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 #define MLOG_MASK_PREFIX ML_REFCOUNT
  20 #include <cluster/masklog.h>
  21 #include "ocfs2.h"
  22 #include "inode.h"
  23 #include "alloc.h"
  24 #include "suballoc.h"
  25 #include "journal.h"
  26 #include "uptodate.h"
  27 #include "super.h"
  28 #include "buffer_head_io.h"
  29 #include "blockcheck.h"
  30 #include "refcounttree.h"
  31 #include "sysfile.h"
  32 #include "dlmglue.h"
  33 #include "extent_map.h"
  34 #include "aops.h"
  35 #include "xattr.h"
  36 #include "namei.h"
  37
  38 #include <linux/bio.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/slab.h>
  41 #include <linux/writeback.h>
  42 #include <linux/pagevec.h>
  43 #include <linux/swap.h>
  44 #include <linux/security.h>
  45 #include <linux/fsnotify.h>
  46 #include <linux/quotaops.h>
  47 #include <linux/namei.h>
  48 #include <linux/mount.h>
  49
  50 struct ocfs2_cow_context {
  51         struct inode *inode;
  52         u32 cow_start;
  53         u32 cow_len;
  54         struct ocfs2_extent_tree data_et;
  55         struct ocfs2_refcount_tree *ref_tree;
  56         struct buffer_head *ref_root_bh;
  57         struct ocfs2_alloc_context *meta_ac;
  58         struct ocfs2_alloc_context *data_ac;
  59         struct ocfs2_cached_dealloc_ctxt dealloc;
  60         void *cow_object;
  61         struct ocfs2_post_refcount *post_refcount;
  62         int extra_credits;
  63         int (*get_clusters)(struct ocfs2_cow_context *context,
  64                             u32 v_cluster, u32 *p_cluster,
  65                             u32 *num_clusters,
  66                             unsigned int *extent_flags);
  67         int (*cow_duplicate_clusters)(handle_t *handle,
  68                                       struct ocfs2_cow_context *context,
  69                                       u32 cpos, u32 old_cluster,
  70                                       u32 new_cluster, u32 new_len);
  71 };
  72
  73 static inline struct ocfs2_refcount_tree *
  74 cache_info_to_refcount(struct ocfs2_caching_info *ci)
  75 {
  76         return container_of(ci, struct ocfs2_refcount_tree, rf_ci);
  77 }
  78
  79 static int ocfs2_validate_refcount_block(struct super_block *sb,
  80                                          struct buffer_head *bh)
  81 {
  82         int rc;
  83         struct ocfs2_refcount_block *rb =
  84                 (struct ocfs2_refcount_block *)bh->b_data;
  85
  86         mlog(0, "Validating refcount block %llu\n",
  87              (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                 mlog(0, "Purge tree %llu\n",
 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         mlog(0, "create tree for inode %lu\n", inode->i_ino);
 578
 579         ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
 580         if (ret) {
 581                 mlog_errno(ret);
 582                 goto out;
 583         }
 584
 585         handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_CREATE_CREDITS);
 586         if (IS_ERR(handle)) {
 587                 ret = PTR_ERR(handle);
 588                 mlog_errno(ret);
 589                 goto out;
 590         }
 591
 592         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 593                                       OCFS2_JOURNAL_ACCESS_WRITE);
 594         if (ret) {
 595                 mlog_errno(ret);
 596                 goto out_commit;
 597         }
 598
 599         ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
 600                                    &suballoc_bit_start, &num_got,
 601                                    &first_blkno);
 602         if (ret) {
 603                 mlog_errno(ret);
 604                 goto out_commit;
 605         }
 606
 607         new_tree = ocfs2_allocate_refcount_tree(osb, first_blkno);
 608         if (!new_tree) {
 609                 ret = -ENOMEM;
 610                 mlog_errno(ret);
 611                 goto out_commit;
 612         }
 613
 614         new_bh = sb_getblk(inode->i_sb, first_blkno);
 615         ocfs2_set_new_buffer_uptodate(&new_tree->rf_ci, new_bh);
 616
 617         ret = ocfs2_journal_access_rb(handle, &new_tree->rf_ci, new_bh,
 618                                       OCFS2_JOURNAL_ACCESS_CREATE);
 619         if (ret) {
 620                 mlog_errno(ret);
 621                 goto out_commit;
 622         }
 623
 624         /* Initialize ocfs2_refcount_block. */
 625         rb = (struct ocfs2_refcount_block *)new_bh->b_data;
 626         memset(rb, 0, inode->i_sb->s_blocksize);
 627         strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
 628         rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
 629         rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
 630         rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
 631         rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
 632         rb->rf_blkno = cpu_to_le64(first_blkno);
 633         rb->rf_count = cpu_to_le32(1);
 634         rb->rf_records.rl_count =
 635                         cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb));
 636         spin_lock(&osb->osb_lock);
 637         rb->rf_generation = osb->s_next_generation++;
 638         spin_unlock(&osb->osb_lock);
 639
 640         ocfs2_journal_dirty(handle, new_bh);
 641
 642         spin_lock(&oi->ip_lock);
 643         oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
 644         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 645         di->i_refcount_loc = cpu_to_le64(first_blkno);
 646         spin_unlock(&oi->ip_lock);
 647
 648         mlog(0, "created tree for inode %lu, refblock %llu\n",
 649              inode->i_ino, (unsigned long long)first_blkno);
 650
 651         ocfs2_journal_dirty(handle, di_bh);
 652
 653         /*
 654          * We have to init the tree lock here since it will use
 655          * the generation number to create it.
 656          */
 657         new_tree->rf_generation = le32_to_cpu(rb->rf_generation);
 658         ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno,
 659                                       new_tree->rf_generation);
 660
 661         spin_lock(&osb->osb_lock);
 662         tree = ocfs2_find_refcount_tree(osb, first_blkno);
 663
 664         /*
 665          * We've just created a new refcount tree in this block.  If
 666          * we found a refcount tree on the ocfs2_super, it must be
 667          * one we just deleted.  We free the old tree before
 668          * inserting the new tree.
 669          */
 670         BUG_ON(tree && tree->rf_generation == new_tree->rf_generation);
 671         if (tree)
 672                 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
 673         ocfs2_insert_refcount_tree(osb, new_tree);
 674         spin_unlock(&osb->osb_lock);
 675         new_tree = NULL;
 676         if (tree)
 677                 ocfs2_refcount_tree_put(tree);
 678
 679 out_commit:
 680         ocfs2_commit_trans(osb, handle);
 681
 682 out:
 683         if (new_tree) {
 684                 ocfs2_metadata_cache_exit(&new_tree->rf_ci);
 685                 kfree(new_tree);
 686         }
 687
 688         brelse(new_bh);
 689         if (meta_ac)
 690                 ocfs2_free_alloc_context(meta_ac);
 691
 692         return ret;
 693 }
 694
 695 static int ocfs2_set_refcount_tree(struct inode *inode,
 696                                    struct buffer_head *di_bh,
 697                                    u64 refcount_loc)
 698 {
 699         int ret;
 700         handle_t *handle = NULL;
 701         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 702         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 703         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 704         struct buffer_head *ref_root_bh = NULL;
 705         struct ocfs2_refcount_block *rb;
 706         struct ocfs2_refcount_tree *ref_tree;
 707
 708         BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
 709
 710         ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
 711                                        &ref_tree, &ref_root_bh);
 712         if (ret) {
 713                 mlog_errno(ret);
 714                 return ret;
 715         }
 716
 717         handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS);
 718         if (IS_ERR(handle)) {
 719                 ret = PTR_ERR(handle);
 720                 mlog_errno(ret);
 721                 goto out;
 722         }
 723
 724         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 725                                       OCFS2_JOURNAL_ACCESS_WRITE);
 726         if (ret) {
 727                 mlog_errno(ret);
 728                 goto out_commit;
 729         }
 730
 731         ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh,
 732                                       OCFS2_JOURNAL_ACCESS_WRITE);
 733         if (ret) {
 734                 mlog_errno(ret);
 735                 goto out_commit;
 736         }
 737
 738         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
 739         le32_add_cpu(&rb->rf_count, 1);
 740
 741         ocfs2_journal_dirty(handle, ref_root_bh);
 742
 743         spin_lock(&oi->ip_lock);
 744         oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
 745         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 746         di->i_refcount_loc = cpu_to_le64(refcount_loc);
 747         spin_unlock(&oi->ip_lock);
 748         ocfs2_journal_dirty(handle, di_bh);
 749
 750 out_commit:
 751         ocfs2_commit_trans(osb, handle);
 752 out:
 753         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
 754         brelse(ref_root_bh);
 755
 756         return ret;
 757 }
 758
 759 int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh)
 760 {
 761         int ret, delete_tree = 0;
 762         handle_t *handle = NULL;
 763         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 764         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 765         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 766         struct ocfs2_refcount_block *rb;
 767         struct inode *alloc_inode = NULL;
 768         struct buffer_head *alloc_bh = NULL;
 769         struct buffer_head *blk_bh = NULL;
 770         struct ocfs2_refcount_tree *ref_tree;
 771         int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS;
 772         u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc);
 773         u16 bit = 0;
 774
 775         if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL))
 776                 return 0;
 777
 778         BUG_ON(!ref_blkno);
 779         ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh);
 780         if (ret) {
 781                 mlog_errno(ret);
 782                 return ret;
 783         }
 784
 785         rb = (struct ocfs2_refcount_block *)blk_bh->b_data;
 786
 787         /*
 788          * If we are the last user, we need to free the block.
 789          * So lock the allocator ahead.
 790          */
 791         if (le32_to_cpu(rb->rf_count) == 1) {
 792                 blk = le64_to_cpu(rb->rf_blkno);
 793                 bit = le16_to_cpu(rb->rf_suballoc_bit);
 794                 if (rb->rf_suballoc_loc)
 795                         bg_blkno = le64_to_cpu(rb->rf_suballoc_loc);
 796                 else
 797                         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
 798
 799                 alloc_inode = ocfs2_get_system_file_inode(osb,
 800                                         EXTENT_ALLOC_SYSTEM_INODE,
 801                                         le16_to_cpu(rb->rf_suballoc_slot));
 802                 if (!alloc_inode) {
 803                         ret = -ENOMEM;
 804                         mlog_errno(ret);
 805                         goto out;
 806                 }
 807                 mutex_lock(&alloc_inode->i_mutex);
 808
 809                 ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1);
 810                 if (ret) {
 811                         mlog_errno(ret);
 812                         goto out_mutex;
 813                 }
 814
 815                 credits += OCFS2_SUBALLOC_FREE;
 816         }
 817
 818         handle = ocfs2_start_trans(osb, credits);
 819         if (IS_ERR(handle)) {
 820                 ret = PTR_ERR(handle);
 821                 mlog_errno(ret);
 822                 goto out_unlock;
 823         }
 824
 825         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 826                                       OCFS2_JOURNAL_ACCESS_WRITE);
 827         if (ret) {
 828                 mlog_errno(ret);
 829                 goto out_commit;
 830         }
 831
 832         ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh,
 833                                       OCFS2_JOURNAL_ACCESS_WRITE);
 834         if (ret) {
 835                 mlog_errno(ret);
 836                 goto out_commit;
 837         }
 838
 839         spin_lock(&oi->ip_lock);
 840         oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL;
 841         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 842         di->i_refcount_loc = 0;
 843         spin_unlock(&oi->ip_lock);
 844         ocfs2_journal_dirty(handle, di_bh);
 845
 846         le32_add_cpu(&rb->rf_count , -1);
 847         ocfs2_journal_dirty(handle, blk_bh);
 848
 849         if (!rb->rf_count) {
 850                 delete_tree = 1;
 851                 ocfs2_erase_refcount_tree_from_list(osb, ref_tree);
 852                 ret = ocfs2_free_suballoc_bits(handle, alloc_inode,
 853                                                alloc_bh, bit, bg_blkno, 1);
 854                 if (ret)
 855                         mlog_errno(ret);
 856         }
 857
 858 out_commit:
 859         ocfs2_commit_trans(osb, handle);
 860 out_unlock:
 861         if (alloc_inode) {
 862                 ocfs2_inode_unlock(alloc_inode, 1);
 863                 brelse(alloc_bh);
 864         }
 865 out_mutex:
 866         if (alloc_inode) {
 867                 mutex_unlock(&alloc_inode->i_mutex);
 868                 iput(alloc_inode);
 869         }
 870 out:
 871         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
 872         if (delete_tree)
 873                 ocfs2_refcount_tree_put(ref_tree);
 874         brelse(blk_bh);
 875
 876         return ret;
 877 }
 878
 879 static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info *ci,
 880                                           struct buffer_head *ref_leaf_bh,
 881                                           u64 cpos, unsigned int len,
 882                                           struct ocfs2_refcount_rec *ret_rec,
 883                                           int *index)
 884 {
 885         int i = 0;
 886         struct ocfs2_refcount_block *rb =
 887                 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
 888         struct ocfs2_refcount_rec *rec = NULL;
 889
 890         for (; i < le16_to_cpu(rb->rf_records.rl_used); i++) {
 891                 rec = &rb->rf_records.rl_recs[i];
 892
 893                 if (le64_to_cpu(rec->r_cpos) +
 894                     le32_to_cpu(rec->r_clusters) <= cpos)
 895                         continue;
 896                 else if (le64_to_cpu(rec->r_cpos) > cpos)
 897                         break;
 898
 899                 /* ok, cpos fail in this rec. Just return. */
 900                 if (ret_rec)
 901                         *ret_rec = *rec;
 902                 goto out;
 903         }
 904
 905         if (ret_rec) {
 906                 /* We meet with a hole here, so fake the rec. */
 907                 ret_rec->r_cpos = cpu_to_le64(cpos);
 908                 ret_rec->r_refcount = 0;
 909                 if (i < le16_to_cpu(rb->rf_records.rl_used) &&
 910                     le64_to_cpu(rec->r_cpos) < cpos + len)
 911                         ret_rec->r_clusters =
 912                                 cpu_to_le32(le64_to_cpu(rec->r_cpos) - cpos);
 913                 else
 914                         ret_rec->r_clusters = cpu_to_le32(len);
 915         }
 916
 917 out:
 918         *index = i;
 919 }
 920
 921 /*
 922  * Try to remove refcount tree. The mechanism is:
 923  * 1) Check whether i_clusters == 0, if no, exit.
 924  * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
 925  * 3) Check whether we have inline xattr stored outside, if yes, exit.
 926  * 4) Remove the tree.
 927  */
 928 int ocfs2_try_remove_refcount_tree(struct inode *inode,
 929                                    struct buffer_head *di_bh)
 930 {
 931         int ret;
 932         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 933         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 934
 935         down_write(&oi->ip_xattr_sem);
 936         down_write(&oi->ip_alloc_sem);
 937
 938         if (oi->ip_clusters)
 939                 goto out;
 940
 941         if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
 942                 goto out;
 943
 944         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
 945             ocfs2_has_inline_xattr_value_outside(inode, di))
 946                 goto out;
 947
 948         ret = ocfs2_remove_refcount_tree(inode, di_bh);
 949         if (ret)
 950                 mlog_errno(ret);
 951 out:
 952         up_write(&oi->ip_alloc_sem);
 953         up_write(&oi->ip_xattr_sem);
 954         return 0;
 955 }
 956
 957 /*
 958  * Find the end range for a leaf refcount block indicated by
 959  * el->l_recs[index].e_blkno.
 960  */
 961 static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info *ci,
 962                                        struct buffer_head *ref_root_bh,
 963                                        struct ocfs2_extent_block *eb,
 964                                        struct ocfs2_extent_list *el,
 965                                        int index,  u32 *cpos_end)
 966 {
 967         int ret, i, subtree_root;
 968         u32 cpos;
 969         u64 blkno;
 970         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
 971         struct ocfs2_path *left_path = NULL, *right_path = NULL;
 972         struct ocfs2_extent_tree et;
 973         struct ocfs2_extent_list *tmp_el;
 974
 975         if (index < le16_to_cpu(el->l_next_free_rec) - 1) {
 976                 /*
 977                  * We have a extent rec after index, so just use the e_cpos
 978                  * of the next extent rec.
 979                  */
 980                 *cpos_end = le32_to_cpu(el->l_recs[index+1].e_cpos);
 981                 return 0;
 982         }
 983
 984         if (!eb || (eb && !eb->h_next_leaf_blk)) {
 985                 /*
 986                  * We are the last extent rec, so any high cpos should
 987                  * be stored in this leaf refcount block.
