fs/f2fs/node.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/node.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/mpage.h>
  11 #include <linux/backing-dev.h>
  12 #include <linux/blkdev.h>
  13 #include <linux/pagevec.h>
  14 #include <linux/swap.h>
  15
  16 #include "f2fs.h"
  17 #include "node.h"
  18 #include "segment.h"
  19 #include "xattr.h"
  20 #include "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                 return -EFSCORRUPTED;
  40         }
  41         return 0;
  42 }
  43
  44 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
  45 {
  46         struct f2fs_nm_info *nm_i = NM_I(sbi);
  47         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
  48         struct sysinfo val;
  49         unsigned long avail_ram;
  50         unsigned long mem_size = 0;
  51         bool res = false;
  52
  53         if (!nm_i)
  54                 return true;
  55
  56         si_meminfo(&val);
  57
  58         /* only uses low memory */
  59         avail_ram = val.totalram - val.totalhigh;
  60
  61         /*
  62          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  63          */
  64         if (type == FREE_NIDS) {
  65                 mem_size = (nm_i->nid_cnt[FREE_NID] *
  66                                 sizeof(struct free_nid)) >> PAGE_SHIFT;
  67                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  68         } else if (type == NAT_ENTRIES) {
  69                 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
  70                                 sizeof(struct nat_entry)) >> PAGE_SHIFT;
  71                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  72                 if (excess_cached_nats(sbi))
  73                         res = false;
  74         } else if (type == DIRTY_DENTS) {
  75                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  76                         return false;
  77                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  78                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  79         } else if (type == INO_ENTRIES) {
  80                 int i;
  81
  82                 for (i = 0; i < MAX_INO_ENTRY; i++)
  83                         mem_size += sbi->im[i].ino_num *
  84                                                 sizeof(struct ino_entry);
  85                 mem_size >>= PAGE_SHIFT;
  86                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  87         } else if (type == EXTENT_CACHE) {
  88                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  89                                 sizeof(struct extent_tree) +
  90                                 atomic_read(&sbi->total_ext_node) *
  91                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  92                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  93         } else if (type == INMEM_PAGES) {
  94                 /* it allows 20% / total_ram for inmemory pages */
  95                 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
  96                 res = mem_size < (val.totalram / 5);
  97         } else if (type == DISCARD_CACHE) {
  98                 mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
  99                                 sizeof(struct discard_cmd)) >> PAGE_SHIFT;
 100                 res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
 101         } else if (type == COMPRESS_PAGE) {
 102 #ifdef CONFIG_F2FS_FS_COMPRESSION
 103                 unsigned long free_ram = val.freeram;
 104
 105                 /*
 106                  * free memory is lower than watermark or cached page count
 107                  * exceed threshold, deny caching compress page.
 108                  */
 109                 res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
 110                         (COMPRESS_MAPPING(sbi)->nrpages <
 111                          free_ram * sbi->compress_percent / 100);
 112 #else
 113                 res = false;
 114 #endif
 115         } else {
 116                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
 117                         return true;
 118         }
 119         return res;
 120 }
 121
 122 static void clear_node_page_dirty(struct page *page)
 123 {
 124         if (PageDirty(page)) {
 125                 f2fs_clear_page_cache_dirty_tag(page);
 126                 clear_page_dirty_for_io(page);
 127                 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
 128         }
 129         ClearPageUptodate(page);
 130 }
 131
 132 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 133 {
 134         return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
 135 }
 136
 137 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 138 {
 139         struct page *src_page;
 140         struct page *dst_page;
 141         pgoff_t dst_off;
 142         void *src_addr;
 143         void *dst_addr;
 144         struct f2fs_nm_info *nm_i = NM_I(sbi);
 145
 146         dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
 147
 148         /* get current nat block page with lock */
 149         src_page = get_current_nat_page(sbi, nid);
 150         if (IS_ERR(src_page))
 151                 return src_page;
 152         dst_page = f2fs_grab_meta_page(sbi, dst_off);
 153         f2fs_bug_on(sbi, PageDirty(src_page));
 154
 155         src_addr = page_address(src_page);
 156         dst_addr = page_address(dst_page);
 157         memcpy(dst_addr, src_addr, PAGE_SIZE);
 158         set_page_dirty(dst_page);
 159         f2fs_put_page(src_page, 1);
 160
 161         set_to_next_nat(nm_i, nid);
 162
 163         return dst_page;
 164 }
 165
 166 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
 167                                                 nid_t nid, bool no_fail)
 168 {
 169         struct nat_entry *new;
 170
 171         new = f2fs_kmem_cache_alloc(nat_entry_slab,
 172                                         GFP_F2FS_ZERO, no_fail, sbi);
 173         if (new) {
 174                 nat_set_nid(new, nid);
 175                 nat_reset_flag(new);
 176         }
 177         return new;
 178 }
 179
 180 static void __free_nat_entry(struct nat_entry *e)
 181 {
 182         kmem_cache_free(nat_entry_slab, e);
 183 }
 184
 185 /* must be locked by nat_tree_lock */
 186 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
 187         struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
 188 {
 189         if (no_fail)
 190                 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
 191         else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
 192                 return NULL;
 193
 194         if (raw_ne)
 195                 node_info_from_raw_nat(&ne->ni, raw_ne);
 196
 197         spin_lock(&nm_i->nat_list_lock);
 198         list_add_tail(&ne->list, &nm_i->nat_entries);
 199         spin_unlock(&nm_i->nat_list_lock);
 200
 201         nm_i->nat_cnt[TOTAL_NAT]++;
 202         nm_i->nat_cnt[RECLAIMABLE_NAT]++;
 203         return ne;
 204 }
 205
 206 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 207 {
 208         struct nat_entry *ne;
 209
 210         ne = radix_tree_lookup(&nm_i->nat_root, n);
 211
 212         /* for recent accessed nat entry, move it to tail of lru list */
 213         if (ne && !get_nat_flag(ne, IS_DIRTY)) {
 214                 spin_lock(&nm_i->nat_list_lock);
 215                 if (!list_empty(&ne->list))
 216                         list_move_tail(&ne->list, &nm_i->nat_entries);
 217                 spin_unlock(&nm_i->nat_list_lock);
 218         }
 219
 220         return ne;
 221 }
 222
 223 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 224                 nid_t start, unsigned int nr, struct nat_entry **ep)
 225 {
 226         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 227 }
 228
 229 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 230 {
 231         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 232         nm_i->nat_cnt[TOTAL_NAT]--;
 233         nm_i->nat_cnt[RECLAIMABLE_NAT]--;
 234         __free_nat_entry(e);
 235 }
 236
 237 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
 238                                                         struct nat_entry *ne)
 239 {
 240         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 241         struct nat_entry_set *head;
 242
 243         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 244         if (!head) {
 245                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
 246                                                 GFP_NOFS, true, NULL);
 247
 248                 INIT_LIST_HEAD(&head->entry_list);
 249                 INIT_LIST_HEAD(&head->set_list);
 250                 head->set = set;
 251                 head->entry_cnt = 0;
 252                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 253         }
 254         return head;
 255 }
 256
 257 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 258                                                 struct nat_entry *ne)
 259 {
 260         struct nat_entry_set *head;
 261         bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
 262
 263         if (!new_ne)
 264                 head = __grab_nat_entry_set(nm_i, ne);
 265
 266         /*
 267          * update entry_cnt in below condition:
 268          * 1. update NEW_ADDR to valid block address;
 269          * 2. update old block address to new one;
 270          */
 271         if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
 272                                 !get_nat_flag(ne, IS_DIRTY)))
 273                 head->entry_cnt++;
 274
 275         set_nat_flag(ne, IS_PREALLOC, new_ne);
 276
 277         if (get_nat_flag(ne, IS_DIRTY))
 278                 goto refresh_list;
 279
 280         nm_i->nat_cnt[DIRTY_NAT]++;
 281         nm_i->nat_cnt[RECLAIMABLE_NAT]--;
 282         set_nat_flag(ne, IS_DIRTY, true);
 283 refresh_list:
 284         spin_lock(&nm_i->nat_list_lock);
 285         if (new_ne)
 286                 list_del_init(&ne->list);
 287         else
 288                 list_move_tail(&ne->list, &head->entry_list);
 289         spin_unlock(&nm_i->nat_list_lock);
 290 }
 291
 292 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 293                 struct nat_entry_set *set, struct nat_entry *ne)
 294 {
 295         spin_lock(&nm_i->nat_list_lock);
 296         list_move_tail(&ne->list, &nm_i->nat_entries);
 297         spin_unlock(&nm_i->nat_list_lock);
 298
 299         set_nat_flag(ne, IS_DIRTY, false);
 300         set->entry_cnt--;
 301         nm_i->nat_cnt[DIRTY_NAT]--;
 302         nm_i->nat_cnt[RECLAIMABLE_NAT]++;
 303 }
 304
 305 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 306                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 307 {
 308         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 309                                                         start, nr);
 310 }
 311
 312 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
 313 {
 314         return NODE_MAPPING(sbi) == page->mapping &&
 315                         IS_DNODE(page) && is_cold_node(page);
 316 }
 317
 318 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
 319 {
 320         spin_lock_init(&sbi->fsync_node_lock);
 321         INIT_LIST_HEAD(&sbi->fsync_node_list);
 322         sbi->fsync_seg_id = 0;
 323         sbi->fsync_node_num = 0;
 324 }
 325
 326 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
 327                                                         struct page *page)
 328 {
 329         struct fsync_node_entry *fn;
 330         unsigned long flags;
 331         unsigned int seq_id;
 332
 333         fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
 334                                         GFP_NOFS, true, NULL);
 335
 336         get_page(page);
 337         fn->page = page;
 338         INIT_LIST_HEAD(&fn->list);
 339
 340         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 341         list_add_tail(&fn->list, &sbi->fsync_node_list);
 342         fn->seq_id = sbi->fsync_seg_id++;
 343         seq_id = fn->seq_id;
 344         sbi->fsync_node_num++;
 345         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 346
 347         return seq_id;
 348 }
 349
 350 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
 351 {
 352         struct fsync_node_entry *fn;
 353         unsigned long flags;
 354
 355         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 356         list_for_each_entry(fn, &sbi->fsync_node_list, list) {
 357                 if (fn->page == page) {
 358                         list_del(&fn->list);
 359                         sbi->fsync_node_num--;
 360                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 361                         kmem_cache_free(fsync_node_entry_slab, fn);
 362                         put_page(page);
 363                         return;
 364                 }
 365         }
 366         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 367         f2fs_bug_on(sbi, 1);
 368 }
 369
 370 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
 371 {
 372         unsigned long flags;
 373
 374         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 375         sbi->fsync_seg_id = 0;
 376         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 377 }
 378
 379 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 380 {
 381         struct f2fs_nm_info *nm_i = NM_I(sbi);
 382         struct nat_entry *e;
 383         bool need = false;
 384
 385         down_read(&nm_i->nat_tree_lock);
 386         e = __lookup_nat_cache(nm_i, nid);
 387         if (e) {
 388                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 389                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 390                         need = true;
 391         }
 392         up_read(&nm_i->nat_tree_lock);
 393         return need;
 394 }
 395
 396 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 397 {
 398         struct f2fs_nm_info *nm_i = NM_I(sbi);
 399         struct nat_entry *e;
 400         bool is_cp = true;
 401
 402         down_read(&nm_i->nat_tree_lock);
 403         e = __lookup_nat_cache(nm_i, nid);
 404         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 405                 is_cp = false;
 406         up_read(&nm_i->nat_tree_lock);
 407         return is_cp;
 408 }
 409
 410 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 411 {
 412         struct f2fs_nm_info *nm_i = NM_I(sbi);
 413         struct nat_entry *e;
 414         bool need_update = true;
 415
 416         down_read(&nm_i->nat_tree_lock);
 417         e = __lookup_nat_cache(nm_i, ino);
 418         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 419                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 420                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 421                 need_update = false;
 422         up_read(&nm_i->nat_tree_lock);
 423         return need_update;
 424 }
 425
 426 /* must be locked by nat_tree_lock */
 427 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 428                                                 struct f2fs_nat_entry *ne)
 429 {
 430         struct f2fs_nm_info *nm_i = NM_I(sbi);
 431         struct nat_entry *new, *e;
 432
 433         new = __alloc_nat_entry(sbi, nid, false);
 434         if (!new)
 435                 return;
 436
 437         down_write(&nm_i->nat_tree_lock);
 438         e = __lookup_nat_cache(nm_i, nid);
 439         if (!e)
 440                 e = __init_nat_entry(nm_i, new, ne, false);
 441         else
 442                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 443                                 nat_get_blkaddr(e) !=
 444                                         le32_to_cpu(ne->block_addr) ||
 445                                 nat_get_version(e) != ne->version);
 446         up_write(&nm_i->nat_tree_lock);
 447         if (e != new)
 448                 __free_nat_entry(new);
 449 }
 450
 451 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 452                         block_t new_blkaddr, bool fsync_done)
 453 {
 454         struct f2fs_nm_info *nm_i = NM_I(sbi);
 455         struct nat_entry *e;
 456         struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
 457
 458         down_write(&nm_i->nat_tree_lock);
 459         e = __lookup_nat_cache(nm_i, ni->nid);
 460         if (!e) {
 461                 e = __init_nat_entry(nm_i, new, NULL, true);
 462                 copy_node_info(&e->ni, ni);
 463                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 464         } else if (new_blkaddr == NEW_ADDR) {
 465                 /*
 466                  * when nid is reallocated,
 467                  * previous nat entry can be remained in nat cache.
