fs/ocfs2/refcounttree.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * refcounttree.c
   4  *
   5  * Copyright (C) 2009 Oracle.  All rights reserved.
   6  */
   7
   8 #include <linux/sort.h>
   9 #include <cluster/masklog.h>
  10 #include "ocfs2.h"
  11 #include "inode.h"
  12 #include "alloc.h"
  13 #include "suballoc.h"
  14 #include "journal.h"
  15 #include "uptodate.h"
  16 #include "super.h"
  17 #include "buffer_head_io.h"
  18 #include "blockcheck.h"
  19 #include "refcounttree.h"
  20 #include "sysfile.h"
  21 #include "dlmglue.h"
  22 #include "extent_map.h"
  23 #include "aops.h"
  24 #include "xattr.h"
  25 #include "namei.h"
  26 #include "ocfs2_trace.h"
  27 #include "file.h"
  28
  29 #include <linux/bio.h>
  30 #include <linux/blkdev.h>
  31 #include <linux/slab.h>
  32 #include <linux/writeback.h>
  33 #include <linux/pagevec.h>
  34 #include <linux/swap.h>
  35 #include <linux/security.h>
  36 #include <linux/fsnotify.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/namei.h>
  39 #include <linux/mount.h>
  40 #include <linux/posix_acl.h>
  41
  42 struct ocfs2_cow_context {
  43         struct inode *inode;
  44         u32 cow_start;
  45         u32 cow_len;
  46         struct ocfs2_extent_tree data_et;
  47         struct ocfs2_refcount_tree *ref_tree;
  48         struct buffer_head *ref_root_bh;
  49         struct ocfs2_alloc_context *meta_ac;
  50         struct ocfs2_alloc_context *data_ac;
  51         struct ocfs2_cached_dealloc_ctxt dealloc;
  52         void *cow_object;
  53         struct ocfs2_post_refcount *post_refcount;
  54         int extra_credits;
  55         int (*get_clusters)(struct ocfs2_cow_context *context,
  56                             u32 v_cluster, u32 *p_cluster,
  57                             u32 *num_clusters,
  58                             unsigned int *extent_flags);
  59         int (*cow_duplicate_clusters)(handle_t *handle,
  60                                       struct inode *inode,
  61                                       u32 cpos, u32 old_cluster,
  62                                       u32 new_cluster, u32 new_len);
  63 };
  64
  65 static inline struct ocfs2_refcount_tree *
  66 cache_info_to_refcount(struct ocfs2_caching_info *ci)
  67 {
  68         return container_of(ci, struct ocfs2_refcount_tree, rf_ci);
  69 }
  70
  71 static int ocfs2_validate_refcount_block(struct super_block *sb,
  72                                          struct buffer_head *bh)
  73 {
  74         int rc;
  75         struct ocfs2_refcount_block *rb =
  76                 (struct ocfs2_refcount_block *)bh->b_data;
  77
  78         trace_ocfs2_validate_refcount_block((unsigned long long)bh->b_blocknr);
  79
  80         BUG_ON(!buffer_uptodate(bh));
  81
  82         /*
  83          * If the ecc fails, we return the error but otherwise
  84          * leave the filesystem running.  We know any error is
  85          * local to this block.
  86          */
  87         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &rb->rf_check);
  88         if (rc) {
  89                 mlog(ML_ERROR, "Checksum failed for refcount block %llu\n",
  90                      (unsigned long long)bh->b_blocknr);
  91                 return rc;
  92         }
  93
  94
  95         if (!OCFS2_IS_VALID_REFCOUNT_BLOCK(rb)) {
  96                 rc = ocfs2_error(sb,
  97                                  "Refcount block #%llu has bad signature %.*s\n",
  98                                  (unsigned long long)bh->b_blocknr, 7,
  99                                  rb->rf_signature);
 100                 goto out;
 101         }
 102
 103         if (le64_to_cpu(rb->rf_blkno) != bh->b_blocknr) {
 104                 rc = ocfs2_error(sb,
 105                                  "Refcount block #%llu has an invalid rf_blkno of %llu\n",
 106                                  (unsigned long long)bh->b_blocknr,
 107                                  (unsigned long long)le64_to_cpu(rb->rf_blkno));
 108                 goto out;
 109         }
 110
 111         if (le32_to_cpu(rb->rf_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 112                 rc = ocfs2_error(sb,
 113                                  "Refcount block #%llu has an invalid rf_fs_generation of #%u\n",
 114                                  (unsigned long long)bh->b_blocknr,
 115                                  le32_to_cpu(rb->rf_fs_generation));
 116                 goto out;
 117         }
 118 out:
 119         return rc;
 120 }
 121
 122 static int ocfs2_read_refcount_block(struct ocfs2_caching_info *ci,
 123                                      u64 rb_blkno,
 124                                      struct buffer_head **bh)
 125 {
 126         int rc;
 127         struct buffer_head *tmp = *bh;
 128
 129         rc = ocfs2_read_block(ci, rb_blkno, &tmp,
 130                               ocfs2_validate_refcount_block);
 131
 132         /* If ocfs2_read_block() got us a new bh, pass it up. */
 133         if (!rc && !*bh)
 134                 *bh = tmp;
 135
 136         return rc;
 137 }
 138
 139 static u64 ocfs2_refcount_cache_owner(struct ocfs2_caching_info *ci)
 140 {
 141         struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 142
 143         return rf->rf_blkno;
 144 }
 145
 146 static struct super_block *
 147 ocfs2_refcount_cache_get_super(struct ocfs2_caching_info *ci)
 148 {
 149         struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 150
 151         return rf->rf_sb;
 152 }
 153
 154 static void ocfs2_refcount_cache_lock(struct ocfs2_caching_info *ci)
 155 __acquires(&rf->rf_lock)
 156 {
 157         struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 158
 159         spin_lock(&rf->rf_lock);
 160 }
 161
 162 static void ocfs2_refcount_cache_unlock(struct ocfs2_caching_info *ci)
 163 __releases(&rf->rf_lock)
 164 {
 165         struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 166
 167         spin_unlock(&rf->rf_lock);
 168 }
 169
 170 static void ocfs2_refcount_cache_io_lock(struct ocfs2_caching_info *ci)
 171 {
 172         struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 173
 174         mutex_lock(&rf->rf_io_mutex);
 175 }
 176
 177 static void ocfs2_refcount_cache_io_unlock(struct ocfs2_caching_info *ci)
 178 {
 179         struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
 180
 181         mutex_unlock(&rf->rf_io_mutex);
 182 }
 183
 184 static const struct ocfs2_caching_operations ocfs2_refcount_caching_ops = {
 185         .co_owner               = ocfs2_refcount_cache_owner,
 186         .co_get_super           = ocfs2_refcount_cache_get_super,
 187         .co_cache_lock          = ocfs2_refcount_cache_lock,
 188         .co_cache_unlock        = ocfs2_refcount_cache_unlock,
 189         .co_io_lock             = ocfs2_refcount_cache_io_lock,
 190         .co_io_unlock           = ocfs2_refcount_cache_io_unlock,
 191 };
 192
 193 static struct ocfs2_refcount_tree *
 194 ocfs2_find_refcount_tree(struct ocfs2_super *osb, u64 blkno)
 195 {
 196         struct rb_node *n = osb->osb_rf_lock_tree.rb_node;
 197         struct ocfs2_refcount_tree *tree = NULL;
 198
 199         while (n) {
 200                 tree = rb_entry(n, struct ocfs2_refcount_tree, rf_node);
 201
 202                 if (blkno < tree->rf_blkno)
 203                         n = n->rb_left;
 204                 else if (blkno > tree->rf_blkno)
 205                         n = n->rb_right;
 206                 else
 207                         return tree;
 208         }
 209
 210         return NULL;
 211 }
 212
 213 /* osb_lock is already locked. */
 214 static void ocfs2_insert_refcount_tree(struct ocfs2_super *osb,
 215                                        struct ocfs2_refcount_tree *new)
 216 {
 217         u64 rf_blkno = new->rf_blkno;
 218         struct rb_node *parent = NULL;
 219         struct rb_node **p = &osb->osb_rf_lock_tree.rb_node;
 220         struct ocfs2_refcount_tree *tmp;
 221
 222         while (*p) {
 223                 parent = *p;
 224
 225                 tmp = rb_entry(parent, struct ocfs2_refcount_tree,
 226                                rf_node);
 227
 228                 if (rf_blkno < tmp->rf_blkno)
 229                         p = &(*p)->rb_left;
 230                 else if (rf_blkno > tmp->rf_blkno)
 231                         p = &(*p)->rb_right;
 232                 else {
 233                         /* This should never happen! */
 234                         mlog(ML_ERROR, "Duplicate refcount block %llu found!\n",
 235                              (unsigned long long)rf_blkno);
 236                         BUG();
 237                 }
 238         }
 239
 240         rb_link_node(&new->rf_node, parent, p);
 241         rb_insert_color(&new->rf_node, &osb->osb_rf_lock_tree);
 242 }
 243
 244 static void ocfs2_free_refcount_tree(struct ocfs2_refcount_tree *tree)
 245 {
 246         ocfs2_metadata_cache_exit(&tree->rf_ci);
 247         ocfs2_simple_drop_lockres(OCFS2_SB(tree->rf_sb), &tree->rf_lockres);
 248         ocfs2_lock_res_free(&tree->rf_lockres);
 249         kfree(tree);
 250 }
 251
 252 static inline void
 253 ocfs2_erase_refcount_tree_from_list_no_lock(struct ocfs2_super *osb,
 254                                         struct ocfs2_refcount_tree *tree)
 255 {
 256         rb_erase(&tree->rf_node, &osb->osb_rf_lock_tree);
 257         if (osb->osb_ref_tree_lru && osb->osb_ref_tree_lru == tree)
 258                 osb->osb_ref_tree_lru = NULL;
 259 }
 260
 261 static void ocfs2_erase_refcount_tree_from_list(struct ocfs2_super *osb,
 262                                         struct ocfs2_refcount_tree *tree)
 263 {
 264         spin_lock(&osb->osb_lock);
 265         ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
 266         spin_unlock(&osb->osb_lock);
 267 }
 268
 269 static void ocfs2_kref_remove_refcount_tree(struct kref *kref)
 270 {
 271         struct ocfs2_refcount_tree *tree =
 272                 container_of(kref, struct ocfs2_refcount_tree, rf_getcnt);
 273
 274         ocfs2_free_refcount_tree(tree);
 275 }
 276
 277 static inline void
 278 ocfs2_refcount_tree_get(struct ocfs2_refcount_tree *tree)
 279 {
 280         kref_get(&tree->rf_getcnt);
 281 }
 282
 283 static inline void
 284 ocfs2_refcount_tree_put(struct ocfs2_refcount_tree *tree)
 285 {
 286         kref_put(&tree->rf_getcnt, ocfs2_kref_remove_refcount_tree);
 287 }
 288
 289 static inline void ocfs2_init_refcount_tree_ci(struct ocfs2_refcount_tree *new,
 290                                                struct super_block *sb)
 291 {
 292         ocfs2_metadata_cache_init(&new->rf_ci, &ocfs2_refcount_caching_ops);
 293         mutex_init(&new->rf_io_mutex);
 294         new->rf_sb = sb;
 295         spin_lock_init(&new->rf_lock);
 296 }
 297
 298 static inline void ocfs2_init_refcount_tree_lock(struct ocfs2_super *osb,
 299                                         struct ocfs2_refcount_tree *new,
 300                                         u64 rf_blkno, u32 generation)
 301 {
 302         init_rwsem(&new->rf_sem);
 303         ocfs2_refcount_lock_res_init(&new->rf_lockres, osb,
 304                                      rf_blkno, generation);
 305 }
 306
 307 static struct ocfs2_refcount_tree*
 308 ocfs2_allocate_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno)
 309 {
 310         struct ocfs2_refcount_tree *new;
 311
 312         new = kzalloc(sizeof(struct ocfs2_refcount_tree), GFP_NOFS);
 313         if (!new)
 314                 return NULL;
 315
 316         new->rf_blkno = rf_blkno;
 317         kref_init(&new->rf_getcnt);
 318         ocfs2_init_refcount_tree_ci(new, osb->sb);
 319
 320         return new;
 321 }
 322
 323 static int ocfs2_get_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno,
 324                                    struct ocfs2_refcount_tree **ret_tree)
 325 {
 326         int ret = 0;
 327         struct ocfs2_refcount_tree *tree, *new = NULL;
 328         struct buffer_head *ref_root_bh = NULL;
 329         struct ocfs2_refcount_block *ref_rb;
 330
 331         spin_lock(&osb->osb_lock);
 332         if (osb->osb_ref_tree_lru &&
 333             osb->osb_ref_tree_lru->rf_blkno == rf_blkno)
 334                 tree = osb->osb_ref_tree_lru;
 335         else
 336                 tree = ocfs2_find_refcount_tree(osb, rf_blkno);
 337         if (tree)
 338                 goto out;
 339
 340         spin_unlock(&osb->osb_lock);
 341
 342         new = ocfs2_allocate_refcount_tree(osb, rf_blkno);
 343         if (!new) {
 344                 ret = -ENOMEM;
 345                 mlog_errno(ret);
 346                 return ret;
 347         }
 348         /*
 349          * We need the generation to create the refcount tree lock and since
 350          * it isn't changed during the tree modification, we are safe here to
 351          * read without protection.
 352          * We also have to purge the cache after we create the lock since the
 353          * refcount block may have the stale data. It can only be trusted when
 354          * we hold the refcount lock.
 355          */
 356         ret = ocfs2_read_refcount_block(&new->rf_ci, rf_blkno, &ref_root_bh);
 357         if (ret) {
 358                 mlog_errno(ret);
 359                 ocfs2_metadata_cache_exit(&new->rf_ci);
 360                 kfree(new);
 361                 return ret;
 362         }
 363
 364         ref_rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
 365         new->rf_generation = le32_to_cpu(ref_rb->rf_generation);
 366         ocfs2_init_refcount_tree_lock(osb, new, rf_blkno,
 367                                       new->rf_generation);
 368         ocfs2_metadata_cache_purge(&new->rf_ci);
 369
 370         spin_lock(&osb->osb_lock);
 371         tree = ocfs2_find_refcount_tree(osb, rf_blkno);
 372         if (tree)
 373                 goto out;
 374
 375         ocfs2_insert_refcount_tree(osb, new);
 376
 377         tree = new;
 378         new = NULL;
 379
 380 out:
 381         *ret_tree = tree;
 382
 383         osb->osb_ref_tree_lru = tree;
 384
 385         spin_unlock(&osb->osb_lock);
 386
 387         if (new)
 388                 ocfs2_free_refcount_tree(new);
 389
 390         brelse(ref_root_bh);
 391         return ret;
 392 }
 393
 394 static int ocfs2_get_refcount_block(struct inode *inode, u64 *ref_blkno)
 395 {
 396         int ret;
 397         struct buffer_head *di_bh = NULL;
 398         struct ocfs2_dinode *di;
 399
 400         ret = ocfs2_read_inode_block(inode, &di_bh);
 401         if (ret) {
 402                 mlog_errno(ret);
 403                 goto out;
 404         }
 405
 406         BUG_ON(!ocfs2_is_refcount_inode(inode));
 407
 408         di = (struct ocfs2_dinode *)di_bh->b_data;
 409         *ref_blkno = le64_to_cpu(di->i_refcount_loc);
 410         brelse(di_bh);
 411 out:
 412         return ret;
 413 }
 414
 415 static int __ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
 416                                       struct ocfs2_refcount_tree *tree, int rw)
 417 {
 418         int ret;
 419
 420         ret = ocfs2_refcount_lock(tree, rw);
 421         if (ret) {
 422                 mlog_errno(ret);
 423                 goto out;
 424         }
 425
 426         if (rw)
 427                 down_write(&tree->rf_sem);
 428         else
 429                 down_read(&tree->rf_sem);
 430
 431 out:
 432         return ret;
 433 }
 434
 435 /*
 436  * Lock the refcount tree pointed by ref_blkno and return the tree.
 437  * In most case, we lock the tree and read the refcount block.
 438  * So read it here if the caller really needs it.
 439  *
 440  * If the tree has been re-created by other node, it will free the
 441  * old one and re-create it.
 442  */
 443 int ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
 444                              u64 ref_blkno, int rw,
 445                              struct ocfs2_refcount_tree **ret_tree,
 446                              struct buffer_head **ref_bh)
 447 {
 448         int ret, delete_tree = 0;
 449         struct ocfs2_refcount_tree *tree = NULL;
 450         struct buffer_head *ref_root_bh = NULL;
 451         struct ocfs2_refcount_block *rb;
 452
 453 again:
 454         ret = ocfs2_get_refcount_tree(osb, ref_blkno, &tree);
 455         if (ret) {
 456                 mlog_errno(ret);
 457                 return ret;
 458         }
 459
 460         ocfs2_refcount_tree_get(tree);
 461
 462         ret = __ocfs2_lock_refcount_tree(osb, tree, rw);
 463         if (ret) {
 464                 mlog_errno(ret);
 465                 ocfs2_refcount_tree_put(tree);
 466                 goto out;
 467         }
 468
 469         ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
 470                                         &ref_root_bh);
 471         if (ret) {
 472                 mlog_errno(ret);
 473                 ocfs2_unlock_refcount_tree(osb, tree, rw);
 474                 goto out;
 475         }
 476
 477         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
 478         /*
 479          * If the refcount block has been freed and re-created, we may need
 480          * to recreate the refcount tree also.
 481          *
 482          * Here we just remove the tree from the rb-tree, and the last
 483          * kref holder will unlock and delete this refcount_tree.
 484          * Then we goto "again" and ocfs2_get_refcount_tree will create
 485          * the new refcount tree for us.
 486          */
 487         if (tree->rf_generation != le32_to_cpu(rb->rf_generation)) {
 488                 if (!tree->rf_removed) {
 489                         ocfs2_erase_refcount_tree_from_list(osb, tree);
 490                         tree->rf_removed = 1;
 491                         delete_tree = 1;
 492                 }
 493
 494                 ocfs2_unlock_refcount_tree(osb, tree, rw);
 495                 /*
 496                  * We get an extra reference when we create the refcount
 497                  * tree, so another put will destroy it.
 498                  */
 499                 if (delete_tree)
 500                         ocfs2_refcount_tree_put(tree);
 501                 brelse(ref_root_bh);
 502                 ref_root_bh = NULL;
 503                 goto again;
 504         }
 505
 506         *ret_tree = tree;
 507         if (ref_bh) {
 508                 *ref_bh = ref_root_bh;
 509                 ref_root_bh = NULL;
 510         }
 511 out:
 512         brelse(ref_root_bh);
 513         return ret;
 514 }
 515
 516 void ocfs2_unlock_refcount_tree(struct ocfs2_super *osb,
 517                                 struct ocfs2_refcount_tree *tree, int rw)
 518 {
 519         if (rw)
 520                 up_write(&tree->rf_sem);
 521         else
 522                 up_read(&tree->rf_sem);
 523
 524         ocfs2_refcount_unlock(tree, rw);
 525         ocfs2_refcount_tree_put(tree);
 526 }
 527
 528 void ocfs2_purge_refcount_trees(struct ocfs2_super *osb)
 529 {
 530         struct rb_node *node;
 531         struct ocfs2_refcount_tree *tree;
 532         struct rb_root *root = &osb->osb_rf_lock_tree;
 533
 534         while ((node = rb_last(root)) != NULL) {
 535                 tree = rb_entry(node, struct ocfs2_refcount_tree, rf_node);
 536
 537                 trace_ocfs2_purge_refcount_trees(
 538                                 (unsigned long long) tree->rf_blkno);
 539
 540                 rb_erase(&tree->rf_node, root);
 541                 ocfs2_free_refcount_tree(tree);
 542         }
 543 }
 544
 545 /*
 546  * Create a refcount tree for an inode.
 547  * We take for granted that the inode is already locked.
