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 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
  30 {
  31         struct f2fs_nm_info *nm_i = NM_I(sbi);
  32         struct sysinfo val;
  33         unsigned long mem_size = 0;
  34         bool res = false;
  35
  36         si_meminfo(&val);
  37         /* give 25%, 25%, 50% memory for each components respectively */
  38         if (type == FREE_NIDS) {
  39                 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >> 12;
  40                 res = mem_size < ((val.totalram * nm_i->ram_thresh / 100) >> 2);
  41         } else if (type == NAT_ENTRIES) {
  42                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >> 12;
  43                 res = mem_size < ((val.totalram * nm_i->ram_thresh / 100) >> 2);
  44         } else if (type == DIRTY_DENTS) {
  45                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  46                 res = mem_size < ((val.totalram * nm_i->ram_thresh / 100) >> 1);
  47         }
  48         return res;
  49 }
  50
  51 static void clear_node_page_dirty(struct page *page)
  52 {
  53         struct address_space *mapping = page->mapping;
  54         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
  55         unsigned int long flags;
  56
  57         if (PageDirty(page)) {
  58                 spin_lock_irqsave(&mapping->tree_lock, flags);
  59                 radix_tree_tag_clear(&mapping->page_tree,
  60                                 page_index(page),
  61                                 PAGECACHE_TAG_DIRTY);
  62                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  63
  64                 clear_page_dirty_for_io(page);
  65                 dec_page_count(sbi, F2FS_DIRTY_NODES);
  66         }
  67         ClearPageUptodate(page);
  68 }
  69
  70 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  71 {
  72         pgoff_t index = current_nat_addr(sbi, nid);
  73         return get_meta_page(sbi, index);
  74 }
  75
  76 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  77 {
  78         struct page *src_page;
  79         struct page *dst_page;
  80         pgoff_t src_off;
  81         pgoff_t dst_off;
  82         void *src_addr;
  83         void *dst_addr;
  84         struct f2fs_nm_info *nm_i = NM_I(sbi);
  85
  86         src_off = current_nat_addr(sbi, nid);
  87         dst_off = next_nat_addr(sbi, src_off);
  88
  89         /* get current nat block page with lock */
  90         src_page = get_meta_page(sbi, src_off);
  91
  92         /* Dirty src_page means that it is already the new target NAT page. */
  93         if (PageDirty(src_page))
  94                 return src_page;
  95
  96         dst_page = grab_meta_page(sbi, dst_off);
  97
  98         src_addr = page_address(src_page);
  99         dst_addr = page_address(dst_page);
 100         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
 101         set_page_dirty(dst_page);
 102         f2fs_put_page(src_page, 1);
 103
 104         set_to_next_nat(nm_i, nid);
 105
 106         return dst_page;
 107 }
 108
 109 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 110 {
 111         return radix_tree_lookup(&nm_i->nat_root, n);
 112 }
 113
 114 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 115                 nid_t start, unsigned int nr, struct nat_entry **ep)
 116 {
 117         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 118 }
 119
 120 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 121 {
 122         list_del(&e->list);
 123         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 124         nm_i->nat_cnt--;
 125         kmem_cache_free(nat_entry_slab, e);
 126 }
 127
 128 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 129 {
 130         struct f2fs_nm_info *nm_i = NM_I(sbi);
 131         struct nat_entry *e;
 132         int is_cp = 1;
 133
 134         read_lock(&nm_i->nat_tree_lock);
 135         e = __lookup_nat_cache(nm_i, nid);
 136         if (e && !e->checkpointed)
 137                 is_cp = 0;
 138         read_unlock(&nm_i->nat_tree_lock);
 139         return is_cp;
 140 }
 141
 142 bool fsync_mark_done(struct f2fs_sb_info *sbi, nid_t nid)
 143 {
 144         struct f2fs_nm_info *nm_i = NM_I(sbi);
 145         struct nat_entry *e;
 146         bool fsync_done = false;
 147
 148         read_lock(&nm_i->nat_tree_lock);
 149         e = __lookup_nat_cache(nm_i, nid);
 150         if (e)
 151                 fsync_done = e->fsync_done;
 152         read_unlock(&nm_i->nat_tree_lock);
 153         return fsync_done;
 154 }
 155
 156 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 157 {
 158         struct nat_entry *new;
 159
 160         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 161         if (!new)
 162                 return NULL;
 163         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 164                 kmem_cache_free(nat_entry_slab, new);
 165                 return NULL;
 166         }
 167         memset(new, 0, sizeof(struct nat_entry));
 168         nat_set_nid(new, nid);
 169         new->checkpointed = true;
 170         list_add_tail(&new->list, &nm_i->nat_entries);
 171         nm_i->nat_cnt++;
 172         return new;
 173 }
 174
 175 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 176                                                 struct f2fs_nat_entry *ne)
 177 {
 178         struct nat_entry *e;
 179 retry:
 180         write_lock(&nm_i->nat_tree_lock);
 181         e = __lookup_nat_cache(nm_i, nid);
 182         if (!e) {
 183                 e = grab_nat_entry(nm_i, nid);
 184                 if (!e) {
 185                         write_unlock(&nm_i->nat_tree_lock);
 186                         goto retry;
 187                 }
 188                 node_info_from_raw_nat(&e->ni, ne);
 189         }
 190         write_unlock(&nm_i->nat_tree_lock);
 191 }
 192
 193 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 194                         block_t new_blkaddr, bool fsync_done)
 195 {
 196         struct f2fs_nm_info *nm_i = NM_I(sbi);
 197         struct nat_entry *e;
 198 retry:
 199         write_lock(&nm_i->nat_tree_lock);
 200         e = __lookup_nat_cache(nm_i, ni->nid);
 201         if (!e) {
 202                 e = grab_nat_entry(nm_i, ni->nid);
 203                 if (!e) {
 204                         write_unlock(&nm_i->nat_tree_lock);
 205                         goto retry;
 206                 }
 207                 e->ni = *ni;
 208                 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
 209         } else if (new_blkaddr == NEW_ADDR) {
 210                 /*
 211                  * when nid is reallocated,
 212                  * previous nat entry can be remained in nat cache.
 213                  * So, reinitialize it with new information.
