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(nm_i) mutex_is_locked(&(nm_i)->build_lock)
  24
  25 static struct kmem_cache *nat_entry_slab;
  26 static struct kmem_cache *free_nid_slab;
  27 static struct kmem_cache *nat_entry_set_slab;
  28 static struct kmem_cache *fsync_node_entry_slab;
  29
  30 /*
  31  * Check whether the given nid is within node id range.
  32  */
  33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
  34 {
  35         if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
  36                 set_sbi_flag(sbi, SBI_NEED_FSCK);
  37                 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
  38                           __func__, nid);
  39                 f2fs_handle_error(sbi, ERROR_CORRUPTED_INODE);
  40                 return -EFSCORRUPTED;
  41         }
  42         return 0;
  43 }
  44
  45 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
  46 {
  47         struct f2fs_nm_info *nm_i = NM_I(sbi);
  48         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
  49         struct sysinfo val;
  50         unsigned long avail_ram;
  51         unsigned long mem_size = 0;
  52         bool res = false;
  53
  54         if (!nm_i)
  55                 return true;
  56
  57         si_meminfo(&val);
  58
  59         /* only uses low memory */
  60         avail_ram = val.totalram - val.totalhigh;
  61
  62         /*
  63          * give 25%, 25%, 50%, 50%, 25%, 25% memory for each components respectively
  64          */
  65         if (type == FREE_NIDS) {
  66                 mem_size = (nm_i->nid_cnt[FREE_NID] *
  67                                 sizeof(struct free_nid)) >> PAGE_SHIFT;
  68                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  69         } else if (type == NAT_ENTRIES) {
  70                 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
  71                                 sizeof(struct nat_entry)) >> PAGE_SHIFT;
  72                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  73                 if (excess_cached_nats(sbi))
  74                         res = false;
  75         } else if (type == DIRTY_DENTS) {
  76                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  77                         return false;
  78                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  79                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  80         } else if (type == INO_ENTRIES) {
  81                 int i;
  82
  83                 for (i = 0; i < MAX_INO_ENTRY; i++)
  84                         mem_size += sbi->im[i].ino_num *
  85                                                 sizeof(struct ino_entry);
  86                 mem_size >>= PAGE_SHIFT;
  87                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  88         } else if (type == READ_EXTENT_CACHE || type == AGE_EXTENT_CACHE) {
  89                 enum extent_type etype = type == READ_EXTENT_CACHE ?
  90                                                 EX_READ : EX_BLOCK_AGE;
  91                 struct extent_tree_info *eti = &sbi->extent_tree[etype];
  92
  93                 mem_size = (atomic_read(&eti->total_ext_tree) *
  94                                 sizeof(struct extent_tree) +
  95                                 atomic_read(&eti->total_ext_node) *
  96                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  97                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  98         } else if (type == DISCARD_CACHE) {
  99                 mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
 100                                 sizeof(struct discard_cmd)) >> PAGE_SHIFT;
 101                 res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
 102         } else if (type == COMPRESS_PAGE) {
 103 #ifdef CONFIG_F2FS_FS_COMPRESSION
 104                 unsigned long free_ram = val.freeram;
 105
 106                 /*
 107                  * free memory is lower than watermark or cached page count
 108                  * exceed threshold, deny caching compress page.
 109                  */
 110                 res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
 111                         (COMPRESS_MAPPING(sbi)->nrpages <
 112                          free_ram * sbi->compress_percent / 100);
 113 #else
 114                 res = false;
 115 #endif
 116         } else {
 117                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
 118                         return true;
 119         }
 120         return res;
 121 }
 122
 123 static void clear_node_folio_dirty(struct folio *folio)
 124 {
 125         if (folio_test_dirty(folio)) {
 126                 f2fs_clear_page_cache_dirty_tag(folio);
 127                 folio_clear_dirty_for_io(folio);
 128                 dec_page_count(F2FS_F_SB(folio), F2FS_DIRTY_NODES);
 129         }
 130         folio_clear_uptodate(folio);
 131 }
 132
 133 static struct folio *get_current_nat_folio(struct f2fs_sb_info *sbi, nid_t nid)
 134 {
 135         return f2fs_get_meta_folio_retry(sbi, current_nat_addr(sbi, nid));
 136 }
 137
 138 static struct folio *get_next_nat_folio(struct f2fs_sb_info *sbi, nid_t nid)
 139 {
 140         struct folio *src_folio;
 141         struct folio *dst_folio;
 142         pgoff_t dst_off;
 143         void *src_addr;
 144         void *dst_addr;
 145         struct f2fs_nm_info *nm_i = NM_I(sbi);
 146
 147         dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
 148
 149         /* get current nat block page with lock */
 150         src_folio = get_current_nat_folio(sbi, nid);
 151         if (IS_ERR(src_folio))
 152                 return src_folio;
 153         dst_folio = f2fs_grab_meta_folio(sbi, dst_off);
 154         f2fs_bug_on(sbi, folio_test_dirty(src_folio));
 155
 156         src_addr = folio_address(src_folio);
 157         dst_addr = folio_address(dst_folio);
 158         memcpy(dst_addr, src_addr, PAGE_SIZE);
 159         folio_mark_dirty(dst_folio);
 160         f2fs_folio_put(src_folio, true);
 161
 162         set_to_next_nat(nm_i, nid);
 163
 164         return dst_folio;
 165 }
 166
 167 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
 168                                                 nid_t nid, bool no_fail)
 169 {
 170         struct nat_entry *new;
 171
 172         new = f2fs_kmem_cache_alloc(nat_entry_slab,
 173                                         GFP_F2FS_ZERO, no_fail, sbi);
 174         if (new) {
 175                 nat_set_nid(new, nid);
 176                 nat_reset_flag(new);
 177         }
 178         return new;
 179 }
 180
 181 static void __free_nat_entry(struct nat_entry *e)
 182 {
 183         kmem_cache_free(nat_entry_slab, e);
 184 }
 185
 186 /* must be locked by nat_tree_lock */
 187 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
 188         struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail, bool init_dirty)
 189 {
 190         if (no_fail)
 191                 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
 192         else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
 193                 return NULL;
 194
 195         if (raw_ne)
 196                 node_info_from_raw_nat(&ne->ni, raw_ne);
 197
 198         if (init_dirty) {
 199                 INIT_LIST_HEAD(&ne->list);
 200                 nm_i->nat_cnt[TOTAL_NAT]++;
 201                 return ne;
 202         }
 203
 204         spin_lock(&nm_i->nat_list_lock);
 205         list_add_tail(&ne->list, &nm_i->nat_entries);
 206         spin_unlock(&nm_i->nat_list_lock);
 207
 208         nm_i->nat_cnt[TOTAL_NAT]++;
 209         nm_i->nat_cnt[RECLAIMABLE_NAT]++;
 210         return ne;
 211 }
 212
 213 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n, bool for_dirty)
 214 {
 215         struct nat_entry *ne;
 216
 217         ne = radix_tree_lookup(&nm_i->nat_root, n);
 218
 219         /*
 220          * for recent accessed nat entry which will not be dirtied soon
 221          * later, move it to tail of lru list.
 222          */
 223         if (ne && !get_nat_flag(ne, IS_DIRTY) && !for_dirty) {
 224                 spin_lock(&nm_i->nat_list_lock);
 225                 if (!list_empty(&ne->list))
 226                         list_move_tail(&ne->list, &nm_i->nat_entries);
 227                 spin_unlock(&nm_i->nat_list_lock);
 228         }
 229
 230         return ne;
 231 }
 232
 233 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 234                 nid_t start, unsigned int nr, struct nat_entry **ep)
 235 {
 236         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 237 }
 238
 239 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 240 {
 241         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 242         nm_i->nat_cnt[TOTAL_NAT]--;
 243         nm_i->nat_cnt[RECLAIMABLE_NAT]--;
 244         __free_nat_entry(e);
 245 }
 246
 247 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
 248                                                         struct nat_entry *ne)
 249 {
 250         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 251         struct nat_entry_set *head;
 252
 253         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 254         if (!head) {
 255                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
 256                                                 GFP_NOFS, true, NULL);
 257
 258                 INIT_LIST_HEAD(&head->entry_list);
 259                 INIT_LIST_HEAD(&head->set_list);
 260                 head->set = set;
 261                 head->entry_cnt = 0;
 262                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 263         }
 264         return head;
 265 }
 266
 267 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 268                 struct nat_entry *ne, bool init_dirty)
 269 {
 270         struct nat_entry_set *head;
 271         bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
 272
 273         if (!new_ne)
 274                 head = __grab_nat_entry_set(nm_i, ne);
 275
 276         /*
 277          * update entry_cnt in below condition:
 278          * 1. update NEW_ADDR to valid block address;
 279          * 2. update old block address to new one;
 280          */
 281         if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
 282                                 !get_nat_flag(ne, IS_DIRTY)))
 283                 head->entry_cnt++;
 284
 285         set_nat_flag(ne, IS_PREALLOC, new_ne);
 286
 287         if (get_nat_flag(ne, IS_DIRTY))
 288                 goto refresh_list;
 289
 290         nm_i->nat_cnt[DIRTY_NAT]++;
 291         if (!init_dirty)
 292                 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
 293         set_nat_flag(ne, IS_DIRTY, true);
 294 refresh_list:
 295         spin_lock(&nm_i->nat_list_lock);
 296         if (new_ne)
 297                 list_del_init(&ne->list);
 298         else
 299                 list_move_tail(&ne->list, &head->entry_list);
 300         spin_unlock(&nm_i->nat_list_lock);
 301 }
 302
 303 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 304                 struct nat_entry_set *set, struct nat_entry *ne)
 305 {
 306         spin_lock(&nm_i->nat_list_lock);
 307         list_move_tail(&ne->list, &nm_i->nat_entries);
 308         spin_unlock(&nm_i->nat_list_lock);
 309
 310         set_nat_flag(ne, IS_DIRTY, false);
 311         set->entry_cnt--;
 312         nm_i->nat_cnt[DIRTY_NAT]--;
 313         nm_i->nat_cnt[RECLAIMABLE_NAT]++;
 314 }
 315
 316 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 317                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 318 {
 319         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 320                                                         start, nr);
 321 }
 322
 323 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct folio *folio)
 324 {
 325         return is_node_folio(folio) && IS_DNODE(folio) && is_cold_node(folio);
 326 }
 327
 328 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
 329 {
 330         spin_lock_init(&sbi->fsync_node_lock);
 331         INIT_LIST_HEAD(&sbi->fsync_node_list);
 332         sbi->fsync_seg_id = 0;
 333         sbi->fsync_node_num = 0;
 334 }
 335
 336 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
 337                 struct folio *folio)
 338 {
 339         struct fsync_node_entry *fn;
 340         unsigned long flags;
 341         unsigned int seq_id;
 342
 343         fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
 344                                         GFP_NOFS, true, NULL);
 345
 346         folio_get(folio);
 347         fn->folio = folio;
 348         INIT_LIST_HEAD(&fn->list);
 349
 350         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 351         list_add_tail(&fn->list, &sbi->fsync_node_list);
 352         fn->seq_id = sbi->fsync_seg_id++;
 353         seq_id = fn->seq_id;
 354         sbi->fsync_node_num++;
 355         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 356
 357         return seq_id;
 358 }
 359
 360 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct folio *folio)
 361 {
 362         struct fsync_node_entry *fn;
 363         unsigned long flags;
 364
 365         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 366         list_for_each_entry(fn, &sbi->fsync_node_list, list) {
 367                 if (fn->folio == folio) {
 368                         list_del(&fn->list);
 369                         sbi->fsync_node_num--;
 370                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 371                         kmem_cache_free(fsync_node_entry_slab, fn);
 372                         folio_put(folio);
 373                         return;
 374                 }
 375         }
 376         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 377         f2fs_bug_on(sbi, 1);
 378 }
 379
 380 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
 381 {
 382         unsigned long flags;
 383
 384         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 385         sbi->fsync_seg_id = 0;
 386         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 387 }
 388
 389 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 390 {
 391         struct f2fs_nm_info *nm_i = NM_I(sbi);
 392         struct nat_entry *e;
 393         bool need = false;
 394
 395         f2fs_down_read(&nm_i->nat_tree_lock);
 396         e = __lookup_nat_cache(nm_i, nid, false);
 397         if (e) {
 398                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 399                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 400                         need = true;
 401         }
 402         f2fs_up_read(&nm_i->nat_tree_lock);
 403         return need;
 404 }
 405
 406 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 407 {
 408         struct f2fs_nm_info *nm_i = NM_I(sbi);
 409         struct nat_entry *e;
 410         bool is_cp = true;
 411
 412         f2fs_down_read(&nm_i->nat_tree_lock);
 413         e = __lookup_nat_cache(nm_i, nid, false);
 414         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 415                 is_cp = false;
 416         f2fs_up_read(&nm_i->nat_tree_lock);
 417         return is_cp;
 418 }
 419
 420 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 421 {
 422         struct f2fs_nm_info *nm_i = NM_I(sbi);
 423         struct nat_entry *e;
 424         bool need_update = true;
 425
 426         f2fs_down_read(&nm_i->nat_tree_lock);
 427         e = __lookup_nat_cache(nm_i, ino, false);
 428         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 429                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 430                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 431                 need_update = false;
 432         f2fs_up_read(&nm_i->nat_tree_lock);
 433         return need_update;
 434 }
 435
 436 /* must be locked by nat_tree_lock */
 437 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 438                                                 struct f2fs_nat_entry *ne)
 439 {
 440         struct f2fs_nm_info *nm_i = NM_I(sbi);
 441         struct nat_entry *new, *e;
 442
 443         /* Let's mitigate lock contention of nat_tree_lock during checkpoint */
 444         if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
 445                 return;
 446
 447         new = __alloc_nat_entry(sbi, nid, false);
 448         if (!new)
 449                 return;
 450
 451         f2fs_down_write(&nm_i->nat_tree_lock);
 452         e = __lookup_nat_cache(nm_i, nid, false);
 453         if (!e)
 454                 e = __init_nat_entry(nm_i, new, ne, false, false);
 455         else
 456                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 457                                 nat_get_blkaddr(e) !=
 458                                         le32_to_cpu(ne->block_addr) ||
 459                                 nat_get_version(e) != ne->version);
 460         f2fs_up_write(&nm_i->nat_tree_lock);
 461         if (e != new)
 462                 __free_nat_entry(new);
 463 }
 464
 465 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 466                         block_t new_blkaddr, bool fsync_done)
 467 {
 468         struct f2fs_nm_info *nm_i = NM_I(sbi);
 469         struct nat_entry *e;
 470         struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
 471         bool init_dirty = false;
 472
 473         f2fs_down_write(&nm_i->nat_tree_lock);
 474         e = __lookup_nat_cache(nm_i, ni->nid, true);
 475         if (!e) {
 476                 init_dirty = true;
 477                 e = __init_nat_entry(nm_i, new, NULL, true, true);
 478                 copy_node_info(&e->ni, ni);
 479                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 480         } else if (new_blkaddr == NEW_ADDR) {
 481                 /*
 482                  * when nid is reallocated,
 483                  * previous nat entry can be remained in nat cache.
