fs/f2fs/node.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/node.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/mpage.h>
  11 #include <linux/sched/mm.h>
  12 #include <linux/blkdev.h>
  13 #include <linux/pagevec.h>
  14 #include <linux/swap.h>
  15
  16 #include "f2fs.h"
  17 #include "node.h"
  18 #include "segment.h"
  19 #include "xattr.h"
  20 #include "iostat.h"
  21 #include <trace/events/f2fs.h>
  22
  23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
  24
  25 static struct kmem_cache *nat_entry_slab;
  26 static struct kmem_cache *free_nid_slab;
  27 static struct kmem_cache *nat_entry_set_slab;
  28 static struct kmem_cache *fsync_node_entry_slab;
  29
  30 /*
  31  * Check whether the given nid is within node id range.
  32  */
  33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
  34 {
  35         if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
  36                 set_sbi_flag(sbi, SBI_NEED_FSCK);
  37                 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
  38                           __func__, nid);
  39                 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         f2fs_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         f2fs_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         f2fs_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         f2fs_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         f2fs_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         f2fs_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         /* Let's mitigate lock contention of nat_tree_lock during checkpoint */
 434         if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
 435                 return;
 436
 437         new = __alloc_nat_entry(sbi, nid, false);
 438         if (!new)
 439                 return;
 440
 441         f2fs_down_write(&nm_i->nat_tree_lock);
 442         e = __lookup_nat_cache(nm_i, nid);
 443         if (!e)
 444                 e = __init_nat_entry(nm_i, new, ne, false);
 445         else
 446                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 447                                 nat_get_blkaddr(e) !=
 448                                         le32_to_cpu(ne->block_addr) ||
 449                                 nat_get_version(e) != ne->version);
 450         f2fs_up_write(&nm_i->nat_tree_lock);
 451         if (e != new)
 452                 __free_nat_entry(new);
 453 }
 454
 455 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 456                         block_t new_blkaddr, bool fsync_done)
 457 {
 458         struct f2fs_nm_info *nm_i = NM_I(sbi);
 459         struct nat_entry *e;
 460         struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
 461
 462         f2fs_down_write(&nm_i->nat_tree_lock);
 463         e = __lookup_nat_cache(nm_i, ni->nid);
 464         if (!e) {
 465                 e = __init_nat_entry(nm_i, new, NULL, true);
 466                 copy_node_info(&e->ni, ni);
 467                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 468         } else if (new_blkaddr == NEW_ADDR) {
 469                 /*
 470                  * when nid is reallocated,
 471                  * previous nat entry can be remained in nat cache.
 472                  * So, reinitialize it with new information.
 473                  */
 474                 copy_node_info(&e->ni, ni);
 475                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 476         }
 477         /* let's free early to reduce memory consumption */
 478         if (e != new)
 479                 __free_nat_entry(new);
 480
 481         /* sanity check */
 482         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 483         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 484                         new_blkaddr == NULL_ADDR);
 485         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 486                         new_blkaddr == NEW_ADDR);
 487         f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
 488                         new_blkaddr == NEW_ADDR);
 489
 490         /* increment version no as node is removed */
 491         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 492                 unsigned char version = nat_get_version(e);
 493
 494                 nat_set_version(e, inc_node_version(version));
 495         }
 496
 497         /* change address */
 498         nat_set_blkaddr(e, new_blkaddr);
 499         if (!__is_valid_data_blkaddr(new_blkaddr))
 500                 set_nat_flag(e, IS_CHECKPOINTED, false);
 501         __set_nat_cache_dirty(nm_i, e);
 502
 503         /* update fsync_mark if its inode nat entry is still alive */
 504         if (ni->nid != ni->ino)
 505                 e = __lookup_nat_cache(nm_i, ni->ino);
 506         if (e) {
 507                 if (fsync_done && ni->nid == ni->ino)
 508                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 509                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 510         }
 511         f2fs_up_write(&nm_i->nat_tree_lock);
 512 }
 513
 514 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 515 {
 516         struct f2fs_nm_info *nm_i = NM_I(sbi);
 517         int nr = nr_shrink;
 518
 519         if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
 520                 return 0;
 521
 522         spin_lock(&nm_i->nat_list_lock);
 523         while (nr_shrink) {
 524                 struct nat_entry *ne;
 525
 526                 if (list_empty(&nm_i->nat_entries))
 527                         break;
 528
 529                 ne = list_first_entry(&nm_i->nat_entries,
 530                                         struct nat_entry, list);
 531                 list_del(&ne->list);
 532                 spin_unlock(&nm_i->nat_list_lock);
 533
 534                 __del_from_nat_cache(nm_i, ne);
 535                 nr_shrink--;
 536
 537                 spin_lock(&nm_i->nat_list_lock);
 538         }
 539         spin_unlock(&nm_i->nat_list_lock);
 540
 541         f2fs_up_write(&nm_i->nat_tree_lock);
 542         return nr - nr_shrink;
 543 }
 544
 545 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
 546                                 struct node_info *ni, bool checkpoint_context)
 547 {
 548         struct f2fs_nm_info *nm_i = NM_I(sbi);
 549         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 550         struct f2fs_journal *journal = curseg->journal;
 551         nid_t start_nid = START_NID(nid);
 552         struct f2fs_nat_block *nat_blk;
 553         struct page *page = NULL;
 554         struct f2fs_nat_entry ne;
 555         struct nat_entry *e;
 556         pgoff_t index;
 557         block_t blkaddr;
 558         int i;
 559
 560         ni->nid = nid;
 561 retry:
 562         /* Check nat cache */
 563         f2fs_down_read(&nm_i->nat_tree_lock);
 564         e = __lookup_nat_cache(nm_i, nid);
 565         if (e) {
 566                 ni->ino = nat_get_ino(e);
 567                 ni->blk_addr = nat_get_blkaddr(e);
 568                 ni->version = nat_get_version(e);
 569                 f2fs_up_read(&nm_i->nat_tree_lock);
 570                 return 0;
 571         }
 572
 573         /*
 574          * Check current segment summary by trying to grab journal_rwsem first.
 575          * This sem is on the critical path on the checkpoint requiring the above
 576          * nat_tree_lock. Therefore, we should retry, if we failed to grab here
 577          * while not bothering checkpoint.
 578          */
 579         if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
 580                 down_read(&curseg->journal_rwsem);
 581         } else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
 582                                 !down_read_trylock(&curseg->journal_rwsem)) {
 583                 f2fs_up_read(&nm_i->nat_tree_lock);
 584                 goto retry;
 585         }
 586
 587         i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 588         if (i >= 0) {
 589                 ne = nat_in_journal(journal, i);
 590                 node_info_from_raw_nat(ni, &ne);
 591         }
 592         up_read(&curseg->journal_rwsem);
 593         if (i >= 0) {
 594                 f2fs_up_read(&nm_i->nat_tree_lock);
 595                 goto cache;
 596         }
 597
 598         /* Fill node_info from nat page */
 599         index = current_nat_addr(sbi, nid);
 600         f2fs_up_read(&nm_i->nat_tree_lock);
 601
 602         page = f2fs_get_meta_page(sbi, index);
 603         if (IS_ERR(page))
 604                 return PTR_ERR(page);
 605
 606         nat_blk = (struct f2fs_nat_block *)page_address(page);
 607         ne = nat_blk->entries[nid - start_nid];
 608         node_info_from_raw_nat(ni, &ne);
 609         f2fs_put_page(page, 1);
 610 cache:
 611         blkaddr = le32_to_cpu(ne.block_addr);
 612         if (__is_valid_data_blkaddr(blkaddr) &&
 613                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
 614                 return -EFAULT;
 615
 616         /* cache nat entry */
 617         cache_nat_entry(sbi, nid, &ne);
 618         return 0;
 619 }
 620
 621 /*
 622  * readahead MAX_RA_NODE number of node pages.
 623  */
 624 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
 625 {
 626         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 627         struct blk_plug plug;
 628         int i, end;
 629         nid_t nid;
 630
 631         blk_start_plug(&plug);
 632
 633         /* Then, try readahead for siblings of the desired node */
 634         end = start + n;
 635         end = min(end, NIDS_PER_BLOCK);
 636         for (i = start; i < end; i++) {
 637                 nid = get_nid(parent, i, false);
 638                 f2fs_ra_node_page(sbi, nid);
 639         }
 640
 641         blk_finish_plug(&plug);
 642 }
 643
 644 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 645 {
 646         const long direct_index = ADDRS_PER_INODE(dn->inode);
 647         const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
 648         const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
 649         unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
 650         int cur_level = dn->cur_level;
 651         int max_level = dn->max_level;
 652         pgoff_t base = 0;
 653
 654         if (!dn->max_level)
 655                 return pgofs + 1;
 656
 657         while (max_level-- > cur_level)
 658                 skipped_unit *= NIDS_PER_BLOCK;
 659
 660         switch (dn->max_level) {
 661         case 3:
 662                 base += 2 * indirect_blks;
 663                 fallthrough;
 664         case 2:
 665                 base += 2 * direct_blks;
 666                 fallthrough;
 667         case 1:
 668                 base += direct_index;
 669                 break;
 670         default:
 671                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 672         }
 673
 674         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 675 }
 676
 677 /*
 678  * The maximum depth is four.
 679  * Offset[0] will have raw inode offset.
 680  */
 681 static int get_node_path(struct inode *inode, long block,
 682                                 int offset[4], unsigned int noffset[4])
 683 {
 684         const long direct_index = ADDRS_PER_INODE(inode);
 685         const long direct_blks = ADDRS_PER_BLOCK(inode);
 686         const long dptrs_per_blk = NIDS_PER_BLOCK;
 687         const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
 688         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 689         int n = 0;
 690         int level = 0;
 691
 692         noffset[0] = 0;
 693
 694         if (block < direct_index) {
 695                 offset[n] = block;
 696                 goto got;
 697         }
 698         block -= direct_index;
 699         if (block < direct_blks) {
 700                 offset[n++] = NODE_DIR1_BLOCK;
 701                 noffset[n] = 1;
 702                 offset[n] = block;
 703                 level = 1;
 704                 goto got;
 705         }
 706         block -= direct_blks;
 707         if (block < direct_blks) {
 708                 offset[n++] = NODE_DIR2_BLOCK;
 709                 noffset[n] = 2;
 710                 offset[n] = block;
 711                 level = 1;
 712                 goto got;
 713         }
 714         block -= direct_blks;
 715         if (block < indirect_blks) {
 716                 offset[n++] = NODE_IND1_BLOCK;
 717                 noffset[n] = 3;
 718                 offset[n++] = block / direct_blks;
 719                 noffset[n] = 4 + offset[n - 1];
 720                 offset[n] = block % direct_blks;
 721                 level = 2;
 722                 goto got;
 723         }
 724         block -= indirect_blks;
 725         if (block < indirect_blks) {
 726                 offset[n++] = NODE_IND2_BLOCK;
 727                 noffset[n] = 4 + dptrs_per_blk;
 728                 offset[n++] = block / direct_blks;
 729                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 730                 offset[n] = block % direct_blks;
 731                 level = 2;
 732                 goto got;
 733         }
 734         block -= indirect_blks;
 735         if (block < dindirect_blks) {
 736                 offset[n++] = NODE_DIND_BLOCK;
 737                 noffset[n] = 5 + (dptrs_per_blk * 2);
 738                 offset[n++] = block / indirect_blks;
 739                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 740                               offset[n - 1] * (dptrs_per_blk + 1);
 741                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 742                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 743                               offset[n - 2] * (dptrs_per_blk + 1) +
 744                               offset[n - 1];
 745                 offset[n] = block % direct_blks;
 746                 level = 3;
 747                 goto got;
 748         } else {
 749                 return -E2BIG;
 750         }
 751 got:
 752         return level;
 753 }
 754
 755 /*
 756  * Caller should call f2fs_put_dnode(dn).