 214                  */
 215                 e->ni = *ni;
 216                 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
 217         }
 218
 219         /* sanity check */
 220         f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
 221         f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
 222                         new_blkaddr == NULL_ADDR);
 223         f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
 224                         new_blkaddr == NEW_ADDR);
 225         f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
 226                         nat_get_blkaddr(e) != NULL_ADDR &&
 227                         new_blkaddr == NEW_ADDR);
 228
 229         /* increament version no as node is removed */
 230         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 231                 unsigned char version = nat_get_version(e);
 232                 nat_set_version(e, inc_node_version(version));
 233         }
 234
 235         /* change address */
 236         nat_set_blkaddr(e, new_blkaddr);
 237         __set_nat_cache_dirty(nm_i, e);
 238
 239         /* update fsync_mark if its inode nat entry is still alive */
 240         e = __lookup_nat_cache(nm_i, ni->ino);
 241         if (e)
 242                 e->fsync_done = fsync_done;
 243         write_unlock(&nm_i->nat_tree_lock);
 244 }
 245
 246 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 247 {
 248         struct f2fs_nm_info *nm_i = NM_I(sbi);
 249
 250         if (available_free_memory(sbi, NAT_ENTRIES))
 251                 return 0;
 252
 253         write_lock(&nm_i->nat_tree_lock);
 254         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 255                 struct nat_entry *ne;
 256                 ne = list_first_entry(&nm_i->nat_entries,
 257                                         struct nat_entry, list);
 258                 __del_from_nat_cache(nm_i, ne);
 259                 nr_shrink--;
 260         }
 261         write_unlock(&nm_i->nat_tree_lock);
 262         return nr_shrink;
 263 }
 264
 265 /*
 266  * This function returns always success
 267  */
 268 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 269 {
 270         struct f2fs_nm_info *nm_i = NM_I(sbi);
 271         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 272         struct f2fs_summary_block *sum = curseg->sum_blk;
 273         nid_t start_nid = START_NID(nid);
 274         struct f2fs_nat_block *nat_blk;
 275         struct page *page = NULL;
 276         struct f2fs_nat_entry ne;
 277         struct nat_entry *e;
 278         int i;
 279
 280         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 281         ni->nid = nid;
 282
 283         /* Check nat cache */
 284         read_lock(&nm_i->nat_tree_lock);
 285         e = __lookup_nat_cache(nm_i, nid);
 286         if (e) {
 287                 ni->ino = nat_get_ino(e);
 288                 ni->blk_addr = nat_get_blkaddr(e);
 289                 ni->version = nat_get_version(e);
 290         }
 291         read_unlock(&nm_i->nat_tree_lock);
 292         if (e)
 293                 return;
 294
 295         /* Check current segment summary */
 296         mutex_lock(&curseg->curseg_mutex);
 297         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 298         if (i >= 0) {
 299                 ne = nat_in_journal(sum, i);
 300                 node_info_from_raw_nat(ni, &ne);
 301         }
 302         mutex_unlock(&curseg->curseg_mutex);
 303         if (i >= 0)
 304                 goto cache;
 305
 306         /* Fill node_info from nat page */
 307         page = get_current_nat_page(sbi, start_nid);
 308         nat_blk = (struct f2fs_nat_block *)page_address(page);
 309         ne = nat_blk->entries[nid - start_nid];
 310         node_info_from_raw_nat(ni, &ne);
 311         f2fs_put_page(page, 1);
 312 cache:
 313         /* cache nat entry */
 314         cache_nat_entry(NM_I(sbi), nid, &ne);
 315 }
 316
 317 /*
 318  * The maximum depth is four.
 319  * Offset[0] will have raw inode offset.
 320  */
 321 static int get_node_path(struct f2fs_inode_info *fi, long block,
 322                                 int offset[4], unsigned int noffset[4])
 323 {
 324         const long direct_index = ADDRS_PER_INODE(fi);
 325         const long direct_blks = ADDRS_PER_BLOCK;
 326         const long dptrs_per_blk = NIDS_PER_BLOCK;
 327         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 328         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 329         int n = 0;
 330         int level = 0;
 331
 332         noffset[0] = 0;
 333
 334         if (block < direct_index) {
 335                 offset[n] = block;
 336                 goto got;
 337         }
 338         block -= direct_index;
 339         if (block < direct_blks) {
 340                 offset[n++] = NODE_DIR1_BLOCK;
 341                 noffset[n] = 1;
 342                 offset[n] = block;
 343                 level = 1;
 344                 goto got;
 345         }
 346         block -= direct_blks;
 347         if (block < direct_blks) {
 348                 offset[n++] = NODE_DIR2_BLOCK;
 349                 noffset[n] = 2;
 350                 offset[n] = block;
 351                 level = 1;
 352                 goto got;
 353         }
 354         block -= direct_blks;
 355         if (block < indirect_blks) {
 356                 offset[n++] = NODE_IND1_BLOCK;
 357                 noffset[n] = 3;
 358                 offset[n++] = block / direct_blks;
 359                 noffset[n] = 4 + offset[n - 1];
 360                 offset[n] = block % direct_blks;
 361                 level = 2;
 362                 goto got;
 363         }
 364         block -= indirect_blks;
 365         if (block < indirect_blks) {
 366                 offset[n++] = NODE_IND2_BLOCK;
 367                 noffset[n] = 4 + dptrs_per_blk;
 368                 offset[n++] = block / direct_blks;
 369                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 370                 offset[n] = block % direct_blks;
 371                 level = 2;
 372                 goto got;
 373         }
 374         block -= indirect_blks;
 375         if (block < dindirect_blks) {
 376                 offset[n++] = NODE_DIND_BLOCK;
 377                 noffset[n] = 5 + (dptrs_per_blk * 2);
 378                 offset[n++] = block / indirect_blks;
 379                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 380                               offset[n - 1] * (dptrs_per_blk + 1);
 381                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 382                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 383                               offset[n - 2] * (dptrs_per_blk + 1) +
 384                               offset[n - 1];
 385                 offset[n] = block % direct_blks;
 386                 level = 3;
 387                 goto got;
 388         } else {
 389                 BUG();
 390         }
 391 got:
 392         return level;
 393 }
 394
 395 /*
 396  * Caller should call f2fs_put_dnode(dn).
 397  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 398  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 399  * In the case of RDONLY_NODE, we don't need to care about mutex.
 400  */
 401 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 402 {
 403         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 404         struct page *npage[4];
 405         struct page *parent;
 406         int offset[4];
 407         unsigned int noffset[4];
 408         nid_t nids[4];
 409         int level, i;
 410         int err = 0;
 411
 412         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 413
 414         nids[0] = dn->inode->i_ino;
 415         npage[0] = dn->inode_page;
 416
 417         if (!npage[0]) {
 418                 npage[0] = get_node_page(sbi, nids[0]);
 419                 if (IS_ERR(npage[0]))
 420                         return PTR_ERR(npage[0]);
 421         }
 422         parent = npage[0];
 423         if (level != 0)
 424                 nids[1] = get_nid(parent, offset[0], true);
 425         dn->inode_page = npage[0];
 426         dn->inode_page_locked = true;
 427
 428         /* get indirect or direct nodes */
 429         for (i = 1; i <= level; i++) {
 430                 bool done = false;
 431
 432                 if (!nids[i] && mode == ALLOC_NODE) {
 433                         /* alloc new node */
 434                         if (!alloc_nid(sbi, &(nids[i]))) {
 435                                 err = -ENOSPC;
 436                                 goto release_pages;
 437                         }
 438
 439                         dn->nid = nids[i];
 440                         npage[i] = new_node_page(dn, noffset[i], NULL);
 441                         if (IS_ERR(npage[i])) {
 442                                 alloc_nid_failed(sbi, nids[i]);
 443                                 err = PTR_ERR(npage[i]);
 444                                 goto release_pages;
 445                         }
 446
 447                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 448                         alloc_nid_done(sbi, nids[i]);
 449                         done = true;
 450                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 451                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 452                         if (IS_ERR(npage[i])) {
 453                                 err = PTR_ERR(npage[i]);
 454                                 goto release_pages;
 455                         }
 456                         done = true;
 457                 }
 458                 if (i == 1) {
 459                         dn->inode_page_locked = false;
 460                         unlock_page(parent);
 461                 } else {
 462                         f2fs_put_page(parent, 1);
 463                 }
 464
 465                 if (!done) {
 466                         npage[i] = get_node_page(sbi, nids[i]);
 467                         if (IS_ERR(npage[i])) {
 468                                 err = PTR_ERR(npage[i]);
 469                                 f2fs_put_page(npage[0], 0);
 470                                 goto release_out;
 471                         }
 472                 }
 473                 if (i < level) {
 474                         parent = npage[i];
 475                         nids[i + 1] = get_nid(parent, offset[i], false);
 476                 }
 477         }
 478         dn->nid = nids[level];
 479         dn->ofs_in_node = offset[level];
 480         dn->node_page = npage[level];
 481         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 482         return 0;
 483
 484 release_pages:
 485         f2fs_put_page(parent, 1);
 486         if (i > 1)
 487                 f2fs_put_page(npage[0], 0);
 488 release_out:
 489         dn->inode_page = NULL;
 490         dn->node_page = NULL;
 491         return err;
 492 }
 493
 494 static void truncate_node(struct dnode_of_data *dn)
 495 {
 496         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 497         struct node_info ni;
 498
 499         get_node_info(sbi, dn->nid, &ni);
 500         if (dn->inode->i_blocks == 0) {
 501                 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
 502                 goto invalidate;
