fs/f2fs/node.c

   1 /*
   2  * fs/f2fs/node.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/mpage.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/swap.h>
  18
  19 #include "f2fs.h"
  20 #include "node.h"
  21 #include "segment.h"
  22 #include <trace/events/f2fs.h>
  23
  24 static struct kmem_cache *nat_entry_slab;
  25 static struct kmem_cache *free_nid_slab;
  26
  27 static void clear_node_page_dirty(struct page *page)
  28 {
  29         struct address_space *mapping = page->mapping;
  30         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
  31         unsigned int long flags;
  32
  33         if (PageDirty(page)) {
  34                 spin_lock_irqsave(&mapping->tree_lock, flags);
  35                 radix_tree_tag_clear(&mapping->page_tree,
  36                                 page_index(page),
  37                                 PAGECACHE_TAG_DIRTY);
  38                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  39
  40                 clear_page_dirty_for_io(page);
  41                 dec_page_count(sbi, F2FS_DIRTY_NODES);
  42         }
  43         ClearPageUptodate(page);
  44 }
  45
  46 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  47 {
  48         pgoff_t index = current_nat_addr(sbi, nid);
  49         return get_meta_page(sbi, index);
  50 }
  51
  52 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  53 {
  54         struct page *src_page;
  55         struct page *dst_page;
  56         pgoff_t src_off;
  57         pgoff_t dst_off;
  58         void *src_addr;
  59         void *dst_addr;
  60         struct f2fs_nm_info *nm_i = NM_I(sbi);
  61
  62         src_off = current_nat_addr(sbi, nid);
  63         dst_off = next_nat_addr(sbi, src_off);
  64
  65         /* get current nat block page with lock */
  66         src_page = get_meta_page(sbi, src_off);
  67
  68         /* Dirty src_page means that it is already the new target NAT page. */
  69         if (PageDirty(src_page))
  70                 return src_page;
  71
  72         dst_page = grab_meta_page(sbi, dst_off);
  73
  74         src_addr = page_address(src_page);
  75         dst_addr = page_address(dst_page);
  76         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
  77         set_page_dirty(dst_page);
  78         f2fs_put_page(src_page, 1);
  79
  80         set_to_next_nat(nm_i, nid);
  81
  82         return dst_page;
  83 }
  84
  85 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
  86 {
  87         return radix_tree_lookup(&nm_i->nat_root, n);
  88 }
  89
  90 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
  91                 nid_t start, unsigned int nr, struct nat_entry **ep)
  92 {
  93         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
  94 }
  95
  96 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
  97 {
  98         list_del(&e->list);
  99         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 100         nm_i->nat_cnt--;
 101         kmem_cache_free(nat_entry_slab, e);
 102 }
 103
 104 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 105 {
 106         struct f2fs_nm_info *nm_i = NM_I(sbi);
 107         struct nat_entry *e;
 108         int is_cp = 1;
 109
 110         read_lock(&nm_i->nat_tree_lock);
 111         e = __lookup_nat_cache(nm_i, nid);
 112         if (e && !e->checkpointed)
 113                 is_cp = 0;
 114         read_unlock(&nm_i->nat_tree_lock);
 115         return is_cp;
 116 }
 117
 118 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 119 {
 120         struct nat_entry *new;
 121
 122         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 123         if (!new)
 124                 return NULL;
 125         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 126                 kmem_cache_free(nat_entry_slab, new);
 127                 return NULL;
 128         }
 129         memset(new, 0, sizeof(struct nat_entry));
 130         nat_set_nid(new, nid);
 131         list_add_tail(&new->list, &nm_i->nat_entries);
 132         nm_i->nat_cnt++;
 133         return new;
 134 }
 135
 136 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 137                                                 struct f2fs_nat_entry *ne)
 138 {
 139         struct nat_entry *e;
 140 retry:
 141         write_lock(&nm_i->nat_tree_lock);
 142         e = __lookup_nat_cache(nm_i, nid);
 143         if (!e) {
 144                 e = grab_nat_entry(nm_i, nid);
 145                 if (!e) {
 146                         write_unlock(&nm_i->nat_tree_lock);
 147                         goto retry;
 148                 }
 149                 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
 150                 nat_set_ino(e, le32_to_cpu(ne->ino));
 151                 nat_set_version(e, ne->version);
 152                 e->checkpointed = true;
 153         }
 154         write_unlock(&nm_i->nat_tree_lock);
 155 }
 156
 157 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 158                         block_t new_blkaddr)
 159 {
 160         struct f2fs_nm_info *nm_i = NM_I(sbi);
 161         struct nat_entry *e;
 162 retry:
 163         write_lock(&nm_i->nat_tree_lock);
 164         e = __lookup_nat_cache(nm_i, ni->nid);
 165         if (!e) {
 166                 e = grab_nat_entry(nm_i, ni->nid);
 167                 if (!e) {
 168                         write_unlock(&nm_i->nat_tree_lock);
 169                         goto retry;
 170                 }
 171                 e->ni = *ni;
 172                 e->checkpointed = true;
 173                 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
 174         } else if (new_blkaddr == NEW_ADDR) {
 175                 /*
 176                  * when nid is reallocated,
 177                  * previous nat entry can be remained in nat cache.
 178                  * So, reinitialize it with new information.
 179                  */
 180                 e->ni = *ni;
 181                 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
 182         }
 183
 184         if (new_blkaddr == NEW_ADDR)
 185                 e->checkpointed = false;
 186
 187         /* sanity check */
 188         f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
 189         f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
 190                         new_blkaddr == NULL_ADDR);
 191         f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
 192                         new_blkaddr == NEW_ADDR);
 193         f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
 194                         nat_get_blkaddr(e) != NULL_ADDR &&
 195                         new_blkaddr == NEW_ADDR);
 196
 197         /* increament version no as node is removed */
 198         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 199                 unsigned char version = nat_get_version(e);
 200                 nat_set_version(e, inc_node_version(version));
 201         }
 202
 203         /* change address */
 204         nat_set_blkaddr(e, new_blkaddr);
 205         __set_nat_cache_dirty(nm_i, e);
 206         write_unlock(&nm_i->nat_tree_lock);
 207 }
 208
 209 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 210 {
 211         struct f2fs_nm_info *nm_i = NM_I(sbi);
 212
 213         if (nm_i->nat_cnt <= NM_WOUT_THRESHOLD)
 214                 return 0;
 215
 216         write_lock(&nm_i->nat_tree_lock);
 217         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 218                 struct nat_entry *ne;
 219                 ne = list_first_entry(&nm_i->nat_entries,
 220                                         struct nat_entry, list);
 221                 __del_from_nat_cache(nm_i, ne);
 222                 nr_shrink--;
 223         }
 224         write_unlock(&nm_i->nat_tree_lock);
 225         return nr_shrink;
 226 }
 227
 228 /*
 229  * This function returns always success
 230  */
 231 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 232 {
 233         struct f2fs_nm_info *nm_i = NM_I(sbi);
 234         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 235         struct f2fs_summary_block *sum = curseg->sum_blk;
 236         nid_t start_nid = START_NID(nid);
 237         struct f2fs_nat_block *nat_blk;
 238         struct page *page = NULL;
 239         struct f2fs_nat_entry ne;
 240         struct nat_entry *e;
 241         int i;
 242
 243         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 244         ni->nid = nid;
 245
 246         /* Check nat cache */
 247         read_lock(&nm_i->nat_tree_lock);
 248         e = __lookup_nat_cache(nm_i, nid);
 249         if (e) {
 250                 ni->ino = nat_get_ino(e);
 251                 ni->blk_addr = nat_get_blkaddr(e);
 252                 ni->version = nat_get_version(e);
 253         }
 254         read_unlock(&nm_i->nat_tree_lock);
 255         if (e)
 256                 return;
 257
 258         /* Check current segment summary */
 259         mutex_lock(&curseg->curseg_mutex);
 260         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 261         if (i >= 0) {
 262                 ne = nat_in_journal(sum, i);
 263                 node_info_from_raw_nat(ni, &ne);
 264         }
 265         mutex_unlock(&curseg->curseg_mutex);
 266         if (i >= 0)
 267                 goto cache;
 268
 269         /* Fill node_info from nat page */
 270         page = get_current_nat_page(sbi, start_nid);
 271         nat_blk = (struct f2fs_nat_block *)page_address(page);
 272         ne = nat_blk->entries[nid - start_nid];
 273         node_info_from_raw_nat(ni, &ne);
 274         f2fs_put_page(page, 1);
 275 cache:
 276         /* cache nat entry */
 277         cache_nat_entry(NM_I(sbi), nid, &ne);
 278 }
 279
 280 /*
 281  * The maximum depth is four.