 757  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 758  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
 759  */
 760 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 761 {
 762         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 763         struct page *npage[4];
 764         struct page *parent = NULL;
 765         int offset[4];
 766         unsigned int noffset[4];
 767         nid_t nids[4];
 768         int level, i = 0;
 769         int err = 0;
 770
 771         level = get_node_path(dn->inode, index, offset, noffset);
 772         if (level < 0)
 773                 return level;
 774
 775         nids[0] = dn->inode->i_ino;
 776         npage[0] = dn->inode_page;
 777
 778         if (!npage[0]) {
 779                 npage[0] = f2fs_get_node_page(sbi, nids[0]);
 780                 if (IS_ERR(npage[0]))
 781                         return PTR_ERR(npage[0]);
 782         }
 783
 784         /* if inline_data is set, should not report any block indices */
 785         if (f2fs_has_inline_data(dn->inode) && index) {
 786                 err = -ENOENT;
 787                 f2fs_put_page(npage[0], 1);
 788                 goto release_out;
 789         }
 790
 791         parent = npage[0];
 792         if (level != 0)
 793                 nids[1] = get_nid(parent, offset[0], true);
 794         dn->inode_page = npage[0];
 795         dn->inode_page_locked = true;
 796
 797         /* get indirect or direct nodes */
 798         for (i = 1; i <= level; i++) {
 799                 bool done = false;
 800
 801                 if (!nids[i] && mode == ALLOC_NODE) {
 802                         /* alloc new node */
 803                         if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
 804                                 err = -ENOSPC;
 805                                 goto release_pages;
 806                         }
 807
 808                         dn->nid = nids[i];
 809                         npage[i] = f2fs_new_node_page(dn, noffset[i]);
 810                         if (IS_ERR(npage[i])) {
 811                                 f2fs_alloc_nid_failed(sbi, nids[i]);
 812                                 err = PTR_ERR(npage[i]);
 813                                 goto release_pages;
 814                         }
 815
 816                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 817                         f2fs_alloc_nid_done(sbi, nids[i]);
 818                         done = true;
 819                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 820                         npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
 821                         if (IS_ERR(npage[i])) {
 822                                 err = PTR_ERR(npage[i]);
 823                                 goto release_pages;
 824                         }
 825                         done = true;
 826                 }
 827                 if (i == 1) {
 828                         dn->inode_page_locked = false;
 829                         unlock_page(parent);
 830                 } else {
 831                         f2fs_put_page(parent, 1);
 832                 }
 833
 834                 if (!done) {
 835                         npage[i] = f2fs_get_node_page(sbi, nids[i]);
 836                         if (IS_ERR(npage[i])) {
 837                                 err = PTR_ERR(npage[i]);
 838                                 f2fs_put_page(npage[0], 0);
 839                                 goto release_out;
 840                         }
 841                 }
 842                 if (i < level) {
 843                         parent = npage[i];
 844                         nids[i + 1] = get_nid(parent, offset[i], false);
 845                 }
 846         }
 847         dn->nid = nids[level];
 848         dn->ofs_in_node = offset[level];
 849         dn->node_page = npage[level];
 850         dn->data_blkaddr = f2fs_data_blkaddr(dn);
 851
 852         if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
 853                                         f2fs_sb_has_readonly(sbi)) {
 854                 unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn);
 855                 block_t blkaddr;
 856
 857                 if (!c_len)
 858                         goto out;
 859
 860                 blkaddr = f2fs_data_blkaddr(dn);
 861                 if (blkaddr == COMPRESS_ADDR)
 862                         blkaddr = data_blkaddr(dn->inode, dn->node_page,
 863                                                 dn->ofs_in_node + 1);
 864
 865                 f2fs_update_extent_tree_range_compressed(dn->inode,
 866                                         index, blkaddr,
 867                                         F2FS_I(dn->inode)->i_cluster_size,
 868                                         c_len);
 869         }
 870 out:
 871         return 0;
 872
 873 release_pages:
 874         f2fs_put_page(parent, 1);
 875         if (i > 1)
 876                 f2fs_put_page(npage[0], 0);
 877 release_out:
 878         dn->inode_page = NULL;
 879         dn->node_page = NULL;
 880         if (err == -ENOENT) {
 881                 dn->cur_level = i;
 882                 dn->max_level = level;
 883                 dn->ofs_in_node = offset[level];
 884         }
 885         return err;
 886 }
 887
 888 static int truncate_node(struct dnode_of_data *dn)
 889 {
 890         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 891         struct node_info ni;
 892         int err;
 893         pgoff_t index;
 894
 895         err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
 896         if (err)
 897                 return err;
 898
 899         /* Deallocate node address */
 900         f2fs_invalidate_blocks(sbi, ni.blk_addr);
 901         dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
 902         set_node_addr(sbi, &ni, NULL_ADDR, false);
 903
 904         if (dn->nid == dn->inode->i_ino) {
 905                 f2fs_remove_orphan_inode(sbi, dn->nid);
 906                 dec_valid_inode_count(sbi);
 907                 f2fs_inode_synced(dn->inode);
 908         }
 909
 910         clear_node_page_dirty(dn->node_page);
 911         set_sbi_flag(sbi, SBI_IS_DIRTY);
 912
 913         index = dn->node_page->index;
 914         f2fs_put_page(dn->node_page, 1);
 915
 916         invalidate_mapping_pages(NODE_MAPPING(sbi),
 917                         index, index);
 918
 919         dn->node_page = NULL;
 920         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 921
 922         return 0;
 923 }
 924
 925 static int truncate_dnode(struct dnode_of_data *dn)
 926 {
 927         struct page *page;
 928         int err;
 929
 930         if (dn->nid == 0)
 931                 return 1;
 932
 933         /* get direct node */
 934         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 935         if (PTR_ERR(page) == -ENOENT)
 936                 return 1;
 937         else if (IS_ERR(page))
 938                 return PTR_ERR(page);
 939
 940         /* Make dnode_of_data for parameter */
 941         dn->node_page = page;
 942         dn->ofs_in_node = 0;
 943         f2fs_truncate_data_blocks(dn);
 944         err = truncate_node(dn);
 945         if (err)
 946                 return err;
 947
 948         return 1;
 949 }
 950
 951 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 952                                                 int ofs, int depth)
 953 {
 954         struct dnode_of_data rdn = *dn;
 955         struct page *page;
 956         struct f2fs_node *rn;
 957         nid_t child_nid;
 958         unsigned int child_nofs;
 959         int freed = 0;
 960         int i, ret;
 961
 962         if (dn->nid == 0)
 963                 return NIDS_PER_BLOCK + 1;
 964
 965         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 966
 967         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 968         if (IS_ERR(page)) {
 969                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 970                 return PTR_ERR(page);
 971         }
 972
 973         f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 974
 975         rn = F2FS_NODE(page);
 976         if (depth < 3) {
 977                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 978                         child_nid = le32_to_cpu(rn->in.nid[i]);
 979                         if (child_nid == 0)
 980                                 continue;
 981                         rdn.nid = child_nid;
 982                         ret = truncate_dnode(&rdn);
 983                         if (ret < 0)
 984                                 goto out_err;
 985                         if (set_nid(page, i, 0, false))
 986                                 dn->node_changed = true;
 987                 }
 988         } else {
 989                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 990                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 991                         child_nid = le32_to_cpu(rn->in.nid[i]);
 992                         if (child_nid == 0) {
 993                                 child_nofs += NIDS_PER_BLOCK + 1;
 994                                 continue;
 995                         }
 996                         rdn.nid = child_nid;
 997                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 998                         if (ret == (NIDS_PER_BLOCK + 1)) {
 999                                 if (set_nid(page, i, 0, false))
1000                                         dn->node_changed = true;
1001                                 child_nofs += ret;
1002                         } else if (ret < 0 && ret != -ENOENT) {
1003                                 goto out_err;
1004                         }
1005                 }
1006                 freed = child_nofs;
1007         }
1008
1009         if (!ofs) {
1010                 /* remove current indirect node */
1011                 dn->node_page = page;
1012                 ret = truncate_node(dn);
1013                 if (ret)
1014                         goto out_err;
1015                 freed++;
1016         } else {
1017                 f2fs_put_page(page, 1);
1018         }
1019         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1020         return freed;
1021
1022 out_err:
1023         f2fs_put_page(page, 1);
1024         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1025         return ret;
1026 }
1027
1028 static int truncate_partial_nodes(struct dnode_of_data *dn,
1029                         struct f2fs_inode *ri, int *offset, int depth)
1030 {
1031         struct page *pages[2];
1032         nid_t nid[3];
1033         nid_t child_nid;
1034         int err = 0;
1035         int i;
1036         int idx = depth - 2;
1037
1038         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1039         if (!nid[0])
1040                 return 0;
1041
1042         /* get indirect nodes in the path */
1043         for (i = 0; i < idx + 1; i++) {
1044                 /* reference count'll be increased */
1045                 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1046                 if (IS_ERR(pages[i])) {
1047                         err = PTR_ERR(pages[i]);
1048                         idx = i - 1;
1049                         goto fail;
1050                 }
1051                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1052         }
1053
1054         f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1055
1056         /* free direct nodes linked to a partial indirect node */
1057         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1058                 child_nid = get_nid(pages[idx], i, false);
1059                 if (!child_nid)
1060                         continue;
1061                 dn->nid = child_nid;
1062                 err = truncate_dnode(dn);
1063                 if (err < 0)
1064                         goto fail;
1065                 if (set_nid(pages[idx], i, 0, false))
1066                         dn->node_changed = true;
1067         }
1068
1069         if (offset[idx + 1] == 0) {
1070                 dn->node_page = pages[idx];
1071                 dn->nid = nid[idx];
1072                 err = truncate_node(dn);
1073                 if (err)
1074                         goto fail;
1075         } else {
1076                 f2fs_put_page(pages[idx], 1);
1077         }
1078         offset[idx]++;
1079         offset[idx + 1] = 0;
1080         idx--;
1081 fail:
1082         for (i = idx; i >= 0; i--)
1083                 f2fs_put_page(pages[i], 1);
1084
1085         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1086
1087         return err;
1088 }
1089
1090 /*
1091  * All the block addresses of data and nodes should be nullified.