 468                  * So, reinitialize it with new information.
 469                  */
 470                 copy_node_info(&e->ni, ni);
 471                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 472         }
 473         /* let's free early to reduce memory consumption */
 474         if (e != new)
 475                 __free_nat_entry(new);
 476
 477         /* sanity check */
 478         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 479         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 480                         new_blkaddr == NULL_ADDR);
 481         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 482                         new_blkaddr == NEW_ADDR);
 483         f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
 484                         new_blkaddr == NEW_ADDR);
 485
 486         /* increment version no as node is removed */
 487         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 488                 unsigned char version = nat_get_version(e);
 489
 490                 nat_set_version(e, inc_node_version(version));
 491         }
 492
 493         /* change address */
 494         nat_set_blkaddr(e, new_blkaddr);
 495         if (!__is_valid_data_blkaddr(new_blkaddr))
 496                 set_nat_flag(e, IS_CHECKPOINTED, false);
 497         __set_nat_cache_dirty(nm_i, e);
 498
 499         /* update fsync_mark if its inode nat entry is still alive */
 500         if (ni->nid != ni->ino)
 501                 e = __lookup_nat_cache(nm_i, ni->ino);
 502         if (e) {
 503                 if (fsync_done && ni->nid == ni->ino)
 504                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 505                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 506         }
 507         up_write(&nm_i->nat_tree_lock);
 508 }
 509
 510 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 511 {
 512         struct f2fs_nm_info *nm_i = NM_I(sbi);
 513         int nr = nr_shrink;
 514
 515         if (!down_write_trylock(&nm_i->nat_tree_lock))
 516                 return 0;
 517
 518         spin_lock(&nm_i->nat_list_lock);
 519         while (nr_shrink) {
 520                 struct nat_entry *ne;
 521
 522                 if (list_empty(&nm_i->nat_entries))
 523                         break;
 524
 525                 ne = list_first_entry(&nm_i->nat_entries,
 526                                         struct nat_entry, list);
 527                 list_del(&ne->list);
 528                 spin_unlock(&nm_i->nat_list_lock);
 529
 530                 __del_from_nat_cache(nm_i, ne);
 531                 nr_shrink--;
 532
 533                 spin_lock(&nm_i->nat_list_lock);
 534         }
 535         spin_unlock(&nm_i->nat_list_lock);
 536
 537         up_write(&nm_i->nat_tree_lock);
 538         return nr - nr_shrink;
 539 }
 540
 541 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
 542                                                 struct node_info *ni)
 543 {
 544         struct f2fs_nm_info *nm_i = NM_I(sbi);
 545         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 546         struct f2fs_journal *journal = curseg->journal;
 547         nid_t start_nid = START_NID(nid);
 548         struct f2fs_nat_block *nat_blk;
 549         struct page *page = NULL;
 550         struct f2fs_nat_entry ne;
 551         struct nat_entry *e;
 552         pgoff_t index;
 553         block_t blkaddr;
 554         int i;
 555
 556         ni->nid = nid;
 557 retry:
 558         /* Check nat cache */
 559         down_read(&nm_i->nat_tree_lock);
 560         e = __lookup_nat_cache(nm_i, nid);
 561         if (e) {
 562                 ni->ino = nat_get_ino(e);
 563                 ni->blk_addr = nat_get_blkaddr(e);
 564                 ni->version = nat_get_version(e);
 565                 up_read(&nm_i->nat_tree_lock);
 566                 return 0;
 567         }
 568
 569         /*
 570          * Check current segment summary by trying to grab journal_rwsem first.
 571          * This sem is on the critical path on the checkpoint requiring the above
 572          * nat_tree_lock. Therefore, we should retry, if we failed to grab here
 573          * while not bothering checkpoint.
 574          */
 575         if (!rwsem_is_locked(&sbi->cp_global_sem)) {
 576                 down_read(&curseg->journal_rwsem);
 577         } else if (!down_read_trylock(&curseg->journal_rwsem)) {
 578                 up_read(&nm_i->nat_tree_lock);
 579                 goto retry;
 580         }
 581
 582         i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 583         if (i >= 0) {
 584                 ne = nat_in_journal(journal, i);
 585                 node_info_from_raw_nat(ni, &ne);
 586         }
 587         up_read(&curseg->journal_rwsem);
 588         if (i >= 0) {
 589                 up_read(&nm_i->nat_tree_lock);
 590                 goto cache;
 591         }
 592
 593         /* Fill node_info from nat page */
 594         index = current_nat_addr(sbi, nid);
 595         up_read(&nm_i->nat_tree_lock);
 596
 597         page = f2fs_get_meta_page(sbi, index);
 598         if (IS_ERR(page))
 599                 return PTR_ERR(page);
 600
 601         nat_blk = (struct f2fs_nat_block *)page_address(page);
 602         ne = nat_blk->entries[nid - start_nid];
 603         node_info_from_raw_nat(ni, &ne);
 604         f2fs_put_page(page, 1);
 605 cache:
 606         blkaddr = le32_to_cpu(ne.block_addr);
 607         if (__is_valid_data_blkaddr(blkaddr) &&
 608                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
 609                 return -EFAULT;
 610
 611         /* cache nat entry */
 612         cache_nat_entry(sbi, nid, &ne);
 613         return 0;
 614 }
 615
 616 /*
 617  * readahead MAX_RA_NODE number of node pages.
 618  */
 619 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
 620 {
 621         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 622         struct blk_plug plug;
 623         int i, end;
 624         nid_t nid;
 625
 626         blk_start_plug(&plug);
 627
 628         /* Then, try readahead for siblings of the desired node */
 629         end = start + n;
 630         end = min(end, NIDS_PER_BLOCK);
 631         for (i = start; i < end; i++) {
 632                 nid = get_nid(parent, i, false);
 633                 f2fs_ra_node_page(sbi, nid);
 634         }
 635
 636         blk_finish_plug(&plug);
 637 }
 638
 639 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 640 {
 641         const long direct_index = ADDRS_PER_INODE(dn->inode);
 642         const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
 643         const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
 644         unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
 645         int cur_level = dn->cur_level;
 646         int max_level = dn->max_level;
 647         pgoff_t base = 0;
 648
 649         if (!dn->max_level)
 650                 return pgofs + 1;
 651
 652         while (max_level-- > cur_level)
 653                 skipped_unit *= NIDS_PER_BLOCK;
 654
 655         switch (dn->max_level) {
 656         case 3:
 657                 base += 2 * indirect_blks;
 658                 fallthrough;
 659         case 2:
 660                 base += 2 * direct_blks;
 661                 fallthrough;
 662         case 1:
 663                 base += direct_index;
 664                 break;
 665         default:
 666                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 667         }
 668
 669         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 670 }
 671
 672 /*
 673  * The maximum depth is four.
 674  * Offset[0] will have raw inode offset.
 675  */
 676 static int get_node_path(struct inode *inode, long block,
 677                                 int offset[4], unsigned int noffset[4])
 678 {
 679         const long direct_index = ADDRS_PER_INODE(inode);
 680         const long direct_blks = ADDRS_PER_BLOCK(inode);
 681         const long dptrs_per_blk = NIDS_PER_BLOCK;
 682         const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
 683         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 684         int n = 0;
 685         int level = 0;
 686
 687         noffset[0] = 0;
 688
 689         if (block < direct_index) {
 690                 offset[n] = block;
 691                 goto got;
 692         }
 693         block -= direct_index;
 694         if (block < direct_blks) {
 695                 offset[n++] = NODE_DIR1_BLOCK;
 696                 noffset[n] = 1;
 697                 offset[n] = block;
 698                 level = 1;
 699                 goto got;
 700         }
 701         block -= direct_blks;
 702         if (block < direct_blks) {
 703                 offset[n++] = NODE_DIR2_BLOCK;
 704                 noffset[n] = 2;
 705                 offset[n] = block;
 706                 level = 1;
 707                 goto got;
 708         }
 709         block -= direct_blks;
 710         if (block < indirect_blks) {
 711                 offset[n++] = NODE_IND1_BLOCK;
 712                 noffset[n] = 3;
 713                 offset[n++] = block / direct_blks;
 714                 noffset[n] = 4 + offset[n - 1];
 715                 offset[n] = block % direct_blks;
 716                 level = 2;
 717                 goto got;
 718         }
 719         block -= indirect_blks;
 720         if (block < indirect_blks) {
 721                 offset[n++] = NODE_IND2_BLOCK;
 722                 noffset[n] = 4 + dptrs_per_blk;
 723                 offset[n++] = block / direct_blks;
 724                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 725                 offset[n] = block % direct_blks;
 726                 level = 2;
 727                 goto got;
 728         }
 729         block -= indirect_blks;
 730         if (block < dindirect_blks) {
 731                 offset[n++] = NODE_DIND_BLOCK;
 732                 noffset[n] = 5 + (dptrs_per_blk * 2);
 733                 offset[n++] = block / indirect_blks;
 734                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 735                               offset[n - 1] * (dptrs_per_blk + 1);
 736                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 737                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 738                               offset[n - 2] * (dptrs_per_blk + 1) +
 739                               offset[n - 1];
 740                 offset[n] = block % direct_blks;
 741                 level = 3;
 742                 goto got;
 743         } else {
 744                 return -E2BIG;
 745         }
 746 got:
 747         return level;
 748 }
 749
 750 /*
 751  * Caller should call f2fs_put_dnode(dn).