 282  * Offset[0] will have raw inode offset.
 283  */
 284 static int get_node_path(struct f2fs_inode_info *fi, long block,
 285                                 int offset[4], unsigned int noffset[4])
 286 {
 287         const long direct_index = ADDRS_PER_INODE(fi);
 288         const long direct_blks = ADDRS_PER_BLOCK;
 289         const long dptrs_per_blk = NIDS_PER_BLOCK;
 290         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 291         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 292         int n = 0;
 293         int level = 0;
 294
 295         noffset[0] = 0;
 296
 297         if (block < direct_index) {
 298                 offset[n] = block;
 299                 goto got;
 300         }
 301         block -= direct_index;
 302         if (block < direct_blks) {
 303                 offset[n++] = NODE_DIR1_BLOCK;
 304                 noffset[n] = 1;
 305                 offset[n] = block;
 306                 level = 1;
 307                 goto got;
 308         }
 309         block -= direct_blks;
 310         if (block < direct_blks) {
 311                 offset[n++] = NODE_DIR2_BLOCK;
 312                 noffset[n] = 2;
 313                 offset[n] = block;
 314                 level = 1;
 315                 goto got;
 316         }
 317         block -= direct_blks;
 318         if (block < indirect_blks) {
 319                 offset[n++] = NODE_IND1_BLOCK;
 320                 noffset[n] = 3;
 321                 offset[n++] = block / direct_blks;
 322                 noffset[n] = 4 + offset[n - 1];
 323                 offset[n] = block % direct_blks;
 324                 level = 2;
 325                 goto got;
 326         }
 327         block -= indirect_blks;
 328         if (block < indirect_blks) {
 329                 offset[n++] = NODE_IND2_BLOCK;
 330                 noffset[n] = 4 + dptrs_per_blk;
 331                 offset[n++] = block / direct_blks;
 332                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 333                 offset[n] = block % direct_blks;
 334                 level = 2;
 335                 goto got;
 336         }
 337         block -= indirect_blks;
 338         if (block < dindirect_blks) {
 339                 offset[n++] = NODE_DIND_BLOCK;
 340                 noffset[n] = 5 + (dptrs_per_blk * 2);
 341                 offset[n++] = block / indirect_blks;
 342                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 343                               offset[n - 1] * (dptrs_per_blk + 1);
 344                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 345                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 346                               offset[n - 2] * (dptrs_per_blk + 1) +
 347                               offset[n - 1];
 348                 offset[n] = block % direct_blks;
 349                 level = 3;
 350                 goto got;
 351         } else {
 352                 BUG();
 353         }
 354 got:
 355         return level;
 356 }
 357
 358 /*
 359  * Caller should call f2fs_put_dnode(dn).
 360  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 361  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 362  * In the case of RDONLY_NODE, we don't need to care about mutex.
 363  */
 364 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 365 {
 366         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 367         struct page *npage[4];
 368         struct page *parent;
 369         int offset[4];
 370         unsigned int noffset[4];
 371         nid_t nids[4];
 372         int level, i;
 373         int err = 0;
 374
 375         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 376
 377         nids[0] = dn->inode->i_ino;
 378         npage[0] = dn->inode_page;
 379
 380         if (!npage[0]) {
 381                 npage[0] = get_node_page(sbi, nids[0]);
 382                 if (IS_ERR(npage[0]))
 383                         return PTR_ERR(npage[0]);
 384         }
 385         parent = npage[0];
 386         if (level != 0)
 387                 nids[1] = get_nid(parent, offset[0], true);
 388         dn->inode_page = npage[0];
 389         dn->inode_page_locked = true;
 390
 391         /* get indirect or direct nodes */
 392         for (i = 1; i <= level; i++) {
 393                 bool done = false;
 394
 395                 if (!nids[i] && mode == ALLOC_NODE) {
 396                         /* alloc new node */
 397                         if (!alloc_nid(sbi, &(nids[i]))) {
 398                                 err = -ENOSPC;
 399                                 goto release_pages;
 400                         }
 401
 402                         dn->nid = nids[i];
 403                         npage[i] = new_node_page(dn, noffset[i], NULL);
 404                         if (IS_ERR(npage[i])) {
 405                                 alloc_nid_failed(sbi, nids[i]);
 406                                 err = PTR_ERR(npage[i]);
 407                                 goto release_pages;
 408                         }
 409
 410                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 411                         alloc_nid_done(sbi, nids[i]);
 412                         done = true;
 413                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 414                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 415                         if (IS_ERR(npage[i])) {
 416                                 err = PTR_ERR(npage[i]);
 417                                 goto release_pages;
 418                         }
 419                         done = true;
 420                 }
 421                 if (i == 1) {
 422                         dn->inode_page_locked = false;
 423                         unlock_page(parent);
 424                 } else {
 425                         f2fs_put_page(parent, 1);
 426                 }
 427
 428                 if (!done) {
 429                         npage[i] = get_node_page(sbi, nids[i]);
 430                         if (IS_ERR(npage[i])) {
 431                                 err = PTR_ERR(npage[i]);
 432                                 f2fs_put_page(npage[0], 0);
 433                                 goto release_out;
 434                         }
 435                 }
 436                 if (i < level) {
 437                         parent = npage[i];
 438                         nids[i + 1] = get_nid(parent, offset[i], false);
 439                 }
 440         }
 441         dn->nid = nids[level];
 442         dn->ofs_in_node = offset[level];
 443         dn->node_page = npage[level];
 444         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 445         return 0;
 446
 447 release_pages:
 448         f2fs_put_page(parent, 1);
 449         if (i > 1)
 450                 f2fs_put_page(npage[0], 0);
 451 release_out:
 452         dn->inode_page = NULL;
 453         dn->node_page = NULL;
 454         return err;
 455 }
 456
 457 static void truncate_node(struct dnode_of_data *dn)
 458 {
 459         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 460         struct node_info ni;
 461
 462         get_node_info(sbi, dn->nid, &ni);
 463         if (dn->inode->i_blocks == 0) {
 464                 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
 465                 goto invalidate;
 466         }
 467         f2fs_bug_on(ni.blk_addr == NULL_ADDR);
 468
 469         /* Deallocate node address */
 470         invalidate_blocks(sbi, ni.blk_addr);
 471         dec_valid_node_count(sbi, dn->inode);
 472         set_node_addr(sbi, &ni, NULL_ADDR);
 473
 474         if (dn->nid == dn->inode->i_ino) {
 475                 remove_orphan_inode(sbi, dn->nid);
 476                 dec_valid_inode_count(sbi);
 477         } else {
 478                 sync_inode_page(dn);
 479         }
 480 invalidate:
 481         clear_node_page_dirty(dn->node_page);
 482         F2FS_SET_SB_DIRT(sbi);
 483
 484         f2fs_put_page(dn->node_page, 1);
 485
 486         invalidate_mapping_pages(NODE_MAPPING(sbi),
 487                         dn->node_page->index, dn->node_page->index);
 488
 489         dn->node_page = NULL;
 490         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 491 }
 492
 493 static int truncate_dnode(struct dnode_of_data *dn)
 494 {
 495         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 496         struct page *page;
 497
 498         if (dn->nid == 0)
 499                 return 1;
 500
 501         /* get direct node */
 502         page = get_node_page(sbi, dn->nid);
 503         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 504                 return 1;
 505         else if (IS_ERR(page))
 506                 return PTR_ERR(page);
 507
 508         /* Make dnode_of_data for parameter */
 509         dn->node_page = page;
 510         dn->ofs_in_node = 0;
 511         truncate_data_blocks(dn);
 512         truncate_node(dn);
 513         return 1;
 514 }
 515
 516 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 517                                                 int ofs, int depth)
 518 {
 519         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 520         struct dnode_of_data rdn = *dn;
 521         struct page *page;
 522         struct f2fs_node *rn;
 523         nid_t child_nid;