1092  */
1093 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1094 {
1095         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1096         int err = 0, cont = 1;
1097         int level, offset[4], noffset[4];
1098         unsigned int nofs = 0;
1099         struct f2fs_inode *ri;
1100         struct dnode_of_data dn;
1101         struct page *page;
1102
1103         trace_f2fs_truncate_inode_blocks_enter(inode, from);
1104
1105         level = get_node_path(inode, from, offset, noffset);
1106         if (level < 0) {
1107                 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1108                 return level;
1109         }
1110
1111         page = f2fs_get_node_page(sbi, inode->i_ino);
1112         if (IS_ERR(page)) {
1113                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1114                 return PTR_ERR(page);
1115         }
1116
1117         set_new_dnode(&dn, inode, page, NULL, 0);
1118         unlock_page(page);
1119
1120         ri = F2FS_INODE(page);
1121         switch (level) {
1122         case 0:
1123         case 1:
1124                 nofs = noffset[1];
1125                 break;
1126         case 2:
1127                 nofs = noffset[1];
1128                 if (!offset[level - 1])
1129                         goto skip_partial;
1130                 err = truncate_partial_nodes(&dn, ri, offset, level);
1131                 if (err < 0 && err != -ENOENT)
1132                         goto fail;
1133                 nofs += 1 + NIDS_PER_BLOCK;
1134                 break;
1135         case 3:
1136                 nofs = 5 + 2 * NIDS_PER_BLOCK;
1137                 if (!offset[level - 1])
1138                         goto skip_partial;
1139                 err = truncate_partial_nodes(&dn, ri, offset, level);
1140                 if (err < 0 && err != -ENOENT)
1141                         goto fail;
1142                 break;
1143         default:
1144                 BUG();
1145         }
1146
1147 skip_partial:
1148         while (cont) {
1149                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1150                 switch (offset[0]) {
1151                 case NODE_DIR1_BLOCK:
1152                 case NODE_DIR2_BLOCK:
1153                         err = truncate_dnode(&dn);
1154                         break;
1155
1156                 case NODE_IND1_BLOCK:
1157                 case NODE_IND2_BLOCK:
1158                         err = truncate_nodes(&dn, nofs, offset[1], 2);
1159                         break;
1160
1161                 case NODE_DIND_BLOCK:
1162                         err = truncate_nodes(&dn, nofs, offset[1], 3);
1163                         cont = 0;
1164                         break;
1165
1166                 default:
1167                         BUG();
1168                 }
1169                 if (err < 0 && err != -ENOENT)
1170                         goto fail;
1171                 if (offset[1] == 0 &&
1172                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1173                         lock_page(page);
1174                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
1175                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1176                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1177                         set_page_dirty(page);
1178                         unlock_page(page);
1179                 }
1180                 offset[1] = 0;
1181                 offset[0]++;
1182                 nofs += err;
1183         }
1184 fail:
1185         f2fs_put_page(page, 0);
1186         trace_f2fs_truncate_inode_blocks_exit(inode, err);
1187         return err > 0 ? 0 : err;
1188 }
1189
1190 /* caller must lock inode page */
1191 int f2fs_truncate_xattr_node(struct inode *inode)
1192 {
1193         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1194         nid_t nid = F2FS_I(inode)->i_xattr_nid;
1195         struct dnode_of_data dn;
1196         struct page *npage;
1197         int err;
1198
1199         if (!nid)
1200                 return 0;
1201
1202         npage = f2fs_get_node_page(sbi, nid);
1203         if (IS_ERR(npage))
1204                 return PTR_ERR(npage);
1205
1206         set_new_dnode(&dn, inode, NULL, npage, nid);
1207         err = truncate_node(&dn);
1208         if (err) {
1209                 f2fs_put_page(npage, 1);
1210                 return err;
1211         }
1212
1213         f2fs_i_xnid_write(inode, 0);
1214
1215         return 0;
1216 }
1217
1218 /*
1219  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1220  * f2fs_unlock_op().
1221  */
1222 int f2fs_remove_inode_page(struct inode *inode)
1223 {
1224         struct dnode_of_data dn;
1225         int err;
1226
1227         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1228         err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1229         if (err)
1230                 return err;
1231
1232         err = f2fs_truncate_xattr_node(inode);
1233         if (err) {
1234                 f2fs_put_dnode(&dn);
1235                 return err;
1236         }
1237
1238         /* remove potential inline_data blocks */
1239         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1240                                 S_ISLNK(inode->i_mode))
1241                 f2fs_truncate_data_blocks_range(&dn, 1);
1242
1243         /* 0 is possible, after f2fs_new_inode() has failed */
1244         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1245                 f2fs_put_dnode(&dn);
1246                 return -EIO;
1247         }
1248
1249         if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1250                 f2fs_warn(F2FS_I_SB(inode),
1251                         "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1252                         inode->i_ino, (unsigned long long)inode->i_blocks);
1253                 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1254         }
1255
1256         /* will put inode & node pages */
1257         err = truncate_node(&dn);
1258         if (err) {
1259                 f2fs_put_dnode(&dn);
1260                 return err;
1261         }
1262         return 0;
1263 }
1264
1265 struct page *f2fs_new_inode_page(struct inode *inode)
1266 {
1267         struct dnode_of_data dn;
1268
1269         /* allocate inode page for new inode */
1270         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1271
1272         /* caller should f2fs_put_page(page, 1); */
1273         return f2fs_new_node_page(&dn, 0);
1274 }
1275
1276 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1277 {
1278         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1279         struct node_info new_ni;
1280         struct page *page;
1281         int err;
1282
1283         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1284                 return ERR_PTR(-EPERM);
1285
1286         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1287         if (!page)
1288                 return ERR_PTR(-ENOMEM);
1289
1290         if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1291                 goto fail;
1292
1293 #ifdef CONFIG_F2FS_CHECK_FS
1294         err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1295         if (err) {
1296                 dec_valid_node_count(sbi, dn->inode, !ofs);
1297                 goto fail;
1298         }
1299         f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1300 #endif
1301         new_ni.nid = dn->nid;
1302         new_ni.ino = dn->inode->i_ino;
1303         new_ni.blk_addr = NULL_ADDR;
1304         new_ni.flag = 0;
1305         new_ni.version = 0;
1306         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1307
1308         f2fs_wait_on_page_writeback(page, NODE, true, true);
1309         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1310         set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1311         if (!PageUptodate(page))
1312                 SetPageUptodate(page);
1313         if (set_page_dirty(page))
1314                 dn->node_changed = true;
1315
1316         if (f2fs_has_xattr_block(ofs))
1317                 f2fs_i_xnid_write(dn->inode, dn->nid);
1318
1319         if (ofs == 0)
1320                 inc_valid_inode_count(sbi);
1321         return page;
1322
1323 fail:
1324         clear_node_page_dirty(page);
1325         f2fs_put_page(page, 1);
1326         return ERR_PTR(err);
1327 }
1328
1329 /*
1330  * Caller should do after getting the following values.
1331  * 0: f2fs_put_page(page, 0)
1332  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1333  */
1334 static int read_node_page(struct page *page, int op_flags)
1335 {
1336         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1337         struct node_info ni;
1338         struct f2fs_io_info fio = {
1339                 .sbi = sbi,
1340                 .type = NODE,
1341                 .op = REQ_OP_READ,
1342                 .op_flags = op_flags,
1343                 .page = page,
1344                 .encrypted_page = NULL,
1345         };
1346         int err;
1347
1348         if (PageUptodate(page)) {
1349                 if (!f2fs_inode_chksum_verify(sbi, page)) {
1350                         ClearPageUptodate(page);
1351                         return -EFSBADCRC;
1352                 }
1353                 return LOCKED_PAGE;
1354         }
1355
1356         err = f2fs_get_node_info(sbi, page->index, &ni, false);
1357         if (err)
1358                 return err;
1359
1360         /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1361         if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR) ||
1362                         is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1363                 ClearPageUptodate(page);
1364                 return -ENOENT;
1365         }
1366
1367         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1368
1369         err = f2fs_submit_page_bio(&fio);
1370
1371         if (!err)
1372                 f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1373
1374         return err;
1375 }
1376
1377 /*
1378  * Readahead a node page
1379  */
1380 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1381 {
1382         struct page *apage;
1383         int err;
1384
1385         if (!nid)
1386                 return;
1387         if (f2fs_check_nid_range(sbi, nid))
1388                 return;
1389
1390         apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1391         if (apage)
1392                 return;
1393
1394         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1395         if (!apage)
1396                 return;
1397
1398         err = read_node_page(apage, REQ_RAHEAD);
1399         f2fs_put_page(apage, err ? 1 : 0);
1400 }
1401
1402 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1403                                         struct page *parent, int start)
1404 {
1405         struct page *page;
1406         int err;
1407
1408         if (!nid)
1409                 return ERR_PTR(-ENOENT);
1410         if (f2fs_check_nid_range(sbi, nid))
1411                 return ERR_PTR(-EINVAL);
1412 repeat:
1413         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1414         if (!page)
1415                 return ERR_PTR(-ENOMEM);
1416
1417         err = read_node_page(page, 0);
1418         if (err < 0) {
1419                 f2fs_put_page(page, 1);
1420                 return ERR_PTR(err);
1421         } else if (err == LOCKED_PAGE) {
1422                 err = 0;
1423                 goto page_hit;
1424         }
1425
1426         if (parent)
1427                 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1428
1429         lock_page(page);
1430
1431         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1432                 f2fs_put_page(page, 1);
1433                 goto repeat;
1434         }
1435
1436         if (unlikely(!PageUptodate(page))) {
1437                 err = -EIO;
1438                 goto out_err;
1439         }
1440
1441         if (!f2fs_inode_chksum_verify(sbi, page)) {
1442                 err = -EFSBADCRC;
1443                 goto out_err;
1444         }
1445 page_hit:
1446         if (unlikely(nid != nid_of_node(page))) {
1447                 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1448                           nid, nid_of_node(page), ino_of_node(page),
1449                           ofs_of_node(page), cpver_of_node(page),
1450                           next_blkaddr_of_node(page));
1451                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1452                 err = -EINVAL;
1453 out_err:
1454                 ClearPageUptodate(page);
1455                 f2fs_put_page(page, 1);
1456                 return ERR_PTR(err);
1457         }
1458         return page;
1459 }
1460
1461 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1462 {
1463         return __get_node_page(sbi, nid, NULL, 0);
1464 }
1465
1466 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1467 {
1468         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1469         nid_t nid = get_nid(parent, start, false);
1470
1471         return __get_node_page(sbi, nid, parent, start);
1472 }
1473
1474 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1475 {
1476         struct inode *inode;
1477         struct page *page;
1478         int ret;
1479
1480         /* should flush inline_data before evict_inode */
1481         inode = ilookup(sbi->sb, ino);
1482         if (!inode)
1483                 return;
1484
1485         page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1486                                         FGP_LOCK|FGP_NOWAIT, 0);
1487         if (!page)
1488                 goto iput_out;
1489
1490         if (!PageUptodate(page))
1491                 goto page_out;
1492
1493         if (!PageDirty(page))
1494                 goto page_out;
1495
1496         if (!clear_page_dirty_for_io(page))
1497                 goto page_out;
1498
1499         ret = f2fs_write_inline_data(inode, page);
