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/sched/mm.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 "iostat.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                 f2fs_handle_error(sbi, ERROR_CORRUPTED_INODE);
  40                 return -EFSCORRUPTED;
  41         }
  42         return 0;
  43 }
  44
  45 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
  46 {
  47         struct f2fs_nm_info *nm_i = NM_I(sbi);
  48         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
  49         struct sysinfo val;
  50         unsigned long avail_ram;
  51         unsigned long mem_size = 0;
  52         bool res = false;
  53
  54         if (!nm_i)
  55                 return true;
  56
  57         si_meminfo(&val);
  58
  59         /* only uses low memory */
  60         avail_ram = val.totalram - val.totalhigh;
  61
  62         /*
  63          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  64          */
  65         if (type == FREE_NIDS) {
  66                 mem_size = (nm_i->nid_cnt[FREE_NID] *
  67                                 sizeof(struct free_nid)) >> PAGE_SHIFT;
  68                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  69         } else if (type == NAT_ENTRIES) {
  70                 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
  71                                 sizeof(struct nat_entry)) >> PAGE_SHIFT;
  72                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  73                 if (excess_cached_nats(sbi))
  74                         res = false;
  75         } else if (type == DIRTY_DENTS) {
  76                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  77                         return false;
  78                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  79                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  80         } else if (type == INO_ENTRIES) {
  81                 int i;
  82
  83                 for (i = 0; i < MAX_INO_ENTRY; i++)
  84                         mem_size += sbi->im[i].ino_num *
  85                                                 sizeof(struct ino_entry);
  86                 mem_size >>= PAGE_SHIFT;
  87                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  88         } else if (type == EXTENT_CACHE) {
  89                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  90                                 sizeof(struct extent_tree) +
  91                                 atomic_read(&sbi->total_ext_node) *
  92                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  93                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  94         } else if (type == DISCARD_CACHE) {
  95                 mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
  96                                 sizeof(struct discard_cmd)) >> PAGE_SHIFT;
  97                 res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
  98         } else if (type == COMPRESS_PAGE) {
  99 #ifdef CONFIG_F2FS_FS_COMPRESSION
 100                 unsigned long free_ram = val.freeram;
 101
 102                 /*
 103                  * free memory is lower than watermark or cached page count
 104                  * exceed threshold, deny caching compress page.
 105                  */
 106                 res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
 107                         (COMPRESS_MAPPING(sbi)->nrpages <
 108                          free_ram * sbi->compress_percent / 100);
 109 #else
 110                 res = false;
 111 #endif
 112         } else {
 113                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
 114                         return true;
 115         }
 116         return res;
 117 }
 118
 119 static void clear_node_page_dirty(struct page *page)
 120 {
 121         if (PageDirty(page)) {
 122                 f2fs_clear_page_cache_dirty_tag(page);
 123                 clear_page_dirty_for_io(page);
 124                 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
 125         }
 126         ClearPageUptodate(page);
 127 }
 128
 129 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 130 {
 131         return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
 132 }
 133
 134 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 135 {
 136         struct page *src_page;
 137         struct page *dst_page;
 138         pgoff_t dst_off;
 139         void *src_addr;
 140         void *dst_addr;
 141         struct f2fs_nm_info *nm_i = NM_I(sbi);
 142
 143         dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
 144
 145         /* get current nat block page with lock */
 146         src_page = get_current_nat_page(sbi, nid);
 147         if (IS_ERR(src_page))
 148                 return src_page;
 149         dst_page = f2fs_grab_meta_page(sbi, dst_off);
 150         f2fs_bug_on(sbi, PageDirty(src_page));
 151
 152         src_addr = page_address(src_page);
 153         dst_addr = page_address(dst_page);
 154         memcpy(dst_addr, src_addr, PAGE_SIZE);
 155         set_page_dirty(dst_page);
 156         f2fs_put_page(src_page, 1);
 157
 158         set_to_next_nat(nm_i, nid);
 159
 160         return dst_page;
 161 }
 162
 163 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
 164                                                 nid_t nid, bool no_fail)
 165 {
 166         struct nat_entry *new;
 167
 168         new = f2fs_kmem_cache_alloc(nat_entry_slab,
 169                                         GFP_F2FS_ZERO, no_fail, sbi);
 170         if (new) {
 171                 nat_set_nid(new, nid);
 172                 nat_reset_flag(new);
 173         }
 174         return new;
 175 }
 176
 177 static void __free_nat_entry(struct nat_entry *e)
 178 {
 179         kmem_cache_free(nat_entry_slab, e);
 180 }
 181
 182 /* must be locked by nat_tree_lock */
 183 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
 184         struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
 185 {
 186         if (no_fail)
 187                 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
 188         else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
 189                 return NULL;
 190
 191         if (raw_ne)
 192                 node_info_from_raw_nat(&ne->ni, raw_ne);
 193
 194         spin_lock(&nm_i->nat_list_lock);
 195         list_add_tail(&ne->list, &nm_i->nat_entries);
 196         spin_unlock(&nm_i->nat_list_lock);
 197
 198         nm_i->nat_cnt[TOTAL_NAT]++;
 199         nm_i->nat_cnt[RECLAIMABLE_NAT]++;
 200         return ne;
 201 }
 202
 203 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 204 {
 205         struct nat_entry *ne;
 206
 207         ne = radix_tree_lookup(&nm_i->nat_root, n);
 208
 209         /* for recent accessed nat entry, move it to tail of lru list */
 210         if (ne && !get_nat_flag(ne, IS_DIRTY)) {
 211                 spin_lock(&nm_i->nat_list_lock);
 212                 if (!list_empty(&ne->list))
 213                         list_move_tail(&ne->list, &nm_i->nat_entries);
 214                 spin_unlock(&nm_i->nat_list_lock);
 215         }
 216
 217         return ne;
 218 }
 219
 220 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 221                 nid_t start, unsigned int nr, struct nat_entry **ep)
 222 {
 223         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 224 }
 225
 226 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 227 {
 228         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 229         nm_i->nat_cnt[TOTAL_NAT]--;
 230         nm_i->nat_cnt[RECLAIMABLE_NAT]--;
 231         __free_nat_entry(e);
 232 }
 233
 234 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
 235                                                         struct nat_entry *ne)
 236 {
 237         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 238         struct nat_entry_set *head;
 239
 240         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 241         if (!head) {
 242                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
 243                                                 GFP_NOFS, true, NULL);
 244
 245                 INIT_LIST_HEAD(&head->entry_list);
 246                 INIT_LIST_HEAD(&head->set_list);
 247                 head->set = set;
 248                 head->entry_cnt = 0;
 249                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 250         }
 251         return head;
 252 }
 253
 254 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 255                                                 struct nat_entry *ne)
 256 {
 257         struct nat_entry_set *head;
 258         bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
 259
 260         if (!new_ne)
 261                 head = __grab_nat_entry_set(nm_i, ne);
 262
 263         /*
 264          * update entry_cnt in below condition:
 265          * 1. update NEW_ADDR to valid block address;
 266          * 2. update old block address to new one;
 267          */
 268         if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
 269                                 !get_nat_flag(ne, IS_DIRTY)))
 270                 head->entry_cnt++;
 271
 272         set_nat_flag(ne, IS_PREALLOC, new_ne);
 273
 274         if (get_nat_flag(ne, IS_DIRTY))
 275                 goto refresh_list;
 276
 277         nm_i->nat_cnt[DIRTY_NAT]++;
 278         nm_i->nat_cnt[RECLAIMABLE_NAT]--;
 279         set_nat_flag(ne, IS_DIRTY, true);
 280 refresh_list:
 281         spin_lock(&nm_i->nat_list_lock);
 282         if (new_ne)
 283                 list_del_init(&ne->list);
 284         else
 285                 list_move_tail(&ne->list, &head->entry_list);
 286         spin_unlock(&nm_i->nat_list_lock);
 287 }
 288
 289 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 290                 struct nat_entry_set *set, struct nat_entry *ne)
 291 {
 292         spin_lock(&nm_i->nat_list_lock);
 293         list_move_tail(&ne->list, &nm_i->nat_entries);
 294         spin_unlock(&nm_i->nat_list_lock);
 295
 296         set_nat_flag(ne, IS_DIRTY, false);
 297         set->entry_cnt--;
 298         nm_i->nat_cnt[DIRTY_NAT]--;
 299         nm_i->nat_cnt[RECLAIMABLE_NAT]++;
 300 }
 301
 302 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 303                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 304 {
 305         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 306                                                         start, nr);
 307 }
 308
 309 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
 310 {
 311         return NODE_MAPPING(sbi) == page->mapping &&
 312                         IS_DNODE(page) && is_cold_node(page);
 313 }
 314
 315 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
 316 {
 317         spin_lock_init(&sbi->fsync_node_lock);
 318         INIT_LIST_HEAD(&sbi->fsync_node_list);
 319         sbi->fsync_seg_id = 0;
 320         sbi->fsync_node_num = 0;
 321 }
 322
 323 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
 324                                                         struct page *page)
 325 {
 326         struct fsync_node_entry *fn;
 327         unsigned long flags;
 328         unsigned int seq_id;
 329
 330         fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
 331                                         GFP_NOFS, true, NULL);
 332
 333         get_page(page);
 334         fn->page = page;
 335         INIT_LIST_HEAD(&fn->list);
 336
 337         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 338         list_add_tail(&fn->list, &sbi->fsync_node_list);
 339         fn->seq_id = sbi->fsync_seg_id++;
 340         seq_id = fn->seq_id;
 341         sbi->fsync_node_num++;
 342         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 343
 344         return seq_id;
 345 }
 346
 347 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
 348 {
 349         struct fsync_node_entry *fn;
 350         unsigned long flags;
 351
 352         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 353         list_for_each_entry(fn, &sbi->fsync_node_list, list) {
 354                 if (fn->page == page) {
 355                         list_del(&fn->list);
 356                         sbi->fsync_node_num--;
 357                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 358                         kmem_cache_free(fsync_node_entry_slab, fn);
 359                         put_page(page);
 360                         return;
 361                 }
 362         }
 363         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 364         f2fs_bug_on(sbi, 1);
 365 }
 366
 367 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
 368 {
 369         unsigned long flags;
 370
 371         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 372         sbi->fsync_seg_id = 0;
 373         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 374 }
 375
 376 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 377 {
 378         struct f2fs_nm_info *nm_i = NM_I(sbi);
 379         struct nat_entry *e;
 380         bool need = false;
 381
 382         f2fs_down_read(&nm_i->nat_tree_lock);
 383         e = __lookup_nat_cache(nm_i, nid);
 384         if (e) {
 385                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 386                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 387                         need = true;
 388         }
 389         f2fs_up_read(&nm_i->nat_tree_lock);
 390         return need;
 391 }
 392
 393 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 394 {
 395         struct f2fs_nm_info *nm_i = NM_I(sbi);
 396         struct nat_entry *e;
 397         bool is_cp = true;
 398
 399         f2fs_down_read(&nm_i->nat_tree_lock);
 400         e = __lookup_nat_cache(nm_i, nid);
 401         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 402                 is_cp = false;
 403         f2fs_up_read(&nm_i->nat_tree_lock);
 404         return is_cp;
 405 }
 406
 407 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 408 {
 409         struct f2fs_nm_info *nm_i = NM_I(sbi);
 410         struct nat_entry *e;
 411         bool need_update = true;
 412
 413         f2fs_down_read(&nm_i->nat_tree_lock);
 414         e = __lookup_nat_cache(nm_i, ino);
 415         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 416                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 417                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 418                 need_update = false;
 419         f2fs_up_read(&nm_i->nat_tree_lock);
 420         return need_update;
 421 }
 422
 423 /* must be locked by nat_tree_lock */
 424 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 425                                                 struct f2fs_nat_entry *ne)
 426 {
 427         struct f2fs_nm_info *nm_i = NM_I(sbi);
 428         struct nat_entry *new, *e;
 429
 430         /* Let's mitigate lock contention of nat_tree_lock during checkpoint */
 431         if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
 432                 return;
 433
 434         new = __alloc_nat_entry(sbi, nid, false);
 435         if (!new)
 436                 return;
 437
 438         f2fs_down_write(&nm_i->nat_tree_lock);
 439         e = __lookup_nat_cache(nm_i, nid);
 440         if (!e)
 441                 e = __init_nat_entry(nm_i, new, ne, false);
 442         else
 443                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 444                                 nat_get_blkaddr(e) !=
 445                                         le32_to_cpu(ne->block_addr) ||
 446                                 nat_get_version(e) != ne->version);
 447         f2fs_up_write(&nm_i->nat_tree_lock);
 448         if (e != new)
 449                 __free_nat_entry(new);
 450 }
 451
 452 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 453                         block_t new_blkaddr, bool fsync_done)
 454 {
 455         struct f2fs_nm_info *nm_i = NM_I(sbi);
 456         struct nat_entry *e;
 457         struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
 458
 459         f2fs_down_write(&nm_i->nat_tree_lock);
 460         e = __lookup_nat_cache(nm_i, ni->nid);
 461         if (!e) {
 462                 e = __init_nat_entry(nm_i, new, NULL, true);
 463                 copy_node_info(&e->ni, ni);
 464                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 465         } else if (new_blkaddr == NEW_ADDR) {
 466                 /*
 467                  * when nid is reallocated,
 468                  * previous nat entry can be remained in nat cache.
