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 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
  25
  26 static struct kmem_cache *nat_entry_slab;
  27 static struct kmem_cache *free_nid_slab;
  28
  29 static void clear_node_page_dirty(struct page *page)
  30 {
  31         struct address_space *mapping = page->mapping;
  32         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
  33         unsigned int long flags;
  34
  35         if (PageDirty(page)) {
  36                 spin_lock_irqsave(&mapping->tree_lock, flags);
  37                 radix_tree_tag_clear(&mapping->page_tree,
  38                                 page_index(page),
  39                                 PAGECACHE_TAG_DIRTY);
  40                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  41
  42                 clear_page_dirty_for_io(page);
  43                 dec_page_count(sbi, F2FS_DIRTY_NODES);
  44         }
  45         ClearPageUptodate(page);
  46 }
  47
  48 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  49 {
  50         pgoff_t index = current_nat_addr(sbi, nid);
  51         return get_meta_page(sbi, index);
  52 }
  53
  54 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  55 {
  56         struct page *src_page;
  57         struct page *dst_page;
  58         pgoff_t src_off;
  59         pgoff_t dst_off;
  60         void *src_addr;
  61         void *dst_addr;
  62         struct f2fs_nm_info *nm_i = NM_I(sbi);
  63
  64         src_off = current_nat_addr(sbi, nid);
  65         dst_off = next_nat_addr(sbi, src_off);
  66
  67         /* get current nat block page with lock */
  68         src_page = get_meta_page(sbi, src_off);
  69
  70         /* Dirty src_page means that it is already the new target NAT page. */
  71         if (PageDirty(src_page))
  72                 return src_page;
  73
  74         dst_page = grab_meta_page(sbi, dst_off);
  75
  76         src_addr = page_address(src_page);
  77         dst_addr = page_address(dst_page);
  78         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
  79         set_page_dirty(dst_page);
  80         f2fs_put_page(src_page, 1);
  81
  82         set_to_next_nat(nm_i, nid);
  83
  84         return dst_page;
  85 }
  86
  87 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
  88 {
  89         return radix_tree_lookup(&nm_i->nat_root, n);
  90 }
  91
  92 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
  93                 nid_t start, unsigned int nr, struct nat_entry **ep)
  94 {
  95         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
  96 }
  97
  98 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
  99 {
 100         list_del(&e->list);
 101         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 102         nm_i->nat_cnt--;
 103         kmem_cache_free(nat_entry_slab, e);
 104 }
 105
 106 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 107 {
 108         struct f2fs_nm_info *nm_i = NM_I(sbi);
 109         struct nat_entry *e;
 110         int is_cp = 1;
 111
 112         read_lock(&nm_i->nat_tree_lock);
 113         e = __lookup_nat_cache(nm_i, nid);
 114         if (e && !e->checkpointed)
 115                 is_cp = 0;
 116         read_unlock(&nm_i->nat_tree_lock);
 117         return is_cp;
 118 }
 119
 120 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 121 {
 122         struct nat_entry *new;
 123
 124         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 125         if (!new)
 126                 return NULL;
 127         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 128                 kmem_cache_free(nat_entry_slab, new);
 129                 return NULL;
 130         }
 131         memset(new, 0, sizeof(struct nat_entry));
 132         nat_set_nid(new, nid);
 133         new->checkpointed = true;
 134         list_add_tail(&new->list, &nm_i->nat_entries);
 135         nm_i->nat_cnt++;
 136         return new;
 137 }
 138
 139 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 140                                                 struct f2fs_nat_entry *ne)
 141 {
 142         struct nat_entry *e;
 143 retry:
 144         write_lock(&nm_i->nat_tree_lock);
 145         e = __lookup_nat_cache(nm_i, nid);
 146         if (!e) {
 147                 e = grab_nat_entry(nm_i, nid);
 148                 if (!e) {
 149                         write_unlock(&nm_i->nat_tree_lock);
 150                         goto retry;
 151                 }
 152                 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
 153                 nat_set_ino(e, le32_to_cpu(ne->ino));
 154                 nat_set_version(e, ne->version);
 155         }
 156         write_unlock(&nm_i->nat_tree_lock);
 157 }
 158
 159 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 160                         block_t new_blkaddr)
 161 {
 162         struct f2fs_nm_info *nm_i = NM_I(sbi);
 163         struct nat_entry *e;
 164 retry:
 165         write_lock(&nm_i->nat_tree_lock);
 166         e = __lookup_nat_cache(nm_i, ni->nid);
 167         if (!e) {
 168                 e = grab_nat_entry(nm_i, ni->nid);
 169                 if (!e) {
 170                         write_unlock(&nm_i->nat_tree_lock);
 171                         goto retry;
 172                 }
 173                 e->ni = *ni;
 174                 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
 175         } else if (new_blkaddr == NEW_ADDR) {
 176                 /*
 177                  * when nid is reallocated,
 178                  * previous nat entry can be remained in nat cache.
 179                  * So, reinitialize it with new information.