 484                  * So, reinitialize it with new information.
 485                  */
 486                 copy_node_info(&e->ni, ni);
 487                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 488         }
 489         /* let's free early to reduce memory consumption */
 490         if (e != new)
 491                 __free_nat_entry(new);
 492
 493         /* sanity check */
 494         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 495         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 496                         new_blkaddr == NULL_ADDR);
 497         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 498                         new_blkaddr == NEW_ADDR);
 499         f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
 500                         new_blkaddr == NEW_ADDR);
 501
 502         /* increment version no as node is removed */
 503         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 504                 unsigned char version = nat_get_version(e);
 505
 506                 nat_set_version(e, inc_node_version(version));
 507         }
 508
 509         /* change address */
 510         nat_set_blkaddr(e, new_blkaddr);
 511         if (!__is_valid_data_blkaddr(new_blkaddr))
 512                 set_nat_flag(e, IS_CHECKPOINTED, false);
 513         __set_nat_cache_dirty(nm_i, e, init_dirty);
 514
 515         /* update fsync_mark if its inode nat entry is still alive */
 516         if (ni->nid != ni->ino)
 517                 e = __lookup_nat_cache(nm_i, ni->ino, false);
 518         if (e) {
 519                 if (fsync_done && ni->nid == ni->ino)
 520                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 521                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 522         }
 523         f2fs_up_write(&nm_i->nat_tree_lock);
 524 }
 525
 526 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 527 {
 528         struct f2fs_nm_info *nm_i = NM_I(sbi);
 529         int nr = nr_shrink;
 530
 531         if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
 532                 return 0;
 533
 534         spin_lock(&nm_i->nat_list_lock);
 535         while (nr_shrink) {
 536                 struct nat_entry *ne;
 537
 538                 if (list_empty(&nm_i->nat_entries))
 539                         break;
 540
 541                 ne = list_first_entry(&nm_i->nat_entries,
 542                                         struct nat_entry, list);
 543                 list_del(&ne->list);
 544                 spin_unlock(&nm_i->nat_list_lock);
 545
 546                 __del_from_nat_cache(nm_i, ne);
 547                 nr_shrink--;
 548
 549                 spin_lock(&nm_i->nat_list_lock);
 550         }
 551         spin_unlock(&nm_i->nat_list_lock);
 552
 553         f2fs_up_write(&nm_i->nat_tree_lock);
 554         return nr - nr_shrink;
 555 }
 556
 557 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
 558                                 struct node_info *ni, bool checkpoint_context)
 559 {
 560         struct f2fs_nm_info *nm_i = NM_I(sbi);
 561         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 562         struct f2fs_journal *journal = curseg->journal;
 563         nid_t start_nid = START_NID(nid);
 564         struct f2fs_nat_block *nat_blk;
 565         struct folio *folio = NULL;
 566         struct f2fs_nat_entry ne;
 567         struct nat_entry *e;
 568         pgoff_t index;
 569         int i;
 570         bool need_cache = true;
 571
 572         ni->flag = 0;
 573         ni->nid = nid;
 574 retry:
 575         /* Check nat cache */
 576         f2fs_down_read(&nm_i->nat_tree_lock);
 577         e = __lookup_nat_cache(nm_i, nid, false);
 578         if (e) {
 579                 ni->ino = nat_get_ino(e);
 580                 ni->blk_addr = nat_get_blkaddr(e);
 581                 ni->version = nat_get_version(e);
 582                 f2fs_up_read(&nm_i->nat_tree_lock);
 583                 if (IS_ENABLED(CONFIG_F2FS_CHECK_FS)) {
 584                         need_cache = false;
 585                         goto sanity_check;
 586                 }
 587                 return 0;
 588         }
 589
 590         /*
 591          * Check current segment summary by trying to grab journal_rwsem first.
 592          * This sem is on the critical path on the checkpoint requiring the above
 593          * nat_tree_lock. Therefore, we should retry, if we failed to grab here
 594          * while not bothering checkpoint.
 595          */
 596         if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
 597                 down_read(&curseg->journal_rwsem);
 598         } else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
 599                                 !down_read_trylock(&curseg->journal_rwsem)) {
 600                 f2fs_up_read(&nm_i->nat_tree_lock);
 601                 goto retry;
 602         }
 603
 604         i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 605         if (i >= 0) {
 606                 ne = nat_in_journal(journal, i);
 607                 node_info_from_raw_nat(ni, &ne);
 608         }
 609         up_read(&curseg->journal_rwsem);
 610         if (i >= 0) {
 611                 f2fs_up_read(&nm_i->nat_tree_lock);
 612                 goto sanity_check;
 613         }
 614
 615         /* Fill node_info from nat page */
 616         index = current_nat_addr(sbi, nid);
 617         f2fs_up_read(&nm_i->nat_tree_lock);
 618
 619         folio = f2fs_get_meta_folio(sbi, index);
 620         if (IS_ERR(folio))
 621                 return PTR_ERR(folio);
 622
 623         nat_blk = folio_address(folio);
 624         ne = nat_blk->entries[nid - start_nid];
 625         node_info_from_raw_nat(ni, &ne);
 626         f2fs_folio_put(folio, true);
 627 sanity_check:
 628         if (__is_valid_data_blkaddr(ni->blk_addr) &&
 629                 !f2fs_is_valid_blkaddr(sbi, ni->blk_addr,
 630                                         DATA_GENERIC_ENHANCE)) {
 631                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 632                 f2fs_err_ratelimited(sbi,
 633                         "f2fs_get_node_info of %pS: inconsistent nat entry, "
 634                         "ino:%u, nid:%u, blkaddr:%u, ver:%u, flag:%u",
 635                         __builtin_return_address(0),
 636                         ni->ino, ni->nid, ni->blk_addr, ni->version, ni->flag);
 637                 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
 638                 return -EFSCORRUPTED;
 639         }
 640
 641         /* cache nat entry */
 642         if (need_cache)
 643                 cache_nat_entry(sbi, nid, &ne);
 644         return 0;
 645 }
 646
 647 /*
 648  * readahead MAX_RA_NODE number of node pages.
 649  */
 650 static void f2fs_ra_node_pages(struct folio *parent, int start, int n)
 651 {
 652         struct f2fs_sb_info *sbi = F2FS_F_SB(parent);
 653         struct blk_plug plug;
 654         int i, end;
 655         nid_t nid;
 656
 657         blk_start_plug(&plug);
 658
 659         /* Then, try readahead for siblings of the desired node */
 660         end = start + n;
 661         end = min(end, (int)NIDS_PER_BLOCK);
 662         for (i = start; i < end; i++) {
 663                 nid = get_nid(parent, i, false);
 664                 f2fs_ra_node_page(sbi, nid);
 665         }
 666
 667         blk_finish_plug(&plug);
 668 }
 669
 670 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 671 {
 672         const long direct_index = ADDRS_PER_INODE(dn->inode);
 673         const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
 674         const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
 675         unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
 676         int cur_level = dn->cur_level;
 677         int max_level = dn->max_level;
 678         pgoff_t base = 0;
 679
 680         if (!dn->max_level)
 681                 return pgofs + 1;
 682
 683         while (max_level-- > cur_level)
 684                 skipped_unit *= NIDS_PER_BLOCK;
 685
 686         switch (dn->max_level) {
 687         case 3:
 688                 base += 2 * indirect_blks;
 689                 fallthrough;
 690         case 2:
 691                 base += 2 * direct_blks;
 692                 fallthrough;
 693         case 1:
 694                 base += direct_index;
 695                 break;
 696         default:
 697                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 698         }
 699
 700         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 701 }
 702
 703 /*
 704  * The maximum depth is four.
 705  * Offset[0] will have raw inode offset.
 706  */
 707 static int get_node_path(struct inode *inode, long block,
 708                                 int offset[4], unsigned int noffset[4])
 709 {
 710         const long direct_index = ADDRS_PER_INODE(inode);
 711         const long direct_blks = ADDRS_PER_BLOCK(inode);
 712         const long dptrs_per_blk = NIDS_PER_BLOCK;
 713         const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
 714         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 715         int n = 0;
 716         int level = 0;
 717
 718         noffset[0] = 0;
 719
 720         if (block < direct_index) {
 721                 offset[n] = block;
 722                 goto got;
 723         }
 724         block -= direct_index;
 725         if (block < direct_blks) {
 726                 offset[n++] = NODE_DIR1_BLOCK;
 727                 noffset[n] = 1;
 728                 offset[n] = block;
 729                 level = 1;
 730                 goto got;
 731         }
 732         block -= direct_blks;
 733         if (block < direct_blks) {
 734                 offset[n++] = NODE_DIR2_BLOCK;
 735                 noffset[n] = 2;
 736                 offset[n] = block;
 737                 level = 1;
 738                 goto got;
 739         }
 740         block -= direct_blks;
 741         if (block < indirect_blks) {
 742                 offset[n++] = NODE_IND1_BLOCK;
 743                 noffset[n] = 3;
 744                 offset[n++] = block / direct_blks;
 745                 noffset[n] = 4 + offset[n - 1];
 746                 offset[n] = block % direct_blks;
 747                 level = 2;
 748                 goto got;
 749         }
 750         block -= indirect_blks;
 751         if (block < indirect_blks) {
 752                 offset[n++] = NODE_IND2_BLOCK;
 753                 noffset[n] = 4 + dptrs_per_blk;
 754                 offset[n++] = block / direct_blks;
 755                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 756                 offset[n] = block % direct_blks;
 757                 level = 2;
 758                 goto got;
 759         }
 760         block -= indirect_blks;
 761         if (block < dindirect_blks) {
 762                 offset[n++] = NODE_DIND_BLOCK;
 763                 noffset[n] = 5 + (dptrs_per_blk * 2);
 764                 offset[n++] = block / indirect_blks;
 765                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 766                               offset[n - 1] * (dptrs_per_blk + 1);
 767                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 768                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 769                               offset[n - 2] * (dptrs_per_blk + 1) +
 770                               offset[n - 1];
 771                 offset[n] = block % direct_blks;
 772                 level = 3;
 773                 goto got;
 774         } else {
 775                 return -E2BIG;
 776         }
 777 got:
 778         return level;
 779 }
 780
 781 static struct folio *f2fs_get_node_folio_ra(struct folio *parent, int start);
 782
 783 /*
 784  * Caller should call f2fs_put_dnode(dn).