 988                  */
 989                 *cpos_end = UINT_MAX;
 990                 return 0;
 991         }
 992
 993         /*
 994          * If the extent block isn't the last one, we have to find
 995          * the subtree root between this extent block and the next
 996          * leaf extent block and get the corresponding e_cpos from
 997          * the subroot. Otherwise we may corrupt the b-tree.
 998          */
 999         ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1000
1001         left_path = ocfs2_new_path_from_et(&et);
1002         if (!left_path) {
1003                 ret = -ENOMEM;
1004                 mlog_errno(ret);
1005                 goto out;
1006         }
1007
1008         cpos = le32_to_cpu(eb->h_list.l_recs[index].e_cpos);
1009         ret = ocfs2_find_path(ci, left_path, cpos);
1010         if (ret) {
1011                 mlog_errno(ret);
1012                 goto out;
1013         }
1014
1015         right_path = ocfs2_new_path_from_path(left_path);
1016         if (!right_path) {
1017                 ret = -ENOMEM;
1018                 mlog_errno(ret);
1019                 goto out;
1020         }
1021
1022         ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, &cpos);
1023         if (ret) {
1024                 mlog_errno(ret);
1025                 goto out;
1026         }
1027
1028         ret = ocfs2_find_path(ci, right_path, cpos);
1029         if (ret) {
1030                 mlog_errno(ret);
1031                 goto out;
1032         }
1033
1034         subtree_root = ocfs2_find_subtree_root(&et, left_path,
1035                                                right_path);
1036
1037         tmp_el = left_path->p_node[subtree_root].el;
1038         blkno = left_path->p_node[subtree_root+1].bh->b_blocknr;
1039         for (i = 0; i < le32_to_cpu(tmp_el->l_next_free_rec); i++) {
1040                 if (le64_to_cpu(tmp_el->l_recs[i].e_blkno) == blkno) {
1041                         *cpos_end = le32_to_cpu(tmp_el->l_recs[i+1].e_cpos);
1042                         break;
1043                 }
1044         }
1045
1046         BUG_ON(i == le32_to_cpu(tmp_el->l_next_free_rec));
1047
1048 out:
1049         ocfs2_free_path(left_path);
1050         ocfs2_free_path(right_path);
1051         return ret;
1052 }
1053
1054 /*
1055  * Given a cpos and len, try to find the refcount record which contains cpos.
1056  * 1. If cpos can be found in one refcount record, return the record.
1057  * 2. If cpos can't be found, return a fake record which start from cpos
1058  *    and end at a small value between cpos+len and start of the next record.
1059  *    This fake record has r_refcount = 0.
1060  */
1061 static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
1062                                   struct buffer_head *ref_root_bh,
1063                                   u64 cpos, unsigned int len,
1064                                   struct ocfs2_refcount_rec *ret_rec,
1065                                   int *index,
1066                                   struct buffer_head **ret_bh)
1067 {
1068         int ret = 0, i, found;
1069         u32 low_cpos, uninitialized_var(cpos_end);
1070         struct ocfs2_extent_list *el;
1071         struct ocfs2_extent_rec *rec = NULL;
1072         struct ocfs2_extent_block *eb = NULL;
1073         struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
1074         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1075         struct ocfs2_refcount_block *rb =
1076                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1077
1078         if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)) {
1079                 ocfs2_find_refcount_rec_in_rl(ci, ref_root_bh, cpos, len,
1080                                               ret_rec, index);
1081                 *ret_bh = ref_root_bh;
1082                 get_bh(ref_root_bh);
1083                 return 0;
1084         }
1085
1086         el = &rb->rf_list;
1087         low_cpos = cpos & OCFS2_32BIT_POS_MASK;
1088
1089         if (el->l_tree_depth) {
1090                 ret = ocfs2_find_leaf(ci, el, low_cpos, &eb_bh);
1091                 if (ret) {
1092                         mlog_errno(ret);
1093                         goto out;
1094                 }
1095
1096                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1097                 el = &eb->h_list;
1098
1099                 if (el->l_tree_depth) {
1100                         ocfs2_error(sb,
1101                         "refcount tree %llu has non zero tree "
1102                         "depth in leaf btree tree block %llu\n",
1103                         (unsigned long long)ocfs2_metadata_cache_owner(ci),
1104                         (unsigned long long)eb_bh->b_blocknr);
1105                         ret = -EROFS;
1106                         goto out;
1107                 }
1108         }
1109
1110         found = 0;
1111         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1112                 rec = &el->l_recs[i];
1113
1114                 if (le32_to_cpu(rec->e_cpos) <= low_cpos) {
1115                         found = 1;
1116                         break;
1117                 }
1118         }
1119
1120         if (found) {
1121                 ret = ocfs2_get_refcount_cpos_end(ci, ref_root_bh,
1122                                                   eb, el, i, &cpos_end);
1123                 if (ret) {
1124                         mlog_errno(ret);
1125                         goto out;
1126                 }
1127
1128                 if (cpos_end < low_cpos + len)
1129                         len = cpos_end - low_cpos;
1130         }
1131
1132         ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
1133                                         &ref_leaf_bh);
1134         if (ret) {
1135                 mlog_errno(ret);
1136                 goto out;
1137         }
1138
1139         ocfs2_find_refcount_rec_in_rl(ci, ref_leaf_bh, cpos, len,
1140                                       ret_rec, index);
1141         *ret_bh = ref_leaf_bh;
1142 out:
1143         brelse(eb_bh);
1144         return ret;
1145 }
1146
1147 enum ocfs2_ref_rec_contig {
1148         REF_CONTIG_NONE = 0,
1149         REF_CONTIG_LEFT,
1150         REF_CONTIG_RIGHT,
1151         REF_CONTIG_LEFTRIGHT,
1152 };
1153
1154 static enum ocfs2_ref_rec_contig
1155         ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block *rb,
1156                                     int index)
1157 {
1158         if ((rb->rf_records.rl_recs[index].r_refcount ==
1159             rb->rf_records.rl_recs[index + 1].r_refcount) &&
1160             (le64_to_cpu(rb->rf_records.rl_recs[index].r_cpos) +
1161             le32_to_cpu(rb->rf_records.rl_recs[index].r_clusters) ==
1162             le64_to_cpu(rb->rf_records.rl_recs[index + 1].r_cpos)))
1163                 return REF_CONTIG_RIGHT;
1164
1165         return REF_CONTIG_NONE;
1166 }
1167
1168 static enum ocfs2_ref_rec_contig
1169         ocfs2_refcount_rec_contig(struct ocfs2_refcount_block *rb,
1170                                   int index)
1171 {
1172         enum ocfs2_ref_rec_contig ret = REF_CONTIG_NONE;
1173
1174         if (index < le16_to_cpu(rb->rf_records.rl_used) - 1)
1175                 ret = ocfs2_refcount_rec_adjacent(rb, index);
1176
1177         if (index > 0) {
1178                 enum ocfs2_ref_rec_contig tmp;
1179
1180                 tmp = ocfs2_refcount_rec_adjacent(rb, index - 1);
1181
1182                 if (tmp == REF_CONTIG_RIGHT) {
1183                         if (ret == REF_CONTIG_RIGHT)
1184                                 ret = REF_CONTIG_LEFTRIGHT;
1185                         else
1186                                 ret = REF_CONTIG_LEFT;
1187                 }
1188         }
1189
1190         return ret;
1191 }
1192
1193 static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block *rb,
1194                                            int index)
1195 {
1196         BUG_ON(rb->rf_records.rl_recs[index].r_refcount !=
1197                rb->rf_records.rl_recs[index+1].r_refcount);
1198
1199         le32_add_cpu(&rb->rf_records.rl_recs[index].r_clusters,
1200                      le32_to_cpu(rb->rf_records.rl_recs[index+1].r_clusters));
1201
1202         if (index < le16_to_cpu(rb->rf_records.rl_used) - 2)
1203                 memmove(&rb->rf_records.rl_recs[index + 1],
1204                         &rb->rf_records.rl_recs[index + 2],
1205                         sizeof(struct ocfs2_refcount_rec) *
1206                         (le16_to_cpu(rb->rf_records.rl_used) - index - 2));
1207
1208         memset(&rb->rf_records.rl_recs[le16_to_cpu(rb->rf_records.rl_used) - 1],
1209                0, sizeof(struct ocfs2_refcount_rec));
1210         le16_add_cpu(&rb->rf_records.rl_used, -1);
1211 }
1212
1213 /*
1214  * Merge the refcount rec if we are contiguous with the adjacent recs.
1215  */
1216 static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
1217                                      int index)
1218 {
1219         enum ocfs2_ref_rec_contig contig =
1220                                 ocfs2_refcount_rec_contig(rb, index);
1221
1222         if (contig == REF_CONTIG_NONE)
1223                 return;
1224
1225         if (contig == REF_CONTIG_LEFT || contig == REF_CONTIG_LEFTRIGHT) {
1226                 BUG_ON(index == 0);
1227                 index--;
1228         }
1229
1230         ocfs2_rotate_refcount_rec_left(rb, index);
1231
1232         if (contig == REF_CONTIG_LEFTRIGHT)
1233                 ocfs2_rotate_refcount_rec_left(rb, index);
1234 }
1235
1236 /*
1237  * Change the refcount indexed by "index" in ref_bh.
1238  * If refcount reaches 0, remove it.
1239  */
1240 static int ocfs2_change_refcount_rec(handle_t *handle,
1241                                      struct ocfs2_caching_info *ci,
1242                                      struct buffer_head *ref_leaf_bh,
1243                                      int index, int merge, int change)
1244 {
1245         int ret;
1246         struct ocfs2_refcount_block *rb =
1247                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1248         struct ocfs2_refcount_list *rl = &rb->rf_records;
1249         struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
1250
1251         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1252                                       OCFS2_JOURNAL_ACCESS_WRITE);
1253         if (ret) {
1254                 mlog_errno(ret);
1255                 goto out;
1256         }
1257
1258         mlog(0, "change index %d, old count %u, change %d\n", index,
1259              le32_to_cpu(rec->r_refcount), change);
1260         le32_add_cpu(&rec->r_refcount, change);
1261
1262         if (!rec->r_refcount) {
1263                 if (index != le16_to_cpu(rl->rl_used) - 1) {
1264                         memmove(rec, rec + 1,
1265                                 (le16_to_cpu(rl->rl_used) - index - 1) *
1266                                 sizeof(struct ocfs2_refcount_rec));
1267                         memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
1268                                0, sizeof(struct ocfs2_refcount_rec));
1269                 }
1270
1271                 le16_add_cpu(&rl->rl_used, -1);
1272         } else if (merge)
1273                 ocfs2_refcount_rec_merge(rb, index);
1274
1275         ocfs2_journal_dirty(handle, ref_leaf_bh);
1276 out:
1277         return ret;
1278 }
1279
1280 static int ocfs2_expand_inline_ref_root(handle_t *handle,
1281                                         struct ocfs2_caching_info *ci,
1282                                         struct buffer_head *ref_root_bh,
1283                                         struct buffer_head **ref_leaf_bh,
1284                                         struct ocfs2_alloc_context *meta_ac)
1285 {
1286         int ret;
1287         u16 suballoc_bit_start;
1288         u32 num_got;
1289         u64 suballoc_loc, blkno;
1290         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1291         struct buffer_head *new_bh = NULL;
1292         struct ocfs2_refcount_block *new_rb;
1293         struct ocfs2_refcount_block *root_rb =
1294                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1295
1296         ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1297                                       OCFS2_JOURNAL_ACCESS_WRITE);
1298         if (ret) {
1299                 mlog_errno(ret);
1300                 goto out;
1301         }
1302
1303         ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1304                                    &suballoc_bit_start, &num_got,
1305                                    &blkno);
1306         if (ret) {
1307                 mlog_errno(ret);
1308                 goto out;
1309         }
1310
1311         new_bh = sb_getblk(sb, blkno);
1312         if (new_bh == NULL) {
1313                 ret = -EIO;
1314                 mlog_errno(ret);
1315                 goto out;
1316         }
1317         ocfs2_set_new_buffer_uptodate(ci, new_bh);
1318
1319         ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1320                                       OCFS2_JOURNAL_ACCESS_CREATE);
1321         if (ret) {
1322                 mlog_errno(ret);
1323                 goto out;
1324         }
1325
1326         /*
1327          * Initialize ocfs2_refcount_block.
1328          * It should contain the same information as the old root.
1329          * so just memcpy it and change the corresponding field.
1330          */
1331         memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
1332
1333         new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1334         new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1335         new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1336         new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1337         new_rb->rf_blkno = cpu_to_le64(blkno);
1338         new_rb->rf_cpos = cpu_to_le32(0);
1339         new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1340         new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1341         ocfs2_journal_dirty(handle, new_bh);
1342
1343         /* Now change the root. */
1344         memset(&root_rb->rf_list, 0, sb->s_blocksize -
1345                offsetof(struct ocfs2_refcount_block, rf_list));
1346         root_rb->rf_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_rb(sb));
1347         root_rb->rf_clusters = cpu_to_le32(1);
1348         root_rb->rf_list.l_next_free_rec = cpu_to_le16(1);
1349         root_rb->rf_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
1350         root_rb->rf_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
1351         root_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_TREE_FL);
1352
1353         ocfs2_journal_dirty(handle, ref_root_bh);
1354
1355         mlog(0, "new leaf block %llu, used %u\n", (unsigned long long)blkno,
1356              le16_to_cpu(new_rb->rf_records.rl_used));
1357
1358         *ref_leaf_bh = new_bh;
1359         new_bh = NULL;
1360 out:
1361         brelse(new_bh);
1362         return ret;
1363 }
1364
1365 static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec *prev,
1366                                            struct ocfs2_refcount_rec *next)
1367 {
1368         if (ocfs2_get_ref_rec_low_cpos(prev) + le32_to_cpu(prev->r_clusters) <=
1369                 ocfs2_get_ref_rec_low_cpos(next))
1370                 return 1;
1371
1372         return 0;
1373 }
1374
1375 static int cmp_refcount_rec_by_low_cpos(const void *a, const void *b)
1376 {
1377         const struct ocfs2_refcount_rec *l = a, *r = b;
1378         u32 l_cpos = ocfs2_get_ref_rec_low_cpos(l);
1379         u32 r_cpos = ocfs2_get_ref_rec_low_cpos(r);
1380
1381         if (l_cpos > r_cpos)
1382                 return 1;
1383         if (l_cpos < r_cpos)
1384                 return -1;
1385         return 0;
1386 }
1387
1388 static int cmp_refcount_rec_by_cpos(const void *a, const void *b)
1389 {
1390         const struct ocfs2_refcount_rec *l = a, *r = b;
1391         u64 l_cpos = le64_to_cpu(l->r_cpos);
1392         u64 r_cpos = le64_to_cpu(r->r_cpos);
1393
1394         if (l_cpos > r_cpos)
1395                 return 1;
1396         if (l_cpos < r_cpos)
1397                 return -1;
1398         return 0;
1399 }
1400
1401 static void swap_refcount_rec(void *a, void *b, int size)
1402 {
1403         struct ocfs2_refcount_rec *l = a, *r = b, tmp;
1404
1405         tmp = *(struct ocfs2_refcount_rec *)l;
1406         *(struct ocfs2_refcount_rec *)l =
1407                         *(struct ocfs2_refcount_rec *)r;
1408         *(struct ocfs2_refcount_rec *)r = tmp;
1409 }
1410
1411 /*
1412  * The refcount cpos are ordered by their 64bit cpos,
1413  * But we will use the low 32 bit to be the e_cpos in the b-tree.
1414  * So we need to make sure that this pos isn't intersected with others.
1415  *
1416  * Note: The refcount block is already sorted by their low 32 bit cpos,
1417  *       So just try the middle pos first, and we will exit when we find
1418  *       the good position.
