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                 if (sbi->sb->s_bdi->dirty_exceeded)
  46                         return false;
  47                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  48                 res = mem_size < ((val.totalram * nm_i->ram_thresh / 100) >> 1);
  49         }
  50         return res;
  51 }
  52
  53 static void clear_node_page_dirty(struct page *page)
  54 {
  55         struct address_space *mapping = page->mapping;
  56         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
  57         unsigned int long flags;
  58
  59         if (PageDirty(page)) {
  60                 spin_lock_irqsave(&mapping->tree_lock, flags);
  61                 radix_tree_tag_clear(&mapping->page_tree,
  62                                 page_index(page),
  63                                 PAGECACHE_TAG_DIRTY);
  64                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  65
  66                 clear_page_dirty_for_io(page);
  67                 dec_page_count(sbi, F2FS_DIRTY_NODES);
  68         }
  69         ClearPageUptodate(page);
  70 }
  71
  72 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  73 {
  74         pgoff_t index = current_nat_addr(sbi, nid);
  75         return get_meta_page(sbi, index);
  76 }
  77
  78 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  79 {
  80         struct page *src_page;
  81         struct page *dst_page;
  82         pgoff_t src_off;
  83         pgoff_t dst_off;
  84         void *src_addr;
  85         void *dst_addr;
  86         struct f2fs_nm_info *nm_i = NM_I(sbi);
  87
  88         src_off = current_nat_addr(sbi, nid);
  89         dst_off = next_nat_addr(sbi, src_off);
  90
  91         /* get current nat block page with lock */
  92         src_page = get_meta_page(sbi, src_off);
  93
  94         /* Dirty src_page means that it is already the new target NAT page. */
  95         if (PageDirty(src_page))
  96                 return src_page;
  97
  98         dst_page = grab_meta_page(sbi, dst_off);
  99
 100         src_addr = page_address(src_page);
 101         dst_addr = page_address(dst_page);
 102         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
 103         set_page_dirty(dst_page);
 104         f2fs_put_page(src_page, 1);
 105
 106         set_to_next_nat(nm_i, nid);
 107
 108         return dst_page;
 109 }
 110
 111 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 112 {
 113         return radix_tree_lookup(&nm_i->nat_root, n);
 114 }
 115
 116 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 117                 nid_t start, unsigned int nr, struct nat_entry **ep)
 118 {
 119         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 120 }
 121
 122 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 123 {
 124         list_del(&e->list);
 125         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 126         nm_i->nat_cnt--;
 127         kmem_cache_free(nat_entry_slab, e);
 128 }
 129
 130 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 131 {
 132         struct f2fs_nm_info *nm_i = NM_I(sbi);
 133         struct nat_entry *e;
 134         int is_cp = 1;
 135
 136         read_lock(&nm_i->nat_tree_lock);
 137         e = __lookup_nat_cache(nm_i, nid);
 138         if (e && !e->checkpointed)
 139                 is_cp = 0;
 140         read_unlock(&nm_i->nat_tree_lock);
 141         return is_cp;
 142 }
 143
 144 bool fsync_mark_done(struct f2fs_sb_info *sbi, nid_t nid)
 145 {
 146         struct f2fs_nm_info *nm_i = NM_I(sbi);
 147         struct nat_entry *e;
 148         bool fsync_done = false;
 149
 150         read_lock(&nm_i->nat_tree_lock);
 151         e = __lookup_nat_cache(nm_i, nid);
 152         if (e)
 153                 fsync_done = e->fsync_done;
 154         read_unlock(&nm_i->nat_tree_lock);
 155         return fsync_done;
 156 }
 157
 158 void fsync_mark_clear(struct f2fs_sb_info *sbi, nid_t nid)
 159 {
 160         struct f2fs_nm_info *nm_i = NM_I(sbi);
 161         struct nat_entry *e;
 162
 163         write_lock(&nm_i->nat_tree_lock);
 164         e = __lookup_nat_cache(nm_i, nid);
 165         if (e)
 166                 e->fsync_done = false;
 167         write_unlock(&nm_i->nat_tree_lock);
 168 }
 169
 170 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 171 {
 172         struct nat_entry *new;
 173
 174         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 175         if (!new)
 176                 return NULL;
 177         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 178                 kmem_cache_free(nat_entry_slab, new);
 179                 return NULL;
 180         }
 181         memset(new, 0, sizeof(struct nat_entry));
 182         nat_set_nid(new, nid);
 183         new->checkpointed = true;
 184         list_add_tail(&new->list, &nm_i->nat_entries);
 185         nm_i->nat_cnt++;
 186         return new;
 187 }
 188
 189 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 190                                                 struct f2fs_nat_entry *ne)
 191 {
 192         struct nat_entry *e;
 193 retry:
 194         write_lock(&nm_i->nat_tree_lock);
 195         e = __lookup_nat_cache(nm_i, nid);
 196         if (!e) {
 197                 e = grab_nat_entry(nm_i, nid);
 198                 if (!e) {
 199                         write_unlock(&nm_i->nat_tree_lock);
 200                         goto retry;
 201                 }
 202                 node_info_from_raw_nat(&e->ni, ne);
 203         }
 204         write_unlock(&nm_i->nat_tree_lock);
 205 }
 206
 207 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 208                         block_t new_blkaddr, bool fsync_done)
 209 {
 210         struct f2fs_nm_info *nm_i = NM_I(sbi);
 211         struct nat_entry *e;
 212 retry:
 213         write_lock(&nm_i->nat_tree_lock);
 214         e = __lookup_nat_cache(nm_i, ni->nid);
 215         if (!e) {
 216                 e = grab_nat_entry(nm_i, ni->nid);
 217                 if (!e) {
 218                         write_unlock(&nm_i->nat_tree_lock);
 219                         goto retry;
 220                 }
 221                 e->ni = *ni;
 222                 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
 223         } else if (new_blkaddr == NEW_ADDR) {
 224                 /*
 225                  * when nid is reallocated,
 226                  * previous nat entry can be remained in nat cache.
 227                  * So, reinitialize it with new information.