1500         inode_dec_dirty_pages(inode);
1501         f2fs_remove_dirty_inode(inode);
1502         if (ret)
1503                 set_page_dirty(page);
1504 page_out:
1505         f2fs_put_page(page, 1);
1506 iput_out:
1507         iput(inode);
1508 }
1509
1510 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1511 {
1512         pgoff_t index;
1513         struct pagevec pvec;
1514         struct page *last_page = NULL;
1515         int nr_pages;
1516
1517         pagevec_init(&pvec);
1518         index = 0;
1519
1520         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1521                                 PAGECACHE_TAG_DIRTY))) {
1522                 int i;
1523
1524                 for (i = 0; i < nr_pages; i++) {
1525                         struct page *page = pvec.pages[i];
1526
1527                         if (unlikely(f2fs_cp_error(sbi))) {
1528                                 f2fs_put_page(last_page, 0);
1529                                 pagevec_release(&pvec);
1530                                 return ERR_PTR(-EIO);
1531                         }
1532
1533                         if (!IS_DNODE(page) || !is_cold_node(page))
1534                                 continue;
1535                         if (ino_of_node(page) != ino)
1536                                 continue;
1537
1538                         lock_page(page);
1539
1540                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1541 continue_unlock:
1542                                 unlock_page(page);
1543                                 continue;
1544                         }
1545                         if (ino_of_node(page) != ino)
1546                                 goto continue_unlock;
1547
1548                         if (!PageDirty(page)) {
1549                                 /* someone wrote it for us */
1550                                 goto continue_unlock;
1551                         }
1552
1553                         if (last_page)
1554                                 f2fs_put_page(last_page, 0);
1555
1556                         get_page(page);
1557                         last_page = page;
1558                         unlock_page(page);
1559                 }
1560                 pagevec_release(&pvec);
1561                 cond_resched();
1562         }
1563         return last_page;
1564 }
1565
1566 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1567                                 struct writeback_control *wbc, bool do_balance,
1568                                 enum iostat_type io_type, unsigned int *seq_id)
1569 {
1570         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1571         nid_t nid;
1572         struct node_info ni;
1573         struct f2fs_io_info fio = {
1574                 .sbi = sbi,
1575                 .ino = ino_of_node(page),
1576                 .type = NODE,
1577                 .op = REQ_OP_WRITE,
1578                 .op_flags = wbc_to_write_flags(wbc),
1579                 .page = page,
1580                 .encrypted_page = NULL,
1581                 .submitted = false,
1582                 .io_type = io_type,
1583                 .io_wbc = wbc,
1584         };
1585         unsigned int seq;
1586
1587         trace_f2fs_writepage(page, NODE);
1588
1589         if (unlikely(f2fs_cp_error(sbi))) {
1590                 ClearPageUptodate(page);
1591                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1592                 unlock_page(page);
1593                 return 0;
1594         }
1595
1596         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1597                 goto redirty_out;
1598
1599         if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1600                         wbc->sync_mode == WB_SYNC_NONE &&
1601                         IS_DNODE(page) && is_cold_node(page))
1602                 goto redirty_out;
1603
1604         /* get old block addr of this node page */
1605         nid = nid_of_node(page);
1606         f2fs_bug_on(sbi, page->index != nid);
1607
1608         if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1609                 goto redirty_out;
1610
1611         if (wbc->for_reclaim) {
1612                 if (!f2fs_down_read_trylock(&sbi->node_write))
1613                         goto redirty_out;
1614         } else {
1615                 f2fs_down_read(&sbi->node_write);
1616         }
1617
1618         /* This page is already truncated */
1619         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1620                 ClearPageUptodate(page);
1621                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1622                 f2fs_up_read(&sbi->node_write);
1623                 unlock_page(page);
1624                 return 0;
1625         }
1626
1627         if (__is_valid_data_blkaddr(ni.blk_addr) &&
1628                 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1629                                         DATA_GENERIC_ENHANCE)) {
1630                 f2fs_up_read(&sbi->node_write);
1631                 goto redirty_out;
1632         }
1633
1634         if (atomic && !test_opt(sbi, NOBARRIER))
1635                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1636
1637         /* should add to global list before clearing PAGECACHE status */
1638         if (f2fs_in_warm_node_list(sbi, page)) {
1639                 seq = f2fs_add_fsync_node_entry(sbi, page);
1640                 if (seq_id)
1641                         *seq_id = seq;
1642         }
1643
1644         set_page_writeback(page);
1645         ClearPageError(page);
1646
1647         fio.old_blkaddr = ni.blk_addr;
1648         f2fs_do_write_node_page(nid, &fio);
1649         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1650         dec_page_count(sbi, F2FS_DIRTY_NODES);
1651         f2fs_up_read(&sbi->node_write);
1652
1653         if (wbc->for_reclaim) {
1654                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1655                 submitted = NULL;
1656         }
1657
1658         unlock_page(page);
1659
1660         if (unlikely(f2fs_cp_error(sbi))) {
1661                 f2fs_submit_merged_write(sbi, NODE);
1662                 submitted = NULL;
1663         }
1664         if (submitted)
1665                 *submitted = fio.submitted;
1666
1667         if (do_balance)
1668                 f2fs_balance_fs(sbi, false);
1669         return 0;
1670
1671 redirty_out:
1672         redirty_page_for_writepage(wbc, page);
1673         return AOP_WRITEPAGE_ACTIVATE;
1674 }
1675
1676 int f2fs_move_node_page(struct page *node_page, int gc_type)
1677 {
1678         int err = 0;
1679
1680         if (gc_type == FG_GC) {
1681                 struct writeback_control wbc = {
1682                         .sync_mode = WB_SYNC_ALL,
1683                         .nr_to_write = 1,
1684                         .for_reclaim = 0,
1685                 };
1686
1687                 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1688
1689                 set_page_dirty(node_page);
1690
1691                 if (!clear_page_dirty_for_io(node_page)) {
1692                         err = -EAGAIN;
1693                         goto out_page;
1694                 }
1695
1696                 if (__write_node_page(node_page, false, NULL,
1697                                         &wbc, false, FS_GC_NODE_IO, NULL)) {
1698                         err = -EAGAIN;
1699                         unlock_page(node_page);
1700                 }
1701                 goto release_page;
1702         } else {
1703                 /* set page dirty and write it */
1704                 if (!PageWriteback(node_page))
1705                         set_page_dirty(node_page);
1706         }
1707 out_page:
1708         unlock_page(node_page);
1709 release_page:
1710         f2fs_put_page(node_page, 0);
1711         return err;
1712 }
1713
1714 static int f2fs_write_node_page(struct page *page,
1715                                 struct writeback_control *wbc)
1716 {
1717         return __write_node_page(page, false, NULL, wbc, false,
1718                                                 FS_NODE_IO, NULL);
1719 }
1720
1721 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1722                         struct writeback_control *wbc, bool atomic,
1723                         unsigned int *seq_id)
1724 {
1725         pgoff_t index;
1726         struct pagevec pvec;
1727         int ret = 0;
1728         struct page *last_page = NULL;
1729         bool marked = false;
1730         nid_t ino = inode->i_ino;
1731         int nr_pages;
1732         int nwritten = 0;
1733
1734         if (atomic) {
1735                 last_page = last_fsync_dnode(sbi, ino);
1736                 if (IS_ERR_OR_NULL(last_page))
1737                         return PTR_ERR_OR_ZERO(last_page);
1738         }
1739 retry:
1740         pagevec_init(&pvec);
1741         index = 0;
1742
1743         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1744                                 PAGECACHE_TAG_DIRTY))) {
1745                 int i;
1746
1747                 for (i = 0; i < nr_pages; i++) {
1748                         struct page *page = pvec.pages[i];
1749                         bool submitted = false;
1750
1751                         if (unlikely(f2fs_cp_error(sbi))) {
1752                                 f2fs_put_page(last_page, 0);
1753                                 pagevec_release(&pvec);
1754                                 ret = -EIO;
1755                                 goto out;
1756                         }
1757
1758                         if (!IS_DNODE(page) || !is_cold_node(page))
1759                                 continue;
1760                         if (ino_of_node(page) != ino)
1761                                 continue;
1762
1763                         lock_page(page);
1764
1765                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1766 continue_unlock:
1767                                 unlock_page(page);
1768                                 continue;
1769                         }
1770                         if (ino_of_node(page) != ino)
1771                                 goto continue_unlock;
1772
1773                         if (!PageDirty(page) && page != last_page) {
1774                                 /* someone wrote it for us */
1775                                 goto continue_unlock;
1776                         }
1777
1778                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1779
1780                         set_fsync_mark(page, 0);
1781                         set_dentry_mark(page, 0);
1782
1783                         if (!atomic || page == last_page) {
1784                                 set_fsync_mark(page, 1);
1785                                 percpu_counter_inc(&sbi->rf_node_block_count);
1786                                 if (IS_INODE(page)) {
1787                                         if (is_inode_flag_set(inode,
1788                                                                 FI_DIRTY_INODE))
1789                                                 f2fs_update_inode(inode, page);
1790                                         set_dentry_mark(page,
1791                                                 f2fs_need_dentry_mark(sbi, ino));
1792                                 }
1793                                 /* may be written by other thread */
1794                                 if (!PageDirty(page))
1795                                         set_page_dirty(page);
1796                         }
1797
1798                         if (!clear_page_dirty_for_io(page))
1799                                 goto continue_unlock;
1800
1801                         ret = __write_node_page(page, atomic &&
1802                                                 page == last_page,
1803                                                 &submitted, wbc, true,
1804                                                 FS_NODE_IO, seq_id);
1805                         if (ret) {
1806                                 unlock_page(page);
1807                                 f2fs_put_page(last_page, 0);
1808                                 break;
1809                         } else if (submitted) {
1810                                 nwritten++;
1811                         }
1812
1813                         if (page == last_page) {
1814                                 f2fs_put_page(page, 0);
1815                                 marked = true;
1816                                 break;
1817                         }
1818                 }
1819                 pagevec_release(&pvec);
1820                 cond_resched();
1821
1822                 if (ret || marked)
1823                         break;
1824         }
1825         if (!ret && atomic && !marked) {
1826                 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1827                            ino, last_page->index);
1828                 lock_page(last_page);
1829                 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1830                 set_page_dirty(last_page);
1831                 unlock_page(last_page);
1832                 goto retry;
1833         }
1834 out:
1835         if (nwritten)
1836                 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1837         return ret ? -EIO : 0;
1838 }
1839
1840 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1841 {
1842         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1843         bool clean;
1844
1845         if (inode->i_ino != ino)
1846                 return 0;
1847
1848         if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1849                 return 0;
1850
1851         spin_lock(&sbi->inode_lock[DIRTY_META]);
1852         clean = list_empty(&F2FS_I(inode)->gdirty_list);
1853         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1854
1855         if (clean)
1856                 return 0;
1857
1858         inode = igrab(inode);
1859         if (!inode)
1860                 return 0;
