fs/f2fs/node.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/node.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/mpage.h>
  11 #include <linux/backing-dev.h>
  12 #include <linux/blkdev.h>
  13 #include <linux/pagevec.h>
  14 #include <linux/swap.h>
  15
  16 #include "f2fs.h"
  17 #include "node.h"
  18 #include "segment.h"
  19 #include "xattr.h"
  20 #include "iostat.h"
  21 #include <trace/events/f2fs.h>
  22
  23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
  24
  25 static struct kmem_cache *nat_entry_slab;
  26 static struct kmem_cache *free_nid_slab;
  27 static struct kmem_cache *nat_entry_set_slab;
  28 static struct kmem_cache *fsync_node_entry_slab;
  29
  30 /*
  31  * Check whether the given nid is within node id range.
  32  */
  33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
  34 {
  35         if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
  36                 set_sbi_flag(sbi, SBI_NEED_FSCK);
  37                 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
  38                           __func__, nid);
  39                 return -EFSCORRUPTED;
  40         }
  41         return 0;
  42 }
  43
  44 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
  45 {
  46         struct f2fs_nm_info *nm_i = NM_I(sbi);
  47         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
  48         struct sysinfo val;
  49         unsigned long avail_ram;
  50         unsigned long mem_size = 0;
  51         bool res = false;
  52
  53         if (!nm_i)
  54                 return true;
  55
  56         si_meminfo(&val);
  57
  58         /* only uses low memory */
  59         avail_ram = val.totalram - val.totalhigh;
  60
  61         /*
  62          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  63          */
  64         if (type == FREE_NIDS) {
  65                 mem_size = (nm_i->nid_cnt[FREE_NID] *
  66                                 sizeof(struct free_nid)) >> PAGE_SHIFT;
  67                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  68         } else if (type == NAT_ENTRIES) {
  69                 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
  70                                 sizeof(struct nat_entry)) >> PAGE_SHIFT;
  71                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  72                 if (excess_cached_nats(sbi))
  73                         res = false;
  74         } else if (type == DIRTY_DENTS) {
  75                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  76                         return false;
  77                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  78                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  79         } else if (type == INO_ENTRIES) {
  80                 int i;
  81
  82                 for (i = 0; i < MAX_INO_ENTRY; i++)
  83                         mem_size += sbi->im[i].ino_num *
  84                                                 sizeof(struct ino_entry);
  85                 mem_size >>= PAGE_SHIFT;
  86                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  87         } else if (type == EXTENT_CACHE) {
  88                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  89                                 sizeof(struct extent_tree) +
  90                                 atomic_read(&sbi->total_ext_node) *
  91                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  92                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  93         } else if (type == INMEM_PAGES) {
  94                 /* it allows 20% / total_ram for inmemory pages */
  95                 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
  96                 res = mem_size < (val.totalram / 5);
  97         } else if (type == DISCARD_CACHE) {
  98                 mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
  99                                 sizeof(struct discard_cmd)) >> PAGE_SHIFT;
 100                 res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
 101         } else if (type == COMPRESS_PAGE) {
 102 #ifdef CONFIG_F2FS_FS_COMPRESSION
 103                 unsigned long free_ram = val.freeram;
 104
 105                 /*
 106                  * free memory is lower than watermark or cached page count
 107                  * exceed threshold, deny caching compress page.
 108                  */
 109                 res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
 110                         (COMPRESS_MAPPING(sbi)->nrpages <
 111                          free_ram * sbi->compress_percent / 100);
 112 #else
 113                 res = false;
 114 #endif
 115         } else {
 116                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
 117                         return true;
 118         }
 119         return res;
 120 }
 121
 122 static void clear_node_page_dirty(struct page *page)
 123 {
 124         if (PageDirty(page)) {
 125                 f2fs_clear_page_cache_dirty_tag(page);
 126                 clear_page_dirty_for_io(page);
 127                 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
 128         }
 129         ClearPageUptodate(page);
 130 }
 131
 132 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 133 {
 134         return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
 135 }
 136
 137 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 138 {
 139         struct page *src_page;
 140         struct page *dst_page;
 141         pgoff_t dst_off;
 142         void *src_addr;
 143         void *dst_addr;
 144         struct f2fs_nm_info *nm_i = NM_I(sbi);
 145
 146         dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
 147
 148         /* get current nat block page with lock */
 149         src_page = get_current_nat_page(sbi, nid);
 150         if (IS_ERR(src_page))
 151                 return src_page;
 152         dst_page = f2fs_grab_meta_page(sbi, dst_off);
 153         f2fs_bug_on(sbi, PageDirty(src_page));
 154
 155         src_addr = page_address(src_page);
 156         dst_addr = page_address(dst_page);
 157         memcpy(dst_addr, src_addr, PAGE_SIZE);
 158         set_page_dirty(dst_page);
 159         f2fs_put_page(src_page, 1);
 160
 161         set_to_next_nat(nm_i, nid);
 162
 163         return dst_page;
 164 }
 165
 166 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
 167                                                 nid_t nid, bool no_fail)
 168 {
 169         struct nat_entry *new;
 170
 171         new = f2fs_kmem_cache_alloc(nat_entry_slab,
 172                                         GFP_F2FS_ZERO, no_fail, sbi);
 173         if (new) {
 174                 nat_set_nid(new, nid);
 175                 nat_reset_flag(new);
 176         }
 177         return new;
 178 }
 179
 180 static void __free_nat_entry(struct nat_entry *e)
 181 {
 182         kmem_cache_free(nat_entry_slab, e);
 183 }
 184
 185 /* must be locked by nat_tree_lock */
 186 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
 187         struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
 188 {
 189         if (no_fail)
 190                 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
 191         else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
 192                 return NULL;
 193
 194         if (raw_ne)
 195                 node_info_from_raw_nat(&ne->ni, raw_ne);
 196
 197         spin_lock(&nm_i->nat_list_lock);
 198         list_add_tail(&ne->list, &nm_i->nat_entries);
 199         spin_unlock(&nm_i->nat_list_lock);
 200
 201         nm_i->nat_cnt[TOTAL_NAT]++;
 202         nm_i->nat_cnt[RECLAIMABLE_NAT]++;
 203         return ne;
 204 }
 205
 206 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 207 {
 208         struct nat_entry *ne;
 209
 210         ne = radix_tree_lookup(&nm_i->nat_root, n);
 211
 212         /* for recent accessed nat entry, move it to tail of lru list */
 213         if (ne && !get_nat_flag(ne, IS_DIRTY)) {
 214                 spin_lock(&nm_i->nat_list_lock);
 215                 if (!list_empty(&ne->list))
 216                         list_move_tail(&ne->list, &nm_i->nat_entries);
 217                 spin_unlock(&nm_i->nat_list_lock);
 218         }
 219
 220         return ne;
 221 }
 222
 223 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 224                 nid_t start, unsigned int nr, struct nat_entry **ep)
 225 {
 226         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 227 }
 228
 229 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 230 {
 231         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 232         nm_i->nat_cnt[TOTAL_NAT]--;
 233         nm_i->nat_cnt[RECLAIMABLE_NAT]--;
 234         __free_nat_entry(e);
 235 }
 236
 237 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
 238                                                         struct nat_entry *ne)
 239 {
 240         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 241         struct nat_entry_set *head;
 242
 243         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 244         if (!head) {
 245                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
 246                                                 GFP_NOFS, true, NULL);
 247
 248                 INIT_LIST_HEAD(&head->entry_list);
 249                 INIT_LIST_HEAD(&head->set_list);
 250                 head->set = set;
 251                 head->entry_cnt = 0;
 252                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 253         }
 254         return head;
 255 }
 256
 257 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 258                                                 struct nat_entry *ne)
 259 {
 260         struct nat_entry_set *head;
 261         bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
 262
 263         if (!new_ne)
 264                 head = __grab_nat_entry_set(nm_i, ne);
 265
 266         /*
 267          * update entry_cnt in below condition:
 268          * 1. update NEW_ADDR to valid block address;
 269          * 2. update old block address to new one;
 270          */
 271         if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
 272                                 !get_nat_flag(ne, IS_DIRTY)))
 273                 head->entry_cnt++;
 274
 275         set_nat_flag(ne, IS_PREALLOC, new_ne);
 276
 277         if (get_nat_flag(ne, IS_DIRTY))
 278                 goto refresh_list;
 279
 280         nm_i->nat_cnt[DIRTY_NAT]++;
 281         nm_i->nat_cnt[RECLAIMABLE_NAT]--;
 282         set_nat_flag(ne, IS_DIRTY, true);
 283 refresh_list:
 284         spin_lock(&nm_i->nat_list_lock);
 285         if (new_ne)
 286                 list_del_init(&ne->list);
 287         else
 288                 list_move_tail(&ne->list, &head->entry_list);
 289         spin_unlock(&nm_i->nat_list_lock);
 290 }
 291
 292 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 293                 struct nat_entry_set *set, struct nat_entry *ne)
 294 {
 295         spin_lock(&nm_i->nat_list_lock);
 296         list_move_tail(&ne->list, &nm_i->nat_entries);
 297         spin_unlock(&nm_i->nat_list_lock);
 298
 299         set_nat_flag(ne, IS_DIRTY, false);
 300         set->entry_cnt--;
 301         nm_i->nat_cnt[DIRTY_NAT]--;
 302         nm_i->nat_cnt[RECLAIMABLE_NAT]++;
 303 }
 304
 305 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 306                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 307 {
 308         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 309                                                         start, nr);
 310 }
 311
 312 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
 313 {
 314         return NODE_MAPPING(sbi) == page->mapping &&
 315                         IS_DNODE(page) && is_cold_node(page);
 316 }
 317
 318 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
 319 {
 320         spin_lock_init(&sbi->fsync_node_lock);
 321         INIT_LIST_HEAD(&sbi->fsync_node_list);
 322         sbi->fsync_seg_id = 0;
 323         sbi->fsync_node_num = 0;
 324 }
 325
 326 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
 327                                                         struct page *page)
 328 {
 329         struct fsync_node_entry *fn;
 330         unsigned long flags;
 331         unsigned int seq_id;
 332
 333         fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
 334                                         GFP_NOFS, true, NULL);
 335
 336         get_page(page);
 337         fn->page = page;
 338         INIT_LIST_HEAD(&fn->list);
 339
 340         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 341         list_add_tail(&fn->list, &sbi->fsync_node_list);
 342         fn->seq_id = sbi->fsync_seg_id++;
 343         seq_id = fn->seq_id;
 344         sbi->fsync_node_num++;
 345         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 346
 347         return seq_id;
 348 }
 349
 350 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
 351 {
 352         struct fsync_node_entry *fn;
 353         unsigned long flags;
 354
 355         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 356         list_for_each_entry(fn, &sbi->fsync_node_list, list) {
 357                 if (fn->page == page) {
 358                         list_del(&fn->list);
 359                         sbi->fsync_node_num--;
 360                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 361                         kmem_cache_free(fsync_node_entry_slab, fn);
 362                         put_page(page);
 363                         return;
 364                 }
 365         }
 366         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 367         f2fs_bug_on(sbi, 1);
 368 }
 369
 370 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
 371 {
 372         unsigned long flags;
 373
 374         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 375         sbi->fsync_seg_id = 0;
 376         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 377 }
 378
 379 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 380 {
 381         struct f2fs_nm_info *nm_i = NM_I(sbi);
 382         struct nat_entry *e;
 383         bool need = false;
 384
 385         down_read(&nm_i->nat_tree_lock);
 386         e = __lookup_nat_cache(nm_i, nid);
 387         if (e) {
 388                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 389                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 390                         need = true;
 391         }
 392         up_read(&nm_i->nat_tree_lock);
 393         return need;
 394 }
 395
 396 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 397 {
 398         struct f2fs_nm_info *nm_i = NM_I(sbi);
 399         struct nat_entry *e;
 400         bool is_cp = true;
 401
 402         down_read(&nm_i->nat_tree_lock);
 403         e = __lookup_nat_cache(nm_i, nid);
 404         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 405                 is_cp = false;
 406         up_read(&nm_i->nat_tree_lock);
 407         return is_cp;
 408 }
 409
 410 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 411 {
 412         struct f2fs_nm_info *nm_i = NM_I(sbi);
 413         struct nat_entry *e;
 414         bool need_update = true;
 415
 416         down_read(&nm_i->nat_tree_lock);
 417         e = __lookup_nat_cache(nm_i, ino);
 418         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 419                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 420                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 421                 need_update = false;
 422         up_read(&nm_i->nat_tree_lock);
 423         return need_update;
 424 }
 425
 426 /* must be locked by nat_tree_lock */
 427 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 428                                                 struct f2fs_nat_entry *ne)
 429 {
 430         struct f2fs_nm_info *nm_i = NM_I(sbi);
 431         struct nat_entry *new, *e;
 432
 433         /* Let's mitigate lock contention of nat_tree_lock during checkpoint */
 434         if (rwsem_is_locked(&sbi->cp_global_sem))
 435                 return;
 436
 437         new = __alloc_nat_entry(sbi, nid, false);
 438         if (!new)
 439                 return;
 440
 441         down_write(&nm_i->nat_tree_lock);
 442         e = __lookup_nat_cache(nm_i, nid);
 443         if (!e)
 444                 e = __init_nat_entry(nm_i, new, ne, false);
 445         else
 446                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 447                                 nat_get_blkaddr(e) !=
 448                                         le32_to_cpu(ne->block_addr) ||
 449                                 nat_get_version(e) != ne->version);
 450         up_write(&nm_i->nat_tree_lock);
 451         if (e != new)
 452                 __free_nat_entry(new);
 453 }
 454
 455 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 456                         block_t new_blkaddr, bool fsync_done)
 457 {
 458         struct f2fs_nm_info *nm_i = NM_I(sbi);
 459         struct nat_entry *e;
 460         struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
 461
 462         down_write(&nm_i->nat_tree_lock);
 463         e = __lookup_nat_cache(nm_i, ni->nid);
 464         if (!e) {
 465                 e = __init_nat_entry(nm_i, new, NULL, true);
 466                 copy_node_info(&e->ni, ni);
 467                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 468         } else if (new_blkaddr == NEW_ADDR) {
 469                 /*
 470                  * when nid is reallocated,
 471                  * previous nat entry can be remained in nat cache.