 228                  */
 229                 e->ni = *ni;
 230                 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
 231         }
 232
 233         /* sanity check */
 234         f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
 235         f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
 236                         new_blkaddr == NULL_ADDR);
 237         f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
 238                         new_blkaddr == NEW_ADDR);
 239         f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
 240                         nat_get_blkaddr(e) != NULL_ADDR &&
 241                         new_blkaddr == NEW_ADDR);
 242
 243         /* increament version no as node is removed */
 244         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 245                 unsigned char version = nat_get_version(e);
 246                 nat_set_version(e, inc_node_version(version));
 247         }
 248
 249         /* change address */
 250         nat_set_blkaddr(e, new_blkaddr);
 251         __set_nat_cache_dirty(nm_i, e);
 252
 253         /* update fsync_mark if its inode nat entry is still alive */
 254         e = __lookup_nat_cache(nm_i, ni->ino);
 255         if (e)
 256                 e->fsync_done = fsync_done;
 257         write_unlock(&nm_i->nat_tree_lock);
 258 }
 259
 260 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 261 {
 262         struct f2fs_nm_info *nm_i = NM_I(sbi);
 263
 264         if (available_free_memory(sbi, NAT_ENTRIES))
 265                 return 0;
 266
 267         write_lock(&nm_i->nat_tree_lock);
 268         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 269                 struct nat_entry *ne;
 270                 ne = list_first_entry(&nm_i->nat_entries,
 271                                         struct nat_entry, list);
 272                 __del_from_nat_cache(nm_i, ne);
 273                 nr_shrink--;
 274         }
 275         write_unlock(&nm_i->nat_tree_lock);
 276         return nr_shrink;
 277 }
 278
 279 /*
 280  * This function returns always success
 281  */
 282 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 283 {
 284         struct f2fs_nm_info *nm_i = NM_I(sbi);
 285         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 286         struct f2fs_summary_block *sum = curseg->sum_blk;
 287         nid_t start_nid = START_NID(nid);
 288         struct f2fs_nat_block *nat_blk;
 289         struct page *page = NULL;
 290         struct f2fs_nat_entry ne;
 291         struct nat_entry *e;
 292         int i;
 293
 294         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 295         ni->nid = nid;
 296
 297         /* Check nat cache */
 298         read_lock(&nm_i->nat_tree_lock);
 299         e = __lookup_nat_cache(nm_i, nid);
 300         if (e) {
 301                 ni->ino = nat_get_ino(e);
 302                 ni->blk_addr = nat_get_blkaddr(e);
 303                 ni->version = nat_get_version(e);
 304         }
 305         read_unlock(&nm_i->nat_tree_lock);
 306         if (e)
 307                 return;
 308
 309         /* Check current segment summary */
 310         mutex_lock(&curseg->curseg_mutex);
 311         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 312         if (i >= 0) {
 313                 ne = nat_in_journal(sum, i);
 314                 node_info_from_raw_nat(ni, &ne);
 315         }
 316         mutex_unlock(&curseg->curseg_mutex);
 317         if (i >= 0)
 318                 goto cache;
 319
 320         /* Fill node_info from nat page */
 321         page = get_current_nat_page(sbi, start_nid);
 322         nat_blk = (struct f2fs_nat_block *)page_address(page);
 323         ne = nat_blk->entries[nid - start_nid];
 324         node_info_from_raw_nat(ni, &ne);
 325         f2fs_put_page(page, 1);
 326 cache:
 327         /* cache nat entry */
 328         cache_nat_entry(NM_I(sbi), nid, &ne);
 329 }
 330
 331 /*
 332  * The maximum depth is four.
 333  * Offset[0] will have raw inode offset.
 334  */
 335 static int get_node_path(struct f2fs_inode_info *fi, long block,
 336                                 int offset[4], unsigned int noffset[4])
 337 {
 338         const long direct_index = ADDRS_PER_INODE(fi);
 339         const long direct_blks = ADDRS_PER_BLOCK;
 340         const long dptrs_per_blk = NIDS_PER_BLOCK;
 341         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 342         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 343         int n = 0;
 344         int level = 0;
 345
 346         noffset[0] = 0;
 347
 348         if (block < direct_index) {
 349                 offset[n] = block;
 350                 goto got;
 351         }
 352         block -= direct_index;
 353         if (block < direct_blks) {
 354                 offset[n++] = NODE_DIR1_BLOCK;
 355                 noffset[n] = 1;
 356                 offset[n] = block;
 357                 level = 1;
 358                 goto got;
 359         }
 360         block -= direct_blks;
 361         if (block < direct_blks) {
 362                 offset[n++] = NODE_DIR2_BLOCK;
 363                 noffset[n] = 2;
 364                 offset[n] = block;
 365                 level = 1;
 366                 goto got;
 367         }
 368         block -= direct_blks;
 369         if (block < indirect_blks) {
 370                 offset[n++] = NODE_IND1_BLOCK;
 371                 noffset[n] = 3;
 372                 offset[n++] = block / direct_blks;
 373                 noffset[n] = 4 + offset[n - 1];
 374                 offset[n] = block % direct_blks;
 375                 level = 2;
 376                 goto got;
 377         }
 378         block -= indirect_blks;
 379         if (block < indirect_blks) {
 380                 offset[n++] = NODE_IND2_BLOCK;
 381                 noffset[n] = 4 + dptrs_per_blk;
 382                 offset[n++] = block / direct_blks;
 383                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 384                 offset[n] = block % direct_blks;
 385                 level = 2;
 386                 goto got;
 387         }
 388         block -= indirect_blks;
 389         if (block < dindirect_blks) {
 390                 offset[n++] = NODE_DIND_BLOCK;
 391                 noffset[n] = 5 + (dptrs_per_blk * 2);
 392                 offset[n++] = block / indirect_blks;
 393                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 394                               offset[n - 1] * (dptrs_per_blk + 1);
 395                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 396                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 397                               offset[n - 2] * (dptrs_per_blk + 1) +
 398                               offset[n - 1];
 399                 offset[n] = block % direct_blks;
 400                 level = 3;
 401                 goto got;
 402         } else {
 403                 BUG();
 404         }
 405 got:
 406         return level;
 407 }
 408
 409 /*
 410  * Caller should call f2fs_put_dnode(dn).
 411  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 412  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 413  * In the case of RDONLY_NODE, we don't need to care about mutex.
 414  */
 415 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 416 {
 417         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 418         struct page *npage[4];
 419         struct page *parent;
 420         int offset[4];
 421         unsigned int noffset[4];
 422         nid_t nids[4];
 423         int level, i;
 424         int err = 0;
 425
 426         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 427
 428         nids[0] = dn->inode->i_ino;
 429         npage[0] = dn->inode_page;
 430
 431         if (!npage[0]) {
 432                 npage[0] = get_node_page(sbi, nids[0]);
 433                 if (IS_ERR(npage[0]))
 434                         return PTR_ERR(npage[0]);
 435         }
 436         parent = npage[0];
 437         if (level != 0)
 438                 nids[1] = get_nid(parent, offset[0], true);
 439         dn->inode_page = npage[0];
 440         dn->inode_page_locked = true;
 441
 442         /* get indirect or direct nodes */
 443         for (i = 1; i <= level; i++) {
 444                 bool done = false;
 445
 446                 if (!nids[i] && mode == ALLOC_NODE) {
 447                         /* alloc new node */
 448                         if (!alloc_nid(sbi, &(nids[i]))) {
 449                                 err = -ENOSPC;
 450                                 goto release_pages;
 451                         }
 452
 453                         dn->nid = nids[i];
 454                         npage[i] = new_node_page(dn, noffset[i], NULL);
 455                         if (IS_ERR(npage[i])) {
 456                                 alloc_nid_failed(sbi, nids[i]);
 457                                 err = PTR_ERR(npage[i]);
 458                                 goto release_pages;
 459                         }
 460
 461                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 462                         alloc_nid_done(sbi, nids[i]);
 463                         done = true;
 464                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 465                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 466                         if (IS_ERR(npage[i])) {
 467                                 err = PTR_ERR(npage[i]);
 468                                 goto release_pages;
 469                         }
 470                         done = true;
 471                 }
 472                 if (i == 1) {
 473                         dn->inode_page_locked = false;
 474                         unlock_page(parent);
 475                 } else {
 476                         f2fs_put_page(parent, 1);
 477                 }
 478
 479                 if (!done) {
 480                         npage[i] = get_node_page(sbi, nids[i]);
 481                         if (IS_ERR(npage[i])) {
 482                                 err = PTR_ERR(npage[i]);
 483                                 f2fs_put_page(npage[0], 0);
 484                                 goto release_out;
 485                         }
 486                 }
 487                 if (i < level) {
 488                         parent = npage[i];
 489                         nids[i + 1] = get_nid(parent, offset[i], false);
 490                 }
 491         }
 492         dn->nid = nids[level];
 493         dn->ofs_in_node = offset[level];
 494         dn->node_page = npage[level];
 495         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 496         return 0;
 497
 498 release_pages:
 499         f2fs_put_page(parent, 1);
 500         if (i > 1)
 501                 f2fs_put_page(npage[0], 0);
 502 release_out:
 503         dn->inode_page = NULL;
 504         dn->node_page = NULL;
 505         return err;
 506 }
 507
 508 static void truncate_node(struct dnode_of_data *dn)
 509 {
 510         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 511         struct node_info ni;
 512
 513         get_node_info(sbi, dn->nid, &ni);
 514         if (dn->inode->i_blocks == 0) {
 515                 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
 516                 goto invalidate;
