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 "trace.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_msg(sbi->sb, KERN_WARNING,
  38                                 "%s: out-of-range nid=%x, run fsck to fix.",
  39                                 __func__, nid);
  40                 return -EINVAL;
  41         }
  42         return 0;
  43 }
  44
  45 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
  46 {
  47         struct f2fs_nm_info *nm_i = NM_I(sbi);
  48         struct sysinfo val;
  49         unsigned long avail_ram;
  50         unsigned long mem_size = 0;
  51         bool res = false;
  52
  53         si_meminfo(&val);
  54
  55         /* only uses low memory */
  56         avail_ram = val.totalram - val.totalhigh;
  57
  58         /*
  59          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  60          */
  61         if (type == FREE_NIDS) {
  62                 mem_size = (nm_i->nid_cnt[FREE_NID] *
  63                                 sizeof(struct free_nid)) >> PAGE_SHIFT;
  64                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  65         } else if (type == NAT_ENTRIES) {
  66                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
  67                                                         PAGE_SHIFT;
  68                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  69                 if (excess_cached_nats(sbi))
  70                         res = false;
  71         } else if (type == DIRTY_DENTS) {
  72                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  73                         return false;
  74                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  75                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  76         } else if (type == INO_ENTRIES) {
  77                 int i;
  78
  79                 for (i = 0; i < MAX_INO_ENTRY; i++)
  80                         mem_size += sbi->im[i].ino_num *
  81                                                 sizeof(struct ino_entry);
  82                 mem_size >>= PAGE_SHIFT;
  83                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  84         } else if (type == EXTENT_CACHE) {
  85                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  86                                 sizeof(struct extent_tree) +
  87                                 atomic_read(&sbi->total_ext_node) *
  88                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  89                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  90         } else if (type == INMEM_PAGES) {
  91                 /* it allows 20% / total_ram for inmemory pages */
  92                 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
  93                 res = mem_size < (val.totalram / 5);
  94         } else {
  95                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
  96                         return true;
  97         }
  98         return res;
  99 }
 100
 101 static void clear_node_page_dirty(struct page *page)
 102 {
 103         if (PageDirty(page)) {
 104                 f2fs_clear_page_cache_dirty_tag(page);
 105                 clear_page_dirty_for_io(page);
 106                 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
 107         }
 108         ClearPageUptodate(page);
 109 }
 110
 111 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 112 {
 113         return f2fs_get_meta_page_nofail(sbi, current_nat_addr(sbi, nid));
 114 }
 115
 116 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 117 {
 118         struct page *src_page;
 119         struct page *dst_page;
 120         pgoff_t dst_off;
 121         void *src_addr;
 122         void *dst_addr;
 123         struct f2fs_nm_info *nm_i = NM_I(sbi);
 124
 125         dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
 126
 127         /* get current nat block page with lock */
 128         src_page = get_current_nat_page(sbi, nid);
 129         if (IS_ERR(src_page))
 130                 return src_page;
 131         dst_page = f2fs_grab_meta_page(sbi, dst_off);
 132         f2fs_bug_on(sbi, PageDirty(src_page));
 133
 134         src_addr = page_address(src_page);
 135         dst_addr = page_address(dst_page);
 136         memcpy(dst_addr, src_addr, PAGE_SIZE);
 137         set_page_dirty(dst_page);
 138         f2fs_put_page(src_page, 1);
 139
 140         set_to_next_nat(nm_i, nid);
 141
 142         return dst_page;
 143 }
 144
 145 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
 146 {
 147         struct nat_entry *new;
 148
 149         if (no_fail)
 150                 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
 151         else
 152                 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
 153         if (new) {
 154                 nat_set_nid(new, nid);
 155                 nat_reset_flag(new);
 156         }
 157         return new;
 158 }
 159
 160 static void __free_nat_entry(struct nat_entry *e)
 161 {
 162         kmem_cache_free(nat_entry_slab, e);
 163 }
 164
 165 /* must be locked by nat_tree_lock */
 166 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
 167         struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
 168 {
 169         if (no_fail)
 170                 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
 171         else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
 172                 return NULL;
 173
 174         if (raw_ne)
 175                 node_info_from_raw_nat(&ne->ni, raw_ne);
 176
 177         spin_lock(&nm_i->nat_list_lock);
 178         list_add_tail(&ne->list, &nm_i->nat_entries);
 179         spin_unlock(&nm_i->nat_list_lock);
 180
 181         nm_i->nat_cnt++;
 182         return ne;
 183 }
 184
 185 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 186 {
 187         struct nat_entry *ne;
 188
 189         ne = radix_tree_lookup(&nm_i->nat_root, n);
 190
 191         /* for recent accessed nat entry, move it to tail of lru list */
 192         if (ne && !get_nat_flag(ne, IS_DIRTY)) {
 193                 spin_lock(&nm_i->nat_list_lock);
 194                 if (!list_empty(&ne->list))
 195                         list_move_tail(&ne->list, &nm_i->nat_entries);
 196                 spin_unlock(&nm_i->nat_list_lock);
 197         }
 198
 199         return ne;
 200 }
 201
 202 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 203                 nid_t start, unsigned int nr, struct nat_entry **ep)
 204 {
 205         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 206 }
 207
 208 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 209 {
 210         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 211         nm_i->nat_cnt--;
 212         __free_nat_entry(e);
 213 }
 214
 215 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
 216                                                         struct nat_entry *ne)
 217 {
 218         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 219         struct nat_entry_set *head;
 220
 221         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 222         if (!head) {
 223                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 224
 225                 INIT_LIST_HEAD(&head->entry_list);
 226                 INIT_LIST_HEAD(&head->set_list);
 227                 head->set = set;
 228                 head->entry_cnt = 0;
 229                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 230         }
 231         return head;
 232 }
 233
 234 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 235                                                 struct nat_entry *ne)
 236 {
 237         struct nat_entry_set *head;
 238         bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
 239
 240         if (!new_ne)
 241                 head = __grab_nat_entry_set(nm_i, ne);
 242
 243         /*
 244          * update entry_cnt in below condition:
 245          * 1. update NEW_ADDR to valid block address;
 246          * 2. update old block address to new one;
 247          */
 248         if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
 249                                 !get_nat_flag(ne, IS_DIRTY)))
 250                 head->entry_cnt++;
 251
 252         set_nat_flag(ne, IS_PREALLOC, new_ne);
 253
 254         if (get_nat_flag(ne, IS_DIRTY))
 255                 goto refresh_list;
 256
 257         nm_i->dirty_nat_cnt++;
 258         set_nat_flag(ne, IS_DIRTY, true);
 259 refresh_list:
 260         spin_lock(&nm_i->nat_list_lock);
 261         if (new_ne)
 262                 list_del_init(&ne->list);
 263         else
 264                 list_move_tail(&ne->list, &head->entry_list);
 265         spin_unlock(&nm_i->nat_list_lock);
 266 }
 267
 268 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 269                 struct nat_entry_set *set, struct nat_entry *ne)
 270 {
 271         spin_lock(&nm_i->nat_list_lock);
 272         list_move_tail(&ne->list, &nm_i->nat_entries);
 273         spin_unlock(&nm_i->nat_list_lock);
 274
 275         set_nat_flag(ne, IS_DIRTY, false);
 276         set->entry_cnt--;
 277         nm_i->dirty_nat_cnt--;
 278 }
 279
 280 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 281                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 282 {
 283         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 284                                                         start, nr);
 285 }
 286
 287 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
 288 {
 289         return NODE_MAPPING(sbi) == page->mapping &&
 290                         IS_DNODE(page) && is_cold_node(page);
 291 }
 292
 293 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
 294 {
 295         spin_lock_init(&sbi->fsync_node_lock);
 296         INIT_LIST_HEAD(&sbi->fsync_node_list);
 297         sbi->fsync_seg_id = 0;
 298         sbi->fsync_node_num = 0;
 299 }
 300
 301 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
 302                                                         struct page *page)
 303 {
 304         struct fsync_node_entry *fn;
 305         unsigned long flags;
 306         unsigned int seq_id;
 307
 308         fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
 309
 310         get_page(page);
 311         fn->page = page;
 312         INIT_LIST_HEAD(&fn->list);
 313
 314         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 315         list_add_tail(&fn->list, &sbi->fsync_node_list);
 316         fn->seq_id = sbi->fsync_seg_id++;
 317         seq_id = fn->seq_id;
 318         sbi->fsync_node_num++;
 319         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 320
 321         return seq_id;
 322 }
 323
 324 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
 325 {
 326         struct fsync_node_entry *fn;
 327         unsigned long flags;
 328
 329         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 330         list_for_each_entry(fn, &sbi->fsync_node_list, list) {
 331                 if (fn->page == page) {
 332                         list_del(&fn->list);
 333                         sbi->fsync_node_num--;
 334                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 335                         kmem_cache_free(fsync_node_entry_slab, fn);
 336                         put_page(page);
 337                         return;
 338                 }
 339         }
 340         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 341         f2fs_bug_on(sbi, 1);
 342 }
 343
 344 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
 345 {
 346         unsigned long flags;
 347
 348         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 349         sbi->fsync_seg_id = 0;
 350         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 351 }
 352
 353 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 354 {
 355         struct f2fs_nm_info *nm_i = NM_I(sbi);
 356         struct nat_entry *e;
 357         bool need = false;
 358
 359         down_read(&nm_i->nat_tree_lock);
 360         e = __lookup_nat_cache(nm_i, nid);
 361         if (e) {
 362                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 363                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 364                         need = true;
 365         }
 366         up_read(&nm_i->nat_tree_lock);
 367         return need;
 368 }
 369
 370 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 371 {
 372         struct f2fs_nm_info *nm_i = NM_I(sbi);
 373         struct nat_entry *e;
 374         bool is_cp = true;
 375
 376         down_read(&nm_i->nat_tree_lock);
 377         e = __lookup_nat_cache(nm_i, nid);
 378         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 379                 is_cp = false;
 380         up_read(&nm_i->nat_tree_lock);
 381         return is_cp;
 382 }
 383
 384 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 385 {
 386         struct f2fs_nm_info *nm_i = NM_I(sbi);
 387         struct nat_entry *e;
 388         bool need_update = true;
 389
 390         down_read(&nm_i->nat_tree_lock);
 391         e = __lookup_nat_cache(nm_i, ino);
 392         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 393                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 394                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 395                 need_update = false;
 396         up_read(&nm_i->nat_tree_lock);
 397         return need_update;
 398 }
 399
 400 /* must be locked by nat_tree_lock */
 401 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 402                                                 struct f2fs_nat_entry *ne)
 403 {
 404         struct f2fs_nm_info *nm_i = NM_I(sbi);
 405         struct nat_entry *new, *e;
 406
 407         new = __alloc_nat_entry(nid, false);
 408         if (!new)
 409                 return;
 410
 411         down_write(&nm_i->nat_tree_lock);
 412         e = __lookup_nat_cache(nm_i, nid);
 413         if (!e)
 414                 e = __init_nat_entry(nm_i, new, ne, false);
 415         else
 416                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 417                                 nat_get_blkaddr(e) !=
 418                                         le32_to_cpu(ne->block_addr) ||
 419                                 nat_get_version(e) != ne->version);
 420         up_write(&nm_i->nat_tree_lock);
 421         if (e != new)
 422                 __free_nat_entry(new);
 423 }
 424
 425 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 426                         block_t new_blkaddr, bool fsync_done)
 427 {
 428         struct f2fs_nm_info *nm_i = NM_I(sbi);
 429         struct nat_entry *e;
 430         struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
 431
 432         down_write(&nm_i->nat_tree_lock);
 433         e = __lookup_nat_cache(nm_i, ni->nid);
 434         if (!e) {
 435                 e = __init_nat_entry(nm_i, new, NULL, true);
 436                 copy_node_info(&e->ni, ni);
 437                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 438         } else if (new_blkaddr == NEW_ADDR) {
 439                 /*
 440                  * when nid is reallocated,
 441                  * previous nat entry can be remained in nat cache.
 442                  * So, reinitialize it with new information.