 524         unsigned int child_nofs;
 525         int freed = 0;
 526         int i, ret;
 527
 528         if (dn->nid == 0)
 529                 return NIDS_PER_BLOCK + 1;
 530
 531         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 532
 533         page = get_node_page(sbi, dn->nid);
 534         if (IS_ERR(page)) {
 535                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 536                 return PTR_ERR(page);
 537         }
 538
 539         rn = F2FS_NODE(page);
 540         if (depth < 3) {
 541                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 542                         child_nid = le32_to_cpu(rn->in.nid[i]);
 543                         if (child_nid == 0)
 544                                 continue;
 545                         rdn.nid = child_nid;
 546                         ret = truncate_dnode(&rdn);
 547                         if (ret < 0)
 548                                 goto out_err;
 549                         set_nid(page, i, 0, false);
 550                 }
 551         } else {
 552                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 553                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 554                         child_nid = le32_to_cpu(rn->in.nid[i]);
 555                         if (child_nid == 0) {
 556                                 child_nofs += NIDS_PER_BLOCK + 1;
 557                                 continue;
 558                         }
 559                         rdn.nid = child_nid;
 560                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 561                         if (ret == (NIDS_PER_BLOCK + 1)) {
 562                                 set_nid(page, i, 0, false);
 563                                 child_nofs += ret;
 564                         } else if (ret < 0 && ret != -ENOENT) {
 565                                 goto out_err;
 566                         }
 567                 }
 568                 freed = child_nofs;
 569         }
 570
 571         if (!ofs) {
 572                 /* remove current indirect node */
 573                 dn->node_page = page;
 574                 truncate_node(dn);
 575                 freed++;
 576         } else {
 577                 f2fs_put_page(page, 1);
 578         }
 579         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 580         return freed;
 581
 582 out_err:
 583         f2fs_put_page(page, 1);
 584         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 585         return ret;
 586 }
 587
 588 static int truncate_partial_nodes(struct dnode_of_data *dn,
 589                         struct f2fs_inode *ri, int *offset, int depth)
 590 {
 591         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 592         struct page *pages[2];
 593         nid_t nid[3];
 594         nid_t child_nid;
 595         int err = 0;
 596         int i;
 597         int idx = depth - 2;
 598
 599         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 600         if (!nid[0])
 601                 return 0;
 602
 603         /* get indirect nodes in the path */
 604         for (i = 0; i < idx + 1; i++) {
 605                 /* refernece count'll be increased */
 606                 pages[i] = get_node_page(sbi, nid[i]);
 607                 if (IS_ERR(pages[i])) {
 608                         err = PTR_ERR(pages[i]);
 609                         idx = i - 1;
 610                         goto fail;
 611                 }
 612                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 613         }
 614
 615         /* free direct nodes linked to a partial indirect node */
 616         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 617                 child_nid = get_nid(pages[idx], i, false);
 618                 if (!child_nid)
 619                         continue;
 620                 dn->nid = child_nid;
 621                 err = truncate_dnode(dn);
 622                 if (err < 0)
 623                         goto fail;
 624                 set_nid(pages[idx], i, 0, false);
 625         }
 626
 627         if (offset[idx + 1] == 0) {
 628                 dn->node_page = pages[idx];
 629                 dn->nid = nid[idx];
 630                 truncate_node(dn);
 631         } else {
 632                 f2fs_put_page(pages[idx], 1);
 633         }
 634         offset[idx]++;
 635         offset[idx + 1] = 0;
 636         idx--;
 637 fail:
 638         for (i = idx; i >= 0; i--)
 639                 f2fs_put_page(pages[i], 1);
 640
 641         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 642
 643         return err;
 644 }
 645
 646 /*
 647  * All the block addresses of data and nodes should be nullified.
 648  */
 649 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 650 {
 651         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 652         int err = 0, cont = 1;
 653         int level, offset[4], noffset[4];
 654         unsigned int nofs = 0;
 655         struct f2fs_inode *ri;
 656         struct dnode_of_data dn;
 657         struct page *page;
 658
 659         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 660
 661         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 662 restart:
 663         page = get_node_page(sbi, inode->i_ino);
 664         if (IS_ERR(page)) {
 665                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 666                 return PTR_ERR(page);
 667         }
 668
 669         set_new_dnode(&dn, inode, page, NULL, 0);
 670         unlock_page(page);
 671
 672         ri = F2FS_INODE(page);
 673         switch (level) {
 674         case 0:
 675         case 1:
 676                 nofs = noffset[1];
 677                 break;
 678         case 2:
 679                 nofs = noffset[1];
 680                 if (!offset[level - 1])
 681                         goto skip_partial;
 682                 err = truncate_partial_nodes(&dn, ri, offset, level);
 683                 if (err < 0 && err != -ENOENT)
 684                         goto fail;
 685                 nofs += 1 + NIDS_PER_BLOCK;
 686                 break;
 687         case 3:
 688                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 689                 if (!offset[level - 1])
 690                         goto skip_partial;
 691                 err = truncate_partial_nodes(&dn, ri, offset, level);
 692                 if (err < 0 && err != -ENOENT)
 693                         goto fail;
 694                 break;
 695         default:
 696                 BUG();
 697         }
 698
 699 skip_partial:
 700         while (cont) {
 701                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 702                 switch (offset[0]) {
 703                 case NODE_DIR1_BLOCK:
 704                 case NODE_DIR2_BLOCK:
 705                         err = truncate_dnode(&dn);
 706                         break;
 707
 708                 case NODE_IND1_BLOCK:
 709                 case NODE_IND2_BLOCK:
 710                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 711                         break;
 712
 713                 case NODE_DIND_BLOCK:
 714                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 715                         cont = 0;
 716                         break;
 717
 718                 default:
 719                         BUG();
 720                 }
 721                 if (err < 0 && err != -ENOENT)
 722                         goto fail;
 723                 if (offset[1] == 0 &&
 724                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 725                         lock_page(page);
 726                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 727                                 f2fs_put_page(page, 1);
 728                                 goto restart;
 729                         }
 730                         wait_on_page_writeback(page);
 731                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 732                         set_page_dirty(page);
 733                         unlock_page(page);
 734                 }
 735                 offset[1] = 0;
 736                 offset[0]++;
 737                 nofs += err;
 738         }
 739 fail:
 740         f2fs_put_page(page, 0);
 741         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 742         return err > 0 ? 0 : err;
 743 }
 744
 745 int truncate_xattr_node(struct inode *inode, struct page *page)
 746 {
 747         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 748         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 749         struct dnode_of_data dn;
 750         struct page *npage;
 751
 752         if (!nid)
 753                 return 0;
 754
 755         npage = get_node_page(sbi, nid);
 756         if (IS_ERR(npage))
 757                 return PTR_ERR(npage);
 758
 759         F2FS_I(inode)->i_xattr_nid = 0;
 760
 761         /* need to do checkpoint during fsync */
 762         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 763
 764         set_new_dnode(&dn, inode, page, npage, nid);
 765
 766         if (page)
 767                 dn.inode_page_locked = true;
 768         truncate_node(&dn);
 769         return 0;
 770 }
 771
 772 /*
 773  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 774  * f2fs_unlock_op().