 785  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 786  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
 787  */
 788 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 789 {
 790         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 791         struct folio *nfolio[4];
 792         struct folio *parent = NULL;
 793         int offset[4];
 794         unsigned int noffset[4];
 795         nid_t nids[4];
 796         int level, i = 0;
 797         int err = 0;
 798
 799         level = get_node_path(dn->inode, index, offset, noffset);
 800         if (level < 0)
 801                 return level;
 802
 803         nids[0] = dn->inode->i_ino;
 804
 805         if (!dn->inode_folio) {
 806                 nfolio[0] = f2fs_get_inode_folio(sbi, nids[0]);
 807                 if (IS_ERR(nfolio[0]))
 808                         return PTR_ERR(nfolio[0]);
 809         } else {
 810                 nfolio[0] = dn->inode_folio;
 811         }
 812
 813         /* if inline_data is set, should not report any block indices */
 814         if (f2fs_has_inline_data(dn->inode) && index) {
 815                 err = -ENOENT;
 816                 f2fs_folio_put(nfolio[0], true);
 817                 goto release_out;
 818         }
 819
 820         parent = nfolio[0];
 821         if (level != 0)
 822                 nids[1] = get_nid(parent, offset[0], true);
 823         dn->inode_folio = nfolio[0];
 824         dn->inode_folio_locked = true;
 825
 826         /* get indirect or direct nodes */
 827         for (i = 1; i <= level; i++) {
 828                 bool done = false;
 829
 830                 if (nids[i] && nids[i] == dn->inode->i_ino) {
 831                         err = -EFSCORRUPTED;
 832                         f2fs_err_ratelimited(sbi,
 833                                 "inode mapping table is corrupted, run fsck to fix it, "
 834                                 "ino:%lu, nid:%u, level:%d, offset:%d",
 835                                 dn->inode->i_ino, nids[i], level, offset[level]);
 836                         set_sbi_flag(sbi, SBI_NEED_FSCK);
 837                         goto release_pages;
 838                 }
 839
 840                 if (!nids[i] && mode == ALLOC_NODE) {
 841                         /* alloc new node */
 842                         if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
 843                                 err = -ENOSPC;
 844                                 goto release_pages;
 845                         }
 846
 847                         dn->nid = nids[i];
 848                         nfolio[i] = f2fs_new_node_folio(dn, noffset[i]);
 849                         if (IS_ERR(nfolio[i])) {
 850                                 f2fs_alloc_nid_failed(sbi, nids[i]);
 851                                 err = PTR_ERR(nfolio[i]);
 852                                 goto release_pages;
 853                         }
 854
 855                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 856                         f2fs_alloc_nid_done(sbi, nids[i]);
 857                         done = true;
 858                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 859                         nfolio[i] = f2fs_get_node_folio_ra(parent, offset[i - 1]);
 860                         if (IS_ERR(nfolio[i])) {
 861                                 err = PTR_ERR(nfolio[i]);
 862                                 goto release_pages;
 863                         }
 864                         done = true;
 865                 }
 866                 if (i == 1) {
 867                         dn->inode_folio_locked = false;
 868                         folio_unlock(parent);
 869                 } else {
 870                         f2fs_folio_put(parent, true);
 871                 }
 872
 873                 if (!done) {
 874                         nfolio[i] = f2fs_get_node_folio(sbi, nids[i]);
 875                         if (IS_ERR(nfolio[i])) {
 876                                 err = PTR_ERR(nfolio[i]);
 877                                 f2fs_folio_put(nfolio[0], false);
 878                                 goto release_out;
 879                         }
 880                 }
 881                 if (i < level) {
 882                         parent = nfolio[i];
 883                         nids[i + 1] = get_nid(parent, offset[i], false);
 884                 }
 885         }
 886         dn->nid = nids[level];
 887         dn->ofs_in_node = offset[level];
 888         dn->node_folio = nfolio[level];
 889         dn->data_blkaddr = f2fs_data_blkaddr(dn);
 890
 891         if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
 892                                         f2fs_sb_has_readonly(sbi)) {
 893                 unsigned int cluster_size = F2FS_I(dn->inode)->i_cluster_size;
 894                 unsigned int ofs_in_node = dn->ofs_in_node;
 895                 pgoff_t fofs = index;
 896                 unsigned int c_len;
 897                 block_t blkaddr;
 898
 899                 /* should align fofs and ofs_in_node to cluster_size */
 900                 if (fofs % cluster_size) {
 901                         fofs = round_down(fofs, cluster_size);
 902                         ofs_in_node = round_down(ofs_in_node, cluster_size);
 903                 }
 904
 905                 c_len = f2fs_cluster_blocks_are_contiguous(dn, ofs_in_node);
 906                 if (!c_len)
 907                         goto out;
 908
 909                 blkaddr = data_blkaddr(dn->inode, dn->node_folio, ofs_in_node);
 910                 if (blkaddr == COMPRESS_ADDR)
 911                         blkaddr = data_blkaddr(dn->inode, dn->node_folio,
 912                                                 ofs_in_node + 1);
 913
 914                 f2fs_update_read_extent_tree_range_compressed(dn->inode,
 915                                         fofs, blkaddr, cluster_size, c_len);
 916         }
 917 out:
 918         return 0;
 919
 920 release_pages:
 921         f2fs_folio_put(parent, true);
 922         if (i > 1)
 923                 f2fs_folio_put(nfolio[0], false);
 924 release_out:
 925         dn->inode_folio = NULL;
 926         dn->node_folio = NULL;
 927         if (err == -ENOENT) {
 928                 dn->cur_level = i;
 929                 dn->max_level = level;
 930                 dn->ofs_in_node = offset[level];
 931         }
 932         return err;
 933 }
 934
 935 static int truncate_node(struct dnode_of_data *dn)
 936 {
 937         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 938         struct node_info ni;
 939         int err;
 940         pgoff_t index;
 941
 942         err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
 943         if (err)
 944                 return err;
 945
 946         if (ni.blk_addr != NEW_ADDR &&
 947                 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC_ENHANCE)) {
 948                 f2fs_err_ratelimited(sbi,
 949                         "nat entry is corrupted, run fsck to fix it, ino:%u, "
 950                         "nid:%u, blkaddr:%u", ni.ino, ni.nid, ni.blk_addr);
 951                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 952                 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
 953                 return -EFSCORRUPTED;
 954         }
 955
 956         /* Deallocate node address */
 957         f2fs_invalidate_blocks(sbi, ni.blk_addr, 1);
 958         dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
 959         set_node_addr(sbi, &ni, NULL_ADDR, false);
 960
 961         if (dn->nid == dn->inode->i_ino) {
 962                 f2fs_remove_orphan_inode(sbi, dn->nid);
 963                 dec_valid_inode_count(sbi);
 964                 f2fs_inode_synced(dn->inode);
 965         }
 966
 967         clear_node_folio_dirty(dn->node_folio);
 968         set_sbi_flag(sbi, SBI_IS_DIRTY);
 969
 970         index = dn->node_folio->index;
 971         f2fs_folio_put(dn->node_folio, true);
 972
 973         invalidate_mapping_pages(NODE_MAPPING(sbi),
 974                         index, index);
 975
 976         dn->node_folio = NULL;
 977         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 978
 979         return 0;
 980 }
 981
 982 static int truncate_dnode(struct dnode_of_data *dn)
 983 {
 984         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 985         struct folio *folio;
 986         int err;
 987
 988         if (dn->nid == 0)
 989                 return 1;
 990
 991         /* get direct node */
 992         folio = f2fs_get_node_folio(sbi, dn->nid);
 993         if (PTR_ERR(folio) == -ENOENT)
 994                 return 1;
 995         else if (IS_ERR(folio))
 996                 return PTR_ERR(folio);
 997
 998         if (IS_INODE(folio) || ino_of_node(folio) != dn->inode->i_ino) {
 999                 f2fs_err(sbi, "incorrect node reference, ino: %lu, nid: %u, ino_of_node: %u",
1000                                 dn->inode->i_ino, dn->nid, ino_of_node(folio));
1001                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1002                 f2fs_handle_error(sbi, ERROR_INVALID_NODE_REFERENCE);
1003                 f2fs_folio_put(folio, true);
1004                 return -EFSCORRUPTED;
1005         }
1006
1007         /* Make dnode_of_data for parameter */
1008         dn->node_folio = folio;
1009         dn->ofs_in_node = 0;
1010         f2fs_truncate_data_blocks_range(dn, ADDRS_PER_BLOCK(dn->inode));
1011         err = truncate_node(dn);
1012         if (err) {
1013                 f2fs_folio_put(folio, true);
1014                 return err;
1015         }
1016
1017         return 1;
1018 }
1019
1020 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
1021                                                 int ofs, int depth)
1022 {
1023         struct dnode_of_data rdn = *dn;
1024         struct folio *folio;
1025         struct f2fs_node *rn;
1026         nid_t child_nid;
1027         unsigned int child_nofs;
1028         int freed = 0;
1029         int i, ret;
1030
1031         if (dn->nid == 0)
1032                 return NIDS_PER_BLOCK + 1;
1033
1034         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
1035
1036         folio = f2fs_get_node_folio(F2FS_I_SB(dn->inode), dn->nid);
1037         if (IS_ERR(folio)) {
1038                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(folio));
1039                 return PTR_ERR(folio);
1040         }
1041
1042         f2fs_ra_node_pages(folio, ofs, NIDS_PER_BLOCK);
1043
1044         rn = F2FS_NODE(folio);
1045         if (depth < 3) {
1046                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
1047                         child_nid = le32_to_cpu(rn->in.nid[i]);
1048                         if (child_nid == 0)
1049                                 continue;
1050                         rdn.nid = child_nid;
1051                         ret = truncate_dnode(&rdn);
1052                         if (ret < 0)
1053                                 goto out_err;
1054                         if (set_nid(folio, i, 0, false))
1055                                 dn->node_changed = true;
1056                 }
1057         } else {
1058                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
1059                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
1060                         child_nid = le32_to_cpu(rn->in.nid[i]);
1061                         if (child_nid == 0) {
1062                                 child_nofs += NIDS_PER_BLOCK + 1;
1063                                 continue;
1064                         }
1065                         rdn.nid = child_nid;
1066                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
1067                         if (ret == (NIDS_PER_BLOCK + 1)) {
1068                                 if (set_nid(folio, i, 0, false))
1069                                         dn->node_changed = true;
1070                                 child_nofs += ret;
1071                         } else if (ret < 0 && ret != -ENOENT) {
1072                                 goto out_err;
1073                         }
1074                 }
1075                 freed = child_nofs;
1076         }
1077
1078         if (!ofs) {
1079                 /* remove current indirect node */
1080                 dn->node_folio = folio;
1081                 ret = truncate_node(dn);
1082                 if (ret)
1083                         goto out_err;
1084                 freed++;
1085         } else {
1086                 f2fs_folio_put(folio, true);
1087         }
1088         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1089         return freed;
1090
1091 out_err:
1092         f2fs_folio_put(folio, true);
1093         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1094         return ret;
1095 }
1096
1097 static int truncate_partial_nodes(struct dnode_of_data *dn,
1098                         struct f2fs_inode *ri, int *offset, int depth)
1099 {
1100         struct folio *folios[2];
1101         nid_t nid[3];
1102         nid_t child_nid;
1103         int err = 0;
1104         int i;
1105         int idx = depth - 2;
1106
1107         nid[0] = get_nid(dn->inode_folio, offset[0], true);
1108         if (!nid[0])
1109                 return 0;
1110
1111         /* get indirect nodes in the path */
1112         for (i = 0; i < idx + 1; i++) {
1113                 /* reference count'll be increased */
1114                 folios[i] = f2fs_get_node_folio(F2FS_I_SB(dn->inode), nid[i]);
1115                 if (IS_ERR(folios[i])) {
1116                         err = PTR_ERR(folios[i]);
1117                         idx = i - 1;
1118                         goto fail;
1119                 }
1120                 nid[i + 1] = get_nid(folios[i], offset[i + 1], false);
1121         }
1122
1123         f2fs_ra_node_pages(folios[idx], offset[idx + 1], NIDS_PER_BLOCK);
1124
1125         /* free direct nodes linked to a partial indirect node */
1126         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1127                 child_nid = get_nid(folios[idx], i, false);
1128                 if (!child_nid)
1129                         continue;
1130                 dn->nid = child_nid;
1131                 err = truncate_dnode(dn);
1132                 if (err < 0)
1133                         goto fail;
1134                 if (set_nid(folios[idx], i, 0, false))
1135                         dn->node_changed = true;
1136         }
1137
1138         if (offset[idx + 1] == 0) {
1139                 dn->node_folio = folios[idx];
1140                 dn->nid = nid[idx];
1141                 err = truncate_node(dn);
1142                 if (err)
1143                         goto fail;
1144         } else {
1145                 f2fs_folio_put(folios[idx], true);
1146         }
1147         offset[idx]++;
1148         offset[idx + 1] = 0;
1149         idx--;
1150 fail:
1151         for (i = idx; i >= 0; i--)
1152                 f2fs_folio_put(folios[i], true);
1153
1154         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1155
1156         return err;
1157 }
1158
1159 /*
1160  * All the block addresses of data and nodes should be nullified.
1161  */
1162 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1163 {
1164         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1165         int err = 0, cont = 1;
1166         int level, offset[4], noffset[4];
1167         unsigned int nofs = 0;
1168         struct f2fs_inode *ri;
1169         struct dnode_of_data dn;
1170         struct folio *folio;
1171
1172         trace_f2fs_truncate_inode_blocks_enter(inode, from);
1173
1174         level = get_node_path(inode, from, offset, noffset);
1175         if (level <= 0) {
1176                 if (!level) {
1177                         level = -EFSCORRUPTED;
1178                         f2fs_err(sbi, "%s: inode ino=%lx has corrupted node block, from:%lu addrs:%u",
1179                                         __func__, inode->i_ino,
1180                                         from, ADDRS_PER_INODE(inode));
1181                         set_sbi_flag(sbi, SBI_NEED_FSCK);
1182                 }
1183                 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1184                 return level;
1185         }
1186
1187         folio = f2fs_get_inode_folio(sbi, inode->i_ino);
1188         if (IS_ERR(folio)) {
1189                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(folio));
1190                 return PTR_ERR(folio);
1191         }
1192
1193         set_new_dnode(&dn, inode, folio, NULL, 0);
1194         folio_unlock(folio);
1195
1196         ri = F2FS_INODE(folio);
1197         switch (level) {
1198         case 0:
1199         case 1:
1200                 nofs = noffset[1];
1201                 break;
1202         case 2:
1203                 nofs = noffset[1];
1204                 if (!offset[level - 1])
1205                         goto skip_partial;
1206                 err = truncate_partial_nodes(&dn, ri, offset, level);
1207                 if (err < 0 && err != -ENOENT)
1208                         goto fail;
1209                 nofs += 1 + NIDS_PER_BLOCK;
1210                 break;
1211         case 3:
1212                 nofs = 5 + 2 * NIDS_PER_BLOCK;
1213                 if (!offset[level - 1])
1214                         goto skip_partial;
1215                 err = truncate_partial_nodes(&dn, ri, offset, level);
1216                 if (err < 0 && err != -ENOENT)
1217                         goto fail;
1218                 break;
1219         default:
1220                 BUG();
1221         }
1222
1223 skip_partial:
1224         while (cont) {
1225                 dn.nid = get_nid(folio, offset[0], true);
1226                 switch (offset[0]) {
1227                 case NODE_DIR1_BLOCK:
1228                 case NODE_DIR2_BLOCK:
1229                         err = truncate_dnode(&dn);
1230                         break;
1231
1232                 case NODE_IND1_BLOCK:
1233                 case NODE_IND2_BLOCK:
1234                         err = truncate_nodes(&dn, nofs, offset[1], 2);
1235                         break;
1236
1237                 case NODE_DIND_BLOCK:
1238                         err = truncate_nodes(&dn, nofs, offset[1], 3);
1239                         cont = 0;
1240                         break;
1241
1242                 default:
1243                         BUG();
1244                 }
1245                 if (err == -ENOENT) {
1246                         set_sbi_flag(F2FS_F_SB(folio), SBI_NEED_FSCK);
1247                         f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1248                         f2fs_err_ratelimited(sbi,
1249                                 "truncate node fail, ino:%lu, nid:%u, "
1250                                 "offset[0]:%d, offset[1]:%d, nofs:%d",
1251                                 inode->i_ino, dn.nid, offset[0],
1252                                 offset[1], nofs);
1253                         err = 0;
1254                 }
1255                 if (err < 0)
1256                         goto fail;
1257                 if (offset[1] == 0 && get_nid(folio, offset[0], true)) {
1258                         folio_lock(folio);
1259                         BUG_ON(!is_node_folio(folio));
1260                         set_nid(folio, offset[0], 0, true);
1261                         folio_unlock(folio);
1262                 }
1263                 offset[1] = 0;
1264                 offset[0]++;
1265                 nofs += err;
1266         }
1267 fail:
1268         f2fs_folio_put(folio, false);
1269         trace_f2fs_truncate_inode_blocks_exit(inode, err);
1270         return err > 0 ? 0 : err;
1271 }
1272
1273 /* caller must lock inode page */
1274 int f2fs_truncate_xattr_node(struct inode *inode)
1275 {
1276         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1277         nid_t nid = F2FS_I(inode)->i_xattr_nid;
1278         struct dnode_of_data dn;
1279         struct folio *nfolio;
1280         int err;
1281
1282         if (!nid)
1283                 return 0;
1284
1285         nfolio = f2fs_get_xnode_folio(sbi, nid);
1286         if (IS_ERR(nfolio))
1287                 return PTR_ERR(nfolio);
1288
1289         set_new_dnode(&dn, inode, NULL, nfolio, nid);
1290         err = truncate_node(&dn);
1291         if (err) {
1292                 f2fs_folio_put(nfolio, true);
1293                 return err;
1294         }
1295
1296         f2fs_i_xnid_write(inode, 0);
1297
1298         return 0;
1299 }
1300
1301 /*
1302  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1303  * f2fs_unlock_op().
