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 "xattr.h"
  23 #include "trace.h"
  24 #include <trace/events/f2fs.h>
  25
  26 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
  27
  28 static struct kmem_cache *nat_entry_slab;
  29 static struct kmem_cache *free_nid_slab;
  30 static struct kmem_cache *nat_entry_set_slab;
  31 static struct kmem_cache *fsync_node_entry_slab;
  32
  33 /*
  34  * Check whether the given nid is within node id range.
  35  */
  36 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
  37 {
  38         if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
  39                 set_sbi_flag(sbi, SBI_NEED_FSCK);
  40                 f2fs_msg(sbi->sb, KERN_WARNING,
  41                                 "%s: out-of-range nid=%x, run fsck to fix.",
  42                                 __func__, nid);
  43                 return -EFSCORRUPTED;
  44         }
  45         return 0;
  46 }
  47
  48 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
  49 {
  50         struct f2fs_nm_info *nm_i = NM_I(sbi);
  51         struct sysinfo val;
  52         unsigned long avail_ram;
  53         unsigned long mem_size = 0;
  54         bool res = false;
  55
  56         si_meminfo(&val);
  57
  58         /* only uses low memory */
  59         avail_ram = val.totalram - val.totalhigh;
  60
  61         /*
  62          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  63          */
  64         if (type == FREE_NIDS) {
  65                 mem_size = (nm_i->nid_cnt[FREE_NID] *
  66                                 sizeof(struct free_nid)) >> PAGE_SHIFT;
  67                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  68         } else if (type == NAT_ENTRIES) {
  69                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
  70                                                         PAGE_SHIFT;
  71                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  72                 if (excess_cached_nats(sbi))
  73                         res = false;
  74         } else if (type == DIRTY_DENTS) {
  75                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  76                         return false;
  77                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  78                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  79         } else if (type == INO_ENTRIES) {
  80                 int i;
  81
  82                 for (i = 0; i < MAX_INO_ENTRY; i++)
  83                         mem_size += sbi->im[i].ino_num *
  84                                                 sizeof(struct ino_entry);
  85                 mem_size >>= PAGE_SHIFT;
  86                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  87         } else if (type == EXTENT_CACHE) {
  88                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  89                                 sizeof(struct extent_tree) +
  90                                 atomic_read(&sbi->total_ext_node) *
  91                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  92                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  93         } else if (type == INMEM_PAGES) {
  94                 /* it allows 20% / total_ram for inmemory pages */
  95                 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
  96                 res = mem_size < (val.totalram / 5);
  97         } else {
  98                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
  99                         return true;
 100         }
 101         return res;
 102 }
 103
 104 static void clear_node_page_dirty(struct page *page)
 105 {
 106         if (PageDirty(page)) {
 107                 f2fs_clear_radix_tree_dirty_tag(page);
 108                 clear_page_dirty_for_io(page);
 109                 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
 110         }
 111         ClearPageUptodate(page);
 112 }
 113
 114 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 115 {
 116         return f2fs_get_meta_page_nofail(sbi, current_nat_addr(sbi, nid));
 117 }
 118
 119 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 120 {
 121         struct page *src_page;
 122         struct page *dst_page;
 123         pgoff_t dst_off;
 124         void *src_addr;
 125         void *dst_addr;
 126         struct f2fs_nm_info *nm_i = NM_I(sbi);
 127
 128         dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
 129
 130         /* get current nat block page with lock */
 131         src_page = get_current_nat_page(sbi, nid);
 132         dst_page = f2fs_grab_meta_page(sbi, dst_off);
 133         f2fs_bug_on(sbi, PageDirty(src_page));
 134
 135         src_addr = page_address(src_page);
 136         dst_addr = page_address(dst_page);
 137         memcpy(dst_addr, src_addr, PAGE_SIZE);
 138         set_page_dirty(dst_page);
 139         f2fs_put_page(src_page, 1);
 140
 141         set_to_next_nat(nm_i, nid);
 142
 143         return dst_page;
 144 }
 145
 146 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
 147 {
 148         struct nat_entry *new;
 149
 150         if (no_fail)
 151                 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
 152         else
 153                 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
 154         if (new) {
 155                 nat_set_nid(new, nid);
 156                 nat_reset_flag(new);
 157         }
 158         return new;
 159 }
 160
 161 static void __free_nat_entry(struct nat_entry *e)
 162 {
 163         kmem_cache_free(nat_entry_slab, e);
 164 }
 165
 166 /* must be locked by nat_tree_lock */
 167 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
 168         struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
 169 {
 170         if (no_fail)
 171                 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
 172         else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
 173                 return NULL;
 174
 175         if (raw_ne)
 176                 node_info_from_raw_nat(&ne->ni, raw_ne);
 177
 178         spin_lock(&nm_i->nat_list_lock);
 179         list_add_tail(&ne->list, &nm_i->nat_entries);
 180         spin_unlock(&nm_i->nat_list_lock);
 181
 182         nm_i->nat_cnt++;
 183         return ne;
 184 }
 185
 186 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 187 {
 188         struct nat_entry *ne;
 189
 190         ne = radix_tree_lookup(&nm_i->nat_root, n);
 191
 192         /* for recent accessed nat entry, move it to tail of lru list */
 193         if (ne && !get_nat_flag(ne, IS_DIRTY)) {
 194                 spin_lock(&nm_i->nat_list_lock);
 195                 if (!list_empty(&ne->list))
 196                         list_move_tail(&ne->list, &nm_i->nat_entries);
 197                 spin_unlock(&nm_i->nat_list_lock);
 198         }
 199
 200         return ne;
 201 }
 202
 203 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 204                 nid_t start, unsigned int nr, struct nat_entry **ep)
 205 {
 206         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 207 }
 208
 209 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 210 {
 211         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 212         nm_i->nat_cnt--;
 213         __free_nat_entry(e);
 214 }
 215
 216 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
 217                                                         struct nat_entry *ne)
 218 {
 219         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 220         struct nat_entry_set *head;
 221
 222         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 223         if (!head) {
 224                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 225
 226                 INIT_LIST_HEAD(&head->entry_list);
 227                 INIT_LIST_HEAD(&head->set_list);
 228                 head->set = set;
 229                 head->entry_cnt = 0;
 230                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 231         }
 232         return head;
 233 }
 234
 235 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 236                                                 struct nat_entry *ne)
 237 {
 238         struct nat_entry_set *head;
 239         bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
 240
 241         if (!new_ne)
 242                 head = __grab_nat_entry_set(nm_i, ne);
 243
 244         /*
 245          * update entry_cnt in below condition:
 246          * 1. update NEW_ADDR to valid block address;
 247          * 2. update old block address to new one;
 248          */
 249         if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
 250                                 !get_nat_flag(ne, IS_DIRTY)))
 251                 head->entry_cnt++;
 252
 253         set_nat_flag(ne, IS_PREALLOC, new_ne);
 254
 255         if (get_nat_flag(ne, IS_DIRTY))
 256                 goto refresh_list;
 257
 258         nm_i->dirty_nat_cnt++;
 259         set_nat_flag(ne, IS_DIRTY, true);
 260 refresh_list:
 261         spin_lock(&nm_i->nat_list_lock);
 262         if (new_ne)
 263                 list_del_init(&ne->list);
 264         else
 265                 list_move_tail(&ne->list, &head->entry_list);
 266         spin_unlock(&nm_i->nat_list_lock);
 267 }
 268
 269 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 270                 struct nat_entry_set *set, struct nat_entry *ne)
 271 {
 272         spin_lock(&nm_i->nat_list_lock);
 273         list_move_tail(&ne->list, &nm_i->nat_entries);
 274         spin_unlock(&nm_i->nat_list_lock);
 275
 276         set_nat_flag(ne, IS_DIRTY, false);
 277         set->entry_cnt--;
 278         nm_i->dirty_nat_cnt--;
 279 }
 280
 281 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 282                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 283 {
 284         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 285                                                         start, nr);
 286 }
 287
 288 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
 289 {
 290         return NODE_MAPPING(sbi) == page->mapping &&
 291                         IS_DNODE(page) && is_cold_node(page);
 292 }
 293
 294 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
 295 {
 296         spin_lock_init(&sbi->fsync_node_lock);
 297         INIT_LIST_HEAD(&sbi->fsync_node_list);
 298         sbi->fsync_seg_id = 0;
 299         sbi->fsync_node_num = 0;
 300 }
 301
 302 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
 303                                                         struct page *page)
 304 {
 305         struct fsync_node_entry *fn;
 306         unsigned long flags;
 307         unsigned int seq_id;
 308
 309         fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
 310
 311         get_page(page);
 312         fn->page = page;
 313         INIT_LIST_HEAD(&fn->list);
 314
 315         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 316         list_add_tail(&fn->list, &sbi->fsync_node_list);
 317         fn->seq_id = sbi->fsync_seg_id++;
 318         seq_id = fn->seq_id;
 319         sbi->fsync_node_num++;
 320         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 321
 322         return seq_id;
 323 }
 324
 325 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
 326 {
 327         struct fsync_node_entry *fn;
 328         unsigned long flags;
 329
 330         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 331         list_for_each_entry(fn, &sbi->fsync_node_list, list) {
 332                 if (fn->page == page) {
 333                         list_del(&fn->list);
 334                         sbi->fsync_node_num--;
 335                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 336                         kmem_cache_free(fsync_node_entry_slab, fn);
 337                         put_page(page);
 338                         return;
 339                 }
 340         }
 341         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 342         f2fs_bug_on(sbi, 1);
 343 }
 344
 345 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
 346 {
 347         unsigned long flags;
 348
 349         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 350         sbi->fsync_seg_id = 0;
 351         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 352 }
 353
 354 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 355 {
 356         struct f2fs_nm_info *nm_i = NM_I(sbi);
 357         struct nat_entry *e;
 358         bool need = false;
 359
 360         down_read(&nm_i->nat_tree_lock);
 361         e = __lookup_nat_cache(nm_i, nid);
 362         if (e) {
 363                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 364                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 365                         need = true;
 366         }
 367         up_read(&nm_i->nat_tree_lock);
 368         return need;
 369 }
 370
 371 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 372 {
 373         struct f2fs_nm_info *nm_i = NM_I(sbi);
 374         struct nat_entry *e;
 375         bool is_cp = true;
 376
 377         down_read(&nm_i->nat_tree_lock);
 378         e = __lookup_nat_cache(nm_i, nid);
 379         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 380                 is_cp = false;
 381         up_read(&nm_i->nat_tree_lock);
 382         return is_cp;
 383 }
 384
 385 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 386 {
 387         struct f2fs_nm_info *nm_i = NM_I(sbi);
 388         struct nat_entry *e;
 389         bool need_update = true;
 390
 391         down_read(&nm_i->nat_tree_lock);
 392         e = __lookup_nat_cache(nm_i, ino);
 393         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 394                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 395                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 396                 need_update = false;
 397         up_read(&nm_i->nat_tree_lock);
 398         return need_update;
 399 }
 400
 401 /* must be locked by nat_tree_lock */
 402 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 403                                                 struct f2fs_nat_entry *ne)
 404 {
 405         struct f2fs_nm_info *nm_i = NM_I(sbi);
 406         struct nat_entry *new, *e;
 407
 408         new = __alloc_nat_entry(nid, false);
 409         if (!new)
 410                 return;
 411
 412         down_write(&nm_i->nat_tree_lock);
 413         e = __lookup_nat_cache(nm_i, nid);
 414         if (!e)
 415                 e = __init_nat_entry(nm_i, new, ne, false);
 416         else
 417                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 418                                 nat_get_blkaddr(e) !=
 419                                         le32_to_cpu(ne->block_addr) ||
 420                                 nat_get_version(e) != ne->version);
 421         up_write(&nm_i->nat_tree_lock);
 422         if (e != new)
 423                 __free_nat_entry(new);
 424 }
 425
 426 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 427                         block_t new_blkaddr, bool fsync_done)
 428 {
 429         struct f2fs_nm_info *nm_i = NM_I(sbi);
 430         struct nat_entry *e;
 431         struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
 432
 433         down_write(&nm_i->nat_tree_lock);
 434         e = __lookup_nat_cache(nm_i, ni->nid);
 435         if (!e) {
 436                 e = __init_nat_entry(nm_i, new, NULL, true);
 437                 copy_node_info(&e->ni, ni);
 438                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 439         } else if (new_blkaddr == NEW_ADDR) {
 440                 /*
 441                  * when nid is reallocated,
 442                  * previous nat entry can be remained in nat cache.