 472                  * So, reinitialize it with new information.
 473                  */
 474                 copy_node_info(&e->ni, ni);
 475                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 476         }
 477         /* let's free early to reduce memory consumption */
 478         if (e != new)
 479                 __free_nat_entry(new);
 480
 481         /* sanity check */
 482         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 483         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 484                         new_blkaddr == NULL_ADDR);
 485         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 486                         new_blkaddr == NEW_ADDR);
 487         f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
 488                         new_blkaddr == NEW_ADDR);
 489
 490         /* increment version no as node is removed */
 491         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 492                 unsigned char version = nat_get_version(e);
 493
 494                 nat_set_version(e, inc_node_version(version));
 495         }
 496
 497         /* change address */
 498         nat_set_blkaddr(e, new_blkaddr);
 499         if (!__is_valid_data_blkaddr(new_blkaddr))
 500                 set_nat_flag(e, IS_CHECKPOINTED, false);
 501         __set_nat_cache_dirty(nm_i, e);
 502
 503         /* update fsync_mark if its inode nat entry is still alive */
 504         if (ni->nid != ni->ino)
 505                 e = __lookup_nat_cache(nm_i, ni->ino);
 506         if (e) {
 507                 if (fsync_done && ni->nid == ni->ino)
 508                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 509                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 510         }
 511         up_write(&nm_i->nat_tree_lock);
 512 }
 513
 514 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 515 {
 516         struct f2fs_nm_info *nm_i = NM_I(sbi);
 517         int nr = nr_shrink;
 518
 519         if (!down_write_trylock(&nm_i->nat_tree_lock))
 520                 return 0;
 521
 522         spin_lock(&nm_i->nat_list_lock);
 523         while (nr_shrink) {
 524                 struct nat_entry *ne;
 525
 526                 if (list_empty(&nm_i->nat_entries))
 527                         break;
 528
 529                 ne = list_first_entry(&nm_i->nat_entries,
 530                                         struct nat_entry, list);
 531                 list_del(&ne->list);
 532                 spin_unlock(&nm_i->nat_list_lock);
 533
 534                 __del_from_nat_cache(nm_i, ne);
 535                 nr_shrink--;
 536
 537                 spin_lock(&nm_i->nat_list_lock);
 538         }
 539         spin_unlock(&nm_i->nat_list_lock);
 540
 541         up_write(&nm_i->nat_tree_lock);
 542         return nr - nr_shrink;
 543 }
 544
 545 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
 546                                 struct node_info *ni, bool checkpoint_context)
 547 {
 548         struct f2fs_nm_info *nm_i = NM_I(sbi);
 549         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 550         struct f2fs_journal *journal = curseg->journal;
 551         nid_t start_nid = START_NID(nid);
 552         struct f2fs_nat_block *nat_blk;
 553         struct page *page = NULL;
 554         struct f2fs_nat_entry ne;
 555         struct nat_entry *e;
 556         pgoff_t index;
 557         block_t blkaddr;
 558         int i;
 559
 560         ni->nid = nid;
 561 retry:
 562         /* Check nat cache */
 563         down_read(&nm_i->nat_tree_lock);
 564         e = __lookup_nat_cache(nm_i, nid);
 565         if (e) {
 566                 ni->ino = nat_get_ino(e);
 567                 ni->blk_addr = nat_get_blkaddr(e);
 568                 ni->version = nat_get_version(e);
 569                 up_read(&nm_i->nat_tree_lock);
 570                 return 0;
 571         }
 572
 573         /*
 574          * Check current segment summary by trying to grab journal_rwsem first.
 575          * This sem is on the critical path on the checkpoint requiring the above
 576          * nat_tree_lock. Therefore, we should retry, if we failed to grab here
 577          * while not bothering checkpoint.
 578          */
 579         if (!rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
 580                 down_read(&curseg->journal_rwsem);
 581         } else if (rwsem_is_contended(&nm_i->nat_tree_lock) ||
 582                                 !down_read_trylock(&curseg->journal_rwsem)) {
 583                 up_read(&nm_i->nat_tree_lock);
 584                 goto retry;
 585         }
 586
 587         i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 588         if (i >= 0) {
 589                 ne = nat_in_journal(journal, i);
 590                 node_info_from_raw_nat(ni, &ne);
 591         }
 592         up_read(&curseg->journal_rwsem);
 593         if (i >= 0) {
 594                 up_read(&nm_i->nat_tree_lock);
 595                 goto cache;
 596         }
 597
 598         /* Fill node_info from nat page */
 599         index = current_nat_addr(sbi, nid);
 600         up_read(&nm_i->nat_tree_lock);
 601
 602         page = f2fs_get_meta_page(sbi, index);
 603         if (IS_ERR(page))
 604                 return PTR_ERR(page);
 605
 606         nat_blk = (struct f2fs_nat_block *)page_address(page);
 607         ne = nat_blk->entries[nid - start_nid];
 608         node_info_from_raw_nat(ni, &ne);
 609         f2fs_put_page(page, 1);
 610 cache:
 611         blkaddr = le32_to_cpu(ne.block_addr);
 612         if (__is_valid_data_blkaddr(blkaddr) &&
 613                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
 614                 return -EFAULT;
 615
 616         /* cache nat entry */
 617         cache_nat_entry(sbi, nid, &ne);
 618         return 0;
 619 }
 620
 621 /*
 622  * readahead MAX_RA_NODE number of node pages.
 623  */
 624 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
 625 {
 626         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 627         struct blk_plug plug;
 628         int i, end;
 629         nid_t nid;
 630
 631         blk_start_plug(&plug);
 632
 633         /* Then, try readahead for siblings of the desired node */
 634         end = start + n;
 635         end = min(end, NIDS_PER_BLOCK);
 636         for (i = start; i < end; i++) {
 637                 nid = get_nid(parent, i, false);
 638                 f2fs_ra_node_page(sbi, nid);
 639         }
 640
 641         blk_finish_plug(&plug);
 642 }
 643
 644 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 645 {
 646         const long direct_index = ADDRS_PER_INODE(dn->inode);
 647         const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
 648         const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
 649         unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
 650         int cur_level = dn->cur_level;
 651         int max_level = dn->max_level;
 652         pgoff_t base = 0;
 653
 654         if (!dn->max_level)
 655                 return pgofs + 1;
 656
 657         while (max_level-- > cur_level)
 658                 skipped_unit *= NIDS_PER_BLOCK;
 659
 660         switch (dn->max_level) {
 661         case 3:
 662                 base += 2 * indirect_blks;
 663                 fallthrough;
 664         case 2:
 665                 base += 2 * direct_blks;
 666                 fallthrough;
 667         case 1:
 668                 base += direct_index;
 669                 break;
 670         default:
 671                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 672         }
 673
 674         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 675 }
 676
 677 /*
 678  * The maximum depth is four.
 679  * Offset[0] will have raw inode offset.
 680  */
 681 static int get_node_path(struct inode *inode, long block,
 682                                 int offset[4], unsigned int noffset[4])
 683 {
 684         const long direct_index = ADDRS_PER_INODE(inode);
 685         const long direct_blks = ADDRS_PER_BLOCK(inode);
 686         const long dptrs_per_blk = NIDS_PER_BLOCK;
 687         const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
 688         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 689         int n = 0;
 690         int level = 0;
 691
 692         noffset[0] = 0;
 693
 694         if (block < direct_index) {
 695                 offset[n] = block;
 696                 goto got;
 697         }
 698         block -= direct_index;
 699         if (block < direct_blks) {
 700                 offset[n++] = NODE_DIR1_BLOCK;
 701                 noffset[n] = 1;
 702                 offset[n] = block;
 703                 level = 1;
 704                 goto got;
 705         }
 706         block -= direct_blks;
 707         if (block < direct_blks) {
 708                 offset[n++] = NODE_DIR2_BLOCK;
 709                 noffset[n] = 2;
 710                 offset[n] = block;
 711                 level = 1;
 712                 goto got;
 713         }
 714         block -= direct_blks;
 715         if (block < indirect_blks) {
 716                 offset[n++] = NODE_IND1_BLOCK;
 717                 noffset[n] = 3;
 718                 offset[n++] = block / direct_blks;
 719                 noffset[n] = 4 + offset[n - 1];
 720                 offset[n] = block % direct_blks;
 721                 level = 2;
 722                 goto got;
 723         }
 724         block -= indirect_blks;
 725         if (block < indirect_blks) {
 726                 offset[n++] = NODE_IND2_BLOCK;
 727                 noffset[n] = 4 + dptrs_per_blk;
 728                 offset[n++] = block / direct_blks;
 729                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 730                 offset[n] = block % direct_blks;
 731                 level = 2;
 732                 goto got;
 733         }
 734         block -= indirect_blks;
 735         if (block < dindirect_blks) {
 736                 offset[n++] = NODE_DIND_BLOCK;
 737                 noffset[n] = 5 + (dptrs_per_blk * 2);
 738                 offset[n++] = block / indirect_blks;
 739                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 740                               offset[n - 1] * (dptrs_per_blk + 1);
 741                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 742                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 743                               offset[n - 2] * (dptrs_per_blk + 1) +
 744                               offset[n - 1];
 745                 offset[n] = block % direct_blks;
 746                 level = 3;
 747                 goto got;
 748         } else {
 749                 return -E2BIG;
 750         }
 751 got:
 752         return level;
 753 }
 754
 755 /*
 756  * Caller should call f2fs_put_dnode(dn).