1304  */
1305 int f2fs_remove_inode_page(struct inode *inode)
1306 {
1307         struct dnode_of_data dn;
1308         int err;
1309
1310         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1311         err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1312         if (err)
1313                 return err;
1314
1315         err = f2fs_truncate_xattr_node(inode);
1316         if (err) {
1317                 f2fs_put_dnode(&dn);
1318                 return err;
1319         }
1320
1321         /* remove potential inline_data blocks */
1322         if (!IS_DEVICE_ALIASING(inode) &&
1323             (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1324              S_ISLNK(inode->i_mode)))
1325                 f2fs_truncate_data_blocks_range(&dn, 1);
1326
1327         /* 0 is possible, after f2fs_new_inode() has failed */
1328         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1329                 f2fs_put_dnode(&dn);
1330                 return -EIO;
1331         }
1332
1333         if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1334                 f2fs_warn(F2FS_I_SB(inode),
1335                         "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1336                         inode->i_ino, (unsigned long long)inode->i_blocks);
1337                 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1338         }
1339
1340         /* will put inode & node pages */
1341         err = truncate_node(&dn);
1342         if (err) {
1343                 f2fs_put_dnode(&dn);
1344                 return err;
1345         }
1346         return 0;
1347 }
1348
1349 struct folio *f2fs_new_inode_folio(struct inode *inode)
1350 {
1351         struct dnode_of_data dn;
1352
1353         /* allocate inode page for new inode */
1354         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1355
1356         /* caller should f2fs_folio_put(folio, true); */
1357         return f2fs_new_node_folio(&dn, 0);
1358 }
1359
1360 struct folio *f2fs_new_node_folio(struct dnode_of_data *dn, unsigned int ofs)
1361 {
1362         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1363         struct node_info new_ni;
1364         struct folio *folio;
1365         int err;
1366
1367         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1368                 return ERR_PTR(-EPERM);
1369
1370         folio = f2fs_grab_cache_folio(NODE_MAPPING(sbi), dn->nid, false);
1371         if (IS_ERR(folio))
1372                 return folio;
1373
1374         if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1375                 goto fail;
1376
1377 #ifdef CONFIG_F2FS_CHECK_FS
1378         err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1379         if (err) {
1380                 dec_valid_node_count(sbi, dn->inode, !ofs);
1381                 goto fail;
1382         }
1383         if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1384                 err = -EFSCORRUPTED;
1385                 dec_valid_node_count(sbi, dn->inode, !ofs);
1386                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1387                 f2fs_warn_ratelimited(sbi,
1388                         "f2fs_new_node_folio: inconsistent nat entry, "
1389                         "ino:%u, nid:%u, blkaddr:%u, ver:%u, flag:%u",
1390                         new_ni.ino, new_ni.nid, new_ni.blk_addr,
1391                         new_ni.version, new_ni.flag);
1392                 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
1393                 goto fail;
1394         }
1395 #endif
1396         new_ni.nid = dn->nid;
1397         new_ni.ino = dn->inode->i_ino;
1398         new_ni.blk_addr = NULL_ADDR;
1399         new_ni.flag = 0;
1400         new_ni.version = 0;
1401         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1402
1403         f2fs_folio_wait_writeback(folio, NODE, true, true);
1404         fill_node_footer(folio, dn->nid, dn->inode->i_ino, ofs, true);
1405         set_cold_node(folio, S_ISDIR(dn->inode->i_mode));
1406         if (!folio_test_uptodate(folio))
1407                 folio_mark_uptodate(folio);
1408         if (folio_mark_dirty(folio))
1409                 dn->node_changed = true;
1410
1411         if (f2fs_has_xattr_block(ofs))
1412                 f2fs_i_xnid_write(dn->inode, dn->nid);
1413
1414         if (ofs == 0)
1415                 inc_valid_inode_count(sbi);
1416         return folio;
1417 fail:
1418         clear_node_folio_dirty(folio);
1419         f2fs_folio_put(folio, true);
1420         return ERR_PTR(err);
1421 }
1422
1423 /*
1424  * Caller should do after getting the following values.
1425  * 0: f2fs_folio_put(folio, false)
1426  * LOCKED_PAGE or error: f2fs_folio_put(folio, true)
1427  */
1428 static int read_node_folio(struct folio *folio, blk_opf_t op_flags)
1429 {
1430         struct f2fs_sb_info *sbi = F2FS_F_SB(folio);
1431         struct node_info ni;
1432         struct f2fs_io_info fio = {
1433                 .sbi = sbi,
1434                 .type = NODE,
1435                 .op = REQ_OP_READ,
1436                 .op_flags = op_flags,
1437                 .folio = folio,
1438                 .encrypted_page = NULL,
1439         };
1440         int err;
1441
1442         if (folio_test_uptodate(folio)) {
1443                 if (!f2fs_inode_chksum_verify(sbi, folio)) {
1444                         folio_clear_uptodate(folio);
1445                         return -EFSBADCRC;
1446                 }
1447                 return LOCKED_PAGE;
1448         }
1449
1450         err = f2fs_get_node_info(sbi, folio->index, &ni, false);
1451         if (err)
1452                 return err;
1453
1454         /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1455         if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1456                 folio_clear_uptodate(folio);
1457                 return -ENOENT;
1458         }
1459
1460         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1461
1462         err = f2fs_submit_page_bio(&fio);
1463
1464         if (!err)
1465                 f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE);
1466
1467         return err;
1468 }
1469
1470 /*
1471  * Readahead a node page
1472  */
1473 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1474 {
1475         struct folio *afolio;
1476         int err;
1477
1478         if (!nid)
1479                 return;
1480         if (f2fs_check_nid_range(sbi, nid))
1481                 return;
1482
1483         afolio = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1484         if (afolio)
1485                 return;
1486
1487         afolio = f2fs_grab_cache_folio(NODE_MAPPING(sbi), nid, false);
1488         if (IS_ERR(afolio))
1489                 return;
1490
1491         err = read_node_folio(afolio, REQ_RAHEAD);
1492         f2fs_folio_put(afolio, err ? true : false);
1493 }
1494
1495 static int sanity_check_node_footer(struct f2fs_sb_info *sbi,
1496                                         struct folio *folio, pgoff_t nid,
1497                                         enum node_type ntype)
1498 {
1499         if (unlikely(nid != nid_of_node(folio) ||
1500                 (ntype == NODE_TYPE_INODE && !IS_INODE(folio)) ||
1501                 (ntype == NODE_TYPE_XATTR &&
1502                 !f2fs_has_xattr_block(ofs_of_node(folio))) ||
1503                 time_to_inject(sbi, FAULT_INCONSISTENT_FOOTER))) {
1504                 f2fs_warn(sbi, "inconsistent node block, node_type:%d, nid:%lu, "
1505                           "node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1506                           ntype, nid, nid_of_node(folio), ino_of_node(folio),
1507                           ofs_of_node(folio), cpver_of_node(folio),
1508                           next_blkaddr_of_node(folio));
1509                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1510                 f2fs_handle_error(sbi, ERROR_INCONSISTENT_FOOTER);
1511                 return -EFSCORRUPTED;
1512         }
1513         return 0;
1514 }
1515
1516 static struct folio *__get_node_folio(struct f2fs_sb_info *sbi, pgoff_t nid,
1517                 struct folio *parent, int start, enum node_type ntype)
1518 {
1519         struct folio *folio;
1520         int err;
1521
1522         if (!nid)
1523                 return ERR_PTR(-ENOENT);
1524         if (f2fs_check_nid_range(sbi, nid))
1525                 return ERR_PTR(-EINVAL);
1526 repeat:
1527         folio = f2fs_grab_cache_folio(NODE_MAPPING(sbi), nid, false);
1528         if (IS_ERR(folio))
1529                 return folio;
1530
1531         err = read_node_folio(folio, 0);
1532         if (err < 0)
1533                 goto out_put_err;
1534         if (err == LOCKED_PAGE)
1535                 goto page_hit;
1536
1537         if (parent)
1538                 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1539
1540         folio_lock(folio);
1541
1542         if (unlikely(!is_node_folio(folio))) {
1543                 f2fs_folio_put(folio, true);
1544                 goto repeat;
1545         }
1546
1547         if (unlikely(!folio_test_uptodate(folio))) {
1548                 err = -EIO;
1549                 goto out_err;
1550         }
1551
1552         if (!f2fs_inode_chksum_verify(sbi, folio)) {
1553                 err = -EFSBADCRC;
1554                 goto out_err;
1555         }
1556 page_hit:
1557         err = sanity_check_node_footer(sbi, folio, nid, ntype);
1558         if (!err)
1559                 return folio;
1560 out_err:
1561         folio_clear_uptodate(folio);
1562 out_put_err:
1563         /* ENOENT comes from read_node_folio which is not an error. */
1564         if (err != -ENOENT)
1565                 f2fs_handle_page_eio(sbi, folio, NODE);
1566         f2fs_folio_put(folio, true);
1567         return ERR_PTR(err);
1568 }
1569
1570 struct folio *f2fs_get_node_folio(struct f2fs_sb_info *sbi, pgoff_t nid)
1571 {
1572         return __get_node_folio(sbi, nid, NULL, 0, NODE_TYPE_REGULAR);
1573 }
1574
1575 struct folio *f2fs_get_inode_folio(struct f2fs_sb_info *sbi, pgoff_t ino)
1576 {
1577         return __get_node_folio(sbi, ino, NULL, 0, NODE_TYPE_INODE);
1578 }
1579
1580 struct folio *f2fs_get_xnode_folio(struct f2fs_sb_info *sbi, pgoff_t xnid)
1581 {
1582         return __get_node_folio(sbi, xnid, NULL, 0, NODE_TYPE_XATTR);
1583 }
1584
1585 static struct folio *f2fs_get_node_folio_ra(struct folio *parent, int start)
1586 {
1587         struct f2fs_sb_info *sbi = F2FS_F_SB(parent);
1588         nid_t nid = get_nid(parent, start, false);
1589
1590         return __get_node_folio(sbi, nid, parent, start, NODE_TYPE_REGULAR);
1591 }
1592
1593 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1594 {
1595         struct inode *inode;
1596         struct folio *folio;
1597         int ret;
1598
1599         /* should flush inline_data before evict_inode */
1600         inode = ilookup(sbi->sb, ino);
1601         if (!inode)
1602                 return;
1603
1604         folio = f2fs_filemap_get_folio(inode->i_mapping, 0,
1605                                         FGP_LOCK|FGP_NOWAIT, 0);
1606         if (IS_ERR(folio))
1607                 goto iput_out;
1608
1609         if (!folio_test_uptodate(folio))
1610                 goto folio_out;
1611
1612         if (!folio_test_dirty(folio))
1613                 goto folio_out;
1614
1615         if (!folio_clear_dirty_for_io(folio))
1616                 goto folio_out;
1617
1618         ret = f2fs_write_inline_data(inode, folio);
1619         inode_dec_dirty_pages(inode);
1620         f2fs_remove_dirty_inode(inode);
1621         if (ret)
1622                 folio_mark_dirty(folio);
1623 folio_out:
1624         f2fs_folio_put(folio, true);
1625 iput_out:
1626         iput(inode);
1627 }
1628
1629 static struct folio *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1630 {
1631         pgoff_t index;
1632         struct folio_batch fbatch;
1633         struct folio *last_folio = NULL;
1634         int nr_folios;
1635
1636         folio_batch_init(&fbatch);
1637         index = 0;
1638
1639         while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1640                                         (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1641                                         &fbatch))) {
1642                 int i;
1643
1644                 for (i = 0; i < nr_folios; i++) {
1645                         struct folio *folio = fbatch.folios[i];
1646
1647                         if (unlikely(f2fs_cp_error(sbi))) {
1648                                 f2fs_folio_put(last_folio, false);
1649                                 folio_batch_release(&fbatch);
1650                                 return ERR_PTR(-EIO);
1651                         }
1652
1653                         if (!IS_DNODE(folio) || !is_cold_node(folio))
1654                                 continue;
1655                         if (ino_of_node(folio) != ino)
1656                                 continue;
1657
1658                         folio_lock(folio);
1659
1660                         if (unlikely(!is_node_folio(folio))) {
1661 continue_unlock:
1662                                 folio_unlock(folio);
1663                                 continue;
1664                         }
1665                         if (ino_of_node(folio) != ino)
1666                                 goto continue_unlock;
1667
1668                         if (!folio_test_dirty(folio)) {
1669                                 /* someone wrote it for us */
1670                                 goto continue_unlock;
1671                         }
1672
1673                         if (last_folio)
1674                                 f2fs_folio_put(last_folio, false);
1675
1676                         folio_get(folio);
1677                         last_folio = folio;
1678                         folio_unlock(folio);
1679                 }
1680                 folio_batch_release(&fbatch);
1681                 cond_resched();
1682         }
1683         return last_folio;
1684 }
1685
1686 static bool __write_node_folio(struct folio *folio, bool atomic, bool *submitted,
1687                                 struct writeback_control *wbc, bool do_balance,
1688                                 enum iostat_type io_type, unsigned int *seq_id)
1689 {
1690         struct f2fs_sb_info *sbi = F2FS_F_SB(folio);
1691         nid_t nid;
1692         struct node_info ni;
1693         struct f2fs_io_info fio = {
1694                 .sbi = sbi,
1695                 .ino = ino_of_node(folio),
1696                 .type = NODE,
1697                 .op = REQ_OP_WRITE,
1698                 .op_flags = wbc_to_write_flags(wbc),
1699                 .folio = folio,
1700                 .encrypted_page = NULL,
1701                 .submitted = 0,
1702                 .io_type = io_type,
1703                 .io_wbc = wbc,
1704         };
1705         unsigned int seq;
1706
1707         trace_f2fs_writepage(folio, NODE);
1708
1709         if (unlikely(f2fs_cp_error(sbi))) {
1710                 /* keep node pages in remount-ro mode */
1711                 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
1712                         goto redirty_out;
1713                 folio_clear_uptodate(folio);
1714                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1715                 folio_unlock(folio);
1716                 return true;
1717         }
1718
1719         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1720                 goto redirty_out;
1721
1722         if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1723                         wbc->sync_mode == WB_SYNC_NONE &&
1724                         IS_DNODE(folio) && is_cold_node(folio))
1725                 goto redirty_out;
1726
1727         /* get old block addr of this node page */
1728         nid = nid_of_node(folio);
1729         f2fs_bug_on(sbi, folio->index != nid);
1730
1731         if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1732                 goto redirty_out;
1733
1734         f2fs_down_read(&sbi->node_write);
1735
1736         /* This page is already truncated */
1737         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1738                 folio_clear_uptodate(folio);
1739                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1740                 f2fs_up_read(&sbi->node_write);
1741                 folio_unlock(folio);
1742                 return true;
1743         }
1744
1745         if (__is_valid_data_blkaddr(ni.blk_addr) &&
1746                 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1747                                         DATA_GENERIC_ENHANCE)) {
1748                 f2fs_up_read(&sbi->node_write);
1749                 goto redirty_out;
1750         }
1751
1752         if (atomic && !test_opt(sbi, NOBARRIER))
1753                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1754