 775  */
 776 void remove_inode_page(struct inode *inode)
 777 {
 778         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 779         struct page *page;
 780         nid_t ino = inode->i_ino;
 781         struct dnode_of_data dn;
 782
 783         page = get_node_page(sbi, ino);
 784         if (IS_ERR(page))
 785                 return;
 786
 787         if (truncate_xattr_node(inode, page)) {
 788                 f2fs_put_page(page, 1);
 789                 return;
 790         }
 791         /* 0 is possible, after f2fs_new_inode() is failed */
 792         f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
 793         set_new_dnode(&dn, inode, page, page, ino);
 794         truncate_node(&dn);
 795 }
 796
 797 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
 798 {
 799         struct dnode_of_data dn;
 800
 801         /* allocate inode page for new inode */
 802         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 803
 804         /* caller should f2fs_put_page(page, 1); */
 805         return new_node_page(&dn, 0, NULL);
 806 }
 807
 808 struct page *new_node_page(struct dnode_of_data *dn,
 809                                 unsigned int ofs, struct page *ipage)
 810 {
 811         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 812         struct node_info old_ni, new_ni;
 813         struct page *page;
 814         int err;
 815
 816         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 817                 return ERR_PTR(-EPERM);
 818
 819         page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
 820         if (!page)
 821                 return ERR_PTR(-ENOMEM);
 822
 823         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 824                 err = -ENOSPC;
 825                 goto fail;
 826         }
 827
 828         get_node_info(sbi, dn->nid, &old_ni);
 829
 830         /* Reinitialize old_ni with new node page */
 831         f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
 832         new_ni = old_ni;
 833         new_ni.ino = dn->inode->i_ino;
 834         set_node_addr(sbi, &new_ni, NEW_ADDR);
 835
 836         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 837         set_cold_node(dn->inode, page);
 838         SetPageUptodate(page);
 839         set_page_dirty(page);
 840
 841         if (ofs == XATTR_NODE_OFFSET)
 842                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 843
 844         dn->node_page = page;
 845         if (ipage)
 846                 update_inode(dn->inode, ipage);
 847         else
 848                 sync_inode_page(dn);
 849         if (ofs == 0)
 850                 inc_valid_inode_count(sbi);
 851
 852         return page;
 853
 854 fail:
 855         clear_node_page_dirty(page);
 856         f2fs_put_page(page, 1);
 857         return ERR_PTR(err);
 858 }
 859
 860 /*
 861  * Caller should do after getting the following values.
 862  * 0: f2fs_put_page(page, 0)
 863  * LOCKED_PAGE: f2fs_put_page(page, 1)
 864  * error: nothing
 865  */
 866 static int read_node_page(struct page *page, int rw)
 867 {
 868         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 869         struct node_info ni;
 870
 871         get_node_info(sbi, page->index, &ni);
 872
 873         if (unlikely(ni.blk_addr == NULL_ADDR)) {
 874                 f2fs_put_page(page, 1);
 875                 return -ENOENT;
 876         }
 877
 878         if (PageUptodate(page))
 879                 return LOCKED_PAGE;
 880
 881         return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
 882 }
 883
 884 /*
 885  * Readahead a node page
 886  */
 887 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 888 {
 889         struct page *apage;
 890         int err;
 891
 892         apage = find_get_page(NODE_MAPPING(sbi), nid);
 893         if (apage && PageUptodate(apage)) {
 894                 f2fs_put_page(apage, 0);
 895                 return;
 896         }
 897         f2fs_put_page(apage, 0);
 898
 899         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
 900         if (!apage)
 901                 return;
 902
 903         err = read_node_page(apage, READA);
 904         if (err == 0)
 905                 f2fs_put_page(apage, 0);
 906         else if (err == LOCKED_PAGE)
 907                 f2fs_put_page(apage, 1);
 908 }
 909
 910 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 911 {
 912         struct page *page;
 913         int err;
 914 repeat:
 915         page = grab_cache_page(NODE_MAPPING(sbi), nid);
 916         if (!page)
 917                 return ERR_PTR(-ENOMEM);
 918
 919         err = read_node_page(page, READ_SYNC);
 920         if (err < 0)
 921                 return ERR_PTR(err);
 922         else if (err == LOCKED_PAGE)
 923                 goto got_it;
 924
 925         lock_page(page);
 926         if (unlikely(!PageUptodate(page))) {
 927                 f2fs_put_page(page, 1);
 928                 return ERR_PTR(-EIO);
 929         }
 930         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 931                 f2fs_put_page(page, 1);
 932                 goto repeat;
 933         }
 934 got_it:
 935         f2fs_bug_on(nid != nid_of_node(page));
 936         mark_page_accessed(page);
 937         return page;
 938 }
 939
 940 /*
 941  * Return a locked page for the desired node page.
 942  * And, readahead MAX_RA_NODE number of node pages.