 757  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 758  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
 759  */
 760 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 761 {
 762         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 763         struct page *npage[4];
 764         struct page *parent = NULL;
 765         int offset[4];
 766         unsigned int noffset[4];
 767         nid_t nids[4];
 768         int level, i = 0;
 769         int err = 0;
 770
 771         level = get_node_path(dn->inode, index, offset, noffset);
 772         if (level < 0)
 773                 return level;
 774
 775         nids[0] = dn->inode->i_ino;
 776         npage[0] = dn->inode_page;
 777
 778         if (!npage[0]) {
 779                 npage[0] = f2fs_get_node_page(sbi, nids[0]);
 780                 if (IS_ERR(npage[0]))
 781                         return PTR_ERR(npage[0]);
 782         }
 783
 784         /* if inline_data is set, should not report any block indices */
 785         if (f2fs_has_inline_data(dn->inode) && index) {
 786                 err = -ENOENT;
 787                 f2fs_put_page(npage[0], 1);
 788                 goto release_out;
 789         }
 790
 791         parent = npage[0];
 792         if (level != 0)
 793                 nids[1] = get_nid(parent, offset[0], true);
 794         dn->inode_page = npage[0];
 795         dn->inode_page_locked = true;
 796
 797         /* get indirect or direct nodes */
 798         for (i = 1; i <= level; i++) {
 799                 bool done = false;
 800
 801                 if (!nids[i] && mode == ALLOC_NODE) {
 802                         /* alloc new node */
 803                         if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
 804                                 err = -ENOSPC;
 805                                 goto release_pages;
 806                         }
 807
 808                         dn->nid = nids[i];
 809                         npage[i] = f2fs_new_node_page(dn, noffset[i]);
 810                         if (IS_ERR(npage[i])) {
 811                                 f2fs_alloc_nid_failed(sbi, nids[i]);
 812                                 err = PTR_ERR(npage[i]);
 813                                 goto release_pages;
 814                         }
 815
 816                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 817                         f2fs_alloc_nid_done(sbi, nids[i]);
 818                         done = true;
 819                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 820                         npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
 821                         if (IS_ERR(npage[i])) {
 822                                 err = PTR_ERR(npage[i]);
 823                                 goto release_pages;
 824                         }
 825                         done = true;
 826                 }
 827                 if (i == 1) {
 828                         dn->inode_page_locked = false;
 829                         unlock_page(parent);
 830                 } else {
 831                         f2fs_put_page(parent, 1);
 832                 }
 833
 834                 if (!done) {
 835                         npage[i] = f2fs_get_node_page(sbi, nids[i]);
 836                         if (IS_ERR(npage[i])) {
 837                                 err = PTR_ERR(npage[i]);
 838                                 f2fs_put_page(npage[0], 0);
 839                                 goto release_out;
 840                         }
 841                 }
 842                 if (i < level) {
 843                         parent = npage[i];
 844                         nids[i + 1] = get_nid(parent, offset[i], false);
 845                 }
 846         }
 847         dn->nid = nids[level];
 848         dn->ofs_in_node = offset[level];
 849         dn->node_page = npage[level];
 850         dn->data_blkaddr = f2fs_data_blkaddr(dn);
 851
 852         if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
 853                                         f2fs_sb_has_readonly(sbi)) {
 854                 unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn);
 855                 block_t blkaddr;
 856
 857                 if (!c_len)
 858                         goto out;
 859
 860                 blkaddr = f2fs_data_blkaddr(dn);
 861                 if (blkaddr == COMPRESS_ADDR)
 862                         blkaddr = data_blkaddr(dn->inode, dn->node_page,
 863                                                 dn->ofs_in_node + 1);
 864
 865                 f2fs_update_extent_tree_range_compressed(dn->inode,
 866                                         index, blkaddr,
 867                                         F2FS_I(dn->inode)->i_cluster_size,
 868                                         c_len);
 869         }
 870 out:
 871         return 0;
 872
 873 release_pages:
 874         f2fs_put_page(parent, 1);
 875         if (i > 1)
 876                 f2fs_put_page(npage[0], 0);
 877 release_out:
 878         dn->inode_page = NULL;
 879         dn->node_page = NULL;
 880         if (err == -ENOENT) {
 881                 dn->cur_level = i;
 882                 dn->max_level = level;
 883                 dn->ofs_in_node = offset[level];
 884         }
 885         return err;
 886 }
 887
 888 static int truncate_node(struct dnode_of_data *dn)
 889 {
 890         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 891         struct node_info ni;
 892         int err;
 893         pgoff_t index;
 894
 895         err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
 896         if (err)
 897                 return err;
 898
 899         /* Deallocate node address */
 900         f2fs_invalidate_blocks(sbi, ni.blk_addr);
 901         dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
 902         set_node_addr(sbi, &ni, NULL_ADDR, false);
 903
 904         if (dn->nid == dn->inode->i_ino) {
 905                 f2fs_remove_orphan_inode(sbi, dn->nid);
 906                 dec_valid_inode_count(sbi);
 907                 f2fs_inode_synced(dn->inode);
 908         }
 909
 910         clear_node_page_dirty(dn->node_page);
 911         set_sbi_flag(sbi, SBI_IS_DIRTY);
 912
 913         index = dn->node_page->index;
 914         f2fs_put_page(dn->node_page, 1);
 915
 916         invalidate_mapping_pages(NODE_MAPPING(sbi),
 917                         index, index);
 918
 919         dn->node_page = NULL;
 920         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 921
 922         return 0;
 923 }
 924
 925 static int truncate_dnode(struct dnode_of_data *dn)
 926 {
 927         struct page *page;
 928         int err;
 929
 930         if (dn->nid == 0)
 931                 return 1;
 932
 933         /* get direct node */
 934         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 935         if (PTR_ERR(page) == -ENOENT)
 936                 return 1;
 937         else if (IS_ERR(page))
 938                 return PTR_ERR(page);
 939
 940         /* Make dnode_of_data for parameter */
 941         dn->node_page = page;
 942         dn->ofs_in_node = 0;
 943         f2fs_truncate_data_blocks(dn);
 944         err = truncate_node(dn);
 945         if (err) {
 946                 f2fs_put_page(page, 1);
 947                 return err;
 948         }
 949
 950         return 1;
 951 }
 952
 953 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 954                                                 int ofs, int depth)
 955 {
 956         struct dnode_of_data rdn = *dn;
 957         struct page *page;
 958         struct f2fs_node *rn;
 959         nid_t child_nid;
 960         unsigned int child_nofs;
 961         int freed = 0;
 962         int i, ret;
 963
 964         if (dn->nid == 0)
 965                 return NIDS_PER_BLOCK + 1;
 966
 967         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 968
 969         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 970         if (IS_ERR(page)) {
 971                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 972                 return PTR_ERR(page);
 973         }
 974
 975         f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 976
 977         rn = F2FS_NODE(page);
 978         if (depth < 3) {
 979                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 980                         child_nid = le32_to_cpu(rn->in.nid[i]);
 981                         if (child_nid == 0)
 982                                 continue;
 983                         rdn.nid = child_nid;
 984                         ret = truncate_dnode(&rdn);
 985                         if (ret < 0)
 986                                 goto out_err;
 987                         if (set_nid(page, i, 0, false))
 988                                 dn->node_changed = true;
 989                 }
 990         } else {
 991                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 992                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 993                         child_nid = le32_to_cpu(rn->in.nid[i]);
 994                         if (child_nid == 0) {
 995                                 child_nofs += NIDS_PER_BLOCK + 1;
 996                                 continue;
 997                         }
 998                         rdn.nid = child_nid;
 999                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
1000                         if (ret == (NIDS_PER_BLOCK + 1)) {
1001                                 if (set_nid(page, i, 0, false))
1002                                         dn->node_changed = true;
1003                                 child_nofs += ret;
1004                         } else if (ret < 0 && ret != -ENOENT) {
1005                                 goto out_err;
1006                         }
1007                 }
1008                 freed = child_nofs;
1009         }
1010
1011         if (!ofs) {
1012                 /* remove current indirect node */
1013                 dn->node_page = page;
1014                 ret = truncate_node(dn);
1015                 if (ret)
1016                         goto out_err;
1017                 freed++;
1018         } else {
1019                 f2fs_put_page(page, 1);
1020         }
1021         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1022         return freed;
1023
1024 out_err:
1025         f2fs_put_page(page, 1);
1026         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1027         return ret;
1028 }
1029
1030 static int truncate_partial_nodes(struct dnode_of_data *dn,
1031                         struct f2fs_inode *ri, int *offset, int depth)
1032 {
1033         struct page *pages[2];
1034         nid_t nid[3];
1035         nid_t child_nid;
1036         int err = 0;
1037         int i;
1038         int idx = depth - 2;
1039
1040         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1041         if (!nid[0])
1042                 return 0;
1043
1044         /* get indirect nodes in the path */
1045         for (i = 0; i < idx + 1; i++) {
1046                 /* reference count'll be increased */
1047                 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1048                 if (IS_ERR(pages[i])) {
1049                         err = PTR_ERR(pages[i]);
1050                         idx = i - 1;
1051                         goto fail;
1052                 }
1053                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1054         }
1055
1056         f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1057
1058         /* free direct nodes linked to a partial indirect node */
1059         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1060                 child_nid = get_nid(pages[idx], i, false);
1061                 if (!child_nid)
1062                         continue;
1063                 dn->nid = child_nid;
1064                 err = truncate_dnode(dn);
1065                 if (err < 0)
1066                         goto fail;
1067                 if (set_nid(pages[idx], i, 0, false))
1068                         dn->node_changed = true;
1069         }
1070
1071         if (offset[idx + 1] == 0) {
1072                 dn->node_page = pages[idx];
1073                 dn->nid = nid[idx];
1074                 err = truncate_node(dn);
1075                 if (err)
1076                         goto fail;
1077         } else {
1078                 f2fs_put_page(pages[idx], 1);
1079         }
1080         offset[idx]++;
1081         offset[idx + 1] = 0;
1082         idx--;
1083 fail:
1084         for (i = idx; i >= 0; i--)
1085                 f2fs_put_page(pages[i], 1);
1086
1087         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1088
1089         return err;
1090 }
1091
1092 /*
1093  * All the block addresses of data and nodes should be nullified.
1094  */
1095 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1096 {
1097         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1098         int err = 0, cont = 1;
1099         int level, offset[4], noffset[4];
1100         unsigned int nofs = 0;
1101         struct f2fs_inode *ri;
1102         struct dnode_of_data dn;
1103         struct page *page;
1104
1105         trace_f2fs_truncate_inode_blocks_enter(inode, from);
1106
1107         level = get_node_path(inode, from, offset, noffset);
1108         if (level < 0) {
1109                 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1110                 return level;
1111         }
1112
1113         page = f2fs_get_node_page(sbi, inode->i_ino);
1114         if (IS_ERR(page)) {
1115                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1116                 return PTR_ERR(page);
1117         }
1118
1119         set_new_dnode(&dn, inode, page, NULL, 0);
1120         unlock_page(page);
1121
1122         ri = F2FS_INODE(page);
1123         switch (level) {
1124         case 0:
1125         case 1:
1126                 nofs = noffset[1];
1127                 break;
1128         case 2:
1129                 nofs = noffset[1];
1130                 if (!offset[level - 1])
1131                         goto skip_partial;
1132                 err = truncate_partial_nodes(&dn, ri, offset, level);
1133                 if (err < 0 && err != -ENOENT)
1134                         goto fail;
1135                 nofs += 1 + NIDS_PER_BLOCK;
1136                 break;
1137         case 3:
1138                 nofs = 5 + 2 * NIDS_PER_BLOCK;
1139                 if (!offset[level - 1])
1140                         goto skip_partial;
1141                 err = truncate_partial_nodes(&dn, ri, offset, level);
1142                 if (err < 0 && err != -ENOENT)
1143                         goto fail;
1144                 break;
1145         default:
1146                 BUG();
1147         }
1148
1149 skip_partial:
1150         while (cont) {
1151                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1152                 switch (offset[0]) {
1153                 case NODE_DIR1_BLOCK:
1154                 case NODE_DIR2_BLOCK:
1155                         err = truncate_dnode(&dn);
1156                         break;
1157
1158                 case NODE_IND1_BLOCK:
1159                 case NODE_IND2_BLOCK:
1160                         err = truncate_nodes(&dn, nofs, offset[1], 2);
1161                         break;
1162
1163                 case NODE_DIND_BLOCK:
1164                         err = truncate_nodes(&dn, nofs, offset[1], 3);
1165                         cont = 0;
1166                         break;
1167
1168                 default:
1169                         BUG();
1170                 }
1171                 if (err < 0 && err != -ENOENT)
1172                         goto fail;
1173                 if (offset[1] == 0 &&
1174                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1175                         lock_page(page);
1176                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
1177                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1178                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1179                         set_page_dirty(page);
1180                         unlock_page(page);
1181                 }
1182                 offset[1] = 0;
1183                 offset[0]++;
1184                 nofs += err;
1185         }
1186 fail:
1187         f2fs_put_page(page, 0);
1188         trace_f2fs_truncate_inode_blocks_exit(inode, err);
1189         return err > 0 ? 0 : err;
1190 }
1191
1192 /* caller must lock inode page */
1193 int f2fs_truncate_xattr_node(struct inode *inode)
1194 {
1195         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1196         nid_t nid = F2FS_I(inode)->i_xattr_nid;
1197         struct dnode_of_data dn;
1198         struct page *npage;
1199         int err;
1200
1201         if (!nid)
1202                 return 0;
1203
1204         npage = f2fs_get_node_page(sbi, nid);
1205         if (IS_ERR(npage))
1206                 return PTR_ERR(npage);
1207
1208         set_new_dnode(&dn, inode, NULL, npage, nid);
1209         err = truncate_node(&dn);
1210         if (err) {
1211                 f2fs_put_page(npage, 1);
1212                 return err;
1213         }
1214
1215         f2fs_i_xnid_write(inode, 0);
1216
1217         return 0;
1218 }
1219
1220 /*
1221  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1222  * f2fs_unlock_op().
