW intentionally.
2641  *
2642  * Normal we will start CoW from the beginning of extent record cotaining cpos.
2643  * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2644  * get good I/O from the resulting extent tree.
2645  */
2646 static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
2647                                            struct ocfs2_extent_list *el,
2648                                            u32 cpos,
2649                                            u32 write_len,
2650                                            u32 max_cpos,
2651                                            u32 *cow_start,
2652                                            u32 *cow_len)
2653 {
2654         int ret = 0;
2655         int tree_height = le16_to_cpu(el->l_tree_depth), i;
2656         struct buffer_head *eb_bh = NULL;
2657         struct ocfs2_extent_block *eb = NULL;
2658         struct ocfs2_extent_rec *rec;
2659         unsigned int want_clusters, rec_end = 0;
2660         int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
2661         int leaf_clusters;
2662
2663         BUG_ON(cpos + write_len > max_cpos);
2664
2665         if (tree_height > 0) {
2666                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
2667                 if (ret) {
2668                         mlog_errno(ret);
2669                         goto out;
2670                 }
2671
2672                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2673                 el = &eb->h_list;
2674
2675                 if (el->l_tree_depth) {
2676                         ocfs2_error(inode->i_sb,
2677                                     "Inode %lu has non zero tree depth in "
2678                                     "leaf block %llu\n", inode->i_ino,
2679                                     (unsigned long long)eb_bh->b_blocknr);
2680                         ret = -EROFS;
2681                         goto out;
2682                 }
2683         }
2684
2685         *cow_len = 0;
2686         for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
2687                 rec = &el->l_recs[i];
2688
2689                 if (ocfs2_is_empty_extent(rec)) {
2690                         mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
2691                                         "index %d\n", inode->i_ino, i);
2692                         continue;
2693                 }
2694
2695                 if (le32_to_cpu(rec->e_cpos) +
2696                     le16_to_cpu(rec->e_leaf_clusters) <= cpos)
2697                         continue;
2698
2699                 if (*cow_len == 0) {
2700                         /*
2701                          * We should find a refcounted record in the
2702                          * first pass.
2703                          */
2704                         BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
2705                         *cow_start = le32_to_cpu(rec->e_cpos);
2706                 }
2707
2708                 /*
2709                  * If we encounter a hole, a non-refcounted record or
2710                  * pass the max_cpos, stop the search.
2711                  */
2712                 if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
2713                     (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
2714                     (max_cpos <= le32_to_cpu(rec->e_cpos)))
2715                         break;
2716
2717                 leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
2718                 rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
2719                 if (rec_end > max_cpos) {
2720                         rec_end = max_cpos;
2721                         leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
2722                 }
2723
2724                 /*
2725                  * How many clusters do we actually need from
2726                  * this extent?  First we see how many we actually
2727                  * need to complete the write.  If that's smaller
2728                  * than contig_clusters, we try for contig_clusters.
2729                  */
2730                 if (!*cow_len)
2731                         want_clusters = write_len;
2732                 else
2733                         want_clusters = (cpos + write_len) -
2734                                 (*cow_start + *cow_len);
2735                 if (want_clusters < contig_clusters)
2736                         want_clusters = contig_clusters;
2737
2738                 /*
2739                  * If the write does not cover the whole extent, we
2740                  * need to calculate how we're going to split the extent.
2741                  * We try to do it on contig_clusters boundaries.
2742                  *
2743                  * Any extent smaller than contig_clusters will be
2744                  * CoWed in its entirety.
2745                  */
2746                 if (leaf_clusters <= contig_clusters)
2747                         *cow_len += leaf_clusters;
2748                 else if (*cow_len || (*cow_start == cpos)) {
2749                         /*
2750                          * This extent needs to be CoW'd from its
2751                          * beginning, so all we have to do is compute
2752                          * how many clusters to grab.  We align
2753                          * want_clusters to the edge of contig_clusters
2754                          * to get better I/O.
2755                          */
2756                         want_clusters = ocfs2_cow_align_length(inode->i_sb,
2757                                                                want_clusters);
2758
2759                         if (leaf_clusters < want_clusters)
2760                                 *cow_len += leaf_clusters;
2761                         else
2762                                 *cow_len += want_clusters;
2763                 } else if ((*cow_start + contig_clusters) >=
2764                            (cpos + write_len)) {
2765                         /*
2766                          * Breaking off contig_clusters at the front
2767                          * of the extent will cover our write.  That's
2768                          * easy.
2769                          */
2770                         *cow_len = contig_clusters;
2771                 } else if ((rec_end - cpos) <= contig_clusters) {
2772                         /*
2773                          * Breaking off contig_clusters at the tail of
2774                          * this extent will cover cpos.
2775                          */
2776                         *cow_start = rec_end - contig_clusters;
2777                         *cow_len = contig_clusters;
2778                 } else if ((rec_end - cpos) <= want_clusters) {
2779                         /*
2780                          * While we can't fit the entire write in this
2781                          * extent, we know that the write goes from cpos
2782                          * to the end of the extent.  Break that off.
2783                          * We try to break it at some multiple of
2784                          * contig_clusters from the front of the extent.
2785                          * Failing that (ie, cpos is within
2786                          * contig_clusters of the front), we'll CoW the
2787                          * entire extent.
2788                          */
2789                         *cow_start = ocfs2_cow_align_start(inode->i_sb,
2790                                                            *cow_start, cpos);
2791                         *cow_len = rec_end - *cow_start;
2792                 } else {
2793                         /*
2794                          * Ok, the entire write lives in the middle of
2795                          * this extent.  Let's try to slice the extent up
2796                          * nicely.  Optimally, our CoW region starts at
2797                          * m*contig_clusters from the beginning of the
2798                          * extent and goes for n*contig_clusters,
2799                          * covering the entire write.
2800                          */
2801                         *cow_start = ocfs2_cow_align_start(inode->i_sb,
2802                                                            *cow_start, cpos);
2803
2804                         want_clusters = (cpos + write_len) - *cow_start;
2805                         want_clusters = ocfs2_cow_align_length(inode->i_sb,
2806                                                                want_clusters);
2807                         if (*cow_start + want_clusters <= rec_end)
2808                                 *cow_len = want_clusters;
2809                         else
2810                                 *cow_len = rec_end - *cow_start;
2811                 }
2812
2813                 /* Have we covered our entire write yet? */
2814                 if ((*cow_start + *cow_len) >= (cpos + write_len))
2815                         break;
2816
2817                 /*
2818                  * If we reach the end of the extent block and don't get enough
2819                  * clusters, continue with the next extent block if possible.
2820                  */
2821                 if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
2822                     eb && eb->h_next_leaf_blk) {
2823                         brelse(eb_bh);
2824                         eb_bh = NULL;
2825
2826                         ret = ocfs2_read_extent_block(INODE_CACHE(inode),
2827                                                le64_to_cpu(eb->h_next_leaf_blk),
2828                                                &eb_bh);
2829                         if (ret) {
2830                                 mlog_errno(ret);
2831                                 goto out;
2832                         }
2833
2834                         eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2835                         el = &eb->h_list;
2836                         i = -1;
2837                 }
2838         }
2839
2840 out:
2841         brelse(eb_bh);
2842         return ret;
2843 }
2844
2845 /*
2846  * Prepare meta_ac, data_ac and calculate credits when we want to add some
2847  * num_clusters in data_tree "et" and change the refcount for the old
2848  * clusters(starting form p_cluster) in the refcount tree.