 469                  * So, reinitialize it with new information.
 470                  */
 471                 copy_node_info(&e->ni, ni);
 472                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 473         }
 474         /* let's free early to reduce memory consumption */
 475         if (e != new)
 476                 __free_nat_entry(new);
 477
 478         /* sanity check */
 479         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 480         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 481                         new_blkaddr == NULL_ADDR);
 482         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 483                         new_blkaddr == NEW_ADDR);
 484         f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
 485                         new_blkaddr == NEW_ADDR);
 486
 487         /* increment version no as node is removed */
 488         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 489                 unsigned char version = nat_get_version(e);
 490
 491                 nat_set_version(e, inc_node_version(version));
 492         }
 493
 494         /* change address */
 495         nat_set_blkaddr(e, new_blkaddr);
 496         if (!__is_valid_data_blkaddr(new_blkaddr))
 497                 set_nat_flag(e, IS_CHECKPOINTED, false);
 498         __set_nat_cache_dirty(nm_i, e);
 499
 500         /* update fsync_mark if its inode nat entry is still alive */
 501         if (ni->nid != ni->ino)
 502                 e = __lookup_nat_cache(nm_i, ni->ino);
 503         if (e) {
 504                 if (fsync_done && ni->nid == ni->ino)
 505                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 506                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 507         }
 508         f2fs_up_write(&nm_i->nat_tree_lock);
 509 }
 510
 511 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 512 {
 513         struct f2fs_nm_info *nm_i = NM_I(sbi);
 514         int nr = nr_shrink;
 515
 516         if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
 517                 return 0;
 518
 519         spin_lock(&nm_i->nat_list_lock);
 520         while (nr_shrink) {
 521                 struct nat_entry *ne;
 522
 523                 if (list_empty(&nm_i->nat_entries))
 524                         break;
 525
 526                 ne = list_first_entry(&nm_i->nat_entries,
 527                                         struct nat_entry, list);
 528                 list_del(&ne->list);
 529                 spin_unlock(&nm_i->nat_list_lock);
 530
 531                 __del_from_nat_cache(nm_i, ne);
 532                 nr_shrink--;
 533
 534                 spin_lock(&nm_i->nat_list_lock);
 535         }
 536         spin_unlock(&nm_i->nat_list_lock);
 537
 538         f2fs_up_write(&nm_i->nat_tree_lock);
 539         return nr - nr_shrink;
 540 }
 541
 542 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
 543                                 struct node_info *ni, bool checkpoint_context)
 544 {
 545         struct f2fs_nm_info *nm_i = NM_I(sbi);
 546         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 547         struct f2fs_journal *journal = curseg->journal;
 548         nid_t start_nid = START_NID(nid);
 549         struct f2fs_nat_block *nat_blk;
 550         struct page *page = NULL;
 551         struct f2fs_nat_entry ne;
 552         struct nat_entry *e;
 553         pgoff_t index;
 554         block_t blkaddr;
 555         int i;
 556
 557         ni->nid = nid;
 558 retry:
 559         /* Check nat cache */
 560         f2fs_down_read(&nm_i->nat_tree_lock);
 561         e = __lookup_nat_cache(nm_i, nid);
 562         if (e) {
 563                 ni->ino = nat_get_ino(e);
 564                 ni->blk_addr = nat_get_blkaddr(e);
 565                 ni->version = nat_get_version(e);
 566                 f2fs_up_read(&nm_i->nat_tree_lock);
 567                 return 0;
 568         }
 569
 570         /*
 571          * Check current segment summary by trying to grab journal_rwsem first.
 572          * This sem is on the critical path on the checkpoint requiring the above
 573          * nat_tree_lock. Therefore, we should retry, if we failed to grab here
 574          * while not bothering checkpoint.
 575          */
 576         if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
 577                 down_read(&curseg->journal_rwsem);
 578         } else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
 579                                 !down_read_trylock(&curseg->journal_rwsem)) {
 580                 f2fs_up_read(&nm_i->nat_tree_lock);
 581                 goto retry;
 582         }
 583
 584         i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 585         if (i >= 0) {
 586                 ne = nat_in_journal(journal, i);
 587                 node_info_from_raw_nat(ni, &ne);
 588         }
 589         up_read(&curseg->journal_rwsem);
 590         if (i >= 0) {
 591                 f2fs_up_read(&nm_i->nat_tree_lock);
 592                 goto cache;
 593         }
 594
 595         /* Fill node_info from nat page */
 596         index = current_nat_addr(sbi, nid);
 597         f2fs_up_read(&nm_i->nat_tree_lock);
 598
 599         page = f2fs_get_meta_page(sbi, index);
 600         if (IS_ERR(page))
 601                 return PTR_ERR(page);
 602
 603         nat_blk = (struct f2fs_nat_block *)page_address(page);
 604         ne = nat_blk->entries[nid - start_nid];
 605         node_info_from_raw_nat(ni, &ne);
 606         f2fs_put_page(page, 1);
 607 cache:
 608         blkaddr = le32_to_cpu(ne.block_addr);
 609         if (__is_valid_data_blkaddr(blkaddr) &&
 610                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
 611                 return -EFAULT;
 612
 613         /* cache nat entry */
 614         cache_nat_entry(sbi, nid, &ne);
 615         return 0;
 616 }
 617
 618 /*
 619  * readahead MAX_RA_NODE number of node pages.
 620  */
 621 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
 622 {
 623         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 624         struct blk_plug plug;
 625         int i, end;
 626         nid_t nid;
 627
 628         blk_start_plug(&plug);
 629
 630         /* Then, try readahead for siblings of the desired node */
 631         end = start + n;
 632         end = min(end, NIDS_PER_BLOCK);
 633         for (i = start; i < end; i++) {
 634                 nid = get_nid(parent, i, false);
 635                 f2fs_ra_node_page(sbi, nid);
 636         }
 637
 638         blk_finish_plug(&plug);
 639 }
 640
 641 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 642 {
 643         const long direct_index = ADDRS_PER_INODE(dn->inode);
 644         const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
 645         const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
 646         unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
 647         int cur_level = dn->cur_level;
 648         int max_level = dn->max_level;
 649         pgoff_t base = 0;
 650
 651         if (!dn->max_level)
 652                 return pgofs + 1;
 653
 654         while (max_level-- > cur_level)
 655                 skipped_unit *= NIDS_PER_BLOCK;
 656
 657         switch (dn->max_level) {
 658         case 3:
 659                 base += 2 * indirect_blks;
 660                 fallthrough;
 661         case 2:
 662                 base += 2 * direct_blks;
 663                 fallthrough;
 664         case 1:
 665                 base += direct_index;
 666                 break;
 667         default:
 668                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 669         }
 670
 671         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 672 }
 673
 674 /*
 675  * The maximum depth is four.
 676  * Offset[0] will have raw inode offset.
 677  */
 678 static int get_node_path(struct inode *inode, long block,
 679                                 int offset[4], unsigned int noffset[4])
 680 {
 681         const long direct_index = ADDRS_PER_INODE(inode);
 682         const long direct_blks = ADDRS_PER_BLOCK(inode);
 683         const long dptrs_per_blk = NIDS_PER_BLOCK;
 684         const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
 685         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 686         int n = 0;
 687         int level = 0;
 688
 689         noffset[0] = 0;
 690
 691         if (block < direct_index) {
 692                 offset[n] = block;
 693                 goto got;
 694         }
 695         block -= direct_index;
 696         if (block < direct_blks) {
 697                 offset[n++] = NODE_DIR1_BLOCK;
 698                 noffset[n] = 1;
 699                 offset[n] = block;
 700                 level = 1;
 701                 goto got;
 702         }
 703         block -= direct_blks;
 704         if (block < direct_blks) {
 705                 offset[n++] = NODE_DIR2_BLOCK;
 706                 noffset[n] = 2;
 707                 offset[n] = block;
 708                 level = 1;
 709                 goto got;
 710         }
 711         block -= direct_blks;
 712         if (block < indirect_blks) {
 713                 offset[n++] = NODE_IND1_BLOCK;
 714                 noffset[n] = 3;
 715                 offset[n++] = block / direct_blks;
 716                 noffset[n] = 4 + offset[n - 1];
 717                 offset[n] = block % direct_blks;
 718                 level = 2;
 719                 goto got;
 720         }
 721         block -= indirect_blks;
 722         if (block < indirect_blks) {
 723                 offset[n++] = NODE_IND2_BLOCK;
 724                 noffset[n] = 4 + dptrs_per_blk;
 725                 offset[n++] = block / direct_blks;
 726                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 727                 offset[n] = block % direct_blks;
 728                 level = 2;
 729                 goto got;
 730         }
 731         block -= indirect_blks;
 732         if (block < dindirect_blks) {
 733                 offset[n++] = NODE_DIND_BLOCK;
 734                 noffset[n] = 5 + (dptrs_per_blk * 2);
 735                 offset[n++] = block / indirect_blks;
 736                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 737                               offset[n - 1] * (dptrs_per_blk + 1);
 738                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 739                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 740                               offset[n - 2] * (dptrs_per_blk + 1) +
 741                               offset[n - 1];
 742                 offset[n] = block % direct_blks;
 743                 level = 3;
 744                 goto got;
 745         } else {
 746                 return -E2BIG;
 747         }
 748 got:
 749         return level;
 750 }
 751
 752 /*
 753  * Caller should call f2fs_put_dnode(dn).
