linux/fs/ocfs2/refcounttree.c
<<
>>
Prefs
   1/* -*- mode: c; c-basic-offset: 8; -*-
   2 * vim: noexpandtab sw=8 ts=8 sts=0:
   3 *
   4 * refcounttree.c
   5 *
   6 * Copyright (C) 2009 Oracle.  All rights reserved.
   7 *
   8 * This program is free software; you can redistribute it and/or
   9 * modify it under the terms of the GNU General Public
  10 * License version 2 as published by the Free Software Foundation.
  11 *
  12 * This program is distributed in the hope that it will be useful,
  13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15 * General Public License for more details.
  16 */
  17
  18#include <linux/sort.h>
  19#include <cluster/masklog.h>
  20#include "ocfs2.h"
  21#include "inode.h"
  22#include "alloc.h"
  23#include "suballoc.h"
  24#include "journal.h"
  25#include "uptodate.h"
  26#include "super.h"
  27#include "buffer_head_io.h"
  28#include "blockcheck.h"
  29#include "refcounttree.h"
  30#include "sysfile.h"
  31#include "dlmglue.h"
  32#include "extent_map.h"
  33#include "aops.h"
  34#include "xattr.h"
  35#include "namei.h"
  36#include "ocfs2_trace.h"
  37#include "file.h"
  38
  39#include <linux/bio.h>
  40#include <linux/blkdev.h>
  41#include <linux/slab.h>
  42#include <linux/writeback.h>
  43#include <linux/pagevec.h>
  44#include <linux/swap.h>
  45#include <linux/security.h>
  46#include <linux/fsnotify.h>
  47#include <linux/quotaops.h>
  48#include <linux/namei.h>
  49#include <linux/mount.h>
  50#include <linux/posix_acl.h>
  51
  52struct ocfs2_cow_context {
  53        struct inode *inode;
  54        u32 cow_start;
  55        u32 cow_len;
  56        struct ocfs2_extent_tree data_et;
  57        struct ocfs2_refcount_tree *ref_tree;
  58        struct buffer_head *ref_root_bh;
  59        struct ocfs2_alloc_context *meta_ac;
  60        struct ocfs2_alloc_context *data_ac;
  61        struct ocfs2_cached_dealloc_ctxt dealloc;
  62        void *cow_object;
  63        struct ocfs2_post_refcount *post_refcount;
  64        int extra_credits;
  65        int (*get_clusters)(struct ocfs2_cow_context *context,
  66                            u32 v_cluster, u32 *p_cluster,
  67                            u32 *num_clusters,
  68                            unsigned int *extent_flags);
  69        int (*cow_duplicate_clusters)(handle_t *handle,
  70                                      struct inode *inode,
  71                                      u32 cpos, u32 old_cluster,
  72                                      u32 new_cluster, u32 new_len);
  73};
  74
  75static inline struct ocfs2_refcount_tree *
  76cache_info_to_refcount(struct ocfs2_caching_info *ci)
  77{
  78        return container_of(ci, struct ocfs2_refcount_tree, rf_ci);
  79}
  80
  81static int ocfs2_validate_refcount_block(struct super_block *sb,
  82                                         struct buffer_head *bh)
  83{
  84        int rc;
  85        struct ocfs2_refcount_block *rb =
  86                (struct ocfs2_refcount_block *)bh->b_data;
  87
  88        trace_ocfs2_validate_refcount_block((unsigned long long)bh->b_blocknr);
  89
  90        BUG_ON(!buffer_uptodate(bh));
  91
  92        /*
  93         * If the ecc fails, we return the error but otherwise
  94         * leave the filesystem running.  We know any error is
  95         * local to this block.
  96         */
  97        rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &rb->rf_check);
  98        if (rc) {
  99                mlog(ML_ERROR, "Checksum failed for refcount block %llu\n",
 100                     (unsigned long long)bh->b_blocknr);
 101                return rc;
 102        }
 103
 104
 105        if (!OCFS2_IS_VALID_REFCOUNT_BLOCK(rb)) {
 106                rc = ocfs2_error(sb,
 107                                 "Refcount block #%llu has bad signature %.*s\n",
 108                                 (unsigned long long)bh->b_blocknr, 7,
 109                                 rb->rf_signature);
 110                goto out;
 111        }
 112
 113        if (le64_to_cpu(rb->rf_blkno) != bh->b_blocknr) {
 114                rc = ocfs2_error(sb,
 115                                 "Refcount block #%llu has an invalid rf_blkno of %llu\n",
 116                                 (unsigned long long)bh->b_blocknr,
 117                                 (unsigned long long)le64_to_cpu(rb->rf_blkno));
 118                goto out;
 119        }
 120
 121        if (le32_to_cpu(rb->rf_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 122                rc = ocfs2_error(sb,
 123                                 "Refcount block #%llu has an invalid rf_fs_generation of #%u\n",
 124                                 (unsigned long long)bh->b_blocknr,
 125                                 le32_to_cpu(rb->rf_fs_generation));
 126                goto out;
 127        }
 128out:
 129        return rc;
 130}
 131
 132static int ocfs2_read_refcount_block(struct ocfs2_caching_info *ci,
 133                                     u64 rb_blkno,
 134                                     struct buffer_head **bh)
 135{
 136        int rc;
 137        struct buffer_head *tmp = *bh;
 138
 139        rc = ocfs2_read_block(ci, rb_blkno, &tmp,
 140                              ocfs2_validate_refcount_block);
 141
 142        /* If ocfs2_read_block() got us a new bh, pass it up. */
 143        if (!rc && !*bh)
 144                *bh = tmp;
 145
 146        return rc;
 147}
 148
 149static u64 ocfs2_refcount_cache_owner(struct ocfs2_caching_info *ci)
 150{
 151        struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 152
 153        return rf->rf_blkno;
 154}
 155
 156static struct super_block *
 157ocfs2_refcount_cache_get_super(struct ocfs2_caching_info *ci)
 158{
 159        struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 160
 161        return rf->rf_sb;
 162}
 163
 164static void ocfs2_refcount_cache_lock(struct ocfs2_caching_info *ci)
 165{
 166        struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 167
 168        spin_lock(&rf->rf_lock);
 169}
 170
 171static void ocfs2_refcount_cache_unlock(struct ocfs2_caching_info *ci)
 172{
 173        struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 174
 175        spin_unlock(&rf->rf_lock);
 176}
 177
 178static void ocfs2_refcount_cache_io_lock(struct ocfs2_caching_info *ci)
 179{
 180        struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 181
 182        mutex_lock(&rf->rf_io_mutex);
 183}
 184
 185static void ocfs2_refcount_cache_io_unlock(struct ocfs2_caching_info *ci)
 186{
 187        struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 188
 189        mutex_unlock(&rf->rf_io_mutex);
 190}
 191
 192static const struct ocfs2_caching_operations ocfs2_refcount_caching_ops = {
 193        .co_owner               = ocfs2_refcount_cache_owner,
 194        .co_get_super           = ocfs2_refcount_cache_get_super,
 195        .co_cache_lock          = ocfs2_refcount_cache_lock,
 196        .co_cache_unlock        = ocfs2_refcount_cache_unlock,
 197        .co_io_lock             = ocfs2_refcount_cache_io_lock,
 198        .co_io_unlock           = ocfs2_refcount_cache_io_unlock,
 199};
 200
 201static struct ocfs2_refcount_tree *
 202ocfs2_find_refcount_tree(struct ocfs2_super *osb, u64 blkno)
 203{
 204        struct rb_node *n = osb->osb_rf_lock_tree.rb_node;
 205        struct ocfs2_refcount_tree *tree = NULL;
 206
 207        while (n) {
 208                tree = rb_entry(n, struct ocfs2_refcount_tree, rf_node);
 209
 210                if (blkno < tree->rf_blkno)
 211                        n = n->rb_left;
 212                else if (blkno > tree->rf_blkno)
 213                        n = n->rb_right;
 214                else
 215                        return tree;
 216        }
 217
 218        return NULL;
 219}
 220
 221/* osb_lock is already locked. */
 222static void ocfs2_insert_refcount_tree(struct ocfs2_super *osb,
 223                                       struct ocfs2_refcount_tree *new)
 224{
 225        u64 rf_blkno = new->rf_blkno;
 226        struct rb_node *parent = NULL;
 227        struct rb_node **p = &osb->osb_rf_lock_tree.rb_node;
 228        struct ocfs2_refcount_tree *tmp;
 229
 230        while (*p) {
 231                parent = *p;
 232
 233                tmp = rb_entry(parent, struct ocfs2_refcount_tree,
 234                               rf_node);
 235
 236                if (rf_blkno < tmp->rf_blkno)
 237                        p = &(*p)->rb_left;
 238                else if (rf_blkno > tmp->rf_blkno)
 239                        p = &(*p)->rb_right;
 240                else {
 241                        /* This should never happen! */
 242                        mlog(ML_ERROR, "Duplicate refcount block %llu found!\n",
 243                             (unsigned long long)rf_blkno);
 244                        BUG();
 245                }
 246        }
 247
 248        rb_link_node(&new->rf_node, parent, p);
 249        rb_insert_color(&new->rf_node, &osb->osb_rf_lock_tree);
 250}
 251
 252static void ocfs2_free_refcount_tree(struct ocfs2_refcount_tree *tree)
 253{
 254        ocfs2_metadata_cache_exit(&tree->rf_ci);
 255        ocfs2_simple_drop_lockres(OCFS2_SB(tree->rf_sb), &tree->rf_lockres);
 256        ocfs2_lock_res_free(&tree->rf_lockres);
 257        kfree(tree);
 258}
 259
 260static inline void
 261ocfs2_erase_refcount_tree_from_list_no_lock(struct ocfs2_super *osb,
 262                                        struct ocfs2_refcount_tree *tree)
 263{
 264        rb_erase(&tree->rf_node, &osb->osb_rf_lock_tree);
 265        if (osb->osb_ref_tree_lru && osb->osb_ref_tree_lru == tree)
 266                osb->osb_ref_tree_lru = NULL;
 267}
 268
 269static void ocfs2_erase_refcount_tree_from_list(struct ocfs2_super *osb,
 270                                        struct ocfs2_refcount_tree *tree)
 271{
 272        spin_lock(&osb->osb_lock);
 273        ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
 274        spin_unlock(&osb->osb_lock);
 275}
 276
 277static void ocfs2_kref_remove_refcount_tree(struct kref *kref)
 278{
 279        struct ocfs2_refcount_tree *tree =
 280                container_of(kref, struct ocfs2_refcount_tree, rf_getcnt);
 281
 282        ocfs2_free_refcount_tree(tree);
 283}
 284
 285static inline void
 286ocfs2_refcount_tree_get(struct ocfs2_refcount_tree *tree)
 287{
 288        kref_get(&tree->rf_getcnt);
 289}
 290
 291static inline void
 292ocfs2_refcount_tree_put(struct ocfs2_refcount_tree *tree)
 293{
 294        kref_put(&tree->rf_getcnt, ocfs2_kref_remove_refcount_tree);
 295}
 296
 297static inline void ocfs2_init_refcount_tree_ci(struct ocfs2_refcount_tree *new,
 298                                               struct super_block *sb)
 299{
 300        ocfs2_metadata_cache_init(&new->rf_ci, &ocfs2_refcount_caching_ops);
 301        mutex_init(&new->rf_io_mutex);
 302        new->rf_sb = sb;
 303        spin_lock_init(&new->rf_lock);
 304}
 305
 306static inline void ocfs2_init_refcount_tree_lock(struct ocfs2_super *osb,
 307                                        struct ocfs2_refcount_tree *new,
 308                                        u64 rf_blkno, u32 generation)
 309{
 310        init_rwsem(&new->rf_sem);
 311        ocfs2_refcount_lock_res_init(&new->rf_lockres, osb,
 312                                     rf_blkno, generation);
 313}
 314
 315static struct ocfs2_refcount_tree*
 316ocfs2_allocate_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno)
 317{
 318        struct ocfs2_refcount_tree *new;
 319
 320        new = kzalloc(sizeof(struct ocfs2_refcount_tree), GFP_NOFS);
 321        if (!new)
 322                return NULL;
 323
 324        new->rf_blkno = rf_blkno;
 325        kref_init(&new->rf_getcnt);
 326        ocfs2_init_refcount_tree_ci(new, osb->sb);
 327
 328        return new;
 329}
 330
 331static int ocfs2_get_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno,
 332                                   struct ocfs2_refcount_tree **ret_tree)
 333{
 334        int ret = 0;
 335        struct ocfs2_refcount_tree *tree, *new = NULL;
 336        struct buffer_head *ref_root_bh = NULL;
 337        struct ocfs2_refcount_block *ref_rb;
 338
 339        spin_lock(&osb->osb_lock);
 340        if (osb->osb_ref_tree_lru &&
 341            osb->osb_ref_tree_lru->rf_blkno == rf_blkno)
 342                tree = osb->osb_ref_tree_lru;
 343        else
 344                tree = ocfs2_find_refcount_tree(osb, rf_blkno);
 345        if (tree)
 346                goto out;
 347
 348        spin_unlock(&osb->osb_lock);
 349
 350        new = ocfs2_allocate_refcount_tree(osb, rf_blkno);
 351        if (!new) {
 352                ret = -ENOMEM;
 353                mlog_errno(ret);
 354                return ret;
 355        }
 356        /*
 357         * We need the generation to create the refcount tree lock and since
 358         * it isn't changed during the tree modification, we are safe here to
 359         * read without protection.
 360         * We also have to purge the cache after we create the lock since the
 361         * refcount block may have the stale data. It can only be trusted when
 362         * we hold the refcount lock.
 363         */
 364        ret = ocfs2_read_refcount_block(&new->rf_ci, rf_blkno, &ref_root_bh);
 365        if (ret) {
 366                mlog_errno(ret);
 367                ocfs2_metadata_cache_exit(&new->rf_ci);
 368                kfree(new);
 369                return ret;
 370        }
 371
 372        ref_rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
 373        new->rf_generation = le32_to_cpu(ref_rb->rf_generation);
 374        ocfs2_init_refcount_tree_lock(osb, new, rf_blkno,
 375                                      new->rf_generation);
 376        ocfs2_metadata_cache_purge(&new->rf_ci);
 377
 378        spin_lock(&osb->osb_lock);
 379        tree = ocfs2_find_refcount_tree(osb, rf_blkno);
 380        if (tree)
 381                goto out;
 382
 383        ocfs2_insert_refcount_tree(osb, new);
 384
 385        tree = new;
 386        new = NULL;
 387
 388out:
 389        *ret_tree = tree;
 390
 391        osb->osb_ref_tree_lru = tree;
 392
 393        spin_unlock(&osb->osb_lock);
 394
 395        if (new)
 396                ocfs2_free_refcount_tree(new);
 397
 398        brelse(ref_root_bh);
 399        return ret;
 400}
 401
 402static int ocfs2_get_refcount_block(struct inode *inode, u64 *ref_blkno)
 403{
 404        int ret;
 405        struct buffer_head *di_bh = NULL;
 406        struct ocfs2_dinode *di;
 407
 408        ret = ocfs2_read_inode_block(inode, &di_bh);
 409        if (ret) {
 410                mlog_errno(ret);
 411                goto out;
 412        }
 413
 414        BUG_ON(!ocfs2_is_refcount_inode(inode));
 415
 416        di = (struct ocfs2_dinode *)di_bh->b_data;
 417        *ref_blkno = le64_to_cpu(di->i_refcount_loc);
 418        brelse(di_bh);
 419out:
 420        return ret;
 421}
 422
 423static int __ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
 424                                      struct ocfs2_refcount_tree *tree, int rw)
 425{
 426        int ret;
 427
 428        ret = ocfs2_refcount_lock(tree, rw);
 429        if (ret) {
 430                mlog_errno(ret);
 431                goto out;
 432        }
 433
 434        if (rw)
 435                down_write(&tree->rf_sem);
 436        else
 437                down_read(&tree->rf_sem);
 438
 439out:
 440        return ret;
 441}
 442
 443/*
 444 * Lock the refcount tree pointed by ref_blkno and return the tree.
 445 * In most case, we lock the tree and read the refcount block.
 446 * So read it here if the caller really needs it.
 447 *
 448 * If the tree has been re-created by other node, it will free the
 449 * old one and re-create it.
 450 */
 451int ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
 452                             u64 ref_blkno, int rw,
 453                             struct ocfs2_refcount_tree **ret_tree,
 454                             struct buffer_head **ref_bh)
 455{
 456        int ret, delete_tree = 0;
 457        struct ocfs2_refcount_tree *tree = NULL;
 458        struct buffer_head *ref_root_bh = NULL;
 459        struct ocfs2_refcount_block *rb;
 460
 461again:
 462        ret = ocfs2_get_refcount_tree(osb, ref_blkno, &tree);
 463        if (ret) {
 464                mlog_errno(ret);
 465                return ret;
 466        }
 467
 468        ocfs2_refcount_tree_get(tree);
 469
 470        ret = __ocfs2_lock_refcount_tree(osb, tree, rw);
 471        if (ret) {
 472                mlog_errno(ret);
 473                ocfs2_refcount_tree_put(tree);
 474                goto out;
 475        }
 476
 477        ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
 478                                        &ref_root_bh);
 479        if (ret) {
 480                mlog_errno(ret);
 481                ocfs2_unlock_refcount_tree(osb, tree, rw);
 482                goto out;
 483        }
 484
 485        rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
 486        /*
 487         * If the refcount block has been freed and re-created, we may need
 488         * to recreate the refcount tree also.
 489         *
 490         * Here we just remove the tree from the rb-tree, and the last
 491         * kref holder will unlock and delete this refcount_tree.
 492         * Then we goto "again" and ocfs2_get_refcount_tree will create
 493         * the new refcount tree for us.
 494         */
 495        if (tree->rf_generation != le32_to_cpu(rb->rf_generation)) {
 496                if (!tree->rf_removed) {
 497                        ocfs2_erase_refcount_tree_from_list(osb, tree);
 498                        tree->rf_removed = 1;
 499                        delete_tree = 1;
 500                }
 501
 502                ocfs2_unlock_refcount_tree(osb, tree, rw);
 503                /*
 504                 * We get an extra reference when we create the refcount
 505                 * tree, so another put will destroy it.
 506                 */
 507                if (delete_tree)
 508                        ocfs2_refcount_tree_put(tree);
 509                brelse(ref_root_bh);
 510                ref_root_bh = NULL;
 511                goto again;
 512        }
 513
 514        *ret_tree = tree;
 515        if (ref_bh) {
 516                *ref_bh = ref_root_bh;
 517                ref_root_bh = NULL;
 518        }
 519out:
 520        brelse(ref_root_bh);
 521        return ret;
 522}
 523
 524void ocfs2_unlock_refcount_tree(struct ocfs2_super *osb,
 525                                struct ocfs2_refcount_tree *tree, int rw)
 526{
 527        if (rw)
 528                up_write(&tree->rf_sem);
 529        else
 530                up_read(&tree->rf_sem);
 531
 532        ocfs2_refcount_unlock(tree, rw);
 533        ocfs2_refcount_tree_put(tree);
 534}
 535
 536void ocfs2_purge_refcount_trees(struct ocfs2_super *osb)
 537{
 538        struct rb_node *node;
 539        struct ocfs2_refcount_tree *tree;
 540        struct rb_root *root = &osb->osb_rf_lock_tree;
 541
 542        while ((node = rb_last(root)) != NULL) {
 543                tree = rb_entry(node, struct ocfs2_refcount_tree, rf_node);
 544
 545                trace_ocfs2_purge_refcount_trees(
 546                                (unsigned long long) tree->rf_blkno);
 547
 548                rb_erase(&tree->rf_node, root);
 549                ocfs2_free_refcount_tree(tree);
 550        }
 551}
 552
 553/*
 554 * Create a refcount tree for an inode.
 555 * We take for granted that the inode is already locked.
