fs/f2fs/node.c

   1 /*
   2  * fs/f2fs/node.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/mpage.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/swap.h>
  18
  19 #include "f2fs.h"
  20 #include "node.h"
  21 #include "segment.h"
  22 #include "trace.h"
  23 #include <trace/events/f2fs.h>
  24
  25 #define on_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
  26
  27 static struct kmem_cache *nat_entry_slab;
  28 static struct kmem_cache *free_nid_slab;
  29 static struct kmem_cache *nat_entry_set_slab;
  30
  31 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
  32 {
  33         struct f2fs_nm_info *nm_i = NM_I(sbi);
  34         struct sysinfo val;
  35         unsigned long avail_ram;
  36         unsigned long mem_size = 0;
  37         bool res = false;
  38
  39         si_meminfo(&val);
  40
  41         /* only uses low memory */
  42         avail_ram = val.totalram - val.totalhigh;
  43
  44         /*
  45          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  46          */
  47         if (type == FREE_NIDS) {
  48                 mem_size = (nm_i->nid_cnt[FREE_NID_LIST] *
  49                                 sizeof(struct free_nid)) >> PAGE_SHIFT;
  50                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  51         } else if (type == NAT_ENTRIES) {
  52                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
  53                                                         PAGE_SHIFT;
  54                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  55                 if (excess_cached_nats(sbi))
  56                         res = false;
  57         } else if (type == DIRTY_DENTS) {
  58                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  59                         return false;
  60                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  61                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  62         } else if (type == INO_ENTRIES) {
  63                 int i;
  64
  65                 for (i = 0; i <= UPDATE_INO; i++)
  66                         mem_size += sbi->im[i].ino_num *
  67                                                 sizeof(struct ino_entry);
  68                 mem_size >>= PAGE_SHIFT;
  69                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  70         } else if (type == EXTENT_CACHE) {
  71                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  72                                 sizeof(struct extent_tree) +
  73                                 atomic_read(&sbi->total_ext_node) *
  74                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  75                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  76         } else {
  77                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
  78                         return true;
  79         }
  80         return res;
  81 }
  82
  83 static void clear_node_page_dirty(struct page *page)
  84 {
  85         struct address_space *mapping = page->mapping;
  86         unsigned int long flags;
  87
  88         if (PageDirty(page)) {
  89                 spin_lock_irqsave(&mapping->tree_lock, flags);
  90                 radix_tree_tag_clear(&mapping->page_tree,
  91                                 page_index(page),
  92                                 PAGECACHE_TAG_DIRTY);
  93                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  94
  95                 clear_page_dirty_for_io(page);
  96                 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
  97         }
  98         ClearPageUptodate(page);
  99 }
 100
 101 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 102 {
 103         pgoff_t index = current_nat_addr(sbi, nid);
 104         return get_meta_page(sbi, index);
 105 }
 106
 107 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 108 {
 109         struct page *src_page;
 110         struct page *dst_page;
 111         pgoff_t src_off;
 112         pgoff_t dst_off;
 113         void *src_addr;
 114         void *dst_addr;
 115         struct f2fs_nm_info *nm_i = NM_I(sbi);
 116
 117         src_off = current_nat_addr(sbi, nid);
 118         dst_off = next_nat_addr(sbi, src_off);
 119
 120         /* get current nat block page with lock */
 121         src_page = get_meta_page(sbi, src_off);
 122         dst_page = grab_meta_page(sbi, dst_off);
 123         f2fs_bug_on(sbi, PageDirty(src_page));
 124
 125         src_addr = page_address(src_page);
 126         dst_addr = page_address(dst_page);
 127         memcpy(dst_addr, src_addr, PAGE_SIZE);
 128         set_page_dirty(dst_page);
 129         f2fs_put_page(src_page, 1);
 130
 131         set_to_next_nat(nm_i, nid);
 132
 133         return dst_page;
 134 }
 135
 136 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 137 {
 138         return radix_tree_lookup(&nm_i->nat_root, n);
 139 }
 140
 141 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 142                 nid_t start, unsigned int nr, struct nat_entry **ep)
 143 {
 144         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 145 }
 146
 147 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 148 {
 149         list_del(&e->list);
 150         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 151         nm_i->nat_cnt--;
 152         kmem_cache_free(nat_entry_slab, e);
 153 }
 154
 155 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 156                                                 struct nat_entry *ne)
 157 {
 158         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 159         struct nat_entry_set *head;
 160
 161         if (get_nat_flag(ne, IS_DIRTY))
 162                 return;
 163
 164         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 165         if (!head) {
 166                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 167
 168                 INIT_LIST_HEAD(&head->entry_list);
 169                 INIT_LIST_HEAD(&head->set_list);
 170                 head->set = set;
 171                 head->entry_cnt = 0;
 172                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 173         }
 174         list_move_tail(&ne->list, &head->entry_list);
 175         nm_i->dirty_nat_cnt++;
 176         head->entry_cnt++;
 177         set_nat_flag(ne, IS_DIRTY, true);
 178 }
 179
 180 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 181                 struct nat_entry_set *set, struct nat_entry *ne)
 182 {
 183         list_move_tail(&ne->list, &nm_i->nat_entries);
 184         set_nat_flag(ne, IS_DIRTY, false);
 185         set->entry_cnt--;
 186         nm_i->dirty_nat_cnt--;
 187 }
 188
 189 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 190                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 191 {
 192         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 193                                                         start, nr);
 194 }
 195
 196 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 197 {
 198         struct f2fs_nm_info *nm_i = NM_I(sbi);
 199         struct nat_entry *e;
 200         bool need = false;
 201
 202         down_read(&nm_i->nat_tree_lock);
 203         e = __lookup_nat_cache(nm_i, nid);
 204         if (e) {
 205                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 206                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 207                         need = true;
 208         }
 209         up_read(&nm_i->nat_tree_lock);
 210         return need;
 211 }
 212
 213 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 214 {
 215         struct f2fs_nm_info *nm_i = NM_I(sbi);
 216         struct nat_entry *e;
 217         bool is_cp = true;
 218
 219         down_read(&nm_i->nat_tree_lock);
 220         e = __lookup_nat_cache(nm_i, nid);
 221         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 222                 is_cp = false;
 223         up_read(&nm_i->nat_tree_lock);
 224         return is_cp;
 225 }
 226
 227 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 228 {
 229         struct f2fs_nm_info *nm_i = NM_I(sbi);
 230         struct nat_entry *e;
 231         bool need_update = true;
 232
 233         down_read(&nm_i->nat_tree_lock);
 234         e = __lookup_nat_cache(nm_i, ino);
 235         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 236                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 237                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 238                 need_update = false;
 239         up_read(&nm_i->nat_tree_lock);
 240         return need_update;
 241 }
 242
 243 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 244                                                                 bool no_fail)
 245 {
 246         struct nat_entry *new;
 247
 248         if (no_fail) {
 249                 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
 250                 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
 251         } else {
 252                 new = kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
 253                 if (!new)
 254                         return NULL;
 255                 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 256                         kmem_cache_free(nat_entry_slab, new);
 257                         return NULL;
 258                 }
 259         }
 260
 261         memset(new, 0, sizeof(struct nat_entry));
 262         nat_set_nid(new, nid);
 263         nat_reset_flag(new);
 264         list_add_tail(&new->list, &nm_i->nat_entries);
 265         nm_i->nat_cnt++;
 266         return new;
 267 }
 268
 269 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 270                                                 struct f2fs_nat_entry *ne)
 271 {
 272         struct f2fs_nm_info *nm_i = NM_I(sbi);
 273         struct nat_entry *e;
 274
 275         e = __lookup_nat_cache(nm_i, nid);
 276         if (!e) {
 277                 e = grab_nat_entry(nm_i, nid, false);
 278                 if (e)
 279                         node_info_from_raw_nat(&e->ni, ne);
 280         } else {
 281                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 282                                 nat_get_blkaddr(e) !=
 283                                         le32_to_cpu(ne->block_addr) ||
 284                                 nat_get_version(e) != ne->version);
 285         }
 286 }
 287
 288 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 289                         block_t new_blkaddr, bool fsync_done)
 290 {
 291         struct f2fs_nm_info *nm_i = NM_I(sbi);
 292         struct nat_entry *e;
 293
 294         down_write(&nm_i->nat_tree_lock);
 295         e = __lookup_nat_cache(nm_i, ni->nid);
 296         if (!e) {
 297                 e = grab_nat_entry(nm_i, ni->nid, true);
 298                 copy_node_info(&e->ni, ni);
 299                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 300         } else if (new_blkaddr == NEW_ADDR) {
 301                 /*
 302                  * when nid is reallocated,
 303                  * previous nat entry can be remained in nat cache.
 304                  * So, reinitialize it with new information.
 305                  */
 306                 copy_node_info(&e->ni, ni);
 307                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 308         }
 309
 310         /* sanity check */
 311         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 312         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 313                         new_blkaddr == NULL_ADDR);
 314         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 315                         new_blkaddr == NEW_ADDR);
 316         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 317                         nat_get_blkaddr(e) != NULL_ADDR &&
 318                         new_blkaddr == NEW_ADDR);
 319
 320         /* increment version no as node is removed */
 321         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 322                 unsigned char version = nat_get_version(e);
 323                 nat_set_version(e, inc_node_version(version));
 324
 325                 /* in order to reuse the nid */
 326                 if (nm_i->next_scan_nid > ni->nid)
 327                         nm_i->next_scan_nid = ni->nid;
 328         }
 329
 330         /* change address */
 331         nat_set_blkaddr(e, new_blkaddr);
 332         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 333                 set_nat_flag(e, IS_CHECKPOINTED, false);
 334         __set_nat_cache_dirty(nm_i, e);
 335
 336         /* update fsync_mark if its inode nat entry is still alive */
 337         if (ni->nid != ni->ino)
 338                 e = __lookup_nat_cache(nm_i, ni->ino);
 339         if (e) {
 340                 if (fsync_done && ni->nid == ni->ino)
 341                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 342                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 343         }
 344         up_write(&nm_i->nat_tree_lock);
 345 }
 346
 347 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 348 {
 349         struct f2fs_nm_info *nm_i = NM_I(sbi);
 350         int nr = nr_shrink;
 351
 352         if (!down_write_trylock(&nm_i->nat_tree_lock))
 353                 return 0;
 354
 355         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 356                 struct nat_entry *ne;
 357                 ne = list_first_entry(&nm_i->nat_entries,
 358                                         struct nat_entry, list);
 359                 __del_from_nat_cache(nm_i, ne);
 360                 nr_shrink--;
 361         }
 362         up_write(&nm_i->nat_tree_lock);
 363         return nr - nr_shrink;
 364 }
 365
 366 /*
 367  * This function always returns success
 368  */
 369 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 370 {
 371         struct f2fs_nm_info *nm_i = NM_I(sbi);
 372         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 373         struct f2fs_journal *journal = curseg->journal;
 374         nid_t start_nid = START_NID(nid);
 375         struct f2fs_nat_block *nat_blk;
 376         struct page *page = NULL;
 377         struct f2fs_nat_entry ne;
 378         struct nat_entry *e;
 379         pgoff_t index;
 380         int i;
 381
 382         ni->nid = nid;
 383
 384         /* Check nat cache */
 385         down_read(&nm_i->nat_tree_lock);
 386         e = __lookup_nat_cache(nm_i, nid);
 387         if (e) {
 388                 ni->ino = nat_get_ino(e);
 389                 ni->blk_addr = nat_get_blkaddr(e);
 390                 ni->version = nat_get_version(e);
 391                 up_read(&nm_i->nat_tree_lock);
 392                 return;
 393         }
 394
 395         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 396
 397         /* Check current segment summary */
 398         down_read(&curseg->journal_rwsem);
 399         i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 400         if (i >= 0) {
 401                 ne = nat_in_journal(journal, i);
 402                 node_info_from_raw_nat(ni, &ne);
 403         }
 404         up_read(&curseg->journal_rwsem);
 405         if (i >= 0) {
 406                 up_read(&nm_i->nat_tree_lock);
 407                 goto cache;
 408         }
 409
 410         /* Fill node_info from nat page */
 411         index = current_nat_addr(sbi, nid);
 412         up_read(&nm_i->nat_tree_lock);
 413
 414         page = get_meta_page(sbi, index);
 415         nat_blk = (struct f2fs_nat_block *)page_address(page);
 416         ne = nat_blk->entries[nid - start_nid];
 417         node_info_from_raw_nat(ni, &ne);
 418         f2fs_put_page(page, 1);
 419 cache:
 420         /* cache nat entry */
 421         down_write(&nm_i->nat_tree_lock);
 422         cache_nat_entry(sbi, nid, &ne);
 423         up_write(&nm_i->nat_tree_lock);
 424 }
 425
 426 /*
 427  * readahead MAX_RA_NODE number of node pages.
