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 "xattr.h"
  23 #include "trace.h"
  24 #include <trace/events/f2fs.h>
  25
  26 #define on_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
  27
  28 static struct kmem_cache *nat_entry_slab;
  29 static struct kmem_cache *free_nid_slab;
  30 static struct kmem_cache *nat_entry_set_slab;
  31
  32 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
  33 {
  34         struct f2fs_nm_info *nm_i = NM_I(sbi);
  35         struct sysinfo val;
  36         unsigned long avail_ram;
  37         unsigned long mem_size = 0;
  38         bool res = false;
  39
  40         si_meminfo(&val);
  41
  42         /* only uses low memory */
  43         avail_ram = val.totalram - val.totalhigh;
  44
  45         /*
  46          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  47          */
  48         if (type == FREE_NIDS) {
  49                 mem_size = (nm_i->nid_cnt[FREE_NID_LIST] *
  50                                 sizeof(struct free_nid)) >> PAGE_SHIFT;
  51                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  52         } else if (type == NAT_ENTRIES) {
  53                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
  54                                                         PAGE_SHIFT;
  55                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  56                 if (excess_cached_nats(sbi))
  57                         res = false;
  58         } else if (type == DIRTY_DENTS) {
  59                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  60                         return false;
  61                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  62                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  63         } else if (type == INO_ENTRIES) {
  64                 int i;
  65
  66                 for (i = 0; i <= UPDATE_INO; i++)
  67                         mem_size += sbi->im[i].ino_num *
  68                                                 sizeof(struct ino_entry);
  69                 mem_size >>= PAGE_SHIFT;
  70                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  71         } else if (type == EXTENT_CACHE) {
  72                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  73                                 sizeof(struct extent_tree) +
  74                                 atomic_read(&sbi->total_ext_node) *
  75                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  76                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  77         } else {
  78                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
  79                         return true;
  80         }
  81         return res;
  82 }
  83
  84 static void clear_node_page_dirty(struct page *page)
  85 {
  86         struct address_space *mapping = page->mapping;
  87         unsigned int long flags;
  88
  89         if (PageDirty(page)) {
  90                 spin_lock_irqsave(&mapping->tree_lock, flags);
  91                 radix_tree_tag_clear(&mapping->page_tree,
  92                                 page_index(page),
  93                                 PAGECACHE_TAG_DIRTY);
  94                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  95
  96                 clear_page_dirty_for_io(page);
  97                 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
  98         }
  99         ClearPageUptodate(page);
 100 }
 101
 102 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 103 {
 104         pgoff_t index = current_nat_addr(sbi, nid);
 105         return get_meta_page(sbi, index);
 106 }
 107
 108 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 109 {
 110         struct page *src_page;
 111         struct page *dst_page;
 112         pgoff_t src_off;
 113         pgoff_t dst_off;
 114         void *src_addr;
 115         void *dst_addr;
 116         struct f2fs_nm_info *nm_i = NM_I(sbi);
 117
 118         src_off = current_nat_addr(sbi, nid);
 119         dst_off = next_nat_addr(sbi, src_off);
 120
 121         /* get current nat block page with lock */
 122         src_page = get_meta_page(sbi, src_off);
 123         dst_page = grab_meta_page(sbi, dst_off);
 124         f2fs_bug_on(sbi, PageDirty(src_page));
 125
 126         src_addr = page_address(src_page);
 127         dst_addr = page_address(dst_page);
 128         memcpy(dst_addr, src_addr, PAGE_SIZE);
 129         set_page_dirty(dst_page);
 130         f2fs_put_page(src_page, 1);
 131
 132         set_to_next_nat(nm_i, nid);
 133
 134         return dst_page;
 135 }
 136
 137 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 138 {
 139         return radix_tree_lookup(&nm_i->nat_root, n);
 140 }
 141
 142 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 143                 nid_t start, unsigned int nr, struct nat_entry **ep)
 144 {
 145         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 146 }
 147
 148 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 149 {
 150         list_del(&e->list);
 151         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 152         nm_i->nat_cnt--;
 153         kmem_cache_free(nat_entry_slab, e);
 154 }
 155
 156 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 157                                                 struct nat_entry *ne)
 158 {
 159         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 160         struct nat_entry_set *head;
 161
 162         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 163         if (!head) {
 164                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 165
 166                 INIT_LIST_HEAD(&head->entry_list);
 167                 INIT_LIST_HEAD(&head->set_list);
 168                 head->set = set;
 169                 head->entry_cnt = 0;
 170                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 171         }
 172
 173         if (get_nat_flag(ne, IS_DIRTY))
 174                 goto refresh_list;
 175
 176         nm_i->dirty_nat_cnt++;
 177         head->entry_cnt++;
 178         set_nat_flag(ne, IS_DIRTY, true);
 179 refresh_list:
 180         if (nat_get_blkaddr(ne) == NEW_ADDR)
 181                 list_del_init(&ne->list);
 182         else
 183                 list_move_tail(&ne->list, &head->entry_list);
 184 }
 185
 186 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 187                 struct nat_entry_set *set, struct nat_entry *ne)
 188 {
 189         list_move_tail(&ne->list, &nm_i->nat_entries);
 190         set_nat_flag(ne, IS_DIRTY, false);
 191         set->entry_cnt--;
 192         nm_i->dirty_nat_cnt--;
 193 }
 194
 195 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 196                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 197 {
 198         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 199                                                         start, nr);
 200 }
 201
 202 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 203 {
 204         struct f2fs_nm_info *nm_i = NM_I(sbi);
 205         struct nat_entry *e;
 206         bool need = false;
 207
 208         down_read(&nm_i->nat_tree_lock);
 209         e = __lookup_nat_cache(nm_i, nid);
 210         if (e) {
 211                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 212                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 213                         need = true;
 214         }
 215         up_read(&nm_i->nat_tree_lock);
 216         return need;
 217 }
 218
 219 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 220 {
 221         struct f2fs_nm_info *nm_i = NM_I(sbi);
 222         struct nat_entry *e;
 223         bool is_cp = true;
 224
 225         down_read(&nm_i->nat_tree_lock);
 226         e = __lookup_nat_cache(nm_i, nid);
 227         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 228                 is_cp = false;
 229         up_read(&nm_i->nat_tree_lock);
 230         return is_cp;
 231 }
 232
 233 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 234 {
 235         struct f2fs_nm_info *nm_i = NM_I(sbi);
 236         struct nat_entry *e;
 237         bool need_update = true;
 238
 239         down_read(&nm_i->nat_tree_lock);
 240         e = __lookup_nat_cache(nm_i, ino);
 241         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 242                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 243                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 244                 need_update = false;
 245         up_read(&nm_i->nat_tree_lock);
 246         return need_update;
 247 }
 248
 249 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 250                                                                 bool no_fail)
 251 {
 252         struct nat_entry *new;
 253
 254         if (no_fail) {
 255                 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
 256                 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
 257         } else {
 258                 new = kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
 259                 if (!new)
 260                         return NULL;
 261                 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 262                         kmem_cache_free(nat_entry_slab, new);
 263                         return NULL;
 264                 }
 265         }
 266
 267         memset(new, 0, sizeof(struct nat_entry));
 268         nat_set_nid(new, nid);
 269         nat_reset_flag(new);
 270         list_add_tail(&new->list, &nm_i->nat_entries);
 271         nm_i->nat_cnt++;
 272         return new;
 273 }
 274
 275 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 276                                                 struct f2fs_nat_entry *ne)
 277 {
 278         struct f2fs_nm_info *nm_i = NM_I(sbi);
 279         struct nat_entry *e;
 280
 281         e = __lookup_nat_cache(nm_i, nid);
 282         if (!e) {
 283                 e = grab_nat_entry(nm_i, nid, false);
 284                 if (e)
 285                         node_info_from_raw_nat(&e->ni, ne);
 286         } else {
 287                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 288                                 nat_get_blkaddr(e) !=
 289                                         le32_to_cpu(ne->block_addr) ||
 290                                 nat_get_version(e) != ne->version);
 291         }
 292 }
 293
 294 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 295                         block_t new_blkaddr, bool fsync_done)
 296 {
 297         struct f2fs_nm_info *nm_i = NM_I(sbi);
 298         struct nat_entry *e;
 299
 300         down_write(&nm_i->nat_tree_lock);
 301         e = __lookup_nat_cache(nm_i, ni->nid);
 302         if (!e) {
 303                 e = grab_nat_entry(nm_i, ni->nid, true);
 304                 copy_node_info(&e->ni, ni);
 305                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 306         } else if (new_blkaddr == NEW_ADDR) {
 307                 /*
 308                  * when nid is reallocated,
 309                  * previous nat entry can be remained in nat cache.
 310                  * So, reinitialize it with new information.