1419  */
1420 static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
1421                                          u32 *split_pos, int *split_index)
1422 {
1423         int num_used = le16_to_cpu(rl->rl_used);
1424         int delta, middle = num_used / 2;
1425
1426         for (delta = 0; delta < middle; delta++) {
1427                 /* Let's check delta earlier than middle */
1428                 if (ocfs2_refcount_rec_no_intersect(
1429                                         &rl->rl_recs[middle - delta - 1],
1430                                         &rl->rl_recs[middle - delta])) {
1431                         *split_index = middle - delta;
1432                         break;
1433                 }
1434
1435                 /* For even counts, don't walk off the end */
1436                 if ((middle + delta + 1) == num_used)
1437                         continue;
1438
1439                 /* Now try delta past middle */
1440                 if (ocfs2_refcount_rec_no_intersect(
1441                                         &rl->rl_recs[middle + delta],
1442                                         &rl->rl_recs[middle + delta + 1])) {
1443                         *split_index = middle + delta + 1;
1444                         break;
1445                 }
1446         }
1447
1448         if (delta >= middle)
1449                 return -ENOSPC;
1450
1451         *split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
1452         return 0;
1453 }
1454
1455 static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
1456                                             struct buffer_head *new_bh,
1457                                             u32 *split_cpos)
1458 {
1459         int split_index = 0, num_moved, ret;
1460         u32 cpos = 0;
1461         struct ocfs2_refcount_block *rb =
1462                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1463         struct ocfs2_refcount_list *rl = &rb->rf_records;
1464         struct ocfs2_refcount_block *new_rb =
1465                         (struct ocfs2_refcount_block *)new_bh->b_data;
1466         struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
1467
1468         mlog(0, "split old leaf refcount block %llu, count = %u, used = %u\n",
1469              (unsigned long long)ref_leaf_bh->b_blocknr,
1470              le32_to_cpu(rl->rl_count), le32_to_cpu(rl->rl_used));
1471
1472         /*
1473          * XXX: Improvement later.
1474          * If we know all the high 32 bit cpos is the same, no need to sort.
1475          *
1476          * In order to make the whole process safe, we do:
1477          * 1. sort the entries by their low 32 bit cpos first so that we can
1478          *    find the split cpos easily.
1479          * 2. call ocfs2_insert_extent to insert the new refcount block.
1480          * 3. move the refcount rec to the new block.
1481          * 4. sort the entries by their 64 bit cpos.
1482          * 5. dirty the new_rb and rb.
1483          */
1484         sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1485              sizeof(struct ocfs2_refcount_rec),
1486              cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
1487
1488         ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
1489         if (ret) {
1490                 mlog_errno(ret);
1491                 return ret;
1492         }
1493
1494         new_rb->rf_cpos = cpu_to_le32(cpos);
1495
1496         /* move refcount records starting from split_index to the new block. */
1497         num_moved = le16_to_cpu(rl->rl_used) - split_index;
1498         memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
1499                num_moved * sizeof(struct ocfs2_refcount_rec));
1500
1501         /*ok, remove the entries we just moved over to the other block. */
1502         memset(&rl->rl_recs[split_index], 0,
1503                num_moved * sizeof(struct ocfs2_refcount_rec));
1504
1505         /* change old and new rl_used accordingly. */
1506         le16_add_cpu(&rl->rl_used, -num_moved);
1507         new_rl->rl_used = cpu_to_le16(num_moved);
1508
1509         sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1510              sizeof(struct ocfs2_refcount_rec),
1511              cmp_refcount_rec_by_cpos, swap_refcount_rec);
1512
1513         sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
1514              sizeof(struct ocfs2_refcount_rec),
1515              cmp_refcount_rec_by_cpos, swap_refcount_rec);
1516
1517         *split_cpos = cpos;
1518         return 0;
1519 }
1520
1521 static int ocfs2_new_leaf_refcount_block(handle_t *handle,
1522                                          struct ocfs2_caching_info *ci,
1523                                          struct buffer_head *ref_root_bh,
1524                                          struct buffer_head *ref_leaf_bh,
1525                                          struct ocfs2_alloc_context *meta_ac)
1526 {
1527         int ret;
1528         u16 suballoc_bit_start;
1529         u32 num_got, new_cpos;
1530         u64 suballoc_loc, blkno;
1531         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1532         struct ocfs2_refcount_block *root_rb =
1533                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1534         struct buffer_head *new_bh = NULL;
1535         struct ocfs2_refcount_block *new_rb;
1536         struct ocfs2_extent_tree ref_et;
1537
1538         BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
1539
1540         ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1541                                       OCFS2_JOURNAL_ACCESS_WRITE);
1542         if (ret) {
1543                 mlog_errno(ret);
1544                 goto out;
1545         }
1546
1547         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1548                                       OCFS2_JOURNAL_ACCESS_WRITE);
1549         if (ret) {
1550                 mlog_errno(ret);
1551                 goto out;
1552         }
1553
1554         ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1555                                    &suballoc_bit_start, &num_got,
1556                                    &blkno);
1557         if (ret) {
1558                 mlog_errno(ret);
1559                 goto out;
1560         }
1561
1562         new_bh = sb_getblk(sb, blkno);
1563         if (new_bh == NULL) {
1564                 ret = -EIO;
1565                 mlog_errno(ret);
1566                 goto out;
1567         }
1568         ocfs2_set_new_buffer_uptodate(ci, new_bh);
1569
1570         ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1571                                       OCFS2_JOURNAL_ACCESS_CREATE);
1572         if (ret) {
1573                 mlog_errno(ret);
1574                 goto out;
1575         }
1576
1577         /* Initialize ocfs2_refcount_block. */
1578         new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1579         memset(new_rb, 0, sb->s_blocksize);
1580         strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
1581         new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1582         new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1583         new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1584         new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1585         new_rb->rf_blkno = cpu_to_le64(blkno);
1586         new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1587         new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1588         new_rb->rf_records.rl_count =
1589                                 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
1590         new_rb->rf_generation = root_rb->rf_generation;
1591
1592         ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
1593         if (ret) {
1594                 mlog_errno(ret);
1595                 goto out;
1596         }
1597
1598         ocfs2_journal_dirty(handle, ref_leaf_bh);
1599         ocfs2_journal_dirty(handle, new_bh);
1600
1601         ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
1602
1603         mlog(0, "insert new leaf block %llu at %u\n",
1604              (unsigned long long)new_bh->b_blocknr, new_cpos);
1605
1606         /* Insert the new leaf block with the specific offset cpos. */
1607         ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
1608                                   1, 0, meta_ac);
1609         if (ret)
1610                 mlog_errno(ret);
1611
1612 out:
1613         brelse(new_bh);
1614         return ret;
1615 }
1616
1617 static int ocfs2_expand_refcount_tree(handle_t *handle,
1618                                       struct ocfs2_caching_info *ci,
1619                                       struct buffer_head *ref_root_bh,
1620                                       struct buffer_head *ref_leaf_bh,
1621                                       struct ocfs2_alloc_context *meta_ac)
1622 {
1623         int ret;
1624         struct buffer_head *expand_bh = NULL;
1625
1626         if (ref_root_bh == ref_leaf_bh) {
1627                 /*
1628                  * the old root bh hasn't been expanded to a b-tree,
1629                  * so expand it first.
1630                  */
1631                 ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
1632                                                    &expand_bh, meta_ac);
1633                 if (ret) {
1634                         mlog_errno(ret);
1635                         goto out;
1636                 }
1637         } else {
1638                 expand_bh = ref_leaf_bh;
1639                 get_bh(expand_bh);
1640         }
1641
1642
1643         /* Now add a new refcount block into the tree.*/
1644         ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
1645                                             expand_bh, meta_ac);
1646         if (ret)
1647                 mlog_errno(ret);
1648 out:
1649         brelse(expand_bh);
1650         return ret;
1651 }
1652
1653 /*
1654  * Adjust the extent rec in b-tree representing ref_leaf_bh.
1655  *
1656  * Only called when we have inserted a new refcount rec at index 0
1657  * which means ocfs2_extent_rec.e_cpos may need some change.
1658  */
1659 static int ocfs2_adjust_refcount_rec(handle_t *handle,
1660                                      struct ocfs2_caching_info *ci,
1661                                      struct buffer_head *ref_root_bh,
1662                                      struct buffer_head *ref_leaf_bh,
1663                                      struct ocfs2_refcount_rec *rec)
1664 {
1665         int ret = 0, i;
1666         u32 new_cpos, old_cpos;
1667         struct ocfs2_path *path = NULL;
1668         struct ocfs2_extent_tree et;
1669         struct ocfs2_refcount_block *rb =
1670                 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1671         struct ocfs2_extent_list *el;
1672
1673         if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
1674                 goto out;
1675
1676         rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1677         old_cpos = le32_to_cpu(rb->rf_cpos);
1678         new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
1679         if (old_cpos <= new_cpos)
1680                 goto out;
1681
1682         ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1683
1684         path = ocfs2_new_path_from_et(&et);
1685         if (!path) {
1686                 ret = -ENOMEM;
1687                 mlog_errno(ret);
1688                 goto out;
1689         }
1690
1691         ret = ocfs2_find_path(ci, path, old_cpos);
1692         if (ret) {
1693                 mlog_errno(ret);
1694                 goto out;
1695         }
1696
1697         /*
1698          * 2 more credits, one for the leaf refcount block, one for
1699          * the extent block contains the extent rec.
1700          */
1701         ret = ocfs2_extend_trans(handle, 2);
1702         if (ret < 0) {
1703                 mlog_errno(ret);
1704                 goto out;
1705         }
1706
1707         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1708                                       OCFS2_JOURNAL_ACCESS_WRITE);
1709         if (ret < 0) {
1710                 mlog_errno(ret);
1711                 goto out;
1712         }
1713
1714         ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
1715                                       OCFS2_JOURNAL_ACCESS_WRITE);
1716         if (ret < 0) {
1717                 mlog_errno(ret);
1718                 goto out;
1719         }
1720
1721         /* change the leaf extent block first. */
1722         el = path_leaf_el(path);
1723
1724         for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
1725                 if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
1726                         break;
1727
1728         BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
1729
1730         el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1731
1732         /* change the r_cpos in the leaf block. */
1733         rb->rf_cpos = cpu_to_le32(new_cpos);
1734
1735         ocfs2_journal_dirty(handle, path_leaf_bh(path));
1736         ocfs2_journal_dirty(handle, ref_leaf_bh);
1737
1738 out:
1739         ocfs2_free_path(path);
1740         return ret;
1741 }
1742
1743 static int ocfs2_insert_refcount_rec(handle_t *handle,
1744                                      struct ocfs2_caching_info *ci,
1745                                      struct buffer_head *ref_root_bh,
1746                                      struct buffer_head *ref_leaf_bh,
1747                                      struct ocfs2_refcount_rec *rec,
1748                                      int index, int merge,
1749                                      struct ocfs2_alloc_context *meta_ac)
1750 {
1751         int ret;
1752         struct ocfs2_refcount_block *rb =
1753                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1754         struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1755         struct buffer_head *new_bh = NULL;
1756
1757         BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1758
1759         if (rf_list->rl_used == rf_list->rl_count) {
1760                 u64 cpos = le64_to_cpu(rec->r_cpos);
1761                 u32 len = le32_to_cpu(rec->r_clusters);
1762
1763                 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1764                                                  ref_leaf_bh, meta_ac);
1765                 if (ret) {
1766                         mlog_errno(ret);
1767                         goto out;
1768                 }
1769
1770                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1771                                              cpos, len, NULL, &index,
1772                                              &new_bh);
1773                 if (ret) {
1774                         mlog_errno(ret);
1775                         goto out;
1776                 }
1777
1778                 ref_leaf_bh = new_bh;
1779                 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1780                 rf_list = &rb->rf_records;
1781         }
1782
1783         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1784                                       OCFS2_JOURNAL_ACCESS_WRITE);
1785         if (ret) {
1786                 mlog_errno(ret);
1787                 goto out;
1788         }
1789
1790         if (index < le16_to_cpu(rf_list->rl_used))
1791                 memmove(&rf_list->rl_recs[index + 1],
1792                         &rf_list->rl_recs[index],
1793                         (le16_to_cpu(rf_list->rl_used) - index) *
1794                          sizeof(struct ocfs2_refcount_rec));
1795
1796         mlog(0, "insert refcount record start %llu, len %u, count %u "
1797              "to leaf block %llu at index %d\n",
1798              (unsigned long long)le64_to_cpu(rec->r_cpos),
1799              le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount),
1800              (unsigned long long)ref_leaf_bh->b_blocknr, index);
1801
1802         rf_list->rl_recs[index] = *rec;
1803
1804         le16_add_cpu(&rf_list->rl_used, 1);
1805
1806         if (merge)
1807                 ocfs2_refcount_rec_merge(rb, index);
1808
1809         ocfs2_journal_dirty(handle, ref_leaf_bh);
1810
1811         if (index == 0) {
1812                 ret = ocfs2_adjust_refcount_rec(handle, ci,
1813                                                 ref_root_bh,
1814                                                 ref_leaf_bh, rec);
1815                 if (ret)
1816                         mlog_errno(ret);
1817         }
1818 out:
1819         brelse(new_bh);
1820         return ret;
1821 }
1822
1823 /*
1824  * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1825  * This is much simple than our b-tree code.
1826  * split_rec is the new refcount rec we want to insert.
1827  * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1828  * increase refcount or decrease a refcount to non-zero).
1829  * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1830  * rec( in case we decrease a refcount to zero).
1831  */
1832 static int ocfs2_split_refcount_rec(handle_t *handle,
1833                                     struct ocfs2_caching_info *ci,
1834                                     struct buffer_head *ref_root_bh,
1835                                     struct buffer_head *ref_leaf_bh,
1836                                     struct ocfs2_refcount_rec *split_rec,
1837                                     int index, int merge,
1838                                     struct ocfs2_alloc_context *meta_ac,
1839                                     struct ocfs2_cached_dealloc_ctxt *dealloc)
1840 {
1841         int ret, recs_need;
1842         u32 len;
1843         struct ocfs2_refcount_block *rb =
1844                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1845         struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1846         struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
1847         struct ocfs2_refcount_rec *tail_rec = NULL;
1848         struct buffer_head *new_bh = NULL;
1849
1850         BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1851
1852         mlog(0, "original r_pos %llu, cluster %u, split %llu, cluster %u\n",
1853              le64_to_cpu(orig_rec->r_cpos), le32_to_cpu(orig_rec->r_clusters),
1854              le64_to_cpu(split_rec->r_cpos),
1855              le32_to_cpu(split_rec->r_clusters));
1856
1857         /*
1858          * If we just need to split the header or tail clusters,
1859          * no more recs are needed, just split is OK.
1860          * Otherwise we at least need one new recs.
1861          */
1862         if (!split_rec->r_refcount &&
1863             (split_rec->r_cpos == orig_rec->r_cpos ||
1864              le64_to_cpu(split_rec->r_cpos) +
1865              le32_to_cpu(split_rec->r_clusters) ==
1866              le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1867                 recs_need = 0;
1868         else
1869                 recs_need = 1;
1870
1871         /*
1872          * We need one more rec if we split in the middle and the new rec have
1873          * some refcount in it.
1874          */
1875         if (split_rec->r_refcount &&
1876             (split_rec->r_cpos != orig_rec->r_cpos &&
1877              le64_to_cpu(split_rec->r_cpos) +
1878              le32_to_cpu(split_rec->r_clusters) !=
1879              le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1880                 recs_need++;
1881
1882         /* If the leaf block don't have enough record, expand it. */
1883         if (le16_to_cpu(rf_list->rl_used) + recs_need >
1884                                          le16_to_cpu(rf_list->rl_count)) {
1885                 struct ocfs2_refcount_rec tmp_rec;
1886                 u64 cpos = le64_to_cpu(orig_rec->r_cpos);
1887                 len = le32_to_cpu(orig_rec->r_clusters);
1888                 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1889                                                  ref_leaf_bh, meta_ac);
1890                 if (ret) {
1891                         mlog_errno(ret);
1892                         goto out;
1893                 }
1894
1895                 /*
1896                  * We have to re-get it since now cpos may be moved to
1897                  * another leaf block.
1898                  */
1899                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1900                                              cpos, len, &tmp_rec, &index,
1901                                              &new_bh);
1902                 if (ret) {
1903                         mlog_errno(ret);
1904                         goto out;
1905                 }
1906
1907                 ref_leaf_bh = new_bh;
1908                 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1909                 rf_list = &rb->rf_records;
1910                 orig_rec = &rf_list->rl_recs[index];
1911         }
1912
1913         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1914                                       OCFS2_JOURNAL_ACCESS_WRITE);
1915         if (ret) {
1916                 mlog_errno(ret);
1917                 goto out;
1918         }
1919
1920         /*
1921          * We have calculated out how many new records we need and store
1922          * in recs_need, so spare enough space first by moving the records
1923          * after "index" to the end.
1924          */
1925         if (index != le16_to_cpu(rf_list->rl_used) - 1)
1926                 memmove(&rf_list->rl_recs[index + 1 + recs_need],
1927                         &rf_list->rl_recs[index + 1],
1928                         (le16_to_cpu(rf_list->rl_used) - index - 1) *
1929                          sizeof(struct ocfs2_refcount_rec));
1930
1931         len = (le64_to_cpu(orig_rec->r_cpos) +
1932               le32_to_cpu(orig_rec->r_clusters)) -
1933               (le64_to_cpu(split_rec->r_cpos) +
1934               le32_to_cpu(split_rec->r_clusters));
1935
1936         /*
1937          * If we have "len", the we will split in the tail and move it
1938          * to the end of the space we have just spared.