2849  *
2850  * Note:
2851  * 1. since we may split the old tree, so we at most will need num_clusters + 2
2852  *    more new leaf records.
2853  * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2854  *    just give data_ac = NULL.
2855  */
2856 static int ocfs2_lock_refcount_allocators(struct super_block *sb,
2857                                         u32 p_cluster, u32 num_clusters,
2858                                         struct ocfs2_extent_tree *et,
2859                                         struct ocfs2_caching_info *ref_ci,
2860                                         struct buffer_head *ref_root_bh,
2861                                         struct ocfs2_alloc_context **meta_ac,
2862                                         struct ocfs2_alloc_context **data_ac,
2863                                         int *credits)
2864 {
2865         int ret = 0, meta_add = 0;
2866         int num_free_extents = ocfs2_num_free_extents(OCFS2_SB(sb), et);
2867
2868         if (num_free_extents < 0) {
2869                 ret = num_free_extents;
2870                 mlog_errno(ret);
2871                 goto out;
2872         }
2873
2874         if (num_free_extents < num_clusters + 2)
2875                 meta_add =
2876                         ocfs2_extend_meta_needed(et->et_root_el);
2877
2878         *credits += ocfs2_calc_extend_credits(sb, et->et_root_el,
2879                                               num_clusters + 2);
2880
2881         ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
2882                                                p_cluster, num_clusters,
2883                                                &meta_add, credits);
2884         if (ret) {
2885                 mlog_errno(ret);
2886                 goto out;
2887         }
2888
2889         trace_ocfs2_lock_refcount_allocators(meta_add, *credits);
2890         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
2891                                                 meta_ac);
2892         if (ret) {
2893                 mlog_errno(ret);
2894                 goto out;
2895         }
2896
2897         if (data_ac) {
2898                 ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
2899                                              data_ac);
2900                 if (ret)
2901                         mlog_errno(ret);
2902         }
2903
2904 out:
2905         if (ret) {
2906                 if (*meta_ac) {
2907                         ocfs2_free_alloc_context(*meta_ac);
2908                         *meta_ac = NULL;
2909                 }
2910         }
2911
2912         return ret;
2913 }
2914
2915 static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
2916 {
2917         BUG_ON(buffer_dirty(bh));
2918
2919         clear_buffer_mapped(bh);
2920
2921         return 0;
2922 }
2923
2924 int ocfs2_duplicate_clusters_by_page(handle_t *handle,
2925                                      struct file *file,
2926                                      u32 cpos, u32 old_cluster,
2927                                      u32 new_cluster, u32 new_len)
2928 {
2929         int ret = 0, partial;
2930         struct inode *inode = file->f_path.dentry->d_inode;
2931         struct ocfs2_caching_info *ci = INODE_CACHE(inode);
2932         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2933         u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
2934         struct page *page;
2935         pgoff_t page_index;
2936         unsigned int from, to, readahead_pages;
2937         loff_t offset, end, map_end;
2938         struct address_space *mapping = inode->i_mapping;
2939
2940         trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
2941                                                new_cluster, new_len);
2942
2943         readahead_pages =
2944                 (ocfs2_cow_contig_clusters(sb) <<
2945                  OCFS2_SB(sb)->s_clustersize_bits) >> PAGE_CACHE_SHIFT;
2946         offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
2947         end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
2948         /*
2949          * We only duplicate pages until we reach the page contains i_size - 1.
2950          * So trim 'end' to i_size.
2951          */
2952         if (end > i_size_read(inode))
2953                 end = i_size_read(inode);
2954
2955         while (offset < end) {
2956                 page_index = offset >> PAGE_CACHE_SHIFT;
2957                 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
2958                 if (map_end > end)
2959                         map_end = end;
2960
2961                 /* from, to is the offset within the page. */
2962                 from = offset & (PAGE_CACHE_SIZE - 1);
2963                 to = PAGE_CACHE_SIZE;
2964                 if (map_end & (PAGE_CACHE_SIZE - 1))
2965                         to = map_end & (PAGE_CACHE_SIZE - 1);
2966
2967                 page = find_or_create_page(mapping, page_index, GFP_NOFS);
2968
2969                 /*
2970                  * In case PAGE_CACHE_SIZE <= CLUSTER_SIZE, This page
2971                  * can't be dirtied before we CoW it out.
2972                  */
2973                 if (PAGE_CACHE_SIZE <= OCFS2_SB(sb)->s_clustersize)
2974                         BUG_ON(PageDirty(page));
2975
2976                 if (PageReadahead(page)) {
2977                         page_cache_async_readahead(mapping,
2978                                                    &file->f_ra, file,
2979                                                    page, page_index,
2980                                                    readahead_pages);
2981                 }
2982
2983                 if (!PageUptodate(page)) {
2984                         ret = block_read_full_page(page, ocfs2_get_block);
2985                         if (ret) {
2986                                 mlog_errno(ret);
2987                                 goto unlock;
2988                         }
2989                         lock_page(page);
2990                 }
2991
2992                 if (page_has_buffers(page)) {
2993                         ret = walk_page_buffers(handle, page_buffers(page),
2994                                                 from, to, &partial,
2995                                                 ocfs2_clear_cow_buffer);
2996                         if (ret) {
2997                                 mlog_errno(ret);
2998                                 goto unlock;
2999                         }
3000                 }
3001
3002                 ocfs2_map_and_dirty_page(inode, handle, from, to,
3003                                          page, 0, &new_block);
3004                 mark_page_accessed(page);
3005 unlock:
3006                 unlock_page(page);
3007                 page_cache_release(page);
3008                 page = NULL;
3009                 offset = map_end;
3010                 if (ret)
3011                         break;
3012         }
3013
3014         return ret;
3015 }
3016
3017 int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
3018                                     struct file *file,
3019                                     u32 cpos, u32 old_cluster,
3020                                     u32 new_cluster, u32 new_len)
3021 {
3022         int ret = 0;
3023         struct inode *inode = file->f_path.dentry->d_inode;
3024         struct super_block *sb = inode->i_sb;
3025         struct ocfs2_caching_info *ci = INODE_CACHE(inode);
3026         int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
3027         u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
3028         u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
3029         struct ocfs2_super *osb = OCFS2_SB(sb);
3030         struct buffer_head *old_bh = NULL;
3031         struct buffer_head *new_bh = NULL;
3032
3033         trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
3034                                                new_cluster, new_len);
3035
3036         for (i = 0; i < blocks; i++, old_block++, new_block++) {
3037                 new_bh = sb_getblk(osb->sb, new_block);
3038                 if (new_bh == NULL) {
3039                         ret = -EIO;
3040                         mlog_errno(ret);
3041                         break;
3042                 }
3043
3044                 ocfs2_set_new_buffer_uptodate(ci, new_bh);
3045
3046                 ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
3047                 if (ret) {
3048                         mlog_errno(ret);
3049                         break;
3050                 }
3051
3052                 ret = ocfs2_journal_access(handle, ci, new_bh,
3053                                            OCFS2_JOURNAL_ACCESS_CREATE);
3054                 if (ret) {
3055                         mlog_errno(ret);
3056                         break;
3057                 }
3058
3059                 memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
3060                 ocfs2_journal_dirty(handle, new_bh);
3061
3062                 brelse(new_bh);
3063                 brelse(old_bh);
3064                 new_bh = NULL;
