fs/f2fs/node.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/node.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/mpage.h>
  11 #include <linux/backing-dev.h>
  12 #include <linux/blkdev.h>
  13 #include <linux/pagevec.h>
  14 #include <linux/swap.h>
  15
  16 #include "f2fs.h"
  17 #include "node.h"
  18 #include "segment.h"
  19 #include "xattr.h"
  20 #include "trace.h"
  21 #include <trace/events/f2fs.h>
  22
  23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
  24
  25 static struct kmem_cache *nat_entry_slab;
  26 static struct kmem_cache *free_nid_slab;
  27 static struct kmem_cache *nat_entry_set_slab;
  28 static struct kmem_cache *fsync_node_entry_slab;
  29
  30 /*
  31  * Check whether the given nid is within node id range.
  32  */
  33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
  34 {
  35         if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
  36                 set_sbi_flag(sbi, SBI_NEED_FSCK);
  37                 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
  38                           __func__, nid);
  39                 return -EFSCORRUPTED;
  40         }
  41         return 0;
  42 }
  43
  44 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
  45 {
  46         struct f2fs_nm_info *nm_i = NM_I(sbi);
  47         struct sysinfo val;
  48         unsigned long avail_ram;
  49         unsigned long mem_size = 0;
  50         bool res = false;
  51
  52         si_meminfo(&val);
  53
  54         /* only uses low memory */
  55         avail_ram = val.totalram - val.totalhigh;
  56
  57         /*
  58          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  59          */
  60         if (type == FREE_NIDS) {
  61                 mem_size = (nm_i->nid_cnt[FREE_NID] *
  62                                 sizeof(struct free_nid)) >> PAGE_SHIFT;
  63                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  64         } else if (type == NAT_ENTRIES) {
  65                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
  66                                                         PAGE_SHIFT;
  67                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  68                 if (excess_cached_nats(sbi))
  69                         res = false;
  70         } else if (type == DIRTY_DENTS) {
  71                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  72                         return false;
  73                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  74                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  75         } else if (type == INO_ENTRIES) {
  76                 int i;
  77
  78                 for (i = 0; i < MAX_INO_ENTRY; i++)
  79                         mem_size += sbi->im[i].ino_num *
  80                                                 sizeof(struct ino_entry);
  81                 mem_size >>= PAGE_SHIFT;
  82                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  83         } else if (type == EXTENT_CACHE) {
  84                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  85                                 sizeof(struct extent_tree) +
  86                                 atomic_read(&sbi->total_ext_node) *
  87                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  88                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  89         } else if (type == INMEM_PAGES) {
  90                 /* it allows 20% / total_ram for inmemory pages */
  91                 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
  92                 res = mem_size < (val.totalram / 5);
  93         } else {
  94                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
  95                         return true;
  96         }
  97         return res;
  98 }
  99
 100 static void clear_node_page_dirty(struct page *page)
 101 {
 102         if (PageDirty(page)) {
 103                 f2fs_clear_page_cache_dirty_tag(page);
 104                 clear_page_dirty_for_io(page);
 105                 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
 106         }
 107         ClearPageUptodate(page);
 108 }
 109
 110 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 111 {
 112         return f2fs_get_meta_page_nofail(sbi, current_nat_addr(sbi, nid));
 113 }
 114
 115 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 116 {
 117         struct page *src_page;
 118         struct page *dst_page;
 119         pgoff_t dst_off;
 120         void *src_addr;
 121         void *dst_addr;
 122         struct f2fs_nm_info *nm_i = NM_I(sbi);
 123
 124         dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
 125
 126         /* get current nat block page with lock */
 127         src_page = get_current_nat_page(sbi, nid);
 128         if (IS_ERR(src_page))
 129                 return src_page;
 130         dst_page = f2fs_grab_meta_page(sbi, dst_off);
 131         f2fs_bug_on(sbi, PageDirty(src_page));
 132
 133         src_addr = page_address(src_page);
 134         dst_addr = page_address(dst_page);
 135         memcpy(dst_addr, src_addr, PAGE_SIZE);
 136         set_page_dirty(dst_page);
 137         f2fs_put_page(src_page, 1);
 138
 139         set_to_next_nat(nm_i, nid);
 140
 141         return dst_page;
 142 }
 143
 144 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
 145 {
 146         struct nat_entry *new;
 147
 148         if (no_fail)
 149                 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
 150         else
 151                 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
 152         if (new) {
 153                 nat_set_nid(new, nid);
 154                 nat_reset_flag(new);
 155         }
 156         return new;
 157 }
 158
 159 static void __free_nat_entry(struct nat_entry *e)
 160 {
 161         kmem_cache_free(nat_entry_slab, e);
 162 }
 163
 164 /* must be locked by nat_tree_lock */
 165 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
 166         struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
 167 {
 168         if (no_fail)
 169                 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
 170         else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
 171                 return NULL;
 172
 173         if (raw_ne)
 174                 node_info_from_raw_nat(&ne->ni, raw_ne);
 175
 176         spin_lock(&nm_i->nat_list_lock);
 177         list_add_tail(&ne->list, &nm_i->nat_entries);
 178         spin_unlock(&nm_i->nat_list_lock);
 179
 180         nm_i->nat_cnt++;
 181         return ne;
 182 }
 183
 184 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 185 {
 186         struct nat_entry *ne;
 187
 188         ne = radix_tree_lookup(&nm_i->nat_root, n);
 189
 190         /* for recent accessed nat entry, move it to tail of lru list */
 191         if (ne && !get_nat_flag(ne, IS_DIRTY)) {
 192                 spin_lock(&nm_i->nat_list_lock);
 193                 if (!list_empty(&ne->list))
 194                         list_move_tail(&ne->list, &nm_i->nat_entries);
 195                 spin_unlock(&nm_i->nat_list_lock);
 196         }
 197
 198         return ne;
 199 }
 200
 201 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 202                 nid_t start, unsigned int nr, struct nat_entry **ep)
 203 {
 204         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 205 }
 206
 207 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 208 {
 209         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 210         nm_i->nat_cnt--;
 211         __free_nat_entry(e);
 212 }
 213
 214 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
 215                                                         struct nat_entry *ne)
 216 {
 217         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 218         struct nat_entry_set *head;
 219
 220         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 221         if (!head) {
 222                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 223
 224                 INIT_LIST_HEAD(&head->entry_list);
 225                 INIT_LIST_HEAD(&head->set_list);
 226                 head->set = set;
 227                 head->entry_cnt = 0;
 228                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 229         }
 230         return head;
 231 }
 232
 233 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 234                                                 struct nat_entry *ne)
 235 {
 236         struct nat_entry_set *head;
 237         bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
 238
 239         if (!new_ne)
 240                 head = __grab_nat_entry_set(nm_i, ne);
 241
 242         /*
 243          * update entry_cnt in below condition:
 244          * 1. update NEW_ADDR to valid block address;
 245          * 2. update old block address to new one;
 246          */
 247         if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
 248                                 !get_nat_flag(ne, IS_DIRTY)))
 249                 head->entry_cnt++;
 250
 251         set_nat_flag(ne, IS_PREALLOC, new_ne);
 252
 253         if (get_nat_flag(ne, IS_DIRTY))
 254                 goto refresh_list;
 255
 256         nm_i->dirty_nat_cnt++;
 257         set_nat_flag(ne, IS_DIRTY, true);
 258 refresh_list:
 259         spin_lock(&nm_i->nat_list_lock);
 260         if (new_ne)
 261                 list_del_init(&ne->list);
 262         else
 263                 list_move_tail(&ne->list, &head->entry_list);
 264         spin_unlock(&nm_i->nat_list_lock);
 265 }
 266
 267 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 268                 struct nat_entry_set *set, struct nat_entry *ne)
 269 {
 270         spin_lock(&nm_i->nat_list_lock);
 271         list_move_tail(&ne->list, &nm_i->nat_entries);
 272         spin_unlock(&nm_i->nat_list_lock);
 273
 274         set_nat_flag(ne, IS_DIRTY, false);
 275         set->entry_cnt--;
 276         nm_i->dirty_nat_cnt--;
 277 }
 278
 279 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 280                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 281 {
 282         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 283                                                         start, nr);
 284 }
 285
 286 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
 287 {
 288         return NODE_MAPPING(sbi) == page->mapping &&
 289                         IS_DNODE(page) && is_cold_node(page);
 290 }
 291
 292 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
 293 {
 294         spin_lock_init(&sbi->fsync_node_lock);
 295         INIT_LIST_HEAD(&sbi->fsync_node_list);
 296         sbi->fsync_seg_id = 0;
 297         sbi->fsync_node_num = 0;
 298 }
 299
 300 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
 301                                                         struct page *page)
 302 {
 303         struct fsync_node_entry *fn;
 304         unsigned long flags;
 305         unsigned int seq_id;
 306
 307         fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
 308
 309         get_page(page);
 310         fn->page = page;
 311         INIT_LIST_HEAD(&fn->list);
 312
 313         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 314         list_add_tail(&fn->list, &sbi->fsync_node_list);
 315         fn->seq_id = sbi->fsync_seg_id++;
 316         seq_id = fn->seq_id;
 317         sbi->fsync_node_num++;
 318         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 319
 320         return seq_id;
 321 }
 322
 323 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
 324 {
 325         struct fsync_node_entry *fn;
 326         unsigned long flags;
 327
 328         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 329         list_for_each_entry(fn, &sbi->fsync_node_list, list) {
 330                 if (fn->page == page) {
 331                         list_del(&fn->list);
 332                         sbi->fsync_node_num--;
 333                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 334                         kmem_cache_free(fsync_node_entry_slab, fn);
 335                         put_page(page);
 336                         return;
 337                 }
 338         }
 339         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 340         f2fs_bug_on(sbi, 1);
 341 }
 342
 343 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
 344 {
 345         unsigned long flags;
 346
 347         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 348         sbi->fsync_seg_id = 0;
 349         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 350 }
 351
 352 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 353 {
 354         struct f2fs_nm_info *nm_i = NM_I(sbi);
 355         struct nat_entry *e;
 356         bool need = false;
 357
 358         down_read(&nm_i->nat_tree_lock);
 359         e = __lookup_nat_cache(nm_i, nid);
 360         if (e) {
 361                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 362                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 363                         need = true;
 364         }
 365         up_read(&nm_i->nat_tree_lock);
 366         return need;
 367 }
 368
 369 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 370 {
 371         struct f2fs_nm_info *nm_i = NM_I(sbi);
 372         struct nat_entry *e;
 373         bool is_cp = true;
 374
 375         down_read(&nm_i->nat_tree_lock);
 376         e = __lookup_nat_cache(nm_i, nid);
 377         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 378                 is_cp = false;
 379         up_read(&nm_i->nat_tree_lock);
 380         return is_cp;
 381 }
 382
 383 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 384 {
 385         struct f2fs_nm_info *nm_i = NM_I(sbi);
 386         struct nat_entry *e;
 387         bool need_update = true;
 388
 389         down_read(&nm_i->nat_tree_lock);
 390         e = __lookup_nat_cache(nm_i, ino);
 391         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 392                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 393                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 394                 need_update = false;
 395         up_read(&nm_i->nat_tree_lock);
 396         return need_update;
 397 }
 398
 399 /* must be locked by nat_tree_lock */
 400 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 401                                                 struct f2fs_nat_entry *ne)
 402 {
 403         struct f2fs_nm_info *nm_i = NM_I(sbi);
 404         struct nat_entry *new, *e;
 405
 406         new = __alloc_nat_entry(nid, false);
 407         if (!new)
 408                 return;
 409
 410         down_write(&nm_i->nat_tree_lock);
 411         e = __lookup_nat_cache(nm_i, nid);
 412         if (!e)
 413                 e = __init_nat_entry(nm_i, new, ne, false);
 414         else
 415                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 416                                 nat_get_blkaddr(e) !=
 417                                         le32_to_cpu(ne->block_addr) ||
 418                                 nat_get_version(e) != ne->version);
 419         up_write(&nm_i->nat_tree_lock);
 420         if (e != new)
 421                 __free_nat_entry(new);
 422 }
 423
 424 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 425                         block_t new_blkaddr, bool fsync_done)
 426 {
 427         struct f2fs_nm_info *nm_i = NM_I(sbi);
 428         struct nat_entry *e;
 429         struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
 430
 431         down_write(&nm_i->nat_tree_lock);
 432         e = __lookup_nat_cache(nm_i, ni->nid);
 433         if (!e) {
 434                 e = __init_nat_entry(nm_i, new, NULL, true);
 435                 copy_node_info(&e->ni, ni);
 436                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 437         } else if (new_blkaddr == NEW_ADDR) {
 438                 /*
 439                  * when nid is reallocated,
 440                  * previous nat entry can be remained in nat cache.