1939          */
1940         if (len) {
1941                 tail_rec = &rf_list->rl_recs[index + recs_need];
1942
1943                 memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
1944                 le64_add_cpu(&tail_rec->r_cpos,
1945                              le32_to_cpu(tail_rec->r_clusters) - len);
1946                 tail_rec->r_clusters = cpu_to_le32(len);
1947         }
1948
1949         /*
1950          * If the split pos isn't the same as the original one, we need to
1951          * split in the head.
1952          *
1953          * Note: We have the chance that split_rec.r_refcount = 0,
1954          * recs_need = 0 and len > 0, which means we just cut the head from
1955          * the orig_rec and in that case we have done some modification in
1956          * orig_rec above, so the check for r_cpos is faked.
1957          */
1958         if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
1959                 len = le64_to_cpu(split_rec->r_cpos) -
1960                       le64_to_cpu(orig_rec->r_cpos);
1961                 orig_rec->r_clusters = cpu_to_le32(len);
1962                 index++;
1963         }
1964
1965         le16_add_cpu(&rf_list->rl_used, recs_need);
1966
1967         if (split_rec->r_refcount) {
1968                 rf_list->rl_recs[index] = *split_rec;
1969                 mlog(0, "insert refcount record start %llu, len %u, count %u "
1970                      "to leaf block %llu at index %d\n",
1971                      (unsigned long long)le64_to_cpu(split_rec->r_cpos),
1972                      le32_to_cpu(split_rec->r_clusters),
1973                      le32_to_cpu(split_rec->r_refcount),
1974                      (unsigned long long)ref_leaf_bh->b_blocknr, index);
1975
1976                 if (merge)
1977                         ocfs2_refcount_rec_merge(rb, index);
1978         }
1979
1980         ocfs2_journal_dirty(handle, ref_leaf_bh);
1981
1982 out:
1983         brelse(new_bh);
1984         return ret;
1985 }
1986
1987 static int __ocfs2_increase_refcount(handle_t *handle,
1988                                      struct ocfs2_caching_info *ci,
1989                                      struct buffer_head *ref_root_bh,
1990                                      u64 cpos, u32 len, int merge,
1991                                      struct ocfs2_alloc_context *meta_ac,
1992                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
1993 {
1994         int ret = 0, index;
1995         struct buffer_head *ref_leaf_bh = NULL;
1996         struct ocfs2_refcount_rec rec;
1997         unsigned int set_len = 0;
1998
1999         mlog(0, "Tree owner %llu, add refcount start %llu, len %u\n",
2000              (unsigned long long)ocfs2_metadata_cache_owner(ci),
2001              (unsigned long long)cpos, len);
2002
2003         while (len) {
2004                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2005                                              cpos, len, &rec, &index,
2006                                              &ref_leaf_bh);
2007                 if (ret) {
2008                         mlog_errno(ret);
2009                         goto out;
2010                 }
2011
2012                 set_len = le32_to_cpu(rec.r_clusters);
2013
2014                 /*
2015                  * Here we may meet with 3 situations:
2016                  *
2017                  * 1. If we find an already existing record, and the length
2018                  *    is the same, cool, we just need to increase the r_refcount
2019                  *    and it is OK.
2020                  * 2. If we find a hole, just insert it with r_refcount = 1.
2021                  * 3. If we are in the middle of one extent record, split
2022                  *    it.
2023                  */
2024                 if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
2025                     set_len <= len) {
2026                         mlog(0, "increase refcount rec, start %llu, len %u, "
2027                              "count %u\n", (unsigned long long)cpos, set_len,
2028                              le32_to_cpu(rec.r_refcount));
2029                         ret = ocfs2_change_refcount_rec(handle, ci,
2030                                                         ref_leaf_bh, index,
2031                                                         merge, 1);
2032                         if (ret) {
2033                                 mlog_errno(ret);
2034                                 goto out;
2035                         }
2036                 } else if (!rec.r_refcount) {
2037                         rec.r_refcount = cpu_to_le32(1);
2038
2039                         mlog(0, "insert refcount rec, start %llu, len %u\n",
2040                              (unsigned long long)le64_to_cpu(rec.r_cpos),
2041                              set_len);
2042                         ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
2043                                                         ref_leaf_bh,
2044                                                         &rec, index,
2045                                                         merge, meta_ac);
2046                         if (ret) {
2047                                 mlog_errno(ret);
2048                                 goto out;
2049                         }
2050                 } else  {
2051                         set_len = min((u64)(cpos + len),
2052                                       le64_to_cpu(rec.r_cpos) + set_len) - cpos;
2053                         rec.r_cpos = cpu_to_le64(cpos);
2054                         rec.r_clusters = cpu_to_le32(set_len);
2055                         le32_add_cpu(&rec.r_refcount, 1);
2056
2057                         mlog(0, "split refcount rec, start %llu, "
2058                              "len %u, count %u\n",
2059                              (unsigned long long)le64_to_cpu(rec.r_cpos),
2060                              set_len, le32_to_cpu(rec.r_refcount));
2061                         ret = ocfs2_split_refcount_rec(handle, ci,
2062                                                        ref_root_bh, ref_leaf_bh,
2063                                                        &rec, index, merge,
2064                                                        meta_ac, dealloc);
2065                         if (ret) {
2066                                 mlog_errno(ret);
2067                                 goto out;
2068                         }
2069                 }
2070
2071                 cpos += set_len;
2072                 len -= set_len;
2073                 brelse(ref_leaf_bh);
2074                 ref_leaf_bh = NULL;
2075         }
2076
2077 out:
2078         brelse(ref_leaf_bh);
2079         return ret;
2080 }
2081
2082 static int ocfs2_remove_refcount_extent(handle_t *handle,
2083                                 struct ocfs2_caching_info *ci,
2084                                 struct buffer_head *ref_root_bh,
2085                                 struct buffer_head *ref_leaf_bh,
2086                                 struct ocfs2_alloc_context *meta_ac,
2087                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
2088 {
2089         int ret;
2090         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2091         struct ocfs2_refcount_block *rb =
2092                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2093         struct ocfs2_extent_tree et;
2094
2095         BUG_ON(rb->rf_records.rl_used);
2096
2097         ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2098         ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
2099                                   1, meta_ac, dealloc);
2100         if (ret) {
2101                 mlog_errno(ret);
2102                 goto out;
2103         }
2104
2105         ocfs2_remove_from_cache(ci, ref_leaf_bh);
2106
2107         /*
2108          * add the freed block to the dealloc so that it will be freed
2109          * when we run dealloc.
2110          */
2111         ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
2112                                         le16_to_cpu(rb->rf_suballoc_slot),
2113                                         le64_to_cpu(rb->rf_suballoc_loc),
2114                                         le64_to_cpu(rb->rf_blkno),
2115                                         le16_to_cpu(rb->rf_suballoc_bit));
2116         if (ret) {
2117                 mlog_errno(ret);
2118                 goto out;
2119         }
2120
2121         ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
2122                                       OCFS2_JOURNAL_ACCESS_WRITE);
2123         if (ret) {
2124                 mlog_errno(ret);
2125                 goto out;
2126         }
2127
2128         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2129
2130         le32_add_cpu(&rb->rf_clusters, -1);
2131
2132         /*
2133          * check whether we need to restore the root refcount block if
2134          * there is no leaf extent block at atll.
2135          */
2136         if (!rb->rf_list.l_next_free_rec) {
2137                 BUG_ON(rb->rf_clusters);
2138
2139                 mlog(0, "reset refcount tree root %llu to be a record block.\n",
2140                      (unsigned long long)ref_root_bh->b_blocknr);
2141
2142                 rb->rf_flags = 0;
2143                 rb->rf_parent = 0;
2144                 rb->rf_cpos = 0;
2145                 memset(&rb->rf_records, 0, sb->s_blocksize -
2146                        offsetof(struct ocfs2_refcount_block, rf_records));
2147                 rb->rf_records.rl_count =
2148                                 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
2149         }
2150
2151         ocfs2_journal_dirty(handle, ref_root_bh);
2152
2153 out:
2154         return ret;
2155 }
2156
2157 int ocfs2_increase_refcount(handle_t *handle,
2158                             struct ocfs2_caching_info *ci,
2159                             struct buffer_head *ref_root_bh,
2160                             u64 cpos, u32 len,
2161                             struct ocfs2_alloc_context *meta_ac,
2162                             struct ocfs2_cached_dealloc_ctxt *dealloc)
2163 {
2164         return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
2165                                          cpos, len, 1,
2166                                          meta_ac, dealloc);
2167 }
2168
2169 static int ocfs2_decrease_refcount_rec(handle_t *handle,
2170                                 struct ocfs2_caching_info *ci,
2171                                 struct buffer_head *ref_root_bh,
2172                                 struct buffer_head *ref_leaf_bh,
2173                                 int index, u64 cpos, unsigned int len,
2174                                 struct ocfs2_alloc_context *meta_ac,
2175                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
2176 {
2177         int ret;
2178         struct ocfs2_refcount_block *rb =
2179                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2180         struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
2181
2182         BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
2183         BUG_ON(cpos + len >
2184                le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
2185
2186         if (cpos == le64_to_cpu(rec->r_cpos) &&
2187             len == le32_to_cpu(rec->r_clusters))
2188                 ret = ocfs2_change_refcount_rec(handle, ci,
2189                                                 ref_leaf_bh, index, 1, -1);
2190         else {
2191                 struct ocfs2_refcount_rec split = *rec;
2192                 split.r_cpos = cpu_to_le64(cpos);
2193                 split.r_clusters = cpu_to_le32(len);
2194
2195                 le32_add_cpu(&split.r_refcount, -1);
2196
2197                 mlog(0, "split refcount rec, start %llu, "
2198                      "len %u, count %u, original start %llu, len %u\n",
2199                      (unsigned long long)le64_to_cpu(split.r_cpos),
2200                      len, le32_to_cpu(split.r_refcount),
2201                      (unsigned long long)le64_to_cpu(rec->r_cpos),
2202                      le32_to_cpu(rec->r_clusters));
2203                 ret = ocfs2_split_refcount_rec(handle, ci,
2204                                                ref_root_bh, ref_leaf_bh,
2205                                                &split, index, 1,
2206                                                meta_ac, dealloc);
2207         }
2208
2209         if (ret) {
2210                 mlog_errno(ret);
2211                 goto out;
2212         }
2213
2214         /* Remove the leaf refcount block if it contains no refcount record. */
2215         if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
2216                 ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
2217                                                    ref_leaf_bh, meta_ac,
2218                                                    dealloc);
2219                 if (ret)
2220                         mlog_errno(ret);
2221         }
2222
2223 out:
2224         return ret;
2225 }
2226
2227 static int __ocfs2_decrease_refcount(handle_t *handle,
2228                                      struct ocfs2_caching_info *ci,
2229                                      struct buffer_head *ref_root_bh,
2230                                      u64 cpos, u32 len,
2231                                      struct ocfs2_alloc_context *meta_ac,
2232                                      struct ocfs2_cached_dealloc_ctxt *dealloc,
2233                                      int delete)
2234 {
2235         int ret = 0, index = 0;
2236         struct ocfs2_refcount_rec rec;
2237         unsigned int r_count = 0, r_len;
2238         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2239         struct buffer_head *ref_leaf_bh = NULL;
2240
2241         mlog(0, "Tree owner %llu, decrease refcount start %llu, "
2242              "len %u, delete %u\n",
2243              (unsigned long long)ocfs2_metadata_cache_owner(ci),
2244              (unsigned long long)cpos, len, delete);
2245
2246         while (len) {
2247                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2248                                              cpos, len, &rec, &index,
2249                                              &ref_leaf_bh);
2250                 if (ret) {
2251                         mlog_errno(ret);
2252                         goto out;
2253                 }
2254
2255                 r_count = le32_to_cpu(rec.r_refcount);
2256                 BUG_ON(r_count == 0);
2257                 if (!delete)
2258                         BUG_ON(r_count > 1);
2259
2260                 r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
2261                               le32_to_cpu(rec.r_clusters)) - cpos;
2262
2263                 ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
2264                                                   ref_leaf_bh, index,
2265                                                   cpos, r_len,
2266                                                   meta_ac, dealloc);
2267                 if (ret) {
2268                         mlog_errno(ret);
2269                         goto out;
2270                 }
2271
2272                 if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
2273                         ret = ocfs2_cache_cluster_dealloc(dealloc,
2274                                           ocfs2_clusters_to_blocks(sb, cpos),
2275                                                           r_len);
2276                         if (ret) {
2277                                 mlog_errno(ret);
2278                                 goto out;
2279                         }
2280                 }
2281
2282                 cpos += r_len;
2283                 len -= r_len;
2284                 brelse(ref_leaf_bh);
2285                 ref_leaf_bh = NULL;
2286         }
2287
2288 out:
2289         brelse(ref_leaf_bh);
2290         return ret;
2291 }
2292
2293 /* Caller must hold refcount tree lock. */
2294 int ocfs2_decrease_refcount(struct inode *inode,
2295                             handle_t *handle, u32 cpos, u32 len,
2296                             struct ocfs2_alloc_context *meta_ac,
2297                             struct ocfs2_cached_dealloc_ctxt *dealloc,
2298                             int delete)
2299 {
2300         int ret;
2301         u64 ref_blkno;
2302         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2303         struct buffer_head *ref_root_bh = NULL;
2304         struct ocfs2_refcount_tree *tree;
2305
2306         BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2307
2308         ret = ocfs2_get_refcount_block(inode, &ref_blkno);
2309         if (ret) {
2310                 mlog_errno(ret);
2311                 goto out;
2312         }
2313
2314         ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
2315         if (ret) {
2316                 mlog_errno(ret);
2317                 goto out;
2318         }
2319
2320         ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
2321                                         &ref_root_bh);
2322         if (ret) {
2323                 mlog_errno(ret);
2324                 goto out;
2325         }
2326
2327         ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
2328                                         cpos, len, meta_ac, dealloc, delete);
2329         if (ret)
2330                 mlog_errno(ret);
2331 out:
2332         brelse(ref_root_bh);
2333         return ret;
2334 }
2335
2336 /*
2337  * Mark the already-existing extent at cpos as refcounted for len clusters.
2338  * This adds the refcount extent flag.
2339  *
2340  * If the existing extent is larger than the request, initiate a
2341  * split. An attempt will be made at merging with adjacent extents.
2342  *
2343  * The caller is responsible for passing down meta_ac if we'll need it.
2344  */
2345 static int ocfs2_mark_extent_refcounted(struct inode *inode,
2346                                 struct ocfs2_extent_tree *et,
2347                                 handle_t *handle, u32 cpos,
2348                                 u32 len, u32 phys,
2349                                 struct ocfs2_alloc_context *meta_ac,
2350                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
2351 {
2352         int ret;
2353
2354         mlog(0, "Inode %lu refcount tree cpos %u, len %u, phys cluster %u\n",
2355              inode->i_ino, cpos, len, phys);
2356
2357         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2358                 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2359                             "tree, but the feature bit is not set in the "
2360                             "super block.", inode->i_ino);
2361                 ret = -EROFS;
2362                 goto out;
2363         }
2364
2365         ret = ocfs2_change_extent_flag(handle, et, cpos,
2366                                        len, phys, meta_ac, dealloc,
2367                                        OCFS2_EXT_REFCOUNTED, 0);
2368         if (ret)
2369                 mlog_errno(ret);
2370
2371 out:
2372         return ret;
2373 }
2374
2375 /*
2376  * Given some contiguous physical clusters, calculate what we need
2377  * for modifying their refcount.
2378  */
2379 static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
2380                                             struct ocfs2_caching_info *ci,
2381                                             struct buffer_head *ref_root_bh,
2382                                             u64 start_cpos,
2383                                             u32 clusters,
2384                                             int *meta_add,
2385                                             int *credits)
2386 {
2387         int ret = 0, index, ref_blocks = 0, recs_add = 0;
2388         u64 cpos = start_cpos;
2389         struct ocfs2_refcount_block *rb;
2390         struct ocfs2_refcount_rec rec;
2391         struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
2392         u32 len;
2393
2394         mlog(0, "start_cpos %llu, clusters %u\n",
2395              (unsigned long long)start_cpos, clusters);
2396         while (clusters) {
2397                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2398                                              cpos, clusters, &rec,
2399                                              &index, &ref_leaf_bh);
2400                 if (ret) {
2401                         mlog_errno(ret);
2402                         goto out;
2403                 }
2404
2405                 if (ref_leaf_bh != prev_bh) {
2406                         /*
2407                          * Now we encounter a new leaf block, so calculate
2408                          * whether we need to extend the old leaf.