 180                  */
 181                 e->ni = *ni;
 182                 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
 183         }
 184
 185         /* sanity check */
 186         f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
 187         f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
 188                         new_blkaddr == NULL_ADDR);
 189         f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
 190                         new_blkaddr == NEW_ADDR);
 191         f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
 192                         nat_get_blkaddr(e) != NULL_ADDR &&
 193                         new_blkaddr == NEW_ADDR);
 194
 195         /* increament version no as node is removed */
 196         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 197                 unsigned char version = nat_get_version(e);
 198                 nat_set_version(e, inc_node_version(version));
 199         }
 200
 201         /* change address */
 202         nat_set_blkaddr(e, new_blkaddr);
 203         __set_nat_cache_dirty(nm_i, e);
 204         write_unlock(&nm_i->nat_tree_lock);
 205 }
 206
 207 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 208 {
 209         struct f2fs_nm_info *nm_i = NM_I(sbi);
 210
 211         if (nm_i->nat_cnt <= NM_WOUT_THRESHOLD || nr_shrink <= 0)
 212                 return 0;
 213
 214         write_lock(&nm_i->nat_tree_lock);
 215         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 216                 struct nat_entry *ne;
 217                 ne = list_first_entry(&nm_i->nat_entries,
 218                                         struct nat_entry, list);
 219                 __del_from_nat_cache(nm_i, ne);
 220                 nr_shrink--;
 221         }
 222         write_unlock(&nm_i->nat_tree_lock);
 223         return nr_shrink;
 224 }
 225
 226 /*
 227  * This function returns always success
 228  */
 229 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 230 {
 231         struct f2fs_nm_info *nm_i = NM_I(sbi);
 232         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 233         struct f2fs_summary_block *sum = curseg->sum_blk;
 234         nid_t start_nid = START_NID(nid);
 235         struct f2fs_nat_block *nat_blk;
 236         struct page *page = NULL;
 237         struct f2fs_nat_entry ne;
 238         struct nat_entry *e;
 239         int i;
 240
 241         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 242         ni->nid = nid;
 243
 244         /* Check nat cache */
 245         read_lock(&nm_i->nat_tree_lock);
 246         e = __lookup_nat_cache(nm_i, nid);
 247         if (e) {
 248                 ni->ino = nat_get_ino(e);
 249                 ni->blk_addr = nat_get_blkaddr(e);
 250                 ni->version = nat_get_version(e);
 251         }
 252         read_unlock(&nm_i->nat_tree_lock);
 253         if (e)
 254                 return;
 255
 256         /* Check current segment summary */
 257         mutex_lock(&curseg->curseg_mutex);
 258         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 259         if (i >= 0) {
 260                 ne = nat_in_journal(sum, i);
 261                 node_info_from_raw_nat(ni, &ne);
 262         }
 263         mutex_unlock(&curseg->curseg_mutex);
 264         if (i >= 0)
 265                 goto cache;
 266
 267         /* Fill node_info from nat page */
 268         page = get_current_nat_page(sbi, start_nid);
 269         nat_blk = (struct f2fs_nat_block *)page_address(page);
 270         ne = nat_blk->entries[nid - start_nid];
 271         node_info_from_raw_nat(ni, &ne);
 272         f2fs_put_page(page, 1);
 273 cache:
 274         /* cache nat entry */
 275         cache_nat_entry(NM_I(sbi), nid, &ne);
 276 }
 277
 278 /*
 279  * The maximum depth is four.
 280  * Offset[0] will have raw inode offset.
 281  */
 282 static int get_node_path(struct f2fs_inode_info *fi, long block,
 283                                 int offset[4], unsigned int noffset[4])
 284 {
 285         const long direct_index = ADDRS_PER_INODE(fi);
 286         const long direct_blks = ADDRS_PER_BLOCK;
 287         const long dptrs_per_blk = NIDS_PER_BLOCK;
 288         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 289         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 290         int n = 0;
 291         int level = 0;
 292
 293         noffset[0] = 0;
 294
 295         if (block < direct_index) {
 296                 offset[n] = block;
 297                 goto got;
 298         }
 299         block -= direct_index;
 300         if (block < direct_blks) {
 301                 offset[n++] = NODE_DIR1_BLOCK;
 302                 noffset[n] = 1;
 303                 offset[n] = block;
 304                 level = 1;
 305                 goto got;
 306         }
 307         block -= direct_blks;
 308         if (block < direct_blks) {
 309                 offset[n++] = NODE_DIR2_BLOCK;
 310                 noffset[n] = 2;
 311                 offset[n] = block;
 312                 level = 1;
 313                 goto got;
 314         }
 315         block -= direct_blks;
 316         if (block < indirect_blks) {
 317                 offset[n++] = NODE_IND1_BLOCK;
 318                 noffset[n] = 3;
 319                 offset[n++] = block / direct_blks;
 320                 noffset[n] = 4 + offset[n - 1];
 321                 offset[n] = block % direct_blks;
 322                 level = 2;
 323                 goto got;
 324         }
 325         block -= indirect_blks;
 326         if (block < indirect_blks) {
 327                 offset[n++] = NODE_IND2_BLOCK;
 328                 noffset[n] = 4 + dptrs_per_blk;
 329                 offset[n++] = block / direct_blks;
 330                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 331                 offset[n] = block % direct_blks;
 332                 level = 2;
 333                 goto got;
 334         }
 335         block -= indirect_blks;
 336         if (block < dindirect_blks) {
 337                 offset[n++] = NODE_DIND_BLOCK;
 338                 noffset[n] = 5 + (dptrs_per_blk * 2);
 339                 offset[n++] = block / indirect_blks;
 340                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 341                               offset[n - 1] * (dptrs_per_blk + 1);
 342                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 343                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 344                               offset[n - 2] * (dptrs_per_blk + 1) +
 345                               offset[n - 1];
 346                 offset[n] = block % direct_blks;
 347                 level = 3;
 348                 goto got;
 349         } else {
 350                 BUG();
 351         }
 352 got:
 353         return level;
 354 }
 355
 356 /*
 357  * Caller should call f2fs_put_dnode(dn).
 358  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 359  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 360  * In the case of RDONLY_NODE, we don't need to care about mutex.