 443                  * So, reinitialize it with new information.
 444                  */
 445                 copy_node_info(&e->ni, ni);
 446                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 447         }
 448         /* let's free early to reduce memory consumption */
 449         if (e != new)
 450                 __free_nat_entry(new);
 451
 452         /* sanity check */
 453         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 454         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 455                         new_blkaddr == NULL_ADDR);
 456         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 457                         new_blkaddr == NEW_ADDR);
 458         f2fs_bug_on(sbi, is_valid_data_blkaddr(sbi, nat_get_blkaddr(e)) &&
 459                         new_blkaddr == NEW_ADDR);
 460
 461         /* increment version no as node is removed */
 462         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 463                 unsigned char version = nat_get_version(e);
 464                 nat_set_version(e, inc_node_version(version));
 465         }
 466
 467         /* change address */
 468         nat_set_blkaddr(e, new_blkaddr);
 469         if (!is_valid_data_blkaddr(sbi, new_blkaddr))
 470                 set_nat_flag(e, IS_CHECKPOINTED, false);
 471         __set_nat_cache_dirty(nm_i, e);
 472
 473         /* update fsync_mark if its inode nat entry is still alive */
 474         if (ni->nid != ni->ino)
 475                 e = __lookup_nat_cache(nm_i, ni->ino);
 476         if (e) {
 477                 if (fsync_done && ni->nid == ni->ino)
 478                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 479                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 480         }
 481         up_write(&nm_i->nat_tree_lock);
 482 }
 483
 484 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 485 {
 486         struct f2fs_nm_info *nm_i = NM_I(sbi);
 487         int nr = nr_shrink;
 488
 489         if (!down_write_trylock(&nm_i->nat_tree_lock))
 490                 return 0;
 491
 492         spin_lock(&nm_i->nat_list_lock);
 493         while (nr_shrink) {
 494                 struct nat_entry *ne;
 495
 496                 if (list_empty(&nm_i->nat_entries))
 497                         break;
 498
 499                 ne = list_first_entry(&nm_i->nat_entries,
 500                                         struct nat_entry, list);
 501                 list_del(&ne->list);
 502                 spin_unlock(&nm_i->nat_list_lock);
 503
 504                 __del_from_nat_cache(nm_i, ne);
 505                 nr_shrink--;
 506
 507                 spin_lock(&nm_i->nat_list_lock);
 508         }
 509         spin_unlock(&nm_i->nat_list_lock);
 510
 511         up_write(&nm_i->nat_tree_lock);
 512         return nr - nr_shrink;
 513 }
 514
 515 /*
 516  * This function always returns success
 517  */
 518 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
 519                                                 struct node_info *ni)
 520 {
 521         struct f2fs_nm_info *nm_i = NM_I(sbi);
 522         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 523         struct f2fs_journal *journal = curseg->journal;
 524         nid_t start_nid = START_NID(nid);
 525         struct f2fs_nat_block *nat_blk;
 526         struct page *page = NULL;
 527         struct f2fs_nat_entry ne;
 528         struct nat_entry *e;
 529         pgoff_t index;
 530         int i;
 531
 532         ni->nid = nid;
 533
 534         /* Check nat cache */
 535         down_read(&nm_i->nat_tree_lock);
 536         e = __lookup_nat_cache(nm_i, nid);
 537         if (e) {
 538                 ni->ino = nat_get_ino(e);
 539                 ni->blk_addr = nat_get_blkaddr(e);
 540                 ni->version = nat_get_version(e);
 541                 up_read(&nm_i->nat_tree_lock);
 542                 return 0;
 543         }
 544
 545         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 546
 547         /* Check current segment summary */
 548         down_read(&curseg->journal_rwsem);
 549         i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 550         if (i >= 0) {
 551                 ne = nat_in_journal(journal, i);
 552                 node_info_from_raw_nat(ni, &ne);
 553         }
 554         up_read(&curseg->journal_rwsem);
 555         if (i >= 0) {
 556                 up_read(&nm_i->nat_tree_lock);
 557                 goto cache;
 558         }
 559
 560         /* Fill node_info from nat page */
 561         index = current_nat_addr(sbi, nid);
 562         up_read(&nm_i->nat_tree_lock);
 563
 564         page = f2fs_get_meta_page(sbi, index);
 565         if (IS_ERR(page))
 566                 return PTR_ERR(page);
 567
 568         nat_blk = (struct f2fs_nat_block *)page_address(page);
 569         ne = nat_blk->entries[nid - start_nid];
 570         node_info_from_raw_nat(ni, &ne);
 571         f2fs_put_page(page, 1);
 572 cache:
 573         /* cache nat entry */
 574         cache_nat_entry(sbi, nid, &ne);
 575         return 0;
 576 }
 577
 578 /*
 579  * readahead MAX_RA_NODE number of node pages.
 580  */
 581 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
 582 {
 583         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 584         struct blk_plug plug;
 585         int i, end;
 586         nid_t nid;
 587
 588         blk_start_plug(&plug);
 589
 590         /* Then, try readahead for siblings of the desired node */
 591         end = start + n;
 592         end = min(end, NIDS_PER_BLOCK);
 593         for (i = start; i < end; i++) {
 594                 nid = get_nid(parent, i, false);
 595                 f2fs_ra_node_page(sbi, nid);
 596         }
 597
 598         blk_finish_plug(&plug);
 599 }
 600
 601 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 602 {
 603         const long direct_index = ADDRS_PER_INODE(dn->inode);
 604         const long direct_blks = ADDRS_PER_BLOCK;
 605         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 606         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 607         int cur_level = dn->cur_level;
 608         int max_level = dn->max_level;
 609         pgoff_t base = 0;
 610
 611         if (!dn->max_level)
 612                 return pgofs + 1;
 613
 614         while (max_level-- > cur_level)
 615                 skipped_unit *= NIDS_PER_BLOCK;
 616
 617         switch (dn->max_level) {
 618         case 3:
 619                 base += 2 * indirect_blks;
 620         case 2:
 621                 base += 2 * direct_blks;
 622         case 1:
 623                 base += direct_index;
 624                 break;
 625         default:
 626                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 627         }
 628
 629         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 630 }
 631
 632 /*
 633  * The maximum depth is four.
 634  * Offset[0] will have raw inode offset.
 635  */
 636 static int get_node_path(struct inode *inode, long block,
 637                                 int offset[4], unsigned int noffset[4])
 638 {
 639         const long direct_index = ADDRS_PER_INODE(inode);
 640         const long direct_blks = ADDRS_PER_BLOCK;
 641         const long dptrs_per_blk = NIDS_PER_BLOCK;
 642         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 643         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 644         int n = 0;
 645         int level = 0;
 646
 647         noffset[0] = 0;
 648
 649         if (block < direct_index) {
 650                 offset[n] = block;
 651                 goto got;
 652         }
 653         block -= direct_index;
 654         if (block < direct_blks) {
 655                 offset[n++] = NODE_DIR1_BLOCK;
 656                 noffset[n] = 1;
 657                 offset[n] = block;
 658                 level = 1;
 659                 goto got;
 660         }
 661         block -= direct_blks;
 662         if (block < direct_blks) {
 663                 offset[n++] = NODE_DIR2_BLOCK;
 664                 noffset[n] = 2;
 665                 offset[n] = block;
 666                 level = 1;
 667                 goto got;
 668         }
 669         block -= direct_blks;
 670         if (block < indirect_blks) {
 671                 offset[n++] = NODE_IND1_BLOCK;
 672                 noffset[n] = 3;
 673                 offset[n++] = block / direct_blks;
 674                 noffset[n] = 4 + offset[n - 1];
 675                 offset[n] = block % direct_blks;
 676                 level = 2;
 677                 goto got;
 678         }
 679         block -= indirect_blks;
 680         if (block < indirect_blks) {
 681                 offset[n++] = NODE_IND2_BLOCK;
 682                 noffset[n] = 4 + dptrs_per_blk;
 683                 offset[n++] = block / direct_blks;
 684                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 685                 offset[n] = block % direct_blks;
 686                 level = 2;
 687                 goto got;
 688         }
 689         block -= indirect_blks;
 690         if (block < dindirect_blks) {
 691                 offset[n++] = NODE_DIND_BLOCK;
 692                 noffset[n] = 5 + (dptrs_per_blk * 2);
 693                 offset[n++] = block / indirect_blks;
 694                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 695                               offset[n - 1] * (dptrs_per_blk + 1);
 696                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 697                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 698                               offset[n - 2] * (dptrs_per_blk + 1) +
 699                               offset[n - 1];
 700                 offset[n] = block % direct_blks;
 701                 level = 3;
 702                 goto got;
 703         } else {
 704                 return -E2BIG;
 705         }
 706 got:
 707         return level;
 708 }
 709
 710 /*
 711  * Caller should call f2fs_put_dnode(dn).
 712  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 713  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 714  * In the case of RDONLY_NODE, we don't need to care about mutex.