 503         }
 504         f2fs_bug_on(ni.blk_addr == NULL_ADDR);
 505
 506         /* Deallocate node address */
 507         invalidate_blocks(sbi, ni.blk_addr);
 508         dec_valid_node_count(sbi, dn->inode);
 509         set_node_addr(sbi, &ni, NULL_ADDR, false);
 510
 511         if (dn->nid == dn->inode->i_ino) {
 512                 remove_orphan_inode(sbi, dn->nid);
 513                 dec_valid_inode_count(sbi);
 514         } else {
 515                 sync_inode_page(dn);
 516         }
 517 invalidate:
 518         clear_node_page_dirty(dn->node_page);
 519         F2FS_SET_SB_DIRT(sbi);
 520
 521         f2fs_put_page(dn->node_page, 1);
 522
 523         invalidate_mapping_pages(NODE_MAPPING(sbi),
 524                         dn->node_page->index, dn->node_page->index);
 525
 526         dn->node_page = NULL;
 527         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 528 }
 529
 530 static int truncate_dnode(struct dnode_of_data *dn)
 531 {
 532         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 533         struct page *page;
 534
 535         if (dn->nid == 0)
 536                 return 1;
 537
 538         /* get direct node */
 539         page = get_node_page(sbi, dn->nid);
 540         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 541                 return 1;
 542         else if (IS_ERR(page))
 543                 return PTR_ERR(page);
 544
 545         /* Make dnode_of_data for parameter */
 546         dn->node_page = page;
 547         dn->ofs_in_node = 0;
 548         truncate_data_blocks(dn);
 549         truncate_node(dn);
 550         return 1;
 551 }
 552
 553 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 554                                                 int ofs, int depth)
 555 {
 556         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 557         struct dnode_of_data rdn = *dn;
 558         struct page *page;
 559         struct f2fs_node *rn;
 560         nid_t child_nid;
 561         unsigned int child_nofs;
 562         int freed = 0;
 563         int i, ret;
 564
 565         if (dn->nid == 0)
 566                 return NIDS_PER_BLOCK + 1;
 567
 568         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 569
 570         page = get_node_page(sbi, dn->nid);
 571         if (IS_ERR(page)) {
 572                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 573                 return PTR_ERR(page);
 574         }
 575
 576         rn = F2FS_NODE(page);
 577         if (depth < 3) {
 578                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 579                         child_nid = le32_to_cpu(rn->in.nid[i]);
 580                         if (child_nid == 0)
 581                                 continue;
 582                         rdn.nid = child_nid;
 583                         ret = truncate_dnode(&rdn);
 584                         if (ret < 0)
 585                                 goto out_err;
 586                         set_nid(page, i, 0, false);
 587                 }
 588         } else {
 589                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 590                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 591                         child_nid = le32_to_cpu(rn->in.nid[i]);
 592                         if (child_nid == 0) {
 593                                 child_nofs += NIDS_PER_BLOCK + 1;
 594                                 continue;
 595                         }
 596                         rdn.nid = child_nid;
 597                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 598                         if (ret == (NIDS_PER_BLOCK + 1)) {
 599                                 set_nid(page, i, 0, false);
 600                                 child_nofs += ret;
 601                         } else if (ret < 0 && ret != -ENOENT) {
 602                                 goto out_err;
 603                         }
 604                 }
 605                 freed = child_nofs;
 606         }
 607
 608         if (!ofs) {
 609                 /* remove current indirect node */
 610                 dn->node_page = page;
 611                 truncate_node(dn);
 612                 freed++;
 613         } else {
 614                 f2fs_put_page(page, 1);
 615         }
 616         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 617         return freed;
 618
 619 out_err:
 620         f2fs_put_page(page, 1);
 621         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 622         return ret;
 623 }
 624
 625 static int truncate_partial_nodes(struct dnode_of_data *dn,
 626                         struct f2fs_inode *ri, int *offset, int depth)
 627 {
 628         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 629         struct page *pages[2];
 630         nid_t nid[3];
 631         nid_t child_nid;
 632         int err = 0;
 633         int i;
 634         int idx = depth - 2;
 635
 636         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 637         if (!nid[0])
 638                 return 0;
 639
 640         /* get indirect nodes in the path */
 641         for (i = 0; i < idx + 1; i++) {
 642                 /* refernece count'll be increased */
 643                 pages[i] = get_node_page(sbi, nid[i]);
 644                 if (IS_ERR(pages[i])) {
 645                         err = PTR_ERR(pages[i]);
 646                         idx = i - 1;
 647                         goto fail;
 648                 }
 649                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 650         }
 651
 652         /* free direct nodes linked to a partial indirect node */
 653         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 654                 child_nid = get_nid(pages[idx], i, false);
 655                 if (!child_nid)
 656                         continue;
 657                 dn->nid = child_nid;
 658                 err = truncate_dnode(dn);
 659                 if (err < 0)
 660                         goto fail;
 661                 set_nid(pages[idx], i, 0, false);
 662         }
 663
 664         if (offset[idx + 1] == 0) {
 665                 dn->node_page = pages[idx];
 666                 dn->nid = nid[idx];
 667                 truncate_node(dn);
 668         } else {
 669                 f2fs_put_page(pages[idx], 1);
 670         }
 671         offset[idx]++;
 672         offset[idx + 1] = 0;
 673         idx--;
 674 fail:
 675         for (i = idx; i >= 0; i--)
 676                 f2fs_put_page(pages[i], 1);
 677
 678         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 679
 680         return err;
 681 }
 682
 683 /*
 684  * All the block addresses of data and nodes should be nullified.
 685  */
 686 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 687 {
 688         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 689         int err = 0, cont = 1;
 690         int level, offset[4], noffset[4];
 691         unsigned int nofs = 0;
 692         struct f2fs_inode *ri;
 693         struct dnode_of_data dn;
 694         struct page *page;
 695
 696         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 697
 698         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 699 restart:
 700         page = get_node_page(sbi, inode->i_ino);
 701         if (IS_ERR(page)) {
 702                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 703                 return PTR_ERR(page);
 704         }
 705
 706         set_new_dnode(&dn, inode, page, NULL, 0);
 707         unlock_page(page);
 708
 709         ri = F2FS_INODE(page);
 710         switch (level) {
 711         case 0:
 712         case 1:
 713                 nofs = noffset[1];
 714                 break;
 715         case 2:
 716                 nofs = noffset[1];
 717                 if (!offset[level - 1])
 718                         goto skip_partial;
 719                 err = truncate_partial_nodes(&dn, ri, offset, level);
 720                 if (err < 0 && err != -ENOENT)
 721                         goto fail;
 722                 nofs += 1 + NIDS_PER_BLOCK;
 723                 break;
 724         case 3:
 725                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 726                 if (!offset[level - 1])
 727                         goto skip_partial;
 728                 err = truncate_partial_nodes(&dn, ri, offset, level);
 729                 if (err < 0 && err != -ENOENT)
 730                         goto fail;
 731                 break;
 732         default:
 733                 BUG();
 734         }
 735
 736 skip_partial:
 737         while (cont) {
 738                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 739                 switch (offset[0]) {
 740                 case NODE_DIR1_BLOCK:
 741                 case NODE_DIR2_BLOCK:
 742                         err = truncate_dnode(&dn);
 743                         break;
 744
 745                 case NODE_IND1_BLOCK:
 746                 case NODE_IND2_BLOCK:
 747                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 748                         break;
 749
 750                 case NODE_DIND_BLOCK:
 751                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 752                         cont = 0;
 753                         break;
 754
 755                 default:
 756                         BUG();
 757                 }
 758                 if (err < 0 && err != -ENOENT)
 759                         goto fail;
 760                 if (offset[1] == 0 &&
 761                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 762                         lock_page(page);
 763                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 764                                 f2fs_put_page(page, 1);
 765                                 goto restart;
 766                         }
 767                         f2fs_wait_on_page_writeback(page, NODE);
 768                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 769                         set_page_dirty(page);
 770                         unlock_page(page);
 771                 }
 772                 offset[1] = 0;
 773                 offset[0]++;
 774                 nofs += err;
 775         }
 776 fail:
 777         f2fs_put_page(page, 0);
 778         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 779         return err > 0 ? 0 : err;
 780 }
 781
 782 int truncate_xattr_node(struct inode *inode, struct page *page)
 783 {
 784         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 785         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 786         struct dnode_of_data dn;
 787         struct page *npage;
 788
 789         if (!nid)
 790                 return 0;
 791
 792         npage = get_node_page(sbi, nid);
 793         if (IS_ERR(npage))
 794                 return PTR_ERR(npage);
 795
 796         F2FS_I(inode)->i_xattr_nid = 0;
 797
 798         /* need to do checkpoint during fsync */
 799         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 800
 801         set_new_dnode(&dn, inode, page, npage, nid);
 802
 803         if (page)
 804                 dn.inode_page_locked = true;
 805         truncate_node(&dn);
 806         return 0;
 807 }
 808
 809 /*
 810  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 811  * f2fs_unlock_op().