 428  */
 429 static void ra_node_pages(struct page *parent, int start, int n)
 430 {
 431         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 432         struct blk_plug plug;
 433         int i, end;
 434         nid_t nid;
 435
 436         blk_start_plug(&plug);
 437
 438         /* Then, try readahead for siblings of the desired node */
 439         end = start + n;
 440         end = min(end, NIDS_PER_BLOCK);
 441         for (i = start; i < end; i++) {
 442                 nid = get_nid(parent, i, false);
 443                 ra_node_page(sbi, nid);
 444         }
 445
 446         blk_finish_plug(&plug);
 447 }
 448
 449 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 450 {
 451         const long direct_index = ADDRS_PER_INODE(dn->inode);
 452         const long direct_blks = ADDRS_PER_BLOCK;
 453         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 454         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 455         int cur_level = dn->cur_level;
 456         int max_level = dn->max_level;
 457         pgoff_t base = 0;
 458
 459         if (!dn->max_level)
 460                 return pgofs + 1;
 461
 462         while (max_level-- > cur_level)
 463                 skipped_unit *= NIDS_PER_BLOCK;
 464
 465         switch (dn->max_level) {
 466         case 3:
 467                 base += 2 * indirect_blks;
 468         case 2:
 469                 base += 2 * direct_blks;
 470         case 1:
 471                 base += direct_index;
 472                 break;
 473         default:
 474                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 475         }
 476
 477         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 478 }
 479
 480 /*
 481  * The maximum depth is four.
 482  * Offset[0] will have raw inode offset.
 483  */
 484 static int get_node_path(struct inode *inode, long block,
 485                                 int offset[4], unsigned int noffset[4])
 486 {
 487         const long direct_index = ADDRS_PER_INODE(inode);
 488         const long direct_blks = ADDRS_PER_BLOCK;
 489         const long dptrs_per_blk = NIDS_PER_BLOCK;
 490         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 491         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 492         int n = 0;
 493         int level = 0;
 494
 495         noffset[0] = 0;
 496
 497         if (block < direct_index) {
 498                 offset[n] = block;
 499                 goto got;
 500         }
 501         block -= direct_index;
 502         if (block < direct_blks) {
 503                 offset[n++] = NODE_DIR1_BLOCK;
 504                 noffset[n] = 1;
 505                 offset[n] = block;
 506                 level = 1;
 507                 goto got;
 508         }
 509         block -= direct_blks;
 510         if (block < direct_blks) {
 511                 offset[n++] = NODE_DIR2_BLOCK;
 512                 noffset[n] = 2;
 513                 offset[n] = block;
 514                 level = 1;
 515                 goto got;
 516         }
 517         block -= direct_blks;
 518         if (block < indirect_blks) {
 519                 offset[n++] = NODE_IND1_BLOCK;
 520                 noffset[n] = 3;
 521                 offset[n++] = block / direct_blks;
 522                 noffset[n] = 4 + offset[n - 1];
 523                 offset[n] = block % direct_blks;
 524                 level = 2;
 525                 goto got;
 526         }
 527         block -= indirect_blks;
 528         if (block < indirect_blks) {
 529                 offset[n++] = NODE_IND2_BLOCK;
 530                 noffset[n] = 4 + dptrs_per_blk;
 531                 offset[n++] = block / direct_blks;
 532                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 533                 offset[n] = block % direct_blks;
 534                 level = 2;
 535                 goto got;
 536         }
 537         block -= indirect_blks;
 538         if (block < dindirect_blks) {
 539                 offset[n++] = NODE_DIND_BLOCK;
 540                 noffset[n] = 5 + (dptrs_per_blk * 2);
 541                 offset[n++] = block / indirect_blks;
 542                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 543                               offset[n - 1] * (dptrs_per_blk + 1);
 544                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 545                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 546                               offset[n - 2] * (dptrs_per_blk + 1) +
 547                               offset[n - 1];
 548                 offset[n] = block % direct_blks;
 549                 level = 3;
 550                 goto got;
 551         } else {
 552                 BUG();
 553         }
 554 got:
 555         return level;
 556 }
 557
 558 /*
 559  * Caller should call f2fs_put_dnode(dn).
 560  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 561  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 562  * In the case of RDONLY_NODE, we don't need to care about mutex.
 563  */
 564 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 565 {
 566         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 567         struct page *npage[4];
 568         struct page *parent = NULL;
 569         int offset[4];
 570         unsigned int noffset[4];
 571         nid_t nids[4];
 572         int level, i = 0;
 573         int err = 0;
 574
 575         level = get_node_path(dn->inode, index, offset, noffset);
 576
 577         nids[0] = dn->inode->i_ino;
 578         npage[0] = dn->inode_page;
 579
 580         if (!npage[0]) {
 581                 npage[0] = get_node_page(sbi, nids[0]);
 582                 if (IS_ERR(npage[0]))
 583                         return PTR_ERR(npage[0]);
 584         }
 585
 586         /* if inline_data is set, should not report any block indices */
 587         if (f2fs_has_inline_data(dn->inode) && index) {
 588                 err = -ENOENT;
 589                 f2fs_put_page(npage[0], 1);
 590                 goto release_out;
 591         }
 592
 593         parent = npage[0];
 594         if (level != 0)
 595                 nids[1] = get_nid(parent, offset[0], true);
 596         dn->inode_page = npage[0];
 597         dn->inode_page_locked = true;
 598
 599         /* get indirect or direct nodes */
 600         for (i = 1; i <= level; i++) {
 601                 bool done = false;
 602
 603                 if (!nids[i] && mode == ALLOC_NODE) {
 604                         /* alloc new node */
 605                         if (!alloc_nid(sbi, &(nids[i]))) {
 606                                 err = -ENOSPC;
 607                                 goto release_pages;
 608                         }
 609
 610                         dn->nid = nids[i];
 611                         npage[i] = new_node_page(dn, noffset[i], NULL);
 612                         if (IS_ERR(npage[i])) {
 613                                 alloc_nid_failed(sbi, nids[i]);
 614                                 err = PTR_ERR(npage[i]);
 615                                 goto release_pages;
 616                         }
 617
 618                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 619                         alloc_nid_done(sbi, nids[i]);
 620                         done = true;
 621                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 622                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 623                         if (IS_ERR(npage[i])) {
 624                                 err = PTR_ERR(npage[i]);
 625                                 goto release_pages;
 626                         }
 627                         done = true;
 628                 }
 629                 if (i == 1) {
 630                         dn->inode_page_locked = false;
 631                         unlock_page(parent);
 632                 } else {
 633                         f2fs_put_page(parent, 1);
 634                 }
 635
 636                 if (!done) {
 637                         npage[i] = get_node_page(sbi, nids[i]);
 638                         if (IS_ERR(npage[i])) {
 639                                 err = PTR_ERR(npage[i]);
 640                                 f2fs_put_page(npage[0], 0);
 641                                 goto release_out;
 642                         }
 643                 }
 644                 if (i < level) {
 645                         parent = npage[i];
 646                         nids[i + 1] = get_nid(parent, offset[i], false);
 647                 }
 648         }
 649         dn->nid = nids[level];
 650         dn->ofs_in_node = offset[level];
 651         dn->node_page = npage[level];
 652         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 653         return 0;
 654
 655 release_pages:
 656         f2fs_put_page(parent, 1);
 657         if (i > 1)
 658                 f2fs_put_page(npage[0], 0);
 659 release_out:
 660         dn->inode_page = NULL;
 661         dn->node_page = NULL;
 662         if (err == -ENOENT) {
 663                 dn->cur_level = i;
 664                 dn->max_level = level;
 665                 dn->ofs_in_node = offset[level];
 666         }
 667         return err;
 668 }
 669
 670 static void truncate_node(struct dnode_of_data *dn)
 671 {
 672         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 673         struct node_info ni;
 674
 675         get_node_info(sbi, dn->nid, &ni);
 676         if (dn->inode->i_blocks == 0) {
 677                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 678                 goto invalidate;
 679         }
 680         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 681
 682         /* Deallocate node address */
 683         invalidate_blocks(sbi, ni.blk_addr);
 684         dec_valid_node_count(sbi, dn->inode);
 685         set_node_addr(sbi, &ni, NULL_ADDR, false);
 686
 687         if (dn->nid == dn->inode->i_ino) {
 688                 remove_orphan_inode(sbi, dn->nid);
 689                 dec_valid_inode_count(sbi);
 690                 f2fs_inode_synced(dn->inode);
 691         }
 692 invalidate:
 693         clear_node_page_dirty(dn->node_page);
 694         set_sbi_flag(sbi, SBI_IS_DIRTY);
 695
 696         f2fs_put_page(dn->node_page, 1);
 697
 698         invalidate_mapping_pages(NODE_MAPPING(sbi),
 699                         dn->node_page->index, dn->node_page->index);
 700
 701         dn->node_page = NULL;
 702         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 703 }
 704
 705 static int truncate_dnode(struct dnode_of_data *dn)
 706 {
 707         struct page *page;
 708
 709         if (dn->nid == 0)
 710                 return 1;
 711
 712         /* get direct node */
 713         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 714         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 715                 return 1;
 716         else if (IS_ERR(page))
 717                 return PTR_ERR(page);
 718
 719         /* Make dnode_of_data for parameter */
 720         dn->node_page = page;
 721         dn->ofs_in_node = 0;
 722         truncate_data_blocks(dn);
 723         truncate_node(dn);
 724         return 1;
 725 }
 726
 727 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 728                                                 int ofs, int depth)
 729 {
 730         struct dnode_of_data rdn = *dn;
 731         struct page *page;
 732         struct f2fs_node *rn;
 733         nid_t child_nid;
 734         unsigned int child_nofs;
 735         int freed = 0;
 736         int i, ret;
 737
 738         if (dn->nid == 0)
 739                 return NIDS_PER_BLOCK + 1;
 740
 741         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 742
 743         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 744         if (IS_ERR(page)) {
 745                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 746                 return PTR_ERR(page);
 747         }
 748
 749         ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 750
 751         rn = F2FS_NODE(page);
