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         f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1236 #endif
1237         new_ni.nid = dn->nid;
1238         new_ni.ino = dn->inode->i_ino;
1239         new_ni.blk_addr = NULL_ADDR;
1240         new_ni.flag = 0;
1241         new_ni.version = 0;
1242         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1243
1244         f2fs_wait_on_page_writeback(page, NODE, true);
1245         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1246         set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1247         if (!PageUptodate(page))
1248                 SetPageUptodate(page);
1249         if (set_page_dirty(page))
1250                 dn->node_changed = true;
1251
1252         if (f2fs_has_xattr_block(ofs))
1253                 f2fs_i_xnid_write(dn->inode, dn->nid);
1254
1255         if (ofs == 0)
1256                 inc_valid_inode_count(sbi);
1257         return page;
1258
1259 fail:
1260         clear_node_page_dirty(page);
1261         f2fs_put_page(page, 1);
1262         return ERR_PTR(err);
1263 }
1264
1265 /*
1266  * Caller should do after getting the following values.
1267  * 0: f2fs_put_page(page, 0)
1268  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1269  */
1270 static int read_node_page(struct page *page, int op_flags)
1271 {
1272         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1273         struct node_info ni;
1274         struct f2fs_io_info fio = {
1275                 .sbi = sbi,
1276                 .type = NODE,
1277                 .op = REQ_OP_READ,
1278                 .op_flags = op_flags,
1279                 .page = page,
1280                 .encrypted_page = NULL,
1281         };
1282         int err;
1283
1284         if (PageUptodate(page)) {
1285                 if (!f2fs_inode_chksum_verify(sbi, page)) {
1286                         ClearPageUptodate(page);
1287                         return -EFSBADCRC;
1288                 }
1289                 return LOCKED_PAGE;
1290         }
1291
1292         err = f2fs_get_node_info(sbi, page->index, &ni);
1293         if (err)
1294                 return err;
1295
1296         if (unlikely(ni.blk_addr == NULL_ADDR) ||
1297                         is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1298                 ClearPageUptodate(page);
1299                 return -ENOENT;
1300         }
1301
1302         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1303         return f2fs_submit_page_bio(&fio);
1304 }
1305
1306 /*
1307  * Readahead a node page
1308  */
1309 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1310 {
1311         struct page *apage;
1312         int err;
1313
1314         if (!nid)
1315                 return;
1316         if (f2fs_check_nid_range(sbi, nid))
1317                 return;
1318
1319         rcu_read_lock();
1320         apage = radix_tree_lookup(&NODE_MAPPING(sbi)->i_pages, nid);
1321         rcu_read_unlock();
1322         if (apage)
1323                 return;
1324
1325         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1326         if (!apage)
1327                 return;
1328
1329         err = read_node_page(apage, REQ_RAHEAD);
1330         f2fs_put_page(apage, err ? 1 : 0);
1331 }
1332
1333 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1334                                         struct page *parent, int start)
1335 {
1336         struct page *page;
1337         int err;
1338
1339         if (!nid)
1340                 return ERR_PTR(-ENOENT);
1341         if (f2fs_check_nid_range(sbi, nid))
1342                 return ERR_PTR(-EINVAL);
1343 repeat:
1344         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1345         if (!page)
1346                 return ERR_PTR(-ENOMEM);
1347
1348         err = read_node_page(page, 0);
1349         if (err < 0) {
1350                 f2fs_put_page(page, 1);
1351                 return ERR_PTR(err);
1352         } else if (err == LOCKED_PAGE) {
1353                 err = 0;
1354                 goto page_hit;
1355         }
1356
1357         if (parent)
1358                 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1359
1360         lock_page(page);
1361
1362         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1363                 f2fs_put_page(page, 1);
1364                 goto repeat;
1365         }
1366
1367         if (unlikely(!PageUptodate(page))) {
1368                 err = -EIO;
1369                 goto out_err;
1370         }
1371
1372         if (!f2fs_inode_chksum_verify(sbi, page)) {
1373                 err = -EFSBADCRC;
1374                 goto out_err;
1375         }
1376 page_hit:
1377         if(unlikely(nid != nid_of_node(page))) {
1378                 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1379                         "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1380                         nid, nid_of_node(page), ino_of_node(page),
1381                         ofs_of_node(page), cpver_of_node(page),
1382                         next_blkaddr_of_node(page));
1383                 err = -EINVAL;
1384 out_err:
1385                 ClearPageUptodate(page);
1386                 f2fs_put_page(page, 1);
1387                 return ERR_PTR(err);
1388         }
1389         return page;
1390 }
1391
1392 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1393 {
1394         return __get_node_page(sbi, nid, NULL, 0);
1395 }
1396
1397 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1398 {
1399         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1400         nid_t nid = get_nid(parent, start, false);
1401
1402         return __get_node_page(sbi, nid, parent, start);
1403 }
1404
1405 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1406 {
1407         struct inode *inode;
1408         struct page *page;
1409         int ret;
1410
1411         /* should flush inline_data before evict_inode */
1412         inode = ilookup(sbi->sb, ino);
1413         if (!inode)
1414                 return;
1415
1416         page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1417                                         FGP_LOCK|FGP_NOWAIT, 0);
1418         if (!page)
1419                 goto iput_out;
1420
1421         if (!PageUptodate(page))
1422                 goto page_out;
1423
1424         if (!PageDirty(page))
1425                 goto page_out;
1426
1427         if (!clear_page_dirty_for_io(page))
1428                 goto page_out;
1429
1430         ret = f2fs_write_inline_data(inode, page);
1431         inode_dec_dirty_pages(inode);
1432         f2fs_remove_dirty_inode(inode);
1433         if (ret)
1434                 set_page_dirty(page);
1435 page_out:
1436         f2fs_put_page(page, 1);
1437 iput_out:
1438         iput(inode);
1439 }
1440
1441 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1442 {
1443         pgoff_t index;
1444         struct pagevec pvec;
1445         struct page *last_page = NULL;
1446         int nr_pages;
1447
1448         pagevec_init(&pvec);
1449         index = 0;
1450
1451         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1452                                 PAGECACHE_TAG_DIRTY))) {
1453                 int i;
1454
1455                 for (i = 0; i < nr_pages; i++) {
1456                         struct page *page = pvec.pages[i];
1457
1458                         if (unlikely(f2fs_cp_error(sbi))) {
1459                                 f2fs_put_page(last_page, 0);
1460                                 pagevec_release(&pvec);
1461                                 return ERR_PTR(-EIO);
1462                         }
1463
1464                         if (!IS_DNODE(page) || !is_cold_node(page))
1465                                 continue;
1466                         if (ino_of_node(page) != ino)
1467                                 continue;
1468
1469                         lock_page(page);
1470
1471                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1472 continue_unlock:
1473                                 unlock_page(page);
1474                                 continue;
1475                         }
1476                         if (ino_of_node(page) != ino)
1477                                 goto continue_unlock;
1478
1479                         if (!PageDirty(page)) {
1480                                 /* someone wrote it for us */
1481                                 goto continue_unlock;
1482                         }
1483
1484                         if (last_page)
1485                                 f2fs_put_page(last_page, 0);
1486
1487                         get_page(page);
1488                         last_page = page;