1755         /* should add to global list before clearing PAGECACHE status */
1756         if (f2fs_in_warm_node_list(sbi, folio)) {
1757                 seq = f2fs_add_fsync_node_entry(sbi, folio);
1758                 if (seq_id)
1759                         *seq_id = seq;
1760         }
1761
1762         folio_start_writeback(folio);
1763
1764         fio.old_blkaddr = ni.blk_addr;
1765         f2fs_do_write_node_page(nid, &fio);
1766         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(folio));
1767         dec_page_count(sbi, F2FS_DIRTY_NODES);
1768         f2fs_up_read(&sbi->node_write);
1769
1770         folio_unlock(folio);
1771
1772         if (unlikely(f2fs_cp_error(sbi))) {
1773                 f2fs_submit_merged_write(sbi, NODE);
1774                 submitted = NULL;
1775         }
1776         if (submitted)
1777                 *submitted = fio.submitted;
1778
1779         if (do_balance)
1780                 f2fs_balance_fs(sbi, false);
1781         return true;
1782
1783 redirty_out:
1784         folio_redirty_for_writepage(wbc, folio);
1785         folio_unlock(folio);
1786         return false;
1787 }
1788
1789 int f2fs_move_node_folio(struct folio *node_folio, int gc_type)
1790 {
1791         int err = 0;
1792
1793         if (gc_type == FG_GC) {
1794                 struct writeback_control wbc = {
1795                         .sync_mode = WB_SYNC_ALL,
1796                         .nr_to_write = 1,
1797                 };
1798
1799                 f2fs_folio_wait_writeback(node_folio, NODE, true, true);
1800
1801                 folio_mark_dirty(node_folio);
1802
1803                 if (!folio_clear_dirty_for_io(node_folio)) {
1804                         err = -EAGAIN;
1805                         goto out_page;
1806                 }
1807
1808                 if (!__write_node_folio(node_folio, false, NULL,
1809                                         &wbc, false, FS_GC_NODE_IO, NULL))
1810                         err = -EAGAIN;
1811                 goto release_page;
1812         } else {
1813                 /* set page dirty and write it */
1814                 if (!folio_test_writeback(node_folio))
1815                         folio_mark_dirty(node_folio);
1816         }
1817 out_page:
1818         folio_unlock(node_folio);
1819 release_page:
1820         f2fs_folio_put(node_folio, false);
1821         return err;
1822 }
1823
1824 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1825                         struct writeback_control *wbc, bool atomic,
1826                         unsigned int *seq_id)
1827 {
1828         pgoff_t index;
1829         struct folio_batch fbatch;
1830         int ret = 0;
1831         struct folio *last_folio = NULL;
1832         bool marked = false;
1833         nid_t ino = inode->i_ino;
1834         int nr_folios;
1835         int nwritten = 0;
1836
1837         if (atomic) {
1838                 last_folio = last_fsync_dnode(sbi, ino);
1839                 if (IS_ERR_OR_NULL(last_folio))
1840                         return PTR_ERR_OR_ZERO(last_folio);
1841         }
1842 retry:
1843         folio_batch_init(&fbatch);
1844         index = 0;
1845
1846         while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1847                                         (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1848                                         &fbatch))) {
1849                 int i;
1850
1851                 for (i = 0; i < nr_folios; i++) {
1852                         struct folio *folio = fbatch.folios[i];
1853                         bool submitted = false;
1854
1855                         if (unlikely(f2fs_cp_error(sbi))) {
1856                                 f2fs_folio_put(last_folio, false);
1857                                 folio_batch_release(&fbatch);
1858                                 ret = -EIO;
1859                                 goto out;
1860                         }
1861
1862                         if (!IS_DNODE(folio) || !is_cold_node(folio))
1863                                 continue;
1864                         if (ino_of_node(folio) != ino)
1865                                 continue;
1866
1867                         folio_lock(folio);
1868
1869                         if (unlikely(!is_node_folio(folio))) {
1870 continue_unlock:
1871                                 folio_unlock(folio);
1872                                 continue;
1873                         }
1874                         if (ino_of_node(folio) != ino)
1875                                 goto continue_unlock;
1876
1877                         if (!folio_test_dirty(folio) && folio != last_folio) {
1878                                 /* someone wrote it for us */
1879                                 goto continue_unlock;
1880                         }
1881
1882                         f2fs_folio_wait_writeback(folio, NODE, true, true);
1883
1884                         set_fsync_mark(folio, 0);
1885                         set_dentry_mark(folio, 0);
1886
1887                         if (!atomic || folio == last_folio) {
1888                                 set_fsync_mark(folio, 1);
1889                                 percpu_counter_inc(&sbi->rf_node_block_count);
1890                                 if (IS_INODE(folio)) {
1891                                         if (is_inode_flag_set(inode,
1892                                                                 FI_DIRTY_INODE))
1893                                                 f2fs_update_inode(inode, folio);
1894                                         set_dentry_mark(folio,
1895                                                 f2fs_need_dentry_mark(sbi, ino));
1896                                 }
1897                                 /* may be written by other thread */
1898                                 if (!folio_test_dirty(folio))
1899                                         folio_mark_dirty(folio);
1900                         }
1901
1902                         if (!folio_clear_dirty_for_io(folio))
1903                                 goto continue_unlock;
1904
1905                         if (!__write_node_folio(folio, atomic &&
1906                                                 folio == last_folio,
1907                                                 &submitted, wbc, true,
1908                                                 FS_NODE_IO, seq_id)) {
1909                                 f2fs_folio_put(last_folio, false);
1910                                 folio_batch_release(&fbatch);
1911                                 ret = -EIO;
1912                                 goto out;
1913                         }
1914                         if (submitted)
1915                                 nwritten++;
1916
1917                         if (folio == last_folio) {
1918                                 f2fs_folio_put(folio, false);
1919                                 folio_batch_release(&fbatch);
1920                                 marked = true;
1921                                 goto out;
1922                         }
1923                 }
1924                 folio_batch_release(&fbatch);
1925                 cond_resched();
1926         }
1927         if (atomic && !marked) {
1928                 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1929                            ino, last_folio->index);
1930                 folio_lock(last_folio);
1931                 f2fs_folio_wait_writeback(last_folio, NODE, true, true);
1932                 folio_mark_dirty(last_folio);
1933                 folio_unlock(last_folio);
1934                 goto retry;
1935         }
1936 out:
1937         if (nwritten)
1938                 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1939         return ret;
1940 }
1941
1942 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1943 {
1944         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1945         bool clean;
1946
1947         if (inode->i_ino != ino)
1948                 return 0;
1949
1950         if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1951                 return 0;
1952
1953         spin_lock(&sbi->inode_lock[DIRTY_META]);
1954         clean = list_empty(&F2FS_I(inode)->gdirty_list);
1955         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1956
1957         if (clean)
1958                 return 0;
1959
1960         inode = igrab(inode);
1961         if (!inode)
1962                 return 0;
1963         return 1;
1964 }
1965
1966 static bool flush_dirty_inode(struct folio *folio)
1967 {
1968         struct f2fs_sb_info *sbi = F2FS_F_SB(folio);
1969         struct inode *inode;
1970         nid_t ino = ino_of_node(folio);
1971
1972         inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1973         if (!inode)
1974                 return false;
1975
1976         f2fs_update_inode(inode, folio);
1977         folio_unlock(folio);
1978
1979         iput(inode);
1980         return true;
1981 }
1982
1983 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1984 {
1985         pgoff_t index = 0;
1986         struct folio_batch fbatch;
1987         int nr_folios;
1988
1989         folio_batch_init(&fbatch);
1990
1991         while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1992                                         (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1993                                         &fbatch))) {
1994                 int i;
1995
1996                 for (i = 0; i < nr_folios; i++) {
1997                         struct folio *folio = fbatch.folios[i];
1998
1999                         if (!IS_INODE(folio))
2000                                 continue;
2001
2002                         folio_lock(folio);
2003
2004                         if (unlikely(!is_node_folio(folio)))
2005                                 goto unlock;
2006                         if (!folio_test_dirty(folio))
2007                                 goto unlock;
2008
2009                         /* flush inline_data, if it's async context. */
2010                         if (folio_test_f2fs_inline(folio)) {
2011                                 folio_clear_f2fs_inline(folio);
2012                                 folio_unlock(folio);
2013                                 flush_inline_data(sbi, ino_of_node(folio));
2014                                 continue;
2015                         }
2016 unlock:
2017                         folio_unlock(folio);
2018                 }
2019                 folio_batch_release(&fbatch);
2020                 cond_resched();
2021         }
2022 }
2023
2024 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
2025                                 struct writeback_control *wbc,
2026                                 bool do_balance, enum iostat_type io_type)
2027 {
2028         pgoff_t index;
2029         struct folio_batch fbatch;
2030         int step = 0;
2031         int nwritten = 0;
2032         int ret = 0;
2033         int nr_folios, done = 0;
2034
2035         folio_batch_init(&fbatch);
2036
2037 next_step:
2038         index = 0;
2039
2040         while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi),
2041                                 &index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
2042                                 &fbatch))) {
2043                 int i;
2044
2045                 for (i = 0; i < nr_folios; i++) {
2046                         struct folio *folio = fbatch.folios[i];
2047                         bool submitted = false;
2048
2049                         /* give a priority to WB_SYNC threads */
2050                         if (atomic_read(&sbi->wb_sync_req[NODE]) &&
2051                                         wbc->sync_mode == WB_SYNC_NONE) {
2052                                 done = 1;
2053                                 break;
2054                         }
2055
2056                         /*
2057                          * flushing sequence with step:
2058                          * 0. indirect nodes
2059                          * 1. dentry dnodes
2060                          * 2. file dnodes
2061                          */
2062                         if (step == 0 && IS_DNODE(folio))
2063                                 continue;
2064                         if (step == 1 && (!IS_DNODE(folio) ||
2065                                                 is_cold_node(folio)))
2066                                 continue;
2067                         if (step == 2 && (!IS_DNODE(folio) ||
2068                                                 !is_cold_node(folio)))
2069                                 continue;
2070 lock_node:
2071                         if (wbc->sync_mode == WB_SYNC_ALL)
2072                                 folio_lock(folio);
2073                         else if (!folio_trylock(folio))
2074                                 continue;
2075
2076                         if (unlikely(!is_node_folio(folio))) {
2077 continue_unlock:
2078                                 folio_unlock(folio);
2079                                 continue;
2080                         }
2081
2082                         if (!folio_test_dirty(folio)) {
2083                                 /* someone wrote it for us */
2084                                 goto continue_unlock;
2085                         }
2086
2087                         /* flush inline_data/inode, if it's async context. */
2088                         if (!do_balance)
2089                                 goto write_node;
2090
2091                         /* flush inline_data */
2092                         if (folio_test_f2fs_inline(folio)) {
2093                                 folio_clear_f2fs_inline(folio);
2094                                 folio_unlock(folio);
2095                                 flush_inline_data(sbi, ino_of_node(folio));
2096                                 goto lock_node;
2097                         }
2098
2099                         /* flush dirty inode */
2100                         if (IS_INODE(folio) && flush_dirty_inode(folio))
2101                                 goto lock_node;
2102 write_node:
2103                         f2fs_folio_wait_writeback(folio, NODE, true, true);
2104
2105                         if (!folio_clear_dirty_for_io(folio))
2106                                 goto continue_unlock;
2107
2108                         set_fsync_mark(folio, 0);
2109                         set_dentry_mark(folio, 0);
2110
2111                         if (!__write_node_folio(folio, false, &submitted,
2112                                         wbc, do_balance, io_type, NULL)) {
2113                                 folio_batch_release(&fbatch);
2114                                 ret = -EIO;
2115                                 goto out;
2116                         }
2117                         if (submitted)
2118                                 nwritten++;
2119
2120                         if (--wbc->nr_to_write == 0)
2121                                 break;
2122                 }
2123                 folio_batch_release(&fbatch);
2124                 cond_resched();
2125
2126                 if (wbc->nr_to_write == 0) {
2127                         step = 2;
2128                         break;
2129                 }
2130         }
2131
2132         if (step < 2) {
2133                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2134                                 wbc->sync_mode == WB_SYNC_NONE && step == 1)
2135                         goto out;
2136                 step++;
2137                 goto next_step;
2138         }
2139 out:
2140         if (nwritten)
2141                 f2fs_submit_merged_write(sbi, NODE);
2142
2143         if (unlikely(f2fs_cp_error(sbi)))
2144                 return -EIO;
2145         return ret;
2146 }
2147
2148 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2149                                                 unsigned int seq_id)
2150 {
2151         struct fsync_node_entry *fn;
2152         struct list_head *head = &sbi->fsync_node_list;
2153         unsigned long flags;
2154         unsigned int cur_seq_id = 0;
2155
2156         while (seq_id && cur_seq_id < seq_id) {
2157                 struct folio *folio;
2158
2159                 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2160                 if (list_empty(head)) {
2161                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2162                         break;
2163                 }
2164                 fn = list_first_entry(head, struct fsync_node_entry, list);
2165                 if (fn->seq_id > seq_id) {
2166                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2167                         break;
2168                 }
2169                 cur_seq_id = fn->seq_id;
2170                 folio = fn->folio;