 517         }
 518         f2fs_bug_on(ni.blk_addr == NULL_ADDR);
 519
 520         /* Deallocate node address */
 521         invalidate_blocks(sbi, ni.blk_addr);
 522         dec_valid_node_count(sbi, dn->inode);
 523         set_node_addr(sbi, &ni, NULL_ADDR, false);
 524
 525         if (dn->nid == dn->inode->i_ino) {
 526                 remove_orphan_inode(sbi, dn->nid);
 527                 dec_valid_inode_count(sbi);
 528         } else {
 529                 sync_inode_page(dn);
 530         }
 531 invalidate:
 532         clear_node_page_dirty(dn->node_page);
 533         F2FS_SET_SB_DIRT(sbi);
 534
 535         f2fs_put_page(dn->node_page, 1);
 536
 537         invalidate_mapping_pages(NODE_MAPPING(sbi),
 538                         dn->node_page->index, dn->node_page->index);
 539
 540         dn->node_page = NULL;
 541         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 542 }
 543
 544 static int truncate_dnode(struct dnode_of_data *dn)
 545 {
 546         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 547         struct page *page;
 548
 549         if (dn->nid == 0)
 550                 return 1;
 551
 552         /* get direct node */
 553         page = get_node_page(sbi, dn->nid);
 554         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 555                 return 1;
 556         else if (IS_ERR(page))
 557                 return PTR_ERR(page);
 558
 559         /* Make dnode_of_data for parameter */
 560         dn->node_page = page;
 561         dn->ofs_in_node = 0;
 562         truncate_data_blocks(dn);
 563         truncate_node(dn);
 564         return 1;
 565 }
 566
 567 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 568                                                 int ofs, int depth)
 569 {
 570         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 571         struct dnode_of_data rdn = *dn;
 572         struct page *page;
 573         struct f2fs_node *rn;
 574         nid_t child_nid;
 575         unsigned int child_nofs;
 576         int freed = 0;
 577         int i, ret;
 578
 579         if (dn->nid == 0)
 580                 return NIDS_PER_BLOCK + 1;
 581
 582         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 583
 584         page = get_node_page(sbi, dn->nid);
 585         if (IS_ERR(page)) {
 586                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 587                 return PTR_ERR(page);
 588         }
 589
 590         rn = F2FS_NODE(page);
 591         if (depth < 3) {
 592                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 593                         child_nid = le32_to_cpu(rn->in.nid[i]);
 594                         if (child_nid == 0)
 595                                 continue;
 596                         rdn.nid = child_nid;
 597                         ret = truncate_dnode(&rdn);
 598                         if (ret < 0)
 599                                 goto out_err;
 600                         set_nid(page, i, 0, false);
 601                 }
 602         } else {
 603                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 604                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 605                         child_nid = le32_to_cpu(rn->in.nid[i]);
 606                         if (child_nid == 0) {
 607                                 child_nofs += NIDS_PER_BLOCK + 1;
 608                                 continue;
 609                         }
 610                         rdn.nid = child_nid;
 611                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 612                         if (ret == (NIDS_PER_BLOCK + 1)) {
 613                                 set_nid(page, i, 0, false);
 614                                 child_nofs += ret;
 615                         } else if (ret < 0 && ret != -ENOENT) {
 616                                 goto out_err;
 617                         }
 618                 }
 619                 freed = child_nofs;
 620         }
 621
 622         if (!ofs) {
 623                 /* remove current indirect node */
 624                 dn->node_page = page;
 625                 truncate_node(dn);
 626                 freed++;
 627         } else {
 628                 f2fs_put_page(page, 1);
 629         }
 630         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 631         return freed;
 632
 633 out_err:
 634         f2fs_put_page(page, 1);
 635         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 636         return ret;
 637 }
 638
 639 static int truncate_partial_nodes(struct dnode_of_data *dn,
 640                         struct f2fs_inode *ri, int *offset, int depth)
 641 {
 642         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 643         struct page *pages[2];
 644         nid_t nid[3];
 645         nid_t child_nid;
 646         int err = 0;
 647         int i;
 648         int idx = depth - 2;
 649
 650         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 651         if (!nid[0])
 652                 return 0;
 653
 654         /* get indirect nodes in the path */
 655         for (i = 0; i < idx + 1; i++) {
 656                 /* refernece count'll be increased */
 657                 pages[i] = get_node_page(sbi, nid[i]);
 658                 if (IS_ERR(pages[i])) {
 659                         err = PTR_ERR(pages[i]);
 660                         idx = i - 1;
 661                         goto fail;
 662                 }
 663                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 664         }
 665
 666         /* free direct nodes linked to a partial indirect node */
 667         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 668                 child_nid = get_nid(pages[idx], i, false);
 669                 if (!child_nid)
 670                         continue;
 671                 dn->nid = child_nid;
 672                 err = truncate_dnode(dn);
 673                 if (err < 0)
 674                         goto fail;
 675                 set_nid(pages[idx], i, 0, false);
 676         }
 677
 678         if (offset[idx + 1] == 0) {
 679                 dn->node_page = pages[idx];
 680                 dn->nid = nid[idx];
 681                 truncate_node(dn);
 682         } else {
 683                 f2fs_put_page(pages[idx], 1);
 684         }
 685         offset[idx]++;
 686         offset[idx + 1] = 0;
 687         idx--;
 688 fail:
 689         for (i = idx; i >= 0; i--)
 690                 f2fs_put_page(pages[i], 1);
 691
 692         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 693
 694         return err;
 695 }
 696
 697 /*
 698  * All the block addresses of data and nodes should be nullified.
 699  */
 700 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 701 {
 702         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 703         int err = 0, cont = 1;
 704         int level, offset[4], noffset[4];
 705         unsigned int nofs = 0;
 706         struct f2fs_inode *ri;
 707         struct dnode_of_data dn;
 708         struct page *page;
 709
 710         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 711
 712         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 713 restart:
 714         page = get_node_page(sbi, inode->i_ino);
 715         if (IS_ERR(page)) {
 716                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 717                 return PTR_ERR(page);
 718         }
 719
 720         set_new_dnode(&dn, inode, page, NULL, 0);
 721         unlock_page(page);
 722
 723         ri = F2FS_INODE(page);
 724         switch (level) {
 725         case 0:
 726         case 1:
 727                 nofs = noffset[1];
 728                 break;
 729         case 2:
 730                 nofs = noffset[1];
 731                 if (!offset[level - 1])
 732                         goto skip_partial;
 733                 err = truncate_partial_nodes(&dn, ri, offset, level);
 734                 if (err < 0 && err != -ENOENT)
 735                         goto fail;
 736                 nofs += 1 + NIDS_PER_BLOCK;
 737                 break;
 738         case 3:
 739                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 740                 if (!offset[level - 1])
 741                         goto skip_partial;
 742                 err = truncate_partial_nodes(&dn, ri, offset, level);
 743                 if (err < 0 && err != -ENOENT)
 744                         goto fail;
 745                 break;
 746         default:
 747                 BUG();
 748         }
 749
 750 skip_partial:
 751         while (cont) {
 752                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 753                 switch (offset[0]) {
 754                 case NODE_DIR1_BLOCK:
 755                 case NODE_DIR2_BLOCK:
 756                         err = truncate_dnode(&dn);
 757                         break;
 758
 759                 case NODE_IND1_BLOCK:
 760                 case NODE_IND2_BLOCK:
 761                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 762                         break;
 763
 764                 case NODE_DIND_BLOCK:
 765                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 766                         cont = 0;
 767                         break;
 768
 769                 default:
 770                         BUG();
 771                 }
 772                 if (err < 0 && err != -ENOENT)
 773                         goto fail;
 774                 if (offset[1] == 0 &&
 775                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 776                         lock_page(page);
 777                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 778                                 f2fs_put_page(page, 1);
 779                                 goto restart;
 780                         }
 781                         f2fs_wait_on_page_writeback(page, NODE);
 782                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 783                         set_page_dirty(page);
 784                         unlock_page(page);
 785                 }
 786                 offset[1] = 0;
 787                 offset[0]++;
 788                 nofs += err;
 789         }
 790 fail:
 791         f2fs_put_page(page, 0);
 792         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 793         return err > 0 ? 0 : err;
 794 }
 795
 796 int truncate_xattr_node(struct inode *inode, struct page *page)
 797 {
 798         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 799         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 800         struct dnode_of_data dn;
 801         struct page *npage;
 802
 803         if (!nid)
 804                 return 0;
 805
 806         npage = get_node_page(sbi, nid);
 807         if (IS_ERR(npage))
 808                 return PTR_ERR(npage);
 809
 810         F2FS_I(inode)->i_xattr_nid = 0;
 811
 812         /* need to do checkpoint during fsync */
 813         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 814
 815         set_new_dnode(&dn, inode, page, npage, nid);
 816
 817         if (page)
 818                 dn.inode_page_locked = true;
 819         truncate_node(&dn);
 820         return 0;
 821 }
 822
 823 /*
 824  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 825  * f2fs_unlock_op().