1489                         unlock_page(page);
1490                 }
1491                 pagevec_release(&pvec);
1492                 cond_resched();
1493         }
1494         return last_page;
1495 }
1496
1497 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1498                                 struct writeback_control *wbc, bool do_balance,
1499                                 enum iostat_type io_type, unsigned int *seq_id)
1500 {
1501         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1502         nid_t nid;
1503         struct node_info ni;
1504         struct f2fs_io_info fio = {
1505                 .sbi = sbi,
1506                 .ino = ino_of_node(page),
1507                 .type = NODE,
1508                 .op = REQ_OP_WRITE,
1509                 .op_flags = wbc_to_write_flags(wbc),
1510                 .page = page,
1511                 .encrypted_page = NULL,
1512                 .submitted = false,
1513                 .io_type = io_type,
1514                 .io_wbc = wbc,
1515         };
1516         unsigned int seq;
1517
1518         trace_f2fs_writepage(page, NODE);
1519
1520         if (unlikely(f2fs_cp_error(sbi)))
1521                 goto redirty_out;
1522
1523         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1524                 goto redirty_out;
1525
1526         if (wbc->sync_mode == WB_SYNC_NONE &&
1527                         IS_DNODE(page) && is_cold_node(page))
1528                 goto redirty_out;
1529
1530         /* get old block addr of this node page */
1531         nid = nid_of_node(page);
1532         f2fs_bug_on(sbi, page->index != nid);
1533
1534         if (f2fs_get_node_info(sbi, nid, &ni))
1535                 goto redirty_out;
1536
1537         if (wbc->for_reclaim) {
1538                 if (!down_read_trylock(&sbi->node_write))
1539                         goto redirty_out;
1540         } else {
1541                 down_read(&sbi->node_write);
1542         }
1543
1544         /* This page is already truncated */
1545         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1546                 ClearPageUptodate(page);
1547                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1548                 up_read(&sbi->node_write);
1549                 unlock_page(page);
1550                 return 0;
1551         }
1552
1553         if (__is_valid_data_blkaddr(ni.blk_addr) &&
1554                 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC)) {
1555                 up_read(&sbi->node_write);
1556                 goto redirty_out;
1557         }
1558
1559         if (atomic && !test_opt(sbi, NOBARRIER))
1560                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1561
1562         set_page_writeback(page);
1563         ClearPageError(page);
1564
1565         if (f2fs_in_warm_node_list(sbi, page)) {
1566                 seq = f2fs_add_fsync_node_entry(sbi, page);
1567                 if (seq_id)
1568                         *seq_id = seq;
1569         }
1570
1571         fio.old_blkaddr = ni.blk_addr;
1572         f2fs_do_write_node_page(nid, &fio);
1573         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1574         dec_page_count(sbi, F2FS_DIRTY_NODES);
1575         up_read(&sbi->node_write);
1576
1577         if (wbc->for_reclaim) {
1578                 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1579                                                 page->index, NODE);
1580                 submitted = NULL;
1581         }
1582
1583         unlock_page(page);
1584
1585         if (unlikely(f2fs_cp_error(sbi))) {
1586                 f2fs_submit_merged_write(sbi, NODE);
1587                 submitted = NULL;
1588         }
1589         if (submitted)
1590                 *submitted = fio.submitted;
1591
1592         if (do_balance)
1593                 f2fs_balance_fs(sbi, false);
1594         return 0;
1595
1596 redirty_out:
1597         redirty_page_for_writepage(wbc, page);
1598         return AOP_WRITEPAGE_ACTIVATE;
1599 }
1600
1601 void f2fs_move_node_page(struct page *node_page, int gc_type)
1602 {
1603         if (gc_type == FG_GC) {
1604                 struct writeback_control wbc = {
1605                         .sync_mode = WB_SYNC_ALL,
1606                         .nr_to_write = 1,
1607                         .for_reclaim = 0,
1608                 };
1609
1610                 set_page_dirty(node_page);
1611                 f2fs_wait_on_page_writeback(node_page, NODE, true);
1612
1613                 f2fs_bug_on(F2FS_P_SB(node_page), PageWriteback(node_page));
1614                 if (!clear_page_dirty_for_io(node_page))
1615                         goto out_page;
1616
1617                 if (__write_node_page(node_page, false, NULL,
1618                                         &wbc, false, FS_GC_NODE_IO, NULL))
1619                         unlock_page(node_page);
1620                 goto release_page;
1621         } else {
1622                 /* set page dirty and write it */
1623                 if (!PageWriteback(node_page))
1624                         set_page_dirty(node_page);
1625         }
1626 out_page:
1627         unlock_page(node_page);
1628 release_page:
1629         f2fs_put_page(node_page, 0);
1630 }
1631
1632 static int f2fs_write_node_page(struct page *page,
1633                                 struct writeback_control *wbc)
1634 {
1635         return __write_node_page(page, false, NULL, wbc, false,
1636                                                 FS_NODE_IO, NULL);
1637 }
1638
1639 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1640                         struct writeback_control *wbc, bool atomic,
1641                         unsigned int *seq_id)
1642 {
1643         pgoff_t index;
1644         pgoff_t last_idx = ULONG_MAX;
1645         struct pagevec pvec;
1646         int ret = 0;
1647         struct page *last_page = NULL;
1648         bool marked = false;
1649         nid_t ino = inode->i_ino;
1650         int nr_pages;
1651
1652         if (atomic) {
1653                 last_page = last_fsync_dnode(sbi, ino);
1654                 if (IS_ERR_OR_NULL(last_page))
1655                         return PTR_ERR_OR_ZERO(last_page);
1656         }
1657 retry:
1658         pagevec_init(&pvec);
1659         index = 0;
1660
1661         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1662                                 PAGECACHE_TAG_DIRTY))) {
1663                 int i;
1664
1665                 for (i = 0; i < nr_pages; i++) {
1666                         struct page *page = pvec.pages[i];
1667                         bool submitted = false;
1668
1669                         if (unlikely(f2fs_cp_error(sbi))) {
1670                                 f2fs_put_page(last_page, 0);
1671                                 pagevec_release(&pvec);
1672                                 ret = -EIO;
1673                                 goto out;
1674                         }
1675
1676                         if (!IS_DNODE(page) || !is_cold_node(page))
1677                                 continue;
1678                         if (ino_of_node(page) != ino)
1679                                 continue;
1680
1681                         lock_page(page);
1682
1683                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1684 continue_unlock:
1685                                 unlock_page(page);
1686                                 continue;
1687                         }
1688                         if (ino_of_node(page) != ino)
1689                                 goto continue_unlock;
1690
1691                         if (!PageDirty(page) && page != last_page) {
1692                                 /* someone wrote it for us */
1693                                 goto continue_unlock;
1694                         }
1695
1696                         f2fs_wait_on_page_writeback(page, NODE, true);
1697                         BUG_ON(PageWriteback(page));
1698
1699                         set_fsync_mark(page, 0);
1700                         set_dentry_mark(page, 0);
1701
1702                         if (!atomic || page == last_page) {
1703                                 set_fsync_mark(page, 1);
1704                                 if (IS_INODE(page)) {
1705                                         if (is_inode_flag_set(inode,
1706                                                                 FI_DIRTY_INODE))
1707                                                 f2fs_update_inode(inode, page);
1708                                         set_dentry_mark(page,
1709                                                 f2fs_need_dentry_mark(sbi, ino));
1710                                 }
1711                                 /*  may be written by other thread */
