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