 361  */
 362 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 363 {
 364         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 365         struct page *npage[4];
 366         struct page *parent;
 367         int offset[4];
 368         unsigned int noffset[4];
 369         nid_t nids[4];
 370         int level, i;
 371         int err = 0;
 372
 373         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 374
 375         nids[0] = dn->inode->i_ino;
 376         npage[0] = dn->inode_page;
 377
 378         if (!npage[0]) {
 379                 npage[0] = get_node_page(sbi, nids[0]);
 380                 if (IS_ERR(npage[0]))
 381                         return PTR_ERR(npage[0]);
 382         }
 383         parent = npage[0];
 384         if (level != 0)
 385                 nids[1] = get_nid(parent, offset[0], true);
 386         dn->inode_page = npage[0];
 387         dn->inode_page_locked = true;
 388
 389         /* get indirect or direct nodes */
 390         for (i = 1; i <= level; i++) {
 391                 bool done = false;
 392
 393                 if (!nids[i] && mode == ALLOC_NODE) {
 394                         /* alloc new node */
 395                         if (!alloc_nid(sbi, &(nids[i]))) {
 396                                 err = -ENOSPC;
 397                                 goto release_pages;
 398                         }
 399
 400                         dn->nid = nids[i];
 401                         npage[i] = new_node_page(dn, noffset[i], NULL);
 402                         if (IS_ERR(npage[i])) {
 403                                 alloc_nid_failed(sbi, nids[i]);
 404                                 err = PTR_ERR(npage[i]);
 405                                 goto release_pages;
 406                         }
 407
 408                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 409                         alloc_nid_done(sbi, nids[i]);
 410                         done = true;
 411                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 412                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 413                         if (IS_ERR(npage[i])) {
 414                                 err = PTR_ERR(npage[i]);
 415                                 goto release_pages;
 416                         }
 417                         done = true;
 418                 }
 419                 if (i == 1) {
 420                         dn->inode_page_locked = false;
 421                         unlock_page(parent);
 422                 } else {
 423                         f2fs_put_page(parent, 1);
 424                 }
 425
 426                 if (!done) {
 427                         npage[i] = get_node_page(sbi, nids[i]);
 428                         if (IS_ERR(npage[i])) {
 429                                 err = PTR_ERR(npage[i]);
 430                                 f2fs_put_page(npage[0], 0);
 431                                 goto release_out;
 432                         }
 433                 }
 434                 if (i < level) {
 435                         parent = npage[i];
 436                         nids[i + 1] = get_nid(parent, offset[i], false);
 437                 }
 438         }
 439         dn->nid = nids[level];
 440         dn->ofs_in_node = offset[level];
 441         dn->node_page = npage[level];
 442         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 443         return 0;
 444
 445 release_pages:
 446         f2fs_put_page(parent, 1);
 447         if (i > 1)
 448                 f2fs_put_page(npage[0], 0);
 449 release_out:
 450         dn->inode_page = NULL;
 451         dn->node_page = NULL;
 452         return err;
 453 }
 454
 455 static void truncate_node(struct dnode_of_data *dn)
 456 {
 457         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 458         struct node_info ni;
 459
 460         get_node_info(sbi, dn->nid, &ni);
 461         if (dn->inode->i_blocks == 0) {
 462                 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
 463                 goto invalidate;
 464         }
 465         f2fs_bug_on(ni.blk_addr == NULL_ADDR);
 466
 467         /* Deallocate node address */
 468         invalidate_blocks(sbi, ni.blk_addr);
 469         dec_valid_node_count(sbi, dn->inode);
 470         set_node_addr(sbi, &ni, NULL_ADDR);
 471
 472         if (dn->nid == dn->inode->i_ino) {
 473                 remove_orphan_inode(sbi, dn->nid);
 474                 dec_valid_inode_count(sbi);
 475         } else {
 476                 sync_inode_page(dn);
 477         }
 478 invalidate:
 479         clear_node_page_dirty(dn->node_page);
 480         F2FS_SET_SB_DIRT(sbi);
 481
 482         f2fs_put_page(dn->node_page, 1);
 483
 484         invalidate_mapping_pages(NODE_MAPPING(sbi),
 485                         dn->node_page->index, dn->node_page->index);
 486
 487         dn->node_page = NULL;
 488         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 489 }
 490
 491 static int truncate_dnode(struct dnode_of_data *dn)
 492 {
 493         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 494         struct page *page;
 495
 496         if (dn->nid == 0)
 497                 return 1;
 498
 499         /* get direct node */
 500         page = get_node_page(sbi, dn->nid);
 501         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 502                 return 1;
 503         else if (IS_ERR(page))
 504                 return PTR_ERR(page);
 505
 506         /* Make dnode_of_data for parameter */
 507         dn->node_page = page;
 508         dn->ofs_in_node = 0;
 509         truncate_data_blocks(dn);
 510         truncate_node(dn);
 511         return 1;
 512 }
 513
 514 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 515                                                 int ofs, int depth)
 516 {
 517         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 518         struct dnode_of_data rdn = *dn;
 519         struct page *page;
 520         struct f2fs_node *rn;
 521         nid_t child_nid;
 522         unsigned int child_nofs;
 523         int freed = 0;
 524         int i, ret;
 525
 526         if (dn->nid == 0)
 527                 return NIDS_PER_BLOCK + 1;
 528
 529         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 530
 531         page = get_node_page(sbi, dn->nid);
 532         if (IS_ERR(page)) {
 533                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 534                 return PTR_ERR(page);
 535         }
 536
 537         rn = F2FS_NODE(page);
 538         if (depth < 3) {
 539                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 540                         child_nid = le32_to_cpu(rn->in.nid[i]);
 541                         if (child_nid == 0)
 542                                 continue;
 543                         rdn.nid = child_nid;
 544                         ret = truncate_dnode(&rdn);
 545                         if (ret < 0)
 546                                 goto out_err;
 547                         set_nid(page, i, 0, false);
 548                 }
 549         } else {
 550                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 551                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 552                         child_nid = le32_to_cpu(rn->in.nid[i]);
 553                         if (child_nid == 0) {
 554                                 child_nofs += NIDS_PER_BLOCK + 1;
 555                                 continue;
 556                         }
 557                         rdn.nid = child_nid;
 558                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 559                         if (ret == (NIDS_PER_BLOCK + 1)) {
 560                                 set_nid(page, i, 0, false);
 561                                 child_nofs += ret;
 562                         } else if (ret < 0 && ret != -ENOENT) {
 563                                 goto out_err;
 564                         }
 565                 }
 566                 freed = child_nofs;
 567         }
 568
 569         if (!ofs) {
 570                 /* remove current indirect node */
 571                 dn->node_page = page;
 572                 truncate_node(dn);
 573                 freed++;
 574         } else {
 575                 f2fs_put_page(page, 1);
 576         }
 577         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 578         return freed;
 579
 580 out_err:
 581         f2fs_put_page(page, 1);
 582         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 583         return ret;
 584 }
 585
 586 static int truncate_partial_nodes(struct dnode_of_data *dn,
 587                         struct f2fs_inode *ri, int *offset, int depth)
 588 {
 589         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 590         struct page *pages[2];
 591         nid_t nid[3];
 592         nid_t child_nid;
 593         int err = 0;
 594         int i;
 595         int idx = depth - 2;
 596
 597         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 598         if (!nid[0])
 599                 return 0;
 600
 601         /* get indirect nodes in the path */
 602         for (i = 0; i < idx + 1; i++) {
 603                 /* refernece count'll be increased */
 604                 pages[i] = get_node_page(sbi, nid[i]);
 605                 if (IS_ERR(pages[i])) {
 606                         err = PTR_ERR(pages[i]);
 607                         idx = i - 1;
 608                         goto fail;
 609                 }
 610                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 611         }
 612
 613         /* free direct nodes linked to a partial indirect node */
 614         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 615                 child_nid = get_nid(pages[idx], i, false);
 616                 if (!child_nid)
 617                         continue;
 618                 dn->nid = child_nid;
 619                 err = truncate_dnode(dn);
 620                 if (err < 0)
 621                         goto fail;
 622                 set_nid(pages[idx], i, 0, false);
 623         }
 624
 625         if (offset[idx + 1] == 0) {
 626                 dn->node_page = pages[idx];
 627                 dn->nid = nid[idx];
 628                 truncate_node(dn);
 629         } else {
 630                 f2fs_put_page(pages[idx], 1);
 631         }
 632         offset[idx]++;
 633         offset[idx + 1] = 0;
 634         idx--;
 635 fail:
 636         for (i = idx; i >= 0; i--)
 637                 f2fs_put_page(pages[i], 1);
 638
 639         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 640
 641         return err;
 642 }
 643
 644 /*
 645  * All the block addresses of data and nodes should be nullified.