1223  */
1224 int f2fs_remove_inode_page(struct inode *inode)
1225 {
1226         struct dnode_of_data dn;
1227         int err;
1228
1229         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1230         err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1231         if (err)
1232                 return err;
1233
1234         err = f2fs_truncate_xattr_node(inode);
1235         if (err) {
1236                 f2fs_put_dnode(&dn);
1237                 return err;
1238         }
1239
1240         /* remove potential inline_data blocks */
1241         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1242                                 S_ISLNK(inode->i_mode))
1243                 f2fs_truncate_data_blocks_range(&dn, 1);
1244
1245         /* 0 is possible, after f2fs_new_inode() has failed */
1246         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1247                 f2fs_put_dnode(&dn);
1248                 return -EIO;
1249         }
1250
1251         if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1252                 f2fs_warn(F2FS_I_SB(inode),
1253                         "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1254                         inode->i_ino, (unsigned long long)inode->i_blocks);
1255                 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1256         }
1257
1258         /* will put inode & node pages */
1259         err = truncate_node(&dn);
1260         if (err) {
1261                 f2fs_put_dnode(&dn);
1262                 return err;
1263         }
1264         return 0;
1265 }
1266
1267 struct page *f2fs_new_inode_page(struct inode *inode)
1268 {
1269         struct dnode_of_data dn;
1270
1271         /* allocate inode page for new inode */
1272         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1273
1274         /* caller should f2fs_put_page(page, 1); */
1275         return f2fs_new_node_page(&dn, 0);
1276 }
1277
1278 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1279 {
1280         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1281         struct node_info new_ni;
1282         struct page *page;
1283         int err;
1284
1285         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1286                 return ERR_PTR(-EPERM);
1287
1288         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1289         if (!page)
1290                 return ERR_PTR(-ENOMEM);
1291
1292         if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1293                 goto fail;
1294
1295 #ifdef CONFIG_F2FS_CHECK_FS
1296         err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1297         if (err) {
1298                 dec_valid_node_count(sbi, dn->inode, !ofs);
1299                 goto fail;
1300         }
1301         if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1302                 err = -EFSCORRUPTED;
1303                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1304                 goto fail;
1305         }
1306 #endif
1307         new_ni.nid = dn->nid;
1308         new_ni.ino = dn->inode->i_ino;
1309         new_ni.blk_addr = NULL_ADDR;
1310         new_ni.flag = 0;
1311         new_ni.version = 0;
1312         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1313
1314         f2fs_wait_on_page_writeback(page, NODE, true, true);
1315         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1316         set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1317         if (!PageUptodate(page))
1318                 SetPageUptodate(page);
1319         if (set_page_dirty(page))
1320                 dn->node_changed = true;
1321
1322         if (f2fs_has_xattr_block(ofs))
1323                 f2fs_i_xnid_write(dn->inode, dn->nid);
1324
1325         if (ofs == 0)
1326                 inc_valid_inode_count(sbi);
1327         return page;
1328
1329 fail:
1330         clear_node_page_dirty(page);
1331         f2fs_put_page(page, 1);
1332         return ERR_PTR(err);
1333 }
1334
1335 /*
1336  * Caller should do after getting the following values.
1337  * 0: f2fs_put_page(page, 0)
1338  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1339  */
1340 static int read_node_page(struct page *page, int op_flags)
1341 {
1342         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1343         struct node_info ni;
1344         struct f2fs_io_info fio = {
1345                 .sbi = sbi,
1346                 .type = NODE,
1347                 .op = REQ_OP_READ,
1348                 .op_flags = op_flags,
1349                 .page = page,
1350                 .encrypted_page = NULL,
1351         };
1352         int err;
1353
1354         if (PageUptodate(page)) {
1355                 if (!f2fs_inode_chksum_verify(sbi, page)) {
1356                         ClearPageUptodate(page);
1357                         return -EFSBADCRC;
1358                 }
1359                 return LOCKED_PAGE;
1360         }
1361
1362         err = f2fs_get_node_info(sbi, page->index, &ni, false);
1363         if (err)
1364                 return err;
1365
1366         /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1367         if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1368                 ClearPageUptodate(page);
1369                 return -ENOENT;
1370         }
1371
1372         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1373
1374         err = f2fs_submit_page_bio(&fio);
1375
1376         if (!err)
1377                 f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1378
1379         return err;
1380 }
1381
1382 /*
1383  * Readahead a node page
1384  */
1385 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1386 {
1387         struct page *apage;
1388         int err;
1389
1390         if (!nid)
1391                 return;
1392         if (f2fs_check_nid_range(sbi, nid))
1393                 return;
1394
1395         apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1396         if (apage)
1397                 return;
1398
1399         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1400         if (!apage)
1401                 return;
1402
1403         err = read_node_page(apage, REQ_RAHEAD);
1404         f2fs_put_page(apage, err ? 1 : 0);
1405 }
1406
1407 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1408                                         struct page *parent, int start)
1409 {
1410         struct page *page;
1411         int err;
1412
1413         if (!nid)
1414                 return ERR_PTR(-ENOENT);
1415         if (f2fs_check_nid_range(sbi, nid))
1416                 return ERR_PTR(-EINVAL);
1417 repeat:
1418         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1419         if (!page)
1420                 return ERR_PTR(-ENOMEM);
1421
1422         err = read_node_page(page, 0);
1423         if (err < 0) {
1424                 f2fs_put_page(page, 1);
1425                 return ERR_PTR(err);
1426         } else if (err == LOCKED_PAGE) {
1427                 err = 0;
1428                 goto page_hit;
1429         }
1430
1431         if (parent)
1432                 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1433
1434         lock_page(page);
1435
1436         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1437                 f2fs_put_page(page, 1);
1438                 goto repeat;
1439         }
1440
1441         if (unlikely(!PageUptodate(page))) {
1442                 err = -EIO;
1443                 goto out_err;
1444         }
1445
1446         if (!f2fs_inode_chksum_verify(sbi, page)) {
1447                 err = -EFSBADCRC;
1448                 goto out_err;
1449         }
1450 page_hit:
1451         if (unlikely(nid != nid_of_node(page))) {
1452                 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1453                           nid, nid_of_node(page), ino_of_node(page),
1454                           ofs_of_node(page), cpver_of_node(page),
1455                           next_blkaddr_of_node(page));
1456                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1457                 err = -EINVAL;
1458 out_err:
1459                 ClearPageUptodate(page);
1460                 f2fs_put_page(page, 1);
1461                 return ERR_PTR(err);
1462         }
1463         return page;
1464 }
1465
1466 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1467 {
1468         return __get_node_page(sbi, nid, NULL, 0);
1469 }
1470
1471 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1472 {
1473         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1474         nid_t nid = get_nid(parent, start, false);
1475
1476         return __get_node_page(sbi, nid, parent, start);
1477 }
1478
1479 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1480 {
1481         struct inode *inode;
1482         struct page *page;
1483         int ret;
1484
1485         /* should flush inline_data before evict_inode */
1486         inode = ilookup(sbi->sb, ino);
1487         if (!inode)
1488                 return;
1489
1490         page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1491                                         FGP_LOCK|FGP_NOWAIT, 0);
1492         if (!page)
1493                 goto iput_out;
1494
1495         if (!PageUptodate(page))
1496                 goto page_out;
1497
1498         if (!PageDirty(page))
1499                 goto page_out;
1500
1501         if (!clear_page_dirty_for_io(page))
1502                 goto page_out;
1503
1504         ret = f2fs_write_inline_data(inode, page);
1505         inode_dec_dirty_pages(inode);
1506         f2fs_remove_dirty_inode(inode);
1507         if (ret)
1508                 set_page_dirty(page);
1509 page_out:
1510         f2fs_put_page(page, 1);
1511 iput_out:
1512         iput(inode);
1513 }
1514
1515 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1516 {
1517         pgoff_t index;
1518         struct pagevec pvec;
1519         struct page *last_page = NULL;
1520         int nr_pages;
1521
1522         pagevec_init(&pvec);
1523         index = 0;
1524
1525         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1526                                 PAGECACHE_TAG_DIRTY))) {
1527                 int i;
1528
1529                 for (i = 0; i < nr_pages; i++) {
1530                         struct page *page = pvec.pages[i];
1531
1532                         if (unlikely(f2fs_cp_error(sbi))) {
1533                                 f2fs_put_page(last_page, 0);
1534                                 pagevec_release(&pvec);
1535                                 return ERR_PTR(-EIO);
1536                         }
1537
1538                         if (!IS_DNODE(page) || !is_cold_node(page))
1539                                 continue;
1540                         if (ino_of_node(page) != ino)
1541                                 continue;
1542
1543                         lock_page(page);
1544
1545                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1546 continue_unlock:
1547                                 unlock_page(page);
1548                                 continue;
1549                         }
1550                         if (ino_of_node(page) != ino)
1551                                 goto continue_unlock;
1552
1553                         if (!PageDirty(page)) {
1554                                 /* someone wrote it for us */
1555                                 goto continue_unlock;
1556                         }
1557
1558                         if (last_page)
1559                                 f2fs_put_page(last_page, 0);
1560
1561                         get_page(page);
1562                         last_page = page;
1563                         unlock_page(page);
1564                 }
1565                 pagevec_release(&pvec);
1566                 cond_resched();
1567         }
1568         return last_page;
1569 }
1570
1571 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1572                                 struct writeback_control *wbc, bool do_balance,
1573                                 enum iostat_type io_type, unsigned int *seq_id)
1574 {
1575         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1576         nid_t nid;
1577         struct node_info ni;
1578         struct f2fs_io_info fio = {
1579                 .sbi = sbi,
1580                 .ino = ino_of_node(page),
1581                 .type = NODE,
1582                 .op = REQ_OP_WRITE,
1583                 .op_flags = wbc_to_write_flags(wbc),
1584                 .page = page,
1585                 .encrypted_page = NULL,
1586                 .submitted = false,
1587                 .io_type = io_type,
1588                 .io_wbc = wbc,
1589         };
1590         unsigned int seq;
1591
1592         trace_f2fs_writepage(page, NODE);
1593
1594         if (unlikely(f2fs_cp_error(sbi))) {
1595                 ClearPageUptodate(page);
1596                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1597                 unlock_page(page);
1598                 return 0;
1599         }
1600
1601         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1602                 goto redirty_out;
1603
1604         if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1605                         wbc->sync_mode == WB_SYNC_NONE &&
1606                         IS_DNODE(page) && is_cold_node(page))
1607                 goto redirty_out;
1608
1609         /* get old block addr of this node page */
1610         nid = nid_of_node(page);
1611         f2fs_bug_on(sbi, page->index != nid);
1612
1613         if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1614                 goto redirty_out;
1615
1616         if (wbc->for_reclaim) {
1617                 if (!down_read_trylock(&sbi->node_write))
1618                         goto redirty_out;
1619         } else {
1620                 down_read(&sbi->node_write);
1621         }
1622
1623         /* This page is already truncated */
1624         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1625                 ClearPageUptodate(page);
1626                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1627                 up_read(&sbi->node_write);
1628                 unlock_page(page);
1629                 return 0;
1630         }
1631
1632         if (__is_valid_data_blkaddr(ni.blk_addr) &&
1633                 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1634                                         DATA_GENERIC_ENHANCE)) {
1635                 up_read(&sbi->node_write);
1636                 goto redirty_out;
1637         }
1638
1639         if (atomic && !test_opt(sbi, NOBARRIER))
1640                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1641
1642         /* should add to global list before clearing PAGECACHE status */
1643         if (f2fs_in_warm_node_list(sbi, page)) {
1644                 seq = f2fs_add_fsync_node_entry(sbi, page);
1645                 if (seq_id)
1646                         *seq_id = seq;
1647         }
1648
1649         set_page_writeback(page);
1650         ClearPageError(page);
1651
1652         fio.old_blkaddr = ni.blk_addr;
1653         f2fs_do_write_node_page(nid, &fio);
1654         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1655         dec_page_count(sbi, F2FS_DIRTY_NODES);
1656         up_read(&sbi->node_write);
1657
1658         if (wbc->for_reclaim) {
1659                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1660                 submitted = NULL;
1661         }
1662
1663         unlock_page(page);
1664
1665         if (unlikely(f2fs_cp_error(sbi))) {
1666                 f2fs_submit_merged_write(sbi, NODE);
1667                 submitted = NULL;
1668         }
1669         if (submitted)
1670                 *submitted = fio.submitted;
1671
1672         if (do_balance)
1673                 f2fs_balance_fs(sbi, false);
1674         return 0;
1675
1676 redirty_out:
1677         redirty_page_for_writepage(wbc, page);
1678         return AOP_WRITEPAGE_ACTIVATE;
1679 }
1680
1681 int f2fs_move_node_page(struct page *node_page, int gc_type)
1682 {
1683         int err = 0;
1684
1685         if (gc_type == FG_GC) {
1686                 struct writeback_control wbc = {
1687                         .sync_mode = WB_SYNC_ALL,
1688                         .nr_to_write = 1,
1689                         .for_reclaim = 0,
1690                 };
1691
1692                 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1693
1694                 set_page_dirty(node_page);
1695
1696                 if (!clear_page_dirty_for_io(node_page)) {
1697                         err = -EAGAIN;
1698                         goto out_page;
1699                 }
1700
1701                 if (__write_node_page(node_page, false, NULL,
1702                                         &wbc, false, FS_GC_NODE_IO, NULL)) {
1703                         err = -EAGAIN;
1704                         unlock_page(node_page);
1705                 }
1706                 goto release_page;
1707         } else {
1708                 /* set page dirty and write it */
1709                 if (!PageWriteback(node_page))