3065                 old_bh = NULL;
3066         }
3067
3068         brelse(new_bh);
3069         brelse(old_bh);
3070         return ret;
3071 }
3072
3073 static int ocfs2_clear_ext_refcount(handle_t *handle,
3074                                     struct ocfs2_extent_tree *et,
3075                                     u32 cpos, u32 p_cluster, u32 len,
3076                                     unsigned int ext_flags,
3077                                     struct ocfs2_alloc_context *meta_ac,
3078                                     struct ocfs2_cached_dealloc_ctxt *dealloc)
3079 {
3080         int ret, index;
3081         struct ocfs2_extent_rec replace_rec;
3082         struct ocfs2_path *path = NULL;
3083         struct ocfs2_extent_list *el;
3084         struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
3085         u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
3086
3087         trace_ocfs2_clear_ext_refcount((unsigned long long)ino,
3088                                        cpos, len, p_cluster, ext_flags);
3089
3090         memset(&replace_rec, 0, sizeof(replace_rec));
3091         replace_rec.e_cpos = cpu_to_le32(cpos);
3092         replace_rec.e_leaf_clusters = cpu_to_le16(len);
3093         replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
3094                                                                    p_cluster));
3095         replace_rec.e_flags = ext_flags;
3096         replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
3097
3098         path = ocfs2_new_path_from_et(et);
3099         if (!path) {
3100                 ret = -ENOMEM;
3101                 mlog_errno(ret);
3102                 goto out;
3103         }
3104
3105         ret = ocfs2_find_path(et->et_ci, path, cpos);
3106         if (ret) {
3107                 mlog_errno(ret);
3108                 goto out;
3109         }
3110
3111         el = path_leaf_el(path);
3112
3113         index = ocfs2_search_extent_list(el, cpos);
3114         if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
3115                 ocfs2_error(sb,
3116                             "Inode %llu has an extent at cpos %u which can no "
3117                             "longer be found.\n",
3118                             (unsigned long long)ino, cpos);
3119                 ret = -EROFS;
3120                 goto out;
3121         }
3122
3123         ret = ocfs2_split_extent(handle, et, path, index,
3124                                  &replace_rec, meta_ac, dealloc);
3125         if (ret)
3126                 mlog_errno(ret);
3127
3128 out:
3129         ocfs2_free_path(path);
3130         return ret;
3131 }
3132
3133 static int ocfs2_replace_clusters(handle_t *handle,
3134                                   struct ocfs2_cow_context *context,
3135                                   u32 cpos, u32 old,
3136                                   u32 new, u32 len,
3137                                   unsigned int ext_flags)
3138 {
3139         int ret;
3140         struct ocfs2_caching_info *ci = context->data_et.et_ci;
3141         u64 ino = ocfs2_metadata_cache_owner(ci);
3142
3143         trace_ocfs2_replace_clusters((unsigned long long)ino,
3144                                      cpos, old, new, len, ext_flags);
3145
3146         /*If the old clusters is unwritten, no need to duplicate. */
3147         if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
3148                 ret = context->cow_duplicate_clusters(handle, context->file,
3149                                                       cpos, old, new, len);
3150                 if (ret) {
3151                         mlog_errno(ret);
3152                         goto out;
3153                 }
3154         }
3155
3156         ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
3157                                        cpos, new, len, ext_flags,
3158                                        context->meta_ac, &context->dealloc);
3159         if (ret)
3160                 mlog_errno(ret);
3161 out:
3162         return ret;
3163 }
3164
3165 int ocfs2_cow_sync_writeback(struct super_block *sb,
3166                              struct inode *inode,
3167                              u32 cpos, u32 num_clusters)
3168 {
3169         int ret = 0;
3170         loff_t offset, end, map_end;
3171         pgoff_t page_index;
3172         struct page *page;
3173
3174         if (ocfs2_should_order_data(inode))
3175                 return 0;
3176
3177         offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
3178         end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
3179
3180         ret = filemap_fdatawrite_range(inode->i_mapping,
3181                                        offset, end - 1);
3182         if (ret < 0) {
3183                 mlog_errno(ret);
3184                 return ret;
3185         }
3186
3187         while (offset < end) {
3188                 page_index = offset >> PAGE_CACHE_SHIFT;
3189                 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
3190                 if (map_end > end)
3191                         map_end = end;
3192
3193                 page = find_or_create_page(inode->i_mapping,
3194                                            page_index, GFP_NOFS);
3195                 BUG_ON(!page);
3196
3197                 wait_on_page_writeback(page);
3198                 if (PageError(page)) {
3199                         ret = -EIO;
3200                         mlog_errno(ret);
3201                 } else
3202                         mark_page_accessed(page);
3203
3204                 unlock_page(page);
3205                 page_cache_release(page);
3206                 page = NULL;
3207                 offset = map_end;
3208                 if (ret)
3209                         break;
3210         }
3211
3212         return ret;
3213 }
3214
3215 static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
3216                                  u32 v_cluster, u32 *p_cluster,
3217                                  u32 *num_clusters,
3218                                  unsigned int *extent_flags)
3219 {
3220         return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
3221                                   num_clusters, extent_flags);
3222 }
3223
3224 static int ocfs2_make_clusters_writable(struct super_block *sb,
3225                                         struct ocfs2_cow_context *context,
3226                                         u32 cpos, u32 p_cluster,
3227                                         u32 num_clusters, unsigned int e_flags)
3228 {
3229         int ret, delete, index, credits =  0;
3230         u32 new_bit, new_len, orig_num_clusters;
3231         unsigned int set_len;
3232         struct ocfs2_super *osb = OCFS2_SB(sb);
3233         handle_t *handle;
3234         struct buffer_head *ref_leaf_bh = NULL;
3235         struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
3236         struct ocfs2_refcount_rec rec;
3237
3238         trace_ocfs2_make_clusters_writable(cpos, p_cluster,
3239                                            num_clusters, e_flags);
3240
3241         ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
3242                                              &context->data_et,
3243                                              ref_ci,
3244                                              context->ref_root_bh,
3245                                              &context->meta_ac,
3246                                              &context->data_ac, &credits);
3247         if (ret) {
3248                 mlog_errno(ret);
3249                 return ret;
3250         }
3251
3252         if (context->post_refcount)
3253                 credits += context->post_refcount->credits;
3254
3255         credits += context->extra_credits;
3256         handle = ocfs2_start_trans(osb, credits);
3257         if (IS_ERR(handle)) {
3258                 ret = PTR_ERR(handle);
3259                 mlog_errno(ret);
3260                 goto out;
3261         }
3262
3263         orig_num_clusters = num_clusters;
3264
3265         while (num_clusters) {
3266                 ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
3267                                              p_cluster, num_clusters,
3268                                              &rec, &index, &ref_leaf_bh);
3269                 if (ret) {
3270                         mlog_errno(ret);
3271                         goto out_commit;
3272                 }
3273
3274                 BUG_ON(!rec.r_refcount);
3275                 set_len = min((u64)p_cluster + num_clusters,
3276                               le64_to_cpu(rec.r_cpos) +
3277                               le32_to_cpu(rec.r_clusters)) - p_cluster;
3278
3279                 /*
3280                  * There are many different situation here.
3281                  * 1. If refcount == 1, remove the flag and don't COW.
3282                  * 2. If refcount > 1, allocate clusters.
3283                  *    Here we may not allocate r_len once at a time, so continue
3284                  *    until we reach num_clusters.