 556 */
 557static int ocfs2_create_refcount_tree(struct inode *inode,
 558                                      struct buffer_head *di_bh)
 559{
 560        int ret;
 561        handle_t *handle = NULL;
 562        struct ocfs2_alloc_context *meta_ac = NULL;
 563        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 564        struct ocfs2_inode_info *oi = OCFS2_I(inode);
 565        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 566        struct buffer_head *new_bh = NULL;
 567        struct ocfs2_refcount_block *rb;
 568        struct ocfs2_refcount_tree *new_tree = NULL, *tree = NULL;
 569        u16 suballoc_bit_start;
 570        u32 num_got;
 571        u64 suballoc_loc, first_blkno;
 572
 573        BUG_ON(ocfs2_is_refcount_inode(inode));
 574
 575        trace_ocfs2_create_refcount_tree(
 576                (unsigned long long)OCFS2_I(inode)->ip_blkno);
 577
 578        ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
 579        if (ret) {
 580                mlog_errno(ret);
 581                goto out;
 582        }
 583
 584        handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_CREATE_CREDITS);
 585        if (IS_ERR(handle)) {
 586                ret = PTR_ERR(handle);
 587                mlog_errno(ret);
 588                goto out;
 589        }
 590
 591        ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 592                                      OCFS2_JOURNAL_ACCESS_WRITE);
 593        if (ret) {
 594                mlog_errno(ret);
 595                goto out_commit;
 596        }
 597
 598        ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
 599                                   &suballoc_bit_start, &num_got,
 600                                   &first_blkno);
 601        if (ret) {
 602                mlog_errno(ret);
 603                goto out_commit;
 604        }
 605
 606        new_tree = ocfs2_allocate_refcount_tree(osb, first_blkno);
 607        if (!new_tree) {
 608                ret = -ENOMEM;
 609                mlog_errno(ret);
 610                goto out_commit;
 611        }
 612
 613        new_bh = sb_getblk(inode->i_sb, first_blkno);
 614        if (!new_bh) {
 615                ret = -ENOMEM;
 616                mlog_errno(ret);
 617                goto out_commit;
 618        }
 619        ocfs2_set_new_buffer_uptodate(&new_tree->rf_ci, new_bh);
 620
 621        ret = ocfs2_journal_access_rb(handle, &new_tree->rf_ci, new_bh,
 622                                      OCFS2_JOURNAL_ACCESS_CREATE);
 623        if (ret) {
 624                mlog_errno(ret);
 625                goto out_commit;
 626        }
 627
 628        /* Initialize ocfs2_refcount_block. */
 629        rb = (struct ocfs2_refcount_block *)new_bh->b_data;
 630        memset(rb, 0, inode->i_sb->s_blocksize);
 631        strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
 632        rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
 633        rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
 634        rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
 635        rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
 636        rb->rf_blkno = cpu_to_le64(first_blkno);
 637        rb->rf_count = cpu_to_le32(1);
 638        rb->rf_records.rl_count =
 639                        cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb));
 640        spin_lock(&osb->osb_lock);
 641        rb->rf_generation = osb->s_next_generation++;
 642        spin_unlock(&osb->osb_lock);
 643
 644        ocfs2_journal_dirty(handle, new_bh);
 645
 646        spin_lock(&oi->ip_lock);
 647        oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
 648        di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 649        di->i_refcount_loc = cpu_to_le64(first_blkno);
 650        spin_unlock(&oi->ip_lock);
 651
 652        trace_ocfs2_create_refcount_tree_blkno((unsigned long long)first_blkno);
 653
 654        ocfs2_journal_dirty(handle, di_bh);
 655
 656        /*
 657         * We have to init the tree lock here since it will use
 658         * the generation number to create it.
 659         */
 660        new_tree->rf_generation = le32_to_cpu(rb->rf_generation);
 661        ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno,
 662                                      new_tree->rf_generation);
 663
 664        spin_lock(&osb->osb_lock);
 665        tree = ocfs2_find_refcount_tree(osb, first_blkno);
 666
 667        /*
 668         * We've just created a new refcount tree in this block.  If
 669         * we found a refcount tree on the ocfs2_super, it must be
 670         * one we just deleted.  We free the old tree before
 671         * inserting the new tree.
 672         */
 673        BUG_ON(tree && tree->rf_generation == new_tree->rf_generation);
 674        if (tree)
 675                ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
 676        ocfs2_insert_refcount_tree(osb, new_tree);
 677        spin_unlock(&osb->osb_lock);
 678        new_tree = NULL;
 679        if (tree)
 680                ocfs2_refcount_tree_put(tree);
 681
 682out_commit:
 683        ocfs2_commit_trans(osb, handle);
 684
 685out:
 686        if (new_tree) {
 687                ocfs2_metadata_cache_exit(&new_tree->rf_ci);
 688                kfree(new_tree);
 689        }
 690
 691        brelse(new_bh);
 692        if (meta_ac)
 693                ocfs2_free_alloc_context(meta_ac);
 694
 695        return ret;
 696}
 697
 698static int ocfs2_set_refcount_tree(struct inode *inode,
 699                                   struct buffer_head *di_bh,
 700                                   u64 refcount_loc)
 701{
 702        int ret;
 703        handle_t *handle = NULL;
 704        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 705        struct ocfs2_inode_info *oi = OCFS2_I(inode);
 706        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 707        struct buffer_head *ref_root_bh = NULL;
 708        struct ocfs2_refcount_block *rb;
 709        struct ocfs2_refcount_tree *ref_tree;
 710
 711        BUG_ON(ocfs2_is_refcount_inode(inode));
 712
 713        ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
 714                                       &ref_tree, &ref_root_bh);
 715        if (ret) {
 716                mlog_errno(ret);
 717                return ret;
 718        }
 719
 720        handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS);
 721        if (IS_ERR(handle)) {
 722                ret = PTR_ERR(handle);
 723                mlog_errno(ret);
 724                goto out;
 725        }
 726
 727        ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 728                                      OCFS2_JOURNAL_ACCESS_WRITE);
 729        if (ret) {
 730                mlog_errno(ret);
 731                goto out_commit;
 732        }
 733
 734        ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh,
 735                                      OCFS2_JOURNAL_ACCESS_WRITE);
 736        if (ret) {
 737                mlog_errno(ret);
 738                goto out_commit;
 739        }
 740
 741        rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
 742        le32_add_cpu(&rb->rf_count, 1);
 743
 744        ocfs2_journal_dirty(handle, ref_root_bh);
 745
 746        spin_lock(&oi->ip_lock);
 747        oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
 748        di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 749        di->i_refcount_loc = cpu_to_le64(refcount_loc);
 750        spin_unlock(&oi->ip_lock);
 751        ocfs2_journal_dirty(handle, di_bh);
 752
 753out_commit:
 754        ocfs2_commit_trans(osb, handle);
 755out:
 756        ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
 757        brelse(ref_root_bh);
 758
 759        return ret;
 760}
 761
 762int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh)
 763{
 764        int ret, delete_tree = 0;
 765        handle_t *handle = NULL;
 766        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 767        struct ocfs2_inode_info *oi = OCFS2_I(inode);
 768        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 769        struct ocfs2_refcount_block *rb;
 770        struct inode *alloc_inode = NULL;
 771        struct buffer_head *alloc_bh = NULL;
 772        struct buffer_head *blk_bh = NULL;
 773        struct ocfs2_refcount_tree *ref_tree;
 774        int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS;
 775        u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc);
 776        u16 bit = 0;
 777
 778        if (!ocfs2_is_refcount_inode(inode))
 779                return 0;
 780
 781        BUG_ON(!ref_blkno);
 782        ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh);
 783        if (ret) {
 784                mlog_errno(ret);
 785                return ret;
 786        }
 787
 788        rb = (struct ocfs2_refcount_block *)blk_bh->b_data;
 789
 790        /*
 791         * If we are the last user, we need to free the block.
 792         * So lock the allocator ahead.
 793         */
 794        if (le32_to_cpu(rb->rf_count) == 1) {
 795                blk = le64_to_cpu(rb->rf_blkno);
 796                bit = le16_to_cpu(rb->rf_suballoc_bit);
 797                if (rb->rf_suballoc_loc)
 798                        bg_blkno = le64_to_cpu(rb->rf_suballoc_loc);
 799                else
 800                        bg_blkno = ocfs2_which_suballoc_group(blk, bit);
 801
 802                alloc_inode = ocfs2_get_system_file_inode(osb,
 803                                        EXTENT_ALLOC_SYSTEM_INODE,
 804                                        le16_to_cpu(rb->rf_suballoc_slot));
 805                if (!alloc_inode) {
 806                        ret = -ENOMEM;
 807                        mlog_errno(ret);
 808                        goto out;
 809                }
 810                inode_lock(alloc_inode);
 811
 812                ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1);
 813                if (ret) {
 814                        mlog_errno(ret);
 815                        goto out_mutex;
 816                }
 817
 818                credits += OCFS2_SUBALLOC_FREE;
 819        }
 820
 821        handle = ocfs2_start_trans(osb, credits);
 822        if (IS_ERR(handle)) {
 823                ret = PTR_ERR(handle);
 824                mlog_errno(ret);
 825                goto out_unlock;
 826        }
 827
 828        ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 829                                      OCFS2_JOURNAL_ACCESS_WRITE);
 830        if (ret) {
 831                mlog_errno(ret);
 832                goto out_commit;
 833        }
 834
 835        ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh,
 836                                      OCFS2_JOURNAL_ACCESS_WRITE);
 837        if (ret) {
 838                mlog_errno(ret);
 839                goto out_commit;
 840        }
 841
 842        spin_lock(&oi->ip_lock);
 843        oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL;
 844        di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 845        di->i_refcount_loc = 0;
 846        spin_unlock(&oi->ip_lock);
 847        ocfs2_journal_dirty(handle, di_bh);
 848
 849        le32_add_cpu(&rb->rf_count , -1);
 850        ocfs2_journal_dirty(handle, blk_bh);
 851
 852        if (!rb->rf_count) {
 853                delete_tree = 1;
 854                ocfs2_erase_refcount_tree_from_list(osb, ref_tree);
 855                ret = ocfs2_free_suballoc_bits(handle, alloc_inode,
 856                                               alloc_bh, bit, bg_blkno, 1);
 857                if (ret)
 858                        mlog_errno(ret);
 859        }
 860
 861out_commit:
 862        ocfs2_commit_trans(osb, handle);
 863out_unlock:
 864        if (alloc_inode) {
 865                ocfs2_inode_unlock(alloc_inode, 1);
 866                brelse(alloc_bh);
 867        }
 868out_mutex:
 869        if (alloc_inode) {
 870                inode_unlock(alloc_inode);
 871                iput(alloc_inode);
 872        }
 873out:
 874        ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
 875        if (delete_tree)
 876                ocfs2_refcount_tree_put(ref_tree);
 877        brelse(blk_bh);
 878
 879        return ret;
 880}
 881
 882static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info *ci,
 883                                          struct buffer_head *ref_leaf_bh,
 884                                          u64 cpos, unsigned int len,
 885                                          struct ocfs2_refcount_rec *ret_rec,
 886                                          int *index)
 887{
 888        int i = 0;
 889        struct ocfs2_refcount_block *rb =
 890                (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
 891        struct ocfs2_refcount_rec *rec = NULL;
 892
 893        for (; i < le16_to_cpu(rb->rf_records.rl_used); i++) {
 894                rec = &rb->rf_records.rl_recs[i];
 895
 896                if (le64_to_cpu(rec->r_cpos) +
 897                    le32_to_cpu(rec->r_clusters) <= cpos)
 898                        continue;
 899                else if (le64_to_cpu(rec->r_cpos) > cpos)
 900                        break;
 901
 902                /* ok, cpos fail in this rec. Just return. */
 903                if (ret_rec)
 904                        *ret_rec = *rec;
 905                goto out;
 906        }
 907
 908        if (ret_rec) {
 909                /* We meet with a hole here, so fake the rec. */
 910                ret_rec->r_cpos = cpu_to_le64(cpos);
 911                ret_rec->r_refcount = 0;
 912                if (i < le16_to_cpu(rb->rf_records.rl_used) &&
 913                    le64_to_cpu(rec->r_cpos) < cpos + len)
 914                        ret_rec->r_clusters =
 915                                cpu_to_le32(le64_to_cpu(rec->r_cpos) - cpos);
 916                else
 917                        ret_rec->r_clusters = cpu_to_le32(len);
 918        }
 919
 920out:
 921        *index = i;
 922}
 923
 924/*
 925 * Try to remove refcount tree. The mechanism is:
 926 * 1) Check whether i_clusters == 0, if no, exit.
 927 * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
 928 * 3) Check whether we have inline xattr stored outside, if yes, exit.
 929 * 4) Remove the tree.
 930 */
 931int ocfs2_try_remove_refcount_tree(struct inode *inode,
 932                                   struct buffer_head *di_bh)
 933{
 934        int ret;
 935        struct ocfs2_inode_info *oi = OCFS2_I(inode);
 936        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 937
 938        down_write(&oi->ip_xattr_sem);
 939        down_write(&oi->ip_alloc_sem);
 940
 941        if (oi->ip_clusters)
 942                goto out;
 943
 944        if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
 945                goto out;
 946
 947        if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
 948            ocfs2_has_inline_xattr_value_outside(inode, di))
 949                goto out;
 950
 951        ret = ocfs2_remove_refcount_tree(inode, di_bh);
 952        if (ret)
 953                mlog_errno(ret);
 954out:
 955        up_write(&oi->ip_alloc_sem);
 956        up_write(&oi->ip_xattr_sem);
 957        return 0;
 958}
 959
 960/*
 961 * Find the end range for a leaf refcount block indicated by
 962 * el->l_recs[index].e_blkno.
 963 */
 964static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info *ci,
 965                                       struct buffer_head *ref_root_bh,
 966                                       struct ocfs2_extent_block *eb,
 967                                       struct ocfs2_extent_list *el,
 968                                       int index,  u32 *cpos_end)
 969{
 970        int ret, i, subtree_root;
 971        u32 cpos;
 972        u64 blkno;
 973        struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
 974        struct ocfs2_path *left_path = NULL, *right_path = NULL;
 975        struct ocfs2_extent_tree et;
 976        struct ocfs2_extent_list *tmp_el;
 977
 978        if (index < le16_to_cpu(el->l_next_free_rec) - 1) {
 979                /*
 980                 * We have a extent rec after index, so just use the e_cpos
 981                 * of the next extent rec.
 982                 */
 983                *cpos_end = le32_to_cpu(el->l_recs[index+1].e_cpos);
 984                return 0;
 985        }
 986
 987        if (!eb || (eb && !eb->h_next_leaf_blk)) {
 988                /*
 989                 * We are the last extent rec, so any high cpos should
 990                 * be stored in this leaf refcount block.
 991                 */
 992                *cpos_end = UINT_MAX;
 993                return 0;
 994        }
 995
 996        /*
 997         * If the extent block isn't the last one, we have to find
 998         * the subtree root between this extent block and the next
 999         * leaf extent block and get the corresponding e_cpos from
1000         * the subroot. Otherwise we may corrupt the b-tree.
1001         */
1002        ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1003
1004        left_path = ocfs2_new_path_from_et(&et);
1005        if (!left_path) {
1006                ret = -ENOMEM;
1007                mlog_errno(ret);
1008                goto out;
1009        }
1010
1011        cpos = le32_to_cpu(eb->h_list.l_recs[index].e_cpos);
1012        ret = ocfs2_find_path(ci, left_path, cpos);
1013        if (ret) {
1014                mlog_errno(ret);
1015                goto out;
1016        }
1017
1018        right_path = ocfs2_new_path_from_path(left_path);
1019        if (!right_path) {
1020                ret = -ENOMEM;
1021                mlog_errno(ret);
1022                goto out;
1023        }
1024
1025        ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, &cpos);
1026        if (ret) {
1027                mlog_errno(ret);
1028                goto out;
1029        }
1030
1031        ret = ocfs2_find_path(ci, right_path, cpos);
1032        if (ret) {
1033                mlog_errno(ret);
1034                goto out;
1035        }
1036
1037        subtree_root = ocfs2_find_subtree_root(&et, left_path,
1038                                               right_path);
1039
1040        tmp_el = left_path->p_node[subtree_root].el;
1041        blkno = left_path->p_node[subtree_root+1].bh->b_blocknr;
1042        for (i = 0; i < le16_to_cpu(tmp_el->l_next_free_rec); i++) {
1043                if (le64_to_cpu(tmp_el->l_recs[i].e_blkno) == blkno) {
1044                        *cpos_end = le32_to_cpu(tmp_el->l_recs[i+1].e_cpos);
1045                        break;
1046                }
1047        }
1048
1049        BUG_ON(i == le16_to_cpu(tmp_el->l_next_free_rec));
1050
1051out:
1052        ocfs2_free_path(left_path);
1053        ocfs2_free_path(right_path);
1054        return ret;
1055}
1056
1057/*
1058 * Given a cpos and len, try to find the refcount record which contains cpos.
1059 * 1. If cpos can be found in one refcount record, return the record.
1060 * 2. If cpos can't be found, return a fake record which start from cpos
1061 *    and end at a small value between cpos+len and start of the next record.
1062 *    This fake record has r_refcount = 0.
1063 */
1064static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
1065                                  struct buffer_head *ref_root_bh,
1066                                  u64 cpos, unsigned int len,
1067                                  struct ocfs2_refcount_rec *ret_rec,
1068                                  int *index,
1069                                  struct buffer_head **ret_bh)
1070{
1071        int ret = 0, i, found;
1072        u32 low_cpos, uninitialized_var(cpos_end);
1073        struct ocfs2_extent_list *el;
1074        struct ocfs2_extent_rec *rec = NULL;
1075        struct ocfs2_extent_block *eb = NULL;
1076        struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
1077        struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1078        struct ocfs2_refcount_block *rb =
1079                        (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1080
1081        if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)) {
1082                ocfs2_find_refcount_rec_in_rl(ci, ref_root_bh, cpos, len,
1083                                              ret_rec, index);
1084                *ret_bh = ref_root_bh;
1085                get_bh(ref_root_bh);
1086                return 0;
1087        }
1088
1089        el = &rb->rf_list;
1090        low_cpos = cpos & OCFS2_32BIT_POS_MASK;
1091
1092        if (el->l_tree_depth) {
1093                ret = ocfs2_find_leaf(ci, el, low_cpos, &eb_bh);
1094                if (ret) {
1095                        mlog_errno(ret);
1096                        goto out;
1097                }
1098
1099                eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1100                el = &eb->h_list;
1101
1102                if (el->l_tree_depth) {
1103                        ret = ocfs2_error(sb,
1104                                          "refcount tree %llu has non zero tree depth in leaf btree tree block %llu\n",
1105                                          (unsigned long long)ocfs2_metadata_cache_owner(ci),
1106                                          (unsigned long long)eb_bh->b_blocknr);
1107                        goto out;
1108                }
1109        }
1110
1111        found = 0;
1112        for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1113                rec = &el->l_recs[i];
1114
1115                if (le32_to_cpu(rec->e_cpos) <= low_cpos) {
1116                        found = 1;
1117                        break;
1118                }
1119        }
1120
1121        if (found) {
1122                ret = ocfs2_get_refcount_cpos_end(ci, ref_root_bh,
1123                                                  eb, el, i, &cpos_end);
1124                if (ret) {
1125                        mlog_errno(ret);
1126                        goto out;
1127                }
1128
1129                if (cpos_end < low_cpos + len)
1130                        len = cpos_end - low_cpos;
1131        }
1132
1133        ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
1134                                        &ref_leaf_bh);
1135        if (ret) {
1136                mlog_errno(ret);
1137                goto out;
1138        }
1139
1140        ocfs2_find_refcount_rec_in_rl(ci, ref_leaf_bh, cpos, len,
1141                                      ret_rec, index);
1142        *ret_bh = ref_leaf_bh;
1143out:
1144        brelse(eb_bh);
1145        return ret;
1146}
1147
1148enum ocfs2_ref_rec_contig {
1149        REF_CONTIG_NONE = 0,
1150        REF_CONTIG_LEFT,
1151        REF_CONTIG_RIGHT,
1152        REF_CONTIG_LEFTRIGHT,
1153};
1154
1155static enum ocfs2_ref_rec_contig
1156        ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block *rb,
1157                                    int index)
1158{
1159        if ((rb->rf_records.rl_recs[index].r_refcount ==
1160            rb->rf_records.rl_recs[index + 1].r_refcount) &&
1161            (le64_to_cpu(rb->rf_records.rl_recs[index].r_cpos) +
1162            le32_to_cpu(rb->rf_records.rl_recs[index].r_clusters) ==
1163            le64_to_cpu(rb->rf_records.rl_recs[index + 1].r_cpos)))
1164                return REF_CONTIG_RIGHT;
1165
1166        return REF_CONTIG_NONE;
1167}
1168
1169static enum ocfs2_ref_rec_contig
1170        ocfs2_refcount_rec_contig(struct ocfs2_refcount_block *rb,
1171                                  int index)
1172{
1173        enum ocfs2_ref_rec_contig ret = REF_CONTIG_NONE;
1174
1175        if (index < le16_to_cpu(rb->rf_records.rl_used) - 1)
1176                ret = ocfs2_refcount_rec_adjacent(rb, index);
1177
1178        if (index > 0) {
1179                enum ocfs2_ref_rec_contig tmp;
1180
1181                tmp = ocfs2_refcount_rec_adjacent(rb, index - 1);
1182
1183                if (tmp == REF_CONTIG_RIGHT) {
1184                        if (ret == REF_CONTIG_RIGHT)
1185                                ret = REF_CONTIG_LEFTRIGHT;
1186                        else
1187                                ret = REF_CONTIG_LEFT;
1188                }
1189        }
1190
1191        return ret;
1192}
1193
1194static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block *rb,
1195                                           int index)
1196{
1197        BUG_ON(rb->rf_records.rl_recs[index].r_refcount !=
1198               rb->rf_records.rl_recs[index+1].r_refcount);
1199
1200        le32_add_cpu(&rb->rf_records.rl_recs[index].r_clusters,
1201                     le32_to_cpu(rb->rf_records.rl_recs[index+1].r_clusters));
1202
1203        if (index < le16_to_cpu(rb->rf_records.rl_used) - 2)
1204                memmove(&rb->rf_records.rl_recs[index + 1],
1205                        &rb->rf_records.rl_recs[index + 2],
1206                        sizeof(struct ocfs2_refcount_rec) *
1207                        (le16_to_cpu(rb->rf_records.rl_used) - index - 2));
1208
1209        memset(&rb->rf_records.rl_recs[le16_to_cpu(rb->rf_records.rl_used) - 1],
1210               0, sizeof(struct ocfs2_refcount_rec));
1211        le16_add_cpu(&rb->rf_records.rl_used, -1);
1212}
1213
1214/*
1215 * Merge the refcount rec if we are contiguous with the adjacent recs.