 646  */
 647 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 648 {
 649         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 650         int err = 0, cont = 1;
 651         int level, offset[4], noffset[4];
 652         unsigned int nofs = 0;
 653         struct f2fs_inode *ri;
 654         struct dnode_of_data dn;
 655         struct page *page;
 656
 657         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 658
 659         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 660 restart:
 661         page = get_node_page(sbi, inode->i_ino);
 662         if (IS_ERR(page)) {
 663                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 664                 return PTR_ERR(page);
 665         }
 666
 667         set_new_dnode(&dn, inode, page, NULL, 0);
 668         unlock_page(page);
 669
 670         ri = F2FS_INODE(page);
 671         switch (level) {
 672         case 0:
 673         case 1:
 674                 nofs = noffset[1];
 675                 break;
 676         case 2:
 677                 nofs = noffset[1];
 678                 if (!offset[level - 1])
 679                         goto skip_partial;
 680                 err = truncate_partial_nodes(&dn, ri, offset, level);
 681                 if (err < 0 && err != -ENOENT)
 682                         goto fail;
 683                 nofs += 1 + NIDS_PER_BLOCK;
 684                 break;
 685         case 3:
 686                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 687                 if (!offset[level - 1])
 688                         goto skip_partial;
 689                 err = truncate_partial_nodes(&dn, ri, offset, level);
 690                 if (err < 0 && err != -ENOENT)
 691                         goto fail;
 692                 break;
 693         default:
 694                 BUG();
 695         }
 696
 697 skip_partial:
 698         while (cont) {
 699                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 700                 switch (offset[0]) {
 701                 case NODE_DIR1_BLOCK:
 702                 case NODE_DIR2_BLOCK:
 703                         err = truncate_dnode(&dn);
 704                         break;
 705
 706                 case NODE_IND1_BLOCK:
 707                 case NODE_IND2_BLOCK:
 708                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 709                         break;
 710
 711                 case NODE_DIND_BLOCK:
 712                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 713                         cont = 0;
 714                         break;
 715
 716                 default:
 717                         BUG();
 718                 }
 719                 if (err < 0 && err != -ENOENT)
 720                         goto fail;
 721                 if (offset[1] == 0 &&
 722                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 723                         lock_page(page);
 724                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 725                                 f2fs_put_page(page, 1);
 726                                 goto restart;
 727                         }
 728                         f2fs_wait_on_page_writeback(page, NODE);
 729                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 730                         set_page_dirty(page);
 731                         unlock_page(page);
 732                 }
 733                 offset[1] = 0;
 734                 offset[0]++;
 735                 nofs += err;
 736         }
 737 fail:
 738         f2fs_put_page(page, 0);
 739         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 740         return err > 0 ? 0 : err;
 741 }
 742
 743 int truncate_xattr_node(struct inode *inode, struct page *page)
 744 {
 745         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 746         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 747         struct dnode_of_data dn;
 748         struct page *npage;
 749
 750         if (!nid)
 751                 return 0;
 752
 753         npage = get_node_page(sbi, nid);
 754         if (IS_ERR(npage))
 755                 return PTR_ERR(npage);
 756
 757         F2FS_I(inode)->i_xattr_nid = 0;
 758
 759         /* need to do checkpoint during fsync */
 760         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 761
 762         set_new_dnode(&dn, inode, page, npage, nid);
 763
 764         if (page)
 765                 dn.inode_page_locked = true;
 766         truncate_node(&dn);
 767         return 0;
 768 }
 769
 770 /*
 771  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 772  * f2fs_unlock_op().
 773  */
 774 void remove_inode_page(struct inode *inode)
 775 {
 776         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 777         struct page *page;
 778         nid_t ino = inode->i_ino;
 779         struct dnode_of_data dn;
 780
 781         page = get_node_page(sbi, ino);
 782         if (IS_ERR(page))
 783                 return;
 784
 785         if (truncate_xattr_node(inode, page)) {
 786                 f2fs_put_page(page, 1);
 787                 return;
 788         }
 789         /* 0 is possible, after f2fs_new_inode() is failed */
 790         f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
 791         set_new_dnode(&dn, inode, page, page, ino);
 792         truncate_node(&dn);
 793 }
 794
 795 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
 796 {
 797         struct dnode_of_data dn;
 798
 799         /* allocate inode page for new inode */
 800         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 801
 802         /* caller should f2fs_put_page(page, 1); */
 803         return new_node_page(&dn, 0, NULL);
 804 }
 805
 806 struct page *new_node_page(struct dnode_of_data *dn,
 807                                 unsigned int ofs, struct page *ipage)
 808 {
 809         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 810         struct node_info old_ni, new_ni;
 811         struct page *page;
 812         int err;
 813
 814         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 815                 return ERR_PTR(-EPERM);
 816
 817         page = grab_cache_page_write_begin(NODE_MAPPING(sbi),
 818                                         dn->nid, AOP_FLAG_NOFS);
 819         if (!page)
 820                 return ERR_PTR(-ENOMEM);
 821
 822         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 823                 err = -ENOSPC;
 824                 goto fail;
 825         }
 826
 827         get_node_info(sbi, dn->nid, &old_ni);
 828
 829         /* Reinitialize old_ni with new node page */
 830         f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
 831         new_ni = old_ni;
 832         new_ni.ino = dn->inode->i_ino;
 833         set_node_addr(sbi, &new_ni, NEW_ADDR);
 834
 835         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 836         set_cold_node(dn->inode, page);
 837         SetPageUptodate(page);
 838         set_page_dirty(page);
 839
 840         if (f2fs_has_xattr_block(ofs))
 841                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 842
 843         dn->node_page = page;
 844         if (ipage)
 845                 update_inode(dn->inode, ipage);
 846         else
 847                 sync_inode_page(dn);
 848         if (ofs == 0)
 849                 inc_valid_inode_count(sbi);
 850
 851         return page;
 852
 853 fail:
 854         clear_node_page_dirty(page);
 855         f2fs_put_page(page, 1);
 856         return ERR_PTR(err);
 857 }
 858
 859 /*
 860  * Caller should do after getting the following values.