 311                  */
 312                 copy_node_info(&e->ni, ni);
 313                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 314         }
 315
 316         /* sanity check */
 317         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 318         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 319                         new_blkaddr == NULL_ADDR);
 320         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 321                         new_blkaddr == NEW_ADDR);
 322         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 323                         nat_get_blkaddr(e) != NULL_ADDR &&
 324                         new_blkaddr == NEW_ADDR);
 325
 326         /* increment version no as node is removed */
 327         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 328                 unsigned char version = nat_get_version(e);
 329                 nat_set_version(e, inc_node_version(version));
 330
 331                 /* in order to reuse the nid */
 332                 if (nm_i->next_scan_nid > ni->nid)
 333                         nm_i->next_scan_nid = ni->nid;
 334         }
 335
 336         /* change address */
 337         nat_set_blkaddr(e, new_blkaddr);
 338         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 339                 set_nat_flag(e, IS_CHECKPOINTED, false);
 340         __set_nat_cache_dirty(nm_i, e);
 341
 342         /* update fsync_mark if its inode nat entry is still alive */
 343         if (ni->nid != ni->ino)
 344                 e = __lookup_nat_cache(nm_i, ni->ino);
 345         if (e) {
 346                 if (fsync_done && ni->nid == ni->ino)
 347                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 348                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 349         }
 350         up_write(&nm_i->nat_tree_lock);
 351 }
 352
 353 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 354 {
 355         struct f2fs_nm_info *nm_i = NM_I(sbi);
 356         int nr = nr_shrink;
 357
 358         if (!down_write_trylock(&nm_i->nat_tree_lock))
 359                 return 0;
 360
 361         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 362                 struct nat_entry *ne;
 363                 ne = list_first_entry(&nm_i->nat_entries,
 364                                         struct nat_entry, list);
 365                 __del_from_nat_cache(nm_i, ne);
 366                 nr_shrink--;
 367         }
 368         up_write(&nm_i->nat_tree_lock);
 369         return nr - nr_shrink;
 370 }
 371
 372 /*
 373  * This function always returns success
 374  */
 375 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 376 {
 377         struct f2fs_nm_info *nm_i = NM_I(sbi);
 378         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 379         struct f2fs_journal *journal = curseg->journal;
 380         nid_t start_nid = START_NID(nid);
 381         struct f2fs_nat_block *nat_blk;
 382         struct page *page = NULL;
 383         struct f2fs_nat_entry ne;
 384         struct nat_entry *e;
 385         pgoff_t index;
 386         int i;
 387
 388         ni->nid = nid;
 389
 390         /* Check nat cache */
 391         down_read(&nm_i->nat_tree_lock);
 392         e = __lookup_nat_cache(nm_i, nid);
 393         if (e) {
 394                 ni->ino = nat_get_ino(e);
 395                 ni->blk_addr = nat_get_blkaddr(e);
 396                 ni->version = nat_get_version(e);
 397                 up_read(&nm_i->nat_tree_lock);
 398                 return;
 399         }
 400
 401         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 402
 403         /* Check current segment summary */
 404         down_read(&curseg->journal_rwsem);
 405         i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 406         if (i >= 0) {
 407                 ne = nat_in_journal(journal, i);
 408                 node_info_from_raw_nat(ni, &ne);
 409         }
 410         up_read(&curseg->journal_rwsem);
 411         if (i >= 0) {
 412                 up_read(&nm_i->nat_tree_lock);
 413                 goto cache;
 414         }
 415
 416         /* Fill node_info from nat page */
 417         index = current_nat_addr(sbi, nid);
 418         up_read(&nm_i->nat_tree_lock);
 419
 420         page = get_meta_page(sbi, index);
 421         nat_blk = (struct f2fs_nat_block *)page_address(page);
 422         ne = nat_blk->entries[nid - start_nid];
 423         node_info_from_raw_nat(ni, &ne);
 424         f2fs_put_page(page, 1);
 425 cache:
 426         /* cache nat entry */
 427         down_write(&nm_i->nat_tree_lock);
 428         cache_nat_entry(sbi, nid, &ne);
 429         up_write(&nm_i->nat_tree_lock);
 430 }
 431
 432 /*
 433  * readahead MAX_RA_NODE number of node pages.
 434  */
 435 static void ra_node_pages(struct page *parent, int start, int n)
 436 {
 437         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 438         struct blk_plug plug;
 439         int i, end;
 440         nid_t nid;
 441
 442         blk_start_plug(&plug);
 443
 444         /* Then, try readahead for siblings of the desired node */
 445         end = start + n;
 446         end = min(end, NIDS_PER_BLOCK);
 447         for (i = start; i < end; i++) {
 448                 nid = get_nid(parent, i, false);
 449                 ra_node_page(sbi, nid);
 450         }
 451
 452         blk_finish_plug(&plug);
 453 }
 454
 455 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 456 {
 457         const long direct_index = ADDRS_PER_INODE(dn->inode);
 458         const long direct_blks = ADDRS_PER_BLOCK;
 459         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 460         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 461         int cur_level = dn->cur_level;
 462         int max_level = dn->max_level;
 463         pgoff_t base = 0;
 464
 465         if (!dn->max_level)
 466                 return pgofs + 1;
 467
 468         while (max_level-- > cur_level)
 469                 skipped_unit *= NIDS_PER_BLOCK;
 470
 471         switch (dn->max_level) {
 472         case 3:
 473                 base += 2 * indirect_blks;
 474         case 2:
 475                 base += 2 * direct_blks;
 476         case 1:
 477                 base += direct_index;
 478                 break;
 479         default:
 480                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 481         }
 482
 483         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 484 }
 485
 486 /*
 487  * The maximum depth is four.
 488  * Offset[0] will have raw inode offset.
 489  */
 490 static int get_node_path(struct inode *inode, long block,
 491                                 int offset[4], unsigned int noffset[4])
 492 {
 493         const long direct_index = ADDRS_PER_INODE(inode);
 494         const long direct_blks = ADDRS_PER_BLOCK;
 495         const long dptrs_per_blk = NIDS_PER_BLOCK;
 496         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 497         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 498         int n = 0;
 499         int level = 0;
 500
 501         noffset[0] = 0;
 502
 503         if (block < direct_index) {
 504                 offset[n] = block;
 505                 goto got;
 506         }
 507         block -= direct_index;
 508         if (block < direct_blks) {
 509                 offset[n++] = NODE_DIR1_BLOCK;
 510                 noffset[n] = 1;
 511                 offset[n] = block;
 512                 level = 1;
 513                 goto got;
 514         }
 515         block -= direct_blks;
 516         if (block < direct_blks) {
 517                 offset[n++] = NODE_DIR2_BLOCK;
 518                 noffset[n] = 2;
 519                 offset[n] = block;
 520                 level = 1;
 521                 goto got;
 522         }
 523         block -= direct_blks;
 524         if (block < indirect_blks) {
 525                 offset[n++] = NODE_IND1_BLOCK;
 526                 noffset[n] = 3;
 527                 offset[n++] = block / direct_blks;
 528                 noffset[n] = 4 + offset[n - 1];
 529                 offset[n] = block % direct_blks;
 530                 level = 2;
 531                 goto got;
 532         }
 533         block -= indirect_blks;
 534         if (block < indirect_blks) {
 535                 offset[n++] = NODE_IND2_BLOCK;
 536                 noffset[n] = 4 + dptrs_per_blk;
 537                 offset[n++] = block / direct_blks;
 538                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 539                 offset[n] = block % direct_blks;
 540                 level = 2;
 541                 goto got;
 542         }
 543         block -= indirect_blks;
 544         if (block < dindirect_blks) {
 545                 offset[n++] = NODE_DIND_BLOCK;
 546                 noffset[n] = 5 + (dptrs_per_blk * 2);
 547                 offset[n++] = block / indirect_blks;
 548                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 549                               offset[n - 1] * (dptrs_per_blk + 1);
 550                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 551                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 552                               offset[n - 2] * (dptrs_per_blk + 1) +
 553                               offset[n - 1];
 554                 offset[n] = block % direct_blks;
 555                 level = 3;
 556                 goto got;
 557         } else {
 558                 return -E2BIG;
 559         }
 560 got:
 561         return level;
 562 }
 563
 564 /*
 565  * Caller should call f2fs_put_dnode(dn).
 566  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 567  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 568  * In the case of RDONLY_NODE, we don't need to care about mutex.