 715  */
 716 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 717 {
 718         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 719         struct page *npage[4];
 720         struct page *parent = NULL;
 721         int offset[4];
 722         unsigned int noffset[4];
 723         nid_t nids[4];
 724         int level, i = 0;
 725         int err = 0;
 726
 727         level = get_node_path(dn->inode, index, offset, noffset);
 728         if (level < 0)
 729                 return level;
 730
 731         nids[0] = dn->inode->i_ino;
 732         npage[0] = dn->inode_page;
 733
 734         if (!npage[0]) {
 735                 npage[0] = f2fs_get_node_page(sbi, nids[0]);
 736                 if (IS_ERR(npage[0]))
 737                         return PTR_ERR(npage[0]);
 738         }
 739
 740         /* if inline_data is set, should not report any block indices */
 741         if (f2fs_has_inline_data(dn->inode) && index) {
 742                 err = -ENOENT;
 743                 f2fs_put_page(npage[0], 1);
 744                 goto release_out;
 745         }
 746
 747         parent = npage[0];
 748         if (level != 0)
 749                 nids[1] = get_nid(parent, offset[0], true);
 750         dn->inode_page = npage[0];
 751         dn->inode_page_locked = true;
 752
 753         /* get indirect or direct nodes */
 754         for (i = 1; i <= level; i++) {
 755                 bool done = false;
 756
 757                 if (!nids[i] && mode == ALLOC_NODE) {
 758                         /* alloc new node */
 759                         if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
 760                                 err = -ENOSPC;
 761                                 goto release_pages;
 762                         }
 763
 764                         dn->nid = nids[i];
 765                         npage[i] = f2fs_new_node_page(dn, noffset[i]);
 766                         if (IS_ERR(npage[i])) {
 767                                 f2fs_alloc_nid_failed(sbi, nids[i]);
 768                                 err = PTR_ERR(npage[i]);
 769                                 goto release_pages;
 770                         }
 771
 772                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 773                         f2fs_alloc_nid_done(sbi, nids[i]);
 774                         done = true;
 775                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 776                         npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
 777                         if (IS_ERR(npage[i])) {
 778                                 err = PTR_ERR(npage[i]);
 779                                 goto release_pages;
 780                         }
 781                         done = true;
 782                 }
 783                 if (i == 1) {
 784                         dn->inode_page_locked = false;
 785                         unlock_page(parent);
 786                 } else {
 787                         f2fs_put_page(parent, 1);
 788                 }
 789
 790                 if (!done) {
 791                         npage[i] = f2fs_get_node_page(sbi, nids[i]);
 792                         if (IS_ERR(npage[i])) {
 793                                 err = PTR_ERR(npage[i]);
 794                                 f2fs_put_page(npage[0], 0);
 795                                 goto release_out;
 796                         }
 797                 }
 798                 if (i < level) {
 799                         parent = npage[i];
 800                         nids[i + 1] = get_nid(parent, offset[i], false);
 801                 }
 802         }
 803         dn->nid = nids[level];
 804         dn->ofs_in_node = offset[level];
 805         dn->node_page = npage[level];
 806         dn->data_blkaddr = datablock_addr(dn->inode,
 807                                 dn->node_page, dn->ofs_in_node);
 808         return 0;
 809
 810 release_pages:
 811         f2fs_put_page(parent, 1);
 812         if (i > 1)
 813                 f2fs_put_page(npage[0], 0);
 814 release_out:
 815         dn->inode_page = NULL;
 816         dn->node_page = NULL;
 817         if (err == -ENOENT) {
 818                 dn->cur_level = i;
 819                 dn->max_level = level;
 820                 dn->ofs_in_node = offset[level];
 821         }
 822         return err;
 823 }
 824
 825 static int truncate_node(struct dnode_of_data *dn)
 826 {
 827         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 828         struct node_info ni;
 829         int err;
 830         pgoff_t index;
 831
 832         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 833         if (err)
 834                 return err;
 835
 836         /* Deallocate node address */
 837         f2fs_invalidate_blocks(sbi, ni.blk_addr);
 838         dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
 839         set_node_addr(sbi, &ni, NULL_ADDR, false);
 840
 841         if (dn->nid == dn->inode->i_ino) {
 842                 f2fs_remove_orphan_inode(sbi, dn->nid);
 843                 dec_valid_inode_count(sbi);
 844                 f2fs_inode_synced(dn->inode);
 845         }
 846
 847         clear_node_page_dirty(dn->node_page);
 848         set_sbi_flag(sbi, SBI_IS_DIRTY);
 849
 850         index = dn->node_page->index;
 851         f2fs_put_page(dn->node_page, 1);
 852
 853         invalidate_mapping_pages(NODE_MAPPING(sbi),
 854                         index, index);
 855
 856         dn->node_page = NULL;
 857         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 858
 859         return 0;
 860 }
 861
 862 static int truncate_dnode(struct dnode_of_data *dn)
 863 {
 864         struct page *page;
 865         int err;
 866
 867         if (dn->nid == 0)
 868                 return 1;
 869
 870         /* get direct node */
 871         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 872         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 873                 return 1;
 874         else if (IS_ERR(page))
 875                 return PTR_ERR(page);
 876
 877         /* Make dnode_of_data for parameter */
 878         dn->node_page = page;
 879         dn->ofs_in_node = 0;
 880         f2fs_truncate_data_blocks(dn);
 881         err = truncate_node(dn);
 882         if (err)
 883                 return err;
 884
 885         return 1;
 886 }
 887
 888 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 889                                                 int ofs, int depth)
 890 {
 891         struct dnode_of_data rdn = *dn;
 892         struct page *page;
 893         struct f2fs_node *rn;
 894         nid_t child_nid;
 895         unsigned int child_nofs;
 896         int freed = 0;
 897         int i, ret;
 898
 899         if (dn->nid == 0)
 900                 return NIDS_PER_BLOCK + 1;
 901
 902         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 903
 904         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 905         if (IS_ERR(page)) {
 906                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 907                 return PTR_ERR(page);
 908         }
 909
 910         f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 911
 912         rn = F2FS_NODE(page);
 913         if (depth < 3) {
 914                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 915                         child_nid = le32_to_cpu(rn->in.nid[i]);
 916                         if (child_nid == 0)
 917                                 continue;
 918                         rdn.nid = child_nid;
 919                         ret = truncate_dnode(&rdn);
 920                         if (ret < 0)
 921                                 goto out_err;
 922                         if (set_nid(page, i, 0, false))
 923                                 dn->node_changed = true;
 924                 }
 925         } else {
 926                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 927                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 928                         child_nid = le32_to_cpu(rn->in.nid[i]);
 929                         if (child_nid == 0) {
 930                                 child_nofs += NIDS_PER_BLOCK + 1;
 931                                 continue;
 932                         }
 933                         rdn.nid = child_nid;
 934                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 935                         if (ret == (NIDS_PER_BLOCK + 1)) {
 936                                 if (set_nid(page, i, 0, false))
 937                                         dn->node_changed = true;
 938                                 child_nofs += ret;
 939                         } else if (ret < 0 && ret != -ENOENT) {
 940                                 goto out_err;
 941                         }
 942                 }
 943                 freed = child_nofs;
 944         }
 945
 946         if (!ofs) {
 947                 /* remove current indirect node */
 948                 dn->node_page = page;
 949                 ret = truncate_node(dn);
 950                 if (ret)
 951                         goto out_err;
 952                 freed++;
 953         } else {
 954                 f2fs_put_page(page, 1);
 955         }
 956         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 957         return freed;
 958
 959 out_err:
 960         f2fs_put_page(page, 1);
 961         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 962         return ret;
 963 }
 964
 965 static int truncate_partial_nodes(struct dnode_of_data *dn,
 966                         struct f2fs_inode *ri, int *offset, int depth)
 967 {
 968         struct page *pages[2];
 969         nid_t nid[3];
 970         nid_t child_nid;
 971         int err = 0;
 972         int i;
 973         int idx = depth - 2;
 974
 975         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 976         if (!nid[0])
 977                 return 0;
 978
 979         /* get indirect nodes in the path */
 980         for (i = 0; i < idx + 1; i++) {
 981                 /* reference count'll be increased */
 982                 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 983                 if (IS_ERR(pages[i])) {
 984                         err = PTR_ERR(pages[i]);
 985                         idx = i - 1;
 986                         goto fail;
 987                 }
 988                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 989         }
 990
 991         f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 992
 993         /* free direct nodes linked to a partial indirect node */
 994         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 995                 child_nid = get_nid(pages[idx], i, false);
 996                 if (!child_nid)
 997                         continue;
 998                 dn->nid = child_nid;
 999                 err = truncate_dnode(dn);
1000                 if (err < 0)
1001                         goto fail;
1002                 if (set_nid(pages[idx], i, 0, false))
1003                         dn->node_changed = true;
1004         }
1005
1006         if (offset[idx + 1] == 0) {
1007                 dn->node_page = pages[idx];
1008                 dn->nid = nid[idx];
1009                 err = truncate_node(dn);
1010                 if (err)
1011                         goto fail;
1012         } else {
1013                 f2fs_put_page(pages[idx], 1);
1014         }
1015         offset[idx]++;
1016         offset[idx + 1] = 0;
1017         idx--;
1018 fail:
1019         for (i = idx; i >= 0; i--)
1020                 f2fs_put_page(pages[i], 1);
1021
1022         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1023
1024         return err;
1025 }
1026
1027 /*
1028  * All the block addresses of data and nodes should be nullified.
1029  */
1030 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1031 {
1032         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1033         int err = 0, cont = 1;
1034         int level, offset[4], noffset[4];
1035         unsigned int nofs = 0;
1036         struct f2fs_inode *ri;
1037         struct dnode_of_data dn;
1038         struct page *page;
1039
1040         trace_f2fs_truncate_inode_blocks_enter(inode, from);
1041
1042         level = get_node_path(inode, from, offset, noffset);
1043         if (level < 0)
1044                 return level;
1045
1046         page = f2fs_get_node_page(sbi, inode->i_ino);
1047         if (IS_ERR(page)) {
1048                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1049                 return PTR_ERR(page);
1050         }
1051
1052         set_new_dnode(&dn, inode, page, NULL, 0);
1053         unlock_page(page);
1054
1055         ri = F2FS_INODE(page);
1056         switch (level) {
1057         case 0:
1058         case 1:
1059                 nofs = noffset[1];
1060                 break;
1061         case 2:
1062                 nofs = noffset[1];
1063                 if (!offset[level - 1])
1064                         goto skip_partial;
1065                 err = truncate_partial_nodes(&dn, ri, offset, level);
1066                 if (err < 0 && err != -ENOENT)
1067                         goto fail;
1068                 nofs += 1 + NIDS_PER_BLOCK;
1069                 break;
1070         case 3:
1071                 nofs = 5 + 2 * NIDS_PER_BLOCK;
1072                 if (!offset[level - 1])
1073                         goto skip_partial;
1074                 err = truncate_partial_nodes(&dn, ri, offset, level);
1075                 if (err < 0 && err != -ENOENT)
1076                         goto fail;
1077                 break;
1078         default:
1079                 BUG();
1080         }
1081
1082 skip_partial:
1083         while (cont) {
1084                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1085                 switch (offset[0]) {
1086                 case NODE_DIR1_BLOCK:
1087                 case NODE_DIR2_BLOCK:
1088                         err = truncate_dnode(&dn);
1089                         break;
1090
1091                 case NODE_IND1_BLOCK:
1092                 case NODE_IND2_BLOCK:
1093                         err = truncate_nodes(&dn, nofs, offset[1], 2);
1094                         break;
1095
1096                 case NODE_DIND_BLOCK:
1097                         err = truncate_nodes(&dn, nofs, offset[1], 3);
1098                         cont = 0;
1099                         break;
1100
1101                 default:
1102                         BUG();
1103                 }
1104                 if (err < 0 && err != -ENOENT)
1105                         goto fail;
1106                 if (offset[1] == 0 &&
1107                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1108                         lock_page(page);
1109                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
1110                         f2fs_wait_on_page_writeback(page, NODE, true);
1111                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1112                         set_page_dirty(page);
1113                         unlock_page(page);
1114                 }
1115                 offset[1] = 0;
1116                 offset[0]++;
1117                 nofs += err;
1118         }
1119 fail:
1120         f2fs_put_page(page, 0);
1121         trace_f2fs_truncate_inode_blocks_exit(inode, err);
1122         return err > 0 ? 0 : err;
1123 }
1124
1125 /* caller must lock inode page */
1126 int f2fs_truncate_xattr_node(struct inode *inode)
1127 {
1128         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1129         nid_t nid = F2FS_I(inode)->i_xattr_nid;
1130         struct dnode_of_data dn;
1131         struct page *npage;
1132         int err;
1133
1134         if (!nid)
1135                 return 0;
1136
1137         npage = f2fs_get_node_page(sbi, nid);
1138         if (IS_ERR(npage))
1139                 return PTR_ERR(npage);
1140
1141         set_new_dnode(&dn, inode, NULL, npage, nid);
1142         err = truncate_node(&dn);
1143         if (err) {
1144                 f2fs_put_page(npage, 1);
1145                 return err;
1146         }
1147
1148         f2fs_i_xnid_write(inode, 0);
1149
1150         return 0;
1151 }
1152
1153 /*
1154  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1155  * f2fs_unlock_op().
1156  */
1157 int f2fs_remove_inode_page(struct inode *inode)
1158 {
1159         struct dnode_of_data dn;
1160         int err;
1161
1162         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1163         err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1164         if (err)
1165                 return err;
1166
1167         err = f2fs_truncate_xattr_node(inode);
1168         if (err) {
1169                 f2fs_put_dnode(&dn);
1170                 return err;
1171         }
1172
1173         /* remove potential inline_data blocks */
1174         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1175                                 S_ISLNK(inode->i_mode))
1176                 f2fs_truncate_data_blocks_range(&dn, 1);
1177
1178         /* 0 is possible, after f2fs_new_inode() has failed */
1179         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1180                 f2fs_put_dnode(&dn);
1181                 return -EIO;
1182         }
1183
1184         if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1185                 f2fs_msg(F2FS_I_SB(inode)->sb, KERN_WARNING,
1186                         "Inconsistent i_blocks, ino:%lu, iblocks:%llu",
1187                         inode->i_ino,
1188                         (unsigned long long)inode->i_blocks);
1189                 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1190         }
1191
1192         /* will put inode & node pages */
1193         err = truncate_node(&dn);
1194         if (err) {
1195                 f2fs_put_dnode(&dn);
1196                 return err;
1197         }
1198         return 0;
1199 }
1200
1201 struct page *f2fs_new_inode_page(struct inode *inode)
1202 {
1203         struct dnode_of_data dn;
1204
1205         /* allocate inode page for new inode */
1206         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1207
1208         /* caller should f2fs_put_page(page, 1); */
1209         return f2fs_new_node_page(&dn, 0);
1210 }
1211
1212 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1213 {
1214         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1215         struct node_info new_ni;
1216         struct page *page;
1217         int err;
1218
1219         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1220                 return ERR_PTR(-EPERM);
1221
1222         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1223         if (!page)
1224                 return ERR_PTR(-ENOMEM);
1225
1226         if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1227                 goto fail;
1228
1229 #ifdef CONFIG_F2FS_CHECK_FS
1230         err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1231         if (err) {
1232                 dec_valid_node_count(sbi, dn->inode, !ofs);
1233                 goto fail;
1234         }
1235         if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1236                 err = -EFSCORRUPTED;
1237                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1238                 goto fail;
1239         }
1240 #endif
1241         new_ni.nid = dn->nid;
1242         new_ni.ino = dn->inode->i_ino;
1243         new_ni.blk_addr = NULL_ADDR;
1244         new_ni.flag = 0;
1245         new_ni.version = 0;
1246         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1247
1248         f2fs_wait_on_page_writeback(page, NODE, true);
1249         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1250         set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1251         if (!PageUptodate(page))
1252                 SetPageUptodate(page);
1253         if (set_page_dirty(page))
1254                 dn->node_changed = true;
1255
1256         if (f2fs_has_xattr_block(ofs))
1257                 f2fs_i_xnid_write(dn->inode, dn->nid);
1258
1259         if (ofs == 0)
1260                 inc_valid_inode_count(sbi);
1261         return page;
1262
1263 fail:
1264         clear_node_page_dirty(page);
1265         f2fs_put_page(page, 1);
1266         return ERR_PTR(err);
1267 }
1268
1269 /*
1270  * Caller should do after getting the following values.