 548  */
 549 static int ocfs2_create_refcount_tree(struct inode *inode,
 550                                       struct buffer_head *di_bh)
 551 {
 552         int ret;
 553         handle_t *handle = NULL;
 554         struct ocfs2_alloc_context *meta_ac = NULL;
 555         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 556         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 557         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 558         struct buffer_head *new_bh = NULL;
 559         struct ocfs2_refcount_block *rb;
 560         struct ocfs2_refcount_tree *new_tree = NULL, *tree = NULL;
 561         u16 suballoc_bit_start;
 562         u32 num_got;
 563         u64 suballoc_loc, first_blkno;
 564
 565         BUG_ON(ocfs2_is_refcount_inode(inode));
 566
 567         trace_ocfs2_create_refcount_tree(
 568                 (unsigned long long)oi->ip_blkno);
 569
 570         ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
 571         if (ret) {
 572                 mlog_errno(ret);
 573                 goto out;
 574         }
 575
 576         handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_CREATE_CREDITS);
 577         if (IS_ERR(handle)) {
 578                 ret = PTR_ERR(handle);
 579                 mlog_errno(ret);
 580                 goto out;
 581         }
 582
 583         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 584                                       OCFS2_JOURNAL_ACCESS_WRITE);
 585         if (ret) {
 586                 mlog_errno(ret);
 587                 goto out_commit;
 588         }
 589
 590         ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
 591                                    &suballoc_bit_start, &num_got,
 592                                    &first_blkno);
 593         if (ret) {
 594                 mlog_errno(ret);
 595                 goto out_commit;
 596         }
 597
 598         new_tree = ocfs2_allocate_refcount_tree(osb, first_blkno);
 599         if (!new_tree) {
 600                 ret = -ENOMEM;
 601                 mlog_errno(ret);
 602                 goto out_commit;
 603         }
 604
 605         new_bh = sb_getblk(inode->i_sb, first_blkno);
 606         if (!new_bh) {
 607                 ret = -ENOMEM;
 608                 mlog_errno(ret);
 609                 goto out_commit;
 610         }
 611         ocfs2_set_new_buffer_uptodate(&new_tree->rf_ci, new_bh);
 612
 613         ret = ocfs2_journal_access_rb(handle, &new_tree->rf_ci, new_bh,
 614                                       OCFS2_JOURNAL_ACCESS_CREATE);
 615         if (ret) {
 616                 mlog_errno(ret);
 617                 goto out_commit;
 618         }
 619
 620         /* Initialize ocfs2_refcount_block. */
 621         rb = (struct ocfs2_refcount_block *)new_bh->b_data;
 622         memset(rb, 0, inode->i_sb->s_blocksize);
 623         strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
 624         rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
 625         rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
 626         rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
 627         rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
 628         rb->rf_blkno = cpu_to_le64(first_blkno);
 629         rb->rf_count = cpu_to_le32(1);
 630         rb->rf_records.rl_count =
 631                         cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb));
 632         spin_lock(&osb->osb_lock);
 633         rb->rf_generation = osb->s_next_generation++;
 634         spin_unlock(&osb->osb_lock);
 635
 636         ocfs2_journal_dirty(handle, new_bh);
 637
 638         spin_lock(&oi->ip_lock);
 639         oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
 640         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 641         di->i_refcount_loc = cpu_to_le64(first_blkno);
 642         spin_unlock(&oi->ip_lock);
 643
 644         trace_ocfs2_create_refcount_tree_blkno((unsigned long long)first_blkno);
 645
 646         ocfs2_journal_dirty(handle, di_bh);
 647
 648         /*
 649          * We have to init the tree lock here since it will use
 650          * the generation number to create it.
 651          */
 652         new_tree->rf_generation = le32_to_cpu(rb->rf_generation);
 653         ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno,
 654                                       new_tree->rf_generation);
 655
 656         spin_lock(&osb->osb_lock);
 657         tree = ocfs2_find_refcount_tree(osb, first_blkno);
 658
 659         /*
 660          * We've just created a new refcount tree in this block.  If
 661          * we found a refcount tree on the ocfs2_super, it must be
 662          * one we just deleted.  We free the old tree before
 663          * inserting the new tree.
 664          */
 665         BUG_ON(tree && tree->rf_generation == new_tree->rf_generation);
 666         if (tree)
 667                 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
 668         ocfs2_insert_refcount_tree(osb, new_tree);
 669         spin_unlock(&osb->osb_lock);
 670         new_tree = NULL;
 671         if (tree)
 672                 ocfs2_refcount_tree_put(tree);
 673
 674 out_commit:
 675         ocfs2_commit_trans(osb, handle);
 676
 677 out:
 678         if (new_tree) {
 679                 ocfs2_metadata_cache_exit(&new_tree->rf_ci);
 680                 kfree(new_tree);
 681         }
 682
 683         brelse(new_bh);
 684         if (meta_ac)
 685                 ocfs2_free_alloc_context(meta_ac);
 686
 687         return ret;
 688 }
 689
 690 static int ocfs2_set_refcount_tree(struct inode *inode,
 691                                    struct buffer_head *di_bh,
 692                                    u64 refcount_loc)
 693 {
 694         int ret;
 695         handle_t *handle = NULL;
 696         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 697         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 698         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 699         struct buffer_head *ref_root_bh = NULL;
 700         struct ocfs2_refcount_block *rb;
 701         struct ocfs2_refcount_tree *ref_tree;
 702
 703         BUG_ON(ocfs2_is_refcount_inode(inode));
 704
 705         ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
 706                                        &ref_tree, &ref_root_bh);
 707         if (ret) {
 708                 mlog_errno(ret);
 709                 return ret;
 710         }
 711
 712         handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS);
 713         if (IS_ERR(handle)) {
 714                 ret = PTR_ERR(handle);
 715                 mlog_errno(ret);
 716                 goto out;
 717         }
 718
 719         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 720                                       OCFS2_JOURNAL_ACCESS_WRITE);
 721         if (ret) {
 722                 mlog_errno(ret);
 723                 goto out_commit;
 724         }
 725
 726         ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh,
 727                                       OCFS2_JOURNAL_ACCESS_WRITE);
 728         if (ret) {
 729                 mlog_errno(ret);
 730                 goto out_commit;
 731         }
 732
 733         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
 734         le32_add_cpu(&rb->rf_count, 1);
 735
 736         ocfs2_journal_dirty(handle, ref_root_bh);
 737
 738         spin_lock(&oi->ip_lock);
 739         oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
 740         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 741         di->i_refcount_loc = cpu_to_le64(refcount_loc);
 742         spin_unlock(&oi->ip_lock);
 743         ocfs2_journal_dirty(handle, di_bh);
 744
 745 out_commit:
 746         ocfs2_commit_trans(osb, handle);
 747 out:
 748         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
 749         brelse(ref_root_bh);
 750
 751         return ret;
 752 }
 753
 754 int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh)
 755 {
 756         int ret, delete_tree = 0;
 757         handle_t *handle = NULL;
 758         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 759         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 760         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 761         struct ocfs2_refcount_block *rb;
 762         struct inode *alloc_inode = NULL;
 763         struct buffer_head *alloc_bh = NULL;
 764         struct buffer_head *blk_bh = NULL;
 765         struct ocfs2_refcount_tree *ref_tree;
 766         int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS;
 767         u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc);
 768         u16 bit = 0;
 769
 770         if (!ocfs2_is_refcount_inode(inode))
 771                 return 0;
 772
 773         BUG_ON(!ref_blkno);
 774         ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh);
 775         if (ret) {
 776                 mlog_errno(ret);
 777                 return ret;
 778         }
 779
 780         rb = (struct ocfs2_refcount_block *)blk_bh->b_data;
 781
 782         /*
 783          * If we are the last user, we need to free the block.
 784          * So lock the allocator ahead.
 785          */
 786         if (le32_to_cpu(rb->rf_count) == 1) {
 787                 blk = le64_to_cpu(rb->rf_blkno);
 788                 bit = le16_to_cpu(rb->rf_suballoc_bit);
 789                 if (rb->rf_suballoc_loc)
 790                         bg_blkno = le64_to_cpu(rb->rf_suballoc_loc);
 791                 else
 792                         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
 793
 794                 alloc_inode = ocfs2_get_system_file_inode(osb,
 795                                         EXTENT_ALLOC_SYSTEM_INODE,
 796                                         le16_to_cpu(rb->rf_suballoc_slot));
 797                 if (!alloc_inode) {
 798                         ret = -ENOMEM;
 799                         mlog_errno(ret);
 800                         goto out;
 801                 }
 802                 inode_lock(alloc_inode);
 803
 804                 ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1);
 805                 if (ret) {
 806                         mlog_errno(ret);
 807                         goto out_mutex;
 808                 }
 809
 810                 credits += OCFS2_SUBALLOC_FREE;
 811         }
 812
 813         handle = ocfs2_start_trans(osb, credits);
 814         if (IS_ERR(handle)) {
 815                 ret = PTR_ERR(handle);
 816                 mlog_errno(ret);
 817                 goto out_unlock;
 818         }
 819
 820         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 821                                       OCFS2_JOURNAL_ACCESS_WRITE);
 822         if (ret) {
 823                 mlog_errno(ret);
 824                 goto out_commit;
 825         }
 826
 827         ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh,
 828                                       OCFS2_JOURNAL_ACCESS_WRITE);
 829         if (ret) {
 830                 mlog_errno(ret);
 831                 goto out_commit;
 832         }
 833
 834         spin_lock(&oi->ip_lock);
 835         oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL;
 836         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
 837         di->i_refcount_loc = 0;
 838         spin_unlock(&oi->ip_lock);
 839         ocfs2_journal_dirty(handle, di_bh);
 840
 841         le32_add_cpu(&rb->rf_count , -1);
 842         ocfs2_journal_dirty(handle, blk_bh);
 843
 844         if (!rb->rf_count) {
 845                 delete_tree = 1;
 846                 ocfs2_erase_refcount_tree_from_list(osb, ref_tree);
 847                 ret = ocfs2_free_suballoc_bits(handle, alloc_inode,
 848                                                alloc_bh, bit, bg_blkno, 1);
 849                 if (ret)
 850                         mlog_errno(ret);
 851         }
 852
 853 out_commit:
 854         ocfs2_commit_trans(osb, handle);
 855 out_unlock:
 856         if (alloc_inode) {
 857                 ocfs2_inode_unlock(alloc_inode, 1);
 858                 brelse(alloc_bh);
 859         }
 860 out_mutex:
 861         if (alloc_inode) {
 862                 inode_unlock(alloc_inode);
 863                 iput(alloc_inode);
 864         }
 865 out:
 866         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
 867         if (delete_tree)
 868                 ocfs2_refcount_tree_put(ref_tree);
 869         brelse(blk_bh);
 870
 871         return ret;
 872 }
 873
 874 static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info *ci,
 875                                           struct buffer_head *ref_leaf_bh,
 876                                           u64 cpos, unsigned int len,
 877                                           struct ocfs2_refcount_rec *ret_rec,
 878                                           int *index)
 879 {
 880         int i = 0;
 881         struct ocfs2_refcount_block *rb =
 882                 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
 883         struct ocfs2_refcount_rec *rec = NULL;
 884
 885         for (; i < le16_to_cpu(rb->rf_records.rl_used); i++) {
 886                 rec = &rb->rf_records.rl_recs[i];
 887
 888                 if (le64_to_cpu(rec->r_cpos) +
 889                     le32_to_cpu(rec->r_clusters) <= cpos)
 890                         continue;
 891                 else if (le64_to_cpu(rec->r_cpos) > cpos)
 892                         break;
 893
 894                 /* ok, cpos fail in this rec. Just return. */
 895                 if (ret_rec)
 896                         *ret_rec = *rec;
 897                 goto out;
 898         }
 899
 900         if (ret_rec) {
 901                 /* We meet with a hole here, so fake the rec. */
 902                 ret_rec->r_cpos = cpu_to_le64(cpos);
 903                 ret_rec->r_refcount = 0;
 904                 if (i < le16_to_cpu(rb->rf_records.rl_used) &&
 905                     le64_to_cpu(rec->r_cpos) < cpos + len)
 906                         ret_rec->r_clusters =
 907                                 cpu_to_le32(le64_to_cpu(rec->r_cpos) - cpos);
 908                 else
 909                         ret_rec->r_clusters = cpu_to_le32(len);
 910         }
 911
 912 out:
 913         *index = i;
 914 }
 915
 916 /*
 917  * Try to remove refcount tree. The mechanism is:
 918  * 1) Check whether i_clusters == 0, if no, exit.
 919  * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
 920  * 3) Check whether we have inline xattr stored outside, if yes, exit.
 921  * 4) Remove the tree.
 922  */
 923 int ocfs2_try_remove_refcount_tree(struct inode *inode,
 924                                    struct buffer_head *di_bh)
 925 {
 926         int ret;
 927         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 928         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 929
 930         down_write(&oi->ip_xattr_sem);
 931         down_write(&oi->ip_alloc_sem);
 932
 933         if (oi->ip_clusters)
 934                 goto out;
 935
 936         if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
 937                 goto out;
 938
 939         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
 940             ocfs2_has_inline_xattr_value_outside(inode, di))
 941                 goto out;
 942
 943         ret = ocfs2_remove_refcount_tree(inode, di_bh);
 944         if (ret)
 945                 mlog_errno(ret);
 946 out:
 947         up_write(&oi->ip_alloc_sem);
 948         up_write(&oi->ip_xattr_sem);
 949         return 0;
 950 }
 951
 952 /*
 953  * Find the end range for a leaf refcount block indicated by
 954  * el->l_recs[index].e_blkno.
 955  */
 956 static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info *ci,
 957                                        struct buffer_head *ref_root_bh,
 958                                        struct ocfs2_extent_block *eb,
 959                                        struct ocfs2_extent_list *el,
 960                                        int index,  u32 *cpos_end)
 961 {
 962         int ret, i, subtree_root;
 963         u32 cpos;
 964         u64 blkno;
 965         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
 966         struct ocfs2_path *left_path = NULL, *right_path = NULL;
 967         struct ocfs2_extent_tree et;
 968         struct ocfs2_extent_list *tmp_el;
 969
 970         if (index < le16_to_cpu(el->l_next_free_rec) - 1) {
 971                 /*
 972                  * We have a extent rec after index, so just use the e_cpos
 973                  * of the next extent rec.
 974                  */
 975                 *cpos_end = le32_to_cpu(el->l_recs[index+1].e_cpos);
 976                 return 0;
 977         }
 978
 979         if (!eb || !eb->h_next_leaf_blk) {
 980                 /*
 981                  * We are the last extent rec, so any high cpos should
 982                  * be stored in this leaf refcount block.
 983                  */
 984                 *cpos_end = UINT_MAX;
 985                 return 0;
 986         }
 987
 988         /*
 989          * If the extent block isn't the last one, we have to find
 990          * the subtree root between this extent block and the next
 991          * leaf extent block and get the corresponding e_cpos from
 992          * the subroot. Otherwise we may corrupt the b-tree.
 993          */
 994         ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
 995
 996         left_path = ocfs2_new_path_from_et(&et);
 997         if (!left_path) {
 998                 ret = -ENOMEM;
 999                 mlog_errno(ret);
1000                 goto out;
1001         }
1002
1003         cpos = le32_to_cpu(eb->h_list.l_recs[index].e_cpos);
1004         ret = ocfs2_find_path(ci, left_path, cpos);
1005         if (ret) {
1006                 mlog_errno(ret);
1007                 goto out;
1008         }
1009
1010         right_path = ocfs2_new_path_from_path(left_path);
1011         if (!right_path) {
1012                 ret = -ENOMEM;
1013                 mlog_errno(ret);
1014                 goto out;
1015         }
1016
1017         ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, &cpos);
1018         if (ret) {
1019                 mlog_errno(ret);
1020                 goto out;
1021         }
1022
1023         ret = ocfs2_find_path(ci, right_path, cpos);
1024         if (ret) {
1025                 mlog_errno(ret);
1026                 goto out;
1027         }
1028
1029         subtree_root = ocfs2_find_subtree_root(&et, left_path,
1030                                                right_path);
1031
1032         tmp_el = left_path->p_node[subtree_root].el;
1033         blkno = left_path->p_node[subtree_root+1].bh->b_blocknr;
1034         for (i = 0; i < le16_to_cpu(tmp_el->l_next_free_rec); i++) {
1035                 if (le64_to_cpu(tmp_el->l_recs[i].e_blkno) == blkno) {
1036                         *cpos_end = le32_to_cpu(tmp_el->l_recs[i+1].e_cpos);
1037                         break;
1038                 }
1039         }
1040
1041         BUG_ON(i == le16_to_cpu(tmp_el->l_next_free_rec));
1042
1043 out:
1044         ocfs2_free_path(left_path);
1045         ocfs2_free_path(right_path);
1046         return ret;
1047 }
1048
1049 /*
1050  * Given a cpos and len, try to find the refcount record which contains cpos.
1051  * 1. If cpos can be found in one refcount record, return the record.
1052  * 2. If cpos can't be found, return a fake record which start from cpos
1053  *    and end at a small value between cpos+len and start of the next record.
1054  *    This fake record has r_refcount = 0.
1055  */
1056 static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
1057                                   struct buffer_head *ref_root_bh,
1058                                   u64 cpos, unsigned int len,
1059                                   struct ocfs2_refcount_rec *ret_rec,
1060                                   int *index,
1061                                   struct buffer_head **ret_bh)
1062 {
1063         int ret = 0, i, found;
1064         u32 low_cpos, cpos_end;
1065         struct ocfs2_extent_list *el;
1066         struct ocfs2_extent_rec *rec = NULL;
1067         struct ocfs2_extent_block *eb = NULL;
1068         struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
1069         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1070         struct ocfs2_refcount_block *rb =
1071                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1072
1073         if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)) {
1074                 ocfs2_find_refcount_rec_in_rl(ci, ref_root_bh, cpos, len,
1075                                               ret_rec, index);
1076                 *ret_bh = ref_root_bh;
1077                 get_bh(ref_root_bh);
1078                 return 0;
1079         }
1080
1081         el = &rb->rf_list;
1082         low_cpos = cpos & OCFS2_32BIT_POS_MASK;
1083
1084         if (el->l_tree_depth) {
1085                 ret = ocfs2_find_leaf(ci, el, low_cpos, &eb_bh);
1086                 if (ret) {
1087                         mlog_errno(ret);
1088                         goto out;
1089                 }
1090
1091                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1092                 el = &eb->h_list;
1093
1094                 if (el->l_tree_depth) {
1095                         ret = ocfs2_error(sb,
1096                                           "refcount tree %llu has non zero tree depth in leaf btree tree block %llu\n",
1097                                           (unsigned long long)ocfs2_metadata_cache_owner(ci),
1098                                           (unsigned long long)eb_bh->b_blocknr);
1099                         goto out;
1100                 }
1101         }
1102
1103         found = 0;
1104         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1105                 rec = &el->l_recs[i];
1106
1107                 if (le32_to_cpu(rec->e_cpos) <= low_cpos) {
1108                         found = 1;
1109                         break;
1110                 }
1111         }
1112
1113         if (found) {
1114                 ret = ocfs2_get_refcount_cpos_end(ci, ref_root_bh,
1115                                                   eb, el, i, &cpos_end);
1116                 if (ret) {
1117                         mlog_errno(ret);
1118                         goto out;
1119                 }
1120
1121                 if (cpos_end < low_cpos + len)
1122                         len = cpos_end - low_cpos;
1123         }
1124
1125         ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
1126                                         &ref_leaf_bh);
1127         if (ret) {
1128                 mlog_errno(ret);
1129                 goto out;
1130         }
1131
1132         ocfs2_find_refcount_rec_in_rl(ci, ref_leaf_bh, cpos, len,
1133                                       ret_rec, index);
1134         *ret_bh = ref_leaf_bh;
1135 out:
1136         brelse(eb_bh);
1137         return ret;
1138 }
1139
1140 enum ocfs2_ref_rec_contig {
1141         REF_CONTIG_NONE = 0,
1142         REF_CONTIG_LEFT,
1143         REF_CONTIG_RIGHT,
1144         REF_CONTIG_LEFTRIGHT,
1145 };
1146
1147 static enum ocfs2_ref_rec_contig
1148         ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block *rb,
1149                                     int index)
1150 {
1151         if ((rb->rf_records.rl_recs[index].r_refcount ==
1152             rb->rf_records.rl_recs[index + 1].r_refcount) &&
1153             (le64_to_cpu(rb->rf_records.rl_recs[index].r_cpos) +
1154             le32_to_cpu(rb->rf_records.rl_recs[index].r_clusters) ==
1155             le64_to_cpu(rb->rf_records.rl_recs[index + 1].r_cpos)))
1156                 return REF_CONTIG_RIGHT;
1157
1158         return REF_CONTIG_NONE;
1159 }
1160
1161 static enum ocfs2_ref_rec_contig
1162         ocfs2_refcount_rec_contig(struct ocfs2_refcount_block *rb,
1163                                   int index)
1164 {
1165         enum ocfs2_ref_rec_contig ret = REF_CONTIG_NONE;
1166
1167         if (index < le16_to_cpu(rb->rf_records.rl_used) - 1)
1168                 ret = ocfs2_refcount_rec_adjacent(rb, index);
1169
1170         if (index > 0) {
1171                 enum ocfs2_ref_rec_contig tmp;
1172
1173                 tmp = ocfs2_refcount_rec_adjacent(rb, index - 1);
1174
1175                 if (tmp == REF_CONTIG_RIGHT) {
1176                         if (ret == REF_CONTIG_RIGHT)
1177                                 ret = REF_CONTIG_LEFTRIGHT;
1178                         else
1179                                 ret = REF_CONTIG_LEFT;
1180                 }
1181         }
1182
1183         return ret;
1184 }
1185
1186 static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block *rb,
1187                                            int index)
1188 {
1189         BUG_ON(rb->rf_records.rl_recs[index].r_refcount !=
1190                rb->rf_records.rl_recs[index+1].r_refcount);
1191
1192         le32_add_cpu(&rb->rf_records.rl_recs[index].r_clusters,
1193                      le32_to_cpu(rb->rf_records.rl_recs[index+1].r_clusters));
1194
1195         if (index < le16_to_cpu(rb->rf_records.rl_used) - 2)
1196                 memmove(&rb->rf_records.rl_recs[index + 1],
1197                         &rb->rf_records.rl_recs[index + 2],
1198                         sizeof(struct ocfs2_refcount_rec) *
1199                         (le16_to_cpu(rb->rf_records.rl_used) - index - 2));
1200
1201         memset(&rb->rf_records.rl_recs[le16_to_cpu(rb->rf_records.rl_used) - 1],
1202                0, sizeof(struct ocfs2_refcount_rec));
1203         le16_add_cpu(&rb->rf_records.rl_used, -1);
1204 }
1205
1206 /*
1207  * Merge the refcount rec if we are contiguous with the adjacent recs.