 752         if (depth < 3) {
 753                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 754                         child_nid = le32_to_cpu(rn->in.nid[i]);
 755                         if (child_nid == 0)
 756                                 continue;
 757                         rdn.nid = child_nid;
 758                         ret = truncate_dnode(&rdn);
 759                         if (ret < 0)
 760                                 goto out_err;
 761                         if (set_nid(page, i, 0, false))
 762                                 dn->node_changed = true;
 763                 }
 764         } else {
 765                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 766                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 767                         child_nid = le32_to_cpu(rn->in.nid[i]);
 768                         if (child_nid == 0) {
 769                                 child_nofs += NIDS_PER_BLOCK + 1;
 770                                 continue;
 771                         }
 772                         rdn.nid = child_nid;
 773                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 774                         if (ret == (NIDS_PER_BLOCK + 1)) {
 775                                 if (set_nid(page, i, 0, false))
 776                                         dn->node_changed = true;
 777                                 child_nofs += ret;
 778                         } else if (ret < 0 && ret != -ENOENT) {
 779                                 goto out_err;
 780                         }
 781                 }
 782                 freed = child_nofs;
 783         }
 784
 785         if (!ofs) {
 786                 /* remove current indirect node */
 787                 dn->node_page = page;
 788                 truncate_node(dn);
 789                 freed++;
 790         } else {
 791                 f2fs_put_page(page, 1);
 792         }
 793         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 794         return freed;
 795
 796 out_err:
 797         f2fs_put_page(page, 1);
 798         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 799         return ret;
 800 }
 801
 802 static int truncate_partial_nodes(struct dnode_of_data *dn,
 803                         struct f2fs_inode *ri, int *offset, int depth)
 804 {
 805         struct page *pages[2];
 806         nid_t nid[3];
 807         nid_t child_nid;
 808         int err = 0;
 809         int i;
 810         int idx = depth - 2;
 811
 812         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 813         if (!nid[0])
 814                 return 0;
 815
 816         /* get indirect nodes in the path */
 817         for (i = 0; i < idx + 1; i++) {
 818                 /* reference count'll be increased */
 819                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 820                 if (IS_ERR(pages[i])) {
 821                         err = PTR_ERR(pages[i]);
 822                         idx = i - 1;
 823                         goto fail;
 824                 }
 825                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 826         }
 827
 828         ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 829
 830         /* free direct nodes linked to a partial indirect node */
 831         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 832                 child_nid = get_nid(pages[idx], i, false);
 833                 if (!child_nid)
 834                         continue;
 835                 dn->nid = child_nid;
 836                 err = truncate_dnode(dn);
 837                 if (err < 0)
 838                         goto fail;
 839                 if (set_nid(pages[idx], i, 0, false))
 840                         dn->node_changed = true;
 841         }
 842
 843         if (offset[idx + 1] == 0) {
 844                 dn->node_page = pages[idx];
 845                 dn->nid = nid[idx];
 846                 truncate_node(dn);
 847         } else {
 848                 f2fs_put_page(pages[idx], 1);
 849         }
 850         offset[idx]++;
 851         offset[idx + 1] = 0;
 852         idx--;
 853 fail:
 854         for (i = idx; i >= 0; i--)
 855                 f2fs_put_page(pages[i], 1);
 856
 857         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 858
 859         return err;
 860 }
 861
 862 /*
 863  * All the block addresses of data and nodes should be nullified.
 864  */
 865 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 866 {
 867         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 868         int err = 0, cont = 1;
 869         int level, offset[4], noffset[4];
 870         unsigned int nofs = 0;
 871         struct f2fs_inode *ri;
 872         struct dnode_of_data dn;
 873         struct page *page;
 874
 875         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 876
 877         level = get_node_path(inode, from, offset, noffset);
 878
 879         page = get_node_page(sbi, inode->i_ino);
 880         if (IS_ERR(page)) {
 881                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 882                 return PTR_ERR(page);
 883         }
 884
 885         set_new_dnode(&dn, inode, page, NULL, 0);
 886         unlock_page(page);
 887
 888         ri = F2FS_INODE(page);
 889         switch (level) {
 890         case 0:
 891         case 1:
 892                 nofs = noffset[1];
 893                 break;
 894         case 2:
 895                 nofs = noffset[1];
 896                 if (!offset[level - 1])
 897                         goto skip_partial;
 898                 err = truncate_partial_nodes(&dn, ri, offset, level);
 899                 if (err < 0 && err != -ENOENT)
 900                         goto fail;
 901                 nofs += 1 + NIDS_PER_BLOCK;
 902                 break;
 903         case 3:
 904                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 905                 if (!offset[level - 1])
 906                         goto skip_partial;
 907                 err = truncate_partial_nodes(&dn, ri, offset, level);
 908                 if (err < 0 && err != -ENOENT)
 909                         goto fail;
 910                 break;
 911         default:
 912                 BUG();
 913         }
 914
 915 skip_partial:
 916         while (cont) {
 917                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 918                 switch (offset[0]) {
 919                 case NODE_DIR1_BLOCK:
 920                 case NODE_DIR2_BLOCK:
 921                         err = truncate_dnode(&dn);
 922                         break;
 923
 924                 case NODE_IND1_BLOCK:
 925                 case NODE_IND2_BLOCK:
 926                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 927                         break;
 928
 929                 case NODE_DIND_BLOCK:
 930                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 931                         cont = 0;
 932                         break;
 933
 934                 default:
 935                         BUG();
 936                 }
 937                 if (err < 0 && err != -ENOENT)
 938                         goto fail;
 939                 if (offset[1] == 0 &&
 940                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 941                         lock_page(page);
 942                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
 943                         f2fs_wait_on_page_writeback(page, NODE, true);
 944                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 945                         set_page_dirty(page);
 946                         unlock_page(page);
 947                 }
 948                 offset[1] = 0;
 949                 offset[0]++;
 950                 nofs += err;
 951         }
 952 fail:
 953         f2fs_put_page(page, 0);
 954         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 955         return err > 0 ? 0 : err;
 956 }
 957
 958 int truncate_xattr_node(struct inode *inode, struct page *page)
 959 {
 960         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 961         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 962         struct dnode_of_data dn;
 963         struct page *npage;
 964
 965         if (!nid)
 966                 return 0;
 967
 968         npage = get_node_page(sbi, nid);
 969         if (IS_ERR(npage))
 970                 return PTR_ERR(npage);
 971
 972         f2fs_i_xnid_write(inode, 0);
 973
 974         set_new_dnode(&dn, inode, page, npage, nid);
 975
 976         if (page)
 977                 dn.inode_page_locked = true;
 978         truncate_node(&dn);
 979         return 0;
 980 }
 981
 982 /*
 983  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 984  * f2fs_unlock_op().
 985  */
 986 int remove_inode_page(struct inode *inode)
 987 {
 988         struct dnode_of_data dn;
 989         int err;
 990
 991         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 992         err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
 993         if (err)
 994                 return err;
 995
 996         err = truncate_xattr_node(inode, dn.inode_page);
 997         if (err) {
 998                 f2fs_put_dnode(&dn);
 999                 return err;
1000         }
1001
1002         /* remove potential inline_data blocks */
1003         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1004                                 S_ISLNK(inode->i_mode))
1005                 truncate_data_blocks_range(&dn, 1);
1006
1007         /* 0 is possible, after f2fs_new_inode() has failed */
1008         f2fs_bug_on(F2FS_I_SB(inode),
1009                         inode->i_blocks != 0 && inode->i_blocks != 1);
1010
1011         /* will put inode & node pages */
1012         truncate_node(&dn);
1013         return 0;
1014 }
1015
1016 struct page *new_inode_page(struct inode *inode)
1017 {
1018         struct dnode_of_data dn;
1019
1020         /* allocate inode page for new inode */
1021         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1022
1023         /* caller should f2fs_put_page(page, 1); */
1024         return new_node_page(&dn, 0, NULL);
1025 }
1026
1027 struct page *new_node_page(struct dnode_of_data *dn,
1028                                 unsigned int ofs, struct page *ipage)
1029 {
1030         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1031         struct node_info new_ni;
1032         struct page *page;
1033         int err;
1034
1035         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1036                 return ERR_PTR(-EPERM);
1037
1038         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1039         if (!page)
1040                 return ERR_PTR(-ENOMEM);
1041
1042         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1043                 err = -ENOSPC;
1044                 goto fail;
1045         }
1046 #ifdef CONFIG_F2FS_CHECK_FS
1047         get_node_info(sbi, dn->nid, &new_ni);
1048         f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1049 #endif
1050         new_ni.nid = dn->nid;
1051         new_ni.ino = dn->inode->i_ino;
1052         new_ni.blk_addr = NULL_ADDR;
1053         new_ni.flag = 0;
1054         new_ni.version = 0;
1055         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1056
1057         f2fs_wait_on_page_writeback(page, NODE, true);
1058         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1059         set_cold_node(dn->inode, page);
1060         if (!PageUptodate(page))
1061                 SetPageUptodate(page);
1062         if (set_page_dirty(page))
1063                 dn->node_changed = true;
1064
1065         if (f2fs_has_xattr_block(ofs))
1066                 f2fs_i_xnid_write(dn->inode, dn->nid);
1067
1068         if (ofs == 0)
1069                 inc_valid_inode_count(sbi);
1070         return page;
1071
1072 fail:
1073         clear_node_page_dirty(page);
1074         f2fs_put_page(page, 1);
1075         return ERR_PTR(err);
1076 }
1077
1078 /*
1079  * Caller should do after getting the following values.