3285                  */
3286                 if (le32_to_cpu(rec.r_refcount) == 1) {
3287                         delete = 0;
3288                         ret = ocfs2_clear_ext_refcount(handle,
3289                                                        &context->data_et,
3290                                                        cpos, p_cluster,
3291                                                        set_len, e_flags,
3292                                                        context->meta_ac,
3293                                                        &context->dealloc);
3294                         if (ret) {
3295                                 mlog_errno(ret);
3296                                 goto out_commit;
3297                         }
3298                 } else {
3299                         delete = 1;
3300
3301                         ret = __ocfs2_claim_clusters(handle,
3302                                                      context->data_ac,
3303                                                      1, set_len,
3304                                                      &new_bit, &new_len);
3305                         if (ret) {
3306                                 mlog_errno(ret);
3307                                 goto out_commit;
3308                         }
3309
3310                         ret = ocfs2_replace_clusters(handle, context,
3311                                                      cpos, p_cluster, new_bit,
3312                                                      new_len, e_flags);
3313                         if (ret) {
3314                                 mlog_errno(ret);
3315                                 goto out_commit;
3316                         }
3317                         set_len = new_len;
3318                 }
3319
3320                 ret = __ocfs2_decrease_refcount(handle, ref_ci,
3321                                                 context->ref_root_bh,
3322                                                 p_cluster, set_len,
3323                                                 context->meta_ac,
3324                                                 &context->dealloc, delete);
3325                 if (ret) {
3326                         mlog_errno(ret);
3327                         goto out_commit;
3328                 }
3329
3330                 cpos += set_len;
3331                 p_cluster += set_len;
3332                 num_clusters -= set_len;
3333                 brelse(ref_leaf_bh);
3334                 ref_leaf_bh = NULL;
3335         }
3336
3337         /* handle any post_cow action. */
3338         if (context->post_refcount && context->post_refcount->func) {
3339                 ret = context->post_refcount->func(context->inode, handle,
3340                                                 context->post_refcount->para);
3341                 if (ret) {
3342                         mlog_errno(ret);
3343                         goto out_commit;
3344                 }
3345         }
3346
3347         /*
3348          * Here we should write the new page out first if we are
3349          * in write-back mode.
3350          */
3351         if (context->get_clusters == ocfs2_di_get_clusters) {
3352                 ret = ocfs2_cow_sync_writeback(sb, context->inode, cpos,
3353                                                orig_num_clusters);
3354                 if (ret)
3355                         mlog_errno(ret);
3356         }
3357
3358 out_commit:
3359         ocfs2_commit_trans(osb, handle);
3360
3361 out:
3362         if (context->data_ac) {
3363                 ocfs2_free_alloc_context(context->data_ac);
3364                 context->data_ac = NULL;
3365         }
3366         if (context->meta_ac) {
3367                 ocfs2_free_alloc_context(context->meta_ac);
3368                 context->meta_ac = NULL;
3369         }
3370         brelse(ref_leaf_bh);
3371
3372         return ret;
3373 }
3374
3375 static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
3376 {
3377         int ret = 0;
3378         struct inode *inode = context->inode;
3379         u32 cow_start = context->cow_start, cow_len = context->cow_len;
3380         u32 p_cluster, num_clusters;
3381         unsigned int ext_flags;
3382         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3383
3384         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
3385                 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
3386                             "tree, but the feature bit is not set in the "
3387                             "super block.", inode->i_ino);
3388                 return -EROFS;
3389         }
3390
3391         ocfs2_init_dealloc_ctxt(&context->dealloc);
3392
3393         while (cow_len) {
3394                 ret = context->get_clusters(context, cow_start, &p_cluster,
3395                                             &num_clusters, &ext_flags);
3396                 if (ret) {
3397                         mlog_errno(ret);
3398                         break;
3399                 }
3400
3401                 BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
3402
3403                 if (cow_len < num_clusters)
3404                         num_clusters = cow_len;
3405
3406                 ret = ocfs2_make_clusters_writable(inode->i_sb, context,
3407                                                    cow_start, p_cluster,
3408                                                    num_clusters, ext_flags);
3409                 if (ret) {
3410                         mlog_errno(ret);
3411                         break;
3412                 }
3413
3414                 cow_len -= num_clusters;
3415                 cow_start += num_clusters;
3416         }
3417
3418         if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
3419                 ocfs2_schedule_truncate_log_flush(osb, 1);
3420                 ocfs2_run_deallocs(osb, &context->dealloc);
3421         }
3422
3423         return ret;
3424 }
3425
3426 static void ocfs2_readahead_for_cow(struct inode *inode,
3427                                     struct file *file,
3428                                     u32 start, u32 len)
3429 {
3430         struct address_space *mapping;
3431         pgoff_t index;
3432         unsigned long num_pages;
3433         int cs_bits = OCFS2_SB(inode->i_sb)->s_clustersize_bits;
3434
3435         if (!file)
3436                 return;
3437
3438         mapping = file->f_mapping;
3439         num_pages = (len << cs_bits) >> PAGE_CACHE_SHIFT;
3440         if (!num_pages)
3441                 num_pages = 1;
3442
3443         index = ((loff_t)start << cs_bits) >> PAGE_CACHE_SHIFT;
3444         page_cache_sync_readahead(mapping, &file->f_ra, file,
3445                                   index, num_pages);
3446 }
3447
3448 /*
3449  * Starting at cpos, try to CoW write_len clusters.  Don't CoW
3450  * past max_cpos.  This will stop when it runs into a hole or an
3451  * unrefcounted extent.
3452  */
3453 static int ocfs2_refcount_cow_hunk(struct inode *inode,
3454                                    struct file *file,
3455                                    struct buffer_head *di_bh,
3456                                    u32 cpos, u32 write_len, u32 max_cpos)
3457 {
3458         int ret;
3459         u32 cow_start = 0, cow_len = 0;
3460         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3461         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3462         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3463         struct buffer_head *ref_root_bh = NULL;
3464         struct ocfs2_refcount_tree *ref_tree;
3465         struct ocfs2_cow_context *context = NULL;
3466
3467         BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3468
3469         ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
3470                                               cpos, write_len, max_cpos,
3471                                               &cow_start, &cow_len);
3472         if (ret) {
3473                 mlog_errno(ret);
3474                 goto out;
3475         }
3476
3477         trace_ocfs2_refcount_cow_hunk(OCFS2_I(inode)->ip_blkno,
3478                                       cpos, write_len, max_cpos,
3479                                       cow_start, cow_len);
3480
3481         BUG_ON(cow_len == 0);
3482
3483         ocfs2_readahead_for_cow(inode, file, cow_start, cow_len);
3484
3485         context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3486         if (!context) {
3487                 ret = -ENOMEM;
3488                 mlog_errno(ret);
3489                 goto out;
3490         }
3491
3492         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
3493                                        1, &ref_tree, &ref_root_bh);
3494         if (ret) {
3495                 mlog_errno(ret);
3496                 goto out;
3497         }
3498
3499         context->inode = inode;
3500         context->cow_start = cow_start;
3501         context->cow_len = cow_len;
3502         context->ref_tree = ref_tree;
3503         context->ref_root_bh = ref_root_bh;
3504         context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
3505         context->get_clusters = ocfs2_di_get_clusters;
3506         context->file = file;
3507
3508         ocfs2_init_dinode_extent_tree(&context->data_et,
3509                                       INODE_CACHE(inode), di_bh);
3510
3511         ret = ocfs2_replace_cow(context);
3512         if (ret)
3513                 mlog_errno(ret);
3514
3515         /*
3516          * truncate the extent map here since no matter whether we meet with
3517          * any error during the action, we shouldn't trust cached extent map
3518          * any more.