1208  */
1209 static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
1210                                      int index)
1211 {
1212         enum ocfs2_ref_rec_contig contig =
1213                                 ocfs2_refcount_rec_contig(rb, index);
1214
1215         if (contig == REF_CONTIG_NONE)
1216                 return;
1217
1218         if (contig == REF_CONTIG_LEFT || contig == REF_CONTIG_LEFTRIGHT) {
1219                 BUG_ON(index == 0);
1220                 index--;
1221         }
1222
1223         ocfs2_rotate_refcount_rec_left(rb, index);
1224
1225         if (contig == REF_CONTIG_LEFTRIGHT)
1226                 ocfs2_rotate_refcount_rec_left(rb, index);
1227 }
1228
1229 /*
1230  * Change the refcount indexed by "index" in ref_bh.
1231  * If refcount reaches 0, remove it.
1232  */
1233 static int ocfs2_change_refcount_rec(handle_t *handle,
1234                                      struct ocfs2_caching_info *ci,
1235                                      struct buffer_head *ref_leaf_bh,
1236                                      int index, int merge, int change)
1237 {
1238         int ret;
1239         struct ocfs2_refcount_block *rb =
1240                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1241         struct ocfs2_refcount_list *rl = &rb->rf_records;
1242         struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
1243
1244         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1245                                       OCFS2_JOURNAL_ACCESS_WRITE);
1246         if (ret) {
1247                 mlog_errno(ret);
1248                 goto out;
1249         }
1250
1251         trace_ocfs2_change_refcount_rec(
1252                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1253                 index, le32_to_cpu(rec->r_refcount), change);
1254         le32_add_cpu(&rec->r_refcount, change);
1255
1256         if (!rec->r_refcount) {
1257                 if (index != le16_to_cpu(rl->rl_used) - 1) {
1258                         memmove(rec, rec + 1,
1259                                 (le16_to_cpu(rl->rl_used) - index - 1) *
1260                                 sizeof(struct ocfs2_refcount_rec));
1261                         memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
1262                                0, sizeof(struct ocfs2_refcount_rec));
1263                 }
1264
1265                 le16_add_cpu(&rl->rl_used, -1);
1266         } else if (merge)
1267                 ocfs2_refcount_rec_merge(rb, index);
1268
1269         ocfs2_journal_dirty(handle, ref_leaf_bh);
1270 out:
1271         return ret;
1272 }
1273
1274 static int ocfs2_expand_inline_ref_root(handle_t *handle,
1275                                         struct ocfs2_caching_info *ci,
1276                                         struct buffer_head *ref_root_bh,
1277                                         struct buffer_head **ref_leaf_bh,
1278                                         struct ocfs2_alloc_context *meta_ac)
1279 {
1280         int ret;
1281         u16 suballoc_bit_start;
1282         u32 num_got;
1283         u64 suballoc_loc, blkno;
1284         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1285         struct buffer_head *new_bh = NULL;
1286         struct ocfs2_refcount_block *new_rb;
1287         struct ocfs2_refcount_block *root_rb =
1288                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1289
1290         ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1291                                       OCFS2_JOURNAL_ACCESS_WRITE);
1292         if (ret) {
1293                 mlog_errno(ret);
1294                 goto out;
1295         }
1296
1297         ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1298                                    &suballoc_bit_start, &num_got,
1299                                    &blkno);
1300         if (ret) {
1301                 mlog_errno(ret);
1302                 goto out;
1303         }
1304
1305         new_bh = sb_getblk(sb, blkno);
1306         if (new_bh == NULL) {
1307                 ret = -ENOMEM;
1308                 mlog_errno(ret);
1309                 goto out;
1310         }
1311         ocfs2_set_new_buffer_uptodate(ci, new_bh);
1312
1313         ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1314                                       OCFS2_JOURNAL_ACCESS_CREATE);
1315         if (ret) {
1316                 mlog_errno(ret);
1317                 goto out;
1318         }
1319
1320         /*
1321          * Initialize ocfs2_refcount_block.
1322          * It should contain the same information as the old root.
1323          * so just memcpy it and change the corresponding field.
1324          */
1325         memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
1326
1327         new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1328         new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1329         new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1330         new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1331         new_rb->rf_blkno = cpu_to_le64(blkno);
1332         new_rb->rf_cpos = cpu_to_le32(0);
1333         new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1334         new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1335         ocfs2_journal_dirty(handle, new_bh);
1336
1337         /* Now change the root. */
1338         memset(&root_rb->rf_list, 0, sb->s_blocksize -
1339                offsetof(struct ocfs2_refcount_block, rf_list));
1340         root_rb->rf_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_rb(sb));
1341         root_rb->rf_clusters = cpu_to_le32(1);
1342         root_rb->rf_list.l_next_free_rec = cpu_to_le16(1);
1343         root_rb->rf_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
1344         root_rb->rf_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
1345         root_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_TREE_FL);
1346
1347         ocfs2_journal_dirty(handle, ref_root_bh);
1348
1349         trace_ocfs2_expand_inline_ref_root((unsigned long long)blkno,
1350                 le16_to_cpu(new_rb->rf_records.rl_used));
1351
1352         *ref_leaf_bh = new_bh;
1353         new_bh = NULL;
1354 out:
1355         brelse(new_bh);
1356         return ret;
1357 }
1358
1359 static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec *prev,
1360                                            struct ocfs2_refcount_rec *next)
1361 {
1362         if (ocfs2_get_ref_rec_low_cpos(prev) + le32_to_cpu(prev->r_clusters) <=
1363                 ocfs2_get_ref_rec_low_cpos(next))
1364                 return 1;
1365
1366         return 0;
1367 }
1368
1369 static int cmp_refcount_rec_by_low_cpos(const void *a, const void *b)
1370 {
1371         const struct ocfs2_refcount_rec *l = a, *r = b;
1372         u32 l_cpos = ocfs2_get_ref_rec_low_cpos(l);
1373         u32 r_cpos = ocfs2_get_ref_rec_low_cpos(r);
1374
1375         if (l_cpos > r_cpos)
1376                 return 1;
1377         if (l_cpos < r_cpos)
1378                 return -1;
1379         return 0;
1380 }
1381
1382 static int cmp_refcount_rec_by_cpos(const void *a, const void *b)
1383 {
1384         const struct ocfs2_refcount_rec *l = a, *r = b;
1385         u64 l_cpos = le64_to_cpu(l->r_cpos);
1386         u64 r_cpos = le64_to_cpu(r->r_cpos);
1387
1388         if (l_cpos > r_cpos)
1389                 return 1;
1390         if (l_cpos < r_cpos)
1391                 return -1;
1392         return 0;
1393 }
1394
1395 static void swap_refcount_rec(void *a, void *b, int size)
1396 {
1397         struct ocfs2_refcount_rec *l = a, *r = b;
1398
1399         swap(*l, *r);
1400 }
1401
1402 /*
1403  * The refcount cpos are ordered by their 64bit cpos,
1404  * But we will use the low 32 bit to be the e_cpos in the b-tree.
1405  * So we need to make sure that this pos isn't intersected with others.
1406  *
1407  * Note: The refcount block is already sorted by their low 32 bit cpos,
1408  *       So just try the middle pos first, and we will exit when we find
1409  *       the good position.
1410  */
1411 static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
1412                                          u32 *split_pos, int *split_index)
1413 {
1414         int num_used = le16_to_cpu(rl->rl_used);
1415         int delta, middle = num_used / 2;
1416
1417         for (delta = 0; delta < middle; delta++) {
1418                 /* Let's check delta earlier than middle */
1419                 if (ocfs2_refcount_rec_no_intersect(
1420                                         &rl->rl_recs[middle - delta - 1],
1421                                         &rl->rl_recs[middle - delta])) {
1422                         *split_index = middle - delta;
1423                         break;
1424                 }
1425
1426                 /* For even counts, don't walk off the end */
1427                 if ((middle + delta + 1) == num_used)
1428                         continue;
1429
1430                 /* Now try delta past middle */
1431                 if (ocfs2_refcount_rec_no_intersect(
1432                                         &rl->rl_recs[middle + delta],
1433                                         &rl->rl_recs[middle + delta + 1])) {
1434                         *split_index = middle + delta + 1;
1435                         break;
1436                 }
1437         }
1438
1439         if (delta >= middle)
1440                 return -ENOSPC;
1441
1442         *split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
1443         return 0;
1444 }
1445
1446 static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
1447                                             struct buffer_head *new_bh,
1448                                             u32 *split_cpos)
1449 {
1450         int split_index = 0, num_moved, ret;
1451         u32 cpos = 0;
1452         struct ocfs2_refcount_block *rb =
1453                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1454         struct ocfs2_refcount_list *rl = &rb->rf_records;
1455         struct ocfs2_refcount_block *new_rb =
1456                         (struct ocfs2_refcount_block *)new_bh->b_data;
1457         struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
1458
1459         trace_ocfs2_divide_leaf_refcount_block(
1460                 (unsigned long long)ref_leaf_bh->b_blocknr,
1461                 le16_to_cpu(rl->rl_count), le16_to_cpu(rl->rl_used));
1462
1463         /*
1464          * XXX: Improvement later.
1465          * If we know all the high 32 bit cpos is the same, no need to sort.
1466          *
1467          * In order to make the whole process safe, we do:
1468          * 1. sort the entries by their low 32 bit cpos first so that we can
1469          *    find the split cpos easily.
1470          * 2. call ocfs2_insert_extent to insert the new refcount block.
1471          * 3. move the refcount rec to the new block.
1472          * 4. sort the entries by their 64 bit cpos.
1473          * 5. dirty the new_rb and rb.
1474          */
1475         sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1476              sizeof(struct ocfs2_refcount_rec),
1477              cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
1478
1479         ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
1480         if (ret) {
1481                 mlog_errno(ret);
1482                 return ret;
1483         }
1484
1485         new_rb->rf_cpos = cpu_to_le32(cpos);
1486
1487         /* move refcount records starting from split_index to the new block. */
1488         num_moved = le16_to_cpu(rl->rl_used) - split_index;
1489         memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
1490                num_moved * sizeof(struct ocfs2_refcount_rec));
1491
1492         /*ok, remove the entries we just moved over to the other block. */
1493         memset(&rl->rl_recs[split_index], 0,
1494                num_moved * sizeof(struct ocfs2_refcount_rec));
1495
1496         /* change old and new rl_used accordingly. */
1497         le16_add_cpu(&rl->rl_used, -num_moved);
1498         new_rl->rl_used = cpu_to_le16(num_moved);
1499
1500         sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1501              sizeof(struct ocfs2_refcount_rec),
1502              cmp_refcount_rec_by_cpos, swap_refcount_rec);
1503
1504         sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
1505              sizeof(struct ocfs2_refcount_rec),
1506              cmp_refcount_rec_by_cpos, swap_refcount_rec);
1507
1508         *split_cpos = cpos;
1509         return 0;
1510 }
1511
1512 static int ocfs2_new_leaf_refcount_block(handle_t *handle,
1513                                          struct ocfs2_caching_info *ci,
1514                                          struct buffer_head *ref_root_bh,
1515                                          struct buffer_head *ref_leaf_bh,
1516                                          struct ocfs2_alloc_context *meta_ac)
1517 {
1518         int ret;
1519         u16 suballoc_bit_start;
1520         u32 num_got, new_cpos;
1521         u64 suballoc_loc, blkno;
1522         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1523         struct ocfs2_refcount_block *root_rb =
1524                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1525         struct buffer_head *new_bh = NULL;
1526         struct ocfs2_refcount_block *new_rb;
1527         struct ocfs2_extent_tree ref_et;
1528
1529         BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
1530
1531         ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1532                                       OCFS2_JOURNAL_ACCESS_WRITE);
1533         if (ret) {
1534                 mlog_errno(ret);
1535                 goto out;
1536         }
1537
1538         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1539                                       OCFS2_JOURNAL_ACCESS_WRITE);
1540         if (ret) {
1541                 mlog_errno(ret);
1542                 goto out;
1543         }
1544
1545         ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1546                                    &suballoc_bit_start, &num_got,
1547                                    &blkno);
1548         if (ret) {
1549                 mlog_errno(ret);
1550                 goto out;
1551         }
1552
1553         new_bh = sb_getblk(sb, blkno);
1554         if (new_bh == NULL) {
1555                 ret = -ENOMEM;
1556                 mlog_errno(ret);
1557                 goto out;
1558         }
1559         ocfs2_set_new_buffer_uptodate(ci, new_bh);
1560
1561         ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1562                                       OCFS2_JOURNAL_ACCESS_CREATE);
1563         if (ret) {
1564                 mlog_errno(ret);
1565                 goto out;
1566         }
1567
1568         /* Initialize ocfs2_refcount_block. */
1569         new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1570         memset(new_rb, 0, sb->s_blocksize);
1571         strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
1572         new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1573         new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1574         new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1575         new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1576         new_rb->rf_blkno = cpu_to_le64(blkno);
1577         new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1578         new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1579         new_rb->rf_records.rl_count =
1580                                 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
1581         new_rb->rf_generation = root_rb->rf_generation;
1582
1583         ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
1584         if (ret) {
1585                 mlog_errno(ret);
1586                 goto out;
1587         }
1588
1589         ocfs2_journal_dirty(handle, ref_leaf_bh);
1590         ocfs2_journal_dirty(handle, new_bh);
1591
1592         ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
1593
1594         trace_ocfs2_new_leaf_refcount_block(
1595                         (unsigned long long)new_bh->b_blocknr, new_cpos);
1596
1597         /* Insert the new leaf block with the specific offset cpos. */
1598         ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
1599                                   1, 0, meta_ac);
1600         if (ret)
1601                 mlog_errno(ret);
1602
1603 out:
1604         brelse(new_bh);
1605         return ret;
1606 }
1607
1608 static int ocfs2_expand_refcount_tree(handle_t *handle,
1609                                       struct ocfs2_caching_info *ci,
1610                                       struct buffer_head *ref_root_bh,
1611                                       struct buffer_head *ref_leaf_bh,
1612                                       struct ocfs2_alloc_context *meta_ac)
1613 {
1614         int ret;
1615         struct buffer_head *expand_bh = NULL;
1616
1617         if (ref_root_bh == ref_leaf_bh) {
1618                 /*
1619                  * the old root bh hasn't been expanded to a b-tree,
1620                  * so expand it first.
1621                  */
1622                 ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
1623                                                    &expand_bh, meta_ac);
1624                 if (ret) {
1625                         mlog_errno(ret);
1626                         goto out;
1627                 }
1628         } else {
1629                 expand_bh = ref_leaf_bh;
1630                 get_bh(expand_bh);
1631         }
1632
1633
1634         /* Now add a new refcount block into the tree.*/
1635         ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
1636                                             expand_bh, meta_ac);
1637         if (ret)
1638                 mlog_errno(ret);
1639 out:
1640         brelse(expand_bh);
1641         return ret;
1642 }
1643
1644 /*
1645  * Adjust the extent rec in b-tree representing ref_leaf_bh.
1646  *
1647  * Only called when we have inserted a new refcount rec at index 0
1648  * which means ocfs2_extent_rec.e_cpos may need some change.
1649  */
1650 static int ocfs2_adjust_refcount_rec(handle_t *handle,
1651                                      struct ocfs2_caching_info *ci,
1652                                      struct buffer_head *ref_root_bh,
1653                                      struct buffer_head *ref_leaf_bh,
1654                                      struct ocfs2_refcount_rec *rec)
1655 {
1656         int ret = 0, i;
1657         u32 new_cpos, old_cpos;
1658         struct ocfs2_path *path = NULL;
1659         struct ocfs2_extent_tree et;
1660         struct ocfs2_refcount_block *rb =
1661                 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1662         struct ocfs2_extent_list *el;
1663
1664         if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
1665                 goto out;
1666
1667         rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1668         old_cpos = le32_to_cpu(rb->rf_cpos);
1669         new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
1670         if (old_cpos <= new_cpos)
1671                 goto out;
1672
1673         ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1674
1675         path = ocfs2_new_path_from_et(&et);
1676         if (!path) {
1677                 ret = -ENOMEM;
1678                 mlog_errno(ret);
1679                 goto out;
1680         }
1681
1682         ret = ocfs2_find_path(ci, path, old_cpos);
1683         if (ret) {
1684                 mlog_errno(ret);
1685                 goto out;
1686         }
1687
1688         /*
1689          * 2 more credits, one for the leaf refcount block, one for
1690          * the extent block contains the extent rec.
1691          */
1692         ret = ocfs2_extend_trans(handle, 2);
1693         if (ret < 0) {
1694                 mlog_errno(ret);
1695                 goto out;
1696         }
1697
1698         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1699                                       OCFS2_JOURNAL_ACCESS_WRITE);
1700         if (ret < 0) {
1701                 mlog_errno(ret);
1702                 goto out;
1703         }
1704
1705         ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
1706                                       OCFS2_JOURNAL_ACCESS_WRITE);
1707         if (ret < 0) {
1708                 mlog_errno(ret);
1709                 goto out;
1710         }
1711
1712         /* change the leaf extent block first. */
1713         el = path_leaf_el(path);
1714
1715         for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
1716                 if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
1717                         break;
1718
1719         BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
1720
1721         el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1722
1723         /* change the r_cpos in the leaf block. */
1724         rb->rf_cpos = cpu_to_le32(new_cpos);
1725
1726         ocfs2_journal_dirty(handle, path_leaf_bh(path));
1727         ocfs2_journal_dirty(handle, ref_leaf_bh);
1728
1729 out:
1730         ocfs2_free_path(path);
1731         return ret;
1732 }
1733
1734 static int ocfs2_insert_refcount_rec(handle_t *handle,
1735                                      struct ocfs2_caching_info *ci,
1736                                      struct buffer_head *ref_root_bh,
1737                                      struct buffer_head *ref_leaf_bh,
1738                                      struct ocfs2_refcount_rec *rec,
1739                                      int index, int merge,
1740                                      struct ocfs2_alloc_context *meta_ac)
1741 {
1742         int ret;
1743         struct ocfs2_refcount_block *rb =
1744                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1745         struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1746         struct buffer_head *new_bh = NULL;
1747
1748         BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1749
1750         if (rf_list->rl_used == rf_list->rl_count) {
1751                 u64 cpos = le64_to_cpu(rec->r_cpos);
1752                 u32 len = le32_to_cpu(rec->r_clusters);
1753
1754                 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1755                                                  ref_leaf_bh, meta_ac);
1756                 if (ret) {
1757                         mlog_errno(ret);
1758                         goto out;
1759                 }
1760
1761                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1762                                              cpos, len, NULL, &index,
1763                                              &new_bh);
1764                 if (ret) {
1765                         mlog_errno(ret);
1766                         goto out;
1767                 }
1768
1769                 ref_leaf_bh = new_bh;
1770                 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1771                 rf_list = &rb->rf_records;
1772         }
1773
1774         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1775                                       OCFS2_JOURNAL_ACCESS_WRITE);
1776         if (ret) {
1777                 mlog_errno(ret);
1778                 goto out;
1779         }
1780
1781         if (index < le16_to_cpu(rf_list->rl_used))
1782                 memmove(&rf_list->rl_recs[index + 1],
1783                         &rf_list->rl_recs[index],
1784                         (le16_to_cpu(rf_list->rl_used) - index) *
1785                          sizeof(struct ocfs2_refcount_rec));
1786
1787         trace_ocfs2_insert_refcount_rec(
1788                 (unsigned long long)ref_leaf_bh->b_blocknr, index,
1789                 (unsigned long long)le64_to_cpu(rec->r_cpos),
1790                 le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount));
1791
1792         rf_list->rl_recs[index] = *rec;
1793
1794         le16_add_cpu(&rf_list->rl_used, 1);
1795
1796         if (merge)
1797                 ocfs2_refcount_rec_merge(rb, index);
1798
1799         ocfs2_journal_dirty(handle, ref_leaf_bh);
1800
1801         if (index == 0) {
1802                 ret = ocfs2_adjust_refcount_rec(handle, ci,
1803                                                 ref_root_bh,
1804                                                 ref_leaf_bh, rec);
1805                 if (ret)
1806                         mlog_errno(ret);
1807         }
1808 out:
1809         brelse(new_bh);
1810         return ret;
1811 }
1812
1813 /*
1814  * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1815  * This is much simple than our b-tree code.