1216 */
1217static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
1218                                     int index)
1219{
1220        enum ocfs2_ref_rec_contig contig =
1221                                ocfs2_refcount_rec_contig(rb, index);
1222
1223        if (contig == REF_CONTIG_NONE)
1224                return;
1225
1226        if (contig == REF_CONTIG_LEFT || contig == REF_CONTIG_LEFTRIGHT) {
1227                BUG_ON(index == 0);
1228                index--;
1229        }
1230
1231        ocfs2_rotate_refcount_rec_left(rb, index);
1232
1233        if (contig == REF_CONTIG_LEFTRIGHT)
1234                ocfs2_rotate_refcount_rec_left(rb, index);
1235}
1236
1237/*
1238 * Change the refcount indexed by "index" in ref_bh.
1239 * If refcount reaches 0, remove it.
1240 */
1241static int ocfs2_change_refcount_rec(handle_t *handle,
1242                                     struct ocfs2_caching_info *ci,
1243                                     struct buffer_head *ref_leaf_bh,
1244                                     int index, int merge, int change)
1245{
1246        int ret;
1247        struct ocfs2_refcount_block *rb =
1248                        (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1249        struct ocfs2_refcount_list *rl = &rb->rf_records;
1250        struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
1251
1252        ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1253                                      OCFS2_JOURNAL_ACCESS_WRITE);
1254        if (ret) {
1255                mlog_errno(ret);
1256                goto out;
1257        }
1258
1259        trace_ocfs2_change_refcount_rec(
1260                (unsigned long long)ocfs2_metadata_cache_owner(ci),
1261                index, le32_to_cpu(rec->r_refcount), change);
1262        le32_add_cpu(&rec->r_refcount, change);
1263
1264        if (!rec->r_refcount) {
1265                if (index != le16_to_cpu(rl->rl_used) - 1) {
1266                        memmove(rec, rec + 1,
1267                                (le16_to_cpu(rl->rl_used) - index - 1) *
1268                                sizeof(struct ocfs2_refcount_rec));
1269                        memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
1270                               0, sizeof(struct ocfs2_refcount_rec));
1271                }
1272
1273                le16_add_cpu(&rl->rl_used, -1);
1274        } else if (merge)
1275                ocfs2_refcount_rec_merge(rb, index);
1276
1277        ocfs2_journal_dirty(handle, ref_leaf_bh);
1278out:
1279        return ret;
1280}
1281
1282static int ocfs2_expand_inline_ref_root(handle_t *handle,
1283                                        struct ocfs2_caching_info *ci,
1284                                        struct buffer_head *ref_root_bh,
1285                                        struct buffer_head **ref_leaf_bh,
1286                                        struct ocfs2_alloc_context *meta_ac)
1287{
1288        int ret;
1289        u16 suballoc_bit_start;
1290        u32 num_got;
1291        u64 suballoc_loc, blkno;
1292        struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1293        struct buffer_head *new_bh = NULL;
1294        struct ocfs2_refcount_block *new_rb;
1295        struct ocfs2_refcount_block *root_rb =
1296                        (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1297
1298        ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1299                                      OCFS2_JOURNAL_ACCESS_WRITE);
1300        if (ret) {
1301                mlog_errno(ret);
1302                goto out;
1303        }
1304
1305        ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1306                                   &suballoc_bit_start, &num_got,
1307                                   &blkno);
1308        if (ret) {
1309                mlog_errno(ret);
1310                goto out;
1311        }
1312
1313        new_bh = sb_getblk(sb, blkno);
1314        if (new_bh == NULL) {
1315                ret = -ENOMEM;
1316                mlog_errno(ret);
1317                goto out;
1318        }
1319        ocfs2_set_new_buffer_uptodate(ci, new_bh);
1320
1321        ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1322                                      OCFS2_JOURNAL_ACCESS_CREATE);
1323        if (ret) {
1324                mlog_errno(ret);
1325                goto out;
1326        }
1327
1328        /*
1329         * Initialize ocfs2_refcount_block.
1330         * It should contain the same information as the old root.
1331         * so just memcpy it and change the corresponding field.
1332         */
1333        memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
1334
1335        new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1336        new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1337        new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1338        new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1339        new_rb->rf_blkno = cpu_to_le64(blkno);
1340        new_rb->rf_cpos = cpu_to_le32(0);
1341        new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1342        new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1343        ocfs2_journal_dirty(handle, new_bh);
1344
1345        /* Now change the root. */
1346        memset(&root_rb->rf_list, 0, sb->s_blocksize -
1347               offsetof(struct ocfs2_refcount_block, rf_list));
1348        root_rb->rf_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_rb(sb));
1349        root_rb->rf_clusters = cpu_to_le32(1);
1350        root_rb->rf_list.l_next_free_rec = cpu_to_le16(1);
1351        root_rb->rf_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
1352        root_rb->rf_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
1353        root_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_TREE_FL);
1354
1355        ocfs2_journal_dirty(handle, ref_root_bh);
1356
1357        trace_ocfs2_expand_inline_ref_root((unsigned long long)blkno,
1358                le16_to_cpu(new_rb->rf_records.rl_used));
1359
1360        *ref_leaf_bh = new_bh;
1361        new_bh = NULL;
1362out:
1363        brelse(new_bh);
1364        return ret;
1365}
1366
1367static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec *prev,
1368                                           struct ocfs2_refcount_rec *next)
1369{
1370        if (ocfs2_get_ref_rec_low_cpos(prev) + le32_to_cpu(prev->r_clusters) <=
1371                ocfs2_get_ref_rec_low_cpos(next))
1372                return 1;
1373
1374        return 0;
1375}
1376
1377static int cmp_refcount_rec_by_low_cpos(const void *a, const void *b)
1378{
1379        const struct ocfs2_refcount_rec *l = a, *r = b;
1380        u32 l_cpos = ocfs2_get_ref_rec_low_cpos(l);
1381        u32 r_cpos = ocfs2_get_ref_rec_low_cpos(r);
1382
1383        if (l_cpos > r_cpos)
1384                return 1;
1385        if (l_cpos < r_cpos)
1386                return -1;
1387        return 0;
1388}
1389
1390static int cmp_refcount_rec_by_cpos(const void *a, const void *b)
1391{
1392        const struct ocfs2_refcount_rec *l = a, *r = b;
1393        u64 l_cpos = le64_to_cpu(l->r_cpos);
1394        u64 r_cpos = le64_to_cpu(r->r_cpos);
1395
1396        if (l_cpos > r_cpos)
1397                return 1;
1398        if (l_cpos < r_cpos)
1399                return -1;
1400        return 0;
1401}
1402
1403static void swap_refcount_rec(void *a, void *b, int size)
1404{
1405        struct ocfs2_refcount_rec *l = a, *r = b;
1406
1407        swap(*l, *r);
1408}
1409
1410/*
1411 * The refcount cpos are ordered by their 64bit cpos,
1412 * But we will use the low 32 bit to be the e_cpos in the b-tree.
1413 * So we need to make sure that this pos isn't intersected with others.
1414 *
1415 * Note: The refcount block is already sorted by their low 32 bit cpos,
1416 *       So just try the middle pos first, and we will exit when we find
1417 *       the good position.
1418 */
1419static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
1420                                         u32 *split_pos, int *split_index)
1421{
1422        int num_used = le16_to_cpu(rl->rl_used);
1423        int delta, middle = num_used / 2;
1424
1425        for (delta = 0; delta < middle; delta++) {
1426                /* Let's check delta earlier than middle */
1427                if (ocfs2_refcount_rec_no_intersect(
1428                                        &rl->rl_recs[middle - delta - 1],
1429                                        &rl->rl_recs[middle - delta])) {
1430                        *split_index = middle - delta;
1431                        break;
1432                }
1433
1434                /* For even counts, don't walk off the end */
1435                if ((middle + delta + 1) == num_used)
1436                        continue;
1437
1438                /* Now try delta past middle */
1439                if (ocfs2_refcount_rec_no_intersect(
1440                                        &rl->rl_recs[middle + delta],
1441                                        &rl->rl_recs[middle + delta + 1])) {
1442                        *split_index = middle + delta + 1;
1443                        break;
1444                }
1445        }
1446
1447        if (delta >= middle)
1448                return -ENOSPC;
1449
1450        *split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
1451        return 0;
1452}
1453
1454static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
1455                                            struct buffer_head *new_bh,
1456                                            u32 *split_cpos)
1457{
1458        int split_index = 0, num_moved, ret;
1459        u32 cpos = 0;
1460        struct ocfs2_refcount_block *rb =
1461                        (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1462        struct ocfs2_refcount_list *rl = &rb->rf_records;
1463        struct ocfs2_refcount_block *new_rb =
1464                        (struct ocfs2_refcount_block *)new_bh->b_data;
1465        struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
1466
1467        trace_ocfs2_divide_leaf_refcount_block(
1468                (unsigned long long)ref_leaf_bh->b_blocknr,
1469                le16_to_cpu(rl->rl_count), le16_to_cpu(rl->rl_used));
1470
1471        /*
1472         * XXX: Improvement later.
1473         * If we know all the high 32 bit cpos is the same, no need to sort.
1474         *
1475         * In order to make the whole process safe, we do:
1476         * 1. sort the entries by their low 32 bit cpos first so that we can
1477         *    find the split cpos easily.
1478         * 2. call ocfs2_insert_extent to insert the new refcount block.
1479         * 3. move the refcount rec to the new block.
1480         * 4. sort the entries by their 64 bit cpos.
1481         * 5. dirty the new_rb and rb.
1482         */
1483        sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1484             sizeof(struct ocfs2_refcount_rec),
1485             cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
1486
1487        ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
1488        if (ret) {
1489                mlog_errno(ret);
1490                return ret;
1491        }
1492
1493        new_rb->rf_cpos = cpu_to_le32(cpos);
1494
1495        /* move refcount records starting from split_index to the new block. */
1496        num_moved = le16_to_cpu(rl->rl_used) - split_index;
1497        memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
1498               num_moved * sizeof(struct ocfs2_refcount_rec));
1499
1500        /*ok, remove the entries we just moved over to the other block. */
1501        memset(&rl->rl_recs[split_index], 0,
1502               num_moved * sizeof(struct ocfs2_refcount_rec));
1503
1504        /* change old and new rl_used accordingly. */
1505        le16_add_cpu(&rl->rl_used, -num_moved);
1506        new_rl->rl_used = cpu_to_le16(num_moved);
1507
1508        sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1509             sizeof(struct ocfs2_refcount_rec),
1510             cmp_refcount_rec_by_cpos, swap_refcount_rec);
1511
1512        sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
1513             sizeof(struct ocfs2_refcount_rec),
1514             cmp_refcount_rec_by_cpos, swap_refcount_rec);
1515
1516        *split_cpos = cpos;
1517        return 0;
1518}
1519
1520static int ocfs2_new_leaf_refcount_block(handle_t *handle,
1521                                         struct ocfs2_caching_info *ci,
1522                                         struct buffer_head *ref_root_bh,
1523                                         struct buffer_head *ref_leaf_bh,
1524                                         struct ocfs2_alloc_context *meta_ac)
1525{
1526        int ret;
1527        u16 suballoc_bit_start;
1528        u32 num_got, new_cpos;
1529        u64 suballoc_loc, blkno;
1530        struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1531        struct ocfs2_refcount_block *root_rb =
1532                        (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1533        struct buffer_head *new_bh = NULL;
1534        struct ocfs2_refcount_block *new_rb;
1535        struct ocfs2_extent_tree ref_et;
1536
1537        BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
1538
1539        ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1540                                      OCFS2_JOURNAL_ACCESS_WRITE);
1541        if (ret) {
1542                mlog_errno(ret);
1543                goto out;
1544        }
1545
1546        ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1547                                      OCFS2_JOURNAL_ACCESS_WRITE);
1548        if (ret) {
1549                mlog_errno(ret);
1550                goto out;
1551        }
1552
1553        ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1554                                   &suballoc_bit_start, &num_got,
1555                                   &blkno);
1556        if (ret) {
1557                mlog_errno(ret);
1558                goto out;
1559        }
1560
1561        new_bh = sb_getblk(sb, blkno);
1562        if (new_bh == NULL) {
1563                ret = -ENOMEM;
1564                mlog_errno(ret);
1565                goto out;
1566        }
1567        ocfs2_set_new_buffer_uptodate(ci, new_bh);
1568
1569        ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1570                                      OCFS2_JOURNAL_ACCESS_CREATE);
1571        if (ret) {
1572                mlog_errno(ret);
1573                goto out;
1574        }
1575
1576        /* Initialize ocfs2_refcount_block. */
1577        new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1578        memset(new_rb, 0, sb->s_blocksize);
1579        strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
1580        new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1581        new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1582        new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1583        new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1584        new_rb->rf_blkno = cpu_to_le64(blkno);
1585        new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1586        new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1587        new_rb->rf_records.rl_count =
1588                                cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
1589        new_rb->rf_generation = root_rb->rf_generation;
1590
1591        ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
1592        if (ret) {
1593                mlog_errno(ret);
1594                goto out;
1595        }
1596
1597        ocfs2_journal_dirty(handle, ref_leaf_bh);
1598        ocfs2_journal_dirty(handle, new_bh);
1599
1600        ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
1601
1602        trace_ocfs2_new_leaf_refcount_block(
1603                        (unsigned long long)new_bh->b_blocknr, new_cpos);
1604
1605        /* Insert the new leaf block with the specific offset cpos. */
1606        ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
1607                                  1, 0, meta_ac);
1608        if (ret)
1609                mlog_errno(ret);
1610
1611out:
1612        brelse(new_bh);
1613        return ret;
1614}
1615
1616static int ocfs2_expand_refcount_tree(handle_t *handle,
1617                                      struct ocfs2_caching_info *ci,
1618                                      struct buffer_head *ref_root_bh,
1619                                      struct buffer_head *ref_leaf_bh,
1620                                      struct ocfs2_alloc_context *meta_ac)
1621{
1622        int ret;
1623        struct buffer_head *expand_bh = NULL;
1624
1625        if (ref_root_bh == ref_leaf_bh) {
1626                /*
1627                 * the old root bh hasn't been expanded to a b-tree,
1628                 * so expand it first.
1629                 */
1630                ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
1631                                                   &expand_bh, meta_ac);
1632                if (ret) {
1633                        mlog_errno(ret);
1634                        goto out;
1635                }
1636        } else {
1637                expand_bh = ref_leaf_bh;
1638                get_bh(expand_bh);
1639        }
1640
1641
1642        /* Now add a new refcount block into the tree.*/
1643        ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
1644                                            expand_bh, meta_ac);
1645        if (ret)
1646                mlog_errno(ret);
1647out:
1648        brelse(expand_bh);
1649        return ret;
1650}
1651
1652/*
1653 * Adjust the extent rec in b-tree representing ref_leaf_bh.
1654 *
1655 * Only called when we have inserted a new refcount rec at index 0
1656 * which means ocfs2_extent_rec.e_cpos may need some change.
1657 */
1658static int ocfs2_adjust_refcount_rec(handle_t *handle,
1659                                     struct ocfs2_caching_info *ci,
1660                                     struct buffer_head *ref_root_bh,
1661                                     struct buffer_head *ref_leaf_bh,
1662                                     struct ocfs2_refcount_rec *rec)
1663{
1664        int ret = 0, i;
1665        u32 new_cpos, old_cpos;
1666        struct ocfs2_path *path = NULL;
1667        struct ocfs2_extent_tree et;
1668        struct ocfs2_refcount_block *rb =
1669                (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1670        struct ocfs2_extent_list *el;
1671
1672        if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
1673                goto out;
1674
1675        rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1676        old_cpos = le32_to_cpu(rb->rf_cpos);
1677        new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
1678        if (old_cpos <= new_cpos)
1679                goto out;
1680
1681        ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1682
1683        path = ocfs2_new_path_from_et(&et);
1684        if (!path) {
1685                ret = -ENOMEM;
1686                mlog_errno(ret);
1687                goto out;
1688        }
1689
1690        ret = ocfs2_find_path(ci, path, old_cpos);
1691        if (ret) {
1692                mlog_errno(ret);
1693                goto out;
1694        }
1695
1696        /*
1697         * 2 more credits, one for the leaf refcount block, one for
1698         * the extent block contains the extent rec.
1699         */
1700        ret = ocfs2_extend_trans(handle, 2);
1701        if (ret < 0) {
1702                mlog_errno(ret);
1703                goto out;
1704        }
1705
1706        ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1707                                      OCFS2_JOURNAL_ACCESS_WRITE);
1708        if (ret < 0) {
1709                mlog_errno(ret);
1710                goto out;
1711        }
1712
1713        ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
1714                                      OCFS2_JOURNAL_ACCESS_WRITE);
1715        if (ret < 0) {
1716                mlog_errno(ret);
1717                goto out;
1718        }
1719
1720        /* change the leaf extent block first. */
1721        el = path_leaf_el(path);
1722
1723        for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
1724                if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
1725                        break;
1726
1727        BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
1728
1729        el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1730
1731        /* change the r_cpos in the leaf block. */
1732        rb->rf_cpos = cpu_to_le32(new_cpos);
1733
1734        ocfs2_journal_dirty(handle, path_leaf_bh(path));
1735        ocfs2_journal_dirty(handle, ref_leaf_bh);
1736
1737out:
1738        ocfs2_free_path(path);
1739        return ret;
1740}
1741
1742static int ocfs2_insert_refcount_rec(handle_t *handle,
1743                                     struct ocfs2_caching_info *ci,
1744                                     struct buffer_head *ref_root_bh,
1745                                     struct buffer_head *ref_leaf_bh,
1746                                     struct ocfs2_refcount_rec *rec,
1747                                     int index, int merge,
1748                                     struct ocfs2_alloc_context *meta_ac)
1749{
1750        int ret;
1751        struct ocfs2_refcount_block *rb =
1752                        (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1753        struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1754        struct buffer_head *new_bh = NULL;
1755
1756        BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1757
1758        if (rf_list->rl_used == rf_list->rl_count) {
1759                u64 cpos = le64_to_cpu(rec->r_cpos);
1760                u32 len = le32_to_cpu(rec->r_clusters);
1761
1762                ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1763                                                 ref_leaf_bh, meta_ac);
1764                if (ret) {
1765                        mlog_errno(ret);
1766                        goto out;
1767                }
1768
1769                ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1770                                             cpos, len, NULL, &index,
1771                                             &new_bh);
1772                if (ret) {
1773                        mlog_errno(ret);
1774                        goto out;
1775                }
1776
1777                ref_leaf_bh = new_bh;
1778                rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1779                rf_list = &rb->rf_records;
1780        }
1781
1782        ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1783                                      OCFS2_JOURNAL_ACCESS_WRITE);
1784        if (ret) {
1785                mlog_errno(ret);
1786                goto out;
1787        }
1788
1789        if (index < le16_to_cpu(rf_list->rl_used))
1790                memmove(&rf_list->rl_recs[index + 1],
1791                        &rf_list->rl_recs[index],
1792                        (le16_to_cpu(rf_list->rl_used) - index) *
1793                         sizeof(struct ocfs2_refcount_rec));
1794
1795        trace_ocfs2_insert_refcount_rec(
1796                (unsigned long long)ref_leaf_bh->b_blocknr, index,
1797                (unsigned long long)le64_to_cpu(rec->r_cpos),
1798                le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount));
1799
1800        rf_list->rl_recs[index] = *rec;
1801
1802        le16_add_cpu(&rf_list->rl_used, 1);
1803
1804        if (merge)
1805                ocfs2_refcount_rec_merge(rb, index);
1806
1807        ocfs2_journal_dirty(handle, ref_leaf_bh);
1808
1809        if (index == 0) {
1810                ret = ocfs2_adjust_refcount_rec(handle, ci,
1811                                                ref_root_bh,
1812                                                ref_leaf_bh, rec);
1813                if (ret)
1814                        mlog_errno(ret);
1815        }
1816out:
1817        brelse(new_bh);
1818        return ret;
1819}
1820
1821/*
1822 * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1823 * This is much simple than our b-tree code.