 569  */
 570 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 571 {
 572         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 573         struct page *npage[4];
 574         struct page *parent = NULL;
 575         int offset[4];
 576         unsigned int noffset[4];
 577         nid_t nids[4];
 578         int level, i = 0;
 579         int err = 0;
 580
 581         level = get_node_path(dn->inode, index, offset, noffset);
 582         if (level < 0)
 583                 return level;
 584
 585         nids[0] = dn->inode->i_ino;
 586         npage[0] = dn->inode_page;
 587
 588         if (!npage[0]) {
 589                 npage[0] = get_node_page(sbi, nids[0]);
 590                 if (IS_ERR(npage[0]))
 591                         return PTR_ERR(npage[0]);
 592         }
 593
 594         /* if inline_data is set, should not report any block indices */
 595         if (f2fs_has_inline_data(dn->inode) && index) {
 596                 err = -ENOENT;
 597                 f2fs_put_page(npage[0], 1);
 598                 goto release_out;
 599         }
 600
 601         parent = npage[0];
 602         if (level != 0)
 603                 nids[1] = get_nid(parent, offset[0], true);
 604         dn->inode_page = npage[0];
 605         dn->inode_page_locked = true;
 606
 607         /* get indirect or direct nodes */
 608         for (i = 1; i <= level; i++) {
 609                 bool done = false;
 610
 611                 if (!nids[i] && mode == ALLOC_NODE) {
 612                         /* alloc new node */
 613                         if (!alloc_nid(sbi, &(nids[i]))) {
 614                                 err = -ENOSPC;
 615                                 goto release_pages;
 616                         }
 617
 618                         dn->nid = nids[i];
 619                         npage[i] = new_node_page(dn, noffset[i]);
 620                         if (IS_ERR(npage[i])) {
 621                                 alloc_nid_failed(sbi, nids[i]);
 622                                 err = PTR_ERR(npage[i]);
 623                                 goto release_pages;
 624                         }
 625
 626                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 627                         alloc_nid_done(sbi, nids[i]);
 628                         done = true;
 629                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 630                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 631                         if (IS_ERR(npage[i])) {
 632                                 err = PTR_ERR(npage[i]);
 633                                 goto release_pages;
 634                         }
 635                         done = true;
 636                 }
 637                 if (i == 1) {
 638                         dn->inode_page_locked = false;
 639                         unlock_page(parent);
 640                 } else {
 641                         f2fs_put_page(parent, 1);
 642                 }
 643
 644                 if (!done) {
 645                         npage[i] = get_node_page(sbi, nids[i]);
 646                         if (IS_ERR(npage[i])) {
 647                                 err = PTR_ERR(npage[i]);
 648                                 f2fs_put_page(npage[0], 0);
 649                                 goto release_out;
 650                         }
 651                 }
 652                 if (i < level) {
 653                         parent = npage[i];
 654                         nids[i + 1] = get_nid(parent, offset[i], false);
 655                 }
 656         }
 657         dn->nid = nids[level];
 658         dn->ofs_in_node = offset[level];
 659         dn->node_page = npage[level];
 660         dn->data_blkaddr = datablock_addr(dn->inode,
 661                                 dn->node_page, dn->ofs_in_node);
 662         return 0;
 663
 664 release_pages:
 665         f2fs_put_page(parent, 1);
 666         if (i > 1)
 667                 f2fs_put_page(npage[0], 0);
 668 release_out:
 669         dn->inode_page = NULL;
 670         dn->node_page = NULL;
 671         if (err == -ENOENT) {
 672                 dn->cur_level = i;
 673                 dn->max_level = level;
 674                 dn->ofs_in_node = offset[level];
 675         }
 676         return err;
 677 }
 678
 679 static void truncate_node(struct dnode_of_data *dn)
 680 {
 681         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 682         struct node_info ni;
 683
 684         get_node_info(sbi, dn->nid, &ni);
 685         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 686
 687         /* Deallocate node address */
 688         invalidate_blocks(sbi, ni.blk_addr);
 689         dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
 690         set_node_addr(sbi, &ni, NULL_ADDR, false);
 691
 692         if (dn->nid == dn->inode->i_ino) {
 693                 remove_orphan_inode(sbi, dn->nid);
 694                 dec_valid_inode_count(sbi);
 695                 f2fs_inode_synced(dn->inode);
 696         }
 697
 698         clear_node_page_dirty(dn->node_page);
 699         set_sbi_flag(sbi, SBI_IS_DIRTY);
 700
 701         f2fs_put_page(dn->node_page, 1);
 702
 703         invalidate_mapping_pages(NODE_MAPPING(sbi),
 704                         dn->node_page->index, dn->node_page->index);
 705
 706         dn->node_page = NULL;
 707         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 708 }
 709
 710 static int truncate_dnode(struct dnode_of_data *dn)
 711 {
 712         struct page *page;
 713
 714         if (dn->nid == 0)
 715                 return 1;
 716
 717         /* get direct node */
 718         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 719         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 720                 return 1;
 721         else if (IS_ERR(page))
 722                 return PTR_ERR(page);
 723
 724         /* Make dnode_of_data for parameter */
 725         dn->node_page = page;
 726         dn->ofs_in_node = 0;
 727         truncate_data_blocks(dn);
 728         truncate_node(dn);
 729         return 1;
 730 }
 731
 732 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 733                                                 int ofs, int depth)
 734 {
 735         struct dnode_of_data rdn = *dn;
 736         struct page *page;
 737         struct f2fs_node *rn;
 738         nid_t child_nid;
 739         unsigned int child_nofs;
 740         int freed = 0;
 741         int i, ret;
 742
 743         if (dn->nid == 0)
 744                 return NIDS_PER_BLOCK + 1;
 745
 746         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 747
 748         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 749         if (IS_ERR(page)) {
 750                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 751                 return PTR_ERR(page);
 752         }
 753
 754         ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 755
 756         rn = F2FS_NODE(page);
 757         if (depth < 3) {
 758                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 759                         child_nid = le32_to_cpu(rn->in.nid[i]);
 760                         if (child_nid == 0)
 761                                 continue;
 762                         rdn.nid = child_nid;
 763                         ret = truncate_dnode(&rdn);
 764                         if (ret < 0)
 765                                 goto out_err;
 766                         if (set_nid(page, i, 0, false))
 767                                 dn->node_changed = true;
 768                 }
 769         } else {
 770                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 771                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 772                         child_nid = le32_to_cpu(rn->in.nid[i]);
 773                         if (child_nid == 0) {
 774                                 child_nofs += NIDS_PER_BLOCK + 1;
 775                                 continue;
 776                         }
 777                         rdn.nid = child_nid;
 778                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 779                         if (ret == (NIDS_PER_BLOCK + 1)) {
 780                                 if (set_nid(page, i, 0, false))
 781                                         dn->node_changed = true;
 782                                 child_nofs += ret;
 783                         } else if (ret < 0 && ret != -ENOENT) {
 784                                 goto out_err;
 785                         }
 786                 }
 787                 freed = child_nofs;
 788         }
 789
 790         if (!ofs) {
 791                 /* remove current indirect node */
 792                 dn->node_page = page;
 793                 truncate_node(dn);
 794                 freed++;
 795         } else {
 796                 f2fs_put_page(page, 1);
 797         }
 798         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 799         return freed;
 800
 801 out_err:
 802         f2fs_put_page(page, 1);
 803         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 804         return ret;
 805 }
 806
 807 static int truncate_partial_nodes(struct dnode_of_data *dn,
 808                         struct f2fs_inode *ri, int *offset, int depth)
 809 {
 810         struct page *pages[2];
 811         nid_t nid[3];
 812         nid_t child_nid;
 813         int err = 0;
 814         int i;
 815         int idx = depth - 2;
 816
 817         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 818         if (!nid[0])
 819                 return 0;
 820
 821         /* get indirect nodes in the path */
 822         for (i = 0; i < idx + 1; i++) {
 823                 /* reference count'll be increased */
 824                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 825                 if (IS_ERR(pages[i])) {
 826                         err = PTR_ERR(pages[i]);
 827                         idx = i - 1;
 828                         goto fail;
 829                 }
 830                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 831         }
 832
 833         ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 834
 835         /* free direct nodes linked to a partial indirect node */
 836         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 837                 child_nid = get_nid(pages[idx], i, false);
 838                 if (!child_nid)
 839                         continue;
 840                 dn->nid = child_nid;
 841                 err = truncate_dnode(dn);
 842                 if (err < 0)
 843                         goto fail;
 844                 if (set_nid(pages[idx], i, 0, false))
 845                         dn->node_changed = true;
 846         }
 847
 848         if (offset[idx + 1] == 0) {
 849                 dn->node_page = pages[idx];
 850                 dn->nid = nid[idx];
 851                 truncate_node(dn);
 852         } else {
 853                 f2fs_put_page(pages[idx], 1);
 854         }
 855         offset[idx]++;
 856         offset[idx + 1] = 0;
 857         idx--;
 858 fail:
 859         for (i = idx; i >= 0; i--)
 860                 f2fs_put_page(pages[i], 1);
 861
 862         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 863
 864         return err;
 865 }
 866
 867 /*
 868  * All the block addresses of data and nodes should be nullified.
 869  */
 870 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 871 {
 872         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 873         int err = 0, cont = 1;
 874         int level, offset[4], noffset[4];
 875         unsigned int nofs = 0;
 876         struct f2fs_inode *ri;
 877         struct dnode_of_data dn;
 878         struct page *page;
 879
 880         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 881
 882         level = get_node_path(inode, from, offset, noffset);
 883         if (level < 0)
 884                 return level;
 885
 886         page = get_node_page(sbi, inode->i_ino);
 887         if (IS_ERR(page)) {
 888                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 889                 return PTR_ERR(page);
 890         }
 891
 892         set_new_dnode(&dn, inode, page, NULL, 0);
 893         unlock_page(page);
 894
 895         ri = F2FS_INODE(page);
 896         switch (level) {
 897         case 0:
 898         case 1:
 899                 nofs = noffset[1];
 900                 break;
 901         case 2:
 902                 nofs = noffset[1];
 903                 if (!offset[level - 1])
 904                         goto skip_partial;
 905                 err = truncate_partial_nodes(&dn, ri, offset, level);
 906                 if (err < 0 && err != -ENOENT)
 907                         goto fail;
 908                 nofs += 1 + NIDS_PER_BLOCK;
 909                 break;
 910         case 3:
 911                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 912                 if (!offset[level - 1])
 913                         goto skip_partial;
 914                 err = truncate_partial_nodes(&dn, ri, offset, level);
 915                 if (err < 0 && err != -ENOENT)
 916                         goto fail;
 917                 break;
 918         default:
 919                 BUG();
 920         }
 921
 922 skip_partial:
 923         while (cont) {
 924                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 925                 switch (offset[0]) {
 926                 case NODE_DIR1_BLOCK:
 927                 case NODE_DIR2_BLOCK:
 928                         err = truncate_dnode(&dn);
 929                         break;
 930
 931                 case NODE_IND1_BLOCK:
 932                 case NODE_IND2_BLOCK:
 933                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 934                         break;
 935
 936                 case NODE_DIND_BLOCK:
 937                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 938                         cont = 0;
 939                         break;
 940
 941                 default:
 942                         BUG();
 943                 }
 944                 if (err < 0 && err != -ENOENT)
 945                         goto fail;
 946                 if (offset[1] == 0 &&
 947                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 948                         lock_page(page);
 949                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
 950                         f2fs_wait_on_page_writeback(page, NODE, true);
 951                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 952                         set_page_dirty(page);
 953                         unlock_page(page);
 954                 }
 955                 offset[1] = 0;
 956                 offset[0]++;
 957                 nofs += err;
 958         }
 959 fail:
 960         f2fs_put_page(page, 0);
 961         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 962         return err > 0 ? 0 : err;
 963 }
 964
 965 int truncate_xattr_node(struct inode *inode, struct page *page)
 966 {
 967         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 968         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 969         struct dnode_of_data dn;
 970         struct page *npage;
 971
 972         if (!nid)
 973                 return 0;
 974
 975         npage = get_node_page(sbi, nid);
 976         if (IS_ERR(npage))
 977                 return PTR_ERR(npage);
 978
 979         f2fs_i_xnid_write(inode, 0);
 980
 981         set_new_dnode(&dn, inode, page, npage, nid);
 982
 983         if (page)
 984                 dn.inode_page_locked = true;
 985         truncate_node(&dn);
 986         return 0;
 987 }
 988
 989 /*
 990  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 991  * f2fs_unlock_op().