1271  * 0: f2fs_put_page(page, 0)
1272  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1273  */
1274 static int read_node_page(struct page *page, int op_flags)
1275 {
1276         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1277         struct node_info ni;
1278         struct f2fs_io_info fio = {
1279                 .sbi = sbi,
1280                 .type = NODE,
1281                 .op = REQ_OP_READ,
1282                 .op_flags = op_flags,
1283                 .page = page,
1284                 .encrypted_page = NULL,
1285         };
1286         int err;
1287
1288         if (PageUptodate(page)) {
1289                 if (!f2fs_inode_chksum_verify(sbi, page)) {
1290                         ClearPageUptodate(page);
1291                         return -EFSBADCRC;
1292                 }
1293                 return LOCKED_PAGE;
1294         }
1295
1296         err = f2fs_get_node_info(sbi, page->index, &ni);
1297         if (err)
1298                 return err;
1299
1300         if (unlikely(ni.blk_addr == NULL_ADDR) ||
1301                         is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1302                 ClearPageUptodate(page);
1303                 return -ENOENT;
1304         }
1305
1306         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1307         return f2fs_submit_page_bio(&fio);
1308 }
1309
1310 /*
1311  * Readahead a node page
1312  */
1313 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1314 {
1315         struct page *apage;
1316         int err;
1317
1318         if (!nid)
1319                 return;
1320         if (f2fs_check_nid_range(sbi, nid))
1321                 return;
1322
1323         rcu_read_lock();
1324         apage = radix_tree_lookup(&NODE_MAPPING(sbi)->i_pages, nid);
1325         rcu_read_unlock();
1326         if (apage)
1327                 return;
1328
1329         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1330         if (!apage)
1331                 return;
1332
1333         err = read_node_page(apage, REQ_RAHEAD);
1334         f2fs_put_page(apage, err ? 1 : 0);
1335 }
1336
1337 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1338                                         struct page *parent, int start)
1339 {
1340         struct page *page;
1341         int err;
1342
1343         if (!nid)
1344                 return ERR_PTR(-ENOENT);
1345         if (f2fs_check_nid_range(sbi, nid))
1346                 return ERR_PTR(-EINVAL);
1347 repeat:
1348         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1349         if (!page)
1350                 return ERR_PTR(-ENOMEM);
1351
1352         err = read_node_page(page, 0);
1353         if (err < 0) {
1354                 f2fs_put_page(page, 1);
1355                 return ERR_PTR(err);
1356         } else if (err == LOCKED_PAGE) {
1357                 err = 0;
1358                 goto page_hit;
1359         }
1360
1361         if (parent)
1362                 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1363
1364         lock_page(page);
1365
1366         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1367                 f2fs_put_page(page, 1);
1368                 goto repeat;
1369         }
1370
1371         if (unlikely(!PageUptodate(page))) {
1372                 err = -EIO;
1373                 goto out_err;
1374         }
1375
1376         if (!f2fs_inode_chksum_verify(sbi, page)) {
1377                 err = -EFSBADCRC;
1378                 goto out_err;
1379         }
1380 page_hit:
1381         if(unlikely(nid != nid_of_node(page))) {
1382                 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1383                         "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1384                         nid, nid_of_node(page), ino_of_node(page),
1385                         ofs_of_node(page), cpver_of_node(page),
1386                         next_blkaddr_of_node(page));
1387                 err = -EINVAL;
1388 out_err:
1389                 ClearPageUptodate(page);
1390                 f2fs_put_page(page, 1);
1391                 return ERR_PTR(err);
1392         }
1393         return page;
1394 }
1395
1396 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1397 {
1398         return __get_node_page(sbi, nid, NULL, 0);
1399 }
1400
1401 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1402 {
1403         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1404         nid_t nid = get_nid(parent, start, false);
1405
1406         return __get_node_page(sbi, nid, parent, start);
1407 }
1408
1409 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1410 {
1411         struct inode *inode;
1412         struct page *page;
1413         int ret;
1414
1415         /* should flush inline_data before evict_inode */
1416         inode = ilookup(sbi->sb, ino);
1417         if (!inode)
1418                 return;
1419
1420         page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1421                                         FGP_LOCK|FGP_NOWAIT, 0);
1422         if (!page)
1423                 goto iput_out;
1424
1425         if (!PageUptodate(page))
1426                 goto page_out;
1427
1428         if (!PageDirty(page))
1429                 goto page_out;
1430
1431         if (!clear_page_dirty_for_io(page))
1432                 goto page_out;
1433
1434         ret = f2fs_write_inline_data(inode, page);
1435         inode_dec_dirty_pages(inode);
1436         f2fs_remove_dirty_inode(inode);
1437         if (ret)
1438                 set_page_dirty(page);
1439 page_out:
1440         f2fs_put_page(page, 1);
1441 iput_out:
1442         iput(inode);
1443 }
1444
1445 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1446 {
1447         pgoff_t index;
1448         struct pagevec pvec;
1449         struct page *last_page = NULL;
1450         int nr_pages;
1451
1452         pagevec_init(&pvec);
1453         index = 0;
1454
1455         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1456                                 PAGECACHE_TAG_DIRTY))) {
1457                 int i;
1458
1459                 for (i = 0; i < nr_pages; i++) {
1460                         struct page *page = pvec.pages[i];
1461
1462                         if (unlikely(f2fs_cp_error(sbi))) {
1463                                 f2fs_put_page(last_page, 0);
1464                                 pagevec_release(&pvec);
1465                                 return ERR_PTR(-EIO);
1466                         }
1467
1468                         if (!IS_DNODE(page) || !is_cold_node(page))
1469                                 continue;
1470                         if (ino_of_node(page) != ino)
1471                                 continue;
1472
1473                         lock_page(page);
1474
1475                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1476 continue_unlock:
1477                                 unlock_page(page);
1478                                 continue;
1479                         }
1480                         if (ino_of_node(page) != ino)
1481                                 goto continue_unlock;
1482
1483                         if (!PageDirty(page)) {
1484                                 /* someone wrote it for us */
1485                                 goto continue_unlock;
1486                         }
1487
1488                         if (last_page)
1489                                 f2fs_put_page(last_page, 0);
1490
1491                         get_page(page);
1492                         last_page = page;
1493                         unlock_page(page);
1494                 }
1495                 pagevec_release(&pvec);
1496                 cond_resched();
1497         }
1498         return last_page;
1499 }
1500
1501 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1502                                 struct writeback_control *wbc, bool do_balance,
1503                                 enum iostat_type io_type, unsigned int *seq_id)
1504 {
1505         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1506         nid_t nid;
1507         struct node_info ni;
1508         struct f2fs_io_info fio = {
1509                 .sbi = sbi,
1510                 .ino = ino_of_node(page),
1511                 .type = NODE,
1512                 .op = REQ_OP_WRITE,
1513                 .op_flags = wbc_to_write_flags(wbc),
1514                 .page = page,
1515                 .encrypted_page = NULL,
1516                 .submitted = false,
1517                 .io_type = io_type,
1518                 .io_wbc = wbc,
1519         };
1520         unsigned int seq;
1521
1522         trace_f2fs_writepage(page, NODE);
1523
1524         if (unlikely(f2fs_cp_error(sbi)))
1525                 goto redirty_out;
1526
1527         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1528                 goto redirty_out;
1529
1530         if (wbc->sync_mode == WB_SYNC_NONE &&
1531                         IS_DNODE(page) && is_cold_node(page))
1532                 goto redirty_out;
1533
1534         /* get old block addr of this node page */
1535         nid = nid_of_node(page);
1536         f2fs_bug_on(sbi, page->index != nid);
1537
1538         if (f2fs_get_node_info(sbi, nid, &ni))
1539                 goto redirty_out;
1540
1541         if (wbc->for_reclaim) {
1542                 if (!down_read_trylock(&sbi->node_write))
1543                         goto redirty_out;
1544         } else {
1545                 down_read(&sbi->node_write);
1546         }
1547
1548         /* This page is already truncated */
1549         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1550                 ClearPageUptodate(page);
1551                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1552                 up_read(&sbi->node_write);
1553                 unlock_page(page);
1554                 return 0;
1555         }
1556
1557         if (__is_valid_data_blkaddr(ni.blk_addr) &&
1558                 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC)) {
1559                 up_read(&sbi->node_write);
1560                 goto redirty_out;
1561         }
1562
1563         if (atomic && !test_opt(sbi, NOBARRIER))
1564                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1565
1566         /* should add to global list before clearing PAGECACHE status */
1567         if (f2fs_in_warm_node_list(sbi, page)) {
1568                 seq = f2fs_add_fsync_node_entry(sbi, page);
1569                 if (seq_id)
1570                         *seq_id = seq;
1571         }
1572
1573         set_page_writeback(page);
1574         ClearPageError(page);
1575
1576         fio.old_blkaddr = ni.blk_addr;
1577         f2fs_do_write_node_page(nid, &fio);
1578         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1579         dec_page_count(sbi, F2FS_DIRTY_NODES);
1580         up_read(&sbi->node_write);
1581
1582         if (wbc->for_reclaim) {
1583                 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1584                                                 page->index, NODE);
1585                 submitted = NULL;
1586         }
1587
1588         unlock_page(page);
1589
1590         if (unlikely(f2fs_cp_error(sbi))) {
1591                 f2fs_submit_merged_write(sbi, NODE);
1592                 submitted = NULL;
1593         }
1594         if (submitted)
1595                 *submitted = fio.submitted;
1596
1597         if (do_balance)
1598                 f2fs_balance_fs(sbi, false);
1599         return 0;
1600
1601 redirty_out:
1602         redirty_page_for_writepage(wbc, page);
1603         return AOP_WRITEPAGE_ACTIVATE;
1604 }
1605
1606 void f2fs_move_node_page(struct page *node_page, int gc_type)
1607 {
1608         if (gc_type == FG_GC) {
1609                 struct writeback_control wbc = {
1610                         .sync_mode = WB_SYNC_ALL,
1611                         .nr_to_write = 1,
1612                         .for_reclaim = 0,
1613                 };
1614
1615                 set_page_dirty(node_page);
1616                 f2fs_wait_on_page_writeback(node_page, NODE, true);
1617
1618                 f2fs_bug_on(F2FS_P_SB(node_page), PageWriteback(node_page));
1619                 if (!clear_page_dirty_for_io(node_page))
1620                         goto out_page;
1621
1622                 if (__write_node_page(node_page, false, NULL,
1623                                         &wbc, false, FS_GC_NODE_IO, NULL))
1624                         unlock_page(node_page);
1625                 goto release_page;
1626         } else {
1627                 /* set page dirty and write it */
1628                 if (!PageWriteback(node_page))
1629                         set_page_dirty(node_page);
1630         }
1631 out_page:
1632         unlock_page(node_page);
1633 release_page:
1634         f2fs_put_page(node_page, 0);
1635 }
1636
1637 static int f2fs_write_node_page(struct page *page,
1638                                 struct writeback_control *wbc)
1639 {
1640         return __write_node_page(page, false, NULL, wbc, false,
1641                                                 FS_NODE_IO, NULL);
1642 }
1643
1644 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1645                         struct writeback_control *wbc, bool atomic,
1646                         unsigned int *seq_id)
1647 {
1648         pgoff_t index;
1649         pgoff_t last_idx = ULONG_MAX;
1650         struct pagevec pvec;
1651         int ret = 0;
1652         struct page *last_page = NULL;
1653         bool marked = false;
1654         nid_t ino = inode->i_ino;
1655         int nr_pages;
1656
1657         if (atomic) {
1658                 last_page = last_fsync_dnode(sbi, ino);
1659                 if (IS_ERR_OR_NULL(last_page))
1660                         return PTR_ERR_OR_ZERO(last_page);
1661         }
1662 retry:
1663         pagevec_init(&pvec);
1664         index = 0;
1665
1666         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1667                                 PAGECACHE_TAG_DIRTY))) {
1668                 int i;
1669
1670                 for (i = 0; i < nr_pages; i++) {
1671                         struct page *page = pvec.pages[i];
1672                         bool submitted = false;