3519          */
3520         ocfs2_extent_map_trunc(inode, cow_start);
3521
3522         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3523         brelse(ref_root_bh);
3524 out:
3525         kfree(context);
3526         return ret;
3527 }
3528
3529 /*
3530  * CoW any and all clusters between cpos and cpos+write_len.
3531  * Don't CoW past max_cpos.  If this returns successfully, all
3532  * clusters between cpos and cpos+write_len are safe to modify.
3533  */
3534 int ocfs2_refcount_cow(struct inode *inode,
3535                        struct file *file,
3536                        struct buffer_head *di_bh,
3537                        u32 cpos, u32 write_len, u32 max_cpos)
3538 {
3539         int ret = 0;
3540         u32 p_cluster, num_clusters;
3541         unsigned int ext_flags;
3542
3543         while (write_len) {
3544                 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3545                                          &num_clusters, &ext_flags);
3546                 if (ret) {
3547                         mlog_errno(ret);
3548                         break;
3549                 }
3550
3551                 if (write_len < num_clusters)
3552                         num_clusters = write_len;
3553
3554                 if (ext_flags & OCFS2_EXT_REFCOUNTED) {
3555                         ret = ocfs2_refcount_cow_hunk(inode, file, di_bh, cpos,
3556                                                       num_clusters, max_cpos);
3557                         if (ret) {
3558                                 mlog_errno(ret);
3559                                 break;
3560                         }
3561                 }
3562
3563                 write_len -= num_clusters;
3564                 cpos += num_clusters;
3565         }
3566
3567         return ret;
3568 }
3569
3570 static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
3571                                           u32 v_cluster, u32 *p_cluster,
3572                                           u32 *num_clusters,
3573                                           unsigned int *extent_flags)
3574 {
3575         struct inode *inode = context->inode;
3576         struct ocfs2_xattr_value_root *xv = context->cow_object;
3577
3578         return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
3579                                         num_clusters, &xv->xr_list,
3580                                         extent_flags);
3581 }
3582
3583 /*
3584  * Given a xattr value root, calculate the most meta/credits we need for
3585  * refcount tree change if we truncate it to 0.
3586  */
3587 int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
3588                                        struct ocfs2_caching_info *ref_ci,
3589                                        struct buffer_head *ref_root_bh,
3590                                        struct ocfs2_xattr_value_root *xv,
3591                                        int *meta_add, int *credits)
3592 {
3593         int ret = 0, index, ref_blocks = 0;
3594         u32 p_cluster, num_clusters;
3595         u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
3596         struct ocfs2_refcount_block *rb;
3597         struct ocfs2_refcount_rec rec;
3598         struct buffer_head *ref_leaf_bh = NULL;
3599
3600         while (cpos < clusters) {
3601                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
3602                                                &num_clusters, &xv->xr_list,
3603                                                NULL);
3604                 if (ret) {
3605                         mlog_errno(ret);
3606                         goto out;
3607                 }
3608
3609                 cpos += num_clusters;
3610
3611                 while (num_clusters) {
3612                         ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
3613                                                      p_cluster, num_clusters,
3614                                                      &rec, &index,
3615                                                      &ref_leaf_bh);
3616                         if (ret) {
3617                                 mlog_errno(ret);
3618                                 goto out;
3619                         }
3620
3621                         BUG_ON(!rec.r_refcount);
3622
3623                         rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
3624
3625                         /*
3626                          * We really don't know whether the other clusters is in
3627                          * this refcount block or not, so just take the worst
3628                          * case that all the clusters are in this block and each
3629                          * one will split a refcount rec, so totally we need
3630                          * clusters * 2 new refcount rec.
3631                          */
3632                         if (le16_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
3633                             le16_to_cpu(rb->rf_records.rl_count))
3634                                 ref_blocks++;
3635
3636                         *credits += 1;
3637                         brelse(ref_leaf_bh);
3638                         ref_leaf_bh = NULL;
3639
3640                         if (num_clusters <= le32_to_cpu(rec.r_clusters))
3641                                 break;
3642                         else
3643                                 num_clusters -= le32_to_cpu(rec.r_clusters);
3644                         p_cluster += num_clusters;
3645                 }
3646         }
3647
3648         *meta_add += ref_blocks;
3649         if (!ref_blocks)
3650                 goto out;
3651
3652         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
3653         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
3654                 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
3655         else {
3656                 struct ocfs2_extent_tree et;
3657
3658                 ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
3659                 *credits += ocfs2_calc_extend_credits(inode->i_sb,
3660                                                       et.et_root_el,
3661                                                       ref_blocks);
3662         }
3663
3664 out:
3665         brelse(ref_leaf_bh);
3666         return ret;
3667 }
3668
3669 /*
3670  * Do CoW for xattr.
3671  */
3672 int ocfs2_refcount_cow_xattr(struct inode *inode,
3673                              struct ocfs2_dinode *di,
3674                              struct ocfs2_xattr_value_buf *vb,
3675                              struct ocfs2_refcount_tree *ref_tree,
3676                              struct buffer_head *ref_root_bh,
3677                              u32 cpos, u32 write_len,
3678                              struct ocfs2_post_refcount *post)
3679 {
3680         int ret;
3681         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
3682         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3683         struct ocfs2_cow_context *context = NULL;
3684         u32 cow_start, cow_len;
3685
3686         BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3687
3688         ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
3689                                               cpos, write_len, UINT_MAX,
3690                                               &cow_start, &cow_len);
3691         if (ret) {
3692                 mlog_errno(ret);
3693                 goto out;
3694         }
3695
3696         BUG_ON(cow_len == 0);
3697
3698         context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3699         if (!context) {
3700                 ret = -ENOMEM;
3701                 mlog_errno(ret);
3702                 goto out;
3703         }
3704
3705         context->inode = inode;
3706         context->cow_start = cow_start;
3707         context->cow_len = cow_len;
3708         context->ref_tree = ref_tree;
3709         context->ref_root_bh = ref_root_bh;
3710         context->cow_object = xv;
3711
3712         context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
3713         /* We need the extra credits for duplicate_clusters by jbd. */
3714         context->extra_credits =
3715                 ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
3716         context->get_clusters = ocfs2_xattr_value_get_clusters;
3717         context->post_refcount = post;
3718
3719         ocfs2_init_xattr_value_extent_tree(&context->data_et,
3720                                            INODE_CACHE(inode), vb);
3721
3722         ret = ocfs2_replace_cow(context);
3723         if (ret)
3724                 mlog_errno(ret);
3725
3726 out:
3727         kfree(context);
3728         return ret;
3729 }
3730
3731 /*
3732  * Insert a new extent into refcount tree and mark a extent rec
3733  * as refcounted in the dinode tree.