1816  * split_rec is the new refcount rec we want to insert.
1817  * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1818  * increase refcount or decrease a refcount to non-zero).
1819  * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1820  * rec( in case we decrease a refcount to zero).
1821  */
1822 static int ocfs2_split_refcount_rec(handle_t *handle,
1823                                     struct ocfs2_caching_info *ci,
1824                                     struct buffer_head *ref_root_bh,
1825                                     struct buffer_head *ref_leaf_bh,
1826                                     struct ocfs2_refcount_rec *split_rec,
1827                                     int index, int merge,
1828                                     struct ocfs2_alloc_context *meta_ac,
1829                                     struct ocfs2_cached_dealloc_ctxt *dealloc)
1830 {
1831         int ret, recs_need;
1832         u32 len;
1833         struct ocfs2_refcount_block *rb =
1834                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1835         struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1836         struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
1837         struct ocfs2_refcount_rec *tail_rec = NULL;
1838         struct buffer_head *new_bh = NULL;
1839
1840         BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1841
1842         trace_ocfs2_split_refcount_rec(le64_to_cpu(orig_rec->r_cpos),
1843                 le32_to_cpu(orig_rec->r_clusters),
1844                 le32_to_cpu(orig_rec->r_refcount),
1845                 le64_to_cpu(split_rec->r_cpos),
1846                 le32_to_cpu(split_rec->r_clusters),
1847                 le32_to_cpu(split_rec->r_refcount));
1848
1849         /*
1850          * If we just need to split the header or tail clusters,
1851          * no more recs are needed, just split is OK.
1852          * Otherwise we at least need one new recs.
1853          */
1854         if (!split_rec->r_refcount &&
1855             (split_rec->r_cpos == orig_rec->r_cpos ||
1856              le64_to_cpu(split_rec->r_cpos) +
1857              le32_to_cpu(split_rec->r_clusters) ==
1858              le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1859                 recs_need = 0;
1860         else
1861                 recs_need = 1;
1862
1863         /*
1864          * We need one more rec if we split in the middle and the new rec have
1865          * some refcount in it.
1866          */
1867         if (split_rec->r_refcount &&
1868             (split_rec->r_cpos != orig_rec->r_cpos &&
1869              le64_to_cpu(split_rec->r_cpos) +
1870              le32_to_cpu(split_rec->r_clusters) !=
1871              le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1872                 recs_need++;
1873
1874         /* If the leaf block don't have enough record, expand it. */
1875         if (le16_to_cpu(rf_list->rl_used) + recs_need >
1876                                          le16_to_cpu(rf_list->rl_count)) {
1877                 struct ocfs2_refcount_rec tmp_rec;
1878                 u64 cpos = le64_to_cpu(orig_rec->r_cpos);
1879                 len = le32_to_cpu(orig_rec->r_clusters);
1880                 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1881                                                  ref_leaf_bh, meta_ac);
1882                 if (ret) {
1883                         mlog_errno(ret);
1884                         goto out;
1885                 }
1886
1887                 /*
1888                  * We have to re-get it since now cpos may be moved to
1889                  * another leaf block.
1890                  */
1891                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1892                                              cpos, len, &tmp_rec, &index,
1893                                              &new_bh);
1894                 if (ret) {
1895                         mlog_errno(ret);
1896                         goto out;
1897                 }
1898
1899                 ref_leaf_bh = new_bh;
1900                 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1901                 rf_list = &rb->rf_records;
1902                 orig_rec = &rf_list->rl_recs[index];
1903         }
1904
1905         ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1906                                       OCFS2_JOURNAL_ACCESS_WRITE);
1907         if (ret) {
1908                 mlog_errno(ret);
1909                 goto out;
1910         }
1911
1912         /*
1913          * We have calculated out how many new records we need and store
1914          * in recs_need, so spare enough space first by moving the records
1915          * after "index" to the end.
1916          */
1917         if (index != le16_to_cpu(rf_list->rl_used) - 1)
1918                 memmove(&rf_list->rl_recs[index + 1 + recs_need],
1919                         &rf_list->rl_recs[index + 1],
1920                         (le16_to_cpu(rf_list->rl_used) - index - 1) *
1921                          sizeof(struct ocfs2_refcount_rec));
1922
1923         len = (le64_to_cpu(orig_rec->r_cpos) +
1924               le32_to_cpu(orig_rec->r_clusters)) -
1925               (le64_to_cpu(split_rec->r_cpos) +
1926               le32_to_cpu(split_rec->r_clusters));
1927
1928         /*
1929          * If we have "len", the we will split in the tail and move it
1930          * to the end of the space we have just spared.
1931          */
1932         if (len) {
1933                 tail_rec = &rf_list->rl_recs[index + recs_need];
1934
1935                 memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
1936                 le64_add_cpu(&tail_rec->r_cpos,
1937                              le32_to_cpu(tail_rec->r_clusters) - len);
1938                 tail_rec->r_clusters = cpu_to_le32(len);
1939         }
1940
1941         /*
1942          * If the split pos isn't the same as the original one, we need to
1943          * split in the head.
1944          *
1945          * Note: We have the chance that split_rec.r_refcount = 0,
1946          * recs_need = 0 and len > 0, which means we just cut the head from
1947          * the orig_rec and in that case we have done some modification in
1948          * orig_rec above, so the check for r_cpos is faked.
1949          */
1950         if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
1951                 len = le64_to_cpu(split_rec->r_cpos) -
1952                       le64_to_cpu(orig_rec->r_cpos);
1953                 orig_rec->r_clusters = cpu_to_le32(len);
1954                 index++;
1955         }
1956
1957         le16_add_cpu(&rf_list->rl_used, recs_need);
1958
1959         if (split_rec->r_refcount) {
1960                 rf_list->rl_recs[index] = *split_rec;
1961                 trace_ocfs2_split_refcount_rec_insert(
1962                         (unsigned long long)ref_leaf_bh->b_blocknr, index,
1963                         (unsigned long long)le64_to_cpu(split_rec->r_cpos),
1964                         le32_to_cpu(split_rec->r_clusters),
1965                         le32_to_cpu(split_rec->r_refcount));
1966
1967                 if (merge)
1968                         ocfs2_refcount_rec_merge(rb, index);
1969         }
1970
1971         ocfs2_journal_dirty(handle, ref_leaf_bh);
1972
1973 out:
1974         brelse(new_bh);
1975         return ret;
1976 }
1977
1978 static int __ocfs2_increase_refcount(handle_t *handle,
1979                                      struct ocfs2_caching_info *ci,
1980                                      struct buffer_head *ref_root_bh,
1981                                      u64 cpos, u32 len, int merge,
1982                                      struct ocfs2_alloc_context *meta_ac,
1983                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
1984 {
1985         int ret = 0, index;
1986         struct buffer_head *ref_leaf_bh = NULL;
1987         struct ocfs2_refcount_rec rec;
1988         unsigned int set_len = 0;
1989
1990         trace_ocfs2_increase_refcount_begin(
1991              (unsigned long long)ocfs2_metadata_cache_owner(ci),
1992              (unsigned long long)cpos, len);
1993
1994         while (len) {
1995                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1996                                              cpos, len, &rec, &index,
1997                                              &ref_leaf_bh);
1998                 if (ret) {
1999                         mlog_errno(ret);
2000                         goto out;
2001                 }
2002
2003                 set_len = le32_to_cpu(rec.r_clusters);
2004
2005                 /*
2006                  * Here we may meet with 3 situations:
2007                  *
2008                  * 1. If we find an already existing record, and the length
2009                  *    is the same, cool, we just need to increase the r_refcount
2010                  *    and it is OK.
2011                  * 2. If we find a hole, just insert it with r_refcount = 1.
2012                  * 3. If we are in the middle of one extent record, split
2013                  *    it.
2014                  */
2015                 if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
2016                     set_len <= len) {
2017                         trace_ocfs2_increase_refcount_change(
2018                                 (unsigned long long)cpos, set_len,
2019                                 le32_to_cpu(rec.r_refcount));
2020                         ret = ocfs2_change_refcount_rec(handle, ci,
2021                                                         ref_leaf_bh, index,
2022                                                         merge, 1);
2023                         if (ret) {
2024                                 mlog_errno(ret);
2025                                 goto out;
2026                         }
2027                 } else if (!rec.r_refcount) {
2028                         rec.r_refcount = cpu_to_le32(1);
2029
2030                         trace_ocfs2_increase_refcount_insert(
2031                              (unsigned long long)le64_to_cpu(rec.r_cpos),
2032                              set_len);
2033                         ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
2034                                                         ref_leaf_bh,
2035                                                         &rec, index,
2036                                                         merge, meta_ac);
2037                         if (ret) {
2038                                 mlog_errno(ret);
2039                                 goto out;
2040                         }
2041                 } else  {
2042                         set_len = min((u64)(cpos + len),
2043                                       le64_to_cpu(rec.r_cpos) + set_len) - cpos;
2044                         rec.r_cpos = cpu_to_le64(cpos);
2045                         rec.r_clusters = cpu_to_le32(set_len);
2046                         le32_add_cpu(&rec.r_refcount, 1);
2047
2048                         trace_ocfs2_increase_refcount_split(
2049                              (unsigned long long)le64_to_cpu(rec.r_cpos),
2050                              set_len, le32_to_cpu(rec.r_refcount));
2051                         ret = ocfs2_split_refcount_rec(handle, ci,
2052                                                        ref_root_bh, ref_leaf_bh,
2053                                                        &rec, index, merge,
2054                                                        meta_ac, dealloc);
2055                         if (ret) {
2056                                 mlog_errno(ret);
2057                                 goto out;
2058                         }
2059                 }
2060
2061                 cpos += set_len;
2062                 len -= set_len;
2063                 brelse(ref_leaf_bh);
2064                 ref_leaf_bh = NULL;
2065         }
2066
2067 out:
2068         brelse(ref_leaf_bh);
2069         return ret;
2070 }
2071
2072 static int ocfs2_remove_refcount_extent(handle_t *handle,
2073                                 struct ocfs2_caching_info *ci,
2074                                 struct buffer_head *ref_root_bh,
2075                                 struct buffer_head *ref_leaf_bh,
2076                                 struct ocfs2_alloc_context *meta_ac,
2077                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
2078 {
2079         int ret;
2080         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2081         struct ocfs2_refcount_block *rb =
2082                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2083         struct ocfs2_extent_tree et;
2084
2085         BUG_ON(rb->rf_records.rl_used);
2086
2087         trace_ocfs2_remove_refcount_extent(
2088                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2089                 (unsigned long long)ref_leaf_bh->b_blocknr,
2090                 le32_to_cpu(rb->rf_cpos));
2091
2092         ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2093         ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
2094                                   1, meta_ac, dealloc);
2095         if (ret) {
2096                 mlog_errno(ret);
2097                 goto out;
2098         }
2099
2100         ocfs2_remove_from_cache(ci, ref_leaf_bh);
2101
2102         /*
2103          * add the freed block to the dealloc so that it will be freed
2104          * when we run dealloc.
2105          */
2106         ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
2107                                         le16_to_cpu(rb->rf_suballoc_slot),
2108                                         le64_to_cpu(rb->rf_suballoc_loc),
2109                                         le64_to_cpu(rb->rf_blkno),
2110                                         le16_to_cpu(rb->rf_suballoc_bit));
2111         if (ret) {
2112                 mlog_errno(ret);
2113                 goto out;
2114         }
2115
2116         ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
2117                                       OCFS2_JOURNAL_ACCESS_WRITE);
2118         if (ret) {
2119                 mlog_errno(ret);
2120                 goto out;
2121         }
2122
2123         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2124
2125         le32_add_cpu(&rb->rf_clusters, -1);
2126
2127         /*
2128          * check whether we need to restore the root refcount block if
2129          * there is no leaf extent block at atll.
2130          */
2131         if (!rb->rf_list.l_next_free_rec) {
2132                 BUG_ON(rb->rf_clusters);
2133
2134                 trace_ocfs2_restore_refcount_block(
2135                      (unsigned long long)ref_root_bh->b_blocknr);
2136
2137                 rb->rf_flags = 0;
2138                 rb->rf_parent = 0;
2139                 rb->rf_cpos = 0;
2140                 memset(&rb->rf_records, 0, sb->s_blocksize -
2141                        offsetof(struct ocfs2_refcount_block, rf_records));
2142                 rb->rf_records.rl_count =
2143                                 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
2144         }
2145
2146         ocfs2_journal_dirty(handle, ref_root_bh);
2147
2148 out:
2149         return ret;
2150 }
2151
2152 int ocfs2_increase_refcount(handle_t *handle,
2153                             struct ocfs2_caching_info *ci,
2154                             struct buffer_head *ref_root_bh,
2155                             u64 cpos, u32 len,
2156                             struct ocfs2_alloc_context *meta_ac,
2157                             struct ocfs2_cached_dealloc_ctxt *dealloc)
2158 {
2159         return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
2160                                          cpos, len, 1,
2161                                          meta_ac, dealloc);
2162 }
2163
2164 static int ocfs2_decrease_refcount_rec(handle_t *handle,
2165                                 struct ocfs2_caching_info *ci,
2166                                 struct buffer_head *ref_root_bh,
2167                                 struct buffer_head *ref_leaf_bh,
2168                                 int index, u64 cpos, unsigned int len,
2169                                 struct ocfs2_alloc_context *meta_ac,
2170                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
2171 {
2172         int ret;
2173         struct ocfs2_refcount_block *rb =
2174                         (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2175         struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
2176
2177         BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
2178         BUG_ON(cpos + len >
2179                le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
2180
2181         trace_ocfs2_decrease_refcount_rec(
2182                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2183                 (unsigned long long)cpos, len);
2184
2185         if (cpos == le64_to_cpu(rec->r_cpos) &&
2186             len == le32_to_cpu(rec->r_clusters))
2187                 ret = ocfs2_change_refcount_rec(handle, ci,
2188                                                 ref_leaf_bh, index, 1, -1);
2189         else {
2190                 struct ocfs2_refcount_rec split = *rec;
2191                 split.r_cpos = cpu_to_le64(cpos);
2192                 split.r_clusters = cpu_to_le32(len);
2193
2194                 le32_add_cpu(&split.r_refcount, -1);
2195
2196                 ret = ocfs2_split_refcount_rec(handle, ci,
2197                                                ref_root_bh, ref_leaf_bh,
2198                                                &split, index, 1,
2199                                                meta_ac, dealloc);
2200         }
2201
2202         if (ret) {
2203                 mlog_errno(ret);
2204                 goto out;
2205         }
2206
2207         /* Remove the leaf refcount block if it contains no refcount record. */
2208         if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
2209                 ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
2210                                                    ref_leaf_bh, meta_ac,
2211                                                    dealloc);
2212                 if (ret)
2213                         mlog_errno(ret);
2214         }
2215
2216 out:
2217         return ret;
2218 }
2219
2220 static int __ocfs2_decrease_refcount(handle_t *handle,
2221                                      struct ocfs2_caching_info *ci,
2222                                      struct buffer_head *ref_root_bh,
2223                                      u64 cpos, u32 len,
2224                                      struct ocfs2_alloc_context *meta_ac,
2225                                      struct ocfs2_cached_dealloc_ctxt *dealloc,
2226                                      int delete)
2227 {
2228         int ret = 0, index = 0;
2229         struct ocfs2_refcount_rec rec;
2230         unsigned int r_count = 0, r_len;
2231         struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2232         struct buffer_head *ref_leaf_bh = NULL;
2233
2234         trace_ocfs2_decrease_refcount(
2235                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2236                 (unsigned long long)cpos, len, delete);
2237
2238         while (len) {
2239                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2240                                              cpos, len, &rec, &index,
2241                                              &ref_leaf_bh);
2242                 if (ret) {
2243                         mlog_errno(ret);
2244                         goto out;
2245                 }
2246
2247                 r_count = le32_to_cpu(rec.r_refcount);
2248                 BUG_ON(r_count == 0);
2249                 if (!delete)
2250                         BUG_ON(r_count > 1);
2251
2252                 r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
2253                               le32_to_cpu(rec.r_clusters)) - cpos;
2254
2255                 ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
2256                                                   ref_leaf_bh, index,
2257                                                   cpos, r_len,
2258                                                   meta_ac, dealloc);
2259                 if (ret) {
2260                         mlog_errno(ret);
2261                         goto out;
2262                 }
2263
2264                 if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
2265                         ret = ocfs2_cache_cluster_dealloc(dealloc,
2266                                           ocfs2_clusters_to_blocks(sb, cpos),
2267                                                           r_len);
2268                         if (ret) {
2269                                 mlog_errno(ret);
2270                                 goto out;
2271                         }
2272                 }
2273
2274                 cpos += r_len;
2275                 len -= r_len;
2276                 brelse(ref_leaf_bh);
2277                 ref_leaf_bh = NULL;
2278         }
2279
2280 out:
2281         brelse(ref_leaf_bh);
2282         return ret;
2283 }
2284
2285 /* Caller must hold refcount tree lock. */
2286 int ocfs2_decrease_refcount(struct inode *inode,
2287                             handle_t *handle, u32 cpos, u32 len,
2288                             struct ocfs2_alloc_context *meta_ac,
2289                             struct ocfs2_cached_dealloc_ctxt *dealloc,
2290                             int delete)
2291 {
2292         int ret;
2293         u64 ref_blkno;
2294         struct buffer_head *ref_root_bh = NULL;
2295         struct ocfs2_refcount_tree *tree;
2296
2297         BUG_ON(!ocfs2_is_refcount_inode(inode));
2298
2299         ret = ocfs2_get_refcount_block(inode, &ref_blkno);
2300         if (ret) {
2301                 mlog_errno(ret);
2302                 goto out;
2303         }
2304
2305         ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
2306         if (ret) {
2307                 mlog_errno(ret);
2308                 goto out;
2309         }
2310
2311         ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
2312                                         &ref_root_bh);
2313         if (ret) {
2314                 mlog_errno(ret);
2315                 goto out;
2316         }
2317
2318         ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
2319                                         cpos, len, meta_ac, dealloc, delete);
2320         if (ret)
2321                 mlog_errno(ret);
2322 out:
2323         brelse(ref_root_bh);
2324         return ret;
2325 }
2326
2327 /*
2328  * Mark the already-existing extent at cpos as refcounted for len clusters.
2329  * This adds the refcount extent flag.
2330  *
2331  * If the existing extent is larger than the request, initiate a
2332  * split. An attempt will be made at merging with adjacent extents.
2333  *
2334  * The caller is responsible for passing down meta_ac if we'll need it.