 441                  * So, reinitialize it with new information.
 442                  */
 443                 copy_node_info(&e->ni, ni);
 444                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 445         }
 446         /* let's free early to reduce memory consumption */
 447         if (e != new)
 448                 __free_nat_entry(new);
 449
 450         /* sanity check */
 451         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 452         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 453                         new_blkaddr == NULL_ADDR);
 454         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 455                         new_blkaddr == NEW_ADDR);
 456         f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
 457                         new_blkaddr == NEW_ADDR);
 458
 459         /* increment version no as node is removed */
 460         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 461                 unsigned char version = nat_get_version(e);
 462                 nat_set_version(e, inc_node_version(version));
 463         }
 464
 465         /* change address */
 466         nat_set_blkaddr(e, new_blkaddr);
 467         if (!__is_valid_data_blkaddr(new_blkaddr))
 468                 set_nat_flag(e, IS_CHECKPOINTED, false);
 469         __set_nat_cache_dirty(nm_i, e);
 470
 471         /* update fsync_mark if its inode nat entry is still alive */
 472         if (ni->nid != ni->ino)
 473                 e = __lookup_nat_cache(nm_i, ni->ino);
 474         if (e) {
 475                 if (fsync_done && ni->nid == ni->ino)
 476                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 477                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 478         }
 479         up_write(&nm_i->nat_tree_lock);
 480 }
 481
 482 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 483 {
 484         struct f2fs_nm_info *nm_i = NM_I(sbi);
 485         int nr = nr_shrink;
 486
 487         if (!down_write_trylock(&nm_i->nat_tree_lock))
 488                 return 0;
 489
 490         spin_lock(&nm_i->nat_list_lock);
 491         while (nr_shrink) {
 492                 struct nat_entry *ne;
 493
 494                 if (list_empty(&nm_i->nat_entries))
 495                         break;
 496
 497                 ne = list_first_entry(&nm_i->nat_entries,
 498                                         struct nat_entry, list);
 499                 list_del(&ne->list);
 500                 spin_unlock(&nm_i->nat_list_lock);
 501
 502                 __del_from_nat_cache(nm_i, ne);
 503                 nr_shrink--;
 504
 505                 spin_lock(&nm_i->nat_list_lock);
 506         }
 507         spin_unlock(&nm_i->nat_list_lock);
 508
 509         up_write(&nm_i->nat_tree_lock);
 510         return nr - nr_shrink;
 511 }
 512
 513 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
 514                                                 struct node_info *ni)
 515 {
 516         struct f2fs_nm_info *nm_i = NM_I(sbi);
 517         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 518         struct f2fs_journal *journal = curseg->journal;
 519         nid_t start_nid = START_NID(nid);
 520         struct f2fs_nat_block *nat_blk;
 521         struct page *page = NULL;
 522         struct f2fs_nat_entry ne;
 523         struct nat_entry *e;
 524         pgoff_t index;
 525         block_t blkaddr;
 526         int i;
 527
 528         ni->nid = nid;
 529
 530         /* Check nat cache */
 531         down_read(&nm_i->nat_tree_lock);
 532         e = __lookup_nat_cache(nm_i, nid);
 533         if (e) {
 534                 ni->ino = nat_get_ino(e);
 535                 ni->blk_addr = nat_get_blkaddr(e);
 536                 ni->version = nat_get_version(e);
 537                 up_read(&nm_i->nat_tree_lock);
 538                 return 0;
 539         }
 540
 541         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 542
 543         /* Check current segment summary */
 544         down_read(&curseg->journal_rwsem);
 545         i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 546         if (i >= 0) {
 547                 ne = nat_in_journal(journal, i);
 548                 node_info_from_raw_nat(ni, &ne);
 549         }
 550         up_read(&curseg->journal_rwsem);
 551         if (i >= 0) {
 552                 up_read(&nm_i->nat_tree_lock);
 553                 goto cache;
 554         }
 555
 556         /* Fill node_info from nat page */
 557         index = current_nat_addr(sbi, nid);
 558         up_read(&nm_i->nat_tree_lock);
 559
 560         page = f2fs_get_meta_page(sbi, index);
 561         if (IS_ERR(page))
 562                 return PTR_ERR(page);
 563
 564         nat_blk = (struct f2fs_nat_block *)page_address(page);
 565         ne = nat_blk->entries[nid - start_nid];
 566         node_info_from_raw_nat(ni, &ne);
 567         f2fs_put_page(page, 1);
 568 cache:
 569         blkaddr = le32_to_cpu(ne.block_addr);
 570         if (__is_valid_data_blkaddr(blkaddr) &&
 571                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
 572                 return -EFAULT;
 573
 574         /* cache nat entry */
 575         cache_nat_entry(sbi, nid, &ne);
 576         return 0;
 577 }
 578
 579 /*
 580  * readahead MAX_RA_NODE number of node pages.
 581  */
 582 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
 583 {
 584         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 585         struct blk_plug plug;
 586         int i, end;
 587         nid_t nid;
 588
 589         blk_start_plug(&plug);
 590
 591         /* Then, try readahead for siblings of the desired node */
 592         end = start + n;
 593         end = min(end, NIDS_PER_BLOCK);
 594         for (i = start; i < end; i++) {
 595                 nid = get_nid(parent, i, false);
 596                 f2fs_ra_node_page(sbi, nid);
 597         }
 598
 599         blk_finish_plug(&plug);
 600 }
 601
 602 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 603 {
 604         const long direct_index = ADDRS_PER_INODE(dn->inode);
 605         const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
 606         const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
 607         unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
 608         int cur_level = dn->cur_level;
 609         int max_level = dn->max_level;
 610         pgoff_t base = 0;
 611
 612         if (!dn->max_level)
 613                 return pgofs + 1;
 614
 615         while (max_level-- > cur_level)
 616                 skipped_unit *= NIDS_PER_BLOCK;
 617
 618         switch (dn->max_level) {
 619         case 3:
 620                 base += 2 * indirect_blks;
 621                 /* fall through */
 622         case 2:
 623                 base += 2 * direct_blks;
 624                 /* fall through */
 625         case 1:
 626                 base += direct_index;
 627                 break;
 628         default:
 629                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 630         }
 631
 632         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 633 }
 634
 635 /*
 636  * The maximum depth is four.
 637  * Offset[0] will have raw inode offset.
 638  */
 639 static int get_node_path(struct inode *inode, long block,
 640                                 int offset[4], unsigned int noffset[4])
 641 {
 642         const long direct_index = ADDRS_PER_INODE(inode);
 643         const long direct_blks = ADDRS_PER_BLOCK(inode);
 644         const long dptrs_per_blk = NIDS_PER_BLOCK;
 645         const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
 646         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 647         int n = 0;
 648         int level = 0;
 649
 650         noffset[0] = 0;
 651
 652         if (block < direct_index) {
 653                 offset[n] = block;
 654                 goto got;
 655         }
 656         block -= direct_index;
 657         if (block < direct_blks) {
 658                 offset[n++] = NODE_DIR1_BLOCK;
 659                 noffset[n] = 1;
 660                 offset[n] = block;
 661                 level = 1;
 662                 goto got;
 663         }
 664         block -= direct_blks;
 665         if (block < direct_blks) {
 666                 offset[n++] = NODE_DIR2_BLOCK;
 667                 noffset[n] = 2;
 668                 offset[n] = block;
 669                 level = 1;
 670                 goto got;
 671         }
 672         block -= direct_blks;
 673         if (block < indirect_blks) {
 674                 offset[n++] = NODE_IND1_BLOCK;
 675                 noffset[n] = 3;
 676                 offset[n++] = block / direct_blks;
 677                 noffset[n] = 4 + offset[n - 1];
 678                 offset[n] = block % direct_blks;
 679                 level = 2;
 680                 goto got;
 681         }
 682         block -= indirect_blks;
 683         if (block < indirect_blks) {
 684                 offset[n++] = NODE_IND2_BLOCK;
 685                 noffset[n] = 4 + dptrs_per_blk;
 686                 offset[n++] = block / direct_blks;
 687                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 688                 offset[n] = block % direct_blks;
 689                 level = 2;
 690                 goto got;
 691         }
 692         block -= indirect_blks;
 693         if (block < dindirect_blks) {
 694                 offset[n++] = NODE_DIND_BLOCK;
 695                 noffset[n] = 5 + (dptrs_per_blk * 2);
 696                 offset[n++] = block / indirect_blks;
 697                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 698                               offset[n - 1] * (dptrs_per_blk + 1);
 699                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 700                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 701                               offset[n - 2] * (dptrs_per_blk + 1) +
 702                               offset[n - 1];
 703                 offset[n] = block % direct_blks;
 704                 level = 3;
 705                 goto got;
 706         } else {
 707                 return -E2BIG;
 708         }
 709 got:
 710         return level;
 711 }
 712
 713 /*
 714  * Caller should call f2fs_put_dnode(dn).