1080  * 0: f2fs_put_page(page, 0)
1081  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1082  */
1083 static int read_node_page(struct page *page, int op_flags)
1084 {
1085         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1086         struct node_info ni;
1087         struct f2fs_io_info fio = {
1088                 .sbi = sbi,
1089                 .type = NODE,
1090                 .op = REQ_OP_READ,
1091                 .op_flags = op_flags,
1092                 .page = page,
1093                 .encrypted_page = NULL,
1094         };
1095
1096         if (PageUptodate(page))
1097                 return LOCKED_PAGE;
1098
1099         get_node_info(sbi, page->index, &ni);
1100
1101         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1102                 ClearPageUptodate(page);
1103                 return -ENOENT;
1104         }
1105
1106         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1107         return f2fs_submit_page_bio(&fio);
1108 }
1109
1110 /*
1111  * Readahead a node page
1112  */
1113 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1114 {
1115         struct page *apage;
1116         int err;
1117
1118         if (!nid)
1119                 return;
1120         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1121
1122         rcu_read_lock();
1123         apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1124         rcu_read_unlock();
1125         if (apage)
1126                 return;
1127
1128         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1129         if (!apage)
1130                 return;
1131
1132         err = read_node_page(apage, REQ_RAHEAD);
1133         f2fs_put_page(apage, err ? 1 : 0);
1134 }
1135
1136 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1137                                         struct page *parent, int start)
1138 {
1139         struct page *page;
1140         int err;
1141
1142         if (!nid)
1143                 return ERR_PTR(-ENOENT);
1144         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1145 repeat:
1146         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1147         if (!page)
1148                 return ERR_PTR(-ENOMEM);
1149
1150         err = read_node_page(page, 0);
1151         if (err < 0) {
1152                 f2fs_put_page(page, 1);
1153                 return ERR_PTR(err);
1154         } else if (err == LOCKED_PAGE) {
1155                 goto page_hit;
1156         }
1157
1158         if (parent)
1159                 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1160
1161         lock_page(page);
1162
1163         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1164                 f2fs_put_page(page, 1);
1165                 goto repeat;
1166         }
1167
1168         if (unlikely(!PageUptodate(page)))
1169                 goto out_err;
1170 page_hit:
1171         if(unlikely(nid != nid_of_node(page))) {
1172                 f2fs_bug_on(sbi, 1);
1173                 ClearPageUptodate(page);
1174 out_err:
1175                 f2fs_put_page(page, 1);
1176                 return ERR_PTR(-EIO);
1177         }
1178         return page;
1179 }
1180
1181 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1182 {
1183         return __get_node_page(sbi, nid, NULL, 0);
1184 }
1185
1186 struct page *get_node_page_ra(struct page *parent, int start)
1187 {
1188         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1189         nid_t nid = get_nid(parent, start, false);
1190
1191         return __get_node_page(sbi, nid, parent, start);
1192 }
1193
1194 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1195 {
1196         struct inode *inode;
1197         struct page *page;
1198         int ret;
1199
1200         /* should flush inline_data before evict_inode */
1201         inode = ilookup(sbi->sb, ino);
1202         if (!inode)
1203                 return;
1204
1205         page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1206         if (!page)
1207                 goto iput_out;
1208
1209         if (!PageUptodate(page))
1210                 goto page_out;
1211
1212         if (!PageDirty(page))
1213                 goto page_out;
1214
1215         if (!clear_page_dirty_for_io(page))
1216                 goto page_out;
1217
1218         ret = f2fs_write_inline_data(inode, page);
1219         inode_dec_dirty_pages(inode);
1220         remove_dirty_inode(inode);
1221         if (ret)
1222                 set_page_dirty(page);
1223 page_out:
1224         f2fs_put_page(page, 1);
1225 iput_out:
1226         iput(inode);
1227 }
1228
1229 void move_node_page(struct page *node_page, int gc_type)
1230 {
1231         if (gc_type == FG_GC) {
1232                 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1233                 struct writeback_control wbc = {
1234                         .sync_mode = WB_SYNC_ALL,
1235                         .nr_to_write = 1,
1236                         .for_reclaim = 0,
1237                 };
1238
1239                 set_page_dirty(node_page);
1240                 f2fs_wait_on_page_writeback(node_page, NODE, true);
1241
1242                 f2fs_bug_on(sbi, PageWriteback(node_page));
1243                 if (!clear_page_dirty_for_io(node_page))
1244                         goto out_page;
1245
1246                 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1247                         unlock_page(node_page);
1248                 goto release_page;
1249         } else {
1250                 /* set page dirty and write it */
1251                 if (!PageWriteback(node_page))
1252                         set_page_dirty(node_page);
1253         }
1254 out_page:
1255         unlock_page(node_page);
1256 release_page:
1257         f2fs_put_page(node_page, 0);
1258 }
1259
1260 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1261 {
1262         pgoff_t index, end;
1263         struct pagevec pvec;
1264         struct page *last_page = NULL;
1265
1266         pagevec_init(&pvec, 0);
1267         index = 0;
1268         end = ULONG_MAX;
1269
1270         while (index <= end) {
1271                 int i, nr_pages;
1272                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1273                                 PAGECACHE_TAG_DIRTY,
1274                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1275                 if (nr_pages == 0)
1276                         break;
1277
1278                 for (i = 0; i < nr_pages; i++) {
1279                         struct page *page = pvec.pages[i];
1280
1281                         if (unlikely(f2fs_cp_error(sbi))) {
1282                                 f2fs_put_page(last_page, 0);
1283                                 pagevec_release(&pvec);
1284                                 return ERR_PTR(-EIO);
1285                         }
1286
1287                         if (!IS_DNODE(page) || !is_cold_node(page))
1288                                 continue;
1289                         if (ino_of_node(page) != ino)
1290                                 continue;
1291
1292                         lock_page(page);
1293
1294                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1295 continue_unlock:
1296                                 unlock_page(page);
1297                                 continue;
1298                         }
1299                         if (ino_of_node(page) != ino)
1300                                 goto continue_unlock;
1301
1302                         if (!PageDirty(page)) {
1303                                 /* someone wrote it for us */
1304                                 goto continue_unlock;
1305                         }
1306
1307                         if (last_page)
1308                                 f2fs_put_page(last_page, 0);
1309
1310                         get_page(page);
1311                         last_page = page;
1312                         unlock_page(page);
1313                 }
1314                 pagevec_release(&pvec);
1315                 cond_resched();
1316         }
1317         return last_page;
1318 }
1319
1320 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1321                                 struct writeback_control *wbc)
1322 {
1323         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1324         nid_t nid;
1325         struct node_info ni;
1326         struct f2fs_io_info fio = {
1327                 .sbi = sbi,
1328                 .type = NODE,
1329                 .op = REQ_OP_WRITE,
1330                 .op_flags = wbc_to_write_flags(wbc),
1331                 .page = page,
1332                 .encrypted_page = NULL,
1333                 .submitted = false,
1334         };
1335
1336         trace_f2fs_writepage(page, NODE);
1337
1338         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1339                 goto redirty_out;
1340         if (unlikely(f2fs_cp_error(sbi)))
1341                 goto redirty_out;
1342
1343         /* get old block addr of this node page */
1344         nid = nid_of_node(page);
1345         f2fs_bug_on(sbi, page->index != nid);
1346
1347         if (wbc->for_reclaim) {
1348                 if (!down_read_trylock(&sbi->node_write))
1349                         goto redirty_out;
1350         } else {
1351                 down_read(&sbi->node_write);
1352         }
1353
1354         get_node_info(sbi, nid, &ni);
1355
1356         /* This page is already truncated */
1357         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1358                 ClearPageUptodate(page);
1359                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1360                 up_read(&sbi->node_write);
1361                 unlock_page(page);
1362                 return 0;
1363         }
1364
1365         if (atomic && !test_opt(sbi, NOBARRIER))
1366                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1367
1368         set_page_writeback(page);
1369         fio.old_blkaddr = ni.blk_addr;
1370         write_node_page(nid, &fio);
1371         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1372         dec_page_count(sbi, F2FS_DIRTY_NODES);
1373         up_read(&sbi->node_write);
1374
1375         if (wbc->for_reclaim) {
1376                 f2fs_submit_merged_bio_cond(sbi, page->mapping->host, 0,
1377                                                 page->index, NODE, WRITE);
1378                 submitted = NULL;
1379         }
1380
1381         unlock_page(page);
1382
1383         if (unlikely(f2fs_cp_error(sbi))) {
1384                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1385                 submitted = NULL;
1386         }
1387         if (submitted)
1388                 *submitted = fio.submitted;
1389
1390         return 0;
1391
1392 redirty_out:
1393         redirty_page_for_writepage(wbc, page);
1394         return AOP_WRITEPAGE_ACTIVATE;
1395 }
1396
1397 static int f2fs_write_node_page(struct page *page,
1398                                 struct writeback_control *wbc)
1399 {
1400         return __write_node_page(page, false, NULL, wbc);
1401 }
1402
1403 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1404                         struct writeback_control *wbc, bool atomic)
1405 {
1406         pgoff_t index, end;
1407         pgoff_t last_idx = ULONG_MAX;
1408         struct pagevec pvec;
1409         int ret = 0;
1410         struct page *last_page = NULL;
1411         bool marked = false;
1412         nid_t ino = inode->i_ino;
1413
1414         if (atomic) {
1415                 last_page = last_fsync_dnode(sbi, ino);
1416                 if (IS_ERR_OR_NULL(last_page))
1417                         return PTR_ERR_OR_ZERO(last_page);
1418         }
1419 retry:
1420         pagevec_init(&pvec, 0);
1421         index = 0;
1422         end = ULONG_MAX;
1423
1424         while (index <= end) {
1425                 int i, nr_pages;
1426                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1427                                 PAGECACHE_TAG_DIRTY,
1428                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1429                 if (nr_pages == 0)
1430                         break;
1431
1432                 for (i = 0; i < nr_pages; i++) {
1433                         struct page *page = pvec.pages[i];
1434                         bool submitted = false;
1435
1436                         if (unlikely(f2fs_cp_error(sbi))) {
1437                                 f2fs_put_page(last_page, 0);
1438                                 pagevec_release(&pvec);
1439                                 ret = -EIO;
1440                                 goto out;
1441                         }
1442
1443                         if (!IS_DNODE(page) || !is_cold_node(page))