 812  */
 813 void remove_inode_page(struct inode *inode)
 814 {
 815         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 816         struct page *page;
 817         nid_t ino = inode->i_ino;
 818         struct dnode_of_data dn;
 819
 820         page = get_node_page(sbi, ino);
 821         if (IS_ERR(page))
 822                 return;
 823
 824         if (truncate_xattr_node(inode, page)) {
 825                 f2fs_put_page(page, 1);
 826                 return;
 827         }
 828         /* 0 is possible, after f2fs_new_inode() is failed */
 829         f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
 830         set_new_dnode(&dn, inode, page, page, ino);
 831         truncate_node(&dn);
 832 }
 833
 834 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
 835 {
 836         struct dnode_of_data dn;
 837
 838         /* allocate inode page for new inode */
 839         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 840
 841         /* caller should f2fs_put_page(page, 1); */
 842         return new_node_page(&dn, 0, NULL);
 843 }
 844
 845 struct page *new_node_page(struct dnode_of_data *dn,
 846                                 unsigned int ofs, struct page *ipage)
 847 {
 848         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 849         struct node_info old_ni, new_ni;
 850         struct page *page;
 851         int err;
 852
 853         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 854                 return ERR_PTR(-EPERM);
 855
 856         page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
 857         if (!page)
 858                 return ERR_PTR(-ENOMEM);
 859
 860         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 861                 err = -ENOSPC;
 862                 goto fail;
 863         }
 864
 865         get_node_info(sbi, dn->nid, &old_ni);
 866
 867         /* Reinitialize old_ni with new node page */
 868         f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
 869         new_ni = old_ni;
 870         new_ni.ino = dn->inode->i_ino;
 871         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
 872
 873         f2fs_wait_on_page_writeback(page, NODE);
 874         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 875         set_cold_node(dn->inode, page);
 876         SetPageUptodate(page);
 877         set_page_dirty(page);
 878
 879         if (f2fs_has_xattr_block(ofs))
 880                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 881
 882         dn->node_page = page;
 883         if (ipage)
 884                 update_inode(dn->inode, ipage);
 885         else
 886                 sync_inode_page(dn);
 887         if (ofs == 0)
 888                 inc_valid_inode_count(sbi);
 889
 890         return page;
 891
 892 fail:
 893         clear_node_page_dirty(page);
 894         f2fs_put_page(page, 1);
 895         return ERR_PTR(err);
 896 }
 897
 898 /*
 899  * Caller should do after getting the following values.
 900  * 0: f2fs_put_page(page, 0)
 901  * LOCKED_PAGE: f2fs_put_page(page, 1)
 902  * error: nothing
 903  */
 904 static int read_node_page(struct page *page, int rw)
 905 {
 906         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 907         struct node_info ni;
 908
 909         get_node_info(sbi, page->index, &ni);
 910
 911         if (unlikely(ni.blk_addr == NULL_ADDR)) {
 912                 f2fs_put_page(page, 1);
 913                 return -ENOENT;
 914         }
 915
 916         if (PageUptodate(page))
 917                 return LOCKED_PAGE;
 918
 919         return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
 920 }
 921
 922 /*
 923  * Readahead a node page
 924  */
 925 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 926 {
 927         struct page *apage;
 928         int err;
 929
 930         apage = find_get_page(NODE_MAPPING(sbi), nid);
 931         if (apage && PageUptodate(apage)) {
 932                 f2fs_put_page(apage, 0);
 933                 return;
 934         }
 935         f2fs_put_page(apage, 0);
 936
 937         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
 938         if (!apage)
 939                 return;
 940
 941         err = read_node_page(apage, READA);
 942         if (err == 0)
 943                 f2fs_put_page(apage, 0);
 944         else if (err == LOCKED_PAGE)
 945                 f2fs_put_page(apage, 1);
 946 }
 947
 948 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 949 {
 950         struct page *page;
 951         int err;
 952 repeat:
 953         page = grab_cache_page(NODE_MAPPING(sbi), nid);
 954         if (!page)
 955                 return ERR_PTR(-ENOMEM);
 956
 957         err = read_node_page(page, READ_SYNC);
 958         if (err < 0)
 959                 return ERR_PTR(err);
 960         else if (err == LOCKED_PAGE)
 961                 goto got_it;
 962
 963         lock_page(page);
 964         if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
 965                 f2fs_put_page(page, 1);
 966                 return ERR_PTR(-EIO);
 967         }
 968         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 969                 f2fs_put_page(page, 1);
 970                 goto repeat;
 971         }
 972 got_it:
 973         mark_page_accessed(page);
 974         return page;
 975 }
 976
 977 /*
 978  * Return a locked page for the desired node page.
 979  * And, readahead MAX_RA_NODE number of node pages.