2409                          */
2410                         if (prev_bh) {
2411                                 rb = (struct ocfs2_refcount_block *)
2412                                                         prev_bh->b_data;
2413
2414                                 if (le64_to_cpu(rb->rf_records.rl_used) +
2415                                     recs_add >
2416                                     le16_to_cpu(rb->rf_records.rl_count))
2417                                         ref_blocks++;
2418                         }
2419
2420                         recs_add = 0;
2421                         *credits += 1;
2422                         brelse(prev_bh);
2423                         prev_bh = ref_leaf_bh;
2424                         get_bh(prev_bh);
2425                 }
2426
2427                 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2428
2429                 mlog(0, "recs_add %d,cpos %llu, clusters %u, rec->r_cpos %llu,"
2430                      "rec->r_clusters %u, rec->r_refcount %u, index %d\n",
2431                      recs_add, (unsigned long long)cpos, clusters,
2432                      (unsigned long long)le64_to_cpu(rec.r_cpos),
2433                      le32_to_cpu(rec.r_clusters),
2434                      le32_to_cpu(rec.r_refcount), index);
2435
2436                 len = min((u64)cpos + clusters, le64_to_cpu(rec.r_cpos) +
2437                           le32_to_cpu(rec.r_clusters)) - cpos;
2438                 /*
2439                  * We record all the records which will be inserted to the
2440                  * same refcount block, so that we can tell exactly whether
2441                  * we need a new refcount block or not.
2442                  *
2443                  * If we will insert a new one, this is easy and only happens
2444                  * during adding refcounted flag to the extent, so we don't
2445                  * have a chance of spliting. We just need one record.
2446                  *
2447                  * If the refcount rec already exists, that would be a little
2448                  * complicated. we may have to:
2449                  * 1) split at the beginning if the start pos isn't aligned.
2450                  *    we need 1 more record in this case.
2451                  * 2) split int the end if the end pos isn't aligned.
2452                  *    we need 1 more record in this case.
2453                  * 3) split in the middle because of file system fragmentation.
2454                  *    we need 2 more records in this case(we can't detect this
2455                  *    beforehand, so always think of the worst case).
2456                  */
2457                 if (rec.r_refcount) {
2458                         recs_add += 2;
2459                         /* Check whether we need a split at the beginning. */
2460                         if (cpos == start_cpos &&
2461                             cpos != le64_to_cpu(rec.r_cpos))
2462                                 recs_add++;
2463
2464                         /* Check whether we need a split in the end. */
2465                         if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
2466                             le32_to_cpu(rec.r_clusters))
2467                                 recs_add++;
2468                 } else
2469                         recs_add++;
2470
2471                 brelse(ref_leaf_bh);
2472                 ref_leaf_bh = NULL;
2473                 clusters -= len;
2474                 cpos += len;
2475         }
2476
2477         if (prev_bh) {
2478                 rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
2479
2480                 if (le64_to_cpu(rb->rf_records.rl_used) + recs_add >
2481                     le16_to_cpu(rb->rf_records.rl_count))
2482                         ref_blocks++;
2483
2484                 *credits += 1;
2485         }
2486
2487         if (!ref_blocks)
2488                 goto out;
2489
2490         mlog(0, "we need ref_blocks %d\n", ref_blocks);
2491         *meta_add += ref_blocks;
2492         *credits += ref_blocks;
2493
2494         /*
2495          * So we may need ref_blocks to insert into the tree.
2496          * That also means we need to change the b-tree and add that number
2497          * of records since we never merge them.
2498          * We need one more block for expansion since the new created leaf
2499          * block is also full and needs split.
2500          */
2501         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2502         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
2503                 struct ocfs2_extent_tree et;
2504
2505                 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2506                 *meta_add += ocfs2_extend_meta_needed(et.et_root_el);
2507                 *credits += ocfs2_calc_extend_credits(sb,
2508                                                       et.et_root_el,
2509                                                       ref_blocks);
2510         } else {
2511                 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
2512                 *meta_add += 1;
2513         }
2514
2515 out:
2516         brelse(ref_leaf_bh);
2517         brelse(prev_bh);
2518         return ret;
2519 }
2520
2521 /*
2522  * For refcount tree, we will decrease some contiguous clusters
2523  * refcount count, so just go through it to see how many blocks
2524  * we gonna touch and whether we need to create new blocks.
2525  *
2526  * Normally the refcount blocks store these refcount should be
2527  * contiguous also, so that we can get the number easily.
2528  * We will at most add split 2 refcount records and 2 more
2529  * refcount blocks, so just check it in a rough way.
2530  *
2531  * Caller must hold refcount tree lock.
2532  */
2533 int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
2534                                           u64 refcount_loc,
2535                                           u64 phys_blkno,
2536                                           u32 clusters,
2537                                           int *credits,
2538                                           int *ref_blocks)
2539 {
2540         int ret;
2541         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2542         struct buffer_head *ref_root_bh = NULL;
2543         struct ocfs2_refcount_tree *tree;
2544         u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
2545
2546         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2547                 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2548                             "tree, but the feature bit is not set in the "
2549                             "super block.", inode->i_ino);
2550                 ret = -EROFS;
2551                 goto out;
2552         }
2553
2554         BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2555
2556         ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
2557                                       refcount_loc, &tree);
2558         if (ret) {
2559                 mlog_errno(ret);
2560                 goto out;
2561         }
2562
2563         ret = ocfs2_read_refcount_block(&tree->rf_ci, refcount_loc,
2564                                         &ref_root_bh);
2565         if (ret) {
2566                 mlog_errno(ret);
2567                 goto out;
2568         }
2569
2570         ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
2571                                                &tree->rf_ci,
2572                                                ref_root_bh,
2573                                                start_cpos, clusters,
2574                                                ref_blocks, credits);
2575         if (ret) {
2576                 mlog_errno(ret);
2577                 goto out;
2578         }
2579
2580         mlog(0, "reserve new metadata %d blocks, credits = %d\n",
2581              *ref_blocks, *credits);
2582
2583 out:
2584         brelse(ref_root_bh);
2585         return ret;
2586 }
2587
2588 #define MAX_CONTIG_BYTES        1048576
2589
2590 static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
2591 {
2592         return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
2593 }
2594
2595 static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
2596 {
2597         return ~(ocfs2_cow_contig_clusters(sb) - 1);
2598 }
2599
2600 /*
2601  * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2602  * find an offset (start + (n * contig_clusters)) that is closest to cpos
2603  * while still being less than or equal to it.
2604  *
2605  * The goal is to break the extent at a multiple of contig_clusters.
2606  */
2607 static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
2608                                                  unsigned int start,
2609                                                  unsigned int cpos)
2610 {
2611         BUG_ON(start > cpos);
2612
2613         return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
2614 }
2615
2616 /*
2617  * Given a cluster count of len, pad it out so that it is a multiple
2618  * of contig_clusters.
2619  */
2620 static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
2621                                                   unsigned int len)
2622 {
2623         unsigned int padded =
2624                 (len + (ocfs2_cow_contig_clusters(sb) - 1)) &
2625                 ocfs2_cow_contig_mask(sb);
2626
2627         /* Did we wrap? */
2628         if (padded < len)
2629                 padded = UINT_MAX;
2630
2631         return padded;
2632 }
2633
2634 /*
2635  * Calculate out the start and number of virtual clusters we need to to CoW.
2636  *
2637  * cpos is vitual start cluster position we want to do CoW in a
2638  * file and write_len is the cluster length.
2639  * max_cpos is the place where we want to stop CoW intentionally.
2640  *
2641  * Normal we will start CoW from the beginning of extent record cotaining cpos.
2642  * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2643  * get good I/O from the resulting extent tree.
2644  */
2645 static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
2646                                            struct ocfs2_extent_list *el,
2647                                            u32 cpos,
2648                                            u32 write_len,
2649                                            u32 max_cpos,
2650                                            u32 *cow_start,
2651                                            u32 *cow_len)
2652 {
2653         int ret = 0;
2654         int tree_height = le16_to_cpu(el->l_tree_depth), i;
2655         struct buffer_head *eb_bh = NULL;
2656         struct ocfs2_extent_block *eb = NULL;
2657         struct ocfs2_extent_rec *rec;
2658         unsigned int want_clusters, rec_end = 0;
2659         int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
2660         int leaf_clusters;
2661
2662         BUG_ON(cpos + write_len > max_cpos);
2663
2664         if (tree_height > 0) {
2665                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
2666                 if (ret) {
2667                         mlog_errno(ret);
2668                         goto out;
2669                 }
2670
2671                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2672                 el = &eb->h_list;
2673
2674                 if (el->l_tree_depth) {
2675                         ocfs2_error(inode->i_sb,
2676                                     "Inode %lu has non zero tree depth in "
2677                                     "leaf block %llu\n", inode->i_ino,
2678                                     (unsigned long long)eb_bh->b_blocknr);
2679                         ret = -EROFS;
2680                         goto out;
2681                 }
2682         }
2683
2684         *cow_len = 0;
2685         for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
2686                 rec = &el->l_recs[i];
2687
2688                 if (ocfs2_is_empty_extent(rec)) {
2689                         mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
2690                                         "index %d\n", inode->i_ino, i);
2691                         continue;
2692                 }
2693
2694                 if (le32_to_cpu(rec->e_cpos) +
2695                     le16_to_cpu(rec->e_leaf_clusters) <= cpos)
2696                         continue;
2697
2698                 if (*cow_len == 0) {
2699                         /*
2700                          * We should find a refcounted record in the
2701                          * first pass.
2702                          */
2703                         BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
2704                         *cow_start = le32_to_cpu(rec->e_cpos);
2705                 }
2706
2707                 /*
2708                  * If we encounter a hole, a non-refcounted record or
2709                  * pass the max_cpos, stop the search.
2710                  */
2711                 if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
2712                     (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
2713                     (max_cpos <= le32_to_cpu(rec->e_cpos)))
2714                         break;
2715
2716                 leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
2717                 rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
2718                 if (rec_end > max_cpos) {
2719                         rec_end = max_cpos;
2720                         leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
2721                 }
2722
2723                 /*
2724                  * How many clusters do we actually need from
2725                  * this extent?  First we see how many we actually
2726                  * need to complete the write.  If that's smaller
2727                  * than contig_clusters, we try for contig_clusters.
2728                  */
2729                 if (!*cow_len)
2730                         want_clusters = write_len;
2731                 else
2732                         want_clusters = (cpos + write_len) -
2733                                 (*cow_start + *cow_len);
2734                 if (want_clusters < contig_clusters)
2735                         want_clusters = contig_clusters;
2736
2737                 /*
2738                  * If the write does not cover the whole extent, we
2739                  * need to calculate how we're going to split the extent.
2740                  * We try to do it on contig_clusters boundaries.
2741                  *
2742                  * Any extent smaller than contig_clusters will be
2743                  * CoWed in its entirety.
2744                  */
2745                 if (leaf_clusters <= contig_clusters)
2746                         *cow_len += leaf_clusters;
2747                 else if (*cow_len || (*cow_start == cpos)) {
2748                         /*
2749                          * This extent needs to be CoW'd from its
2750                          * beginning, so all we have to do is compute
2751                          * how many clusters to grab.  We align
2752                          * want_clusters to the edge of contig_clusters
2753                          * to get better I/O.
2754                          */
2755                         want_clusters = ocfs2_cow_align_length(inode->i_sb,
2756                                                                want_clusters);
2757
2758                         if (leaf_clusters < want_clusters)
2759                                 *cow_len += leaf_clusters;
2760                         else
2761                                 *cow_len += want_clusters;
2762                 } else if ((*cow_start + contig_clusters) >=
2763                            (cpos + write_len)) {
2764                         /*
2765                          * Breaking off contig_clusters at the front
2766                          * of the extent will cover our write.  That's
2767                          * easy.
2768                          */
2769                         *cow_len = contig_clusters;
2770                 } else if ((rec_end - cpos) <= contig_clusters) {
2771                         /*
2772                          * Breaking off contig_clusters at the tail of
2773                          * this extent will cover cpos.
2774                          */
2775                         *cow_start = rec_end - contig_clusters;
2776                         *cow_len = contig_clusters;
2777                 } else if ((rec_end - cpos) <= want_clusters) {
2778                         /*
2779                          * While we can't fit the entire write in this
2780                          * extent, we know that the write goes from cpos
2781                          * to the end of the extent.  Break that off.
2782                          * We try to break it at some multiple of
2783                          * contig_clusters from the front of the extent.
2784                          * Failing that (ie, cpos is within
2785                          * contig_clusters of the front), we'll CoW the
2786                          * entire extent.
2787                          */
2788                         *cow_start = ocfs2_cow_align_start(inode->i_sb,
2789                                                            *cow_start, cpos);
2790                         *cow_len = rec_end - *cow_start;
2791                 } else {
2792                         /*
2793                          * Ok, the entire write lives in the middle of
2794                          * this extent.  Let's try to slice the extent up
2795                          * nicely.  Optimally, our CoW region starts at
2796                          * m*contig_clusters from the beginning of the
2797                          * extent and goes for n*contig_clusters,
2798                          * covering the entire write.
2799                          */
2800                         *cow_start = ocfs2_cow_align_start(inode->i_sb,
2801                                                            *cow_start, cpos);
2802
2803                         want_clusters = (cpos + write_len) - *cow_start;
2804                         want_clusters = ocfs2_cow_align_length(inode->i_sb,
2805                                                                want_clusters);
2806                         if (*cow_start + want_clusters <= rec_end)
2807                                 *cow_len = want_clusters;
2808                         else
2809                                 *cow_len = rec_end - *cow_start;
2810                 }
2811
2812                 /* Have we covered our entire write yet? */
2813                 if ((*cow_start + *cow_len) >= (cpos + write_len))
2814                         break;
2815
2816                 /*
2817                  * If we reach the end of the extent block and don't get enough
2818                  * clusters, continue with the next extent block if possible.
2819                  */
2820                 if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
2821                     eb && eb->h_next_leaf_blk) {
2822                         brelse(eb_bh);
2823                         eb_bh = NULL;
2824
2825                         ret = ocfs2_read_extent_block(INODE_CACHE(inode),
2826                                                le64_to_cpu(eb->h_next_leaf_blk),
2827                                                &eb_bh);
2828                         if (ret) {
2829                                 mlog_errno(ret);
2830                                 goto out;
2831                         }
2832
2833                         eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2834                         el = &eb->h_list;
2835                         i = -1;
2836                 }
2837         }
2838
2839 out:
2840         brelse(eb_bh);
2841         return ret;
2842 }
2843
2844 /*
2845  * Prepare meta_ac, data_ac and calculate credits when we want to add some
2846  * num_clusters in data_tree "et" and change the refcount for the old
2847  * clusters(starting form p_cluster) in the refcount tree.
2848  *
2849  * Note:
2850  * 1. since we may split the old tree, so we at most will need num_clusters + 2
2851  *    more new leaf records.
2852  * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2853  *    just give data_ac = NULL.