 443                  */
 444                 copy_node_info(&e->ni, ni);
 445                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 446         }
 447         /* let's free early to reduce memory consumption */
 448         if (e != new)
 449                 __free_nat_entry(new);
 450
 451         /* sanity check */
 452         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 453         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 454                         new_blkaddr == NULL_ADDR);
 455         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 456                         new_blkaddr == NEW_ADDR);
 457         f2fs_bug_on(sbi, is_valid_data_blkaddr(sbi, nat_get_blkaddr(e)) &&
 458                         new_blkaddr == NEW_ADDR);
 459
 460         /* increment version no as node is removed */
 461         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 462                 unsigned char version = nat_get_version(e);
 463                 nat_set_version(e, inc_node_version(version));
 464         }
 465
 466         /* change address */
 467         nat_set_blkaddr(e, new_blkaddr);
 468         if (!is_valid_data_blkaddr(sbi, new_blkaddr))
 469                 set_nat_flag(e, IS_CHECKPOINTED, false);
 470         __set_nat_cache_dirty(nm_i, e);
 471
 472         /* update fsync_mark if its inode nat entry is still alive */
 473         if (ni->nid != ni->ino)
 474                 e = __lookup_nat_cache(nm_i, ni->ino);
 475         if (e) {
 476                 if (fsync_done && ni->nid == ni->ino)
 477                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 478                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 479         }
 480         up_write(&nm_i->nat_tree_lock);
 481 }
 482
 483 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 484 {
 485         struct f2fs_nm_info *nm_i = NM_I(sbi);
 486         int nr = nr_shrink;
 487
 488         if (!down_write_trylock(&nm_i->nat_tree_lock))
 489                 return 0;
 490
 491         spin_lock(&nm_i->nat_list_lock);
 492         while (nr_shrink) {
 493                 struct nat_entry *ne;
 494
 495                 if (list_empty(&nm_i->nat_entries))
 496                         break;
 497
 498                 ne = list_first_entry(&nm_i->nat_entries,
 499                                         struct nat_entry, list);
 500                 list_del(&ne->list);
 501                 spin_unlock(&nm_i->nat_list_lock);
 502
 503                 __del_from_nat_cache(nm_i, ne);
 504                 nr_shrink--;
 505
 506                 spin_lock(&nm_i->nat_list_lock);
 507         }
 508         spin_unlock(&nm_i->nat_list_lock);
 509
 510         up_write(&nm_i->nat_tree_lock);
 511         return nr - nr_shrink;
 512 }
 513
 514 /*
 515  * This function always returns success
 516  */
 517 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
 518                                                 struct node_info *ni)
 519 {
 520         struct f2fs_nm_info *nm_i = NM_I(sbi);
 521         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 522         struct f2fs_journal *journal = curseg->journal;
 523         nid_t start_nid = START_NID(nid);
 524         struct f2fs_nat_block *nat_blk;
 525         struct page *page = NULL;
 526         struct f2fs_nat_entry ne;
 527         struct nat_entry *e;
 528         pgoff_t index;
 529         int i;
 530
 531         ni->nid = nid;
 532
 533         /* Check nat cache */
 534         down_read(&nm_i->nat_tree_lock);
 535         e = __lookup_nat_cache(nm_i, nid);
 536         if (e) {
 537                 ni->ino = nat_get_ino(e);
 538                 ni->blk_addr = nat_get_blkaddr(e);
 539                 ni->version = nat_get_version(e);
 540                 up_read(&nm_i->nat_tree_lock);
 541                 return 0;
 542         }
 543
 544         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 545
 546         /* Check current segment summary */
 547         down_read(&curseg->journal_rwsem);
 548         i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 549         if (i >= 0) {
 550                 ne = nat_in_journal(journal, i);
 551                 node_info_from_raw_nat(ni, &ne);
 552         }
 553         up_read(&curseg->journal_rwsem);
 554         if (i >= 0) {
 555                 up_read(&nm_i->nat_tree_lock);
 556                 goto cache;
 557         }
 558
 559         /* Fill node_info from nat page */
 560         index = current_nat_addr(sbi, nid);
 561         up_read(&nm_i->nat_tree_lock);
 562
 563         page = f2fs_get_meta_page(sbi, index);
 564         if (IS_ERR(page))
 565                 return PTR_ERR(page);
 566
 567         nat_blk = (struct f2fs_nat_block *)page_address(page);
 568         ne = nat_blk->entries[nid - start_nid];
 569         node_info_from_raw_nat(ni, &ne);
 570         f2fs_put_page(page, 1);
 571 cache:
 572         /* cache nat entry */
 573         cache_nat_entry(sbi, nid, &ne);
 574         return 0;
 575 }
 576
 577 /*
 578  * readahead MAX_RA_NODE number of node pages.
 579  */
 580 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
 581 {
 582         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 583         struct blk_plug plug;
 584         int i, end;
 585         nid_t nid;
 586
 587         blk_start_plug(&plug);
 588
 589         /* Then, try readahead for siblings of the desired node */
 590         end = start + n;
 591         end = min(end, NIDS_PER_BLOCK);
 592         for (i = start; i < end; i++) {
 593                 nid = get_nid(parent, i, false);
 594                 f2fs_ra_node_page(sbi, nid);
 595         }
 596
 597         blk_finish_plug(&plug);
 598 }
 599
 600 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 601 {
 602         const long direct_index = ADDRS_PER_INODE(dn->inode);
 603         const long direct_blks = ADDRS_PER_BLOCK;
 604         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 605         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 606         int cur_level = dn->cur_level;
 607         int max_level = dn->max_level;
 608         pgoff_t base = 0;
 609
 610         if (!dn->max_level)
 611                 return pgofs + 1;
 612
 613         while (max_level-- > cur_level)
 614                 skipped_unit *= NIDS_PER_BLOCK;
 615
 616         switch (dn->max_level) {
 617         case 3:
 618                 base += 2 * indirect_blks;
 619         case 2:
 620                 base += 2 * direct_blks;
 621         case 1:
 622                 base += direct_index;
 623                 break;
 624         default:
 625                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 626         }
 627
 628         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 629 }
 630
 631 /*
 632  * The maximum depth is four.
 633  * Offset[0] will have raw inode offset.
 634  */
 635 static int get_node_path(struct inode *inode, long block,
 636                                 int offset[4], unsigned int noffset[4])
 637 {
 638         const long direct_index = ADDRS_PER_INODE(inode);
 639         const long direct_blks = ADDRS_PER_BLOCK;
 640         const long dptrs_per_blk = NIDS_PER_BLOCK;
 641         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 642         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 643         int n = 0;
 644         int level = 0;
 645
 646         noffset[0] = 0;
 647
 648         if (block < direct_index) {
 649                 offset[n] = block;
 650                 goto got;
 651         }
 652         block -= direct_index;
 653         if (block < direct_blks) {
 654                 offset[n++] = NODE_DIR1_BLOCK;
 655                 noffset[n] = 1;
 656                 offset[n] = block;
 657                 level = 1;
 658                 goto got;
 659         }
 660         block -= direct_blks;
 661         if (block < direct_blks) {
 662                 offset[n++] = NODE_DIR2_BLOCK;
 663                 noffset[n] = 2;
 664                 offset[n] = block;
 665                 level = 1;
 666                 goto got;
 667         }
 668         block -= direct_blks;
 669         if (block < indirect_blks) {
 670                 offset[n++] = NODE_IND1_BLOCK;
 671                 noffset[n] = 3;
 672                 offset[n++] = block / direct_blks;
 673                 noffset[n] = 4 + offset[n - 1];
 674                 offset[n] = block % direct_blks;
 675                 level = 2;
 676                 goto got;
 677         }
 678         block -= indirect_blks;
 679         if (block < indirect_blks) {
 680                 offset[n++] = NODE_IND2_BLOCK;
 681                 noffset[n] = 4 + dptrs_per_blk;
 682                 offset[n++] = block / direct_blks;
 683                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 684                 offset[n] = block % direct_blks;
 685                 level = 2;
 686                 goto got;
 687         }
 688         block -= indirect_blks;
 689         if (block < dindirect_blks) {
 690                 offset[n++] = NODE_DIND_BLOCK;
 691                 noffset[n] = 5 + (dptrs_per_blk * 2);
 692                 offset[n++] = block / indirect_blks;
 693                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 694                               offset[n - 1] * (dptrs_per_blk + 1);
 695                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 696                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 697                               offset[n - 2] * (dptrs_per_blk + 1) +
 698                               offset[n - 1];
 699                 offset[n] = block % direct_blks;
 700                 level = 3;
 701                 goto got;
 702         } else {
 703                 return -E2BIG;
 704         }
 705 got:
 706         return level;
 707 }
 708
 709 /*
 710  * Caller should call f2fs_put_dnode(dn).
 711  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 712  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 713  * In the case of RDONLY_NODE, we don't need to care about mutex.