1673
1674                         if (unlikely(f2fs_cp_error(sbi))) {
1675                                 f2fs_put_page(last_page, 0);
1676                                 pagevec_release(&pvec);
1677                                 ret = -EIO;
1678                                 goto out;
1679                         }
1680
1681                         if (!IS_DNODE(page) || !is_cold_node(page))
1682                                 continue;
1683                         if (ino_of_node(page) != ino)
1684                                 continue;
1685
1686                         lock_page(page);
1687
1688                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1689 continue_unlock:
1690                                 unlock_page(page);
1691                                 continue;
1692                         }
1693                         if (ino_of_node(page) != ino)
1694                                 goto continue_unlock;
1695
1696                         if (!PageDirty(page) && page != last_page) {
1697                                 /* someone wrote it for us */
1698                                 goto continue_unlock;
1699                         }
1700
1701                         f2fs_wait_on_page_writeback(page, NODE, true);
1702                         BUG_ON(PageWriteback(page));
1703
1704                         set_fsync_mark(page, 0);
1705                         set_dentry_mark(page, 0);
1706
1707                         if (!atomic || page == last_page) {
1708                                 set_fsync_mark(page, 1);
1709                                 if (IS_INODE(page)) {
1710                                         if (is_inode_flag_set(inode,
1711                                                                 FI_DIRTY_INODE))
1712                                                 f2fs_update_inode(inode, page);
1713                                         set_dentry_mark(page,
1714                                                 f2fs_need_dentry_mark(sbi, ino));
1715                                 }
1716                                 /*  may be written by other thread */
1717                                 if (!PageDirty(page))
1718                                         set_page_dirty(page);
1719                         }
1720
1721                         if (!clear_page_dirty_for_io(page))
1722                                 goto continue_unlock;
1723
1724                         ret = __write_node_page(page, atomic &&
1725                                                 page == last_page,
1726                                                 &submitted, wbc, true,
1727                                                 FS_NODE_IO, seq_id);
1728                         if (ret) {
1729                                 unlock_page(page);
1730                                 f2fs_put_page(last_page, 0);
1731                                 break;
1732                         } else if (submitted) {
1733                                 last_idx = page->index;
1734                         }
1735
1736                         if (page == last_page) {
1737                                 f2fs_put_page(page, 0);
1738                                 marked = true;
1739                                 break;
1740                         }
1741                 }
1742                 pagevec_release(&pvec);
1743                 cond_resched();
1744
1745                 if (ret || marked)
1746                         break;
1747         }
1748         if (!ret && atomic && !marked) {
1749                 f2fs_msg(sbi->sb, KERN_DEBUG,
1750                         "Retry to write fsync mark: ino=%u, idx=%lx",
1751                                         ino, last_page->index);
1752                 lock_page(last_page);
1753                 f2fs_wait_on_page_writeback(last_page, NODE, true);
1754                 set_page_dirty(last_page);
1755                 unlock_page(last_page);
1756                 goto retry;
1757         }
1758 out:
1759         if (last_idx != ULONG_MAX)
1760                 f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1761         return ret ? -EIO: 0;
1762 }
1763
1764 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1765                                 struct writeback_control *wbc,
1766                                 bool do_balance, enum iostat_type io_type)
1767 {
1768         pgoff_t index;
1769         struct pagevec pvec;
1770         int step = 0;
1771         int nwritten = 0;
1772         int ret = 0;
1773         int nr_pages, done = 0;
1774
1775         pagevec_init(&pvec);
1776
1777 next_step:
1778         index = 0;
1779
1780         while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1781                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1782                 int i;
1783
1784                 for (i = 0; i < nr_pages; i++) {
1785                         struct page *page = pvec.pages[i];
1786                         bool submitted = false;
1787
1788                         /* give a priority to WB_SYNC threads */
1789                         if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1790                                         wbc->sync_mode == WB_SYNC_NONE) {
1791                                 done = 1;
1792                                 break;
1793                         }
1794
1795                         /*
1796                          * flushing sequence with step:
1797                          * 0. indirect nodes
1798                          * 1. dentry dnodes
1799                          * 2. file dnodes
1800                          */
1801                         if (step == 0 && IS_DNODE(page))
1802                                 continue;
1803                         if (step == 1 && (!IS_DNODE(page) ||
1804                                                 is_cold_node(page)))
1805                                 continue;
1806                         if (step == 2 && (!IS_DNODE(page) ||
1807                                                 !is_cold_node(page)))
1808                                 continue;
1809 lock_node:
1810                         if (wbc->sync_mode == WB_SYNC_ALL)
1811                                 lock_page(page);
1812                         else if (!trylock_page(page))
1813                                 continue;
1814
1815                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1816 continue_unlock:
1817                                 unlock_page(page);
1818                                 continue;
1819                         }
1820
1821                         if (!PageDirty(page)) {
1822                                 /* someone wrote it for us */
1823                                 goto continue_unlock;
1824                         }
1825
1826                         /* flush inline_data */
1827                         if (is_inline_node(page)) {
1828                                 clear_inline_node(page);
1829                                 unlock_page(page);
1830                                 flush_inline_data(sbi, ino_of_node(page));
1831                                 goto lock_node;
1832                         }
1833
1834                         f2fs_wait_on_page_writeback(page, NODE, true);
1835
1836                         BUG_ON(PageWriteback(page));
1837                         if (!clear_page_dirty_for_io(page))
1838                                 goto continue_unlock;
1839
1840                         set_fsync_mark(page, 0);
1841                         set_dentry_mark(page, 0);
1842
1843                         ret = __write_node_page(page, false, &submitted,
1844                                                 wbc, do_balance, io_type, NULL);
1845                         if (ret)
1846                                 unlock_page(page);
1847                         else if (submitted)
1848                                 nwritten++;
1849
1850                         if (--wbc->nr_to_write == 0)
1851                                 break;
1852                 }
1853                 pagevec_release(&pvec);
1854                 cond_resched();
1855
1856                 if (wbc->nr_to_write == 0) {
1857                         step = 2;
1858                         break;
1859                 }
1860         }
1861
1862         if (step < 2) {
1863                 if (wbc->sync_mode == WB_SYNC_NONE && step == 1)
1864                         goto out;
1865                 step++;
1866                 goto next_step;
1867         }
1868 out:
1869         if (nwritten)
1870                 f2fs_submit_merged_write(sbi, NODE);
1871
1872         if (unlikely(f2fs_cp_error(sbi)))
1873                 return -EIO;
1874         return ret;
1875 }
1876
1877 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1878                                                 unsigned int seq_id)
1879 {
1880         struct fsync_node_entry *fn;
1881         struct page *page;
1882         struct list_head *head = &sbi->fsync_node_list;
1883         unsigned long flags;
1884         unsigned int cur_seq_id = 0;
1885         int ret2, ret = 0;
1886
1887         while (seq_id && cur_seq_id < seq_id) {
1888                 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
1889                 if (list_empty(head)) {
1890                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1891                         break;
1892                 }
1893                 fn = list_first_entry(head, struct fsync_node_entry, list);
1894                 if (fn->seq_id > seq_id) {
1895                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1896                         break;
1897                 }
1898                 cur_seq_id = fn->seq_id;
1899                 page = fn->page;
1900                 get_page(page);
1901                 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1902
1903                 f2fs_wait_on_page_writeback(page, NODE, true);
1904                 if (TestClearPageError(page))
1905                         ret = -EIO;
1906
1907                 put_page(page);
1908
1909                 if (ret)
1910                         break;
1911         }
1912
1913         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1914         if (!ret)
1915                 ret = ret2;
1916
1917         return ret;
1918 }
1919
1920 static int f2fs_write_node_pages(struct address_space *mapping,
1921                             struct writeback_control *wbc)
1922 {
1923         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1924         struct blk_plug plug;
1925         long diff;
1926
1927         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1928                 goto skip_write;
1929
1930         /* balancing f2fs's metadata in background */
1931         f2fs_balance_fs_bg(sbi);
1932
1933         /* collect a number of dirty node pages and write together */
1934         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1935                 goto skip_write;
1936
1937         if (wbc->sync_mode == WB_SYNC_ALL)
1938                 atomic_inc(&sbi->wb_sync_req[NODE]);
1939         else if (atomic_read(&sbi->wb_sync_req[NODE]))
1940                 goto skip_write;
1941
1942         trace_f2fs_writepages(mapping->host, wbc, NODE);
1943
1944         diff = nr_pages_to_write(sbi, NODE, wbc);
1945         blk_start_plug(&plug);
1946         f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1947         blk_finish_plug(&plug);
1948         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1949
1950         if (wbc->sync_mode == WB_SYNC_ALL)
1951                 atomic_dec(&sbi->wb_sync_req[NODE]);
1952         return 0;
1953
1954 skip_write:
1955         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1956         trace_f2fs_writepages(mapping->host, wbc, NODE);
1957         return 0;
1958 }
1959
1960 static int f2fs_set_node_page_dirty(struct page *page)
1961 {
1962         trace_f2fs_set_page_dirty(page, NODE);
1963
1964         if (!PageUptodate(page))
1965                 SetPageUptodate(page);
1966 #ifdef CONFIG_F2FS_CHECK_FS
1967         if (IS_INODE(page))
1968                 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
1969 #endif
1970         if (!PageDirty(page)) {
1971                 __set_page_dirty_nobuffers(page);
1972                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1973                 SetPagePrivate(page);
1974                 f2fs_trace_pid(page);
1975                 return 1;
1976         }
1977         return 0;
1978 }
1979
1980 /*
1981  * Structure of the f2fs node operations
1982  */
1983 const struct address_space_operations f2fs_node_aops = {
1984         .writepage      = f2fs_write_node_page,
1985         .writepages     = f2fs_write_node_pages,
1986         .set_page_dirty = f2fs_set_node_page_dirty,
1987         .invalidatepage = f2fs_invalidate_page,
1988         .releasepage    = f2fs_release_page,
1989 #ifdef CONFIG_MIGRATION
1990         .migratepage    = f2fs_migrate_page,
1991 #endif
1992 };
1993
1994 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1995                                                 nid_t n)
1996 {
1997         return radix_tree_lookup(&nm_i->free_nid_root, n);
1998 }
1999
2000 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2001                         struct free_nid *i, enum nid_state state)
2002 {
2003         struct f2fs_nm_info *nm_i = NM_I(sbi);
2004
2005         int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2006         if (err)
2007                 return err;
2008
2009         f2fs_bug_on(sbi, state != i->state);
2010         nm_i->nid_cnt[state]++;
2011         if (state == FREE_NID)
2012                 list_add_tail(&i->list, &nm_i->free_nid_list);
2013         return 0;
2014 }
2015
2016 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2017                         struct free_nid *i, enum nid_state state)
2018 {
2019         struct f2fs_nm_info *nm_i = NM_I(sbi);
2020
2021         f2fs_bug_on(sbi, state != i->state);
2022         nm_i->nid_cnt[state]--;
2023         if (state == FREE_NID)
2024                 list_del(&i->list);