1712                                 if (!PageDirty(page))
1713                                         set_page_dirty(page);
1714                         }
1715
1716                         if (!clear_page_dirty_for_io(page))
1717                                 goto continue_unlock;
1718
1719                         ret = __write_node_page(page, atomic &&
1720                                                 page == last_page,
1721                                                 &submitted, wbc, true,
1722                                                 FS_NODE_IO, seq_id);
1723                         if (ret) {
1724                                 unlock_page(page);
1725                                 f2fs_put_page(last_page, 0);
1726                                 break;
1727                         } else if (submitted) {
1728                                 last_idx = page->index;
1729                         }
1730
1731                         if (page == last_page) {
1732                                 f2fs_put_page(page, 0);
1733                                 marked = true;
1734                                 break;
1735                         }
1736                 }
1737                 pagevec_release(&pvec);
1738                 cond_resched();
1739
1740                 if (ret || marked)
1741                         break;
1742         }
1743         if (!ret && atomic && !marked) {
1744                 f2fs_msg(sbi->sb, KERN_DEBUG,
1745                         "Retry to write fsync mark: ino=%u, idx=%lx",
1746                                         ino, last_page->index);
1747                 lock_page(last_page);
1748                 f2fs_wait_on_page_writeback(last_page, NODE, true);
1749                 set_page_dirty(last_page);
1750                 unlock_page(last_page);
1751                 goto retry;
1752         }
1753 out:
1754         if (last_idx != ULONG_MAX)
1755                 f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1756         return ret ? -EIO: 0;
1757 }
1758
1759 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1760                                 struct writeback_control *wbc,
1761                                 bool do_balance, enum iostat_type io_type)
1762 {
1763         pgoff_t index;
1764         struct pagevec pvec;
1765         int step = 0;
1766         int nwritten = 0;
1767         int ret = 0;
1768         int nr_pages, done = 0;
1769
1770         pagevec_init(&pvec);
1771
1772 next_step:
1773         index = 0;
1774
1775         while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1776                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1777                 int i;
1778
1779                 for (i = 0; i < nr_pages; i++) {
1780                         struct page *page = pvec.pages[i];
1781                         bool submitted = false;
1782
1783                         /* give a priority to WB_SYNC threads */
1784                         if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1785                                         wbc->sync_mode == WB_SYNC_NONE) {
1786                                 done = 1;
1787                                 break;
1788                         }
1789
1790                         /*
1791                          * flushing sequence with step:
1792                          * 0. indirect nodes
1793                          * 1. dentry dnodes
1794                          * 2. file dnodes
1795                          */
1796                         if (step == 0 && IS_DNODE(page))
1797                                 continue;
1798                         if (step == 1 && (!IS_DNODE(page) ||
1799                                                 is_cold_node(page)))
1800                                 continue;
1801                         if (step == 2 && (!IS_DNODE(page) ||
1802                                                 !is_cold_node(page)))
1803                                 continue;
1804 lock_node:
1805                         if (wbc->sync_mode == WB_SYNC_ALL)
1806                                 lock_page(page);
1807                         else if (!trylock_page(page))
1808                                 continue;
1809
1810                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1811 continue_unlock:
1812                                 unlock_page(page);
1813                                 continue;
1814                         }
1815
1816                         if (!PageDirty(page)) {
1817                                 /* someone wrote it for us */
1818                                 goto continue_unlock;
1819                         }
1820
1821                         /* flush inline_data */
1822                         if (is_inline_node(page)) {
1823                                 clear_inline_node(page);
1824                                 unlock_page(page);
1825                                 flush_inline_data(sbi, ino_of_node(page));
1826                                 goto lock_node;
1827                         }
1828
1829                         f2fs_wait_on_page_writeback(page, NODE, true);
1830
1831                         BUG_ON(PageWriteback(page));
1832                         if (!clear_page_dirty_for_io(page))
1833                                 goto continue_unlock;
1834
1835                         set_fsync_mark(page, 0);
1836                         set_dentry_mark(page, 0);
1837
1838                         ret = __write_node_page(page, false, &submitted,
1839                                                 wbc, do_balance, io_type, NULL);
1840                         if (ret)
1841                                 unlock_page(page);
1842                         else if (submitted)
1843                                 nwritten++;
1844
1845                         if (--wbc->nr_to_write == 0)
1846                                 break;
1847                 }
1848                 pagevec_release(&pvec);
1849                 cond_resched();
1850
1851                 if (wbc->nr_to_write == 0) {
1852                         step = 2;
1853                         break;
1854                 }
1855         }
1856
1857         if (step < 2) {
1858                 if (wbc->sync_mode == WB_SYNC_NONE && step == 1)
1859                         goto out;
1860                 step++;
1861                 goto next_step;
1862         }
1863 out:
1864         if (nwritten)
1865                 f2fs_submit_merged_write(sbi, NODE);
1866
1867         if (unlikely(f2fs_cp_error(sbi)))
1868                 return -EIO;
1869         return ret;
1870 }
1871
1872 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1873                                                 unsigned int seq_id)
1874 {
1875         struct fsync_node_entry *fn;
1876         struct page *page;
1877         struct list_head *head = &sbi->fsync_node_list;
1878         unsigned long flags;
1879         unsigned int cur_seq_id = 0;
1880         int ret2, ret = 0;
1881
1882         while (seq_id && cur_seq_id < seq_id) {
1883                 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
1884                 if (list_empty(head)) {
1885                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1886                         break;
1887                 }
1888                 fn = list_first_entry(head, struct fsync_node_entry, list);
1889                 if (fn->seq_id > seq_id) {
1890                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1891                         break;
1892                 }
1893                 cur_seq_id = fn->seq_id;
1894                 page = fn->page;
1895                 get_page(page);
1896                 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1897
1898                 f2fs_wait_on_page_writeback(page, NODE, true);
1899                 if (TestClearPageError(page))
1900                         ret = -EIO;
1901
1902                 put_page(page);
1903
1904                 if (ret)
1905                         break;
1906         }
1907
1908         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1909         if (!ret)
1910                 ret = ret2;
1911
1912         return ret;
1913 }
1914
1915 static int f2fs_write_node_pages(struct address_space *mapping,
1916                             struct writeback_control *wbc)
1917 {
1918         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1919         struct blk_plug plug;
1920         long diff;
1921
1922         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1923                 goto skip_write;
1924
1925         /* balancing f2fs's metadata in background */