 943  */
 944 struct page *get_node_page_ra(struct page *parent, int start)
 945 {
 946         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 947         struct blk_plug plug;
 948         struct page *page;
 949         int err, i, end;
 950         nid_t nid;
 951
 952         /* First, try getting the desired direct node. */
 953         nid = get_nid(parent, start, false);
 954         if (!nid)
 955                 return ERR_PTR(-ENOENT);
 956 repeat:
 957         page = grab_cache_page(NODE_MAPPING(sbi), nid);
 958         if (!page)
 959                 return ERR_PTR(-ENOMEM);
 960
 961         err = read_node_page(page, READ_SYNC);
 962         if (err < 0)
 963                 return ERR_PTR(err);
 964         else if (err == LOCKED_PAGE)
 965                 goto page_hit;
 966
 967         blk_start_plug(&plug);
 968
 969         /* Then, try readahead for siblings of the desired node */
 970         end = start + MAX_RA_NODE;
 971         end = min(end, NIDS_PER_BLOCK);
 972         for (i = start + 1; i < end; i++) {
 973                 nid = get_nid(parent, i, false);
 974                 if (!nid)
 975                         continue;
 976                 ra_node_page(sbi, nid);
 977         }
 978
 979         blk_finish_plug(&plug);
 980
 981         lock_page(page);
 982         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 983                 f2fs_put_page(page, 1);
 984                 goto repeat;
 985         }
 986 page_hit:
 987         if (unlikely(!PageUptodate(page))) {
 988                 f2fs_put_page(page, 1);
 989                 return ERR_PTR(-EIO);
 990         }
 991         mark_page_accessed(page);
 992         return page;
 993 }
 994
 995 void sync_inode_page(struct dnode_of_data *dn)
 996 {
 997         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
 998                 update_inode(dn->inode, dn->node_page);
 999         } else if (dn->inode_page) {
1000                 if (!dn->inode_page_locked)
1001                         lock_page(dn->inode_page);
1002                 update_inode(dn->inode, dn->inode_page);
1003                 if (!dn->inode_page_locked)
1004                         unlock_page(dn->inode_page);
1005         } else {
1006                 update_inode_page(dn->inode);
1007         }
1008 }
1009
1010 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1011                                         struct writeback_control *wbc)
1012 {
1013         pgoff_t index, end;
1014         struct pagevec pvec;
1015         int step = ino ? 2 : 0;
1016         int nwritten = 0, wrote = 0;
1017
1018         pagevec_init(&pvec, 0);
1019
1020 next_step:
1021         index = 0;
1022         end = LONG_MAX;
1023
1024         while (index <= end) {
1025                 int i, nr_pages;
1026                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1027                                 PAGECACHE_TAG_DIRTY,
1028                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1029                 if (nr_pages == 0)
1030                         break;
1031
1032                 for (i = 0; i < nr_pages; i++) {
1033                         struct page *page = pvec.pages[i];
1034
1035                         /*
1036                          * flushing sequence with step:
1037                          * 0. indirect nodes
1038                          * 1. dentry dnodes
1039                          * 2. file dnodes
1040                          */
1041                         if (step == 0 && IS_DNODE(page))
1042                                 continue;
1043                         if (step == 1 && (!IS_DNODE(page) ||
1044                                                 is_cold_node(page)))
1045                                 continue;
1046                         if (step == 2 && (!IS_DNODE(page) ||
1047                                                 !is_cold_node(page)))
1048                                 continue;
1049
1050                         /*
1051                          * If an fsync mode,
1052                          * we should not skip writing node pages.
1053                          */
1054                         if (ino && ino_of_node(page) == ino)
1055                                 lock_page(page);
1056                         else if (!trylock_page(page))
1057                                 continue;
1058
1059                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1060 continue_unlock:
1061                                 unlock_page(page);
1062                                 continue;
1063                         }
1064                         if (ino && ino_of_node(page) != ino)
1065                                 goto continue_unlock;
1066
1067                         if (!PageDirty(page)) {
1068                                 /* someone wrote it for us */
1069                                 goto continue_unlock;
1070                         }
1071
1072                         if (!clear_page_dirty_for_io(page))
1073                                 goto continue_unlock;
1074
1075                         /* called by fsync() */
1076                         if (ino && IS_DNODE(page)) {
1077                                 int mark = !is_checkpointed_node(sbi, ino);
1078                                 set_fsync_mark(page, 1);
1079                                 if (IS_INODE(page))
1080                                         set_dentry_mark(page, mark);
1081                                 nwritten++;
1082                         } else {
1083                                 set_fsync_mark(page, 0);
1084                                 set_dentry_mark(page, 0);
1085                         }
1086                         NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1087                         wrote++;
1088
1089                         if (--wbc->nr_to_write == 0)
1090                                 break;
1091                 }
1092                 pagevec_release(&pvec);
1093                 cond_resched();
1094
1095                 if (wbc->nr_to_write == 0) {
1096                         step = 2;
1097                         break;
1098                 }
1099         }
1100
1101         if (step < 2) {
1102                 step++;
1103                 goto next_step;
1104         }
1105
1106         if (wrote)
1107                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1108         return nwritten;
1109 }
1110
1111 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1112 {
1113         pgoff_t index = 0, end = LONG_MAX;
1114         struct pagevec pvec;
1115         int ret2 = 0, ret = 0;
1116
1117         pagevec_init(&pvec, 0);
1118
1119         while (index <= end) {
1120                 int i, nr_pages;
1121                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1122                                 PAGECACHE_TAG_WRITEBACK,
1123                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1124                 if (nr_pages == 0)
1125                         break;
1126
1127                 for (i = 0; i < nr_pages; i++) {
1128                         struct page *page = pvec.pages[i];
1129
1130                         /* until radix tree lookup accepts end_index */
1131                         if (unlikely(page->index > end))
1132                                 continue;
1133
1134                         if (ino && ino_of_node(page) == ino) {
1135                                 wait_on_page_writeback(page);
1136                                 if (TestClearPageError(page))
1137                                         ret = -EIO;
1138                         }
1139                 }
1140                 pagevec_release(&pvec);
1141                 cond_resched();
1142         }
1143
1144         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1145                 ret2 = -ENOSPC;
1146         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1147                 ret2 = -EIO;
1148         if (!ret)
1149                 ret = ret2;
1150         return ret;
1151 }
1152
1153 static int f2fs_write_node_page(struct page *page,
1154                                 struct writeback_control *wbc)
1155 {
1156         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1157         nid_t nid;
1158         block_t new_addr;
1159         struct node_info ni;
1160         struct f2fs_io_info fio = {
1161                 .type = NODE,
1162                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1163         };
1164
1165         if (unlikely(sbi->por_doing))
1166                 goto redirty_out;
1167
1168         wait_on_page_writeback(page);
1169
1170         /* get old block addr of this node page */
1171         nid = nid_of_node(page);
1172         f2fs_bug_on(page->index != nid);
1173
1174         get_node_info(sbi, nid, &ni);
1175
1176         /* This page is already truncated */
1177         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1178                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1179                 unlock_page(page);
1180                 return 0;
1181         }
1182
1183         if (wbc->for_reclaim)
1184                 goto redirty_out;
1185
1186         mutex_lock(&sbi->node_write);
1187         set_page_writeback(page);
1188         write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1189         set_node_addr(sbi, &ni, new_addr);
1190         dec_page_count(sbi, F2FS_DIRTY_NODES);
1191         mutex_unlock(&sbi->node_write);
1192         unlock_page(page);
1193         return 0;
1194
1195 redirty_out:
1196         dec_page_count(sbi, F2FS_DIRTY_NODES);
1197         wbc->pages_skipped++;
1198         set_page_dirty(page);
1199         return AOP_WRITEPAGE_ACTIVATE;
1200 }
1201
1202 /*
1203  * It is very important to gather dirty pages and write at once, so that we can
1204  * submit a big bio without interfering other data writes.
1205  * Be default, 512 pages (2MB) * 3 node types, is more reasonable.