 992  */
 993 int remove_inode_page(struct inode *inode)
 994 {
 995         struct dnode_of_data dn;
 996         int err;
 997
 998         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 999         err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1000         if (err)
1001                 return err;
1002
1003         err = truncate_xattr_node(inode, dn.inode_page);
1004         if (err) {
1005                 f2fs_put_dnode(&dn);
1006                 return err;
1007         }
1008
1009         /* remove potential inline_data blocks */
1010         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1011                                 S_ISLNK(inode->i_mode))
1012                 truncate_data_blocks_range(&dn, 1);
1013
1014         /* 0 is possible, after f2fs_new_inode() has failed */
1015         f2fs_bug_on(F2FS_I_SB(inode),
1016                         inode->i_blocks != 0 && inode->i_blocks != 8);
1017
1018         /* will put inode & node pages */
1019         truncate_node(&dn);
1020         return 0;
1021 }
1022
1023 struct page *new_inode_page(struct inode *inode)
1024 {
1025         struct dnode_of_data dn;
1026
1027         /* allocate inode page for new inode */
1028         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1029
1030         /* caller should f2fs_put_page(page, 1); */
1031         return new_node_page(&dn, 0);
1032 }
1033
1034 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1035 {
1036         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1037         struct node_info new_ni;
1038         struct page *page;
1039         int err;
1040
1041         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1042                 return ERR_PTR(-EPERM);
1043
1044         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1045         if (!page)
1046                 return ERR_PTR(-ENOMEM);
1047
1048         if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1049                 goto fail;
1050
1051 #ifdef CONFIG_F2FS_CHECK_FS
1052         get_node_info(sbi, dn->nid, &new_ni);
1053         f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1054 #endif
1055         new_ni.nid = dn->nid;
1056         new_ni.ino = dn->inode->i_ino;
1057         new_ni.blk_addr = NULL_ADDR;
1058         new_ni.flag = 0;
1059         new_ni.version = 0;
1060         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1061
1062         f2fs_wait_on_page_writeback(page, NODE, true);
1063         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1064         set_cold_node(dn->inode, page);
1065         if (!PageUptodate(page))
1066                 SetPageUptodate(page);
1067         if (set_page_dirty(page))
1068                 dn->node_changed = true;
1069
1070         if (f2fs_has_xattr_block(ofs))
1071                 f2fs_i_xnid_write(dn->inode, dn->nid);
1072
1073         if (ofs == 0)
1074                 inc_valid_inode_count(sbi);
1075         return page;
1076
1077 fail:
1078         clear_node_page_dirty(page);
1079         f2fs_put_page(page, 1);
1080         return ERR_PTR(err);
1081 }
1082
1083 /*
1084  * Caller should do after getting the following values.
1085  * 0: f2fs_put_page(page, 0)
1086  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1087  */
1088 static int read_node_page(struct page *page, int op_flags)
1089 {
1090         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1091         struct node_info ni;
1092         struct f2fs_io_info fio = {
1093                 .sbi = sbi,
1094                 .type = NODE,
1095                 .op = REQ_OP_READ,
1096                 .op_flags = op_flags,
1097                 .page = page,
1098                 .encrypted_page = NULL,
1099         };
1100
1101         if (PageUptodate(page))
1102                 return LOCKED_PAGE;
1103
1104         get_node_info(sbi, page->index, &ni);
1105
1106         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1107                 ClearPageUptodate(page);
1108                 return -ENOENT;
1109         }
1110
1111         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1112         return f2fs_submit_page_bio(&fio);
1113 }
1114
1115 /*
1116  * Readahead a node page
1117  */
1118 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1119 {
1120         struct page *apage;
1121         int err;
1122
1123         if (!nid)
1124                 return;
1125         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1126
1127         rcu_read_lock();
1128         apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1129         rcu_read_unlock();
1130         if (apage)
1131                 return;
1132
1133         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1134         if (!apage)
1135                 return;
1136
1137         err = read_node_page(apage, REQ_RAHEAD);
1138         f2fs_put_page(apage, err ? 1 : 0);
1139 }
1140
1141 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1142                                         struct page *parent, int start)
1143 {
1144         struct page *page;
1145         int err;
1146
1147         if (!nid)
1148                 return ERR_PTR(-ENOENT);
1149         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1150 repeat:
1151         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1152         if (!page)
1153                 return ERR_PTR(-ENOMEM);
1154
1155         err = read_node_page(page, 0);
1156         if (err < 0) {
1157                 f2fs_put_page(page, 1);
1158                 return ERR_PTR(err);
1159         } else if (err == LOCKED_PAGE) {
1160                 err = 0;
1161                 goto page_hit;
1162         }
1163
1164         if (parent)
1165                 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1166
1167         lock_page(page);
1168
1169         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1170                 f2fs_put_page(page, 1);
1171                 goto repeat;
1172         }
1173
1174         if (unlikely(!PageUptodate(page))) {
1175                 err = -EIO;
1176                 goto out_err;
1177         }
1178
1179         if (!f2fs_inode_chksum_verify(sbi, page)) {
1180                 err = -EBADMSG;
1181                 goto out_err;
1182         }
1183 page_hit:
1184         if(unlikely(nid != nid_of_node(page))) {
1185                 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1186                         "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1187                         nid, nid_of_node(page), ino_of_node(page),
1188                         ofs_of_node(page), cpver_of_node(page),
1189                         next_blkaddr_of_node(page));
1190                 err = -EINVAL;
1191 out_err:
1192                 ClearPageUptodate(page);
1193                 f2fs_put_page(page, 1);
1194                 return ERR_PTR(err);
1195         }
1196         return page;
1197 }
1198
1199 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1200 {
1201         return __get_node_page(sbi, nid, NULL, 0);
1202 }
1203
1204 struct page *get_node_page_ra(struct page *parent, int start)
1205 {
1206         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1207         nid_t nid = get_nid(parent, start, false);
1208
1209         return __get_node_page(sbi, nid, parent, start);
1210 }
1211
1212 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1213 {
1214         struct inode *inode;
1215         struct page *page;
1216         int ret;
1217
1218         /* should flush inline_data before evict_inode */
1219         inode = ilookup(sbi->sb, ino);
1220         if (!inode)
1221                 return;
1222
1223         page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1224         if (!page)
1225                 goto iput_out;
1226
1227         if (!PageUptodate(page))
1228                 goto page_out;
1229
1230         if (!PageDirty(page))
1231                 goto page_out;
1232
1233         if (!clear_page_dirty_for_io(page))
1234                 goto page_out;
1235
1236         ret = f2fs_write_inline_data(inode, page);
1237         inode_dec_dirty_pages(inode);
1238         remove_dirty_inode(inode);
1239         if (ret)
1240                 set_page_dirty(page);
1241 page_out:
1242         f2fs_put_page(page, 1);
1243 iput_out:
1244         iput(inode);
1245 }
1246
1247 void move_node_page(struct page *node_page, int gc_type)
1248 {
1249         if (gc_type == FG_GC) {
1250                 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1251                 struct writeback_control wbc = {
1252                         .sync_mode = WB_SYNC_ALL,
1253                         .nr_to_write = 1,
1254                         .for_reclaim = 0,
1255                 };
1256
1257                 set_page_dirty(node_page);
1258                 f2fs_wait_on_page_writeback(node_page, NODE, true);
1259
1260                 f2fs_bug_on(sbi, PageWriteback(node_page));
1261                 if (!clear_page_dirty_for_io(node_page))
1262                         goto out_page;
1263
1264                 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1265                         unlock_page(node_page);
1266                 goto release_page;
1267         } else {
1268                 /* set page dirty and write it */
1269                 if (!PageWriteback(node_page))
1270                         set_page_dirty(node_page);
1271         }
1272 out_page:
1273         unlock_page(node_page);
1274 release_page:
1275         f2fs_put_page(node_page, 0);
1276 }
1277
1278 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1279 {
1280         pgoff_t index, end;
1281         struct pagevec pvec;
1282         struct page *last_page = NULL;
1283
1284         pagevec_init(&pvec, 0);
1285         index = 0;
1286         end = ULONG_MAX;
1287
1288         while (index <= end) {
1289                 int i, nr_pages;
1290                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1291                                 PAGECACHE_TAG_DIRTY,
1292                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1293                 if (nr_pages == 0)
1294                         break;
1295
1296                 for (i = 0; i < nr_pages; i++) {
1297                         struct page *page = pvec.pages[i];
1298
1299                         if (unlikely(f2fs_cp_error(sbi))) {
1300                                 f2fs_put_page(last_page, 0);
1301                                 pagevec_release(&pvec);
1302                                 return ERR_PTR(-EIO);
1303                         }
1304
1305                         if (!IS_DNODE(page) || !is_cold_node(page))
1306                                 continue;
1307                         if (ino_of_node(page) != ino)
1308                                 continue;
1309
1310                         lock_page(page);
1311
1312                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1313 continue_unlock:
1314                                 unlock_page(page);
1315                                 continue;
1316                         }
1317                         if (ino_of_node(page) != ino)
1318                                 goto continue_unlock;
1319
1320                         if (!PageDirty(page)) {
1321                                 /* someone wrote it for us */
1322                                 goto continue_unlock;
1323                         }
1324
1325                         if (last_page)
1326                                 f2fs_put_page(last_page, 0);
1327
1328                         get_page(page);
1329                         last_page = page;
1330                         unlock_page(page);
1331                 }
1332                 pagevec_release(&pvec);
1333                 cond_resched();
1334         }
1335         return last_page;