2335  */
2336 static int ocfs2_mark_extent_refcounted(struct inode *inode,
2337                                 struct ocfs2_extent_tree *et,
2338                                 handle_t *handle, u32 cpos,
2339                                 u32 len, u32 phys,
2340                                 struct ocfs2_alloc_context *meta_ac,
2341                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
2342 {
2343         int ret;
2344
2345         trace_ocfs2_mark_extent_refcounted(OCFS2_I(inode)->ip_blkno,
2346                                            cpos, len, phys);
2347
2348         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2349                 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",
2350                                   inode->i_ino);
2351                 goto out;
2352         }
2353
2354         ret = ocfs2_change_extent_flag(handle, et, cpos,
2355                                        len, phys, meta_ac, dealloc,
2356                                        OCFS2_EXT_REFCOUNTED, 0);
2357         if (ret)
2358                 mlog_errno(ret);
2359
2360 out:
2361         return ret;
2362 }
2363
2364 /*
2365  * Given some contiguous physical clusters, calculate what we need
2366  * for modifying their refcount.
2367  */
2368 static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
2369                                             struct ocfs2_caching_info *ci,
2370                                             struct buffer_head *ref_root_bh,
2371                                             u64 start_cpos,
2372                                             u32 clusters,
2373                                             int *meta_add,
2374                                             int *credits)
2375 {
2376         int ret = 0, index, ref_blocks = 0, recs_add = 0;
2377         u64 cpos = start_cpos;
2378         struct ocfs2_refcount_block *rb;
2379         struct ocfs2_refcount_rec rec;
2380         struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
2381         u32 len;
2382
2383         while (clusters) {
2384                 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2385                                              cpos, clusters, &rec,
2386                                              &index, &ref_leaf_bh);
2387                 if (ret) {
2388                         mlog_errno(ret);
2389                         goto out;
2390                 }
2391
2392                 if (ref_leaf_bh != prev_bh) {
2393                         /*
2394                          * Now we encounter a new leaf block, so calculate
2395                          * whether we need to extend the old leaf.
2396                          */
2397                         if (prev_bh) {
2398                                 rb = (struct ocfs2_refcount_block *)
2399                                                         prev_bh->b_data;
2400
2401                                 if (le16_to_cpu(rb->rf_records.rl_used) +
2402                                     recs_add >
2403                                     le16_to_cpu(rb->rf_records.rl_count))
2404                                         ref_blocks++;
2405                         }
2406
2407                         recs_add = 0;
2408                         *credits += 1;
2409                         brelse(prev_bh);
2410                         prev_bh = ref_leaf_bh;
2411                         get_bh(prev_bh);
2412                 }
2413
2414                 trace_ocfs2_calc_refcount_meta_credits_iterate(
2415                                 recs_add, (unsigned long long)cpos, clusters,
2416                                 (unsigned long long)le64_to_cpu(rec.r_cpos),
2417                                 le32_to_cpu(rec.r_clusters),
2418                                 le32_to_cpu(rec.r_refcount), index);
2419
2420                 len = min((u64)cpos + clusters, le64_to_cpu(rec.r_cpos) +
2421                           le32_to_cpu(rec.r_clusters)) - cpos;
2422                 /*
2423                  * We record all the records which will be inserted to the
2424                  * same refcount block, so that we can tell exactly whether
2425                  * we need a new refcount block or not.
2426                  *
2427                  * If we will insert a new one, this is easy and only happens
2428                  * during adding refcounted flag to the extent, so we don't
2429                  * have a chance of spliting. We just need one record.
2430                  *
2431                  * If the refcount rec already exists, that would be a little
2432                  * complicated. we may have to:
2433                  * 1) split at the beginning if the start pos isn't aligned.
2434                  *    we need 1 more record in this case.
2435                  * 2) split int the end if the end pos isn't aligned.
2436                  *    we need 1 more record in this case.
2437                  * 3) split in the middle because of file system fragmentation.
2438                  *    we need 2 more records in this case(we can't detect this
2439                  *    beforehand, so always think of the worst case).
2440                  */
2441                 if (rec.r_refcount) {
2442                         recs_add += 2;
2443                         /* Check whether we need a split at the beginning. */
2444                         if (cpos == start_cpos &&
2445                             cpos != le64_to_cpu(rec.r_cpos))
2446                                 recs_add++;
2447
2448                         /* Check whether we need a split in the end. */
2449                         if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
2450                             le32_to_cpu(rec.r_clusters))
2451                                 recs_add++;
2452                 } else
2453                         recs_add++;
2454
2455                 brelse(ref_leaf_bh);
2456                 ref_leaf_bh = NULL;
2457                 clusters -= len;
2458                 cpos += len;
2459         }
2460
2461         if (prev_bh) {
2462                 rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
2463
2464                 if (le16_to_cpu(rb->rf_records.rl_used) + recs_add >
2465                     le16_to_cpu(rb->rf_records.rl_count))
2466                         ref_blocks++;
2467
2468                 *credits += 1;
2469         }
2470
2471         if (!ref_blocks)
2472                 goto out;
2473
2474         *meta_add += ref_blocks;
2475         *credits += ref_blocks;
2476
2477         /*
2478          * So we may need ref_blocks to insert into the tree.
2479          * That also means we need to change the b-tree and add that number
2480          * of records since we never merge them.
2481          * We need one more block for expansion since the new created leaf
2482          * block is also full and needs split.
2483          */
2484         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2485         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
2486                 struct ocfs2_extent_tree et;
2487
2488                 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2489                 *meta_add += ocfs2_extend_meta_needed(et.et_root_el);
2490                 *credits += ocfs2_calc_extend_credits(sb,
2491                                                       et.et_root_el);
2492         } else {
2493                 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
2494                 *meta_add += 1;
2495         }
2496
2497 out:
2498
2499         trace_ocfs2_calc_refcount_meta_credits(
2500                 (unsigned long long)start_cpos, clusters,
2501                 *meta_add, *credits);
2502         brelse(ref_leaf_bh);
2503         brelse(prev_bh);
2504         return ret;
2505 }
2506
2507 /*
2508  * For refcount tree, we will decrease some contiguous clusters
2509  * refcount count, so just go through it to see how many blocks
2510  * we gonna touch and whether we need to create new blocks.
2511  *
2512  * Normally the refcount blocks store these refcount should be
2513  * contiguous also, so that we can get the number easily.
2514  * We will at most add split 2 refcount records and 2 more
2515  * refcount blocks, so just check it in a rough way.
2516  *
2517  * Caller must hold refcount tree lock.
2518  */
2519 int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
2520                                           u64 refcount_loc,
2521                                           u64 phys_blkno,
2522                                           u32 clusters,
2523                                           int *credits,
2524                                           int *ref_blocks)
2525 {
2526         int ret;
2527         struct buffer_head *ref_root_bh = NULL;
2528         struct ocfs2_refcount_tree *tree;
2529         u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
2530
2531         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2532                 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",
2533                                   inode->i_ino);
2534                 goto out;
2535         }
2536
2537         BUG_ON(!ocfs2_is_refcount_inode(inode));
2538
2539         ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
2540                                       refcount_loc, &tree);
2541         if (ret) {
2542                 mlog_errno(ret);
2543                 goto out;
2544         }
2545
2546         ret = ocfs2_read_refcount_block(&tree->rf_ci, refcount_loc,
2547                                         &ref_root_bh);
2548         if (ret) {
2549                 mlog_errno(ret);
2550                 goto out;
2551         }
2552
2553         ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
2554                                                &tree->rf_ci,
2555                                                ref_root_bh,
2556                                                start_cpos, clusters,
2557                                                ref_blocks, credits);
2558         if (ret) {
2559                 mlog_errno(ret);
2560                 goto out;
2561         }
2562
2563         trace_ocfs2_prepare_refcount_change_for_del(*ref_blocks, *credits);
2564
2565 out:
2566         brelse(ref_root_bh);
2567         return ret;
2568 }
2569
2570 #define MAX_CONTIG_BYTES        1048576
2571
2572 static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
2573 {
2574         return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
2575 }
2576
2577 static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
2578 {
2579         return ~(ocfs2_cow_contig_clusters(sb) - 1);
2580 }
2581
2582 /*
2583  * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2584  * find an offset (start + (n * contig_clusters)) that is closest to cpos
2585  * while still being less than or equal to it.
2586  *
2587  * The goal is to break the extent at a multiple of contig_clusters.
2588  */
2589 static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
2590                                                  unsigned int start,
2591                                                  unsigned int cpos)
2592 {
2593         BUG_ON(start > cpos);
2594
2595         return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
2596 }
2597
2598 /*
2599  * Given a cluster count of len, pad it out so that it is a multiple
2600  * of contig_clusters.
2601  */
2602 static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
2603                                                   unsigned int len)
2604 {
2605         unsigned int padded =
2606                 (len + (ocfs2_cow_contig_clusters(sb) - 1)) &
2607                 ocfs2_cow_contig_mask(sb);
2608
2609         /* Did we wrap? */
2610         if (padded < len)
2611                 padded = UINT_MAX;
2612
2613         return padded;
2614 }
2615
2616 /*
2617  * Calculate out the start and number of virtual clusters we need to CoW.
2618  *
2619  * cpos is vitual start cluster position we want to do CoW in a
2620  * file and write_len is the cluster length.
2621  * max_cpos is the place where we want to stop CoW intentionally.
2622  *
2623  * Normal we will start CoW from the beginning of extent record cotaining cpos.
2624  * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2625  * get good I/O from the resulting extent tree.
2626  */
2627 static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
2628                                            struct ocfs2_extent_list *el,
2629                                            u32 cpos,
2630                                            u32 write_len,
2631                                            u32 max_cpos,
2632                                            u32 *cow_start,
2633                                            u32 *cow_len)
2634 {
2635         int ret = 0;
2636         int tree_height = le16_to_cpu(el->l_tree_depth), i;
2637         struct buffer_head *eb_bh = NULL;
2638         struct ocfs2_extent_block *eb = NULL;
2639         struct ocfs2_extent_rec *rec;
2640         unsigned int want_clusters, rec_end = 0;
2641         int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
2642         int leaf_clusters;
2643
2644         BUG_ON(cpos + write_len > max_cpos);
2645
2646         if (tree_height > 0) {
2647                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
2648                 if (ret) {
2649                         mlog_errno(ret);
2650                         goto out;
2651                 }
2652
2653                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2654                 el = &eb->h_list;
2655
2656                 if (el->l_tree_depth) {
2657                         ret = ocfs2_error(inode->i_sb,
2658                                           "Inode %lu has non zero tree depth in leaf block %llu\n",
2659                                           inode->i_ino,
2660                                           (unsigned long long)eb_bh->b_blocknr);
2661                         goto out;
2662                 }
2663         }
2664
2665         *cow_len = 0;
2666         for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
2667                 rec = &el->l_recs[i];
2668
2669                 if (ocfs2_is_empty_extent(rec)) {
2670                         mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
2671                                         "index %d\n", inode->i_ino, i);
2672                         continue;
2673                 }
2674
2675                 if (le32_to_cpu(rec->e_cpos) +
2676                     le16_to_cpu(rec->e_leaf_clusters) <= cpos)
2677                         continue;
2678
2679                 if (*cow_len == 0) {
2680                         /*
2681                          * We should find a refcounted record in the
2682                          * first pass.
2683                          */
2684                         BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
2685                         *cow_start = le32_to_cpu(rec->e_cpos);
2686                 }
2687
2688                 /*
2689                  * If we encounter a hole, a non-refcounted record or
2690                  * pass the max_cpos, stop the search.
2691                  */
2692                 if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
2693                     (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
2694                     (max_cpos <= le32_to_cpu(rec->e_cpos)))
2695                         break;
2696
2697                 leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
2698                 rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
2699                 if (rec_end > max_cpos) {
2700                         rec_end = max_cpos;
2701                         leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
2702                 }
2703
2704                 /*
2705                  * How many clusters do we actually need from
2706                  * this extent?  First we see how many we actually
2707                  * need to complete the write.  If that's smaller
2708                  * than contig_clusters, we try for contig_clusters.
2709                  */
2710                 if (!*cow_len)
2711                         want_clusters = write_len;
2712                 else
2713                         want_clusters = (cpos + write_len) -
2714                                 (*cow_start + *cow_len);
2715                 if (want_clusters < contig_clusters)
2716                         want_clusters = contig_clusters;
2717
2718                 /*
2719                  * If the write does not cover the whole extent, we
2720                  * need to calculate how we're going to split the extent.
2721                  * We try to do it on contig_clusters boundaries.
2722                  *
2723                  * Any extent smaller than contig_clusters will be
2724                  * CoWed in its entirety.
2725                  */
2726                 if (leaf_clusters <= contig_clusters)
2727                         *cow_len += leaf_clusters;
2728                 else if (*cow_len || (*cow_start == cpos)) {
2729                         /*
2730                          * This extent needs to be CoW'd from its
2731                          * beginning, so all we have to do is compute
2732                          * how many clusters to grab.  We align
2733                          * want_clusters to the edge of contig_clusters
2734                          * to get better I/O.
2735                          */
2736                         want_clusters = ocfs2_cow_align_length(inode->i_sb,
2737                                                                want_clusters);
2738
2739                         if (leaf_clusters < want_clusters)
2740                                 *cow_len += leaf_clusters;
2741                         else
2742                                 *cow_len += want_clusters;
2743                 } else if ((*cow_start + contig_clusters) >=
2744                            (cpos + write_len)) {
2745                         /*
2746                          * Breaking off contig_clusters at the front
2747                          * of the extent will cover our write.  That's
2748                          * easy.
2749                          */
2750                         *cow_len = contig_clusters;
2751                 } else if ((rec_end - cpos) <= contig_clusters) {
2752                         /*
2753                          * Breaking off contig_clusters at the tail of
2754                          * this extent will cover cpos.
2755                          */
2756                         *cow_start = rec_end - contig_clusters;
2757                         *cow_len = contig_clusters;
2758                 } else if ((rec_end - cpos) <= want_clusters) {
2759                         /*
2760                          * While we can't fit the entire write in this
2761                          * extent, we know that the write goes from cpos
2762                          * to the end of the extent.  Break that off.
2763                          * We try to break it at some multiple of
2764                          * contig_clusters from the front of the extent.
2765                          * Failing that (ie, cpos is within
2766                          * contig_clusters of the front), we'll CoW the
2767                          * entire extent.
2768                          */
2769                         *cow_start = ocfs2_cow_align_start(inode->i_sb,
2770                                                            *cow_start, cpos);
2771                         *cow_len = rec_end - *cow_start;
2772                 } else {
2773                         /*
2774                          * Ok, the entire write lives in the middle of
2775                          * this extent.  Let's try to slice the extent up
2776                          * nicely.  Optimally, our CoW region starts at
2777                          * m*contig_clusters from the beginning of the
2778                          * extent and goes for n*contig_clusters,
2779                          * covering the entire write.
2780                          */
2781                         *cow_start = ocfs2_cow_align_start(inode->i_sb,
2782                                                            *cow_start, cpos);
2783
2784                         want_clusters = (cpos + write_len) - *cow_start;
2785                         want_clusters = ocfs2_cow_align_length(inode->i_sb,
2786                                                                want_clusters);
2787                         if (*cow_start + want_clusters <= rec_end)
2788                                 *cow_len = want_clusters;
2789                         else
2790                                 *cow_len = rec_end - *cow_start;
2791                 }
2792
2793                 /* Have we covered our entire write yet? */
2794                 if ((*cow_start + *cow_len) >= (cpos + write_len))
2795                         break;
2796
2797                 /*
2798                  * If we reach the end of the extent block and don't get enough
2799                  * clusters, continue with the next extent block if possible.
2800                  */
2801                 if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
2802                     eb && eb->h_next_leaf_blk) {
2803                         brelse(eb_bh);
2804                         eb_bh = NULL;
2805
2806                         ret = ocfs2_read_extent_block(INODE_CACHE(inode),
2807                                                le64_to_cpu(eb->h_next_leaf_blk),
2808                                                &eb_bh);
2809                         if (ret) {
2810                                 mlog_errno(ret);
2811                                 goto out;
2812                         }
2813
2814                         eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2815                         el = &eb->h_list;
2816                         i = -1;
2817                 }
2818         }
2819
2820 out:
2821         brelse(eb_bh);
2822         return ret;
2823 }
2824
2825 /*
2826  * Prepare meta_ac, data_ac and calculate credits when we want to add some
2827  * num_clusters in data_tree "et" and change the refcount for the old
2828  * clusters(starting form p_cluster) in the refcount tree.
2829  *
2830  * Note:
2831  * 1. since we may split the old tree, so we at most will need num_clusters + 2
2832  *    more new leaf records.
2833  * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2834  *    just give data_ac = NULL.
2835  */
2836 static int ocfs2_lock_refcount_allocators(struct super_block *sb,
2837                                         u32 p_cluster, u32 num_clusters,
2838                                         struct ocfs2_extent_tree *et,
2839                                         struct ocfs2_caching_info *ref_ci,
2840                                         struct buffer_head *ref_root_bh,
2841                                         struct ocfs2_alloc_context **meta_ac,
2842                                         struct ocfs2_alloc_context **data_ac,
2843                                         int *credits)
2844 {
2845         int ret = 0, meta_add = 0;
2846         int num_free_extents = ocfs2_num_free_extents(et);
2847
2848         if (num_free_extents < 0) {
2849                 ret = num_free_extents;
2850                 mlog_errno(ret);
2851                 goto out;
2852         }
2853
2854         if (num_free_extents < num_clusters + 2)
2855                 meta_add =
2856                         ocfs2_extend_meta_needed(et->et_root_el);
2857
2858         *credits += ocfs2_calc_extend_credits(sb, et->et_root_el);
2859
2860         ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
2861                                                p_cluster, num_clusters,
2862                                                &meta_add, credits);
2863         if (ret) {
2864                 mlog_errno(ret);
2865                 goto out;
2866         }
2867
2868         trace_ocfs2_lock_refcount_allocators(meta_add, *credits);
2869         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
2870                                                 meta_ac);
2871         if (ret) {
2872                 mlog_errno(ret);
2873                 goto out;
2874         }
2875
2876         if (data_ac) {
2877                 ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
2878                                              data_ac);
2879                 if (ret)
2880                         mlog_errno(ret);
2881         }
2882
2883 out:
2884         if (ret) {
2885                 if (*meta_ac) {
2886                         ocfs2_free_alloc_context(*meta_ac);
2887                         *meta_ac = NULL;
2888                 }
2889         }
2890
2891         return ret;
2892 }
2893
2894 static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
2895 {
2896         BUG_ON(buffer_dirty(bh));
2897
2898         clear_buffer_mapped(bh);
2899
2900         return 0;
2901 }
2902
2903 int ocfs2_duplicate_clusters_by_page(handle_t *handle,
2904                                      struct inode *inode,
2905                                      u32 cpos, u32 old_cluster,
2906                                      u32 new_cluster, u32 new_len)
2907 {
2908         int ret = 0, partial;
2909         struct super_block *sb = inode->i_sb;
2910         u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
2911         struct page *page;
2912         pgoff_t page_index;
2913         unsigned int from, to;
2914         loff_t offset, end, map_end;
2915         struct address_space *mapping = inode->i_mapping;
2916
2917         trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
2918                                                new_cluster, new_len);
2919
2920         offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
2921         end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
2922         /*
2923          * We only duplicate pages until we reach the page contains i_size - 1.
2924          * So trim 'end' to i_size.
2925          */
2926         if (end > i_size_read(inode))
2927                 end = i_size_read(inode);
2928
2929         while (offset < end) {
2930                 page_index = offset >> PAGE_SHIFT;
2931                 map_end = ((loff_t)page_index + 1) << PAGE_SHIFT;
2932                 if (map_end > end)
2933                         map_end = end;
2934
2935                 /* from, to is the offset within the page. */
2936                 from = offset & (PAGE_SIZE - 1);
2937                 to = PAGE_SIZE;
2938                 if (map_end & (PAGE_SIZE - 1))
2939                         to = map_end & (PAGE_SIZE - 1);
2940
2941 retry:
2942                 page = find_or_create_page(mapping, page_index, GFP_NOFS);
2943                 if (!page) {
2944                         ret = -ENOMEM;
2945                         mlog_errno(ret);
2946                         break;
2947                 }
2948
2949                 /*
2950                  * In case PAGE_SIZE <= CLUSTER_SIZE, we do not expect a dirty
2951                  * page, so write it back.