3734  */
3735 int ocfs2_add_refcount_flag(struct inode *inode,
3736                             struct ocfs2_extent_tree *data_et,
3737                             struct ocfs2_caching_info *ref_ci,
3738                             struct buffer_head *ref_root_bh,
3739                             u32 cpos, u32 p_cluster, u32 num_clusters,
3740                             struct ocfs2_cached_dealloc_ctxt *dealloc,
3741                             struct ocfs2_post_refcount *post)
3742 {
3743         int ret;
3744         handle_t *handle;
3745         int credits = 1, ref_blocks = 0;
3746         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3747         struct ocfs2_alloc_context *meta_ac = NULL;
3748
3749         ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
3750                                                ref_ci, ref_root_bh,
3751                                                p_cluster, num_clusters,
3752                                                &ref_blocks, &credits);
3753         if (ret) {
3754                 mlog_errno(ret);
3755                 goto out;
3756         }
3757
3758         trace_ocfs2_add_refcount_flag(ref_blocks, credits);
3759
3760         if (ref_blocks) {
3761                 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
3762                                                         ref_blocks, &meta_ac);
3763                 if (ret) {
3764                         mlog_errno(ret);
3765                         goto out;
3766                 }
3767         }
3768
3769         if (post)
3770                 credits += post->credits;
3771
3772         handle = ocfs2_start_trans(osb, credits);
3773         if (IS_ERR(handle)) {
3774                 ret = PTR_ERR(handle);
3775                 mlog_errno(ret);
3776                 goto out;
3777         }
3778
3779         ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
3780                                            cpos, num_clusters, p_cluster,
3781                                            meta_ac, dealloc);
3782         if (ret) {
3783                 mlog_errno(ret);
3784                 goto out_commit;
3785         }
3786
3787         ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3788                                         p_cluster, num_clusters, 0,
3789                                         meta_ac, dealloc);
3790         if (ret) {
3791                 mlog_errno(ret);
3792                 goto out_commit;
3793         }
3794
3795         if (post && post->func) {
3796                 ret = post->func(inode, handle, post->para);
3797                 if (ret)
3798                         mlog_errno(ret);
3799         }
3800
3801 out_commit:
3802         ocfs2_commit_trans(osb, handle);
3803 out:
3804         if (meta_ac)
3805                 ocfs2_free_alloc_context(meta_ac);
3806         return ret;
3807 }
3808
3809 static int ocfs2_change_ctime(struct inode *inode,
3810                               struct buffer_head *di_bh)
3811 {
3812         int ret;
3813         handle_t *handle;
3814         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3815
3816         handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
3817                                    OCFS2_INODE_UPDATE_CREDITS);
3818         if (IS_ERR(handle)) {
3819                 ret = PTR_ERR(handle);
3820                 mlog_errno(ret);
3821                 goto out;
3822         }
3823
3824         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
3825                                       OCFS2_JOURNAL_ACCESS_WRITE);
3826         if (ret) {
3827                 mlog_errno(ret);
3828                 goto out_commit;
3829         }
3830
3831         inode->i_ctime = CURRENT_TIME;
3832         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3833         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3834
3835         ocfs2_journal_dirty(handle, di_bh);
3836
3837 out_commit:
3838         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3839 out:
3840         return ret;
3841 }
3842
3843 static int ocfs2_attach_refcount_tree(struct inode *inode,
3844                                       struct buffer_head *di_bh)
3845 {
3846         int ret, data_changed = 0;
3847         struct buffer_head *ref_root_bh = NULL;
3848         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3849         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3850         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3851         struct ocfs2_refcount_tree *ref_tree;
3852         unsigned int ext_flags;
3853         loff_t size;
3854         u32 cpos, num_clusters, clusters, p_cluster;
3855         struct ocfs2_cached_dealloc_ctxt dealloc;
3856         struct ocfs2_extent_tree di_et;
3857
3858         ocfs2_init_dealloc_ctxt(&dealloc);
3859
3860         if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) {
3861                 ret = ocfs2_create_refcount_tree(inode, di_bh);
3862                 if (ret) {
3863                         mlog_errno(ret);
3864                         goto out;
3865                 }
3866         }
3867
3868         BUG_ON(!di->i_refcount_loc);
3869         ret = ocfs2_lock_refcount_tree(osb,
3870                                        le64_to_cpu(di->i_refcount_loc), 1,
3871                                        &ref_tree, &ref_root_bh);
3872         if (ret) {
3873                 mlog_errno(ret);
3874                 goto out;
3875         }
3876
3877         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
3878                 goto attach_xattr;
3879
3880         ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
3881
3882         size = i_size_read(inode);
3883         clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
3884
3885         cpos = 0;
3886         while (cpos < clusters) {
3887                 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3888                                          &num_clusters, &ext_flags);
3889
3890                 if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
3891                         ret = ocfs2_add_refcount_flag(inode, &di_et,
3892                                                       &ref_tree->rf_ci,
3893                                                       ref_root_bh, cpos,
3894                                                       p_cluster, num_clusters,
3895                                                       &dealloc, NULL);
3896                         if (ret) {
3897                                 mlog_errno(ret);
3898                                 goto unlock;
3899                         }
3900
3901                         data_changed = 1;
3902                 }
3903                 cpos += num_clusters;
3904         }
3905
3906 attach_xattr:
3907         if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
3908                 ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
3909                                                        &ref_tree->rf_ci,
3910                                                        ref_root_bh,
3911                                                        &dealloc);
3912                 if (ret) {
3913                         mlog_errno(ret);
3914                         goto unlock;
3915                 }
3916         }
3917
3918         if (data_changed) {
3919                 ret = ocfs2_change_ctime(inode, di_bh);
3920                 if (ret)
3921                         mlog_errno(ret);
3922         }
3923
3924 unlock:
3925         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3926         brelse(ref_root_bh);
3927
3928         if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
3929                 ocfs2_schedule_truncate_log_flush(osb, 1);
3930                 ocfs2_run_deallocs(osb, &dealloc);
3931         }
3932 out:
3933         /*
3934          * Empty the extent map so that we may get the right extent
3935          * record from the disk.