 752  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 753  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
 754  */
 755 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 756 {
 757         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 758         struct page *npage[4];
 759         struct page *parent = NULL;
 760         int offset[4];
 761         unsigned int noffset[4];
 762         nid_t nids[4];
 763         int level, i = 0;
 764         int err = 0;
 765
 766         level = get_node_path(dn->inode, index, offset, noffset);
 767         if (level < 0)
 768                 return level;
 769
 770         nids[0] = dn->inode->i_ino;
 771         npage[0] = dn->inode_page;
 772
 773         if (!npage[0]) {
 774                 npage[0] = f2fs_get_node_page(sbi, nids[0]);
 775                 if (IS_ERR(npage[0]))
 776                         return PTR_ERR(npage[0]);
 777         }
 778
 779         /* if inline_data is set, should not report any block indices */
 780         if (f2fs_has_inline_data(dn->inode) && index) {
 781                 err = -ENOENT;
 782                 f2fs_put_page(npage[0], 1);
 783                 goto release_out;
 784         }
 785
 786         parent = npage[0];
 787         if (level != 0)
 788                 nids[1] = get_nid(parent, offset[0], true);
 789         dn->inode_page = npage[0];
 790         dn->inode_page_locked = true;
 791
 792         /* get indirect or direct nodes */
 793         for (i = 1; i <= level; i++) {
 794                 bool done = false;
 795
 796                 if (!nids[i] && mode == ALLOC_NODE) {
 797                         /* alloc new node */
 798                         if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
 799                                 err = -ENOSPC;
 800                                 goto release_pages;
 801                         }
 802
 803                         dn->nid = nids[i];
 804                         npage[i] = f2fs_new_node_page(dn, noffset[i]);
 805                         if (IS_ERR(npage[i])) {
 806                                 f2fs_alloc_nid_failed(sbi, nids[i]);
 807                                 err = PTR_ERR(npage[i]);
 808                                 goto release_pages;
 809                         }
 810
 811                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 812                         f2fs_alloc_nid_done(sbi, nids[i]);
 813                         done = true;
 814                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 815                         npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
 816                         if (IS_ERR(npage[i])) {
 817                                 err = PTR_ERR(npage[i]);
 818                                 goto release_pages;
 819                         }
 820                         done = true;
 821                 }
 822                 if (i == 1) {
 823                         dn->inode_page_locked = false;
 824                         unlock_page(parent);
 825                 } else {
 826                         f2fs_put_page(parent, 1);
 827                 }
 828
 829                 if (!done) {
 830                         npage[i] = f2fs_get_node_page(sbi, nids[i]);
 831                         if (IS_ERR(npage[i])) {
 832                                 err = PTR_ERR(npage[i]);
 833                                 f2fs_put_page(npage[0], 0);
 834                                 goto release_out;
 835                         }
 836                 }
 837                 if (i < level) {
 838                         parent = npage[i];
 839                         nids[i + 1] = get_nid(parent, offset[i], false);
 840                 }
 841         }
 842         dn->nid = nids[level];
 843         dn->ofs_in_node = offset[level];
 844         dn->node_page = npage[level];
 845         dn->data_blkaddr = f2fs_data_blkaddr(dn);
 846
 847         if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
 848                                         f2fs_sb_has_readonly(sbi)) {
 849                 unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn);
 850                 block_t blkaddr;
 851
 852                 if (!c_len)
 853                         goto out;
 854
 855                 blkaddr = f2fs_data_blkaddr(dn);
 856                 if (blkaddr == COMPRESS_ADDR)
 857                         blkaddr = data_blkaddr(dn->inode, dn->node_page,
 858                                                 dn->ofs_in_node + 1);
 859
 860                 f2fs_update_extent_tree_range_compressed(dn->inode,
 861                                         index, blkaddr,
 862                                         F2FS_I(dn->inode)->i_cluster_size,
 863                                         c_len);
 864         }
 865 out:
 866         return 0;
 867
 868 release_pages:
 869         f2fs_put_page(parent, 1);
 870         if (i > 1)
 871                 f2fs_put_page(npage[0], 0);
 872 release_out:
 873         dn->inode_page = NULL;
 874         dn->node_page = NULL;
 875         if (err == -ENOENT) {
 876                 dn->cur_level = i;
 877                 dn->max_level = level;
 878                 dn->ofs_in_node = offset[level];
 879         }
 880         return err;
 881 }
 882
 883 static int truncate_node(struct dnode_of_data *dn)
 884 {
 885         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 886         struct node_info ni;
 887         int err;
 888         pgoff_t index;
 889
 890         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 891         if (err)
 892                 return err;
 893
 894         /* Deallocate node address */
 895         f2fs_invalidate_blocks(sbi, ni.blk_addr);
 896         dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
 897         set_node_addr(sbi, &ni, NULL_ADDR, false);
 898
 899         if (dn->nid == dn->inode->i_ino) {
 900                 f2fs_remove_orphan_inode(sbi, dn->nid);
 901                 dec_valid_inode_count(sbi);
 902                 f2fs_inode_synced(dn->inode);
 903         }
 904
 905         clear_node_page_dirty(dn->node_page);
 906         set_sbi_flag(sbi, SBI_IS_DIRTY);
 907
 908         index = dn->node_page->index;
 909         f2fs_put_page(dn->node_page, 1);
 910
 911         invalidate_mapping_pages(NODE_MAPPING(sbi),
 912                         index, index);
 913
 914         dn->node_page = NULL;
 915         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 916
 917         return 0;
 918 }
 919
 920 static int truncate_dnode(struct dnode_of_data *dn)
 921 {
 922         struct page *page;
 923         int err;
 924
 925         if (dn->nid == 0)
 926                 return 1;
 927
 928         /* get direct node */
 929         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 930         if (PTR_ERR(page) == -ENOENT)
 931                 return 1;
 932         else if (IS_ERR(page))
 933                 return PTR_ERR(page);
 934
 935         /* Make dnode_of_data for parameter */
 936         dn->node_page = page;
 937         dn->ofs_in_node = 0;
 938         f2fs_truncate_data_blocks(dn);
 939         err = truncate_node(dn);
 940         if (err)
 941                 return err;
 942
 943         return 1;
 944 }
 945
 946 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 947                                                 int ofs, int depth)
 948 {
 949         struct dnode_of_data rdn = *dn;
 950         struct page *page;
 951         struct f2fs_node *rn;
 952         nid_t child_nid;
 953         unsigned int child_nofs;
 954         int freed = 0;
 955         int i, ret;
 956
 957         if (dn->nid == 0)
 958                 return NIDS_PER_BLOCK + 1;
 959
 960         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 961
 962         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 963         if (IS_ERR(page)) {
 964                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 965                 return PTR_ERR(page);
 966         }
 967
 968         f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 969
 970         rn = F2FS_NODE(page);
 971         if (depth < 3) {
 972                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 973                         child_nid = le32_to_cpu(rn->in.nid[i]);
 974                         if (child_nid == 0)
 975                                 continue;
 976                         rdn.nid = child_nid;
 977                         ret = truncate_dnode(&rdn);
 978                         if (ret < 0)
 979                                 goto out_err;
 980                         if (set_nid(page, i, 0, false))
 981                                 dn->node_changed = true;
 982                 }
 983         } else {
 984                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 985                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 986                         child_nid = le32_to_cpu(rn->in.nid[i]);
 987                         if (child_nid == 0) {
 988                                 child_nofs += NIDS_PER_BLOCK + 1;
 989                                 continue;
 990                         }
 991                         rdn.nid = child_nid;
 992                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 993                         if (ret == (NIDS_PER_BLOCK + 1)) {
 994                                 if (set_nid(page, i, 0, false))
 995                                         dn->node_changed = true;
 996                                 child_nofs += ret;
 997                         } else if (ret < 0 && ret != -ENOENT) {
 998                                 goto out_err;
 999                         }
1000                 }
1001                 freed = child_nofs;
1002         }
1003
1004         if (!ofs) {
1005                 /* remove current indirect node */
1006                 dn->node_page = page;
1007                 ret = truncate_node(dn);
1008                 if (ret)
1009                         goto out_err;
1010                 freed++;
1011         } else {
1012                 f2fs_put_page(page, 1);
1013         }
1014         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1015         return freed;
1016
1017 out_err:
1018         f2fs_put_page(page, 1);
1019         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1020         return ret;
1021 }
1022
1023 static int truncate_partial_nodes(struct dnode_of_data *dn,
1024                         struct f2fs_inode *ri, int *offset, int depth)
1025 {
1026         struct page *pages[2];
1027         nid_t nid[3];
1028         nid_t child_nid;
1029         int err = 0;
1030         int i;
1031         int idx = depth - 2;
1032
1033         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1034         if (!nid[0])
1035                 return 0;
1036
1037         /* get indirect nodes in the path */
1038         for (i = 0; i < idx + 1; i++) {
1039                 /* reference count'll be increased */
1040                 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1041                 if (IS_ERR(pages[i])) {
1042                         err = PTR_ERR(pages[i]);
1043                         idx = i - 1;
1044                         goto fail;
1045                 }
1046                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1047         }
1048
1049         f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1050
1051         /* free direct nodes linked to a partial indirect node */
1052         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1053                 child_nid = get_nid(pages[idx], i, false);
1054                 if (!child_nid)
1055                         continue;
1056                 dn->nid = child_nid;
1057                 err = truncate_dnode(dn);
1058                 if (err < 0)
1059                         goto fail;
1060                 if (set_nid(pages[idx], i, 0, false))
1061                         dn->node_changed = true;
1062         }
1063
1064         if (offset[idx + 1] == 0) {
1065                 dn->node_page = pages[idx];
1066                 dn->nid = nid[idx];
1067                 err = truncate_node(dn);
1068                 if (err)
1069                         goto fail;
1070         } else {
1071                 f2fs_put_page(pages[idx], 1);
1072         }
1073         offset[idx]++;
1074         offset[idx + 1] = 0;
1075         idx--;
1076 fail:
1077         for (i = idx; i >= 0; i--)
1078                 f2fs_put_page(pages[i], 1);
1079
1080         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1081
1082         return err;
1083 }
1084
1085 /*
1086  * All the block addresses of data and nodes should be nullified.
1087  */
1088 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1089 {
1090         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1091         int err = 0, cont = 1;
1092         int level, offset[4], noffset[4];
1093         unsigned int nofs = 0;
1094         struct f2fs_inode *ri;
1095         struct dnode_of_data dn;
1096         struct page *page;
1097
1098         trace_f2fs_truncate_inode_blocks_enter(inode, from);
1099
1100         level = get_node_path(inode, from, offset, noffset);
1101         if (level < 0) {
1102                 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1103                 return level;
1104         }
1105
1106         page = f2fs_get_node_page(sbi, inode->i_ino);
1107         if (IS_ERR(page)) {
1108                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1109                 return PTR_ERR(page);
1110         }
1111
1112         set_new_dnode(&dn, inode, page, NULL, 0);
1113         unlock_page(page);
1114
1115         ri = F2FS_INODE(page);
1116         switch (level) {
1117         case 0:
1118         case 1:
1119                 nofs = noffset[1];
1120                 break;
1121         case 2:
1122                 nofs = noffset[1];
1123                 if (!offset[level - 1])
1124                         goto skip_partial;
1125                 err = truncate_partial_nodes(&dn, ri, offset, level);
1126                 if (err < 0 && err != -ENOENT)
1127                         goto fail;
1128                 nofs += 1 + NIDS_PER_BLOCK;
1129                 break;
1130         case 3:
1131                 nofs = 5 + 2 * NIDS_PER_BLOCK;
1132                 if (!offset[level - 1])
1133                         goto skip_partial;
1134                 err = truncate_partial_nodes(&dn, ri, offset, level);
1135                 if (err < 0 && err != -ENOENT)
1136                         goto fail;
1137                 break;
1138         default:
1139                 BUG();
1140         }
1141
1142 skip_partial:
1143         while (cont) {
1144                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1145                 switch (offset[0]) {
1146                 case NODE_DIR1_BLOCK:
1147                 case NODE_DIR2_BLOCK:
1148                         err = truncate_dnode(&dn);
1149                         break;
1150
1151                 case NODE_IND1_BLOCK:
1152                 case NODE_IND2_BLOCK:
1153                         err = truncate_nodes(&dn, nofs, offset[1], 2);
1154                         break;
1155
1156                 case NODE_DIND_BLOCK:
1157                         err = truncate_nodes(&dn, nofs, offset[1], 3);
1158                         cont = 0;
1159                         break;
1160
1161                 default:
1162                         BUG();
1163                 }
1164                 if (err < 0 && err != -ENOENT)
1165                         goto fail;
1166                 if (offset[1] == 0 &&
1167                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1168                         lock_page(page);
1169                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
1170                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1171                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1172                         set_page_dirty(page);
1173                         unlock_page(page);
1174                 }
1175                 offset[1] = 0;
1176                 offset[0]++;
1177                 nofs += err;
1178         }
1179 fail:
1180         f2fs_put_page(page, 0);
1181         trace_f2fs_truncate_inode_blocks_exit(inode, err);
1182         return err > 0 ? 0 : err;
1183 }
1184
1185 /* caller must lock inode page */
1186 int f2fs_truncate_xattr_node(struct inode *inode)
1187 {
1188         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1189         nid_t nid = F2FS_I(inode)->i_xattr_nid;
1190         struct dnode_of_data dn;
1191         struct page *npage;
1192         int err;
1193
1194         if (!nid)
1195                 return 0;
1196
1197         npage = f2fs_get_node_page(sbi, nid);
1198         if (IS_ERR(npage))
1199                 return PTR_ERR(npage);
1200
1201         set_new_dnode(&dn, inode, NULL, npage, nid);
1202         err = truncate_node(&dn);
1203         if (err) {
1204                 f2fs_put_page(npage, 1);
1205                 return err;
1206         }
1207
1208         f2fs_i_xnid_write(inode, 0);
1209
1210         return 0;
1211 }
1212
1213 /*
1214  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1215  * f2fs_unlock_op().
1216  */
1217 int f2fs_remove_inode_page(struct inode *inode)
1218 {
1219         struct dnode_of_data dn;
1220         int err;
1221
1222         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1223         err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1224         if (err)
1225                 return err;
1226
1227         err = f2fs_truncate_xattr_node(inode);
1228         if (err) {
1229                 f2fs_put_dnode(&dn);
1230                 return err;
1231         }
1232
1233         /* remove potential inline_data blocks */
1234         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1235                                 S_ISLNK(inode->i_mode))
1236                 f2fs_truncate_data_blocks_range(&dn, 1);
1237
1238         /* 0 is possible, after f2fs_new_inode() has failed */
1239         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1240                 f2fs_put_dnode(&dn);
1241                 return -EIO;
1242         }
1243
1244         if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1245                 f2fs_warn(F2FS_I_SB(inode),
1246                         "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1247                         inode->i_ino, (unsigned long long)inode->i_blocks);
1248                 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1249         }
1250
1251         /* will put inode & node pages */
1252         err = truncate_node(&dn);
1253         if (err) {
1254                 f2fs_put_dnode(&dn);
1255                 return err;
1256         }
1257         return 0;
1258 }
1259
1260 struct page *f2fs_new_inode_page(struct inode *inode)
1261 {
1262         struct dnode_of_data dn;
1263
1264         /* allocate inode page for new inode */
1265         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1266
1267         /* caller should f2fs_put_page(page, 1); */
1268         return f2fs_new_node_page(&dn, 0);
1269 }
1270
1271 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1272 {
1273         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1274         struct node_info new_ni;
1275         struct page *page;
1276         int err;
1277
1278         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1279                 return ERR_PTR(-EPERM);
1280
1281         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1282         if (!page)
1283                 return ERR_PTR(-ENOMEM);
1284
1285         if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1286                 goto fail;
1287
1288 #ifdef CONFIG_F2FS_CHECK_FS
1289         err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1290         if (err) {
1291                 dec_valid_node_count(sbi, dn->inode, !ofs);
1292                 goto fail;
1293         }
1294         f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1295 #endif
1296         new_ni.nid = dn->nid;
1297         new_ni.ino = dn->inode->i_ino;
1298         new_ni.blk_addr = NULL_ADDR;
1299         new_ni.flag = 0;
1300         new_ni.version = 0;
1301         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1302
1303         f2fs_wait_on_page_writeback(page, NODE, true, true);
1304         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1305         set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1306         if (!PageUptodate(page))
1307                 SetPageUptodate(page);
1308         if (set_page_dirty(page))
1309                 dn->node_changed = true;
1310
1311         if (f2fs_has_xattr_block(ofs))
1312                 f2fs_i_xnid_write(dn->inode, dn->nid);
1313
1314         if (ofs == 0)
1315                 inc_valid_inode_count(sbi);
1316         return page;
1317
1318 fail:
1319         clear_node_page_dirty(page);
1320         f2fs_put_page(page, 1);
1321         return ERR_PTR(err);
1322 }
1323
1324 /*
1325  * Caller should do after getting the following values.