 826  */
 827 void remove_inode_page(struct inode *inode)
 828 {
 829         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 830         struct page *page;
 831         nid_t ino = inode->i_ino;
 832         struct dnode_of_data dn;
 833
 834         page = get_node_page(sbi, ino);
 835         if (IS_ERR(page))
 836                 return;
 837
 838         if (truncate_xattr_node(inode, page)) {
 839                 f2fs_put_page(page, 1);
 840                 return;
 841         }
 842         /* 0 is possible, after f2fs_new_inode() is failed */
 843         f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
 844         set_new_dnode(&dn, inode, page, page, ino);
 845         truncate_node(&dn);
 846 }
 847
 848 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
 849 {
 850         struct dnode_of_data dn;
 851
 852         /* allocate inode page for new inode */
 853         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 854
 855         /* caller should f2fs_put_page(page, 1); */
 856         return new_node_page(&dn, 0, NULL);
 857 }
 858
 859 struct page *new_node_page(struct dnode_of_data *dn,
 860                                 unsigned int ofs, struct page *ipage)
 861 {
 862         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 863         struct node_info old_ni, new_ni;
 864         struct page *page;
 865         int err;
 866
 867         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 868                 return ERR_PTR(-EPERM);
 869
 870         page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
 871         if (!page)
 872                 return ERR_PTR(-ENOMEM);
 873
 874         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 875                 err = -ENOSPC;
 876                 goto fail;
 877         }
 878
 879         get_node_info(sbi, dn->nid, &old_ni);
 880
 881         /* Reinitialize old_ni with new node page */
 882         f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
 883         new_ni = old_ni;
 884         new_ni.ino = dn->inode->i_ino;
 885         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
 886
 887         f2fs_wait_on_page_writeback(page, NODE);
 888         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 889         set_cold_node(dn->inode, page);
 890         SetPageUptodate(page);
 891         set_page_dirty(page);
 892
 893         if (f2fs_has_xattr_block(ofs))
 894                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 895
 896         dn->node_page = page;
 897         if (ipage)
 898                 update_inode(dn->inode, ipage);
 899         else
 900                 sync_inode_page(dn);
 901         if (ofs == 0)
 902                 inc_valid_inode_count(sbi);
 903
 904         return page;
 905
 906 fail:
 907         clear_node_page_dirty(page);
 908         f2fs_put_page(page, 1);
 909         return ERR_PTR(err);
 910 }
 911
 912 /*
 913  * Caller should do after getting the following values.
 914  * 0: f2fs_put_page(page, 0)
 915  * LOCKED_PAGE: f2fs_put_page(page, 1)
 916  * error: nothing
 917  */
 918 static int read_node_page(struct page *page, int rw)
 919 {
 920         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 921         struct node_info ni;
 922
 923         get_node_info(sbi, page->index, &ni);
 924
 925         if (unlikely(ni.blk_addr == NULL_ADDR)) {
 926                 f2fs_put_page(page, 1);
 927                 return -ENOENT;
 928         }
 929
 930         if (PageUptodate(page))
 931                 return LOCKED_PAGE;
 932
 933         return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
 934 }
 935
 936 /*
 937  * Readahead a node page
 938  */
 939 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 940 {
 941         struct page *apage;
 942         int err;
 943
 944         apage = find_get_page(NODE_MAPPING(sbi), nid);
 945         if (apage && PageUptodate(apage)) {
 946                 f2fs_put_page(apage, 0);
 947                 return;
 948         }
 949         f2fs_put_page(apage, 0);
 950
 951         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
 952         if (!apage)
 953                 return;
 954
 955         err = read_node_page(apage, READA);
 956         if (err == 0)
 957                 f2fs_put_page(apage, 0);
 958         else if (err == LOCKED_PAGE)
 959                 f2fs_put_page(apage, 1);
 960 }
 961
 962 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 963 {
 964         struct page *page;
 965         int err;
 966 repeat:
 967         page = grab_cache_page(NODE_MAPPING(sbi), nid);
 968         if (!page)
 969                 return ERR_PTR(-ENOMEM);
 970
 971         err = read_node_page(page, READ_SYNC);
 972         if (err < 0)
 973                 return ERR_PTR(err);
 974         else if (err == LOCKED_PAGE)
 975                 goto got_it;
 976
 977         lock_page(page);
 978         if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
 979                 f2fs_put_page(page, 1);
 980                 return ERR_PTR(-EIO);
 981         }
 982         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 983                 f2fs_put_page(page, 1);
 984                 goto repeat;
 985         }
 986 got_it:
 987         return page;
 988 }
 989
 990 /*
 991  * Return a locked page for the desired node page.
 992  * And, readahead MAX_RA_NODE number of node pages.
 993  */
 994 struct page *get_node_page_ra(struct page *parent, int start)
 995 {
 996         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 997         struct blk_plug plug;
 998         struct page *page;
 999         int err, i, end;
1000         nid_t nid;
1001
1002         /* First, try getting the desired direct node. */
1003         nid = get_nid(parent, start, false);
1004         if (!nid)
1005                 return ERR_PTR(-ENOENT);
1006 repeat:
1007         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1008         if (!page)
1009                 return ERR_PTR(-ENOMEM);
1010
1011         err = read_node_page(page, READ_SYNC);
1012         if (err < 0)
1013                 return ERR_PTR(err);
1014         else if (err == LOCKED_PAGE)
1015                 goto page_hit;
1016
1017         blk_start_plug(&plug);
1018
1019         /* Then, try readahead for siblings of the desired node */
1020         end = start + MAX_RA_NODE;
1021         end = min(end, NIDS_PER_BLOCK);
1022         for (i = start + 1; i < end; i++) {
1023                 nid = get_nid(parent, i, false);
1024                 if (!nid)
1025                         continue;
1026                 ra_node_page(sbi, nid);
1027         }
1028
1029         blk_finish_plug(&plug);
1030
1031         lock_page(page);
1032         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1033                 f2fs_put_page(page, 1);
1034                 goto repeat;
1035         }
1036 page_hit:
1037         if (unlikely(!PageUptodate(page))) {
1038                 f2fs_put_page(page, 1);
1039                 return ERR_PTR(-EIO);
1040         }
1041         return page;
1042 }
1043
1044 void sync_inode_page(struct dnode_of_data *dn)
1045 {
1046         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1047                 update_inode(dn->inode, dn->node_page);
1048         } else if (dn->inode_page) {
1049                 if (!dn->inode_page_locked)
1050                         lock_page(dn->inode_page);
1051                 update_inode(dn->inode, dn->inode_page);
1052                 if (!dn->inode_page_locked)
1053                         unlock_page(dn->inode_page);
1054         } else {
1055                 update_inode_page(dn->inode);
1056         }
1057 }
1058
1059 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1060                                         struct writeback_control *wbc)
1061 {
1062         pgoff_t index, end;
1063         struct pagevec pvec;
1064         int step = ino ? 2 : 0;
1065         int nwritten = 0, wrote = 0;
1066
1067         pagevec_init(&pvec, 0);
1068
1069 next_step:
1070         index = 0;
1071         end = LONG_MAX;
1072
1073         while (index <= end) {
1074                 int i, nr_pages;
1075                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1076                                 PAGECACHE_TAG_DIRTY,
1077                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1078                 if (nr_pages == 0)
1079                         break;
1080
1081                 for (i = 0; i < nr_pages; i++) {
1082                         struct page *page = pvec.pages[i];
1083
1084                         /*
1085                          * flushing sequence with step:
1086                          * 0. indirect nodes
1087                          * 1. dentry dnodes
1088                          * 2. file dnodes
1089                          */
1090                         if (step == 0 && IS_DNODE(page))
1091                                 continue;
1092                         if (step == 1 && (!IS_DNODE(page) ||
1093                                                 is_cold_node(page)))
1094                                 continue;
1095                         if (step == 2 && (!IS_DNODE(page) ||
1096                                                 !is_cold_node(page)))
1097                                 continue;
1098
1099                         /*
1100                          * If an fsync mode,
1101                          * we should not skip writing node pages.