2171                 folio_get(folio);
2172                 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2173
2174                 f2fs_folio_wait_writeback(folio, NODE, true, false);
2175
2176                 folio_put(folio);
2177         }
2178
2179         return filemap_check_errors(NODE_MAPPING(sbi));
2180 }
2181
2182 static int f2fs_write_node_pages(struct address_space *mapping,
2183                             struct writeback_control *wbc)
2184 {
2185         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2186         struct blk_plug plug;
2187         long diff;
2188
2189         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2190                 goto skip_write;
2191
2192         /* balancing f2fs's metadata in background */
2193         f2fs_balance_fs_bg(sbi, true);
2194
2195         /* collect a number of dirty node pages and write together */
2196         if (wbc->sync_mode != WB_SYNC_ALL &&
2197                         get_pages(sbi, F2FS_DIRTY_NODES) <
2198                                         nr_pages_to_skip(sbi, NODE))
2199                 goto skip_write;
2200
2201         if (wbc->sync_mode == WB_SYNC_ALL)
2202                 atomic_inc(&sbi->wb_sync_req[NODE]);
2203         else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2204                 /* to avoid potential deadlock */
2205                 if (current->plug)
2206                         blk_finish_plug(current->plug);
2207                 goto skip_write;
2208         }
2209
2210         trace_f2fs_writepages(mapping->host, wbc, NODE);
2211
2212         diff = nr_pages_to_write(sbi, NODE, wbc);
2213         blk_start_plug(&plug);
2214         f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2215         blk_finish_plug(&plug);
2216         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2217
2218         if (wbc->sync_mode == WB_SYNC_ALL)
2219                 atomic_dec(&sbi->wb_sync_req[NODE]);
2220         return 0;
2221
2222 skip_write:
2223         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2224         trace_f2fs_writepages(mapping->host, wbc, NODE);
2225         return 0;
2226 }
2227
2228 static bool f2fs_dirty_node_folio(struct address_space *mapping,
2229                 struct folio *folio)
2230 {
2231         trace_f2fs_set_page_dirty(folio, NODE);
2232
2233         if (!folio_test_uptodate(folio))
2234                 folio_mark_uptodate(folio);
2235 #ifdef CONFIG_F2FS_CHECK_FS
2236         if (IS_INODE(folio))
2237                 f2fs_inode_chksum_set(F2FS_M_SB(mapping), folio);
2238 #endif
2239         if (filemap_dirty_folio(mapping, folio)) {
2240                 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2241                 folio_set_f2fs_reference(folio);
2242                 return true;
2243         }
2244         return false;
2245 }
2246
2247 /*
2248  * Structure of the f2fs node operations
2249  */
2250 const struct address_space_operations f2fs_node_aops = {
2251         .writepages     = f2fs_write_node_pages,
2252         .dirty_folio    = f2fs_dirty_node_folio,
2253         .invalidate_folio = f2fs_invalidate_folio,
2254         .release_folio  = f2fs_release_folio,
2255         .migrate_folio  = filemap_migrate_folio,
2256 };
2257
2258 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2259                                                 nid_t n)
2260 {
2261         return radix_tree_lookup(&nm_i->free_nid_root, n);
2262 }
2263
2264 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2265                                 struct free_nid *i)
2266 {
2267         struct f2fs_nm_info *nm_i = NM_I(sbi);
2268         int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2269
2270         if (err)
2271                 return err;
2272
2273         nm_i->nid_cnt[FREE_NID]++;
2274         list_add_tail(&i->list, &nm_i->free_nid_list);
2275         return 0;
2276 }
2277
2278 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2279                         struct free_nid *i, enum nid_state state)
2280 {
2281         struct f2fs_nm_info *nm_i = NM_I(sbi);
2282
2283         f2fs_bug_on(sbi, state != i->state);
2284         nm_i->nid_cnt[state]--;
2285         if (state == FREE_NID)
2286                 list_del(&i->list);
2287         radix_tree_delete(&nm_i->free_nid_root, i->nid);
2288 }
2289
2290 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2291                         enum nid_state org_state, enum nid_state dst_state)
2292 {
2293         struct f2fs_nm_info *nm_i = NM_I(sbi);
2294
2295         f2fs_bug_on(sbi, org_state != i->state);
2296         i->state = dst_state;
2297         nm_i->nid_cnt[org_state]--;
2298         nm_i->nid_cnt[dst_state]++;
2299
2300         switch (dst_state) {
2301         case PREALLOC_NID:
2302                 list_del(&i->list);
2303                 break;
2304         case FREE_NID:
2305                 list_add_tail(&i->list, &nm_i->free_nid_list);
2306                 break;
2307         default:
2308                 BUG_ON(1);
2309         }
2310 }
2311
2312 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2313                                                         bool set, bool build)
2314 {
2315         struct f2fs_nm_info *nm_i = NM_I(sbi);
2316         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2317         unsigned int nid_ofs = nid - START_NID(nid);
2318
2319         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2320                 return;
2321
2322         if (set) {
2323                 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2324                         return;
2325                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2326                 nm_i->free_nid_count[nat_ofs]++;
2327         } else {
2328                 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2329                         return;
2330                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2331                 if (!build)
2332                         nm_i->free_nid_count[nat_ofs]--;
2333         }
2334 }
2335
2336 /* return if the nid is recognized as free */
2337 static bool add_free_nid(struct f2fs_sb_info *sbi,
2338                                 nid_t nid, bool build, bool update)
2339 {
2340         struct f2fs_nm_info *nm_i = NM_I(sbi);
2341         struct free_nid *i, *e;
2342         struct nat_entry *ne;
2343         int err;
2344         bool ret = false;
2345
2346         /* 0 nid should not be used */
2347         if (unlikely(nid == 0))
2348                 return false;
2349
2350         if (unlikely(f2fs_check_nid_range(sbi, nid)))
2351                 return false;
2352
2353         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2354         i->nid = nid;
2355         i->state = FREE_NID;
2356
2357         err = radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2358         f2fs_bug_on(sbi, err);
2359
2360         err = -EINVAL;
2361
2362         spin_lock(&nm_i->nid_list_lock);
2363
2364         if (build) {
2365                 /*
2366                  *   Thread A             Thread B
2367                  *  - f2fs_create
2368                  *   - f2fs_new_inode
2369                  *    - f2fs_alloc_nid
2370                  *     - __insert_nid_to_list(PREALLOC_NID)
2371                  *                     - f2fs_balance_fs_bg
2372                  *                      - f2fs_build_free_nids
2373                  *                       - __f2fs_build_free_nids
2374                  *                        - scan_nat_page
2375                  *                         - add_free_nid
2376                  *                          - __lookup_nat_cache
2377                  *  - f2fs_add_link
2378                  *   - f2fs_init_inode_metadata
2379                  *    - f2fs_new_inode_folio
2380                  *     - f2fs_new_node_folio
2381                  *      - set_node_addr
2382                  *  - f2fs_alloc_nid_done
2383                  *   - __remove_nid_from_list(PREALLOC_NID)
2384                  *                         - __insert_nid_to_list(FREE_NID)
2385                  */
2386                 ne = __lookup_nat_cache(nm_i, nid, false);
2387                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2388                                 nat_get_blkaddr(ne) != NULL_ADDR))
2389                         goto err_out;
2390
2391                 e = __lookup_free_nid_list(nm_i, nid);
2392                 if (e) {
2393                         if (e->state == FREE_NID)
2394                                 ret = true;
2395                         goto err_out;
2396                 }
2397         }
2398         ret = true;
2399         err = __insert_free_nid(sbi, i);
2400 err_out:
2401         if (update) {
2402                 update_free_nid_bitmap(sbi, nid, ret, build);
2403                 if (!build)
2404                         nm_i->available_nids++;
2405         }
2406         spin_unlock(&nm_i->nid_list_lock);
2407         radix_tree_preload_end();
2408
2409         if (err)
2410                 kmem_cache_free(free_nid_slab, i);
2411         return ret;
2412 }
2413
2414 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2415 {
2416         struct f2fs_nm_info *nm_i = NM_I(sbi);
2417         struct free_nid *i;
2418         bool need_free = false;
2419
2420         spin_lock(&nm_i->nid_list_lock);
2421         i = __lookup_free_nid_list(nm_i, nid);
2422         if (i && i->state == FREE_NID) {
2423                 __remove_free_nid(sbi, i, FREE_NID);
2424                 need_free = true;
2425         }
2426         spin_unlock(&nm_i->nid_list_lock);
2427
2428         if (need_free)
2429                 kmem_cache_free(free_nid_slab, i);
2430 }
2431
2432 static int scan_nat_page(struct f2fs_sb_info *sbi,
2433                         struct f2fs_nat_block *nat_blk, nid_t start_nid)
2434 {
2435         struct f2fs_nm_info *nm_i = NM_I(sbi);
2436         block_t blk_addr;
2437         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2438         int i;
2439
2440         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2441
2442         i = start_nid % NAT_ENTRY_PER_BLOCK;
2443
2444         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2445                 if (unlikely(start_nid >= nm_i->max_nid))
2446                         break;
2447
2448                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2449
2450                 if (blk_addr == NEW_ADDR)
2451                         return -EFSCORRUPTED;
2452
2453                 if (blk_addr == NULL_ADDR) {
2454                         add_free_nid(sbi, start_nid, true, true);
2455                 } else {
2456                         spin_lock(&NM_I(sbi)->nid_list_lock);
2457                         update_free_nid_bitmap(sbi, start_nid, false, true);
2458                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2459                 }
2460         }
2461
2462         return 0;
2463 }
2464
2465 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2466 {
2467         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2468         struct f2fs_journal *journal = curseg->journal;
2469         int i;
2470
2471         down_read(&curseg->journal_rwsem);
2472         for (i = 0; i < nats_in_cursum(journal); i++) {
2473                 block_t addr;
2474                 nid_t nid;
2475
2476                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2477                 nid = le32_to_cpu(nid_in_journal(journal, i));
2478                 if (addr == NULL_ADDR)
2479                         add_free_nid(sbi, nid, true, false);
2480                 else
2481                         remove_free_nid(sbi, nid);
2482         }
2483         up_read(&curseg->journal_rwsem);
2484 }
2485
2486 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2487 {
2488         struct f2fs_nm_info *nm_i = NM_I(sbi);
2489         unsigned int i, idx;
2490         nid_t nid;
2491
2492         f2fs_down_read(&nm_i->nat_tree_lock);
2493
2494         for (i = 0; i < nm_i->nat_blocks; i++) {
2495                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2496                         continue;
2497                 if (!nm_i->free_nid_count[i])
2498                         continue;
2499                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2500                         idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2501                                                 NAT_ENTRY_PER_BLOCK, idx);
2502                         if (idx >= NAT_ENTRY_PER_BLOCK)
2503                                 break;
2504
2505                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
2506                         add_free_nid(sbi, nid, true, false);
2507
2508                         if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2509                                 goto out;
2510                 }
2511         }
2512 out:
2513         scan_curseg_cache(sbi);
2514
2515         f2fs_up_read(&nm_i->nat_tree_lock);
2516 }
2517
2518 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2519                                                 bool sync, bool mount)
2520 {
2521         struct f2fs_nm_info *nm_i = NM_I(sbi);
2522         int i = 0, ret;
2523         nid_t nid = nm_i->next_scan_nid;
2524
2525         if (unlikely(nid >= nm_i->max_nid))
2526                 nid = 0;
2527
2528         if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2529                 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2530
2531         /* Enough entries */
2532         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2533                 return 0;
2534
2535         if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2536                 return 0;
2537
2538         if (!mount) {
2539                 /* try to find free nids in free_nid_bitmap */
2540                 scan_free_nid_bits(sbi);
2541
2542                 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2543                         return 0;
2544         }
2545
2546         /* readahead nat pages to be scanned */
2547         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2548                                                         META_NAT, true);
2549
2550         f2fs_down_read(&nm_i->nat_tree_lock);
2551
2552         while (1) {
2553                 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2554                                                 nm_i->nat_block_bitmap)) {
2555                         struct folio *folio = get_current_nat_folio(sbi, nid);
2556
2557                         if (IS_ERR(folio)) {
2558                                 ret = PTR_ERR(folio);
2559                         } else {
2560                                 ret = scan_nat_page(sbi, folio_address(folio),
2561                                                 nid);
2562                                 f2fs_folio_put(folio, true);
2563                         }
2564
2565                         if (ret) {
2566                                 f2fs_up_read(&nm_i->nat_tree_lock);
2567
2568                                 if (ret == -EFSCORRUPTED) {
2569                                         f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2570                                         set_sbi_flag(sbi, SBI_NEED_FSCK);
2571                                         f2fs_handle_error(sbi,
2572                                                 ERROR_INCONSISTENT_NAT);
2573                                 }
2574
2575                                 return ret;
2576                         }
2577                 }
2578
2579                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2580                 if (unlikely(nid >= nm_i->max_nid))
2581                         nid = 0;
2582
2583                 if (++i >= FREE_NID_PAGES)
2584                         break;
2585         }
2586
2587         /* go to the next free nat pages to find free nids abundantly */
2588         nm_i->next_scan_nid = nid;
2589
2590         /* find free nids from current sum_pages */
2591         scan_curseg_cache(sbi);
2592
2593         f2fs_up_read(&nm_i->nat_tree_lock);
2594
2595         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2596                                         nm_i->ra_nid_pages, META_NAT, false);
2597
2598         return 0;
2599 }
2600
2601 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2602 {
2603         int ret;
2604
2605         mutex_lock(&NM_I(sbi)->build_lock);
2606         ret = __f2fs_build_free_nids(sbi, sync, mount);
2607         mutex_unlock(&NM_I(sbi)->build_lock);
2608
2609         return ret;
2610 }
2611
2612 /*
2613  * If this function returns success, caller can obtain a new nid
2614  * from second parameter of this function.
2615  * The returned nid could be used ino as well as nid when inode is created.