 754  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 755  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
 756  */
 757 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 758 {
 759         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 760         struct page *npage[4];
 761         struct page *parent = NULL;
 762         int offset[4];
 763         unsigned int noffset[4];
 764         nid_t nids[4];
 765         int level, i = 0;
 766         int err = 0;
 767
 768         level = get_node_path(dn->inode, index, offset, noffset);
 769         if (level < 0)
 770                 return level;
 771
 772         nids[0] = dn->inode->i_ino;
 773         npage[0] = dn->inode_page;
 774
 775         if (!npage[0]) {
 776                 npage[0] = f2fs_get_node_page(sbi, nids[0]);
 777                 if (IS_ERR(npage[0]))
 778                         return PTR_ERR(npage[0]);
 779         }
 780
 781         /* if inline_data is set, should not report any block indices */
 782         if (f2fs_has_inline_data(dn->inode) && index) {
 783                 err = -ENOENT;
 784                 f2fs_put_page(npage[0], 1);
 785                 goto release_out;
 786         }
 787
 788         parent = npage[0];
 789         if (level != 0)
 790                 nids[1] = get_nid(parent, offset[0], true);
 791         dn->inode_page = npage[0];
 792         dn->inode_page_locked = true;
 793
 794         /* get indirect or direct nodes */
 795         for (i = 1; i <= level; i++) {
 796                 bool done = false;
 797
 798                 if (!nids[i] && mode == ALLOC_NODE) {
 799                         /* alloc new node */
 800                         if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
 801                                 err = -ENOSPC;
 802                                 goto release_pages;
 803                         }
 804
 805                         dn->nid = nids[i];
 806                         npage[i] = f2fs_new_node_page(dn, noffset[i]);
 807                         if (IS_ERR(npage[i])) {
 808                                 f2fs_alloc_nid_failed(sbi, nids[i]);
 809                                 err = PTR_ERR(npage[i]);
 810                                 goto release_pages;
 811                         }
 812
 813                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 814                         f2fs_alloc_nid_done(sbi, nids[i]);
 815                         done = true;
 816                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 817                         npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
 818                         if (IS_ERR(npage[i])) {
 819                                 err = PTR_ERR(npage[i]);
 820                                 goto release_pages;
 821                         }
 822                         done = true;
 823                 }
 824                 if (i == 1) {
 825                         dn->inode_page_locked = false;
 826                         unlock_page(parent);
 827                 } else {
 828                         f2fs_put_page(parent, 1);
 829                 }
 830
 831                 if (!done) {
 832                         npage[i] = f2fs_get_node_page(sbi, nids[i]);
 833                         if (IS_ERR(npage[i])) {
 834                                 err = PTR_ERR(npage[i]);
 835                                 f2fs_put_page(npage[0], 0);
 836                                 goto release_out;
 837                         }
 838                 }
 839                 if (i < level) {
 840                         parent = npage[i];
 841                         nids[i + 1] = get_nid(parent, offset[i], false);
 842                 }
 843         }
 844         dn->nid = nids[level];
 845         dn->ofs_in_node = offset[level];
 846         dn->node_page = npage[level];
 847         dn->data_blkaddr = f2fs_data_blkaddr(dn);
 848
 849         if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
 850                                         f2fs_sb_has_readonly(sbi)) {
 851                 unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn);
 852                 block_t blkaddr;
 853
 854                 if (!c_len)
 855                         goto out;
 856
 857                 blkaddr = f2fs_data_blkaddr(dn);
 858                 if (blkaddr == COMPRESS_ADDR)
 859                         blkaddr = data_blkaddr(dn->inode, dn->node_page,
 860                                                 dn->ofs_in_node + 1);
 861
 862                 f2fs_update_extent_tree_range_compressed(dn->inode,
 863                                         index, blkaddr,
 864                                         F2FS_I(dn->inode)->i_cluster_size,
 865                                         c_len);
 866         }
 867 out:
 868         return 0;
 869
 870 release_pages:
 871         f2fs_put_page(parent, 1);
 872         if (i > 1)
 873                 f2fs_put_page(npage[0], 0);
 874 release_out:
 875         dn->inode_page = NULL;
 876         dn->node_page = NULL;
 877         if (err == -ENOENT) {
 878                 dn->cur_level = i;
 879                 dn->max_level = level;
 880                 dn->ofs_in_node = offset[level];
 881         }
 882         return err;
 883 }
 884
 885 static int truncate_node(struct dnode_of_data *dn)
 886 {
 887         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 888         struct node_info ni;
 889         int err;
 890         pgoff_t index;
 891
 892         err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
 893         if (err)
 894                 return err;
 895
 896         /* Deallocate node address */
 897         f2fs_invalidate_blocks(sbi, ni.blk_addr);
 898         dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
 899         set_node_addr(sbi, &ni, NULL_ADDR, false);
 900
 901         if (dn->nid == dn->inode->i_ino) {
 902                 f2fs_remove_orphan_inode(sbi, dn->nid);
 903                 dec_valid_inode_count(sbi);
 904                 f2fs_inode_synced(dn->inode);
 905         }
 906
 907         clear_node_page_dirty(dn->node_page);
 908         set_sbi_flag(sbi, SBI_IS_DIRTY);
 909
 910         index = dn->node_page->index;
 911         f2fs_put_page(dn->node_page, 1);
 912
 913         invalidate_mapping_pages(NODE_MAPPING(sbi),
 914                         index, index);
 915
 916         dn->node_page = NULL;
 917         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 918
 919         return 0;
 920 }
 921
 922 static int truncate_dnode(struct dnode_of_data *dn)
 923 {
 924         struct page *page;
 925         int err;
 926
 927         if (dn->nid == 0)
 928                 return 1;
 929
 930         /* get direct node */
 931         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 932         if (PTR_ERR(page) == -ENOENT)
 933                 return 1;
 934         else if (IS_ERR(page))
 935                 return PTR_ERR(page);
 936
 937         /* Make dnode_of_data for parameter */
 938         dn->node_page = page;
 939         dn->ofs_in_node = 0;
 940         f2fs_truncate_data_blocks(dn);
 941         err = truncate_node(dn);
 942         if (err)
 943                 return err;
 944
 945         return 1;
 946 }
 947
 948 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 949                                                 int ofs, int depth)
 950 {
 951         struct dnode_of_data rdn = *dn;
 952         struct page *page;
 953         struct f2fs_node *rn;
 954         nid_t child_nid;
 955         unsigned int child_nofs;
 956         int freed = 0;
 957         int i, ret;
 958
 959         if (dn->nid == 0)
 960                 return NIDS_PER_BLOCK + 1;
 961
 962         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 963
 964         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 965         if (IS_ERR(page)) {
 966                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 967                 return PTR_ERR(page);
 968         }
 969
 970         f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 971
 972         rn = F2FS_NODE(page);
 973         if (depth < 3) {
 974                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 975                         child_nid = le32_to_cpu(rn->in.nid[i]);
 976                         if (child_nid == 0)
 977                                 continue;
 978                         rdn.nid = child_nid;
 979                         ret = truncate_dnode(&rdn);
 980                         if (ret < 0)
 981                                 goto out_err;
 982                         if (set_nid(page, i, 0, false))
 983                                 dn->node_changed = true;
 984                 }
 985         } else {
 986                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 987                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 988                         child_nid = le32_to_cpu(rn->in.nid[i]);
 989                         if (child_nid == 0) {
 990                                 child_nofs += NIDS_PER_BLOCK + 1;
 991                                 continue;
 992                         }
 993                         rdn.nid = child_nid;
 994                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 995                         if (ret == (NIDS_PER_BLOCK + 1)) {
 996                                 if (set_nid(page, i, 0, false))
 997                                         dn->node_changed = true;
 998                                 child_nofs += ret;
 999                         } else if (ret < 0 && ret != -ENOENT) {
1000                                 goto out_err;
1001                         }
1002                 }
1003                 freed = child_nofs;
1004         }
1005
1006         if (!ofs) {
1007                 /* remove current indirect node */
1008                 dn->node_page = page;
1009                 ret = truncate_node(dn);
1010                 if (ret)
1011                         goto out_err;
1012                 freed++;
1013         } else {
1014                 f2fs_put_page(page, 1);
1015         }
1016         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1017         return freed;
1018
1019 out_err:
1020         f2fs_put_page(page, 1);
1021         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1022         return ret;
1023 }
1024
1025 static int truncate_partial_nodes(struct dnode_of_data *dn,
1026                         struct f2fs_inode *ri, int *offset, int depth)
1027 {
1028         struct page *pages[2];
1029         nid_t nid[3];
1030         nid_t child_nid;
1031         int err = 0;
1032         int i;
1033         int idx = depth - 2;
1034
1035         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1036         if (!nid[0])
1037                 return 0;
1038
1039         /* get indirect nodes in the path */
1040         for (i = 0; i < idx + 1; i++) {
1041                 /* reference count'll be increased */
1042                 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1043                 if (IS_ERR(pages[i])) {
1044                         err = PTR_ERR(pages[i]);
1045                         idx = i - 1;
1046                         goto fail;
1047                 }
1048                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1049         }
1050
1051         f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1052
1053         /* free direct nodes linked to a partial indirect node */
1054         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1055                 child_nid = get_nid(pages[idx], i, false);
1056                 if (!child_nid)
1057                         continue;
1058                 dn->nid = child_nid;
1059                 err = truncate_dnode(dn);
1060                 if (err < 0)
1061                         goto fail;
1062                 if (set_nid(pages[idx], i, 0, false))
1063                         dn->node_changed = true;
1064         }
1065
1066         if (offset[idx + 1] == 0) {
1067                 dn->node_page = pages[idx];
1068                 dn->nid = nid[idx];
1069                 err = truncate_node(dn);
1070                 if (err)
1071                         goto fail;
1072         } else {
1073                 f2fs_put_page(pages[idx], 1);
1074         }
1075         offset[idx]++;
1076         offset[idx + 1] = 0;
1077         idx--;
1078 fail:
1079         for (i = idx; i >= 0; i--)
1080                 f2fs_put_page(pages[i], 1);
1081
1082         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1083
1084         return err;
1085 }
1086
1087 /*
1088  * All the block addresses of data and nodes should be nullified.
1089  */
1090 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1091 {
1092         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1093         int err = 0, cont = 1;
1094         int level, offset[4], noffset[4];
1095         unsigned int nofs = 0;
1096         struct f2fs_inode *ri;
1097         struct dnode_of_data dn;
1098         struct page *page;
1099
1100         trace_f2fs_truncate_inode_blocks_enter(inode, from);
1101
1102         level = get_node_path(inode, from, offset, noffset);
1103         if (level < 0) {
1104                 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1105                 return level;
1106         }
1107
1108         page = f2fs_get_node_page(sbi, inode->i_ino);
1109         if (IS_ERR(page)) {
1110                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1111                 return PTR_ERR(page);
1112         }
1113
1114         set_new_dnode(&dn, inode, page, NULL, 0);
1115         unlock_page(page);
1116
1117         ri = F2FS_INODE(page);
1118         switch (level) {
1119         case 0:
1120         case 1:
1121                 nofs = noffset[1];
1122                 break;
1123         case 2:
1124                 nofs = noffset[1];
1125                 if (!offset[level - 1])
1126                         goto skip_partial;
1127                 err = truncate_partial_nodes(&dn, ri, offset, level);
1128                 if (err < 0 && err != -ENOENT)
1129                         goto fail;
1130                 nofs += 1 + NIDS_PER_BLOCK;
1131                 break;
1132         case 3:
1133                 nofs = 5 + 2 * NIDS_PER_BLOCK;
1134                 if (!offset[level - 1])
1135                         goto skip_partial;
1136                 err = truncate_partial_nodes(&dn, ri, offset, level);
1137                 if (err < 0 && err != -ENOENT)
1138                         goto fail;
1139                 break;
1140         default:
1141                 BUG();
1142         }
1143
1144 skip_partial:
1145         while (cont) {
1146                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1147                 switch (offset[0]) {
1148                 case NODE_DIR1_BLOCK:
1149                 case NODE_DIR2_BLOCK:
1150                         err = truncate_dnode(&dn);
1151                         break;
1152
1153                 case NODE_IND1_BLOCK:
1154                 case NODE_IND2_BLOCK:
1155                         err = truncate_nodes(&dn, nofs, offset[1], 2);
1156                         break;
1157
1158                 case NODE_DIND_BLOCK:
1159                         err = truncate_nodes(&dn, nofs, offset[1], 3);
1160                         cont = 0;
1161                         break;
1162
1163                 default:
1164                         BUG();
1165                 }
1166                 if (err < 0 && err != -ENOENT)
1167                         goto fail;
1168                 if (offset[1] == 0 &&
1169                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1170                         lock_page(page);
1171                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
1172                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1173                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1174                         set_page_dirty(page);
1175                         unlock_page(page);
1176                 }
1177                 offset[1] = 0;
1178                 offset[0]++;
1179                 nofs += err;
1180         }
1181 fail:
1182         f2fs_put_page(page, 0);
1183         trace_f2fs_truncate_inode_blocks_exit(inode, err);
1184         return err > 0 ? 0 : err;
1185 }
1186
1187 /* caller must lock inode page */
1188 int f2fs_truncate_xattr_node(struct inode *inode)
1189 {
1190         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1191         nid_t nid = F2FS_I(inode)->i_xattr_nid;
1192         struct dnode_of_data dn;
1193         struct page *npage;
1194         int err;
1195
1196         if (!nid)
1197                 return 0;
1198
1199         npage = f2fs_get_node_page(sbi, nid);
1200         if (IS_ERR(npage))
1201                 return PTR_ERR(npage);
1202
1203         set_new_dnode(&dn, inode, NULL, npage, nid);
1204         err = truncate_node(&dn);
1205         if (err) {
1206                 f2fs_put_page(npage, 1);
1207                 return err;
1208         }
1209
1210         f2fs_i_xnid_write(inode, 0);
1211
1212         return 0;
1213 }
1214
1215 /*
1216  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1217  * f2fs_unlock_op().
1218  */
1219 int f2fs_remove_inode_page(struct inode *inode)
1220 {
1221         struct dnode_of_data dn;
1222         int err;
1223
1224         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1225         err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1226         if (err)
1227                 return err;
1228
1229         err = f2fs_truncate_xattr_node(inode);
1230         if (err) {
1231                 f2fs_put_dnode(&dn);
1232                 return err;
1233         }
1234
1235         /* remove potential inline_data blocks */
1236         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1237                                 S_ISLNK(inode->i_mode))
1238                 f2fs_truncate_data_blocks_range(&dn, 1);
1239
1240         /* 0 is possible, after f2fs_new_inode() has failed */
1241         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1242                 f2fs_put_dnode(&dn);
1243                 return -EIO;
1244         }
1245
1246         if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1247                 f2fs_warn(F2FS_I_SB(inode),
1248                         "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1249                         inode->i_ino, (unsigned long long)inode->i_blocks);
1250                 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1251         }
1252
1253         /* will put inode & node pages */
1254         err = truncate_node(&dn);
1255         if (err) {
1256                 f2fs_put_dnode(&dn);
1257                 return err;
1258         }
1259         return 0;
1260 }
1261
1262 struct page *f2fs_new_inode_page(struct inode *inode)
1263 {
1264         struct dnode_of_data dn;
1265
1266         /* allocate inode page for new inode */
1267         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1268
1269         /* caller should f2fs_put_page(page, 1); */
1270         return f2fs_new_node_page(&dn, 0);
1271 }
1272
1273 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1274 {
1275         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1276         struct node_info new_ni;
1277         struct page *page;
1278         int err;
1279
1280         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1281                 return ERR_PTR(-EPERM);
1282
1283         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1284         if (!page)
1285                 return ERR_PTR(-ENOMEM);
1286
1287         if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1288                 goto fail;
1289
1290 #ifdef CONFIG_F2FS_CHECK_FS
1291         err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1292         if (err) {
1293                 dec_valid_node_count(sbi, dn->inode, !ofs);
1294                 goto fail;
1295         }
1296         if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1297                 err = -EFSCORRUPTED;
1298                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1299                 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1300                 goto fail;
1301         }
1302 #endif
1303         new_ni.nid = dn->nid;
1304         new_ni.ino = dn->inode->i_ino;
1305         new_ni.blk_addr = NULL_ADDR;
1306         new_ni.flag = 0;
1307         new_ni.version = 0;
1308         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1309
1310         f2fs_wait_on_page_writeback(page, NODE, true, true);
1311         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1312         set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1313         if (!PageUptodate(page))
1314                 SetPageUptodate(page);
1315         if (set_page_dirty(page))
1316                 dn->node_changed = true;
1317
1318         if (f2fs_has_xattr_block(ofs))
1319                 f2fs_i_xnid_write(dn->inode, dn->nid);
1320
1321         if (ofs == 0)
1322                 inc_valid_inode_count(sbi);
1323         return page;
1324
1325 fail:
1326         clear_node_page_dirty(page);
1327         f2fs_put_page(page, 1);
1328         return ERR_PTR(err);
1329 }
1330
1331 /*
1332  * Caller should do after getting the following values.