 715  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 716  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
 717  */
 718 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 719 {
 720         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 721         struct page *npage[4];
 722         struct page *parent = NULL;
 723         int offset[4];
 724         unsigned int noffset[4];
 725         nid_t nids[4];
 726         int level, i = 0;
 727         int err = 0;
 728
 729         level = get_node_path(dn->inode, index, offset, noffset);
 730         if (level < 0)
 731                 return level;
 732
 733         nids[0] = dn->inode->i_ino;
 734         npage[0] = dn->inode_page;
 735
 736         if (!npage[0]) {
 737                 npage[0] = f2fs_get_node_page(sbi, nids[0]);
 738                 if (IS_ERR(npage[0]))
 739                         return PTR_ERR(npage[0]);
 740         }
 741
 742         /* if inline_data is set, should not report any block indices */
 743         if (f2fs_has_inline_data(dn->inode) && index) {
 744                 err = -ENOENT;
 745                 f2fs_put_page(npage[0], 1);
 746                 goto release_out;
 747         }
 748
 749         parent = npage[0];
 750         if (level != 0)
 751                 nids[1] = get_nid(parent, offset[0], true);
 752         dn->inode_page = npage[0];
 753         dn->inode_page_locked = true;
 754
 755         /* get indirect or direct nodes */
 756         for (i = 1; i <= level; i++) {
 757                 bool done = false;
 758
 759                 if (!nids[i] && mode == ALLOC_NODE) {
 760                         /* alloc new node */
 761                         if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
 762                                 err = -ENOSPC;
 763                                 goto release_pages;
 764                         }
 765
 766                         dn->nid = nids[i];
 767                         npage[i] = f2fs_new_node_page(dn, noffset[i]);
 768                         if (IS_ERR(npage[i])) {
 769                                 f2fs_alloc_nid_failed(sbi, nids[i]);
 770                                 err = PTR_ERR(npage[i]);
 771                                 goto release_pages;
 772                         }
 773
 774                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 775                         f2fs_alloc_nid_done(sbi, nids[i]);
 776                         done = true;
 777                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 778                         npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
 779                         if (IS_ERR(npage[i])) {
 780                                 err = PTR_ERR(npage[i]);
 781                                 goto release_pages;
 782                         }
 783                         done = true;
 784                 }
 785                 if (i == 1) {
 786                         dn->inode_page_locked = false;
 787                         unlock_page(parent);
 788                 } else {
 789                         f2fs_put_page(parent, 1);
 790                 }
 791
 792                 if (!done) {
 793                         npage[i] = f2fs_get_node_page(sbi, nids[i]);
 794                         if (IS_ERR(npage[i])) {
 795                                 err = PTR_ERR(npage[i]);
 796                                 f2fs_put_page(npage[0], 0);
 797                                 goto release_out;
 798                         }
 799                 }
 800                 if (i < level) {
 801                         parent = npage[i];
 802                         nids[i + 1] = get_nid(parent, offset[i], false);
 803                 }
 804         }
 805         dn->nid = nids[level];
 806         dn->ofs_in_node = offset[level];
 807         dn->node_page = npage[level];
 808         dn->data_blkaddr = f2fs_data_blkaddr(dn);
 809         return 0;
 810
 811 release_pages:
 812         f2fs_put_page(parent, 1);
 813         if (i > 1)
 814                 f2fs_put_page(npage[0], 0);
 815 release_out:
 816         dn->inode_page = NULL;
 817         dn->node_page = NULL;
 818         if (err == -ENOENT) {
 819                 dn->cur_level = i;
 820                 dn->max_level = level;
 821                 dn->ofs_in_node = offset[level];
 822         }
 823         return err;
 824 }
 825
 826 static int truncate_node(struct dnode_of_data *dn)
 827 {
 828         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 829         struct node_info ni;
 830         int err;
 831         pgoff_t index;
 832
 833         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 834         if (err)
 835                 return err;
 836
 837         /* Deallocate node address */
 838         f2fs_invalidate_blocks(sbi, ni.blk_addr);
 839         dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
 840         set_node_addr(sbi, &ni, NULL_ADDR, false);
 841
 842         if (dn->nid == dn->inode->i_ino) {
 843                 f2fs_remove_orphan_inode(sbi, dn->nid);
 844                 dec_valid_inode_count(sbi);
 845                 f2fs_inode_synced(dn->inode);
 846         }
 847
 848         clear_node_page_dirty(dn->node_page);
 849         set_sbi_flag(sbi, SBI_IS_DIRTY);
 850
 851         index = dn->node_page->index;
 852         f2fs_put_page(dn->node_page, 1);
 853
 854         invalidate_mapping_pages(NODE_MAPPING(sbi),
 855                         index, index);
 856
 857         dn->node_page = NULL;
 858         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 859
 860         return 0;
 861 }
 862
 863 static int truncate_dnode(struct dnode_of_data *dn)
 864 {
 865         struct page *page;
 866         int err;
 867
 868         if (dn->nid == 0)
 869                 return 1;
 870
 871         /* get direct node */
 872         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 873         if (PTR_ERR(page) == -ENOENT)
 874                 return 1;
 875         else if (IS_ERR(page))
 876                 return PTR_ERR(page);
 877
 878         /* Make dnode_of_data for parameter */
 879         dn->node_page = page;
 880         dn->ofs_in_node = 0;
 881         f2fs_truncate_data_blocks(dn);
 882         err = truncate_node(dn);
 883         if (err)
 884                 return err;
 885
 886         return 1;
 887 }
 888
 889 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 890                                                 int ofs, int depth)
 891 {
 892         struct dnode_of_data rdn = *dn;
 893         struct page *page;
 894         struct f2fs_node *rn;
 895         nid_t child_nid;
 896         unsigned int child_nofs;
 897         int freed = 0;
 898         int i, ret;
 899
 900         if (dn->nid == 0)
 901                 return NIDS_PER_BLOCK + 1;
 902
 903         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 904
 905         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 906         if (IS_ERR(page)) {
 907                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 908                 return PTR_ERR(page);
 909         }
 910
 911         f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 912
 913         rn = F2FS_NODE(page);
 914         if (depth < 3) {
 915                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 916                         child_nid = le32_to_cpu(rn->in.nid[i]);
 917                         if (child_nid == 0)
 918                                 continue;
 919                         rdn.nid = child_nid;
 920                         ret = truncate_dnode(&rdn);
 921                         if (ret < 0)
 922                                 goto out_err;
 923                         if (set_nid(page, i, 0, false))
 924                                 dn->node_changed = true;
 925                 }
 926         } else {
 927                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 928                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 929                         child_nid = le32_to_cpu(rn->in.nid[i]);
 930                         if (child_nid == 0) {
 931                                 child_nofs += NIDS_PER_BLOCK + 1;
 932                                 continue;
 933                         }
 934                         rdn.nid = child_nid;
 935                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 936                         if (ret == (NIDS_PER_BLOCK + 1)) {
 937                                 if (set_nid(page, i, 0, false))
 938                                         dn->node_changed = true;
 939                                 child_nofs += ret;
 940                         } else if (ret < 0 && ret != -ENOENT) {
 941                                 goto out_err;
 942                         }
 943                 }
 944                 freed = child_nofs;
 945         }
 946
 947         if (!ofs) {
 948                 /* remove current indirect node */
 949                 dn->node_page = page;
 950                 ret = truncate_node(dn);
 951                 if (ret)
 952                         goto out_err;
 953                 freed++;
 954         } else {
 955                 f2fs_put_page(page, 1);
 956         }
 957         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 958         return freed;
 959
 960 out_err:
 961         f2fs_put_page(page, 1);
 962         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 963         return ret;
 964 }
 965
 966 static int truncate_partial_nodes(struct dnode_of_data *dn,
 967                         struct f2fs_inode *ri, int *offset, int depth)
 968 {
 969         struct page *pages[2];
 970         nid_t nid[3];
 971         nid_t child_nid;
 972         int err = 0;
 973         int i;
 974         int idx = depth - 2;
 975
 976         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 977         if (!nid[0])
 978                 return 0;
 979
 980         /* get indirect nodes in the path */
 981         for (i = 0; i < idx + 1; i++) {
 982                 /* reference count'll be increased */
 983                 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 984                 if (IS_ERR(pages[i])) {
 985                         err = PTR_ERR(pages[i]);
 986                         idx = i - 1;
 987                         goto fail;
 988                 }
 989                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 990         }
 991
 992         f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 993
 994         /* free direct nodes linked to a partial indirect node */
 995         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 996                 child_nid = get_nid(pages[idx], i, false);
 997                 if (!child_nid)
 998                         continue;
 999                 dn->nid = child_nid;
1000                 err = truncate_dnode(dn);
1001                 if (err < 0)
1002                         goto fail;
1003                 if (set_nid(pages[idx], i, 0, false))
1004                         dn->node_changed = true;
1005         }
1006
1007         if (offset[idx + 1] == 0) {
1008                 dn->node_page = pages[idx];
1009                 dn->nid = nid[idx];
1010                 err = truncate_node(dn);
1011                 if (err)
1012                         goto fail;
1013         } else {
1014                 f2fs_put_page(pages[idx], 1);
1015         }
1016         offset[idx]++;
1017         offset[idx + 1] = 0;
1018         idx--;
1019 fail:
1020         for (i = idx; i >= 0; i--)
1021                 f2fs_put_page(pages[i], 1);
1022
1023         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1024
1025         return err;
1026 }
1027
1028 /*
1029  * All the block addresses of data and nodes should be nullified.
1030  */
1031 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1032 {
1033         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1034         int err = 0, cont = 1;
1035         int level, offset[4], noffset[4];
1036         unsigned int nofs = 0;
1037         struct f2fs_inode *ri;
1038         struct dnode_of_data dn;
1039         struct page *page;
1040
1041         trace_f2fs_truncate_inode_blocks_enter(inode, from);
1042
1043         level = get_node_path(inode, from, offset, noffset);
1044         if (level < 0)
1045                 return level;
1046
1047         page = f2fs_get_node_page(sbi, inode->i_ino);
1048         if (IS_ERR(page)) {
1049                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1050                 return PTR_ERR(page);
1051         }
1052
1053         set_new_dnode(&dn, inode, page, NULL, 0);
1054         unlock_page(page);
1055
1056         ri = F2FS_INODE(page);
1057         switch (level) {
1058         case 0:
1059         case 1:
1060                 nofs = noffset[1];
1061                 break;
1062         case 2:
1063                 nofs = noffset[1];
1064                 if (!offset[level - 1])
1065                         goto skip_partial;
1066                 err = truncate_partial_nodes(&dn, ri, offset, level);
1067                 if (err < 0 && err != -ENOENT)
1068                         goto fail;
1069                 nofs += 1 + NIDS_PER_BLOCK;
1070                 break;
1071         case 3:
1072                 nofs = 5 + 2 * NIDS_PER_BLOCK;
1073                 if (!offset[level - 1])
1074                         goto skip_partial;
1075                 err = truncate_partial_nodes(&dn, ri, offset, level);
1076                 if (err < 0 && err != -ENOENT)
1077                         goto fail;
1078                 break;
1079         default:
1080                 BUG();
1081         }
1082
1083 skip_partial:
1084         while (cont) {
1085                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1086                 switch (offset[0]) {
1087                 case NODE_DIR1_BLOCK:
1088                 case NODE_DIR2_BLOCK:
1089                         err = truncate_dnode(&dn);
1090                         break;
1091
1092                 case NODE_IND1_BLOCK:
1093                 case NODE_IND2_BLOCK:
1094                         err = truncate_nodes(&dn, nofs, offset[1], 2);
1095                         break;
1096
1097                 case NODE_DIND_BLOCK:
1098                         err = truncate_nodes(&dn, nofs, offset[1], 3);
1099                         cont = 0;
1100                         break;
1101
1102                 default:
1103                         BUG();
1104                 }
1105                 if (err < 0 && err != -ENOENT)
1106                         goto fail;
1107                 if (offset[1] == 0 &&
1108                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1109                         lock_page(page);
1110                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
1111                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1112                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1113                         set_page_dirty(page);
1114                         unlock_page(page);
1115                 }
1116                 offset[1] = 0;
1117                 offset[0]++;
1118                 nofs += err;
1119         }
1120 fail:
1121         f2fs_put_page(page, 0);
1122         trace_f2fs_truncate_inode_blocks_exit(inode, err);
1123         return err > 0 ? 0 : err;
1124 }
1125
1126 /* caller must lock inode page */
1127 int f2fs_truncate_xattr_node(struct inode *inode)
1128 {
1129         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1130         nid_t nid = F2FS_I(inode)->i_xattr_nid;
1131         struct dnode_of_data dn;
1132         struct page *npage;
1133         int err;
1134
1135         if (!nid)
1136                 return 0;
1137
1138         npage = f2fs_get_node_page(sbi, nid);
1139         if (IS_ERR(npage))
1140                 return PTR_ERR(npage);
1141
1142         set_new_dnode(&dn, inode, NULL, npage, nid);
1143         err = truncate_node(&dn);
1144         if (err) {
1145                 f2fs_put_page(npage, 1);
1146                 return err;
1147         }
1148
1149         f2fs_i_xnid_write(inode, 0);
1150
1151         return 0;
1152 }
1153
1154 /*
1155  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1156  * f2fs_unlock_op().