1926         f2fs_balance_fs_bg(sbi);
1927
1928         /* collect a number of dirty node pages and write together */
1929         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1930                 goto skip_write;
1931
1932         if (wbc->sync_mode == WB_SYNC_ALL)
1933                 atomic_inc(&sbi->wb_sync_req[NODE]);
1934         else if (atomic_read(&sbi->wb_sync_req[NODE]))
1935                 goto skip_write;
1936
1937         trace_f2fs_writepages(mapping->host, wbc, NODE);
1938
1939         diff = nr_pages_to_write(sbi, NODE, wbc);
1940         blk_start_plug(&plug);
1941         f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1942         blk_finish_plug(&plug);
1943         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1944
1945         if (wbc->sync_mode == WB_SYNC_ALL)
1946                 atomic_dec(&sbi->wb_sync_req[NODE]);
1947         return 0;
1948
1949 skip_write:
1950         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1951         trace_f2fs_writepages(mapping->host, wbc, NODE);
1952         return 0;
1953 }
1954
1955 static int f2fs_set_node_page_dirty(struct page *page)
1956 {
1957         trace_f2fs_set_page_dirty(page, NODE);
1958
1959         if (!PageUptodate(page))
1960                 SetPageUptodate(page);
1961 #ifdef CONFIG_F2FS_CHECK_FS
1962         if (IS_INODE(page))
1963                 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
1964 #endif
1965         if (!PageDirty(page)) {
1966                 __set_page_dirty_nobuffers(page);
1967                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1968                 SetPagePrivate(page);
1969                 f2fs_trace_pid(page);
1970                 return 1;
1971         }
1972         return 0;
1973 }
1974
1975 /*
1976  * Structure of the f2fs node operations
1977  */
1978 const struct address_space_operations f2fs_node_aops = {
1979         .writepage      = f2fs_write_node_page,
1980         .writepages     = f2fs_write_node_pages,
1981         .set_page_dirty = f2fs_set_node_page_dirty,
1982         .invalidatepage = f2fs_invalidate_page,
1983         .releasepage    = f2fs_release_page,
1984 #ifdef CONFIG_MIGRATION
1985         .migratepage    = f2fs_migrate_page,
1986 #endif
1987 };
1988
1989 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1990                                                 nid_t n)
1991 {
1992         return radix_tree_lookup(&nm_i->free_nid_root, n);
1993 }
1994
1995 static int __insert_free_nid(struct f2fs_sb_info *sbi,
1996                         struct free_nid *i, enum nid_state state)
1997 {
1998         struct f2fs_nm_info *nm_i = NM_I(sbi);
1999
2000         int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2001         if (err)
2002                 return err;
2003
2004         f2fs_bug_on(sbi, state != i->state);
2005         nm_i->nid_cnt[state]++;
2006         if (state == FREE_NID)
2007                 list_add_tail(&i->list, &nm_i->free_nid_list);
2008         return 0;
2009 }
2010
2011 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2012                         struct free_nid *i, enum nid_state state)
2013 {
2014         struct f2fs_nm_info *nm_i = NM_I(sbi);
2015
2016         f2fs_bug_on(sbi, state != i->state);
2017         nm_i->nid_cnt[state]--;
2018         if (state == FREE_NID)
2019                 list_del(&i->list);
2020         radix_tree_delete(&nm_i->free_nid_root, i->nid);
2021 }
2022
2023 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2024                         enum nid_state org_state, enum nid_state dst_state)
2025 {
2026         struct f2fs_nm_info *nm_i = NM_I(sbi);
2027
2028         f2fs_bug_on(sbi, org_state != i->state);
2029         i->state = dst_state;
2030         nm_i->nid_cnt[org_state]--;
2031         nm_i->nid_cnt[dst_state]++;
2032
2033         switch (dst_state) {
2034         case PREALLOC_NID:
2035                 list_del(&i->list);
2036                 break;
2037         case FREE_NID:
2038                 list_add_tail(&i->list, &nm_i->free_nid_list);
2039                 break;
2040         default:
2041                 BUG_ON(1);
2042         }
2043 }
2044
2045 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2046                                                         bool set, bool build)
2047 {
2048         struct f2fs_nm_info *nm_i = NM_I(sbi);
2049         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2050         unsigned int nid_ofs = nid - START_NID(nid);
2051
2052         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2053                 return;
2054
2055         if (set) {
2056                 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2057                         return;
2058                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2059                 nm_i->free_nid_count[nat_ofs]++;
2060         } else {
2061                 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2062                         return;
2063                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2064                 if (!build)
2065                         nm_i->free_nid_count[nat_ofs]--;
2066         }
2067 }
2068
2069 /* return if the nid is recognized as free */
2070 static bool add_free_nid(struct f2fs_sb_info *sbi,
2071                                 nid_t nid, bool build, bool update)
2072 {
2073         struct f2fs_nm_info *nm_i = NM_I(sbi);
2074         struct free_nid *i, *e;
2075         struct nat_entry *ne;
2076         int err = -EINVAL;
2077         bool ret = false;
2078
2079         /* 0 nid should not be used */
2080         if (unlikely(nid == 0))
2081                 return false;
2082
2083         if (unlikely(f2fs_check_nid_range(sbi, nid)))
2084                 return false;
2085
2086         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2087         i->nid = nid;
2088         i->state = FREE_NID;
2089
2090         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2091
2092         spin_lock(&nm_i->nid_list_lock);
2093
2094         if (build) {
2095                 /*
2096                  *   Thread A             Thread B
2097                  *  - f2fs_create
2098                  *   - f2fs_new_inode
2099                  *    - f2fs_alloc_nid
2100                  *     - __insert_nid_to_list(PREALLOC_NID)
2101                  *                     - f2fs_balance_fs_bg
2102                  *                      - f2fs_build_free_nids
2103                  *                       - __f2fs_build_free_nids
2104                  *                        - scan_nat_page
2105                  *                         - add_free_nid
2106                  *                          - __lookup_nat_cache
2107                  *  - f2fs_add_link
2108                  *   - f2fs_init_inode_metadata
2109                  *    - f2fs_new_inode_page
2110                  *     - f2fs_new_node_page
2111                  *      - set_node_addr
2112                  *  - f2fs_alloc_nid_done
2113                  *   - __remove_nid_from_list(PREALLOC_NID)
2114                  *                         - __insert_nid_to_list(FREE_NID)
2115                  */
2116                 ne = __lookup_nat_cache(nm_i, nid);
2117                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2118                                 nat_get_blkaddr(ne) != NULL_ADDR))
2119                         goto err_out;
2120
2121                 e = __lookup_free_nid_list(nm_i, nid);
2122                 if (e) {
2123                         if (e->state == FREE_NID)
2124                                 ret = true;
2125                         goto err_out;
2126                 }
2127         }
2128         ret = true;
2129         err = __insert_free_nid(sbi, i, FREE_NID);
2130 err_out:
2131         if (update) {
2132                 update_free_nid_bitmap(sbi, nid, ret, build);
2133                 if (!build)
2134                         nm_i->available_nids++;
2135         }
2136         spin_unlock(&nm_i->nid_list_lock);
2137         radix_tree_preload_end();
2138
2139         if (err)
2140                 kmem_cache_free(free_nid_slab, i);
2141         return ret;
2142 }
2143
2144 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2145 {
2146         struct f2fs_nm_info *nm_i = NM_I(sbi);
2147         struct free_nid *i;
2148         bool need_free = false;
2149
2150         spin_lock(&nm_i->nid_list_lock);
2151         i = __lookup_free_nid_list(nm_i, nid);
2152         if (i && i->state == FREE_NID) {
2153                 __remove_free_nid(sbi, i, FREE_NID);
2154                 need_free = true;
2155         }
2156         spin_unlock(&nm_i->nid_list_lock);
2157
2158         if (need_free)