1333  * 0: f2fs_put_page(page, 0)
1334  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1335  */
1336 static int read_node_page(struct page *page, blk_opf_t op_flags)
1337 {
1338         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1339         struct node_info ni;
1340         struct f2fs_io_info fio = {
1341                 .sbi = sbi,
1342                 .type = NODE,
1343                 .op = REQ_OP_READ,
1344                 .op_flags = op_flags,
1345                 .page = page,
1346                 .encrypted_page = NULL,
1347         };
1348         int err;
1349
1350         if (PageUptodate(page)) {
1351                 if (!f2fs_inode_chksum_verify(sbi, page)) {
1352                         ClearPageUptodate(page);
1353                         return -EFSBADCRC;
1354                 }
1355                 return LOCKED_PAGE;
1356         }
1357
1358         err = f2fs_get_node_info(sbi, page->index, &ni, false);
1359         if (err)
1360                 return err;
1361
1362         /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1363         if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1364                 ClearPageUptodate(page);
1365                 return -ENOENT;
1366         }
1367
1368         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1369
1370         err = f2fs_submit_page_bio(&fio);
1371
1372         if (!err)
1373                 f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE);
1374
1375         return err;
1376 }
1377
1378 /*
1379  * Readahead a node page
1380  */
1381 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1382 {
1383         struct page *apage;
1384         int err;
1385
1386         if (!nid)
1387                 return;
1388         if (f2fs_check_nid_range(sbi, nid))
1389                 return;
1390
1391         apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1392         if (apage)
1393                 return;
1394
1395         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1396         if (!apage)
1397                 return;
1398
1399         err = read_node_page(apage, REQ_RAHEAD);
1400         f2fs_put_page(apage, err ? 1 : 0);
1401 }
1402
1403 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1404                                         struct page *parent, int start)
1405 {
1406         struct page *page;
1407         int err;
1408
1409         if (!nid)
1410                 return ERR_PTR(-ENOENT);
1411         if (f2fs_check_nid_range(sbi, nid))
1412                 return ERR_PTR(-EINVAL);
1413 repeat:
1414         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1415         if (!page)
1416                 return ERR_PTR(-ENOMEM);
1417
1418         err = read_node_page(page, 0);
1419         if (err < 0) {
1420                 goto out_put_err;
1421         } else if (err == LOCKED_PAGE) {
1422                 err = 0;
1423                 goto page_hit;
1424         }
1425
1426         if (parent)
1427                 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1428
1429         lock_page(page);
1430
1431         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1432                 f2fs_put_page(page, 1);
1433                 goto repeat;
1434         }
1435
1436         if (unlikely(!PageUptodate(page))) {
1437                 err = -EIO;
1438                 goto out_err;
1439         }
1440
1441         if (!f2fs_inode_chksum_verify(sbi, page)) {
1442                 err = -EFSBADCRC;
1443                 goto out_err;
1444         }
1445 page_hit:
1446         if (likely(nid == nid_of_node(page)))
1447                 return page;
1448
1449         f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1450                           nid, nid_of_node(page), ino_of_node(page),
1451                           ofs_of_node(page), cpver_of_node(page),
1452                           next_blkaddr_of_node(page));
1453         set_sbi_flag(sbi, SBI_NEED_FSCK);
1454         err = -EINVAL;
1455 out_err:
1456         ClearPageUptodate(page);
1457 out_put_err:
1458         /* ENOENT comes from read_node_page which is not an error. */
1459         if (err != -ENOENT)
1460                 f2fs_handle_page_eio(sbi, page->index, NODE);
1461         f2fs_put_page(page, 1);
1462         return ERR_PTR(err);
1463 }
1464
1465 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1466 {
1467         return __get_node_page(sbi, nid, NULL, 0);
1468 }
1469
1470 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1471 {
1472         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1473         nid_t nid = get_nid(parent, start, false);
1474
1475         return __get_node_page(sbi, nid, parent, start);
1476 }
1477
1478 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1479 {
1480         struct inode *inode;
1481         struct page *page;
1482         int ret;
1483
1484         /* should flush inline_data before evict_inode */
1485         inode = ilookup(sbi->sb, ino);
1486         if (!inode)
1487                 return;
1488
1489         page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1490                                         FGP_LOCK|FGP_NOWAIT, 0);
1491         if (!page)
1492                 goto iput_out;
1493
1494         if (!PageUptodate(page))
1495                 goto page_out;
1496
1497         if (!PageDirty(page))
1498                 goto page_out;
1499
1500         if (!clear_page_dirty_for_io(page))
1501                 goto page_out;
1502
1503         ret = f2fs_write_inline_data(inode, page);
1504         inode_dec_dirty_pages(inode);
1505         f2fs_remove_dirty_inode(inode);
1506         if (ret)
1507                 set_page_dirty(page);
1508 page_out:
1509         f2fs_put_page(page, 1);
1510 iput_out:
1511         iput(inode);
1512 }
1513
1514 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1515 {
1516         pgoff_t index;
1517         struct pagevec pvec;
1518         struct page *last_page = NULL;
1519         int nr_pages;
1520
1521         pagevec_init(&pvec);
1522         index = 0;
1523
1524         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1525                                 PAGECACHE_TAG_DIRTY))) {
1526                 int i;
1527
1528                 for (i = 0; i < nr_pages; i++) {
1529                         struct page *page = pvec.pages[i];
1530
1531                         if (unlikely(f2fs_cp_error(sbi))) {
1532                                 f2fs_put_page(last_page, 0);
1533                                 pagevec_release(&pvec);
1534                                 return ERR_PTR(-EIO);
1535                         }
1536
1537                         if (!IS_DNODE(page) || !is_cold_node(page))
1538                                 continue;
1539                         if (ino_of_node(page) != ino)
1540                                 continue;
1541
1542                         lock_page(page);
1543
1544                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1545 continue_unlock:
1546                                 unlock_page(page);
1547                                 continue;
1548                         }
1549                         if (ino_of_node(page) != ino)
1550                                 goto continue_unlock;
1551
1552                         if (!PageDirty(page)) {
1553                                 /* someone wrote it for us */
1554                                 goto continue_unlock;
1555                         }
1556
1557                         if (last_page)
1558                                 f2fs_put_page(last_page, 0);
1559
1560                         get_page(page);
1561                         last_page = page;
1562                         unlock_page(page);
1563                 }
1564                 pagevec_release(&pvec);
1565                 cond_resched();
1566         }
1567         return last_page;
1568 }
1569
1570 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1571                                 struct writeback_control *wbc, bool do_balance,
1572                                 enum iostat_type io_type, unsigned int *seq_id)
1573 {
1574         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1575         nid_t nid;
1576         struct node_info ni;
1577         struct f2fs_io_info fio = {
1578                 .sbi = sbi,
1579                 .ino = ino_of_node(page),
1580                 .type = NODE,
1581                 .op = REQ_OP_WRITE,
1582                 .op_flags = wbc_to_write_flags(wbc),
1583                 .page = page,
1584                 .encrypted_page = NULL,
1585                 .submitted = false,
1586                 .io_type = io_type,
1587                 .io_wbc = wbc,
1588         };
1589         unsigned int seq;
1590
1591         trace_f2fs_writepage(page, NODE);
1592
1593         if (unlikely(f2fs_cp_error(sbi))) {
1594                 ClearPageUptodate(page);
1595                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1596                 unlock_page(page);
1597                 return 0;
1598         }
1599
1600         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1601                 goto redirty_out;
1602
1603         if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1604                         wbc->sync_mode == WB_SYNC_NONE &&
1605                         IS_DNODE(page) && is_cold_node(page))
1606                 goto redirty_out;
1607
1608         /* get old block addr of this node page */
1609         nid = nid_of_node(page);
1610         f2fs_bug_on(sbi, page->index != nid);
1611
1612         if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1613                 goto redirty_out;
1614
1615         if (wbc->for_reclaim) {
1616                 if (!f2fs_down_read_trylock(&sbi->node_write))
1617                         goto redirty_out;
1618         } else {
1619                 f2fs_down_read(&sbi->node_write);
1620         }
1621
1622         /* This page is already truncated */
1623         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1624                 ClearPageUptodate(page);
1625                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1626                 f2fs_up_read(&sbi->node_write);
1627                 unlock_page(page);
1628                 return 0;
1629         }
1630
1631         if (__is_valid_data_blkaddr(ni.blk_addr) &&
1632                 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1633                                         DATA_GENERIC_ENHANCE)) {
1634                 f2fs_up_read(&sbi->node_write);
1635                 goto redirty_out;
1636         }
1637
1638         if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_sb_has_blkzoned(sbi))
1639                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1640
1641         /* should add to global list before clearing PAGECACHE status */
1642         if (f2fs_in_warm_node_list(sbi, page)) {
1643                 seq = f2fs_add_fsync_node_entry(sbi, page);
1644                 if (seq_id)
1645                         *seq_id = seq;
1646         }
1647
1648         set_page_writeback(page);
1649         ClearPageError(page);
1650
1651         fio.old_blkaddr = ni.blk_addr;
1652         f2fs_do_write_node_page(nid, &fio);
1653         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1654         dec_page_count(sbi, F2FS_DIRTY_NODES);
1655         f2fs_up_read(&sbi->node_write);
1656
1657         if (wbc->for_reclaim) {
1658                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1659                 submitted = NULL;
1660         }
1661
1662         unlock_page(page);
1663
1664         if (unlikely(f2fs_cp_error(sbi))) {
1665                 f2fs_submit_merged_write(sbi, NODE);
1666                 submitted = NULL;
1667         }
1668         if (submitted)
1669                 *submitted = fio.submitted;
1670
1671         if (do_balance)
1672                 f2fs_balance_fs(sbi, false);
1673         return 0;
1674
1675 redirty_out:
1676         redirty_page_for_writepage(wbc, page);
1677         return AOP_WRITEPAGE_ACTIVATE;
1678 }
1679
1680 int f2fs_move_node_page(struct page *node_page, int gc_type)
1681 {
1682         int err = 0;
1683
1684         if (gc_type == FG_GC) {
1685                 struct writeback_control wbc = {
1686                         .sync_mode = WB_SYNC_ALL,
1687                         .nr_to_write = 1,
1688                         .for_reclaim = 0,
1689                 };
1690
1691                 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1692
1693                 set_page_dirty(node_page);
1694
1695                 if (!clear_page_dirty_for_io(node_page)) {
1696                         err = -EAGAIN;
1697                         goto out_page;
1698                 }
1699
1700                 if (__write_node_page(node_page, false, NULL,
1701                                         &wbc, false, FS_GC_NODE_IO, NULL)) {
1702                         err = -EAGAIN;
1703                         unlock_page(node_page);
1704                 }
1705                 goto release_page;
1706         } else {
1707                 /* set page dirty and write it */
1708                 if (!PageWriteback(node_page))
1709                         set_page_dirty(node_page);
1710         }
1711 out_page:
1712         unlock_page(node_page);
1713 release_page:
1714         f2fs_put_page(node_page, 0);
1715         return err;
1716 }
1717
1718 static int f2fs_write_node_page(struct page *page,
1719                                 struct writeback_control *wbc)
1720 {
1721         return __write_node_page(page, false, NULL, wbc, false,
1722                                                 FS_NODE_IO, NULL);
1723 }
1724
1725 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1726                         struct writeback_control *wbc, bool atomic,
1727                         unsigned int *seq_id)
1728 {
1729         pgoff_t index;
1730         struct pagevec pvec;
1731         int ret = 0;
1732         struct page *last_page = NULL;
1733         bool marked = false;
1734         nid_t ino = inode->i_ino;
1735         int nr_pages;
1736         int nwritten = 0;
1737
1738         if (atomic) {
1739                 last_page = last_fsync_dnode(sbi, ino);
1740                 if (IS_ERR_OR_NULL(last_page))
1741                         return PTR_ERR_OR_ZERO(last_page);
1742         }
1743 retry:
1744         pagevec_init(&pvec);
1745         index = 0;
1746
1747         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1748                                 PAGECACHE_TAG_DIRTY))) {
1749                 int i;
1750
1751                 for (i = 0; i < nr_pages; i++) {
1752                         struct page *page = pvec.pages[i];
1753                         bool submitted = false;
1754
1755                         if (unlikely(f2fs_cp_error(sbi))) {
1756                                 f2fs_put_page(last_page, 0);
1757                                 pagevec_release(&pvec);
1758                                 ret = -EIO;
1759                                 goto out;
1760                         }
1761
1762                         if (!IS_DNODE(page) || !is_cold_node(page))
1763                                 continue;
1764                         if (ino_of_node(page) != ino)
1765                                 continue;
1766
1767                         lock_page(page);
1768
1769                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1770 continue_unlock:
1771                                 unlock_page(page);
1772                                 continue;
1773                         }
1774                         if (ino_of_node(page) != ino)
1775                                 goto continue_unlock;
1776