1157  */
1158 int f2fs_remove_inode_page(struct inode *inode)
1159 {
1160         struct dnode_of_data dn;
1161         int err;
1162
1163         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1164         err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1165         if (err)
1166                 return err;
1167
1168         err = f2fs_truncate_xattr_node(inode);
1169         if (err) {
1170                 f2fs_put_dnode(&dn);
1171                 return err;
1172         }
1173
1174         /* remove potential inline_data blocks */
1175         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1176                                 S_ISLNK(inode->i_mode))
1177                 f2fs_truncate_data_blocks_range(&dn, 1);
1178
1179         /* 0 is possible, after f2fs_new_inode() has failed */
1180         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1181                 f2fs_put_dnode(&dn);
1182                 return -EIO;
1183         }
1184
1185         if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1186                 f2fs_warn(F2FS_I_SB(inode),
1187                         "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1188                         inode->i_ino, (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, 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
1304         err = f2fs_submit_page_bio(&fio);
1305
1306         if (!err)
1307                 f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1308
1309         return err;
1310 }
1311
1312 /*
1313  * Readahead a node page
1314  */
1315 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1316 {
1317         struct page *apage;
1318         int err;
1319
1320         if (!nid)
1321                 return;
1322         if (f2fs_check_nid_range(sbi, nid))
1323                 return;
1324
1325         apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1326         if (apage)
1327                 return;
1328
1329         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1330         if (!apage)
1331                 return;
1332
1333         err = read_node_page(apage, REQ_RAHEAD);
1334         f2fs_put_page(apage, err ? 1 : 0);
1335 }
1336
1337 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1338                                         struct page *parent, int start)
1339 {
1340         struct page *page;
1341         int err;
1342
1343         if (!nid)
1344                 return ERR_PTR(-ENOENT);
1345         if (f2fs_check_nid_range(sbi, nid))
1346                 return ERR_PTR(-EINVAL);
1347 repeat:
1348         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1349         if (!page)
1350                 return ERR_PTR(-ENOMEM);
1351
1352         err = read_node_page(page, 0);
1353         if (err < 0) {
1354                 f2fs_put_page(page, 1);
1355                 return ERR_PTR(err);
1356         } else if (err == LOCKED_PAGE) {
1357                 err = 0;
1358                 goto page_hit;
1359         }
1360
1361         if (parent)
1362                 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1363
1364         lock_page(page);
1365
1366         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1367                 f2fs_put_page(page, 1);
1368                 goto repeat;
1369         }
1370
1371         if (unlikely(!PageUptodate(page))) {
1372                 err = -EIO;
1373                 goto out_err;
1374         }
1375
1376         if (!f2fs_inode_chksum_verify(sbi, page)) {
1377                 err = -EFSBADCRC;
1378                 goto out_err;
1379         }
1380 page_hit:
1381         if(unlikely(nid != nid_of_node(page))) {
1382                 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1383                           nid, nid_of_node(page), ino_of_node(page),
1384                           ofs_of_node(page), cpver_of_node(page),
1385                           next_blkaddr_of_node(page));
1386                 err = -EINVAL;
1387 out_err:
1388                 ClearPageUptodate(page);
1389                 f2fs_put_page(page, 1);
1390                 return ERR_PTR(err);
1391         }
1392         return page;
1393 }
1394
1395 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1396 {
1397         return __get_node_page(sbi, nid, NULL, 0);
1398 }
1399
1400 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1401 {
1402         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1403         nid_t nid = get_nid(parent, start, false);
1404
1405         return __get_node_page(sbi, nid, parent, start);
1406 }
1407
1408 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1409 {
1410         struct inode *inode;
1411         struct page *page;
1412         int ret;
1413
1414         /* should flush inline_data before evict_inode */
1415         inode = ilookup(sbi->sb, ino);
1416         if (!inode)
1417                 return;
1418
1419         page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1420                                         FGP_LOCK|FGP_NOWAIT, 0);
1421         if (!page)
1422                 goto iput_out;
1423
1424         if (!PageUptodate(page))
1425                 goto page_out;
1426
1427         if (!PageDirty(page))
1428                 goto page_out;
1429
1430         if (!clear_page_dirty_for_io(page))
1431                 goto page_out;
1432
1433         ret = f2fs_write_inline_data(inode, page);
1434         inode_dec_dirty_pages(inode);
1435         f2fs_remove_dirty_inode(inode);
1436         if (ret)
1437                 set_page_dirty(page);
1438 page_out:
1439         f2fs_put_page(page, 1);
1440 iput_out:
1441         iput(inode);
1442 }
1443
1444 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1445 {
1446         pgoff_t index;
1447         struct pagevec pvec;
1448         struct page *last_page = NULL;
1449         int nr_pages;
1450
1451         pagevec_init(&pvec);
1452         index = 0;
1453
1454         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1455                                 PAGECACHE_TAG_DIRTY))) {
1456                 int i;
1457
1458                 for (i = 0; i < nr_pages; i++) {
1459                         struct page *page = pvec.pages[i];
1460
1461                         if (unlikely(f2fs_cp_error(sbi))) {
1462                                 f2fs_put_page(last_page, 0);
1463                                 pagevec_release(&pvec);
1464                                 return ERR_PTR(-EIO);
1465                         }
1466
1467                         if (!IS_DNODE(page) || !is_cold_node(page))
1468                                 continue;
1469                         if (ino_of_node(page) != ino)
1470                                 continue;
1471
1472                         lock_page(page);
1473
1474                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1475 continue_unlock:
1476                                 unlock_page(page);
1477                                 continue;
1478                         }
1479                         if (ino_of_node(page) != ino)
1480                                 goto continue_unlock;
1481
1482                         if (!PageDirty(page)) {
1483                                 /* someone wrote it for us */
1484                                 goto continue_unlock;
1485                         }
1486
1487                         if (last_page)
1488                                 f2fs_put_page(last_page, 0);
1489
1490                         get_page(page);
1491                         last_page = page;
1492                         unlock_page(page);
1493                 }
1494                 pagevec_release(&pvec);
1495                 cond_resched();
1496         }
1497         return last_page;
1498 }
1499
1500 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1501                                 struct writeback_control *wbc, bool do_balance,
1502                                 enum iostat_type io_type, unsigned int *seq_id)
1503 {
1504         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1505         nid_t nid;
1506         struct node_info ni;
1507         struct f2fs_io_info fio = {
1508                 .sbi = sbi,
1509                 .ino = ino_of_node(page),
1510                 .type = NODE,
1511                 .op = REQ_OP_WRITE,
1512                 .op_flags = wbc_to_write_flags(wbc),
1513                 .page = page,
1514                 .encrypted_page = NULL,
1515                 .submitted = false,
1516                 .io_type = io_type,
1517                 .io_wbc = wbc,
1518         };
1519         unsigned int seq;
1520
1521         trace_f2fs_writepage(page, NODE);
1522
1523         if (unlikely(f2fs_cp_error(sbi))) {
1524                 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
1525                         ClearPageUptodate(page);
1526                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1527                         unlock_page(page);
1528                         return 0;
1529                 }
1530                 goto redirty_out;
1531         }
1532
1533         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1534                 goto redirty_out;
1535
1536         if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1537                         wbc->sync_mode == WB_SYNC_NONE &&
1538                         IS_DNODE(page) && is_cold_node(page))
1539                 goto redirty_out;
1540
1541         /* get old block addr of this node page */
1542         nid = nid_of_node(page);
1543         f2fs_bug_on(sbi, page->index != nid);
1544
1545         if (f2fs_get_node_info(sbi, nid, &ni))
1546                 goto redirty_out;
1547
1548         if (wbc->for_reclaim) {
1549                 if (!down_read_trylock(&sbi->node_write))
1550                         goto redirty_out;
1551         } else {
1552                 down_read(&sbi->node_write);
1553         }
1554
1555         /* This page is already truncated */
1556         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1557                 ClearPageUptodate(page);
1558                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1559                 up_read(&sbi->node_write);
1560                 unlock_page(page);
1561                 return 0;
1562         }
1563
1564         if (__is_valid_data_blkaddr(ni.blk_addr) &&
1565                 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1566                                         DATA_GENERIC_ENHANCE)) {
1567                 up_read(&sbi->node_write);
1568                 goto redirty_out;
1569         }
1570
1571         if (atomic && !test_opt(sbi, NOBARRIER))
1572                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1573
1574         /* should add to global list before clearing PAGECACHE status */
1575         if (f2fs_in_warm_node_list(sbi, page)) {
1576                 seq = f2fs_add_fsync_node_entry(sbi, page);
1577                 if (seq_id)
1578                         *seq_id = seq;
1579         }
1580
1581         set_page_writeback(page);
1582         ClearPageError(page);
1583
1584         fio.old_blkaddr = ni.blk_addr;
1585         f2fs_do_write_node_page(nid, &fio);
1586         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1587         dec_page_count(sbi, F2FS_DIRTY_NODES);
1588         up_read(&sbi->node_write);
1589
1590         if (wbc->for_reclaim) {
1591                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1592                 submitted = NULL;
1593         }
1594
1595         unlock_page(page);
1596
1597         if (unlikely(f2fs_cp_error(sbi))) {
1598                 f2fs_submit_merged_write(sbi, NODE);
1599                 submitted = NULL;
1600         }
1601         if (submitted)
1602                 *submitted = fio.submitted;
1603
1604         if (do_balance)
1605                 f2fs_balance_fs(sbi, false);
1606         return 0;
1607
1608 redirty_out:
1609         redirty_page_for_writepage(wbc, page);
1610         return AOP_WRITEPAGE_ACTIVATE;
1611 }
1612
1613 int f2fs_move_node_page(struct page *node_page, int gc_type)
1614 {
1615         int err = 0;
1616
1617         if (gc_type == FG_GC) {
1618                 struct writeback_control wbc = {
1619                         .sync_mode = WB_SYNC_ALL,
1620                         .nr_to_write = 1,
1621                         .for_reclaim = 0,
1622                 };
1623
1624                 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1625
1626                 set_page_dirty(node_page);
1627
1628                 if (!clear_page_dirty_for_io(node_page)) {
1629                         err = -EAGAIN;
1630                         goto out_page;
1631                 }
1632
1633                 if (__write_node_page(node_page, false, NULL,
1634                                         &wbc, false, FS_GC_NODE_IO, NULL)) {
1635                         err = -EAGAIN;
1636                         unlock_page(node_page);
1637                 }
1638                 goto release_page;
1639         } else {
1640                 /* set page dirty and write it */
1641                 if (!PageWriteback(node_page))
1642                         set_page_dirty(node_page);
1643         }
1644 out_page:
1645         unlock_page(node_page);
1646 release_page:
1647         f2fs_put_page(node_page, 0);
1648         return err;
1649 }
1650
1651 static int f2fs_write_node_page(struct page *page,
1652                                 struct writeback_control *wbc)
1653 {
1654         return __write_node_page(page, false, NULL, wbc, false,
1655                                                 FS_NODE_IO, NULL);
1656 }
1657
1658 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1659                         struct writeback_control *wbc, bool atomic,
1660                         unsigned int *seq_id)
1661 {
1662         pgoff_t index;
1663         struct pagevec pvec;
1664         int ret = 0;
1665         struct page *last_page = NULL;
1666         bool marked = false;
1667         nid_t ino = inode->i_ino;
1668         int nr_pages;
1669         int nwritten = 0;
1670
1671         if (atomic) {
1672                 last_page = last_fsync_dnode(sbi, ino);
1673                 if (IS_ERR_OR_NULL(last_page))
1674                         return PTR_ERR_OR_ZERO(last_page);
1675         }
1676 retry:
1677         pagevec_init(&pvec);
1678         index = 0;
1679
1680         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1681                                 PAGECACHE_TAG_DIRTY))) {
1682                 int i;
1683
1684                 for (i = 0; i < nr_pages; i++) {
1685                         struct page *page = pvec.pages[i];
1686                         bool submitted = false;
1687
1688                         if (unlikely(f2fs_cp_error(sbi))) {
1689                                 f2fs_put_page(last_page, 0);
1690                                 pagevec_release(&pvec);
1691                                 ret = -EIO;
1692                                 goto out;
1693                         }
1694
1695                         if (!IS_DNODE(page) || !is_cold_node(page))
1696                                 continue;
1697                         if (ino_of_node(page) != ino)
1698                                 continue;
1699
1700                         lock_page(page);
1701
1702                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1703 continue_unlock:
1704                                 unlock_page(page);