1336 }
1337
1338 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1339                                 struct writeback_control *wbc, bool do_balance,
1340                                 enum iostat_type io_type)
1341 {
1342         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1343         nid_t nid;
1344         struct node_info ni;
1345         struct f2fs_io_info fio = {
1346                 .sbi = sbi,
1347                 .type = NODE,
1348                 .op = REQ_OP_WRITE,
1349                 .op_flags = wbc_to_write_flags(wbc),
1350                 .page = page,
1351                 .encrypted_page = NULL,
1352                 .submitted = false,
1353                 .io_type = io_type,
1354         };
1355
1356         trace_f2fs_writepage(page, NODE);
1357
1358         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1359                 goto redirty_out;
1360         if (unlikely(f2fs_cp_error(sbi)))
1361                 goto redirty_out;
1362
1363         /* get old block addr of this node page */
1364         nid = nid_of_node(page);
1365         f2fs_bug_on(sbi, page->index != nid);
1366
1367         if (wbc->for_reclaim) {
1368                 if (!down_read_trylock(&sbi->node_write))
1369                         goto redirty_out;
1370         } else {
1371                 down_read(&sbi->node_write);
1372         }
1373
1374         get_node_info(sbi, nid, &ni);
1375
1376         /* This page is already truncated */
1377         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1378                 ClearPageUptodate(page);
1379                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1380                 up_read(&sbi->node_write);
1381                 unlock_page(page);
1382                 return 0;
1383         }
1384
1385         if (atomic && !test_opt(sbi, NOBARRIER))
1386                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1387
1388         set_page_writeback(page);
1389         fio.old_blkaddr = ni.blk_addr;
1390         write_node_page(nid, &fio);
1391         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1392         dec_page_count(sbi, F2FS_DIRTY_NODES);
1393         up_read(&sbi->node_write);
1394
1395         if (wbc->for_reclaim) {
1396                 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1397                                                 page->index, NODE);
1398                 submitted = NULL;
1399         }
1400
1401         unlock_page(page);
1402
1403         if (unlikely(f2fs_cp_error(sbi))) {
1404                 f2fs_submit_merged_write(sbi, NODE);
1405                 submitted = NULL;
1406         }
1407         if (submitted)
1408                 *submitted = fio.submitted;
1409
1410         if (do_balance)
1411                 f2fs_balance_fs(sbi, false);
1412         return 0;
1413
1414 redirty_out:
1415         redirty_page_for_writepage(wbc, page);
1416         return AOP_WRITEPAGE_ACTIVATE;
1417 }
1418
1419 static int f2fs_write_node_page(struct page *page,
1420                                 struct writeback_control *wbc)
1421 {
1422         return __write_node_page(page, false, NULL, wbc, false, FS_NODE_IO);
1423 }
1424
1425 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1426                         struct writeback_control *wbc, bool atomic)
1427 {
1428         pgoff_t index, end;
1429         pgoff_t last_idx = ULONG_MAX;
1430         struct pagevec pvec;
1431         int ret = 0;
1432         struct page *last_page = NULL;
1433         bool marked = false;
1434         nid_t ino = inode->i_ino;
1435
1436         if (atomic) {
1437                 last_page = last_fsync_dnode(sbi, ino);
1438                 if (IS_ERR_OR_NULL(last_page))
1439                         return PTR_ERR_OR_ZERO(last_page);
1440         }
1441 retry:
1442         pagevec_init(&pvec, 0);
1443         index = 0;
1444         end = ULONG_MAX;
1445
1446         while (index <= end) {
1447                 int i, nr_pages;
1448                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1449                                 PAGECACHE_TAG_DIRTY,
1450                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1451                 if (nr_pages == 0)
1452                         break;
1453
1454                 for (i = 0; i < nr_pages; i++) {
1455                         struct page *page = pvec.pages[i];
1456                         bool submitted = false;
1457
1458                         if (unlikely(f2fs_cp_error(sbi))) {
1459                                 f2fs_put_page(last_page, 0);
1460                                 pagevec_release(&pvec);
1461                                 ret = -EIO;
1462                                 goto out;
1463                         }
1464
1465                         if (!IS_DNODE(page) || !is_cold_node(page))
1466                                 continue;
1467                         if (ino_of_node(page) != ino)
1468                                 continue;
1469
1470                         lock_page(page);
1471
1472                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1473 continue_unlock:
1474                                 unlock_page(page);
1475                                 continue;
1476                         }
1477                         if (ino_of_node(page) != ino)
1478                                 goto continue_unlock;
1479
1480                         if (!PageDirty(page) && page != last_page) {
1481                                 /* someone wrote it for us */
1482                                 goto continue_unlock;
1483                         }
1484
1485                         f2fs_wait_on_page_writeback(page, NODE, true);
1486                         BUG_ON(PageWriteback(page));
1487
1488                         set_fsync_mark(page, 0);
1489                         set_dentry_mark(page, 0);
1490
1491                         if (!atomic || page == last_page) {
1492                                 set_fsync_mark(page, 1);
1493                                 if (IS_INODE(page)) {
1494                                         if (is_inode_flag_set(inode,
1495                                                                 FI_DIRTY_INODE))
1496                                                 update_inode(inode, page);
1497                                         set_dentry_mark(page,
1498                                                 need_dentry_mark(sbi, ino));
1499                                 }
1500                                 /*  may be written by other thread */
1501                                 if (!PageDirty(page))
1502                                         set_page_dirty(page);
1503                         }
1504
1505                         if (!clear_page_dirty_for_io(page))
1506                                 goto continue_unlock;
1507
1508                         ret = __write_node_page(page, atomic &&
1509                                                 page == last_page,
1510                                                 &submitted, wbc, true,
1511                                                 FS_NODE_IO);
1512                         if (ret) {
1513                                 unlock_page(page);
1514                                 f2fs_put_page(last_page, 0);
1515                                 break;
1516                         } else if (submitted) {
1517                                 last_idx = page->index;
1518                         }
1519
1520                         if (page == last_page) {
1521                                 f2fs_put_page(page, 0);
1522                                 marked = true;
1523                                 break;
1524                         }
1525                 }
1526                 pagevec_release(&pvec);
1527                 cond_resched();
1528
1529                 if (ret || marked)
1530                         break;
1531         }
1532         if (!ret && atomic && !marked) {
1533                 f2fs_msg(sbi->sb, KERN_DEBUG,
1534                         "Retry to write fsync mark: ino=%u, idx=%lx",
1535                                         ino, last_page->index);
1536                 lock_page(last_page);
1537                 f2fs_wait_on_page_writeback(last_page, NODE, true);
1538                 set_page_dirty(last_page);
1539                 unlock_page(last_page);
1540                 goto retry;
1541         }
1542 out:
1543         if (last_idx != ULONG_MAX)
1544                 f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1545         return ret ? -EIO: 0;
1546 }
1547
1548 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc,
1549                                 bool do_balance, enum iostat_type io_type)
1550 {
1551         pgoff_t index, end;
1552         struct pagevec pvec;
1553         int step = 0;
1554         int nwritten = 0;
1555         int ret = 0;
1556
1557         pagevec_init(&pvec, 0);
1558
1559 next_step:
1560         index = 0;
1561         end = ULONG_MAX;
1562
1563         while (index <= end) {
1564                 int i, nr_pages;
1565                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1566                                 PAGECACHE_TAG_DIRTY,
1567                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1568                 if (nr_pages == 0)
1569                         break;
1570
1571                 for (i = 0; i < nr_pages; i++) {
1572                         struct page *page = pvec.pages[i];
1573                         bool submitted = false;
1574
1575                         if (unlikely(f2fs_cp_error(sbi))) {
1576                                 pagevec_release(&pvec);
1577                                 ret = -EIO;
1578                                 goto out;
1579                         }
1580
1581                         /*
1582                          * flushing sequence with step:
1583                          * 0. indirect nodes
1584                          * 1. dentry dnodes
1585                          * 2. file dnodes
1586                          */
1587                         if (step == 0 && IS_DNODE(page))
1588                                 continue;
1589                         if (step == 1 && (!IS_DNODE(page) ||
1590                                                 is_cold_node(page)))
1591                                 continue;
1592                         if (step == 2 && (!IS_DNODE(page) ||
1593                                                 !is_cold_node(page)))
1594                                 continue;
1595 lock_node:
1596                         if (!trylock_page(page))
1597                                 continue;
1598
1599                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1600 continue_unlock:
1601                                 unlock_page(page);
1602                                 continue;
1603                         }
1604
1605                         if (!PageDirty(page)) {
1606                                 /* someone wrote it for us */
1607                                 goto continue_unlock;
1608                         }
1609
1610                         /* flush inline_data */
1611                         if (is_inline_node(page)) {
1612                                 clear_inline_node(page);
1613                                 unlock_page(page);
1614                                 flush_inline_data(sbi, ino_of_node(page));
1615                                 goto lock_node;
1616                         }
1617
1618                         f2fs_wait_on_page_writeback(page, NODE, true);
1619
1620                         BUG_ON(PageWriteback(page));
1621                         if (!clear_page_dirty_for_io(page))
1622                                 goto continue_unlock;
1623