1777                         if (!PageDirty(page) && page != last_page) {
1778                                 /* someone wrote it for us */
1779                                 goto continue_unlock;
1780                         }
1781
1782                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1783
1784                         set_fsync_mark(page, 0);
1785                         set_dentry_mark(page, 0);
1786
1787                         if (!atomic || page == last_page) {
1788                                 set_fsync_mark(page, 1);
1789                                 percpu_counter_inc(&sbi->rf_node_block_count);
1790                                 if (IS_INODE(page)) {
1791                                         if (is_inode_flag_set(inode,
1792                                                                 FI_DIRTY_INODE))
1793                                                 f2fs_update_inode(inode, page);
1794                                         set_dentry_mark(page,
1795                                                 f2fs_need_dentry_mark(sbi, ino));
1796                                 }
1797                                 /* may be written by other thread */
1798                                 if (!PageDirty(page))
1799                                         set_page_dirty(page);
1800                         }
1801
1802                         if (!clear_page_dirty_for_io(page))
1803                                 goto continue_unlock;
1804
1805                         ret = __write_node_page(page, atomic &&
1806                                                 page == last_page,
1807                                                 &submitted, wbc, true,
1808                                                 FS_NODE_IO, seq_id);
1809                         if (ret) {
1810                                 unlock_page(page);
1811                                 f2fs_put_page(last_page, 0);
1812                                 break;
1813                         } else if (submitted) {
1814                                 nwritten++;
1815                         }
1816
1817                         if (page == last_page) {
1818                                 f2fs_put_page(page, 0);
1819                                 marked = true;
1820                                 break;
1821                         }
1822                 }
1823                 pagevec_release(&pvec);
1824                 cond_resched();
1825
1826                 if (ret || marked)
1827                         break;
1828         }
1829         if (!ret && atomic && !marked) {
1830                 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1831                            ino, last_page->index);
1832                 lock_page(last_page);
1833                 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1834                 set_page_dirty(last_page);
1835                 unlock_page(last_page);
1836                 goto retry;
1837         }
1838 out:
1839         if (nwritten)
1840                 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1841         return ret ? -EIO : 0;
1842 }
1843
1844 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1845 {
1846         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1847         bool clean;
1848
1849         if (inode->i_ino != ino)
1850                 return 0;
1851
1852         if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1853                 return 0;
1854
1855         spin_lock(&sbi->inode_lock[DIRTY_META]);
1856         clean = list_empty(&F2FS_I(inode)->gdirty_list);
1857         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1858
1859         if (clean)
1860                 return 0;
1861
1862         inode = igrab(inode);
1863         if (!inode)
1864                 return 0;
1865         return 1;
1866 }
1867
1868 static bool flush_dirty_inode(struct page *page)
1869 {
1870         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1871         struct inode *inode;
1872         nid_t ino = ino_of_node(page);
1873
1874         inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1875         if (!inode)
1876                 return false;
1877
1878         f2fs_update_inode(inode, page);
1879         unlock_page(page);
1880
1881         iput(inode);
1882         return true;
1883 }
1884
1885 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1886 {
1887         pgoff_t index = 0;
1888         struct pagevec pvec;
1889         int nr_pages;
1890
1891         pagevec_init(&pvec);
1892
1893         while ((nr_pages = pagevec_lookup_tag(&pvec,
1894                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1895                 int i;
1896
1897                 for (i = 0; i < nr_pages; i++) {
1898                         struct page *page = pvec.pages[i];
1899
1900                         if (!IS_DNODE(page))
1901                                 continue;
1902
1903                         lock_page(page);
1904
1905                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1906 continue_unlock:
1907                                 unlock_page(page);
1908                                 continue;
1909                         }
1910
1911                         if (!PageDirty(page)) {
1912                                 /* someone wrote it for us */
1913                                 goto continue_unlock;
1914                         }
1915
1916                         /* flush inline_data, if it's async context. */
1917                         if (page_private_inline(page)) {
1918                                 clear_page_private_inline(page);
1919                                 unlock_page(page);
1920                                 flush_inline_data(sbi, ino_of_node(page));
1921                                 continue;
1922                         }
1923                         unlock_page(page);
1924                 }
1925                 pagevec_release(&pvec);
1926                 cond_resched();
1927         }
1928 }
1929
1930 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1931                                 struct writeback_control *wbc,
1932                                 bool do_balance, enum iostat_type io_type)
1933 {
1934         pgoff_t index;
1935         struct pagevec pvec;
1936         int step = 0;
1937         int nwritten = 0;
1938         int ret = 0;
1939         int nr_pages, done = 0;
1940
1941         pagevec_init(&pvec);
1942
1943 next_step:
1944         index = 0;
1945
1946         while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1947                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1948                 int i;
1949
1950                 for (i = 0; i < nr_pages; i++) {
1951                         struct page *page = pvec.pages[i];
1952                         bool submitted = false;
1953
1954                         /* give a priority to WB_SYNC threads */
1955                         if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1956                                         wbc->sync_mode == WB_SYNC_NONE) {
1957                                 done = 1;
1958                                 break;
1959                         }
1960
1961                         /*
1962                          * flushing sequence with step:
1963                          * 0. indirect nodes
1964                          * 1. dentry dnodes
1965                          * 2. file dnodes
1966                          */
1967                         if (step == 0 && IS_DNODE(page))
1968                                 continue;
1969                         if (step == 1 && (!IS_DNODE(page) ||
1970                                                 is_cold_node(page)))
1971                                 continue;
1972                         if (step == 2 && (!IS_DNODE(page) ||
1973                                                 !is_cold_node(page)))
1974                                 continue;
1975 lock_node:
1976                         if (wbc->sync_mode == WB_SYNC_ALL)
1977                                 lock_page(page);
1978                         else if (!trylock_page(page))
1979                                 continue;
1980
1981                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1982 continue_unlock:
1983                                 unlock_page(page);
1984                                 continue;
1985                         }
1986
1987                         if (!PageDirty(page)) {
1988                                 /* someone wrote it for us */
1989                                 goto continue_unlock;
1990                         }
1991
1992                         /* flush inline_data/inode, if it's async context. */
1993                         if (!do_balance)
1994                                 goto write_node;
1995
1996                         /* flush inline_data */
1997                         if (page_private_inline(page)) {
1998                                 clear_page_private_inline(page);
1999                                 unlock_page(page);
2000                                 flush_inline_data(sbi, ino_of_node(page));
2001                                 goto lock_node;
2002                         }
2003
2004                         /* flush dirty inode */
2005                         if (IS_INODE(page) && flush_dirty_inode(page))
2006                                 goto lock_node;
2007 write_node:
2008                         f2fs_wait_on_page_writeback(page, NODE, true, true);
2009
2010                         if (!clear_page_dirty_for_io(page))
2011                                 goto continue_unlock;
2012
2013                         set_fsync_mark(page, 0);
2014                         set_dentry_mark(page, 0);
2015
2016                         ret = __write_node_page(page, false, &submitted,
2017                                                 wbc, do_balance, io_type, NULL);
2018                         if (ret)
2019                                 unlock_page(page);
2020                         else if (submitted)
2021                                 nwritten++;
2022
2023                         if (--wbc->nr_to_write == 0)
2024                                 break;
2025                 }
2026                 pagevec_release(&pvec);
2027                 cond_resched();
2028
2029                 if (wbc->nr_to_write == 0) {
2030                         step = 2;
2031                         break;
2032                 }
2033         }
2034
2035         if (step < 2) {
2036                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2037                                 wbc->sync_mode == WB_SYNC_NONE && step == 1)
2038                         goto out;
2039                 step++;
2040                 goto next_step;
2041         }
2042 out:
2043         if (nwritten)
2044                 f2fs_submit_merged_write(sbi, NODE);
2045
2046         if (unlikely(f2fs_cp_error(sbi)))
2047                 return -EIO;
2048         return ret;
2049 }
2050
2051 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2052                                                 unsigned int seq_id)
2053 {
2054         struct fsync_node_entry *fn;
2055         struct page *page;
2056         struct list_head *head = &sbi->fsync_node_list;
2057         unsigned long flags;
2058         unsigned int cur_seq_id = 0;
2059         int ret2, ret = 0;
2060
2061         while (seq_id && cur_seq_id < seq_id) {
2062                 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2063                 if (list_empty(head)) {
2064                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2065                         break;
2066                 }
2067                 fn = list_first_entry(head, struct fsync_node_entry, list);
2068                 if (fn->seq_id > seq_id) {
2069                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2070                         break;
2071                 }
2072                 cur_seq_id = fn->seq_id;
2073                 page = fn->page;
2074                 get_page(page);
2075                 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2076
2077                 f2fs_wait_on_page_writeback(page, NODE, true, false);
2078                 if (TestClearPageError(page))
2079                         ret = -EIO;
2080
2081                 put_page(page);
2082
2083                 if (ret)
2084                         break;
2085         }
2086
2087         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2088         if (!ret)
2089                 ret = ret2;
2090
2091         return ret;
2092 }
2093
2094 static int f2fs_write_node_pages(struct address_space *mapping,
2095                             struct writeback_control *wbc)
2096 {
2097         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2098         struct blk_plug plug;
2099         long diff;
2100
2101         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2102                 goto skip_write;
2103
2104         /* balancing f2fs's metadata in background */
2105         f2fs_balance_fs_bg(sbi, true);
2106
2107         /* collect a number of dirty node pages and write together */
2108         if (wbc->sync_mode != WB_SYNC_ALL &&
2109                         get_pages(sbi, F2FS_DIRTY_NODES) <
2110                                         nr_pages_to_skip(sbi, NODE))
2111                 goto skip_write;
2112
2113         if (wbc->sync_mode == WB_SYNC_ALL)
2114                 atomic_inc(&sbi->wb_sync_req[NODE]);
2115         else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2116                 /* to avoid potential deadlock */
2117                 if (current->plug)
2118                         blk_finish_plug(current->plug);
2119                 goto skip_write;
2120         }
2121
2122         trace_f2fs_writepages(mapping->host, wbc, NODE);
2123
2124         diff = nr_pages_to_write(sbi, NODE, wbc);
2125         blk_start_plug(&plug);
2126         f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2127         blk_finish_plug(&plug);
2128         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2129
2130         if (wbc->sync_mode == WB_SYNC_ALL)
2131                 atomic_dec(&sbi->wb_sync_req[NODE]);
2132         return 0;
2133
2134 skip_write:
2135         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2136         trace_f2fs_writepages(mapping->host, wbc, NODE);
2137         return 0;
2138 }
2139
2140 static bool f2fs_dirty_node_folio(struct address_space *mapping,
2141                 struct folio *folio)
2142 {
2143         trace_f2fs_set_page_dirty(&folio->page, NODE);
2144
2145         if (!folio_test_uptodate(folio))
2146                 folio_mark_uptodate(folio);
2147 #ifdef CONFIG_F2FS_CHECK_FS
2148         if (IS_INODE(&folio->page))
2149                 f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page);
2150 #endif
2151         if (filemap_dirty_folio(mapping, folio)) {
2152                 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2153                 set_page_private_reference(&folio->page);
2154                 return true;
2155         }
2156         return false;
2157 }
2158
2159 /*
2160  * Structure of the f2fs node operations
2161  */
2162 const struct address_space_operations f2fs_node_aops = {
2163         .writepage      = f2fs_write_node_page,
2164         .writepages     = f2fs_write_node_pages,
2165         .dirty_folio    = f2fs_dirty_node_folio,
2166         .invalidate_folio = f2fs_invalidate_folio,
2167         .release_folio  = f2fs_release_folio,
2168         .migrate_folio  = filemap_migrate_folio,
2169 };
2170
2171 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2172                                                 nid_t n)