1710                         set_page_dirty(node_page);
1711         }
1712 out_page:
1713         unlock_page(node_page);
1714 release_page:
1715         f2fs_put_page(node_page, 0);
1716         return err;
1717 }
1718
1719 static int f2fs_write_node_page(struct page *page,
1720                                 struct writeback_control *wbc)
1721 {
1722         return __write_node_page(page, false, NULL, wbc, false,
1723                                                 FS_NODE_IO, NULL);
1724 }
1725
1726 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1727                         struct writeback_control *wbc, bool atomic,
1728                         unsigned int *seq_id)
1729 {
1730         pgoff_t index;
1731         struct pagevec pvec;
1732         int ret = 0;
1733         struct page *last_page = NULL;
1734         bool marked = false;
1735         nid_t ino = inode->i_ino;
1736         int nr_pages;
1737         int nwritten = 0;
1738
1739         if (atomic) {
1740                 last_page = last_fsync_dnode(sbi, ino);
1741                 if (IS_ERR_OR_NULL(last_page))
1742                         return PTR_ERR_OR_ZERO(last_page);
1743         }
1744 retry:
1745         pagevec_init(&pvec);
1746         index = 0;
1747
1748         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1749                                 PAGECACHE_TAG_DIRTY))) {
1750                 int i;
1751
1752                 for (i = 0; i < nr_pages; i++) {
1753                         struct page *page = pvec.pages[i];
1754                         bool submitted = false;
1755
1756                         if (unlikely(f2fs_cp_error(sbi))) {
1757                                 f2fs_put_page(last_page, 0);
1758                                 pagevec_release(&pvec);
1759                                 ret = -EIO;
1760                                 goto out;
1761                         }
1762
1763                         if (!IS_DNODE(page) || !is_cold_node(page))
1764                                 continue;
1765                         if (ino_of_node(page) != ino)
1766                                 continue;
1767
1768                         lock_page(page);
1769
1770                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1771 continue_unlock:
1772                                 unlock_page(page);
1773                                 continue;
1774                         }
1775                         if (ino_of_node(page) != ino)
1776                                 goto continue_unlock;
1777
1778                         if (!PageDirty(page) && page != last_page) {
1779                                 /* someone wrote it for us */
1780                                 goto continue_unlock;
1781                         }
1782
1783                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1784
1785                         set_fsync_mark(page, 0);
1786                         set_dentry_mark(page, 0);
1787
1788                         if (!atomic || page == last_page) {
1789                                 set_fsync_mark(page, 1);
1790                                 if (IS_INODE(page)) {
1791                                         if (is_inode_flag_set(inode,
1792                                                                 FI_DIRTY_INODE))
1793                                                 f2fs_update_inode(inode, page);
1794                                         set_dentry_mark(page,
1795                                                 f2fs_need_dentry_mark(sbi, ino));
1796                                 }
1797                                 /* may be written by other thread */
1798                                 if (!PageDirty(page))
1799                                         set_page_dirty(page);
1800                         }
1801
1802                         if (!clear_page_dirty_for_io(page))
1803                                 goto continue_unlock;
1804
1805                         ret = __write_node_page(page, atomic &&
1806                                                 page == last_page,
1807                                                 &submitted, wbc, true,
1808                                                 FS_NODE_IO, seq_id);
1809                         if (ret) {
1810                                 unlock_page(page);
1811                                 f2fs_put_page(last_page, 0);
1812                                 break;
1813                         } else if (submitted) {
1814                                 nwritten++;
1815                         }
1816
1817                         if (page == last_page) {
1818                                 f2fs_put_page(page, 0);
1819                                 marked = true;
1820                                 break;
1821                         }
1822                 }
1823                 pagevec_release(&pvec);
1824                 cond_resched();
1825
1826                 if (ret || marked)
1827                         break;
1828         }
1829         if (!ret && atomic && !marked) {
1830                 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1831                            ino, last_page->index);
1832                 lock_page(last_page);
1833                 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1834                 set_page_dirty(last_page);
1835                 unlock_page(last_page);
1836                 goto retry;
1837         }
1838 out:
1839         if (nwritten)
1840                 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1841         return ret ? -EIO : 0;
1842 }
1843
1844 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1845 {
1846         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1847         bool clean;
1848
1849         if (inode->i_ino != ino)
1850                 return 0;
1851
1852         if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1853                 return 0;
1854
1855         spin_lock(&sbi->inode_lock[DIRTY_META]);
1856         clean = list_empty(&F2FS_I(inode)->gdirty_list);
1857         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1858
1859         if (clean)
1860                 return 0;
1861
1862         inode = igrab(inode);
1863         if (!inode)
1864                 return 0;
1865         return 1;
1866 }
1867
1868 static bool flush_dirty_inode(struct page *page)
1869 {
1870         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1871         struct inode *inode;
1872         nid_t ino = ino_of_node(page);
1873
1874         inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1875         if (!inode)
1876                 return false;
1877
1878         f2fs_update_inode(inode, page);
1879         unlock_page(page);
1880
1881         iput(inode);
1882         return true;
1883 }
1884
1885 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1886 {
1887         pgoff_t index = 0;
1888         struct pagevec pvec;
1889         int nr_pages;
1890
1891         pagevec_init(&pvec);
1892
1893         while ((nr_pages = pagevec_lookup_tag(&pvec,
1894                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1895                 int i;
1896
1897                 for (i = 0; i < nr_pages; i++) {
1898                         struct page *page = pvec.pages[i];
1899
1900                         if (!IS_DNODE(page))
1901                                 continue;
1902
1903                         lock_page(page);
1904
1905                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1906 continue_unlock:
1907                                 unlock_page(page);
1908                                 continue;
1909                         }
1910
1911                         if (!PageDirty(page)) {
1912                                 /* someone wrote it for us */
1913                                 goto continue_unlock;
1914                         }
1915
1916                         /* flush inline_data, if it's async context. */
1917                         if (page_private_inline(page)) {
1918                                 clear_page_private_inline(page);
1919                                 unlock_page(page);
1920                                 flush_inline_data(sbi, ino_of_node(page));
1921                                 continue;
1922                         }
1923                         unlock_page(page);
1924                 }
1925                 pagevec_release(&pvec);
1926                 cond_resched();
1927         }
1928 }
1929
1930 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1931                                 struct writeback_control *wbc,
1932                                 bool do_balance, enum iostat_type io_type)
1933 {
1934         pgoff_t index;
1935         struct pagevec pvec;
1936         int step = 0;
1937         int nwritten = 0;
1938         int ret = 0;
1939         int nr_pages, done = 0;
1940
1941         pagevec_init(&pvec);
1942
1943 next_step:
1944         index = 0;
1945
1946         while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1947                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1948                 int i;
1949
1950                 for (i = 0; i < nr_pages; i++) {
1951                         struct page *page = pvec.pages[i];
1952                         bool submitted = false;
1953                         bool may_dirty = true;
1954
1955                         /* give a priority to WB_SYNC threads */
1956                         if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1957                                         wbc->sync_mode == WB_SYNC_NONE) {
1958                                 done = 1;
1959                                 break;
1960                         }
1961
1962                         /*
1963                          * flushing sequence with step:
1964                          * 0. indirect nodes
1965                          * 1. dentry dnodes
1966                          * 2. file dnodes
1967                          */
1968                         if (step == 0 && IS_DNODE(page))
1969                                 continue;
1970                         if (step == 1 && (!IS_DNODE(page) ||
1971                                                 is_cold_node(page)))
1972                                 continue;
1973                         if (step == 2 && (!IS_DNODE(page) ||
1974                                                 !is_cold_node(page)))
1975                                 continue;
1976 lock_node:
1977                         if (wbc->sync_mode == WB_SYNC_ALL)
1978                                 lock_page(page);
1979                         else if (!trylock_page(page))
1980                                 continue;
1981
1982                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1983 continue_unlock:
1984                                 unlock_page(page);
1985                                 continue;
1986                         }
1987
1988                         if (!PageDirty(page)) {
1989                                 /* someone wrote it for us */
1990                                 goto continue_unlock;
1991                         }
1992
1993                         /* flush inline_data/inode, if it's async context. */
1994                         if (!do_balance)
1995                                 goto write_node;
1996
1997                         /* flush inline_data */
1998                         if (page_private_inline(page)) {
1999                                 clear_page_private_inline(page);
2000                                 unlock_page(page);
2001                                 flush_inline_data(sbi, ino_of_node(page));
2002                                 goto lock_node;
2003                         }
2004
2005                         /* flush dirty inode */
2006                         if (IS_INODE(page) && may_dirty) {
2007                                 may_dirty = false;
2008                                 if (flush_dirty_inode(page))
2009                                         goto lock_node;
2010                         }
2011 write_node:
2012                         f2fs_wait_on_page_writeback(page, NODE, true, true);
2013
2014                         if (!clear_page_dirty_for_io(page))
2015                                 goto continue_unlock;
2016
2017                         set_fsync_mark(page, 0);
2018                         set_dentry_mark(page, 0);
2019
2020                         ret = __write_node_page(page, false, &submitted,
2021                                                 wbc, do_balance, io_type, NULL);
2022                         if (ret)
2023                                 unlock_page(page);
2024                         else if (submitted)
2025                                 nwritten++;
2026
2027                         if (--wbc->nr_to_write == 0)
2028                                 break;
2029                 }
2030                 pagevec_release(&pvec);
2031                 cond_resched();
2032
2033                 if (wbc->nr_to_write == 0) {
2034                         step = 2;
2035                         break;
2036                 }
2037         }
2038
2039         if (step < 2) {
2040                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2041                                 wbc->sync_mode == WB_SYNC_NONE && step == 1)
2042                         goto out;
2043                 step++;
2044                 goto next_step;
2045         }
2046 out:
2047         if (nwritten)
2048                 f2fs_submit_merged_write(sbi, NODE);
2049
2050         if (unlikely(f2fs_cp_error(sbi)))
2051                 return -EIO;
2052         return ret;
2053 }
2054
2055 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2056                                                 unsigned int seq_id)
2057 {
2058         struct fsync_node_entry *fn;
2059         struct page *page;
2060         struct list_head *head = &sbi->fsync_node_list;
2061         unsigned long flags;
2062         unsigned int cur_seq_id = 0;
2063         int ret2, ret = 0;
2064
2065         while (seq_id && cur_seq_id < seq_id) {
2066                 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2067                 if (list_empty(head)) {
2068                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2069                         break;
2070                 }
2071                 fn = list_first_entry(head, struct fsync_node_entry, list);
2072                 if (fn->seq_id > seq_id) {
2073                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2074                         break;
2075                 }
2076                 cur_seq_id = fn->seq_id;
2077                 page = fn->page;
2078                 get_page(page);
2079                 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2080
2081                 f2fs_wait_on_page_writeback(page, NODE, true, false);
2082                 if (TestClearPageError(page))
2083                         ret = -EIO;
2084
2085                 put_page(page);
2086
2087                 if (ret)
2088                         break;
2089         }
2090
2091         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2092         if (!ret)
2093                 ret = ret2;
2094
2095         return ret;
2096 }
2097
2098 static int f2fs_write_node_pages(struct address_space *mapping,
2099                             struct writeback_control *wbc)
2100 {
2101         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2102         struct blk_plug plug;
2103         long diff;
2104
2105         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2106                 goto skip_write;
2107
2108         /* balancing f2fs's metadata in background */
2109         f2fs_balance_fs_bg(sbi, true);
2110
2111         /* collect a number of dirty node pages and write together */
2112         if (wbc->sync_mode != WB_SYNC_ALL &&
2113                         get_pages(sbi, F2FS_DIRTY_NODES) <
2114                                         nr_pages_to_skip(sbi, NODE))
2115                 goto skip_write;
2116
2117         if (wbc->sync_mode == WB_SYNC_ALL)
2118                 atomic_inc(&sbi->wb_sync_req[NODE]);
2119         else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2120                 /* to avoid potential deadlock */
2121                 if (current->plug)
2122                         blk_finish_plug(current->plug);
2123                 goto skip_write;
2124         }
2125
2126         trace_f2fs_writepages(mapping->host, wbc, NODE);
2127
2128         diff = nr_pages_to_write(sbi, NODE, wbc);
2129         blk_start_plug(&plug);
2130         f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2131         blk_finish_plug(&plug);
2132         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2133
2134         if (wbc->sync_mode == WB_SYNC_ALL)
2135                 atomic_dec(&sbi->wb_sync_req[NODE]);
2136         return 0;
2137
2138 skip_write:
2139         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2140         trace_f2fs_writepages(mapping->host, wbc, NODE);