 861  * 0: f2fs_put_page(page, 0)
 862  * LOCKED_PAGE: f2fs_put_page(page, 1)
 863  * error: nothing
 864  */
 865 static int read_node_page(struct page *page, int rw)
 866 {
 867         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 868         struct node_info ni;
 869
 870         get_node_info(sbi, page->index, &ni);
 871
 872         if (unlikely(ni.blk_addr == NULL_ADDR)) {
 873                 f2fs_put_page(page, 1);
 874                 return -ENOENT;
 875         }
 876
 877         if (PageUptodate(page))
 878                 return LOCKED_PAGE;
 879
 880         return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
 881 }
 882
 883 /*
 884  * Readahead a node page
 885  */
 886 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 887 {
 888         struct page *apage;
 889         int err;
 890
 891         apage = find_get_page(NODE_MAPPING(sbi), nid);
 892         if (apage && PageUptodate(apage)) {
 893                 f2fs_put_page(apage, 0);
 894                 return;
 895         }
 896         f2fs_put_page(apage, 0);
 897
 898         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
 899         if (!apage)
 900                 return;
 901
 902         err = read_node_page(apage, READA);
 903         if (err == 0)
 904                 f2fs_put_page(apage, 0);
 905         else if (err == LOCKED_PAGE)
 906                 f2fs_put_page(apage, 1);
 907 }
 908
 909 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 910 {
 911         struct page *page;
 912         int err;
 913 repeat:
 914         page = grab_cache_page_write_begin(NODE_MAPPING(sbi),
 915                                         nid, AOP_FLAG_NOFS);
 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                                 f2fs_wait_on_page_writeback(page, NODE);
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         f2fs_wait_on_page_writeback(page, NODE);
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         account_page_redirty(page);
1199         set_page_dirty(page);
1200         return AOP_WRITEPAGE_ACTIVATE;
1201 }
1202
1203 static int f2fs_write_node_pages(struct address_space *mapping,
1204                             struct writeback_control *wbc)
1205 {
1206         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1207         long diff;
1208
1209         /* balancing f2fs's metadata in background */
1210         f2fs_balance_fs_bg(sbi);
1211
1212         /* collect a number of dirty node pages and write together */
1213         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1214                 goto skip_write;
1215
1216         diff = nr_pages_to_write(sbi, NODE, wbc);
1217         wbc->sync_mode = WB_SYNC_NONE;
1218         sync_node_pages(sbi, 0, wbc);
1219         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1220         return 0;
1221
1222 skip_write:
1223         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1224         return 0;
1225 }
1226
1227 static int f2fs_set_node_page_dirty(struct page *page)
1228 {
1229         struct address_space *mapping = page->mapping;
1230         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1231
1232         trace_f2fs_set_page_dirty(page, NODE);
1233
1234         SetPageUptodate(page);
1235         if (!PageDirty(page)) {
1236                 __set_page_dirty_nobuffers(page);
1237                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1238                 SetPagePrivate(page);
1239                 return 1;
1240         }
1241         return 0;
1242 }
1243
1244 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1245                                       unsigned int length)
1246 {
1247         struct inode *inode = page->mapping->host;
1248         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1249         if (PageDirty(page))
1250                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1251         ClearPagePrivate(page);
1252 }
1253
1254 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1255 {
1256         ClearPagePrivate(page);
1257         return 1;
1258 }
1259
1260 /*
1261  * Structure of the f2fs node operations
1262  */
1263 const struct address_space_operations f2fs_node_aops = {
1264         .writepage      = f2fs_write_node_page,
1265         .writepages     = f2fs_write_node_pages,
1266         .set_page_dirty = f2fs_set_node_page_dirty,
1267         .invalidatepage = f2fs_invalidate_node_page,
1268         .releasepage    = f2fs_release_node_page,
1269 };
1270
1271 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1272                                                 nid_t n)
1273 {
1274         return radix_tree_lookup(&nm_i->free_nid_root, n);
1275 }
1276
1277 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1278                                                 struct free_nid *i)
1279 {
1280         list_del(&i->list);
1281         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1282         kmem_cache_free(free_nid_slab, i);
1283 }
1284
1285 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build)
1286 {
1287         struct free_nid *i;
1288         struct nat_entry *ne;
1289         bool allocated = false;
1290
1291         if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1292                 return -1;
1293
1294         /* 0 nid should not be used */
1295         if (unlikely(nid == 0))
1296                 return 0;
1297
1298         if (build) {
1299                 /* do not add allocated nids */
1300                 read_lock(&nm_i->nat_tree_lock);
1301                 ne = __lookup_nat_cache(nm_i, nid);
1302                 if (ne &&
1303                         (!ne->checkpointed || nat_get_blkaddr(ne) != NULL_ADDR))
1304                         allocated = true;
1305                 read_unlock(&nm_i->nat_tree_lock);
1306                 if (allocated)
1307                         return 0;
1308         }
1309
1310         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1311         i->nid = nid;
1312         i->state = NID_NEW;
1313
1314         spin_lock(&nm_i->free_nid_list_lock);
1315         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1316                 spin_unlock(&nm_i->free_nid_list_lock);
1317                 kmem_cache_free(free_nid_slab, i);
1318                 return 0;
1319         }
1320         list_add_tail(&i->list, &nm_i->free_nid_list);
1321         nm_i->fcnt++;
1322         spin_unlock(&nm_i->free_nid_list_lock);
1323         return 1;
1324 }
1325
1326 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1327 {
1328         struct free_nid *i;
1329         spin_lock(&nm_i->free_nid_list_lock);
1330         i = __lookup_free_nid_list(nm_i, nid);
1331         if (i && i->state == NID_NEW) {
1332                 __del_from_free_nid_list(nm_i, i);
1333                 nm_i->fcnt--;
1334         }
1335         spin_unlock(&nm_i->free_nid_list_lock);
1336 }
1337
1338 static void scan_nat_page(struct f2fs_nm_info *nm_i,
1339                         struct page *nat_page, nid_t start_nid)
1340 {
1341         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1342         block_t blk_addr;
1343         int i;
1344
1345         i = start_nid % NAT_ENTRY_PER_BLOCK;
1346
1347         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1348
1349                 if (unlikely(start_nid >= nm_i->max_nid))
1350                         break;
1351
1352                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1353                 f2fs_bug_on(blk_addr == NEW_ADDR);
1354                 if (blk_addr == NULL_ADDR) {
1355                         if (add_free_nid(nm_i, start_nid, true) < 0)
1356                                 break;
1357                 }
1358         }
1359 }
1360
1361 static void build_free_nids(struct f2fs_sb_info *sbi)
1362 {
1363         struct f2fs_nm_info *nm_i = NM_I(sbi);
1364         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1365         struct f2fs_summary_block *sum = curseg->sum_blk;
1366         int i = 0;
1367         nid_t nid = nm_i->next_scan_nid;
1368
1369         /* Enough entries */
1370         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1371                 return;
1372
1373         /* readahead nat pages to be scanned */
1374         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1375
1376         while (1) {
1377                 struct page *page = get_current_nat_page(sbi, nid);
1378
1379                 scan_nat_page(nm_i, page, nid);
1380                 f2fs_put_page(page, 1);
1381
1382                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1383                 if (unlikely(nid >= nm_i->max_nid))
1384                         nid = 0;
1385
1386                 if (i++ == FREE_NID_PAGES)
1387                         break;
1388         }
1389
1390         /* go to the next free nat pages to find free nids abundantly */
1391         nm_i->next_scan_nid = nid;
1392
1393         /* find free nids from current sum_pages */
1394         mutex_lock(&curseg->curseg_mutex);
1395         for (i = 0; i < nats_in_cursum(sum); i++) {
1396                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1397                 nid = le32_to_cpu(nid_in_journal(sum, i));
1398                 if (addr == NULL_ADDR)
1399                         add_free_nid(nm_i, nid, true);
1400                 else
1401                         remove_free_nid(nm_i, nid);
1402         }
1403         mutex_unlock(&curseg->curseg_mutex);
1404 }
1405
1406 /*
1407  * If this function returns success, caller can obtain a new nid
1408  * from second parameter of this function.