2159                 kmem_cache_free(free_nid_slab, i);
2160 }
2161
2162 static int scan_nat_page(struct f2fs_sb_info *sbi,
2163                         struct page *nat_page, nid_t start_nid)
2164 {
2165         struct f2fs_nm_info *nm_i = NM_I(sbi);
2166         struct f2fs_nat_block *nat_blk = page_address(nat_page);
2167         block_t blk_addr;
2168         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2169         int i;
2170
2171         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2172
2173         i = start_nid % NAT_ENTRY_PER_BLOCK;
2174
2175         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2176                 if (unlikely(start_nid >= nm_i->max_nid))
2177                         break;
2178
2179                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2180
2181                 if (blk_addr == NEW_ADDR)
2182                         return -EINVAL;
2183
2184                 if (blk_addr == NULL_ADDR) {
2185                         add_free_nid(sbi, start_nid, true, true);
2186                 } else {
2187                         spin_lock(&NM_I(sbi)->nid_list_lock);
2188                         update_free_nid_bitmap(sbi, start_nid, false, true);
2189                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2190                 }
2191         }
2192
2193         return 0;
2194 }
2195
2196 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2197 {
2198         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2199         struct f2fs_journal *journal = curseg->journal;
2200         int i;
2201
2202         down_read(&curseg->journal_rwsem);
2203         for (i = 0; i < nats_in_cursum(journal); i++) {
2204                 block_t addr;
2205                 nid_t nid;
2206
2207                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2208                 nid = le32_to_cpu(nid_in_journal(journal, i));
2209                 if (addr == NULL_ADDR)
2210                         add_free_nid(sbi, nid, true, false);
2211                 else
2212                         remove_free_nid(sbi, nid);
2213         }
2214         up_read(&curseg->journal_rwsem);
2215 }
2216
2217 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2218 {
2219         struct f2fs_nm_info *nm_i = NM_I(sbi);
2220         unsigned int i, idx;
2221         nid_t nid;
2222
2223         down_read(&nm_i->nat_tree_lock);
2224
2225         for (i = 0; i < nm_i->nat_blocks; i++) {
2226                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2227                         continue;
2228                 if (!nm_i->free_nid_count[i])
2229                         continue;
2230                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2231                         idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2232                                                 NAT_ENTRY_PER_BLOCK, idx);
2233                         if (idx >= NAT_ENTRY_PER_BLOCK)
2234                                 break;
2235
2236                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
2237                         add_free_nid(sbi, nid, true, false);
2238
2239                         if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2240                                 goto out;
2241                 }
2242         }
2243 out:
2244         scan_curseg_cache(sbi);
2245
2246         up_read(&nm_i->nat_tree_lock);
2247 }
2248
2249 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2250                                                 bool sync, bool mount)
2251 {
2252         struct f2fs_nm_info *nm_i = NM_I(sbi);
2253         int i = 0, ret;
2254         nid_t nid = nm_i->next_scan_nid;
2255
2256         if (unlikely(nid >= nm_i->max_nid))
2257                 nid = 0;
2258
2259         /* Enough entries */
2260         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2261                 return 0;
2262
2263         if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2264                 return 0;
2265
2266         if (!mount) {
2267                 /* try to find free nids in free_nid_bitmap */
2268                 scan_free_nid_bits(sbi);
2269
2270                 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2271                         return 0;
2272         }
2273
2274         /* readahead nat pages to be scanned */
2275         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2276                                                         META_NAT, true);
2277
2278         down_read(&nm_i->nat_tree_lock);
2279
2280         while (1) {
2281                 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2282                                                 nm_i->nat_block_bitmap)) {
2283                         struct page *page = get_current_nat_page(sbi, nid);
2284
2285                         ret = scan_nat_page(sbi, page, nid);
2286                         f2fs_put_page(page, 1);
2287
2288                         if (ret) {
2289                                 up_read(&nm_i->nat_tree_lock);
2290                                 f2fs_bug_on(sbi, !mount);
2291                                 f2fs_msg(sbi->sb, KERN_ERR,
2292                                         "NAT is corrupt, run fsck to fix it");
2293                                 return -EINVAL;
2294                         }
2295                 }
2296
2297                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2298                 if (unlikely(nid >= nm_i->max_nid))
2299                         nid = 0;
2300
2301                 if (++i >= FREE_NID_PAGES)
2302                         break;
2303         }
2304
2305         /* go to the next free nat pages to find free nids abundantly */
2306         nm_i->next_scan_nid = nid;
2307
2308         /* find free nids from current sum_pages */
2309         scan_curseg_cache(sbi);
2310
2311         up_read(&nm_i->nat_tree_lock);
2312
2313         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2314                                         nm_i->ra_nid_pages, META_NAT, false);
2315
2316         return 0;
2317 }
2318
2319 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2320 {
2321         int ret;
2322
2323         mutex_lock(&NM_I(sbi)->build_lock);
2324         ret = __f2fs_build_free_nids(sbi, sync, mount);
2325         mutex_unlock(&NM_I(sbi)->build_lock);
2326
2327         return ret;
2328 }
2329
2330 /*
2331  * If this function returns success, caller can obtain a new nid
2332  * from second parameter of this function.
2333  * The returned nid could be used ino as well as nid when inode is created.
2334  */
2335 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2336 {
2337         struct f2fs_nm_info *nm_i = NM_I(sbi);
2338         struct free_nid *i = NULL;
2339 retry:
2340         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2341                 f2fs_show_injection_info(FAULT_ALLOC_NID);
2342                 return false;
2343         }
2344
2345         spin_lock(&nm_i->nid_list_lock);
2346
2347         if (unlikely(nm_i->available_nids == 0)) {
2348                 spin_unlock(&nm_i->nid_list_lock);
2349                 return false;
2350         }
2351
2352         /* We should not use stale free nids created by f2fs_build_free_nids */
2353         if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2354                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2355                 i = list_first_entry(&nm_i->free_nid_list,
2356                                         struct free_nid, list);
2357                 *nid = i->nid;
2358
2359                 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2360                 nm_i->available_nids--;
2361
2362                 update_free_nid_bitmap(sbi, *nid, false, false);
2363
2364                 spin_unlock(&nm_i->nid_list_lock);
2365                 return true;
2366         }
2367         spin_unlock(&nm_i->nid_list_lock);
2368
2369         /* Let's scan nat pages and its caches to get free nids */
2370         f2fs_build_free_nids(sbi, true, false);
2371         goto retry;
2372 }
2373
2374 /*
2375  * f2fs_alloc_nid() should be called prior to this function.
2376  */
2377 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2378 {
2379         struct f2fs_nm_info *nm_i = NM_I(sbi);
2380         struct free_nid *i;
2381
2382         spin_lock(&nm_i->nid_list_lock);
2383         i = __lookup_free_nid_list(nm_i, nid);
2384         f2fs_bug_on(sbi, !i);
2385         __remove_free_nid(sbi, i, PREALLOC_NID);
2386         spin_unlock(&nm_i->nid_list_lock);
2387
2388         kmem_cache_free(free_nid_slab, i);
2389 }
2390
2391 /*
2392  * f2fs_alloc_nid() should be called prior to this function.