1824 * split_rec is the new refcount rec we want to insert.
1825 * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1826 * increase refcount or decrease a refcount to non-zero).
1827 * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1828 * rec( in case we decrease a refcount to zero).
1829 */
1830static int ocfs2_split_refcount_rec(handle_t *handle,
1831                                    struct ocfs2_caching_info *ci,
1832                                    struct buffer_head *ref_root_bh,
1833                                    struct buffer_head *ref_leaf_bh,
1834                                    struct ocfs2_refcount_rec *split_rec,
1835                                    int index, int merge,
1836                                    struct ocfs2_alloc_context *meta_ac,
1837                                    struct ocfs2_cached_dealloc_ctxt *dealloc)
1838{
1839        int ret, recs_need;
1840        u32 len;
1841        struct ocfs2_refcount_block *rb =
1842                        (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1843        struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1844        struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
1845        struct ocfs2_refcount_rec *tail_rec = NULL;
1846        struct buffer_head *new_bh = NULL;
1847
1848        BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1849
1850        trace_ocfs2_split_refcount_rec(le64_to_cpu(orig_rec->r_cpos),
1851                le32_to_cpu(orig_rec->r_clusters),
1852                le32_to_cpu(orig_rec->r_refcount),
1853                le64_to_cpu(split_rec->r_cpos),
1854                le32_to_cpu(split_rec->r_clusters),
1855                le32_to_cpu(split_rec->r_refcount));
1856
1857        /*
1858         * If we just need to split the header or tail clusters,
1859         * no more recs are needed, just split is OK.
1860         * Otherwise we at least need one new recs.
1861         */
1862        if (!split_rec->r_refcount &&
1863            (split_rec->r_cpos == orig_rec->r_cpos ||
1864             le64_to_cpu(split_rec->r_cpos) +
1865             le32_to_cpu(split_rec->r_clusters) ==
1866             le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1867                recs_need = 0;
1868        else
1869                recs_need = 1;
1870
1871        /*
1872         * We need one more rec if we split in the middle and the new rec have
1873         * some refcount in it.
1874         */
1875        if (split_rec->r_refcount &&
1876            (split_rec->r_cpos != orig_rec->r_cpos &&
1877             le64_to_cpu(split_rec->r_cpos) +
1878             le32_to_cpu(split_rec->r_clusters) !=
1879             le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1880                recs_need++;
1881
1882        /* If the leaf block don't have enough record, expand it. */
1883        if (le16_to_cpu(rf_list->rl_used) + recs_need >
1884                                         le16_to_cpu(rf_list->rl_count)) {
1885                struct ocfs2_refcount_rec tmp_rec;
1886                u64 cpos = le64_to_cpu(orig_rec->r_cpos);
1887                len = le32_to_cpu(orig_rec->r_clusters);
1888                ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1889                                                 ref_leaf_bh, meta_ac);
1890                if (ret) {
1891                        mlog_errno(ret);
1892                        goto out;
1893                }
1894
1895                /*
1896                 * We have to re-get it since now cpos may be moved to
1897                 * another leaf block.
1898                 */
1899                ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1900                                             cpos, len, &tmp_rec, &index,
1901                                             &new_bh);
1902                if (ret) {
1903                        mlog_errno(ret);
1904                        goto out;
1905                }
1906
1907                ref_leaf_bh = new_bh;
1908                rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1909                rf_list = &rb->rf_records;
1910                orig_rec = &rf_list->rl_recs[index];
1911        }
1912
1913        ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1914                                      OCFS2_JOURNAL_ACCESS_WRITE);
1915        if (ret) {
1916                mlog_errno(ret);
1917                goto out;
1918        }
1919
1920        /*
1921         * We have calculated out how many new records we need and store
1922         * in recs_need, so spare enough space first by moving the records
1923         * after "index" to the end.
1924         */
1925        if (index != le16_to_cpu(rf_list->rl_used) - 1)
1926                memmove(&rf_list->rl_recs[index + 1 + recs_need],
1927                        &rf_list->rl_recs[index + 1],
1928                        (le16_to_cpu(rf_list->rl_used) - index - 1) *
1929                         sizeof(struct ocfs2_refcount_rec));
1930
1931        len = (le64_to_cpu(orig_rec->r_cpos) +
1932              le32_to_cpu(orig_rec->r_clusters)) -
1933              (le64_to_cpu(split_rec->r_cpos) +
1934              le32_to_cpu(split_rec->r_clusters));
1935
1936        /*
1937         * If we have "len", the we will split in the tail and move it
1938         * to the end of the space we have just spared.
1939         */
1940        if (len) {
1941                tail_rec = &rf_list->rl_recs[index + recs_need];
1942
1943                memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
1944                le64_add_cpu(&tail_rec->r_cpos,
1945                             le32_to_cpu(tail_rec->r_clusters) - len);
1946                tail_rec->r_clusters = cpu_to_le32(len);
1947        }
1948
1949        /*
1950         * If the split pos isn't the same as the original one, we need to
1951         * split in the head.
1952         *
1953         * Note: We have the chance that split_rec.r_refcount = 0,
1954         * recs_need = 0 and len > 0, which means we just cut the head from
1955         * the orig_rec and in that case we have done some modification in
1956         * orig_rec above, so the check for r_cpos is faked.
1957         */
1958        if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
1959                len = le64_to_cpu(split_rec->r_cpos) -
1960                      le64_to_cpu(orig_rec->r_cpos);
1961                orig_rec->r_clusters = cpu_to_le32(len);
1962                index++;
1963        }
1964
1965        le16_add_cpu(&rf_list->rl_used, recs_need);
1966
1967        if (split_rec->r_refcount) {
1968                rf_list->rl_recs[index] = *split_rec;
1969                trace_ocfs2_split_refcount_rec_insert(
1970                        (unsigned long long)ref_leaf_bh->b_blocknr, index,
1971                        (unsigned long long)le64_to_cpu(split_rec->r_cpos),
1972                        le32_to_cpu(split_rec->r_clusters),
1973                        le32_to_cpu(split_rec->r_refcount));
1974
1975                if (merge)
1976                        ocfs2_refcount_rec_merge(rb, index);
1977        }
1978
1979        ocfs2_journal_dirty(handle, ref_leaf_bh);
1980
1981out:
1982        brelse(new_bh);
1983        return ret;
1984}
1985
1986static int __ocfs2_increase_refcount(handle_t *handle,
1987                                     struct ocfs2_caching_info *ci,
1988                                     struct buffer_head *ref_root_bh,
1989                                     u64 cpos, u32 len, int merge,
1990                                     struct ocfs2_alloc_context *meta_ac,
1991                                     struct ocfs2_cached_dealloc_ctxt *dealloc)
1992{
1993        int ret = 0, index;
1994        struct buffer_head *ref_leaf_bh = NULL;
1995        struct ocfs2_refcount_rec rec;
1996        unsigned int set_len = 0;
1997
1998        trace_ocfs2_increase_refcount_begin(
1999             (unsigned long long)ocfs2_metadata_cache_owner(ci),
2000             (unsigned long long)cpos, len);
2001
2002        while (len) {
2003                ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2004                                             cpos, len, &rec, &index,
2005                                             &ref_leaf_bh);
2006                if (ret) {
2007                        mlog_errno(ret);
2008                        goto out;
2009                }
2010
2011                set_len = le32_to_cpu(rec.r_clusters);
2012
2013                /*
2014                 * Here we may meet with 3 situations:
2015                 *
2016                 * 1. If we find an already existing record, and the length
2017                 *    is the same, cool, we just need to increase the r_refcount
2018                 *    and it is OK.
2019                 * 2. If we find a hole, just insert it with r_refcount = 1.
2020                 * 3. If we are in the middle of one extent record, split
2021                 *    it.
2022                 */
2023                if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
2024                    set_len <= len) {
2025                        trace_ocfs2_increase_refcount_change(
2026                                (unsigned long long)cpos, set_len,
2027                                le32_to_cpu(rec.r_refcount));
2028                        ret = ocfs2_change_refcount_rec(handle, ci,
2029                                                        ref_leaf_bh, index,
2030                                                        merge, 1);
2031                        if (ret) {
2032                                mlog_errno(ret);
2033                                goto out;
2034                        }
2035                } else if (!rec.r_refcount) {
2036                        rec.r_refcount = cpu_to_le32(1);
2037
2038                        trace_ocfs2_increase_refcount_insert(
2039                             (unsigned long long)le64_to_cpu(rec.r_cpos),
2040                             set_len);
2041                        ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
2042                                                        ref_leaf_bh,
2043                                                        &rec, index,
2044                                                        merge, meta_ac);
2045                        if (ret) {
2046                                mlog_errno(ret);
2047                                goto out;
2048                        }
2049                } else  {
2050                        set_len = min((u64)(cpos + len),
2051                                      le64_to_cpu(rec.r_cpos) + set_len) - cpos;
2052                        rec.r_cpos = cpu_to_le64(cpos);
2053                        rec.r_clusters = cpu_to_le32(set_len);
2054                        le32_add_cpu(&rec.r_refcount, 1);
2055
2056                        trace_ocfs2_increase_refcount_split(
2057                             (unsigned long long)le64_to_cpu(rec.r_cpos),
2058                             set_len, le32_to_cpu(rec.r_refcount));
2059                        ret = ocfs2_split_refcount_rec(handle, ci,
2060                                                       ref_root_bh, ref_leaf_bh,
2061                                                       &rec, index, merge,
2062                                                       meta_ac, dealloc);
2063                        if (ret) {
2064                                mlog_errno(ret);
2065                                goto out;
2066                        }
2067                }
2068
2069                cpos += set_len;
2070                len -= set_len;
2071                brelse(ref_leaf_bh);
2072                ref_leaf_bh = NULL;
2073        }
2074
2075out:
2076        brelse(ref_leaf_bh);
2077        return ret;
2078}
2079
2080static int ocfs2_remove_refcount_extent(handle_t *handle,
2081                                struct ocfs2_caching_info *ci,
2082                                struct buffer_head *ref_root_bh,
2083                                struct buffer_head *ref_leaf_bh,
2084                                struct ocfs2_alloc_context *meta_ac,
2085                                struct ocfs2_cached_dealloc_ctxt *dealloc)
2086{
2087        int ret;
2088        struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2089        struct ocfs2_refcount_block *rb =
2090                        (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2091        struct ocfs2_extent_tree et;
2092
2093        BUG_ON(rb->rf_records.rl_used);
2094
2095        trace_ocfs2_remove_refcount_extent(
2096                (unsigned long long)ocfs2_metadata_cache_owner(ci),
2097                (unsigned long long)ref_leaf_bh->b_blocknr,
2098                le32_to_cpu(rb->rf_cpos));
2099
2100        ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2101        ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
2102                                  1, meta_ac, dealloc);
2103        if (ret) {
2104                mlog_errno(ret);
2105                goto out;
2106        }
2107
2108        ocfs2_remove_from_cache(ci, ref_leaf_bh);
2109
2110        /*
2111         * add the freed block to the dealloc so that it will be freed
2112         * when we run dealloc.
2113         */
2114        ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
2115                                        le16_to_cpu(rb->rf_suballoc_slot),
2116                                        le64_to_cpu(rb->rf_suballoc_loc),
2117                                        le64_to_cpu(rb->rf_blkno),
2118                                        le16_to_cpu(rb->rf_suballoc_bit));
2119        if (ret) {
2120                mlog_errno(ret);
2121                goto out;
2122        }
2123
2124        ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
2125                                      OCFS2_JOURNAL_ACCESS_WRITE);
2126        if (ret) {
2127                mlog_errno(ret);
2128                goto out;
2129        }
2130
2131        rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2132
2133        le32_add_cpu(&rb->rf_clusters, -1);
2134
2135        /*
2136         * check whether we need to restore the root refcount block if
2137         * there is no leaf extent block at atll.
2138         */
2139        if (!rb->rf_list.l_next_free_rec) {
2140                BUG_ON(rb->rf_clusters);
2141
2142                trace_ocfs2_restore_refcount_block(
2143                     (unsigned long long)ref_root_bh->b_blocknr);
2144
2145                rb->rf_flags = 0;
2146                rb->rf_parent = 0;
2147                rb->rf_cpos = 0;
2148                memset(&rb->rf_records, 0, sb->s_blocksize -
2149                       offsetof(struct ocfs2_refcount_block, rf_records));
2150                rb->rf_records.rl_count =
2151                                cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
2152        }
2153
2154        ocfs2_journal_dirty(handle, ref_root_bh);
2155
2156out:
2157        return ret;
2158}
2159
2160int ocfs2_increase_refcount(handle_t *handle,
2161                            struct ocfs2_caching_info *ci,
2162                            struct buffer_head *ref_root_bh,
2163                            u64 cpos, u32 len,
2164                            struct ocfs2_alloc_context *meta_ac,
2165                            struct ocfs2_cached_dealloc_ctxt *dealloc)
2166{
2167        return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
2168                                         cpos, len, 1,
2169                                         meta_ac, dealloc);
2170}
2171
2172static int ocfs2_decrease_refcount_rec(handle_t *handle,
2173                                struct ocfs2_caching_info *ci,
2174                                struct buffer_head *ref_root_bh,
2175                                struct buffer_head *ref_leaf_bh,
2176                                int index, u64 cpos, unsigned int len,
2177                                struct ocfs2_alloc_context *meta_ac,
2178                                struct ocfs2_cached_dealloc_ctxt *dealloc)
2179{
2180        int ret;
2181        struct ocfs2_refcount_block *rb =
2182                        (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2183        struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
2184
2185        BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
2186        BUG_ON(cpos + len >
2187               le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
2188
2189        trace_ocfs2_decrease_refcount_rec(
2190                (unsigned long long)ocfs2_metadata_cache_owner(ci),
2191                (unsigned long long)cpos, len);
2192
2193        if (cpos == le64_to_cpu(rec->r_cpos) &&
2194            len == le32_to_cpu(rec->r_clusters))
2195                ret = ocfs2_change_refcount_rec(handle, ci,
2196                                                ref_leaf_bh, index, 1, -1);
2197        else {
2198                struct ocfs2_refcount_rec split = *rec;
2199                split.r_cpos = cpu_to_le64(cpos);
2200                split.r_clusters = cpu_to_le32(len);
2201
2202                le32_add_cpu(&split.r_refcount, -1);
2203
2204                ret = ocfs2_split_refcount_rec(handle, ci,
2205                                               ref_root_bh, ref_leaf_bh,
2206                                               &split, index, 1,
2207                                               meta_ac, dealloc);
2208        }
2209
2210        if (ret) {
2211                mlog_errno(ret);
2212                goto out;
2213        }
2214
2215        /* Remove the leaf refcount block if it contains no refcount record. */
2216        if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
2217                ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
2218                                                   ref_leaf_bh, meta_ac,
2219                                                   dealloc);
2220                if (ret)
2221                        mlog_errno(ret);
2222        }
2223
2224out:
2225        return ret;
2226}
2227
2228static int __ocfs2_decrease_refcount(handle_t *handle,
2229                                     struct ocfs2_caching_info *ci,
2230                                     struct buffer_head *ref_root_bh,
2231                                     u64 cpos, u32 len,
2232                                     struct ocfs2_alloc_context *meta_ac,
2233                                     struct ocfs2_cached_dealloc_ctxt *dealloc,
2234                                     int delete)
2235{
2236        int ret = 0, index = 0;
2237        struct ocfs2_refcount_rec rec;
2238        unsigned int r_count = 0, r_len;
2239        struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2240        struct buffer_head *ref_leaf_bh = NULL;
2241
2242        trace_ocfs2_decrease_refcount(
2243                (unsigned long long)ocfs2_metadata_cache_owner(ci),
2244                (unsigned long long)cpos, len, delete);
2245
2246        while (len) {
2247                ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2248                                             cpos, len, &rec, &index,
2249                                             &ref_leaf_bh);
2250                if (ret) {
2251                        mlog_errno(ret);
2252                        goto out;
2253                }
2254
2255                r_count = le32_to_cpu(rec.r_refcount);
2256                BUG_ON(r_count == 0);
2257                if (!delete)
2258                        BUG_ON(r_count > 1);
2259
2260                r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
2261                              le32_to_cpu(rec.r_clusters)) - cpos;
2262
2263                ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
2264                                                  ref_leaf_bh, index,
2265                                                  cpos, r_len,
2266                                                  meta_ac, dealloc);
2267                if (ret) {
2268                        mlog_errno(ret);
2269                        goto out;
2270                }
2271
2272                if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
2273                        ret = ocfs2_cache_cluster_dealloc(dealloc,
2274                                          ocfs2_clusters_to_blocks(sb, cpos),
2275                                                          r_len);
2276                        if (ret) {
2277                                mlog_errno(ret);
2278                                goto out;
2279                        }
2280                }
2281
2282                cpos += r_len;
2283                len -= r_len;
2284                brelse(ref_leaf_bh);
2285                ref_leaf_bh = NULL;
2286        }
2287
2288out:
2289        brelse(ref_leaf_bh);
2290        return ret;
2291}
2292
2293/* Caller must hold refcount tree lock. */
2294int 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 buffer_head *ref_root_bh = NULL;
2303        struct ocfs2_refcount_tree *tree;
2304
2305        BUG_ON(!ocfs2_is_refcount_inode(inode));
2306
2307        ret = ocfs2_get_refcount_block(inode, &ref_blkno);
2308        if (ret) {
2309                mlog_errno(ret);
2310                goto out;
2311        }
2312
2313        ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
2314        if (ret) {
2315                mlog_errno(ret);
2316                goto out;
2317        }
2318
2319        ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
2320                                        &ref_root_bh);
2321        if (ret) {
2322                mlog_errno(ret);
2323                goto out;
2324        }
2325
2326        ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
2327                                        cpos, len, meta_ac, dealloc, delete);
2328        if (ret)
2329                mlog_errno(ret);
2330out:
2331        brelse(ref_root_bh);
2332        return ret;
2333}
2334
2335/*
2336 * Mark the already-existing extent at cpos as refcounted for len clusters.
2337 * This adds the refcount extent flag.
2338 *
2339 * If the existing extent is larger than the request, initiate a
2340 * split. An attempt will be made at merging with adjacent extents.
2341 *
2342 * The caller is responsible for passing down meta_ac if we'll need it.
2343 */
2344static int ocfs2_mark_extent_refcounted(struct inode *inode,
2345                                struct ocfs2_extent_tree *et,
2346                                handle_t *handle, u32 cpos,
2347                                u32 len, u32 phys,
2348                                struct ocfs2_alloc_context *meta_ac,
2349                                struct ocfs2_cached_dealloc_ctxt *dealloc)
2350{
2351        int ret;
2352
2353        trace_ocfs2_mark_extent_refcounted(OCFS2_I(inode)->ip_blkno,
2354                                           cpos, len, phys);
2355
2356        if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2357                ret = ocfs2_error(inode->i_sb, "Inode %lu want to use refcount tree, but the feature bit is not set in the super block\n",
2358                                  inode->i_ino);
2359                goto out;
2360        }
2361
2362        ret = ocfs2_change_extent_flag(handle, et, cpos,
2363                                       len, phys, meta_ac, dealloc,
2364                                       OCFS2_EXT_REFCOUNTED, 0);
2365        if (ret)
2366                mlog_errno(ret);
2367
2368out:
2369        return ret;
2370}
2371
2372/*
2373 * Given some contiguous physical clusters, calculate what we need
2374 * for modifying their refcount.