 714  */
 715 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 716 {
 717         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 718         struct page *npage[4];
 719         struct page *parent = NULL;
 720         int offset[4];
 721         unsigned int noffset[4];
 722         nid_t nids[4];
 723         int level, i = 0;
 724         int err = 0;
 725
 726         level = get_node_path(dn->inode, index, offset, noffset);
 727         if (level < 0)
 728                 return level;
 729
 730         nids[0] = dn->inode->i_ino;
 731         npage[0] = dn->inode_page;
 732
 733         if (!npage[0]) {
 734                 npage[0] = f2fs_get_node_page(sbi, nids[0]);
 735                 if (IS_ERR(npage[0]))
 736                         return PTR_ERR(npage[0]);
 737         }
 738
 739         /* if inline_data is set, should not report any block indices */
 740         if (f2fs_has_inline_data(dn->inode) && index) {
 741                 err = -ENOENT;
 742                 f2fs_put_page(npage[0], 1);
 743                 goto release_out;
 744         }
 745
 746         parent = npage[0];
 747         if (level != 0)
 748                 nids[1] = get_nid(parent, offset[0], true);
 749         dn->inode_page = npage[0];
 750         dn->inode_page_locked = true;
 751
 752         /* get indirect or direct nodes */
 753         for (i = 1; i <= level; i++) {
 754                 bool done = false;
 755
 756                 if (!nids[i] && mode == ALLOC_NODE) {
 757                         /* alloc new node */
 758                         if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
 759                                 err = -ENOSPC;
 760                                 goto release_pages;
 761                         }
 762
 763                         dn->nid = nids[i];
 764                         npage[i] = f2fs_new_node_page(dn, noffset[i]);
 765                         if (IS_ERR(npage[i])) {
 766                                 f2fs_alloc_nid_failed(sbi, nids[i]);
 767                                 err = PTR_ERR(npage[i]);
 768                                 goto release_pages;
 769                         }
 770
 771                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 772                         f2fs_alloc_nid_done(sbi, nids[i]);
 773                         done = true;
 774                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 775                         npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
 776                         if (IS_ERR(npage[i])) {
 777                                 err = PTR_ERR(npage[i]);
 778                                 goto release_pages;
 779                         }
 780                         done = true;
 781                 }
 782                 if (i == 1) {
 783                         dn->inode_page_locked = false;
 784                         unlock_page(parent);
 785                 } else {
 786                         f2fs_put_page(parent, 1);
 787                 }
 788
 789                 if (!done) {
 790                         npage[i] = f2fs_get_node_page(sbi, nids[i]);
 791                         if (IS_ERR(npage[i])) {
 792                                 err = PTR_ERR(npage[i]);
 793                                 f2fs_put_page(npage[0], 0);
 794                                 goto release_out;
 795                         }
 796                 }
 797                 if (i < level) {
 798                         parent = npage[i];
 799                         nids[i + 1] = get_nid(parent, offset[i], false);
 800                 }
 801         }
 802         dn->nid = nids[level];
 803         dn->ofs_in_node = offset[level];
 804         dn->node_page = npage[level];
 805         dn->data_blkaddr = datablock_addr(dn->inode,
 806                                 dn->node_page, dn->ofs_in_node);
 807         return 0;
 808
 809 release_pages:
 810         f2fs_put_page(parent, 1);
 811         if (i > 1)
 812                 f2fs_put_page(npage[0], 0);
 813 release_out:
 814         dn->inode_page = NULL;
 815         dn->node_page = NULL;
 816         if (err == -ENOENT) {
 817                 dn->cur_level = i;
 818                 dn->max_level = level;
 819                 dn->ofs_in_node = offset[level];
 820         }
 821         return err;
 822 }
 823
 824 static int truncate_node(struct dnode_of_data *dn)
 825 {
 826         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 827         struct node_info ni;
 828         int err;
 829         pgoff_t index;
 830
 831         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 832         if (err)
 833                 return err;
 834
 835         /* Deallocate node address */
 836         f2fs_invalidate_blocks(sbi, ni.blk_addr);
 837         dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
 838         set_node_addr(sbi, &ni, NULL_ADDR, false);
 839
 840         if (dn->nid == dn->inode->i_ino) {
 841                 f2fs_remove_orphan_inode(sbi, dn->nid);
 842                 dec_valid_inode_count(sbi);
 843                 f2fs_inode_synced(dn->inode);
 844         }
 845
 846         clear_node_page_dirty(dn->node_page);
 847         set_sbi_flag(sbi, SBI_IS_DIRTY);
 848
 849         index = dn->node_page->index;
 850         f2fs_put_page(dn->node_page, 1);
 851
 852         invalidate_mapping_pages(NODE_MAPPING(sbi),
 853                         index, index);
 854
 855         dn->node_page = NULL;
 856         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 857
 858         return 0;
 859 }
 860
 861 static int truncate_dnode(struct dnode_of_data *dn)
 862 {
 863         struct page *page;
 864         int err;
 865
 866         if (dn->nid == 0)
 867                 return 1;
 868
 869         /* get direct node */
 870         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 871         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 872                 return 1;
 873         else if (IS_ERR(page))
 874                 return PTR_ERR(page);
 875
 876         /* Make dnode_of_data for parameter */
 877         dn->node_page = page;
 878         dn->ofs_in_node = 0;
 879         f2fs_truncate_data_blocks(dn);
 880         err = truncate_node(dn);
 881         if (err)
 882                 return err;
 883
 884         return 1;
 885 }
 886
 887 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 888                                                 int ofs, int depth)
 889 {
 890         struct dnode_of_data rdn = *dn;
 891         struct page *page;
 892         struct f2fs_node *rn;
 893         nid_t child_nid;
 894         unsigned int child_nofs;
 895         int freed = 0;
 896         int i, ret;
 897
 898         if (dn->nid == 0)
 899                 return NIDS_PER_BLOCK + 1;
 900
 901         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 902
 903         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 904         if (IS_ERR(page)) {
 905                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 906                 return PTR_ERR(page);
 907         }
 908
 909         f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 910
 911         rn = F2FS_NODE(page);
 912         if (depth < 3) {
 913                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 914                         child_nid = le32_to_cpu(rn->in.nid[i]);
 915                         if (child_nid == 0)
 916                                 continue;
 917                         rdn.nid = child_nid;
 918                         ret = truncate_dnode(&rdn);
 919                         if (ret < 0)
 920                                 goto out_err;
 921                         if (set_nid(page, i, 0, false))
 922                                 dn->node_changed = true;
 923                 }
 924         } else {
 925                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 926                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 927                         child_nid = le32_to_cpu(rn->in.nid[i]);
 928                         if (child_nid == 0) {
 929                                 child_nofs += NIDS_PER_BLOCK + 1;
 930                                 continue;
 931                         }
 932                         rdn.nid = child_nid;
 933                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 934                         if (ret == (NIDS_PER_BLOCK + 1)) {
 935                                 if (set_nid(page, i, 0, false))
 936                                         dn->node_changed = true;
 937                                 child_nofs += ret;
 938                         } else if (ret < 0 && ret != -ENOENT) {
 939                                 goto out_err;
 940                         }
 941                 }
 942                 freed = child_nofs;
 943         }
 944
 945         if (!ofs) {
 946                 /* remove current indirect node */
 947                 dn->node_page = page;
 948                 ret = truncate_node(dn);
 949                 if (ret)
 950                         goto out_err;
 951                 freed++;
 952         } else {
 953                 f2fs_put_page(page, 1);
 954         }
 955         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 956         return freed;
 957
 958 out_err:
 959         f2fs_put_page(page, 1);
 960         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 961         return ret;
 962 }
 963
 964 static int truncate_partial_nodes(struct dnode_of_data *dn,
 965                         struct f2fs_inode *ri, int *offset, int depth)
 966 {
 967         struct page *pages[2];
 968         nid_t nid[3];
 969         nid_t child_nid;
 970         int err = 0;
 971         int i;
 972         int idx = depth - 2;
 973
 974         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 975         if (!nid[0])
 976                 return 0;
 977
 978         /* get indirect nodes in the path */
 979         for (i = 0; i < idx + 1; i++) {
 980                 /* reference count'll be increased */
 981                 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 982                 if (IS_ERR(pages[i])) {
 983                         err = PTR_ERR(pages[i]);
 984                         idx = i - 1;
 985                         goto fail;
 986                 }
 987                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 988         }
 989
 990         f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 991
 992         /* free direct nodes linked to a partial indirect node */
 993         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 994                 child_nid = get_nid(pages[idx], i, false);
 995                 if (!child_nid)
 996                         continue;
 997                 dn->nid = child_nid;
 998                 err = truncate_dnode(dn);
 999                 if (err < 0)
1000                         goto fail;
1001                 if (set_nid(pages[idx], i, 0, false))
1002                         dn->node_changed = true;
1003         }
1004
1005         if (offset[idx + 1] == 0) {
1006                 dn->node_page = pages[idx];
1007                 dn->nid = nid[idx];
1008                 err = truncate_node(dn);
1009                 if (err)
1010                         goto fail;
1011         } else {
1012                 f2fs_put_page(pages[idx], 1);
1013         }
1014         offset[idx]++;
1015         offset[idx + 1] = 0;
1016         idx--;
1017 fail:
1018         for (i = idx; i >= 0; i--)
1019                 f2fs_put_page(pages[i], 1);
1020
1021         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1022
1023         return err;
1024 }
1025
1026 /*
1027  * All the block addresses of data and nodes should be nullified.
1028  */
1029 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1030 {
1031         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1032         int err = 0, cont = 1;
1033         int level, offset[4], noffset[4];
1034         unsigned int nofs = 0;
1035         struct f2fs_inode *ri;
1036         struct dnode_of_data dn;
1037         struct page *page;
1038
1039         trace_f2fs_truncate_inode_blocks_enter(inode, from);
1040
1041         level = get_node_path(inode, from, offset, noffset);
1042         if (level < 0)
1043                 return level;
1044
1045         page = f2fs_get_node_page(sbi, inode->i_ino);
1046         if (IS_ERR(page)) {
1047                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1048                 return PTR_ERR(page);
1049         }
1050
1051         set_new_dnode(&dn, inode, page, NULL, 0);
1052         unlock_page(page);
1053
1054         ri = F2FS_INODE(page);
1055         switch (level) {
1056         case 0:
1057         case 1:
1058                 nofs = noffset[1];
1059                 break;
1060         case 2:
1061                 nofs = noffset[1];
1062                 if (!offset[level - 1])
1063                         goto skip_partial;
1064                 err = truncate_partial_nodes(&dn, ri, offset, level);
1065                 if (err < 0 && err != -ENOENT)
1066                         goto fail;
1067                 nofs += 1 + NIDS_PER_BLOCK;
1068                 break;
1069         case 3:
1070                 nofs = 5 + 2 * NIDS_PER_BLOCK;
1071                 if (!offset[level - 1])
1072                         goto skip_partial;
1073                 err = truncate_partial_nodes(&dn, ri, offset, level);
1074                 if (err < 0 && err != -ENOENT)
1075                         goto fail;
1076                 break;
1077         default:
1078                 BUG();
1079         }
1080
1081 skip_partial:
1082         while (cont) {
1083                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1084                 switch (offset[0]) {
1085                 case NODE_DIR1_BLOCK:
1086                 case NODE_DIR2_BLOCK:
1087                         err = truncate_dnode(&dn);
1088                         break;
1089
1090                 case NODE_IND1_BLOCK:
1091                 case NODE_IND2_BLOCK:
1092                         err = truncate_nodes(&dn, nofs, offset[1], 2);
1093                         break;
1094
1095                 case NODE_DIND_BLOCK:
1096                         err = truncate_nodes(&dn, nofs, offset[1], 3);
1097                         cont = 0;
1098                         break;
1099
1100                 default:
1101                         BUG();
1102                 }
1103                 if (err < 0 && err != -ENOENT)
1104                         goto fail;
1105                 if (offset[1] == 0 &&
1106                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1107                         lock_page(page);
1108                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
1109                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1110                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1111                         set_page_dirty(page);
1112                         unlock_page(page);
1113                 }
1114                 offset[1] = 0;
1115                 offset[0]++;
1116                 nofs += err;
1117         }
1118 fail:
1119         f2fs_put_page(page, 0);
1120         trace_f2fs_truncate_inode_blocks_exit(inode, err);
1121         return err > 0 ? 0 : err;
1122 }
1123
1124 /* caller must lock inode page */
1125 int f2fs_truncate_xattr_node(struct inode *inode)
1126 {
1127         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1128         nid_t nid = F2FS_I(inode)->i_xattr_nid;
1129         struct dnode_of_data dn;
1130         struct page *npage;
1131         int err;
1132
1133         if (!nid)
1134                 return 0;
1135
1136         npage = f2fs_get_node_page(sbi, nid);
1137         if (IS_ERR(npage))
1138                 return PTR_ERR(npage);
1139
1140         set_new_dnode(&dn, inode, NULL, npage, nid);
1141         err = truncate_node(&dn);
1142         if (err) {
1143                 f2fs_put_page(npage, 1);
1144                 return err;
1145         }
1146
1147         f2fs_i_xnid_write(inode, 0);
1148
1149         return 0;
1150 }
1151
1152 /*
1153  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1154  * f2fs_unlock_op().
1155  */
1156 int f2fs_remove_inode_page(struct inode *inode)
1157 {
1158         struct dnode_of_data dn;
1159         int err;
1160
1161         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1162         err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1163         if (err)
1164                 return err;
1165
1166         err = f2fs_truncate_xattr_node(inode);
1167         if (err) {
1168                 f2fs_put_dnode(&dn);
1169                 return err;
1170         }
1171
1172         /* remove potential inline_data blocks */
1173         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1174                                 S_ISLNK(inode->i_mode))
1175                 f2fs_truncate_data_blocks_range(&dn, 1);
1176
1177         /* 0 is possible, after f2fs_new_inode() has failed */
1178         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1179                 f2fs_put_dnode(&dn);
1180                 return -EIO;
1181         }
1182         f2fs_bug_on(F2FS_I_SB(inode),
1183                         inode->i_blocks != 0 && inode->i_blocks != 8);
1184
1185         /* will put inode & node pages */
1186         err = truncate_node(&dn);
1187         if (err) {
1188                 f2fs_put_dnode(&dn);
1189                 return err;
1190         }
1191         return 0;
1192 }
1193
1194 struct page *f2fs_new_inode_page(struct inode *inode)
1195 {
1196         struct dnode_of_data dn;
1197
1198         /* allocate inode page for new inode */
1199         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1200
1201         /* caller should f2fs_put_page(page, 1); */
1202         return f2fs_new_node_page(&dn, 0);
1203 }
1204
1205 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1206 {
1207         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1208         struct node_info new_ni;
1209         struct page *page;
1210         int err;
1211
1212         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1213                 return ERR_PTR(-EPERM);
1214
1215         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1216         if (!page)
1217                 return ERR_PTR(-ENOMEM);
1218
1219         if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1220                 goto fail;
1221
1222 #ifdef CONFIG_F2FS_CHECK_FS
1223         err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1224         if (err) {
1225                 dec_valid_node_count(sbi, dn->inode, !ofs);
1226                 goto fail;
1227         }
1228         f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1229 #endif
1230         new_ni.nid = dn->nid;
1231         new_ni.ino = dn->inode->i_ino;
1232         new_ni.blk_addr = NULL_ADDR;
1233         new_ni.flag = 0;
1234         new_ni.version = 0;
1235         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1236
1237         f2fs_wait_on_page_writeback(page, NODE, true, true);
1238         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1239         set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1240         if (!PageUptodate(page))
1241                 SetPageUptodate(page);
1242         if (set_page_dirty(page))
1243                 dn->node_changed = true;
1244
1245         if (f2fs_has_xattr_block(ofs))
1246                 f2fs_i_xnid_write(dn->inode, dn->nid);
1247
1248         if (ofs == 0)
1249                 inc_valid_inode_count(sbi);
1250         return page;
1251
1252 fail:
1253         clear_node_page_dirty(page);
1254         f2fs_put_page(page, 1);
1255         return ERR_PTR(err);
1256 }
1257
1258 /*
1259  * Caller should do after getting the following values.