2854  */
2855 static int ocfs2_lock_refcount_allocators(struct super_block *sb,
2856                                         u32 p_cluster, u32 num_clusters,
2857                                         struct ocfs2_extent_tree *et,
2858                                         struct ocfs2_caching_info *ref_ci,
2859                                         struct buffer_head *ref_root_bh,
2860                                         struct ocfs2_alloc_context **meta_ac,
2861                                         struct ocfs2_alloc_context **data_ac,
2862                                         int *credits)
2863 {
2864         int ret = 0, meta_add = 0;
2865         int num_free_extents = ocfs2_num_free_extents(OCFS2_SB(sb), et);
2866
2867         if (num_free_extents < 0) {
2868                 ret = num_free_extents;
2869                 mlog_errno(ret);
2870                 goto out;
2871         }
2872
2873         if (num_free_extents < num_clusters + 2)
2874                 meta_add =
2875                         ocfs2_extend_meta_needed(et->et_root_el);
2876
2877         *credits += ocfs2_calc_extend_credits(sb, et->et_root_el,
2878                                               num_clusters + 2);
2879
2880         ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
2881                                                p_cluster, num_clusters,
2882                                                &meta_add, credits);
2883         if (ret) {
2884                 mlog_errno(ret);
2885                 goto out;
2886         }
2887
2888         mlog(0, "reserve new metadata %d, clusters %u, credits = %d\n",
2889              meta_add, num_clusters, *credits);
2890         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
2891                                                 meta_ac);
2892         if (ret) {
2893                 mlog_errno(ret);
2894                 goto out;
2895         }
2896
2897         if (data_ac) {
2898                 ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
2899                                              data_ac);
2900                 if (ret)
2901                         mlog_errno(ret);
2902         }
2903
2904 out:
2905         if (ret) {
2906                 if (*meta_ac) {
2907                         ocfs2_free_alloc_context(*meta_ac);
2908                         *meta_ac = NULL;
2909                 }
2910         }
2911
2912         return ret;
2913 }
2914
2915 static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
2916 {
2917         BUG_ON(buffer_dirty(bh));
2918
2919         clear_buffer_mapped(bh);
2920
2921         return 0;
2922 }
2923
2924 static int ocfs2_duplicate_clusters_by_page(handle_t *handle,
2925                                             struct ocfs2_cow_context *context,
2926                                             u32 cpos, u32 old_cluster,
2927                                             u32 new_cluster, u32 new_len)
2928 {
2929         int ret = 0, partial;
2930         struct ocfs2_caching_info *ci = context->data_et.et_ci;
2931         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2932         u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
2933         struct page *page;
2934         pgoff_t page_index;
2935         unsigned int from, to;
2936         loff_t offset, end, map_end;
2937         struct address_space *mapping = context->inode->i_mapping;
2938
2939         mlog(0, "old_cluster %u, new %u, len %u at offset %u\n", old_cluster,
2940              new_cluster, new_len, cpos);
2941
2942         offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
2943         end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
2944         /*
2945          * We only duplicate pages until we reach the page contains i_size - 1.
2946          * So trim 'end' to i_size.
2947          */
2948         if (end > i_size_read(context->inode))
2949                 end = i_size_read(context->inode);
2950
2951         while (offset < end) {
2952                 page_index = offset >> PAGE_CACHE_SHIFT;
2953                 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
2954                 if (map_end > end)
2955                         map_end = end;
2956
2957                 /* from, to is the offset within the page. */
2958                 from = offset & (PAGE_CACHE_SIZE - 1);
2959                 to = PAGE_CACHE_SIZE;
2960                 if (map_end & (PAGE_CACHE_SIZE - 1))
2961                         to = map_end & (PAGE_CACHE_SIZE - 1);
2962
2963                 page = find_or_create_page(mapping, page_index, GFP_NOFS);
2964
2965                 /*
2966                  * In case PAGE_CACHE_SIZE <= CLUSTER_SIZE, This page
2967                  * can't be dirtied before we CoW it out.
2968                  */
2969                 if (PAGE_CACHE_SIZE <= OCFS2_SB(sb)->s_clustersize)
2970                         BUG_ON(PageDirty(page));
2971
2972                 if (!PageUptodate(page)) {
2973                         ret = block_read_full_page(page, ocfs2_get_block);
2974                         if (ret) {
2975                                 mlog_errno(ret);
2976                                 goto unlock;
2977                         }
2978                         lock_page(page);
2979                 }
2980
2981                 if (page_has_buffers(page)) {
2982                         ret = walk_page_buffers(handle, page_buffers(page),
2983                                                 from, to, &partial,
2984                                                 ocfs2_clear_cow_buffer);
2985                         if (ret) {
2986                                 mlog_errno(ret);
2987                                 goto unlock;
2988                         }
2989                 }
2990
2991                 ocfs2_map_and_dirty_page(context->inode,
2992                                          handle, from, to,
2993                                          page, 0, &new_block);
2994                 mark_page_accessed(page);
2995 unlock:
2996                 unlock_page(page);
2997                 page_cache_release(page);
2998                 page = NULL;
2999                 offset = map_end;
3000                 if (ret)
3001                         break;
3002         }
3003
3004         return ret;
3005 }
3006
3007 static int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
3008                                            struct ocfs2_cow_context *context,
3009                                            u32 cpos, u32 old_cluster,
3010                                            u32 new_cluster, u32 new_len)
3011 {
3012         int ret = 0;
3013         struct super_block *sb = context->inode->i_sb;
3014         struct ocfs2_caching_info *ci = context->data_et.et_ci;
3015         int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
3016         u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
3017         u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
3018         struct ocfs2_super *osb = OCFS2_SB(sb);
3019         struct buffer_head *old_bh = NULL;
3020         struct buffer_head *new_bh = NULL;
3021
3022         mlog(0, "old_cluster %u, new %u, len %u\n", old_cluster,
3023              new_cluster, new_len);
3024
3025         for (i = 0; i < blocks; i++, old_block++, new_block++) {
3026                 new_bh = sb_getblk(osb->sb, new_block);
3027                 if (new_bh == NULL) {
3028                         ret = -EIO;
3029                         mlog_errno(ret);
3030                         break;
3031                 }
3032
3033                 ocfs2_set_new_buffer_uptodate(ci, new_bh);
3034
3035                 ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
3036                 if (ret) {
3037                         mlog_errno(ret);
3038                         break;
3039                 }
3040
3041                 ret = ocfs2_journal_access(handle, ci, new_bh,
3042                                            OCFS2_JOURNAL_ACCESS_CREATE);
3043                 if (ret) {
3044                         mlog_errno(ret);
3045                         break;
3046                 }
3047
3048                 memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
3049                 ocfs2_journal_dirty(handle, new_bh);
3050
3051                 brelse(new_bh);
3052                 brelse(old_bh);
3053                 new_bh = NULL;
3054                 old_bh = NULL;
3055         }
3056
3057         brelse(new_bh);
3058         brelse(old_bh);
3059         return ret;
3060 }
3061
3062 static int ocfs2_clear_ext_refcount(handle_t *handle,
3063                                     struct ocfs2_extent_tree *et,
3064                                     u32 cpos, u32 p_cluster, u32 len,
3065                                     unsigned int ext_flags,
3066                                     struct ocfs2_alloc_context *meta_ac,
3067                                     struct ocfs2_cached_dealloc_ctxt *dealloc)
3068 {
3069         int ret, index;
3070         struct ocfs2_extent_rec replace_rec;
3071         struct ocfs2_path *path = NULL;
3072         struct ocfs2_extent_list *el;
3073         struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
3074         u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
3075
3076         mlog(0, "inode %llu cpos %u, len %u, p_cluster %u, ext_flags %u\n",
3077              (unsigned long long)ino, cpos, len, p_cluster, ext_flags);
3078
3079         memset(&replace_rec, 0, sizeof(replace_rec));
3080         replace_rec.e_cpos = cpu_to_le32(cpos);
3081         replace_rec.e_leaf_clusters = cpu_to_le16(len);
3082         replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
3083                                                                    p_cluster));
3084         replace_rec.e_flags = ext_flags;
3085         replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
3086
3087         path = ocfs2_new_path_from_et(et);
3088         if (!path) {
3089                 ret = -ENOMEM;
3090                 mlog_errno(ret);
3091                 goto out;
3092         }
3093
3094         ret = ocfs2_find_path(et->et_ci, path, cpos);
3095         if (ret) {
3096                 mlog_errno(ret);
3097                 goto out;
3098         }
3099
3100         el = path_leaf_el(path);
3101
3102         index = ocfs2_search_extent_list(el, cpos);
3103         if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
3104                 ocfs2_error(sb,
3105                             "Inode %llu has an extent at cpos %u which can no "
3106                             "longer be found.\n",
3107                             (unsigned long long)ino, cpos);
3108                 ret = -EROFS;
3109                 goto out;
3110         }
3111
3112         ret = ocfs2_split_extent(handle, et, path, index,
3113                                  &replace_rec, meta_ac, dealloc);
3114         if (ret)
3115                 mlog_errno(ret);
3116
3117 out:
3118         ocfs2_free_path(path);
3119         return ret;
3120 }
3121
3122 static int ocfs2_replace_clusters(handle_t *handle,
3123                                   struct ocfs2_cow_context *context,
3124                                   u32 cpos, u32 old,
3125                                   u32 new, u32 len,
3126                                   unsigned int ext_flags)
3127 {
3128         int ret;
3129         struct ocfs2_caching_info *ci = context->data_et.et_ci;
3130         u64 ino = ocfs2_metadata_cache_owner(ci);
3131
3132         mlog(0, "inode %llu, cpos %u, old %u, new %u, len %u, ext_flags %u\n",
3133              (unsigned long long)ino, cpos, old, new, len, ext_flags);
3134
3135         /*If the old clusters is unwritten, no need to duplicate. */
3136         if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
3137                 ret = context->cow_duplicate_clusters(handle, context, cpos,
3138                                                       old, new, len);
3139                 if (ret) {
3140                         mlog_errno(ret);
3141                         goto out;
3142                 }
3143         }
3144
3145         ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
3146                                        cpos, new, len, ext_flags,
3147                                        context->meta_ac, &context->dealloc);
3148         if (ret)
3149                 mlog_errno(ret);
3150 out:
3151         return ret;
3152 }
3153
3154 static int ocfs2_cow_sync_writeback(struct super_block *sb,
3155                                     struct ocfs2_cow_context *context,
3156                                     u32 cpos, u32 num_clusters)
3157 {
3158         int ret = 0;
3159         loff_t offset, end, map_end;
3160         pgoff_t page_index;
3161         struct page *page;
3162
3163         if (ocfs2_should_order_data(context->inode))
3164                 return 0;
3165
3166         offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
3167         end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
3168
3169         ret = filemap_fdatawrite_range(context->inode->i_mapping,
3170                                        offset, end - 1);
3171         if (ret < 0) {
3172                 mlog_errno(ret);
3173                 return ret;
3174         }
3175
3176         while (offset < end) {
3177                 page_index = offset >> PAGE_CACHE_SHIFT;
3178                 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
3179                 if (map_end > end)
3180                         map_end = end;
3181
3182                 page = find_or_create_page(context->inode->i_mapping,
3183                                            page_index, GFP_NOFS);
3184                 BUG_ON(!page);
3185
3186                 wait_on_page_writeback(page);
3187                 if (PageError(page)) {
3188                         ret = -EIO;
3189                         mlog_errno(ret);
3190                 } else
3191                         mark_page_accessed(page);
3192
3193                 unlock_page(page);
3194                 page_cache_release(page);
3195                 page = NULL;
3196                 offset = map_end;
3197                 if (ret)
3198                         break;
3199         }
3200
3201         return ret;
3202 }
3203
3204 static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
3205                                  u32 v_cluster, u32 *p_cluster,
3206                                  u32 *num_clusters,
3207                                  unsigned int *extent_flags)
3208 {
3209         return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
3210                                   num_clusters, extent_flags);
3211 }
3212
3213 static int ocfs2_make_clusters_writable(struct super_block *sb,
3214                                         struct ocfs2_cow_context *context,
3215                                         u32 cpos, u32 p_cluster,
3216                                         u32 num_clusters, unsigned int e_flags)
3217 {
3218         int ret, delete, index, credits =  0;
3219         u32 new_bit, new_len;
3220         unsigned int set_len;
3221         struct ocfs2_super *osb = OCFS2_SB(sb);
3222         handle_t *handle;
3223         struct buffer_head *ref_leaf_bh = NULL;
3224         struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
3225         struct ocfs2_refcount_rec rec;
3226
3227         mlog(0, "cpos %u, p_cluster %u, num_clusters %u, e_flags %u\n",
3228              cpos, p_cluster, num_clusters, e_flags);
3229
3230         ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
3231                                              &context->data_et,
3232                                              ref_ci,
3233                                              context->ref_root_bh,
3234                                              &context->meta_ac,
3235                                              &context->data_ac, &credits);
3236         if (ret) {
3237                 mlog_errno(ret);
3238                 return ret;
3239         }
3240
3241         if (context->post_refcount)
3242                 credits += context->post_refcount->credits;
3243
3244         credits += context->extra_credits;
3245         handle = ocfs2_start_trans(osb, credits);
3246         if (IS_ERR(handle)) {
3247                 ret = PTR_ERR(handle);
3248                 mlog_errno(ret);
3249                 goto out;
3250         }
3251
3252         while (num_clusters) {
3253                 ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
3254                                              p_cluster, num_clusters,
3255                                              &rec, &index, &ref_leaf_bh);
3256                 if (ret) {
3257                         mlog_errno(ret);
3258                         goto out_commit;
3259                 }
3260
3261                 BUG_ON(!rec.r_refcount);
3262                 set_len = min((u64)p_cluster + num_clusters,
3263                               le64_to_cpu(rec.r_cpos) +
3264                               le32_to_cpu(rec.r_clusters)) - p_cluster;
3265
3266                 /*
3267                  * There are many different situation here.
3268                  * 1. If refcount == 1, remove the flag and don't COW.
3269                  * 2. If refcount > 1, allocate clusters.
3270                  *    Here we may not allocate r_len once at a time, so continue
3271                  *    until we reach num_clusters.
3272                  */
3273                 if (le32_to_cpu(rec.r_refcount) == 1) {
3274                         delete = 0;
3275                         ret = ocfs2_clear_ext_refcount(handle,
3276                                                        &context->data_et,
3277                                                        cpos, p_cluster,
3278                                                        set_len, e_flags,
3279                                                        context->meta_ac,
3280                                                        &context->dealloc);
3281                         if (ret) {
3282                                 mlog_errno(ret);
3283                                 goto out_commit;
3284                         }
3285                 } else {
3286                         delete = 1;
3287
3288                         ret = __ocfs2_claim_clusters(handle,
3289                                                      context->data_ac,
3290                                                      1, set_len,
3291                                                      &new_bit, &new_len);
3292                         if (ret) {
3293                                 mlog_errno(ret);
3294                                 goto out_commit;
3295                         }
3296
3297                         ret = ocfs2_replace_clusters(handle, context,
3298                                                      cpos, p_cluster, new_bit,
3299                                                      new_len, e_flags);
3300                         if (ret) {
3301                                 mlog_errno(ret);
3302                                 goto out_commit;
3303                         }
3304                         set_len = new_len;
3305                 }
3306
3307                 ret = __ocfs2_decrease_refcount(handle, ref_ci,
3308                                                 context->ref_root_bh,
3309                                                 p_cluster, set_len,
3310                                                 context->meta_ac,
3311                                                 &context->dealloc, delete);
3312                 if (ret) {
3313                         mlog_errno(ret);
3314                         goto out_commit;
3315                 }
3316
3317                 cpos += set_len;
3318                 p_cluster += set_len;
3319                 num_clusters -= set_len;
3320                 brelse(ref_leaf_bh);
3321                 ref_leaf_bh = NULL;
3322         }
3323
3324         /* handle any post_cow action. */
3325         if (context->post_refcount && context->post_refcount->func) {
3326                 ret = context->post_refcount->func(context->inode, handle,
3327                                                 context->post_refcount->para);
3328                 if (ret) {
3329                         mlog_errno(ret);
3330                         goto out_commit;
3331                 }
3332         }
3333
3334         /*
3335          * Here we should write the new page out first if we are
3336          * in write-back mode.
3337          */
3338         if (context->get_clusters == ocfs2_di_get_clusters) {
3339                 ret = ocfs2_cow_sync_writeback(sb, context, cpos, num_clusters);
3340                 if (ret)
3341                         mlog_errno(ret);
3342         }
3343
3344 out_commit:
3345         ocfs2_commit_trans(osb, handle);
3346
3347 out:
3348         if (context->data_ac) {
3349                 ocfs2_free_alloc_context(context->data_ac);
3350                 context->data_ac = NULL;
3351         }
3352         if (context->meta_ac) {
3353                 ocfs2_free_alloc_context(context->meta_ac);
3354                 context->meta_ac = NULL;
3355         }
3356         brelse(ref_leaf_bh);
3357
3358         return ret;
3359 }
3360
3361 static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
3362 {
3363         int ret = 0;
3364         struct inode *inode = context->inode;
3365         u32 cow_start = context->cow_start, cow_len = context->cow_len;
3366         u32 p_cluster, num_clusters;
3367         unsigned int ext_flags;
3368         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3369
3370         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
3371                 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
3372                             "tree, but the feature bit is not set in the "
3373                             "super block.", inode->i_ino);
3374                 return -EROFS;
3375         }
3376
3377         ocfs2_init_dealloc_ctxt(&context->dealloc);
3378
3379         while (cow_len) {
3380                 ret = context->get_clusters(context, cow_start, &p_cluster,
3381                                             &num_clusters, &ext_flags);
3382                 if (ret) {
3383                         mlog_errno(ret);
3384                         break;
3385                 }
3386
3387                 BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
3388
3389                 if (cow_len < num_clusters)
3390                         num_clusters = cow_len;
3391
3392                 ret = ocfs2_make_clusters_writable(inode->i_sb, context,
3393                                                    cow_start, p_cluster,
3394                                                    num_clusters, ext_flags);
3395                 if (ret) {
3396                         mlog_errno(ret);
3397                         break;
3398                 }
3399
3400                 cow_len -= num_clusters;
3401                 cow_start += num_clusters;
3402         }
3403
3404         if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
3405                 ocfs2_schedule_truncate_log_flush(osb, 1);
3406                 ocfs2_run_deallocs(osb, &context->dealloc);
3407         }
3408
3409         return ret;
3410 }
3411
3412 /*
3413  * Starting at cpos, try to CoW write_len clusters.  Don't CoW
3414  * past max_cpos.  This will stop when it runs into a hole or an
3415  * unrefcounted extent.