1326  * 0: f2fs_put_page(page, 0)
1327  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1328  */
1329 static int read_node_page(struct page *page, int op_flags)
1330 {
1331         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1332         struct node_info ni;
1333         struct f2fs_io_info fio = {
1334                 .sbi = sbi,
1335                 .type = NODE,
1336                 .op = REQ_OP_READ,
1337                 .op_flags = op_flags,
1338                 .page = page,
1339                 .encrypted_page = NULL,
1340         };
1341         int err;
1342
1343         if (PageUptodate(page)) {
1344                 if (!f2fs_inode_chksum_verify(sbi, page)) {
1345                         ClearPageUptodate(page);
1346                         return -EFSBADCRC;
1347                 }
1348                 return LOCKED_PAGE;
1349         }
1350
1351         err = f2fs_get_node_info(sbi, page->index, &ni);
1352         if (err)
1353                 return err;
1354
1355         /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1356         if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR) ||
1357                         is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1358                 ClearPageUptodate(page);
1359                 return -ENOENT;
1360         }
1361
1362         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1363
1364         err = f2fs_submit_page_bio(&fio);
1365
1366         if (!err)
1367                 f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1368
1369         return err;
1370 }
1371
1372 /*
1373  * Readahead a node page
1374  */
1375 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1376 {
1377         struct page *apage;
1378         int err;
1379
1380         if (!nid)
1381                 return;
1382         if (f2fs_check_nid_range(sbi, nid))
1383                 return;
1384
1385         apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1386         if (apage)
1387                 return;
1388
1389         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1390         if (!apage)
1391                 return;
1392
1393         err = read_node_page(apage, REQ_RAHEAD);
1394         f2fs_put_page(apage, err ? 1 : 0);
1395 }
1396
1397 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1398                                         struct page *parent, int start)
1399 {
1400         struct page *page;
1401         int err;
1402
1403         if (!nid)
1404                 return ERR_PTR(-ENOENT);
1405         if (f2fs_check_nid_range(sbi, nid))
1406                 return ERR_PTR(-EINVAL);
1407 repeat:
1408         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1409         if (!page)
1410                 return ERR_PTR(-ENOMEM);
1411
1412         err = read_node_page(page, 0);
1413         if (err < 0) {
1414                 f2fs_put_page(page, 1);
1415                 return ERR_PTR(err);
1416         } else if (err == LOCKED_PAGE) {
1417                 err = 0;
1418                 goto page_hit;
1419         }
1420
1421         if (parent)
1422                 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1423
1424         lock_page(page);
1425
1426         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1427                 f2fs_put_page(page, 1);
1428                 goto repeat;
1429         }
1430
1431         if (unlikely(!PageUptodate(page))) {
1432                 err = -EIO;
1433                 goto out_err;
1434         }
1435
1436         if (!f2fs_inode_chksum_verify(sbi, page)) {
1437                 err = -EFSBADCRC;
1438                 goto out_err;
1439         }
1440 page_hit:
1441         if (unlikely(nid != nid_of_node(page))) {
1442                 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1443                           nid, nid_of_node(page), ino_of_node(page),
1444                           ofs_of_node(page), cpver_of_node(page),
1445                           next_blkaddr_of_node(page));
1446                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1447                 err = -EINVAL;
1448 out_err:
1449                 ClearPageUptodate(page);
1450                 f2fs_put_page(page, 1);
1451                 return ERR_PTR(err);
1452         }
1453         return page;
1454 }
1455
1456 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1457 {
1458         return __get_node_page(sbi, nid, NULL, 0);
1459 }
1460
1461 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1462 {
1463         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1464         nid_t nid = get_nid(parent, start, false);
1465
1466         return __get_node_page(sbi, nid, parent, start);
1467 }
1468
1469 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1470 {
1471         struct inode *inode;
1472         struct page *page;
1473         int ret;
1474
1475         /* should flush inline_data before evict_inode */
1476         inode = ilookup(sbi->sb, ino);
1477         if (!inode)
1478                 return;
1479
1480         page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1481                                         FGP_LOCK|FGP_NOWAIT, 0);
1482         if (!page)
1483                 goto iput_out;
1484
1485         if (!PageUptodate(page))
1486                 goto page_out;
1487
1488         if (!PageDirty(page))
1489                 goto page_out;
1490
1491         if (!clear_page_dirty_for_io(page))
1492                 goto page_out;
1493
1494         ret = f2fs_write_inline_data(inode, page);
1495         inode_dec_dirty_pages(inode);
1496         f2fs_remove_dirty_inode(inode);
1497         if (ret)
1498                 set_page_dirty(page);
1499 page_out:
1500         f2fs_put_page(page, 1);
1501 iput_out:
1502         iput(inode);
1503 }
1504
1505 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1506 {
1507         pgoff_t index;
1508         struct pagevec pvec;
1509         struct page *last_page = NULL;
1510         int nr_pages;
1511
1512         pagevec_init(&pvec);
1513         index = 0;
1514
1515         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1516                                 PAGECACHE_TAG_DIRTY))) {
1517                 int i;
1518
1519                 for (i = 0; i < nr_pages; i++) {
1520                         struct page *page = pvec.pages[i];
1521
1522                         if (unlikely(f2fs_cp_error(sbi))) {
1523                                 f2fs_put_page(last_page, 0);
1524                                 pagevec_release(&pvec);
1525                                 return ERR_PTR(-EIO);
1526                         }
1527
1528                         if (!IS_DNODE(page) || !is_cold_node(page))
1529                                 continue;
1530                         if (ino_of_node(page) != ino)
1531                                 continue;
1532
1533                         lock_page(page);
1534
1535                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1536 continue_unlock:
1537                                 unlock_page(page);
1538                                 continue;
1539                         }
1540                         if (ino_of_node(page) != ino)
1541                                 goto continue_unlock;
1542
1543                         if (!PageDirty(page)) {
1544                                 /* someone wrote it for us */
1545                                 goto continue_unlock;
1546                         }
1547
1548                         if (last_page)
1549                                 f2fs_put_page(last_page, 0);
1550
1551                         get_page(page);
1552                         last_page = page;
1553                         unlock_page(page);
1554                 }
1555                 pagevec_release(&pvec);
1556                 cond_resched();
1557         }
1558         return last_page;
1559 }
1560
1561 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1562                                 struct writeback_control *wbc, bool do_balance,
1563                                 enum iostat_type io_type, unsigned int *seq_id)
1564 {
1565         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1566         nid_t nid;
1567         struct node_info ni;
1568         struct f2fs_io_info fio = {
1569                 .sbi = sbi,
1570                 .ino = ino_of_node(page),
1571                 .type = NODE,
1572                 .op = REQ_OP_WRITE,
1573                 .op_flags = wbc_to_write_flags(wbc),
1574                 .page = page,
1575                 .encrypted_page = NULL,
1576                 .submitted = false,
1577                 .io_type = io_type,
1578                 .io_wbc = wbc,
1579         };
1580         unsigned int seq;
1581
1582         trace_f2fs_writepage(page, NODE);
1583
1584         if (unlikely(f2fs_cp_error(sbi))) {
1585                 ClearPageUptodate(page);
1586                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1587                 unlock_page(page);
1588                 return 0;
1589         }
1590
1591         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1592                 goto redirty_out;
1593
1594         if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1595                         wbc->sync_mode == WB_SYNC_NONE &&
1596                         IS_DNODE(page) && is_cold_node(page))
1597                 goto redirty_out;
1598
1599         /* get old block addr of this node page */
1600         nid = nid_of_node(page);
1601         f2fs_bug_on(sbi, page->index != nid);
1602
1603         if (f2fs_get_node_info(sbi, nid, &ni))
1604                 goto redirty_out;
1605
1606         if (wbc->for_reclaim) {
1607                 if (!down_read_trylock(&sbi->node_write))
1608                         goto redirty_out;
1609         } else {
1610                 down_read(&sbi->node_write);
1611         }
1612
1613         /* This page is already truncated */
1614         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1615                 ClearPageUptodate(page);
1616                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1617                 up_read(&sbi->node_write);
1618                 unlock_page(page);
1619                 return 0;
1620         }
1621
1622         if (__is_valid_data_blkaddr(ni.blk_addr) &&
1623                 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1624                                         DATA_GENERIC_ENHANCE)) {
1625                 up_read(&sbi->node_write);
1626                 goto redirty_out;
1627         }
1628
1629         if (atomic && !test_opt(sbi, NOBARRIER))
1630                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1631
1632         /* should add to global list before clearing PAGECACHE status */
1633         if (f2fs_in_warm_node_list(sbi, page)) {
1634                 seq = f2fs_add_fsync_node_entry(sbi, page);
1635                 if (seq_id)
1636                         *seq_id = seq;
1637         }
1638
1639         set_page_writeback(page);
1640         ClearPageError(page);
1641
1642         fio.old_blkaddr = ni.blk_addr;
1643         f2fs_do_write_node_page(nid, &fio);
1644         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1645         dec_page_count(sbi, F2FS_DIRTY_NODES);
1646         up_read(&sbi->node_write);
1647
1648         if (wbc->for_reclaim) {
1649                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1650                 submitted = NULL;
1651         }
1652
1653         unlock_page(page);
1654
1655         if (unlikely(f2fs_cp_error(sbi))) {
1656                 f2fs_submit_merged_write(sbi, NODE);
1657                 submitted = NULL;
1658         }
1659         if (submitted)
1660                 *submitted = fio.submitted;
1661
1662         if (do_balance)
1663                 f2fs_balance_fs(sbi, false);
1664         return 0;
1665
1666 redirty_out:
1667         redirty_page_for_writepage(wbc, page);
1668         return AOP_WRITEPAGE_ACTIVATE;
1669 }
1670
1671 int f2fs_move_node_page(struct page *node_page, int gc_type)
1672 {
1673         int err = 0;
1674
1675         if (gc_type == FG_GC) {
1676                 struct writeback_control wbc = {
1677                         .sync_mode = WB_SYNC_ALL,
1678                         .nr_to_write = 1,
1679                         .for_reclaim = 0,
1680                 };
1681
1682                 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1683
1684                 set_page_dirty(node_page);
1685
1686                 if (!clear_page_dirty_for_io(node_page)) {
1687                         err = -EAGAIN;
1688                         goto out_page;
1689                 }
1690
1691                 if (__write_node_page(node_page, false, NULL,
1692                                         &wbc, false, FS_GC_NODE_IO, NULL)) {
1693                         err = -EAGAIN;
1694                         unlock_page(node_page);
1695                 }
1696                 goto release_page;
1697         } else {
1698                 /* set page dirty and write it */
1699                 if (!PageWriteback(node_page))
1700                         set_page_dirty(node_page);
1701         }
1702 out_page:
1703         unlock_page(node_page);
1704 release_page:
1705         f2fs_put_page(node_page, 0);
1706         return err;
1707 }
1708
1709 static int f2fs_write_node_page(struct page *page,
1710                                 struct writeback_control *wbc)
1711 {
1712         return __write_node_page(page, false, NULL, wbc, false,
1713                                                 FS_NODE_IO, NULL);
1714 }
1715
1716 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1717                         struct writeback_control *wbc, bool atomic,
1718                         unsigned int *seq_id)
1719 {
1720         pgoff_t index;
1721         struct pagevec pvec;
1722         int ret = 0;
1723         struct page *last_page = NULL;
1724         bool marked = false;
1725         nid_t ino = inode->i_ino;
1726         int nr_pages;
1727         int nwritten = 0;
1728
1729         if (atomic) {
1730                 last_page = last_fsync_dnode(sbi, ino);
1731                 if (IS_ERR_OR_NULL(last_page))
1732                         return PTR_ERR_OR_ZERO(last_page);
1733         }
1734 retry:
1735         pagevec_init(&pvec);
1736         index = 0;
1737
1738         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1739                                 PAGECACHE_TAG_DIRTY))) {
1740                 int i;
1741
1742                 for (i = 0; i < nr_pages; i++) {
1743                         struct page *page = pvec.pages[i];
1744                         bool submitted = false;
1745
1746                         if (unlikely(f2fs_cp_error(sbi))) {
1747                                 f2fs_put_page(last_page, 0);
1748                                 pagevec_release(&pvec);
1749                                 ret = -EIO;
1750                                 goto out;
1751                         }
1752
1753                         if (!IS_DNODE(page) || !is_cold_node(page))
1754                                 continue;
1755                         if (ino_of_node(page) != ino)
1756                                 continue;
1757
1758                         lock_page(page);
1759
1760                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1761 continue_unlock:
1762                                 unlock_page(page);
1763                                 continue;
1764                         }
1765                         if (ino_of_node(page) != ino)
1766                                 goto continue_unlock;
1767