1861         return 1;
1862 }
1863
1864 static bool flush_dirty_inode(struct page *page)
1865 {
1866         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1867         struct inode *inode;
1868         nid_t ino = ino_of_node(page);
1869
1870         inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1871         if (!inode)
1872                 return false;
1873
1874         f2fs_update_inode(inode, page);
1875         unlock_page(page);
1876
1877         iput(inode);
1878         return true;
1879 }
1880
1881 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1882 {
1883         pgoff_t index = 0;
1884         struct pagevec pvec;
1885         int nr_pages;
1886
1887         pagevec_init(&pvec);
1888
1889         while ((nr_pages = pagevec_lookup_tag(&pvec,
1890                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1891                 int i;
1892
1893                 for (i = 0; i < nr_pages; i++) {
1894                         struct page *page = pvec.pages[i];
1895
1896                         if (!IS_DNODE(page))
1897                                 continue;
1898
1899                         lock_page(page);
1900
1901                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1902 continue_unlock:
1903                                 unlock_page(page);
1904                                 continue;
1905                         }
1906
1907                         if (!PageDirty(page)) {
1908                                 /* someone wrote it for us */
1909                                 goto continue_unlock;
1910                         }
1911
1912                         /* flush inline_data, if it's async context. */
1913                         if (page_private_inline(page)) {
1914                                 clear_page_private_inline(page);
1915                                 unlock_page(page);
1916                                 flush_inline_data(sbi, ino_of_node(page));
1917                                 continue;
1918                         }
1919                         unlock_page(page);
1920                 }
1921                 pagevec_release(&pvec);
1922                 cond_resched();
1923         }
1924 }
1925
1926 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1927                                 struct writeback_control *wbc,
1928                                 bool do_balance, enum iostat_type io_type)
1929 {
1930         pgoff_t index;
1931         struct pagevec pvec;
1932         int step = 0;
1933         int nwritten = 0;
1934         int ret = 0;
1935         int nr_pages, done = 0;
1936
1937         pagevec_init(&pvec);
1938
1939 next_step:
1940         index = 0;
1941
1942         while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1943                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1944                 int i;
1945
1946                 for (i = 0; i < nr_pages; i++) {
1947                         struct page *page = pvec.pages[i];
1948                         bool submitted = false;
1949                         bool may_dirty = true;
1950
1951                         /* give a priority to WB_SYNC threads */
1952                         if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1953                                         wbc->sync_mode == WB_SYNC_NONE) {
1954                                 done = 1;
1955                                 break;
1956                         }
1957
1958                         /*
1959                          * flushing sequence with step:
1960                          * 0. indirect nodes
1961                          * 1. dentry dnodes
1962                          * 2. file dnodes
1963                          */
1964                         if (step == 0 && IS_DNODE(page))
1965                                 continue;
1966                         if (step == 1 && (!IS_DNODE(page) ||
1967                                                 is_cold_node(page)))
1968                                 continue;
1969                         if (step == 2 && (!IS_DNODE(page) ||
1970                                                 !is_cold_node(page)))
1971                                 continue;
1972 lock_node:
1973                         if (wbc->sync_mode == WB_SYNC_ALL)
1974                                 lock_page(page);
1975                         else if (!trylock_page(page))
1976                                 continue;
1977
1978                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1979 continue_unlock:
1980                                 unlock_page(page);
1981                                 continue;
1982                         }
1983
1984                         if (!PageDirty(page)) {
1985                                 /* someone wrote it for us */
1986                                 goto continue_unlock;
1987                         }
1988
1989                         /* flush inline_data/inode, if it's async context. */
1990                         if (!do_balance)
1991                                 goto write_node;
1992
1993                         /* flush inline_data */
1994                         if (page_private_inline(page)) {
1995                                 clear_page_private_inline(page);
1996                                 unlock_page(page);
1997                                 flush_inline_data(sbi, ino_of_node(page));
1998                                 goto lock_node;
1999                         }
2000
2001                         /* flush dirty inode */
2002                         if (IS_INODE(page) && may_dirty) {
2003                                 may_dirty = false;
2004                                 if (flush_dirty_inode(page))
2005                                         goto lock_node;
2006                         }
2007 write_node:
2008                         f2fs_wait_on_page_writeback(page, NODE, true, true);
2009
2010                         if (!clear_page_dirty_for_io(page))
2011                                 goto continue_unlock;
2012
2013                         set_fsync_mark(page, 0);
2014                         set_dentry_mark(page, 0);
2015
2016                         ret = __write_node_page(page, false, &submitted,
2017                                                 wbc, do_balance, io_type, NULL);
2018                         if (ret)
2019                                 unlock_page(page);
2020                         else if (submitted)
2021                                 nwritten++;
2022
2023                         if (--wbc->nr_to_write == 0)
2024                                 break;
2025                 }
2026                 pagevec_release(&pvec);
2027                 cond_resched();
2028
2029                 if (wbc->nr_to_write == 0) {
2030                         step = 2;
2031                         break;
2032                 }
2033         }
2034
2035         if (step < 2) {
2036                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2037                                 wbc->sync_mode == WB_SYNC_NONE && step == 1)
2038                         goto out;
2039                 step++;
2040                 goto next_step;
2041         }
2042 out:
2043         if (nwritten)
2044                 f2fs_submit_merged_write(sbi, NODE);
2045
2046         if (unlikely(f2fs_cp_error(sbi)))
2047                 return -EIO;
2048         return ret;
2049 }
2050
2051 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2052                                                 unsigned int seq_id)
2053 {
2054         struct fsync_node_entry *fn;
2055         struct page *page;
2056         struct list_head *head = &sbi->fsync_node_list;
2057         unsigned long flags;
2058         unsigned int cur_seq_id = 0;
2059         int ret2, ret = 0;
2060
2061         while (seq_id && cur_seq_id < seq_id) {
2062                 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2063                 if (list_empty(head)) {
2064                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2065                         break;
2066                 }
2067                 fn = list_first_entry(head, struct fsync_node_entry, list);
2068                 if (fn->seq_id > seq_id) {
2069                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2070                         break;
2071                 }
2072                 cur_seq_id = fn->seq_id;
2073                 page = fn->page;
2074                 get_page(page);
2075                 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2076
2077                 f2fs_wait_on_page_writeback(page, NODE, true, false);
2078                 if (TestClearPageError(page))
2079                         ret = -EIO;
2080
2081                 put_page(page);
2082
2083                 if (ret)
2084                         break;
2085         }
2086
2087         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2088         if (!ret)
2089                 ret = ret2;
2090
2091         return ret;
2092 }
2093
2094 static int f2fs_write_node_pages(struct address_space *mapping,
2095                             struct writeback_control *wbc)
2096 {
2097         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2098         struct blk_plug plug;
2099         long diff;
2100
2101         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2102                 goto skip_write;
2103
2104         /* balancing f2fs's metadata in background */
2105         f2fs_balance_fs_bg(sbi, true);
2106
2107         /* collect a number of dirty node pages and write together */
2108         if (wbc->sync_mode != WB_SYNC_ALL &&
2109                         get_pages(sbi, F2FS_DIRTY_NODES) <
2110                                         nr_pages_to_skip(sbi, NODE))
2111                 goto skip_write;
2112
2113         if (wbc->sync_mode == WB_SYNC_ALL)
2114                 atomic_inc(&sbi->wb_sync_req[NODE]);
2115         else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2116                 /* to avoid potential deadlock */
2117                 if (current->plug)
2118                         blk_finish_plug(current->plug);
2119                 goto skip_write;
2120         }
2121
2122         trace_f2fs_writepages(mapping->host, wbc, NODE);
2123
2124         diff = nr_pages_to_write(sbi, NODE, wbc);
2125         blk_start_plug(&plug);
2126         f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2127         blk_finish_plug(&plug);
2128         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2129
2130         if (wbc->sync_mode == WB_SYNC_ALL)
2131                 atomic_dec(&sbi->wb_sync_req[NODE]);
2132         return 0;
2133
2134 skip_write:
2135         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2136         trace_f2fs_writepages(mapping->host, wbc, NODE);
2137         return 0;
2138 }
2139
2140 static bool f2fs_dirty_node_folio(struct address_space *mapping,
2141                 struct folio *folio)
2142 {
2143         trace_f2fs_set_page_dirty(&folio->page, NODE);
2144
2145         if (!folio_test_uptodate(folio))
2146                 folio_mark_uptodate(folio);
2147 #ifdef CONFIG_F2FS_CHECK_FS
2148         if (IS_INODE(&folio->page))
2149                 f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page);
2150 #endif
2151         if (!folio_test_dirty(folio)) {
2152                 filemap_dirty_folio(mapping, folio);
2153                 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2154                 set_page_private_reference(&folio->page);
2155                 return true;
2156         }
2157         return false;
2158 }
2159
2160 /*
2161  * Structure of the f2fs node operations
2162  */
2163 const struct address_space_operations f2fs_node_aops = {
2164         .writepage      = f2fs_write_node_page,
2165         .writepages     = f2fs_write_node_pages,
2166         .dirty_folio    = f2fs_dirty_node_folio,
2167         .invalidate_folio = f2fs_invalidate_folio,
2168         .releasepage    = f2fs_release_page,
2169 #ifdef CONFIG_MIGRATION
2170         .migratepage    = f2fs_migrate_page,
2171 #endif
2172 };
2173
2174 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2175                                                 nid_t n)
2176 {
2177         return radix_tree_lookup(&nm_i->free_nid_root, n);
2178 }
2179
2180 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2181                                 struct free_nid *i)
2182 {
2183         struct f2fs_nm_info *nm_i = NM_I(sbi);
2184         int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2185
2186         if (err)
2187                 return err;
2188
2189         nm_i->nid_cnt[FREE_NID]++;
2190         list_add_tail(&i->list, &nm_i->free_nid_list);
2191         return 0;
2192 }
2193
2194 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2195                         struct free_nid *i, enum nid_state state)
2196 {
2197         struct f2fs_nm_info *nm_i = NM_I(sbi);
2198
2199         f2fs_bug_on(sbi, state != i->state);
2200         nm_i->nid_cnt[state]--;
2201         if (state == FREE_NID)
2202                 list_del(&i->list);
2203         radix_tree_delete(&nm_i->free_nid_root, i->nid);
2204 }
2205
2206 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2207                         enum nid_state org_state, enum nid_state dst_state)
2208 {
2209         struct f2fs_nm_info *nm_i = NM_I(sbi);
2210
2211         f2fs_bug_on(sbi, org_state != i->state);
2212         i->state = dst_state;
2213         nm_i->nid_cnt[org_state]--;
2214         nm_i->nid_cnt[dst_state]++;
2215
2216         switch (dst_state) {
2217         case PREALLOC_NID:
2218                 list_del(&i->list);
2219                 break;
2220         case FREE_NID:
2221                 list_add_tail(&i->list, &nm_i->free_nid_list);
2222                 break;
2223         default:
2224                 BUG_ON(1);
2225         }
2226 }
2227
2228 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2229 {