1260  * 0: f2fs_put_page(page, 0)
1261  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1262  */
1263 static int read_node_page(struct page *page, int op_flags)
1264 {
1265         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1266         struct node_info ni;
1267         struct f2fs_io_info fio = {
1268                 .sbi = sbi,
1269                 .type = NODE,
1270                 .op = REQ_OP_READ,
1271                 .op_flags = op_flags,
1272                 .page = page,
1273                 .encrypted_page = NULL,
1274         };
1275         int err;
1276
1277         if (PageUptodate(page)) {
1278 #ifdef CONFIG_F2FS_CHECK_FS
1279                 f2fs_bug_on(sbi, !f2fs_inode_chksum_verify(sbi, page));
1280 #endif
1281                 return LOCKED_PAGE;
1282         }
1283
1284         err = f2fs_get_node_info(sbi, page->index, &ni);
1285         if (err)
1286                 return err;
1287
1288         if (unlikely(ni.blk_addr == NULL_ADDR) ||
1289                         is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1290                 ClearPageUptodate(page);
1291                 return -ENOENT;
1292         }
1293
1294         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1295         return f2fs_submit_page_bio(&fio);
1296 }
1297
1298 /*
1299  * Readahead a node page
1300  */
1301 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1302 {
1303         struct page *apage;
1304         int err;
1305
1306         if (!nid)
1307                 return;
1308         if (f2fs_check_nid_range(sbi, nid))
1309                 return;
1310
1311         apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1312         if (apage)
1313                 return;
1314
1315         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1316         if (!apage)
1317                 return;
1318
1319         err = read_node_page(apage, REQ_RAHEAD);
1320         f2fs_put_page(apage, err ? 1 : 0);
1321 }
1322
1323 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1324                                         struct page *parent, int start)
1325 {
1326         struct page *page;
1327         int err;
1328
1329         if (!nid)
1330                 return ERR_PTR(-ENOENT);
1331         if (f2fs_check_nid_range(sbi, nid))
1332                 return ERR_PTR(-EINVAL);
1333 repeat:
1334         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1335         if (!page)
1336                 return ERR_PTR(-ENOMEM);
1337
1338         err = read_node_page(page, 0);
1339         if (err < 0) {
1340                 f2fs_put_page(page, 1);
1341                 return ERR_PTR(err);
1342         } else if (err == LOCKED_PAGE) {
1343                 err = 0;
1344                 goto page_hit;
1345         }
1346
1347         if (parent)
1348                 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1349
1350         lock_page(page);
1351
1352         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1353                 f2fs_put_page(page, 1);
1354                 goto repeat;
1355         }
1356
1357         if (unlikely(!PageUptodate(page))) {
1358                 err = -EIO;
1359                 goto out_err;
1360         }
1361
1362         if (!f2fs_inode_chksum_verify(sbi, page)) {
1363                 err = -EBADMSG;
1364                 goto out_err;
1365         }
1366 page_hit:
1367         if(unlikely(nid != nid_of_node(page))) {
1368                 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1369                         "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1370                         nid, nid_of_node(page), ino_of_node(page),
1371                         ofs_of_node(page), cpver_of_node(page),
1372                         next_blkaddr_of_node(page));
1373                 err = -EINVAL;
1374 out_err:
1375                 ClearPageUptodate(page);
1376                 f2fs_put_page(page, 1);
1377                 return ERR_PTR(err);
1378         }
1379         return page;
1380 }
1381
1382 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1383 {
1384         return __get_node_page(sbi, nid, NULL, 0);
1385 }
1386
1387 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1388 {
1389         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1390         nid_t nid = get_nid(parent, start, false);
1391
1392         return __get_node_page(sbi, nid, parent, start);
1393 }
1394
1395 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1396 {
1397         struct inode *inode;
1398         struct page *page;
1399         int ret;
1400
1401         /* should flush inline_data before evict_inode */
1402         inode = ilookup(sbi->sb, ino);
1403         if (!inode)
1404                 return;
1405
1406         page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1407                                         FGP_LOCK|FGP_NOWAIT, 0);
1408         if (!page)
1409                 goto iput_out;
1410
1411         if (!PageUptodate(page))
1412                 goto page_out;
1413
1414         if (!PageDirty(page))
1415                 goto page_out;
1416
1417         if (!clear_page_dirty_for_io(page))
1418                 goto page_out;
1419
1420         ret = f2fs_write_inline_data(inode, page);
1421         inode_dec_dirty_pages(inode);
1422         f2fs_remove_dirty_inode(inode);
1423         if (ret)
1424                 set_page_dirty(page);
1425 page_out:
1426         f2fs_put_page(page, 1);
1427 iput_out:
1428         iput(inode);
1429 }
1430
1431 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1432 {
1433         pgoff_t index;
1434         struct pagevec pvec;
1435         struct page *last_page = NULL;
1436         int nr_pages;
1437
1438         pagevec_init(&pvec);
1439         index = 0;
1440
1441         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1442                                 PAGECACHE_TAG_DIRTY))) {
1443                 int i;
1444
1445                 for (i = 0; i < nr_pages; i++) {
1446                         struct page *page = pvec.pages[i];
1447
1448                         if (unlikely(f2fs_cp_error(sbi))) {
1449                                 f2fs_put_page(last_page, 0);
1450                                 pagevec_release(&pvec);
1451                                 return ERR_PTR(-EIO);
1452                         }
1453
1454                         if (!IS_DNODE(page) || !is_cold_node(page))
1455                                 continue;
1456                         if (ino_of_node(page) != ino)
1457                                 continue;
1458
1459                         lock_page(page);
1460
1461                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1462 continue_unlock:
1463                                 unlock_page(page);
1464                                 continue;
1465                         }
1466                         if (ino_of_node(page) != ino)
1467                                 goto continue_unlock;
1468
1469                         if (!PageDirty(page)) {
1470                                 /* someone wrote it for us */
1471                                 goto continue_unlock;
1472                         }
1473
1474                         if (last_page)
1475                                 f2fs_put_page(last_page, 0);
1476
1477                         get_page(page);
1478                         last_page = page;
1479                         unlock_page(page);
1480                 }
1481                 pagevec_release(&pvec);
1482                 cond_resched();
1483         }
1484         return last_page;
1485 }
1486
1487 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1488                                 struct writeback_control *wbc, bool do_balance,
1489                                 enum iostat_type io_type, unsigned int *seq_id)
1490 {
1491         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1492         nid_t nid;
1493         struct node_info ni;
1494         struct f2fs_io_info fio = {
1495                 .sbi = sbi,
1496                 .ino = ino_of_node(page),
1497                 .type = NODE,
1498                 .op = REQ_OP_WRITE,
1499                 .op_flags = wbc_to_write_flags(wbc),
1500                 .page = page,
1501                 .encrypted_page = NULL,
1502                 .submitted = false,
1503                 .io_type = io_type,
1504                 .io_wbc = wbc,
1505         };
1506         unsigned int seq;
1507
1508         trace_f2fs_writepage(page, NODE);
1509
1510         if (unlikely(f2fs_cp_error(sbi)))
1511                 goto redirty_out;
1512
1513         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1514                 goto redirty_out;
1515
1516         if (wbc->sync_mode == WB_SYNC_NONE &&
1517                         IS_DNODE(page) && is_cold_node(page))
1518                 goto redirty_out;
1519
1520         /* get old block addr of this node page */
1521         nid = nid_of_node(page);
1522         f2fs_bug_on(sbi, page->index != nid);
1523
1524         if (f2fs_get_node_info(sbi, nid, &ni))
1525                 goto redirty_out;
1526
1527         if (wbc->for_reclaim) {
1528                 if (!down_read_trylock(&sbi->node_write))
1529                         goto redirty_out;
1530         } else {
1531                 down_read(&sbi->node_write);
1532         }
1533
1534         /* This page is already truncated */
1535         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1536                 ClearPageUptodate(page);
1537                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1538                 up_read(&sbi->node_write);
1539                 unlock_page(page);
1540                 return 0;
1541         }
1542
1543         if (__is_valid_data_blkaddr(ni.blk_addr) &&
1544                 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC)) {
1545                 up_read(&sbi->node_write);
1546                 goto redirty_out;
1547         }
1548
1549         if (atomic && !test_opt(sbi, NOBARRIER))
1550                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1551
1552         set_page_writeback(page);
1553         ClearPageError(page);
1554
1555         if (f2fs_in_warm_node_list(sbi, page)) {
1556                 seq = f2fs_add_fsync_node_entry(sbi, page);
1557                 if (seq_id)
1558                         *seq_id = seq;
1559         }
1560
1561         fio.old_blkaddr = ni.blk_addr;
1562         f2fs_do_write_node_page(nid, &fio);
1563         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1564         dec_page_count(sbi, F2FS_DIRTY_NODES);
1565         up_read(&sbi->node_write);
1566
1567         if (wbc->for_reclaim) {
1568                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1569                 submitted = NULL;
1570         }
1571
1572         unlock_page(page);
1573
1574         if (unlikely(f2fs_cp_error(sbi))) {
1575                 f2fs_submit_merged_write(sbi, NODE);
1576                 submitted = NULL;
1577         }
1578         if (submitted)
1579                 *submitted = fio.submitted;
1580
1581         if (do_balance)
1582                 f2fs_balance_fs(sbi, false);
1583         return 0;
1584
1585 redirty_out:
1586         redirty_page_for_writepage(wbc, page);
1587         return AOP_WRITEPAGE_ACTIVATE;
1588 }
1589
1590 int f2fs_move_node_page(struct page *node_page, int gc_type)
1591 {
1592         int err = 0;
1593
1594         if (gc_type == FG_GC) {
1595                 struct writeback_control wbc = {
1596                         .sync_mode = WB_SYNC_ALL,
1597                         .nr_to_write = 1,
1598                         .for_reclaim = 0,
1599                 };
1600
1601                 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1602
1603                 set_page_dirty(node_page);
1604
1605                 if (!clear_page_dirty_for_io(node_page)) {
1606                         err = -EAGAIN;
1607                         goto out_page;
1608                 }
1609
1610                 if (__write_node_page(node_page, false, NULL,
1611                                         &wbc, false, FS_GC_NODE_IO, NULL)) {
1612                         err = -EAGAIN;
1613                         unlock_page(node_page);
1614                 }
1615                 goto release_page;
1616         } else {
1617                 /* set page dirty and write it */
1618                 if (!PageWriteback(node_page))
1619                         set_page_dirty(node_page);
1620         }
1621 out_page:
1622         unlock_page(node_page);
1623 release_page:
1624         f2fs_put_page(node_page, 0);
1625         return err;
1626 }
1627
1628 static int f2fs_write_node_page(struct page *page,
1629                                 struct writeback_control *wbc)
1630 {
1631         return __write_node_page(page, false, NULL, wbc, false,
1632                                                 FS_NODE_IO, NULL);
1633 }
1634
1635 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1636                         struct writeback_control *wbc, bool atomic,
1637                         unsigned int *seq_id)
1638 {
1639         pgoff_t index;
1640         struct pagevec pvec;
1641         int ret = 0;
1642         struct page *last_page = NULL;
1643         bool marked = false;
1644         nid_t ino = inode->i_ino;
1645         int nr_pages;
1646         int nwritten = 0;
1647
1648         if (atomic) {
1649                 last_page = last_fsync_dnode(sbi, ino);
1650                 if (IS_ERR_OR_NULL(last_page))
1651                         return PTR_ERR_OR_ZERO(last_page);
1652         }
1653 retry:
1654         pagevec_init(&pvec);
1655         index = 0;
1656
1657         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1658                                 PAGECACHE_TAG_DIRTY))) {
1659                 int i;
1660
1661                 for (i = 0; i < nr_pages; i++) {
1662                         struct page *page = pvec.pages[i];
1663                         bool submitted = false;
1664
1665                         if (unlikely(f2fs_cp_error(sbi))) {