1705                                 continue;
1706                         }
1707                         if (ino_of_node(page) != ino)
1708                                 goto continue_unlock;
1709
1710                         if (!PageDirty(page) && page != last_page) {
1711                                 /* someone wrote it for us */
1712                                 goto continue_unlock;
1713                         }
1714
1715                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1716
1717                         set_fsync_mark(page, 0);
1718                         set_dentry_mark(page, 0);
1719
1720                         if (!atomic || page == last_page) {
1721                                 set_fsync_mark(page, 1);
1722                                 if (IS_INODE(page)) {
1723                                         if (is_inode_flag_set(inode,
1724                                                                 FI_DIRTY_INODE))
1725                                                 f2fs_update_inode(inode, page);
1726                                         set_dentry_mark(page,
1727                                                 f2fs_need_dentry_mark(sbi, ino));
1728                                 }
1729                                 /*  may be written by other thread */
1730                                 if (!PageDirty(page))
1731                                         set_page_dirty(page);
1732                         }
1733
1734                         if (!clear_page_dirty_for_io(page))
1735                                 goto continue_unlock;
1736
1737                         ret = __write_node_page(page, atomic &&
1738                                                 page == last_page,
1739                                                 &submitted, wbc, true,
1740                                                 FS_NODE_IO, seq_id);
1741                         if (ret) {
1742                                 unlock_page(page);
1743                                 f2fs_put_page(last_page, 0);
1744                                 break;
1745                         } else if (submitted) {
1746                                 nwritten++;
1747                         }
1748
1749                         if (page == last_page) {
1750                                 f2fs_put_page(page, 0);
1751                                 marked = true;
1752                                 break;
1753                         }
1754                 }
1755                 pagevec_release(&pvec);
1756                 cond_resched();
1757
1758                 if (ret || marked)
1759                         break;
1760         }
1761         if (!ret && atomic && !marked) {
1762                 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1763                            ino, last_page->index);
1764                 lock_page(last_page);
1765                 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1766                 set_page_dirty(last_page);
1767                 unlock_page(last_page);
1768                 goto retry;
1769         }
1770 out:
1771         if (nwritten)
1772                 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1773         return ret ? -EIO: 0;
1774 }
1775
1776 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1777 {
1778         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1779         bool clean;
1780
1781         if (inode->i_ino != ino)
1782                 return 0;
1783
1784         if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1785                 return 0;
1786
1787         spin_lock(&sbi->inode_lock[DIRTY_META]);
1788         clean = list_empty(&F2FS_I(inode)->gdirty_list);
1789         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1790
1791         if (clean)
1792                 return 0;
1793
1794         inode = igrab(inode);
1795         if (!inode)
1796                 return 0;
1797         return 1;
1798 }
1799
1800 static bool flush_dirty_inode(struct page *page)
1801 {
1802         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1803         struct inode *inode;
1804         nid_t ino = ino_of_node(page);
1805
1806         inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1807         if (!inode)
1808                 return false;
1809
1810         f2fs_update_inode(inode, page);
1811         unlock_page(page);
1812
1813         iput(inode);
1814         return true;
1815 }
1816
1817 int f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1818 {
1819         pgoff_t index = 0;
1820         struct pagevec pvec;
1821         int nr_pages;
1822         int ret = 0;
1823
1824         pagevec_init(&pvec);
1825
1826         while ((nr_pages = pagevec_lookup_tag(&pvec,
1827                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1828                 int i;
1829
1830                 for (i = 0; i < nr_pages; i++) {
1831                         struct page *page = pvec.pages[i];
1832
1833                         if (!IS_DNODE(page))
1834                                 continue;
1835
1836                         lock_page(page);
1837
1838                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1839 continue_unlock:
1840                                 unlock_page(page);
1841                                 continue;
1842                         }
1843
1844                         if (!PageDirty(page)) {
1845                                 /* someone wrote it for us */
1846                                 goto continue_unlock;
1847                         }
1848
1849                         /* flush inline_data, if it's async context. */
1850                         if (is_inline_node(page)) {
1851                                 clear_inline_node(page);
1852                                 unlock_page(page);
1853                                 flush_inline_data(sbi, ino_of_node(page));
1854                                 continue;
1855                         }
1856                         unlock_page(page);
1857                 }
1858                 pagevec_release(&pvec);
1859                 cond_resched();
1860         }
1861         return ret;
1862 }
1863
1864 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1865                                 struct writeback_control *wbc,
1866                                 bool do_balance, enum iostat_type io_type)
1867 {
1868         pgoff_t index;
1869         struct pagevec pvec;
1870         int step = 0;
1871         int nwritten = 0;
1872         int ret = 0;
1873         int nr_pages, done = 0;
1874
1875         pagevec_init(&pvec);
1876
1877 next_step:
1878         index = 0;
1879
1880         while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1881                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1882                 int i;
1883
1884                 for (i = 0; i < nr_pages; i++) {
1885                         struct page *page = pvec.pages[i];
1886                         bool submitted = false;
1887                         bool may_dirty = true;
1888
1889                         /* give a priority to WB_SYNC threads */
1890                         if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1891                                         wbc->sync_mode == WB_SYNC_NONE) {
1892                                 done = 1;
1893                                 break;
1894                         }
1895
1896                         /*
1897                          * flushing sequence with step:
1898                          * 0. indirect nodes
1899                          * 1. dentry dnodes
1900                          * 2. file dnodes
1901                          */
1902                         if (step == 0 && IS_DNODE(page))
1903                                 continue;
1904                         if (step == 1 && (!IS_DNODE(page) ||
1905                                                 is_cold_node(page)))
1906                                 continue;
1907                         if (step == 2 && (!IS_DNODE(page) ||
1908                                                 !is_cold_node(page)))
1909                                 continue;
1910 lock_node:
1911                         if (wbc->sync_mode == WB_SYNC_ALL)
1912                                 lock_page(page);
1913                         else if (!trylock_page(page))
1914                                 continue;
1915
1916                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1917 continue_unlock:
1918                                 unlock_page(page);
1919                                 continue;
1920                         }
1921
1922                         if (!PageDirty(page)) {
1923                                 /* someone wrote it for us */
1924                                 goto continue_unlock;
1925                         }
1926
1927                         /* flush inline_data, if it's async context. */
1928                         if (do_balance && is_inline_node(page)) {
1929                                 clear_inline_node(page);
1930                                 unlock_page(page);
1931                                 flush_inline_data(sbi, ino_of_node(page));
1932                                 goto lock_node;
1933                         }
1934
1935                         /* flush dirty inode */
1936                         if (IS_INODE(page) && may_dirty) {
1937                                 may_dirty = false;
1938                                 if (flush_dirty_inode(page))
1939                                         goto lock_node;
1940                         }
1941
1942                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1943
1944                         if (!clear_page_dirty_for_io(page))
1945                                 goto continue_unlock;
1946
1947                         set_fsync_mark(page, 0);
1948                         set_dentry_mark(page, 0);
1949
1950                         ret = __write_node_page(page, false, &submitted,
1951                                                 wbc, do_balance, io_type, NULL);
1952                         if (ret)
1953                                 unlock_page(page);
1954                         else if (submitted)
1955                                 nwritten++;
1956
1957                         if (--wbc->nr_to_write == 0)
1958                                 break;
1959                 }
1960                 pagevec_release(&pvec);
1961                 cond_resched();
1962
1963                 if (wbc->nr_to_write == 0) {
1964                         step = 2;
1965                         break;
1966                 }
1967         }
1968
1969         if (step < 2) {
1970                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1971                                 wbc->sync_mode == WB_SYNC_NONE && step == 1)
1972                         goto out;
1973                 step++;
1974                 goto next_step;
1975         }
1976 out:
1977         if (nwritten)
1978                 f2fs_submit_merged_write(sbi, NODE);
1979
1980         if (unlikely(f2fs_cp_error(sbi)))
1981                 return -EIO;
1982         return ret;
1983 }
1984
1985 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1986                                                 unsigned int seq_id)
1987 {
1988         struct fsync_node_entry *fn;
1989         struct page *page;
1990         struct list_head *head = &sbi->fsync_node_list;
1991         unsigned long flags;
1992         unsigned int cur_seq_id = 0;
1993         int ret2, ret = 0;
1994
1995         while (seq_id && cur_seq_id < seq_id) {
1996                 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
1997                 if (list_empty(head)) {
1998                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1999                         break;
2000                 }
2001                 fn = list_first_entry(head, struct fsync_node_entry, list);
2002                 if (fn->seq_id > seq_id) {
2003                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2004                         break;
2005                 }
2006                 cur_seq_id = fn->seq_id;
2007                 page = fn->page;
2008                 get_page(page);
2009                 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2010
2011                 f2fs_wait_on_page_writeback(page, NODE, true, false);
2012                 if (TestClearPageError(page))
2013                         ret = -EIO;
2014
2015                 put_page(page);
2016
2017                 if (ret)
2018                         break;
2019         }
2020
2021         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2022         if (!ret)
2023                 ret = ret2;
2024
2025         return ret;
2026 }
2027
2028 static int f2fs_write_node_pages(struct address_space *mapping,
2029                             struct writeback_control *wbc)
2030 {
2031         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2032         struct blk_plug plug;
2033         long diff;
2034
2035         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2036                 goto skip_write;
2037
2038         /* balancing f2fs's metadata in background */
2039         f2fs_balance_fs_bg(sbi, true);
2040
2041         /* collect a number of dirty node pages and write together */
2042         if (wbc->sync_mode != WB_SYNC_ALL &&
2043                         get_pages(sbi, F2FS_DIRTY_NODES) <
2044                                         nr_pages_to_skip(sbi, NODE))
2045                 goto skip_write;
2046
2047         if (wbc->sync_mode == WB_SYNC_ALL)
2048                 atomic_inc(&sbi->wb_sync_req[NODE]);
2049         else if (atomic_read(&sbi->wb_sync_req[NODE]))
2050                 goto skip_write;
2051
2052         trace_f2fs_writepages(mapping->host, wbc, NODE);
2053
2054         diff = nr_pages_to_write(sbi, NODE, wbc);
2055         blk_start_plug(&plug);
2056         f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2057         blk_finish_plug(&plug);
2058         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2059
2060         if (wbc->sync_mode == WB_SYNC_ALL)
2061                 atomic_dec(&sbi->wb_sync_req[NODE]);
2062         return 0;
2063
2064 skip_write:
2065         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2066         trace_f2fs_writepages(mapping->host, wbc, NODE);
2067         return 0;
2068 }
2069
2070 static int f2fs_set_node_page_dirty(struct page *page)
2071 {
2072         trace_f2fs_set_page_dirty(page, NODE);
2073
2074         if (!PageUptodate(page))
2075                 SetPageUptodate(page);
2076 #ifdef CONFIG_F2FS_CHECK_FS
2077         if (IS_INODE(page))
2078                 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2079 #endif
2080         if (!PageDirty(page)) {
2081                 __set_page_dirty_nobuffers(page);
2082                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2083                 f2fs_set_page_private(page, 0);
2084                 f2fs_trace_pid(page);
2085                 return 1;
2086         }
2087         return 0;
2088 }
2089
2090 /*
2091  * Structure of the f2fs node operations
2092  */
2093 const struct address_space_operations f2fs_node_aops = {
2094         .writepage      = f2fs_write_node_page,
2095         .writepages     = f2fs_write_node_pages,
2096         .set_page_dirty = f2fs_set_node_page_dirty,