2616  */
2617 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2618 {
2619         struct f2fs_nm_info *nm_i = NM_I(sbi);
2620         struct free_nid *i = NULL;
2621 retry:
2622         if (time_to_inject(sbi, FAULT_ALLOC_NID))
2623                 return false;
2624
2625         spin_lock(&nm_i->nid_list_lock);
2626
2627         if (unlikely(nm_i->available_nids == 0)) {
2628                 spin_unlock(&nm_i->nid_list_lock);
2629                 return false;
2630         }
2631
2632         /* We should not use stale free nids created by f2fs_build_free_nids */
2633         if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2634                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2635                 i = list_first_entry(&nm_i->free_nid_list,
2636                                         struct free_nid, list);
2637                 *nid = i->nid;
2638
2639                 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2640                 nm_i->available_nids--;
2641
2642                 update_free_nid_bitmap(sbi, *nid, false, false);
2643
2644                 spin_unlock(&nm_i->nid_list_lock);
2645                 return true;
2646         }
2647         spin_unlock(&nm_i->nid_list_lock);
2648
2649         /* Let's scan nat pages and its caches to get free nids */
2650         if (!f2fs_build_free_nids(sbi, true, false))
2651                 goto retry;
2652         return false;
2653 }
2654
2655 /*
2656  * f2fs_alloc_nid() should be called prior to this function.
2657  */
2658 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2659 {
2660         struct f2fs_nm_info *nm_i = NM_I(sbi);
2661         struct free_nid *i;
2662
2663         spin_lock(&nm_i->nid_list_lock);
2664         i = __lookup_free_nid_list(nm_i, nid);
2665         f2fs_bug_on(sbi, !i);
2666         __remove_free_nid(sbi, i, PREALLOC_NID);
2667         spin_unlock(&nm_i->nid_list_lock);
2668
2669         kmem_cache_free(free_nid_slab, i);
2670 }
2671
2672 /*
2673  * f2fs_alloc_nid() should be called prior to this function.
2674  */
2675 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2676 {
2677         struct f2fs_nm_info *nm_i = NM_I(sbi);
2678         struct free_nid *i;
2679         bool need_free = false;
2680
2681         if (!nid)
2682                 return;
2683
2684         spin_lock(&nm_i->nid_list_lock);
2685         i = __lookup_free_nid_list(nm_i, nid);
2686         f2fs_bug_on(sbi, !i);
2687
2688         if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2689                 __remove_free_nid(sbi, i, PREALLOC_NID);
2690                 need_free = true;
2691         } else {
2692                 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2693         }
2694
2695         nm_i->available_nids++;
2696
2697         update_free_nid_bitmap(sbi, nid, true, false);
2698
2699         spin_unlock(&nm_i->nid_list_lock);
2700
2701         if (need_free)
2702                 kmem_cache_free(free_nid_slab, i);
2703 }
2704
2705 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2706 {
2707         struct f2fs_nm_info *nm_i = NM_I(sbi);
2708         int nr = nr_shrink;
2709
2710         if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2711                 return 0;
2712
2713         if (!mutex_trylock(&nm_i->build_lock))
2714                 return 0;
2715
2716         while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2717                 struct free_nid *i, *next;
2718                 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2719
2720                 spin_lock(&nm_i->nid_list_lock);
2721                 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2722                         if (!nr_shrink || !batch ||
2723                                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2724                                 break;
2725                         __remove_free_nid(sbi, i, FREE_NID);
2726                         kmem_cache_free(free_nid_slab, i);
2727                         nr_shrink--;
2728                         batch--;
2729                 }
2730                 spin_unlock(&nm_i->nid_list_lock);
2731         }
2732
2733         mutex_unlock(&nm_i->build_lock);
2734
2735         return nr - nr_shrink;
2736 }
2737
2738 int f2fs_recover_inline_xattr(struct inode *inode, struct folio *folio)
2739 {
2740         void *src_addr, *dst_addr;
2741         size_t inline_size;
2742         struct folio *ifolio;
2743         struct f2fs_inode *ri;
2744
2745         ifolio = f2fs_get_inode_folio(F2FS_I_SB(inode), inode->i_ino);
2746         if (IS_ERR(ifolio))
2747                 return PTR_ERR(ifolio);
2748
2749         ri = F2FS_INODE(folio);
2750         if (ri->i_inline & F2FS_INLINE_XATTR) {
2751                 if (!f2fs_has_inline_xattr(inode)) {
2752                         set_inode_flag(inode, FI_INLINE_XATTR);
2753                         stat_inc_inline_xattr(inode);
2754                 }
2755         } else {
2756                 if (f2fs_has_inline_xattr(inode)) {
2757                         stat_dec_inline_xattr(inode);
2758                         clear_inode_flag(inode, FI_INLINE_XATTR);
2759                 }
2760                 goto update_inode;
2761         }
2762
2763         dst_addr = inline_xattr_addr(inode, ifolio);
2764         src_addr = inline_xattr_addr(inode, folio);
2765         inline_size = inline_xattr_size(inode);
2766
2767         f2fs_folio_wait_writeback(ifolio, NODE, true, true);
2768         memcpy(dst_addr, src_addr, inline_size);
2769 update_inode:
2770         f2fs_update_inode(inode, ifolio);
2771         f2fs_folio_put(ifolio, true);
2772         return 0;
2773 }
2774
2775 int f2fs_recover_xattr_data(struct inode *inode, struct folio *folio)
2776 {
2777         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2778         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2779         nid_t new_xnid;
2780         struct dnode_of_data dn;
2781         struct node_info ni;
2782         struct folio *xfolio;
2783         int err;
2784
2785         if (!prev_xnid)
2786                 goto recover_xnid;
2787
2788         /* 1: invalidate the previous xattr nid */
2789         err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2790         if (err)
2791                 return err;
2792
2793         f2fs_invalidate_blocks(sbi, ni.blk_addr, 1);
2794         dec_valid_node_count(sbi, inode, false);
2795         set_node_addr(sbi, &ni, NULL_ADDR, false);
2796
2797 recover_xnid:
2798         /* 2: update xattr nid in inode */
2799         if (!f2fs_alloc_nid(sbi, &new_xnid))
2800                 return -ENOSPC;
2801
2802         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2803         xfolio = f2fs_new_node_folio(&dn, XATTR_NODE_OFFSET);
2804         if (IS_ERR(xfolio)) {
2805                 f2fs_alloc_nid_failed(sbi, new_xnid);
2806                 return PTR_ERR(xfolio);
2807         }
2808
2809         f2fs_alloc_nid_done(sbi, new_xnid);
2810         f2fs_update_inode_page(inode);
2811
2812         /* 3: update and set xattr node page dirty */
2813         if (folio) {
2814                 memcpy(F2FS_NODE(xfolio), F2FS_NODE(folio),
2815                                 VALID_XATTR_BLOCK_SIZE);
2816                 folio_mark_dirty(xfolio);
2817         }
2818         f2fs_folio_put(xfolio, true);
2819
2820         return 0;
2821 }
2822
2823 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct folio *folio)
2824 {
2825         struct f2fs_inode *src, *dst;
2826         nid_t ino = ino_of_node(folio);
2827         struct node_info old_ni, new_ni;
2828         struct folio *ifolio;
2829         int err;
2830
2831         err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2832         if (err)
2833                 return err;
2834
2835         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2836                 return -EINVAL;
2837 retry:
2838         ifolio = f2fs_grab_cache_folio(NODE_MAPPING(sbi), ino, false);
2839         if (IS_ERR(ifolio)) {
2840                 memalloc_retry_wait(GFP_NOFS);
2841                 goto retry;
2842         }
2843
2844         /* Should not use this inode from free nid list */
2845         remove_free_nid(sbi, ino);
2846
2847         if (!folio_test_uptodate(ifolio))
2848                 folio_mark_uptodate(ifolio);
2849         fill_node_footer(ifolio, ino, ino, 0, true);
2850         set_cold_node(ifolio, false);
2851
2852         src = F2FS_INODE(folio);
2853         dst = F2FS_INODE(ifolio);
2854
2855         memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2856         dst->i_size = 0;
2857         dst->i_blocks = cpu_to_le64(1);
2858         dst->i_links = cpu_to_le32(1);
2859         dst->i_xattr_nid = 0;
2860         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2861         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2862                 dst->i_extra_isize = src->i_extra_isize;
2863
2864                 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2865                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2866                                                         i_inline_xattr_size))
2867                         dst->i_inline_xattr_size = src->i_inline_xattr_size;
2868
2869                 if (f2fs_sb_has_project_quota(sbi) &&
2870                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2871                                                                 i_projid))
2872                         dst->i_projid = src->i_projid;
2873
2874                 if (f2fs_sb_has_inode_crtime(sbi) &&
2875                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2876                                                         i_crtime_nsec)) {
2877                         dst->i_crtime = src->i_crtime;
2878                         dst->i_crtime_nsec = src->i_crtime_nsec;
2879                 }
2880         }
2881
2882         new_ni = old_ni;
2883         new_ni.ino = ino;
2884
2885         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2886                 WARN_ON(1);
2887         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2888         inc_valid_inode_count(sbi);
2889         folio_mark_dirty(ifolio);
2890         f2fs_folio_put(ifolio, true);
2891         return 0;
2892 }
2893
2894 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2895                         unsigned int segno, struct f2fs_summary_block *sum)
2896 {
2897         struct f2fs_node *rn;
2898         struct f2fs_summary *sum_entry;
2899         block_t addr;
2900         int i, idx, last_offset, nrpages;
2901
2902         /* scan the node segment */
2903         last_offset = BLKS_PER_SEG(sbi);
2904         addr = START_BLOCK(sbi, segno);
2905         sum_entry = &sum->entries[0];
2906
2907         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2908                 nrpages = bio_max_segs(last_offset - i);
2909
2910                 /* readahead node pages */
2911                 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2912
2913                 for (idx = addr; idx < addr + nrpages; idx++) {
2914                         struct folio *folio = f2fs_get_tmp_folio(sbi, idx);
2915
2916                         if (IS_ERR(folio))
2917                                 return PTR_ERR(folio);
2918
2919                         rn = F2FS_NODE(folio);
2920                         sum_entry->nid = rn->footer.nid;
2921                         sum_entry->version = 0;
2922                         sum_entry->ofs_in_node = 0;
2923                         sum_entry++;
2924                         f2fs_folio_put(folio, true);
2925                 }
2926
2927                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2928                                                         addr + nrpages);
2929         }
2930         return 0;
2931 }
2932
2933 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2934 {
2935         struct f2fs_nm_info *nm_i = NM_I(sbi);
2936         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2937         struct f2fs_journal *journal = curseg->journal;
2938         int i;
2939         bool init_dirty;
2940
2941         down_write(&curseg->journal_rwsem);
2942         for (i = 0; i < nats_in_cursum(journal); i++) {
2943                 struct nat_entry *ne;
2944                 struct f2fs_nat_entry raw_ne;
2945                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2946
2947                 if (f2fs_check_nid_range(sbi, nid))
2948                         continue;
2949
2950                 init_dirty = false;
2951
2952                 raw_ne = nat_in_journal(journal, i);
2953
2954                 ne = __lookup_nat_cache(nm_i, nid, true);
2955                 if (!ne) {
2956                         init_dirty = true;
2957                         ne = __alloc_nat_entry(sbi, nid, true);
2958                         __init_nat_entry(nm_i, ne, &raw_ne, true, true);
2959                 }
2960
2961                 /*
2962                  * if a free nat in journal has not been used after last
2963                  * checkpoint, we should remove it from available nids,
2964                  * since later we will add it again.
2965                  */
2966                 if (!get_nat_flag(ne, IS_DIRTY) &&
2967                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2968                         spin_lock(&nm_i->nid_list_lock);
2969                         nm_i->available_nids--;
2970                         spin_unlock(&nm_i->nid_list_lock);
2971                 }
2972
2973                 __set_nat_cache_dirty(nm_i, ne, init_dirty);
2974         }
2975         update_nats_in_cursum(journal, -i);
2976         up_write(&curseg->journal_rwsem);
2977 }
2978
2979 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2980                                                 struct list_head *head, int max)
2981 {
2982         struct nat_entry_set *cur;
2983
2984         if (nes->entry_cnt >= max)
2985                 goto add_out;
2986
2987         list_for_each_entry(cur, head, set_list) {
2988                 if (cur->entry_cnt >= nes->entry_cnt) {
2989                         list_add(&nes->set_list, cur->set_list.prev);
2990                         return;
2991                 }
2992         }
2993 add_out:
2994         list_add_tail(&nes->set_list, head);
2995 }
2996
2997 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2998                 const struct f2fs_nat_block *nat_blk)
2999 {
3000         struct f2fs_nm_info *nm_i = NM_I(sbi);
3001         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
3002         int valid = 0;
3003         int i = 0;
3004
3005         if (!enabled_nat_bits(sbi, NULL))
3006                 return;
3007
3008         if (nat_index == 0) {
3009                 valid = 1;
3010                 i = 1;
3011         }
3012         for (; i < NAT_ENTRY_PER_BLOCK; i++) {
3013                 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
3014                         valid++;
3015         }
3016         if (valid == 0) {
3017                 __set_bit_le(nat_index, nm_i->empty_nat_bits);
3018                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
3019                 return;
3020         }
3021
3022         __clear_bit_le(nat_index, nm_i->empty_nat_bits);
3023         if (valid == NAT_ENTRY_PER_BLOCK)
3024                 __set_bit_le(nat_index, nm_i->full_nat_bits);
3025         else
3026                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
3027 }
3028
3029 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
3030                 struct nat_entry_set *set, struct cp_control *cpc)
3031 {
3032         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3033         struct f2fs_journal *journal = curseg->journal;
3034         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
3035         bool to_journal = true;
3036         struct f2fs_nat_block *nat_blk;
3037         struct nat_entry *ne, *cur;
3038         struct folio *folio = NULL;
3039
3040         /*
3041          * there are two steps to flush nat entries:
3042          * #1, flush nat entries to journal in current hot data summary block.
3043          * #2, flush nat entries to nat page.
