Pileus Git - ~andy/linux/blob - fs/ocfs2/refcounttree.c

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