2097         .invalidatepage = f2fs_invalidate_page,
2098         .releasepage    = f2fs_release_page,
2099 #ifdef CONFIG_MIGRATION
2100         .migratepage    = f2fs_migrate_page,
2101 #endif
2102 };
2103
2104 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2105                                                 nid_t n)
2106 {
2107         return radix_tree_lookup(&nm_i->free_nid_root, n);
2108 }
2109
2110 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2111                         struct free_nid *i, enum nid_state state)
2112 {
2113         struct f2fs_nm_info *nm_i = NM_I(sbi);
2114
2115         int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2116         if (err)
2117                 return err;
2118
2119         f2fs_bug_on(sbi, state != i->state);
2120         nm_i->nid_cnt[state]++;
2121         if (state == FREE_NID)
2122                 list_add_tail(&i->list, &nm_i->free_nid_list);
2123         return 0;
2124 }
2125
2126 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2127                         struct free_nid *i, enum nid_state state)
2128 {
2129         struct f2fs_nm_info *nm_i = NM_I(sbi);
2130
2131         f2fs_bug_on(sbi, state != i->state);
2132         nm_i->nid_cnt[state]--;
2133         if (state == FREE_NID)
2134                 list_del(&i->list);
2135         radix_tree_delete(&nm_i->free_nid_root, i->nid);
2136 }
2137
2138 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2139                         enum nid_state org_state, enum nid_state dst_state)
2140 {
2141         struct f2fs_nm_info *nm_i = NM_I(sbi);
2142
2143         f2fs_bug_on(sbi, org_state != i->state);
2144         i->state = dst_state;
2145         nm_i->nid_cnt[org_state]--;
2146         nm_i->nid_cnt[dst_state]++;
2147
2148         switch (dst_state) {
2149         case PREALLOC_NID:
2150                 list_del(&i->list);
2151                 break;
2152         case FREE_NID:
2153                 list_add_tail(&i->list, &nm_i->free_nid_list);
2154                 break;
2155         default:
2156                 BUG_ON(1);
2157         }
2158 }
2159
2160 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2161                                                         bool set, bool build)
2162 {
2163         struct f2fs_nm_info *nm_i = NM_I(sbi);
2164         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2165         unsigned int nid_ofs = nid - START_NID(nid);
2166
2167         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2168                 return;
2169
2170         if (set) {
2171                 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2172                         return;
2173                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2174                 nm_i->free_nid_count[nat_ofs]++;
2175         } else {
2176                 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2177                         return;
2178                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2179                 if (!build)
2180                         nm_i->free_nid_count[nat_ofs]--;
2181         }
2182 }
2183
2184 /* return if the nid is recognized as free */
2185 static bool add_free_nid(struct f2fs_sb_info *sbi,
2186                                 nid_t nid, bool build, bool update)
2187 {
2188         struct f2fs_nm_info *nm_i = NM_I(sbi);
2189         struct free_nid *i, *e;
2190         struct nat_entry *ne;
2191         int err = -EINVAL;
2192         bool ret = false;
2193
2194         /* 0 nid should not be used */
2195         if (unlikely(nid == 0))
2196                 return false;
2197
2198         if (unlikely(f2fs_check_nid_range(sbi, nid)))
2199                 return false;
2200
2201         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2202         i->nid = nid;
2203         i->state = FREE_NID;
2204
2205         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2206
2207         spin_lock(&nm_i->nid_list_lock);
2208
2209         if (build) {
2210                 /*
2211                  *   Thread A             Thread B
2212                  *  - f2fs_create
2213                  *   - f2fs_new_inode
2214                  *    - f2fs_alloc_nid
2215                  *     - __insert_nid_to_list(PREALLOC_NID)
2216                  *                     - f2fs_balance_fs_bg
2217                  *                      - f2fs_build_free_nids
2218                  *                       - __f2fs_build_free_nids
2219                  *                        - scan_nat_page
2220                  *                         - add_free_nid
2221                  *                          - __lookup_nat_cache
2222                  *  - f2fs_add_link
2223                  *   - f2fs_init_inode_metadata
2224                  *    - f2fs_new_inode_page
2225                  *     - f2fs_new_node_page
2226                  *      - set_node_addr
2227                  *  - f2fs_alloc_nid_done
2228                  *   - __remove_nid_from_list(PREALLOC_NID)
2229                  *                         - __insert_nid_to_list(FREE_NID)
2230                  */
2231                 ne = __lookup_nat_cache(nm_i, nid);
2232                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2233                                 nat_get_blkaddr(ne) != NULL_ADDR))
2234                         goto err_out;
2235
2236                 e = __lookup_free_nid_list(nm_i, nid);
2237                 if (e) {
2238                         if (e->state == FREE_NID)
2239                                 ret = true;
2240                         goto err_out;
2241                 }
2242         }
2243         ret = true;
2244         err = __insert_free_nid(sbi, i, FREE_NID);
2245 err_out:
2246         if (update) {
2247                 update_free_nid_bitmap(sbi, nid, ret, build);
2248                 if (!build)
2249                         nm_i->available_nids++;
2250         }
2251         spin_unlock(&nm_i->nid_list_lock);
2252         radix_tree_preload_end();
2253
2254         if (err)
2255                 kmem_cache_free(free_nid_slab, i);
2256         return ret;
2257 }
2258
2259 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2260 {
2261         struct f2fs_nm_info *nm_i = NM_I(sbi);
2262         struct free_nid *i;
2263         bool need_free = false;
2264
2265         spin_lock(&nm_i->nid_list_lock);
2266         i = __lookup_free_nid_list(nm_i, nid);
2267         if (i && i->state == FREE_NID) {
2268                 __remove_free_nid(sbi, i, FREE_NID);
2269                 need_free = true;
2270         }
2271         spin_unlock(&nm_i->nid_list_lock);
2272
2273         if (need_free)
2274                 kmem_cache_free(free_nid_slab, i);
2275 }
2276
2277 static int scan_nat_page(struct f2fs_sb_info *sbi,
2278                         struct page *nat_page, nid_t start_nid)
2279 {
2280         struct f2fs_nm_info *nm_i = NM_I(sbi);
2281         struct f2fs_nat_block *nat_blk = page_address(nat_page);
2282         block_t blk_addr;
2283         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2284         int i;
2285
2286         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2287
2288         i = start_nid % NAT_ENTRY_PER_BLOCK;
2289
2290         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2291                 if (unlikely(start_nid >= nm_i->max_nid))
2292                         break;
2293
2294                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2295
2296                 if (blk_addr == NEW_ADDR)
2297                         return -EINVAL;
2298
2299                 if (blk_addr == NULL_ADDR) {
2300                         add_free_nid(sbi, start_nid, true, true);
2301                 } else {
2302                         spin_lock(&NM_I(sbi)->nid_list_lock);
2303                         update_free_nid_bitmap(sbi, start_nid, false, true);
2304                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2305                 }
2306         }
2307
2308         return 0;
2309 }
2310
2311 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2312 {
2313         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2314         struct f2fs_journal *journal = curseg->journal;
2315         int i;
2316
2317         down_read(&curseg->journal_rwsem);
2318         for (i = 0; i < nats_in_cursum(journal); i++) {
2319                 block_t addr;
2320                 nid_t nid;
2321
2322                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2323                 nid = le32_to_cpu(nid_in_journal(journal, i));
2324                 if (addr == NULL_ADDR)
2325                         add_free_nid(sbi, nid, true, false);
2326                 else
2327                         remove_free_nid(sbi, nid);
2328         }
2329         up_read(&curseg->journal_rwsem);
2330 }
2331
2332 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2333 {
2334         struct f2fs_nm_info *nm_i = NM_I(sbi);
2335         unsigned int i, idx;
2336         nid_t nid;
2337
2338         down_read(&nm_i->nat_tree_lock);
2339
2340         for (i = 0; i < nm_i->nat_blocks; i++) {
2341                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2342                         continue;
2343                 if (!nm_i->free_nid_count[i])
2344                         continue;
2345                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2346                         idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2347                                                 NAT_ENTRY_PER_BLOCK, idx);
2348                         if (idx >= NAT_ENTRY_PER_BLOCK)
2349                                 break;
2350
2351                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
2352                         add_free_nid(sbi, nid, true, false);
2353
2354                         if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2355                                 goto out;
2356                 }
2357         }
2358 out:
2359         scan_curseg_cache(sbi);
2360
2361         up_read(&nm_i->nat_tree_lock);
2362 }
2363
2364 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2365                                                 bool sync, bool mount)
2366 {
2367         struct f2fs_nm_info *nm_i = NM_I(sbi);
2368         int i = 0, ret;
2369         nid_t nid = nm_i->next_scan_nid;
2370
2371         if (unlikely(nid >= nm_i->max_nid))
2372                 nid = 0;
2373
2374         /* Enough entries */
2375         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2376                 return 0;
2377
2378         if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2379                 return 0;
2380
2381         if (!mount) {
2382                 /* try to find free nids in free_nid_bitmap */
2383                 scan_free_nid_bits(sbi);
2384
2385                 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2386                         return 0;
2387         }
2388
2389         /* readahead nat pages to be scanned */
2390         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2391                                                         META_NAT, true);
2392
2393         down_read(&nm_i->nat_tree_lock);
2394
2395         while (1) {
2396                 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2397                                                 nm_i->nat_block_bitmap)) {
2398                         struct page *page = get_current_nat_page(sbi, nid);
2399
2400                         if (IS_ERR(page)) {
2401                                 ret = PTR_ERR(page);
2402                         } else {
2403                                 ret = scan_nat_page(sbi, page, nid);
2404                                 f2fs_put_page(page, 1);
2405                         }
2406
2407                         if (ret) {
2408                                 up_read(&nm_i->nat_tree_lock);
2409                                 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2410                                 return ret;
2411                         }
2412                 }
2413
2414                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2415                 if (unlikely(nid >= nm_i->max_nid))
2416                         nid = 0;
2417
2418                 if (++i >= FREE_NID_PAGES)
2419                         break;
2420         }
2421
2422         /* go to the next free nat pages to find free nids abundantly */
2423         nm_i->next_scan_nid = nid;
2424
2425         /* find free nids from current sum_pages */
2426         scan_curseg_cache(sbi);
2427
2428         up_read(&nm_i->nat_tree_lock);
2429
2430         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2431                                         nm_i->ra_nid_pages, META_NAT, false);
2432
2433         return 0;
2434 }
2435
2436 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2437 {
2438         int ret;
2439
2440         mutex_lock(&NM_I(sbi)->build_lock);
2441         ret = __f2fs_build_free_nids(sbi, sync, mount);
2442         mutex_unlock(&NM_I(sbi)->build_lock);
2443
2444         return ret;
2445 }
2446
2447 /*
2448  * If this function returns success, caller can obtain a new nid
2449  * from second parameter of this function.
2450  * The returned nid could be used ino as well as nid when inode is created.
2451  */
2452 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2453 {
2454         struct f2fs_nm_info *nm_i = NM_I(sbi);
2455         struct free_nid *i = NULL;
2456 retry:
2457         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2458                 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2459                 return false;
2460         }
2461
2462         spin_lock(&nm_i->nid_list_lock);
2463
2464         if (unlikely(nm_i->available_nids == 0)) {
2465                 spin_unlock(&nm_i->nid_list_lock);
2466                 return false;
2467         }
2468
2469         /* We should not use stale free nids created by f2fs_build_free_nids */
2470         if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2471                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2472                 i = list_first_entry(&nm_i->free_nid_list,
2473                                         struct free_nid, list);
2474                 *nid = i->nid;
2475
2476                 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2477                 nm_i->available_nids--;
2478
2479                 update_free_nid_bitmap(sbi, *nid, false, false);
2480
2481                 spin_unlock(&nm_i->nid_list_lock);
2482                 return true;
2483         }
2484         spin_unlock(&nm_i->nid_list_lock);
2485
2486         /* Let's scan nat pages and its caches to get free nids */
2487         if (!f2fs_build_free_nids(sbi, true, false))
2488                 goto retry;
2489         return false;
2490 }
2491
2492 /*
2493  * f2fs_alloc_nid() should be called prior to this function.
2494  */
2495 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2496 {
2497         struct f2fs_nm_info *nm_i = NM_I(sbi);
2498         struct free_nid *i;
2499
2500         spin_lock(&nm_i->nid_list_lock);
2501         i = __lookup_free_nid_list(nm_i, nid);
2502         f2fs_bug_on(sbi, !i);
2503         __remove_free_nid(sbi, i, PREALLOC_NID);
2504         spin_unlock(&nm_i->nid_list_lock);
2505
2506         kmem_cache_free(free_nid_slab, i);
2507 }
2508
2509 /*
2510  * f2fs_alloc_nid() should be called prior to this function.