1444                                 continue;
1445                         if (ino_of_node(page) != ino)
1446                                 continue;
1447
1448                         lock_page(page);
1449
1450                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1451 continue_unlock:
1452                                 unlock_page(page);
1453                                 continue;
1454                         }
1455                         if (ino_of_node(page) != ino)
1456                                 goto continue_unlock;
1457
1458                         if (!PageDirty(page) && page != last_page) {
1459                                 /* someone wrote it for us */
1460                                 goto continue_unlock;
1461                         }
1462
1463                         f2fs_wait_on_page_writeback(page, NODE, true);
1464                         BUG_ON(PageWriteback(page));
1465
1466                         set_fsync_mark(page, 0);
1467                         set_dentry_mark(page, 0);
1468
1469                         if (!atomic || page == last_page) {
1470                                 set_fsync_mark(page, 1);
1471                                 if (IS_INODE(page)) {
1472                                         if (is_inode_flag_set(inode,
1473                                                                 FI_DIRTY_INODE))
1474                                                 update_inode(inode, page);
1475                                         set_dentry_mark(page,
1476                                                 need_dentry_mark(sbi, ino));
1477                                 }
1478                                 /*  may be written by other thread */
1479                                 if (!PageDirty(page))
1480                                         set_page_dirty(page);
1481                         }
1482
1483                         if (!clear_page_dirty_for_io(page))
1484                                 goto continue_unlock;
1485
1486                         ret = __write_node_page(page, atomic &&
1487                                                 page == last_page,
1488                                                 &submitted, wbc);
1489                         if (ret) {
1490                                 unlock_page(page);
1491                                 f2fs_put_page(last_page, 0);
1492                                 break;
1493                         } else if (submitted) {
1494                                 last_idx = page->index;
1495                         }
1496
1497                         if (page == last_page) {
1498                                 f2fs_put_page(page, 0);
1499                                 marked = true;
1500                                 break;
1501                         }
1502                 }
1503                 pagevec_release(&pvec);
1504                 cond_resched();
1505
1506                 if (ret || marked)
1507                         break;
1508         }
1509         if (!ret && atomic && !marked) {
1510                 f2fs_msg(sbi->sb, KERN_DEBUG,
1511                         "Retry to write fsync mark: ino=%u, idx=%lx",
1512                                         ino, last_page->index);
1513                 lock_page(last_page);
1514                 f2fs_wait_on_page_writeback(last_page, NODE, true);
1515                 set_page_dirty(last_page);
1516                 unlock_page(last_page);
1517                 goto retry;
1518         }
1519 out:
1520         if (last_idx != ULONG_MAX)
1521                 f2fs_submit_merged_bio_cond(sbi, NULL, ino, last_idx,
1522                                                         NODE, WRITE);
1523         return ret ? -EIO: 0;
1524 }
1525
1526 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1527 {
1528         pgoff_t index, end;
1529         struct pagevec pvec;
1530         int step = 0;
1531         int nwritten = 0;
1532         int ret = 0;
1533
1534         pagevec_init(&pvec, 0);
1535
1536 next_step:
1537         index = 0;
1538         end = ULONG_MAX;
1539
1540         while (index <= end) {
1541                 int i, nr_pages;
1542                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1543                                 PAGECACHE_TAG_DIRTY,
1544                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1545                 if (nr_pages == 0)
1546                         break;
1547
1548                 for (i = 0; i < nr_pages; i++) {
1549                         struct page *page = pvec.pages[i];
1550                         bool submitted = false;
1551
1552                         if (unlikely(f2fs_cp_error(sbi))) {
1553                                 pagevec_release(&pvec);
1554                                 ret = -EIO;
1555                                 goto out;
1556                         }
1557
1558                         /*
1559                          * flushing sequence with step:
1560                          * 0. indirect nodes
1561                          * 1. dentry dnodes
1562                          * 2. file dnodes
1563                          */
1564                         if (step == 0 && IS_DNODE(page))
1565                                 continue;
1566                         if (step == 1 && (!IS_DNODE(page) ||
1567                                                 is_cold_node(page)))
1568                                 continue;
1569                         if (step == 2 && (!IS_DNODE(page) ||
1570                                                 !is_cold_node(page)))
1571                                 continue;
1572 lock_node:
1573                         if (!trylock_page(page))
1574                                 continue;
1575
1576                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1577 continue_unlock:
1578                                 unlock_page(page);
1579                                 continue;
1580                         }
1581
1582                         if (!PageDirty(page)) {
1583                                 /* someone wrote it for us */
1584                                 goto continue_unlock;
1585                         }
1586
1587                         /* flush inline_data */
1588                         if (is_inline_node(page)) {
1589                                 clear_inline_node(page);
1590                                 unlock_page(page);
1591                                 flush_inline_data(sbi, ino_of_node(page));
1592                                 goto lock_node;
1593                         }
1594
1595                         f2fs_wait_on_page_writeback(page, NODE, true);
1596
1597                         BUG_ON(PageWriteback(page));
1598                         if (!clear_page_dirty_for_io(page))
1599                                 goto continue_unlock;
1600
1601                         set_fsync_mark(page, 0);
1602                         set_dentry_mark(page, 0);
1603
1604                         ret = __write_node_page(page, false, &submitted, wbc);
1605                         if (ret)
1606                                 unlock_page(page);
1607                         else if (submitted)
1608                                 nwritten++;
1609
1610                         if (--wbc->nr_to_write == 0)
1611                                 break;
1612                 }
1613                 pagevec_release(&pvec);
1614                 cond_resched();
1615
1616                 if (wbc->nr_to_write == 0) {
1617                         step = 2;
1618                         break;
1619                 }
1620         }
1621
1622         if (step < 2) {
1623                 step++;
1624                 goto next_step;
1625         }
1626 out:
1627         if (nwritten)
1628                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1629         return ret;
1630 }
1631
1632 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1633 {
1634         pgoff_t index = 0, end = ULONG_MAX;
1635         struct pagevec pvec;
1636         int ret2, ret = 0;
1637
1638         pagevec_init(&pvec, 0);
1639
1640         while (index <= end) {
1641                 int i, nr_pages;
1642                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1643                                 PAGECACHE_TAG_WRITEBACK,
1644                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1645                 if (nr_pages == 0)
1646                         break;
1647
1648                 for (i = 0; i < nr_pages; i++) {
1649                         struct page *page = pvec.pages[i];
1650
1651                         /* until radix tree lookup accepts end_index */
1652                         if (unlikely(page->index > end))
1653                                 continue;
1654
1655                         if (ino && ino_of_node(page) == ino) {
1656                                 f2fs_wait_on_page_writeback(page, NODE, true);
1657                                 if (TestClearPageError(page))
1658                                         ret = -EIO;
1659                         }
1660                 }
1661                 pagevec_release(&pvec);
1662                 cond_resched();
1663         }
1664
1665         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1666         if (!ret)
1667                 ret = ret2;
1668         return ret;
1669 }
1670
1671 static int f2fs_write_node_pages(struct address_space *mapping,
1672                             struct writeback_control *wbc)
1673 {
1674         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1675         struct blk_plug plug;
1676         long diff;
1677
1678         /* balancing f2fs's metadata in background */
1679         f2fs_balance_fs_bg(sbi);
1680
1681         /* collect a number of dirty node pages and write together */
1682         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1683                 goto skip_write;
1684
1685         trace_f2fs_writepages(mapping->host, wbc, NODE);
1686
1687         diff = nr_pages_to_write(sbi, NODE, wbc);
1688         wbc->sync_mode = WB_SYNC_NONE;
1689         blk_start_plug(&plug);
1690         sync_node_pages(sbi, wbc);
1691         blk_finish_plug(&plug);
1692         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1693         return 0;
1694
1695 skip_write:
1696         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1697         trace_f2fs_writepages(mapping->host, wbc, NODE);
1698         return 0;
1699 }
1700
1701 static int f2fs_set_node_page_dirty(struct page *page)
1702 {
1703         trace_f2fs_set_page_dirty(page, NODE);
1704
1705         if (!PageUptodate(page))
1706                 SetPageUptodate(page);
1707         if (!PageDirty(page)) {
1708                 f2fs_set_page_dirty_nobuffers(page);
1709                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1710                 SetPagePrivate(page);
1711                 f2fs_trace_pid(page);
1712                 return 1;
1713         }
1714         return 0;
1715 }
1716
1717 /*
1718  * Structure of the f2fs node operations
1719  */
1720 const struct address_space_operations f2fs_node_aops = {
1721         .writepage      = f2fs_write_node_page,
1722         .writepages     = f2fs_write_node_pages,
1723         .set_page_dirty = f2fs_set_node_page_dirty,
1724         .invalidatepage = f2fs_invalidate_page,
1725         .releasepage    = f2fs_release_page,
1726 #ifdef CONFIG_MIGRATION
1727         .migratepage    = f2fs_migrate_page,
1728 #endif
1729 };
1730
1731 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1732                                                 nid_t n)
1733 {
1734         return radix_tree_lookup(&nm_i->free_nid_root, n);
1735 }
1736
1737 static int __insert_nid_to_list(struct f2fs_sb_info *sbi,
1738                         struct free_nid *i, enum nid_list list, bool new)
1739 {
1740         struct f2fs_nm_info *nm_i = NM_I(sbi);
1741
1742         if (new) {
1743                 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1744                 if (err)
1745                         return err;
1746         }
1747
1748         f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1749                                                 i->state != NID_ALLOC);
1750         nm_i->nid_cnt[list]++;
1751         list_add_tail(&i->list, &nm_i->nid_list[list]);
1752         return 0;
1753 }
1754
1755 static void __remove_nid_from_list(struct f2fs_sb_info *sbi,
1756                         struct free_nid *i, enum nid_list list, bool reuse)
1757 {
1758         struct f2fs_nm_info *nm_i = NM_I(sbi);
1759