2230         struct f2fs_nm_info *nm_i = NM_I(sbi);
2231         unsigned int i;
2232         bool ret = true;
2233
2234         f2fs_down_read(&nm_i->nat_tree_lock);
2235         for (i = 0; i < nm_i->nat_blocks; i++) {
2236                 if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2237                         ret = false;
2238                         break;
2239                 }
2240         }
2241         f2fs_up_read(&nm_i->nat_tree_lock);
2242
2243         return ret;
2244 }
2245
2246 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2247                                                         bool set, bool build)
2248 {
2249         struct f2fs_nm_info *nm_i = NM_I(sbi);
2250         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2251         unsigned int nid_ofs = nid - START_NID(nid);
2252
2253         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2254                 return;
2255
2256         if (set) {
2257                 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2258                         return;
2259                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2260                 nm_i->free_nid_count[nat_ofs]++;
2261         } else {
2262                 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2263                         return;
2264                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2265                 if (!build)
2266                         nm_i->free_nid_count[nat_ofs]--;
2267         }
2268 }
2269
2270 /* return if the nid is recognized as free */
2271 static bool add_free_nid(struct f2fs_sb_info *sbi,
2272                                 nid_t nid, bool build, bool update)
2273 {
2274         struct f2fs_nm_info *nm_i = NM_I(sbi);
2275         struct free_nid *i, *e;
2276         struct nat_entry *ne;
2277         int err = -EINVAL;
2278         bool ret = false;
2279
2280         /* 0 nid should not be used */
2281         if (unlikely(nid == 0))
2282                 return false;
2283
2284         if (unlikely(f2fs_check_nid_range(sbi, nid)))
2285                 return false;
2286
2287         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2288         i->nid = nid;
2289         i->state = FREE_NID;
2290
2291         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2292
2293         spin_lock(&nm_i->nid_list_lock);
2294
2295         if (build) {
2296                 /*
2297                  *   Thread A             Thread B
2298                  *  - f2fs_create
2299                  *   - f2fs_new_inode
2300                  *    - f2fs_alloc_nid
2301                  *     - __insert_nid_to_list(PREALLOC_NID)
2302                  *                     - f2fs_balance_fs_bg
2303                  *                      - f2fs_build_free_nids
2304                  *                       - __f2fs_build_free_nids
2305                  *                        - scan_nat_page
2306                  *                         - add_free_nid
2307                  *                          - __lookup_nat_cache
2308                  *  - f2fs_add_link
2309                  *   - f2fs_init_inode_metadata
2310                  *    - f2fs_new_inode_page
2311                  *     - f2fs_new_node_page
2312                  *      - set_node_addr
2313                  *  - f2fs_alloc_nid_done
2314                  *   - __remove_nid_from_list(PREALLOC_NID)
2315                  *                         - __insert_nid_to_list(FREE_NID)
2316                  */
2317                 ne = __lookup_nat_cache(nm_i, nid);
2318                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2319                                 nat_get_blkaddr(ne) != NULL_ADDR))
2320                         goto err_out;
2321
2322                 e = __lookup_free_nid_list(nm_i, nid);
2323                 if (e) {
2324                         if (e->state == FREE_NID)
2325                                 ret = true;
2326                         goto err_out;
2327                 }
2328         }
2329         ret = true;
2330         err = __insert_free_nid(sbi, i);
2331 err_out:
2332         if (update) {
2333                 update_free_nid_bitmap(sbi, nid, ret, build);
2334                 if (!build)
2335                         nm_i->available_nids++;
2336         }
2337         spin_unlock(&nm_i->nid_list_lock);
2338         radix_tree_preload_end();
2339
2340         if (err)
2341                 kmem_cache_free(free_nid_slab, i);
2342         return ret;
2343 }
2344
2345 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2346 {
2347         struct f2fs_nm_info *nm_i = NM_I(sbi);
2348         struct free_nid *i;
2349         bool need_free = false;
2350
2351         spin_lock(&nm_i->nid_list_lock);
2352         i = __lookup_free_nid_list(nm_i, nid);
2353         if (i && i->state == FREE_NID) {
2354                 __remove_free_nid(sbi, i, FREE_NID);
2355                 need_free = true;
2356         }
2357         spin_unlock(&nm_i->nid_list_lock);
2358
2359         if (need_free)
2360                 kmem_cache_free(free_nid_slab, i);
2361 }
2362
2363 static int scan_nat_page(struct f2fs_sb_info *sbi,
2364                         struct page *nat_page, nid_t start_nid)
2365 {
2366         struct f2fs_nm_info *nm_i = NM_I(sbi);
2367         struct f2fs_nat_block *nat_blk = page_address(nat_page);
2368         block_t blk_addr;
2369         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2370         int i;
2371
2372         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2373
2374         i = start_nid % NAT_ENTRY_PER_BLOCK;
2375
2376         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2377                 if (unlikely(start_nid >= nm_i->max_nid))
2378                         break;
2379
2380                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2381
2382                 if (blk_addr == NEW_ADDR)
2383                         return -EINVAL;
2384
2385                 if (blk_addr == NULL_ADDR) {
2386                         add_free_nid(sbi, start_nid, true, true);
2387                 } else {
2388                         spin_lock(&NM_I(sbi)->nid_list_lock);
2389                         update_free_nid_bitmap(sbi, start_nid, false, true);
2390                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2391                 }
2392         }
2393
2394         return 0;
2395 }
2396
2397 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2398 {
2399         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2400         struct f2fs_journal *journal = curseg->journal;
2401         int i;
2402
2403         down_read(&curseg->journal_rwsem);
2404         for (i = 0; i < nats_in_cursum(journal); i++) {
2405                 block_t addr;
2406                 nid_t nid;
2407
2408                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2409                 nid = le32_to_cpu(nid_in_journal(journal, i));
2410                 if (addr == NULL_ADDR)
2411                         add_free_nid(sbi, nid, true, false);
2412                 else
2413                         remove_free_nid(sbi, nid);
2414         }
2415         up_read(&curseg->journal_rwsem);
2416 }
2417
2418 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2419 {
2420         struct f2fs_nm_info *nm_i = NM_I(sbi);
2421         unsigned int i, idx;
2422         nid_t nid;
2423
2424         f2fs_down_read(&nm_i->nat_tree_lock);
2425
2426         for (i = 0; i < nm_i->nat_blocks; i++) {
2427                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2428                         continue;
2429                 if (!nm_i->free_nid_count[i])
2430                         continue;
2431                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2432                         idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2433                                                 NAT_ENTRY_PER_BLOCK, idx);
2434                         if (idx >= NAT_ENTRY_PER_BLOCK)
2435                                 break;
2436
2437                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
2438                         add_free_nid(sbi, nid, true, false);
2439
2440                         if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2441                                 goto out;
2442                 }
2443         }
2444 out:
2445         scan_curseg_cache(sbi);
2446
2447         f2fs_up_read(&nm_i->nat_tree_lock);
2448 }
2449
2450 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2451                                                 bool sync, bool mount)
2452 {
2453         struct f2fs_nm_info *nm_i = NM_I(sbi);
2454         int i = 0, ret;
2455         nid_t nid = nm_i->next_scan_nid;
2456
2457         if (unlikely(nid >= nm_i->max_nid))
2458                 nid = 0;
2459
2460         if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2461                 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2462
2463         /* Enough entries */
2464         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2465                 return 0;
2466
2467         if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2468                 return 0;
2469
2470         if (!mount) {
2471                 /* try to find free nids in free_nid_bitmap */
2472                 scan_free_nid_bits(sbi);
2473
2474                 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2475                         return 0;
2476         }
2477
2478         /* readahead nat pages to be scanned */
2479         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2480                                                         META_NAT, true);
2481
2482         f2fs_down_read(&nm_i->nat_tree_lock);
2483
2484         while (1) {
2485                 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2486                                                 nm_i->nat_block_bitmap)) {
2487                         struct page *page = get_current_nat_page(sbi, nid);
2488
2489                         if (IS_ERR(page)) {
2490                                 ret = PTR_ERR(page);
2491                         } else {
2492                                 ret = scan_nat_page(sbi, page, nid);
2493                                 f2fs_put_page(page, 1);
2494                         }
2495
2496                         if (ret) {
2497                                 f2fs_up_read(&nm_i->nat_tree_lock);
2498                                 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2499                                 return ret;
2500                         }
2501                 }
2502
2503                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2504                 if (unlikely(nid >= nm_i->max_nid))
2505                         nid = 0;
2506
2507                 if (++i >= FREE_NID_PAGES)
2508                         break;
2509         }
2510
2511         /* go to the next free nat pages to find free nids abundantly */
2512         nm_i->next_scan_nid = nid;
2513
2514         /* find free nids from current sum_pages */
2515         scan_curseg_cache(sbi);
2516
2517         f2fs_up_read(&nm_i->nat_tree_lock);
2518
2519         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2520                                         nm_i->ra_nid_pages, META_NAT, false);
2521
2522         return 0;
2523 }
2524
2525 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2526 {
2527         int ret;
2528
2529         mutex_lock(&NM_I(sbi)->build_lock);
2530         ret = __f2fs_build_free_nids(sbi, sync, mount);
2531         mutex_unlock(&NM_I(sbi)->build_lock);
2532
2533         return ret;
2534 }
2535
2536 /*
2537  * If this function returns success, caller can obtain a new nid
2538  * from second parameter of this function.
2539  * The returned nid could be used ino as well as nid when inode is created.
2540  */
2541 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2542 {
2543         struct f2fs_nm_info *nm_i = NM_I(sbi);
2544         struct free_nid *i = NULL;
2545 retry:
2546         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2547                 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2548                 return false;
2549         }
2550
2551         spin_lock(&nm_i->nid_list_lock);
2552
2553         if (unlikely(nm_i->available_nids == 0)) {
2554                 spin_unlock(&nm_i->nid_list_lock);
2555                 return false;
2556         }
2557
2558         /* We should not use stale free nids created by f2fs_build_free_nids */
2559         if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2560                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2561                 i = list_first_entry(&nm_i->free_nid_list,
2562                                         struct free_nid, list);
2563                 *nid = i->nid;
2564
2565                 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2566                 nm_i->available_nids--;
2567
2568                 update_free_nid_bitmap(sbi, *nid, false, false);
2569
2570                 spin_unlock(&nm_i->nid_list_lock);
2571                 return true;
2572         }
2573         spin_unlock(&nm_i->nid_list_lock);
2574
2575         /* Let's scan nat pages and its caches to get free nids */
2576         if (!f2fs_build_free_nids(sbi, true, false))
2577                 goto retry;
2578         return false;
2579 }
2580
2581 /*
2582  * f2fs_alloc_nid() should be called prior to this function.
2583  */
2584 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2585 {
2586         struct f2fs_nm_info *nm_i = NM_I(sbi);
2587         struct free_nid *i;
2588
2589         spin_lock(&nm_i->nid_list_lock);
2590         i = __lookup_free_nid_list(nm_i, nid);
2591         f2fs_bug_on(sbi, !i);
2592         __remove_free_nid(sbi, i, PREALLOC_NID);
2593         spin_unlock(&nm_i->nid_list_lock);
2594
2595         kmem_cache_free(free_nid_slab, i);
2596 }
2597
2598 /*
2599  * f2fs_alloc_nid() should be called prior to this function.