1102                          */
1103                         if (ino && ino_of_node(page) == ino)
1104                                 lock_page(page);
1105                         else if (!trylock_page(page))
1106                                 continue;
1107
1108                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1109 continue_unlock:
1110                                 unlock_page(page);
1111                                 continue;
1112                         }
1113                         if (ino && ino_of_node(page) != ino)
1114                                 goto continue_unlock;
1115
1116                         if (!PageDirty(page)) {
1117                                 /* someone wrote it for us */
1118                                 goto continue_unlock;
1119                         }
1120
1121                         if (!clear_page_dirty_for_io(page))
1122                                 goto continue_unlock;
1123
1124                         /* called by fsync() */
1125                         if (ino && IS_DNODE(page)) {
1126                                 int mark = !is_checkpointed_node(sbi, ino);
1127                                 set_fsync_mark(page, 1);
1128                                 if (IS_INODE(page))
1129                                         set_dentry_mark(page, mark);
1130                                 nwritten++;
1131                         } else {
1132                                 set_fsync_mark(page, 0);
1133                                 set_dentry_mark(page, 0);
1134                         }
1135                         NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1136                         wrote++;
1137
1138                         if (--wbc->nr_to_write == 0)
1139                                 break;
1140                 }
1141                 pagevec_release(&pvec);
1142                 cond_resched();
1143
1144                 if (wbc->nr_to_write == 0) {
1145                         step = 2;
1146                         break;
1147                 }
1148         }
1149
1150         if (step < 2) {
1151                 step++;
1152                 goto next_step;
1153         }
1154
1155         if (wrote)
1156                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1157         return nwritten;
1158 }
1159
1160 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1161 {
1162         pgoff_t index = 0, end = LONG_MAX;
1163         struct pagevec pvec;
1164         int ret2 = 0, ret = 0;
1165
1166         pagevec_init(&pvec, 0);
1167
1168         while (index <= end) {
1169                 int i, nr_pages;
1170                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1171                                 PAGECACHE_TAG_WRITEBACK,
1172                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1173                 if (nr_pages == 0)
1174                         break;
1175
1176                 for (i = 0; i < nr_pages; i++) {
1177                         struct page *page = pvec.pages[i];
1178
1179                         /* until radix tree lookup accepts end_index */
1180                         if (unlikely(page->index > end))
1181                                 continue;
1182
1183                         if (ino && ino_of_node(page) == ino) {
1184                                 f2fs_wait_on_page_writeback(page, NODE);
1185                                 if (TestClearPageError(page))
1186                                         ret = -EIO;
1187                         }
1188                 }
1189                 pagevec_release(&pvec);
1190                 cond_resched();
1191         }
1192
1193         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1194                 ret2 = -ENOSPC;
1195         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1196                 ret2 = -EIO;
1197         if (!ret)
1198                 ret = ret2;
1199         return ret;
1200 }
1201
1202 static int f2fs_write_node_page(struct page *page,
1203                                 struct writeback_control *wbc)
1204 {
1205         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1206         nid_t nid;
1207         block_t new_addr;
1208         struct node_info ni;
1209         struct f2fs_io_info fio = {
1210                 .type = NODE,
1211                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1212         };
1213
1214         trace_f2fs_writepage(page, NODE);
1215
1216         if (unlikely(sbi->por_doing))
1217                 goto redirty_out;
1218
1219         f2fs_wait_on_page_writeback(page, NODE);
1220
1221         /* get old block addr of this node page */
1222         nid = nid_of_node(page);
1223         f2fs_bug_on(page->index != nid);
1224
1225         get_node_info(sbi, nid, &ni);
1226
1227         /* This page is already truncated */
1228         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1229                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1230                 unlock_page(page);
1231                 return 0;
1232         }
1233
1234         if (wbc->for_reclaim)
1235                 goto redirty_out;
1236
1237         mutex_lock(&sbi->node_write);
1238         set_page_writeback(page);
1239         write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1240         set_node_addr(sbi, &ni, new_addr, is_fsync_dnode(page));
1241         dec_page_count(sbi, F2FS_DIRTY_NODES);
1242         mutex_unlock(&sbi->node_write);
1243         unlock_page(page);
1244         return 0;
1245
1246 redirty_out:
1247         redirty_page_for_writepage(wbc, page);
1248         return AOP_WRITEPAGE_ACTIVATE;
1249 }
1250
1251 static int f2fs_write_node_pages(struct address_space *mapping,
1252                             struct writeback_control *wbc)
1253 {
1254         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1255         long diff;
1256
1257         trace_f2fs_writepages(mapping->host, wbc, NODE);
1258
1259         /* balancing f2fs's metadata in background */
1260         f2fs_balance_fs_bg(sbi);
1261
1262         /* collect a number of dirty node pages and write together */
1263         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1264                 goto skip_write;
1265
1266         diff = nr_pages_to_write(sbi, NODE, wbc);
1267         wbc->sync_mode = WB_SYNC_NONE;
1268         sync_node_pages(sbi, 0, wbc);
1269         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1270         return 0;
1271
1272 skip_write:
1273         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1274         return 0;
1275 }
1276
1277 static int f2fs_set_node_page_dirty(struct page *page)
1278 {
1279         struct address_space *mapping = page->mapping;
1280         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1281
1282         trace_f2fs_set_page_dirty(page, NODE);
1283
1284         SetPageUptodate(page);
1285         if (!PageDirty(page)) {
1286                 __set_page_dirty_nobuffers(page);
1287                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1288                 SetPagePrivate(page);
1289                 return 1;
1290         }
1291         return 0;
1292 }
1293
1294 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1295                                       unsigned int length)
1296 {
1297         struct inode *inode = page->mapping->host;
1298         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1299         if (PageDirty(page))
1300                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1301         ClearPagePrivate(page);
1302 }
1303
1304 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1305 {
1306         ClearPagePrivate(page);
1307         return 1;
1308 }