 980  */
 981 struct page *get_node_page_ra(struct page *parent, int start)
 982 {
 983         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 984         struct blk_plug plug;
 985         struct page *page;
 986         int err, i, end;
 987         nid_t nid;
 988
 989         /* First, try getting the desired direct node. */
 990         nid = get_nid(parent, start, false);
 991         if (!nid)
 992                 return ERR_PTR(-ENOENT);
 993 repeat:
 994         page = grab_cache_page(NODE_MAPPING(sbi), nid);
 995         if (!page)
 996                 return ERR_PTR(-ENOMEM);
 997
 998         err = read_node_page(page, READ_SYNC);
 999         if (err < 0)
1000                 return ERR_PTR(err);
1001         else if (err == LOCKED_PAGE)
1002                 goto page_hit;
1003
1004         blk_start_plug(&plug);
1005
1006         /* Then, try readahead for siblings of the desired node */
1007         end = start + MAX_RA_NODE;
1008         end = min(end, NIDS_PER_BLOCK);
1009         for (i = start + 1; i < end; i++) {
1010                 nid = get_nid(parent, i, false);
1011                 if (!nid)
1012                         continue;
1013                 ra_node_page(sbi, nid);
1014         }
1015
1016         blk_finish_plug(&plug);
1017
1018         lock_page(page);
1019         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1020                 f2fs_put_page(page, 1);
1021                 goto repeat;
1022         }
1023 page_hit:
1024         if (unlikely(!PageUptodate(page))) {
1025                 f2fs_put_page(page, 1);
1026                 return ERR_PTR(-EIO);
1027         }
1028         mark_page_accessed(page);
1029         return page;
1030 }
1031
1032 void sync_inode_page(struct dnode_of_data *dn)
1033 {
1034         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1035                 update_inode(dn->inode, dn->node_page);
1036         } else if (dn->inode_page) {
1037                 if (!dn->inode_page_locked)
1038                         lock_page(dn->inode_page);
1039                 update_inode(dn->inode, dn->inode_page);
1040                 if (!dn->inode_page_locked)
1041                         unlock_page(dn->inode_page);
1042         } else {
1043                 update_inode_page(dn->inode);
1044         }
1045 }
1046
1047 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1048                                         struct writeback_control *wbc)
1049 {
1050         pgoff_t index, end;
1051         struct pagevec pvec;
1052         int step = ino ? 2 : 0;
1053         int nwritten = 0, wrote = 0;
1054
1055         pagevec_init(&pvec, 0);
1056
1057 next_step:
1058         index = 0;
1059         end = LONG_MAX;
1060
1061         while (index <= end) {
1062                 int i, nr_pages;
1063                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1064                                 PAGECACHE_TAG_DIRTY,
1065                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1066                 if (nr_pages == 0)
1067                         break;
1068
1069                 for (i = 0; i < nr_pages; i++) {
1070                         struct page *page = pvec.pages[i];
1071
1072                         /*
1073                          * flushing sequence with step:
1074                          * 0. indirect nodes
1075                          * 1. dentry dnodes
1076                          * 2. file dnodes
1077                          */
1078                         if (step == 0 && IS_DNODE(page))
1079                                 continue;
1080                         if (step == 1 && (!IS_DNODE(page) ||
1081                                                 is_cold_node(page)))
1082                                 continue;
1083                         if (step == 2 && (!IS_DNODE(page) ||
1084                                                 !is_cold_node(page)))
1085                                 continue;
1086
1087                         /*
1088                          * If an fsync mode,
1089                          * we should not skip writing node pages.
1090                          */
1091                         if (ino && ino_of_node(page) == ino)
1092                                 lock_page(page);
1093                         else if (!trylock_page(page))
1094                                 continue;
1095
1096                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1097 continue_unlock:
1098                                 unlock_page(page);
1099                                 continue;
1100                         }
1101                         if (ino && ino_of_node(page) != ino)
1102                                 goto continue_unlock;
1103
1104                         if (!PageDirty(page)) {
1105                                 /* someone wrote it for us */
1106                                 goto continue_unlock;
1107                         }
1108
1109                         if (!clear_page_dirty_for_io(page))
1110                                 goto continue_unlock;
1111
1112                         /* called by fsync() */
1113                         if (ino && IS_DNODE(page)) {
1114                                 int mark = !is_checkpointed_node(sbi, ino);
1115                                 set_fsync_mark(page, 1);
1116                                 if (IS_INODE(page))
1117                                         set_dentry_mark(page, mark);
1118                                 nwritten++;
1119                         } else {
1120                                 set_fsync_mark(page, 0);
1121                                 set_dentry_mark(page, 0);
1122                         }
1123                         NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1124                         wrote++;
1125
1126                         if (--wbc->nr_to_write == 0)
1127                                 break;
1128                 }
1129                 pagevec_release(&pvec);
1130                 cond_resched();
1131
1132                 if (wbc->nr_to_write == 0) {
1133                         step = 2;
1134                         break;
1135                 }
1136         }
1137
1138         if (step < 2) {
1139                 step++;
1140                 goto next_step;
1141         }
1142
1143         if (wrote)
1144                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1145         return nwritten;
1146 }
1147
1148 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1149 {
1150         pgoff_t index = 0, end = LONG_MAX;
1151         struct pagevec pvec;
1152         int ret2 = 0, ret = 0;
1153
1154         pagevec_init(&pvec, 0);
1155
1156         while (index <= end) {
1157                 int i, nr_pages;
1158                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1159                                 PAGECACHE_TAG_WRITEBACK,
1160                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1161                 if (nr_pages == 0)
1162                         break;
1163
1164                 for (i = 0; i < nr_pages; i++) {
1165                         struct page *page = pvec.pages[i];
1166
1167                         /* until radix tree lookup accepts end_index */
1168                         if (unlikely(page->index > end))
1169                                 continue;
1170
1171                         if (ino && ino_of_node(page) == ino) {
1172                                 f2fs_wait_on_page_writeback(page, NODE);
1173                                 if (TestClearPageError(page))
1174                                         ret = -EIO;
1175                         }
1176                 }
1177                 pagevec_release(&pvec);
1178                 cond_resched();
1179         }
1180
1181         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1182                 ret2 = -ENOSPC;
1183         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1184                 ret2 = -EIO;
1185         if (!ret)
1186                 ret = ret2;
1187         return ret;
1188 }
1189
1190 static int f2fs_write_node_page(struct page *page,
1191                                 struct writeback_control *wbc)
1192 {
1193         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1194         nid_t nid;
1195         block_t new_addr;
1196         struct node_info ni;
1197         struct f2fs_io_info fio = {
1198                 .type = NODE,
1199                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1200         };
1201
1202         trace_f2fs_writepage(page, NODE);
1203
1204         if (unlikely(sbi->por_doing))
1205                 goto redirty_out;
1206
1207         f2fs_wait_on_page_writeback(page, NODE);
1208
1209         /* get old block addr of this node page */
1210         nid = nid_of_node(page);
1211         f2fs_bug_on(page->index != nid);
1212
1213         get_node_info(sbi, nid, &ni);
1214
1215         /* This page is already truncated */
1216         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1217                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1218                 unlock_page(page);
1219                 return 0;
1220         }
1221
1222         if (wbc->for_reclaim)
1223                 goto redirty_out;
1224
1225         mutex_lock(&sbi->node_write);
1226         set_page_writeback(page);
1227         write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1228         set_node_addr(sbi, &ni, new_addr, is_fsync_dnode(page));
1229         dec_page_count(sbi, F2FS_DIRTY_NODES);
1230         mutex_unlock(&sbi->node_write);
1231         unlock_page(page);
1232         return 0;
1233
1234 redirty_out:
1235         redirty_page_for_writepage(wbc, page);
1236         return AOP_WRITEPAGE_ACTIVATE;
1237 }