1206  */
1207 #define COLLECT_DIRTY_NODES     1536
1208 static int f2fs_write_node_pages(struct address_space *mapping,
1209                             struct writeback_control *wbc)
1210 {
1211         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1212         long nr_to_write = wbc->nr_to_write;
1213
1214         /* balancing f2fs's metadata in background */
1215         f2fs_balance_fs_bg(sbi);
1216
1217         /* collect a number of dirty node pages and write together */
1218         if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1219                 return 0;
1220
1221         /* if mounting is failed, skip writing node pages */
1222         wbc->nr_to_write = 3 * max_hw_blocks(sbi);
1223         wbc->sync_mode = WB_SYNC_NONE;
1224         sync_node_pages(sbi, 0, wbc);
1225         wbc->nr_to_write = nr_to_write - (3 * max_hw_blocks(sbi) -
1226                                                 wbc->nr_to_write);
1227         return 0;
1228 }
1229
1230 static int f2fs_set_node_page_dirty(struct page *page)
1231 {
1232         struct address_space *mapping = page->mapping;
1233         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1234
1235         trace_f2fs_set_page_dirty(page, NODE);
1236
1237         SetPageUptodate(page);
1238         if (!PageDirty(page)) {
1239                 __set_page_dirty_nobuffers(page);
1240                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1241                 SetPagePrivate(page);
1242                 return 1;
1243         }
1244         return 0;
1245 }
1246
1247 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1248                                       unsigned int length)
1249 {
1250         struct inode *inode = page->mapping->host;
1251         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1252         if (PageDirty(page))
1253                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1254         ClearPagePrivate(page);
1255 }
1256
1257 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1258 {
1259         ClearPagePrivate(page);
1260         return 1;
1261 }
1262
1263 /*
1264  * Structure of the f2fs node operations
1265  */
1266 const struct address_space_operations f2fs_node_aops = {
1267         .writepage      = f2fs_write_node_page,
1268         .writepages     = f2fs_write_node_pages,
1269         .set_page_dirty = f2fs_set_node_page_dirty,
1270         .invalidatepage = f2fs_invalidate_node_page,
1271         .releasepage    = f2fs_release_node_page,
1272 };
1273
1274 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1275 {
1276         struct list_head *this;
1277         struct free_nid *i;
1278         list_for_each(this, head) {
1279                 i = list_entry(this, struct free_nid, list);
1280                 if (i->nid == n)
1281                         return i;
1282         }
1283         return NULL;
1284 }
1285
1286 static void __del_from_free_nid_list(struct free_nid *i)
1287 {
1288         list_del(&i->list);
1289         kmem_cache_free(free_nid_slab, i);
1290 }
1291
1292 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build)
1293 {
1294         struct free_nid *i;
1295         struct nat_entry *ne;
1296         bool allocated = false;
1297
1298         if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1299                 return -1;
1300
1301         /* 0 nid should not be used */
1302         if (unlikely(nid == 0))
1303                 return 0;
1304
1305         if (build) {
1306                 /* do not add allocated nids */
1307                 read_lock(&nm_i->nat_tree_lock);
1308                 ne = __lookup_nat_cache(nm_i, nid);
1309                 if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1310                         allocated = true;
1311                 read_unlock(&nm_i->nat_tree_lock);
1312                 if (allocated)
1313                         return 0;
1314         }
1315
1316         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1317         i->nid = nid;
1318         i->state = NID_NEW;
1319
1320         spin_lock(&nm_i->free_nid_list_lock);
1321         if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1322                 spin_unlock(&nm_i->free_nid_list_lock);
1323                 kmem_cache_free(free_nid_slab, i);
1324                 return 0;
1325         }
1326         list_add_tail(&i->list, &nm_i->free_nid_list);
1327         nm_i->fcnt++;
1328         spin_unlock(&nm_i->free_nid_list_lock);
1329         return 1;
1330 }
1331
1332 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1333 {
1334         struct free_nid *i;
1335         spin_lock(&nm_i->free_nid_list_lock);
1336         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1337         if (i && i->state == NID_NEW) {
1338                 __del_from_free_nid_list(i);
1339                 nm_i->fcnt--;
1340         }
1341         spin_unlock(&nm_i->free_nid_list_lock);
1342 }
1343
1344 static void scan_nat_page(struct f2fs_nm_info *nm_i,
1345                         struct page *nat_page, nid_t start_nid)
1346 {
1347         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1348         block_t blk_addr;
1349         int i;
1350
1351         i = start_nid % NAT_ENTRY_PER_BLOCK;
1352
1353         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1354
1355                 if (unlikely(start_nid >= nm_i->max_nid))
1356                         break;
1357
1358                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1359                 f2fs_bug_on(blk_addr == NEW_ADDR);
1360                 if (blk_addr == NULL_ADDR) {
1361                         if (add_free_nid(nm_i, start_nid, true) < 0)
1362                                 break;
1363                 }
1364         }
1365 }
1366
1367 static void build_free_nids(struct f2fs_sb_info *sbi)
1368 {
1369         struct f2fs_nm_info *nm_i = NM_I(sbi);
1370         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1371         struct f2fs_summary_block *sum = curseg->sum_blk;
1372         int i = 0;
1373         nid_t nid = nm_i->next_scan_nid;
1374
1375         /* Enough entries */
1376         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1377                 return;
1378
1379         /* readahead nat pages to be scanned */
1380         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1381
1382         while (1) {
1383                 struct page *page = get_current_nat_page(sbi, nid);
1384
1385                 scan_nat_page(nm_i, page, nid);
1386                 f2fs_put_page(page, 1);
1387
1388                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1389                 if (unlikely(nid >= nm_i->max_nid))
1390                         nid = 0;
1391
1392                 if (i++ == FREE_NID_PAGES)
1393                         break;
1394         }
1395
1396         /* go to the next free nat pages to find free nids abundantly */
1397         nm_i->next_scan_nid = nid;
1398
1399         /* find free nids from current sum_pages */
1400         mutex_lock(&curseg->curseg_mutex);
1401         for (i = 0; i < nats_in_cursum(sum); i++) {
1402                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1403                 nid = le32_to_cpu(nid_in_journal(sum, i));
1404                 if (addr == NULL_ADDR)
1405                         add_free_nid(nm_i, nid, true);
1406                 else
1407                         remove_free_nid(nm_i, nid);
1408         }
1409         mutex_unlock(&curseg->curseg_mutex);
1410 }
1411
1412 /*
1413  * If this function returns success, caller can obtain a new nid
1414  * from second parameter of this function.
1415  * The returned nid could be used ino as well as nid when inode is created.
1416  */
1417 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1418 {
1419         struct f2fs_nm_info *nm_i = NM_I(sbi);
1420         struct free_nid *i = NULL;
1421         struct list_head *this;
1422 retry:
1423         if (unlikely(sbi->total_valid_node_count + 1 >= nm_i->max_nid))
1424                 return false;
1425
1426         spin_lock(&nm_i->free_nid_list_lock);
1427
1428         /* We should not use stale free nids created by build_free_nids */
1429         if (nm_i->fcnt && !sbi->on_build_free_nids) {
1430                 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1431                 list_for_each(this, &nm_i->free_nid_list) {
1432                         i = list_entry(this, struct free_nid, list);
1433                         if (i->state == NID_NEW)
1434                                 break;
1435                 }
1436
1437                 f2fs_bug_on(i->state != NID_NEW);
1438                 *nid = i->nid;
1439                 i->state = NID_ALLOC;
1440                 nm_i->fcnt--;
1441                 spin_unlock(&nm_i->free_nid_list_lock);
1442                 return true;
1443         }
1444         spin_unlock(&nm_i->free_nid_list_lock);
1445
1446         /* Let's scan nat pages and its caches to get free nids */
1447         mutex_lock(&nm_i->build_lock);
1448         sbi->on_build_free_nids = true;
1449         build_free_nids(sbi);
1450         sbi->on_build_free_nids = false;
1451         mutex_unlock(&nm_i->build_lock);
1452         goto retry;
1453 }
1454
1455 /*
1456  * alloc_nid() should be called prior to this function.
1457  */
1458 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1459 {
1460         struct f2fs_nm_info *nm_i = NM_I(sbi);
1461         struct free_nid *i;
1462
1463         spin_lock(&nm_i->free_nid_list_lock);
1464         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1465         f2fs_bug_on(!i || i->state != NID_ALLOC);
1466         __del_from_free_nid_list(i);
1467         spin_unlock(&nm_i->free_nid_list_lock);
1468 }
1469
1470 /*
1471  * alloc_nid() should be called prior to this function.