1768                         if (!PageDirty(page) && page != last_page) {
1769                                 /* someone wrote it for us */
1770                                 goto continue_unlock;
1771                         }
1772
1773                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1774
1775                         set_fsync_mark(page, 0);
1776                         set_dentry_mark(page, 0);
1777
1778                         if (!atomic || page == last_page) {
1779                                 set_fsync_mark(page, 1);
1780                                 if (IS_INODE(page)) {
1781                                         if (is_inode_flag_set(inode,
1782                                                                 FI_DIRTY_INODE))
1783                                                 f2fs_update_inode(inode, page);
1784                                         set_dentry_mark(page,
1785                                                 f2fs_need_dentry_mark(sbi, ino));
1786                                 }
1787                                 /* may be written by other thread */
1788                                 if (!PageDirty(page))
1789                                         set_page_dirty(page);
1790                         }
1791
1792                         if (!clear_page_dirty_for_io(page))
1793                                 goto continue_unlock;
1794
1795                         ret = __write_node_page(page, atomic &&
1796                                                 page == last_page,
1797                                                 &submitted, wbc, true,
1798                                                 FS_NODE_IO, seq_id);
1799                         if (ret) {
1800                                 unlock_page(page);
1801                                 f2fs_put_page(last_page, 0);
1802                                 break;
1803                         } else if (submitted) {
1804                                 nwritten++;
1805                         }
1806
1807                         if (page == last_page) {
1808                                 f2fs_put_page(page, 0);
1809                                 marked = true;
1810                                 break;
1811                         }
1812                 }
1813                 pagevec_release(&pvec);
1814                 cond_resched();
1815
1816                 if (ret || marked)
1817                         break;
1818         }
1819         if (!ret && atomic && !marked) {
1820                 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1821                            ino, last_page->index);
1822                 lock_page(last_page);
1823                 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1824                 set_page_dirty(last_page);
1825                 unlock_page(last_page);
1826                 goto retry;
1827         }
1828 out:
1829         if (nwritten)
1830                 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1831         return ret ? -EIO : 0;
1832 }
1833
1834 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1835 {
1836         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1837         bool clean;
1838
1839         if (inode->i_ino != ino)
1840                 return 0;
1841
1842         if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1843                 return 0;
1844
1845         spin_lock(&sbi->inode_lock[DIRTY_META]);
1846         clean = list_empty(&F2FS_I(inode)->gdirty_list);
1847         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1848
1849         if (clean)
1850                 return 0;
1851
1852         inode = igrab(inode);
1853         if (!inode)
1854                 return 0;
1855         return 1;
1856 }
1857
1858 static bool flush_dirty_inode(struct page *page)
1859 {
1860         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1861         struct inode *inode;
1862         nid_t ino = ino_of_node(page);
1863
1864         inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1865         if (!inode)
1866                 return false;
1867
1868         f2fs_update_inode(inode, page);
1869         unlock_page(page);
1870
1871         iput(inode);
1872         return true;
1873 }
1874
1875 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1876 {
1877         pgoff_t index = 0;
1878         struct pagevec pvec;
1879         int nr_pages;
1880
1881         pagevec_init(&pvec);
1882
1883         while ((nr_pages = pagevec_lookup_tag(&pvec,
1884                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1885                 int i;
1886
1887                 for (i = 0; i < nr_pages; i++) {
1888                         struct page *page = pvec.pages[i];
1889
1890                         if (!IS_DNODE(page))
1891                                 continue;
1892
1893                         lock_page(page);
1894
1895                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1896 continue_unlock:
1897                                 unlock_page(page);
1898                                 continue;
1899                         }
1900
1901                         if (!PageDirty(page)) {
1902                                 /* someone wrote it for us */
1903                                 goto continue_unlock;
1904                         }
1905
1906                         /* flush inline_data, if it's async context. */
1907                         if (page_private_inline(page)) {
1908                                 clear_page_private_inline(page);
1909                                 unlock_page(page);
1910                                 flush_inline_data(sbi, ino_of_node(page));
1911                                 continue;
1912                         }
1913                         unlock_page(page);
1914                 }
1915                 pagevec_release(&pvec);
1916                 cond_resched();
1917         }
1918 }
1919
1920 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1921                                 struct writeback_control *wbc,
1922                                 bool do_balance, enum iostat_type io_type)
1923 {
1924         pgoff_t index;
1925         struct pagevec pvec;
1926         int step = 0;
1927         int nwritten = 0;
1928         int ret = 0;
1929         int nr_pages, done = 0;
1930
1931         pagevec_init(&pvec);
1932
1933 next_step:
1934         index = 0;
1935
1936         while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1937                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1938                 int i;
1939
1940                 for (i = 0; i < nr_pages; i++) {
1941                         struct page *page = pvec.pages[i];
1942                         bool submitted = false;
1943                         bool may_dirty = true;
1944
1945                         /* give a priority to WB_SYNC threads */
1946                         if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1947                                         wbc->sync_mode == WB_SYNC_NONE) {
1948                                 done = 1;
1949                                 break;
1950                         }
1951
1952                         /*
1953                          * flushing sequence with step:
1954                          * 0. indirect nodes
1955                          * 1. dentry dnodes
1956                          * 2. file dnodes
1957                          */
1958                         if (step == 0 && IS_DNODE(page))
1959                                 continue;
1960                         if (step == 1 && (!IS_DNODE(page) ||
1961                                                 is_cold_node(page)))
1962                                 continue;
1963                         if (step == 2 && (!IS_DNODE(page) ||
1964                                                 !is_cold_node(page)))
1965                                 continue;
1966 lock_node:
1967                         if (wbc->sync_mode == WB_SYNC_ALL)
1968                                 lock_page(page);
1969                         else if (!trylock_page(page))
1970                                 continue;
1971
1972                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1973 continue_unlock:
1974                                 unlock_page(page);
1975                                 continue;
1976                         }
1977
1978                         if (!PageDirty(page)) {
1979                                 /* someone wrote it for us */
1980                                 goto continue_unlock;
1981                         }
1982
1983                         /* flush inline_data/inode, if it's async context. */
1984                         if (!do_balance)
1985                                 goto write_node;
1986
1987                         /* flush inline_data */
1988                         if (page_private_inline(page)) {
1989                                 clear_page_private_inline(page);
1990                                 unlock_page(page);
1991                                 flush_inline_data(sbi, ino_of_node(page));
1992                                 goto lock_node;
1993                         }
1994
1995                         /* flush dirty inode */
1996                         if (IS_INODE(page) && may_dirty) {
1997                                 may_dirty = false;
1998                                 if (flush_dirty_inode(page))
1999                                         goto lock_node;
2000                         }
2001 write_node:
2002                         f2fs_wait_on_page_writeback(page, NODE, true, true);
2003
2004                         if (!clear_page_dirty_for_io(page))
2005                                 goto continue_unlock;
2006
2007                         set_fsync_mark(page, 0);
2008                         set_dentry_mark(page, 0);
2009
2010                         ret = __write_node_page(page, false, &submitted,
2011                                                 wbc, do_balance, io_type, NULL);
2012                         if (ret)
2013                                 unlock_page(page);
2014                         else if (submitted)
2015                                 nwritten++;
2016
2017                         if (--wbc->nr_to_write == 0)
2018                                 break;
2019                 }
2020                 pagevec_release(&pvec);
2021                 cond_resched();
2022
2023                 if (wbc->nr_to_write == 0) {
2024                         step = 2;
2025                         break;
2026                 }
2027         }
2028
2029         if (step < 2) {
2030                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2031                                 wbc->sync_mode == WB_SYNC_NONE && step == 1)
2032                         goto out;
2033                 step++;
2034                 goto next_step;
2035         }
2036 out:
2037         if (nwritten)
2038                 f2fs_submit_merged_write(sbi, NODE);
2039
2040         if (unlikely(f2fs_cp_error(sbi)))
2041                 return -EIO;
2042         return ret;
2043 }
2044
2045 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2046                                                 unsigned int seq_id)
2047 {
2048         struct fsync_node_entry *fn;
2049         struct page *page;
2050         struct list_head *head = &sbi->fsync_node_list;
2051         unsigned long flags;
2052         unsigned int cur_seq_id = 0;
2053         int ret2, ret = 0;
2054
2055         while (seq_id && cur_seq_id < seq_id) {
2056                 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2057                 if (list_empty(head)) {
2058                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2059                         break;
2060                 }
2061                 fn = list_first_entry(head, struct fsync_node_entry, list);
2062                 if (fn->seq_id > seq_id) {
2063                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2064                         break;
2065                 }
2066                 cur_seq_id = fn->seq_id;
2067                 page = fn->page;
2068                 get_page(page);
2069                 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2070
2071                 f2fs_wait_on_page_writeback(page, NODE, true, false);
2072                 if (TestClearPageError(page))
2073                         ret = -EIO;
2074
2075                 put_page(page);
2076
2077                 if (ret)
2078                         break;
2079         }
2080
2081         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2082         if (!ret)
2083                 ret = ret2;
2084
2085         return ret;
2086 }
2087
2088 static int f2fs_write_node_pages(struct address_space *mapping,
2089                             struct writeback_control *wbc)
2090 {
2091         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2092         struct blk_plug plug;
2093         long diff;
2094
2095         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2096                 goto skip_write;
2097
2098         /* balancing f2fs's metadata in background */
2099         f2fs_balance_fs_bg(sbi, true);
2100
2101         /* collect a number of dirty node pages and write together */
2102         if (wbc->sync_mode != WB_SYNC_ALL &&
2103                         get_pages(sbi, F2FS_DIRTY_NODES) <
2104                                         nr_pages_to_skip(sbi, NODE))
2105                 goto skip_write;
2106
2107         if (wbc->sync_mode == WB_SYNC_ALL)
2108                 atomic_inc(&sbi->wb_sync_req[NODE]);
2109         else if (atomic_read(&sbi->wb_sync_req[NODE]))
2110                 goto skip_write;
2111
2112         trace_f2fs_writepages(mapping->host, wbc, NODE);
2113
2114         diff = nr_pages_to_write(sbi, NODE, wbc);
2115         blk_start_plug(&plug);
2116         f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2117         blk_finish_plug(&plug);
2118         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2119
2120         if (wbc->sync_mode == WB_SYNC_ALL)
2121                 atomic_dec(&sbi->wb_sync_req[NODE]);
2122         return 0;
2123
2124 skip_write:
2125         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2126         trace_f2fs_writepages(mapping->host, wbc, NODE);
2127         return 0;
2128 }
2129
2130 static int f2fs_set_node_page_dirty(struct page *page)
2131 {
2132         trace_f2fs_set_page_dirty(page, NODE);
2133
2134         if (!PageUptodate(page))
2135                 SetPageUptodate(page);
2136 #ifdef CONFIG_F2FS_CHECK_FS
2137         if (IS_INODE(page))
2138                 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2139 #endif
2140         if (!PageDirty(page)) {
2141                 __set_page_dirty_nobuffers(page);
2142                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2143                 set_page_private_reference(page);
2144                 return 1;
2145         }
2146         return 0;
2147 }
2148
2149 /*
2150  * Structure of the f2fs node operations
2151  */
2152 const struct address_space_operations f2fs_node_aops = {
2153         .writepage      = f2fs_write_node_page,
2154         .writepages     = f2fs_write_node_pages,
2155         .set_page_dirty = f2fs_set_node_page_dirty,
2156         .invalidatepage = f2fs_invalidate_page,
2157         .releasepage    = f2fs_release_page,
2158 #ifdef CONFIG_MIGRATION
2159         .migratepage    = f2fs_migrate_page,
2160 #endif
2161 };
2162
2163 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2164                                                 nid_t n)
2165 {