1666                                 f2fs_put_page(last_page, 0);
1667                                 pagevec_release(&pvec);
1668                                 ret = -EIO;
1669                                 goto out;
1670                         }
1671
1672                         if (!IS_DNODE(page) || !is_cold_node(page))
1673                                 continue;
1674                         if (ino_of_node(page) != ino)
1675                                 continue;
1676
1677                         lock_page(page);
1678
1679                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1680 continue_unlock:
1681                                 unlock_page(page);
1682                                 continue;
1683                         }
1684                         if (ino_of_node(page) != ino)
1685                                 goto continue_unlock;
1686
1687                         if (!PageDirty(page) && page != last_page) {
1688                                 /* someone wrote it for us */
1689                                 goto continue_unlock;
1690                         }
1691
1692                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1693
1694                         set_fsync_mark(page, 0);
1695                         set_dentry_mark(page, 0);
1696
1697                         if (!atomic || page == last_page) {
1698                                 set_fsync_mark(page, 1);
1699                                 if (IS_INODE(page)) {
1700                                         if (is_inode_flag_set(inode,
1701                                                                 FI_DIRTY_INODE))
1702                                                 f2fs_update_inode(inode, page);
1703                                         set_dentry_mark(page,
1704                                                 f2fs_need_dentry_mark(sbi, ino));
1705                                 }
1706                                 /*  may be written by other thread */
1707                                 if (!PageDirty(page))
1708                                         set_page_dirty(page);
1709                         }
1710
1711                         if (!clear_page_dirty_for_io(page))
1712                                 goto continue_unlock;
1713
1714                         ret = __write_node_page(page, atomic &&
1715                                                 page == last_page,
1716                                                 &submitted, wbc, true,
1717                                                 FS_NODE_IO, seq_id);
1718                         if (ret) {
1719                                 unlock_page(page);
1720                                 f2fs_put_page(last_page, 0);
1721                                 break;
1722                         } else if (submitted) {
1723                                 nwritten++;
1724                         }
1725
1726                         if (page == last_page) {
1727                                 f2fs_put_page(page, 0);
1728                                 marked = true;
1729                                 break;
1730                         }
1731                 }
1732                 pagevec_release(&pvec);
1733                 cond_resched();
1734
1735                 if (ret || marked)
1736                         break;
1737         }
1738         if (!ret && atomic && !marked) {
1739                 f2fs_msg(sbi->sb, KERN_DEBUG,
1740                         "Retry to write fsync mark: ino=%u, idx=%lx",
1741                                         ino, last_page->index);
1742                 lock_page(last_page);
1743                 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1744                 set_page_dirty(last_page);
1745                 unlock_page(last_page);
1746                 goto retry;
1747         }
1748 out:
1749         if (nwritten)
1750                 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1751         return ret ? -EIO: 0;
1752 }
1753
1754 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1755                                 struct writeback_control *wbc,
1756                                 bool do_balance, enum iostat_type io_type)
1757 {
1758         pgoff_t index;
1759         struct pagevec pvec;
1760         int step = 0;
1761         int nwritten = 0;
1762         int ret = 0;
1763         int nr_pages, done = 0;
1764
1765         pagevec_init(&pvec);
1766
1767 next_step:
1768         index = 0;
1769
1770         while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1771                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1772                 int i;
1773
1774                 for (i = 0; i < nr_pages; i++) {
1775                         struct page *page = pvec.pages[i];
1776                         bool submitted = false;
1777
1778                         /* give a priority to WB_SYNC threads */
1779                         if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1780                                         wbc->sync_mode == WB_SYNC_NONE) {
1781                                 done = 1;
1782                                 break;
1783                         }
1784
1785                         /*
1786                          * flushing sequence with step:
1787                          * 0. indirect nodes
1788                          * 1. dentry dnodes
1789                          * 2. file dnodes
1790                          */
1791                         if (step == 0 && IS_DNODE(page))
1792                                 continue;
1793                         if (step == 1 && (!IS_DNODE(page) ||
1794                                                 is_cold_node(page)))
1795                                 continue;
1796                         if (step == 2 && (!IS_DNODE(page) ||
1797                                                 !is_cold_node(page)))
1798                                 continue;
1799 lock_node:
1800                         if (wbc->sync_mode == WB_SYNC_ALL)
1801                                 lock_page(page);
1802                         else if (!trylock_page(page))
1803                                 continue;
1804
1805                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1806 continue_unlock:
1807                                 unlock_page(page);
1808                                 continue;
1809                         }
1810
1811                         if (!PageDirty(page)) {
1812                                 /* someone wrote it for us */
1813                                 goto continue_unlock;
1814                         }
1815
1816                         /* flush inline_data */
1817                         if (is_inline_node(page)) {
1818                                 clear_inline_node(page);
1819                                 unlock_page(page);
1820                                 flush_inline_data(sbi, ino_of_node(page));
1821                                 goto lock_node;
1822                         }
1823
1824                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1825
1826                         if (!clear_page_dirty_for_io(page))
1827                                 goto continue_unlock;
1828
1829                         set_fsync_mark(page, 0);
1830                         set_dentry_mark(page, 0);
1831
1832                         ret = __write_node_page(page, false, &submitted,
1833                                                 wbc, do_balance, io_type, NULL);
1834                         if (ret)
1835                                 unlock_page(page);
1836                         else if (submitted)
1837                                 nwritten++;
1838
1839                         if (--wbc->nr_to_write == 0)
1840                                 break;
1841                 }
1842                 pagevec_release(&pvec);
1843                 cond_resched();
1844
1845                 if (wbc->nr_to_write == 0) {
1846                         step = 2;
1847                         break;
1848                 }
1849         }
1850
1851         if (step < 2) {
1852                 if (wbc->sync_mode == WB_SYNC_NONE && step == 1)
1853                         goto out;
1854                 step++;
1855                 goto next_step;
1856         }
1857 out:
1858         if (nwritten)
1859                 f2fs_submit_merged_write(sbi, NODE);
1860
1861         if (unlikely(f2fs_cp_error(sbi)))
1862                 return -EIO;
1863         return ret;
1864 }
1865
1866 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1867                                                 unsigned int seq_id)
1868 {
1869         struct fsync_node_entry *fn;
1870         struct page *page;
1871         struct list_head *head = &sbi->fsync_node_list;
1872         unsigned long flags;
1873         unsigned int cur_seq_id = 0;
1874         int ret2, ret = 0;
1875
1876         while (seq_id && cur_seq_id < seq_id) {
1877                 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
1878                 if (list_empty(head)) {
1879                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1880                         break;
1881                 }
1882                 fn = list_first_entry(head, struct fsync_node_entry, list);
1883                 if (fn->seq_id > seq_id) {
1884                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1885                         break;
1886                 }
1887                 cur_seq_id = fn->seq_id;
1888                 page = fn->page;
1889                 get_page(page);
1890                 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1891
1892                 f2fs_wait_on_page_writeback(page, NODE, true, false);
1893                 if (TestClearPageError(page))
1894                         ret = -EIO;
1895
1896                 put_page(page);
1897
1898                 if (ret)
1899                         break;
1900         }
1901
1902         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1903         if (!ret)
1904                 ret = ret2;
1905
1906         return ret;
1907 }
1908
1909 static int f2fs_write_node_pages(struct address_space *mapping,
1910                             struct writeback_control *wbc)
1911 {
1912         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1913         struct blk_plug plug;
1914         long diff;
1915
1916         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1917                 goto skip_write;
1918
1919         /* balancing f2fs's metadata in background */
1920         f2fs_balance_fs_bg(sbi);
1921
1922         /* collect a number of dirty node pages and write together */
1923         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1924                 goto skip_write;
1925
1926         if (wbc->sync_mode == WB_SYNC_ALL)
1927                 atomic_inc(&sbi->wb_sync_req[NODE]);
1928         else if (atomic_read(&sbi->wb_sync_req[NODE]))
1929                 goto skip_write;
1930
1931         trace_f2fs_writepages(mapping->host, wbc, NODE);
1932
1933         diff = nr_pages_to_write(sbi, NODE, wbc);
1934         blk_start_plug(&plug);
1935         f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1936         blk_finish_plug(&plug);
1937         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1938
1939         if (wbc->sync_mode == WB_SYNC_ALL)
1940                 atomic_dec(&sbi->wb_sync_req[NODE]);
1941         return 0;
1942
1943 skip_write:
1944         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1945         trace_f2fs_writepages(mapping->host, wbc, NODE);
1946         return 0;
1947 }
1948
1949 static int f2fs_set_node_page_dirty(struct page *page)
1950 {
1951         trace_f2fs_set_page_dirty(page, NODE);
1952
1953         if (!PageUptodate(page))
1954                 SetPageUptodate(page);
1955 #ifdef CONFIG_F2FS_CHECK_FS
1956         if (IS_INODE(page))
1957                 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
1958 #endif
1959         if (!PageDirty(page)) {
1960                 __set_page_dirty_nobuffers(page);
1961                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1962                 SetPagePrivate(page);
1963                 f2fs_trace_pid(page);
1964                 return 1;
1965         }
1966         return 0;
1967 }
1968
1969 /*
1970  * Structure of the f2fs node operations
1971  */
1972 const struct address_space_operations f2fs_node_aops = {
1973         .writepage      = f2fs_write_node_page,
1974         .writepages     = f2fs_write_node_pages,
1975         .set_page_dirty = f2fs_set_node_page_dirty,
1976         .invalidatepage = f2fs_invalidate_page,
1977         .releasepage    = f2fs_release_page,
1978 #ifdef CONFIG_MIGRATION
1979         .migratepage    = f2fs_migrate_page,
1980 #endif
1981 };
1982
1983 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1984                                                 nid_t n)
1985 {
1986         return radix_tree_lookup(&nm_i->free_nid_root, n);
1987 }
1988
1989 static int __insert_free_nid(struct f2fs_sb_info *sbi,
1990                         struct free_nid *i, enum nid_state state)
1991 {
1992         struct f2fs_nm_info *nm_i = NM_I(sbi);
1993
1994         int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1995         if (err)
1996                 return err;
1997
1998         f2fs_bug_on(sbi, state != i->state);
1999         nm_i->nid_cnt[state]++;
2000         if (state == FREE_NID)
2001                 list_add_tail(&i->list, &nm_i->free_nid_list);
2002         return 0;
2003 }
2004
2005 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2006                         struct free_nid *i, enum nid_state state)
2007 {
2008         struct f2fs_nm_info *nm_i = NM_I(sbi);
2009
2010         f2fs_bug_on(sbi, state != i->state);
2011         nm_i->nid_cnt[state]--;
2012         if (state == FREE_NID)
2013                 list_del(&i->list);
2014         radix_tree_delete(&nm_i->free_nid_root, i->nid);
2015 }
2016
