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