2173 {
2174         return radix_tree_lookup(&nm_i->free_nid_root, n);
2175 }
2176
2177 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2178                                 struct free_nid *i)
2179 {
2180         struct f2fs_nm_info *nm_i = NM_I(sbi);
2181         int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2182
2183         if (err)
2184                 return err;
2185
2186         nm_i->nid_cnt[FREE_NID]++;
2187         list_add_tail(&i->list, &nm_i->free_nid_list);
2188         return 0;
2189 }
2190
2191 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2192                         struct free_nid *i, enum nid_state state)
2193 {
2194         struct f2fs_nm_info *nm_i = NM_I(sbi);
2195
2196         f2fs_bug_on(sbi, state != i->state);
2197         nm_i->nid_cnt[state]--;
2198         if (state == FREE_NID)
2199                 list_del(&i->list);
2200         radix_tree_delete(&nm_i->free_nid_root, i->nid);
2201 }
2202
2203 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2204                         enum nid_state org_state, enum nid_state dst_state)
2205 {
2206         struct f2fs_nm_info *nm_i = NM_I(sbi);
2207
2208         f2fs_bug_on(sbi, org_state != i->state);
2209         i->state = dst_state;
2210         nm_i->nid_cnt[org_state]--;
2211         nm_i->nid_cnt[dst_state]++;
2212
2213         switch (dst_state) {
2214         case PREALLOC_NID:
2215                 list_del(&i->list);
2216                 break;
2217         case FREE_NID:
2218                 list_add_tail(&i->list, &nm_i->free_nid_list);
2219                 break;
2220         default:
2221                 BUG_ON(1);
2222         }
2223 }
2224
2225 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2226 {
2227         struct f2fs_nm_info *nm_i = NM_I(sbi);
2228         unsigned int i;
2229         bool ret = true;
2230
2231         f2fs_down_read(&nm_i->nat_tree_lock);
2232         for (i = 0; i < nm_i->nat_blocks; i++) {
2233                 if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2234                         ret = false;
2235                         break;
2236                 }
2237         }
2238         f2fs_up_read(&nm_i->nat_tree_lock);
2239
2240         return ret;
2241 }
2242
2243 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2244                                                         bool set, bool build)
2245 {
2246         struct f2fs_nm_info *nm_i = NM_I(sbi);
2247         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2248         unsigned int nid_ofs = nid - START_NID(nid);
2249
2250         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2251                 return;
2252
2253         if (set) {
2254                 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2255                         return;
2256                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2257                 nm_i->free_nid_count[nat_ofs]++;
2258         } else {
2259                 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2260                         return;
2261                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2262                 if (!build)
2263                         nm_i->free_nid_count[nat_ofs]--;
2264         }
2265 }
2266
2267 /* return if the nid is recognized as free */
2268 static bool add_free_nid(struct f2fs_sb_info *sbi,
2269                                 nid_t nid, bool build, bool update)
2270 {
2271         struct f2fs_nm_info *nm_i = NM_I(sbi);
2272         struct free_nid *i, *e;
2273         struct nat_entry *ne;
2274         int err = -EINVAL;
2275         bool ret = false;
2276
2277         /* 0 nid should not be used */
2278         if (unlikely(nid == 0))
2279                 return false;
2280
2281         if (unlikely(f2fs_check_nid_range(sbi, nid)))
2282                 return false;
2283
2284         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2285         i->nid = nid;
2286         i->state = FREE_NID;
2287
2288         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2289
2290         spin_lock(&nm_i->nid_list_lock);
2291
2292         if (build) {
2293                 /*
2294                  *   Thread A             Thread B
2295                  *  - f2fs_create
2296                  *   - f2fs_new_inode
2297                  *    - f2fs_alloc_nid
2298                  *     - __insert_nid_to_list(PREALLOC_NID)
2299                  *                     - f2fs_balance_fs_bg
2300                  *                      - f2fs_build_free_nids
2301                  *                       - __f2fs_build_free_nids
2302                  *                        - scan_nat_page
2303                  *                         - add_free_nid
2304                  *                          - __lookup_nat_cache
2305                  *  - f2fs_add_link
2306                  *   - f2fs_init_inode_metadata
2307                  *    - f2fs_new_inode_page
2308                  *     - f2fs_new_node_page
2309                  *      - set_node_addr
2310                  *  - f2fs_alloc_nid_done
2311                  *   - __remove_nid_from_list(PREALLOC_NID)
2312                  *                         - __insert_nid_to_list(FREE_NID)
2313                  */
2314                 ne = __lookup_nat_cache(nm_i, nid);
2315                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2316                                 nat_get_blkaddr(ne) != NULL_ADDR))
2317                         goto err_out;
2318
2319                 e = __lookup_free_nid_list(nm_i, nid);
2320                 if (e) {
2321                         if (e->state == FREE_NID)
2322                                 ret = true;
2323                         goto err_out;
2324                 }
2325         }
2326         ret = true;
2327         err = __insert_free_nid(sbi, i);
2328 err_out:
2329         if (update) {
2330                 update_free_nid_bitmap(sbi, nid, ret, build);
2331                 if (!build)
2332                         nm_i->available_nids++;
2333         }
2334         spin_unlock(&nm_i->nid_list_lock);
2335         radix_tree_preload_end();
2336
2337         if (err)
2338                 kmem_cache_free(free_nid_slab, i);
2339         return ret;
2340 }
2341
2342 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2343 {
2344         struct f2fs_nm_info *nm_i = NM_I(sbi);
2345         struct free_nid *i;
2346         bool need_free = false;
2347
2348         spin_lock(&nm_i->nid_list_lock);
2349         i = __lookup_free_nid_list(nm_i, nid);
2350         if (i && i->state == FREE_NID) {
2351                 __remove_free_nid(sbi, i, FREE_NID);
2352                 need_free = true;
2353         }
2354         spin_unlock(&nm_i->nid_list_lock);
2355
2356         if (need_free)
2357                 kmem_cache_free(free_nid_slab, i);
2358 }
2359
2360 static int scan_nat_page(struct f2fs_sb_info *sbi,
2361                         struct page *nat_page, nid_t start_nid)
2362 {
2363         struct f2fs_nm_info *nm_i = NM_I(sbi);
2364         struct f2fs_nat_block *nat_blk = page_address(nat_page);
2365         block_t blk_addr;
2366         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2367         int i;
2368
2369         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2370
2371         i = start_nid % NAT_ENTRY_PER_BLOCK;
2372
2373         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2374                 if (unlikely(start_nid >= nm_i->max_nid))
2375                         break;
2376
2377                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2378
2379                 if (blk_addr == NEW_ADDR)
2380                         return -EINVAL;
2381
2382                 if (blk_addr == NULL_ADDR) {
2383                         add_free_nid(sbi, start_nid, true, true);
2384                 } else {
2385                         spin_lock(&NM_I(sbi)->nid_list_lock);
2386                         update_free_nid_bitmap(sbi, start_nid, false, true);
2387                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2388                 }
2389         }
2390
2391         return 0;
2392 }
2393
2394 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2395 {
2396         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2397         struct f2fs_journal *journal = curseg->journal;
2398         int i;
2399
2400         down_read(&curseg->journal_rwsem);
2401         for (i = 0; i < nats_in_cursum(journal); i++) {
2402                 block_t addr;
2403                 nid_t nid;
2404
2405                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2406                 nid = le32_to_cpu(nid_in_journal(journal, i));
2407                 if (addr == NULL_ADDR)
2408                         add_free_nid(sbi, nid, true, false);
2409                 else
2410                         remove_free_nid(sbi, nid);
2411         }
2412         up_read(&curseg->journal_rwsem);
2413 }
2414
2415 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2416 {
2417         struct f2fs_nm_info *nm_i = NM_I(sbi);
2418         unsigned int i, idx;
2419         nid_t nid;
2420
2421         f2fs_down_read(&nm_i->nat_tree_lock);
2422
2423         for (i = 0; i < nm_i->nat_blocks; i++) {
2424                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2425                         continue;
2426                 if (!nm_i->free_nid_count[i])
2427                         continue;
2428                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2429                         idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2430                                                 NAT_ENTRY_PER_BLOCK, idx);
2431                         if (idx >= NAT_ENTRY_PER_BLOCK)
2432                                 break;
2433
2434                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
2435                         add_free_nid(sbi, nid, true, false);
2436
2437                         if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2438                                 goto out;
2439                 }
2440         }
2441 out:
2442         scan_curseg_cache(sbi);
2443
2444         f2fs_up_read(&nm_i->nat_tree_lock);
2445 }
2446
2447 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2448                                                 bool sync, bool mount)
2449 {
2450         struct f2fs_nm_info *nm_i = NM_I(sbi);
2451         int i = 0, ret;
2452         nid_t nid = nm_i->next_scan_nid;
2453
2454         if (unlikely(nid >= nm_i->max_nid))
2455                 nid = 0;
2456
2457         if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2458                 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2459
2460         /* Enough entries */
2461         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2462                 return 0;
2463
2464         if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2465                 return 0;
2466
2467         if (!mount) {
2468                 /* try to find free nids in free_nid_bitmap */
2469                 scan_free_nid_bits(sbi);
2470
2471                 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2472                         return 0;
2473         }
2474
2475         /* readahead nat pages to be scanned */
2476         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2477                                                         META_NAT, true);
2478
2479         f2fs_down_read(&nm_i->nat_tree_lock);
2480
2481         while (1) {
2482                 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2483                                                 nm_i->nat_block_bitmap)) {
2484                         struct page *page = get_current_nat_page(sbi, nid);
2485
2486                         if (IS_ERR(page)) {
2487                                 ret = PTR_ERR(page);
2488                         } else {
2489                                 ret = scan_nat_page(sbi, page, nid);
2490                                 f2fs_put_page(page, 1);
2491                         }
2492
2493                         if (ret) {
2494                                 f2fs_up_read(&nm_i->nat_tree_lock);
2495                                 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2496                                 return ret;
2497                         }
2498                 }
2499
2500                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2501                 if (unlikely(nid >= nm_i->max_nid))
2502                         nid = 0;
2503
2504                 if (++i >= FREE_NID_PAGES)
2505                         break;
2506         }
2507
2508         /* go to the next free nat pages to find free nids abundantly */
2509         nm_i->next_scan_nid = nid;
2510
2511         /* find free nids from current sum_pages */
2512         scan_curseg_cache(sbi);
2513
2514         f2fs_up_read(&nm_i->nat_tree_lock);
2515
2516         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2517                                         nm_i->ra_nid_pages, META_NAT, false);
2518
2519         return 0;
2520 }
2521
2522 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2523 {
2524         int ret;
2525
2526         mutex_lock(&NM_I(sbi)->build_lock);
2527         ret = __f2fs_build_free_nids(sbi, sync, mount);
2528         mutex_unlock(&NM_I(sbi)->build_lock);
2529
2530         return ret;
2531 }
2532
2533 /*
2534  * If this function returns success, caller can obtain a new nid
2535  * from second parameter of this function.
2536  * The returned nid could be used ino as well as nid when inode is created.
2537  */
2538 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2539 {
2540         struct f2fs_nm_info *nm_i = NM_I(sbi);
2541         struct free_nid *i = NULL;
2542 retry:
2543         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2544                 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2545                 return false;
2546         }
2547
2548         spin_lock(&nm_i->nid_list_lock);
2549
2550         if (unlikely(nm_i->available_nids == 0)) {
2551                 spin_unlock(&nm_i->nid_list_lock);
2552                 return false;
2553         }
2554
2555         /* We should not use stale free nids created by f2fs_build_free_nids */
2556         if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2557                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2558                 i = list_first_entry(&nm_i->free_nid_list,
2559                                         struct free_nid, list);
2560                 *nid = i->nid;
2561
2562                 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2563                 nm_i->available_nids--;
2564
2565                 update_free_nid_bitmap(sbi, *nid, false, false);
2566
2567                 spin_unlock(&nm_i->nid_list_lock);
2568                 return true;
2569         }
2570         spin_unlock(&nm_i->nid_list_lock);
2571
2572         /* Let's scan nat pages and its caches to get free nids */
2573         if (!f2fs_build_free_nids(sbi, true, false))
2574                 goto retry;
2575         return false;
2576 }
2577
2578 /*
2579  * f2fs_alloc_nid() should be called prior to this function.
2580  */
2581 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2582 {
2583         struct f2fs_nm_info *nm_i = NM_I(sbi);
2584         struct free_nid *i;
2585
2586         spin_lock(&nm_i->nid_list_lock);
2587         i = __lookup_free_nid_list(nm_i, nid);
2588         f2fs_bug_on(sbi, !i);
2589         __remove_free_nid(sbi, i, PREALLOC_NID);
2590         spin_unlock(&nm_i->nid_list_lock);
2591
2592         kmem_cache_free(free_nid_slab, i);
2593 }
2594
2595 /*
2596  * f2fs_alloc_nid() should be called prior to this function.