1238
1239 static int f2fs_write_node_pages(struct address_space *mapping,
1240                             struct writeback_control *wbc)
1241 {
1242         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1243         long diff;
1244
1245         trace_f2fs_writepages(mapping->host, wbc, NODE);
1246
1247         /* balancing f2fs's metadata in background */
1248         f2fs_balance_fs_bg(sbi);
1249
1250         /* collect a number of dirty node pages and write together */
1251         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1252                 goto skip_write;
1253
1254         diff = nr_pages_to_write(sbi, NODE, wbc);
1255         wbc->sync_mode = WB_SYNC_NONE;
1256         sync_node_pages(sbi, 0, wbc);
1257         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1258         return 0;
1259
1260 skip_write:
1261         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1262         return 0;
1263 }
1264
1265 static int f2fs_set_node_page_dirty(struct page *page)
1266 {
1267         struct address_space *mapping = page->mapping;
1268         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1269
1270         trace_f2fs_set_page_dirty(page, NODE);
1271
1272         SetPageUptodate(page);
1273         if (!PageDirty(page)) {
1274                 __set_page_dirty_nobuffers(page);
1275                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1276                 SetPagePrivate(page);
1277                 return 1;
1278         }
1279         return 0;
1280 }
1281
1282 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1283                                       unsigned int length)
1284 {
1285         struct inode *inode = page->mapping->host;
1286         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1287         if (PageDirty(page))
1288                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1289         ClearPagePrivate(page);
1290 }
1291
1292 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1293 {
1294         ClearPagePrivate(page);
1295         return 1;
1296 }
1297
1298 /*
1299  * Structure of the f2fs node operations
1300  */
1301 const struct address_space_operations f2fs_node_aops = {
1302         .writepage      = f2fs_write_node_page,
1303         .writepages     = f2fs_write_node_pages,
1304         .set_page_dirty = f2fs_set_node_page_dirty,
1305         .invalidatepage = f2fs_invalidate_node_page,
1306         .releasepage    = f2fs_release_node_page,
1307 };
1308
1309 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1310                                                 nid_t n)
1311 {
1312         return radix_tree_lookup(&nm_i->free_nid_root, n);
1313 }
1314
1315 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1316                                                 struct free_nid *i)
1317 {
1318         list_del(&i->list);
1319         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1320 }
1321
1322 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1323 {
1324         struct f2fs_nm_info *nm_i = NM_I(sbi);
1325         struct free_nid *i;
1326         struct nat_entry *ne;
1327         bool allocated = false;
1328
1329         if (!available_free_memory(sbi, FREE_NIDS))
1330                 return -1;
1331
1332         /* 0 nid should not be used */
1333         if (unlikely(nid == 0))
1334                 return 0;
1335
1336         if (build) {
1337                 /* do not add allocated nids */
1338                 read_lock(&nm_i->nat_tree_lock);
1339                 ne = __lookup_nat_cache(nm_i, nid);
1340                 if (ne &&
1341                         (!ne->checkpointed || nat_get_blkaddr(ne) != NULL_ADDR))
1342                         allocated = true;
1343                 read_unlock(&nm_i->nat_tree_lock);
1344                 if (allocated)
1345                         return 0;
1346         }
1347
1348         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1349         i->nid = nid;
1350         i->state = NID_NEW;
1351
1352         spin_lock(&nm_i->free_nid_list_lock);
1353         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1354                 spin_unlock(&nm_i->free_nid_list_lock);
1355                 kmem_cache_free(free_nid_slab, i);
1356                 return 0;
1357         }
1358         list_add_tail(&i->list, &nm_i->free_nid_list);
1359         nm_i->fcnt++;
1360         spin_unlock(&nm_i->free_nid_list_lock);
1361         return 1;
1362 }
1363
1364 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1365 {
1366         struct free_nid *i;
1367         bool need_free = false;
1368
1369         spin_lock(&nm_i->free_nid_list_lock);
1370         i = __lookup_free_nid_list(nm_i, nid);
1371         if (i && i->state == NID_NEW) {
1372                 __del_from_free_nid_list(nm_i, i);
1373                 nm_i->fcnt--;
1374                 need_free = true;
1375         }
1376         spin_unlock(&nm_i->free_nid_list_lock);
1377
1378         if (need_free)
1379                 kmem_cache_free(free_nid_slab, i);
1380 }
1381
1382 static void scan_nat_page(struct f2fs_sb_info *sbi,
1383                         struct page *nat_page, nid_t start_nid)
1384 {
1385         struct f2fs_nm_info *nm_i = NM_I(sbi);
1386         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1387         block_t blk_addr;
1388         int i;
1389
1390         i = start_nid % NAT_ENTRY_PER_BLOCK;
1391
1392         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1393
1394                 if (unlikely(start_nid >= nm_i->max_nid))
1395                         break;
1396
1397                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1398                 f2fs_bug_on(blk_addr == NEW_ADDR);
1399                 if (blk_addr == NULL_ADDR) {
1400                         if (add_free_nid(sbi, start_nid, true) < 0)
1401                                 break;
1402                 }
1403         }
1404 }
1405
1406 static void build_free_nids(struct f2fs_sb_info *sbi)
1407 {
1408         struct f2fs_nm_info *nm_i = NM_I(sbi);
1409         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1410         struct f2fs_summary_block *sum = curseg->sum_blk;
1411         int i = 0;
1412         nid_t nid = nm_i->next_scan_nid;
1413
1414         /* Enough entries */
1415         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1416                 return;
1417
1418         /* readahead nat pages to be scanned */
1419         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1420
1421         while (1) {
1422                 struct page *page = get_current_nat_page(sbi, nid);
1423
1424                 scan_nat_page(sbi, page, nid);
1425                 f2fs_put_page(page, 1);
1426
1427                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1428                 if (unlikely(nid >= nm_i->max_nid))
1429                         nid = 0;
1430
1431                 if (i++ == FREE_NID_PAGES)
1432                         break;
1433         }
1434
1435         /* go to the next free nat pages to find free nids abundantly */
1436         nm_i->next_scan_nid = nid;
1437
1438         /* find free nids from current sum_pages */
1439         mutex_lock(&curseg->curseg_mutex);
1440         for (i = 0; i < nats_in_cursum(sum); i++) {
1441                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1442                 nid = le32_to_cpu(nid_in_journal(sum, i));
1443                 if (addr == NULL_ADDR)
1444                         add_free_nid(sbi, nid, true);
1445                 else
1446                         remove_free_nid(nm_i, nid);
1447         }
1448         mutex_unlock(&curseg->curseg_mutex);
1449 }
1450
1451 /*
1452  * If this function returns success, caller can obtain a new nid
1453  * from second parameter of this function.
1454  * The returned nid could be used ino as well as nid when inode is created.
1455  */
1456 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1457 {
1458         struct f2fs_nm_info *nm_i = NM_I(sbi);
1459         struct free_nid *i = NULL;
1460 retry:
1461         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1462                 return false;
1463
1464         spin_lock(&nm_i->free_nid_list_lock);
1465
1466         /* We should not use stale free nids created by build_free_nids */
1467         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1468                 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1469                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1470                         if (i->state == NID_NEW)
1471                                 break;
1472
1473                 f2fs_bug_on(i->state != NID_NEW);
1474                 *nid = i->nid;
1475                 i->state = NID_ALLOC;
1476                 nm_i->fcnt--;
1477                 spin_unlock(&nm_i->free_nid_list_lock);
1478                 return true;
1479         }
1480         spin_unlock(&nm_i->free_nid_list_lock);
1481
1482         /* Let's scan nat pages and its caches to get free nids */
1483         mutex_lock(&nm_i->build_lock);
1484         build_free_nids(sbi);
1485         mutex_unlock(&nm_i->build_lock);
1486         goto retry;
1487 }
1488
1489 /*
1490  * alloc_nid() should be called prior to this function.
1491  */
1492 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1493 {
1494         struct f2fs_nm_info *nm_i = NM_I(sbi);
1495         struct free_nid *i;
1496
1497         spin_lock(&nm_i->free_nid_list_lock);
1498         i = __lookup_free_nid_list(nm_i, nid);
1499         f2fs_bug_on(!i || i->state != NID_ALLOC);
1500         __del_from_free_nid_list(nm_i, i);
1501         spin_unlock(&nm_i->free_nid_list_lock);
1502
1503         kmem_cache_free(free_nid_slab, i);
1504 }
1505
1506 /*
1507  * alloc_nid() should be called prior to this function.