1409  * The returned nid could be used ino as well as nid when inode is created.
1410  */
1411 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1412 {
1413         struct f2fs_nm_info *nm_i = NM_I(sbi);
1414         struct free_nid *i = NULL;
1415         struct list_head *this;
1416 retry:
1417         if (unlikely(sbi->total_valid_node_count + 1 >= nm_i->max_nid))
1418                 return false;
1419
1420         spin_lock(&nm_i->free_nid_list_lock);
1421
1422         /* We should not use stale free nids created by build_free_nids */
1423         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1424                 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1425                 list_for_each(this, &nm_i->free_nid_list) {
1426                         i = list_entry(this, struct free_nid, list);
1427                         if (i->state == NID_NEW)
1428                                 break;
1429                 }
1430
1431                 f2fs_bug_on(i->state != NID_NEW);
1432                 *nid = i->nid;
1433                 i->state = NID_ALLOC;
1434                 nm_i->fcnt--;
1435                 spin_unlock(&nm_i->free_nid_list_lock);
1436                 return true;
1437         }
1438         spin_unlock(&nm_i->free_nid_list_lock);
1439
1440         /* Let's scan nat pages and its caches to get free nids */
1441         mutex_lock(&nm_i->build_lock);
1442         build_free_nids(sbi);
1443         mutex_unlock(&nm_i->build_lock);
1444         goto retry;
1445 }
1446
1447 /*
1448  * alloc_nid() should be called prior to this function.
1449  */
1450 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1451 {
1452         struct f2fs_nm_info *nm_i = NM_I(sbi);
1453         struct free_nid *i;
1454
1455         spin_lock(&nm_i->free_nid_list_lock);
1456         i = __lookup_free_nid_list(nm_i, nid);
1457         f2fs_bug_on(!i || i->state != NID_ALLOC);
1458         __del_from_free_nid_list(nm_i, i);
1459         spin_unlock(&nm_i->free_nid_list_lock);
1460 }
1461
1462 /*
1463  * alloc_nid() should be called prior to this function.
1464  */
1465 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1466 {
1467         struct f2fs_nm_info *nm_i = NM_I(sbi);
1468         struct free_nid *i;
1469
1470         if (!nid)
1471                 return;
1472
1473         spin_lock(&nm_i->free_nid_list_lock);
1474         i = __lookup_free_nid_list(nm_i, nid);
1475         f2fs_bug_on(!i || i->state != NID_ALLOC);
1476         if (nm_i->fcnt > 2 * MAX_FREE_NIDS) {
1477                 __del_from_free_nid_list(nm_i, i);
1478         } else {
1479                 i->state = NID_NEW;
1480                 nm_i->fcnt++;
1481         }
1482         spin_unlock(&nm_i->free_nid_list_lock);
1483 }
1484
1485 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1486                 struct f2fs_summary *sum, struct node_info *ni,
1487                 block_t new_blkaddr)
1488 {
1489         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1490         set_node_addr(sbi, ni, new_blkaddr);
1491         clear_node_page_dirty(page);
1492 }
1493
1494 void recover_inline_xattr(struct inode *inode, struct page *page)
1495 {
1496         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1497         void *src_addr, *dst_addr;
1498         size_t inline_size;
1499         struct page *ipage;
1500         struct f2fs_inode *ri;
1501
1502         if (!f2fs_has_inline_xattr(inode))
1503                 return;
1504
1505         if (!IS_INODE(page))
1506                 return;
1507
1508         ri = F2FS_INODE(page);
1509         if (!(ri->i_inline & F2FS_INLINE_XATTR))
1510                 return;
1511
1512         ipage = get_node_page(sbi, inode->i_ino);
1513         f2fs_bug_on(IS_ERR(ipage));
1514
1515         dst_addr = inline_xattr_addr(ipage);
1516         src_addr = inline_xattr_addr(page);
1517         inline_size = inline_xattr_size(inode);
1518
1519         memcpy(dst_addr, src_addr, inline_size);
1520
1521         update_inode(inode, ipage);
1522         f2fs_put_page(ipage, 1);
1523 }
1524
1525 bool recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1526 {
1527         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1528         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1529         nid_t new_xnid = nid_of_node(page);
1530         struct node_info ni;
1531
1532         recover_inline_xattr(inode, page);
1533
1534         if (!f2fs_has_xattr_block(ofs_of_node(page)))
1535                 return false;
1536
1537         /* 1: invalidate the previous xattr nid */
1538         if (!prev_xnid)
1539                 goto recover_xnid;
1540
1541         /* Deallocate node address */
1542         get_node_info(sbi, prev_xnid, &ni);
1543         f2fs_bug_on(ni.blk_addr == NULL_ADDR);
1544         invalidate_blocks(sbi, ni.blk_addr);
1545         dec_valid_node_count(sbi, inode);
1546         set_node_addr(sbi, &ni, NULL_ADDR);
1547
1548 recover_xnid:
1549         /* 2: allocate new xattr nid */
1550         if (unlikely(!inc_valid_node_count(sbi, inode)))
1551                 f2fs_bug_on(1);
1552
1553         remove_free_nid(NM_I(sbi), new_xnid);
1554         get_node_info(sbi, new_xnid, &ni);
1555         ni.ino = inode->i_ino;
1556         set_node_addr(sbi, &ni, NEW_ADDR);
1557         F2FS_I(inode)->i_xattr_nid = new_xnid;
1558
1559         /* 3: update xattr blkaddr */
1560         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1561         set_node_addr(sbi, &ni, blkaddr);
1562
1563         update_inode_page(inode);
1564         return true;
1565 }
1566
1567 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1568 {
1569         struct f2fs_inode *src, *dst;
1570         nid_t ino = ino_of_node(page);
1571         struct node_info old_ni, new_ni;
1572         struct page *ipage;
1573
1574         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1575         if (!ipage)