2597  */
2598 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2599 {
2600         struct f2fs_nm_info *nm_i = NM_I(sbi);
2601         struct free_nid *i;
2602         bool need_free = false;
2603
2604         if (!nid)
2605                 return;
2606
2607         spin_lock(&nm_i->nid_list_lock);
2608         i = __lookup_free_nid_list(nm_i, nid);
2609         f2fs_bug_on(sbi, !i);
2610
2611         if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2612                 __remove_free_nid(sbi, i, PREALLOC_NID);
2613                 need_free = true;
2614         } else {
2615                 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2616         }
2617
2618         nm_i->available_nids++;
2619
2620         update_free_nid_bitmap(sbi, nid, true, false);
2621
2622         spin_unlock(&nm_i->nid_list_lock);
2623
2624         if (need_free)
2625                 kmem_cache_free(free_nid_slab, i);
2626 }
2627
2628 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2629 {
2630         struct f2fs_nm_info *nm_i = NM_I(sbi);
2631         int nr = nr_shrink;
2632
2633         if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2634                 return 0;
2635
2636         if (!mutex_trylock(&nm_i->build_lock))
2637                 return 0;
2638
2639         while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2640                 struct free_nid *i, *next;
2641                 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2642
2643                 spin_lock(&nm_i->nid_list_lock);
2644                 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2645                         if (!nr_shrink || !batch ||
2646                                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2647                                 break;
2648                         __remove_free_nid(sbi, i, FREE_NID);
2649                         kmem_cache_free(free_nid_slab, i);
2650                         nr_shrink--;
2651                         batch--;
2652                 }
2653                 spin_unlock(&nm_i->nid_list_lock);
2654         }
2655
2656         mutex_unlock(&nm_i->build_lock);
2657
2658         return nr - nr_shrink;
2659 }
2660
2661 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2662 {
2663         void *src_addr, *dst_addr;
2664         size_t inline_size;
2665         struct page *ipage;
2666         struct f2fs_inode *ri;
2667
2668         ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2669         if (IS_ERR(ipage))
2670                 return PTR_ERR(ipage);
2671
2672         ri = F2FS_INODE(page);
2673         if (ri->i_inline & F2FS_INLINE_XATTR) {
2674                 if (!f2fs_has_inline_xattr(inode)) {
2675                         set_inode_flag(inode, FI_INLINE_XATTR);
2676                         stat_inc_inline_xattr(inode);
2677                 }
2678         } else {
2679                 if (f2fs_has_inline_xattr(inode)) {
2680                         stat_dec_inline_xattr(inode);
2681                         clear_inode_flag(inode, FI_INLINE_XATTR);
2682                 }
2683                 goto update_inode;
2684         }
2685
2686         dst_addr = inline_xattr_addr(inode, ipage);
2687         src_addr = inline_xattr_addr(inode, page);
2688         inline_size = inline_xattr_size(inode);
2689
2690         f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2691         memcpy(dst_addr, src_addr, inline_size);
2692 update_inode:
2693         f2fs_update_inode(inode, ipage);
2694         f2fs_put_page(ipage, 1);
2695         return 0;
2696 }
2697
2698 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2699 {
2700         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2701         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2702         nid_t new_xnid;
2703         struct dnode_of_data dn;
2704         struct node_info ni;
2705         struct page *xpage;
2706         int err;
2707
2708         if (!prev_xnid)
2709                 goto recover_xnid;
2710
2711         /* 1: invalidate the previous xattr nid */
2712         err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2713         if (err)
2714                 return err;
2715
2716         f2fs_invalidate_blocks(sbi, ni.blk_addr);
2717         dec_valid_node_count(sbi, inode, false);
2718         set_node_addr(sbi, &ni, NULL_ADDR, false);
2719
2720 recover_xnid:
2721         /* 2: update xattr nid in inode */
2722         if (!f2fs_alloc_nid(sbi, &new_xnid))
2723                 return -ENOSPC;
2724
2725         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2726         xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2727         if (IS_ERR(xpage)) {
2728                 f2fs_alloc_nid_failed(sbi, new_xnid);
2729                 return PTR_ERR(xpage);
2730         }
2731
2732         f2fs_alloc_nid_done(sbi, new_xnid);
2733         f2fs_update_inode_page(inode);
2734
2735         /* 3: update and set xattr node page dirty */
2736         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2737
2738         set_page_dirty(xpage);
2739         f2fs_put_page(xpage, 1);
2740
2741         return 0;
2742 }
2743
2744 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2745 {
2746         struct f2fs_inode *src, *dst;
2747         nid_t ino = ino_of_node(page);
2748         struct node_info old_ni, new_ni;
2749         struct page *ipage;
2750         int err;
2751
2752         err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2753         if (err)
2754                 return err;
2755
2756         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2757                 return -EINVAL;
2758 retry:
2759         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2760         if (!ipage) {
2761                 memalloc_retry_wait(GFP_NOFS);
2762                 goto retry;
2763         }
2764
2765         /* Should not use this inode from free nid list */
2766         remove_free_nid(sbi, ino);
2767
2768         if (!PageUptodate(ipage))
2769                 SetPageUptodate(ipage);
2770         fill_node_footer(ipage, ino, ino, 0, true);
2771         set_cold_node(ipage, false);
2772
2773         src = F2FS_INODE(page);
2774         dst = F2FS_INODE(ipage);
2775
2776         memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2777         dst->i_size = 0;
2778         dst->i_blocks = cpu_to_le64(1);
2779         dst->i_links = cpu_to_le32(1);
2780         dst->i_xattr_nid = 0;
2781         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2782         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2783                 dst->i_extra_isize = src->i_extra_isize;
2784
2785                 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2786                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2787                                                         i_inline_xattr_size))
2788                         dst->i_inline_xattr_size = src->i_inline_xattr_size;
2789
2790                 if (f2fs_sb_has_project_quota(sbi) &&
2791                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2792                                                                 i_projid))
2793                         dst->i_projid = src->i_projid;
2794
2795                 if (f2fs_sb_has_inode_crtime(sbi) &&
2796                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2797                                                         i_crtime_nsec)) {
2798                         dst->i_crtime = src->i_crtime;
2799                         dst->i_crtime_nsec = src->i_crtime_nsec;
2800                 }
2801         }
2802
2803         new_ni = old_ni;
2804         new_ni.ino = ino;
2805
2806         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2807                 WARN_ON(1);
2808         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2809         inc_valid_inode_count(sbi);
2810         set_page_dirty(ipage);
2811         f2fs_put_page(ipage, 1);
2812         return 0;
2813 }
2814
2815 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2816                         unsigned int segno, struct f2fs_summary_block *sum)
2817 {
2818         struct f2fs_node *rn;
2819         struct f2fs_summary *sum_entry;
2820         block_t addr;
2821         int i, idx, last_offset, nrpages;
2822
2823         /* scan the node segment */
2824         last_offset = sbi->blocks_per_seg;
2825         addr = START_BLOCK(sbi, segno);
2826         sum_entry = &sum->entries[0];
2827
2828         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2829                 nrpages = bio_max_segs(last_offset - i);
2830
2831                 /* readahead node pages */
2832                 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2833
2834                 for (idx = addr; idx < addr + nrpages; idx++) {
2835                         struct page *page = f2fs_get_tmp_page(sbi, idx);
2836
2837                         if (IS_ERR(page))
2838                                 return PTR_ERR(page);
2839
2840                         rn = F2FS_NODE(page);
2841                         sum_entry->nid = rn->footer.nid;
2842                         sum_entry->version = 0;
2843                         sum_entry->ofs_in_node = 0;
2844                         sum_entry++;
2845                         f2fs_put_page(page, 1);
2846                 }
2847
2848                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2849                                                         addr + nrpages);
2850         }
2851         return 0;
2852 }
2853
2854 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2855 {
2856         struct f2fs_nm_info *nm_i = NM_I(sbi);
2857         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2858         struct f2fs_journal *journal = curseg->journal;
2859         int i;
2860
2861         down_write(&curseg->journal_rwsem);
2862         for (i = 0; i < nats_in_cursum(journal); i++) {
2863                 struct nat_entry *ne;
2864                 struct f2fs_nat_entry raw_ne;
2865                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2866
2867                 if (f2fs_check_nid_range(sbi, nid))
2868                         continue;
2869
2870                 raw_ne = nat_in_journal(journal, i);
2871
2872                 ne = __lookup_nat_cache(nm_i, nid);
2873                 if (!ne) {
2874                         ne = __alloc_nat_entry(sbi, nid, true);
2875                         __init_nat_entry(nm_i, ne, &raw_ne, true);
2876                 }
2877
2878                 /*
2879                  * if a free nat in journal has not been used after last
2880                  * checkpoint, we should remove it from available nids,
2881                  * since later we will add it again.
2882                  */
2883                 if (!get_nat_flag(ne, IS_DIRTY) &&
2884                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2885                         spin_lock(&nm_i->nid_list_lock);
2886                         nm_i->available_nids--;
2887                         spin_unlock(&nm_i->nid_list_lock);
2888                 }
2889
2890                 __set_nat_cache_dirty(nm_i, ne);
2891         }
2892         update_nats_in_cursum(journal, -i);
2893         up_write(&curseg->journal_rwsem);
2894 }
2895
2896 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2897                                                 struct list_head *head, int max)
2898 {
2899         struct nat_entry_set *cur;
2900
2901         if (nes->entry_cnt >= max)
2902                 goto add_out;
2903
2904         list_for_each_entry(cur, head, set_list) {
2905                 if (cur->entry_cnt >= nes->entry_cnt) {
2906                         list_add(&nes->set_list, cur->set_list.prev);
2907                         return;
2908                 }
2909         }
2910 add_out:
2911         list_add_tail(&nes->set_list, head);
2912 }
2913
2914 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2915                                                         unsigned int valid)
2916 {
2917         if (valid == 0) {
2918                 __set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2919                 __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2920                 return;
2921         }
2922
2923         __clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2924         if (valid == NAT_ENTRY_PER_BLOCK)
2925                 __set_bit_le(nat_ofs, nm_i->full_nat_bits);
2926         else
2927                 __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2928 }
2929
2930 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2931                                                 struct page *page)
2932 {
2933         struct f2fs_nm_info *nm_i = NM_I(sbi);
2934         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2935         struct f2fs_nat_block *nat_blk = page_address(page);
2936         int valid = 0;
2937         int i = 0;
2938
2939         if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2940                 return;
2941
2942         if (nat_index == 0) {
2943                 valid = 1;
2944                 i = 1;
2945         }
2946         for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2947                 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2948                         valid++;
2949         }
2950
2951         __update_nat_bits(nm_i, nat_index, valid);
2952 }
2953
2954 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2955 {
2956         struct f2fs_nm_info *nm_i = NM_I(sbi);
2957         unsigned int nat_ofs;
2958
2959         f2fs_down_read(&nm_i->nat_tree_lock);
2960
2961         for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
2962                 unsigned int valid = 0, nid_ofs = 0;
2963
2964                 /* handle nid zero due to it should never be used */
2965                 if (unlikely(nat_ofs == 0)) {
2966                         valid = 1;
2967                         nid_ofs = 1;
2968                 }
2969
2970                 for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
2971                         if (!test_bit_le(nid_ofs,
2972                                         nm_i->free_nid_bitmap[nat_ofs]))
2973                                 valid++;
2974                 }
2975
2976                 __update_nat_bits(nm_i, nat_ofs, valid);
2977         }
2978
2979         f2fs_up_read(&nm_i->nat_tree_lock);
2980 }
2981
2982 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2983                 struct nat_entry_set *set, struct cp_control *cpc)
2984 {
2985         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2986         struct f2fs_journal *journal = curseg->journal;
2987         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2988         bool to_journal = true;
2989         struct f2fs_nat_block *nat_blk;
2990         struct nat_entry *ne, *cur;
2991         struct page *page = NULL;
2992
2993         /*
2994          * there are two steps to flush nat entries:
2995          * #1, flush nat entries to journal in current hot data summary block.
2996          * #2, flush nat entries to nat page.
2997          */
2998         if ((cpc->reason & CP_UMOUNT) ||
2999                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3000                 to_journal = false;
3001
3002         if (to_journal) {
3003                 down_write(&curseg->journal_rwsem);
3004         } else {
3005                 page = get_next_nat_page(sbi, start_nid);
3006                 if (IS_ERR(page))
3007                         return PTR_ERR(page);
3008
3009                 nat_blk = page_address(page);
3010                 f2fs_bug_on(sbi, !nat_blk);
3011         }
3012
3013         /* flush dirty nats in nat entry set */
3014         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3015                 struct f2fs_nat_entry *raw_ne;
3016                 nid_t nid = nat_get_nid(ne);
3017                 int offset;
3018
3019                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3020
3021                 if (to_journal) {
3022                         offset = f2fs_lookup_journal_in_cursum(journal,
3023                                                         NAT_JOURNAL, nid, 1);
3024                         f2fs_bug_on(sbi, offset < 0);
3025                         raw_ne = &nat_in_journal(journal, offset);
3026                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
3027                 } else {
3028                         raw_ne = &nat_blk->entries[nid - start_nid];
3029                 }
3030                 raw_nat_from_node_info(raw_ne, &ne->ni);
3031                 nat_reset_flag(ne);
3032                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
3033                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
3034                         add_free_nid(sbi, nid, false, true);
3035                 } else {
3036                         spin_lock(&NM_I(sbi)->nid_list_lock);
3037                         update_free_nid_bitmap(sbi, nid, false, false);
3038                         spin_unlock(&NM_I(sbi)->nid_list_lock);
3039                 }
3040         }
3041
3042         if (to_journal) {
3043                 up_write(&curseg->journal_rwsem);
3044         } else {
3045                 update_nat_bits(sbi, start_nid, page);
3046                 f2fs_put_page(page, 1);
3047         }
3048
3049         /* Allow dirty nats by node block allocation in write_begin */
3050         if (!set->entry_cnt) {
3051                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3052                 kmem_cache_free(nat_entry_set_slab, set);
3053         }
3054         return 0;
3055 }
3056
3057 /*
3058  * This function is called during the checkpointing process.