2600  */
2601 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2602 {
2603         struct f2fs_nm_info *nm_i = NM_I(sbi);
2604         struct free_nid *i;
2605         bool need_free = false;
2606
2607         if (!nid)
2608                 return;
2609
2610         spin_lock(&nm_i->nid_list_lock);
2611         i = __lookup_free_nid_list(nm_i, nid);
2612         f2fs_bug_on(sbi, !i);
2613
2614         if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2615                 __remove_free_nid(sbi, i, PREALLOC_NID);
2616                 need_free = true;
2617         } else {
2618                 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2619         }
2620
2621         nm_i->available_nids++;
2622
2623         update_free_nid_bitmap(sbi, nid, true, false);
2624
2625         spin_unlock(&nm_i->nid_list_lock);
2626
2627         if (need_free)
2628                 kmem_cache_free(free_nid_slab, i);
2629 }
2630
2631 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2632 {
2633         struct f2fs_nm_info *nm_i = NM_I(sbi);
2634         int nr = nr_shrink;
2635
2636         if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2637                 return 0;
2638
2639         if (!mutex_trylock(&nm_i->build_lock))
2640                 return 0;
2641
2642         while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2643                 struct free_nid *i, *next;
2644                 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2645
2646                 spin_lock(&nm_i->nid_list_lock);
2647                 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2648                         if (!nr_shrink || !batch ||
2649                                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2650                                 break;
2651                         __remove_free_nid(sbi, i, FREE_NID);
2652                         kmem_cache_free(free_nid_slab, i);
2653                         nr_shrink--;
2654                         batch--;
2655                 }
2656                 spin_unlock(&nm_i->nid_list_lock);
2657         }
2658
2659         mutex_unlock(&nm_i->build_lock);
2660
2661         return nr - nr_shrink;
2662 }
2663
2664 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2665 {
2666         void *src_addr, *dst_addr;
2667         size_t inline_size;
2668         struct page *ipage;
2669         struct f2fs_inode *ri;
2670
2671         ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2672         if (IS_ERR(ipage))
2673                 return PTR_ERR(ipage);
2674
2675         ri = F2FS_INODE(page);
2676         if (ri->i_inline & F2FS_INLINE_XATTR) {
2677                 if (!f2fs_has_inline_xattr(inode)) {
2678                         set_inode_flag(inode, FI_INLINE_XATTR);
2679                         stat_inc_inline_xattr(inode);
2680                 }
2681         } else {
2682                 if (f2fs_has_inline_xattr(inode)) {
2683                         stat_dec_inline_xattr(inode);
2684                         clear_inode_flag(inode, FI_INLINE_XATTR);
2685                 }
2686                 goto update_inode;
2687         }
2688
2689         dst_addr = inline_xattr_addr(inode, ipage);
2690         src_addr = inline_xattr_addr(inode, page);
2691         inline_size = inline_xattr_size(inode);
2692
2693         f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2694         memcpy(dst_addr, src_addr, inline_size);
2695 update_inode:
2696         f2fs_update_inode(inode, ipage);
2697         f2fs_put_page(ipage, 1);
2698         return 0;
2699 }
2700
2701 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2702 {
2703         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2704         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2705         nid_t new_xnid;
2706         struct dnode_of_data dn;
2707         struct node_info ni;
2708         struct page *xpage;
2709         int err;
2710
2711         if (!prev_xnid)
2712                 goto recover_xnid;
2713
2714         /* 1: invalidate the previous xattr nid */
2715         err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2716         if (err)
2717                 return err;
2718
2719         f2fs_invalidate_blocks(sbi, ni.blk_addr);
2720         dec_valid_node_count(sbi, inode, false);
2721         set_node_addr(sbi, &ni, NULL_ADDR, false);
2722
2723 recover_xnid:
2724         /* 2: update xattr nid in inode */
2725         if (!f2fs_alloc_nid(sbi, &new_xnid))
2726                 return -ENOSPC;
2727
2728         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2729         xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2730         if (IS_ERR(xpage)) {
2731                 f2fs_alloc_nid_failed(sbi, new_xnid);
2732                 return PTR_ERR(xpage);
2733         }
2734
2735         f2fs_alloc_nid_done(sbi, new_xnid);
2736         f2fs_update_inode_page(inode);
2737
2738         /* 3: update and set xattr node page dirty */
2739         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2740
2741         set_page_dirty(xpage);
2742         f2fs_put_page(xpage, 1);
2743
2744         return 0;
2745 }
2746
2747 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2748 {
2749         struct f2fs_inode *src, *dst;
2750         nid_t ino = ino_of_node(page);
2751         struct node_info old_ni, new_ni;
2752         struct page *ipage;
2753         int err;
2754
2755         err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2756         if (err)
2757                 return err;
2758
2759         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2760                 return -EINVAL;
2761 retry:
2762         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2763         if (!ipage) {
2764                 memalloc_retry_wait(GFP_NOFS);
2765                 goto retry;
2766         }
2767
2768         /* Should not use this inode from free nid list */
2769         remove_free_nid(sbi, ino);
2770
2771         if (!PageUptodate(ipage))
2772                 SetPageUptodate(ipage);
2773         fill_node_footer(ipage, ino, ino, 0, true);
2774         set_cold_node(ipage, false);
2775
2776         src = F2FS_INODE(page);
2777         dst = F2FS_INODE(ipage);
2778
2779         memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2780         dst->i_size = 0;
2781         dst->i_blocks = cpu_to_le64(1);
2782         dst->i_links = cpu_to_le32(1);
2783         dst->i_xattr_nid = 0;
2784         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2785         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2786                 dst->i_extra_isize = src->i_extra_isize;
2787
2788                 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2789                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2790                                                         i_inline_xattr_size))
2791                         dst->i_inline_xattr_size = src->i_inline_xattr_size;
2792
2793                 if (f2fs_sb_has_project_quota(sbi) &&
2794                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2795                                                                 i_projid))
2796                         dst->i_projid = src->i_projid;
2797
2798                 if (f2fs_sb_has_inode_crtime(sbi) &&
2799                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2800                                                         i_crtime_nsec)) {
2801                         dst->i_crtime = src->i_crtime;
2802                         dst->i_crtime_nsec = src->i_crtime_nsec;
2803                 }
2804         }
2805
2806         new_ni = old_ni;
2807         new_ni.ino = ino;
2808
2809         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2810                 WARN_ON(1);
2811         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2812         inc_valid_inode_count(sbi);
2813         set_page_dirty(ipage);
2814         f2fs_put_page(ipage, 1);
2815         return 0;
2816 }
2817
2818 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2819                         unsigned int segno, struct f2fs_summary_block *sum)
2820 {
2821         struct f2fs_node *rn;
2822         struct f2fs_summary *sum_entry;
2823         block_t addr;
2824         int i, idx, last_offset, nrpages;
2825
2826         /* scan the node segment */
2827         last_offset = sbi->blocks_per_seg;
2828         addr = START_BLOCK(sbi, segno);
2829         sum_entry = &sum->entries[0];
2830
2831         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2832                 nrpages = bio_max_segs(last_offset - i);
2833
2834                 /* readahead node pages */
2835                 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2836
2837                 for (idx = addr; idx < addr + nrpages; idx++) {
2838                         struct page *page = f2fs_get_tmp_page(sbi, idx);
2839
2840                         if (IS_ERR(page))
2841                                 return PTR_ERR(page);
2842
2843                         rn = F2FS_NODE(page);
2844                         sum_entry->nid = rn->footer.nid;
2845                         sum_entry->version = 0;
2846                         sum_entry->ofs_in_node = 0;
2847                         sum_entry++;
2848                         f2fs_put_page(page, 1);
2849                 }
2850
2851                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2852                                                         addr + nrpages);
2853         }
2854         return 0;
2855 }
2856
2857 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2858 {
2859         struct f2fs_nm_info *nm_i = NM_I(sbi);
2860         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2861         struct f2fs_journal *journal = curseg->journal;
2862         int i;
2863
2864         down_write(&curseg->journal_rwsem);
2865         for (i = 0; i < nats_in_cursum(journal); i++) {
2866                 struct nat_entry *ne;
2867                 struct f2fs_nat_entry raw_ne;
2868                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2869
2870                 if (f2fs_check_nid_range(sbi, nid))
2871                         continue;
2872
2873                 raw_ne = nat_in_journal(journal, i);
2874
2875                 ne = __lookup_nat_cache(nm_i, nid);
2876                 if (!ne) {
2877                         ne = __alloc_nat_entry(sbi, nid, true);
2878                         __init_nat_entry(nm_i, ne, &raw_ne, true);
2879                 }
2880
2881                 /*
2882                  * if a free nat in journal has not been used after last
2883                  * checkpoint, we should remove it from available nids,
2884                  * since later we will add it again.
2885                  */
2886                 if (!get_nat_flag(ne, IS_DIRTY) &&
2887                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2888                         spin_lock(&nm_i->nid_list_lock);
2889                         nm_i->available_nids--;
2890                         spin_unlock(&nm_i->nid_list_lock);
2891                 }
2892
2893                 __set_nat_cache_dirty(nm_i, ne);
2894         }
2895         update_nats_in_cursum(journal, -i);
2896         up_write(&curseg->journal_rwsem);
2897 }
2898
2899 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2900                                                 struct list_head *head, int max)
2901 {
2902         struct nat_entry_set *cur;
2903
2904         if (nes->entry_cnt >= max)
2905                 goto add_out;
2906
2907         list_for_each_entry(cur, head, set_list) {
2908                 if (cur->entry_cnt >= nes->entry_cnt) {
2909                         list_add(&nes->set_list, cur->set_list.prev);
2910                         return;
2911                 }
2912         }
2913 add_out:
2914         list_add_tail(&nes->set_list, head);
2915 }
2916
2917 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2918                                                         unsigned int valid)
2919 {
2920         if (valid == 0) {
2921                 __set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2922                 __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2923                 return;
2924         }
2925
2926         __clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2927         if (valid == NAT_ENTRY_PER_BLOCK)
2928                 __set_bit_le(nat_ofs, nm_i->full_nat_bits);
2929         else
2930                 __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2931 }
2932
2933 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2934                                                 struct page *page)
2935 {
2936         struct f2fs_nm_info *nm_i = NM_I(sbi);
2937         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2938         struct f2fs_nat_block *nat_blk = page_address(page);
2939         int valid = 0;
2940         int i = 0;
2941
2942         if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2943                 return;
2944
2945         if (nat_index == 0) {
2946                 valid = 1;
2947                 i = 1;
2948         }
2949         for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2950                 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2951                         valid++;
2952         }
2953
2954         __update_nat_bits(nm_i, nat_index, valid);
2955 }
2956
2957 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2958 {
2959         struct f2fs_nm_info *nm_i = NM_I(sbi);
2960         unsigned int nat_ofs;
2961
2962         f2fs_down_read(&nm_i->nat_tree_lock);
2963
2964         for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
2965                 unsigned int valid = 0, nid_ofs = 0;
2966
2967                 /* handle nid zero due to it should never be used */
2968                 if (unlikely(nat_ofs == 0)) {
2969                         valid = 1;
2970                         nid_ofs = 1;
2971                 }
2972
2973                 for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
2974                         if (!test_bit_le(nid_ofs,
2975                                         nm_i->free_nid_bitmap[nat_ofs]))
2976                                 valid++;
2977                 }
2978
2979                 __update_nat_bits(nm_i, nat_ofs, valid);
2980         }
2981
2982         f2fs_up_read(&nm_i->nat_tree_lock);
2983 }
2984
2985 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2986                 struct nat_entry_set *set, struct cp_control *cpc)
2987 {
2988         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2989         struct f2fs_journal *journal = curseg->journal;
2990         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2991         bool to_journal = true;
2992         struct f2fs_nat_block *nat_blk;
2993         struct nat_entry *ne, *cur;
2994         struct page *page = NULL;
2995
2996         /*
2997          * there are two steps to flush nat entries:
2998          * #1, flush nat entries to journal in current hot data summary block.
2999          * #2, flush nat entries to nat page.
3000          */
3001         if ((cpc->reason & CP_UMOUNT) ||
3002                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3003                 to_journal = false;
3004
3005         if (to_journal) {
3006                 down_write(&curseg->journal_rwsem);
3007         } else {
3008                 page = get_next_nat_page(sbi, start_nid);
3009                 if (IS_ERR(page))
3010                         return PTR_ERR(page);
3011
3012                 nat_blk = page_address(page);
3013                 f2fs_bug_on(sbi, !nat_blk);
3014         }
3015
3016         /* flush dirty nats in nat entry set */
3017         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3018                 struct f2fs_nat_entry *raw_ne;
3019                 nid_t nid = nat_get_nid(ne);
3020                 int offset;
3021
3022                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3023
3024                 if (to_journal) {
3025                         offset = f2fs_lookup_journal_in_cursum(journal,
3026                                                         NAT_JOURNAL, nid, 1);
3027                         f2fs_bug_on(sbi, offset < 0);
3028                         raw_ne = &nat_in_journal(journal, offset);
3029                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
3030                 } else {
3031                         raw_ne = &nat_blk->entries[nid - start_nid];
3032                 }
3033                 raw_nat_from_node_info(raw_ne, &ne->ni);
3034                 nat_reset_flag(ne);
3035                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
3036                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
3037                         add_free_nid(sbi, nid, false, true);
3038                 } else {
3039                         spin_lock(&NM_I(sbi)->nid_list_lock);
3040                         update_free_nid_bitmap(sbi, nid, false, false);
3041                         spin_unlock(&NM_I(sbi)->nid_list_lock);
3042                 }
3043         }
3044
3045         if (to_journal) {
3046                 up_write(&curseg->journal_rwsem);
3047         } else {
3048                 update_nat_bits(sbi, start_nid, page);
3049                 f2fs_put_page(page, 1);
3050         }
3051
3052         /* Allow dirty nats by node block allocation in write_begin */
3053         if (!set->entry_cnt) {
3054                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3055                 kmem_cache_free(nat_entry_set_slab, set);
3056         }
3057         return 0;
3058 }
3059
3060 /*
3061  * This function is called during the checkpointing process.
3062  */
3063 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3064 {
3065         struct f2fs_nm_info *nm_i = NM_I(sbi);
3066         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3067         struct f2fs_journal *journal = curseg->journal;
3068         struct nat_entry_set *setvec[SETVEC_SIZE];
3069         struct nat_entry_set *set, *tmp;
3070         unsigned int found;
3071         nid_t set_idx = 0;
3072         LIST_HEAD(sets);
3073         int err = 0;
3074
3075         /*
3076          * during unmount, let's flush nat_bits before checking
3077          * nat_cnt[DIRTY_NAT].