2017 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2018                         enum nid_state org_state, enum nid_state dst_state)
2019 {
2020         struct f2fs_nm_info *nm_i = NM_I(sbi);
2021
2022         f2fs_bug_on(sbi, org_state != i->state);
2023         i->state = dst_state;
2024         nm_i->nid_cnt[org_state]--;
2025         nm_i->nid_cnt[dst_state]++;
2026
2027         switch (dst_state) {
2028         case PREALLOC_NID:
2029                 list_del(&i->list);
2030                 break;
2031         case FREE_NID:
2032                 list_add_tail(&i->list, &nm_i->free_nid_list);
2033                 break;
2034         default:
2035                 BUG_ON(1);
2036         }
2037 }
2038
2039 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2040                                                         bool set, bool build)
2041 {
2042         struct f2fs_nm_info *nm_i = NM_I(sbi);
2043         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2044         unsigned int nid_ofs = nid - START_NID(nid);
2045
2046         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2047                 return;
2048
2049         if (set) {
2050                 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2051                         return;
2052                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2053                 nm_i->free_nid_count[nat_ofs]++;
2054         } else {
2055                 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2056                         return;
2057                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2058                 if (!build)
2059                         nm_i->free_nid_count[nat_ofs]--;
2060         }
2061 }
2062
2063 /* return if the nid is recognized as free */
2064 static bool add_free_nid(struct f2fs_sb_info *sbi,
2065                                 nid_t nid, bool build, bool update)
2066 {
2067         struct f2fs_nm_info *nm_i = NM_I(sbi);
2068         struct free_nid *i, *e;
2069         struct nat_entry *ne;
2070         int err = -EINVAL;
2071         bool ret = false;
2072
2073         /* 0 nid should not be used */
2074         if (unlikely(nid == 0))
2075                 return false;
2076
2077         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2078         i->nid = nid;
2079         i->state = FREE_NID;
2080
2081         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2082
2083         spin_lock(&nm_i->nid_list_lock);
2084
2085         if (build) {
2086                 /*
2087                  *   Thread A             Thread B
2088                  *  - f2fs_create
2089                  *   - f2fs_new_inode
2090                  *    - f2fs_alloc_nid
2091                  *     - __insert_nid_to_list(PREALLOC_NID)
2092                  *                     - f2fs_balance_fs_bg
2093                  *                      - f2fs_build_free_nids
2094                  *                       - __f2fs_build_free_nids
2095                  *                        - scan_nat_page
2096                  *                         - add_free_nid
2097                  *                          - __lookup_nat_cache
2098                  *  - f2fs_add_link
2099                  *   - f2fs_init_inode_metadata
2100                  *    - f2fs_new_inode_page
2101                  *     - f2fs_new_node_page
2102                  *      - set_node_addr
2103                  *  - f2fs_alloc_nid_done
2104                  *   - __remove_nid_from_list(PREALLOC_NID)
2105                  *                         - __insert_nid_to_list(FREE_NID)
2106                  */
2107                 ne = __lookup_nat_cache(nm_i, nid);
2108                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2109                                 nat_get_blkaddr(ne) != NULL_ADDR))
2110                         goto err_out;
2111
2112                 e = __lookup_free_nid_list(nm_i, nid);
2113                 if (e) {
2114                         if (e->state == FREE_NID)
2115                                 ret = true;
2116                         goto err_out;
2117                 }
2118         }
2119         ret = true;
2120         err = __insert_free_nid(sbi, i, FREE_NID);
2121 err_out:
2122         if (update) {
2123                 update_free_nid_bitmap(sbi, nid, ret, build);
2124                 if (!build)
2125                         nm_i->available_nids++;
2126         }
2127         spin_unlock(&nm_i->nid_list_lock);
2128         radix_tree_preload_end();
2129
2130         if (err)
2131                 kmem_cache_free(free_nid_slab, i);
2132         return ret;
2133 }
2134
2135 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2136 {
2137         struct f2fs_nm_info *nm_i = NM_I(sbi);
2138         struct free_nid *i;
2139         bool need_free = false;
2140
2141         spin_lock(&nm_i->nid_list_lock);
2142         i = __lookup_free_nid_list(nm_i, nid);
2143         if (i && i->state == FREE_NID) {
2144                 __remove_free_nid(sbi, i, FREE_NID);
2145                 need_free = true;
2146         }
2147         spin_unlock(&nm_i->nid_list_lock);
2148
2149         if (need_free)
2150                 kmem_cache_free(free_nid_slab, i);
2151 }
2152
2153 static int scan_nat_page(struct f2fs_sb_info *sbi,
2154                         struct page *nat_page, nid_t start_nid)
2155 {
2156         struct f2fs_nm_info *nm_i = NM_I(sbi);
2157         struct f2fs_nat_block *nat_blk = page_address(nat_page);
2158         block_t blk_addr;
2159         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2160         int i;
2161
2162         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2163
2164         i = start_nid % NAT_ENTRY_PER_BLOCK;
2165
2166         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2167                 if (unlikely(start_nid >= nm_i->max_nid))
2168                         break;
2169
2170                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2171
2172                 if (blk_addr == NEW_ADDR)
2173                         return -EINVAL;
2174
2175                 if (blk_addr == NULL_ADDR) {
2176                         add_free_nid(sbi, start_nid, true, true);
2177                 } else {
2178                         spin_lock(&NM_I(sbi)->nid_list_lock);
2179                         update_free_nid_bitmap(sbi, start_nid, false, true);
2180                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2181                 }
2182         }
2183
2184         return 0;
2185 }
2186
2187 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2188 {
2189         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2190         struct f2fs_journal *journal = curseg->journal;
2191         int i;
2192
2193         down_read(&curseg->journal_rwsem);
2194         for (i = 0; i < nats_in_cursum(journal); i++) {
2195                 block_t addr;
2196                 nid_t nid;
2197
2198                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2199                 nid = le32_to_cpu(nid_in_journal(journal, i));
2200                 if (addr == NULL_ADDR)
2201                         add_free_nid(sbi, nid, true, false);
2202                 else
2203                         remove_free_nid(sbi, nid);
2204         }
2205         up_read(&curseg->journal_rwsem);
2206 }
2207
2208 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2209 {
2210         struct f2fs_nm_info *nm_i = NM_I(sbi);
2211         unsigned int i, idx;
2212         nid_t nid;
2213
2214         down_read(&nm_i->nat_tree_lock);
2215
2216         for (i = 0; i < nm_i->nat_blocks; i++) {
2217                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2218                         continue;
2219                 if (!nm_i->free_nid_count[i])
2220                         continue;
2221                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2222                         idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2223                                                 NAT_ENTRY_PER_BLOCK, idx);
2224                         if (idx >= NAT_ENTRY_PER_BLOCK)
2225                                 break;
2226
2227                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
2228                         add_free_nid(sbi, nid, true, false);
2229
2230                         if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2231                                 goto out;
2232                 }
2233         }
2234 out:
2235         scan_curseg_cache(sbi);
2236
2237         up_read(&nm_i->nat_tree_lock);
2238 }
2239
2240 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2241                                                 bool sync, bool mount)
2242 {
2243         struct f2fs_nm_info *nm_i = NM_I(sbi);
2244         int i = 0, ret;
2245         nid_t nid = nm_i->next_scan_nid;
2246
2247         if (unlikely(nid >= nm_i->max_nid))
2248                 nid = 0;
2249
2250         /* Enough entries */
2251         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2252                 return 0;
2253
2254         if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2255                 return 0;
2256
2257         if (!mount) {
2258                 /* try to find free nids in free_nid_bitmap */
2259                 scan_free_nid_bits(sbi);
2260
2261                 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2262                         return 0;
2263         }
2264
2265         /* readahead nat pages to be scanned */
2266         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2267                                                         META_NAT, true);
2268
2269         down_read(&nm_i->nat_tree_lock);
2270
2271         while (1) {
2272                 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2273                                                 nm_i->nat_block_bitmap)) {
2274                         struct page *page = get_current_nat_page(sbi, nid);
2275
2276                         if (IS_ERR(page)) {
2277                                 ret = PTR_ERR(page);
2278                         } else {
2279                                 ret = scan_nat_page(sbi, page, nid);
2280                                 f2fs_put_page(page, 1);
2281                         }
2282
2283                         if (ret) {
2284                                 up_read(&nm_i->nat_tree_lock);
2285                                 f2fs_bug_on(sbi, !mount);
2286                                 f2fs_msg(sbi->sb, KERN_ERR,
2287                                         "NAT is corrupt, run fsck to fix it");
2288                                 return ret;
2289                         }
2290                 }
2291
2292                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2293                 if (unlikely(nid >= nm_i->max_nid))
2294                         nid = 0;
2295
2296                 if (++i >= FREE_NID_PAGES)
2297                         break;
2298         }
2299
2300         /* go to the next free nat pages to find free nids abundantly */
2301         nm_i->next_scan_nid = nid;
2302
2303         /* find free nids from current sum_pages */
2304         scan_curseg_cache(sbi);
2305
2306         up_read(&nm_i->nat_tree_lock);
2307
2308         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2309                                         nm_i->ra_nid_pages, META_NAT, false);
2310
2311         return 0;
2312 }
2313
2314 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2315 {
2316         int ret;
2317
2318         mutex_lock(&NM_I(sbi)->build_lock);
2319         ret = __f2fs_build_free_nids(sbi, sync, mount);
2320         mutex_unlock(&NM_I(sbi)->build_lock);
2321
2322         return ret;
2323 }
2324
2325 /*
2326  * If this function returns success, caller can obtain a new nid
2327  * from second parameter of this function.
2328  * The returned nid could be used ino as well as nid when inode is created.
2329  */
2330 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2331 {
2332         struct f2fs_nm_info *nm_i = NM_I(sbi);
2333         struct free_nid *i = NULL;
2334 retry:
2335         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2336                 f2fs_show_injection_info(FAULT_ALLOC_NID);
2337                 return false;
2338         }
2339
2340         spin_lock(&nm_i->nid_list_lock);
2341
2342         if (unlikely(nm_i->available_nids == 0)) {
2343                 spin_unlock(&nm_i->nid_list_lock);
2344                 return false;
2345         }
2346
2347         /* We should not use stale free nids created by f2fs_build_free_nids */
2348         if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2349                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2350                 i = list_first_entry(&nm_i->free_nid_list,
2351                                         struct free_nid, list);
2352                 *nid = i->nid;
2353
2354                 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2355                 nm_i->available_nids--;
2356
2357                 update_free_nid_bitmap(sbi, *nid, false, false);
2358
2359                 spin_unlock(&nm_i->nid_list_lock);
2360                 return true;
2361         }
2362         spin_unlock(&nm_i->nid_list_lock);
2363
2364         /* Let's scan nat pages and its caches to get free nids */
2365         if (!f2fs_build_free_nids(sbi, true, false))
2366                 goto retry;
2367         return false;
2368 }
2369
2370 /*
2371  * f2fs_alloc_nid() should be called prior to this function.
2372  */
2373 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2374 {
2375         struct f2fs_nm_info *nm_i = NM_I(sbi);
2376         struct free_nid *i;
2377
2378         spin_lock(&nm_i->nid_list_lock);
2379         i = __lookup_free_nid_list(nm_i, nid);
2380         f2fs_bug_on(sbi, !i);
2381         __remove_free_nid(sbi, i, PREALLOC_NID);
2382         spin_unlock(&nm_i->nid_list_lock);
2383
2384         kmem_cache_free(free_nid_slab, i);
2385 }
2386
2387 /*
2388  * f2fs_alloc_nid() should be called prior to this function.