1472  */
1473 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1474 {
1475         struct f2fs_nm_info *nm_i = NM_I(sbi);
1476         struct free_nid *i;
1477
1478         if (!nid)
1479                 return;
1480
1481         spin_lock(&nm_i->free_nid_list_lock);
1482         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1483         f2fs_bug_on(!i || i->state != NID_ALLOC);
1484         if (nm_i->fcnt > 2 * MAX_FREE_NIDS) {
1485                 __del_from_free_nid_list(i);
1486         } else {
1487                 i->state = NID_NEW;
1488                 nm_i->fcnt++;
1489         }
1490         spin_unlock(&nm_i->free_nid_list_lock);
1491 }
1492
1493 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1494                 struct f2fs_summary *sum, struct node_info *ni,
1495                 block_t new_blkaddr)
1496 {
1497         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1498         set_node_addr(sbi, ni, new_blkaddr);
1499         clear_node_page_dirty(page);
1500 }
1501
1502 bool recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1503 {
1504         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1505         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1506         nid_t new_xnid = nid_of_node(page);
1507         struct node_info ni;
1508
1509         if (ofs_of_node(page) != XATTR_NODE_OFFSET)
1510                 return false;
1511
1512         /* 1: invalidate the previous xattr nid */
1513         if (!prev_xnid)
1514                 goto recover_xnid;
1515
1516         /* Deallocate node address */
1517         get_node_info(sbi, prev_xnid, &ni);
1518         f2fs_bug_on(ni.blk_addr == NULL_ADDR);
1519         invalidate_blocks(sbi, ni.blk_addr);
1520         dec_valid_node_count(sbi, inode);
1521         set_node_addr(sbi, &ni, NULL_ADDR);
1522
1523 recover_xnid:
1524         /* 2: allocate new xattr nid */
1525         if (unlikely(!inc_valid_node_count(sbi, inode)))
1526                 f2fs_bug_on(1);
1527
1528         remove_free_nid(NM_I(sbi), new_xnid);
1529         get_node_info(sbi, new_xnid, &ni);
1530         ni.ino = inode->i_ino;
1531         set_node_addr(sbi, &ni, NEW_ADDR);
1532         F2FS_I(inode)->i_xattr_nid = new_xnid;
1533
1534         /* 3: update xattr blkaddr */
1535         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1536         set_node_addr(sbi, &ni, blkaddr);
1537
1538         update_inode_page(inode);
1539         return true;
1540 }
1541
1542 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1543 {
1544         struct f2fs_inode *src, *dst;
1545         nid_t ino = ino_of_node(page);
1546         struct node_info old_ni, new_ni;
1547         struct page *ipage;
1548
1549         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1550         if (!ipage)
1551                 return -ENOMEM;
1552
1553         /* Should not use this inode  from free nid list */
1554         remove_free_nid(NM_I(sbi), ino);
1555
1556         get_node_info(sbi, ino, &old_ni);
1557         SetPageUptodate(ipage);
1558         fill_node_footer(ipage, ino, ino, 0, true);
1559
1560         src = F2FS_INODE(page);
1561         dst = F2FS_INODE(ipage);
1562
1563         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1564         dst->i_size = 0;
1565         dst->i_blocks = cpu_to_le64(1);
1566         dst->i_links = cpu_to_le32(1);
1567         dst->i_xattr_nid = 0;
1568
1569         new_ni = old_ni;
1570         new_ni.ino = ino;
1571
1572         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1573                 WARN_ON(1);
1574         set_node_addr(sbi, &new_ni, NEW_ADDR);
1575         inc_valid_inode_count(sbi);
1576         f2fs_put_page(ipage, 1);
1577         return 0;
1578 }
1579
1580 /*
1581  * ra_sum_pages() merge contiguous pages into one bio and submit.
1582  * these pre-readed pages are linked in pages list.
1583  */
1584 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
1585                                 int start, int nrpages)
1586 {
1587         struct page *page;
1588         int page_idx = start;
1589         struct f2fs_io_info fio = {
1590                 .type = META,
1591                 .rw = READ_SYNC | REQ_META | REQ_PRIO
1592         };
1593
1594         for (; page_idx < start + nrpages; page_idx++) {
1595                 /* alloc temporal page for read node summary info*/
1596                 page = alloc_page(GFP_F2FS_ZERO);
1597                 if (!page) {
1598                         struct page *tmp;
1599                         list_for_each_entry_safe(page, tmp, pages, lru) {
1600                                 list_del(&page->lru);
1601                                 unlock_page(page);
1602                                 __free_pages(page, 0);
1603                         }
1604                         return -ENOMEM;
1605                 }
1606
1607                 lock_page(page);
1608                 page->index = page_idx;
1609                 list_add_tail(&page->lru, pages);
1610         }
1611
1612         list_for_each_entry(page, pages, lru)
1613                 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1614
1615         f2fs_submit_merged_bio(sbi, META, READ);
1616         return 0;
1617 }
1618
1619 int restore_node_summary(struct f2fs_sb_info *sbi,
1620                         unsigned int segno, struct f2fs_summary_block *sum)
1621 {
1622         struct f2fs_node *rn;
1623         struct f2fs_summary *sum_entry;
1624         struct page *page, *tmp;
1625         block_t addr;
1626         int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1627         int i, last_offset, nrpages, err = 0;
1628         LIST_HEAD(page_list);
1629
1630         /* scan the node segment */
1631         last_offset = sbi->blocks_per_seg;
1632         addr = START_BLOCK(sbi, segno);
1633         sum_entry = &sum->entries[0];
1634
1635         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1636                 nrpages = min(last_offset - i, bio_blocks);
1637
1638                 /* read ahead node pages */
1639                 err = ra_sum_pages(sbi, &page_list, addr, nrpages);
1640                 if (err)
1641                         return err;
1642
1643                 list_for_each_entry_safe(page, tmp, &page_list, lru) {
1644
1645                         lock_page(page);
1646                         if (unlikely(!PageUptodate(page))) {
1647                                 err = -EIO;
1648                         } else {
1649                                 rn = F2FS_NODE(page);
1650                                 sum_entry->nid = rn->footer.nid;
1651                                 sum_entry->version = 0;
1652                                 sum_entry->ofs_in_node = 0;
1653                                 sum_entry++;
1654                         }
1655
1656                         list_del(&page->lru);
1657                         unlock_page(page);
1658                         __free_pages(page, 0);
1659                 }
1660         }
1661         return err;
1662 }
1663
1664 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1665 {
1666         struct f2fs_nm_info *nm_i = NM_I(sbi);
1667         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1668         struct f2fs_summary_block *sum = curseg->sum_blk;
1669         int i;
1670
1671         mutex_lock(&curseg->curseg_mutex);
1672
1673         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1674                 mutex_unlock(&curseg->curseg_mutex);
1675                 return false;
1676         }
1677
1678         for (i = 0; i < nats_in_cursum(sum); i++) {
1679                 struct nat_entry *ne;
1680                 struct f2fs_nat_entry raw_ne;
1681                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1682
1683                 raw_ne = nat_in_journal(sum, i);
1684 retry:
1685                 write_lock(&nm_i->nat_tree_lock);
1686                 ne = __lookup_nat_cache(nm_i, nid);
1687                 if (ne) {
1688                         __set_nat_cache_dirty(nm_i, ne);
1689                         write_unlock(&nm_i->nat_tree_lock);
1690                         continue;
1691                 }
1692                 ne = grab_nat_entry(nm_i, nid);
1693                 if (!ne) {
1694                         write_unlock(&nm_i->nat_tree_lock);
1695                         goto retry;
1696                 }
1697                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1698                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1699                 nat_set_version(ne, raw_ne.version);
1700                 __set_nat_cache_dirty(nm_i, ne);
1701                 write_unlock(&nm_i->nat_tree_lock);
1702         }
1703         update_nats_in_cursum(sum, -i);
1704         mutex_unlock(&curseg->curseg_mutex);
1705         return true;
1706 }
1707
1708 /*
1709  * This function is called during the checkpointing process.