1624                         set_fsync_mark(page, 0);
1625                         set_dentry_mark(page, 0);
1626
1627                         ret = __write_node_page(page, false, &submitted,
1628                                                 wbc, do_balance, io_type);
1629                         if (ret)
1630                                 unlock_page(page);
1631                         else if (submitted)
1632                                 nwritten++;
1633
1634                         if (--wbc->nr_to_write == 0)
1635                                 break;
1636                 }
1637                 pagevec_release(&pvec);
1638                 cond_resched();
1639
1640                 if (wbc->nr_to_write == 0) {
1641                         step = 2;
1642                         break;
1643                 }
1644         }
1645
1646         if (step < 2) {
1647                 step++;
1648                 goto next_step;
1649         }
1650 out:
1651         if (nwritten)
1652                 f2fs_submit_merged_write(sbi, NODE);
1653         return ret;
1654 }
1655
1656 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1657 {
1658         pgoff_t index = 0, end = ULONG_MAX;
1659         struct pagevec pvec;
1660         int ret2, ret = 0;
1661
1662         pagevec_init(&pvec, 0);
1663
1664         while (index <= end) {
1665                 int i, nr_pages;
1666                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1667                                 PAGECACHE_TAG_WRITEBACK,
1668                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1669                 if (nr_pages == 0)
1670                         break;
1671
1672                 for (i = 0; i < nr_pages; i++) {
1673                         struct page *page = pvec.pages[i];
1674
1675                         /* until radix tree lookup accepts end_index */
1676                         if (unlikely(page->index > end))
1677                                 continue;
1678
1679                         if (ino && ino_of_node(page) == ino) {
1680                                 f2fs_wait_on_page_writeback(page, NODE, true);
1681                                 if (TestClearPageError(page))
1682                                         ret = -EIO;
1683                         }
1684                 }
1685                 pagevec_release(&pvec);
1686                 cond_resched();
1687         }
1688
1689         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1690         if (!ret)
1691                 ret = ret2;
1692         return ret;
1693 }
1694
1695 static int f2fs_write_node_pages(struct address_space *mapping,
1696                             struct writeback_control *wbc)
1697 {
1698         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1699         struct blk_plug plug;
1700         long diff;
1701
1702         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1703                 goto skip_write;
1704
1705         /* balancing f2fs's metadata in background */
1706         f2fs_balance_fs_bg(sbi);
1707
1708         /* collect a number of dirty node pages and write together */
1709         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1710                 goto skip_write;
1711
1712         trace_f2fs_writepages(mapping->host, wbc, NODE);
1713
1714         diff = nr_pages_to_write(sbi, NODE, wbc);
1715         wbc->sync_mode = WB_SYNC_NONE;
1716         blk_start_plug(&plug);
1717         sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1718         blk_finish_plug(&plug);
1719         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1720         return 0;
1721
1722 skip_write:
1723         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1724         trace_f2fs_writepages(mapping->host, wbc, NODE);
1725         return 0;
1726 }
1727
1728 static int f2fs_set_node_page_dirty(struct page *page)
1729 {
1730         trace_f2fs_set_page_dirty(page, NODE);
1731
1732         if (!PageUptodate(page))
1733                 SetPageUptodate(page);
1734         if (!PageDirty(page)) {
1735                 f2fs_set_page_dirty_nobuffers(page);
1736                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1737                 SetPagePrivate(page);
1738                 f2fs_trace_pid(page);
1739                 return 1;
1740         }
1741         return 0;
1742 }
1743
1744 /*
1745  * Structure of the f2fs node operations
1746  */
1747 const struct address_space_operations f2fs_node_aops = {
1748         .writepage      = f2fs_write_node_page,
1749         .writepages     = f2fs_write_node_pages,
1750         .set_page_dirty = f2fs_set_node_page_dirty,
1751         .invalidatepage = f2fs_invalidate_page,
1752         .releasepage    = f2fs_release_page,
1753 #ifdef CONFIG_MIGRATION
1754         .migratepage    = f2fs_migrate_page,
1755 #endif
1756 };
1757
1758 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1759                                                 nid_t n)
1760 {
1761         return radix_tree_lookup(&nm_i->free_nid_root, n);
1762 }
1763
1764 static int __insert_nid_to_list(struct f2fs_sb_info *sbi,
1765                         struct free_nid *i, enum nid_list list, bool new)
1766 {
1767         struct f2fs_nm_info *nm_i = NM_I(sbi);
1768
1769         if (new) {
1770                 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1771                 if (err)
1772                         return err;
1773         }
1774
1775         f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1776                                                 i->state != NID_ALLOC);
1777         nm_i->nid_cnt[list]++;
1778         list_add_tail(&i->list, &nm_i->nid_list[list]);
1779         return 0;
1780 }
1781
1782 static void __remove_nid_from_list(struct f2fs_sb_info *sbi,
1783                         struct free_nid *i, enum nid_list list, bool reuse)
1784 {
1785         struct f2fs_nm_info *nm_i = NM_I(sbi);
1786
1787         f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1788                                                 i->state != NID_ALLOC);
1789         nm_i->nid_cnt[list]--;
1790         list_del(&i->list);
1791         if (!reuse)
1792                 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1793 }
1794
1795 /* return if the nid is recognized as free */
1796 static bool add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1797 {
1798         struct f2fs_nm_info *nm_i = NM_I(sbi);
1799         struct free_nid *i, *e;
1800         struct nat_entry *ne;
1801         int err = -EINVAL;
1802         bool ret = false;
1803
1804         /* 0 nid should not be used */
1805         if (unlikely(nid == 0))
1806                 return false;
1807
1808         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1809         i->nid = nid;
1810         i->state = NID_NEW;
1811
1812         if (radix_tree_preload(GFP_NOFS))
1813                 goto err;
1814
1815         spin_lock(&nm_i->nid_list_lock);
1816
1817         if (build) {
1818                 /*
1819                  *   Thread A             Thread B
1820                  *  - f2fs_create
1821                  *   - f2fs_new_inode
1822                  *    - alloc_nid
1823                  *     - __insert_nid_to_list(ALLOC_NID_LIST)
1824                  *                     - f2fs_balance_fs_bg
1825                  *                      - build_free_nids
1826                  *                       - __build_free_nids
1827                  *                        - scan_nat_page
1828                  *                         - add_free_nid
1829                  *                          - __lookup_nat_cache
1830                  *  - f2fs_add_link
1831                  *   - init_inode_metadata
1832                  *    - new_inode_page
1833                  *     - new_node_page
1834                  *      - set_node_addr
1835                  *  - alloc_nid_done
1836                  *   - __remove_nid_from_list(ALLOC_NID_LIST)
1837                  *                         - __insert_nid_to_list(FREE_NID_LIST)
1838                  */
1839                 ne = __lookup_nat_cache(nm_i, nid);
1840                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1841                                 nat_get_blkaddr(ne) != NULL_ADDR))
1842                         goto err_out;
1843
1844                 e = __lookup_free_nid_list(nm_i, nid);
1845                 if (e) {
1846                         if (e->state == NID_NEW)
1847                                 ret = true;
1848                         goto err_out;
1849                 }
1850         }
1851         ret = true;
1852         err = __insert_nid_to_list(sbi, i, FREE_NID_LIST, true);
1853 err_out:
1854         spin_unlock(&nm_i->nid_list_lock);
1855         radix_tree_preload_end();
1856 err:
1857         if (err)
1858                 kmem_cache_free(free_nid_slab, i);
1859         return ret;
1860 }
1861
1862 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1863 {
1864         struct f2fs_nm_info *nm_i = NM_I(sbi);
1865         struct free_nid *i;
1866         bool need_free = false;
1867
1868         spin_lock(&nm_i->nid_list_lock);
1869         i = __lookup_free_nid_list(nm_i, nid);
1870         if (i && i->state == NID_NEW) {
1871                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
1872                 need_free = true;
1873         }
1874         spin_unlock(&nm_i->nid_list_lock);
1875
1876         if (need_free)
1877                 kmem_cache_free(free_nid_slab, i);
1878 }
1879
1880 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
1881                                                         bool set, bool build)
1882 {
1883         struct f2fs_nm_info *nm_i = NM_I(sbi);
1884         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
1885         unsigned int nid_ofs = nid - START_NID(nid);
1886
1887         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1888                 return;
1889
1890         if (set)
1891                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1892         else
1893                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1894
1895         if (set)
1896                 nm_i->free_nid_count[nat_ofs]++;
1897         else if (!build)
1898                 nm_i->free_nid_count[nat_ofs]--;
1899 }
1900
1901 static void scan_nat_page(struct f2fs_sb_info *sbi,
1902                         struct page *nat_page, nid_t start_nid)
1903 {
1904         struct f2fs_nm_info *nm_i = NM_I(sbi);
1905         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1906         block_t blk_addr;
1907         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
1908         int i;
1909
1910         if (test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1911                 return;
1912
1913         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
1914
1915         i = start_nid % NAT_ENTRY_PER_BLOCK;
1916
1917         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1918                 bool freed = false;
1919
1920                 if (unlikely(start_nid >= nm_i->max_nid))
1921                         break;
1922
1923                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1924                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1925                 if (blk_addr == NULL_ADDR)
1926                         freed = add_free_nid(sbi, start_nid, true);
1927                 spin_lock(&NM_I(sbi)->nid_list_lock);
1928                 update_free_nid_bitmap(sbi, start_nid, freed, true);
1929                 spin_unlock(&NM_I(sbi)->nid_list_lock);
1930         }
1931 }
1932
1933 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
1934 {
1935         struct f2fs_nm_info *nm_i = NM_I(sbi);