2375 */
2376static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
2377                                            struct ocfs2_caching_info *ci,
2378                                            struct buffer_head *ref_root_bh,
2379                                            u64 start_cpos,
2380                                            u32 clusters,
2381                                            int *meta_add,
2382                                            int *credits)
2383{
2384        int ret = 0, index, ref_blocks = 0, recs_add = 0;
2385        u64 cpos = start_cpos;
2386        struct ocfs2_refcount_block *rb;
2387        struct ocfs2_refcount_rec rec;
2388        struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
2389        u32 len;
2390
2391        while (clusters) {
2392                ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2393                                             cpos, clusters, &rec,
2394                                             &index, &ref_leaf_bh);
2395                if (ret) {
2396                        mlog_errno(ret);
2397                        goto out;
2398                }
2399
2400                if (ref_leaf_bh != prev_bh) {
2401                        /*
2402                         * Now we encounter a new leaf block, so calculate
2403                         * whether we need to extend the old leaf.
2404                         */
2405                        if (prev_bh) {
2406                                rb = (struct ocfs2_refcount_block *)
2407                                                        prev_bh->b_data;
2408
2409                                if (le16_to_cpu(rb->rf_records.rl_used) +
2410                                    recs_add >
2411                                    le16_to_cpu(rb->rf_records.rl_count))
2412                                        ref_blocks++;
2413                        }
2414
2415                        recs_add = 0;
2416                        *credits += 1;
2417                        brelse(prev_bh);
2418                        prev_bh = ref_leaf_bh;
2419                        get_bh(prev_bh);
2420                }
2421
2422                trace_ocfs2_calc_refcount_meta_credits_iterate(
2423                                recs_add, (unsigned long long)cpos, clusters,
2424                                (unsigned long long)le64_to_cpu(rec.r_cpos),
2425                                le32_to_cpu(rec.r_clusters),
2426                                le32_to_cpu(rec.r_refcount), index);
2427
2428                len = min((u64)cpos + clusters, le64_to_cpu(rec.r_cpos) +
2429                          le32_to_cpu(rec.r_clusters)) - cpos;
2430                /*
2431                 * We record all the records which will be inserted to the
2432                 * same refcount block, so that we can tell exactly whether
2433                 * we need a new refcount block or not.
2434                 *
2435                 * If we will insert a new one, this is easy and only happens
2436                 * during adding refcounted flag to the extent, so we don't
2437                 * have a chance of spliting. We just need one record.
2438                 *
2439                 * If the refcount rec already exists, that would be a little
2440                 * complicated. we may have to:
2441                 * 1) split at the beginning if the start pos isn't aligned.
2442                 *    we need 1 more record in this case.
2443                 * 2) split int the end if the end pos isn't aligned.
2444                 *    we need 1 more record in this case.
2445                 * 3) split in the middle because of file system fragmentation.
2446                 *    we need 2 more records in this case(we can't detect this
2447                 *    beforehand, so always think of the worst case).
2448                 */
2449                if (rec.r_refcount) {
2450                        recs_add += 2;
2451                        /* Check whether we need a split at the beginning. */
2452                        if (cpos == start_cpos &&
2453                            cpos != le64_to_cpu(rec.r_cpos))
2454                                recs_add++;
2455
2456                        /* Check whether we need a split in the end. */
2457                        if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
2458                            le32_to_cpu(rec.r_clusters))
2459                                recs_add++;
2460                } else
2461                        recs_add++;
2462
2463                brelse(ref_leaf_bh);
2464                ref_leaf_bh = NULL;
2465                clusters -= len;
2466                cpos += len;
2467        }
2468
2469        if (prev_bh) {
2470                rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
2471
2472                if (le16_to_cpu(rb->rf_records.rl_used) + recs_add >
2473                    le16_to_cpu(rb->rf_records.rl_count))
2474                        ref_blocks++;
2475
2476                *credits += 1;
2477        }
2478
2479        if (!ref_blocks)
2480                goto out;
2481
2482        *meta_add += ref_blocks;
2483        *credits += ref_blocks;
2484
2485        /*
2486         * So we may need ref_blocks to insert into the tree.
2487         * That also means we need to change the b-tree and add that number
2488         * of records since we never merge them.
2489         * We need one more block for expansion since the new created leaf
2490         * block is also full and needs split.
2491         */
2492        rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2493        if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
2494                struct ocfs2_extent_tree et;
2495
2496                ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2497                *meta_add += ocfs2_extend_meta_needed(et.et_root_el);
2498                *credits += ocfs2_calc_extend_credits(sb,
2499                                                      et.et_root_el);
2500        } else {
2501                *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
2502                *meta_add += 1;
2503        }
2504
2505out:
2506
2507        trace_ocfs2_calc_refcount_meta_credits(
2508                (unsigned long long)start_cpos, clusters,
2509                *meta_add, *credits);
2510        brelse(ref_leaf_bh);
2511        brelse(prev_bh);
2512        return ret;
2513}
2514
2515/*
2516 * For refcount tree, we will decrease some contiguous clusters
2517 * refcount count, so just go through it to see how many blocks
2518 * we gonna touch and whether we need to create new blocks.
2519 *
2520 * Normally the refcount blocks store these refcount should be
2521 * contiguous also, so that we can get the number easily.
2522 * We will at most add split 2 refcount records and 2 more
2523 * refcount blocks, so just check it in a rough way.
2524 *
2525 * Caller must hold refcount tree lock.
2526 */
2527int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
2528                                          u64 refcount_loc,
2529                                          u64 phys_blkno,
2530                                          u32 clusters,
2531                                          int *credits,
2532                                          int *ref_blocks)
2533{
2534        int ret;
2535        struct buffer_head *ref_root_bh = NULL;
2536        struct ocfs2_refcount_tree *tree;
2537        u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
2538
2539        if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2540                ret = ocfs2_error(inode->i_sb, "Inode %lu want to use refcount tree, but the feature bit is not set in the super block\n",
2541                                  inode->i_ino);
2542                goto out;
2543        }
2544
2545        BUG_ON(!ocfs2_is_refcount_inode(inode));
2546
2547        ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
2548                                      refcount_loc, &tree);
2549        if (ret) {
2550                mlog_errno(ret);
2551                goto out;
2552        }
2553
2554        ret = ocfs2_read_refcount_block(&tree->rf_ci, refcount_loc,
2555                                        &ref_root_bh);
2556        if (ret) {
2557                mlog_errno(ret);
2558                goto out;
2559        }
2560
2561        ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
2562                                               &tree->rf_ci,
2563                                               ref_root_bh,
2564                                               start_cpos, clusters,
2565                                               ref_blocks, credits);
2566        if (ret) {
2567                mlog_errno(ret);
2568                goto out;
2569        }
2570
2571        trace_ocfs2_prepare_refcount_change_for_del(*ref_blocks, *credits);
2572
2573out:
2574        brelse(ref_root_bh);
2575        return ret;
2576}
2577
2578#define MAX_CONTIG_BYTES        1048576
2579
2580static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
2581{
2582        return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
2583}
2584
2585static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
2586{
2587        return ~(ocfs2_cow_contig_clusters(sb) - 1);
2588}
2589
2590/*
2591 * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2592 * find an offset (start + (n * contig_clusters)) that is closest to cpos
2593 * while still being less than or equal to it.
2594 *
2595 * The goal is to break the extent at a multiple of contig_clusters.
2596 */
2597static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
2598                                                 unsigned int start,
2599                                                 unsigned int cpos)
2600{
2601        BUG_ON(start > cpos);
2602
2603        return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
2604}
2605
2606/*
2607 * Given a cluster count of len, pad it out so that it is a multiple
2608 * of contig_clusters.
2609 */
2610static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
2611                                                  unsigned int len)
2612{
2613        unsigned int padded =
2614                (len + (ocfs2_cow_contig_clusters(sb) - 1)) &
2615                ocfs2_cow_contig_mask(sb);
2616
2617        /* Did we wrap? */
2618        if (padded < len)
2619                padded = UINT_MAX;
2620
2621        return padded;
2622}
2623
2624/*
2625 * Calculate out the start and number of virtual clusters we need to to CoW.
2626 *
2627 * cpos is vitual start cluster position we want to do CoW in a
2628 * file and write_len is the cluster length.
2629 * max_cpos is the place where we want to stop CoW intentionally.
2630 *
2631 * Normal we will start CoW from the beginning of extent record cotaining cpos.
2632 * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2633 * get good I/O from the resulting extent tree.
2634 */
2635static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
2636                                           struct ocfs2_extent_list *el,
2637                                           u32 cpos,
2638                                           u32 write_len,
2639                                           u32 max_cpos,
2640                                           u32 *cow_start,
2641                                           u32 *cow_len)
2642{
2643        int ret = 0;
2644        int tree_height = le16_to_cpu(el->l_tree_depth), i;
2645        struct buffer_head *eb_bh = NULL;
2646        struct ocfs2_extent_block *eb = NULL;
2647        struct ocfs2_extent_rec *rec;
2648        unsigned int want_clusters, rec_end = 0;
2649        int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
2650        int leaf_clusters;
2651
2652        BUG_ON(cpos + write_len > max_cpos);
2653
2654        if (tree_height > 0) {
2655                ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
2656                if (ret) {
2657                        mlog_errno(ret);
2658                        goto out;
2659                }
2660
2661                eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2662                el = &eb->h_list;
2663
2664                if (el->l_tree_depth) {
2665                        ret = ocfs2_error(inode->i_sb,
2666                                          "Inode %lu has non zero tree depth in leaf block %llu\n",
2667                                          inode->i_ino,
2668                                          (unsigned long long)eb_bh->b_blocknr);
2669                        goto out;
2670                }
2671        }
2672
2673        *cow_len = 0;
2674        for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
2675                rec = &el->l_recs[i];
2676
2677                if (ocfs2_is_empty_extent(rec)) {
2678                        mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
2679                                        "index %d\n", inode->i_ino, i);
2680                        continue;
2681                }
2682
2683                if (le32_to_cpu(rec->e_cpos) +
2684                    le16_to_cpu(rec->e_leaf_clusters) <= cpos)
2685                        continue;
2686
2687                if (*cow_len == 0) {
2688                        /*
2689                         * We should find a refcounted record in the
2690                         * first pass.
2691                         */
2692                        BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
2693                        *cow_start = le32_to_cpu(rec->e_cpos);
2694                }
2695
2696                /*
2697                 * If we encounter a hole, a non-refcounted record or
2698                 * pass the max_cpos, stop the search.
2699                 */
2700                if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
2701                    (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
2702                    (max_cpos <= le32_to_cpu(rec->e_cpos)))
2703                        break;
2704
2705                leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
2706                rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
2707                if (rec_end > max_cpos) {
2708                        rec_end = max_cpos;
2709                        leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
2710                }
2711
2712                /*
2713                 * How many clusters do we actually need from
2714                 * this extent?  First we see how many we actually
2715                 * need to complete the write.  If that's smaller
2716                 * than contig_clusters, we try for contig_clusters.
2717                 */
2718                if (!*cow_len)
2719                        want_clusters = write_len;
2720                else
2721                        want_clusters = (cpos + write_len) -
2722                                (*cow_start + *cow_len);
2723                if (want_clusters < contig_clusters)
2724                        want_clusters = contig_clusters;
2725
2726                /*
2727                 * If the write does not cover the whole extent, we
2728                 * need to calculate how we're going to split the extent.
2729                 * We try to do it on contig_clusters boundaries.
2730                 *
2731                 * Any extent smaller than contig_clusters will be
2732                 * CoWed in its entirety.
2733                 */
2734                if (leaf_clusters <= contig_clusters)
2735                        *cow_len += leaf_clusters;
2736                else if (*cow_len || (*cow_start == cpos)) {
2737                        /*
2738                         * This extent needs to be CoW'd from its
2739                         * beginning, so all we have to do is compute
2740                         * how many clusters to grab.  We align
2741                         * want_clusters to the edge of contig_clusters
2742                         * to get better I/O.
2743                         */
2744                        want_clusters = ocfs2_cow_align_length(inode->i_sb,
2745                                                               want_clusters);
2746
2747                        if (leaf_clusters < want_clusters)
2748                                *cow_len += leaf_clusters;
2749                        else
2750                                *cow_len += want_clusters;
2751                } else if ((*cow_start + contig_clusters) >=
2752                           (cpos + write_len)) {
2753                        /*
2754                         * Breaking off contig_clusters at the front
2755                         * of the extent will cover our write.  That's
2756                         * easy.
2757                         */
2758                        *cow_len = contig_clusters;
2759                } else if ((rec_end - cpos) <= contig_clusters) {
2760                        /*
2761                         * Breaking off contig_clusters at the tail of
2762                         * this extent will cover cpos.
2763                         */
2764                        *cow_start = rec_end - contig_clusters;
2765                        *cow_len = contig_clusters;
2766                } else if ((rec_end - cpos) <= want_clusters) {
2767                        /*
2768                         * While we can't fit the entire write in this
2769                         * extent, we know that the write goes from cpos
2770                         * to the end of the extent.  Break that off.
2771                         * We try to break it at some multiple of
2772                         * contig_clusters from the front of the extent.
2773                         * Failing that (ie, cpos is within
2774                         * contig_clusters of the front), we'll CoW the
2775                         * entire extent.
2776                         */
2777                        *cow_start = ocfs2_cow_align_start(inode->i_sb,
2778                                                           *cow_start, cpos);
2779                        *cow_len = rec_end - *cow_start;
2780                } else {
2781                        /*
2782                         * Ok, the entire write lives in the middle of
2783                         * this extent.  Let's try to slice the extent up
2784                         * nicely.  Optimally, our CoW region starts at
2785                         * m*contig_clusters from the beginning of the
2786                         * extent and goes for n*contig_clusters,
2787                         * covering the entire write.
2788                         */
2789                        *cow_start = ocfs2_cow_align_start(inode->i_sb,
2790                                                           *cow_start, cpos);
2791
2792                        want_clusters = (cpos + write_len) - *cow_start;
2793                        want_clusters = ocfs2_cow_align_length(inode->i_sb,
2794                                                               want_clusters);
2795                        if (*cow_start + want_clusters <= rec_end)
2796                                *cow_len = want_clusters;
2797                        else
2798                                *cow_len = rec_end - *cow_start;
2799                }
2800
2801                /* Have we covered our entire write yet? */
2802                if ((*cow_start + *cow_len) >= (cpos + write_len))
2803                        break;
2804
2805                /*
2806                 * If we reach the end of the extent block and don't get enough
2807                 * clusters, continue with the next extent block if possible.
2808                 */
2809                if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
2810                    eb && eb->h_next_leaf_blk) {
2811                        brelse(eb_bh);
2812                        eb_bh = NULL;
2813
2814                        ret = ocfs2_read_extent_block(INODE_CACHE(inode),
2815                                               le64_to_cpu(eb->h_next_leaf_blk),
2816                                               &eb_bh);
2817                        if (ret) {
2818                                mlog_errno(ret);
2819                                goto out;
2820                        }
2821
2822                        eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2823                        el = &eb->h_list;
2824                        i = -1;
2825                }
2826        }
2827
2828out:
2829        brelse(eb_bh);
2830        return ret;
2831}
2832
2833/*
2834 * Prepare meta_ac, data_ac and calculate credits when we want to add some
2835 * num_clusters in data_tree "et" and change the refcount for the old
2836 * clusters(starting form p_cluster) in the refcount tree.
2837 *
2838 * Note:
2839 * 1. since we may split the old tree, so we at most will need num_clusters + 2
2840 *    more new leaf records.
2841 * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2842 *    just give data_ac = NULL.
2843 */
2844static int ocfs2_lock_refcount_allocators(struct super_block *sb,
2845                                        u32 p_cluster, u32 num_clusters,
2846                                        struct ocfs2_extent_tree *et,
2847                                        struct ocfs2_caching_info *ref_ci,
2848                                        struct buffer_head *ref_root_bh,
2849                                        struct ocfs2_alloc_context **meta_ac,
2850                                        struct ocfs2_alloc_context **data_ac,
2851                                        int *credits)
2852{
2853        int ret = 0, meta_add = 0;
2854        int num_free_extents = ocfs2_num_free_extents(OCFS2_SB(sb), et);
2855
2856        if (num_free_extents < 0) {
2857                ret = num_free_extents;
2858                mlog_errno(ret);
2859                goto out;
2860        }
2861
2862        if (num_free_extents < num_clusters + 2)
2863                meta_add =
2864                        ocfs2_extend_meta_needed(et->et_root_el);
2865
2866        *credits += ocfs2_calc_extend_credits(sb, et->et_root_el);
2867
2868        ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
2869                                               p_cluster, num_clusters,
2870                                               &meta_add, credits);
2871        if (ret) {
2872                mlog_errno(ret);
2873                goto out;
2874        }
2875
2876        trace_ocfs2_lock_refcount_allocators(meta_add, *credits);
2877        ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
2878                                                meta_ac);
2879        if (ret) {
2880                mlog_errno(ret);
2881                goto out;
2882        }
2883
2884        if (data_ac) {
2885                ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
2886                                             data_ac);
2887                if (ret)
2888                        mlog_errno(ret);
2889        }
2890
2891out:
2892        if (ret) {
2893                if (*meta_ac) {
2894                        ocfs2_free_alloc_context(*meta_ac);
2895                        *meta_ac = NULL;
2896                }
2897        }
2898
2899        return ret;
2900}
2901
2902static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
2903{
2904        BUG_ON(buffer_dirty(bh));
2905
2906        clear_buffer_mapped(bh);
2907
2908        return 0;
2909}
2910
2911int ocfs2_duplicate_clusters_by_page(handle_t *handle,
2912                                     struct inode *inode,
2913                                     u32 cpos, u32 old_cluster,
2914                                     u32 new_cluster, u32 new_len)
2915{
2916        int ret = 0, partial;
2917        struct super_block *sb = inode->i_sb;
2918        u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
2919        struct page *page;
2920        pgoff_t page_index;
2921        unsigned int from, to;
2922        loff_t offset, end, map_end;
2923        struct address_space *mapping = inode->i_mapping;
2924
2925        trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
2926                                               new_cluster, new_len);
2927
2928        offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
2929        end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
2930        /*
2931         * We only duplicate pages until we reach the page contains i_size - 1.
2932         * So trim 'end' to i_size.
2933         */
2934        if (end > i_size_read(inode))
2935                end = i_size_read(inode);
2936
2937        while (offset < end) {
2938                page_index = offset >> PAGE_SHIFT;
2939                map_end = ((loff_t)page_index + 1) << PAGE_SHIFT;
2940                if (map_end > end)
2941                        map_end = end;
2942
2943                /* from, to is the offset within the page. */
2944                from = offset & (PAGE_SIZE - 1);
2945                to = PAGE_SIZE;
2946                if (map_end & (PAGE_SIZE - 1))
2947                        to = map_end & (PAGE_SIZE - 1);
2948
2949                page = find_or_create_page(mapping, page_index, GFP_NOFS);
2950                if (!page) {
2951                        ret = -ENOMEM;
2952                        mlog_errno(ret);
2953                        break;
2954                }
2955
2956                /*
2957                 * In case PAGE_SIZE <= CLUSTER_SIZE, This page
2958                 * can't be dirtied before we CoW it out.