1710  */
1711 void flush_nat_entries(struct f2fs_sb_info *sbi)
1712 {
1713         struct f2fs_nm_info *nm_i = NM_I(sbi);
1714         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1715         struct f2fs_summary_block *sum = curseg->sum_blk;
1716         struct list_head *cur, *n;
1717         struct page *page = NULL;
1718         struct f2fs_nat_block *nat_blk = NULL;
1719         nid_t start_nid = 0, end_nid = 0;
1720         bool flushed;
1721
1722         flushed = flush_nats_in_journal(sbi);
1723
1724         if (!flushed)
1725                 mutex_lock(&curseg->curseg_mutex);
1726
1727         /* 1) flush dirty nat caches */
1728         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1729                 struct nat_entry *ne;
1730                 nid_t nid;
1731                 struct f2fs_nat_entry raw_ne;
1732                 int offset = -1;
1733                 block_t new_blkaddr;
1734
1735                 ne = list_entry(cur, struct nat_entry, list);
1736                 nid = nat_get_nid(ne);
1737
1738                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1739                         continue;
1740                 if (flushed)
1741                         goto to_nat_page;
1742
1743                 /* if there is room for nat enries in curseg->sumpage */
1744                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1745                 if (offset >= 0) {
1746                         raw_ne = nat_in_journal(sum, offset);
1747                         goto flush_now;
1748                 }
1749 to_nat_page:
1750                 if (!page || (start_nid > nid || nid > end_nid)) {
1751                         if (page) {
1752                                 f2fs_put_page(page, 1);
1753                                 page = NULL;
1754                         }
1755                         start_nid = START_NID(nid);
1756                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1757
1758                         /*
1759                          * get nat block with dirty flag, increased reference
1760                          * count, mapped and lock
1761                          */
1762                         page = get_next_nat_page(sbi, start_nid);
1763                         nat_blk = page_address(page);
1764                 }
1765
1766                 f2fs_bug_on(!nat_blk);
1767                 raw_ne = nat_blk->entries[nid - start_nid];
1768 flush_now:
1769                 new_blkaddr = nat_get_blkaddr(ne);
1770
1771                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1772                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1773                 raw_ne.version = nat_get_version(ne);
1774
1775                 if (offset < 0) {
1776                         nat_blk->entries[nid - start_nid] = raw_ne;
1777                 } else {
1778                         nat_in_journal(sum, offset) = raw_ne;
1779                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1780                 }
1781
1782                 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1783                                 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1784                         write_lock(&nm_i->nat_tree_lock);
1785                         __del_from_nat_cache(nm_i, ne);
1786                         write_unlock(&nm_i->nat_tree_lock);
1787                 } else {
1788                         write_lock(&nm_i->nat_tree_lock);
1789                         __clear_nat_cache_dirty(nm_i, ne);
1790                         ne->checkpointed = true;
1791                         write_unlock(&nm_i->nat_tree_lock);
1792                 }
1793         }
1794         if (!flushed)
1795                 mutex_unlock(&curseg->curseg_mutex);
1796         f2fs_put_page(page, 1);
1797
1798         /* 2) shrink nat caches if necessary */
1799         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1800 }
1801
1802 static int init_node_manager(struct f2fs_sb_info *sbi)
1803 {
1804         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1805         struct f2fs_nm_info *nm_i = NM_I(sbi);
1806         unsigned char *version_bitmap;
1807         unsigned int nat_segs, nat_blocks;
1808
1809         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1810
1811         /* segment_count_nat includes pair segment so divide to 2. */
1812         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1813         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1814         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1815         nm_i->fcnt = 0;
1816         nm_i->nat_cnt = 0;
1817
1818         INIT_LIST_HEAD(&nm_i->free_nid_list);
1819         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1820         INIT_LIST_HEAD(&nm_i->nat_entries);
1821         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1822
1823         mutex_init(&nm_i->build_lock);
1824         spin_lock_init(&nm_i->free_nid_list_lock);
1825         rwlock_init(&nm_i->nat_tree_lock);
1826
1827         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1828         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1829         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1830         if (!version_bitmap)
1831                 return -EFAULT;
1832
1833         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1834                                         GFP_KERNEL);
1835         if (!nm_i->nat_bitmap)
1836                 return -ENOMEM;
1837         return 0;
1838 }
1839
1840 int build_node_manager(struct f2fs_sb_info *sbi)
1841 {
1842         int err;
1843
1844         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1845         if (!sbi->nm_info)
1846                 return -ENOMEM;
1847
1848         err = init_node_manager(sbi);
1849         if (err)
1850                 return err;
1851
1852         build_free_nids(sbi);
1853         return 0;
1854 }
1855
1856 void destroy_node_manager(struct f2fs_sb_info *sbi)
1857 {
1858         struct f2fs_nm_info *nm_i = NM_I(sbi);
1859         struct free_nid *i, *next_i;
1860         struct nat_entry *natvec[NATVEC_SIZE];
1861         nid_t nid = 0;
1862         unsigned int found;
1863
1864         if (!nm_i)
1865                 return;
1866
1867         /* destroy free nid list */
1868         spin_lock(&nm_i->free_nid_list_lock);
1869         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1870                 f2fs_bug_on(i->state == NID_ALLOC);
1871                 __del_from_free_nid_list(i);
1872                 nm_i->fcnt--;
1873         }
1874         f2fs_bug_on(nm_i->fcnt);
1875         spin_unlock(&nm_i->free_nid_list_lock);
1876
1877         /* destroy nat cache */
1878         write_lock(&nm_i->nat_tree_lock);
1879         while ((found = __gang_lookup_nat_cache(nm_i,
1880                                         nid, NATVEC_SIZE, natvec))) {
1881                 unsigned idx;
1882                 for (idx = 0; idx < found; idx++) {
1883                         struct nat_entry *e = natvec[idx];
1884                         nid = nat_get_nid(e) + 1;
1885                         __del_from_nat_cache(nm_i, e);
1886                 }
1887         }
1888         f2fs_bug_on(nm_i->nat_cnt);
1889         write_unlock(&nm_i->nat_tree_lock);
1890
1891         kfree(nm_i->nat_bitmap);
1892         sbi->nm_info = NULL;
1893         kfree(nm_i);
1894 }
1895
1896 int __init create_node_manager_caches(void)
1897 {
1898         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1899                         sizeof(struct nat_entry), NULL);
1900         if (!nat_entry_slab)
1901                 return -ENOMEM;
1902
1903         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1904                         sizeof(struct free_nid), NULL);
1905         if (!free_nid_slab) {
1906                 kmem_cache_destroy(nat_entry_slab);
1907                 return -ENOMEM;
1908         }
1909         return 0;
1910 }
1911
1912 void destroy_node_manager_caches(void)
1913 {
1914         kmem_cache_destroy(free_nid_slab);
1915         kmem_cache_destroy(nat_entry_slab);
1916 }