1508  */
1509 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1510 {
1511         struct f2fs_nm_info *nm_i = NM_I(sbi);
1512         struct free_nid *i;
1513         bool need_free = false;
1514
1515         if (!nid)
1516                 return;
1517
1518         spin_lock(&nm_i->free_nid_list_lock);
1519         i = __lookup_free_nid_list(nm_i, nid);
1520         f2fs_bug_on(!i || i->state != NID_ALLOC);
1521         if (!available_free_memory(sbi, FREE_NIDS)) {
1522                 __del_from_free_nid_list(nm_i, i);
1523                 need_free = true;
1524         } else {
1525                 i->state = NID_NEW;
1526                 nm_i->fcnt++;
1527         }
1528         spin_unlock(&nm_i->free_nid_list_lock);
1529
1530         if (need_free)
1531                 kmem_cache_free(free_nid_slab, i);
1532 }
1533
1534 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1535                 struct f2fs_summary *sum, struct node_info *ni,
1536                 block_t new_blkaddr)
1537 {
1538         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1539         set_node_addr(sbi, ni, new_blkaddr, false);
1540         clear_node_page_dirty(page);
1541 }
1542
1543 static void recover_inline_xattr(struct inode *inode, struct page *page)
1544 {
1545         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1546         void *src_addr, *dst_addr;
1547         size_t inline_size;
1548         struct page *ipage;
1549         struct f2fs_inode *ri;
1550
1551         if (!f2fs_has_inline_xattr(inode))
1552                 return;
1553
1554         if (!IS_INODE(page))
1555                 return;
1556
1557         ri = F2FS_INODE(page);
1558         if (!(ri->i_inline & F2FS_INLINE_XATTR))
1559                 return;
1560
1561         ipage = get_node_page(sbi, inode->i_ino);
1562         f2fs_bug_on(IS_ERR(ipage));
1563
1564         dst_addr = inline_xattr_addr(ipage);
1565         src_addr = inline_xattr_addr(page);
1566         inline_size = inline_xattr_size(inode);
1567
1568         f2fs_wait_on_page_writeback(ipage, NODE);
1569         memcpy(dst_addr, src_addr, inline_size);
1570
1571         update_inode(inode, ipage);
1572         f2fs_put_page(ipage, 1);
1573 }
1574
1575 bool recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1576 {
1577         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1578         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1579         nid_t new_xnid = nid_of_node(page);
1580         struct node_info ni;
1581
1582         recover_inline_xattr(inode, page);
1583
1584         if (!f2fs_has_xattr_block(ofs_of_node(page)))
1585                 return false;
1586
1587         /* 1: invalidate the previous xattr nid */
1588         if (!prev_xnid)
1589                 goto recover_xnid;
1590
1591         /* Deallocate node address */
1592         get_node_info(sbi, prev_xnid, &ni);
1593         f2fs_bug_on(ni.blk_addr == NULL_ADDR);
1594         invalidate_blocks(sbi, ni.blk_addr);
1595         dec_valid_node_count(sbi, inode);
1596         set_node_addr(sbi, &ni, NULL_ADDR, false);
1597
1598 recover_xnid:
1599         /* 2: allocate new xattr nid */
1600         if (unlikely(!inc_valid_node_count(sbi, inode)))
1601                 f2fs_bug_on(1);
1602
1603         remove_free_nid(NM_I(sbi), new_xnid);
1604         get_node_info(sbi, new_xnid, &ni);
1605         ni.ino = inode->i_ino;
1606         set_node_addr(sbi, &ni, NEW_ADDR, false);
1607         F2FS_I(inode)->i_xattr_nid = new_xnid;
1608
1609         /* 3: update xattr blkaddr */
1610         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1611         set_node_addr(sbi, &ni, blkaddr, false);
1612
1613         update_inode_page(inode);
1614         return true;
1615 }
1616
1617 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1618 {
1619         struct f2fs_inode *src, *dst;
1620         nid_t ino = ino_of_node(page);
1621         struct node_info old_ni, new_ni;
1622         struct page *ipage;
1623
1624         get_node_info(sbi, ino, &old_ni);
1625
1626         if (unlikely(old_ni.blk_addr != NULL_ADDR))
1627                 return -EINVAL;
1628
1629         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1630         if (!ipage)
1631                 return -ENOMEM;
1632
1633         /* Should not use this inode  from free nid list */
1634         remove_free_nid(NM_I(sbi), ino);
1635
1636         SetPageUptodate(ipage);
1637         fill_node_footer(ipage, ino, ino, 0, true);
1638
1639         src = F2FS_INODE(page);
1640         dst = F2FS_INODE(ipage);
1641
1642         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1643         dst->i_size = 0;
1644         dst->i_blocks = cpu_to_le64(1);
1645         dst->i_links = cpu_to_le32(1);
1646         dst->i_xattr_nid = 0;
1647
1648         new_ni = old_ni;
1649         new_ni.ino = ino;
1650
1651         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1652                 WARN_ON(1);
1653         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1654         inc_valid_inode_count(sbi);
1655         f2fs_put_page(ipage, 1);
1656         return 0;
1657 }
1658
1659 /*
1660  * ra_sum_pages() merge contiguous pages into one bio and submit.
1661  * these pre-readed pages are linked in pages list.
1662  */
1663 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
1664                                 int start, int nrpages)
1665 {
1666         struct page *page;
1667         int page_idx = start;
1668         struct f2fs_io_info fio = {
1669                 .type = META,
1670                 .rw = READ_SYNC | REQ_META | REQ_PRIO
1671         };
1672
1673         for (; page_idx < start + nrpages; page_idx++) {
1674                 /* alloc temporal page for read node summary info*/
1675                 page = alloc_page(GFP_F2FS_ZERO);
1676                 if (!page)
1677                         break;
1678
1679                 lock_page(page);
1680                 page->index = page_idx;
1681                 list_add_tail(&page->lru, pages);
1682         }
1683
1684         list_for_each_entry(page, pages, lru)
1685                 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1686
1687         f2fs_submit_merged_bio(sbi, META, READ);
1688
1689         return page_idx - start;
1690 }
1691
1692 int restore_node_summary(struct f2fs_sb_info *sbi,
1693                         unsigned int segno, struct f2fs_summary_block *sum)
1694 {
1695         struct f2fs_node *rn;
1696         struct f2fs_summary *sum_entry;
1697         struct page *page, *tmp;
1698         block_t addr;
1699         int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1700         int i, last_offset, nrpages, err = 0;
1701         LIST_HEAD(page_list);
1702
1703         /* scan the node segment */
1704         last_offset = sbi->blocks_per_seg;
1705         addr = START_BLOCK(sbi, segno);
1706         sum_entry = &sum->entries[0];
1707
1708         for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) {
1709                 nrpages = min(last_offset - i, bio_blocks);
1710
1711                 /* read ahead node pages */
1712                 nrpages = ra_sum_pages(sbi, &page_list, addr, nrpages);
1713                 if (!nrpages)
1714                         return -ENOMEM;
1715
1716                 list_for_each_entry_safe(page, tmp, &page_list, lru) {
1717                         if (err)
1718                                 goto skip;
1719
1720                         lock_page(page);
1721                         if (unlikely(!PageUptodate(page))) {
1722                                 err = -EIO;
1723                         } else {
1724                                 rn = F2FS_NODE(page);
1725                                 sum_entry->nid = rn->footer.nid;
1726                                 sum_entry->version = 0;
1727                                 sum_entry->ofs_in_node = 0;
1728                                 sum_entry++;
1729                         }
1730                         unlock_page(page);
1731 skip:
1732                         list_del(&page->lru);
1733                         __free_pages(page, 0);
1734                 }
1735         }
1736         return err;
1737 }
1738
1739 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1740 {
1741         struct f2fs_nm_info *nm_i = NM_I(sbi);
1742         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1743         struct f2fs_summary_block *sum = curseg->sum_blk;
1744         int i;
1745
1746         mutex_lock(&curseg->curseg_mutex);
1747
1748         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1749                 mutex_unlock(&curseg->curseg_mutex);
1750                 return false;
1751         }
1752
1753         for (i = 0; i < nats_in_cursum(sum); i++) {
1754                 struct nat_entry *ne;
1755                 struct f2fs_nat_entry raw_ne;
1756                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1757
1758                 raw_ne = nat_in_journal(sum, i);
1759 retry:
1760                 write_lock(&nm_i->nat_tree_lock);
1761                 ne = __lookup_nat_cache(nm_i, nid);
1762                 if (ne) {
1763                         __set_nat_cache_dirty(nm_i, ne);
1764                         write_unlock(&nm_i->nat_tree_lock);
1765                         continue;
1766                 }
1767                 ne = grab_nat_entry(nm_i, nid);
1768                 if (!ne) {
1769                         write_unlock(&nm_i->nat_tree_lock);
1770                         goto retry;
1771                 }
1772                 node_info_from_raw_nat(&ne->ni, &raw_ne);
1773                 __set_nat_cache_dirty(nm_i, ne);
1774                 write_unlock(&nm_i->nat_tree_lock);
1775         }
1776         update_nats_in_cursum(sum, -i);
1777         mutex_unlock(&curseg->curseg_mutex);
1778         return true;
1779 }
1780
1781 /*
1782  * This function is called during the checkpointing process.