2393  */
2394 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2395 {
2396         struct f2fs_nm_info *nm_i = NM_I(sbi);
2397         struct free_nid *i;
2398         bool need_free = false;
2399
2400         if (!nid)
2401                 return;
2402
2403         spin_lock(&nm_i->nid_list_lock);
2404         i = __lookup_free_nid_list(nm_i, nid);
2405         f2fs_bug_on(sbi, !i);
2406
2407         if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2408                 __remove_free_nid(sbi, i, PREALLOC_NID);
2409                 need_free = true;
2410         } else {
2411                 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2412         }
2413
2414         nm_i->available_nids++;
2415
2416         update_free_nid_bitmap(sbi, nid, true, false);
2417
2418         spin_unlock(&nm_i->nid_list_lock);
2419
2420         if (need_free)
2421                 kmem_cache_free(free_nid_slab, i);
2422 }
2423
2424 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2425 {
2426         struct f2fs_nm_info *nm_i = NM_I(sbi);
2427         struct free_nid *i, *next;
2428         int nr = nr_shrink;
2429
2430         if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2431                 return 0;
2432
2433         if (!mutex_trylock(&nm_i->build_lock))
2434                 return 0;
2435
2436         spin_lock(&nm_i->nid_list_lock);
2437         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2438                 if (nr_shrink <= 0 ||
2439                                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2440                         break;
2441
2442                 __remove_free_nid(sbi, i, FREE_NID);
2443                 kmem_cache_free(free_nid_slab, i);
2444                 nr_shrink--;
2445         }
2446         spin_unlock(&nm_i->nid_list_lock);
2447         mutex_unlock(&nm_i->build_lock);
2448
2449         return nr - nr_shrink;
2450 }
2451
2452 void f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2453 {
2454         void *src_addr, *dst_addr;
2455         size_t inline_size;
2456         struct page *ipage;
2457         struct f2fs_inode *ri;
2458
2459         ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2460         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2461
2462         ri = F2FS_INODE(page);
2463         if (ri->i_inline & F2FS_INLINE_XATTR) {
2464                 set_inode_flag(inode, FI_INLINE_XATTR);
2465         } else {
2466                 clear_inode_flag(inode, FI_INLINE_XATTR);
2467                 goto update_inode;
2468         }
2469
2470         dst_addr = inline_xattr_addr(inode, ipage);
2471         src_addr = inline_xattr_addr(inode, page);
2472         inline_size = inline_xattr_size(inode);
2473
2474         f2fs_wait_on_page_writeback(ipage, NODE, true);
2475         memcpy(dst_addr, src_addr, inline_size);
2476 update_inode:
2477         f2fs_update_inode(inode, ipage);
2478         f2fs_put_page(ipage, 1);
2479 }
2480
2481 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2482 {
2483         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2484         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2485         nid_t new_xnid;
2486         struct dnode_of_data dn;
2487         struct node_info ni;
2488         struct page *xpage;
2489         int err;
2490
2491         if (!prev_xnid)
2492                 goto recover_xnid;
2493
2494         /* 1: invalidate the previous xattr nid */
2495         err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2496         if (err)
2497                 return err;
2498
2499         f2fs_invalidate_blocks(sbi, ni.blk_addr);
2500         dec_valid_node_count(sbi, inode, false);
2501         set_node_addr(sbi, &ni, NULL_ADDR, false);
2502
2503 recover_xnid:
2504         /* 2: update xattr nid in inode */
2505         if (!f2fs_alloc_nid(sbi, &new_xnid))
2506                 return -ENOSPC;
2507
2508         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2509         xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2510         if (IS_ERR(xpage)) {
2511                 f2fs_alloc_nid_failed(sbi, new_xnid);
2512                 return PTR_ERR(xpage);
2513         }
2514
2515         f2fs_alloc_nid_done(sbi, new_xnid);
2516         f2fs_update_inode_page(inode);
2517
2518         /* 3: update and set xattr node page dirty */
2519         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2520
2521         set_page_dirty(xpage);
2522         f2fs_put_page(xpage, 1);
2523
2524         return 0;
2525 }
2526
2527 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2528 {
2529         struct f2fs_inode *src, *dst;
2530         nid_t ino = ino_of_node(page);
2531         struct node_info old_ni, new_ni;
2532         struct page *ipage;
2533         int err;
2534
2535         err = f2fs_get_node_info(sbi, ino, &old_ni);
2536         if (err)
2537                 return err;
2538
2539         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2540                 return -EINVAL;
2541 retry:
2542         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2543         if (!ipage) {
2544                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2545                 goto retry;
2546         }
2547
2548         /* Should not use this inode from free nid list */
2549         remove_free_nid(sbi, ino);
2550
2551         if (!PageUptodate(ipage))
2552                 SetPageUptodate(ipage);
2553         fill_node_footer(ipage, ino, ino, 0, true);
2554         set_cold_node(ipage, false);
2555
2556         src = F2FS_INODE(page);
2557         dst = F2FS_INODE(ipage);
2558
2559         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2560         dst->i_size = 0;
2561         dst->i_blocks = cpu_to_le64(1);
2562         dst->i_links = cpu_to_le32(1);
2563         dst->i_xattr_nid = 0;
2564         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2565         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2566                 dst->i_extra_isize = src->i_extra_isize;
2567
2568                 if (f2fs_sb_has_flexible_inline_xattr(sbi->sb) &&
2569                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2570                                                         i_inline_xattr_size))
2571                         dst->i_inline_xattr_size = src->i_inline_xattr_size;
2572
2573                 if (f2fs_sb_has_project_quota(sbi->sb) &&
2574                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2575                                                                 i_projid))
2576                         dst->i_projid = src->i_projid;
2577
2578                 if (f2fs_sb_has_inode_crtime(sbi->sb) &&
2579                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2580                                                         i_crtime_nsec)) {
2581                         dst->i_crtime = src->i_crtime;
2582                         dst->i_crtime_nsec = src->i_crtime_nsec;
2583                 }
2584         }
2585
2586         new_ni = old_ni;
2587         new_ni.ino = ino;
2588
2589         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2590                 WARN_ON(1);
2591         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2592         inc_valid_inode_count(sbi);
2593         set_page_dirty(ipage);
2594         f2fs_put_page(ipage, 1);
2595         return 0;
2596 }
2597
2598 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2599                         unsigned int segno, struct f2fs_summary_block *sum)
2600 {
2601         struct f2fs_node *rn;
2602         struct f2fs_summary *sum_entry;
2603         block_t addr;
2604         int i, idx, last_offset, nrpages;
2605
2606         /* scan the node segment */
2607         last_offset = sbi->blocks_per_seg;
2608         addr = START_BLOCK(sbi, segno);
2609         sum_entry = &sum->entries[0];
2610
2611         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2612                 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2613
2614                 /* readahead node pages */
2615                 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2616
2617                 for (idx = addr; idx < addr + nrpages; idx++) {
2618                         struct page *page = f2fs_get_tmp_page(sbi, idx);
2619
2620                         if (IS_ERR(page))
2621                                 return PTR_ERR(page);
2622
2623                         rn = F2FS_NODE(page);
2624                         sum_entry->nid = rn->footer.nid;
2625                         sum_entry->version = 0;
2626                         sum_entry->ofs_in_node = 0;
2627                         sum_entry++;
2628                         f2fs_put_page(page, 1);
2629                 }
2630
2631                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2632                                                         addr + nrpages);
2633         }
2634         return 0;
2635 }
2636
2637 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2638 {
2639         struct f2fs_nm_info *nm_i = NM_I(sbi);
2640         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2641         struct f2fs_journal *journal = curseg->journal;
2642         int i;
2643
2644         down_write(&curseg->journal_rwsem);
2645         for (i = 0; i < nats_in_cursum(journal); i++) {
2646                 struct nat_entry *ne;
2647                 struct f2fs_nat_entry raw_ne;
2648                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2649
2650                 raw_ne = nat_in_journal(journal, i);
2651
2652                 ne = __lookup_nat_cache(nm_i, nid);
2653                 if (!ne) {
2654                         ne = __alloc_nat_entry(nid, true);
2655                         __init_nat_entry(nm_i, ne, &raw_ne, true);
2656                 }
2657
2658                 /*
2659                  * if a free nat in journal has not been used after last
2660                  * checkpoint, we should remove it from available nids,
2661                  * since later we will add it again.
2662                  */
2663                 if (!get_nat_flag(ne, IS_DIRTY) &&
2664                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2665                         spin_lock(&nm_i->nid_list_lock);
2666                         nm_i->available_nids--;
2667                         spin_unlock(&nm_i->nid_list_lock);
2668                 }
2669
2670                 __set_nat_cache_dirty(nm_i, ne);
2671         }
2672         update_nats_in_cursum(journal, -i);
2673         up_write(&curseg->journal_rwsem);
2674 }
2675
2676 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2677                                                 struct list_head *head, int max)
2678 {
2679         struct nat_entry_set *cur;
2680
2681         if (nes->entry_cnt >= max)
2682                 goto add_out;
2683
2684         list_for_each_entry(cur, head, set_list) {
2685                 if (cur->entry_cnt >= nes->entry_cnt) {
2686                         list_add(&nes->set_list, cur->set_list.prev);
2687                         return;
2688                 }
2689         }
2690 add_out:
2691         list_add_tail(&nes->set_list, head);
2692 }
2693
2694 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2695                                                 struct page *page)
2696 {
2697         struct f2fs_nm_info *nm_i = NM_I(sbi);
2698         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2699         struct f2fs_nat_block *nat_blk = page_address(page);
2700         int valid = 0;
2701         int i = 0;
2702
2703         if (!enabled_nat_bits(sbi, NULL))
2704                 return;
2705
2706         if (nat_index == 0) {
2707                 valid = 1;
2708                 i = 1;
2709         }
2710         for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2711                 if (nat_blk->entries[i].block_addr != NULL_ADDR)
2712                         valid++;
2713         }
2714         if (valid == 0) {
2715                 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2716                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2717                 return;
2718         }
2719
2720         __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2721         if (valid == NAT_ENTRY_PER_BLOCK)
2722                 __set_bit_le(nat_index, nm_i->full_nat_bits);
2723         else
2724                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2725 }
2726
2727 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2728                 struct nat_entry_set *set, struct cp_control *cpc)
2729 {
2730         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2731         struct f2fs_journal *journal = curseg->journal;
2732         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2733         bool to_journal = true;
2734         struct f2fs_nat_block *nat_blk;
2735         struct nat_entry *ne, *cur;
2736         struct page *page = NULL;
2737
2738         /*
2739          * there are two steps to flush nat entries:
2740          * #1, flush nat entries to journal in current hot data summary block.