2141         return 0;
2142 }
2143
2144 static int f2fs_set_node_page_dirty(struct page *page)
2145 {
2146         trace_f2fs_set_page_dirty(page, NODE);
2147
2148         if (!PageUptodate(page))
2149                 SetPageUptodate(page);
2150 #ifdef CONFIG_F2FS_CHECK_FS
2151         if (IS_INODE(page))
2152                 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2153 #endif
2154         if (!PageDirty(page)) {
2155                 __set_page_dirty_nobuffers(page);
2156                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2157                 set_page_private_reference(page);
2158                 return 1;
2159         }
2160         return 0;
2161 }
2162
2163 /*
2164  * Structure of the f2fs node operations
2165  */
2166 const struct address_space_operations f2fs_node_aops = {
2167         .writepage      = f2fs_write_node_page,
2168         .writepages     = f2fs_write_node_pages,
2169         .set_page_dirty = f2fs_set_node_page_dirty,
2170         .invalidatepage = f2fs_invalidate_page,
2171         .releasepage    = f2fs_release_page,
2172 #ifdef CONFIG_MIGRATION
2173         .migratepage    = f2fs_migrate_page,
2174 #endif
2175 };
2176
2177 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2178                                                 nid_t n)
2179 {
2180         return radix_tree_lookup(&nm_i->free_nid_root, n);
2181 }
2182
2183 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2184                                 struct free_nid *i)
2185 {
2186         struct f2fs_nm_info *nm_i = NM_I(sbi);
2187         int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2188
2189         if (err)
2190                 return err;
2191
2192         nm_i->nid_cnt[FREE_NID]++;
2193         list_add_tail(&i->list, &nm_i->free_nid_list);
2194         return 0;
2195 }
2196
2197 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2198                         struct free_nid *i, enum nid_state state)
2199 {
2200         struct f2fs_nm_info *nm_i = NM_I(sbi);
2201
2202         f2fs_bug_on(sbi, state != i->state);
2203         nm_i->nid_cnt[state]--;
2204         if (state == FREE_NID)
2205                 list_del(&i->list);
2206         radix_tree_delete(&nm_i->free_nid_root, i->nid);
2207 }
2208
2209 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2210                         enum nid_state org_state, enum nid_state dst_state)
2211 {
2212         struct f2fs_nm_info *nm_i = NM_I(sbi);
2213
2214         f2fs_bug_on(sbi, org_state != i->state);
2215         i->state = dst_state;
2216         nm_i->nid_cnt[org_state]--;
2217         nm_i->nid_cnt[dst_state]++;
2218
2219         switch (dst_state) {
2220         case PREALLOC_NID:
2221                 list_del(&i->list);
2222                 break;
2223         case FREE_NID:
2224                 list_add_tail(&i->list, &nm_i->free_nid_list);
2225                 break;
2226         default:
2227                 BUG_ON(1);
2228         }
2229 }
2230
2231 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2232 {
2233         struct f2fs_nm_info *nm_i = NM_I(sbi);
2234         unsigned int i;
2235         bool ret = true;
2236
2237         down_read(&nm_i->nat_tree_lock);
2238         for (i = 0; i < nm_i->nat_blocks; i++) {
2239                 if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2240                         ret = false;
2241                         break;
2242                 }
2243         }
2244         up_read(&nm_i->nat_tree_lock);
2245
2246         return ret;
2247 }
2248
2249 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2250                                                         bool set, bool build)
2251 {
2252         struct f2fs_nm_info *nm_i = NM_I(sbi);
2253         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2254         unsigned int nid_ofs = nid - START_NID(nid);
2255
2256         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2257                 return;
2258
2259         if (set) {
2260                 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2261                         return;
2262                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2263                 nm_i->free_nid_count[nat_ofs]++;
2264         } else {
2265                 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2266                         return;
2267                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2268                 if (!build)
2269                         nm_i->free_nid_count[nat_ofs]--;
2270         }
2271 }
2272
2273 /* return if the nid is recognized as free */
2274 static bool add_free_nid(struct f2fs_sb_info *sbi,
2275                                 nid_t nid, bool build, bool update)
2276 {
2277         struct f2fs_nm_info *nm_i = NM_I(sbi);
2278         struct free_nid *i, *e;
2279         struct nat_entry *ne;
2280         int err = -EINVAL;
2281         bool ret = false;
2282
2283         /* 0 nid should not be used */
2284         if (unlikely(nid == 0))
2285                 return false;
2286
2287         if (unlikely(f2fs_check_nid_range(sbi, nid)))
2288                 return false;
2289
2290         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2291         i->nid = nid;
2292         i->state = FREE_NID;
2293
2294         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2295
2296         spin_lock(&nm_i->nid_list_lock);
2297
2298         if (build) {
2299                 /*
2300                  *   Thread A             Thread B
2301                  *  - f2fs_create
2302                  *   - f2fs_new_inode
2303                  *    - f2fs_alloc_nid
2304                  *     - __insert_nid_to_list(PREALLOC_NID)
2305                  *                     - f2fs_balance_fs_bg
2306                  *                      - f2fs_build_free_nids
2307                  *                       - __f2fs_build_free_nids
2308                  *                        - scan_nat_page
2309                  *                         - add_free_nid
2310                  *                          - __lookup_nat_cache
2311                  *  - f2fs_add_link
2312                  *   - f2fs_init_inode_metadata
2313                  *    - f2fs_new_inode_page
2314                  *     - f2fs_new_node_page
2315                  *      - set_node_addr
2316                  *  - f2fs_alloc_nid_done
2317                  *   - __remove_nid_from_list(PREALLOC_NID)
2318                  *                         - __insert_nid_to_list(FREE_NID)
2319                  */
2320                 ne = __lookup_nat_cache(nm_i, nid);
2321                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2322                                 nat_get_blkaddr(ne) != NULL_ADDR))
2323                         goto err_out;
2324
2325                 e = __lookup_free_nid_list(nm_i, nid);
2326                 if (e) {
2327                         if (e->state == FREE_NID)
2328                                 ret = true;
2329                         goto err_out;
2330                 }
2331         }
2332         ret = true;
2333         err = __insert_free_nid(sbi, i);
2334 err_out:
2335         if (update) {
2336                 update_free_nid_bitmap(sbi, nid, ret, build);
2337                 if (!build)
2338                         nm_i->available_nids++;
2339         }
2340         spin_unlock(&nm_i->nid_list_lock);
2341         radix_tree_preload_end();
2342
2343         if (err)
2344                 kmem_cache_free(free_nid_slab, i);
2345         return ret;
2346 }
2347
2348 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2349 {
2350         struct f2fs_nm_info *nm_i = NM_I(sbi);
2351         struct free_nid *i;
2352         bool need_free = false;
2353
2354         spin_lock(&nm_i->nid_list_lock);
2355         i = __lookup_free_nid_list(nm_i, nid);
2356         if (i && i->state == FREE_NID) {
2357                 __remove_free_nid(sbi, i, FREE_NID);
2358                 need_free = true;
2359         }
2360         spin_unlock(&nm_i->nid_list_lock);
2361
2362         if (need_free)
2363                 kmem_cache_free(free_nid_slab, i);
2364 }
2365
2366 static int scan_nat_page(struct f2fs_sb_info *sbi,
2367                         struct page *nat_page, nid_t start_nid)
2368 {
2369         struct f2fs_nm_info *nm_i = NM_I(sbi);
2370         struct f2fs_nat_block *nat_blk = page_address(nat_page);
2371         block_t blk_addr;
2372         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2373         int i;
2374
2375         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2376
2377         i = start_nid % NAT_ENTRY_PER_BLOCK;
2378
2379         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2380                 if (unlikely(start_nid >= nm_i->max_nid))
2381                         break;
2382
2383                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2384
2385                 if (blk_addr == NEW_ADDR)
2386                         return -EINVAL;
2387
2388                 if (blk_addr == NULL_ADDR) {
2389                         add_free_nid(sbi, start_nid, true, true);
2390                 } else {
2391                         spin_lock(&NM_I(sbi)->nid_list_lock);
2392                         update_free_nid_bitmap(sbi, start_nid, false, true);
2393                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2394                 }
2395         }
2396
2397         return 0;
2398 }
2399
2400 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2401 {
2402         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2403         struct f2fs_journal *journal = curseg->journal;
2404         int i;
2405
2406         down_read(&curseg->journal_rwsem);
2407         for (i = 0; i < nats_in_cursum(journal); i++) {
2408                 block_t addr;
2409                 nid_t nid;
2410
2411                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2412                 nid = le32_to_cpu(nid_in_journal(journal, i));
2413                 if (addr == NULL_ADDR)
2414                         add_free_nid(sbi, nid, true, false);
2415                 else
2416                         remove_free_nid(sbi, nid);
2417         }
2418         up_read(&curseg->journal_rwsem);
2419 }
2420
2421 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2422 {
2423         struct f2fs_nm_info *nm_i = NM_I(sbi);
2424         unsigned int i, idx;
2425         nid_t nid;
2426
2427         down_read(&nm_i->nat_tree_lock);
2428
2429         for (i = 0; i < nm_i->nat_blocks; i++) {
2430                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2431                         continue;
2432                 if (!nm_i->free_nid_count[i])
2433                         continue;
2434                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2435                         idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2436                                                 NAT_ENTRY_PER_BLOCK, idx);
2437                         if (idx >= NAT_ENTRY_PER_BLOCK)
2438                                 break;
2439
2440                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
2441                         add_free_nid(sbi, nid, true, false);
2442
2443                         if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2444                                 goto out;
2445                 }
2446         }
2447 out:
2448         scan_curseg_cache(sbi);
2449
2450         up_read(&nm_i->nat_tree_lock);
2451 }
2452
2453 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2454                                                 bool sync, bool mount)
2455 {
2456         struct f2fs_nm_info *nm_i = NM_I(sbi);
2457         int i = 0, ret;
2458         nid_t nid = nm_i->next_scan_nid;
2459
2460         if (unlikely(nid >= nm_i->max_nid))
2461                 nid = 0;
2462
2463         if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2464                 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2465
2466         /* Enough entries */
2467         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2468                 return 0;
2469
2470         if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2471                 return 0;
2472
2473         if (!mount) {
2474                 /* try to find free nids in free_nid_bitmap */
2475                 scan_free_nid_bits(sbi);
2476
2477                 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2478                         return 0;
2479         }
2480
2481         /* readahead nat pages to be scanned */
2482         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2483                                                         META_NAT, true);
2484
2485         down_read(&nm_i->nat_tree_lock);
2486
2487         while (1) {
2488                 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2489                                                 nm_i->nat_block_bitmap)) {
2490                         struct page *page = get_current_nat_page(sbi, nid);
2491
2492                         if (IS_ERR(page)) {
2493                                 ret = PTR_ERR(page);
2494                         } else {
2495                                 ret = scan_nat_page(sbi, page, nid);
2496                                 f2fs_put_page(page, 1);
2497                         }
2498
2499                         if (ret) {
2500                                 up_read(&nm_i->nat_tree_lock);
2501                                 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2502                                 return ret;
2503                         }
2504                 }
2505
2506                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2507                 if (unlikely(nid >= nm_i->max_nid))
2508                         nid = 0;
2509
2510                 if (++i >= FREE_NID_PAGES)
2511                         break;
2512         }
2513
2514         /* go to the next free nat pages to find free nids abundantly */
2515         nm_i->next_scan_nid = nid;
2516
2517         /* find free nids from current sum_pages */
2518         scan_curseg_cache(sbi);
2519
2520         up_read(&nm_i->nat_tree_lock);
2521
2522         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2523                                         nm_i->ra_nid_pages, META_NAT, false);
2524
2525         return 0;
2526 }
2527
2528 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2529 {
2530         int ret;
2531
2532         mutex_lock(&NM_I(sbi)->build_lock);
2533         ret = __f2fs_build_free_nids(sbi, sync, mount);
2534         mutex_unlock(&NM_I(sbi)->build_lock);
2535
2536         return ret;
2537 }
2538
2539 /*
2540  * If this function returns success, caller can obtain a new nid
2541  * from second parameter of this function.
2542  * The returned nid could be used ino as well as nid when inode is created.
2543  */
2544 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2545 {
2546         struct f2fs_nm_info *nm_i = NM_I(sbi);
2547         struct free_nid *i = NULL;
2548 retry:
2549         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2550                 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2551                 return false;
2552         }
2553
2554         spin_lock(&nm_i->nid_list_lock);
2555
2556         if (unlikely(nm_i->available_nids == 0)) {
2557                 spin_unlock(&nm_i->nid_list_lock);
2558                 return false;
2559         }
2560
2561         /* We should not use stale free nids created by f2fs_build_free_nids */
2562         if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2563                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2564                 i = list_first_entry(&nm_i->free_nid_list,
2565                                         struct free_nid, list);
2566                 *nid = i->nid;
2567
2568                 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2569                 nm_i->available_nids--;
2570
2571                 update_free_nid_bitmap(sbi, *nid, false, false);
2572
2573                 spin_unlock(&nm_i->nid_list_lock);
2574                 return true;
2575         }
2576         spin_unlock(&nm_i->nid_list_lock);
2577
2578         /* Let's scan nat pages and its caches to get free nids */
2579         if (!f2fs_build_free_nids(sbi, true, false))
2580                 goto retry;
2581         return false;
2582 }
2583
2584 /*
2585  * f2fs_alloc_nid() should be called prior to this function.
2586  */
2587 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2588 {
2589         struct f2fs_nm_info *nm_i = NM_I(sbi);
2590         struct free_nid *i;
2591
2592         spin_lock(&nm_i->nid_list_lock);
2593         i = __lookup_free_nid_list(nm_i, nid);
2594         f2fs_bug_on(sbi, !i);
2595         __remove_free_nid(sbi, i, PREALLOC_NID);
2596         spin_unlock(&nm_i->nid_list_lock);
2597
2598         kmem_cache_free(free_nid_slab, i);
2599 }
2600
2601 /*
2602  * f2fs_alloc_nid() should be called prior to this function.