2025         radix_tree_delete(&nm_i->free_nid_root, i->nid);
2026 }
2027
2028 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2029                         enum nid_state org_state, enum nid_state dst_state)
2030 {
2031         struct f2fs_nm_info *nm_i = NM_I(sbi);
2032
2033         f2fs_bug_on(sbi, org_state != i->state);
2034         i->state = dst_state;
2035         nm_i->nid_cnt[org_state]--;
2036         nm_i->nid_cnt[dst_state]++;
2037
2038         switch (dst_state) {
2039         case PREALLOC_NID:
2040                 list_del(&i->list);
2041                 break;
2042         case FREE_NID:
2043                 list_add_tail(&i->list, &nm_i->free_nid_list);
2044                 break;
2045         default:
2046                 BUG_ON(1);
2047         }
2048 }
2049
2050 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2051                                                         bool set, bool build)
2052 {
2053         struct f2fs_nm_info *nm_i = NM_I(sbi);
2054         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2055         unsigned int nid_ofs = nid - START_NID(nid);
2056
2057         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2058                 return;
2059
2060         if (set) {
2061                 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2062                         return;
2063                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2064                 nm_i->free_nid_count[nat_ofs]++;
2065         } else {
2066                 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2067                         return;
2068                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2069                 if (!build)
2070                         nm_i->free_nid_count[nat_ofs]--;
2071         }
2072 }
2073
2074 /* return if the nid is recognized as free */
2075 static bool add_free_nid(struct f2fs_sb_info *sbi,
2076                                 nid_t nid, bool build, bool update)
2077 {
2078         struct f2fs_nm_info *nm_i = NM_I(sbi);
2079         struct free_nid *i, *e;
2080         struct nat_entry *ne;
2081         int err = -EINVAL;
2082         bool ret = false;
2083
2084         /* 0 nid should not be used */
2085         if (unlikely(nid == 0))
2086                 return false;
2087
2088         if (unlikely(f2fs_check_nid_range(sbi, nid)))
2089                 return false;
2090
2091         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2092         i->nid = nid;
2093         i->state = FREE_NID;
2094
2095         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2096
2097         spin_lock(&nm_i->nid_list_lock);
2098
2099         if (build) {
2100                 /*
2101                  *   Thread A             Thread B
2102                  *  - f2fs_create
2103                  *   - f2fs_new_inode
2104                  *    - f2fs_alloc_nid
2105                  *     - __insert_nid_to_list(PREALLOC_NID)
2106                  *                     - f2fs_balance_fs_bg
2107                  *                      - f2fs_build_free_nids
2108                  *                       - __f2fs_build_free_nids
2109                  *                        - scan_nat_page
2110                  *                         - add_free_nid
2111                  *                          - __lookup_nat_cache
2112                  *  - f2fs_add_link
2113                  *   - f2fs_init_inode_metadata
2114                  *    - f2fs_new_inode_page
2115                  *     - f2fs_new_node_page
2116                  *      - set_node_addr
2117                  *  - f2fs_alloc_nid_done
2118                  *   - __remove_nid_from_list(PREALLOC_NID)
2119                  *                         - __insert_nid_to_list(FREE_NID)
2120                  */
2121                 ne = __lookup_nat_cache(nm_i, nid);
2122                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2123                                 nat_get_blkaddr(ne) != NULL_ADDR))
2124                         goto err_out;
2125
2126                 e = __lookup_free_nid_list(nm_i, nid);
2127                 if (e) {
2128                         if (e->state == FREE_NID)
2129                                 ret = true;
2130                         goto err_out;
2131                 }
2132         }
2133         ret = true;
2134         err = __insert_free_nid(sbi, i, FREE_NID);
2135 err_out:
2136         if (update) {
2137                 update_free_nid_bitmap(sbi, nid, ret, build);
2138                 if (!build)
2139                         nm_i->available_nids++;
2140         }
2141         spin_unlock(&nm_i->nid_list_lock);
2142         radix_tree_preload_end();
2143
2144         if (err)
2145                 kmem_cache_free(free_nid_slab, i);
2146         return ret;
2147 }
2148
2149 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2150 {
2151         struct f2fs_nm_info *nm_i = NM_I(sbi);
2152         struct free_nid *i;
2153         bool need_free = false;
2154
2155         spin_lock(&nm_i->nid_list_lock);
2156         i = __lookup_free_nid_list(nm_i, nid);
2157         if (i && i->state == FREE_NID) {
2158                 __remove_free_nid(sbi, i, FREE_NID);
2159                 need_free = true;
2160         }
2161         spin_unlock(&nm_i->nid_list_lock);
2162
2163         if (need_free)
2164                 kmem_cache_free(free_nid_slab, i);
2165 }
2166
2167 static int scan_nat_page(struct f2fs_sb_info *sbi,
2168                         struct page *nat_page, nid_t start_nid)
2169 {
2170         struct f2fs_nm_info *nm_i = NM_I(sbi);
2171         struct f2fs_nat_block *nat_blk = page_address(nat_page);
2172         block_t blk_addr;
2173         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2174         int i;
2175
2176         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2177
2178         i = start_nid % NAT_ENTRY_PER_BLOCK;
2179
2180         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2181                 if (unlikely(start_nid >= nm_i->max_nid))
2182                         break;
2183
2184                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2185
2186                 if (blk_addr == NEW_ADDR)
2187                         return -EINVAL;
2188
2189                 if (blk_addr == NULL_ADDR) {
2190                         add_free_nid(sbi, start_nid, true, true);
2191                 } else {
2192                         spin_lock(&NM_I(sbi)->nid_list_lock);
2193                         update_free_nid_bitmap(sbi, start_nid, false, true);
2194                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2195                 }
2196         }
2197
2198         return 0;
2199 }
2200
2201 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2202 {
2203         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2204         struct f2fs_journal *journal = curseg->journal;
2205         int i;
2206
2207         down_read(&curseg->journal_rwsem);
2208         for (i = 0; i < nats_in_cursum(journal); i++) {
2209                 block_t addr;
2210                 nid_t nid;
2211
2212                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2213                 nid = le32_to_cpu(nid_in_journal(journal, i));
2214                 if (addr == NULL_ADDR)
2215                         add_free_nid(sbi, nid, true, false);
2216                 else
2217                         remove_free_nid(sbi, nid);
2218         }
2219         up_read(&curseg->journal_rwsem);
2220 }
2221
2222 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2223 {
2224         struct f2fs_nm_info *nm_i = NM_I(sbi);
2225         unsigned int i, idx;
2226         nid_t nid;
2227
2228         down_read(&nm_i->nat_tree_lock);
2229
2230         for (i = 0; i < nm_i->nat_blocks; i++) {
2231                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2232                         continue;
2233                 if (!nm_i->free_nid_count[i])
2234                         continue;
2235                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2236                         idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2237                                                 NAT_ENTRY_PER_BLOCK, idx);
2238                         if (idx >= NAT_ENTRY_PER_BLOCK)
2239                                 break;
2240
2241                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
2242                         add_free_nid(sbi, nid, true, false);
2243
2244                         if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2245                                 goto out;
2246                 }
2247         }
2248 out:
2249         scan_curseg_cache(sbi);
2250
2251         up_read(&nm_i->nat_tree_lock);
2252 }
2253
2254 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2255                                                 bool sync, bool mount)
2256 {
2257         struct f2fs_nm_info *nm_i = NM_I(sbi);
2258         int i = 0, ret;
2259         nid_t nid = nm_i->next_scan_nid;
2260
2261         if (unlikely(nid >= nm_i->max_nid))
2262                 nid = 0;
2263
2264         if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2265                 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2266
2267         /* Enough entries */
2268         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2269                 return 0;
2270
2271         if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2272                 return 0;
2273
2274         if (!mount) {
2275                 /* try to find free nids in free_nid_bitmap */
2276                 scan_free_nid_bits(sbi);
2277
2278                 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2279                         return 0;
2280         }
2281
2282         /* readahead nat pages to be scanned */
2283         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2284                                                         META_NAT, true);
2285
2286         down_read(&nm_i->nat_tree_lock);
2287
2288         while (1) {
2289                 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2290                                                 nm_i->nat_block_bitmap)) {
2291                         struct page *page = get_current_nat_page(sbi, nid);
2292
2293                         ret = scan_nat_page(sbi, page, nid);
2294                         f2fs_put_page(page, 1);
2295
2296                         if (ret) {
2297                                 up_read(&nm_i->nat_tree_lock);
2298                                 f2fs_bug_on(sbi, !mount);
2299                                 f2fs_msg(sbi->sb, KERN_ERR,
2300                                         "NAT is corrupt, run fsck to fix it");
2301                                 return -EINVAL;
2302                         }
2303                 }
2304
2305                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2306                 if (unlikely(nid >= nm_i->max_nid))
2307                         nid = 0;
2308
2309                 if (++i >= FREE_NID_PAGES)
2310                         break;
2311         }
2312
2313         /* go to the next free nat pages to find free nids abundantly */
2314         nm_i->next_scan_nid = nid;
2315
2316         /* find free nids from current sum_pages */
2317         scan_curseg_cache(sbi);
2318
2319         up_read(&nm_i->nat_tree_lock);
2320
2321         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2322                                         nm_i->ra_nid_pages, META_NAT, false);
2323
2324         return 0;
2325 }
2326
2327 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2328 {
2329         int ret;
2330
2331         mutex_lock(&NM_I(sbi)->build_lock);
2332         ret = __f2fs_build_free_nids(sbi, sync, mount);
2333         mutex_unlock(&NM_I(sbi)->build_lock);
2334
2335         return ret;
2336 }
2337
2338 /*
2339  * If this function returns success, caller can obtain a new nid
2340  * from second parameter of this function.
2341  * The returned nid could be used ino as well as nid when inode is created.
2342  */
2343 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2344 {
2345         struct f2fs_nm_info *nm_i = NM_I(sbi);
2346         struct free_nid *i = NULL;
2347 retry:
2348         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2349                 f2fs_show_injection_info(FAULT_ALLOC_NID);
2350                 return false;
2351         }
2352
2353         spin_lock(&nm_i->nid_list_lock);
2354
2355         if (unlikely(nm_i->available_nids == 0)) {
2356                 spin_unlock(&nm_i->nid_list_lock);
2357                 return false;
2358         }
2359
2360         /* We should not use stale free nids created by f2fs_build_free_nids */
2361         if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2362                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2363                 i = list_first_entry(&nm_i->free_nid_list,
2364                                         struct free_nid, list);
2365                 *nid = i->nid;
2366
2367                 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2368                 nm_i->available_nids--;
2369
2370                 update_free_nid_bitmap(sbi, *nid, false, false);
2371
2372                 spin_unlock(&nm_i->nid_list_lock);
2373                 return true;
2374         }
2375         spin_unlock(&nm_i->nid_list_lock);
2376
2377         /* Let's scan nat pages and its caches to get free nids */
2378         if (!f2fs_build_free_nids(sbi, true, false))
2379                 goto retry;
2380         return false;
2381 }
2382
2383 /*
2384  * f2fs_alloc_nid() should be called prior to this function.
2385  */
2386 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2387 {
2388         struct f2fs_nm_info *nm_i = NM_I(sbi);
2389         struct free_nid *i;
2390
2391         spin_lock(&nm_i->nid_list_lock);
2392         i = __lookup_free_nid_list(nm_i, nid);
2393         f2fs_bug_on(sbi, !i);
2394         __remove_free_nid(sbi, i, PREALLOC_NID);
2395         spin_unlock(&nm_i->nid_list_lock);
2396
2397         kmem_cache_free(free_nid_slab, i);
2398 }
2399
2400 /*
2401  * f2fs_alloc_nid() should be called prior to this function.