3936          */
3937         ocfs2_extent_map_trunc(inode, 0);
3938
3939         return ret;
3940 }
3941
3942 static int ocfs2_add_refcounted_extent(struct inode *inode,
3943                                    struct ocfs2_extent_tree *et,
3944                                    struct ocfs2_caching_info *ref_ci,
3945                                    struct buffer_head *ref_root_bh,
3946                                    u32 cpos, u32 p_cluster, u32 num_clusters,
3947                                    unsigned int ext_flags,
3948                                    struct ocfs2_cached_dealloc_ctxt *dealloc)
3949 {
3950         int ret;
3951         handle_t *handle;
3952         int credits = 0;
3953         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3954         struct ocfs2_alloc_context *meta_ac = NULL;
3955
3956         ret = ocfs2_lock_refcount_allocators(inode->i_sb,
3957                                              p_cluster, num_clusters,
3958                                              et, ref_ci,
3959                                              ref_root_bh, &meta_ac,
3960                                              NULL, &credits);
3961         if (ret) {
3962                 mlog_errno(ret);
3963                 goto out;
3964         }
3965
3966         handle = ocfs2_start_trans(osb, credits);
3967         if (IS_ERR(handle)) {
3968                 ret = PTR_ERR(handle);
3969                 mlog_errno(ret);
3970                 goto out;
3971         }
3972
3973         ret = ocfs2_insert_extent(handle, et, cpos,
3974                         ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
3975                         num_clusters, ext_flags, meta_ac);
3976         if (ret) {
3977                 mlog_errno(ret);
3978                 goto out_commit;
3979         }
3980
3981         ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3982                                       p_cluster, num_clusters,
3983                                       meta_ac, dealloc);
3984         if (ret)
3985                 mlog_errno(ret);
3986
3987 out_commit:
3988         ocfs2_commit_trans(osb, handle);
3989 out:
3990         if (meta_ac)
3991                 ocfs2_free_alloc_context(meta_ac);
3992         return ret;
3993 }
3994
3995 static int ocfs2_duplicate_inline_data(struct inode *s_inode,
3996                                        struct buffer_head *s_bh,
3997                                        struct inode *t_inode,
3998                                        struct buffer_head *t_bh)
3999 {
4000         int ret;
4001         handle_t *handle;
4002         struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4003         struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4004         struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
4005
4006         BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
4007
4008         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
4009         if (IS_ERR(handle)) {
4010                 ret = PTR_ERR(handle);
4011                 mlog_errno(ret);
4012                 goto out;
4013         }
4014
4015         ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4016                                       OCFS2_JOURNAL_ACCESS_WRITE);
4017         if (ret) {
4018                 mlog_errno(ret);
4019                 goto out_commit;
4020         }
4021
4022         t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
4023         memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
4024                le16_to_cpu(s_di->id2.i_data.id_count));
4025         spin_lock(&OCFS2_I(t_inode)->ip_lock);
4026         OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
4027         t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
4028         spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4029
4030         ocfs2_journal_dirty(handle, t_bh);
4031
4032 out_commit:
4033         ocfs2_commit_trans(osb, handle);
4034 out:
4035         return ret;
4036 }
4037
4038 static int ocfs2_duplicate_extent_list(struct inode *s_inode,
4039                                 struct inode *t_inode,
4040                                 struct buffer_head *t_bh,
4041                                 struct ocfs2_caching_info *ref_ci,
4042                                 struct buffer_head *ref_root_bh,
4043                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
4044 {
4045         int ret = 0;
4046         u32 p_cluster, num_clusters, clusters, cpos;
4047         loff_t size;
4048         unsigned int ext_flags;
4049         struct ocfs2_extent_tree et;
4050
4051         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
4052
4053         size = i_size_read(s_inode);
4054         clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
4055
4056         cpos = 0;
4057         while (cpos < clusters) {
4058                 ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
4059                                          &num_clusters, &ext_flags);
4060
4061                 if (p_cluster) {
4062                         ret = ocfs2_add_refcounted_extent(t_inode, &et,
4063                                                           ref_ci, ref_root_bh,
4064                                                           cpos, p_cluster,
4065                                                           num_clusters,
4066                                                           ext_flags,
4067                                                           dealloc);
4068                         if (ret) {
4069                                 mlog_errno(ret);
4070                                 goto out;
4071                         }
4072                 }
4073
4074                 cpos += num_clusters;
4075         }
4076
4077 out:
4078         return ret;
4079 }
4080
4081 /*
4082  * change the new file's attributes to the src.
4083  *
4084  * reflink creates a snapshot of a file, that means the attributes
4085  * must be identical except for three exceptions - nlink, ino, and ctime.
4086  */
4087 static int ocfs2_complete_reflink(struct inode *s_inode,
4088                                   struct buffer_head *s_bh,
4089                                   struct inode *t_inode,
4090                                   struct buffer_head *t_bh,
4091                                   bool preserve)
4092 {
4093         int ret;
4094         handle_t *handle;
4095         struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4096         struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
4097         loff_t size = i_size_read(s_inode);
4098
4099         handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
4100                                    OCFS2_INODE_UPDATE_CREDITS);
4101         if (IS_ERR(handle)) {
4102                 ret = PTR_ERR(handle);
4103                 mlog_errno(ret);
4104                 return ret;
4105         }
4106
4107         ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4108                                       OCFS2_JOURNAL_ACCESS_WRITE);
4109         if (ret) {
4110                 mlog_errno(ret);
4111                 goto out_commit;
4112         }
4113
4114         spin_lock(&OCFS2_I(t_inode)->ip_lock);
4115         OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
4116         OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
4117         OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
4118         spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4119         i_size_write(t_inode, size);
4120         t_inode->i_blocks = s_inode->i_blocks;
4121
4122         di->i_xattr_inline_size = s_di->i_xattr_inline_size;
4123         di->i_clusters = s_di->i_clusters;
4124         di->i_size = s_di->i_size;
4125         di->i_dyn_features = s_di->i_dyn_features;
4126         di->i_attr = s_di->i_attr;
4127
4128         if (preserve) {
4129                 t_inode->i_uid = s_inode->i_uid;
4130                 t_inode->i_gid = s_inode->i_gid;
4131                 t_inode->i_mode = s_inode->i_mode;
4132                 di->i_uid = s_di->i_uid;
4133                 di->i_gid = s_di->i_gid;
4134                 di->i_mode = s_di->i_mode;
4135
4136                 /*
4137                  * update time.
4138                  * we want mtime to appear identical to the source and
4139                  * update ctime.
4140                  */
4141                 t_inode->i_ctime = CURRENT_TIME;
4142
4143                 di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
4144                 di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
4145
4146                 t_inode->i_mtime = s_inode->i_mtime;
4147                 di->i_mtime = s_di->i_mtime;
4148                 di->i_mtime_nsec = s_di->i_mtime_nsec;
4149         }
4150
4151         ocfs2_journal_dirty(handle, t_bh);
4152
4153 out_commit:
4154         ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
4155         return ret;
4156 }
4157
4158 static int ocfs2_create_reflink_node(struct inode *s_inode,
4159                                      struct buffer_head *s_bh,
4160                                      struct inode *t_inode,
4161                                      struct buffer_head *t_bh,
4162                                      bool preserve)
4163 {
4164         int ret;
4165         struct buffer_head *ref_root_bh = NULL;
4166         struct ocfs2_cached_dealloc_ctxt dealloc;
4167         struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4168         struct ocfs2_refcount_block *rb;
4169         struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
4170         struct ocfs2_refcount_tree *ref_tree;
4171
4172         ocfs2_init_dealloc_ctxt(&dealloc);
4173
4174         ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4175                                       le64_to_cpu(di->i_refcount_loc));
4176         if (ret) {
4177                 mlog_errno(ret);
4178                 goto out;
4179         }
4180
4181         if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4182                 ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
4183                                                   t_inode, t_bh);
4184                 if (ret)
4185                         mlog_errno(ret);