2952                  */
2953                 if (PAGE_SIZE <= OCFS2_SB(sb)->s_clustersize) {
2954                         if (PageDirty(page)) {
2955                                 unlock_page(page);
2956                                 put_page(page);
2957
2958                                 ret = filemap_write_and_wait_range(mapping,
2959                                                 offset, map_end - 1);
2960                                 goto retry;
2961                         }
2962                 }
2963
2964                 if (!PageUptodate(page)) {
2965                         struct folio *folio = page_folio(page);
2966
2967                         ret = block_read_full_folio(folio, ocfs2_get_block);
2968                         if (ret) {
2969                                 mlog_errno(ret);
2970                                 goto unlock;
2971                         }
2972                         folio_lock(folio);
2973                 }
2974
2975                 if (page_has_buffers(page)) {
2976                         ret = walk_page_buffers(handle, page_buffers(page),
2977                                                 from, to, &partial,
2978                                                 ocfs2_clear_cow_buffer);
2979                         if (ret) {
2980                                 mlog_errno(ret);
2981                                 goto unlock;
2982                         }
2983                 }
2984
2985                 ocfs2_map_and_dirty_page(inode,
2986                                          handle, from, to,
2987                                          page, 0, &new_block);
2988                 mark_page_accessed(page);
2989 unlock:
2990                 unlock_page(page);
2991                 put_page(page);
2992                 page = NULL;
2993                 offset = map_end;
2994                 if (ret)
2995                         break;
2996         }
2997
2998         return ret;
2999 }
3000
3001 int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
3002                                     struct inode *inode,
3003                                     u32 cpos, u32 old_cluster,
3004                                     u32 new_cluster, u32 new_len)
3005 {
3006         int ret = 0;
3007         struct super_block *sb = inode->i_sb;
3008         struct ocfs2_caching_info *ci = INODE_CACHE(inode);
3009         int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
3010         u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
3011         u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
3012         struct ocfs2_super *osb = OCFS2_SB(sb);
3013         struct buffer_head *old_bh = NULL;
3014         struct buffer_head *new_bh = NULL;
3015
3016         trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
3017                                                new_cluster, new_len);
3018
3019         for (i = 0; i < blocks; i++, old_block++, new_block++) {
3020                 new_bh = sb_getblk(osb->sb, new_block);
3021                 if (new_bh == NULL) {
3022                         ret = -ENOMEM;
3023                         mlog_errno(ret);
3024                         break;
3025                 }
3026
3027                 ocfs2_set_new_buffer_uptodate(ci, new_bh);
3028
3029                 ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
3030                 if (ret) {
3031                         mlog_errno(ret);
3032                         break;
3033                 }
3034
3035                 ret = ocfs2_journal_access(handle, ci, new_bh,
3036                                            OCFS2_JOURNAL_ACCESS_CREATE);
3037                 if (ret) {
3038                         mlog_errno(ret);
3039                         break;
3040                 }
3041
3042                 memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
3043                 ocfs2_journal_dirty(handle, new_bh);
3044
3045                 brelse(new_bh);
3046                 brelse(old_bh);
3047                 new_bh = NULL;
3048                 old_bh = NULL;
3049         }
3050
3051         brelse(new_bh);
3052         brelse(old_bh);
3053         return ret;
3054 }
3055
3056 static int ocfs2_clear_ext_refcount(handle_t *handle,
3057                                     struct ocfs2_extent_tree *et,
3058                                     u32 cpos, u32 p_cluster, u32 len,
3059                                     unsigned int ext_flags,
3060                                     struct ocfs2_alloc_context *meta_ac,
3061                                     struct ocfs2_cached_dealloc_ctxt *dealloc)
3062 {
3063         int ret, index;
3064         struct ocfs2_extent_rec replace_rec;
3065         struct ocfs2_path *path = NULL;
3066         struct ocfs2_extent_list *el;
3067         struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
3068         u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
3069
3070         trace_ocfs2_clear_ext_refcount((unsigned long long)ino,
3071                                        cpos, len, p_cluster, ext_flags);
3072
3073         memset(&replace_rec, 0, sizeof(replace_rec));
3074         replace_rec.e_cpos = cpu_to_le32(cpos);
3075         replace_rec.e_leaf_clusters = cpu_to_le16(len);
3076         replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
3077                                                                    p_cluster));
3078         replace_rec.e_flags = ext_flags;
3079         replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
3080
3081         path = ocfs2_new_path_from_et(et);
3082         if (!path) {
3083                 ret = -ENOMEM;
3084                 mlog_errno(ret);
3085                 goto out;
3086         }
3087
3088         ret = ocfs2_find_path(et->et_ci, path, cpos);
3089         if (ret) {
3090                 mlog_errno(ret);
3091                 goto out;
3092         }
3093
3094         el = path_leaf_el(path);
3095
3096         index = ocfs2_search_extent_list(el, cpos);
3097         if (index == -1) {
3098                 ret = ocfs2_error(sb,
3099                                   "Inode %llu has an extent at cpos %u which can no longer be found\n",
3100                                   (unsigned long long)ino, cpos);
3101                 goto out;
3102         }
3103
3104         ret = ocfs2_split_extent(handle, et, path, index,
3105                                  &replace_rec, meta_ac, dealloc);
3106         if (ret)
3107                 mlog_errno(ret);
3108
3109 out:
3110         ocfs2_free_path(path);
3111         return ret;
3112 }
3113
3114 static int ocfs2_replace_clusters(handle_t *handle,
3115                                   struct ocfs2_cow_context *context,
3116                                   u32 cpos, u32 old,
3117                                   u32 new, u32 len,
3118                                   unsigned int ext_flags)
3119 {
3120         int ret;
3121         struct ocfs2_caching_info *ci = context->data_et.et_ci;
3122         u64 ino = ocfs2_metadata_cache_owner(ci);
3123
3124         trace_ocfs2_replace_clusters((unsigned long long)ino,
3125                                      cpos, old, new, len, ext_flags);
3126
3127         /*If the old clusters is unwritten, no need to duplicate. */
3128         if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
3129                 ret = context->cow_duplicate_clusters(handle, context->inode,
3130                                                       cpos, old, new, len);
3131                 if (ret) {
3132                         mlog_errno(ret);
3133                         goto out;
3134                 }
3135         }
3136
3137         ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
3138                                        cpos, new, len, ext_flags,
3139                                        context->meta_ac, &context->dealloc);
3140         if (ret)
3141                 mlog_errno(ret);
3142 out:
3143         return ret;
3144 }
3145
3146 int ocfs2_cow_sync_writeback(struct super_block *sb,
3147                              struct inode *inode,
3148                              u32 cpos, u32 num_clusters)
3149 {
3150         int ret;
3151         loff_t start, end;
3152
3153         if (ocfs2_should_order_data(inode))
3154                 return 0;
3155
3156         start = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
3157         end = start + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits) - 1;
3158
3159         ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
3160         if (ret < 0)
3161                 mlog_errno(ret);
3162
3163         return ret;
3164 }
3165
3166 static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
3167                                  u32 v_cluster, u32 *p_cluster,
3168                                  u32 *num_clusters,
3169                                  unsigned int *extent_flags)
3170 {
3171         return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
3172                                   num_clusters, extent_flags);
3173 }
3174
3175 static int ocfs2_make_clusters_writable(struct super_block *sb,
3176                                         struct ocfs2_cow_context *context,
3177                                         u32 cpos, u32 p_cluster,
3178                                         u32 num_clusters, unsigned int e_flags)
3179 {
3180         int ret, delete, index, credits =  0;
3181         u32 new_bit, new_len, orig_num_clusters;
3182         unsigned int set_len;
3183         struct ocfs2_super *osb = OCFS2_SB(sb);
3184         handle_t *handle;
3185         struct buffer_head *ref_leaf_bh = NULL;
3186         struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
3187         struct ocfs2_refcount_rec rec;
3188
3189         trace_ocfs2_make_clusters_writable(cpos, p_cluster,
3190                                            num_clusters, e_flags);
3191
3192         ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
3193                                              &context->data_et,
3194                                              ref_ci,
3195                                              context->ref_root_bh,
3196                                              &context->meta_ac,
3197                                              &context->data_ac, &credits);
3198         if (ret) {
3199                 mlog_errno(ret);
3200                 return ret;
3201         }
3202
3203         if (context->post_refcount)
3204                 credits += context->post_refcount->credits;
3205
3206         credits += context->extra_credits;
3207         handle = ocfs2_start_trans(osb, credits);
3208         if (IS_ERR(handle)) {
3209                 ret = PTR_ERR(handle);
3210                 mlog_errno(ret);
3211                 goto out;
3212         }
3213
3214         orig_num_clusters = num_clusters;
3215
3216         while (num_clusters) {
3217                 ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
3218                                              p_cluster, num_clusters,
3219                                              &rec, &index, &ref_leaf_bh);
3220                 if (ret) {
3221                         mlog_errno(ret);
3222                         goto out_commit;
3223                 }
3224
3225                 BUG_ON(!rec.r_refcount);
3226                 set_len = min((u64)p_cluster + num_clusters,
3227                               le64_to_cpu(rec.r_cpos) +
3228                               le32_to_cpu(rec.r_clusters)) - p_cluster;
3229
3230                 /*
3231                  * There are many different situation here.
3232                  * 1. If refcount == 1, remove the flag and don't COW.
3233                  * 2. If refcount > 1, allocate clusters.
3234                  *    Here we may not allocate r_len once at a time, so continue
3235                  *    until we reach num_clusters.
3236                  */
3237                 if (le32_to_cpu(rec.r_refcount) == 1) {
3238                         delete = 0;
3239                         ret = ocfs2_clear_ext_refcount(handle,
3240                                                        &context->data_et,
3241                                                        cpos, p_cluster,
3242                                                        set_len, e_flags,
3243                                                        context->meta_ac,
3244                                                        &context->dealloc);
3245                         if (ret) {
3246                                 mlog_errno(ret);
3247                                 goto out_commit;
3248                         }
3249                 } else {
3250                         delete = 1;
3251
3252                         ret = __ocfs2_claim_clusters(handle,
3253                                                      context->data_ac,
3254                                                      1, set_len,
3255                                                      &new_bit, &new_len);
3256                         if (ret) {
3257                                 mlog_errno(ret);
3258                                 goto out_commit;
3259                         }
3260
3261                         ret = ocfs2_replace_clusters(handle, context,
3262                                                      cpos, p_cluster, new_bit,
3263                                                      new_len, e_flags);
3264                         if (ret) {
3265                                 mlog_errno(ret);
3266                                 goto out_commit;
3267                         }
3268                         set_len = new_len;
3269                 }
3270
3271                 ret = __ocfs2_decrease_refcount(handle, ref_ci,
3272                                                 context->ref_root_bh,
3273                                                 p_cluster, set_len,
3274                                                 context->meta_ac,
3275                                                 &context->dealloc, delete);
3276                 if (ret) {
3277                         mlog_errno(ret);
3278                         goto out_commit;
3279                 }
3280
3281                 cpos += set_len;
3282                 p_cluster += set_len;
3283                 num_clusters -= set_len;
3284                 brelse(ref_leaf_bh);
3285                 ref_leaf_bh = NULL;
3286         }
3287
3288         /* handle any post_cow action. */
3289         if (context->post_refcount && context->post_refcount->func) {
3290                 ret = context->post_refcount->func(context->inode, handle,
3291                                                 context->post_refcount->para);
3292                 if (ret) {
3293                         mlog_errno(ret);
3294                         goto out_commit;
3295                 }
3296         }
3297
3298         /*
3299          * Here we should write the new page out first if we are
3300          * in write-back mode.
3301          */
3302         if (context->get_clusters == ocfs2_di_get_clusters) {
3303                 ret = ocfs2_cow_sync_writeback(sb, context->inode, cpos,
3304                                                orig_num_clusters);
3305                 if (ret)
3306                         mlog_errno(ret);
3307         }
3308
3309 out_commit:
3310         ocfs2_commit_trans(osb, handle);
3311
3312 out:
3313         if (context->data_ac) {
3314                 ocfs2_free_alloc_context(context->data_ac);
3315                 context->data_ac = NULL;
3316         }
3317         if (context->meta_ac) {
3318                 ocfs2_free_alloc_context(context->meta_ac);
3319                 context->meta_ac = NULL;
3320         }
3321         brelse(ref_leaf_bh);
3322
3323         return ret;
3324 }
3325
3326 static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
3327 {
3328         int ret = 0;
3329         struct inode *inode = context->inode;
3330         u32 cow_start = context->cow_start, cow_len = context->cow_len;
3331         u32 p_cluster, num_clusters;
3332         unsigned int ext_flags;
3333         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3334
3335         if (!ocfs2_refcount_tree(osb)) {
3336                 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",
3337                                    inode->i_ino);
3338         }
3339
3340         ocfs2_init_dealloc_ctxt(&context->dealloc);
3341
3342         while (cow_len) {
3343                 ret = context->get_clusters(context, cow_start, &p_cluster,
3344                                             &num_clusters, &ext_flags);
3345                 if (ret) {
3346                         mlog_errno(ret);
3347                         break;
3348                 }
3349
3350                 BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
3351
3352                 if (cow_len < num_clusters)
3353                         num_clusters = cow_len;
3354
3355                 ret = ocfs2_make_clusters_writable(inode->i_sb, context,
3356                                                    cow_start, p_cluster,
3357                                                    num_clusters, ext_flags);
3358                 if (ret) {
3359                         mlog_errno(ret);
3360                         break;
3361                 }
3362
3363                 cow_len -= num_clusters;
3364                 cow_start += num_clusters;
3365         }
3366
3367         if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
3368                 ocfs2_schedule_truncate_log_flush(osb, 1);
3369                 ocfs2_run_deallocs(osb, &context->dealloc);
3370         }
3371
3372         return ret;
3373 }
3374
3375 /*
3376  * Starting at cpos, try to CoW write_len clusters.  Don't CoW
3377  * past max_cpos.  This will stop when it runs into a hole or an
3378  * unrefcounted extent.
3379  */
3380 static int ocfs2_refcount_cow_hunk(struct inode *inode,
3381                                    struct buffer_head *di_bh,
3382                                    u32 cpos, u32 write_len, u32 max_cpos)
3383 {
3384         int ret;
3385         u32 cow_start = 0, cow_len = 0;
3386         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3387         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3388         struct buffer_head *ref_root_bh = NULL;
3389         struct ocfs2_refcount_tree *ref_tree;
3390         struct ocfs2_cow_context *context = NULL;
3391
3392         BUG_ON(!ocfs2_is_refcount_inode(inode));
3393
3394         ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
3395                                               cpos, write_len, max_cpos,
3396                                               &cow_start, &cow_len);
3397         if (ret) {
3398                 mlog_errno(ret);
3399                 goto out;
3400         }
3401
3402         trace_ocfs2_refcount_cow_hunk(OCFS2_I(inode)->ip_blkno,
3403                                       cpos, write_len, max_cpos,
3404                                       cow_start, cow_len);
3405
3406         BUG_ON(cow_len == 0);
3407
3408         context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3409         if (!context) {
3410                 ret = -ENOMEM;
3411                 mlog_errno(ret);
3412                 goto out;
3413         }
3414
3415         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
3416                                        1, &ref_tree, &ref_root_bh);
3417         if (ret) {
3418                 mlog_errno(ret);
3419                 goto out;
3420         }
3421
3422         context->inode = inode;
3423         context->cow_start = cow_start;
3424         context->cow_len = cow_len;
3425         context->ref_tree = ref_tree;
3426         context->ref_root_bh = ref_root_bh;
3427         context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
3428         context->get_clusters = ocfs2_di_get_clusters;
3429
3430         ocfs2_init_dinode_extent_tree(&context->data_et,
3431                                       INODE_CACHE(inode), di_bh);
3432
3433         ret = ocfs2_replace_cow(context);
3434         if (ret)
3435                 mlog_errno(ret);
3436
3437         /*
3438          * truncate the extent map here since no matter whether we meet with
3439          * any error during the action, we shouldn't trust cached extent map
3440          * any more.
3441          */
3442         ocfs2_extent_map_trunc(inode, cow_start);
3443
3444         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3445         brelse(ref_root_bh);
3446 out:
3447         kfree(context);
3448         return ret;
3449 }
3450
3451 /*
3452  * CoW any and all clusters between cpos and cpos+write_len.
3453  * Don't CoW past max_cpos.  If this returns successfully, all
3454  * clusters between cpos and cpos+write_len are safe to modify.
3455  */
3456 int ocfs2_refcount_cow(struct inode *inode,
3457                        struct buffer_head *di_bh,
3458                        u32 cpos, u32 write_len, u32 max_cpos)
3459 {
3460         int ret = 0;
3461         u32 p_cluster, num_clusters;
3462         unsigned int ext_flags;
3463
3464         while (write_len) {
3465                 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3466                                          &num_clusters, &ext_flags);
3467                 if (ret) {
3468                         mlog_errno(ret);
3469                         break;
3470                 }
3471
3472                 if (write_len < num_clusters)
3473                         num_clusters = write_len;
3474
3475                 if (ext_flags & OCFS2_EXT_REFCOUNTED) {
3476                         ret = ocfs2_refcount_cow_hunk(inode, di_bh, cpos,
3477                                                       num_clusters, max_cpos);
3478                         if (ret) {
3479                                 mlog_errno(ret);
3480                                 break;
3481                         }
3482                 }
3483
3484                 write_len -= num_clusters;
3485                 cpos += num_clusters;
3486         }
3487
3488         return ret;
3489 }
3490
3491 static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
3492                                           u32 v_cluster, u32 *p_cluster,
3493                                           u32 *num_clusters,
3494                                           unsigned int *extent_flags)
3495 {
3496         struct inode *inode = context->inode;
3497         struct ocfs2_xattr_value_root *xv = context->cow_object;
3498
3499         return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
3500                                         num_clusters, &xv->xr_list,
3501                                         extent_flags);
3502 }
3503
3504 /*
3505  * Given a xattr value root, calculate the most meta/credits we need for
3506  * refcount tree change if we truncate it to 0.
3507  */
3508 int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
3509                                        struct ocfs2_caching_info *ref_ci,
3510                                        struct buffer_head *ref_root_bh,
3511                                        struct ocfs2_xattr_value_root *xv,
3512                                        int *meta_add, int *credits)
3513 {
3514         int ret = 0, index, ref_blocks = 0;
3515         u32 p_cluster, num_clusters;
3516         u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
3517         struct ocfs2_refcount_block *rb;
3518         struct ocfs2_refcount_rec rec;
3519         struct buffer_head *ref_leaf_bh = NULL;
3520
3521         while (cpos < clusters) {
3522                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
3523                                                &num_clusters, &xv->xr_list,
3524                                                NULL);
3525                 if (ret) {
3526                         mlog_errno(ret);
3527                         goto out;
3528                 }
3529
3530                 cpos += num_clusters;
3531
3532                 while (num_clusters) {
3533                         ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
3534                                                      p_cluster, num_clusters,
3535                                                      &rec, &index,
3536                                                      &ref_leaf_bh);
3537                         if (ret) {
3538                                 mlog_errno(ret);
3539                                 goto out;
3540                         }
3541
3542                         BUG_ON(!rec.r_refcount);
3543
3544                         rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
3545
3546                         /*
3547                          * We really don't know whether the other clusters is in
3548                          * this refcount block or not, so just take the worst
3549                          * case that all the clusters are in this block and each
3550                          * one will split a refcount rec, so totally we need
3551                          * clusters * 2 new refcount rec.
3552                          */
3553                         if (le16_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
3554                             le16_to_cpu(rb->rf_records.rl_count))
3555                                 ref_blocks++;
3556
3557                         *credits += 1;
3558                         brelse(ref_leaf_bh);
3559                         ref_leaf_bh = NULL;
3560
3561                         if (num_clusters <= le32_to_cpu(rec.r_clusters))
3562                                 break;
3563                         else
3564                                 num_clusters -= le32_to_cpu(rec.r_clusters);
3565                         p_cluster += num_clusters;
3566                 }
3567         }
3568
3569         *meta_add += ref_blocks;
3570         if (!ref_blocks)
3571                 goto out;
3572
3573         rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
3574         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
3575                 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
3576         else {
3577                 struct ocfs2_extent_tree et;
3578
3579                 ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
3580                 *credits += ocfs2_calc_extend_credits(inode->i_sb,
3581                                                       et.et_root_el);
3582         }
3583
3584 out:
3585         brelse(ref_leaf_bh);
3586         return ret;
3587 }
3588
3589 /*
3590  * Do CoW for xattr.