1309
1310 /*
1311  * Structure of the f2fs node operations
1312  */
1313 const struct address_space_operations f2fs_node_aops = {
1314         .writepage      = f2fs_write_node_page,
1315         .writepages     = f2fs_write_node_pages,
1316         .set_page_dirty = f2fs_set_node_page_dirty,
1317         .invalidatepage = f2fs_invalidate_node_page,
1318         .releasepage    = f2fs_release_node_page,
1319 };
1320
1321 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1322                                                 nid_t n)
1323 {
1324         return radix_tree_lookup(&nm_i->free_nid_root, n);
1325 }
1326
1327 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1328                                                 struct free_nid *i)
1329 {
1330         list_del(&i->list);
1331         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1332 }
1333
1334 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1335 {
1336         struct f2fs_nm_info *nm_i = NM_I(sbi);
1337         struct free_nid *i;
1338         struct nat_entry *ne;
1339         bool allocated = false;
1340
1341         if (!available_free_memory(sbi, FREE_NIDS))
1342                 return -1;
1343
1344         /* 0 nid should not be used */
1345         if (unlikely(nid == 0))
1346                 return 0;
1347
1348         if (build) {
1349                 /* do not add allocated nids */
1350                 read_lock(&nm_i->nat_tree_lock);
1351                 ne = __lookup_nat_cache(nm_i, nid);
1352                 if (ne &&
1353                         (!ne->checkpointed || nat_get_blkaddr(ne) != NULL_ADDR))
1354                         allocated = true;
1355                 read_unlock(&nm_i->nat_tree_lock);
1356                 if (allocated)
1357                         return 0;
1358         }
1359
1360         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1361         i->nid = nid;
1362         i->state = NID_NEW;
1363
1364         spin_lock(&nm_i->free_nid_list_lock);
1365         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1366                 spin_unlock(&nm_i->free_nid_list_lock);
1367                 kmem_cache_free(free_nid_slab, i);
1368                 return 0;
1369         }
1370         list_add_tail(&i->list, &nm_i->free_nid_list);
1371         nm_i->fcnt++;
1372         spin_unlock(&nm_i->free_nid_list_lock);
1373         return 1;
1374 }
1375
1376 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1377 {
1378         struct free_nid *i;
1379         bool need_free = false;
1380
1381         spin_lock(&nm_i->free_nid_list_lock);
1382         i = __lookup_free_nid_list(nm_i, nid);
1383         if (i && i->state == NID_NEW) {
1384                 __del_from_free_nid_list(nm_i, i);
1385                 nm_i->fcnt--;
1386                 need_free = true;
1387         }
1388         spin_unlock(&nm_i->free_nid_list_lock);
1389
1390         if (need_free)
1391                 kmem_cache_free(free_nid_slab, i);
1392 }
1393
1394 static void scan_nat_page(struct f2fs_sb_info *sbi,
1395                         struct page *nat_page, nid_t start_nid)
1396 {
1397         struct f2fs_nm_info *nm_i = NM_I(sbi);
1398         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1399         block_t blk_addr;
1400         int i;
1401
1402         i = start_nid % NAT_ENTRY_PER_BLOCK;
1403
1404         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1405
1406                 if (unlikely(start_nid >= nm_i->max_nid))
1407                         break;
1408
1409                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1410                 f2fs_bug_on(blk_addr == NEW_ADDR);
1411                 if (blk_addr == NULL_ADDR) {
1412                         if (add_free_nid(sbi, start_nid, true) < 0)
1413                                 break;
1414                 }
1415         }
1416 }
1417
1418 static void build_free_nids(struct f2fs_sb_info *sbi)
1419 {
1420         struct f2fs_nm_info *nm_i = NM_I(sbi);
1421         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1422         struct f2fs_summary_block *sum = curseg->sum_blk;
1423         int i = 0;
1424         nid_t nid = nm_i->next_scan_nid;
1425
1426         /* Enough entries */
1427         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1428                 return;
1429
1430         /* readahead nat pages to be scanned */
1431         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1432
1433         while (1) {
1434                 struct page *page = get_current_nat_page(sbi, nid);
1435
1436                 scan_nat_page(sbi, page, nid);
1437                 f2fs_put_page(page, 1);
1438
1439                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1440                 if (unlikely(nid >= nm_i->max_nid))
1441                         nid = 0;
1442
1443                 if (i++ == FREE_NID_PAGES)
1444                         break;
1445         }
1446
1447         /* go to the next free nat pages to find free nids abundantly */
1448         nm_i->next_scan_nid = nid;
1449
1450         /* find free nids from current sum_pages */
1451         mutex_lock(&curseg->curseg_mutex);
1452         for (i = 0; i < nats_in_cursum(sum); i++) {
1453                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1454                 nid = le32_to_cpu(nid_in_journal(sum, i));
1455                 if (addr == NULL_ADDR)
1456                         add_free_nid(sbi, nid, true);
1457                 else
1458                         remove_free_nid(nm_i, nid);
1459         }
1460         mutex_unlock(&curseg->curseg_mutex);
1461 }
1462
1463 /*
1464  * If this function returns success, caller can obtain a new nid
1465  * from second parameter of this function.
1466  * The returned nid could be used ino as well as nid when inode is created.
1467  */
1468 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1469 {
1470         struct f2fs_nm_info *nm_i = NM_I(sbi);
1471         struct free_nid *i = NULL;
1472 retry:
1473         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1474                 return false;
1475
1476         spin_lock(&nm_i->free_nid_list_lock);
1477
1478         /* We should not use stale free nids created by build_free_nids */
1479         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1480                 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1481                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1482                         if (i->state == NID_NEW)
1483                                 break;
1484
1485                 f2fs_bug_on(i->state != NID_NEW);
1486                 *nid = i->nid;
1487                 i->state = NID_ALLOC;
1488                 nm_i->fcnt--;
1489                 spin_unlock(&nm_i->free_nid_list_lock);
1490                 return true;
1491         }
1492         spin_unlock(&nm_i->free_nid_list_lock);
1493
1494         /* Let's scan nat pages and its caches to get free nids */
1495         mutex_lock(&nm_i->build_lock);
1496         build_free_nids(sbi);
1497         mutex_unlock(&nm_i->build_lock);
1498         goto retry;
1499 }
1500
1501 /*
1502  * alloc_nid() should be called prior to this function.
1503  */
1504 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1505 {
1506         struct f2fs_nm_info *nm_i = NM_I(sbi);
1507         struct free_nid *i;
1508
1509         spin_lock(&nm_i->free_nid_list_lock);
1510         i = __lookup_free_nid_list(nm_i, nid);
1511         f2fs_bug_on(!i || i->state != NID_ALLOC);
1512         __del_from_free_nid_list(nm_i, i);
1513         spin_unlock(&nm_i->free_nid_list_lock);
1514
1515         kmem_cache_free(free_nid_slab, i);
1516 }
1517
1518 /*
1519  * alloc_nid() should be called prior to this function.