1576                 return -ENOMEM;
1577
1578         /* Should not use this inode  from free nid list */
1579         remove_free_nid(NM_I(sbi), ino);
1580
1581         get_node_info(sbi, ino, &old_ni);
1582         SetPageUptodate(ipage);
1583         fill_node_footer(ipage, ino, ino, 0, true);
1584
1585         src = F2FS_INODE(page);
1586         dst = F2FS_INODE(ipage);
1587
1588         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1589         dst->i_size = 0;
1590         dst->i_blocks = cpu_to_le64(1);
1591         dst->i_links = cpu_to_le32(1);
1592         dst->i_xattr_nid = 0;
1593
1594         new_ni = old_ni;
1595         new_ni.ino = ino;
1596
1597         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1598                 WARN_ON(1);
1599         set_node_addr(sbi, &new_ni, NEW_ADDR);
1600         inc_valid_inode_count(sbi);
1601         f2fs_put_page(ipage, 1);
1602         return 0;
1603 }
1604
1605 /*
1606  * ra_sum_pages() merge contiguous pages into one bio and submit.
1607  * these pre-readed pages are linked in pages list.
1608  */
1609 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
1610                                 int start, int nrpages)
1611 {
1612         struct page *page;
1613         int page_idx = start;
1614         struct f2fs_io_info fio = {
1615                 .type = META,
1616                 .rw = READ_SYNC | REQ_META | REQ_PRIO
1617         };
1618
1619         for (; page_idx < start + nrpages; page_idx++) {
1620                 /* alloc temporal page for read node summary info*/
1621                 page = alloc_page(GFP_F2FS_ZERO);
1622                 if (!page)
1623                         break;
1624
1625                 lock_page(page);
1626                 page->index = page_idx;
1627                 list_add_tail(&page->lru, pages);
1628         }
1629
1630         list_for_each_entry(page, pages, lru)
1631                 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1632
1633         f2fs_submit_merged_bio(sbi, META, READ);
1634
1635         return page_idx - start;
1636 }
1637
1638 int restore_node_summary(struct f2fs_sb_info *sbi,
1639                         unsigned int segno, struct f2fs_summary_block *sum)
1640 {
1641         struct f2fs_node *rn;
1642         struct f2fs_summary *sum_entry;
1643         struct page *page, *tmp;
1644         block_t addr;
1645         int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1646         int i, last_offset, nrpages, err = 0;
1647         LIST_HEAD(page_list);
1648
1649         /* scan the node segment */
1650         last_offset = sbi->blocks_per_seg;
1651         addr = START_BLOCK(sbi, segno);
1652         sum_entry = &sum->entries[0];
1653
1654         for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) {
1655                 nrpages = min(last_offset - i, bio_blocks);
1656
1657                 /* read ahead node pages */
1658                 nrpages = ra_sum_pages(sbi, &page_list, addr, nrpages);
1659                 if (!nrpages)
1660                         return -ENOMEM;
1661
1662                 list_for_each_entry_safe(page, tmp, &page_list, lru) {
1663                         if (err)
1664                                 goto skip;
1665
1666                         lock_page(page);
1667                         if (unlikely(!PageUptodate(page))) {
1668                                 err = -EIO;
1669                         } else {
1670                                 rn = F2FS_NODE(page);
1671                                 sum_entry->nid = rn->footer.nid;
1672                                 sum_entry->version = 0;
1673                                 sum_entry->ofs_in_node = 0;
1674                                 sum_entry++;
1675                         }
1676                         unlock_page(page);
1677 skip:
1678                         list_del(&page->lru);
1679                         __free_pages(page, 0);
1680                 }
1681         }
1682         return err;
1683 }
1684
1685 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1686 {
1687         struct f2fs_nm_info *nm_i = NM_I(sbi);
1688         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1689         struct f2fs_summary_block *sum = curseg->sum_blk;
1690         int i;
1691
1692         mutex_lock(&curseg->curseg_mutex);
1693
1694         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1695                 mutex_unlock(&curseg->curseg_mutex);
1696                 return false;
1697         }
1698
1699         for (i = 0; i < nats_in_cursum(sum); i++) {
1700                 struct nat_entry *ne;
1701                 struct f2fs_nat_entry raw_ne;
1702                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1703
1704                 raw_ne = nat_in_journal(sum, i);
1705 retry:
1706                 write_lock(&nm_i->nat_tree_lock);
1707                 ne = __lookup_nat_cache(nm_i, nid);
1708                 if (ne) {
1709                         __set_nat_cache_dirty(nm_i, ne);
1710                         write_unlock(&nm_i->nat_tree_lock);
1711                         continue;
1712                 }
1713                 ne = grab_nat_entry(nm_i, nid);
1714                 if (!ne) {
1715                         write_unlock(&nm_i->nat_tree_lock);
1716                         goto retry;
1717                 }
1718                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1719                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1720                 nat_set_version(ne, raw_ne.version);
1721                 __set_nat_cache_dirty(nm_i, ne);
1722                 write_unlock(&nm_i->nat_tree_lock);
1723         }
1724         update_nats_in_cursum(sum, -i);
1725         mutex_unlock(&curseg->curseg_mutex);
1726         return true;
1727 }
1728
1729 /*
1730  * This function is called during the checkpointing process.