2959                 */
2960                if (PAGE_SIZE <= OCFS2_SB(sb)->s_clustersize)
2961                        BUG_ON(PageDirty(page));
2962
2963                if (!PageUptodate(page)) {
2964                        ret = block_read_full_page(page, ocfs2_get_block);
2965                        if (ret) {
2966                                mlog_errno(ret);
2967                                goto unlock;
2968                        }
2969                        lock_page(page);
2970                }
2971
2972                if (page_has_buffers(page)) {
2973                        ret = walk_page_buffers(handle, page_buffers(page),
2974                                                from, to, &partial,
2975                                                ocfs2_clear_cow_buffer);
2976                        if (ret) {
2977                                mlog_errno(ret);
2978                                goto unlock;
2979                        }
2980                }
2981
2982                ocfs2_map_and_dirty_page(inode,
2983                                         handle, from, to,
2984                                         page, 0, &new_block);
2985                mark_page_accessed(page);
2986unlock:
2987                unlock_page(page);
2988                put_page(page);
2989                page = NULL;
2990                offset = map_end;
2991                if (ret)
2992                        break;
2993        }
2994
2995        return ret;
2996}
2997
2998int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
2999                                    struct inode *inode,
3000                                    u32 cpos, u32 old_cluster,
3001                                    u32 new_cluster, u32 new_len)
3002{
3003        int ret = 0;
3004        struct super_block *sb = inode->i_sb;
3005        struct ocfs2_caching_info *ci = INODE_CACHE(inode);
3006        int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
3007        u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
3008        u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
3009        struct ocfs2_super *osb = OCFS2_SB(sb);
3010        struct buffer_head *old_bh = NULL;
3011        struct buffer_head *new_bh = NULL;
3012
3013        trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
3014                                               new_cluster, new_len);
3015
3016        for (i = 0; i < blocks; i++, old_block++, new_block++) {
3017                new_bh = sb_getblk(osb->sb, new_block);
3018                if (new_bh == NULL) {
3019                        ret = -ENOMEM;
3020                        mlog_errno(ret);
3021                        break;
3022                }
3023
3024                ocfs2_set_new_buffer_uptodate(ci, new_bh);
3025
3026                ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
3027                if (ret) {
3028                        mlog_errno(ret);
3029                        break;
3030                }
3031
3032                ret = ocfs2_journal_access(handle, ci, new_bh,
3033                                           OCFS2_JOURNAL_ACCESS_CREATE);
3034                if (ret) {
3035                        mlog_errno(ret);
3036                        break;
3037                }
3038
3039                memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
3040                ocfs2_journal_dirty(handle, new_bh);
3041
3042                brelse(new_bh);
3043                brelse(old_bh);
3044                new_bh = NULL;
3045                old_bh = NULL;
3046        }
3047
3048        brelse(new_bh);
3049        brelse(old_bh);
3050        return ret;
3051}
3052
3053static int ocfs2_clear_ext_refcount(handle_t *handle,
3054                                    struct ocfs2_extent_tree *et,
3055                                    u32 cpos, u32 p_cluster, u32 len,
3056                                    unsigned int ext_flags,
3057                                    struct ocfs2_alloc_context *meta_ac,
3058                                    struct ocfs2_cached_dealloc_ctxt *dealloc)
3059{
3060        int ret, index;
3061        struct ocfs2_extent_rec replace_rec;
3062        struct ocfs2_path *path = NULL;
3063        struct ocfs2_extent_list *el;
3064        struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
3065        u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
3066
3067        trace_ocfs2_clear_ext_refcount((unsigned long long)ino,
3068                                       cpos, len, p_cluster, ext_flags);
3069
3070        memset(&replace_rec, 0, sizeof(replace_rec));
3071        replace_rec.e_cpos = cpu_to_le32(cpos);
3072        replace_rec.e_leaf_clusters = cpu_to_le16(len);
3073        replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
3074                                                                   p_cluster));
3075        replace_rec.e_flags = ext_flags;
3076        replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
3077
3078        path = ocfs2_new_path_from_et(et);
3079        if (!path) {
3080                ret = -ENOMEM;
3081                mlog_errno(ret);
3082                goto out;
3083        }
3084
3085        ret = ocfs2_find_path(et->et_ci, path, cpos);
3086        if (ret) {
3087                mlog_errno(ret);
3088                goto out;
3089        }
3090
3091        el = path_leaf_el(path);
3092
3093        index = ocfs2_search_extent_list(el, cpos);
3094        if (index == -1) {
3095                ret = ocfs2_error(sb,
3096                                  "Inode %llu has an extent at cpos %u which can no longer be found\n",
3097                                  (unsigned long long)ino, cpos);
3098                goto out;
3099        }
3100
3101        ret = ocfs2_split_extent(handle, et, path, index,
3102                                 &replace_rec, meta_ac, dealloc);
3103        if (ret)
3104                mlog_errno(ret);
3105
3106out:
3107        ocfs2_free_path(path);
3108        return ret;
3109}
3110
3111static int ocfs2_replace_clusters(handle_t *handle,
3112                                  struct ocfs2_cow_context *context,
3113                                  u32 cpos, u32 old,
3114                                  u32 new, u32 len,
3115                                  unsigned int ext_flags)
3116{
3117        int ret;
3118        struct ocfs2_caching_info *ci = context->data_et.et_ci;
3119        u64 ino = ocfs2_metadata_cache_owner(ci);
3120
3121        trace_ocfs2_replace_clusters((unsigned long long)ino,
3122                                     cpos, old, new, len, ext_flags);
3123
3124        /*If the old clusters is unwritten, no need to duplicate. */
3125        if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
3126                ret = context->cow_duplicate_clusters(handle, context->inode,
3127                                                      cpos, old, new, len);
3128                if (ret) {
3129                        mlog_errno(ret);
3130                        goto out;
3131                }
3132        }
3133
3134        ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
3135                                       cpos, new, len, ext_flags,
3136                                       context->meta_ac, &context->dealloc);
3137        if (ret)
3138                mlog_errno(ret);
3139out:
3140        return ret;
3141}
3142
3143int ocfs2_cow_sync_writeback(struct super_block *sb,
3144                             struct inode *inode,
3145                             u32 cpos, u32 num_clusters)
3146{
3147        int ret = 0;
3148        loff_t offset, end, map_end;
3149        pgoff_t page_index;
3150        struct page *page;
3151
3152        if (ocfs2_should_order_data(inode))
3153                return 0;
3154
3155        offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
3156        end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
3157
3158        ret = filemap_fdatawrite_range(inode->i_mapping,
3159                                       offset, end - 1);
3160        if (ret < 0) {
3161                mlog_errno(ret);
3162                return ret;
3163        }
3164
3165        while (offset < end) {
3166                page_index = offset >> PAGE_SHIFT;
3167                map_end = ((loff_t)page_index + 1) << PAGE_SHIFT;
3168                if (map_end > end)
3169                        map_end = end;
3170
3171                page = find_or_create_page(inode->i_mapping,
3172                                           page_index, GFP_NOFS);
3173                BUG_ON(!page);
3174
3175                wait_on_page_writeback(page);
3176                if (PageError(page)) {
3177                        ret = -EIO;
3178                        mlog_errno(ret);
3179                } else
3180                        mark_page_accessed(page);
3181
3182                unlock_page(page);
3183                put_page(page);
3184                page = NULL;
3185                offset = map_end;
3186                if (ret)
3187                        break;
3188        }
3189
3190        return ret;
3191}
3192
3193static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
3194                                 u32 v_cluster, u32 *p_cluster,
3195                                 u32 *num_clusters,
3196                                 unsigned int *extent_flags)
3197{
3198        return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
3199                                  num_clusters, extent_flags);
3200}
3201
3202static int ocfs2_make_clusters_writable(struct super_block *sb,
3203                                        struct ocfs2_cow_context *context,
3204                                        u32 cpos, u32 p_cluster,
3205                                        u32 num_clusters, unsigned int e_flags)
3206{
3207        int ret, delete, index, credits =  0;
3208        u32 new_bit, new_len, orig_num_clusters;
3209        unsigned int set_len;
3210        struct ocfs2_super *osb = OCFS2_SB(sb);
3211        handle_t *handle;
3212        struct buffer_head *ref_leaf_bh = NULL;
3213        struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
3214        struct ocfs2_refcount_rec rec;
3215
3216        trace_ocfs2_make_clusters_writable(cpos, p_cluster,
3217                                           num_clusters, e_flags);
3218
3219        ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
3220                                             &context->data_et,
3221                                             ref_ci,
3222                                             context->ref_root_bh,
3223                                             &context->meta_ac,
3224                                             &context->data_ac, &credits);
3225        if (ret) {
3226                mlog_errno(ret);
3227                return ret;
3228        }
3229
3230        if (context->post_refcount)
3231                credits += context->post_refcount->credits;
3232
3233        credits += context->extra_credits;
3234        handle = ocfs2_start_trans(osb, credits);
3235        if (IS_ERR(handle)) {
3236                ret = PTR_ERR(handle);
3237                mlog_errno(ret);
3238                goto out;
3239        }
3240
3241        orig_num_clusters = num_clusters;
3242
3243        while (num_clusters) {
3244                ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
3245                                             p_cluster, num_clusters,
3246                                             &rec, &index, &ref_leaf_bh);
3247                if (ret) {
3248                        mlog_errno(ret);
3249                        goto out_commit;
3250                }
3251
3252                BUG_ON(!rec.r_refcount);
3253                set_len = min((u64)p_cluster + num_clusters,
3254                              le64_to_cpu(rec.r_cpos) +
3255                              le32_to_cpu(rec.r_clusters)) - p_cluster;
3256
3257                /*
3258                 * There are many different situation here.
3259                 * 1. If refcount == 1, remove the flag and don't COW.
3260                 * 2. If refcount > 1, allocate clusters.
3261                 *    Here we may not allocate r_len once at a time, so continue
3262                 *    until we reach num_clusters.
3263                 */
3264                if (le32_to_cpu(rec.r_refcount) == 1) {
3265                        delete = 0;
3266                        ret = ocfs2_clear_ext_refcount(handle,
3267                                                       &context->data_et,
3268                                                       cpos, p_cluster,
3269                                                       set_len, e_flags,
3270                                                       context->meta_ac,
3271                                                       &context->dealloc);
3272                        if (ret) {
3273                                mlog_errno(ret);
3274                                goto out_commit;
3275                        }
3276                } else {
3277                        delete = 1;
3278
3279                        ret = __ocfs2_claim_clusters(handle,
3280                                                     context->data_ac,
3281                                                     1, set_len,
3282                                                     &new_bit, &new_len);
3283                        if (ret) {
3284                                mlog_errno(ret);
3285                                goto out_commit;
3286                        }
3287
3288                        ret = ocfs2_replace_clusters(handle, context,
3289                                                     cpos, p_cluster, new_bit,
3290                                                     new_len, e_flags);
3291                        if (ret) {
3292                                mlog_errno(ret);
3293                                goto out_commit;
3294                        }
3295                        set_len = new_len;
3296                }
3297
3298                ret = __ocfs2_decrease_refcount(handle, ref_ci,
3299                                                context->ref_root_bh,
3300                                                p_cluster, set_len,
3301                                                context->meta_ac,
3302                                                &context->dealloc, delete);
3303                if (ret) {
3304                        mlog_errno(ret);
3305                        goto out_commit;
3306                }
3307
3308                cpos += set_len;
3309                p_cluster += set_len;
3310                num_clusters -= set_len;
3311                brelse(ref_leaf_bh);
3312                ref_leaf_bh = NULL;
3313        }
3314
3315        /* handle any post_cow action. */
3316        if (context->post_refcount && context->post_refcount->func) {
3317                ret = context->post_refcount->func(context->inode, handle,
3318                                                context->post_refcount->para);
3319                if (ret) {
3320                        mlog_errno(ret);
3321                        goto out_commit;
3322                }
3323        }
3324
3325        /*
3326         * Here we should write the new page out first if we are
3327         * in write-back mode.
3328         */
3329        if (context->get_clusters == ocfs2_di_get_clusters) {
3330                ret = ocfs2_cow_sync_writeback(sb, context->inode, cpos,
3331                                               orig_num_clusters);
3332                if (ret)
3333                        mlog_errno(ret);
3334        }
3335
3336out_commit:
3337        ocfs2_commit_trans(osb, handle);
3338
3339out:
3340        if (context->data_ac) {
3341                ocfs2_free_alloc_context(context->data_ac);
3342                context->data_ac = NULL;
3343        }
3344        if (context->meta_ac) {
3345                ocfs2_free_alloc_context(context->meta_ac);
3346                context->meta_ac = NULL;
3347        }
3348        brelse(ref_leaf_bh);
3349
3350        return ret;
3351}
3352
3353static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
3354{
3355        int ret = 0;
3356        struct inode *inode = context->inode;
3357        u32 cow_start = context->cow_start, cow_len = context->cow_len;
3358        u32 p_cluster, num_clusters;
3359        unsigned int ext_flags;
3360        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3361
3362        if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
3363                return ocfs2_error(inode->i_sb, "Inode %lu want to use refcount tree, but the feature bit is not set in the super block\n",
3364                                   inode->i_ino);
3365        }
3366
3367        ocfs2_init_dealloc_ctxt(&context->dealloc);
3368
3369        while (cow_len) {
3370                ret = context->get_clusters(context, cow_start, &p_cluster,
3371                                            &num_clusters, &ext_flags);
3372                if (ret) {
3373                        mlog_errno(ret);
3374                        break;
3375                }
3376
3377                BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
3378
3379                if (cow_len < num_clusters)
3380                        num_clusters = cow_len;
3381
3382                ret = ocfs2_make_clusters_writable(inode->i_sb, context,
3383                                                   cow_start, p_cluster,
3384                                                   num_clusters, ext_flags);
3385                if (ret) {
3386                        mlog_errno(ret);
3387                        break;
3388                }
3389
3390                cow_len -= num_clusters;
3391                cow_start += num_clusters;
3392        }
3393
3394        if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
3395                ocfs2_schedule_truncate_log_flush(osb, 1);
3396                ocfs2_run_deallocs(osb, &context->dealloc);
3397        }
3398
3399        return ret;
3400}
3401
3402/*
3403 * Starting at cpos, try to CoW write_len clusters.  Don't CoW
3404 * past max_cpos.  This will stop when it runs into a hole or an
3405 * unrefcounted extent.
3406 */
3407static int ocfs2_refcount_cow_hunk(struct inode *inode,
3408                                   struct buffer_head *di_bh,
3409                                   u32 cpos, u32 write_len, u32 max_cpos)
3410{
3411        int ret;
3412        u32 cow_start = 0, cow_len = 0;
3413        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3414        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3415        struct buffer_head *ref_root_bh = NULL;
3416        struct ocfs2_refcount_tree *ref_tree;
3417        struct ocfs2_cow_context *context = NULL;
3418
3419        BUG_ON(!ocfs2_is_refcount_inode(inode));
3420
3421        ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
3422                                              cpos, write_len, max_cpos,
3423                                              &cow_start, &cow_len);
3424        if (ret) {
3425                mlog_errno(ret);
3426                goto out;
3427        }
3428
3429        trace_ocfs2_refcount_cow_hunk(OCFS2_I(inode)->ip_blkno,
3430                                      cpos, write_len, max_cpos,
3431                                      cow_start, cow_len);
3432
3433        BUG_ON(cow_len == 0);
3434
3435        context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3436        if (!context) {
3437                ret = -ENOMEM;
3438                mlog_errno(ret);
3439                goto out;
3440        }
3441
3442        ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
3443                                       1, &ref_tree, &ref_root_bh);
3444        if (ret) {
3445                mlog_errno(ret);
3446                goto out;
3447        }
3448
3449        context->inode = inode;
3450        context->cow_start = cow_start;
3451        context->cow_len = cow_len;
3452        context->ref_tree = ref_tree;
3453        context->ref_root_bh = ref_root_bh;
3454        context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
3455        context->get_clusters = ocfs2_di_get_clusters;
3456
3457        ocfs2_init_dinode_extent_tree(&context->data_et,
3458                                      INODE_CACHE(inode), di_bh);
3459
3460        ret = ocfs2_replace_cow(context);
3461        if (ret)
3462                mlog_errno(ret);
3463
3464        /*
3465         * truncate the extent map here since no matter whether we meet with
3466         * any error during the action, we shouldn't trust cached extent map
3467         * any more.
3468         */
3469        ocfs2_extent_map_trunc(inode, cow_start);
3470
3471        ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3472        brelse(ref_root_bh);
3473out:
3474        kfree(context);
3475        return ret;
3476}
3477
3478/*
3479 * CoW any and all clusters between cpos and cpos+write_len.
3480 * Don't CoW past max_cpos.  If this returns successfully, all
3481 * clusters between cpos and cpos+write_len are safe to modify.
3482 */
3483int ocfs2_refcount_cow(struct inode *inode,
3484                       struct buffer_head *di_bh,
3485                       u32 cpos, u32 write_len, u32 max_cpos)
3486{
3487        int ret = 0;
3488        u32 p_cluster, num_clusters;
3489        unsigned int ext_flags;
3490
3491        while (write_len) {
3492                ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3493                                         &num_clusters, &ext_flags);
3494                if (ret) {
3495                        mlog_errno(ret);
3496                        break;
3497                }
3498
3499                if (write_len < num_clusters)
3500                        num_clusters = write_len;
3501
3502                if (ext_flags & OCFS2_EXT_REFCOUNTED) {
3503                        ret = ocfs2_refcount_cow_hunk(inode, di_bh, cpos,
3504                                                      num_clusters, max_cpos);
3505                        if (ret) {
3506                                mlog_errno(ret);
3507                                break;
3508                        }
3509                }
3510
3511                write_len -= num_clusters;
3512                cpos += num_clusters;
3513        }
3514
3515        return ret;
3516}
3517
3518static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
3519                                          u32 v_cluster, u32 *p_cluster,
3520                                          u32 *num_clusters,
3521                                          unsigned int *extent_flags)
3522{
3523        struct inode *inode = context->inode;
3524        struct ocfs2_xattr_value_root *xv = context->cow_object;
3525
3526        return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
3527                                        num_clusters, &xv->xr_list,
3528                                        extent_flags);
3529}
3530
3531/*
3532 * Given a xattr value root, calculate the most meta/credits we need for
3533 * refcount tree change if we truncate it to 0.
3534 */
3535int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
3536                                       struct ocfs2_caching_info *ref_ci,
3537                                       struct buffer_head *ref_root_bh,
3538                                       struct ocfs2_xattr_value_root *xv,
3539                                       int *meta_add, int *credits)
3540{
3541        int ret = 0, index, ref_blocks = 0;
3542        u32 p_cluster, num_clusters;
3543        u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
3544        struct ocfs2_refcount_block *rb;
3545        struct ocfs2_refcount_rec rec;
3546        struct buffer_head *ref_leaf_bh = NULL;
3547
3548        while (cpos < clusters) {
3549                ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
3550                                               &num_clusters, &xv->xr_list,
3551                                               NULL);
3552                if (ret) {
3553                        mlog_errno(ret);
3554                        goto out;
3555                }
3556
3557                cpos += num_clusters;
3558
3559                while (num_clusters) {
3560                        ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
3561                                                     p_cluster, num_clusters,
3562                                                     &rec, &index,
3563                                                     &ref_leaf_bh);
3564                        if (ret) {
3565                                mlog_errno(ret);
3566                                goto out;
3567                        }
3568
3569                        BUG_ON(!rec.r_refcount);
3570
3571                        rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
3572
3573                        /*
3574                         * We really don't know whether the other clusters is in
3575                         * this refcount block or not, so just take the worst
3576                         * case that all the clusters are in this block and each
3577                         * one will split a refcount rec, so totally we need
3578                         * clusters * 2 new refcount rec.
3579                         */
3580                        if (le16_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
3581                            le16_to_cpu(rb->rf_records.rl_count))
3582                                ref_blocks++;
3583
3584                        *credits += 1;
3585                        brelse(ref_leaf_bh);
3586                        ref_leaf_bh = NULL;
3587
3588                        if (num_clusters <= le32_to_cpu(rec.r_clusters))
3589                                break;
3590                        else
3591                                num_clusters -= le32_to_cpu(rec.r_clusters);
3592                        p_cluster += num_clusters;
3593                }
3594        }
3595
3596        *meta_add += ref_blocks;
3597        if (!ref_blocks)
3598                goto out;
3599
3600        rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
3601        if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
3602                *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
3603        else {
3604                struct ocfs2_extent_tree et;
3605
3606                ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
3607                *credits += ocfs2_calc_extend_credits(inode->i_sb,
3608                                                      et.et_root_el);
3609        }
3610
3611out:
3612        brelse(ref_leaf_bh);
3613        return ret;
3614}
3615
3616/*
3617 * Do CoW for xattr.