2402  */
2403 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2404 {
2405         struct f2fs_nm_info *nm_i = NM_I(sbi);
2406         struct free_nid *i;
2407         bool need_free = false;
2408
2409         if (!nid)
2410                 return;
2411
2412         spin_lock(&nm_i->nid_list_lock);
2413         i = __lookup_free_nid_list(nm_i, nid);
2414         f2fs_bug_on(sbi, !i);
2415
2416         if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2417                 __remove_free_nid(sbi, i, PREALLOC_NID);
2418                 need_free = true;
2419         } else {
2420                 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2421         }
2422
2423         nm_i->available_nids++;
2424
2425         update_free_nid_bitmap(sbi, nid, true, false);
2426
2427         spin_unlock(&nm_i->nid_list_lock);
2428
2429         if (need_free)
2430                 kmem_cache_free(free_nid_slab, i);
2431 }
2432
2433 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2434 {
2435         struct f2fs_nm_info *nm_i = NM_I(sbi);
2436         struct free_nid *i, *next;
2437         int nr = nr_shrink;
2438
2439         if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2440                 return 0;
2441
2442         if (!mutex_trylock(&nm_i->build_lock))
2443                 return 0;
2444
2445         spin_lock(&nm_i->nid_list_lock);
2446         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2447                 if (nr_shrink <= 0 ||
2448                                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2449                         break;
2450
2451                 __remove_free_nid(sbi, i, FREE_NID);
2452                 kmem_cache_free(free_nid_slab, i);
2453                 nr_shrink--;
2454         }
2455         spin_unlock(&nm_i->nid_list_lock);
2456         mutex_unlock(&nm_i->build_lock);
2457
2458         return nr - nr_shrink;
2459 }
2460
2461 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2462 {
2463         void *src_addr, *dst_addr;
2464         size_t inline_size;
2465         struct page *ipage;
2466         struct f2fs_inode *ri;
2467
2468         ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2469         if (IS_ERR(ipage))
2470                 return PTR_ERR(ipage);
2471
2472         ri = F2FS_INODE(page);
2473         if (ri->i_inline & F2FS_INLINE_XATTR) {
2474                 set_inode_flag(inode, FI_INLINE_XATTR);
2475         } else {
2476                 clear_inode_flag(inode, FI_INLINE_XATTR);
2477                 goto update_inode;
2478         }
2479
2480         dst_addr = inline_xattr_addr(inode, ipage);
2481         src_addr = inline_xattr_addr(inode, page);
2482         inline_size = inline_xattr_size(inode);
2483
2484         f2fs_wait_on_page_writeback(ipage, NODE, true);
2485         memcpy(dst_addr, src_addr, inline_size);
2486 update_inode:
2487         f2fs_update_inode(inode, ipage);
2488         f2fs_put_page(ipage, 1);
2489         return 0;
2490 }
2491
2492 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2493 {
2494         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2495         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2496         nid_t new_xnid;
2497         struct dnode_of_data dn;
2498         struct node_info ni;
2499         struct page *xpage;
2500         int err;
2501
2502         if (!prev_xnid)
2503                 goto recover_xnid;
2504
2505         /* 1: invalidate the previous xattr nid */
2506         err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2507         if (err)
2508                 return err;
2509
2510         f2fs_invalidate_blocks(sbi, ni.blk_addr);
2511         dec_valid_node_count(sbi, inode, false);
2512         set_node_addr(sbi, &ni, NULL_ADDR, false);
2513
2514 recover_xnid:
2515         /* 2: update xattr nid in inode */
2516         if (!f2fs_alloc_nid(sbi, &new_xnid))
2517                 return -ENOSPC;
2518
2519         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2520         xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2521         if (IS_ERR(xpage)) {
2522                 f2fs_alloc_nid_failed(sbi, new_xnid);
2523                 return PTR_ERR(xpage);
2524         }
2525
2526         f2fs_alloc_nid_done(sbi, new_xnid);
2527         f2fs_update_inode_page(inode);
2528
2529         /* 3: update and set xattr node page dirty */
2530         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2531
2532         set_page_dirty(xpage);
2533         f2fs_put_page(xpage, 1);
2534
2535         return 0;
2536 }
2537
2538 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2539 {
2540         struct f2fs_inode *src, *dst;
2541         nid_t ino = ino_of_node(page);
2542         struct node_info old_ni, new_ni;
2543         struct page *ipage;
2544         int err;
2545
2546         err = f2fs_get_node_info(sbi, ino, &old_ni);
2547         if (err)
2548                 return err;
2549
2550         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2551                 return -EINVAL;
2552 retry:
2553         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2554         if (!ipage) {
2555                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2556                 goto retry;
2557         }
2558
2559         /* Should not use this inode from free nid list */
2560         remove_free_nid(sbi, ino);
2561
2562         if (!PageUptodate(ipage))
2563                 SetPageUptodate(ipage);
2564         fill_node_footer(ipage, ino, ino, 0, true);
2565         set_cold_node(ipage, false);
2566
2567         src = F2FS_INODE(page);
2568         dst = F2FS_INODE(ipage);
2569
2570         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2571         dst->i_size = 0;
2572         dst->i_blocks = cpu_to_le64(1);
2573         dst->i_links = cpu_to_le32(1);
2574         dst->i_xattr_nid = 0;
2575         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2576         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2577                 dst->i_extra_isize = src->i_extra_isize;
2578
2579                 if (f2fs_sb_has_flexible_inline_xattr(sbi->sb) &&
2580                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2581                                                         i_inline_xattr_size))
2582                         dst->i_inline_xattr_size = src->i_inline_xattr_size;
2583
2584                 if (f2fs_sb_has_project_quota(sbi->sb) &&
2585                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2586                                                                 i_projid))
2587                         dst->i_projid = src->i_projid;
2588
2589                 if (f2fs_sb_has_inode_crtime(sbi->sb) &&
2590                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2591                                                         i_crtime_nsec)) {
2592                         dst->i_crtime = src->i_crtime;
2593                         dst->i_crtime_nsec = src->i_crtime_nsec;
2594                 }
2595         }
2596
2597         new_ni = old_ni;
2598         new_ni.ino = ino;
2599
2600         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2601                 WARN_ON(1);
2602         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2603         inc_valid_inode_count(sbi);
2604         set_page_dirty(ipage);
2605         f2fs_put_page(ipage, 1);
2606         return 0;
2607 }
2608
2609 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2610                         unsigned int segno, struct f2fs_summary_block *sum)
2611 {
2612         struct f2fs_node *rn;
2613         struct f2fs_summary *sum_entry;
2614         block_t addr;
2615         int i, idx, last_offset, nrpages;
2616
2617         /* scan the node segment */
2618         last_offset = sbi->blocks_per_seg;
2619         addr = START_BLOCK(sbi, segno);
2620         sum_entry = &sum->entries[0];
2621
2622         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2623                 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2624
2625                 /* readahead node pages */
2626                 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2627
2628                 for (idx = addr; idx < addr + nrpages; idx++) {
2629                         struct page *page = f2fs_get_tmp_page(sbi, idx);
2630
2631                         if (IS_ERR(page))
2632                                 return PTR_ERR(page);
2633
2634                         rn = F2FS_NODE(page);
2635                         sum_entry->nid = rn->footer.nid;
2636                         sum_entry->version = 0;
2637                         sum_entry->ofs_in_node = 0;
2638                         sum_entry++;
2639                         f2fs_put_page(page, 1);
2640                 }
2641
2642                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2643                                                         addr + nrpages);
2644         }
2645         return 0;
2646 }
2647
2648 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2649 {
2650         struct f2fs_nm_info *nm_i = NM_I(sbi);
2651         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2652         struct f2fs_journal *journal = curseg->journal;
2653         int i;
2654
2655         down_write(&curseg->journal_rwsem);
2656         for (i = 0; i < nats_in_cursum(journal); i++) {
2657                 struct nat_entry *ne;
2658                 struct f2fs_nat_entry raw_ne;
2659                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2660
2661                 if (f2fs_check_nid_range(sbi, nid))
2662                         continue;
2663
2664                 raw_ne = nat_in_journal(journal, i);
2665
2666                 ne = __lookup_nat_cache(nm_i, nid);
2667                 if (!ne) {
2668                         ne = __alloc_nat_entry(nid, true);
2669                         __init_nat_entry(nm_i, ne, &raw_ne, true);
2670                 }
2671
2672                 /*
2673                  * if a free nat in journal has not been used after last
2674                  * checkpoint, we should remove it from available nids,
2675                  * since later we will add it again.
2676                  */
2677                 if (!get_nat_flag(ne, IS_DIRTY) &&
2678                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2679                         spin_lock(&nm_i->nid_list_lock);
2680                         nm_i->available_nids--;
2681                         spin_unlock(&nm_i->nid_list_lock);
2682                 }
2683
2684                 __set_nat_cache_dirty(nm_i, ne);
2685         }
2686         update_nats_in_cursum(journal, -i);
2687         up_write(&curseg->journal_rwsem);
2688 }
2689
2690 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2691                                                 struct list_head *head, int max)
2692 {
2693         struct nat_entry_set *cur;
2694
2695         if (nes->entry_cnt >= max)
2696                 goto add_out;
2697
2698         list_for_each_entry(cur, head, set_list) {
2699                 if (cur->entry_cnt >= nes->entry_cnt) {
2700                         list_add(&nes->set_list, cur->set_list.prev);
2701                         return;
2702                 }
2703         }
2704 add_out:
2705         list_add_tail(&nes->set_list, head);
2706 }
2707
2708 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2709                                                 struct page *page)
2710 {
2711         struct f2fs_nm_info *nm_i = NM_I(sbi);
2712         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2713         struct f2fs_nat_block *nat_blk = page_address(page);
2714         int valid = 0;
2715         int i = 0;
2716
2717         if (!enabled_nat_bits(sbi, NULL))
2718                 return;
2719
2720         if (nat_index == 0) {
2721                 valid = 1;
2722                 i = 1;
2723         }
2724         for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2725                 if (nat_blk->entries[i].block_addr != NULL_ADDR)
2726                         valid++;
2727         }
2728         if (valid == 0) {
2729                 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2730                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2731                 return;
2732         }
2733
2734         __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2735         if (valid == NAT_ENTRY_PER_BLOCK)
2736                 __set_bit_le(nat_index, nm_i->full_nat_bits);
2737         else
2738                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2739 }
2740
2741 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2742                 struct nat_entry_set *set, struct cp_control *cpc)
2743 {
2744         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2745         struct f2fs_journal *journal = curseg->journal;
2746         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2747         bool to_journal = true;
2748         struct f2fs_nat_block *nat_blk;
2749         struct nat_entry *ne, *cur;
2750         struct page *page = NULL;
2751
2752         /*
2753          * there are two steps to flush nat entries:
2754          * #1, flush nat entries to journal in current hot data summary block.
2755          * #2, flush nat entries to nat page.
2756          */
2757         if (enabled_nat_bits(sbi, cpc) ||
2758                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2759                 to_journal = false;
2760
2761         if (to_journal) {
2762                 down_write(&curseg->journal_rwsem);
2763         } else {
2764                 page = get_next_nat_page(sbi, start_nid);
2765                 nat_blk = page_address(page);
2766                 f2fs_bug_on(sbi, !nat_blk);
2767         }
2768
2769         /* flush dirty nats in nat entry set */
2770         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2771                 struct f2fs_nat_entry *raw_ne;
2772                 nid_t nid = nat_get_nid(ne);
2773                 int offset;
2774
2775                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2776
2777                 if (to_journal) {
2778                         offset = f2fs_lookup_journal_in_cursum(journal,
2779                                                         NAT_JOURNAL, nid, 1);
2780                         f2fs_bug_on(sbi, offset < 0);
2781                         raw_ne = &nat_in_journal(journal, offset);
2782                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2783                 } else {
2784                         raw_ne = &nat_blk->entries[nid - start_nid];
2785                 }
2786                 raw_nat_from_node_info(raw_ne, &ne->ni);
2787                 nat_reset_flag(ne);
2788                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2789                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2790                         add_free_nid(sbi, nid, false, true);
2791                 } else {
2792                         spin_lock(&NM_I(sbi)->nid_list_lock);
2793                         update_free_nid_bitmap(sbi, nid, false, false);
2794                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2795                 }
2796         }
2797
2798         if (to_journal) {
2799                 up_write(&curseg->journal_rwsem);
2800         } else {
2801                 __update_nat_bits(sbi, start_nid, page);
2802                 f2fs_put_page(page, 1);
2803         }
2804
2805         /* Allow dirty nats by node block allocation in write_begin */
2806         if (!set->entry_cnt) {
2807                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2808                 kmem_cache_free(nat_entry_set_slab, set);
2809         }
2810 }
2811
2812 /*
2813  * This function is called during the checkpointing process.