2166         return radix_tree_lookup(&nm_i->free_nid_root, n);
2167 }
2168
2169 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2170                                 struct free_nid *i)
2171 {
2172         struct f2fs_nm_info *nm_i = NM_I(sbi);
2173         int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2174
2175         if (err)
2176                 return err;
2177
2178         nm_i->nid_cnt[FREE_NID]++;
2179         list_add_tail(&i->list, &nm_i->free_nid_list);
2180         return 0;
2181 }
2182
2183 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2184                         struct free_nid *i, enum nid_state state)
2185 {
2186         struct f2fs_nm_info *nm_i = NM_I(sbi);
2187
2188         f2fs_bug_on(sbi, state != i->state);
2189         nm_i->nid_cnt[state]--;
2190         if (state == FREE_NID)
2191                 list_del(&i->list);
2192         radix_tree_delete(&nm_i->free_nid_root, i->nid);
2193 }
2194
2195 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2196                         enum nid_state org_state, enum nid_state dst_state)
2197 {
2198         struct f2fs_nm_info *nm_i = NM_I(sbi);
2199
2200         f2fs_bug_on(sbi, org_state != i->state);
2201         i->state = dst_state;
2202         nm_i->nid_cnt[org_state]--;
2203         nm_i->nid_cnt[dst_state]++;
2204
2205         switch (dst_state) {
2206         case PREALLOC_NID:
2207                 list_del(&i->list);
2208                 break;
2209         case FREE_NID:
2210                 list_add_tail(&i->list, &nm_i->free_nid_list);
2211                 break;
2212         default:
2213                 BUG_ON(1);
2214         }
2215 }
2216
2217 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2218 {
2219         struct f2fs_nm_info *nm_i = NM_I(sbi);
2220         unsigned int i;
2221         bool ret = true;
2222
2223         down_read(&nm_i->nat_tree_lock);
2224         for (i = 0; i < nm_i->nat_blocks; i++) {
2225                 if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2226                         ret = false;
2227                         break;
2228                 }
2229         }
2230         up_read(&nm_i->nat_tree_lock);
2231
2232         return ret;
2233 }
2234
2235 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2236                                                         bool set, bool build)
2237 {
2238         struct f2fs_nm_info *nm_i = NM_I(sbi);
2239         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2240         unsigned int nid_ofs = nid - START_NID(nid);
2241
2242         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2243                 return;
2244
2245         if (set) {
2246                 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2247                         return;
2248                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2249                 nm_i->free_nid_count[nat_ofs]++;
2250         } else {
2251                 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2252                         return;
2253                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2254                 if (!build)
2255                         nm_i->free_nid_count[nat_ofs]--;
2256         }
2257 }
2258
2259 /* return if the nid is recognized as free */
2260 static bool add_free_nid(struct f2fs_sb_info *sbi,
2261                                 nid_t nid, bool build, bool update)
2262 {
2263         struct f2fs_nm_info *nm_i = NM_I(sbi);
2264         struct free_nid *i, *e;
2265         struct nat_entry *ne;
2266         int err = -EINVAL;
2267         bool ret = false;
2268
2269         /* 0 nid should not be used */
2270         if (unlikely(nid == 0))
2271                 return false;
2272
2273         if (unlikely(f2fs_check_nid_range(sbi, nid)))
2274                 return false;
2275
2276         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2277         i->nid = nid;
2278         i->state = FREE_NID;
2279
2280         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2281
2282         spin_lock(&nm_i->nid_list_lock);
2283
2284         if (build) {
2285                 /*
2286                  *   Thread A             Thread B
2287                  *  - f2fs_create
2288                  *   - f2fs_new_inode
2289                  *    - f2fs_alloc_nid
2290                  *     - __insert_nid_to_list(PREALLOC_NID)
2291                  *                     - f2fs_balance_fs_bg
2292                  *                      - f2fs_build_free_nids
2293                  *                       - __f2fs_build_free_nids
2294                  *                        - scan_nat_page
2295                  *                         - add_free_nid
2296                  *                          - __lookup_nat_cache
2297                  *  - f2fs_add_link
2298                  *   - f2fs_init_inode_metadata
2299                  *    - f2fs_new_inode_page
2300                  *     - f2fs_new_node_page
2301                  *      - set_node_addr
2302                  *  - f2fs_alloc_nid_done
2303                  *   - __remove_nid_from_list(PREALLOC_NID)
2304                  *                         - __insert_nid_to_list(FREE_NID)
2305                  */
2306                 ne = __lookup_nat_cache(nm_i, nid);
2307                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2308                                 nat_get_blkaddr(ne) != NULL_ADDR))
2309                         goto err_out;
2310
2311                 e = __lookup_free_nid_list(nm_i, nid);
2312                 if (e) {
2313                         if (e->state == FREE_NID)
2314                                 ret = true;
2315                         goto err_out;
2316                 }
2317         }
2318         ret = true;
2319         err = __insert_free_nid(sbi, i);
2320 err_out:
2321         if (update) {
2322                 update_free_nid_bitmap(sbi, nid, ret, build);
2323                 if (!build)
2324                         nm_i->available_nids++;
2325         }
2326         spin_unlock(&nm_i->nid_list_lock);
2327         radix_tree_preload_end();
2328
2329         if (err)
2330                 kmem_cache_free(free_nid_slab, i);
2331         return ret;
2332 }
2333
2334 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2335 {
2336         struct f2fs_nm_info *nm_i = NM_I(sbi);
2337         struct free_nid *i;
2338         bool need_free = false;
2339
2340         spin_lock(&nm_i->nid_list_lock);
2341         i = __lookup_free_nid_list(nm_i, nid);
2342         if (i && i->state == FREE_NID) {
2343                 __remove_free_nid(sbi, i, FREE_NID);
2344                 need_free = true;
2345         }
2346         spin_unlock(&nm_i->nid_list_lock);
2347
2348         if (need_free)
2349                 kmem_cache_free(free_nid_slab, i);
2350 }
2351
2352 static int scan_nat_page(struct f2fs_sb_info *sbi,
2353                         struct page *nat_page, nid_t start_nid)
2354 {
2355         struct f2fs_nm_info *nm_i = NM_I(sbi);
2356         struct f2fs_nat_block *nat_blk = page_address(nat_page);
2357         block_t blk_addr;
2358         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2359         int i;
2360
2361         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2362
2363         i = start_nid % NAT_ENTRY_PER_BLOCK;
2364
2365         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2366                 if (unlikely(start_nid >= nm_i->max_nid))
2367                         break;
2368
2369                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2370
2371                 if (blk_addr == NEW_ADDR)
2372                         return -EINVAL;
2373
2374                 if (blk_addr == NULL_ADDR) {
2375                         add_free_nid(sbi, start_nid, true, true);
2376                 } else {
2377                         spin_lock(&NM_I(sbi)->nid_list_lock);
2378                         update_free_nid_bitmap(sbi, start_nid, false, true);
2379                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2380                 }
2381         }
2382
2383         return 0;
2384 }
2385
2386 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2387 {
2388         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2389         struct f2fs_journal *journal = curseg->journal;
2390         int i;
2391
2392         down_read(&curseg->journal_rwsem);
2393         for (i = 0; i < nats_in_cursum(journal); i++) {
2394                 block_t addr;
2395                 nid_t nid;
2396
2397                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2398                 nid = le32_to_cpu(nid_in_journal(journal, i));
2399                 if (addr == NULL_ADDR)
2400                         add_free_nid(sbi, nid, true, false);
2401                 else
2402                         remove_free_nid(sbi, nid);
2403         }
2404         up_read(&curseg->journal_rwsem);
2405 }
2406
2407 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2408 {
2409         struct f2fs_nm_info *nm_i = NM_I(sbi);
2410         unsigned int i, idx;
2411         nid_t nid;
2412
2413         down_read(&nm_i->nat_tree_lock);
2414
2415         for (i = 0; i < nm_i->nat_blocks; i++) {
2416                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2417                         continue;
2418                 if (!nm_i->free_nid_count[i])
2419                         continue;
2420                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2421                         idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2422                                                 NAT_ENTRY_PER_BLOCK, idx);
2423                         if (idx >= NAT_ENTRY_PER_BLOCK)
2424                                 break;
2425
2426                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
2427                         add_free_nid(sbi, nid, true, false);
2428
2429                         if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2430                                 goto out;
2431                 }
2432         }
2433 out:
2434         scan_curseg_cache(sbi);
2435
2436         up_read(&nm_i->nat_tree_lock);
2437 }
2438
2439 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2440                                                 bool sync, bool mount)
2441 {
2442         struct f2fs_nm_info *nm_i = NM_I(sbi);
2443         int i = 0, ret;
2444         nid_t nid = nm_i->next_scan_nid;
2445
2446         if (unlikely(nid >= nm_i->max_nid))
2447                 nid = 0;
2448
2449         if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2450                 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2451
2452         /* Enough entries */
2453         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2454                 return 0;
2455
2456         if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2457                 return 0;
2458
2459         if (!mount) {
2460                 /* try to find free nids in free_nid_bitmap */
2461                 scan_free_nid_bits(sbi);
2462
2463                 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2464                         return 0;
2465         }
2466
2467         /* readahead nat pages to be scanned */
2468         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2469                                                         META_NAT, true);
2470
2471         down_read(&nm_i->nat_tree_lock);
2472
2473         while (1) {
2474                 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2475                                                 nm_i->nat_block_bitmap)) {
2476                         struct page *page = get_current_nat_page(sbi, nid);
2477
2478                         if (IS_ERR(page)) {
2479                                 ret = PTR_ERR(page);
2480                         } else {
2481                                 ret = scan_nat_page(sbi, page, nid);
2482                                 f2fs_put_page(page, 1);
2483                         }
2484
2485                         if (ret) {
2486                                 up_read(&nm_i->nat_tree_lock);
2487                                 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2488                                 return ret;
2489                         }
2490                 }
2491
2492                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2493                 if (unlikely(nid >= nm_i->max_nid))
2494                         nid = 0;
2495
2496                 if (++i >= FREE_NID_PAGES)
2497                         break;
2498         }
2499
2500         /* go to the next free nat pages to find free nids abundantly */
2501         nm_i->next_scan_nid = nid;
2502
2503         /* find free nids from current sum_pages */
2504         scan_curseg_cache(sbi);
2505
2506         up_read(&nm_i->nat_tree_lock);
2507
2508         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2509                                         nm_i->ra_nid_pages, META_NAT, false);
2510
2511         return 0;
2512 }
2513
2514 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2515 {
2516         int ret;
2517
2518         mutex_lock(&NM_I(sbi)->build_lock);
2519         ret = __f2fs_build_free_nids(sbi, sync, mount);
2520         mutex_unlock(&NM_I(sbi)->build_lock);
2521
2522         return ret;
2523 }
2524
2525 /*
2526  * If this function returns success, caller can obtain a new nid
2527  * from second parameter of this function.
2528  * The returned nid could be used ino as well as nid when inode is created.
2529  */
2530 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2531 {
2532         struct f2fs_nm_info *nm_i = NM_I(sbi);
2533         struct free_nid *i = NULL;
2534 retry:
2535         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2536                 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2537                 return false;
2538         }
2539
2540         spin_lock(&nm_i->nid_list_lock);
2541
2542         if (unlikely(nm_i->available_nids == 0)) {
2543                 spin_unlock(&nm_i->nid_list_lock);
2544                 return false;
2545         }
2546
2547         /* We should not use stale free nids created by f2fs_build_free_nids */
2548         if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2549                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2550                 i = list_first_entry(&nm_i->free_nid_list,
2551                                         struct free_nid, list);
2552                 *nid = i->nid;
2553
2554                 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2555                 nm_i->available_nids--;
2556
2557                 update_free_nid_bitmap(sbi, *nid, false, false);
2558
2559                 spin_unlock(&nm_i->nid_list_lock);
2560                 return true;
2561         }
2562         spin_unlock(&nm_i->nid_list_lock);
2563
2564         /* Let's scan nat pages and its caches to get free nids */
2565         if (!f2fs_build_free_nids(sbi, true, false))
2566                 goto retry;
2567         return false;
2568 }
2569
2570 /*
2571  * f2fs_alloc_nid() should be called prior to this function.
2572  */
2573 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2574 {
2575         struct f2fs_nm_info *nm_i = NM_I(sbi);
2576         struct free_nid *i;
2577
2578         spin_lock(&nm_i->nid_list_lock);
2579         i = __lookup_free_nid_list(nm_i, nid);
2580         f2fs_bug_on(sbi, !i);
2581         __remove_free_nid(sbi, i, PREALLOC_NID);
2582         spin_unlock(&nm_i->nid_list_lock);
2583
2584         kmem_cache_free(free_nid_slab, i);
2585 }
2586
2587 /*
2588  * f2fs_alloc_nid() should be called prior to this function.