3078          */
3079         if (cpc->reason & CP_UMOUNT) {
3080                 f2fs_down_write(&nm_i->nat_tree_lock);
3081                 remove_nats_in_journal(sbi);
3082                 f2fs_up_write(&nm_i->nat_tree_lock);
3083         }
3084
3085         if (!nm_i->nat_cnt[DIRTY_NAT])
3086                 return 0;
3087
3088         f2fs_down_write(&nm_i->nat_tree_lock);
3089
3090         /*
3091          * if there are no enough space in journal to store dirty nat
3092          * entries, remove all entries from journal and merge them
3093          * into nat entry set.
3094          */
3095         if (cpc->reason & CP_UMOUNT ||
3096                 !__has_cursum_space(journal,
3097                         nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3098                 remove_nats_in_journal(sbi);
3099
3100         while ((found = __gang_lookup_nat_set(nm_i,
3101                                         set_idx, SETVEC_SIZE, setvec))) {
3102                 unsigned idx;
3103
3104                 set_idx = setvec[found - 1]->set + 1;
3105                 for (idx = 0; idx < found; idx++)
3106                         __adjust_nat_entry_set(setvec[idx], &sets,
3107                                                 MAX_NAT_JENTRIES(journal));
3108         }
3109
3110         /* flush dirty nats in nat entry set */
3111         list_for_each_entry_safe(set, tmp, &sets, set_list) {
3112                 err = __flush_nat_entry_set(sbi, set, cpc);
3113                 if (err)
3114                         break;
3115         }
3116
3117         f2fs_up_write(&nm_i->nat_tree_lock);
3118         /* Allow dirty nats by node block allocation in write_begin */
3119
3120         return err;
3121 }
3122
3123 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3124 {
3125         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3126         struct f2fs_nm_info *nm_i = NM_I(sbi);
3127         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3128         unsigned int i;
3129         __u64 cp_ver = cur_cp_version(ckpt);
3130         block_t nat_bits_addr;
3131
3132         nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3133         nm_i->nat_bits = f2fs_kvzalloc(sbi,
3134                         nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3135         if (!nm_i->nat_bits)
3136                 return -ENOMEM;
3137
3138         nm_i->full_nat_bits = nm_i->nat_bits + 8;
3139         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3140
3141         if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3142                 return 0;
3143
3144         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3145                                                 nm_i->nat_bits_blocks;
3146         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3147                 struct page *page;
3148
3149                 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3150                 if (IS_ERR(page))
3151                         return PTR_ERR(page);
3152
3153                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3154                                         page_address(page), F2FS_BLKSIZE);
3155                 f2fs_put_page(page, 1);
3156         }
3157
3158         cp_ver |= (cur_cp_crc(ckpt) << 32);
3159         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3160                 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3161                 f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3162                         cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3163                 return 0;
3164         }
3165
3166         f2fs_notice(sbi, "Found nat_bits in checkpoint");
3167         return 0;
3168 }
3169
3170 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3171 {
3172         struct f2fs_nm_info *nm_i = NM_I(sbi);
3173         unsigned int i = 0;
3174         nid_t nid, last_nid;
3175
3176         if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3177                 return;
3178
3179         for (i = 0; i < nm_i->nat_blocks; i++) {
3180                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3181                 if (i >= nm_i->nat_blocks)
3182                         break;
3183
3184                 __set_bit_le(i, nm_i->nat_block_bitmap);
3185
3186                 nid = i * NAT_ENTRY_PER_BLOCK;
3187                 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3188
3189                 spin_lock(&NM_I(sbi)->nid_list_lock);
3190                 for (; nid < last_nid; nid++)
3191                         update_free_nid_bitmap(sbi, nid, true, true);
3192                 spin_unlock(&NM_I(sbi)->nid_list_lock);
3193         }
3194
3195         for (i = 0; i < nm_i->nat_blocks; i++) {
3196                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3197                 if (i >= nm_i->nat_blocks)
3198                         break;
3199
3200                 __set_bit_le(i, nm_i->nat_block_bitmap);
3201         }
3202 }
3203
3204 static int init_node_manager(struct f2fs_sb_info *sbi)
3205 {
3206         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3207         struct f2fs_nm_info *nm_i = NM_I(sbi);
3208         unsigned char *version_bitmap;
3209         unsigned int nat_segs;
3210         int err;
3211
3212         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3213
3214         /* segment_count_nat includes pair segment so divide to 2. */
3215         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3216         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3217         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3218
3219         /* not used nids: 0, node, meta, (and root counted as valid node) */
3220         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3221                                                 F2FS_RESERVED_NODE_NUM;
3222         nm_i->nid_cnt[FREE_NID] = 0;
3223         nm_i->nid_cnt[PREALLOC_NID] = 0;
3224         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3225         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3226         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3227         nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3228
3229         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3230         INIT_LIST_HEAD(&nm_i->free_nid_list);
3231         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3232         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3233         INIT_LIST_HEAD(&nm_i->nat_entries);
3234         spin_lock_init(&nm_i->nat_list_lock);
3235
3236         mutex_init(&nm_i->build_lock);
3237         spin_lock_init(&nm_i->nid_list_lock);
3238         init_f2fs_rwsem(&nm_i->nat_tree_lock);
3239
3240         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3241         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3242         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3243         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3244                                         GFP_KERNEL);
3245         if (!nm_i->nat_bitmap)
3246                 return -ENOMEM;
3247
3248         err = __get_nat_bitmaps(sbi);
3249         if (err)
3250                 return err;
3251
3252 #ifdef CONFIG_F2FS_CHECK_FS
3253         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3254                                         GFP_KERNEL);
3255         if (!nm_i->nat_bitmap_mir)
3256                 return -ENOMEM;
3257 #endif
3258
3259         return 0;
3260 }
3261
3262 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3263 {
3264         struct f2fs_nm_info *nm_i = NM_I(sbi);
3265         int i;
3266
3267         nm_i->free_nid_bitmap =
3268                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3269                                               nm_i->nat_blocks),
3270                               GFP_KERNEL);
3271         if (!nm_i->free_nid_bitmap)
3272                 return -ENOMEM;
3273
3274         for (i = 0; i < nm_i->nat_blocks; i++) {
3275                 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3276                         f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3277                 if (!nm_i->free_nid_bitmap[i])
3278                         return -ENOMEM;
3279         }
3280
3281         nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3282                                                                 GFP_KERNEL);
3283         if (!nm_i->nat_block_bitmap)
3284                 return -ENOMEM;
3285
3286         nm_i->free_nid_count =
3287                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3288                                               nm_i->nat_blocks),
3289                               GFP_KERNEL);
3290         if (!nm_i->free_nid_count)
3291                 return -ENOMEM;
3292         return 0;
3293 }
3294
3295 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3296 {
3297         int err;
3298
3299         sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3300                                                         GFP_KERNEL);
3301         if (!sbi->nm_info)
3302                 return -ENOMEM;
3303
3304         err = init_node_manager(sbi);
3305         if (err)
3306                 return err;
3307
3308         err = init_free_nid_cache(sbi);
3309         if (err)
3310                 return err;
3311
3312         /* load free nid status from nat_bits table */
3313         load_free_nid_bitmap(sbi);
3314
3315         return f2fs_build_free_nids(sbi, true, true);
3316 }
3317
3318 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3319 {
3320         struct f2fs_nm_info *nm_i = NM_I(sbi);
3321         struct free_nid *i, *next_i;
3322         struct nat_entry *natvec[NATVEC_SIZE];
3323         struct nat_entry_set *setvec[SETVEC_SIZE];
3324         nid_t nid = 0;
3325         unsigned int found;
3326
3327         if (!nm_i)
3328                 return;
3329
3330         /* destroy free nid list */
3331         spin_lock(&nm_i->nid_list_lock);
3332         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3333                 __remove_free_nid(sbi, i, FREE_NID);
3334                 spin_unlock(&nm_i->nid_list_lock);
3335                 kmem_cache_free(free_nid_slab, i);
3336                 spin_lock(&nm_i->nid_list_lock);
3337         }
3338         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3339         f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3340         f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3341         spin_unlock(&nm_i->nid_list_lock);
3342
3343         /* destroy nat cache */
3344         f2fs_down_write(&nm_i->nat_tree_lock);
3345         while ((found = __gang_lookup_nat_cache(nm_i,
3346                                         nid, NATVEC_SIZE, natvec))) {
3347                 unsigned idx;
3348
3349                 nid = nat_get_nid(natvec[found - 1]) + 1;
3350                 for (idx = 0; idx < found; idx++) {
3351                         spin_lock(&nm_i->nat_list_lock);
3352                         list_del(&natvec[idx]->list);
3353                         spin_unlock(&nm_i->nat_list_lock);
3354
3355                         __del_from_nat_cache(nm_i, natvec[idx]);
3356                 }
3357         }
3358         f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3359
3360         /* destroy nat set cache */
3361         nid = 0;
3362         while ((found = __gang_lookup_nat_set(nm_i,
3363                                         nid, SETVEC_SIZE, setvec))) {
3364                 unsigned idx;
3365
3366                 nid = setvec[found - 1]->set + 1;
3367                 for (idx = 0; idx < found; idx++) {
3368                         /* entry_cnt is not zero, when cp_error was occurred */
3369                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3370                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3371                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3372                 }
3373         }
3374         f2fs_up_write(&nm_i->nat_tree_lock);
3375
3376         kvfree(nm_i->nat_block_bitmap);
3377         if (nm_i->free_nid_bitmap) {
3378                 int i;
3379
3380                 for (i = 0; i < nm_i->nat_blocks; i++)
3381                         kvfree(nm_i->free_nid_bitmap[i]);
3382                 kvfree(nm_i->free_nid_bitmap);
3383         }
3384         kvfree(nm_i->free_nid_count);
3385
3386         kvfree(nm_i->nat_bitmap);
3387         kvfree(nm_i->nat_bits);
3388 #ifdef CONFIG_F2FS_CHECK_FS
3389         kvfree(nm_i->nat_bitmap_mir);
3390 #endif
3391         sbi->nm_info = NULL;
3392         kfree(nm_i);
3393 }
3394
3395 int __init f2fs_create_node_manager_caches(void)
3396 {
3397         nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3398                         sizeof(struct nat_entry));
3399         if (!nat_entry_slab)
3400                 goto fail;
3401
3402         free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3403                         sizeof(struct free_nid));
3404         if (!free_nid_slab)
3405                 goto destroy_nat_entry;
3406
3407         nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3408                         sizeof(struct nat_entry_set));
3409         if (!nat_entry_set_slab)
3410                 goto destroy_free_nid;
3411
3412         fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3413                         sizeof(struct fsync_node_entry));
3414         if (!fsync_node_entry_slab)
3415                 goto destroy_nat_entry_set;
3416         return 0;
3417
3418 destroy_nat_entry_set:
3419         kmem_cache_destroy(nat_entry_set_slab);
3420 destroy_free_nid:
3421         kmem_cache_destroy(free_nid_slab);
3422 destroy_nat_entry:
3423         kmem_cache_destroy(nat_entry_slab);
3424 fail:
3425         return -ENOMEM;
3426 }
3427
3428 void f2fs_destroy_node_manager_caches(void)
3429 {
3430         kmem_cache_destroy(fsync_node_entry_slab);
3431         kmem_cache_destroy(nat_entry_set_slab);
3432         kmem_cache_destroy(free_nid_slab);
3433         kmem_cache_destroy(nat_entry_slab);
3434 }