2603  */
2604 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2605 {
2606         struct f2fs_nm_info *nm_i = NM_I(sbi);
2607         struct free_nid *i;
2608         bool need_free = false;
2609
2610         if (!nid)
2611                 return;
2612
2613         spin_lock(&nm_i->nid_list_lock);
2614         i = __lookup_free_nid_list(nm_i, nid);
2615         f2fs_bug_on(sbi, !i);
2616
2617         if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2618                 __remove_free_nid(sbi, i, PREALLOC_NID);
2619                 need_free = true;
2620         } else {
2621                 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2622         }
2623
2624         nm_i->available_nids++;
2625
2626         update_free_nid_bitmap(sbi, nid, true, false);
2627
2628         spin_unlock(&nm_i->nid_list_lock);
2629
2630         if (need_free)
2631                 kmem_cache_free(free_nid_slab, i);
2632 }
2633
2634 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2635 {
2636         struct f2fs_nm_info *nm_i = NM_I(sbi);
2637         int nr = nr_shrink;
2638
2639         if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2640                 return 0;
2641
2642         if (!mutex_trylock(&nm_i->build_lock))
2643                 return 0;
2644
2645         while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2646                 struct free_nid *i, *next;
2647                 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2648
2649                 spin_lock(&nm_i->nid_list_lock);
2650                 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2651                         if (!nr_shrink || !batch ||
2652                                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2653                                 break;
2654                         __remove_free_nid(sbi, i, FREE_NID);
2655                         kmem_cache_free(free_nid_slab, i);
2656                         nr_shrink--;
2657                         batch--;
2658                 }
2659                 spin_unlock(&nm_i->nid_list_lock);
2660         }
2661
2662         mutex_unlock(&nm_i->build_lock);
2663
2664         return nr - nr_shrink;
2665 }
2666
2667 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2668 {
2669         void *src_addr, *dst_addr;
2670         size_t inline_size;
2671         struct page *ipage;
2672         struct f2fs_inode *ri;
2673
2674         ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2675         if (IS_ERR(ipage))
2676                 return PTR_ERR(ipage);
2677
2678         ri = F2FS_INODE(page);
2679         if (ri->i_inline & F2FS_INLINE_XATTR) {
2680                 if (!f2fs_has_inline_xattr(inode)) {
2681                         set_inode_flag(inode, FI_INLINE_XATTR);
2682                         stat_inc_inline_xattr(inode);
2683                 }
2684         } else {
2685                 if (f2fs_has_inline_xattr(inode)) {
2686                         stat_dec_inline_xattr(inode);
2687                         clear_inode_flag(inode, FI_INLINE_XATTR);
2688                 }
2689                 goto update_inode;
2690         }
2691
2692         dst_addr = inline_xattr_addr(inode, ipage);
2693         src_addr = inline_xattr_addr(inode, page);
2694         inline_size = inline_xattr_size(inode);
2695
2696         f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2697         memcpy(dst_addr, src_addr, inline_size);
2698 update_inode:
2699         f2fs_update_inode(inode, ipage);
2700         f2fs_put_page(ipage, 1);
2701         return 0;
2702 }
2703
2704 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2705 {
2706         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2707         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2708         nid_t new_xnid;
2709         struct dnode_of_data dn;
2710         struct node_info ni;
2711         struct page *xpage;
2712         int err;
2713
2714         if (!prev_xnid)
2715                 goto recover_xnid;
2716
2717         /* 1: invalidate the previous xattr nid */
2718         err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2719         if (err)
2720                 return err;
2721
2722         f2fs_invalidate_blocks(sbi, ni.blk_addr);
2723         dec_valid_node_count(sbi, inode, false);
2724         set_node_addr(sbi, &ni, NULL_ADDR, false);
2725
2726 recover_xnid:
2727         /* 2: update xattr nid in inode */
2728         if (!f2fs_alloc_nid(sbi, &new_xnid))
2729                 return -ENOSPC;
2730
2731         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2732         xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2733         if (IS_ERR(xpage)) {
2734                 f2fs_alloc_nid_failed(sbi, new_xnid);
2735                 return PTR_ERR(xpage);
2736         }
2737
2738         f2fs_alloc_nid_done(sbi, new_xnid);
2739         f2fs_update_inode_page(inode);
2740
2741         /* 3: update and set xattr node page dirty */
2742         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2743
2744         set_page_dirty(xpage);
2745         f2fs_put_page(xpage, 1);
2746
2747         return 0;
2748 }
2749
2750 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2751 {
2752         struct f2fs_inode *src, *dst;
2753         nid_t ino = ino_of_node(page);
2754         struct node_info old_ni, new_ni;
2755         struct page *ipage;
2756         int err;
2757
2758         err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2759         if (err)
2760                 return err;
2761
2762         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2763                 return -EINVAL;
2764 retry:
2765         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2766         if (!ipage) {
2767                 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2768                 goto retry;
2769         }
2770
2771         /* Should not use this inode from free nid list */
2772         remove_free_nid(sbi, ino);
2773
2774         if (!PageUptodate(ipage))
2775                 SetPageUptodate(ipage);
2776         fill_node_footer(ipage, ino, ino, 0, true);
2777         set_cold_node(ipage, false);
2778
2779         src = F2FS_INODE(page);
2780         dst = F2FS_INODE(ipage);
2781
2782         memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2783         dst->i_size = 0;
2784         dst->i_blocks = cpu_to_le64(1);
2785         dst->i_links = cpu_to_le32(1);
2786         dst->i_xattr_nid = 0;
2787         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2788         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2789                 dst->i_extra_isize = src->i_extra_isize;
2790
2791                 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2792                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2793                                                         i_inline_xattr_size))
2794                         dst->i_inline_xattr_size = src->i_inline_xattr_size;
2795
2796                 if (f2fs_sb_has_project_quota(sbi) &&
2797                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2798                                                                 i_projid))
2799                         dst->i_projid = src->i_projid;
2800
2801                 if (f2fs_sb_has_inode_crtime(sbi) &&
2802                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2803                                                         i_crtime_nsec)) {
2804                         dst->i_crtime = src->i_crtime;
2805                         dst->i_crtime_nsec = src->i_crtime_nsec;
2806                 }
2807         }
2808
2809         new_ni = old_ni;
2810         new_ni.ino = ino;
2811
2812         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2813                 WARN_ON(1);
2814         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2815         inc_valid_inode_count(sbi);
2816         set_page_dirty(ipage);
2817         f2fs_put_page(ipage, 1);
2818         return 0;
2819 }
2820
2821 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2822                         unsigned int segno, struct f2fs_summary_block *sum)
2823 {
2824         struct f2fs_node *rn;
2825         struct f2fs_summary *sum_entry;
2826         block_t addr;
2827         int i, idx, last_offset, nrpages;
2828
2829         /* scan the node segment */
2830         last_offset = sbi->blocks_per_seg;
2831         addr = START_BLOCK(sbi, segno);
2832         sum_entry = &sum->entries[0];
2833
2834         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2835                 nrpages = bio_max_segs(last_offset - i);
2836
2837                 /* readahead node pages */
2838                 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2839
2840                 for (idx = addr; idx < addr + nrpages; idx++) {
2841                         struct page *page = f2fs_get_tmp_page(sbi, idx);
2842
2843                         if (IS_ERR(page))
2844                                 return PTR_ERR(page);
2845
2846                         rn = F2FS_NODE(page);
2847                         sum_entry->nid = rn->footer.nid;
2848                         sum_entry->version = 0;
2849                         sum_entry->ofs_in_node = 0;
2850                         sum_entry++;
2851                         f2fs_put_page(page, 1);
2852                 }
2853
2854                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2855                                                         addr + nrpages);
2856         }
2857         return 0;
2858 }
2859
2860 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2861 {
2862         struct f2fs_nm_info *nm_i = NM_I(sbi);
2863         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2864         struct f2fs_journal *journal = curseg->journal;
2865         int i;
2866
2867         down_write(&curseg->journal_rwsem);
2868         for (i = 0; i < nats_in_cursum(journal); i++) {
2869                 struct nat_entry *ne;
2870                 struct f2fs_nat_entry raw_ne;
2871                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2872
2873                 if (f2fs_check_nid_range(sbi, nid))
2874                         continue;
2875
2876                 raw_ne = nat_in_journal(journal, i);
2877
2878                 ne = __lookup_nat_cache(nm_i, nid);
2879                 if (!ne) {
2880                         ne = __alloc_nat_entry(sbi, nid, true);
2881                         __init_nat_entry(nm_i, ne, &raw_ne, true);
2882                 }
2883
2884                 /*
2885                  * if a free nat in journal has not been used after last
2886                  * checkpoint, we should remove it from available nids,
2887                  * since later we will add it again.
2888                  */
2889                 if (!get_nat_flag(ne, IS_DIRTY) &&
2890                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2891                         spin_lock(&nm_i->nid_list_lock);
2892                         nm_i->available_nids--;
2893                         spin_unlock(&nm_i->nid_list_lock);
2894                 }
2895
2896                 __set_nat_cache_dirty(nm_i, ne);
2897         }
2898         update_nats_in_cursum(journal, -i);
2899         up_write(&curseg->journal_rwsem);
2900 }
2901
2902 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2903                                                 struct list_head *head, int max)
2904 {
2905         struct nat_entry_set *cur;
2906
2907         if (nes->entry_cnt >= max)
2908                 goto add_out;
2909
2910         list_for_each_entry(cur, head, set_list) {
2911                 if (cur->entry_cnt >= nes->entry_cnt) {
2912                         list_add(&nes->set_list, cur->set_list.prev);
2913                         return;
2914                 }
2915         }
2916 add_out:
2917         list_add_tail(&nes->set_list, head);
2918 }
2919
2920 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2921                                                         unsigned int valid)
2922 {
2923         if (valid == 0) {
2924                 __set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2925                 __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2926                 return;
2927         }
2928
2929         __clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2930         if (valid == NAT_ENTRY_PER_BLOCK)
2931                 __set_bit_le(nat_ofs, nm_i->full_nat_bits);
2932         else
2933                 __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2934 }
2935
2936 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2937                                                 struct page *page)
2938 {
2939         struct f2fs_nm_info *nm_i = NM_I(sbi);
2940         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2941         struct f2fs_nat_block *nat_blk = page_address(page);
2942         int valid = 0;
2943         int i = 0;
2944
2945         if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2946                 return;
2947
2948         if (nat_index == 0) {
2949                 valid = 1;
2950                 i = 1;
2951         }
2952         for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2953                 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2954                         valid++;
2955         }
2956
2957         __update_nat_bits(nm_i, nat_index, valid);
2958 }
2959
2960 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2961 {
2962         struct f2fs_nm_info *nm_i = NM_I(sbi);
2963         unsigned int nat_ofs;
2964
2965         down_read(&nm_i->nat_tree_lock);
2966
2967         for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
2968                 unsigned int valid = 0, nid_ofs = 0;
2969
2970                 /* handle nid zero due to it should never be used */
2971                 if (unlikely(nat_ofs == 0)) {
2972                         valid = 1;
2973                         nid_ofs = 1;
2974                 }
2975
2976                 for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
2977                         if (!test_bit_le(nid_ofs,
2978                                         nm_i->free_nid_bitmap[nat_ofs]))
2979                                 valid++;
2980                 }
2981
2982                 __update_nat_bits(nm_i, nat_ofs, valid);
2983         }
2984
2985         up_read(&nm_i->nat_tree_lock);
2986 }
2987
2988 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2989                 struct nat_entry_set *set, struct cp_control *cpc)
2990 {
2991         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2992         struct f2fs_journal *journal = curseg->journal;
2993         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2994         bool to_journal = true;
2995         struct f2fs_nat_block *nat_blk;
2996         struct nat_entry *ne, *cur;
2997         struct page *page = NULL;
2998
2999         /*
3000          * there are two steps to flush nat entries:
3001          * #1, flush nat entries to journal in current hot data summary block.
3002          * #2, flush nat entries to nat page.