3618 */
3619int ocfs2_refcount_cow_xattr(struct inode *inode,
3620                             struct ocfs2_dinode *di,
3621                             struct ocfs2_xattr_value_buf *vb,
3622                             struct ocfs2_refcount_tree *ref_tree,
3623                             struct buffer_head *ref_root_bh,
3624                             u32 cpos, u32 write_len,
3625                             struct ocfs2_post_refcount *post)
3626{
3627        int ret;
3628        struct ocfs2_xattr_value_root *xv = vb->vb_xv;
3629        struct ocfs2_cow_context *context = NULL;
3630        u32 cow_start, cow_len;
3631
3632        BUG_ON(!ocfs2_is_refcount_inode(inode));
3633
3634        ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
3635                                              cpos, write_len, UINT_MAX,
3636                                              &cow_start, &cow_len);
3637        if (ret) {
3638                mlog_errno(ret);
3639                goto out;
3640        }
3641
3642        BUG_ON(cow_len == 0);
3643
3644        context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3645        if (!context) {
3646                ret = -ENOMEM;
3647                mlog_errno(ret);
3648                goto out;
3649        }
3650
3651        context->inode = inode;
3652        context->cow_start = cow_start;
3653        context->cow_len = cow_len;
3654        context->ref_tree = ref_tree;
3655        context->ref_root_bh = ref_root_bh;
3656        context->cow_object = xv;
3657
3658        context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
3659        /* We need the extra credits for duplicate_clusters by jbd. */
3660        context->extra_credits =
3661                ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
3662        context->get_clusters = ocfs2_xattr_value_get_clusters;
3663        context->post_refcount = post;
3664
3665        ocfs2_init_xattr_value_extent_tree(&context->data_et,
3666                                           INODE_CACHE(inode), vb);
3667
3668        ret = ocfs2_replace_cow(context);
3669        if (ret)
3670                mlog_errno(ret);
3671
3672out:
3673        kfree(context);
3674        return ret;
3675}
3676
3677/*
3678 * Insert a new extent into refcount tree and mark a extent rec
3679 * as refcounted in the dinode tree.
3680 */
3681int ocfs2_add_refcount_flag(struct inode *inode,
3682                            struct ocfs2_extent_tree *data_et,
3683                            struct ocfs2_caching_info *ref_ci,
3684                            struct buffer_head *ref_root_bh,
3685                            u32 cpos, u32 p_cluster, u32 num_clusters,
3686                            struct ocfs2_cached_dealloc_ctxt *dealloc,
3687                            struct ocfs2_post_refcount *post)
3688{
3689        int ret;
3690        handle_t *handle;
3691        int credits = 1, ref_blocks = 0;
3692        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3693        struct ocfs2_alloc_context *meta_ac = NULL;
3694
3695        /* We need to be able to handle at least an extent tree split. */
3696        ref_blocks = ocfs2_extend_meta_needed(data_et->et_root_el);
3697
3698        ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
3699                                               ref_ci, ref_root_bh,
3700                                               p_cluster, num_clusters,
3701                                               &ref_blocks, &credits);
3702        if (ret) {
3703                mlog_errno(ret);
3704                goto out;
3705        }
3706
3707        trace_ocfs2_add_refcount_flag(ref_blocks, credits);
3708
3709        if (ref_blocks) {
3710                ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
3711                                                        ref_blocks, &meta_ac);
3712                if (ret) {
3713                        mlog_errno(ret);
3714                        goto out;
3715                }
3716        }
3717
3718        if (post)
3719                credits += post->credits;
3720
3721        handle = ocfs2_start_trans(osb, credits);
3722        if (IS_ERR(handle)) {
3723                ret = PTR_ERR(handle);
3724                mlog_errno(ret);
3725                goto out;
3726        }
3727
3728        ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
3729                                           cpos, num_clusters, p_cluster,
3730                                           meta_ac, dealloc);
3731        if (ret) {
3732                mlog_errno(ret);
3733                goto out_commit;
3734        }
3735
3736        ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3737                                        p_cluster, num_clusters, 0,
3738                                        meta_ac, dealloc);
3739        if (ret) {
3740                mlog_errno(ret);
3741                goto out_commit;
3742        }
3743
3744        if (post && post->func) {
3745                ret = post->func(inode, handle, post->para);
3746                if (ret)
3747                        mlog_errno(ret);
3748        }
3749
3750out_commit:
3751        ocfs2_commit_trans(osb, handle);
3752out:
3753        if (meta_ac)
3754                ocfs2_free_alloc_context(meta_ac);
3755        return ret;
3756}
3757
3758static int ocfs2_change_ctime(struct inode *inode,
3759                              struct buffer_head *di_bh)
3760{
3761        int ret;
3762        handle_t *handle;
3763        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3764
3765        handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
3766                                   OCFS2_INODE_UPDATE_CREDITS);
3767        if (IS_ERR(handle)) {
3768                ret = PTR_ERR(handle);
3769                mlog_errno(ret);
3770                goto out;
3771        }
3772
3773        ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
3774                                      OCFS2_JOURNAL_ACCESS_WRITE);
3775        if (ret) {
3776                mlog_errno(ret);
3777                goto out_commit;
3778        }
3779
3780        inode->i_ctime = current_time(inode);
3781        di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3782        di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3783
3784        ocfs2_journal_dirty(handle, di_bh);
3785
3786out_commit:
3787        ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3788out:
3789        return ret;
3790}
3791
3792static int ocfs2_attach_refcount_tree(struct inode *inode,
3793                                      struct buffer_head *di_bh)
3794{
3795        int ret, data_changed = 0;
3796        struct buffer_head *ref_root_bh = NULL;
3797        struct ocfs2_inode_info *oi = OCFS2_I(inode);
3798        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3799        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3800        struct ocfs2_refcount_tree *ref_tree;
3801        unsigned int ext_flags;
3802        loff_t size;
3803        u32 cpos, num_clusters, clusters, p_cluster;
3804        struct ocfs2_cached_dealloc_ctxt dealloc;
3805        struct ocfs2_extent_tree di_et;
3806
3807        ocfs2_init_dealloc_ctxt(&dealloc);
3808
3809        if (!ocfs2_is_refcount_inode(inode)) {
3810                ret = ocfs2_create_refcount_tree(inode, di_bh);
3811                if (ret) {
3812                        mlog_errno(ret);
3813                        goto out;
3814                }
3815        }
3816
3817        BUG_ON(!di->i_refcount_loc);
3818        ret = ocfs2_lock_refcount_tree(osb,
3819                                       le64_to_cpu(di->i_refcount_loc), 1,
3820                                       &ref_tree, &ref_root_bh);
3821        if (ret) {
3822                mlog_errno(ret);
3823                goto out;
3824        }
3825
3826        if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
3827                goto attach_xattr;
3828
3829        ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
3830
3831        size = i_size_read(inode);
3832        clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
3833
3834        cpos = 0;
3835        while (cpos < clusters) {
3836                ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3837                                         &num_clusters, &ext_flags);
3838                if (ret) {
3839                        mlog_errno(ret);
3840                        goto unlock;
3841                }
3842                if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
3843                        ret = ocfs2_add_refcount_flag(inode, &di_et,
3844                                                      &ref_tree->rf_ci,
3845                                                      ref_root_bh, cpos,
3846                                                      p_cluster, num_clusters,
3847                                                      &dealloc, NULL);
3848                        if (ret) {
3849                                mlog_errno(ret);
3850                                goto unlock;
3851                        }
3852
3853                        data_changed = 1;
3854                }
3855                cpos += num_clusters;
3856        }
3857
3858attach_xattr:
3859        if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
3860                ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
3861                                                       &ref_tree->rf_ci,
3862                                                       ref_root_bh,
3863                                                       &dealloc);
3864                if (ret) {
3865                        mlog_errno(ret);
3866                        goto unlock;
3867                }
3868        }
3869
3870        if (data_changed) {
3871                ret = ocfs2_change_ctime(inode, di_bh);
3872                if (ret)
3873                        mlog_errno(ret);
3874        }
3875
3876unlock:
3877        ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3878        brelse(ref_root_bh);
3879
3880        if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
3881                ocfs2_schedule_truncate_log_flush(osb, 1);
3882                ocfs2_run_deallocs(osb, &dealloc);
3883        }
3884out:
3885        /*
3886         * Empty the extent map so that we may get the right extent
3887         * record from the disk.
3888         */
3889        ocfs2_extent_map_trunc(inode, 0);
3890
3891        return ret;
3892}
3893
3894static int ocfs2_add_refcounted_extent(struct inode *inode,
3895                                   struct ocfs2_extent_tree *et,
3896                                   struct ocfs2_caching_info *ref_ci,
3897                                   struct buffer_head *ref_root_bh,
3898                                   u32 cpos, u32 p_cluster, u32 num_clusters,
3899                                   unsigned int ext_flags,
3900                                   struct ocfs2_cached_dealloc_ctxt *dealloc)
3901{
3902        int ret;
3903        handle_t *handle;
3904        int credits = 0;
3905        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3906        struct ocfs2_alloc_context *meta_ac = NULL;
3907
3908        ret = ocfs2_lock_refcount_allocators(inode->i_sb,
3909                                             p_cluster, num_clusters,
3910                                             et, ref_ci,
3911                                             ref_root_bh, &meta_ac,
3912                                             NULL, &credits);
3913        if (ret) {
3914                mlog_errno(ret);
3915                goto out;
3916        }
3917
3918        handle = ocfs2_start_trans(osb, credits);
3919        if (IS_ERR(handle)) {
3920                ret = PTR_ERR(handle);
3921                mlog_errno(ret);
3922                goto out;
3923        }
3924
3925        ret = ocfs2_insert_extent(handle, et, cpos,
3926                        ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
3927                        num_clusters, ext_flags, meta_ac);
3928        if (ret) {
3929                mlog_errno(ret);
3930                goto out_commit;
3931        }
3932
3933        ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3934                                      p_cluster, num_clusters,
3935                                      meta_ac, dealloc);
3936        if (ret) {
3937                mlog_errno(ret);
3938                goto out_commit;
3939        }
3940
3941        ret = dquot_alloc_space_nodirty(inode,
3942                ocfs2_clusters_to_bytes(osb->sb, num_clusters));
3943        if (ret)
3944                mlog_errno(ret);
3945
3946out_commit:
3947        ocfs2_commit_trans(osb, handle);
3948out:
3949        if (meta_ac)
3950                ocfs2_free_alloc_context(meta_ac);
3951        return ret;
3952}
3953
3954static int ocfs2_duplicate_inline_data(struct inode *s_inode,
3955                                       struct buffer_head *s_bh,
3956                                       struct inode *t_inode,
3957                                       struct buffer_head *t_bh)
3958{
3959        int ret;
3960        handle_t *handle;
3961        struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
3962        struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
3963        struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
3964
3965        BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
3966
3967        handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
3968        if (IS_ERR(handle)) {
3969                ret = PTR_ERR(handle);
3970                mlog_errno(ret);
3971                goto out;
3972        }
3973
3974        ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
3975                                      OCFS2_JOURNAL_ACCESS_WRITE);
3976        if (ret) {
3977                mlog_errno(ret);
3978                goto out_commit;
3979        }
3980
3981        t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
3982        memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
3983               le16_to_cpu(s_di->id2.i_data.id_count));
3984        spin_lock(&OCFS2_I(t_inode)->ip_lock);
3985        OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
3986        t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
3987        spin_unlock(&OCFS2_I(t_inode)->ip_lock);
3988
3989        ocfs2_journal_dirty(handle, t_bh);
3990
3991out_commit:
3992        ocfs2_commit_trans(osb, handle);
3993out:
3994        return ret;
3995}
3996
3997static int ocfs2_duplicate_extent_list(struct inode *s_inode,
3998                                struct inode *t_inode,
3999                                struct buffer_head *t_bh,
4000                                struct ocfs2_caching_info *ref_ci,
4001                                struct buffer_head *ref_root_bh,
4002                                struct ocfs2_cached_dealloc_ctxt *dealloc)
4003{
4004        int ret = 0;
4005        u32 p_cluster, num_clusters, clusters, cpos;
4006        loff_t size;
4007        unsigned int ext_flags;
4008        struct ocfs2_extent_tree et;
4009
4010        ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
4011
4012        size = i_size_read(s_inode);
4013        clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
4014
4015        cpos = 0;
4016        while (cpos < clusters) {
4017                ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
4018                                         &num_clusters, &ext_flags);
4019                if (ret) {
4020                        mlog_errno(ret);
4021                        goto out;
4022                }
4023                if (p_cluster) {
4024                        ret = ocfs2_add_refcounted_extent(t_inode, &et,
4025                                                          ref_ci, ref_root_bh,
4026                                                          cpos, p_cluster,
4027                                                          num_clusters,
4028                                                          ext_flags,
4029                                                          dealloc);
4030                        if (ret) {
4031                                mlog_errno(ret);
4032                                goto out;
4033                        }
4034                }
4035
4036                cpos += num_clusters;
4037        }
4038
4039out:
4040        return ret;
4041}
4042
4043/*
4044 * change the new file's attributes to the src.
4045 *
4046 * reflink creates a snapshot of a file, that means the attributes
4047 * must be identical except for three exceptions - nlink, ino, and ctime.
4048 */
4049static int ocfs2_complete_reflink(struct inode *s_inode,
4050                                  struct buffer_head *s_bh,
4051                                  struct inode *t_inode,
4052                                  struct buffer_head *t_bh,
4053                                  bool preserve)
4054{
4055        int ret;
4056        handle_t *handle;
4057        struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4058        struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
4059        loff_t size = i_size_read(s_inode);
4060
4061        handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
4062                                   OCFS2_INODE_UPDATE_CREDITS);
4063        if (IS_ERR(handle)) {
4064                ret = PTR_ERR(handle);
4065                mlog_errno(ret);
4066                return ret;
4067        }
4068
4069        ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4070                                      OCFS2_JOURNAL_ACCESS_WRITE);
4071        if (ret) {
4072                mlog_errno(ret);
4073                goto out_commit;
4074        }
4075
4076        spin_lock(&OCFS2_I(t_inode)->ip_lock);
4077        OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
4078        OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
4079        OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
4080        spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4081        i_size_write(t_inode, size);
4082        t_inode->i_blocks = s_inode->i_blocks;
4083
4084        di->i_xattr_inline_size = s_di->i_xattr_inline_size;
4085        di->i_clusters = s_di->i_clusters;
4086        di->i_size = s_di->i_size;
4087        di->i_dyn_features = s_di->i_dyn_features;
4088        di->i_attr = s_di->i_attr;
4089
4090        if (preserve) {
4091                t_inode->i_uid = s_inode->i_uid;
4092                t_inode->i_gid = s_inode->i_gid;
4093                t_inode->i_mode = s_inode->i_mode;
4094                di->i_uid = s_di->i_uid;
4095                di->i_gid = s_di->i_gid;
4096                di->i_mode = s_di->i_mode;
4097
4098                /*
4099                 * update time.
4100                 * we want mtime to appear identical to the source and
4101                 * update ctime.
4102                 */
4103                t_inode->i_ctime = current_time(t_inode);
4104
4105                di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
4106                di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
4107
4108                t_inode->i_mtime = s_inode->i_mtime;
4109                di->i_mtime = s_di->i_mtime;
4110                di->i_mtime_nsec = s_di->i_mtime_nsec;
4111        }
4112
4113        ocfs2_journal_dirty(handle, t_bh);
4114
4115out_commit:
4116        ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
4117        return ret;
4118}
4119
4120static int ocfs2_create_reflink_node(struct inode *s_inode,
4121                                     struct buffer_head *s_bh,
4122                                     struct inode *t_inode,
4123                                     struct buffer_head *t_bh,
4124                                     bool preserve)
4125{
4126        int ret;
4127        struct buffer_head *ref_root_bh = NULL;
4128        struct ocfs2_cached_dealloc_ctxt dealloc;
4129        struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4130        struct ocfs2_refcount_block *rb;
4131        struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
4132        struct ocfs2_refcount_tree *ref_tree;
4133
4134        ocfs2_init_dealloc_ctxt(&dealloc);
4135
4136        ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4137                                      le64_to_cpu(di->i_refcount_loc));
4138        if (ret) {
4139                mlog_errno(ret);
4140                goto out;
4141        }
4142
4143        if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4144                ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
4145                                                  t_inode, t_bh);
4146                if (ret)
4147                        mlog_errno(ret);
4148                goto out;
4149        }
4150
4151        ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
4152                                       1, &ref_tree, &ref_root_bh);
4153        if (ret) {
4154                mlog_errno(ret);
4155                goto out;
4156        }
4157        rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
4158
4159        ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
4160                                          &ref_tree->rf_ci, ref_root_bh,
4161                                          &dealloc);
4162        if (ret) {
4163                mlog_errno(ret);
4164                goto out_unlock_refcount;
4165        }
4166
4167out_unlock_refcount:
4168        ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4169        brelse(ref_root_bh);
4170out:
4171        if (ocfs2_dealloc_has_cluster(&dealloc)) {
4172                ocfs2_schedule_truncate_log_flush(osb, 1);
4173                ocfs2_run_deallocs(osb, &dealloc);
4174        }
4175
4176        return ret;
4177}
4178
4179static int __ocfs2_reflink(struct dentry *old_dentry,
4180                           struct buffer_head *old_bh,
4181                           struct inode *new_inode,
4182                           bool preserve)
4183{
4184        int ret;
4185        struct inode *inode = d_inode(old_dentry);
4186        struct buffer_head *new_bh = NULL;
4187
4188        if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
4189                ret = -EINVAL;
4190                mlog_errno(ret);
4191                goto out;
4192        }
4193
4194        ret = filemap_fdatawrite(inode->i_mapping);
4195        if (ret) {
4196                mlog_errno(ret);
4197                goto out;
4198        }
4199
4200        ret = ocfs2_attach_refcount_tree(inode, old_bh);
4201        if (ret) {
4202                mlog_errno(ret);
4203                goto out;
4204        }
4205
4206        inode_lock_nested(new_inode, I_MUTEX_CHILD);
4207        ret = ocfs2_inode_lock_nested(new_inode, &new_bh, 1,
4208                                      OI_LS_REFLINK_TARGET);
4209        if (ret) {
4210                mlog_errno(ret);
4211                goto out_unlock;
4212        }
4213
4214        ret = ocfs2_create_reflink_node(inode, old_bh,
4215                                        new_inode, new_bh, preserve);
4216        if (ret) {
4217                mlog_errno(ret);
4218                goto inode_unlock;
4219        }
4220
4221        if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
4222                ret = ocfs2_reflink_xattrs(inode, old_bh,
4223                                           new_inode, new_bh,
4224                                           preserve);
4225                if (ret) {
4226                        mlog_errno(ret);
4227                        goto inode_unlock;
4228                }
4229        }
4230
4231        ret = ocfs2_complete_reflink(inode, old_bh,
4232                                     new_inode, new_bh, preserve);
4233        if (ret)
4234                mlog_errno(ret);
4235
4236inode_unlock:
4237        ocfs2_inode_unlock(new_inode, 1);
4238        brelse(new_bh);
4239out_unlock:
4240        inode_unlock(new_inode);
4241out:
4242        if (!ret) {
4243                ret = filemap_fdatawait(inode->i_mapping);
4244                if (ret)
4245                        mlog_errno(ret);
4246        }
4247        return ret;
4248}
4249
4250static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
4251                         struct dentry *new_dentry, bool preserve)
4252{
4253        int error;
4254        struct inode *inode = d_inode(old_dentry);
4255        struct buffer_head *old_bh = NULL;
4256        struct inode *new_orphan_inode = NULL;
4257
4258        if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4259                return -EOPNOTSUPP;
4260
4261
4262        error = ocfs2_create_inode_in_orphan(dir, inode->i_mode,
4263                                             &new_orphan_inode);
4264        if (error) {
4265                mlog_errno(error);
4266                goto out;
4267        }
4268
4269        error = ocfs2_rw_lock(inode, 1);
4270        if (error) {
4271                mlog_errno(error);
4272                goto out;
4273        }
4274
4275        error = ocfs2_inode_lock(inode, &old_bh, 1);
4276        if (error) {
4277                mlog_errno(error);
4278                ocfs2_rw_unlock(inode, 1);
4279                goto out;
4280        }
4281
4282        down_write(&OCFS2_I(inode)->ip_xattr_sem);
4283        down_write(&OCFS2_I(inode)->ip_alloc_sem);
4284        error = __ocfs2_reflink(old_dentry, old_bh,
4285                                new_orphan_inode, preserve);
4286        up_write(&OCFS2_I(inode)->ip_alloc_sem);
4287        up_write(&OCFS2_I(inode)->ip_xattr_sem);
4288
4289        ocfs2_inode_unlock(inode, 1);
4290        ocfs2_rw_unlock(inode, 1);
4291        brelse(old_bh);
4292
4293        if (error) {
4294                mlog_errno(error);
4295                goto out;
4296        }
4297
4298        /* If the security isn't preserved, we need to re-initialize them. */
4299        if (!preserve) {
4300                error = ocfs2_init_security_and_acl(dir, new_orphan_inode,
4301                                                    &new_dentry->d_name);
4302                if (error)
4303                        mlog_errno(error);
4304        }
4305out:
4306        if (!error) {
4307                error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
4308                                                       new_dentry);
4309                if (error)
4310                        mlog_errno(error);
4311        }
4312
4313        if (new_orphan_inode) {
4314                /*
4315                 * We need to open_unlock the inode no matter whether we
4316                 * succeed or not, so that other nodes can delete it later.