1520  */
1521 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1522 {
1523         struct f2fs_nm_info *nm_i = NM_I(sbi);
1524         struct free_nid *i;
1525         bool need_free = false;
1526
1527         if (!nid)
1528                 return;
1529
1530         spin_lock(&nm_i->free_nid_list_lock);
1531         i = __lookup_free_nid_list(nm_i, nid);
1532         f2fs_bug_on(!i || i->state != NID_ALLOC);
1533         if (!available_free_memory(sbi, FREE_NIDS)) {
1534                 __del_from_free_nid_list(nm_i, i);
1535                 need_free = true;
1536         } else {
1537                 i->state = NID_NEW;
1538                 nm_i->fcnt++;
1539         }
1540         spin_unlock(&nm_i->free_nid_list_lock);
1541
1542         if (need_free)
1543                 kmem_cache_free(free_nid_slab, i);
1544 }
1545
1546 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1547                 struct f2fs_summary *sum, struct node_info *ni,
1548                 block_t new_blkaddr)
1549 {
1550         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1551         set_node_addr(sbi, ni, new_blkaddr, false);
1552         clear_node_page_dirty(page);
1553 }
1554
1555 static void recover_inline_xattr(struct inode *inode, struct page *page)
1556 {
1557         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1558         void *src_addr, *dst_addr;
1559         size_t inline_size;
1560         struct page *ipage;
1561         struct f2fs_inode *ri;
1562
1563         if (!f2fs_has_inline_xattr(inode))
1564                 return;
1565
1566         if (!IS_INODE(page))
1567                 return;
1568
1569         ri = F2FS_INODE(page);
1570         if (!(ri->i_inline & F2FS_INLINE_XATTR))
1571                 return;
1572
1573         ipage = get_node_page(sbi, inode->i_ino);
1574         f2fs_bug_on(IS_ERR(ipage));
1575
1576         dst_addr = inline_xattr_addr(ipage);
1577         src_addr = inline_xattr_addr(page);
1578         inline_size = inline_xattr_size(inode);
1579
1580         f2fs_wait_on_page_writeback(ipage, NODE);
1581         memcpy(dst_addr, src_addr, inline_size);
1582
1583         update_inode(inode, ipage);
1584         f2fs_put_page(ipage, 1);
1585 }
1586
1587 bool recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1588 {
1589         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1590         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1591         nid_t new_xnid = nid_of_node(page);
1592         struct node_info ni;
1593
1594         recover_inline_xattr(inode, page);
1595
1596         if (!f2fs_has_xattr_block(ofs_of_node(page)))
1597                 return false;
1598
1599         /* 1: invalidate the previous xattr nid */
1600         if (!prev_xnid)
1601                 goto recover_xnid;
1602
1603         /* Deallocate node address */
1604         get_node_info(sbi, prev_xnid, &ni);
1605         f2fs_bug_on(ni.blk_addr == NULL_ADDR);
1606         invalidate_blocks(sbi, ni.blk_addr);
1607         dec_valid_node_count(sbi, inode);
1608         set_node_addr(sbi, &ni, NULL_ADDR, false);
1609
1610 recover_xnid:
1611         /* 2: allocate new xattr nid */
1612         if (unlikely(!inc_valid_node_count(sbi, inode)))
1613                 f2fs_bug_on(1);
1614
1615         remove_free_nid(NM_I(sbi), new_xnid);
1616         get_node_info(sbi, new_xnid, &ni);
1617         ni.ino = inode->i_ino;
1618         set_node_addr(sbi, &ni, NEW_ADDR, false);
1619         F2FS_I(inode)->i_xattr_nid = new_xnid;
1620
1621         /* 3: update xattr blkaddr */
1622         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1623         set_node_addr(sbi, &ni, blkaddr, false);
1624
1625         update_inode_page(inode);
1626         return true;
1627 }
1628
1629 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1630 {
1631         struct f2fs_inode *src, *dst;
1632         nid_t ino = ino_of_node(page);
1633         struct node_info old_ni, new_ni;
1634         struct page *ipage;
1635
1636         get_node_info(sbi, ino, &old_ni);
1637
1638         if (unlikely(old_ni.blk_addr != NULL_ADDR))
1639                 return -EINVAL;
1640
1641         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1642         if (!ipage)
1643                 return -ENOMEM;
1644
1645         /* Should not use this inode  from free nid list */
1646         remove_free_nid(NM_I(sbi), ino);
1647
1648         SetPageUptodate(ipage);
1649         fill_node_footer(ipage, ino, ino, 0, true);
1650
1651         src = F2FS_INODE(page);
1652         dst = F2FS_INODE(ipage);
1653
1654         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1655         dst->i_size = 0;
1656         dst->i_blocks = cpu_to_le64(1);
1657         dst->i_links = cpu_to_le32(1);
1658         dst->i_xattr_nid = 0;
1659
1660         new_ni = old_ni;
1661         new_ni.ino = ino;
1662
1663         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1664                 WARN_ON(1);
1665         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1666         inc_valid_inode_count(sbi);
1667         f2fs_put_page(ipage, 1);
1668         return 0;
1669 }
1670
1671 /*
1672  * ra_sum_pages() merge contiguous pages into one bio and submit.
1673  * these pre-readed pages are alloced in bd_inode's mapping tree.
1674  */
1675 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct page **pages,
1676                                 int start, int nrpages)
1677 {
1678         struct inode *inode = sbi->sb->s_bdev->bd_inode;
1679         struct address_space *mapping = inode->i_mapping;
1680         int i, page_idx = start;
1681         struct f2fs_io_info fio = {
1682                 .type = META,
1683                 .rw = READ_SYNC | REQ_META | REQ_PRIO
1684         };
1685
1686         for (i = 0; page_idx < start + nrpages; page_idx++, i++) {
1687                 /* alloc page in bd_inode for reading node summary info */
1688                 pages[i] = grab_cache_page(mapping, page_idx);
1689                 if (!pages[i])
1690                         break;
1691                 f2fs_submit_page_mbio(sbi, pages[i], page_idx, &fio);
1692         }
1693
1694         f2fs_submit_merged_bio(sbi, META, READ);
1695         return i;
1696 }
1697
1698 int restore_node_summary(struct f2fs_sb_info *sbi,
1699                         unsigned int segno, struct f2fs_summary_block *sum)
1700 {
1701         struct f2fs_node *rn;
1702         struct f2fs_summary *sum_entry;
1703         struct inode *inode = sbi->sb->s_bdev->bd_inode;
1704         block_t addr;
1705         int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1706         struct page *pages[bio_blocks];
1707         int i, idx, last_offset, nrpages, err = 0;
1708
1709         /* scan the node segment */
1710         last_offset = sbi->blocks_per_seg;
1711         addr = START_BLOCK(sbi, segno);
1712         sum_entry = &sum->entries[0];
1713
1714         for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) {
1715                 nrpages = min(last_offset - i, bio_blocks);
1716
1717                 /* read ahead node pages */
1718                 nrpages = ra_sum_pages(sbi, pages, addr, nrpages);
1719                 if (!nrpages)
1720                         return -ENOMEM;
1721
1722                 for (idx = 0; idx < nrpages; idx++) {
1723                         if (err)
1724                                 goto skip;
1725
1726                         lock_page(pages[idx]);
1727                         if (unlikely(!PageUptodate(pages[idx]))) {
1728                                 err = -EIO;
1729                         } else {
1730                                 rn = F2FS_NODE(pages[idx]);
1731                                 sum_entry->nid = rn->footer.nid;
1732                                 sum_entry->version = 0;
1733                                 sum_entry->ofs_in_node = 0;
1734                                 sum_entry++;
1735                         }
1736                         unlock_page(pages[idx]);
1737 skip:
1738                         page_cache_release(pages[idx]);
1739                 }
1740
1741                 invalidate_mapping_pages(inode->i_mapping, addr,
1742                                                         addr + nrpages);
1743         }
1744         return err;
1745 }
1746
1747 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1748 {
1749         struct f2fs_nm_info *nm_i = NM_I(sbi);
1750         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1751         struct f2fs_summary_block *sum = curseg->sum_blk;
1752         int i;
1753
1754         mutex_lock(&curseg->curseg_mutex);
1755
1756         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1757                 mutex_unlock(&curseg->curseg_mutex);
1758                 return false;
1759         }
1760
1761         for (i = 0; i < nats_in_cursum(sum); i++) {
1762                 struct nat_entry *ne;
1763                 struct f2fs_nat_entry raw_ne;
1764                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1765
1766                 raw_ne = nat_in_journal(sum, i);
1767 retry:
1768                 write_lock(&nm_i->nat_tree_lock);
1769                 ne = __lookup_nat_cache(nm_i, nid);
1770                 if (ne) {
1771                         __set_nat_cache_dirty(nm_i, ne);
1772                         write_unlock(&nm_i->nat_tree_lock);
1773                         continue;
1774                 }
1775                 ne = grab_nat_entry(nm_i, nid);
1776                 if (!ne) {
1777                         write_unlock(&nm_i->nat_tree_lock);
1778                         goto retry;
1779                 }
1780                 node_info_from_raw_nat(&ne->ni, &raw_ne);
1781                 __set_nat_cache_dirty(nm_i, ne);
1782                 write_unlock(&nm_i->nat_tree_lock);
1783         }
1784         update_nats_in_cursum(sum, -i);
1785         mutex_unlock(&curseg->curseg_mutex);
1786         return true;
1787 }
1788
1789 /*
1790  * This function is called during the checkpointing process.