2389  */
2390 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2391 {
2392         struct f2fs_nm_info *nm_i = NM_I(sbi);
2393         struct free_nid *i;
2394         bool need_free = false;
2395
2396         if (!nid)
2397                 return;
2398
2399         spin_lock(&nm_i->nid_list_lock);
2400         i = __lookup_free_nid_list(nm_i, nid);
2401         f2fs_bug_on(sbi, !i);
2402
2403         if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2404                 __remove_free_nid(sbi, i, PREALLOC_NID);
2405                 need_free = true;
2406         } else {
2407                 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2408         }
2409
2410         nm_i->available_nids++;
2411
2412         update_free_nid_bitmap(sbi, nid, true, false);
2413
2414         spin_unlock(&nm_i->nid_list_lock);
2415
2416         if (need_free)
2417                 kmem_cache_free(free_nid_slab, i);
2418 }
2419
2420 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2421 {
2422         struct f2fs_nm_info *nm_i = NM_I(sbi);
2423         struct free_nid *i, *next;
2424         int nr = nr_shrink;
2425
2426         if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2427                 return 0;
2428
2429         if (!mutex_trylock(&nm_i->build_lock))
2430                 return 0;
2431
2432         spin_lock(&nm_i->nid_list_lock);
2433         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2434                 if (nr_shrink <= 0 ||
2435                                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2436                         break;
2437
2438                 __remove_free_nid(sbi, i, FREE_NID);
2439                 kmem_cache_free(free_nid_slab, i);
2440                 nr_shrink--;
2441         }
2442         spin_unlock(&nm_i->nid_list_lock);
2443         mutex_unlock(&nm_i->build_lock);
2444
2445         return nr - nr_shrink;
2446 }
2447
2448 void f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2449 {
2450         void *src_addr, *dst_addr;
2451         size_t inline_size;
2452         struct page *ipage;
2453         struct f2fs_inode *ri;
2454
2455         ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2456         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2457
2458         ri = F2FS_INODE(page);
2459         if (ri->i_inline & F2FS_INLINE_XATTR) {
2460                 set_inode_flag(inode, FI_INLINE_XATTR);
2461         } else {
2462                 clear_inode_flag(inode, FI_INLINE_XATTR);
2463                 goto update_inode;
2464         }
2465
2466         dst_addr = inline_xattr_addr(inode, ipage);
2467         src_addr = inline_xattr_addr(inode, page);
2468         inline_size = inline_xattr_size(inode);
2469
2470         f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2471         memcpy(dst_addr, src_addr, inline_size);
2472 update_inode:
2473         f2fs_update_inode(inode, ipage);
2474         f2fs_put_page(ipage, 1);
2475 }
2476
2477 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2478 {
2479         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2480         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2481         nid_t new_xnid;
2482         struct dnode_of_data dn;
2483         struct node_info ni;
2484         struct page *xpage;
2485         int err;
2486
2487         if (!prev_xnid)
2488                 goto recover_xnid;
2489
2490         /* 1: invalidate the previous xattr nid */
2491         err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2492         if (err)
2493                 return err;
2494
2495         f2fs_invalidate_blocks(sbi, ni.blk_addr);
2496         dec_valid_node_count(sbi, inode, false);
2497         set_node_addr(sbi, &ni, NULL_ADDR, false);
2498
2499 recover_xnid:
2500         /* 2: update xattr nid in inode */
2501         if (!f2fs_alloc_nid(sbi, &new_xnid))
2502                 return -ENOSPC;
2503
2504         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2505         xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2506         if (IS_ERR(xpage)) {
2507                 f2fs_alloc_nid_failed(sbi, new_xnid);
2508                 return PTR_ERR(xpage);
2509         }
2510
2511         f2fs_alloc_nid_done(sbi, new_xnid);
2512         f2fs_update_inode_page(inode);
2513
2514         /* 3: update and set xattr node page dirty */
2515         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2516
2517         set_page_dirty(xpage);
2518         f2fs_put_page(xpage, 1);
2519
2520         return 0;
2521 }
2522
2523 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2524 {
2525         struct f2fs_inode *src, *dst;
2526         nid_t ino = ino_of_node(page);
2527         struct node_info old_ni, new_ni;
2528         struct page *ipage;
2529         int err;
2530
2531         err = f2fs_get_node_info(sbi, ino, &old_ni);
2532         if (err)
2533                 return err;
2534
2535         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2536                 return -EINVAL;
2537 retry:
2538         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2539         if (!ipage) {
2540                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2541                 goto retry;
2542         }
2543
2544         /* Should not use this inode from free nid list */
2545         remove_free_nid(sbi, ino);
2546
2547         if (!PageUptodate(ipage))
2548                 SetPageUptodate(ipage);
2549         fill_node_footer(ipage, ino, ino, 0, true);
2550         set_cold_node(ipage, false);
2551
2552         src = F2FS_INODE(page);
2553         dst = F2FS_INODE(ipage);
2554
2555         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2556         dst->i_size = 0;
2557         dst->i_blocks = cpu_to_le64(1);
2558         dst->i_links = cpu_to_le32(1);
2559         dst->i_xattr_nid = 0;
2560         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2561         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2562                 dst->i_extra_isize = src->i_extra_isize;
2563
2564                 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2565                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2566                                                         i_inline_xattr_size))
2567                         dst->i_inline_xattr_size = src->i_inline_xattr_size;
2568
2569                 if (f2fs_sb_has_project_quota(sbi) &&
2570                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2571                                                                 i_projid))
2572                         dst->i_projid = src->i_projid;
2573
2574                 if (f2fs_sb_has_inode_crtime(sbi) &&
2575                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2576                                                         i_crtime_nsec)) {
2577                         dst->i_crtime = src->i_crtime;
2578                         dst->i_crtime_nsec = src->i_crtime_nsec;
2579                 }
2580         }
2581
2582         new_ni = old_ni;
2583         new_ni.ino = ino;
2584
2585         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2586                 WARN_ON(1);
2587         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2588         inc_valid_inode_count(sbi);
2589         set_page_dirty(ipage);
2590         f2fs_put_page(ipage, 1);
2591         return 0;
2592 }
2593
2594 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2595                         unsigned int segno, struct f2fs_summary_block *sum)
2596 {
2597         struct f2fs_node *rn;
2598         struct f2fs_summary *sum_entry;
2599         block_t addr;
2600         int i, idx, last_offset, nrpages;
2601
2602         /* scan the node segment */
2603         last_offset = sbi->blocks_per_seg;
2604         addr = START_BLOCK(sbi, segno);
2605         sum_entry = &sum->entries[0];
2606
2607         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2608                 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2609
2610                 /* readahead node pages */
2611                 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2612
2613                 for (idx = addr; idx < addr + nrpages; idx++) {
2614                         struct page *page = f2fs_get_tmp_page(sbi, idx);
2615
2616                         if (IS_ERR(page))
2617                                 return PTR_ERR(page);
2618
2619                         rn = F2FS_NODE(page);
2620                         sum_entry->nid = rn->footer.nid;
2621                         sum_entry->version = 0;
2622                         sum_entry->ofs_in_node = 0;
2623                         sum_entry++;
2624                         f2fs_put_page(page, 1);
2625                 }
2626
2627                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2628                                                         addr + nrpages);
2629         }
2630         return 0;
2631 }
2632
2633 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2634 {
2635         struct f2fs_nm_info *nm_i = NM_I(sbi);
2636         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2637         struct f2fs_journal *journal = curseg->journal;
2638         int i;
2639
2640         down_write(&curseg->journal_rwsem);
2641         for (i = 0; i < nats_in_cursum(journal); i++) {
2642                 struct nat_entry *ne;
2643                 struct f2fs_nat_entry raw_ne;
2644                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2645
2646                 raw_ne = nat_in_journal(journal, i);
2647
2648                 ne = __lookup_nat_cache(nm_i, nid);
2649                 if (!ne) {
2650                         ne = __alloc_nat_entry(nid, true);
2651                         __init_nat_entry(nm_i, ne, &raw_ne, true);
2652                 }
2653
2654                 /*
2655                  * if a free nat in journal has not been used after last
2656                  * checkpoint, we should remove it from available nids,
2657                  * since later we will add it again.
2658                  */
2659                 if (!get_nat_flag(ne, IS_DIRTY) &&
2660                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2661                         spin_lock(&nm_i->nid_list_lock);
2662                         nm_i->available_nids--;
2663                         spin_unlock(&nm_i->nid_list_lock);
2664                 }
2665
2666                 __set_nat_cache_dirty(nm_i, ne);
2667         }
2668         update_nats_in_cursum(journal, -i);
2669         up_write(&curseg->journal_rwsem);
2670 }
2671
2672 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2673                                                 struct list_head *head, int max)
2674 {
2675         struct nat_entry_set *cur;
2676
2677         if (nes->entry_cnt >= max)
2678                 goto add_out;
2679
2680         list_for_each_entry(cur, head, set_list) {
2681                 if (cur->entry_cnt >= nes->entry_cnt) {
2682                         list_add(&nes->set_list, cur->set_list.prev);
2683                         return;
2684                 }
2685         }
2686 add_out:
2687         list_add_tail(&nes->set_list, head);
2688 }
2689
2690 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2691                                                 struct page *page)
2692 {
2693         struct f2fs_nm_info *nm_i = NM_I(sbi);
2694         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2695         struct f2fs_nat_block *nat_blk = page_address(page);
2696         int valid = 0;
2697         int i = 0;
2698
2699         if (!enabled_nat_bits(sbi, NULL))
2700                 return;
2701
2702         if (nat_index == 0) {
2703                 valid = 1;
2704                 i = 1;
2705         }
2706         for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2707                 if (nat_blk->entries[i].block_addr != NULL_ADDR)
2708                         valid++;
2709         }
2710         if (valid == 0) {
2711                 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2712                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2713                 return;
2714         }
2715
2716         __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2717         if (valid == NAT_ENTRY_PER_BLOCK)
2718                 __set_bit_le(nat_index, nm_i->full_nat_bits);
2719         else
2720                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2721 }
2722
2723 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2724                 struct nat_entry_set *set, struct cp_control *cpc)
2725 {
2726         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2727         struct f2fs_journal *journal = curseg->journal;
2728         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2729         bool to_journal = true;
2730         struct f2fs_nat_block *nat_blk;
2731         struct nat_entry *ne, *cur;
2732         struct page *page = NULL;
2733
2734         /*
2735          * there are two steps to flush nat entries:
2736          * #1, flush nat entries to journal in current hot data summary block.
2737          * #2, flush nat entries to nat page.
2738          */
2739         if (enabled_nat_bits(sbi, cpc) ||
2740                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2741                 to_journal = false;
2742
2743         if (to_journal) {
2744                 down_write(&curseg->journal_rwsem);
2745         } else {
2746                 page = get_next_nat_page(sbi, start_nid);
2747                 if (IS_ERR(page))
2748                         return PTR_ERR(page);
2749
2750                 nat_blk = page_address(page);
2751                 f2fs_bug_on(sbi, !nat_blk);
2752         }
2753
2754         /* flush dirty nats in nat entry set */
2755         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2756                 struct f2fs_nat_entry *raw_ne;
2757                 nid_t nid = nat_get_nid(ne);
2758                 int offset;
2759
2760                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2761
2762                 if (to_journal) {
2763                         offset = f2fs_lookup_journal_in_cursum(journal,
2764                                                         NAT_JOURNAL, nid, 1);
2765                         f2fs_bug_on(sbi, offset < 0);
2766                         raw_ne = &nat_in_journal(journal, offset);
2767                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2768                 } else {
2769                         raw_ne = &nat_blk->entries[nid - start_nid];
2770                 }
2771                 raw_nat_from_node_info(raw_ne, &ne->ni);
2772                 nat_reset_flag(ne);
2773                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2774                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2775                         add_free_nid(sbi, nid, false, true);
2776                 } else {
2777                         spin_lock(&NM_I(sbi)->nid_list_lock);
2778                         update_free_nid_bitmap(sbi, nid, false, false);
2779                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2780                 }
2781         }
2782
2783         if (to_journal) {
2784                 up_write(&curseg->journal_rwsem);
2785         } else {
2786                 __update_nat_bits(sbi, start_nid, page);
2787                 f2fs_put_page(page, 1);
2788         }
2789
2790         /* Allow dirty nats by node block allocation in write_begin */
2791         if (!set->entry_cnt) {
2792                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2793                 kmem_cache_free(nat_entry_set_slab, set);
2794         }
2795         return 0;
2796 }
2797
2798 /*
2799  * This function is called during the checkpointing process.