3591  */
3592 int ocfs2_refcount_cow_xattr(struct inode *inode,
3593                              struct ocfs2_dinode *di,
3594                              struct ocfs2_xattr_value_buf *vb,
3595                              struct ocfs2_refcount_tree *ref_tree,
3596                              struct buffer_head *ref_root_bh,
3597                              u32 cpos, u32 write_len,
3598                              struct ocfs2_post_refcount *post)
3599 {
3600         int ret;
3601         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
3602         struct ocfs2_cow_context *context = NULL;
3603         u32 cow_start, cow_len;
3604
3605         BUG_ON(!ocfs2_is_refcount_inode(inode));
3606
3607         ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
3608                                               cpos, write_len, UINT_MAX,
3609                                               &cow_start, &cow_len);
3610         if (ret) {
3611                 mlog_errno(ret);
3612                 goto out;
3613         }
3614
3615         BUG_ON(cow_len == 0);
3616
3617         context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3618         if (!context) {
3619                 ret = -ENOMEM;
3620                 mlog_errno(ret);
3621                 goto out;
3622         }
3623
3624         context->inode = inode;
3625         context->cow_start = cow_start;
3626         context->cow_len = cow_len;
3627         context->ref_tree = ref_tree;
3628         context->ref_root_bh = ref_root_bh;
3629         context->cow_object = xv;
3630
3631         context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
3632         /* We need the extra credits for duplicate_clusters by jbd. */
3633         context->extra_credits =
3634                 ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
3635         context->get_clusters = ocfs2_xattr_value_get_clusters;
3636         context->post_refcount = post;
3637
3638         ocfs2_init_xattr_value_extent_tree(&context->data_et,
3639                                            INODE_CACHE(inode), vb);
3640
3641         ret = ocfs2_replace_cow(context);
3642         if (ret)
3643                 mlog_errno(ret);
3644
3645 out:
3646         kfree(context);
3647         return ret;
3648 }
3649
3650 /*
3651  * Insert a new extent into refcount tree and mark a extent rec
3652  * as refcounted in the dinode tree.
3653  */
3654 int ocfs2_add_refcount_flag(struct inode *inode,
3655                             struct ocfs2_extent_tree *data_et,
3656                             struct ocfs2_caching_info *ref_ci,
3657                             struct buffer_head *ref_root_bh,
3658                             u32 cpos, u32 p_cluster, u32 num_clusters,
3659                             struct ocfs2_cached_dealloc_ctxt *dealloc,
3660                             struct ocfs2_post_refcount *post)
3661 {
3662         int ret;
3663         handle_t *handle;
3664         int credits = 1, ref_blocks = 0;
3665         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3666         struct ocfs2_alloc_context *meta_ac = NULL;
3667
3668         /* We need to be able to handle at least an extent tree split. */
3669         ref_blocks = ocfs2_extend_meta_needed(data_et->et_root_el);
3670
3671         ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
3672                                                ref_ci, ref_root_bh,
3673                                                p_cluster, num_clusters,
3674                                                &ref_blocks, &credits);
3675         if (ret) {
3676                 mlog_errno(ret);
3677                 goto out;
3678         }
3679
3680         trace_ocfs2_add_refcount_flag(ref_blocks, credits);
3681
3682         if (ref_blocks) {
3683                 ret = ocfs2_reserve_new_metadata_blocks(osb,
3684                                                         ref_blocks, &meta_ac);
3685                 if (ret) {
3686                         mlog_errno(ret);
3687                         goto out;
3688                 }
3689         }
3690
3691         if (post)
3692                 credits += post->credits;
3693
3694         handle = ocfs2_start_trans(osb, credits);
3695         if (IS_ERR(handle)) {
3696                 ret = PTR_ERR(handle);
3697                 mlog_errno(ret);
3698                 goto out;
3699         }
3700
3701         ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
3702                                            cpos, num_clusters, p_cluster,
3703                                            meta_ac, dealloc);
3704         if (ret) {
3705                 mlog_errno(ret);
3706                 goto out_commit;
3707         }
3708
3709         ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3710                                         p_cluster, num_clusters, 0,
3711                                         meta_ac, dealloc);
3712         if (ret) {
3713                 mlog_errno(ret);
3714                 goto out_commit;
3715         }
3716
3717         if (post && post->func) {
3718                 ret = post->func(inode, handle, post->para);
3719                 if (ret)
3720                         mlog_errno(ret);
3721         }
3722
3723 out_commit:
3724         ocfs2_commit_trans(osb, handle);
3725 out:
3726         if (meta_ac)
3727                 ocfs2_free_alloc_context(meta_ac);
3728         return ret;
3729 }
3730
3731 static int ocfs2_change_ctime(struct inode *inode,
3732                               struct buffer_head *di_bh)
3733 {
3734         int ret;
3735         handle_t *handle;
3736         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3737
3738         handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
3739                                    OCFS2_INODE_UPDATE_CREDITS);
3740         if (IS_ERR(handle)) {
3741                 ret = PTR_ERR(handle);
3742                 mlog_errno(ret);
3743                 goto out;
3744         }
3745
3746         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
3747                                       OCFS2_JOURNAL_ACCESS_WRITE);
3748         if (ret) {
3749                 mlog_errno(ret);
3750                 goto out_commit;
3751         }
3752
3753         inode_set_ctime_current(inode);
3754         di->i_ctime = cpu_to_le64(inode_get_ctime(inode).tv_sec);
3755         di->i_ctime_nsec = cpu_to_le32(inode_get_ctime(inode).tv_nsec);
3756
3757         ocfs2_journal_dirty(handle, di_bh);
3758
3759 out_commit:
3760         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3761 out:
3762         return ret;
3763 }
3764
3765 static int ocfs2_attach_refcount_tree(struct inode *inode,
3766                                       struct buffer_head *di_bh)
3767 {
3768         int ret, data_changed = 0;
3769         struct buffer_head *ref_root_bh = NULL;
3770         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3771         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3772         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3773         struct ocfs2_refcount_tree *ref_tree;
3774         unsigned int ext_flags;
3775         loff_t size;
3776         u32 cpos, num_clusters, clusters, p_cluster;
3777         struct ocfs2_cached_dealloc_ctxt dealloc;
3778         struct ocfs2_extent_tree di_et;
3779
3780         ocfs2_init_dealloc_ctxt(&dealloc);
3781
3782         if (!ocfs2_is_refcount_inode(inode)) {
3783                 ret = ocfs2_create_refcount_tree(inode, di_bh);
3784                 if (ret) {
3785                         mlog_errno(ret);
3786                         goto out;
3787                 }
3788         }
3789
3790         BUG_ON(!di->i_refcount_loc);
3791         ret = ocfs2_lock_refcount_tree(osb,
3792                                        le64_to_cpu(di->i_refcount_loc), 1,
3793                                        &ref_tree, &ref_root_bh);
3794         if (ret) {
3795                 mlog_errno(ret);
3796                 goto out;
3797         }
3798
3799         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
3800                 goto attach_xattr;
3801
3802         ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
3803
3804         size = i_size_read(inode);
3805         clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
3806
3807         cpos = 0;
3808         while (cpos < clusters) {
3809                 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3810                                          &num_clusters, &ext_flags);
3811                 if (ret) {
3812                         mlog_errno(ret);
3813                         goto unlock;
3814                 }
3815                 if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
3816                         ret = ocfs2_add_refcount_flag(inode, &di_et,
3817                                                       &ref_tree->rf_ci,
3818                                                       ref_root_bh, cpos,
3819                                                       p_cluster, num_clusters,
3820                                                       &dealloc, NULL);
3821                         if (ret) {
3822                                 mlog_errno(ret);
3823                                 goto unlock;
3824                         }
3825
3826                         data_changed = 1;
3827                 }
3828                 cpos += num_clusters;
3829         }
3830
3831 attach_xattr:
3832         if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
3833                 ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
3834                                                        &ref_tree->rf_ci,
3835                                                        ref_root_bh,
3836                                                        &dealloc);
3837                 if (ret) {
3838                         mlog_errno(ret);
3839                         goto unlock;
3840                 }
3841         }
3842
3843         if (data_changed) {
3844                 ret = ocfs2_change_ctime(inode, di_bh);
3845                 if (ret)
3846                         mlog_errno(ret);
3847         }
3848
3849 unlock:
3850         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3851         brelse(ref_root_bh);
3852
3853         if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
3854                 ocfs2_schedule_truncate_log_flush(osb, 1);
3855                 ocfs2_run_deallocs(osb, &dealloc);
3856         }
3857 out:
3858         /*
3859          * Empty the extent map so that we may get the right extent
3860          * record from the disk.
3861          */
3862         ocfs2_extent_map_trunc(inode, 0);
3863
3864         return ret;
3865 }
3866
3867 static int ocfs2_add_refcounted_extent(struct inode *inode,
3868                                    struct ocfs2_extent_tree *et,
3869                                    struct ocfs2_caching_info *ref_ci,
3870                                    struct buffer_head *ref_root_bh,
3871                                    u32 cpos, u32 p_cluster, u32 num_clusters,
3872                                    unsigned int ext_flags,
3873                                    struct ocfs2_cached_dealloc_ctxt *dealloc)
3874 {
3875         int ret;
3876         handle_t *handle;
3877         int credits = 0;
3878         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3879         struct ocfs2_alloc_context *meta_ac = NULL;
3880
3881         ret = ocfs2_lock_refcount_allocators(inode->i_sb,
3882                                              p_cluster, num_clusters,
3883                                              et, ref_ci,
3884                                              ref_root_bh, &meta_ac,
3885                                              NULL, &credits);
3886         if (ret) {
3887                 mlog_errno(ret);
3888                 goto out;
3889         }
3890
3891         handle = ocfs2_start_trans(osb, credits);
3892         if (IS_ERR(handle)) {
3893                 ret = PTR_ERR(handle);
3894                 mlog_errno(ret);
3895                 goto out;
3896         }
3897
3898         ret = ocfs2_insert_extent(handle, et, cpos,
3899                         ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
3900                         num_clusters, ext_flags, meta_ac);
3901         if (ret) {
3902                 mlog_errno(ret);
3903                 goto out_commit;
3904         }
3905
3906         ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3907                                       p_cluster, num_clusters,
3908                                       meta_ac, dealloc);
3909         if (ret) {
3910                 mlog_errno(ret);
3911                 goto out_commit;
3912         }
3913
3914         ret = dquot_alloc_space_nodirty(inode,
3915                 ocfs2_clusters_to_bytes(osb->sb, num_clusters));
3916         if (ret)
3917                 mlog_errno(ret);
3918
3919 out_commit:
3920         ocfs2_commit_trans(osb, handle);
3921 out:
3922         if (meta_ac)
3923                 ocfs2_free_alloc_context(meta_ac);
3924         return ret;
3925 }
3926
3927 static int ocfs2_duplicate_inline_data(struct inode *s_inode,
3928                                        struct buffer_head *s_bh,
3929                                        struct inode *t_inode,
3930                                        struct buffer_head *t_bh)
3931 {
3932         int ret;
3933         handle_t *handle;
3934         struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
3935         struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
3936         struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
3937
3938         BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
3939
3940         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
3941         if (IS_ERR(handle)) {
3942                 ret = PTR_ERR(handle);
3943                 mlog_errno(ret);
3944                 goto out;
3945         }
3946
3947         ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
3948                                       OCFS2_JOURNAL_ACCESS_WRITE);
3949         if (ret) {
3950                 mlog_errno(ret);
3951                 goto out_commit;
3952         }
3953
3954         t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
3955         memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
3956                le16_to_cpu(s_di->id2.i_data.id_count));
3957         spin_lock(&OCFS2_I(t_inode)->ip_lock);
3958         OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
3959         t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
3960         spin_unlock(&OCFS2_I(t_inode)->ip_lock);
3961
3962         ocfs2_journal_dirty(handle, t_bh);
3963
3964 out_commit:
3965         ocfs2_commit_trans(osb, handle);
3966 out:
3967         return ret;
3968 }
3969
3970 static int ocfs2_duplicate_extent_list(struct inode *s_inode,
3971                                 struct inode *t_inode,
3972                                 struct buffer_head *t_bh,
3973                                 struct ocfs2_caching_info *ref_ci,
3974                                 struct buffer_head *ref_root_bh,
3975                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
3976 {
3977         int ret = 0;
3978         u32 p_cluster, num_clusters, clusters, cpos;
3979         loff_t size;
3980         unsigned int ext_flags;
3981         struct ocfs2_extent_tree et;
3982
3983         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
3984
3985         size = i_size_read(s_inode);
3986         clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
3987
3988         cpos = 0;
3989         while (cpos < clusters) {
3990                 ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
3991                                          &num_clusters, &ext_flags);
3992                 if (ret) {
3993                         mlog_errno(ret);
3994                         goto out;
3995                 }
3996                 if (p_cluster) {
3997                         ret = ocfs2_add_refcounted_extent(t_inode, &et,
3998                                                           ref_ci, ref_root_bh,
3999                                                           cpos, p_cluster,
4000                                                           num_clusters,
4001                                                           ext_flags,
4002                                                           dealloc);
4003                         if (ret) {
4004                                 mlog_errno(ret);
4005                                 goto out;
4006                         }
4007                 }
4008
4009                 cpos += num_clusters;
4010         }
4011
4012 out:
4013         return ret;
4014 }
4015
4016 /*
4017  * change the new file's attributes to the src.
4018  *
4019  * reflink creates a snapshot of a file, that means the attributes
4020  * must be identical except for three exceptions - nlink, ino, and ctime.
4021  */
4022 static int ocfs2_complete_reflink(struct inode *s_inode,
4023                                   struct buffer_head *s_bh,
4024                                   struct inode *t_inode,
4025                                   struct buffer_head *t_bh,
4026                                   bool preserve)
4027 {
4028         int ret;
4029         handle_t *handle;
4030         struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4031         struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
4032         loff_t size = i_size_read(s_inode);
4033
4034         handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
4035                                    OCFS2_INODE_UPDATE_CREDITS);
4036         if (IS_ERR(handle)) {
4037                 ret = PTR_ERR(handle);
4038                 mlog_errno(ret);
4039                 return ret;
4040         }
4041
4042         ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4043                                       OCFS2_JOURNAL_ACCESS_WRITE);
4044         if (ret) {
4045                 mlog_errno(ret);
4046                 goto out_commit;
4047         }
4048
4049         spin_lock(&OCFS2_I(t_inode)->ip_lock);
4050         OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
4051         OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
4052         OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
4053         spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4054         i_size_write(t_inode, size);
4055         t_inode->i_blocks = s_inode->i_blocks;
4056
4057         di->i_xattr_inline_size = s_di->i_xattr_inline_size;
4058         di->i_clusters = s_di->i_clusters;
4059         di->i_size = s_di->i_size;
4060         di->i_dyn_features = s_di->i_dyn_features;
4061         di->i_attr = s_di->i_attr;
4062
4063         if (preserve) {
4064                 t_inode->i_uid = s_inode->i_uid;
4065                 t_inode->i_gid = s_inode->i_gid;
4066                 t_inode->i_mode = s_inode->i_mode;
4067                 di->i_uid = s_di->i_uid;
4068                 di->i_gid = s_di->i_gid;
4069                 di->i_mode = s_di->i_mode;
4070
4071                 /*
4072                  * update time.
4073                  * we want mtime to appear identical to the source and
4074                  * update ctime.
4075                  */
4076                 inode_set_ctime_current(t_inode);
4077
4078                 di->i_ctime = cpu_to_le64(inode_get_ctime(t_inode).tv_sec);
4079                 di->i_ctime_nsec = cpu_to_le32(inode_get_ctime(t_inode).tv_nsec);
4080
4081                 t_inode->i_mtime = s_inode->i_mtime;
4082                 di->i_mtime = s_di->i_mtime;
4083                 di->i_mtime_nsec = s_di->i_mtime_nsec;
4084         }
4085
4086         ocfs2_journal_dirty(handle, t_bh);
4087
4088 out_commit:
4089         ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
4090         return ret;
4091 }
4092
4093 static int ocfs2_create_reflink_node(struct inode *s_inode,
4094                                      struct buffer_head *s_bh,
4095                                      struct inode *t_inode,
4096                                      struct buffer_head *t_bh,
4097                                      bool preserve)
4098 {
4099         int ret;
4100         struct buffer_head *ref_root_bh = NULL;
4101         struct ocfs2_cached_dealloc_ctxt dealloc;
4102         struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4103         struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
4104         struct ocfs2_refcount_tree *ref_tree;
4105
4106         ocfs2_init_dealloc_ctxt(&dealloc);
4107
4108         ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4109                                       le64_to_cpu(di->i_refcount_loc));
4110         if (ret) {
4111                 mlog_errno(ret);
4112                 goto out;
4113         }
4114
4115         if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4116                 ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
4117                                                   t_inode, t_bh);
4118                 if (ret)
4119                         mlog_errno(ret);
4120                 goto out;
4121         }
4122
4123         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
4124                                        1, &ref_tree, &ref_root_bh);
4125         if (ret) {
4126                 mlog_errno(ret);
4127                 goto out;
4128         }
4129
4130         ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
4131                                           &ref_tree->rf_ci, ref_root_bh,
4132                                           &dealloc);
4133         if (ret) {
4134                 mlog_errno(ret);
4135                 goto out_unlock_refcount;
4136         }
4137
4138 out_unlock_refcount:
4139         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4140         brelse(ref_root_bh);
4141 out:
4142         if (ocfs2_dealloc_has_cluster(&dealloc)) {
4143                 ocfs2_schedule_truncate_log_flush(osb, 1);
4144                 ocfs2_run_deallocs(osb, &dealloc);
4145         }
4146
4147         return ret;
4148 }
4149
4150 static int __ocfs2_reflink(struct dentry *old_dentry,
4151                            struct buffer_head *old_bh,
4152                            struct inode *new_inode,
4153                            bool preserve)
4154 {
4155         int ret;
4156         struct inode *inode = d_inode(old_dentry);
4157         struct buffer_head *new_bh = NULL;
4158
4159         if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
4160                 ret = -EINVAL;
4161                 mlog_errno(ret);
4162                 goto out;
4163         }
4164
4165         ret = filemap_fdatawrite(inode->i_mapping);
4166         if (ret) {
4167                 mlog_errno(ret);
4168                 goto out;
4169         }
4170
4171         ret = ocfs2_attach_refcount_tree(inode, old_bh);
4172         if (ret) {
4173                 mlog_errno(ret);
4174                 goto out;
4175         }
4176
4177         inode_lock_nested(new_inode, I_MUTEX_CHILD);
4178         ret = ocfs2_inode_lock_nested(new_inode, &new_bh, 1,
4179                                       OI_LS_REFLINK_TARGET);
4180         if (ret) {
4181                 mlog_errno(ret);
4182                 goto out_unlock;
4183         }
4184
4185         ret = ocfs2_create_reflink_node(inode, old_bh,
4186                                         new_inode, new_bh, preserve);
4187         if (ret) {
4188                 mlog_errno(ret);
4189                 goto inode_unlock;
4190         }
4191
4192         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
4193                 ret = ocfs2_reflink_xattrs(inode, old_bh,
4194                                            new_inode, new_bh,
4195                                            preserve);
4196                 if (ret) {
4197                         mlog_errno(ret);
4198                         goto inode_unlock;
4199                 }
4200         }
4201
4202         ret = ocfs2_complete_reflink(inode, old_bh,
4203                                      new_inode, new_bh, preserve);
4204         if (ret)
4205                 mlog_errno(ret);
4206
4207 inode_unlock:
4208         ocfs2_inode_unlock(new_inode, 1);
4209         brelse(new_bh);
4210 out_unlock:
4211         inode_unlock(new_inode);
4212 out:
4213         if (!ret) {
4214                 ret = filemap_fdatawait(inode->i_mapping);
4215                 if (ret)
4216                         mlog_errno(ret);
4217         }
4218         return ret;
4219 }
4220
4221 static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
4222                          struct dentry *new_dentry, bool preserve)
4223 {
4224         int error, had_lock;
4225         struct inode *inode = d_inode(old_dentry);
4226         struct buffer_head *old_bh = NULL;
4227         struct inode *new_orphan_inode = NULL;
4228         struct ocfs2_lock_holder oh;
4229
4230         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4231                 return -EOPNOTSUPP;
4232
4233
4234         error = ocfs2_create_inode_in_orphan(dir, inode->i_mode,
4235                                              &new_orphan_inode);
4236         if (error) {
4237                 mlog_errno(error);
4238                 goto out;
4239         }
4240
4241         error = ocfs2_rw_lock(inode, 1);
4242         if (error) {
4243                 mlog_errno(error);
4244                 goto out;
4245         }
4246
4247         error = ocfs2_inode_lock(inode, &old_bh, 1);
4248         if (error) {
4249                 mlog_errno(error);
4250                 ocfs2_rw_unlock(inode, 1);
4251                 goto out;
4252         }
4253
4254         down_write(&OCFS2_I(inode)->ip_xattr_sem);
4255         down_write(&OCFS2_I(inode)->ip_alloc_sem);
4256         error = __ocfs2_reflink(old_dentry, old_bh,
4257                                 new_orphan_inode, preserve);
4258         up_write(&OCFS2_I(inode)->ip_alloc_sem);
4259         up_write(&OCFS2_I(inode)->ip_xattr_sem);
4260
4261         ocfs2_inode_unlock(inode, 1);
4262         ocfs2_rw_unlock(inode, 1);
4263         brelse(old_bh);
4264
4265         if (error) {
4266                 mlog_errno(error);
4267                 goto out;
4268         }
4269
4270         had_lock = ocfs2_inode_lock_tracker(new_orphan_inode, NULL, 1,
4271                                             &oh);
4272         if (had_lock < 0) {
4273                 error = had_lock;
4274                 mlog_errno(error);
4275                 goto out;
4276         }
4277
4278         /* If the security isn't preserved, we need to re-initialize them. */
4279         if (!preserve) {
4280                 error = ocfs2_init_security_and_acl(dir, new_orphan_inode,
4281                                                     &new_dentry->d_name);
4282                 if (error)
4283                         mlog_errno(error);
4284         }
4285         if (!error) {
4286                 error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
4287                                                        new_dentry);
4288                 if (error)
4289                         mlog_errno(error);
4290         }
4291         ocfs2_inode_unlock_tracker(new_orphan_inode, 1, &oh, had_lock);
4292
4293 out:
4294         if (new_orphan_inode) {
4295                 /*
4296                  * We need to open_unlock the inode no matter whether we
4297                  * succeed or not, so that other nodes can delete it later.