2814  */
2815 void f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2816 {
2817         struct f2fs_nm_info *nm_i = NM_I(sbi);
2818         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2819         struct f2fs_journal *journal = curseg->journal;
2820         struct nat_entry_set *setvec[SETVEC_SIZE];
2821         struct nat_entry_set *set, *tmp;
2822         unsigned int found;
2823         nid_t set_idx = 0;
2824         LIST_HEAD(sets);
2825
2826         /* during unmount, let's flush nat_bits before checking dirty_nat_cnt */
2827         if (enabled_nat_bits(sbi, cpc)) {
2828                 down_write(&nm_i->nat_tree_lock);
2829                 remove_nats_in_journal(sbi);
2830                 up_write(&nm_i->nat_tree_lock);
2831         }
2832
2833         if (!nm_i->dirty_nat_cnt)
2834                 return;
2835
2836         down_write(&nm_i->nat_tree_lock);
2837
2838         /*
2839          * if there are no enough space in journal to store dirty nat
2840          * entries, remove all entries from journal and merge them
2841          * into nat entry set.
2842          */
2843         if (enabled_nat_bits(sbi, cpc) ||
2844                 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2845                 remove_nats_in_journal(sbi);
2846
2847         while ((found = __gang_lookup_nat_set(nm_i,
2848                                         set_idx, SETVEC_SIZE, setvec))) {
2849                 unsigned idx;
2850                 set_idx = setvec[found - 1]->set + 1;
2851                 for (idx = 0; idx < found; idx++)
2852                         __adjust_nat_entry_set(setvec[idx], &sets,
2853                                                 MAX_NAT_JENTRIES(journal));
2854         }
2855
2856         /* flush dirty nats in nat entry set */
2857         list_for_each_entry_safe(set, tmp, &sets, set_list)
2858                 __flush_nat_entry_set(sbi, set, cpc);
2859
2860         up_write(&nm_i->nat_tree_lock);
2861         /* Allow dirty nats by node block allocation in write_begin */
2862 }
2863
2864 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2865 {
2866         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2867         struct f2fs_nm_info *nm_i = NM_I(sbi);
2868         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2869         unsigned int i;
2870         __u64 cp_ver = cur_cp_version(ckpt);
2871         block_t nat_bits_addr;
2872
2873         if (!enabled_nat_bits(sbi, NULL))
2874                 return 0;
2875
2876         nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2877         nm_i->nat_bits = f2fs_kzalloc(sbi,
2878                         nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2879         if (!nm_i->nat_bits)
2880                 return -ENOMEM;
2881
2882         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2883                                                 nm_i->nat_bits_blocks;
2884         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2885                 struct page *page;
2886
2887                 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
2888                 if (IS_ERR(page)) {
2889                         disable_nat_bits(sbi, true);
2890                         return PTR_ERR(page);
2891                 }
2892
2893                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2894                                         page_address(page), F2FS_BLKSIZE);
2895                 f2fs_put_page(page, 1);
2896         }
2897
2898         cp_ver |= (cur_cp_crc(ckpt) << 32);
2899         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2900                 disable_nat_bits(sbi, true);
2901                 return 0;
2902         }
2903
2904         nm_i->full_nat_bits = nm_i->nat_bits + 8;
2905         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2906
2907         f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2908         return 0;
2909 }
2910
2911 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2912 {
2913         struct f2fs_nm_info *nm_i = NM_I(sbi);
2914         unsigned int i = 0;
2915         nid_t nid, last_nid;
2916
2917         if (!enabled_nat_bits(sbi, NULL))
2918                 return;
2919
2920         for (i = 0; i < nm_i->nat_blocks; i++) {
2921                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2922                 if (i >= nm_i->nat_blocks)
2923                         break;
2924
2925                 __set_bit_le(i, nm_i->nat_block_bitmap);
2926
2927                 nid = i * NAT_ENTRY_PER_BLOCK;
2928                 last_nid = nid + NAT_ENTRY_PER_BLOCK;
2929
2930                 spin_lock(&NM_I(sbi)->nid_list_lock);
2931                 for (; nid < last_nid; nid++)
2932                         update_free_nid_bitmap(sbi, nid, true, true);
2933                 spin_unlock(&NM_I(sbi)->nid_list_lock);
2934         }
2935
2936         for (i = 0; i < nm_i->nat_blocks; i++) {
2937                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2938                 if (i >= nm_i->nat_blocks)
2939                         break;
2940
2941                 __set_bit_le(i, nm_i->nat_block_bitmap);
2942         }
2943 }
2944
2945 static int init_node_manager(struct f2fs_sb_info *sbi)
2946 {
2947         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2948         struct f2fs_nm_info *nm_i = NM_I(sbi);
2949         unsigned char *version_bitmap;
2950         unsigned int nat_segs;
2951         int err;
2952
2953         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2954
2955         /* segment_count_nat includes pair segment so divide to 2. */
2956         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2957         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2958         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2959
2960         /* not used nids: 0, node, meta, (and root counted as valid node) */
2961         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2962                                 sbi->nquota_files - F2FS_RESERVED_NODE_NUM;
2963         nm_i->nid_cnt[FREE_NID] = 0;
2964         nm_i->nid_cnt[PREALLOC_NID] = 0;
2965         nm_i->nat_cnt = 0;
2966         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2967         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2968         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2969
2970         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2971         INIT_LIST_HEAD(&nm_i->free_nid_list);
2972         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2973         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2974         INIT_LIST_HEAD(&nm_i->nat_entries);
2975         spin_lock_init(&nm_i->nat_list_lock);
2976
2977         mutex_init(&nm_i->build_lock);
2978         spin_lock_init(&nm_i->nid_list_lock);
2979         init_rwsem(&nm_i->nat_tree_lock);
2980
2981         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2982         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2983         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2984         if (!version_bitmap)
2985                 return -EFAULT;
2986
2987         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2988                                         GFP_KERNEL);
2989         if (!nm_i->nat_bitmap)
2990                 return -ENOMEM;
2991
2992         err = __get_nat_bitmaps(sbi);
2993         if (err)
2994                 return err;
2995
2996 #ifdef CONFIG_F2FS_CHECK_FS
2997         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2998                                         GFP_KERNEL);
2999         if (!nm_i->nat_bitmap_mir)
3000                 return -ENOMEM;
3001 #endif
3002
3003         return 0;
3004 }
3005
3006 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3007 {
3008         struct f2fs_nm_info *nm_i = NM_I(sbi);
3009         int i;
3010
3011         nm_i->free_nid_bitmap =
3012                 f2fs_kzalloc(sbi, array_size(sizeof(unsigned char *),
3013                                              nm_i->nat_blocks),
3014                              GFP_KERNEL);
3015         if (!nm_i->free_nid_bitmap)
3016                 return -ENOMEM;
3017
3018         for (i = 0; i < nm_i->nat_blocks; i++) {
3019                 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3020                         f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3021                 if (!nm_i->free_nid_bitmap[i])
3022                         return -ENOMEM;
3023         }
3024
3025         nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3026                                                                 GFP_KERNEL);
3027         if (!nm_i->nat_block_bitmap)
3028                 return -ENOMEM;
3029
3030         nm_i->free_nid_count =
3031                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3032                                               nm_i->nat_blocks),
3033                               GFP_KERNEL);
3034         if (!nm_i->free_nid_count)
3035                 return -ENOMEM;
3036         return 0;
3037 }
3038
3039 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3040 {
3041         int err;
3042
3043         sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3044                                                         GFP_KERNEL);
3045         if (!sbi->nm_info)
3046                 return -ENOMEM;
3047
3048         err = init_node_manager(sbi);
3049         if (err)
3050                 return err;
3051
3052         err = init_free_nid_cache(sbi);
3053         if (err)
3054                 return err;
3055
3056         /* load free nid status from nat_bits table */
3057         load_free_nid_bitmap(sbi);
3058
3059         return f2fs_build_free_nids(sbi, true, true);
3060 }
3061
3062 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3063 {
3064         struct f2fs_nm_info *nm_i = NM_I(sbi);
3065         struct free_nid *i, *next_i;
3066         struct nat_entry *natvec[NATVEC_SIZE];
3067         struct nat_entry_set *setvec[SETVEC_SIZE];
3068         nid_t nid = 0;
3069         unsigned int found;
3070
3071         if (!nm_i)
3072                 return;
3073
3074         /* destroy free nid list */
3075         spin_lock(&nm_i->nid_list_lock);
3076         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3077                 __remove_free_nid(sbi, i, FREE_NID);
3078                 spin_unlock(&nm_i->nid_list_lock);
3079                 kmem_cache_free(free_nid_slab, i);
3080                 spin_lock(&nm_i->nid_list_lock);
3081         }
3082         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3083         f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3084         f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3085         spin_unlock(&nm_i->nid_list_lock);
3086
3087         /* destroy nat cache */
3088         down_write(&nm_i->nat_tree_lock);
3089         while ((found = __gang_lookup_nat_cache(nm_i,
3090                                         nid, NATVEC_SIZE, natvec))) {
3091                 unsigned idx;
3092
3093                 nid = nat_get_nid(natvec[found - 1]) + 1;
3094                 for (idx = 0; idx < found; idx++) {
3095                         spin_lock(&nm_i->nat_list_lock);
3096                         list_del(&natvec[idx]->list);
3097                         spin_unlock(&nm_i->nat_list_lock);
3098
3099                         __del_from_nat_cache(nm_i, natvec[idx]);
3100                 }
3101         }
3102         f2fs_bug_on(sbi, nm_i->nat_cnt);
3103
3104         /* destroy nat set cache */
3105         nid = 0;
3106         while ((found = __gang_lookup_nat_set(nm_i,
3107                                         nid, SETVEC_SIZE, setvec))) {
3108                 unsigned idx;
3109
3110                 nid = setvec[found - 1]->set + 1;
3111                 for (idx = 0; idx < found; idx++) {
3112                         /* entry_cnt is not zero, when cp_error was occurred */
3113                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3114                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3115                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3116                 }
3117         }
3118         up_write(&nm_i->nat_tree_lock);
3119
3120         kvfree(nm_i->nat_block_bitmap);
3121         if (nm_i->free_nid_bitmap) {
3122                 int i;
3123
3124                 for (i = 0; i < nm_i->nat_blocks; i++)
3125                         kvfree(nm_i->free_nid_bitmap[i]);
3126                 kfree(nm_i->free_nid_bitmap);
3127         }
3128         kvfree(nm_i->free_nid_count);
3129
3130         kfree(nm_i->nat_bitmap);
3131         kfree(nm_i->nat_bits);
3132 #ifdef CONFIG_F2FS_CHECK_FS
3133         kfree(nm_i->nat_bitmap_mir);
3134 #endif
3135         sbi->nm_info = NULL;
3136         kfree(nm_i);
3137 }
3138
3139 int __init f2fs_create_node_manager_caches(void)
3140 {
3141         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
3142                         sizeof(struct nat_entry));
3143         if (!nat_entry_slab)
3144                 goto fail;
3145
3146         free_nid_slab = f2fs_kmem_cache_create("free_nid",
3147                         sizeof(struct free_nid));
3148         if (!free_nid_slab)
3149                 goto destroy_nat_entry;
3150
3151         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
3152                         sizeof(struct nat_entry_set));
3153         if (!nat_entry_set_slab)
3154                 goto destroy_free_nid;
3155
3156         fsync_node_entry_slab = f2fs_kmem_cache_create("fsync_node_entry",
3157                         sizeof(struct fsync_node_entry));
3158         if (!fsync_node_entry_slab)
3159                 goto destroy_nat_entry_set;
3160         return 0;
3161
3162 destroy_nat_entry_set:
3163         kmem_cache_destroy(nat_entry_set_slab);
3164 destroy_free_nid:
3165         kmem_cache_destroy(free_nid_slab);
3166 destroy_nat_entry:
3167         kmem_cache_destroy(nat_entry_slab);
3168 fail:
3169         return -ENOMEM;
3170 }
3171
3172 void f2fs_destroy_node_manager_caches(void)
3173 {
3174         kmem_cache_destroy(fsync_node_entry_slab);
3175         kmem_cache_destroy(nat_entry_set_slab);
3176         kmem_cache_destroy(free_nid_slab);
3177         kmem_cache_destroy(nat_entry_slab);
3178 }