3003          */
3004         if ((cpc->reason & CP_UMOUNT) ||
3005                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3006                 to_journal = false;
3007
3008         if (to_journal) {
3009                 down_write(&curseg->journal_rwsem);
3010         } else {
3011                 page = get_next_nat_page(sbi, start_nid);
3012                 if (IS_ERR(page))
3013                         return PTR_ERR(page);
3014
3015                 nat_blk = page_address(page);
3016                 f2fs_bug_on(sbi, !nat_blk);
3017         }
3018
3019         /* flush dirty nats in nat entry set */
3020         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3021                 struct f2fs_nat_entry *raw_ne;
3022                 nid_t nid = nat_get_nid(ne);
3023                 int offset;
3024
3025                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3026
3027                 if (to_journal) {
3028                         offset = f2fs_lookup_journal_in_cursum(journal,
3029                                                         NAT_JOURNAL, nid, 1);
3030                         f2fs_bug_on(sbi, offset < 0);
3031                         raw_ne = &nat_in_journal(journal, offset);
3032                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
3033                 } else {
3034                         raw_ne = &nat_blk->entries[nid - start_nid];
3035                 }
3036                 raw_nat_from_node_info(raw_ne, &ne->ni);
3037                 nat_reset_flag(ne);
3038                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
3039                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
3040                         add_free_nid(sbi, nid, false, true);
3041                 } else {
3042                         spin_lock(&NM_I(sbi)->nid_list_lock);
3043                         update_free_nid_bitmap(sbi, nid, false, false);
3044                         spin_unlock(&NM_I(sbi)->nid_list_lock);
3045                 }
3046         }
3047
3048         if (to_journal) {
3049                 up_write(&curseg->journal_rwsem);
3050         } else {
3051                 update_nat_bits(sbi, start_nid, page);
3052                 f2fs_put_page(page, 1);
3053         }
3054
3055         /* Allow dirty nats by node block allocation in write_begin */
3056         if (!set->entry_cnt) {
3057                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3058                 kmem_cache_free(nat_entry_set_slab, set);
3059         }
3060         return 0;
3061 }
3062
3063 /*
3064  * This function is called during the checkpointing process.
3065  */
3066 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3067 {
3068         struct f2fs_nm_info *nm_i = NM_I(sbi);
3069         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3070         struct f2fs_journal *journal = curseg->journal;
3071         struct nat_entry_set *setvec[SETVEC_SIZE];
3072         struct nat_entry_set *set, *tmp;
3073         unsigned int found;
3074         nid_t set_idx = 0;
3075         LIST_HEAD(sets);
3076         int err = 0;
3077
3078         /*
3079          * during unmount, let's flush nat_bits before checking
3080          * nat_cnt[DIRTY_NAT].
3081          */
3082         if (cpc->reason & CP_UMOUNT) {
3083                 down_write(&nm_i->nat_tree_lock);
3084                 remove_nats_in_journal(sbi);
3085                 up_write(&nm_i->nat_tree_lock);
3086         }
3087
3088         if (!nm_i->nat_cnt[DIRTY_NAT])
3089                 return 0;
3090
3091         down_write(&nm_i->nat_tree_lock);
3092
3093         /*
3094          * if there are no enough space in journal to store dirty nat
3095          * entries, remove all entries from journal and merge them
3096          * into nat entry set.
3097          */
3098         if (cpc->reason & CP_UMOUNT ||
3099                 !__has_cursum_space(journal,
3100                         nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3101                 remove_nats_in_journal(sbi);
3102
3103         while ((found = __gang_lookup_nat_set(nm_i,
3104                                         set_idx, SETVEC_SIZE, setvec))) {
3105                 unsigned idx;
3106
3107                 set_idx = setvec[found - 1]->set + 1;
3108                 for (idx = 0; idx < found; idx++)
3109                         __adjust_nat_entry_set(setvec[idx], &sets,
3110                                                 MAX_NAT_JENTRIES(journal));
3111         }
3112
3113         /* flush dirty nats in nat entry set */
3114         list_for_each_entry_safe(set, tmp, &sets, set_list) {
3115                 err = __flush_nat_entry_set(sbi, set, cpc);
3116                 if (err)
3117                         break;
3118         }
3119
3120         up_write(&nm_i->nat_tree_lock);
3121         /* Allow dirty nats by node block allocation in write_begin */
3122
3123         return err;
3124 }
3125
3126 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3127 {
3128         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3129         struct f2fs_nm_info *nm_i = NM_I(sbi);
3130         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3131         unsigned int i;
3132         __u64 cp_ver = cur_cp_version(ckpt);
3133         block_t nat_bits_addr;
3134
3135         nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3136         nm_i->nat_bits = f2fs_kvzalloc(sbi,
3137                         nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3138         if (!nm_i->nat_bits)
3139                 return -ENOMEM;
3140
3141         nm_i->full_nat_bits = nm_i->nat_bits + 8;
3142         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3143
3144         if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3145                 return 0;
3146
3147         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3148                                                 nm_i->nat_bits_blocks;
3149         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3150                 struct page *page;
3151
3152                 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3153                 if (IS_ERR(page))
3154                         return PTR_ERR(page);
3155
3156                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3157                                         page_address(page), F2FS_BLKSIZE);
3158                 f2fs_put_page(page, 1);
3159         }
3160
3161         cp_ver |= (cur_cp_crc(ckpt) << 32);
3162         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3163                 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3164                 f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3165                         cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3166                 return 0;
3167         }
3168
3169         f2fs_notice(sbi, "Found nat_bits in checkpoint");
3170         return 0;
3171 }
3172
3173 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3174 {
3175         struct f2fs_nm_info *nm_i = NM_I(sbi);
3176         unsigned int i = 0;
3177         nid_t nid, last_nid;
3178
3179         if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3180                 return;
3181
3182         for (i = 0; i < nm_i->nat_blocks; i++) {
3183                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3184                 if (i >= nm_i->nat_blocks)
3185                         break;
3186
3187                 __set_bit_le(i, nm_i->nat_block_bitmap);
3188
3189                 nid = i * NAT_ENTRY_PER_BLOCK;
3190                 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3191
3192                 spin_lock(&NM_I(sbi)->nid_list_lock);
3193                 for (; nid < last_nid; nid++)
3194                         update_free_nid_bitmap(sbi, nid, true, true);
3195                 spin_unlock(&NM_I(sbi)->nid_list_lock);
3196         }
3197
3198         for (i = 0; i < nm_i->nat_blocks; i++) {
3199                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3200                 if (i >= nm_i->nat_blocks)
3201                         break;
3202
3203                 __set_bit_le(i, nm_i->nat_block_bitmap);
3204         }
3205 }
3206
3207 static int init_node_manager(struct f2fs_sb_info *sbi)
3208 {
3209         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3210         struct f2fs_nm_info *nm_i = NM_I(sbi);
3211         unsigned char *version_bitmap;
3212         unsigned int nat_segs;
3213         int err;
3214
3215         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3216
3217         /* segment_count_nat includes pair segment so divide to 2. */
3218         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3219         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3220         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3221
3222         /* not used nids: 0, node, meta, (and root counted as valid node) */
3223         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3224                                                 F2FS_RESERVED_NODE_NUM;
3225         nm_i->nid_cnt[FREE_NID] = 0;
3226         nm_i->nid_cnt[PREALLOC_NID] = 0;
3227         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3228         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3229         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3230
3231         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3232         INIT_LIST_HEAD(&nm_i->free_nid_list);
3233         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3234         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3235         INIT_LIST_HEAD(&nm_i->nat_entries);
3236         spin_lock_init(&nm_i->nat_list_lock);
3237
3238         mutex_init(&nm_i->build_lock);
3239         spin_lock_init(&nm_i->nid_list_lock);
3240         init_rwsem(&nm_i->nat_tree_lock);
3241
3242         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3243         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3244         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3245         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3246                                         GFP_KERNEL);
3247         if (!nm_i->nat_bitmap)
3248                 return -ENOMEM;
3249
3250         err = __get_nat_bitmaps(sbi);
3251         if (err)
3252                 return err;
3253
3254 #ifdef CONFIG_F2FS_CHECK_FS
3255         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3256                                         GFP_KERNEL);
3257         if (!nm_i->nat_bitmap_mir)
3258                 return -ENOMEM;
3259 #endif
3260
3261         return 0;
3262 }
3263
3264 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3265 {
3266         struct f2fs_nm_info *nm_i = NM_I(sbi);
3267         int i;
3268
3269         nm_i->free_nid_bitmap =
3270                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3271                                               nm_i->nat_blocks),
3272                               GFP_KERNEL);
3273         if (!nm_i->free_nid_bitmap)
3274                 return -ENOMEM;
3275
3276         for (i = 0; i < nm_i->nat_blocks; i++) {
3277                 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3278                         f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3279                 if (!nm_i->free_nid_bitmap[i])
3280                         return -ENOMEM;
3281         }
3282
3283         nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3284                                                                 GFP_KERNEL);
3285         if (!nm_i->nat_block_bitmap)
3286                 return -ENOMEM;
3287
3288         nm_i->free_nid_count =
3289                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3290                                               nm_i->nat_blocks),
3291                               GFP_KERNEL);
3292         if (!nm_i->free_nid_count)
3293                 return -ENOMEM;
3294         return 0;
3295 }
3296
3297 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3298 {
3299         int err;
3300
3301         sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3302                                                         GFP_KERNEL);
3303         if (!sbi->nm_info)
3304                 return -ENOMEM;
3305
3306         err = init_node_manager(sbi);
3307         if (err)
3308                 return err;
3309
3310         err = init_free_nid_cache(sbi);
3311         if (err)
3312                 return err;
3313
3314         /* load free nid status from nat_bits table */
3315         load_free_nid_bitmap(sbi);
3316
3317         return f2fs_build_free_nids(sbi, true, true);
3318 }
3319
3320 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3321 {
3322         struct f2fs_nm_info *nm_i = NM_I(sbi);
3323         struct free_nid *i, *next_i;
3324         struct nat_entry *natvec[NATVEC_SIZE];
3325         struct nat_entry_set *setvec[SETVEC_SIZE];
3326         nid_t nid = 0;
3327         unsigned int found;
3328
3329         if (!nm_i)
3330                 return;
3331
3332         /* destroy free nid list */
3333         spin_lock(&nm_i->nid_list_lock);
3334         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3335                 __remove_free_nid(sbi, i, FREE_NID);
3336                 spin_unlock(&nm_i->nid_list_lock);
3337                 kmem_cache_free(free_nid_slab, i);
3338                 spin_lock(&nm_i->nid_list_lock);
3339         }
3340         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3341         f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3342         f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3343         spin_unlock(&nm_i->nid_list_lock);
3344
3345         /* destroy nat cache */
3346         down_write(&nm_i->nat_tree_lock);
3347         while ((found = __gang_lookup_nat_cache(nm_i,
3348                                         nid, NATVEC_SIZE, natvec))) {
3349                 unsigned idx;
3350
3351                 nid = nat_get_nid(natvec[found - 1]) + 1;
3352                 for (idx = 0; idx < found; idx++) {
3353                         spin_lock(&nm_i->nat_list_lock);
3354                         list_del(&natvec[idx]->list);
3355                         spin_unlock(&nm_i->nat_list_lock);
3356
3357                         __del_from_nat_cache(nm_i, natvec[idx]);
3358                 }
3359         }
3360         f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3361
3362         /* destroy nat set cache */
3363         nid = 0;
3364         while ((found = __gang_lookup_nat_set(nm_i,
3365                                         nid, SETVEC_SIZE, setvec))) {
3366                 unsigned idx;
3367
3368                 nid = setvec[found - 1]->set + 1;
3369                 for (idx = 0; idx < found; idx++) {
3370                         /* entry_cnt is not zero, when cp_error was occurred */
3371                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3372                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3373                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3374                 }
3375         }
3376         up_write(&nm_i->nat_tree_lock);
3377
3378         kvfree(nm_i->nat_block_bitmap);
3379         if (nm_i->free_nid_bitmap) {
3380                 int i;
3381
3382                 for (i = 0; i < nm_i->nat_blocks; i++)
3383                         kvfree(nm_i->free_nid_bitmap[i]);
3384                 kvfree(nm_i->free_nid_bitmap);
3385         }
3386         kvfree(nm_i->free_nid_count);
3387
3388         kvfree(nm_i->nat_bitmap);
3389         kvfree(nm_i->nat_bits);
3390 #ifdef CONFIG_F2FS_CHECK_FS
3391         kvfree(nm_i->nat_bitmap_mir);
3392 #endif
3393         sbi->nm_info = NULL;
3394         kfree(nm_i);
3395 }
3396
3397 int __init f2fs_create_node_manager_caches(void)
3398 {
3399         nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3400                         sizeof(struct nat_entry));
3401         if (!nat_entry_slab)
3402                 goto fail;
3403
3404         free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3405                         sizeof(struct free_nid));
3406         if (!free_nid_slab)
3407                 goto destroy_nat_entry;
3408
3409         nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3410                         sizeof(struct nat_entry_set));
3411         if (!nat_entry_set_slab)
3412                 goto destroy_free_nid;
3413
3414         fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3415                         sizeof(struct fsync_node_entry));
3416         if (!fsync_node_entry_slab)
3417                 goto destroy_nat_entry_set;
3418         return 0;
3419
3420 destroy_nat_entry_set:
3421         kmem_cache_destroy(nat_entry_set_slab);
3422 destroy_free_nid:
3423         kmem_cache_destroy(free_nid_slab);
3424 destroy_nat_entry:
3425         kmem_cache_destroy(nat_entry_slab);
3426 fail:
3427         return -ENOMEM;
3428 }
3429
3430 void f2fs_destroy_node_manager_caches(void)
3431 {
3432         kmem_cache_destroy(fsync_node_entry_slab);
3433         kmem_cache_destroy(nat_entry_set_slab);
3434         kmem_cache_destroy(free_nid_slab);
3435         kmem_cache_destroy(nat_entry_slab);
3436 }