4186                 goto out;
4187         }
4188
4189         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
4190                                        1, &ref_tree, &ref_root_bh);
4191         if (ret) {
4192                 mlog_errno(ret);
4193                 goto out;
4194         }
4195         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
4196
4197         ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
4198                                           &ref_tree->rf_ci, ref_root_bh,
4199                                           &dealloc);
4200         if (ret) {
4201                 mlog_errno(ret);
4202                 goto out_unlock_refcount;
4203         }
4204
4205 out_unlock_refcount:
4206         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4207         brelse(ref_root_bh);
4208 out:
4209         if (ocfs2_dealloc_has_cluster(&dealloc)) {
4210                 ocfs2_schedule_truncate_log_flush(osb, 1);
4211                 ocfs2_run_deallocs(osb, &dealloc);
4212         }
4213
4214         return ret;
4215 }
4216
4217 static int __ocfs2_reflink(struct dentry *old_dentry,
4218                            struct buffer_head *old_bh,
4219                            struct inode *new_inode,
4220                            bool preserve)
4221 {
4222         int ret;
4223         struct inode *inode = old_dentry->d_inode;
4224         struct buffer_head *new_bh = NULL;
4225
4226         if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
4227                 ret = -EINVAL;
4228                 mlog_errno(ret);
4229                 goto out;
4230         }
4231
4232         ret = filemap_fdatawrite(inode->i_mapping);
4233         if (ret) {
4234                 mlog_errno(ret);
4235                 goto out;
4236         }
4237
4238         ret = ocfs2_attach_refcount_tree(inode, old_bh);
4239         if (ret) {
4240                 mlog_errno(ret);
4241                 goto out;
4242         }
4243
4244         mutex_lock_nested(&new_inode->i_mutex, I_MUTEX_CHILD);
4245         ret = ocfs2_inode_lock_nested(new_inode, &new_bh, 1,
4246                                       OI_LS_REFLINK_TARGET);
4247         if (ret) {
4248                 mlog_errno(ret);
4249                 goto out_unlock;
4250         }
4251
4252         ret = ocfs2_create_reflink_node(inode, old_bh,
4253                                         new_inode, new_bh, preserve);
4254         if (ret) {
4255                 mlog_errno(ret);
4256                 goto inode_unlock;
4257         }
4258
4259         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
4260                 ret = ocfs2_reflink_xattrs(inode, old_bh,
4261                                            new_inode, new_bh,
4262                                            preserve);
4263                 if (ret) {
4264                         mlog_errno(ret);
4265                         goto inode_unlock;
4266                 }
4267         }
4268
4269         ret = ocfs2_complete_reflink(inode, old_bh,
4270                                      new_inode, new_bh, preserve);
4271         if (ret)
4272                 mlog_errno(ret);
4273
4274 inode_unlock:
4275         ocfs2_inode_unlock(new_inode, 1);
4276         brelse(new_bh);
4277 out_unlock:
4278         mutex_unlock(&new_inode->i_mutex);
4279 out:
4280         if (!ret) {
4281                 ret = filemap_fdatawait(inode->i_mapping);
4282                 if (ret)
4283                         mlog_errno(ret);
4284         }
4285         return ret;
4286 }
4287
4288 static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
4289                          struct dentry *new_dentry, bool preserve)
4290 {
4291         int error;
4292         struct inode *inode = old_dentry->d_inode;
4293         struct buffer_head *old_bh = NULL;
4294         struct inode *new_orphan_inode = NULL;
4295
4296         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4297                 return -EOPNOTSUPP;
4298
4299         error = ocfs2_create_inode_in_orphan(dir, inode->i_mode,
4300                                              &new_orphan_inode);
4301         if (error) {
4302                 mlog_errno(error);
4303                 goto out;
4304         }
4305
4306         error = ocfs2_inode_lock(inode, &old_bh, 1);
4307         if (error) {
4308                 mlog_errno(error);
4309                 goto out;
4310         }
4311
4312         down_write(&OCFS2_I(inode)->ip_xattr_sem);
4313         down_write(&OCFS2_I(inode)->ip_alloc_sem);
4314         error = __ocfs2_reflink(old_dentry, old_bh,
4315                                 new_orphan_inode, preserve);
4316         up_write(&OCFS2_I(inode)->ip_alloc_sem);
4317         up_write(&OCFS2_I(inode)->ip_xattr_sem);
4318
4319         ocfs2_inode_unlock(inode, 1);
4320         brelse(old_bh);
4321
4322         if (error) {
4323                 mlog_errno(error);
4324                 goto out;
4325         }
4326
4327         /* If the security isn't preserved, we need to re-initialize them. */
4328         if (!preserve) {
4329                 error = ocfs2_init_security_and_acl(dir, new_orphan_inode,
4330                                                     &new_dentry->d_name);
4331                 if (error)
4332                         mlog_errno(error);
4333         }
4334 out:
4335         if (!error) {
4336                 error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
4337                                                        new_dentry);
4338                 if (error)
4339                         mlog_errno(error);
4340         }
4341
4342         if (new_orphan_inode) {
4343                 /*
4344                  * We need to open_unlock the inode no matter whether we
4345                  * succeed or not, so that other nodes can delete it later.
4346                  */
4347                 ocfs2_open_unlock(new_orphan_inode);
4348                 if (error)
4349                         iput(new_orphan_inode);
4350         }
4351
4352         return error;
4353 }
4354
4355 /*
4356  * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
4357  * sys_reflink().  This will go away when vfs_reflink() exists in
4358  * fs/namei.c.
4359  */
4360
4361 /* copied from may_create in VFS. */
4362 static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
4363 {
4364         if (child->d_inode)
4365                 return -EEXIST;
4366         if (IS_DEADDIR(dir))
4367                 return -ENOENT;
4368         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
4369 }
4370
4371 /**
4372  * ocfs2_vfs_reflink - Create a reference-counted link
4373  *
4374  * @old_dentry:        source dentry + inode
4375  * @dir:       directory to create the target
4376  * @new_dentry:        target dentry
4377  * @preserve:  if true, preserve all file attributes
4378  */
4379 static int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
4380                              struct dentry *new_dentry, bool preserve)
4381 {
4382         struct inode *inode = old_dentry->d_inode;
4383         int error;
4384
4385         if (!inode)
4386                 return -ENOENT;
4387
4388         error = ocfs2_may_create(dir, new_dentry);
4389         if (error)
4390                 return error;
4391
4392         if (dir->i_sb != inode->i_sb)
4393                 return -EXDEV;
4394
4395         /*
4396          * A reflink to an append-only or immutable file cannot be created.
4397          */
4398         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4399                 return -EPERM;
4400
4401         /* Only regular files can be reflinked. */
4402         if (!S_ISREG(inode->i_mode))
4403                 return -EPERM;
4404
4405         /*
4406          * If the caller wants to preserve ownership, they require the
4407          * rights to do so.
4408          */
4409         if (preserve) {
4410                 if ((current_fsuid() != inode->i_uid) && !capable(CAP_CHOWN))
4411                         return -EPERM;
4412                 if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
4413                         return -EPERM;
4414         }
4415
4416         /*
4417          * If the caller is modifying any aspect of the attributes, they
4418          * are not creating a snapshot.  They need read permission on the
4419          * file.
4420          */
4421         if (!preserve) {
4422                 error = inode_permission(inode, MAY_READ);
4423                 if (error)
4424                         return error;
4425         }
4426
4427         mutex_lock(&inode->i_mutex);
4428         dquot_initialize(dir);
4429         error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
4430         mutex_unlock(&inode->i_mutex);
4431         if (!error)
4432                 fsnotify_create(dir, new_dentry);
4433         return error;
4434 }
4435 /*
4436  * Most codes are copied from sys_linkat.
4437  */
4438 int ocfs2_reflink_ioctl(struct inode *inode,
4439                         const char __user *oldname,
4440                         const char __user *newname,
4441                         bool preserve)
4442 {
4443         struct dentry *new_dentry;
4444         struct path old_path, new_path;
4445         int error;
4446
4447         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4448                 return -EOPNOTSUPP;
4449
4450         error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
4451         if (error) {
4452                 mlog_errno(error);
4453                 return error;
4454         }
4455
4456         new_dentry = user_path_create(AT_FDCWD, newname, &new_path, 0);
4457         error = PTR_ERR(new_dentry);
4458         if (IS_ERR(new_dentry)) {
4459                 mlog_errno(error);
4460                 goto out;
4461         }
4462
4463         error = -EXDEV;
4464         if (old_path.mnt != new_path.mnt) {
4465                 mlog_errno(error);
4466                 goto out_dput;
4467         }
4468
4469         error = ocfs2_vfs_reflink(old_path.dentry,
4470                                   new_path.dentry->d_inode,
4471                                   new_dentry, preserve);
4472 out_dput:
4473         done_path_create(&new_path, new_dentry);
4474 out:
4475         path_put(&old_path);
4476
4477         return error;
4478 }