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