2589  */
2590 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2591 {
2592         struct f2fs_nm_info *nm_i = NM_I(sbi);
2593         struct free_nid *i;
2594         bool need_free = false;
2595
2596         if (!nid)
2597                 return;
2598
2599         spin_lock(&nm_i->nid_list_lock);
2600         i = __lookup_free_nid_list(nm_i, nid);
2601         f2fs_bug_on(sbi, !i);
2602
2603         if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2604                 __remove_free_nid(sbi, i, PREALLOC_NID);
2605                 need_free = true;
2606         } else {
2607                 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2608         }
2609
2610         nm_i->available_nids++;
2611
2612         update_free_nid_bitmap(sbi, nid, true, false);
2613
2614         spin_unlock(&nm_i->nid_list_lock);
2615
2616         if (need_free)
2617                 kmem_cache_free(free_nid_slab, i);
2618 }
2619
2620 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2621 {
2622         struct f2fs_nm_info *nm_i = NM_I(sbi);
2623         int nr = nr_shrink;
2624
2625         if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2626                 return 0;
2627
2628         if (!mutex_trylock(&nm_i->build_lock))
2629                 return 0;
2630
2631         while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2632                 struct free_nid *i, *next;
2633                 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2634
2635                 spin_lock(&nm_i->nid_list_lock);
2636                 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2637                         if (!nr_shrink || !batch ||
2638                                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2639                                 break;
2640                         __remove_free_nid(sbi, i, FREE_NID);
2641                         kmem_cache_free(free_nid_slab, i);
2642                         nr_shrink--;
2643                         batch--;
2644                 }
2645                 spin_unlock(&nm_i->nid_list_lock);
2646         }
2647
2648         mutex_unlock(&nm_i->build_lock);
2649
2650         return nr - nr_shrink;
2651 }
2652
2653 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2654 {
2655         void *src_addr, *dst_addr;
2656         size_t inline_size;
2657         struct page *ipage;
2658         struct f2fs_inode *ri;
2659
2660         ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2661         if (IS_ERR(ipage))
2662                 return PTR_ERR(ipage);
2663
2664         ri = F2FS_INODE(page);
2665         if (ri->i_inline & F2FS_INLINE_XATTR) {
2666                 if (!f2fs_has_inline_xattr(inode)) {
2667                         set_inode_flag(inode, FI_INLINE_XATTR);
2668                         stat_inc_inline_xattr(inode);
2669                 }
2670         } else {
2671                 if (f2fs_has_inline_xattr(inode)) {
2672                         stat_dec_inline_xattr(inode);
2673                         clear_inode_flag(inode, FI_INLINE_XATTR);
2674                 }
2675                 goto update_inode;
2676         }
2677
2678         dst_addr = inline_xattr_addr(inode, ipage);
2679         src_addr = inline_xattr_addr(inode, page);
2680         inline_size = inline_xattr_size(inode);
2681
2682         f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2683         memcpy(dst_addr, src_addr, inline_size);
2684 update_inode:
2685         f2fs_update_inode(inode, ipage);
2686         f2fs_put_page(ipage, 1);
2687         return 0;
2688 }
2689
2690 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2691 {
2692         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2693         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2694         nid_t new_xnid;
2695         struct dnode_of_data dn;
2696         struct node_info ni;
2697         struct page *xpage;
2698         int err;
2699
2700         if (!prev_xnid)
2701                 goto recover_xnid;
2702
2703         /* 1: invalidate the previous xattr nid */
2704         err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2705         if (err)
2706                 return err;
2707
2708         f2fs_invalidate_blocks(sbi, ni.blk_addr);
2709         dec_valid_node_count(sbi, inode, false);
2710         set_node_addr(sbi, &ni, NULL_ADDR, false);
2711
2712 recover_xnid:
2713         /* 2: update xattr nid in inode */
2714         if (!f2fs_alloc_nid(sbi, &new_xnid))
2715                 return -ENOSPC;
2716
2717         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2718         xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2719         if (IS_ERR(xpage)) {
2720                 f2fs_alloc_nid_failed(sbi, new_xnid);
2721                 return PTR_ERR(xpage);
2722         }
2723
2724         f2fs_alloc_nid_done(sbi, new_xnid);
2725         f2fs_update_inode_page(inode);
2726
2727         /* 3: update and set xattr node page dirty */
2728         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2729
2730         set_page_dirty(xpage);
2731         f2fs_put_page(xpage, 1);
2732
2733         return 0;
2734 }
2735
2736 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2737 {
2738         struct f2fs_inode *src, *dst;
2739         nid_t ino = ino_of_node(page);
2740         struct node_info old_ni, new_ni;
2741         struct page *ipage;
2742         int err;
2743
2744         err = f2fs_get_node_info(sbi, ino, &old_ni);
2745         if (err)
2746                 return err;
2747
2748         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2749                 return -EINVAL;
2750 retry:
2751         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2752         if (!ipage) {
2753                 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2754                 goto retry;
2755         }
2756
2757         /* Should not use this inode from free nid list */
2758         remove_free_nid(sbi, ino);
2759
2760         if (!PageUptodate(ipage))
2761                 SetPageUptodate(ipage);
2762         fill_node_footer(ipage, ino, ino, 0, true);
2763         set_cold_node(ipage, false);
2764
2765         src = F2FS_INODE(page);
2766         dst = F2FS_INODE(ipage);
2767
2768         memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2769         dst->i_size = 0;
2770         dst->i_blocks = cpu_to_le64(1);
2771         dst->i_links = cpu_to_le32(1);
2772         dst->i_xattr_nid = 0;
2773         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2774         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2775                 dst->i_extra_isize = src->i_extra_isize;
2776
2777                 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2778                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2779                                                         i_inline_xattr_size))
2780                         dst->i_inline_xattr_size = src->i_inline_xattr_size;
2781
2782                 if (f2fs_sb_has_project_quota(sbi) &&
2783                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2784                                                                 i_projid))
2785                         dst->i_projid = src->i_projid;
2786
2787                 if (f2fs_sb_has_inode_crtime(sbi) &&
2788                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2789                                                         i_crtime_nsec)) {
2790                         dst->i_crtime = src->i_crtime;
2791                         dst->i_crtime_nsec = src->i_crtime_nsec;
2792                 }
2793         }
2794
2795         new_ni = old_ni;
2796         new_ni.ino = ino;
2797
2798         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2799                 WARN_ON(1);
2800         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2801         inc_valid_inode_count(sbi);
2802         set_page_dirty(ipage);
2803         f2fs_put_page(ipage, 1);
2804         return 0;
2805 }
2806
2807 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2808                         unsigned int segno, struct f2fs_summary_block *sum)
2809 {
2810         struct f2fs_node *rn;
2811         struct f2fs_summary *sum_entry;
2812         block_t addr;
2813         int i, idx, last_offset, nrpages;
2814
2815         /* scan the node segment */
2816         last_offset = sbi->blocks_per_seg;
2817         addr = START_BLOCK(sbi, segno);
2818         sum_entry = &sum->entries[0];
2819
2820         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2821                 nrpages = bio_max_segs(last_offset - i);
2822
2823                 /* readahead node pages */
2824                 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2825
2826                 for (idx = addr; idx < addr + nrpages; idx++) {
2827                         struct page *page = f2fs_get_tmp_page(sbi, idx);
2828
2829                         if (IS_ERR(page))
2830                                 return PTR_ERR(page);
2831
2832                         rn = F2FS_NODE(page);
2833                         sum_entry->nid = rn->footer.nid;
2834                         sum_entry->version = 0;
2835                         sum_entry->ofs_in_node = 0;
2836                         sum_entry++;
2837                         f2fs_put_page(page, 1);
2838                 }
2839
2840                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2841                                                         addr + nrpages);
2842         }
2843         return 0;
2844 }
2845
2846 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2847 {
2848         struct f2fs_nm_info *nm_i = NM_I(sbi);
2849         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2850         struct f2fs_journal *journal = curseg->journal;
2851         int i;
2852
2853         down_write(&curseg->journal_rwsem);
2854         for (i = 0; i < nats_in_cursum(journal); i++) {
2855                 struct nat_entry *ne;
2856                 struct f2fs_nat_entry raw_ne;
2857                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2858
2859                 if (f2fs_check_nid_range(sbi, nid))
2860                         continue;
2861
2862                 raw_ne = nat_in_journal(journal, i);
2863
2864                 ne = __lookup_nat_cache(nm_i, nid);
2865                 if (!ne) {
2866                         ne = __alloc_nat_entry(sbi, nid, true);
2867                         __init_nat_entry(nm_i, ne, &raw_ne, true);
2868                 }
2869
2870                 /*
2871                  * if a free nat in journal has not been used after last
2872                  * checkpoint, we should remove it from available nids,
2873                  * since later we will add it again.
2874                  */
2875                 if (!get_nat_flag(ne, IS_DIRTY) &&
2876                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2877                         spin_lock(&nm_i->nid_list_lock);
2878                         nm_i->available_nids--;
2879                         spin_unlock(&nm_i->nid_list_lock);
2880                 }
2881
2882                 __set_nat_cache_dirty(nm_i, ne);
2883         }
2884         update_nats_in_cursum(journal, -i);
2885         up_write(&curseg->journal_rwsem);
2886 }
2887
2888 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2889                                                 struct list_head *head, int max)
2890 {
2891         struct nat_entry_set *cur;
2892
2893         if (nes->entry_cnt >= max)
2894                 goto add_out;
2895
2896         list_for_each_entry(cur, head, set_list) {
2897                 if (cur->entry_cnt >= nes->entry_cnt) {
2898                         list_add(&nes->set_list, cur->set_list.prev);
2899                         return;
2900                 }
2901         }
2902 add_out:
2903         list_add_tail(&nes->set_list, head);
2904 }
2905
2906 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2907                                                         unsigned int valid)
2908 {
2909         if (valid == 0) {
2910                 __set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2911                 __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2912                 return;
2913         }
2914
2915         __clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2916         if (valid == NAT_ENTRY_PER_BLOCK)
2917                 __set_bit_le(nat_ofs, nm_i->full_nat_bits);
2918         else
2919                 __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2920 }
2921
2922 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2923                                                 struct page *page)
2924 {
2925         struct f2fs_nm_info *nm_i = NM_I(sbi);
2926         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2927         struct f2fs_nat_block *nat_blk = page_address(page);
2928         int valid = 0;
2929         int i = 0;
2930
2931         if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2932                 return;
2933
2934         if (nat_index == 0) {
2935                 valid = 1;
2936                 i = 1;
2937         }
2938         for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2939                 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2940                         valid++;
2941         }
2942
2943         __update_nat_bits(nm_i, nat_index, valid);
2944 }
2945
2946 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2947 {
2948         struct f2fs_nm_info *nm_i = NM_I(sbi);
2949         unsigned int nat_ofs;
2950
2951         down_read(&nm_i->nat_tree_lock);
2952
2953         for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
2954                 unsigned int valid = 0, nid_ofs = 0;
2955
2956                 /* handle nid zero due to it should never be used */
2957                 if (unlikely(nat_ofs == 0)) {
2958                         valid = 1;
2959                         nid_ofs = 1;
2960                 }
2961
2962                 for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
2963                         if (!test_bit_le(nid_ofs,
2964                                         nm_i->free_nid_bitmap[nat_ofs]))
2965                                 valid++;
2966                 }
2967
2968                 __update_nat_bits(nm_i, nat_ofs, valid);
2969         }
2970
2971         up_read(&nm_i->nat_tree_lock);
2972 }
2973
2974 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2975                 struct nat_entry_set *set, struct cp_control *cpc)
2976 {
2977         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2978         struct f2fs_journal *journal = curseg->journal;
2979         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2980         bool to_journal = true;
2981         struct f2fs_nat_block *nat_blk;
2982         struct nat_entry *ne, *cur;
2983         struct page *page = NULL;
2984
2985         /*
2986          * there are two steps to flush nat entries:
2987          * #1, flush nat entries to journal in current hot data summary block.
2988          * #2, flush nat entries to nat page.
2989          */
2990         if ((cpc->reason & CP_UMOUNT) ||
2991                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2992                 to_journal = false;
2993
2994         if (to_journal) {
2995                 down_write(&curseg->journal_rwsem);
2996         } else {
2997                 page = get_next_nat_page(sbi, start_nid);
2998                 if (IS_ERR(page))
2999                         return PTR_ERR(page);
3000
3001                 nat_blk = page_address(page);
3002                 f2fs_bug_on(sbi, !nat_blk);
3003         }
3004
3005         /* flush dirty nats in nat entry set */
3006         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3007                 struct f2fs_nat_entry *raw_ne;
3008                 nid_t nid = nat_get_nid(ne);
3009                 int offset;
3010
3011                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3012
3013                 if (to_journal) {
3014                         offset = f2fs_lookup_journal_in_cursum(journal,
3015                                                         NAT_JOURNAL, nid, 1);
3016                         f2fs_bug_on(sbi, offset < 0);
3017                         raw_ne = &nat_in_journal(journal, offset);
3018                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
3019                 } else {
3020                         raw_ne = &nat_blk->entries[nid - start_nid];
3021                 }
3022                 raw_nat_from_node_info(raw_ne, &ne->ni);
3023                 nat_reset_flag(ne);
3024                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
3025                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
3026                         add_free_nid(sbi, nid, false, true);
3027                 } else {
3028                         spin_lock(&NM_I(sbi)->nid_list_lock);
3029                         update_free_nid_bitmap(sbi, nid, false, false);
3030                         spin_unlock(&NM_I(sbi)->nid_list_lock);
3031                 }
3032         }
3033
3034         if (to_journal) {
3035                 up_write(&curseg->journal_rwsem);
3036         } else {
3037                 update_nat_bits(sbi, start_nid, page);
3038                 f2fs_put_page(page, 1);
3039         }
3040
3041         /* Allow dirty nats by node block allocation in write_begin */
3042         if (!set->entry_cnt) {
3043                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3044                 kmem_cache_free(nat_entry_set_slab, set);
3045         }
3046         return 0;
3047 }
3048
3049 /*
3050  * This function is called during the checkpointing process.