2741          * #2, flush nat entries to nat page.
2742          */
2743         if (enabled_nat_bits(sbi, cpc) ||
2744                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2745                 to_journal = false;
2746
2747         if (to_journal) {
2748                 down_write(&curseg->journal_rwsem);
2749         } else {
2750                 page = get_next_nat_page(sbi, start_nid);
2751                 nat_blk = page_address(page);
2752                 f2fs_bug_on(sbi, !nat_blk);
2753         }
2754
2755         /* flush dirty nats in nat entry set */
2756         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2757                 struct f2fs_nat_entry *raw_ne;
2758                 nid_t nid = nat_get_nid(ne);
2759                 int offset;
2760
2761                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2762
2763                 if (to_journal) {
2764                         offset = f2fs_lookup_journal_in_cursum(journal,
2765                                                         NAT_JOURNAL, nid, 1);
2766                         f2fs_bug_on(sbi, offset < 0);
2767                         raw_ne = &nat_in_journal(journal, offset);
2768                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2769                 } else {
2770                         raw_ne = &nat_blk->entries[nid - start_nid];
2771                 }
2772                 raw_nat_from_node_info(raw_ne, &ne->ni);
2773                 nat_reset_flag(ne);
2774                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2775                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2776                         add_free_nid(sbi, nid, false, true);
2777                 } else {
2778                         spin_lock(&NM_I(sbi)->nid_list_lock);
2779                         update_free_nid_bitmap(sbi, nid, false, false);
2780                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2781                 }
2782         }
2783
2784         if (to_journal) {
2785                 up_write(&curseg->journal_rwsem);
2786         } else {
2787                 __update_nat_bits(sbi, start_nid, page);
2788                 f2fs_put_page(page, 1);
2789         }
2790
2791         /* Allow dirty nats by node block allocation in write_begin */
2792         if (!set->entry_cnt) {
2793                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2794                 kmem_cache_free(nat_entry_set_slab, set);
2795         }
2796 }
2797
2798 /*
2799  * This function is called during the checkpointing process.
2800  */
2801 void f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2802 {
2803         struct f2fs_nm_info *nm_i = NM_I(sbi);
2804         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2805         struct f2fs_journal *journal = curseg->journal;
2806         struct nat_entry_set *setvec[SETVEC_SIZE];
2807         struct nat_entry_set *set, *tmp;
2808         unsigned int found;
2809         nid_t set_idx = 0;
2810         LIST_HEAD(sets);
2811
2812         /* during unmount, let's flush nat_bits before checking dirty_nat_cnt */
2813         if (enabled_nat_bits(sbi, cpc)) {
2814                 down_write(&nm_i->nat_tree_lock);
2815                 remove_nats_in_journal(sbi);
2816                 up_write(&nm_i->nat_tree_lock);
2817         }
2818
2819         if (!nm_i->dirty_nat_cnt)
2820                 return;
2821
2822         down_write(&nm_i->nat_tree_lock);
2823
2824         /*
2825          * if there are no enough space in journal to store dirty nat
2826          * entries, remove all entries from journal and merge them
2827          * into nat entry set.
2828          */
2829         if (enabled_nat_bits(sbi, cpc) ||
2830                 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2831                 remove_nats_in_journal(sbi);
2832
2833         while ((found = __gang_lookup_nat_set(nm_i,
2834                                         set_idx, SETVEC_SIZE, setvec))) {
2835                 unsigned idx;
2836                 set_idx = setvec[found - 1]->set + 1;
2837                 for (idx = 0; idx < found; idx++)
2838                         __adjust_nat_entry_set(setvec[idx], &sets,
2839                                                 MAX_NAT_JENTRIES(journal));
2840         }
2841
2842         /* flush dirty nats in nat entry set */
2843         list_for_each_entry_safe(set, tmp, &sets, set_list)
2844                 __flush_nat_entry_set(sbi, set, cpc);
2845
2846         up_write(&nm_i->nat_tree_lock);
2847         /* Allow dirty nats by node block allocation in write_begin */
2848 }
2849
2850 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2851 {
2852         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2853         struct f2fs_nm_info *nm_i = NM_I(sbi);
2854         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2855         unsigned int i;
2856         __u64 cp_ver = cur_cp_version(ckpt);
2857         block_t nat_bits_addr;
2858
2859         if (!enabled_nat_bits(sbi, NULL))
2860                 return 0;
2861
2862         nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2863         nm_i->nat_bits = f2fs_kzalloc(sbi,
2864                         nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2865         if (!nm_i->nat_bits)
2866                 return -ENOMEM;
2867
2868         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2869                                                 nm_i->nat_bits_blocks;
2870         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2871                 struct page *page;
2872
2873                 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
2874                 if (IS_ERR(page)) {
2875                         disable_nat_bits(sbi, true);
2876                         return PTR_ERR(page);
2877                 }
2878
2879                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2880                                         page_address(page), F2FS_BLKSIZE);
2881                 f2fs_put_page(page, 1);
2882         }
2883
2884         cp_ver |= (cur_cp_crc(ckpt) << 32);
2885         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2886                 disable_nat_bits(sbi, true);
2887                 return 0;
2888         }
2889
2890         nm_i->full_nat_bits = nm_i->nat_bits + 8;
2891         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2892
2893         f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2894         return 0;
2895 }
2896
2897 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2898 {
2899         struct f2fs_nm_info *nm_i = NM_I(sbi);
2900         unsigned int i = 0;
2901         nid_t nid, last_nid;
2902
2903         if (!enabled_nat_bits(sbi, NULL))
2904                 return;
2905
2906         for (i = 0; i < nm_i->nat_blocks; i++) {
2907                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2908                 if (i >= nm_i->nat_blocks)
2909                         break;
2910
2911                 __set_bit_le(i, nm_i->nat_block_bitmap);
2912
2913                 nid = i * NAT_ENTRY_PER_BLOCK;
2914                 last_nid = nid + NAT_ENTRY_PER_BLOCK;
2915
2916                 spin_lock(&NM_I(sbi)->nid_list_lock);
2917                 for (; nid < last_nid; nid++)
2918                         update_free_nid_bitmap(sbi, nid, true, true);
2919                 spin_unlock(&NM_I(sbi)->nid_list_lock);
2920         }
2921
2922         for (i = 0; i < nm_i->nat_blocks; i++) {
2923                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2924                 if (i >= nm_i->nat_blocks)
2925                         break;
2926
2927                 __set_bit_le(i, nm_i->nat_block_bitmap);
2928         }
2929 }
2930
2931 static int init_node_manager(struct f2fs_sb_info *sbi)
2932 {
2933         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2934         struct f2fs_nm_info *nm_i = NM_I(sbi);
2935         unsigned char *version_bitmap;
2936         unsigned int nat_segs;
2937         int err;
2938
2939         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2940
2941         /* segment_count_nat includes pair segment so divide to 2. */
2942         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2943         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2944         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2945
2946         /* not used nids: 0, node, meta, (and root counted as valid node) */