1791  */
1792 void flush_nat_entries(struct f2fs_sb_info *sbi)
1793 {
1794         struct f2fs_nm_info *nm_i = NM_I(sbi);
1795         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1796         struct f2fs_summary_block *sum = curseg->sum_blk;
1797         struct nat_entry *ne, *cur;
1798         struct page *page = NULL;
1799         struct f2fs_nat_block *nat_blk = NULL;
1800         nid_t start_nid = 0, end_nid = 0;
1801         bool flushed;
1802
1803         flushed = flush_nats_in_journal(sbi);
1804
1805         if (!flushed)
1806                 mutex_lock(&curseg->curseg_mutex);
1807
1808         /* 1) flush dirty nat caches */
1809         list_for_each_entry_safe(ne, cur, &nm_i->dirty_nat_entries, list) {
1810                 nid_t nid;
1811                 struct f2fs_nat_entry raw_ne;
1812                 int offset = -1;
1813
1814                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1815                         continue;
1816
1817                 nid = nat_get_nid(ne);
1818
1819                 if (flushed)
1820                         goto to_nat_page;
1821
1822                 /* if there is room for nat enries in curseg->sumpage */
1823                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1824                 if (offset >= 0) {
1825                         raw_ne = nat_in_journal(sum, offset);
1826                         goto flush_now;
1827                 }
1828 to_nat_page:
1829                 if (!page || (start_nid > nid || nid > end_nid)) {
1830                         if (page) {
1831                                 f2fs_put_page(page, 1);
1832                                 page = NULL;
1833                         }
1834                         start_nid = START_NID(nid);
1835                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1836
1837                         /*
1838                          * get nat block with dirty flag, increased reference
1839                          * count, mapped and lock
1840                          */
1841                         page = get_next_nat_page(sbi, start_nid);
1842                         nat_blk = page_address(page);
1843                 }
1844
1845                 f2fs_bug_on(!nat_blk);
1846                 raw_ne = nat_blk->entries[nid - start_nid];
1847 flush_now:
1848                 raw_nat_from_node_info(&raw_ne, &ne->ni);
1849
1850                 if (offset < 0) {
1851                         nat_blk->entries[nid - start_nid] = raw_ne;
1852                 } else {
1853                         nat_in_journal(sum, offset) = raw_ne;
1854                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1855                 }
1856
1857                 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1858                                 add_free_nid(sbi, nid, false) <= 0) {
1859                         write_lock(&nm_i->nat_tree_lock);
1860                         __del_from_nat_cache(nm_i, ne);
1861                         write_unlock(&nm_i->nat_tree_lock);
1862                 } else {
1863                         write_lock(&nm_i->nat_tree_lock);
1864                         __clear_nat_cache_dirty(nm_i, ne);
1865                         write_unlock(&nm_i->nat_tree_lock);
1866                 }
1867         }
1868         if (!flushed)
1869                 mutex_unlock(&curseg->curseg_mutex);
1870         f2fs_put_page(page, 1);
1871 }
1872
1873 static int init_node_manager(struct f2fs_sb_info *sbi)
1874 {
1875         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1876         struct f2fs_nm_info *nm_i = NM_I(sbi);
1877         unsigned char *version_bitmap;
1878         unsigned int nat_segs, nat_blocks;
1879
1880         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1881
1882         /* segment_count_nat includes pair segment so divide to 2. */
1883         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1884         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1885
1886         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1887
1888         /* not used nids: 0, node, meta, (and root counted as valid node) */
1889         nm_i->available_nids = nm_i->max_nid - 3;
1890         nm_i->fcnt = 0;
1891         nm_i->nat_cnt = 0;
1892         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1893
1894         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1895         INIT_LIST_HEAD(&nm_i->free_nid_list);
1896         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1897         INIT_LIST_HEAD(&nm_i->nat_entries);
1898         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1899
1900         mutex_init(&nm_i->build_lock);
1901         spin_lock_init(&nm_i->free_nid_list_lock);
1902         rwlock_init(&nm_i->nat_tree_lock);
1903
1904         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1905         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1906         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1907         if (!version_bitmap)
1908                 return -EFAULT;
1909
1910         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1911                                         GFP_KERNEL);
1912         if (!nm_i->nat_bitmap)
1913                 return -ENOMEM;
1914         return 0;
1915 }
1916
1917 int build_node_manager(struct f2fs_sb_info *sbi)
1918 {
1919         int err;
1920
1921         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1922         if (!sbi->nm_info)
1923                 return -ENOMEM;
1924
1925         err = init_node_manager(sbi);
1926         if (err)
1927                 return err;
1928
1929         build_free_nids(sbi);
1930         return 0;
1931 }
1932
1933 void destroy_node_manager(struct f2fs_sb_info *sbi)
1934 {
1935         struct f2fs_nm_info *nm_i = NM_I(sbi);
1936         struct free_nid *i, *next_i;
1937         struct nat_entry *natvec[NATVEC_SIZE];
1938         nid_t nid = 0;
1939         unsigned int found;
1940
1941         if (!nm_i)
1942                 return;
1943
1944         /* destroy free nid list */
1945         spin_lock(&nm_i->free_nid_list_lock);
1946         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1947                 f2fs_bug_on(i->state == NID_ALLOC);
1948                 __del_from_free_nid_list(nm_i, i);
1949                 nm_i->fcnt--;
1950                 spin_unlock(&nm_i->free_nid_list_lock);
1951                 kmem_cache_free(free_nid_slab, i);
1952                 spin_lock(&nm_i->free_nid_list_lock);
1953         }
1954         f2fs_bug_on(nm_i->fcnt);
1955         spin_unlock(&nm_i->free_nid_list_lock);
1956
1957         /* destroy nat cache */
1958         write_lock(&nm_i->nat_tree_lock);
1959         while ((found = __gang_lookup_nat_cache(nm_i,
1960                                         nid, NATVEC_SIZE, natvec))) {
1961                 unsigned idx;
1962                 nid = nat_get_nid(natvec[found - 1]) + 1;
1963                 for (idx = 0; idx < found; idx++)
1964                         __del_from_nat_cache(nm_i, natvec[idx]);
1965         }
1966         f2fs_bug_on(nm_i->nat_cnt);
1967         write_unlock(&nm_i->nat_tree_lock);
1968
1969         kfree(nm_i->nat_bitmap);
1970         sbi->nm_info = NULL;
1971         kfree(nm_i);
1972 }
1973
1974 int __init create_node_manager_caches(void)
1975 {
1976         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1977                         sizeof(struct nat_entry));
1978         if (!nat_entry_slab)
1979                 return -ENOMEM;
1980
1981         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1982                         sizeof(struct free_nid));
1983         if (!free_nid_slab) {
1984                 kmem_cache_destroy(nat_entry_slab);
1985                 return -ENOMEM;
1986         }
1987         return 0;
1988 }
1989
1990 void destroy_node_manager_caches(void)
1991 {
1992         kmem_cache_destroy(free_nid_slab);
1993         kmem_cache_destroy(nat_entry_slab);
1994 }