3416  */
3417 static int ocfs2_refcount_cow_hunk(struct inode *inode,
3418                                    struct buffer_head *di_bh,
3419                                    u32 cpos, u32 write_len, u32 max_cpos)
3420 {
3421         int ret;
3422         u32 cow_start = 0, cow_len = 0;
3423         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3424         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3425         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3426         struct buffer_head *ref_root_bh = NULL;
3427         struct ocfs2_refcount_tree *ref_tree;
3428         struct ocfs2_cow_context *context = NULL;
3429
3430         BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3431
3432         ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
3433                                               cpos, write_len, max_cpos,
3434                                               &cow_start, &cow_len);
3435         if (ret) {
3436                 mlog_errno(ret);
3437                 goto out;
3438         }
3439
3440         mlog(0, "CoW inode %lu, cpos %u, write_len %u, cow_start %u, "
3441              "cow_len %u\n", inode->i_ino,
3442              cpos, write_len, cow_start, cow_len);
3443
3444         BUG_ON(cow_len == 0);
3445
3446         context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3447         if (!context) {
3448                 ret = -ENOMEM;
3449                 mlog_errno(ret);
3450                 goto out;
3451         }
3452
3453         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
3454                                        1, &ref_tree, &ref_root_bh);
3455         if (ret) {
3456                 mlog_errno(ret);
3457                 goto out;
3458         }
3459
3460         context->inode = inode;
3461         context->cow_start = cow_start;
3462         context->cow_len = cow_len;
3463         context->ref_tree = ref_tree;
3464         context->ref_root_bh = ref_root_bh;
3465         context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
3466         context->get_clusters = ocfs2_di_get_clusters;
3467
3468         ocfs2_init_dinode_extent_tree(&context->data_et,
3469                                       INODE_CACHE(inode), di_bh);
3470
3471         ret = ocfs2_replace_cow(context);
3472         if (ret)
3473                 mlog_errno(ret);
3474
3475         /*
3476          * truncate the extent map here since no matter whether we meet with
3477          * any error during the action, we shouldn't trust cached extent map
3478          * any more.
3479          */
3480         ocfs2_extent_map_trunc(inode, cow_start);
3481
3482         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3483         brelse(ref_root_bh);
3484 out:
3485         kfree(context);
3486         return ret;
3487 }
3488
3489 /*
3490  * CoW any and all clusters between cpos and cpos+write_len.
3491  * Don't CoW past max_cpos.  If this returns successfully, all
3492  * clusters between cpos and cpos+write_len are safe to modify.
3493  */
3494 int ocfs2_refcount_cow(struct inode *inode,
3495                        struct buffer_head *di_bh,
3496                        u32 cpos, u32 write_len, u32 max_cpos)
3497 {
3498         int ret = 0;
3499         u32 p_cluster, num_clusters;
3500         unsigned int ext_flags;
3501
3502         while (write_len) {
3503                 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3504                                          &num_clusters, &ext_flags);
3505                 if (ret) {
3506                         mlog_errno(ret);
3507                         break;
3508                 }
3509
3510                 if (write_len < num_clusters)
3511                         num_clusters = write_len;
3512
3513                 if (ext_flags & OCFS2_EXT_REFCOUNTED) {
3514                         ret = ocfs2_refcount_cow_hunk(inode, di_bh, cpos,
3515                                                       num_clusters, max_cpos);
3516                         if (ret) {
3517                                 mlog_errno(ret);
3518                                 break;
3519                         }
3520                 }
3521
3522                 write_len -= num_clusters;
3523                 cpos += num_clusters;
3524         }
3525
3526         return ret;
3527 }
3528
3529 static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
3530                                           u32 v_cluster, u32 *p_cluster,
3531                                           u32 *num_clusters,
3532                                           unsigned int *extent_flags)
3533 {
3534         struct inode *inode = context->inode;
3535         struct ocfs2_xattr_value_root *xv = context->cow_object;
3536
3537         return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
3538                                         num_clusters, &xv->xr_list,
3539                                         extent_flags);
3540 }
3541
3542 /*
3543  * Given a xattr value root, calculate the most meta/credits we need for
3544  * refcount tree change if we truncate it to 0.
3545  */
3546 int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
3547                                        struct ocfs2_caching_info *ref_ci,
3548                                        struct buffer_head *ref_root_bh,
3549                                        struct ocfs2_xattr_value_root *xv,
3550                                        int *meta_add, int *credits)
3551 {
3552         int ret = 0, index, ref_blocks = 0;
3553         u32 p_cluster, num_clusters;
3554         u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
3555         struct ocfs2_refcount_block *rb;
3556         struct ocfs2_refcount_rec rec;
3557         struct buffer_head *ref_leaf_bh = NULL;
3558
3559         while (cpos < clusters) {
3560                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
3561                                                &num_clusters, &xv->xr_list,
3562                                                NULL);
3563                 if (ret) {
3564                         mlog_errno(ret);
3565                         goto out;
3566                 }
3567
3568                 cpos += num_clusters;
3569
3570                 while (num_clusters) {
3571                         ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
3572                                                      p_cluster, num_clusters,
3573                                                      &rec, &index,
3574                                                      &ref_leaf_bh);
3575                         if (ret) {
3576                                 mlog_errno(ret);
3577                                 goto out;
3578                         }
3579
3580                         BUG_ON(!rec.r_refcount);
3581
3582                         rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
3583
3584                         /*
3585                          * We really don't know whether the other clusters is in
3586                          * this refcount block or not, so just take the worst
3587                          * case that all the clusters are in this block and each
3588                          * one will split a refcount rec, so totally we need
3589                          * clusters * 2 new refcount rec.
3590                          */
3591                         if (le64_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
3592                             le16_to_cpu(rb->rf_records.rl_count))
3593                                 ref_blocks++;
3594
3595                         *credits += 1;
3596                         brelse(ref_leaf_bh);
3597                         ref_leaf_bh = NULL;
3598
3599                         if (num_clusters <= le32_to_cpu(rec.r_clusters))
3600                                 break;
3601                         else
3602                                 num_clusters -= le32_to_cpu(rec.r_clusters);
3603                         p_cluster += num_clusters;
3604                 }
3605         }
3606
3607         *meta_add += ref_blocks;
3608         if (!ref_blocks)
3609                 goto out;
3610
3611         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
3612         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
3613                 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
3614         else {
3615                 struct ocfs2_extent_tree et;
3616
3617                 ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
3618                 *credits += ocfs2_calc_extend_credits(inode->i_sb,
3619                                                       et.et_root_el,
3620                                                       ref_blocks);
3621         }
3622
3623 out:
3624         brelse(ref_leaf_bh);
3625         return ret;
3626 }
3627
3628 /*
3629  * Do CoW for xattr.
3630  */
3631 int ocfs2_refcount_cow_xattr(struct inode *inode,
3632                              struct ocfs2_dinode *di,
3633                              struct ocfs2_xattr_value_buf *vb,
3634                              struct ocfs2_refcount_tree *ref_tree,
3635                              struct buffer_head *ref_root_bh,
3636                              u32 cpos, u32 write_len,
3637                              struct ocfs2_post_refcount *post)
3638 {
3639         int ret;
3640         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
3641         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3642         struct ocfs2_cow_context *context = NULL;
3643         u32 cow_start, cow_len;
3644
3645         BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3646
3647         ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
3648                                               cpos, write_len, UINT_MAX,
3649                                               &cow_start, &cow_len);
3650         if (ret) {
3651                 mlog_errno(ret);
3652                 goto out;
3653         }
3654
3655         BUG_ON(cow_len == 0);
3656
3657         context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3658         if (!context) {
3659                 ret = -ENOMEM;
3660                 mlog_errno(ret);
3661                 goto out;
3662         }
3663
3664         context->inode = inode;
3665         context->cow_start = cow_start;
3666         context->cow_len = cow_len;
3667         context->ref_tree = ref_tree;
3668         context->ref_root_bh = ref_root_bh;;
3669         context->cow_object = xv;
3670
3671         context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
3672         /* We need the extra credits for duplicate_clusters by jbd. */
3673         context->extra_credits =
3674                 ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
3675         context->get_clusters = ocfs2_xattr_value_get_clusters;
3676         context->post_refcount = post;
3677
3678         ocfs2_init_xattr_value_extent_tree(&context->data_et,
3679                                            INODE_CACHE(inode), vb);
3680
3681         ret = ocfs2_replace_cow(context);
3682         if (ret)
3683                 mlog_errno(ret);
3684
3685 out:
3686         kfree(context);
3687         return ret;
3688 }
3689
3690 /*
3691  * Insert a new extent into refcount tree and mark a extent rec
3692  * as refcounted in the dinode tree.
3693  */
3694 int ocfs2_add_refcount_flag(struct inode *inode,
3695                             struct ocfs2_extent_tree *data_et,
3696                             struct ocfs2_caching_info *ref_ci,
3697                             struct buffer_head *ref_root_bh,
3698                             u32 cpos, u32 p_cluster, u32 num_clusters,
3699                             struct ocfs2_cached_dealloc_ctxt *dealloc,
3700                             struct ocfs2_post_refcount *post)
3701 {
3702         int ret;
3703         handle_t *handle;
3704         int credits = 1, ref_blocks = 0;
3705         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3706         struct ocfs2_alloc_context *meta_ac = NULL;
3707
3708         ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
3709                                                ref_ci, ref_root_bh,
3710                                                p_cluster, num_clusters,
3711                                                &ref_blocks, &credits);
3712         if (ret) {
3713                 mlog_errno(ret);
3714                 goto out;
3715         }
3716
3717         mlog(0, "reserve new metadata %d, credits = %d\n",
3718              ref_blocks, credits);
3719
3720         if (ref_blocks) {
3721                 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
3722                                                         ref_blocks, &meta_ac);
3723                 if (ret) {
3724                         mlog_errno(ret);
3725                         goto out;
3726                 }
3727         }
3728
3729         if (post)
3730                 credits += post->credits;
3731
3732         handle = ocfs2_start_trans(osb, credits);
3733         if (IS_ERR(handle)) {
3734                 ret = PTR_ERR(handle);
3735                 mlog_errno(ret);
3736                 goto out;
3737         }
3738
3739         ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
3740                                            cpos, num_clusters, p_cluster,
3741                                            meta_ac, dealloc);
3742         if (ret) {
3743                 mlog_errno(ret);
3744                 goto out_commit;
3745         }
3746
3747         ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3748                                         p_cluster, num_clusters, 0,
3749                                         meta_ac, dealloc);
3750         if (ret) {
3751                 mlog_errno(ret);
3752                 goto out_commit;
3753         }
3754
3755         if (post && post->func) {
3756                 ret = post->func(inode, handle, post->para);
3757                 if (ret)
3758                         mlog_errno(ret);
3759         }
3760
3761 out_commit:
3762         ocfs2_commit_trans(osb, handle);
3763 out:
3764         if (meta_ac)
3765                 ocfs2_free_alloc_context(meta_ac);
3766         return ret;
3767 }
3768
3769 static int ocfs2_change_ctime(struct inode *inode,
3770                               struct buffer_head *di_bh)
3771 {
3772         int ret;
3773         handle_t *handle;
3774         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3775
3776         handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
3777                                    OCFS2_INODE_UPDATE_CREDITS);
3778         if (IS_ERR(handle)) {
3779                 ret = PTR_ERR(handle);
3780                 mlog_errno(ret);
3781                 goto out;
3782         }
3783
3784         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
3785                                       OCFS2_JOURNAL_ACCESS_WRITE);
3786         if (ret) {
3787                 mlog_errno(ret);
3788                 goto out_commit;
3789         }
3790
3791         inode->i_ctime = CURRENT_TIME;
3792         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3793         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3794
3795         ocfs2_journal_dirty(handle, di_bh);
3796
3797 out_commit:
3798         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3799 out:
3800         return ret;
3801 }
3802
3803 static int ocfs2_attach_refcount_tree(struct inode *inode,
3804                                       struct buffer_head *di_bh)
3805 {
3806         int ret, data_changed = 0;
3807         struct buffer_head *ref_root_bh = NULL;
3808         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3809         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3810         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3811         struct ocfs2_refcount_tree *ref_tree;
3812         unsigned int ext_flags;
3813         loff_t size;
3814         u32 cpos, num_clusters, clusters, p_cluster;
3815         struct ocfs2_cached_dealloc_ctxt dealloc;
3816         struct ocfs2_extent_tree di_et;
3817
3818         ocfs2_init_dealloc_ctxt(&dealloc);
3819
3820         if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) {
3821                 ret = ocfs2_create_refcount_tree(inode, di_bh);
3822                 if (ret) {
3823                         mlog_errno(ret);
3824                         goto out;
3825                 }
3826         }
3827
3828         BUG_ON(!di->i_refcount_loc);
3829         ret = ocfs2_lock_refcount_tree(osb,
3830                                        le64_to_cpu(di->i_refcount_loc), 1,
3831                                        &ref_tree, &ref_root_bh);
3832         if (ret) {
3833                 mlog_errno(ret);
3834                 goto out;
3835         }
3836
3837         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
3838                 goto attach_xattr;
3839
3840         ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
3841
3842         size = i_size_read(inode);
3843         clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
3844
3845         cpos = 0;
3846         while (cpos < clusters) {
3847                 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3848                                          &num_clusters, &ext_flags);
3849
3850                 if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
3851                         ret = ocfs2_add_refcount_flag(inode, &di_et,
3852                                                       &ref_tree->rf_ci,
3853                                                       ref_root_bh, cpos,
3854                                                       p_cluster, num_clusters,
3855                                                       &dealloc, NULL);
3856                         if (ret) {
3857                                 mlog_errno(ret);
3858                                 goto unlock;
3859                         }
3860
3861                         data_changed = 1;
3862                 }
3863                 cpos += num_clusters;
3864         }
3865
3866 attach_xattr:
3867         if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
3868                 ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
3869                                                        &ref_tree->rf_ci,
3870                                                        ref_root_bh,
3871                                                        &dealloc);
3872                 if (ret) {
3873                         mlog_errno(ret);
3874                         goto unlock;
3875                 }
3876         }
3877
3878         if (data_changed) {
3879                 ret = ocfs2_change_ctime(inode, di_bh);
3880                 if (ret)
3881                         mlog_errno(ret);
3882         }
3883
3884 unlock:
3885         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3886         brelse(ref_root_bh);
3887
3888         if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
3889                 ocfs2_schedule_truncate_log_flush(osb, 1);
3890                 ocfs2_run_deallocs(osb, &dealloc);
3891         }
3892 out:
3893         /*
3894          * Empty the extent map so that we may get the right extent
3895          * record from the disk.