1760         f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1761                                                 i->state != NID_ALLOC);
1762         nm_i->nid_cnt[list]--;
1763         list_del(&i->list);
1764         if (!reuse)
1765                 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1766 }
1767
1768 /* return if the nid is recognized as free */
1769 static bool add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1770 {
1771         struct f2fs_nm_info *nm_i = NM_I(sbi);
1772         struct free_nid *i, *e;
1773         struct nat_entry *ne;
1774         int err = -EINVAL;
1775         bool ret = false;
1776
1777         /* 0 nid should not be used */
1778         if (unlikely(nid == 0))
1779                 return false;
1780
1781         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1782         i->nid = nid;
1783         i->state = NID_NEW;
1784
1785         if (radix_tree_preload(GFP_NOFS))
1786                 goto err;
1787
1788         spin_lock(&nm_i->nid_list_lock);
1789
1790         if (build) {
1791                 /*
1792                  *   Thread A             Thread B
1793                  *  - f2fs_create
1794                  *   - f2fs_new_inode
1795                  *    - alloc_nid
1796                  *     - __insert_nid_to_list(ALLOC_NID_LIST)
1797                  *                     - f2fs_balance_fs_bg
1798                  *                      - build_free_nids
1799                  *                       - __build_free_nids
1800                  *                        - scan_nat_page
1801                  *                         - add_free_nid
1802                  *                          - __lookup_nat_cache
1803                  *  - f2fs_add_link
1804                  *   - init_inode_metadata
1805                  *    - new_inode_page
1806                  *     - new_node_page
1807                  *      - set_node_addr
1808                  *  - alloc_nid_done
1809                  *   - __remove_nid_from_list(ALLOC_NID_LIST)
1810                  *                         - __insert_nid_to_list(FREE_NID_LIST)
1811                  */
1812                 ne = __lookup_nat_cache(nm_i, nid);
1813                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1814                                 nat_get_blkaddr(ne) != NULL_ADDR))
1815                         goto err_out;
1816
1817                 e = __lookup_free_nid_list(nm_i, nid);
1818                 if (e) {
1819                         if (e->state == NID_NEW)
1820                                 ret = true;
1821                         goto err_out;
1822                 }
1823         }
1824         ret = true;
1825         err = __insert_nid_to_list(sbi, i, FREE_NID_LIST, true);
1826 err_out:
1827         spin_unlock(&nm_i->nid_list_lock);
1828         radix_tree_preload_end();
1829 err:
1830         if (err)
1831                 kmem_cache_free(free_nid_slab, i);
1832         return ret;
1833 }
1834
1835 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1836 {
1837         struct f2fs_nm_info *nm_i = NM_I(sbi);
1838         struct free_nid *i;
1839         bool need_free = false;
1840
1841         spin_lock(&nm_i->nid_list_lock);
1842         i = __lookup_free_nid_list(nm_i, nid);
1843         if (i && i->state == NID_NEW) {
1844                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
1845                 need_free = true;
1846         }
1847         spin_unlock(&nm_i->nid_list_lock);
1848
1849         if (need_free)
1850                 kmem_cache_free(free_nid_slab, i);
1851 }
1852
1853 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
1854                                                         bool set, bool build)
1855 {
1856         struct f2fs_nm_info *nm_i = NM_I(sbi);
1857         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
1858         unsigned int nid_ofs = nid - START_NID(nid);
1859
1860         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1861                 return;
1862
1863         if (set)
1864                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1865         else
1866                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1867
1868         if (set)
1869                 nm_i->free_nid_count[nat_ofs]++;
1870         else if (!build)
1871                 nm_i->free_nid_count[nat_ofs]--;
1872 }
1873
1874 static void scan_nat_page(struct f2fs_sb_info *sbi,
1875                         struct page *nat_page, nid_t start_nid)
1876 {
1877         struct f2fs_nm_info *nm_i = NM_I(sbi);
1878         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1879         block_t blk_addr;
1880         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
1881         int i;
1882
1883         if (test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1884                 return;
1885
1886         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
1887
1888         i = start_nid % NAT_ENTRY_PER_BLOCK;
1889
1890         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1891                 bool freed = false;
1892
1893                 if (unlikely(start_nid >= nm_i->max_nid))
1894                         break;
1895
1896                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1897                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1898                 if (blk_addr == NULL_ADDR)
1899                         freed = add_free_nid(sbi, start_nid, true);
1900                 spin_lock(&NM_I(sbi)->nid_list_lock);
1901                 update_free_nid_bitmap(sbi, start_nid, freed, true);
1902                 spin_unlock(&NM_I(sbi)->nid_list_lock);
1903         }
1904 }
1905
1906 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
1907 {
1908         struct f2fs_nm_info *nm_i = NM_I(sbi);
1909         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1910         struct f2fs_journal *journal = curseg->journal;
1911         unsigned int i, idx;
1912
1913         down_read(&nm_i->nat_tree_lock);
1914
1915         for (i = 0; i < nm_i->nat_blocks; i++) {
1916                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
1917                         continue;
1918                 if (!nm_i->free_nid_count[i])
1919                         continue;
1920                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
1921                         nid_t nid;
1922
1923                         if (!test_bit_le(idx, nm_i->free_nid_bitmap[i]))
1924                                 continue;
1925
1926                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
1927                         add_free_nid(sbi, nid, true);
1928
1929                         if (nm_i->nid_cnt[FREE_NID_LIST] >= MAX_FREE_NIDS)
1930                                 goto out;
1931                 }
1932         }
1933 out:
1934         down_read(&curseg->journal_rwsem);
1935         for (i = 0; i < nats_in_cursum(journal); i++) {
1936                 block_t addr;
1937                 nid_t nid;
1938
1939                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1940                 nid = le32_to_cpu(nid_in_journal(journal, i));
1941                 if (addr == NULL_ADDR)
1942                         add_free_nid(sbi, nid, true);
1943                 else
1944                         remove_free_nid(sbi, nid);
1945         }
1946         up_read(&curseg->journal_rwsem);
1947         up_read(&nm_i->nat_tree_lock);
1948 }
1949
1950 static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
1951 {
1952         struct f2fs_nm_info *nm_i = NM_I(sbi);
1953         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1954         struct f2fs_journal *journal = curseg->journal;
1955         int i = 0;
1956         nid_t nid = nm_i->next_scan_nid;
1957
1958         if (unlikely(nid >= nm_i->max_nid))
1959                 nid = 0;
1960
1961         /* Enough entries */
1962         if (nm_i->nid_cnt[FREE_NID_LIST] >= NAT_ENTRY_PER_BLOCK)
1963                 return;
1964
1965         if (!sync && !available_free_memory(sbi, FREE_NIDS))
1966                 return;
1967
1968         if (!mount) {
1969                 /* try to find free nids in free_nid_bitmap */
1970                 scan_free_nid_bits(sbi);
1971
1972                 if (nm_i->nid_cnt[FREE_NID_LIST])
1973                         return;
1974         }
1975
1976         /* readahead nat pages to be scanned */
1977         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1978                                                         META_NAT, true);
1979
1980         down_read(&nm_i->nat_tree_lock);
1981
1982         while (1) {
1983                 struct page *page = get_current_nat_page(sbi, nid);
1984
1985                 scan_nat_page(sbi, page, nid);
1986                 f2fs_put_page(page, 1);
1987
1988                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1989                 if (unlikely(nid >= nm_i->max_nid))
1990                         nid = 0;
1991
1992                 if (++i >= FREE_NID_PAGES)
1993                         break;
1994         }
1995
1996         /* go to the next free nat pages to find free nids abundantly */
1997         nm_i->next_scan_nid = nid;
1998
1999         /* find free nids from current sum_pages */
2000         down_read(&curseg->journal_rwsem);
2001         for (i = 0; i < nats_in_cursum(journal); i++) {
2002                 block_t addr;
2003
2004                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2005                 nid = le32_to_cpu(nid_in_journal(journal, i));
2006                 if (addr == NULL_ADDR)
2007                         add_free_nid(sbi, nid, true);
2008                 else
2009                         remove_free_nid(sbi, nid);
2010         }
2011         up_read(&curseg->journal_rwsem);
2012         up_read(&nm_i->nat_tree_lock);
2013
2014         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2015                                         nm_i->ra_nid_pages, META_NAT, false);
2016 }
2017
2018 void build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2019 {
2020         mutex_lock(&NM_I(sbi)->build_lock);
2021         __build_free_nids(sbi, sync, mount);
2022         mutex_unlock(&NM_I(sbi)->build_lock);
2023 }
2024
2025 /*
2026  * If this function returns success, caller can obtain a new nid
2027  * from second parameter of this function.
2028  * The returned nid could be used ino as well as nid when inode is created.
2029  */
2030 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2031 {
2032         struct f2fs_nm_info *nm_i = NM_I(sbi);
2033         struct free_nid *i = NULL;
2034 retry:
2035 #ifdef CONFIG_F2FS_FAULT_INJECTION
2036         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2037                 f2fs_show_injection_info(FAULT_ALLOC_NID);
2038                 return false;
2039         }
2040 #endif
2041         spin_lock(&nm_i->nid_list_lock);
2042
2043         if (unlikely(nm_i->available_nids == 0)) {
2044                 spin_unlock(&nm_i->nid_list_lock);
2045                 return false;
2046         }
2047
2048         /* We should not use stale free nids created by build_free_nids */
2049         if (nm_i->nid_cnt[FREE_NID_LIST] && !on_build_free_nids(nm_i)) {
2050                 f2fs_bug_on(sbi, list_empty(&nm_i->nid_list[FREE_NID_LIST]));
2051                 i = list_first_entry(&nm_i->nid_list[FREE_NID_LIST],
2052                                         struct free_nid, list);
2053                 *nid = i->nid;
2054
2055                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, true);
2056                 i->state = NID_ALLOC;
2057                 __insert_nid_to_list(sbi, i, ALLOC_NID_LIST, false);
2058                 nm_i->available_nids--;
2059
2060                 update_free_nid_bitmap(sbi, *nid, false, false);
2061
2062                 spin_unlock(&nm_i->nid_list_lock);
2063                 return true;
2064         }
2065         spin_unlock(&nm_i->nid_list_lock);
2066
2067         /* Let's scan nat pages and its caches to get free nids */
2068         build_free_nids(sbi, true, false);
2069         goto retry;
2070 }
2071
2072 /*
2073  * alloc_nid() should be called prior to this function.
2074  */
2075 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2076 {
2077         struct f2fs_nm_info *nm_i = NM_I(sbi);
2078         struct free_nid *i;
2079
2080         spin_lock(&nm_i->nid_list_lock);
2081         i = __lookup_free_nid_list(nm_i, nid);
2082         f2fs_bug_on(sbi, !i);
2083         __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
2084         spin_unlock(&nm_i->nid_list_lock);
2085
2086         kmem_cache_free(free_nid_slab, i);
2087 }
2088
2089 /*
2090  * alloc_nid() should be called prior to this function.