3059  */
3060 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3061 {
3062         struct f2fs_nm_info *nm_i = NM_I(sbi);
3063         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3064         struct f2fs_journal *journal = curseg->journal;
3065         struct nat_entry_set *setvec[SETVEC_SIZE];
3066         struct nat_entry_set *set, *tmp;
3067         unsigned int found;
3068         nid_t set_idx = 0;
3069         LIST_HEAD(sets);
3070         int err = 0;
3071
3072         /*
3073          * during unmount, let's flush nat_bits before checking
3074          * nat_cnt[DIRTY_NAT].
3075          */
3076         if (cpc->reason & CP_UMOUNT) {
3077                 f2fs_down_write(&nm_i->nat_tree_lock);
3078                 remove_nats_in_journal(sbi);
3079                 f2fs_up_write(&nm_i->nat_tree_lock);
3080         }
3081
3082         if (!nm_i->nat_cnt[DIRTY_NAT])
3083                 return 0;
3084
3085         f2fs_down_write(&nm_i->nat_tree_lock);
3086
3087         /*
3088          * if there are no enough space in journal to store dirty nat
3089          * entries, remove all entries from journal and merge them
3090          * into nat entry set.
3091          */
3092         if (cpc->reason & CP_UMOUNT ||
3093                 !__has_cursum_space(journal,
3094                         nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3095                 remove_nats_in_journal(sbi);
3096
3097         while ((found = __gang_lookup_nat_set(nm_i,
3098                                         set_idx, SETVEC_SIZE, setvec))) {
3099                 unsigned idx;
3100
3101                 set_idx = setvec[found - 1]->set + 1;
3102                 for (idx = 0; idx < found; idx++)
3103                         __adjust_nat_entry_set(setvec[idx], &sets,
3104                                                 MAX_NAT_JENTRIES(journal));
3105         }
3106
3107         /* flush dirty nats in nat entry set */
3108         list_for_each_entry_safe(set, tmp, &sets, set_list) {
3109                 err = __flush_nat_entry_set(sbi, set, cpc);
3110                 if (err)
3111                         break;
3112         }
3113
3114         f2fs_up_write(&nm_i->nat_tree_lock);
3115         /* Allow dirty nats by node block allocation in write_begin */
3116
3117         return err;
3118 }
3119
3120 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3121 {
3122         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3123         struct f2fs_nm_info *nm_i = NM_I(sbi);
3124         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3125         unsigned int i;
3126         __u64 cp_ver = cur_cp_version(ckpt);
3127         block_t nat_bits_addr;
3128
3129         nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3130         nm_i->nat_bits = f2fs_kvzalloc(sbi,
3131                         nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3132         if (!nm_i->nat_bits)
3133                 return -ENOMEM;
3134
3135         nm_i->full_nat_bits = nm_i->nat_bits + 8;
3136         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3137
3138         if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3139                 return 0;
3140
3141         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3142                                                 nm_i->nat_bits_blocks;
3143         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3144                 struct page *page;
3145
3146                 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3147                 if (IS_ERR(page))
3148                         return PTR_ERR(page);
3149
3150                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3151                                         page_address(page), F2FS_BLKSIZE);
3152                 f2fs_put_page(page, 1);
3153         }
3154
3155         cp_ver |= (cur_cp_crc(ckpt) << 32);
3156         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3157                 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3158                 f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3159                         cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3160                 return 0;
3161         }
3162
3163         f2fs_notice(sbi, "Found nat_bits in checkpoint");
3164         return 0;
3165 }
3166
3167 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3168 {
3169         struct f2fs_nm_info *nm_i = NM_I(sbi);
3170         unsigned int i = 0;
3171         nid_t nid, last_nid;
3172
3173         if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3174                 return;
3175
3176         for (i = 0; i < nm_i->nat_blocks; i++) {
3177                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3178                 if (i >= nm_i->nat_blocks)
3179                         break;
3180
3181                 __set_bit_le(i, nm_i->nat_block_bitmap);
3182
3183                 nid = i * NAT_ENTRY_PER_BLOCK;
3184                 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3185
3186                 spin_lock(&NM_I(sbi)->nid_list_lock);
3187                 for (; nid < last_nid; nid++)
3188                         update_free_nid_bitmap(sbi, nid, true, true);
3189                 spin_unlock(&NM_I(sbi)->nid_list_lock);
3190         }
3191
3192         for (i = 0; i < nm_i->nat_blocks; i++) {
3193                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3194                 if (i >= nm_i->nat_blocks)
3195                         break;
3196
3197                 __set_bit_le(i, nm_i->nat_block_bitmap);
3198         }
3199 }
3200
3201 static int init_node_manager(struct f2fs_sb_info *sbi)
3202 {
3203         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3204         struct f2fs_nm_info *nm_i = NM_I(sbi);
3205         unsigned char *version_bitmap;
3206         unsigned int nat_segs;
3207         int err;
3208
3209         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3210
3211         /* segment_count_nat includes pair segment so divide to 2. */
3212         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3213         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3214         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3215
3216         /* not used nids: 0, node, meta, (and root counted as valid node) */
3217         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3218                                                 F2FS_RESERVED_NODE_NUM;
3219         nm_i->nid_cnt[FREE_NID] = 0;
3220         nm_i->nid_cnt[PREALLOC_NID] = 0;
3221         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3222         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3223         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3224         nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3225
3226         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3227         INIT_LIST_HEAD(&nm_i->free_nid_list);
3228         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3229         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3230         INIT_LIST_HEAD(&nm_i->nat_entries);
3231         spin_lock_init(&nm_i->nat_list_lock);
3232
3233         mutex_init(&nm_i->build_lock);
3234         spin_lock_init(&nm_i->nid_list_lock);
3235         init_f2fs_rwsem(&nm_i->nat_tree_lock);
3236
3237         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3238         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3239         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3240         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3241                                         GFP_KERNEL);
3242         if (!nm_i->nat_bitmap)
3243                 return -ENOMEM;
3244
3245         err = __get_nat_bitmaps(sbi);
3246         if (err)
3247                 return err;
3248
3249 #ifdef CONFIG_F2FS_CHECK_FS
3250         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3251                                         GFP_KERNEL);
3252         if (!nm_i->nat_bitmap_mir)
3253                 return -ENOMEM;
3254 #endif
3255
3256         return 0;
3257 }
3258
3259 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3260 {
3261         struct f2fs_nm_info *nm_i = NM_I(sbi);
3262         int i;
3263
3264         nm_i->free_nid_bitmap =
3265                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3266                                               nm_i->nat_blocks),
3267                               GFP_KERNEL);
3268         if (!nm_i->free_nid_bitmap)
3269                 return -ENOMEM;
3270
3271         for (i = 0; i < nm_i->nat_blocks; i++) {
3272                 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3273                         f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3274                 if (!nm_i->free_nid_bitmap[i])
3275                         return -ENOMEM;
3276         }
3277
3278         nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3279                                                                 GFP_KERNEL);
3280         if (!nm_i->nat_block_bitmap)
3281                 return -ENOMEM;
3282
3283         nm_i->free_nid_count =
3284                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3285                                               nm_i->nat_blocks),
3286                               GFP_KERNEL);
3287         if (!nm_i->free_nid_count)
3288                 return -ENOMEM;
3289         return 0;
3290 }
3291
3292 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3293 {
3294         int err;
3295
3296         sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3297                                                         GFP_KERNEL);
3298         if (!sbi->nm_info)
3299                 return -ENOMEM;
3300
3301         err = init_node_manager(sbi);
3302         if (err)
3303                 return err;
3304
3305         err = init_free_nid_cache(sbi);
3306         if (err)
3307                 return err;
3308
3309         /* load free nid status from nat_bits table */
3310         load_free_nid_bitmap(sbi);
3311
3312         return f2fs_build_free_nids(sbi, true, true);
3313 }
3314
3315 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3316 {
3317         struct f2fs_nm_info *nm_i = NM_I(sbi);
3318         struct free_nid *i, *next_i;
3319         struct nat_entry *natvec[NATVEC_SIZE];
3320         struct nat_entry_set *setvec[SETVEC_SIZE];
3321         nid_t nid = 0;
3322         unsigned int found;
3323
3324         if (!nm_i)
3325                 return;
3326
3327         /* destroy free nid list */
3328         spin_lock(&nm_i->nid_list_lock);
3329         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3330                 __remove_free_nid(sbi, i, FREE_NID);
3331                 spin_unlock(&nm_i->nid_list_lock);
3332                 kmem_cache_free(free_nid_slab, i);
3333                 spin_lock(&nm_i->nid_list_lock);
3334         }
3335         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3336         f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3337         f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3338         spin_unlock(&nm_i->nid_list_lock);
3339
3340         /* destroy nat cache */
3341         f2fs_down_write(&nm_i->nat_tree_lock);
3342         while ((found = __gang_lookup_nat_cache(nm_i,
3343                                         nid, NATVEC_SIZE, natvec))) {
3344                 unsigned idx;
3345
3346                 nid = nat_get_nid(natvec[found - 1]) + 1;
3347                 for (idx = 0; idx < found; idx++) {
3348                         spin_lock(&nm_i->nat_list_lock);
3349                         list_del(&natvec[idx]->list);
3350                         spin_unlock(&nm_i->nat_list_lock);
3351
3352                         __del_from_nat_cache(nm_i, natvec[idx]);
3353                 }
3354         }
3355         f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3356
3357         /* destroy nat set cache */
3358         nid = 0;
3359         while ((found = __gang_lookup_nat_set(nm_i,
3360                                         nid, SETVEC_SIZE, setvec))) {
3361                 unsigned idx;
3362
3363                 nid = setvec[found - 1]->set + 1;
3364                 for (idx = 0; idx < found; idx++) {
3365                         /* entry_cnt is not zero, when cp_error was occurred */
3366                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3367                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3368                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3369                 }
3370         }
3371         f2fs_up_write(&nm_i->nat_tree_lock);
3372
3373         kvfree(nm_i->nat_block_bitmap);
3374         if (nm_i->free_nid_bitmap) {
3375                 int i;
3376
3377                 for (i = 0; i < nm_i->nat_blocks; i++)
3378                         kvfree(nm_i->free_nid_bitmap[i]);
3379                 kvfree(nm_i->free_nid_bitmap);
3380         }
3381         kvfree(nm_i->free_nid_count);
3382
3383         kvfree(nm_i->nat_bitmap);
3384         kvfree(nm_i->nat_bits);
3385 #ifdef CONFIG_F2FS_CHECK_FS
3386         kvfree(nm_i->nat_bitmap_mir);
3387 #endif
3388         sbi->nm_info = NULL;
3389         kfree(nm_i);
3390 }
3391
3392 int __init f2fs_create_node_manager_caches(void)
3393 {
3394         nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3395                         sizeof(struct nat_entry));
3396         if (!nat_entry_slab)
3397                 goto fail;
3398
3399         free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3400                         sizeof(struct free_nid));
3401         if (!free_nid_slab)
3402                 goto destroy_nat_entry;
3403
3404         nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3405                         sizeof(struct nat_entry_set));
3406         if (!nat_entry_set_slab)
3407                 goto destroy_free_nid;
3408
3409         fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3410                         sizeof(struct fsync_node_entry));
3411         if (!fsync_node_entry_slab)
3412                 goto destroy_nat_entry_set;
3413         return 0;
3414
3415 destroy_nat_entry_set:
3416         kmem_cache_destroy(nat_entry_set_slab);
3417 destroy_free_nid:
3418         kmem_cache_destroy(free_nid_slab);
3419 destroy_nat_entry:
3420         kmem_cache_destroy(nat_entry_slab);
3421 fail:
3422         return -ENOMEM;
3423 }
3424
3425 void f2fs_destroy_node_manager_caches(void)
3426 {
3427         kmem_cache_destroy(fsync_node_entry_slab);
3428         kmem_cache_destroy(nat_entry_set_slab);
3429         kmem_cache_destroy(free_nid_slab);
3430         kmem_cache_destroy(nat_entry_slab);
3431 }