1783  */
1784 void flush_nat_entries(struct f2fs_sb_info *sbi)
1785 {
1786         struct f2fs_nm_info *nm_i = NM_I(sbi);
1787         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1788         struct f2fs_summary_block *sum = curseg->sum_blk;
1789         struct nat_entry *ne, *cur;
1790         struct page *page = NULL;
1791         struct f2fs_nat_block *nat_blk = NULL;
1792         nid_t start_nid = 0, end_nid = 0;
1793         bool flushed;
1794
1795         flushed = flush_nats_in_journal(sbi);
1796
1797         if (!flushed)
1798                 mutex_lock(&curseg->curseg_mutex);
1799
1800         /* 1) flush dirty nat caches */
1801         list_for_each_entry_safe(ne, cur, &nm_i->dirty_nat_entries, list) {
1802                 nid_t nid;
1803                 struct f2fs_nat_entry raw_ne;
1804                 int offset = -1;
1805
1806                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1807                         continue;
1808
1809                 nid = nat_get_nid(ne);
1810
1811                 if (flushed)
1812                         goto to_nat_page;
1813
1814                 /* if there is room for nat enries in curseg->sumpage */
1815                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1816                 if (offset >= 0) {
1817                         raw_ne = nat_in_journal(sum, offset);
1818                         goto flush_now;
1819                 }
1820 to_nat_page:
1821                 if (!page || (start_nid > nid || nid > end_nid)) {
1822                         if (page) {
1823                                 f2fs_put_page(page, 1);
1824                                 page = NULL;
1825                         }
1826                         start_nid = START_NID(nid);
1827                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1828
1829                         /*
1830                          * get nat block with dirty flag, increased reference
1831                          * count, mapped and lock
1832                          */
1833                         page = get_next_nat_page(sbi, start_nid);
1834                         nat_blk = page_address(page);
1835                 }
1836
1837                 f2fs_bug_on(!nat_blk);
1838                 raw_ne = nat_blk->entries[nid - start_nid];
1839 flush_now:
1840                 raw_nat_from_node_info(&raw_ne, &ne->ni);
1841
1842                 if (offset < 0) {
1843                         nat_blk->entries[nid - start_nid] = raw_ne;
1844                 } else {
1845                         nat_in_journal(sum, offset) = raw_ne;
1846                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1847                 }
1848
1849                 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1850                                 add_free_nid(sbi, nid, false) <= 0) {
1851                         write_lock(&nm_i->nat_tree_lock);
1852                         __del_from_nat_cache(nm_i, ne);
1853                         write_unlock(&nm_i->nat_tree_lock);
1854                 } else {
1855                         write_lock(&nm_i->nat_tree_lock);
1856                         __clear_nat_cache_dirty(nm_i, ne);
1857                         write_unlock(&nm_i->nat_tree_lock);
1858                 }
1859         }
1860         if (!flushed)
1861                 mutex_unlock(&curseg->curseg_mutex);
1862         f2fs_put_page(page, 1);
1863 }
1864
1865 static int init_node_manager(struct f2fs_sb_info *sbi)
1866 {
1867         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1868         struct f2fs_nm_info *nm_i = NM_I(sbi);
1869         unsigned char *version_bitmap;
1870         unsigned int nat_segs, nat_blocks;
1871
1872         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1873
1874         /* segment_count_nat includes pair segment so divide to 2. */
1875         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1876         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1877
1878         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1879
1880         /* not used nids: 0, node, meta, (and root counted as valid node) */
1881         nm_i->available_nids = nm_i->max_nid - 3;
1882         nm_i->fcnt = 0;
1883         nm_i->nat_cnt = 0;
1884         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1885
1886         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1887         INIT_LIST_HEAD(&nm_i->free_nid_list);
1888         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1889         INIT_LIST_HEAD(&nm_i->nat_entries);
1890         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1891
1892         mutex_init(&nm_i->build_lock);
1893         spin_lock_init(&nm_i->free_nid_list_lock);
1894         rwlock_init(&nm_i->nat_tree_lock);
1895
1896         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1897         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1898         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1899         if (!version_bitmap)
1900                 return -EFAULT;
1901
1902         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1903                                         GFP_KERNEL);
1904         if (!nm_i->nat_bitmap)
1905                 return -ENOMEM;
1906         return 0;
1907 }
1908
1909 int build_node_manager(struct f2fs_sb_info *sbi)
1910 {
1911         int err;
1912
1913         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1914         if (!sbi->nm_info)
1915                 return -ENOMEM;
1916
1917         err = init_node_manager(sbi);
1918         if (err)
1919                 return err;
1920
1921         build_free_nids(sbi);
1922         return 0;
1923 }
1924
1925 void destroy_node_manager(struct f2fs_sb_info *sbi)
1926 {
1927         struct f2fs_nm_info *nm_i = NM_I(sbi);
1928         struct free_nid *i, *next_i;
1929         struct nat_entry *natvec[NATVEC_SIZE];
1930         nid_t nid = 0;
1931         unsigned int found;
1932
1933         if (!nm_i)
1934                 return;
1935
1936         /* destroy free nid list */
1937         spin_lock(&nm_i->free_nid_list_lock);
1938         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1939                 f2fs_bug_on(i->state == NID_ALLOC);
1940                 __del_from_free_nid_list(nm_i, i);
1941                 nm_i->fcnt--;
1942                 spin_unlock(&nm_i->free_nid_list_lock);
1943                 kmem_cache_free(free_nid_slab, i);
1944                 spin_lock(&nm_i->free_nid_list_lock);
1945         }
1946         f2fs_bug_on(nm_i->fcnt);
1947         spin_unlock(&nm_i->free_nid_list_lock);
1948
1949         /* destroy nat cache */
1950         write_lock(&nm_i->nat_tree_lock);
1951         while ((found = __gang_lookup_nat_cache(nm_i,
1952                                         nid, NATVEC_SIZE, natvec))) {
1953                 unsigned idx;
1954                 nid = nat_get_nid(natvec[found - 1]) + 1;
1955                 for (idx = 0; idx < found; idx++)
1956                         __del_from_nat_cache(nm_i, natvec[idx]);
1957         }
1958         f2fs_bug_on(nm_i->nat_cnt);
1959         write_unlock(&nm_i->nat_tree_lock);
1960
1961         kfree(nm_i->nat_bitmap);
1962         sbi->nm_info = NULL;
1963         kfree(nm_i);
1964 }
1965
1966 int __init create_node_manager_caches(void)
1967 {
1968         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1969                         sizeof(struct nat_entry));
1970         if (!nat_entry_slab)
1971                 return -ENOMEM;
1972
1973         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1974                         sizeof(struct free_nid));
1975         if (!free_nid_slab) {
1976                 kmem_cache_destroy(nat_entry_slab);
1977                 return -ENOMEM;
1978         }
1979         return 0;
1980 }
1981
1982 void destroy_node_manager_caches(void)
1983 {
1984         kmem_cache_destroy(free_nid_slab);
1985         kmem_cache_destroy(nat_entry_slab);
1986 }