2091  */
2092 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2093 {
2094         struct f2fs_nm_info *nm_i = NM_I(sbi);
2095         struct free_nid *i;
2096         bool need_free = false;
2097
2098         if (!nid)
2099                 return;
2100
2101         spin_lock(&nm_i->nid_list_lock);
2102         i = __lookup_free_nid_list(nm_i, nid);
2103         f2fs_bug_on(sbi, !i);
2104
2105         if (!available_free_memory(sbi, FREE_NIDS)) {
2106                 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
2107                 need_free = true;
2108         } else {
2109                 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, true);
2110                 i->state = NID_NEW;
2111                 __insert_nid_to_list(sbi, i, FREE_NID_LIST, false);
2112         }
2113
2114         nm_i->available_nids++;
2115
2116         update_free_nid_bitmap(sbi, nid, true, false);
2117
2118         spin_unlock(&nm_i->nid_list_lock);
2119
2120         if (need_free)
2121                 kmem_cache_free(free_nid_slab, i);
2122 }
2123
2124 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2125 {
2126         struct f2fs_nm_info *nm_i = NM_I(sbi);
2127         struct free_nid *i, *next;
2128         int nr = nr_shrink;
2129
2130         if (nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
2131                 return 0;
2132
2133         if (!mutex_trylock(&nm_i->build_lock))
2134                 return 0;
2135
2136         spin_lock(&nm_i->nid_list_lock);
2137         list_for_each_entry_safe(i, next, &nm_i->nid_list[FREE_NID_LIST],
2138                                                                         list) {
2139                 if (nr_shrink <= 0 ||
2140                                 nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
2141                         break;
2142
2143                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2144                 kmem_cache_free(free_nid_slab, i);
2145                 nr_shrink--;
2146         }
2147         spin_unlock(&nm_i->nid_list_lock);
2148         mutex_unlock(&nm_i->build_lock);
2149
2150         return nr - nr_shrink;
2151 }
2152
2153 void recover_inline_xattr(struct inode *inode, struct page *page)
2154 {
2155         void *src_addr, *dst_addr;
2156         size_t inline_size;
2157         struct page *ipage;
2158         struct f2fs_inode *ri;
2159
2160         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2161         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2162
2163         ri = F2FS_INODE(page);
2164         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2165                 clear_inode_flag(inode, FI_INLINE_XATTR);
2166                 goto update_inode;
2167         }
2168
2169         dst_addr = inline_xattr_addr(ipage);
2170         src_addr = inline_xattr_addr(page);
2171         inline_size = inline_xattr_size(inode);
2172
2173         f2fs_wait_on_page_writeback(ipage, NODE, true);
2174         memcpy(dst_addr, src_addr, inline_size);
2175 update_inode:
2176         update_inode(inode, ipage);
2177         f2fs_put_page(ipage, 1);
2178 }
2179
2180 int recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2181 {
2182         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2183         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2184         nid_t new_xnid = nid_of_node(page);
2185         struct node_info ni;
2186         struct page *xpage;
2187
2188         if (!prev_xnid)
2189                 goto recover_xnid;
2190
2191         /* 1: invalidate the previous xattr nid */
2192         get_node_info(sbi, prev_xnid, &ni);
2193         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2194         invalidate_blocks(sbi, ni.blk_addr);
2195         dec_valid_node_count(sbi, inode);
2196         set_node_addr(sbi, &ni, NULL_ADDR, false);
2197
2198 recover_xnid:
2199         /* 2: update xattr nid in inode */
2200         remove_free_nid(sbi, new_xnid);
2201         f2fs_i_xnid_write(inode, new_xnid);
2202         if (unlikely(!inc_valid_node_count(sbi, inode)))
2203                 f2fs_bug_on(sbi, 1);
2204         update_inode_page(inode);
2205
2206         /* 3: update and set xattr node page dirty */
2207         xpage = grab_cache_page(NODE_MAPPING(sbi), new_xnid);
2208         if (!xpage)
2209                 return -ENOMEM;
2210
2211         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), PAGE_SIZE);
2212
2213         get_node_info(sbi, new_xnid, &ni);
2214         ni.ino = inode->i_ino;
2215         set_node_addr(sbi, &ni, NEW_ADDR, false);
2216         set_page_dirty(xpage);
2217         f2fs_put_page(xpage, 1);
2218
2219         return 0;
2220 }
2221
2222 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2223 {
2224         struct f2fs_inode *src, *dst;
2225         nid_t ino = ino_of_node(page);
2226         struct node_info old_ni, new_ni;
2227         struct page *ipage;
2228
2229         get_node_info(sbi, ino, &old_ni);
2230
2231         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2232                 return -EINVAL;
2233 retry:
2234         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2235         if (!ipage) {
2236                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2237                 goto retry;
2238         }
2239
2240         /* Should not use this inode from free nid list */
2241         remove_free_nid(sbi, ino);
2242
2243         if (!PageUptodate(ipage))
2244                 SetPageUptodate(ipage);
2245         fill_node_footer(ipage, ino, ino, 0, true);
2246
2247         src = F2FS_INODE(page);
2248         dst = F2FS_INODE(ipage);
2249
2250         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2251         dst->i_size = 0;
2252         dst->i_blocks = cpu_to_le64(1);
2253         dst->i_links = cpu_to_le32(1);
2254         dst->i_xattr_nid = 0;
2255         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2256
2257         new_ni = old_ni;
2258         new_ni.ino = ino;
2259
2260         if (unlikely(!inc_valid_node_count(sbi, NULL)))
2261                 WARN_ON(1);
2262         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2263         inc_valid_inode_count(sbi);
2264         set_page_dirty(ipage);
2265         f2fs_put_page(ipage, 1);
2266         return 0;
2267 }
2268
2269 int restore_node_summary(struct f2fs_sb_info *sbi,
2270                         unsigned int segno, struct f2fs_summary_block *sum)
2271 {
2272         struct f2fs_node *rn;
2273         struct f2fs_summary *sum_entry;
2274         block_t addr;
2275         int i, idx, last_offset, nrpages;
2276
2277         /* scan the node segment */
2278         last_offset = sbi->blocks_per_seg;
2279         addr = START_BLOCK(sbi, segno);
2280         sum_entry = &sum->entries[0];
2281
2282         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2283                 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2284
2285                 /* readahead node pages */
2286                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2287
2288                 for (idx = addr; idx < addr + nrpages; idx++) {
2289                         struct page *page = get_tmp_page(sbi, idx);
2290
2291                         rn = F2FS_NODE(page);
2292                         sum_entry->nid = rn->footer.nid;
2293                         sum_entry->version = 0;
2294                         sum_entry->ofs_in_node = 0;
2295                         sum_entry++;
2296                         f2fs_put_page(page, 1);
2297                 }
2298
2299                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2300                                                         addr + nrpages);
2301         }
2302         return 0;
2303 }
2304
2305 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2306 {
2307         struct f2fs_nm_info *nm_i = NM_I(sbi);
2308         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2309         struct f2fs_journal *journal = curseg->journal;
2310         int i;
2311
2312         down_write(&curseg->journal_rwsem);
2313         for (i = 0; i < nats_in_cursum(journal); i++) {
2314                 struct nat_entry *ne;
2315                 struct f2fs_nat_entry raw_ne;
2316                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2317
2318                 raw_ne = nat_in_journal(journal, i);
2319
2320                 ne = __lookup_nat_cache(nm_i, nid);
2321                 if (!ne) {
2322                         ne = grab_nat_entry(nm_i, nid, true);
2323                         node_info_from_raw_nat(&ne->ni, &raw_ne);
2324                 }
2325
2326                 /*
2327                  * if a free nat in journal has not been used after last
2328                  * checkpoint, we should remove it from available nids,
2329                  * since later we will add it again.
2330                  */
2331                 if (!get_nat_flag(ne, IS_DIRTY) &&
2332                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2333                         spin_lock(&nm_i->nid_list_lock);
2334                         nm_i->available_nids--;
2335                         spin_unlock(&nm_i->nid_list_lock);
2336                 }
2337
2338                 __set_nat_cache_dirty(nm_i, ne);
2339         }
2340         update_nats_in_cursum(journal, -i);
2341         up_write(&curseg->journal_rwsem);
2342 }
2343
2344 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2345                                                 struct list_head *head, int max)
2346 {
2347         struct nat_entry_set *cur;
2348
2349         if (nes->entry_cnt >= max)
2350                 goto add_out;
2351
2352         list_for_each_entry(cur, head, set_list) {
2353                 if (cur->entry_cnt >= nes->entry_cnt) {
2354                         list_add(&nes->set_list, cur->set_list.prev);
2355                         return;
2356                 }
2357         }
2358 add_out:
2359         list_add_tail(&nes->set_list, head);
2360 }
2361
2362 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2363                                                 struct page *page)
2364 {
2365         struct f2fs_nm_info *nm_i = NM_I(sbi);
2366         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2367         struct f2fs_nat_block *nat_blk = page_address(page);
2368         int valid = 0;
2369         int i;
2370
2371         if (!enabled_nat_bits(sbi, NULL))
2372                 return;
2373
2374         for (i = 0; i < NAT_ENTRY_PER_BLOCK; i++) {
2375                 if (start_nid == 0 && i == 0)
2376                         valid++;
2377                 if (nat_blk->entries[i].block_addr)
2378                         valid++;
2379         }
2380         if (valid == 0) {
2381                 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2382                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2383                 return;
2384         }
2385
2386         __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2387         if (valid == NAT_ENTRY_PER_BLOCK)
2388                 __set_bit_le(nat_index, nm_i->full_nat_bits);
2389         else
2390                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2391 }
2392
2393 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2394                 struct nat_entry_set *set, struct cp_control *cpc)
2395 {
2396         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2397         struct f2fs_journal *journal = curseg->journal;
2398         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2399         bool to_journal = true;
2400         struct f2fs_nat_block *nat_blk;
2401         struct nat_entry *ne, *cur;
2402         struct page *page = NULL;
2403
2404         /*
2405          * there are two steps to flush nat entries:
2406          * #1, flush nat entries to journal in current hot data summary block.
2407          * #2, flush nat entries to nat page.
2408          */
2409         if (enabled_nat_bits(sbi, cpc) ||
2410                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2411                 to_journal = false;
2412
2413         if (to_journal) {
2414                 down_write(&curseg->journal_rwsem);
2415         } else {
2416                 page = get_next_nat_page(sbi, start_nid);
2417                 nat_blk = page_address(page);
2418                 f2fs_bug_on(sbi, !nat_blk);
2419         }
2420
2421         /* flush dirty nats in nat entry set */
2422         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2423                 struct f2fs_nat_entry *raw_ne;
2424                 nid_t nid = nat_get_nid(ne);
2425                 int offset;
2426
2427                 if (nat_get_blkaddr(ne) == NEW_ADDR)
2428                         continue;
2429
2430                 if (to_journal) {
2431                         offset = lookup_journal_in_cursum(journal,
2432                                                         NAT_JOURNAL, nid, 1);
2433                         f2fs_bug_on(sbi, offset < 0);
2434                         raw_ne = &nat_in_journal(journal, offset);
2435                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2436                 } else {
2437                         raw_ne = &nat_blk->entries[nid - start_nid];
2438                 }
2439                 raw_nat_from_node_info(raw_ne, &ne->ni);
2440                 nat_reset_flag(ne);
2441                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2442                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2443                         add_free_nid(sbi, nid, false);
2444                         spin_lock(&NM_I(sbi)->nid_list_lock);
2445                         NM_I(sbi)->available_nids++;
2446                         update_free_nid_bitmap(sbi, nid, true, false);
2447                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2448                 } else {
2449                         spin_lock(&NM_I(sbi)->nid_list_lock);
2450                         update_free_nid_bitmap(sbi, nid, false, false);
2451                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2452                 }
2453         }
2454
2455         if (to_journal) {
2456                 up_write(&curseg->journal_rwsem);
2457         } else {
2458                 __update_nat_bits(sbi, start_nid, page);
2459                 f2fs_put_page(page, 1);
2460         }
2461
2462         /* Allow dirty nats by node block allocation in write_begin */
2463         if (!set->entry_cnt) {
2464                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2465                 kmem_cache_free(nat_entry_set_slab, set);
2466         }
2467 }
2468
2469 /*
2470  * This function is called during the checkpointing process.
2471  */
2472 void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2473 {
2474         struct f2fs_nm_info *nm_i = NM_I(sbi);
2475         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2476         struct f2fs_journal *journal = curseg->journal;
2477         struct nat_entry_set *setvec[SETVEC_SIZE];
2478         struct nat_entry_set *set, *tmp;
2479         unsigned int found;
2480         nid_t set_idx = 0;
2481         LIST_HEAD(sets);
2482
2483         if (!nm_i->dirty_nat_cnt)
2484                 return;
2485
2486         down_write(&nm_i->nat_tree_lock);
2487
2488         /*
2489          * if there are no enough space in journal to store dirty nat
2490          * entries, remove all entries from journal and merge them
2491          * into nat entry set.