2511  */
2512 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2513 {
2514         struct f2fs_nm_info *nm_i = NM_I(sbi);
2515         struct free_nid *i;
2516         bool need_free = false;
2517
2518         if (!nid)
2519                 return;
2520
2521         spin_lock(&nm_i->nid_list_lock);
2522         i = __lookup_free_nid_list(nm_i, nid);
2523         f2fs_bug_on(sbi, !i);
2524
2525         if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2526                 __remove_free_nid(sbi, i, PREALLOC_NID);
2527                 need_free = true;
2528         } else {
2529                 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2530         }
2531
2532         nm_i->available_nids++;
2533
2534         update_free_nid_bitmap(sbi, nid, true, false);
2535
2536         spin_unlock(&nm_i->nid_list_lock);
2537
2538         if (need_free)
2539                 kmem_cache_free(free_nid_slab, i);
2540 }
2541
2542 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2543 {
2544         struct f2fs_nm_info *nm_i = NM_I(sbi);
2545         int nr = nr_shrink;
2546
2547         if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2548                 return 0;
2549
2550         if (!mutex_trylock(&nm_i->build_lock))
2551                 return 0;
2552
2553         while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2554                 struct free_nid *i, *next;
2555                 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2556
2557                 spin_lock(&nm_i->nid_list_lock);
2558                 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2559                         if (!nr_shrink || !batch ||
2560                                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2561                                 break;
2562                         __remove_free_nid(sbi, i, FREE_NID);
2563                         kmem_cache_free(free_nid_slab, i);
2564                         nr_shrink--;
2565                         batch--;
2566                 }
2567                 spin_unlock(&nm_i->nid_list_lock);
2568         }
2569
2570         mutex_unlock(&nm_i->build_lock);
2571
2572         return nr - nr_shrink;
2573 }
2574
2575 void f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2576 {
2577         void *src_addr, *dst_addr;
2578         size_t inline_size;
2579         struct page *ipage;
2580         struct f2fs_inode *ri;
2581
2582         ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2583         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2584
2585         ri = F2FS_INODE(page);
2586         if (ri->i_inline & F2FS_INLINE_XATTR) {
2587                 set_inode_flag(inode, FI_INLINE_XATTR);
2588         } else {
2589                 clear_inode_flag(inode, FI_INLINE_XATTR);
2590                 goto update_inode;
2591         }
2592
2593         dst_addr = inline_xattr_addr(inode, ipage);
2594         src_addr = inline_xattr_addr(inode, page);
2595         inline_size = inline_xattr_size(inode);
2596
2597         f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2598         memcpy(dst_addr, src_addr, inline_size);
2599 update_inode:
2600         f2fs_update_inode(inode, ipage);
2601         f2fs_put_page(ipage, 1);
2602 }
2603
2604 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2605 {
2606         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2607         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2608         nid_t new_xnid;
2609         struct dnode_of_data dn;
2610         struct node_info ni;
2611         struct page *xpage;
2612         int err;
2613
2614         if (!prev_xnid)
2615                 goto recover_xnid;
2616
2617         /* 1: invalidate the previous xattr nid */
2618         err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2619         if (err)
2620                 return err;
2621
2622         f2fs_invalidate_blocks(sbi, ni.blk_addr);
2623         dec_valid_node_count(sbi, inode, false);
2624         set_node_addr(sbi, &ni, NULL_ADDR, false);
2625
2626 recover_xnid:
2627         /* 2: update xattr nid in inode */
2628         if (!f2fs_alloc_nid(sbi, &new_xnid))
2629                 return -ENOSPC;
2630
2631         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2632         xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2633         if (IS_ERR(xpage)) {
2634                 f2fs_alloc_nid_failed(sbi, new_xnid);
2635                 return PTR_ERR(xpage);
2636         }
2637
2638         f2fs_alloc_nid_done(sbi, new_xnid);
2639         f2fs_update_inode_page(inode);
2640
2641         /* 3: update and set xattr node page dirty */
2642         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2643
2644         set_page_dirty(xpage);
2645         f2fs_put_page(xpage, 1);
2646
2647         return 0;
2648 }
2649
2650 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2651 {
2652         struct f2fs_inode *src, *dst;
2653         nid_t ino = ino_of_node(page);
2654         struct node_info old_ni, new_ni;
2655         struct page *ipage;
2656         int err;
2657
2658         err = f2fs_get_node_info(sbi, ino, &old_ni);
2659         if (err)
2660                 return err;
2661
2662         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2663                 return -EINVAL;
2664 retry:
2665         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2666         if (!ipage) {
2667                 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2668                 goto retry;
2669         }
2670
2671         /* Should not use this inode from free nid list */
2672         remove_free_nid(sbi, ino);
2673
2674         if (!PageUptodate(ipage))
2675                 SetPageUptodate(ipage);
2676         fill_node_footer(ipage, ino, ino, 0, true);
2677         set_cold_node(ipage, false);
2678
2679         src = F2FS_INODE(page);
2680         dst = F2FS_INODE(ipage);
2681
2682         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2683         dst->i_size = 0;
2684         dst->i_blocks = cpu_to_le64(1);
2685         dst->i_links = cpu_to_le32(1);
2686         dst->i_xattr_nid = 0;
2687         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2688         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2689                 dst->i_extra_isize = src->i_extra_isize;
2690
2691                 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2692                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2693                                                         i_inline_xattr_size))
2694                         dst->i_inline_xattr_size = src->i_inline_xattr_size;
2695
2696                 if (f2fs_sb_has_project_quota(sbi) &&
2697                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2698                                                                 i_projid))
2699                         dst->i_projid = src->i_projid;
2700
2701                 if (f2fs_sb_has_inode_crtime(sbi) &&
2702                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2703                                                         i_crtime_nsec)) {
2704                         dst->i_crtime = src->i_crtime;
2705                         dst->i_crtime_nsec = src->i_crtime_nsec;
2706                 }
2707         }
2708
2709         new_ni = old_ni;
2710         new_ni.ino = ino;
2711
2712         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2713                 WARN_ON(1);
2714         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2715         inc_valid_inode_count(sbi);
2716         set_page_dirty(ipage);
2717         f2fs_put_page(ipage, 1);
2718         return 0;
2719 }
2720
2721 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2722                         unsigned int segno, struct f2fs_summary_block *sum)
2723 {
2724         struct f2fs_node *rn;
2725         struct f2fs_summary *sum_entry;
2726         block_t addr;
2727         int i, idx, last_offset, nrpages;
2728
2729         /* scan the node segment */
2730         last_offset = sbi->blocks_per_seg;
2731         addr = START_BLOCK(sbi, segno);
2732         sum_entry = &sum->entries[0];
2733
2734         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2735                 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2736
2737                 /* readahead node pages */
2738                 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2739
2740                 for (idx = addr; idx < addr + nrpages; idx++) {
2741                         struct page *page = f2fs_get_tmp_page(sbi, idx);
2742
2743                         if (IS_ERR(page))
2744                                 return PTR_ERR(page);
2745
2746                         rn = F2FS_NODE(page);
2747                         sum_entry->nid = rn->footer.nid;
2748                         sum_entry->version = 0;
2749                         sum_entry->ofs_in_node = 0;
2750                         sum_entry++;
2751                         f2fs_put_page(page, 1);
2752                 }
2753
2754                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2755                                                         addr + nrpages);
2756         }
2757         return 0;
2758 }
2759
2760 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2761 {
2762         struct f2fs_nm_info *nm_i = NM_I(sbi);
2763         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2764         struct f2fs_journal *journal = curseg->journal;
2765         int i;
2766
2767         down_write(&curseg->journal_rwsem);
2768         for (i = 0; i < nats_in_cursum(journal); i++) {
2769                 struct nat_entry *ne;
2770                 struct f2fs_nat_entry raw_ne;
2771                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2772
2773                 raw_ne = nat_in_journal(journal, i);
2774
2775                 ne = __lookup_nat_cache(nm_i, nid);
2776                 if (!ne) {
2777                         ne = __alloc_nat_entry(nid, true);
2778                         __init_nat_entry(nm_i, ne, &raw_ne, true);
2779                 }
2780
2781                 /*
2782                  * if a free nat in journal has not been used after last
2783                  * checkpoint, we should remove it from available nids,
2784                  * since later we will add it again.
2785                  */
2786                 if (!get_nat_flag(ne, IS_DIRTY) &&
2787                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2788                         spin_lock(&nm_i->nid_list_lock);
2789                         nm_i->available_nids--;
2790                         spin_unlock(&nm_i->nid_list_lock);
2791                 }
2792
2793                 __set_nat_cache_dirty(nm_i, ne);
2794         }
2795         update_nats_in_cursum(journal, -i);
2796         up_write(&curseg->journal_rwsem);
2797 }
2798
2799 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2800                                                 struct list_head *head, int max)
2801 {
2802         struct nat_entry_set *cur;
2803
2804         if (nes->entry_cnt >= max)
2805                 goto add_out;
2806
2807         list_for_each_entry(cur, head, set_list) {
2808                 if (cur->entry_cnt >= nes->entry_cnt) {
2809                         list_add(&nes->set_list, cur->set_list.prev);
2810                         return;
2811                 }
2812         }
2813 add_out:
2814         list_add_tail(&nes->set_list, head);
2815 }
2816
2817 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2818                                                 struct page *page)
2819 {
2820         struct f2fs_nm_info *nm_i = NM_I(sbi);
2821         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2822         struct f2fs_nat_block *nat_blk = page_address(page);
2823         int valid = 0;
2824         int i = 0;
2825
2826         if (!enabled_nat_bits(sbi, NULL))
2827                 return;
2828
2829         if (nat_index == 0) {
2830                 valid = 1;
2831                 i = 1;
2832         }
2833         for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2834                 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2835                         valid++;
2836         }
2837         if (valid == 0) {
2838                 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2839                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2840                 return;
2841         }
2842
2843         __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2844         if (valid == NAT_ENTRY_PER_BLOCK)
2845                 __set_bit_le(nat_index, nm_i->full_nat_bits);
2846         else
2847                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2848 }
2849
2850 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2851                 struct nat_entry_set *set, struct cp_control *cpc)
2852 {
2853         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2854         struct f2fs_journal *journal = curseg->journal;
2855         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2856         bool to_journal = true;
2857         struct f2fs_nat_block *nat_blk;
2858         struct nat_entry *ne, *cur;
2859         struct page *page = NULL;
2860
2861         /*
2862          * there are two steps to flush nat entries:
2863          * #1, flush nat entries to journal in current hot data summary block.
2864          * #2, flush nat entries to nat page.
2865          */
2866         if (enabled_nat_bits(sbi, cpc) ||
2867                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2868                 to_journal = false;
2869
2870         if (to_journal) {
2871                 down_write(&curseg->journal_rwsem);
2872         } else {
2873                 page = get_next_nat_page(sbi, start_nid);
2874                 if (IS_ERR(page))
2875                         return PTR_ERR(page);
2876
2877                 nat_blk = page_address(page);
2878                 f2fs_bug_on(sbi, !nat_blk);
2879         }
2880
2881         /* flush dirty nats in nat entry set */
2882         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2883                 struct f2fs_nat_entry *raw_ne;
2884                 nid_t nid = nat_get_nid(ne);
2885                 int offset;
2886
2887                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2888
2889                 if (to_journal) {
2890                         offset = f2fs_lookup_journal_in_cursum(journal,
2891                                                         NAT_JOURNAL, nid, 1);
2892                         f2fs_bug_on(sbi, offset < 0);
2893                         raw_ne = &nat_in_journal(journal, offset);
2894                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2895                 } else {
2896                         raw_ne = &nat_blk->entries[nid - start_nid];
2897                 }
2898                 raw_nat_from_node_info(raw_ne, &ne->ni);
2899                 nat_reset_flag(ne);
2900                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2901                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2902                         add_free_nid(sbi, nid, false, true);
2903                 } else {
2904                         spin_lock(&NM_I(sbi)->nid_list_lock);
2905                         update_free_nid_bitmap(sbi, nid, false, false);
2906                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2907                 }
2908         }
2909
2910         if (to_journal) {
2911                 up_write(&curseg->journal_rwsem);
2912         } else {
2913                 __update_nat_bits(sbi, start_nid, page);
2914                 f2fs_put_page(page, 1);
2915         }
2916
2917         /* Allow dirty nats by node block allocation in write_begin */
2918         if (!set->entry_cnt) {
2919                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2920                 kmem_cache_free(nat_entry_set_slab, set);
2921         }
2922         return 0;
2923 }
2924
2925 /*
2926  * This function is called during the checkpointing process.