3051  */
3052 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3053 {
3054         struct f2fs_nm_info *nm_i = NM_I(sbi);
3055         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3056         struct f2fs_journal *journal = curseg->journal;
3057         struct nat_entry_set *setvec[SETVEC_SIZE];
3058         struct nat_entry_set *set, *tmp;
3059         unsigned int found;
3060         nid_t set_idx = 0;
3061         LIST_HEAD(sets);
3062         int err = 0;
3063
3064         /*
3065          * during unmount, let's flush nat_bits before checking
3066          * nat_cnt[DIRTY_NAT].
3067          */
3068         if (cpc->reason & CP_UMOUNT) {
3069                 down_write(&nm_i->nat_tree_lock);
3070                 remove_nats_in_journal(sbi);
3071                 up_write(&nm_i->nat_tree_lock);
3072         }
3073
3074         if (!nm_i->nat_cnt[DIRTY_NAT])
3075                 return 0;
3076
3077         down_write(&nm_i->nat_tree_lock);
3078
3079         /*
3080          * if there are no enough space in journal to store dirty nat
3081          * entries, remove all entries from journal and merge them
3082          * into nat entry set.
3083          */
3084         if (cpc->reason & CP_UMOUNT ||
3085                 !__has_cursum_space(journal,
3086                         nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3087                 remove_nats_in_journal(sbi);
3088
3089         while ((found = __gang_lookup_nat_set(nm_i,
3090                                         set_idx, SETVEC_SIZE, setvec))) {
3091                 unsigned idx;
3092
3093                 set_idx = setvec[found - 1]->set + 1;
3094                 for (idx = 0; idx < found; idx++)
3095                         __adjust_nat_entry_set(setvec[idx], &sets,
3096                                                 MAX_NAT_JENTRIES(journal));
3097         }
3098
3099         /* flush dirty nats in nat entry set */
3100         list_for_each_entry_safe(set, tmp, &sets, set_list) {
3101                 err = __flush_nat_entry_set(sbi, set, cpc);
3102                 if (err)
3103                         break;
3104         }
3105
3106         up_write(&nm_i->nat_tree_lock);
3107         /* Allow dirty nats by node block allocation in write_begin */
3108
3109         return err;
3110 }
3111
3112 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3113 {
3114         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3115         struct f2fs_nm_info *nm_i = NM_I(sbi);
3116         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3117         unsigned int i;
3118         __u64 cp_ver = cur_cp_version(ckpt);
3119         block_t nat_bits_addr;
3120
3121         nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3122         nm_i->nat_bits = f2fs_kvzalloc(sbi,
3123                         nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3124         if (!nm_i->nat_bits)
3125                 return -ENOMEM;
3126
3127         nm_i->full_nat_bits = nm_i->nat_bits + 8;
3128         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3129
3130         if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3131                 return 0;
3132
3133         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3134                                                 nm_i->nat_bits_blocks;
3135         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3136                 struct page *page;
3137
3138                 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3139                 if (IS_ERR(page))
3140                         return PTR_ERR(page);
3141
3142                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3143                                         page_address(page), F2FS_BLKSIZE);
3144                 f2fs_put_page(page, 1);
3145         }
3146
3147         cp_ver |= (cur_cp_crc(ckpt) << 32);
3148         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3149                 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3150                 f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3151                         cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3152                 return 0;
3153         }
3154
3155         f2fs_notice(sbi, "Found nat_bits in checkpoint");
3156         return 0;
3157 }
3158
3159 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3160 {
3161         struct f2fs_nm_info *nm_i = NM_I(sbi);
3162         unsigned int i = 0;
3163         nid_t nid, last_nid;
3164
3165         if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3166                 return;
3167
3168         for (i = 0; i < nm_i->nat_blocks; i++) {
3169                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3170                 if (i >= nm_i->nat_blocks)
3171                         break;
3172
3173                 __set_bit_le(i, nm_i->nat_block_bitmap);
3174
3175                 nid = i * NAT_ENTRY_PER_BLOCK;
3176                 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3177
3178                 spin_lock(&NM_I(sbi)->nid_list_lock);
3179                 for (; nid < last_nid; nid++)
3180                         update_free_nid_bitmap(sbi, nid, true, true);
3181                 spin_unlock(&NM_I(sbi)->nid_list_lock);
3182         }
3183
3184         for (i = 0; i < nm_i->nat_blocks; i++) {
3185                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3186                 if (i >= nm_i->nat_blocks)
3187                         break;
3188
3189                 __set_bit_le(i, nm_i->nat_block_bitmap);
3190         }
3191 }
3192
3193 static int init_node_manager(struct f2fs_sb_info *sbi)
3194 {
3195         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3196         struct f2fs_nm_info *nm_i = NM_I(sbi);
3197         unsigned char *version_bitmap;
3198         unsigned int nat_segs;
3199         int err;
3200
3201         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3202
3203         /* segment_count_nat includes pair segment so divide to 2. */
3204         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3205         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3206         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3207
3208         /* not used nids: 0, node, meta, (and root counted as valid node) */
3209         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3210                                                 F2FS_RESERVED_NODE_NUM;
3211         nm_i->nid_cnt[FREE_NID] = 0;
3212         nm_i->nid_cnt[PREALLOC_NID] = 0;
3213         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3214         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3215         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3216
3217         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3218         INIT_LIST_HEAD(&nm_i->free_nid_list);
3219         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3220         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3221         INIT_LIST_HEAD(&nm_i->nat_entries);
3222         spin_lock_init(&nm_i->nat_list_lock);
3223
3224         mutex_init(&nm_i->build_lock);
3225         spin_lock_init(&nm_i->nid_list_lock);
3226         init_rwsem(&nm_i->nat_tree_lock);
3227
3228         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3229         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3230         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3231         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3232                                         GFP_KERNEL);
3233         if (!nm_i->nat_bitmap)
3234                 return -ENOMEM;
3235
3236         err = __get_nat_bitmaps(sbi);
3237         if (err)
3238                 return err;
3239
3240 #ifdef CONFIG_F2FS_CHECK_FS
3241         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3242                                         GFP_KERNEL);
3243         if (!nm_i->nat_bitmap_mir)
3244                 return -ENOMEM;
3245 #endif
3246
3247         return 0;
3248 }
3249
3250 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3251 {
3252         struct f2fs_nm_info *nm_i = NM_I(sbi);
3253         int i;
3254
3255         nm_i->free_nid_bitmap =
3256                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3257                                               nm_i->nat_blocks),
3258                               GFP_KERNEL);
3259         if (!nm_i->free_nid_bitmap)
3260                 return -ENOMEM;
3261
3262         for (i = 0; i < nm_i->nat_blocks; i++) {
3263                 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3264                         f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3265                 if (!nm_i->free_nid_bitmap[i])
3266                         return -ENOMEM;
3267         }
3268
3269         nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3270                                                                 GFP_KERNEL);
3271         if (!nm_i->nat_block_bitmap)
3272                 return -ENOMEM;
3273
3274         nm_i->free_nid_count =
3275                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3276                                               nm_i->nat_blocks),
3277                               GFP_KERNEL);
3278         if (!nm_i->free_nid_count)
3279                 return -ENOMEM;
3280         return 0;
3281 }
3282
3283 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3284 {
3285         int err;
3286
3287         sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3288                                                         GFP_KERNEL);
3289         if (!sbi->nm_info)
3290                 return -ENOMEM;
3291
3292         err = init_node_manager(sbi);
3293         if (err)
3294                 return err;
3295
3296         err = init_free_nid_cache(sbi);
3297         if (err)
3298                 return err;
3299
3300         /* load free nid status from nat_bits table */
3301         load_free_nid_bitmap(sbi);
3302
3303         return f2fs_build_free_nids(sbi, true, true);
3304 }
3305
3306 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3307 {
3308         struct f2fs_nm_info *nm_i = NM_I(sbi);
3309         struct free_nid *i, *next_i;
3310         struct nat_entry *natvec[NATVEC_SIZE];
3311         struct nat_entry_set *setvec[SETVEC_SIZE];
3312         nid_t nid = 0;
3313         unsigned int found;
3314
3315         if (!nm_i)
3316                 return;
3317
3318         /* destroy free nid list */
3319         spin_lock(&nm_i->nid_list_lock);
3320         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3321                 __remove_free_nid(sbi, i, FREE_NID);
3322                 spin_unlock(&nm_i->nid_list_lock);
3323                 kmem_cache_free(free_nid_slab, i);
3324                 spin_lock(&nm_i->nid_list_lock);
3325         }
3326         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3327         f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3328         f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3329         spin_unlock(&nm_i->nid_list_lock);
3330
3331         /* destroy nat cache */
3332         down_write(&nm_i->nat_tree_lock);
3333         while ((found = __gang_lookup_nat_cache(nm_i,
3334                                         nid, NATVEC_SIZE, natvec))) {
3335                 unsigned idx;
3336
3337                 nid = nat_get_nid(natvec[found - 1]) + 1;
3338                 for (idx = 0; idx < found; idx++) {
3339                         spin_lock(&nm_i->nat_list_lock);
3340                         list_del(&natvec[idx]->list);
3341                         spin_unlock(&nm_i->nat_list_lock);
3342
3343                         __del_from_nat_cache(nm_i, natvec[idx]);
3344                 }
3345         }
3346         f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3347
3348         /* destroy nat set cache */
3349         nid = 0;
3350         while ((found = __gang_lookup_nat_set(nm_i,
3351                                         nid, SETVEC_SIZE, setvec))) {
3352                 unsigned idx;
3353
3354                 nid = setvec[found - 1]->set + 1;
3355                 for (idx = 0; idx < found; idx++) {
3356                         /* entry_cnt is not zero, when cp_error was occurred */
3357                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3358                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3359                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3360                 }
3361         }
3362         up_write(&nm_i->nat_tree_lock);
3363
3364         kvfree(nm_i->nat_block_bitmap);
3365         if (nm_i->free_nid_bitmap) {
3366                 int i;
3367
3368                 for (i = 0; i < nm_i->nat_blocks; i++)
3369                         kvfree(nm_i->free_nid_bitmap[i]);
3370                 kvfree(nm_i->free_nid_bitmap);
3371         }
3372         kvfree(nm_i->free_nid_count);
3373
3374         kvfree(nm_i->nat_bitmap);
3375         kvfree(nm_i->nat_bits);
3376 #ifdef CONFIG_F2FS_CHECK_FS
3377         kvfree(nm_i->nat_bitmap_mir);
3378 #endif
3379         sbi->nm_info = NULL;
3380         kfree(nm_i);
3381 }
3382
3383 int __init f2fs_create_node_manager_caches(void)
3384 {
3385         nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3386                         sizeof(struct nat_entry));
3387         if (!nat_entry_slab)
3388                 goto fail;
3389
3390         free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3391                         sizeof(struct free_nid));
3392         if (!free_nid_slab)
3393                 goto destroy_nat_entry;
3394
3395         nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3396                         sizeof(struct nat_entry_set));
3397         if (!nat_entry_set_slab)
3398                 goto destroy_free_nid;
3399
3400         fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3401                         sizeof(struct fsync_node_entry));
3402         if (!fsync_node_entry_slab)
3403                 goto destroy_nat_entry_set;
3404         return 0;
3405
3406 destroy_nat_entry_set:
3407         kmem_cache_destroy(nat_entry_set_slab);
3408 destroy_free_nid:
3409         kmem_cache_destroy(free_nid_slab);
3410 destroy_nat_entry:
3411         kmem_cache_destroy(nat_entry_slab);
3412 fail:
3413         return -ENOMEM;
3414 }
3415
3416 void f2fs_destroy_node_manager_caches(void)
3417 {
3418         kmem_cache_destroy(fsync_node_entry_slab);
3419         kmem_cache_destroy(nat_entry_set_slab);
3420         kmem_cache_destroy(free_nid_slab);
3421         kmem_cache_destroy(nat_entry_slab);
3422 }