2492          */
2493         if (enabled_nat_bits(sbi, cpc) ||
2494                 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2495                 remove_nats_in_journal(sbi);
2496
2497         while ((found = __gang_lookup_nat_set(nm_i,
2498                                         set_idx, SETVEC_SIZE, setvec))) {
2499                 unsigned idx;
2500                 set_idx = setvec[found - 1]->set + 1;
2501                 for (idx = 0; idx < found; idx++)
2502                         __adjust_nat_entry_set(setvec[idx], &sets,
2503                                                 MAX_NAT_JENTRIES(journal));
2504         }
2505
2506         /* flush dirty nats in nat entry set */
2507         list_for_each_entry_safe(set, tmp, &sets, set_list)
2508                 __flush_nat_entry_set(sbi, set, cpc);
2509
2510         up_write(&nm_i->nat_tree_lock);
2511         /* Allow dirty nats by node block allocation in write_begin */
2512 }
2513
2514 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2515 {
2516         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2517         struct f2fs_nm_info *nm_i = NM_I(sbi);
2518         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2519         unsigned int i;
2520         __u64 cp_ver = cur_cp_version(ckpt);
2521         block_t nat_bits_addr;
2522
2523         if (!enabled_nat_bits(sbi, NULL))
2524                 return 0;
2525
2526         nm_i->nat_bits_blocks = F2FS_BYTES_TO_BLK((nat_bits_bytes << 1) + 8 +
2527                                                 F2FS_BLKSIZE - 1);
2528         nm_i->nat_bits = kzalloc(nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS,
2529                                                 GFP_KERNEL);
2530         if (!nm_i->nat_bits)
2531                 return -ENOMEM;
2532
2533         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2534                                                 nm_i->nat_bits_blocks;
2535         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2536                 struct page *page = get_meta_page(sbi, nat_bits_addr++);
2537
2538                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2539                                         page_address(page), F2FS_BLKSIZE);
2540                 f2fs_put_page(page, 1);
2541         }
2542
2543         cp_ver |= (cur_cp_crc(ckpt) << 32);
2544         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2545                 disable_nat_bits(sbi, true);
2546                 return 0;
2547         }
2548
2549         nm_i->full_nat_bits = nm_i->nat_bits + 8;
2550         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2551
2552         f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2553         return 0;
2554 }
2555
2556 inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2557 {
2558         struct f2fs_nm_info *nm_i = NM_I(sbi);
2559         unsigned int i = 0;
2560         nid_t nid, last_nid;
2561
2562         if (!enabled_nat_bits(sbi, NULL))
2563                 return;
2564
2565         for (i = 0; i < nm_i->nat_blocks; i++) {
2566                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2567                 if (i >= nm_i->nat_blocks)
2568                         break;
2569
2570                 __set_bit_le(i, nm_i->nat_block_bitmap);
2571
2572                 nid = i * NAT_ENTRY_PER_BLOCK;
2573                 last_nid = (i + 1) * NAT_ENTRY_PER_BLOCK;
2574
2575                 spin_lock(&NM_I(sbi)->nid_list_lock);
2576                 for (; nid < last_nid; nid++)
2577                         update_free_nid_bitmap(sbi, nid, true, true);
2578                 spin_unlock(&NM_I(sbi)->nid_list_lock);
2579         }
2580
2581         for (i = 0; i < nm_i->nat_blocks; i++) {
2582                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2583                 if (i >= nm_i->nat_blocks)
2584                         break;
2585
2586                 __set_bit_le(i, nm_i->nat_block_bitmap);
2587         }
2588 }
2589
2590 static int init_node_manager(struct f2fs_sb_info *sbi)
2591 {
2592         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2593         struct f2fs_nm_info *nm_i = NM_I(sbi);
2594         unsigned char *version_bitmap;
2595         unsigned int nat_segs;
2596         int err;
2597
2598         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2599
2600         /* segment_count_nat includes pair segment so divide to 2. */
2601         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2602         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2603         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2604
2605         /* not used nids: 0, node, meta, (and root counted as valid node) */
2606         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2607                                                         F2FS_RESERVED_NODE_NUM;
2608         nm_i->nid_cnt[FREE_NID_LIST] = 0;
2609         nm_i->nid_cnt[ALLOC_NID_LIST] = 0;
2610         nm_i->nat_cnt = 0;
2611         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2612         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2613         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2614
2615         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2616         INIT_LIST_HEAD(&nm_i->nid_list[FREE_NID_LIST]);
2617         INIT_LIST_HEAD(&nm_i->nid_list[ALLOC_NID_LIST]);
2618         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2619         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2620         INIT_LIST_HEAD(&nm_i->nat_entries);
2621
2622         mutex_init(&nm_i->build_lock);
2623         spin_lock_init(&nm_i->nid_list_lock);
2624         init_rwsem(&nm_i->nat_tree_lock);
2625
2626         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2627         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2628         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2629         if (!version_bitmap)
2630                 return -EFAULT;
2631
2632         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2633                                         GFP_KERNEL);
2634         if (!nm_i->nat_bitmap)
2635                 return -ENOMEM;
2636
2637         err = __get_nat_bitmaps(sbi);
2638         if (err)
2639                 return err;
2640
2641 #ifdef CONFIG_F2FS_CHECK_FS
2642         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2643                                         GFP_KERNEL);
2644         if (!nm_i->nat_bitmap_mir)
2645                 return -ENOMEM;
2646 #endif
2647
2648         return 0;
2649 }
2650
2651 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2652 {
2653         struct f2fs_nm_info *nm_i = NM_I(sbi);
2654
2655         nm_i->free_nid_bitmap = f2fs_kvzalloc(nm_i->nat_blocks *
2656                                         NAT_ENTRY_BITMAP_SIZE, GFP_KERNEL);
2657         if (!nm_i->free_nid_bitmap)
2658                 return -ENOMEM;
2659
2660         nm_i->nat_block_bitmap = f2fs_kvzalloc(nm_i->nat_blocks / 8,
2661                                                                 GFP_KERNEL);
2662         if (!nm_i->nat_block_bitmap)
2663                 return -ENOMEM;
2664
2665         nm_i->free_nid_count = f2fs_kvzalloc(nm_i->nat_blocks *
2666                                         sizeof(unsigned short), GFP_KERNEL);
2667         if (!nm_i->free_nid_count)
2668                 return -ENOMEM;
2669         return 0;
2670 }
2671
2672 int build_node_manager(struct f2fs_sb_info *sbi)
2673 {
2674         int err;
2675
2676         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2677         if (!sbi->nm_info)
2678                 return -ENOMEM;
2679
2680         err = init_node_manager(sbi);
2681         if (err)
2682                 return err;
2683
2684         err = init_free_nid_cache(sbi);
2685         if (err)
2686                 return err;
2687
2688         /* load free nid status from nat_bits table */
2689         load_free_nid_bitmap(sbi);
2690
2691         build_free_nids(sbi, true, true);
2692         return 0;
2693 }
2694
2695 void destroy_node_manager(struct f2fs_sb_info *sbi)
2696 {
2697         struct f2fs_nm_info *nm_i = NM_I(sbi);
2698         struct free_nid *i, *next_i;
2699         struct nat_entry *natvec[NATVEC_SIZE];
2700         struct nat_entry_set *setvec[SETVEC_SIZE];
2701         nid_t nid = 0;
2702         unsigned int found;
2703
2704         if (!nm_i)
2705                 return;
2706
2707         /* destroy free nid list */
2708         spin_lock(&nm_i->nid_list_lock);
2709         list_for_each_entry_safe(i, next_i, &nm_i->nid_list[FREE_NID_LIST],
2710                                                                         list) {
2711                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2712                 spin_unlock(&nm_i->nid_list_lock);
2713                 kmem_cache_free(free_nid_slab, i);
2714                 spin_lock(&nm_i->nid_list_lock);
2715         }
2716         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID_LIST]);
2717         f2fs_bug_on(sbi, nm_i->nid_cnt[ALLOC_NID_LIST]);
2718         f2fs_bug_on(sbi, !list_empty(&nm_i->nid_list[ALLOC_NID_LIST]));
2719         spin_unlock(&nm_i->nid_list_lock);
2720
2721         /* destroy nat cache */
2722         down_write(&nm_i->nat_tree_lock);
2723         while ((found = __gang_lookup_nat_cache(nm_i,
2724                                         nid, NATVEC_SIZE, natvec))) {
2725                 unsigned idx;
2726
2727                 nid = nat_get_nid(natvec[found - 1]) + 1;
2728                 for (idx = 0; idx < found; idx++)
2729                         __del_from_nat_cache(nm_i, natvec[idx]);
2730         }
2731         f2fs_bug_on(sbi, nm_i->nat_cnt);
2732
2733         /* destroy nat set cache */
2734         nid = 0;
2735         while ((found = __gang_lookup_nat_set(nm_i,
2736                                         nid, SETVEC_SIZE, setvec))) {
2737                 unsigned idx;
2738
2739                 nid = setvec[found - 1]->set + 1;
2740                 for (idx = 0; idx < found; idx++) {
2741                         /* entry_cnt is not zero, when cp_error was occurred */
2742                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2743                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2744                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2745                 }
2746         }
2747         up_write(&nm_i->nat_tree_lock);
2748
2749         kvfree(nm_i->nat_block_bitmap);
2750         kvfree(nm_i->free_nid_bitmap);
2751         kvfree(nm_i->free_nid_count);
2752
2753         kfree(nm_i->nat_bitmap);
2754         kfree(nm_i->nat_bits);
2755 #ifdef CONFIG_F2FS_CHECK_FS
2756         kfree(nm_i->nat_bitmap_mir);
2757 #endif
2758         sbi->nm_info = NULL;
2759         kfree(nm_i);
2760 }
2761
2762 int __init create_node_manager_caches(void)
2763 {
2764         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2765                         sizeof(struct nat_entry));
2766         if (!nat_entry_slab)
2767                 goto fail;
2768
2769         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2770                         sizeof(struct free_nid));
2771         if (!free_nid_slab)
2772                 goto destroy_nat_entry;
2773
2774         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2775                         sizeof(struct nat_entry_set));
2776         if (!nat_entry_set_slab)
2777                 goto destroy_free_nid;
2778         return 0;
2779
2780 destroy_free_nid:
2781         kmem_cache_destroy(free_nid_slab);
2782 destroy_nat_entry:
2783         kmem_cache_destroy(nat_entry_slab);
2784 fail:
2785         return -ENOMEM;
2786 }
2787
2788 void destroy_node_manager_caches(void)
2789 {
2790         kmem_cache_destroy(nat_entry_set_slab);
2791         kmem_cache_destroy(free_nid_slab);
2792         kmem_cache_destroy(nat_entry_slab);
2793 }