1936         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1937         struct f2fs_journal *journal = curseg->journal;
1938         unsigned int i, idx;
1939
1940         down_read(&nm_i->nat_tree_lock);
1941
1942         for (i = 0; i < nm_i->nat_blocks; i++) {
1943                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
1944                         continue;
1945                 if (!nm_i->free_nid_count[i])
1946                         continue;
1947                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
1948                         nid_t nid;
1949
1950                         if (!test_bit_le(idx, nm_i->free_nid_bitmap[i]))
1951                                 continue;
1952
1953                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
1954                         add_free_nid(sbi, nid, true);
1955
1956                         if (nm_i->nid_cnt[FREE_NID_LIST] >= MAX_FREE_NIDS)
1957                                 goto out;
1958                 }
1959         }
1960 out:
1961         down_read(&curseg->journal_rwsem);
1962         for (i = 0; i < nats_in_cursum(journal); i++) {
1963                 block_t addr;
1964                 nid_t nid;
1965
1966                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1967                 nid = le32_to_cpu(nid_in_journal(journal, i));
1968                 if (addr == NULL_ADDR)
1969                         add_free_nid(sbi, nid, true);
1970                 else
1971                         remove_free_nid(sbi, nid);
1972         }
1973         up_read(&curseg->journal_rwsem);
1974         up_read(&nm_i->nat_tree_lock);
1975 }
1976
1977 static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
1978 {
1979         struct f2fs_nm_info *nm_i = NM_I(sbi);
1980         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1981         struct f2fs_journal *journal = curseg->journal;
1982         int i = 0;
1983         nid_t nid = nm_i->next_scan_nid;
1984
1985         if (unlikely(nid >= nm_i->max_nid))
1986                 nid = 0;
1987
1988         /* Enough entries */
1989         if (nm_i->nid_cnt[FREE_NID_LIST] >= NAT_ENTRY_PER_BLOCK)
1990                 return;
1991
1992         if (!sync && !available_free_memory(sbi, FREE_NIDS))
1993                 return;
1994
1995         if (!mount) {
1996                 /* try to find free nids in free_nid_bitmap */
1997                 scan_free_nid_bits(sbi);
1998
1999                 if (nm_i->nid_cnt[FREE_NID_LIST])
2000                         return;
2001         }
2002
2003         /* readahead nat pages to be scanned */
2004         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2005                                                         META_NAT, true);
2006
2007         down_read(&nm_i->nat_tree_lock);
2008
2009         while (1) {
2010                 struct page *page = get_current_nat_page(sbi, nid);
2011
2012                 scan_nat_page(sbi, page, nid);
2013                 f2fs_put_page(page, 1);
2014
2015                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2016                 if (unlikely(nid >= nm_i->max_nid))
2017                         nid = 0;
2018
2019                 if (++i >= FREE_NID_PAGES)
2020                         break;
2021         }
2022
2023         /* go to the next free nat pages to find free nids abundantly */
2024         nm_i->next_scan_nid = nid;
2025
2026         /* find free nids from current sum_pages */
2027         down_read(&curseg->journal_rwsem);
2028         for (i = 0; i < nats_in_cursum(journal); i++) {
2029                 block_t addr;
2030
2031                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2032                 nid = le32_to_cpu(nid_in_journal(journal, i));
2033                 if (addr == NULL_ADDR)
2034                         add_free_nid(sbi, nid, true);
2035                 else
2036                         remove_free_nid(sbi, nid);
2037         }
2038         up_read(&curseg->journal_rwsem);
2039         up_read(&nm_i->nat_tree_lock);
2040
2041         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2042                                         nm_i->ra_nid_pages, META_NAT, false);
2043 }
2044
2045 void build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2046 {
2047         mutex_lock(&NM_I(sbi)->build_lock);
2048         __build_free_nids(sbi, sync, mount);
2049         mutex_unlock(&NM_I(sbi)->build_lock);
2050 }
2051
2052 /*
2053  * If this function returns success, caller can obtain a new nid
2054  * from second parameter of this function.
2055  * The returned nid could be used ino as well as nid when inode is created.
2056  */
2057 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2058 {
2059         struct f2fs_nm_info *nm_i = NM_I(sbi);
2060         struct free_nid *i = NULL;
2061 retry:
2062 #ifdef CONFIG_F2FS_FAULT_INJECTION
2063         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2064                 f2fs_show_injection_info(FAULT_ALLOC_NID);
2065                 return false;
2066         }
2067 #endif
2068         spin_lock(&nm_i->nid_list_lock);
2069
2070         if (unlikely(nm_i->available_nids == 0)) {
2071                 spin_unlock(&nm_i->nid_list_lock);
2072                 return false;
2073         }
2074
2075         /* We should not use stale free nids created by build_free_nids */
2076         if (nm_i->nid_cnt[FREE_NID_LIST] && !on_build_free_nids(nm_i)) {
2077                 f2fs_bug_on(sbi, list_empty(&nm_i->nid_list[FREE_NID_LIST]));
2078                 i = list_first_entry(&nm_i->nid_list[FREE_NID_LIST],
2079                                         struct free_nid, list);
2080                 *nid = i->nid;
2081
2082                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, true);
2083                 i->state = NID_ALLOC;
2084                 __insert_nid_to_list(sbi, i, ALLOC_NID_LIST, false);
2085                 nm_i->available_nids--;
2086
2087                 update_free_nid_bitmap(sbi, *nid, false, false);
2088
2089                 spin_unlock(&nm_i->nid_list_lock);
2090                 return true;
2091         }
2092         spin_unlock(&nm_i->nid_list_lock);
2093
2094         /* Let's scan nat pages and its caches to get free nids */
2095         build_free_nids(sbi, true, false);
2096         goto retry;
2097 }
2098
2099 /*
2100  * alloc_nid() should be called prior to this function.
2101  */
2102 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2103 {
2104         struct f2fs_nm_info *nm_i = NM_I(sbi);
2105         struct free_nid *i;
2106
2107         spin_lock(&nm_i->nid_list_lock);
2108         i = __lookup_free_nid_list(nm_i, nid);
2109         f2fs_bug_on(sbi, !i);
2110         __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
2111         spin_unlock(&nm_i->nid_list_lock);
2112
2113         kmem_cache_free(free_nid_slab, i);
2114 }
2115
2116 /*
2117  * alloc_nid() should be called prior to this function.
2118  */
2119 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2120 {
2121         struct f2fs_nm_info *nm_i = NM_I(sbi);
2122         struct free_nid *i;
2123         bool need_free = false;
2124
2125         if (!nid)
2126                 return;
2127
2128         spin_lock(&nm_i->nid_list_lock);
2129         i = __lookup_free_nid_list(nm_i, nid);
2130         f2fs_bug_on(sbi, !i);
2131
2132         if (!available_free_memory(sbi, FREE_NIDS)) {
2133                 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
2134                 need_free = true;
2135         } else {
2136                 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, true);
2137                 i->state = NID_NEW;
2138                 __insert_nid_to_list(sbi, i, FREE_NID_LIST, false);
2139         }
2140
2141         nm_i->available_nids++;
2142
2143         update_free_nid_bitmap(sbi, nid, true, false);
2144
2145         spin_unlock(&nm_i->nid_list_lock);
2146
2147         if (need_free)
2148                 kmem_cache_free(free_nid_slab, i);
2149 }
2150
2151 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2152 {
2153         struct f2fs_nm_info *nm_i = NM_I(sbi);
2154         struct free_nid *i, *next;
2155         int nr = nr_shrink;
2156
2157         if (nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
2158                 return 0;
2159
2160         if (!mutex_trylock(&nm_i->build_lock))
2161                 return 0;
2162
2163         spin_lock(&nm_i->nid_list_lock);
2164         list_for_each_entry_safe(i, next, &nm_i->nid_list[FREE_NID_LIST],
2165                                                                         list) {
2166                 if (nr_shrink <= 0 ||
2167                                 nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
2168                         break;
2169
2170                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2171                 kmem_cache_free(free_nid_slab, i);
2172                 nr_shrink--;
2173         }
2174         spin_unlock(&nm_i->nid_list_lock);
2175         mutex_unlock(&nm_i->build_lock);
2176
2177         return nr - nr_shrink;
2178 }
2179
2180 void recover_inline_xattr(struct inode *inode, struct page *page)
2181 {
2182         void *src_addr, *dst_addr;
2183         size_t inline_size;
2184         struct page *ipage;
2185         struct f2fs_inode *ri;
2186
2187         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2188         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2189
2190         ri = F2FS_INODE(page);
2191         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2192                 clear_inode_flag(inode, FI_INLINE_XATTR);
2193                 goto update_inode;
2194         }
2195
2196         dst_addr = inline_xattr_addr(ipage);
2197         src_addr = inline_xattr_addr(page);
2198         inline_size = inline_xattr_size(inode);
2199
2200         f2fs_wait_on_page_writeback(ipage, NODE, true);
2201         memcpy(dst_addr, src_addr, inline_size);
2202 update_inode:
2203         update_inode(inode, ipage);
2204         f2fs_put_page(ipage, 1);
2205 }
2206
2207 int recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2208 {
2209         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2210         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2211         nid_t new_xnid;
2212         struct dnode_of_data dn;
2213         struct node_info ni;
2214         struct page *xpage;
2215
2216         if (!prev_xnid)
2217                 goto recover_xnid;
2218
2219         /* 1: invalidate the previous xattr nid */
2220         get_node_info(sbi, prev_xnid, &ni);
2221         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2222         invalidate_blocks(sbi, ni.blk_addr);
2223         dec_valid_node_count(sbi, inode, false);
2224         set_node_addr(sbi, &ni, NULL_ADDR, false);
2225
2226 recover_xnid:
2227         /* 2: update xattr nid in inode */
2228         if (!alloc_nid(sbi, &new_xnid))
2229                 return -ENOSPC;
2230
2231         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2232         xpage = new_node_page(&dn, XATTR_NODE_OFFSET);
2233         if (IS_ERR(xpage)) {
2234                 alloc_nid_failed(sbi, new_xnid);
2235                 return PTR_ERR(xpage);
2236         }
2237
2238         alloc_nid_done(sbi, new_xnid);
2239         update_inode_page(inode);
2240
2241         /* 3: update and set xattr node page dirty */
2242         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2243
2244         set_page_dirty(xpage);
2245         f2fs_put_page(xpage, 1);
2246
2247         return 0;
2248 }
2249
2250 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2251 {
2252         struct f2fs_inode *src, *dst;
2253         nid_t ino = ino_of_node(page);
2254         struct node_info old_ni, new_ni;
2255         struct page *ipage;
2256
2257         get_node_info(sbi, ino, &old_ni);