2800  */
2801 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2802 {
2803         struct f2fs_nm_info *nm_i = NM_I(sbi);
2804         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2805         struct f2fs_journal *journal = curseg->journal;
2806         struct nat_entry_set *setvec[SETVEC_SIZE];
2807         struct nat_entry_set *set, *tmp;
2808         unsigned int found;
2809         nid_t set_idx = 0;
2810         LIST_HEAD(sets);
2811         int err = 0;
2812
2813         /* during unmount, let's flush nat_bits before checking dirty_nat_cnt */
2814         if (enabled_nat_bits(sbi, cpc)) {
2815                 down_write(&nm_i->nat_tree_lock);
2816                 remove_nats_in_journal(sbi);
2817                 up_write(&nm_i->nat_tree_lock);
2818         }
2819
2820         if (!nm_i->dirty_nat_cnt)
2821                 return 0;
2822
2823         down_write(&nm_i->nat_tree_lock);
2824
2825         /*
2826          * if there are no enough space in journal to store dirty nat
2827          * entries, remove all entries from journal and merge them
2828          * into nat entry set.
2829          */
2830         if (enabled_nat_bits(sbi, cpc) ||
2831                 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2832                 remove_nats_in_journal(sbi);
2833
2834         while ((found = __gang_lookup_nat_set(nm_i,
2835                                         set_idx, SETVEC_SIZE, setvec))) {
2836                 unsigned idx;
2837                 set_idx = setvec[found - 1]->set + 1;
2838                 for (idx = 0; idx < found; idx++)
2839                         __adjust_nat_entry_set(setvec[idx], &sets,
2840                                                 MAX_NAT_JENTRIES(journal));
2841         }
2842
2843         /* flush dirty nats in nat entry set */
2844         list_for_each_entry_safe(set, tmp, &sets, set_list) {
2845                 err = __flush_nat_entry_set(sbi, set, cpc);
2846                 if (err)
2847                         break;
2848         }
2849
2850         up_write(&nm_i->nat_tree_lock);
2851         /* Allow dirty nats by node block allocation in write_begin */
2852
2853         return err;
2854 }
2855
2856 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2857 {
2858         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2859         struct f2fs_nm_info *nm_i = NM_I(sbi);
2860         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2861         unsigned int i;
2862         __u64 cp_ver = cur_cp_version(ckpt);
2863         block_t nat_bits_addr;
2864
2865         if (!enabled_nat_bits(sbi, NULL))
2866                 return 0;
2867
2868         nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2869         nm_i->nat_bits = f2fs_kzalloc(sbi,
2870                         nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2871         if (!nm_i->nat_bits)
2872                 return -ENOMEM;
2873
2874         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2875                                                 nm_i->nat_bits_blocks;
2876         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2877                 struct page *page;
2878
2879                 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
2880                 if (IS_ERR(page))
2881                         return PTR_ERR(page);
2882
2883                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2884                                         page_address(page), F2FS_BLKSIZE);
2885                 f2fs_put_page(page, 1);
2886         }
2887
2888         cp_ver |= (cur_cp_crc(ckpt) << 32);
2889         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2890                 disable_nat_bits(sbi, true);
2891                 return 0;
2892         }
2893
2894         nm_i->full_nat_bits = nm_i->nat_bits + 8;
2895         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2896
2897         f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2898         return 0;
2899 }
2900
2901 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2902 {
2903         struct f2fs_nm_info *nm_i = NM_I(sbi);
2904         unsigned int i = 0;
2905         nid_t nid, last_nid;
2906
2907         if (!enabled_nat_bits(sbi, NULL))
2908                 return;
2909
2910         for (i = 0; i < nm_i->nat_blocks; i++) {
2911                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2912                 if (i >= nm_i->nat_blocks)
2913                         break;
2914
2915                 __set_bit_le(i, nm_i->nat_block_bitmap);
2916
2917                 nid = i * NAT_ENTRY_PER_BLOCK;
2918                 last_nid = nid + NAT_ENTRY_PER_BLOCK;
2919
2920                 spin_lock(&NM_I(sbi)->nid_list_lock);
2921                 for (; nid < last_nid; nid++)
2922                         update_free_nid_bitmap(sbi, nid, true, true);
2923                 spin_unlock(&NM_I(sbi)->nid_list_lock);
2924         }
2925
2926         for (i = 0; i < nm_i->nat_blocks; i++) {
2927                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2928                 if (i >= nm_i->nat_blocks)
2929                         break;
2930
2931                 __set_bit_le(i, nm_i->nat_block_bitmap);
2932         }
2933 }
2934
2935 static int init_node_manager(struct f2fs_sb_info *sbi)
2936 {
2937         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2938         struct f2fs_nm_info *nm_i = NM_I(sbi);
2939         unsigned char *version_bitmap;
2940         unsigned int nat_segs;
2941         int err;
2942
2943         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2944
2945         /* segment_count_nat includes pair segment so divide to 2. */
2946         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2947         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2948         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2949
2950         /* not used nids: 0, node, meta, (and root counted as valid node) */
2951         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2952                                 sbi->nquota_files - F2FS_RESERVED_NODE_NUM;
2953         nm_i->nid_cnt[FREE_NID] = 0;
2954         nm_i->nid_cnt[PREALLOC_NID] = 0;
2955         nm_i->nat_cnt = 0;
2956         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2957         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2958         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2959
2960         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2961         INIT_LIST_HEAD(&nm_i->free_nid_list);
2962         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2963         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2964         INIT_LIST_HEAD(&nm_i->nat_entries);
2965         spin_lock_init(&nm_i->nat_list_lock);
2966
2967         mutex_init(&nm_i->build_lock);
2968         spin_lock_init(&nm_i->nid_list_lock);
2969         init_rwsem(&nm_i->nat_tree_lock);
2970
2971         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2972         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2973         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2974         if (!version_bitmap)
2975                 return -EFAULT;
2976
2977         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2978                                         GFP_KERNEL);
2979         if (!nm_i->nat_bitmap)
2980                 return -ENOMEM;
2981
2982         err = __get_nat_bitmaps(sbi);
2983         if (err)
2984                 return err;
2985
2986 #ifdef CONFIG_F2FS_CHECK_FS
2987         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2988                                         GFP_KERNEL);
2989         if (!nm_i->nat_bitmap_mir)
2990                 return -ENOMEM;
2991 #endif
2992
2993         return 0;
2994 }
2995
2996 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2997 {
2998         struct f2fs_nm_info *nm_i = NM_I(sbi);
2999         int i;
3000
3001         nm_i->free_nid_bitmap =
3002                 f2fs_kzalloc(sbi, array_size(sizeof(unsigned char *),
3003                                              nm_i->nat_blocks),
3004                              GFP_KERNEL);
3005         if (!nm_i->free_nid_bitmap)
3006                 return -ENOMEM;
3007
3008         for (i = 0; i < nm_i->nat_blocks; i++) {
3009                 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3010                         f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3011                 if (!nm_i->free_nid_bitmap[i])
3012                         return -ENOMEM;
3013         }
3014
3015         nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3016                                                                 GFP_KERNEL);
3017         if (!nm_i->nat_block_bitmap)
3018                 return -ENOMEM;
3019
3020         nm_i->free_nid_count =
3021                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3022                                               nm_i->nat_blocks),
3023                               GFP_KERNEL);
3024         if (!nm_i->free_nid_count)
3025                 return -ENOMEM;
3026         return 0;
3027 }
3028
3029 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3030 {
3031         int err;
3032
3033         sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3034                                                         GFP_KERNEL);
3035         if (!sbi->nm_info)
3036                 return -ENOMEM;
3037
3038         err = init_node_manager(sbi);
3039         if (err)
3040                 return err;
3041
3042         err = init_free_nid_cache(sbi);
3043         if (err)
3044                 return err;
3045
3046         /* load free nid status from nat_bits table */
3047         load_free_nid_bitmap(sbi);
3048
3049         return f2fs_build_free_nids(sbi, true, true);
3050 }
3051
3052 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3053 {
3054         struct f2fs_nm_info *nm_i = NM_I(sbi);
3055         struct free_nid *i, *next_i;
3056         struct nat_entry *natvec[NATVEC_SIZE];
3057         struct nat_entry_set *setvec[SETVEC_SIZE];
3058         nid_t nid = 0;
3059         unsigned int found;
3060
3061         if (!nm_i)
3062                 return;
3063
3064         /* destroy free nid list */
3065         spin_lock(&nm_i->nid_list_lock);
3066         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3067                 __remove_free_nid(sbi, i, FREE_NID);
3068                 spin_unlock(&nm_i->nid_list_lock);
3069                 kmem_cache_free(free_nid_slab, i);
3070                 spin_lock(&nm_i->nid_list_lock);
3071         }
3072         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3073         f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3074         f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3075         spin_unlock(&nm_i->nid_list_lock);
3076
3077         /* destroy nat cache */
3078         down_write(&nm_i->nat_tree_lock);
3079         while ((found = __gang_lookup_nat_cache(nm_i,
3080                                         nid, NATVEC_SIZE, natvec))) {
3081                 unsigned idx;
3082
3083                 nid = nat_get_nid(natvec[found - 1]) + 1;
3084                 for (idx = 0; idx < found; idx++) {
3085                         spin_lock(&nm_i->nat_list_lock);
3086                         list_del(&natvec[idx]->list);
3087                         spin_unlock(&nm_i->nat_list_lock);
3088
3089                         __del_from_nat_cache(nm_i, natvec[idx]);
3090                 }
3091         }
3092         f2fs_bug_on(sbi, nm_i->nat_cnt);
3093
3094         /* destroy nat set cache */
3095         nid = 0;
3096         while ((found = __gang_lookup_nat_set(nm_i,
3097                                         nid, SETVEC_SIZE, setvec))) {
3098                 unsigned idx;
3099
3100                 nid = setvec[found - 1]->set + 1;
3101                 for (idx = 0; idx < found; idx++) {
3102                         /* entry_cnt is not zero, when cp_error was occurred */
3103                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3104                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3105                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3106                 }
3107         }
3108         up_write(&nm_i->nat_tree_lock);
3109
3110         kvfree(nm_i->nat_block_bitmap);
3111         if (nm_i->free_nid_bitmap) {
3112                 int i;
3113
3114                 for (i = 0; i < nm_i->nat_blocks; i++)
3115                         kvfree(nm_i->free_nid_bitmap[i]);
3116                 kvfree(nm_i->free_nid_bitmap);
3117         }
3118         kvfree(nm_i->free_nid_count);
3119
3120         kvfree(nm_i->nat_bitmap);
3121         kvfree(nm_i->nat_bits);
3122 #ifdef CONFIG_F2FS_CHECK_FS
3123         kvfree(nm_i->nat_bitmap_mir);
3124 #endif
3125         sbi->nm_info = NULL;
3126         kvfree(nm_i);
3127 }
3128
3129 int __init f2fs_create_node_manager_caches(void)
3130 {
3131         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
3132                         sizeof(struct nat_entry));
3133         if (!nat_entry_slab)
3134                 goto fail;
3135
3136         free_nid_slab = f2fs_kmem_cache_create("free_nid",
3137                         sizeof(struct free_nid));
3138         if (!free_nid_slab)
3139                 goto destroy_nat_entry;
3140
3141         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
3142                         sizeof(struct nat_entry_set));
3143         if (!nat_entry_set_slab)
3144                 goto destroy_free_nid;
3145
3146         fsync_node_entry_slab = f2fs_kmem_cache_create("fsync_node_entry",
3147                         sizeof(struct fsync_node_entry));
3148         if (!fsync_node_entry_slab)
3149                 goto destroy_nat_entry_set;
3150         return 0;
3151
3152 destroy_nat_entry_set:
3153         kmem_cache_destroy(nat_entry_set_slab);
3154 destroy_free_nid:
3155         kmem_cache_destroy(free_nid_slab);
3156 destroy_nat_entry:
3157         kmem_cache_destroy(nat_entry_slab);
3158 fail:
3159         return -ENOMEM;
3160 }
3161
3162 void f2fs_destroy_node_manager_caches(void)
3163 {
3164         kmem_cache_destroy(fsync_node_entry_slab);
3165         kmem_cache_destroy(nat_entry_set_slab);
3166         kmem_cache_destroy(free_nid_slab);
3167         kmem_cache_destroy(nat_entry_slab);
3168 }