4317                 */
4318                ocfs2_open_unlock(new_orphan_inode);
4319                if (error)
4320                        iput(new_orphan_inode);
4321        }
4322
4323        return error;
4324}
4325
4326/*
4327 * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
4328 * sys_reflink().  This will go away when vfs_reflink() exists in
4329 * fs/namei.c.
4330 */
4331
4332/* copied from may_create in VFS. */
4333static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
4334{
4335        if (d_really_is_positive(child))
4336                return -EEXIST;
4337        if (IS_DEADDIR(dir))
4338                return -ENOENT;
4339        return inode_permission(dir, MAY_WRITE | MAY_EXEC);
4340}
4341
4342/**
4343 * ocfs2_vfs_reflink - Create a reference-counted link
4344 *
4345 * @old_dentry:        source dentry + inode
4346 * @dir:       directory to create the target
4347 * @new_dentry:        target dentry
4348 * @preserve:  if true, preserve all file attributes
4349 */
4350static int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
4351                             struct dentry *new_dentry, bool preserve)
4352{
4353        struct inode *inode = d_inode(old_dentry);
4354        int error;
4355
4356        if (!inode)
4357                return -ENOENT;
4358
4359        error = ocfs2_may_create(dir, new_dentry);
4360        if (error)
4361                return error;
4362
4363        if (dir->i_sb != inode->i_sb)
4364                return -EXDEV;
4365
4366        /*
4367         * A reflink to an append-only or immutable file cannot be created.
4368         */
4369        if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4370                return -EPERM;
4371
4372        /* Only regular files can be reflinked. */
4373        if (!S_ISREG(inode->i_mode))
4374                return -EPERM;
4375
4376        /*
4377         * If the caller wants to preserve ownership, they require the
4378         * rights to do so.
4379         */
4380        if (preserve) {
4381                if (!uid_eq(current_fsuid(), inode->i_uid) && !capable(CAP_CHOWN))
4382                        return -EPERM;
4383                if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
4384                        return -EPERM;
4385        }
4386
4387        /*
4388         * If the caller is modifying any aspect of the attributes, they
4389         * are not creating a snapshot.  They need read permission on the
4390         * file.
4391         */
4392        if (!preserve) {
4393                error = inode_permission(inode, MAY_READ);
4394                if (error)
4395                        return error;
4396        }
4397
4398        inode_lock(inode);
4399        error = dquot_initialize(dir);
4400        if (!error)
4401                error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
4402        inode_unlock(inode);
4403        if (!error)
4404                fsnotify_create(dir, new_dentry);
4405        return error;
4406}
4407/*
4408 * Most codes are copied from sys_linkat.
4409 */
4410int ocfs2_reflink_ioctl(struct inode *inode,
4411                        const char __user *oldname,
4412                        const char __user *newname,
4413                        bool preserve)
4414{
4415        struct dentry *new_dentry;
4416        struct path old_path, new_path;
4417        int error;
4418
4419        if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4420                return -EOPNOTSUPP;
4421
4422        error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
4423        if (error) {
4424                mlog_errno(error);
4425                return error;
4426        }
4427
4428        new_dentry = user_path_create(AT_FDCWD, newname, &new_path, 0);
4429        error = PTR_ERR(new_dentry);
4430        if (IS_ERR(new_dentry)) {
4431                mlog_errno(error);
4432                goto out;
4433        }
4434
4435        error = -EXDEV;
4436        if (old_path.mnt != new_path.mnt) {
4437                mlog_errno(error);
4438                goto out_dput;
4439        }
4440
4441        error = ocfs2_vfs_reflink(old_path.dentry,
4442                                  d_inode(new_path.dentry),
4443                                  new_dentry, preserve);
4444out_dput:
4445        done_path_create(&new_path, new_dentry);
4446out:
4447        path_put(&old_path);
4448
4449        return error;
4450}
4451
4452/* Update destination inode size, if necessary. */
4453static int ocfs2_reflink_update_dest(struct inode *dest,
4454                                     struct buffer_head *d_bh,
4455                                     loff_t newlen)
4456{
4457        handle_t *handle;
4458        int ret;
4459
4460        dest->i_blocks = ocfs2_inode_sector_count(dest);
4461
4462        if (newlen <= i_size_read(dest))
4463                return 0;
4464
4465        handle = ocfs2_start_trans(OCFS2_SB(dest->i_sb),
4466                                   OCFS2_INODE_UPDATE_CREDITS);
4467        if (IS_ERR(handle)) {
4468                ret = PTR_ERR(handle);
4469                mlog_errno(ret);
4470                return ret;
4471        }
4472
4473        /* Extend i_size if needed. */
4474        spin_lock(&OCFS2_I(dest)->ip_lock);
4475        if (newlen > i_size_read(dest))
4476                i_size_write(dest, newlen);
4477        spin_unlock(&OCFS2_I(dest)->ip_lock);
4478        dest->i_ctime = dest->i_mtime = current_time(dest);
4479
4480        ret = ocfs2_mark_inode_dirty(handle, dest, d_bh);
4481        if (ret) {
4482                mlog_errno(ret);
4483                goto out_commit;
4484        }
4485
4486out_commit:
4487        ocfs2_commit_trans(OCFS2_SB(dest->i_sb), handle);
4488        return ret;
4489}
4490
4491/* Remap the range pos_in:len in s_inode to pos_out:len in t_inode. */
4492static int ocfs2_reflink_remap_extent(struct inode *s_inode,
4493                                      struct buffer_head *s_bh,
4494                                      loff_t pos_in,
4495                                      struct inode *t_inode,
4496                                      struct buffer_head *t_bh,
4497                                      loff_t pos_out,
4498                                      loff_t len,
4499                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
4500{
4501        struct ocfs2_extent_tree s_et;
4502        struct ocfs2_extent_tree t_et;
4503        struct ocfs2_dinode *dis;
4504        struct buffer_head *ref_root_bh = NULL;
4505        struct ocfs2_refcount_tree *ref_tree;
4506        struct ocfs2_super *osb;
4507        loff_t pstart, plen;
4508        u32 p_cluster, num_clusters, slast, spos, tpos;
4509        unsigned int ext_flags;
4510        int ret = 0;
4511
4512        osb = OCFS2_SB(s_inode->i_sb);
4513        dis = (struct ocfs2_dinode *)s_bh->b_data;
4514        ocfs2_init_dinode_extent_tree(&s_et, INODE_CACHE(s_inode), s_bh);
4515        ocfs2_init_dinode_extent_tree(&t_et, INODE_CACHE(t_inode), t_bh);
4516
4517        spos = ocfs2_bytes_to_clusters(s_inode->i_sb, pos_in);
4518        tpos = ocfs2_bytes_to_clusters(t_inode->i_sb, pos_out);
4519        slast = ocfs2_clusters_for_bytes(s_inode->i_sb, pos_in + len);
4520
4521        while (spos < slast) {
4522                if (fatal_signal_pending(current)) {
4523                        ret = -EINTR;
4524                        goto out;
4525                }
4526
4527                /* Look up the extent. */
4528                ret = ocfs2_get_clusters(s_inode, spos, &p_cluster,
4529                                         &num_clusters, &ext_flags);
4530                if (ret) {
4531                        mlog_errno(ret);
4532                        goto out;
4533                }
4534
4535                num_clusters = min_t(u32, num_clusters, slast - spos);
4536
4537                /* Punch out the dest range. */
4538                pstart = ocfs2_clusters_to_bytes(t_inode->i_sb, tpos);
4539                plen = ocfs2_clusters_to_bytes(t_inode->i_sb, num_clusters);
4540                ret = ocfs2_remove_inode_range(t_inode, t_bh, pstart, plen);
4541                if (ret) {
4542                        mlog_errno(ret);
4543                        goto out;
4544                }
4545
4546                if (p_cluster == 0)
4547                        goto next_loop;
4548
4549                /* Lock the refcount btree... */
4550                ret = ocfs2_lock_refcount_tree(osb,
4551                                               le64_to_cpu(dis->i_refcount_loc),
4552                                               1, &ref_tree, &ref_root_bh);
4553                if (ret) {
4554                        mlog_errno(ret);
4555                        goto out;
4556                }
4557
4558                /* Mark s_inode's extent as refcounted. */
4559                if (!(ext_flags & OCFS2_EXT_REFCOUNTED)) {
4560                        ret = ocfs2_add_refcount_flag(s_inode, &s_et,
4561                                                      &ref_tree->rf_ci,
4562                                                      ref_root_bh, spos,
4563                                                      p_cluster, num_clusters,
4564                                                      dealloc, NULL);
4565                        if (ret) {
4566                                mlog_errno(ret);
4567                                goto out_unlock_refcount;
4568                        }
4569                }
4570
4571                /* Map in the new extent. */
4572                ext_flags |= OCFS2_EXT_REFCOUNTED;
4573                ret = ocfs2_add_refcounted_extent(t_inode, &t_et,
4574                                                  &ref_tree->rf_ci,
4575                                                  ref_root_bh,
4576                                                  tpos, p_cluster,
4577                                                  num_clusters,
4578                                                  ext_flags,
4579                                                  dealloc);
4580                if (ret) {
4581                        mlog_errno(ret);
4582                        goto out_unlock_refcount;
4583                }
4584
4585                ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4586                brelse(ref_root_bh);
4587next_loop:
4588                spos += num_clusters;
4589                tpos += num_clusters;
4590        }
4591
4592out:
4593        return ret;
4594out_unlock_refcount:
4595        ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4596        brelse(ref_root_bh);
4597        return ret;
4598}
4599
4600/* Set up refcount tree and remap s_inode to t_inode. */
4601static int ocfs2_reflink_remap_blocks(struct inode *s_inode,
4602                                      struct buffer_head *s_bh,
4603                                      loff_t pos_in,
4604                                      struct inode *t_inode,
4605                                      struct buffer_head *t_bh,
4606                                      loff_t pos_out,
4607                                      loff_t len)
4608{
4609        struct ocfs2_cached_dealloc_ctxt dealloc;
4610        struct ocfs2_super *osb;
4611        struct ocfs2_dinode *dis;
4612        struct ocfs2_dinode *dit;
4613        int ret;
4614
4615        osb = OCFS2_SB(s_inode->i_sb);
4616        dis = (struct ocfs2_dinode *)s_bh->b_data;
4617        dit = (struct ocfs2_dinode *)t_bh->b_data;
4618        ocfs2_init_dealloc_ctxt(&dealloc);
4619
4620        /*
4621         * If we're reflinking the entire file and the source is inline
4622         * data, just copy the contents.
4623         */
4624        if (pos_in == pos_out && pos_in == 0 && len == i_size_read(s_inode) &&
4625            i_size_read(t_inode) <= len &&
4626            (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) {
4627                ret = ocfs2_duplicate_inline_data(s_inode, s_bh, t_inode, t_bh);
4628                if (ret)
4629                        mlog_errno(ret);
4630                goto out;
4631        }
4632
4633        /*
4634         * If both inodes belong to two different refcount groups then
4635         * forget it because we don't know how (or want) to go merging
4636         * refcount trees.
4637         */
4638        ret = -EOPNOTSUPP;
4639        if (ocfs2_is_refcount_inode(s_inode) &&
4640            ocfs2_is_refcount_inode(t_inode) &&
4641            le64_to_cpu(dis->i_refcount_loc) !=
4642            le64_to_cpu(dit->i_refcount_loc))
4643                goto out;
4644
4645        /* Neither inode has a refcount tree.  Add one to s_inode. */
4646        if (!ocfs2_is_refcount_inode(s_inode) &&
4647            !ocfs2_is_refcount_inode(t_inode)) {
4648                ret = ocfs2_create_refcount_tree(s_inode, s_bh);
4649                if (ret) {
4650                        mlog_errno(ret);
4651                        goto out;
4652                }
4653        }
4654
4655        /* Ensure that both inodes end up with the same refcount tree. */
4656        if (!ocfs2_is_refcount_inode(s_inode)) {
4657                ret = ocfs2_set_refcount_tree(s_inode, s_bh,
4658                                              le64_to_cpu(dit->i_refcount_loc));
4659                if (ret) {
4660                        mlog_errno(ret);
4661                        goto out;
4662                }
4663        }
4664        if (!ocfs2_is_refcount_inode(t_inode)) {
4665                ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4666                                              le64_to_cpu(dis->i_refcount_loc));
4667                if (ret) {
4668                        mlog_errno(ret);
4669                        goto out;
4670                }
4671        }
4672
4673        /* Turn off inline data in the dest file. */
4674        if (OCFS2_I(t_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4675                ret = ocfs2_convert_inline_data_to_extents(t_inode, t_bh);
4676                if (ret) {
4677                        mlog_errno(ret);
4678                        goto out;
4679                }
4680        }
4681
4682        /* Actually remap extents now. */
4683        ret = ocfs2_reflink_remap_extent(s_inode, s_bh, pos_in, t_inode, t_bh,
4684                                         pos_out, len, &dealloc);
4685        if (ret) {
4686                mlog_errno(ret);
4687                goto out;
4688        }
4689
4690out:
4691        if (ocfs2_dealloc_has_cluster(&dealloc)) {
4692                ocfs2_schedule_truncate_log_flush(osb, 1);
4693                ocfs2_run_deallocs(osb, &dealloc);
4694        }
4695
4696        return ret;
4697}
4698
4699/* Lock an inode and grab a bh pointing to the inode. */
4700static int ocfs2_reflink_inodes_lock(struct inode *s_inode,
4701                                     struct buffer_head **bh1,
4702                                     struct inode *t_inode,
4703                                     struct buffer_head **bh2)
4704{
4705        struct inode *inode1;
4706        struct inode *inode2;
4707        struct ocfs2_inode_info *oi1;
4708        struct ocfs2_inode_info *oi2;
4709        bool same_inode = (s_inode == t_inode);
4710        int status;
4711
4712        /* First grab the VFS and rw locks. */
4713        lock_two_nondirectories(s_inode, t_inode);
4714        inode1 = s_inode;
4715        inode2 = t_inode;
4716        if (inode1->i_ino > inode2->i_ino)
4717                swap(inode1, inode2);
4718
4719        status = ocfs2_rw_lock(inode1, 1);
4720        if (status) {
4721                mlog_errno(status);
4722                goto out_i1;
4723        }
4724        if (!same_inode) {
4725                status = ocfs2_rw_lock(inode2, 1);
4726                if (status) {
4727                        mlog_errno(status);
4728                        goto out_i2;
4729                }
4730        }
4731
4732        /* Now go for the cluster locks */
4733        oi1 = OCFS2_I(inode1);
4734        oi2 = OCFS2_I(inode2);
4735
4736        trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
4737                                (unsigned long long)oi2->ip_blkno);
4738
4739        if (*bh1)
4740                *bh1 = NULL;
4741        if (*bh2)
4742                *bh2 = NULL;
4743
4744        /* We always want to lock the one with the lower lockid first. */
4745        if (oi1->ip_blkno > oi2->ip_blkno)
4746                mlog_errno(-ENOLCK);
4747
4748        /* lock id1 */
4749        status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_REFLINK_TARGET);
4750        if (status < 0) {
4751                if (status != -ENOENT)
4752                        mlog_errno(status);
4753                goto out_rw2;
4754        }
4755
4756        /* lock id2 */
4757        if (!same_inode) {
4758                status = ocfs2_inode_lock_nested(inode2, bh2, 1,
4759                                                 OI_LS_REFLINK_TARGET);
4760                if (status < 0) {
4761                        if (status != -ENOENT)
4762                                mlog_errno(status);
4763                        goto out_cl1;
4764                }
4765        } else
4766                *bh2 = *bh1;
4767
4768        trace_ocfs2_double_lock_end(
4769                        (unsigned long long)OCFS2_I(inode1)->ip_blkno,
4770                        (unsigned long long)OCFS2_I(inode2)->ip_blkno);
4771
4772        return 0;
4773
4774out_cl1:
4775        ocfs2_inode_unlock(inode1, 1);
4776        brelse(*bh1);
4777        *bh1 = NULL;
4778out_rw2:
4779        ocfs2_rw_unlock(inode2, 1);
4780out_i2:
4781        ocfs2_rw_unlock(inode1, 1);
4782out_i1:
4783        unlock_two_nondirectories(s_inode, t_inode);
4784        return status;
4785}
4786
4787/* Unlock both inodes and release buffers. */
4788static void ocfs2_reflink_inodes_unlock(struct inode *s_inode,
4789                                        struct buffer_head *s_bh,
4790                                        struct inode *t_inode,
4791                                        struct buffer_head *t_bh)
4792{
4793        ocfs2_inode_unlock(s_inode, 1);
4794        ocfs2_rw_unlock(s_inode, 1);
4795        brelse(s_bh);
4796        if (s_inode != t_inode) {
4797                ocfs2_inode_unlock(t_inode, 1);
4798                ocfs2_rw_unlock(t_inode, 1);
4799                brelse(t_bh);
4800        }
4801        unlock_two_nondirectories(s_inode, t_inode);
4802}
4803
4804/* Link a range of blocks from one file to another. */
4805int ocfs2_reflink_remap_range(struct file *file_in,
4806                              loff_t pos_in,
4807                              struct file *file_out,
4808                              loff_t pos_out,
4809                              u64 len,
4810                              bool is_dedupe)
4811{
4812        struct inode *inode_in = file_inode(file_in);
4813        struct inode *inode_out = file_inode(file_out);
4814        struct ocfs2_super *osb = OCFS2_SB(inode_in->i_sb);
4815        struct buffer_head *in_bh = NULL, *out_bh = NULL;
4816        bool same_inode = (inode_in == inode_out);
4817        ssize_t ret;
4818
4819        if (!ocfs2_refcount_tree(osb))
4820                return -EOPNOTSUPP;
4821        if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
4822                return -EROFS;
4823
4824        /* Lock both files against IO */
4825        ret = ocfs2_reflink_inodes_lock(inode_in, &in_bh, inode_out, &out_bh);
4826        if (ret)
4827                return ret;
4828
4829        /* Check file eligibility and prepare for block sharing. */
4830        ret = -EINVAL;
4831        if ((OCFS2_I(inode_in)->ip_flags & OCFS2_INODE_SYSTEM_FILE) ||
4832            (OCFS2_I(inode_out)->ip_flags & OCFS2_INODE_SYSTEM_FILE))
4833                goto out_unlock;
4834
4835        ret = vfs_clone_file_prep_inodes(inode_in, pos_in, inode_out, pos_out,
4836                        &len, is_dedupe);
4837        if (ret <= 0)
4838                goto out_unlock;
4839
4840        /* Lock out changes to the allocation maps and remap. */
4841        down_write(&OCFS2_I(inode_in)->ip_alloc_sem);
4842        if (!same_inode)
4843                down_write_nested(&OCFS2_I(inode_out)->ip_alloc_sem,
4844                                  SINGLE_DEPTH_NESTING);
4845
4846        ret = ocfs2_reflink_remap_blocks(inode_in, in_bh, pos_in, inode_out,
4847                                         out_bh, pos_out, len);
4848
4849        /* Zap any page cache for the destination file's range. */
4850        if (!ret)
4851                truncate_inode_pages_range(&inode_out->i_data, pos_out,
4852                                           PAGE_ALIGN(pos_out + len) - 1);
4853
4854        up_write(&OCFS2_I(inode_in)->ip_alloc_sem);
4855        if (!same_inode)
4856                up_write(&OCFS2_I(inode_out)->ip_alloc_sem);
4857        if (ret) {
4858                mlog_errno(ret);
4859                goto out_unlock;
4860        }
4861
4862        /*
4863         * Empty the extent map so that we may get the right extent
4864         * record from the disk.
4865         */
4866        ocfs2_extent_map_trunc(inode_in, 0);
4867        ocfs2_extent_map_trunc(inode_out, 0);
4868
4869        ret = ocfs2_reflink_update_dest(inode_out, out_bh, pos_out + len);
4870        if (ret) {
4871                mlog_errno(ret);
4872                goto out_unlock;
4873        }
4874
4875        ocfs2_reflink_inodes_unlock(inode_in, in_bh, inode_out, out_bh);
4876        return 0;
4877
4878out_unlock:
4879        ocfs2_reflink_inodes_unlock(inode_in, in_bh, inode_out, out_bh);
4880        return ret;
4881}
4882