3896          */
3897         ocfs2_extent_map_trunc(inode, 0);
3898
3899         return ret;
3900 }
3901
3902 static int ocfs2_add_refcounted_extent(struct inode *inode,
3903                                    struct ocfs2_extent_tree *et,
3904                                    struct ocfs2_caching_info *ref_ci,
3905                                    struct buffer_head *ref_root_bh,
3906                                    u32 cpos, u32 p_cluster, u32 num_clusters,
3907                                    unsigned int ext_flags,
3908                                    struct ocfs2_cached_dealloc_ctxt *dealloc)
3909 {
3910         int ret;
3911         handle_t *handle;
3912         int credits = 0;
3913         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3914         struct ocfs2_alloc_context *meta_ac = NULL;
3915
3916         ret = ocfs2_lock_refcount_allocators(inode->i_sb,
3917                                              p_cluster, num_clusters,
3918                                              et, ref_ci,
3919                                              ref_root_bh, &meta_ac,
3920                                              NULL, &credits);
3921         if (ret) {
3922                 mlog_errno(ret);
3923                 goto out;
3924         }
3925
3926         handle = ocfs2_start_trans(osb, credits);
3927         if (IS_ERR(handle)) {
3928                 ret = PTR_ERR(handle);
3929                 mlog_errno(ret);
3930                 goto out;
3931         }
3932
3933         ret = ocfs2_insert_extent(handle, et, cpos,
3934                         ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
3935                         num_clusters, ext_flags, meta_ac);
3936         if (ret) {
3937                 mlog_errno(ret);
3938                 goto out_commit;
3939         }
3940
3941         ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3942                                       p_cluster, num_clusters,
3943                                       meta_ac, dealloc);
3944         if (ret)
3945                 mlog_errno(ret);
3946
3947 out_commit:
3948         ocfs2_commit_trans(osb, handle);
3949 out:
3950         if (meta_ac)
3951                 ocfs2_free_alloc_context(meta_ac);
3952         return ret;
3953 }
3954
3955 static int ocfs2_duplicate_inline_data(struct inode *s_inode,
3956                                        struct buffer_head *s_bh,
3957                                        struct inode *t_inode,
3958                                        struct buffer_head *t_bh)
3959 {
3960         int ret;
3961         handle_t *handle;
3962         struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
3963         struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
3964         struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
3965
3966         BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
3967
3968         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
3969         if (IS_ERR(handle)) {
3970                 ret = PTR_ERR(handle);
3971                 mlog_errno(ret);
3972                 goto out;
3973         }
3974
3975         ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
3976                                       OCFS2_JOURNAL_ACCESS_WRITE);
3977         if (ret) {
3978                 mlog_errno(ret);
3979                 goto out_commit;
3980         }
3981
3982         t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
3983         memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
3984                le16_to_cpu(s_di->id2.i_data.id_count));
3985         spin_lock(&OCFS2_I(t_inode)->ip_lock);
3986         OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
3987         t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
3988         spin_unlock(&OCFS2_I(t_inode)->ip_lock);
3989
3990         ocfs2_journal_dirty(handle, t_bh);
3991
3992 out_commit:
3993         ocfs2_commit_trans(osb, handle);
3994 out:
3995         return ret;
3996 }
3997
3998 static int ocfs2_duplicate_extent_list(struct inode *s_inode,
3999                                 struct inode *t_inode,
4000                                 struct buffer_head *t_bh,
4001                                 struct ocfs2_caching_info *ref_ci,
4002                                 struct buffer_head *ref_root_bh,
4003                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
4004 {
4005         int ret = 0;
4006         u32 p_cluster, num_clusters, clusters, cpos;
4007         loff_t size;
4008         unsigned int ext_flags;
4009         struct ocfs2_extent_tree et;
4010
4011         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
4012
4013         size = i_size_read(s_inode);
4014         clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
4015
4016         cpos = 0;
4017         while (cpos < clusters) {
4018                 ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
4019                                          &num_clusters, &ext_flags);
4020
4021                 if (p_cluster) {
4022                         ret = ocfs2_add_refcounted_extent(t_inode, &et,
4023                                                           ref_ci, ref_root_bh,
4024                                                           cpos, p_cluster,
4025                                                           num_clusters,
4026                                                           ext_flags,
4027                                                           dealloc);
4028                         if (ret) {
4029                                 mlog_errno(ret);
4030                                 goto out;
4031                         }
4032                 }
4033
4034                 cpos += num_clusters;
4035         }
4036
4037 out:
4038         return ret;
4039 }
4040
4041 /*
4042  * change the new file's attributes to the src.
4043  *
4044  * reflink creates a snapshot of a file, that means the attributes
4045  * must be identical except for three exceptions - nlink, ino, and ctime.
4046  */
4047 static int ocfs2_complete_reflink(struct inode *s_inode,
4048                                   struct buffer_head *s_bh,
4049                                   struct inode *t_inode,
4050                                   struct buffer_head *t_bh,
4051                                   bool preserve)
4052 {
4053         int ret;
4054         handle_t *handle;
4055         struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4056         struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
4057         loff_t size = i_size_read(s_inode);
4058
4059         handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
4060                                    OCFS2_INODE_UPDATE_CREDITS);
4061         if (IS_ERR(handle)) {
4062                 ret = PTR_ERR(handle);
4063                 mlog_errno(ret);
4064                 return ret;
4065         }
4066
4067         ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4068                                       OCFS2_JOURNAL_ACCESS_WRITE);
4069         if (ret) {
4070                 mlog_errno(ret);
4071                 goto out_commit;
4072         }
4073
4074         spin_lock(&OCFS2_I(t_inode)->ip_lock);
4075         OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
4076         OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
4077         OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
4078         spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4079         i_size_write(t_inode, size);
4080         t_inode->i_blocks = s_inode->i_blocks;
4081
4082         di->i_xattr_inline_size = s_di->i_xattr_inline_size;
4083         di->i_clusters = s_di->i_clusters;
4084         di->i_size = s_di->i_size;
4085         di->i_dyn_features = s_di->i_dyn_features;
4086         di->i_attr = s_di->i_attr;
4087
4088         if (preserve) {
4089                 t_inode->i_uid = s_inode->i_uid;
4090                 t_inode->i_gid = s_inode->i_gid;
4091                 t_inode->i_mode = s_inode->i_mode;
4092                 di->i_uid = s_di->i_uid;
4093                 di->i_gid = s_di->i_gid;
4094                 di->i_mode = s_di->i_mode;
4095
4096                 /*
4097                  * update time.
4098                  * we want mtime to appear identical to the source and
4099                  * update ctime.
4100                  */
4101                 t_inode->i_ctime = CURRENT_TIME;
4102
4103                 di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
4104                 di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
4105
4106                 t_inode->i_mtime = s_inode->i_mtime;
4107                 di->i_mtime = s_di->i_mtime;
4108                 di->i_mtime_nsec = s_di->i_mtime_nsec;
4109         }
4110
4111         ocfs2_journal_dirty(handle, t_bh);
4112
4113 out_commit:
4114         ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
4115         return ret;
4116 }
4117
4118 static int ocfs2_create_reflink_node(struct inode *s_inode,
4119                                      struct buffer_head *s_bh,
4120                                      struct inode *t_inode,
4121                                      struct buffer_head *t_bh,
4122                                      bool preserve)
4123 {
4124         int ret;
4125         struct buffer_head *ref_root_bh = NULL;
4126         struct ocfs2_cached_dealloc_ctxt dealloc;
4127         struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4128         struct ocfs2_refcount_block *rb;
4129         struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
4130         struct ocfs2_refcount_tree *ref_tree;
4131
4132         ocfs2_init_dealloc_ctxt(&dealloc);
4133
4134         ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4135                                       le64_to_cpu(di->i_refcount_loc));
4136         if (ret) {
4137                 mlog_errno(ret);
4138                 goto out;
4139         }
4140
4141         if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4142                 ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
4143                                                   t_inode, t_bh);
4144                 if (ret)
4145                         mlog_errno(ret);
4146                 goto out;
4147         }
4148
4149         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
4150                                        1, &ref_tree, &ref_root_bh);
4151         if (ret) {
4152                 mlog_errno(ret);
4153                 goto out;
4154         }
4155         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
4156
4157         ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
4158                                           &ref_tree->rf_ci, ref_root_bh,
4159                                           &dealloc);
4160         if (ret) {
4161                 mlog_errno(ret);
4162                 goto out_unlock_refcount;
4163         }
4164
4165 out_unlock_refcount:
4166         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4167         brelse(ref_root_bh);
4168 out:
4169         if (ocfs2_dealloc_has_cluster(&dealloc)) {
4170                 ocfs2_schedule_truncate_log_flush(osb, 1);
4171                 ocfs2_run_deallocs(osb, &dealloc);
4172         }
4173
4174         return ret;
4175 }
4176
4177 static int __ocfs2_reflink(struct dentry *old_dentry,
4178                            struct buffer_head *old_bh,
4179                            struct inode *new_inode,
4180                            bool preserve)
4181 {
4182         int ret;
4183         struct inode *inode = old_dentry->d_inode;
4184         struct buffer_head *new_bh = NULL;
4185
4186         if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
4187                 ret = -EINVAL;
4188                 mlog_errno(ret);
4189                 goto out;
4190         }
4191
4192         ret = filemap_fdatawrite(inode->i_mapping);
4193         if (ret) {
4194                 mlog_errno(ret);
4195                 goto out;
4196         }
4197
4198         ret = ocfs2_attach_refcount_tree(inode, old_bh);
4199         if (ret) {
4200                 mlog_errno(ret);
4201                 goto out;
4202         }
4203
4204         mutex_lock_nested(&new_inode->i_mutex, I_MUTEX_CHILD);
4205         ret = ocfs2_inode_lock_nested(new_inode, &new_bh, 1,
4206                                       OI_LS_REFLINK_TARGET);
4207         if (ret) {
4208                 mlog_errno(ret);
4209                 goto out_unlock;
4210         }
4211
4212         ret = ocfs2_create_reflink_node(inode, old_bh,
4213                                         new_inode, new_bh, preserve);
4214         if (ret) {
4215                 mlog_errno(ret);
4216                 goto inode_unlock;
4217         }
4218
4219         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
4220                 ret = ocfs2_reflink_xattrs(inode, old_bh,
4221                                            new_inode, new_bh,
4222                                            preserve);
4223                 if (ret) {
4224                         mlog_errno(ret);
4225                         goto inode_unlock;
4226                 }
4227         }
4228
4229         ret = ocfs2_complete_reflink(inode, old_bh,
4230                                      new_inode, new_bh, preserve);
4231         if (ret)
4232                 mlog_errno(ret);
4233
4234 inode_unlock:
4235         ocfs2_inode_unlock(new_inode, 1);
4236         brelse(new_bh);
4237 out_unlock:
4238         mutex_unlock(&new_inode->i_mutex);
4239 out:
4240         if (!ret) {
4241                 ret = filemap_fdatawait(inode->i_mapping);
4242                 if (ret)
4243                         mlog_errno(ret);
4244         }
4245         return ret;
4246 }
4247
4248 static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
4249                          struct dentry *new_dentry, bool preserve)
4250 {
4251         int error;
4252         struct inode *inode = old_dentry->d_inode;
4253         struct buffer_head *old_bh = NULL;
4254         struct inode *new_orphan_inode = NULL;
4255
4256         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4257                 return -EOPNOTSUPP;
4258
4259         error = ocfs2_create_inode_in_orphan(dir, inode->i_mode,
4260                                              &new_orphan_inode);
4261         if (error) {
4262                 mlog_errno(error);
4263                 goto out;
4264         }
4265
4266         error = ocfs2_inode_lock(inode, &old_bh, 1);
4267         if (error) {
4268                 mlog_errno(error);
4269                 goto out;
4270         }
4271
4272         down_write(&OCFS2_I(inode)->ip_xattr_sem);
4273         down_write(&OCFS2_I(inode)->ip_alloc_sem);
4274         error = __ocfs2_reflink(old_dentry, old_bh,
4275                                 new_orphan_inode, preserve);
4276         up_write(&OCFS2_I(inode)->ip_alloc_sem);
4277         up_write(&OCFS2_I(inode)->ip_xattr_sem);
4278
4279         ocfs2_inode_unlock(inode, 1);
4280         brelse(old_bh);
4281
4282         if (error) {
4283                 mlog_errno(error);
4284                 goto out;
4285         }
4286
4287         /* If the security isn't preserved, we need to re-initialize them. */
4288         if (!preserve) {
4289                 error = ocfs2_init_security_and_acl(dir, new_orphan_inode);
4290                 if (error)
4291                         mlog_errno(error);
4292         }
4293 out:
4294         if (!error) {
4295                 error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
4296                                                        new_dentry);
4297                 if (error)
4298                         mlog_errno(error);
4299         }
4300
4301         if (new_orphan_inode) {
4302                 /*
4303                  * We need to open_unlock the inode no matter whether we
4304                  * succeed or not, so that other nodes can delete it later.
4305                  */
4306                 ocfs2_open_unlock(new_orphan_inode);
4307                 if (error)
4308                         iput(new_orphan_inode);
4309         }
4310
4311         return error;
4312 }
4313
4314 /*
4315  * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
4316  * sys_reflink().  This will go away when vfs_reflink() exists in
4317  * fs/namei.c.
4318  */
4319
4320 /* copied from may_create in VFS. */
4321 static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
4322 {
4323         if (child->d_inode)
4324                 return -EEXIST;
4325         if (IS_DEADDIR(dir))
4326                 return -ENOENT;
4327         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
4328 }
4329
4330 /* copied from user_path_parent. */
4331 static int ocfs2_user_path_parent(const char __user *path,
4332                                   struct nameidata *nd, char **name)
4333 {
4334         char *s = getname(path);
4335         int error;
4336
4337         if (IS_ERR(s))
4338                 return PTR_ERR(s);
4339
4340         error = path_lookup(s, LOOKUP_PARENT, nd);
4341         if (error)
4342                 putname(s);
4343         else
4344                 *name = s;
4345
4346         return error;
4347 }
4348
4349 /**
4350  * ocfs2_vfs_reflink - Create a reference-counted link
4351  *
4352  * @old_dentry:        source dentry + inode
4353  * @dir:       directory to create the target
4354  * @new_dentry:        target dentry
4355  * @preserve:  if true, preserve all file attributes
4356  */
4357 static int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
4358                              struct dentry *new_dentry, bool preserve)
4359 {
4360         struct inode *inode = old_dentry->d_inode;
4361         int error;
4362
4363         if (!inode)
4364                 return -ENOENT;
4365
4366         error = ocfs2_may_create(dir, new_dentry);
4367         if (error)
4368                 return error;
4369
4370         if (dir->i_sb != inode->i_sb)
4371                 return -EXDEV;
4372
4373         /*
4374          * A reflink to an append-only or immutable file cannot be created.
4375          */
4376         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4377                 return -EPERM;
4378
4379         /* Only regular files can be reflinked. */
4380         if (!S_ISREG(inode->i_mode))
4381                 return -EPERM;
4382
4383         /*
4384          * If the caller wants to preserve ownership, they require the
4385          * rights to do so.
4386          */
4387         if (preserve) {
4388                 if ((current_fsuid() != inode->i_uid) && !capable(CAP_CHOWN))
4389                         return -EPERM;
4390                 if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
4391                         return -EPERM;
4392         }
4393
4394         /*
4395          * If the caller is modifying any aspect of the attributes, they
4396          * are not creating a snapshot.  They need read permission on the
4397          * file.
4398          */
4399         if (!preserve) {
4400                 error = inode_permission(inode, MAY_READ);
4401                 if (error)
4402                         return error;
4403         }
4404
4405         mutex_lock(&inode->i_mutex);
4406         dquot_initialize(dir);
4407         error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
4408         mutex_unlock(&inode->i_mutex);
4409         if (!error)
4410                 fsnotify_create(dir, new_dentry);
4411         return error;
4412 }
4413 /*
4414  * Most codes are copied from sys_linkat.
4415  */
4416 int ocfs2_reflink_ioctl(struct inode *inode,
4417                         const char __user *oldname,
4418                         const char __user *newname,
4419                         bool preserve)
4420 {
4421         struct dentry *new_dentry;
4422         struct nameidata nd;
4423         struct path old_path;
4424         int error;
4425         char *to = NULL;
4426
4427         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4428                 return -EOPNOTSUPP;
4429
4430         error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
4431         if (error) {
4432                 mlog_errno(error);
4433                 return error;
4434         }
4435
4436         error = ocfs2_user_path_parent(newname, &nd, &to);
4437         if (error) {
4438                 mlog_errno(error);
4439                 goto out;
4440         }
4441
4442         error = -EXDEV;
4443         if (old_path.mnt != nd.path.mnt)
4444                 goto out_release;
4445         new_dentry = lookup_create(&nd, 0);
4446         error = PTR_ERR(new_dentry);
4447         if (IS_ERR(new_dentry)) {
4448                 mlog_errno(error);
4449                 goto out_unlock;
4450         }
4451
4452         error = mnt_want_write(nd.path.mnt);
4453         if (error) {
4454                 mlog_errno(error);
4455                 goto out_dput;
4456         }
4457
4458         error = ocfs2_vfs_reflink(old_path.dentry,
4459                                   nd.path.dentry->d_inode,
4460                                   new_dentry, preserve);
4461         mnt_drop_write(nd.path.mnt);
4462 out_dput:
4463         dput(new_dentry);
4464 out_unlock:
4465         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
4466 out_release:
4467         path_put(&nd.path);
4468         putname(to);
4469 out:
4470         path_put(&old_path);
4471
4472         return error;
4473 }