4298                  */
4299                 ocfs2_open_unlock(new_orphan_inode);
4300                 if (error)
4301                         iput(new_orphan_inode);
4302         }
4303
4304         return error;
4305 }
4306
4307 /*
4308  * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
4309  * sys_reflink().  This will go away when vfs_reflink() exists in
4310  * fs/namei.c.
4311  */
4312
4313 /* copied from may_create in VFS. */
4314 static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
4315 {
4316         if (d_really_is_positive(child))
4317                 return -EEXIST;
4318         if (IS_DEADDIR(dir))
4319                 return -ENOENT;
4320         return inode_permission(&nop_mnt_idmap, dir, MAY_WRITE | MAY_EXEC);
4321 }
4322
4323 /**
4324  * ocfs2_vfs_reflink - Create a reference-counted link
4325  *
4326  * @old_dentry:        source dentry + inode
4327  * @dir:       directory to create the target
4328  * @new_dentry:        target dentry
4329  * @preserve:  if true, preserve all file attributes
4330  */
4331 static int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
4332                              struct dentry *new_dentry, bool preserve)
4333 {
4334         struct inode *inode = d_inode(old_dentry);
4335         int error;
4336
4337         if (!inode)
4338                 return -ENOENT;
4339
4340         error = ocfs2_may_create(dir, new_dentry);
4341         if (error)
4342                 return error;
4343
4344         if (dir->i_sb != inode->i_sb)
4345                 return -EXDEV;
4346
4347         /*
4348          * A reflink to an append-only or immutable file cannot be created.
4349          */
4350         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4351                 return -EPERM;
4352
4353         /* Only regular files can be reflinked. */
4354         if (!S_ISREG(inode->i_mode))
4355                 return -EPERM;
4356
4357         /*
4358          * If the caller wants to preserve ownership, they require the
4359          * rights to do so.
4360          */
4361         if (preserve) {
4362                 if (!uid_eq(current_fsuid(), inode->i_uid) && !capable(CAP_CHOWN))
4363                         return -EPERM;
4364                 if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
4365                         return -EPERM;
4366         }
4367
4368         /*
4369          * If the caller is modifying any aspect of the attributes, they
4370          * are not creating a snapshot.  They need read permission on the
4371          * file.
4372          */
4373         if (!preserve) {
4374                 error = inode_permission(&nop_mnt_idmap, inode, MAY_READ);
4375                 if (error)
4376                         return error;
4377         }
4378
4379         inode_lock(inode);
4380         error = dquot_initialize(dir);
4381         if (!error)
4382                 error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
4383         inode_unlock(inode);
4384         if (!error)
4385                 fsnotify_create(dir, new_dentry);
4386         return error;
4387 }
4388 /*
4389  * Most codes are copied from sys_linkat.
4390  */
4391 int ocfs2_reflink_ioctl(struct inode *inode,
4392                         const char __user *oldname,
4393                         const char __user *newname,
4394                         bool preserve)
4395 {
4396         struct dentry *new_dentry;
4397         struct path old_path, new_path;
4398         int error;
4399
4400         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4401                 return -EOPNOTSUPP;
4402
4403         error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
4404         if (error) {
4405                 mlog_errno(error);
4406                 return error;
4407         }
4408
4409         new_dentry = user_path_create(AT_FDCWD, newname, &new_path, 0);
4410         error = PTR_ERR(new_dentry);
4411         if (IS_ERR(new_dentry)) {
4412                 mlog_errno(error);
4413                 goto out;
4414         }
4415
4416         error = -EXDEV;
4417         if (old_path.mnt != new_path.mnt) {
4418                 mlog_errno(error);
4419                 goto out_dput;
4420         }
4421
4422         error = ocfs2_vfs_reflink(old_path.dentry,
4423                                   d_inode(new_path.dentry),
4424                                   new_dentry, preserve);
4425 out_dput:
4426         done_path_create(&new_path, new_dentry);
4427 out:
4428         path_put(&old_path);
4429
4430         return error;
4431 }
4432
4433 /* Update destination inode size, if necessary. */
4434 int ocfs2_reflink_update_dest(struct inode *dest,
4435                               struct buffer_head *d_bh,
4436                               loff_t newlen)
4437 {
4438         handle_t *handle;
4439         int ret;
4440
4441         dest->i_blocks = ocfs2_inode_sector_count(dest);
4442
4443         if (newlen <= i_size_read(dest))
4444                 return 0;
4445
4446         handle = ocfs2_start_trans(OCFS2_SB(dest->i_sb),
4447                                    OCFS2_INODE_UPDATE_CREDITS);
4448         if (IS_ERR(handle)) {
4449                 ret = PTR_ERR(handle);
4450                 mlog_errno(ret);
4451                 return ret;
4452         }
4453
4454         /* Extend i_size if needed. */
4455         spin_lock(&OCFS2_I(dest)->ip_lock);
4456         if (newlen > i_size_read(dest))
4457                 i_size_write(dest, newlen);
4458         spin_unlock(&OCFS2_I(dest)->ip_lock);
4459         dest->i_mtime = inode_set_ctime_current(dest);
4460
4461         ret = ocfs2_mark_inode_dirty(handle, dest, d_bh);
4462         if (ret) {
4463                 mlog_errno(ret);
4464                 goto out_commit;
4465         }
4466
4467 out_commit:
4468         ocfs2_commit_trans(OCFS2_SB(dest->i_sb), handle);
4469         return ret;
4470 }
4471
4472 /* Remap the range pos_in:len in s_inode to pos_out:len in t_inode. */
4473 static loff_t ocfs2_reflink_remap_extent(struct inode *s_inode,
4474                                          struct buffer_head *s_bh,
4475                                          loff_t pos_in,
4476                                          struct inode *t_inode,
4477                                          struct buffer_head *t_bh,
4478                                          loff_t pos_out,
4479                                          loff_t len,
4480                                          struct ocfs2_cached_dealloc_ctxt *dealloc)
4481 {
4482         struct ocfs2_extent_tree s_et;
4483         struct ocfs2_extent_tree t_et;
4484         struct ocfs2_dinode *dis;
4485         struct buffer_head *ref_root_bh = NULL;
4486         struct ocfs2_refcount_tree *ref_tree;
4487         struct ocfs2_super *osb;
4488         loff_t remapped_bytes = 0;
4489         loff_t pstart, plen;
4490         u32 p_cluster, num_clusters, slast, spos, tpos, remapped_clus = 0;
4491         unsigned int ext_flags;
4492         int ret = 0;
4493
4494         osb = OCFS2_SB(s_inode->i_sb);
4495         dis = (struct ocfs2_dinode *)s_bh->b_data;
4496         ocfs2_init_dinode_extent_tree(&s_et, INODE_CACHE(s_inode), s_bh);
4497         ocfs2_init_dinode_extent_tree(&t_et, INODE_CACHE(t_inode), t_bh);
4498
4499         spos = ocfs2_bytes_to_clusters(s_inode->i_sb, pos_in);
4500         tpos = ocfs2_bytes_to_clusters(t_inode->i_sb, pos_out);
4501         slast = ocfs2_clusters_for_bytes(s_inode->i_sb, pos_in + len);
4502
4503         while (spos < slast) {
4504                 if (fatal_signal_pending(current)) {
4505                         ret = -EINTR;
4506                         goto out;
4507                 }
4508
4509                 /* Look up the extent. */
4510                 ret = ocfs2_get_clusters(s_inode, spos, &p_cluster,
4511                                          &num_clusters, &ext_flags);
4512                 if (ret) {
4513                         mlog_errno(ret);
4514                         goto out;
4515                 }
4516
4517                 num_clusters = min_t(u32, num_clusters, slast - spos);
4518
4519                 /* Punch out the dest range. */
4520                 pstart = ocfs2_clusters_to_bytes(t_inode->i_sb, tpos);
4521                 plen = ocfs2_clusters_to_bytes(t_inode->i_sb, num_clusters);
4522                 ret = ocfs2_remove_inode_range(t_inode, t_bh, pstart, plen);
4523                 if (ret) {
4524                         mlog_errno(ret);
4525                         goto out;
4526                 }
4527
4528                 if (p_cluster == 0)
4529                         goto next_loop;
4530
4531                 /* Lock the refcount btree... */
4532                 ret = ocfs2_lock_refcount_tree(osb,
4533                                                le64_to_cpu(dis->i_refcount_loc),
4534                                                1, &ref_tree, &ref_root_bh);
4535                 if (ret) {
4536                         mlog_errno(ret);
4537                         goto out;
4538                 }
4539
4540                 /* Mark s_inode's extent as refcounted. */
4541                 if (!(ext_flags & OCFS2_EXT_REFCOUNTED)) {
4542                         ret = ocfs2_add_refcount_flag(s_inode, &s_et,
4543                                                       &ref_tree->rf_ci,
4544                                                       ref_root_bh, spos,
4545                                                       p_cluster, num_clusters,
4546                                                       dealloc, NULL);
4547                         if (ret) {
4548                                 mlog_errno(ret);
4549                                 goto out_unlock_refcount;
4550                         }
4551                 }
4552
4553                 /* Map in the new extent. */
4554                 ext_flags |= OCFS2_EXT_REFCOUNTED;
4555                 ret = ocfs2_add_refcounted_extent(t_inode, &t_et,
4556                                                   &ref_tree->rf_ci,
4557                                                   ref_root_bh,
4558                                                   tpos, p_cluster,
4559                                                   num_clusters,
4560                                                   ext_flags,
4561                                                   dealloc);
4562                 if (ret) {
4563                         mlog_errno(ret);
4564                         goto out_unlock_refcount;
4565                 }
4566
4567                 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4568                 brelse(ref_root_bh);
4569 next_loop:
4570                 spos += num_clusters;
4571                 tpos += num_clusters;
4572                 remapped_clus += num_clusters;
4573         }
4574
4575         goto out;
4576 out_unlock_refcount:
4577         ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4578         brelse(ref_root_bh);
4579 out:
4580         remapped_bytes = ocfs2_clusters_to_bytes(t_inode->i_sb, remapped_clus);
4581         remapped_bytes = min_t(loff_t, len, remapped_bytes);
4582
4583         return remapped_bytes > 0 ? remapped_bytes : ret;
4584 }
4585
4586 /* Set up refcount tree and remap s_inode to t_inode. */
4587 loff_t ocfs2_reflink_remap_blocks(struct inode *s_inode,
4588                                   struct buffer_head *s_bh,
4589                                   loff_t pos_in,
4590                                   struct inode *t_inode,
4591                                   struct buffer_head *t_bh,
4592                                   loff_t pos_out,
4593                                   loff_t len)
4594 {
4595         struct ocfs2_cached_dealloc_ctxt dealloc;
4596         struct ocfs2_super *osb;
4597         struct ocfs2_dinode *dis;
4598         struct ocfs2_dinode *dit;
4599         loff_t ret;
4600
4601         osb = OCFS2_SB(s_inode->i_sb);
4602         dis = (struct ocfs2_dinode *)s_bh->b_data;
4603         dit = (struct ocfs2_dinode *)t_bh->b_data;
4604         ocfs2_init_dealloc_ctxt(&dealloc);
4605
4606         /*
4607          * If we're reflinking the entire file and the source is inline
4608          * data, just copy the contents.
4609          */
4610         if (pos_in == pos_out && pos_in == 0 && len == i_size_read(s_inode) &&
4611             i_size_read(t_inode) <= len &&
4612             (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) {
4613                 ret = ocfs2_duplicate_inline_data(s_inode, s_bh, t_inode, t_bh);
4614                 if (ret)
4615                         mlog_errno(ret);
4616                 goto out;
4617         }
4618
4619         /*
4620          * If both inodes belong to two different refcount groups then
4621          * forget it because we don't know how (or want) to go merging
4622          * refcount trees.
4623          */
4624         ret = -EOPNOTSUPP;
4625         if (ocfs2_is_refcount_inode(s_inode) &&
4626             ocfs2_is_refcount_inode(t_inode) &&
4627             le64_to_cpu(dis->i_refcount_loc) !=
4628             le64_to_cpu(dit->i_refcount_loc))
4629                 goto out;
4630
4631         /* Neither inode has a refcount tree.  Add one to s_inode. */
4632         if (!ocfs2_is_refcount_inode(s_inode) &&
4633             !ocfs2_is_refcount_inode(t_inode)) {
4634                 ret = ocfs2_create_refcount_tree(s_inode, s_bh);
4635                 if (ret) {
4636                         mlog_errno(ret);
4637                         goto out;
4638                 }
4639         }
4640
4641         /* Ensure that both inodes end up with the same refcount tree. */
4642         if (!ocfs2_is_refcount_inode(s_inode)) {
4643                 ret = ocfs2_set_refcount_tree(s_inode, s_bh,
4644                                               le64_to_cpu(dit->i_refcount_loc));
4645                 if (ret) {
4646                         mlog_errno(ret);
4647                         goto out;
4648                 }
4649         }
4650         if (!ocfs2_is_refcount_inode(t_inode)) {
4651                 ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4652                                               le64_to_cpu(dis->i_refcount_loc));
4653                 if (ret) {
4654                         mlog_errno(ret);
4655                         goto out;
4656                 }
4657         }
4658
4659         /* Turn off inline data in the dest file. */
4660         if (OCFS2_I(t_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4661                 ret = ocfs2_convert_inline_data_to_extents(t_inode, t_bh);
4662                 if (ret) {
4663                         mlog_errno(ret);
4664                         goto out;
4665                 }
4666         }
4667
4668         /* Actually remap extents now. */
4669         ret = ocfs2_reflink_remap_extent(s_inode, s_bh, pos_in, t_inode, t_bh,
4670                                          pos_out, len, &dealloc);
4671         if (ret < 0) {
4672                 mlog_errno(ret);
4673                 goto out;
4674         }
4675
4676 out:
4677         if (ocfs2_dealloc_has_cluster(&dealloc)) {
4678                 ocfs2_schedule_truncate_log_flush(osb, 1);
4679                 ocfs2_run_deallocs(osb, &dealloc);
4680         }
4681
4682         return ret;
4683 }
4684
4685 /* Lock an inode and grab a bh pointing to the inode. */
4686 int ocfs2_reflink_inodes_lock(struct inode *s_inode,
4687                               struct buffer_head **bh_s,
4688                               struct inode *t_inode,
4689                               struct buffer_head **bh_t)
4690 {
4691         struct inode *inode1 = s_inode;
4692         struct inode *inode2 = t_inode;
4693         struct ocfs2_inode_info *oi1;
4694         struct ocfs2_inode_info *oi2;
4695         struct buffer_head *bh1 = NULL;
4696         struct buffer_head *bh2 = NULL;
4697         bool same_inode = (s_inode == t_inode);
4698         bool need_swap = (inode1->i_ino > inode2->i_ino);
4699         int status;
4700
4701         /* First grab the VFS and rw locks. */
4702         lock_two_nondirectories(s_inode, t_inode);
4703         if (need_swap)
4704                 swap(inode1, inode2);
4705
4706         status = ocfs2_rw_lock(inode1, 1);
4707         if (status) {
4708                 mlog_errno(status);
4709                 goto out_i1;
4710         }
4711         if (!same_inode) {
4712                 status = ocfs2_rw_lock(inode2, 1);
4713                 if (status) {
4714                         mlog_errno(status);
4715                         goto out_i2;
4716                 }
4717         }
4718
4719         /* Now go for the cluster locks */
4720         oi1 = OCFS2_I(inode1);
4721         oi2 = OCFS2_I(inode2);
4722
4723         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
4724                                 (unsigned long long)oi2->ip_blkno);
4725
4726         /* We always want to lock the one with the lower lockid first. */
4727         if (oi1->ip_blkno > oi2->ip_blkno)
4728                 mlog_errno(-ENOLCK);
4729
4730         /* lock id1 */
4731         status = ocfs2_inode_lock_nested(inode1, &bh1, 1,
4732                                          OI_LS_REFLINK_TARGET);
4733         if (status < 0) {
4734                 if (status != -ENOENT)
4735                         mlog_errno(status);
4736                 goto out_rw2;
4737         }
4738
4739         /* lock id2 */
4740         if (!same_inode) {
4741                 status = ocfs2_inode_lock_nested(inode2, &bh2, 1,
4742                                                  OI_LS_REFLINK_TARGET);
4743                 if (status < 0) {
4744                         if (status != -ENOENT)
4745                                 mlog_errno(status);
4746                         goto out_cl1;
4747                 }
4748         } else {
4749                 bh2 = bh1;
4750         }
4751
4752         /*
4753          * If we swapped inode order above, we have to swap the buffer heads
4754          * before passing them back to the caller.
4755          */
4756         if (need_swap)
4757                 swap(bh1, bh2);
4758         *bh_s = bh1;
4759         *bh_t = bh2;
4760
4761         trace_ocfs2_double_lock_end(
4762                         (unsigned long long)oi1->ip_blkno,
4763                         (unsigned long long)oi2->ip_blkno);
4764
4765         return 0;
4766
4767 out_cl1:
4768         ocfs2_inode_unlock(inode1, 1);
4769         brelse(bh1);
4770 out_rw2:
4771         ocfs2_rw_unlock(inode2, 1);
4772 out_i2:
4773         ocfs2_rw_unlock(inode1, 1);
4774 out_i1:
4775         unlock_two_nondirectories(s_inode, t_inode);
4776         return status;
4777 }
4778
4779 /* Unlock both inodes and release buffers. */
4780 void ocfs2_reflink_inodes_unlock(struct inode *s_inode,
4781                                  struct buffer_head *s_bh,
4782                                  struct inode *t_inode,
4783                                  struct buffer_head *t_bh)
4784 {
4785         ocfs2_inode_unlock(s_inode, 1);
4786         ocfs2_rw_unlock(s_inode, 1);
4787         brelse(s_bh);
4788         if (s_inode != t_inode) {
4789                 ocfs2_inode_unlock(t_inode, 1);
4790                 ocfs2_rw_unlock(t_inode, 1);
4791                 brelse(t_bh);
4792         }
4793         unlock_two_nondirectories(s_inode, t_inode);
4794 }