2947         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2948                                 sbi->nquota_files - F2FS_RESERVED_NODE_NUM;
2949         nm_i->nid_cnt[FREE_NID] = 0;
2950         nm_i->nid_cnt[PREALLOC_NID] = 0;
2951         nm_i->nat_cnt = 0;
2952         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2953         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2954         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2955
2956         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2957         INIT_LIST_HEAD(&nm_i->free_nid_list);
2958         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2959         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2960         INIT_LIST_HEAD(&nm_i->nat_entries);
2961         spin_lock_init(&nm_i->nat_list_lock);
2962
2963         mutex_init(&nm_i->build_lock);
2964         spin_lock_init(&nm_i->nid_list_lock);
2965         init_rwsem(&nm_i->nat_tree_lock);
2966
2967         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2968         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2969         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2970         if (!version_bitmap)
2971                 return -EFAULT;
2972
2973         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2974                                         GFP_KERNEL);
2975         if (!nm_i->nat_bitmap)
2976                 return -ENOMEM;
2977
2978         err = __get_nat_bitmaps(sbi);
2979         if (err)
2980                 return err;
2981
2982 #ifdef CONFIG_F2FS_CHECK_FS
2983         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2984                                         GFP_KERNEL);
2985         if (!nm_i->nat_bitmap_mir)
2986                 return -ENOMEM;
2987 #endif
2988
2989         return 0;
2990 }
2991
2992 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2993 {
2994         struct f2fs_nm_info *nm_i = NM_I(sbi);
2995         int i;
2996
2997         nm_i->free_nid_bitmap =
2998                 f2fs_kzalloc(sbi, array_size(sizeof(unsigned char *),
2999                                              nm_i->nat_blocks),
3000                              GFP_KERNEL);
3001         if (!nm_i->free_nid_bitmap)
3002                 return -ENOMEM;
3003
3004         for (i = 0; i < nm_i->nat_blocks; i++) {
3005                 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3006                         f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3007                 if (!nm_i->free_nid_bitmap[i])
3008                         return -ENOMEM;
3009         }
3010
3011         nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3012                                                                 GFP_KERNEL);
3013         if (!nm_i->nat_block_bitmap)
3014                 return -ENOMEM;
3015
3016         nm_i->free_nid_count =
3017                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3018                                               nm_i->nat_blocks),
3019                               GFP_KERNEL);
3020         if (!nm_i->free_nid_count)
3021                 return -ENOMEM;
3022         return 0;
3023 }
3024
3025 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3026 {
3027         int err;
3028
3029         sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3030                                                         GFP_KERNEL);
3031         if (!sbi->nm_info)
3032                 return -ENOMEM;
3033
3034         err = init_node_manager(sbi);
3035         if (err)
3036                 return err;
3037
3038         err = init_free_nid_cache(sbi);
3039         if (err)
3040                 return err;
3041
3042         /* load free nid status from nat_bits table */
3043         load_free_nid_bitmap(sbi);
3044
3045         return f2fs_build_free_nids(sbi, true, true);
3046 }
3047
3048 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3049 {
3050         struct f2fs_nm_info *nm_i = NM_I(sbi);
3051         struct free_nid *i, *next_i;
3052         struct nat_entry *natvec[NATVEC_SIZE];
3053         struct nat_entry_set *setvec[SETVEC_SIZE];
3054         nid_t nid = 0;
3055         unsigned int found;
3056
3057         if (!nm_i)
3058                 return;
3059
3060         /* destroy free nid list */
3061         spin_lock(&nm_i->nid_list_lock);
3062         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3063                 __remove_free_nid(sbi, i, FREE_NID);
3064                 spin_unlock(&nm_i->nid_list_lock);
3065                 kmem_cache_free(free_nid_slab, i);
3066                 spin_lock(&nm_i->nid_list_lock);
3067         }
3068         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3069         f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3070         f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3071         spin_unlock(&nm_i->nid_list_lock);
3072
3073         /* destroy nat cache */
3074         down_write(&nm_i->nat_tree_lock);
3075         while ((found = __gang_lookup_nat_cache(nm_i,
3076                                         nid, NATVEC_SIZE, natvec))) {
3077                 unsigned idx;
3078
3079                 nid = nat_get_nid(natvec[found - 1]) + 1;
3080                 for (idx = 0; idx < found; idx++) {
3081                         spin_lock(&nm_i->nat_list_lock);
3082                         list_del(&natvec[idx]->list);
3083                         spin_unlock(&nm_i->nat_list_lock);
3084
3085                         __del_from_nat_cache(nm_i, natvec[idx]);
3086                 }
3087         }
3088         f2fs_bug_on(sbi, nm_i->nat_cnt);
3089
3090         /* destroy nat set cache */
3091         nid = 0;
3092         while ((found = __gang_lookup_nat_set(nm_i,
3093                                         nid, SETVEC_SIZE, setvec))) {
3094                 unsigned idx;
3095
3096                 nid = setvec[found - 1]->set + 1;
3097                 for (idx = 0; idx < found; idx++) {
3098                         /* entry_cnt is not zero, when cp_error was occurred */
3099                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3100                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3101                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3102                 }
3103         }
3104         up_write(&nm_i->nat_tree_lock);
3105
3106         kvfree(nm_i->nat_block_bitmap);
3107         if (nm_i->free_nid_bitmap) {
3108                 int i;
3109
3110                 for (i = 0; i < nm_i->nat_blocks; i++)
3111                         kvfree(nm_i->free_nid_bitmap[i]);
3112                 kfree(nm_i->free_nid_bitmap);
3113         }
3114         kvfree(nm_i->free_nid_count);
3115
3116         kfree(nm_i->nat_bitmap);
3117         kfree(nm_i->nat_bits);
3118 #ifdef CONFIG_F2FS_CHECK_FS
3119         kfree(nm_i->nat_bitmap_mir);
3120 #endif
3121         sbi->nm_info = NULL;
3122         kfree(nm_i);
3123 }
3124
3125 int __init f2fs_create_node_manager_caches(void)
3126 {
3127         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
3128                         sizeof(struct nat_entry));
3129         if (!nat_entry_slab)
3130                 goto fail;
3131
3132         free_nid_slab = f2fs_kmem_cache_create("free_nid",
3133                         sizeof(struct free_nid));
3134         if (!free_nid_slab)
3135                 goto destroy_nat_entry;
3136
3137         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
3138                         sizeof(struct nat_entry_set));
3139         if (!nat_entry_set_slab)
3140                 goto destroy_free_nid;
3141
3142         fsync_node_entry_slab = f2fs_kmem_cache_create("fsync_node_entry",
3143                         sizeof(struct fsync_node_entry));
3144         if (!fsync_node_entry_slab)
3145                 goto destroy_nat_entry_set;
3146         return 0;
3147
3148 destroy_nat_entry_set:
3149         kmem_cache_destroy(nat_entry_set_slab);
3150 destroy_free_nid:
3151         kmem_cache_destroy(free_nid_slab);
3152 destroy_nat_entry:
3153         kmem_cache_destroy(nat_entry_slab);
3154 fail:
3155         return -ENOMEM;
3156 }
3157
3158 void f2fs_destroy_node_manager_caches(void)
3159 {
3160         kmem_cache_destroy(fsync_node_entry_slab);
3161         kmem_cache_destroy(nat_entry_set_slab);
3162         kmem_cache_destroy(free_nid_slab);
3163         kmem_cache_destroy(nat_entry_slab);
3164 }