2927  */
2928 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2929 {
2930         struct f2fs_nm_info *nm_i = NM_I(sbi);
2931         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2932         struct f2fs_journal *journal = curseg->journal;
2933         struct nat_entry_set *setvec[SETVEC_SIZE];
2934         struct nat_entry_set *set, *tmp;
2935         unsigned int found;
2936         nid_t set_idx = 0;
2937         LIST_HEAD(sets);
2938         int err = 0;
2939
2940         /* during unmount, let's flush nat_bits before checking dirty_nat_cnt */
2941         if (enabled_nat_bits(sbi, cpc)) {
2942                 down_write(&nm_i->nat_tree_lock);
2943                 remove_nats_in_journal(sbi);
2944                 up_write(&nm_i->nat_tree_lock);
2945         }
2946
2947         if (!nm_i->dirty_nat_cnt)
2948                 return 0;
2949
2950         down_write(&nm_i->nat_tree_lock);
2951
2952         /*
2953          * if there are no enough space in journal to store dirty nat
2954          * entries, remove all entries from journal and merge them
2955          * into nat entry set.
2956          */
2957         if (enabled_nat_bits(sbi, cpc) ||
2958                 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2959                 remove_nats_in_journal(sbi);
2960
2961         while ((found = __gang_lookup_nat_set(nm_i,
2962                                         set_idx, SETVEC_SIZE, setvec))) {
2963                 unsigned idx;
2964                 set_idx = setvec[found - 1]->set + 1;
2965                 for (idx = 0; idx < found; idx++)
2966                         __adjust_nat_entry_set(setvec[idx], &sets,
2967                                                 MAX_NAT_JENTRIES(journal));
2968         }
2969
2970         /* flush dirty nats in nat entry set */
2971         list_for_each_entry_safe(set, tmp, &sets, set_list) {
2972                 err = __flush_nat_entry_set(sbi, set, cpc);
2973                 if (err)
2974                         break;
2975         }
2976
2977         up_write(&nm_i->nat_tree_lock);
2978         /* Allow dirty nats by node block allocation in write_begin */
2979
2980         return err;
2981 }
2982
2983 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2984 {
2985         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2986         struct f2fs_nm_info *nm_i = NM_I(sbi);
2987         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2988         unsigned int i;
2989         __u64 cp_ver = cur_cp_version(ckpt);
2990         block_t nat_bits_addr;
2991
2992         if (!enabled_nat_bits(sbi, NULL))
2993                 return 0;
2994
2995         nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2996         nm_i->nat_bits = f2fs_kvzalloc(sbi,
2997                         nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2998         if (!nm_i->nat_bits)
2999                 return -ENOMEM;
3000
3001         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3002                                                 nm_i->nat_bits_blocks;
3003         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3004                 struct page *page;
3005
3006                 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3007                 if (IS_ERR(page))
3008                         return PTR_ERR(page);
3009
3010                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3011                                         page_address(page), F2FS_BLKSIZE);
3012                 f2fs_put_page(page, 1);
3013         }
3014
3015         cp_ver |= (cur_cp_crc(ckpt) << 32);
3016         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3017                 disable_nat_bits(sbi, true);
3018                 return 0;
3019         }
3020
3021         nm_i->full_nat_bits = nm_i->nat_bits + 8;
3022         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3023
3024         f2fs_notice(sbi, "Found nat_bits in checkpoint");
3025         return 0;
3026 }
3027
3028 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3029 {
3030         struct f2fs_nm_info *nm_i = NM_I(sbi);
3031         unsigned int i = 0;
3032         nid_t nid, last_nid;
3033
3034         if (!enabled_nat_bits(sbi, NULL))
3035                 return;
3036
3037         for (i = 0; i < nm_i->nat_blocks; i++) {
3038                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3039                 if (i >= nm_i->nat_blocks)
3040                         break;
3041
3042                 __set_bit_le(i, nm_i->nat_block_bitmap);
3043
3044                 nid = i * NAT_ENTRY_PER_BLOCK;
3045                 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3046
3047                 spin_lock(&NM_I(sbi)->nid_list_lock);
3048                 for (; nid < last_nid; nid++)
3049                         update_free_nid_bitmap(sbi, nid, true, true);
3050                 spin_unlock(&NM_I(sbi)->nid_list_lock);
3051         }
3052
3053         for (i = 0; i < nm_i->nat_blocks; i++) {
3054                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3055                 if (i >= nm_i->nat_blocks)
3056                         break;
3057
3058                 __set_bit_le(i, nm_i->nat_block_bitmap);
3059         }
3060 }
3061
3062 static int init_node_manager(struct f2fs_sb_info *sbi)
3063 {
3064         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3065         struct f2fs_nm_info *nm_i = NM_I(sbi);
3066         unsigned char *version_bitmap;
3067         unsigned int nat_segs;
3068         int err;
3069
3070         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3071
3072         /* segment_count_nat includes pair segment so divide to 2. */
3073         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3074         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3075         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3076
3077         /* not used nids: 0, node, meta, (and root counted as valid node) */
3078         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3079                                                 F2FS_RESERVED_NODE_NUM;
3080         nm_i->nid_cnt[FREE_NID] = 0;
3081         nm_i->nid_cnt[PREALLOC_NID] = 0;
3082         nm_i->nat_cnt = 0;
3083         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3084         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3085         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3086
3087         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3088         INIT_LIST_HEAD(&nm_i->free_nid_list);
3089         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3090         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3091         INIT_LIST_HEAD(&nm_i->nat_entries);
3092         spin_lock_init(&nm_i->nat_list_lock);
3093
3094         mutex_init(&nm_i->build_lock);
3095         spin_lock_init(&nm_i->nid_list_lock);
3096         init_rwsem(&nm_i->nat_tree_lock);
3097
3098         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3099         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3100         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3101         if (!version_bitmap)
3102                 return -EFAULT;
3103
3104         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3105                                         GFP_KERNEL);
3106         if (!nm_i->nat_bitmap)
3107                 return -ENOMEM;
3108
3109         err = __get_nat_bitmaps(sbi);
3110         if (err)
3111                 return err;
3112
3113 #ifdef CONFIG_F2FS_CHECK_FS
3114         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3115                                         GFP_KERNEL);
3116         if (!nm_i->nat_bitmap_mir)
3117                 return -ENOMEM;
3118 #endif
3119
3120         return 0;
3121 }
3122
3123 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3124 {
3125         struct f2fs_nm_info *nm_i = NM_I(sbi);
3126         int i;
3127
3128         nm_i->free_nid_bitmap =
3129                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3130                                               nm_i->nat_blocks),
3131                               GFP_KERNEL);
3132         if (!nm_i->free_nid_bitmap)
3133                 return -ENOMEM;
3134
3135         for (i = 0; i < nm_i->nat_blocks; i++) {
3136                 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3137                         f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3138                 if (!nm_i->free_nid_bitmap[i])
3139                         return -ENOMEM;
3140         }
3141
3142         nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3143                                                                 GFP_KERNEL);
3144         if (!nm_i->nat_block_bitmap)
3145                 return -ENOMEM;
3146
3147         nm_i->free_nid_count =
3148                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3149                                               nm_i->nat_blocks),
3150                               GFP_KERNEL);
3151         if (!nm_i->free_nid_count)
3152                 return -ENOMEM;
3153         return 0;
3154 }
3155
3156 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3157 {
3158         int err;
3159
3160         sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3161                                                         GFP_KERNEL);
3162         if (!sbi->nm_info)
3163                 return -ENOMEM;
3164
3165         err = init_node_manager(sbi);
3166         if (err)
3167                 return err;
3168
3169         err = init_free_nid_cache(sbi);
3170         if (err)
3171                 return err;
3172
3173         /* load free nid status from nat_bits table */
3174         load_free_nid_bitmap(sbi);
3175
3176         return f2fs_build_free_nids(sbi, true, true);
3177 }
3178
3179 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3180 {
3181         struct f2fs_nm_info *nm_i = NM_I(sbi);
3182         struct free_nid *i, *next_i;
3183         struct nat_entry *natvec[NATVEC_SIZE];
3184         struct nat_entry_set *setvec[SETVEC_SIZE];
3185         nid_t nid = 0;
3186         unsigned int found;
3187
3188         if (!nm_i)
3189                 return;
3190
3191         /* destroy free nid list */
3192         spin_lock(&nm_i->nid_list_lock);
3193         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3194                 __remove_free_nid(sbi, i, FREE_NID);
3195                 spin_unlock(&nm_i->nid_list_lock);
3196                 kmem_cache_free(free_nid_slab, i);
3197                 spin_lock(&nm_i->nid_list_lock);
3198         }
3199         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3200         f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3201         f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3202         spin_unlock(&nm_i->nid_list_lock);
3203
3204         /* destroy nat cache */
3205         down_write(&nm_i->nat_tree_lock);
3206         while ((found = __gang_lookup_nat_cache(nm_i,
3207                                         nid, NATVEC_SIZE, natvec))) {
3208                 unsigned idx;
3209
3210                 nid = nat_get_nid(natvec[found - 1]) + 1;
3211                 for (idx = 0; idx < found; idx++) {
3212                         spin_lock(&nm_i->nat_list_lock);
3213                         list_del(&natvec[idx]->list);
3214                         spin_unlock(&nm_i->nat_list_lock);
3215
3216                         __del_from_nat_cache(nm_i, natvec[idx]);
3217                 }
3218         }
3219         f2fs_bug_on(sbi, nm_i->nat_cnt);
3220
3221         /* destroy nat set cache */
3222         nid = 0;
3223         while ((found = __gang_lookup_nat_set(nm_i,
3224                                         nid, SETVEC_SIZE, setvec))) {
3225                 unsigned idx;
3226
3227                 nid = setvec[found - 1]->set + 1;
3228                 for (idx = 0; idx < found; idx++) {
3229                         /* entry_cnt is not zero, when cp_error was occurred */
3230                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3231                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3232                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3233                 }
3234         }
3235         up_write(&nm_i->nat_tree_lock);
3236
3237         kvfree(nm_i->nat_block_bitmap);
3238         if (nm_i->free_nid_bitmap) {
3239                 int i;
3240
3241                 for (i = 0; i < nm_i->nat_blocks; i++)
3242                         kvfree(nm_i->free_nid_bitmap[i]);
3243                 kvfree(nm_i->free_nid_bitmap);
3244         }
3245         kvfree(nm_i->free_nid_count);
3246
3247         kvfree(nm_i->nat_bitmap);
3248         kvfree(nm_i->nat_bits);
3249 #ifdef CONFIG_F2FS_CHECK_FS
3250         kvfree(nm_i->nat_bitmap_mir);
3251 #endif
3252         sbi->nm_info = NULL;
3253         kvfree(nm_i);
3254 }
3255
3256 int __init f2fs_create_node_manager_caches(void)
3257 {
3258         nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3259                         sizeof(struct nat_entry));
3260         if (!nat_entry_slab)
3261                 goto fail;
3262
3263         free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3264                         sizeof(struct free_nid));
3265         if (!free_nid_slab)
3266                 goto destroy_nat_entry;
3267
3268         nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3269                         sizeof(struct nat_entry_set));
3270         if (!nat_entry_set_slab)
3271                 goto destroy_free_nid;
3272
3273         fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3274                         sizeof(struct fsync_node_entry));
3275         if (!fsync_node_entry_slab)
3276                 goto destroy_nat_entry_set;
3277         return 0;
3278
3279 destroy_nat_entry_set:
3280         kmem_cache_destroy(nat_entry_set_slab);
3281 destroy_free_nid:
3282         kmem_cache_destroy(free_nid_slab);
3283 destroy_nat_entry:
3284         kmem_cache_destroy(nat_entry_slab);
3285 fail:
3286         return -ENOMEM;
3287 }
3288
3289 void f2fs_destroy_node_manager_caches(void)
3290 {
3291         kmem_cache_destroy(fsync_node_entry_slab);
3292         kmem_cache_destroy(nat_entry_set_slab);
3293         kmem_cache_destroy(free_nid_slab);
3294         kmem_cache_destroy(nat_entry_slab);
3295 }