1731  */
1732 void flush_nat_entries(struct f2fs_sb_info *sbi)
1733 {
1734         struct f2fs_nm_info *nm_i = NM_I(sbi);
1735         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1736         struct f2fs_summary_block *sum = curseg->sum_blk;
1737         struct list_head *cur, *n;
1738         struct page *page = NULL;
1739         struct f2fs_nat_block *nat_blk = NULL;
1740         nid_t start_nid = 0, end_nid = 0;
1741         bool flushed;
1742
1743         flushed = flush_nats_in_journal(sbi);
1744
1745         if (!flushed)
1746                 mutex_lock(&curseg->curseg_mutex);
1747
1748         /* 1) flush dirty nat caches */
1749         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1750                 struct nat_entry *ne;
1751                 nid_t nid;
1752                 struct f2fs_nat_entry raw_ne;
1753                 int offset = -1;
1754                 block_t new_blkaddr;
1755
1756                 ne = list_entry(cur, struct nat_entry, list);
1757                 nid = nat_get_nid(ne);
1758
1759                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1760                         continue;
1761                 if (flushed)
1762                         goto to_nat_page;
1763
1764                 /* if there is room for nat enries in curseg->sumpage */
1765                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1766                 if (offset >= 0) {
1767                         raw_ne = nat_in_journal(sum, offset);
1768                         goto flush_now;
1769                 }
1770 to_nat_page:
1771                 if (!page || (start_nid > nid || nid > end_nid)) {
1772                         if (page) {
1773                                 f2fs_put_page(page, 1);
1774                                 page = NULL;
1775                         }
1776                         start_nid = START_NID(nid);
1777                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1778
1779                         /*
1780                          * get nat block with dirty flag, increased reference
1781                          * count, mapped and lock
1782                          */
1783                         page = get_next_nat_page(sbi, start_nid);
1784                         nat_blk = page_address(page);
1785                 }
1786
1787                 f2fs_bug_on(!nat_blk);
1788                 raw_ne = nat_blk->entries[nid - start_nid];
1789 flush_now:
1790                 new_blkaddr = nat_get_blkaddr(ne);
1791
1792                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1793                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1794                 raw_ne.version = nat_get_version(ne);
1795
1796                 if (offset < 0) {
1797                         nat_blk->entries[nid - start_nid] = raw_ne;
1798                 } else {
1799                         nat_in_journal(sum, offset) = raw_ne;
1800                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1801                 }
1802
1803                 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1804                                 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1805                         write_lock(&nm_i->nat_tree_lock);
1806                         __del_from_nat_cache(nm_i, ne);
1807                         write_unlock(&nm_i->nat_tree_lock);
1808                 } else {
1809                         write_lock(&nm_i->nat_tree_lock);
1810                         __clear_nat_cache_dirty(nm_i, ne);
1811                         write_unlock(&nm_i->nat_tree_lock);
1812                 }
1813         }
1814         if (!flushed)
1815                 mutex_unlock(&curseg->curseg_mutex);
1816         f2fs_put_page(page, 1);
1817
1818         /* 2) shrink nat caches if necessary */
1819         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1820 }
1821
1822 static int init_node_manager(struct f2fs_sb_info *sbi)
1823 {
1824         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1825         struct f2fs_nm_info *nm_i = NM_I(sbi);
1826         unsigned char *version_bitmap;
1827         unsigned int nat_segs, nat_blocks;
1828
1829         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1830
1831         /* segment_count_nat includes pair segment so divide to 2. */
1832         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1833         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1834
1835         /* not used nids: 0, node, meta, (and root counted as valid node) */
1836         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks - 3;
1837         nm_i->fcnt = 0;
1838         nm_i->nat_cnt = 0;
1839
1840         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1841         INIT_LIST_HEAD(&nm_i->free_nid_list);
1842         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1843         INIT_LIST_HEAD(&nm_i->nat_entries);
1844         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1845
1846         mutex_init(&nm_i->build_lock);
1847         spin_lock_init(&nm_i->free_nid_list_lock);
1848         rwlock_init(&nm_i->nat_tree_lock);
1849
1850         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1851         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1852         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1853         if (!version_bitmap)
1854                 return -EFAULT;
1855
1856         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1857                                         GFP_KERNEL);
1858         if (!nm_i->nat_bitmap)
1859                 return -ENOMEM;
1860         return 0;
1861 }
1862
1863 int build_node_manager(struct f2fs_sb_info *sbi)
1864 {
1865         int err;
1866
1867         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1868         if (!sbi->nm_info)
1869                 return -ENOMEM;
1870
1871         err = init_node_manager(sbi);
1872         if (err)
1873                 return err;
1874
1875         build_free_nids(sbi);
1876         return 0;
1877 }
1878
1879 void destroy_node_manager(struct f2fs_sb_info *sbi)
1880 {
1881         struct f2fs_nm_info *nm_i = NM_I(sbi);
1882         struct free_nid *i, *next_i;
1883         struct nat_entry *natvec[NATVEC_SIZE];
1884         nid_t nid = 0;
1885         unsigned int found;
1886
1887         if (!nm_i)
1888                 return;
1889
1890         /* destroy free nid list */
1891         spin_lock(&nm_i->free_nid_list_lock);
1892         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1893                 f2fs_bug_on(i->state == NID_ALLOC);
1894                 __del_from_free_nid_list(nm_i, i);
1895                 nm_i->fcnt--;
1896         }
1897         f2fs_bug_on(nm_i->fcnt);
1898         spin_unlock(&nm_i->free_nid_list_lock);
1899
1900         /* destroy nat cache */
1901         write_lock(&nm_i->nat_tree_lock);
1902         while ((found = __gang_lookup_nat_cache(nm_i,
1903                                         nid, NATVEC_SIZE, natvec))) {
1904                 unsigned idx;
1905                 nid = nat_get_nid(natvec[found - 1]) + 1;
1906                 for (idx = 0; idx < found; idx++)
1907                         __del_from_nat_cache(nm_i, natvec[idx]);
1908         }
1909         f2fs_bug_on(nm_i->nat_cnt);
1910         write_unlock(&nm_i->nat_tree_lock);
1911
1912         kfree(nm_i->nat_bitmap);
1913         sbi->nm_info = NULL;
1914         kfree(nm_i);
1915 }
1916
1917 int __init create_node_manager_caches(void)
1918 {
1919         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1920                         sizeof(struct nat_entry));
1921         if (!nat_entry_slab)
1922                 return -ENOMEM;
1923
1924         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1925                         sizeof(struct free_nid));
1926         if (!free_nid_slab) {
1927                 kmem_cache_destroy(nat_entry_slab);
1928                 return -ENOMEM;
1929         }
1930         return 0;
1931 }
1932
1933 void destroy_node_manager_caches(void)
1934 {
1935         kmem_cache_destroy(free_nid_slab);
1936         kmem_cache_destroy(nat_entry_slab);
1937 }