3044          */
3045         if (enabled_nat_bits(sbi, cpc) ||
3046                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3047                 to_journal = false;
3048
3049         if (to_journal) {
3050                 down_write(&curseg->journal_rwsem);
3051         } else {
3052                 folio = get_next_nat_folio(sbi, start_nid);
3053                 if (IS_ERR(folio))
3054                         return PTR_ERR(folio);
3055
3056                 nat_blk = folio_address(folio);
3057                 f2fs_bug_on(sbi, !nat_blk);
3058         }
3059
3060         /* flush dirty nats in nat entry set */
3061         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3062                 struct f2fs_nat_entry *raw_ne;
3063                 nid_t nid = nat_get_nid(ne);
3064                 int offset;
3065
3066                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3067
3068                 if (to_journal) {
3069                         offset = f2fs_lookup_journal_in_cursum(journal,
3070                                                         NAT_JOURNAL, nid, 1);
3071                         f2fs_bug_on(sbi, offset < 0);
3072                         raw_ne = &nat_in_journal(journal, offset);
3073                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
3074                 } else {
3075                         raw_ne = &nat_blk->entries[nid - start_nid];
3076                 }
3077                 raw_nat_from_node_info(raw_ne, &ne->ni);
3078                 nat_reset_flag(ne);
3079                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
3080                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
3081                         add_free_nid(sbi, nid, false, true);
3082                 } else {
3083                         spin_lock(&NM_I(sbi)->nid_list_lock);
3084                         update_free_nid_bitmap(sbi, nid, false, false);
3085                         spin_unlock(&NM_I(sbi)->nid_list_lock);
3086                 }
3087         }
3088
3089         if (to_journal) {
3090                 up_write(&curseg->journal_rwsem);
3091         } else {
3092                 __update_nat_bits(sbi, start_nid, nat_blk);
3093                 f2fs_folio_put(folio, true);
3094         }
3095
3096         /* Allow dirty nats by node block allocation in write_begin */
3097         if (!set->entry_cnt) {
3098                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3099                 kmem_cache_free(nat_entry_set_slab, set);
3100         }
3101         return 0;
3102 }
3103
3104 /*
3105  * This function is called during the checkpointing process.
3106  */
3107 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3108 {
3109         struct f2fs_nm_info *nm_i = NM_I(sbi);
3110         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3111         struct f2fs_journal *journal = curseg->journal;
3112         struct nat_entry_set *setvec[NAT_VEC_SIZE];
3113         struct nat_entry_set *set, *tmp;
3114         unsigned int found;
3115         nid_t set_idx = 0;
3116         LIST_HEAD(sets);
3117         int err = 0;
3118
3119         /*
3120          * during unmount, let's flush nat_bits before checking
3121          * nat_cnt[DIRTY_NAT].
3122          */
3123         if (enabled_nat_bits(sbi, cpc)) {
3124                 f2fs_down_write(&nm_i->nat_tree_lock);
3125                 remove_nats_in_journal(sbi);
3126                 f2fs_up_write(&nm_i->nat_tree_lock);
3127         }
3128
3129         if (!nm_i->nat_cnt[DIRTY_NAT])
3130                 return 0;
3131
3132         f2fs_down_write(&nm_i->nat_tree_lock);
3133
3134         /*
3135          * if there are no enough space in journal to store dirty nat
3136          * entries, remove all entries from journal and merge them
3137          * into nat entry set.
3138          */
3139         if (enabled_nat_bits(sbi, cpc) ||
3140                 !__has_cursum_space(journal,
3141                         nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3142                 remove_nats_in_journal(sbi);
3143
3144         while ((found = __gang_lookup_nat_set(nm_i,
3145                                         set_idx, NAT_VEC_SIZE, setvec))) {
3146                 unsigned idx;
3147
3148                 set_idx = setvec[found - 1]->set + 1;
3149                 for (idx = 0; idx < found; idx++)
3150                         __adjust_nat_entry_set(setvec[idx], &sets,
3151                                                 MAX_NAT_JENTRIES(journal));
3152         }
3153
3154         /* flush dirty nats in nat entry set */
3155         list_for_each_entry_safe(set, tmp, &sets, set_list) {
3156                 err = __flush_nat_entry_set(sbi, set, cpc);
3157                 if (err)
3158                         break;
3159         }
3160
3161         f2fs_up_write(&nm_i->nat_tree_lock);
3162         /* Allow dirty nats by node block allocation in write_begin */
3163
3164         return err;
3165 }
3166
3167 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3168 {
3169         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3170         struct f2fs_nm_info *nm_i = NM_I(sbi);
3171         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3172         unsigned int i;
3173         __u64 cp_ver = cur_cp_version(ckpt);
3174         block_t nat_bits_addr;
3175
3176         if (!enabled_nat_bits(sbi, NULL))
3177                 return 0;
3178
3179         nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3180         nm_i->nat_bits = f2fs_kvzalloc(sbi,
3181                         F2FS_BLK_TO_BYTES(nm_i->nat_bits_blocks), GFP_KERNEL);
3182         if (!nm_i->nat_bits)
3183                 return -ENOMEM;
3184
3185         nat_bits_addr = __start_cp_addr(sbi) + BLKS_PER_SEG(sbi) -
3186                                                 nm_i->nat_bits_blocks;
3187         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3188                 struct folio *folio;
3189
3190                 folio = f2fs_get_meta_folio(sbi, nat_bits_addr++);
3191                 if (IS_ERR(folio))
3192                         return PTR_ERR(folio);
3193
3194                 memcpy(nm_i->nat_bits + F2FS_BLK_TO_BYTES(i),
3195                                         folio_address(folio), F2FS_BLKSIZE);
3196                 f2fs_folio_put(folio, true);
3197         }
3198
3199         cp_ver |= (cur_cp_crc(ckpt) << 32);
3200         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3201                 disable_nat_bits(sbi, true);
3202                 return 0;
3203         }
3204
3205         nm_i->full_nat_bits = nm_i->nat_bits + 8;
3206         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3207
3208         f2fs_notice(sbi, "Found nat_bits in checkpoint");
3209         return 0;
3210 }
3211
3212 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3213 {
3214         struct f2fs_nm_info *nm_i = NM_I(sbi);
3215         unsigned int i = 0;
3216         nid_t nid, last_nid;
3217
3218         if (!enabled_nat_bits(sbi, NULL))
3219                 return;
3220
3221         for (i = 0; i < nm_i->nat_blocks; i++) {
3222                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3223                 if (i >= nm_i->nat_blocks)
3224                         break;
3225
3226                 __set_bit_le(i, nm_i->nat_block_bitmap);
3227
3228                 nid = i * NAT_ENTRY_PER_BLOCK;
3229                 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3230
3231                 spin_lock(&NM_I(sbi)->nid_list_lock);
3232                 for (; nid < last_nid; nid++)
3233                         update_free_nid_bitmap(sbi, nid, true, true);
3234                 spin_unlock(&NM_I(sbi)->nid_list_lock);
3235         }
3236
3237         for (i = 0; i < nm_i->nat_blocks; i++) {
3238                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3239                 if (i >= nm_i->nat_blocks)
3240                         break;
3241
3242                 __set_bit_le(i, nm_i->nat_block_bitmap);
3243         }
3244 }
3245
3246 static int init_node_manager(struct f2fs_sb_info *sbi)
3247 {
3248         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3249         struct f2fs_nm_info *nm_i = NM_I(sbi);
3250         unsigned char *version_bitmap;
3251         unsigned int nat_segs;
3252         int err;
3253
3254         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3255
3256         /* segment_count_nat includes pair segment so divide to 2. */
3257         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3258         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3259         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3260
3261         /* not used nids: 0, node, meta, (and root counted as valid node) */
3262         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3263                                                 F2FS_RESERVED_NODE_NUM;
3264         nm_i->nid_cnt[FREE_NID] = 0;
3265         nm_i->nid_cnt[PREALLOC_NID] = 0;
3266         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3267         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3268         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3269         nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3270
3271         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3272         INIT_LIST_HEAD(&nm_i->free_nid_list);
3273         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3274         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3275         INIT_LIST_HEAD(&nm_i->nat_entries);
3276         spin_lock_init(&nm_i->nat_list_lock);
3277
3278         mutex_init(&nm_i->build_lock);
3279         spin_lock_init(&nm_i->nid_list_lock);
3280         init_f2fs_rwsem(&nm_i->nat_tree_lock);
3281
3282         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3283         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3284         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3285         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3286                                         GFP_KERNEL);
3287         if (!nm_i->nat_bitmap)
3288                 return -ENOMEM;
3289
3290         if (!test_opt(sbi, NAT_BITS))
3291                 disable_nat_bits(sbi, true);
3292
3293         err = __get_nat_bitmaps(sbi);
3294         if (err)
3295                 return err;
3296
3297 #ifdef CONFIG_F2FS_CHECK_FS
3298         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3299                                         GFP_KERNEL);
3300         if (!nm_i->nat_bitmap_mir)
3301                 return -ENOMEM;
3302 #endif
3303
3304         return 0;
3305 }
3306
3307 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3308 {
3309         struct f2fs_nm_info *nm_i = NM_I(sbi);
3310         int i;
3311
3312         nm_i->free_nid_bitmap =
3313                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3314                                               nm_i->nat_blocks),
3315                               GFP_KERNEL);
3316         if (!nm_i->free_nid_bitmap)
3317                 return -ENOMEM;
3318
3319         for (i = 0; i < nm_i->nat_blocks; i++) {
3320                 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3321                         f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3322                 if (!nm_i->free_nid_bitmap[i])
3323                         return -ENOMEM;
3324         }
3325
3326         nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3327                                                                 GFP_KERNEL);
3328         if (!nm_i->nat_block_bitmap)
3329                 return -ENOMEM;
3330
3331         nm_i->free_nid_count =
3332                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3333                                               nm_i->nat_blocks),
3334                               GFP_KERNEL);
3335         if (!nm_i->free_nid_count)
3336                 return -ENOMEM;
3337         return 0;
3338 }
3339
3340 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3341 {
3342         int err;
3343
3344         sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3345                                                         GFP_KERNEL);
3346         if (!sbi->nm_info)
3347                 return -ENOMEM;
3348
3349         err = init_node_manager(sbi);
3350         if (err)
3351                 return err;
3352
3353         err = init_free_nid_cache(sbi);
3354         if (err)
3355                 return err;
3356
3357         /* load free nid status from nat_bits table */
3358         load_free_nid_bitmap(sbi);
3359
3360         return f2fs_build_free_nids(sbi, true, true);
3361 }
3362
3363 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3364 {
3365         struct f2fs_nm_info *nm_i = NM_I(sbi);
3366         struct free_nid *i, *next_i;
3367         void *vec[NAT_VEC_SIZE];
3368         struct nat_entry **natvec = (struct nat_entry **)vec;
3369         struct nat_entry_set **setvec = (struct nat_entry_set **)vec;
3370         nid_t nid = 0;
3371         unsigned int found;
3372
3373         if (!nm_i)
3374                 return;
3375
3376         /* destroy free nid list */
3377         spin_lock(&nm_i->nid_list_lock);
3378         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3379                 __remove_free_nid(sbi, i, FREE_NID);
3380                 spin_unlock(&nm_i->nid_list_lock);
3381                 kmem_cache_free(free_nid_slab, i);
3382                 spin_lock(&nm_i->nid_list_lock);
3383         }
3384         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3385         f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3386         f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3387         spin_unlock(&nm_i->nid_list_lock);
3388
3389         /* destroy nat cache */
3390         f2fs_down_write(&nm_i->nat_tree_lock);
3391         while ((found = __gang_lookup_nat_cache(nm_i,
3392                                         nid, NAT_VEC_SIZE, natvec))) {
3393                 unsigned idx;
3394
3395                 nid = nat_get_nid(natvec[found - 1]) + 1;
3396                 for (idx = 0; idx < found; idx++) {
3397                         spin_lock(&nm_i->nat_list_lock);
3398                         list_del(&natvec[idx]->list);
3399                         spin_unlock(&nm_i->nat_list_lock);
3400
3401                         __del_from_nat_cache(nm_i, natvec[idx]);
3402                 }
3403         }
3404         f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3405
3406         /* destroy nat set cache */
3407         nid = 0;
3408         memset(vec, 0, sizeof(void *) * NAT_VEC_SIZE);
3409         while ((found = __gang_lookup_nat_set(nm_i,
3410                                         nid, NAT_VEC_SIZE, setvec))) {
3411                 unsigned idx;
3412
3413                 nid = setvec[found - 1]->set + 1;
3414                 for (idx = 0; idx < found; idx++) {
3415                         /* entry_cnt is not zero, when cp_error was occurred */
3416                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3417                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3418                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3419                 }
3420         }
3421         f2fs_up_write(&nm_i->nat_tree_lock);
3422
3423         kvfree(nm_i->nat_block_bitmap);
3424         if (nm_i->free_nid_bitmap) {
3425                 int i;
3426
3427                 for (i = 0; i < nm_i->nat_blocks; i++)
3428                         kvfree(nm_i->free_nid_bitmap[i]);
3429                 kvfree(nm_i->free_nid_bitmap);
3430         }
3431         kvfree(nm_i->free_nid_count);
3432
3433         kfree(nm_i->nat_bitmap);
3434         kvfree(nm_i->nat_bits);
3435 #ifdef CONFIG_F2FS_CHECK_FS
3436         kfree(nm_i->nat_bitmap_mir);
3437 #endif
3438         sbi->nm_info = NULL;
3439         kfree(nm_i);
3440 }
3441
3442 int __init f2fs_create_node_manager_caches(void)
3443 {
3444         nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3445                         sizeof(struct nat_entry));
3446         if (!nat_entry_slab)
3447                 goto fail;
3448
3449         free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3450                         sizeof(struct free_nid));
3451         if (!free_nid_slab)
3452                 goto destroy_nat_entry;
3453
3454         nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3455                         sizeof(struct nat_entry_set));
3456         if (!nat_entry_set_slab)
3457                 goto destroy_free_nid;
3458
3459         fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3460                         sizeof(struct fsync_node_entry));
3461         if (!fsync_node_entry_slab)
3462                 goto destroy_nat_entry_set;
3463         return 0;
3464
3465 destroy_nat_entry_set:
3466         kmem_cache_destroy(nat_entry_set_slab);
3467 destroy_free_nid:
3468         kmem_cache_destroy(free_nid_slab);
3469 destroy_nat_entry:
3470         kmem_cache_destroy(nat_entry_slab);
3471 fail:
3472         return -ENOMEM;
3473 }
3474
3475 void f2fs_destroy_node_manager_caches(void)
3476 {
3477         kmem_cache_destroy(fsync_node_entry_slab);
3478         kmem_cache_destroy(nat_entry_set_slab);
3479         kmem_cache_destroy(free_nid_slab);
3480         kmem_cache_destroy(nat_entry_slab);
3481 }