fs/ocfs2/refcounttree.c

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