2258
2259         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2260                 return -EINVAL;
2261 retry:
2262         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2263         if (!ipage) {
2264                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2265                 goto retry;
2266         }
2267
2268         /* Should not use this inode from free nid list */
2269         remove_free_nid(sbi, ino);
2270
2271         if (!PageUptodate(ipage))
2272                 SetPageUptodate(ipage);
2273         fill_node_footer(ipage, ino, ino, 0, true);
2274
2275         src = F2FS_INODE(page);
2276         dst = F2FS_INODE(ipage);
2277
2278         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2279         dst->i_size = 0;
2280         dst->i_blocks = cpu_to_le64(1);
2281         dst->i_links = cpu_to_le32(1);
2282         dst->i_xattr_nid = 0;
2283         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2284         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2285                 dst->i_extra_isize = src->i_extra_isize;
2286                 if (f2fs_sb_has_project_quota(sbi->sb) &&
2287                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2288                                                                 i_projid))
2289                         dst->i_projid = src->i_projid;
2290         }
2291
2292         new_ni = old_ni;
2293         new_ni.ino = ino;
2294
2295         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2296                 WARN_ON(1);
2297         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2298         inc_valid_inode_count(sbi);
2299         set_page_dirty(ipage);
2300         f2fs_put_page(ipage, 1);
2301         return 0;
2302 }
2303
2304 int restore_node_summary(struct f2fs_sb_info *sbi,
2305                         unsigned int segno, struct f2fs_summary_block *sum)
2306 {
2307         struct f2fs_node *rn;
2308         struct f2fs_summary *sum_entry;
2309         block_t addr;
2310         int i, idx, last_offset, nrpages;
2311
2312         /* scan the node segment */
2313         last_offset = sbi->blocks_per_seg;
2314         addr = START_BLOCK(sbi, segno);
2315         sum_entry = &sum->entries[0];
2316
2317         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2318                 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2319
2320                 /* readahead node pages */
2321                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2322
2323                 for (idx = addr; idx < addr + nrpages; idx++) {
2324                         struct page *page = get_tmp_page(sbi, idx);
2325
2326                         rn = F2FS_NODE(page);
2327                         sum_entry->nid = rn->footer.nid;
2328                         sum_entry->version = 0;
2329                         sum_entry->ofs_in_node = 0;
2330                         sum_entry++;
2331                         f2fs_put_page(page, 1);
2332                 }
2333
2334                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2335                                                         addr + nrpages);
2336         }
2337         return 0;
2338 }
2339
2340 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2341 {
2342         struct f2fs_nm_info *nm_i = NM_I(sbi);
2343         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2344         struct f2fs_journal *journal = curseg->journal;
2345         int i;
2346
2347         down_write(&curseg->journal_rwsem);
2348         for (i = 0; i < nats_in_cursum(journal); i++) {
2349                 struct nat_entry *ne;
2350                 struct f2fs_nat_entry raw_ne;
2351                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2352
2353                 raw_ne = nat_in_journal(journal, i);
2354
2355                 ne = __lookup_nat_cache(nm_i, nid);
2356                 if (!ne) {
2357                         ne = grab_nat_entry(nm_i, nid, true);
2358                         node_info_from_raw_nat(&ne->ni, &raw_ne);
2359                 }
2360
2361                 /*
2362                  * if a free nat in journal has not been used after last
2363                  * checkpoint, we should remove it from available nids,
2364                  * since later we will add it again.
2365                  */
2366                 if (!get_nat_flag(ne, IS_DIRTY) &&
2367                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2368                         spin_lock(&nm_i->nid_list_lock);
2369                         nm_i->available_nids--;
2370                         spin_unlock(&nm_i->nid_list_lock);
2371                 }
2372
2373                 __set_nat_cache_dirty(nm_i, ne);
2374         }
2375         update_nats_in_cursum(journal, -i);
2376         up_write(&curseg->journal_rwsem);
2377 }
2378
2379 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2380                                                 struct list_head *head, int max)
2381 {
2382         struct nat_entry_set *cur;
2383
2384         if (nes->entry_cnt >= max)
2385                 goto add_out;
2386
2387         list_for_each_entry(cur, head, set_list) {
2388                 if (cur->entry_cnt >= nes->entry_cnt) {
2389                         list_add(&nes->set_list, cur->set_list.prev);
2390                         return;
2391                 }
2392         }
2393 add_out:
2394         list_add_tail(&nes->set_list, head);
2395 }
2396
2397 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2398                                                 struct page *page)
2399 {
2400         struct f2fs_nm_info *nm_i = NM_I(sbi);
2401         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2402         struct f2fs_nat_block *nat_blk = page_address(page);
2403         int valid = 0;
2404         int i;
2405
2406         if (!enabled_nat_bits(sbi, NULL))
2407                 return;
2408
2409         for (i = 0; i < NAT_ENTRY_PER_BLOCK; i++) {
2410                 if (start_nid == 0 && i == 0)
2411                         valid++;
2412                 if (nat_blk->entries[i].block_addr)
2413                         valid++;
2414         }
2415         if (valid == 0) {
2416                 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2417                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2418                 return;
2419         }
2420
2421         __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2422         if (valid == NAT_ENTRY_PER_BLOCK)
2423                 __set_bit_le(nat_index, nm_i->full_nat_bits);
2424         else
2425                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2426 }
2427
2428 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2429                 struct nat_entry_set *set, struct cp_control *cpc)
2430 {
2431         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2432         struct f2fs_journal *journal = curseg->journal;
2433         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2434         bool to_journal = true;
2435         struct f2fs_nat_block *nat_blk;
2436         struct nat_entry *ne, *cur;
2437         struct page *page = NULL;
2438
2439         /*
2440          * there are two steps to flush nat entries:
2441          * #1, flush nat entries to journal in current hot data summary block.
2442          * #2, flush nat entries to nat page.
2443          */
2444         if (enabled_nat_bits(sbi, cpc) ||
2445                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2446                 to_journal = false;
2447
2448         if (to_journal) {
2449                 down_write(&curseg->journal_rwsem);
2450         } else {
2451                 page = get_next_nat_page(sbi, start_nid);
2452                 nat_blk = page_address(page);
2453                 f2fs_bug_on(sbi, !nat_blk);
2454         }
2455
2456         /* flush dirty nats in nat entry set */
2457         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2458                 struct f2fs_nat_entry *raw_ne;
2459                 nid_t nid = nat_get_nid(ne);
2460                 int offset;
2461
2462                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2463
2464                 if (to_journal) {
2465                         offset = lookup_journal_in_cursum(journal,
2466                                                         NAT_JOURNAL, nid, 1);
2467                         f2fs_bug_on(sbi, offset < 0);
2468                         raw_ne = &nat_in_journal(journal, offset);
2469                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2470                 } else {
2471                         raw_ne = &nat_blk->entries[nid - start_nid];
2472                 }
2473                 raw_nat_from_node_info(raw_ne, &ne->ni);
2474                 nat_reset_flag(ne);
2475                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2476                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2477                         add_free_nid(sbi, nid, false);
2478                         spin_lock(&NM_I(sbi)->nid_list_lock);
2479                         NM_I(sbi)->available_nids++;
2480                         update_free_nid_bitmap(sbi, nid, true, false);
2481                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2482                 } else {
2483                         spin_lock(&NM_I(sbi)->nid_list_lock);
2484                         update_free_nid_bitmap(sbi, nid, false, false);
2485                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2486                 }
2487         }
2488
2489         if (to_journal) {
2490                 up_write(&curseg->journal_rwsem);
2491         } else {
2492                 __update_nat_bits(sbi, start_nid, page);
2493                 f2fs_put_page(page, 1);
2494         }
2495
2496         /* Allow dirty nats by node block allocation in write_begin */
2497         if (!set->entry_cnt) {
2498                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2499                 kmem_cache_free(nat_entry_set_slab, set);
2500         }
2501 }
2502
2503 /*
2504  * This function is called during the checkpointing process.
2505  */
2506 void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2507 {
2508         struct f2fs_nm_info *nm_i = NM_I(sbi);
2509         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2510         struct f2fs_journal *journal = curseg->journal;
2511         struct nat_entry_set *setvec[SETVEC_SIZE];
2512         struct nat_entry_set *set, *tmp;
2513         unsigned int found;
2514         nid_t set_idx = 0;
2515         LIST_HEAD(sets);
2516
2517         if (!nm_i->dirty_nat_cnt)
2518                 return;
2519
2520         down_write(&nm_i->nat_tree_lock);
2521
2522         /*
2523          * if there are no enough space in journal to store dirty nat
2524          * entries, remove all entries from journal and merge them
2525          * into nat entry set.
2526          */
2527         if (enabled_nat_bits(sbi, cpc) ||
2528                 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2529                 remove_nats_in_journal(sbi);
2530
2531         while ((found = __gang_lookup_nat_set(nm_i,
2532                                         set_idx, SETVEC_SIZE, setvec))) {
2533                 unsigned idx;
2534                 set_idx = setvec[found - 1]->set + 1;
2535                 for (idx = 0; idx < found; idx++)
2536                         __adjust_nat_entry_set(setvec[idx], &sets,
2537                                                 MAX_NAT_JENTRIES(journal));
2538         }
2539
2540         /* flush dirty nats in nat entry set */
2541         list_for_each_entry_safe(set, tmp, &sets, set_list)
2542                 __flush_nat_entry_set(sbi, set, cpc);
2543
2544         up_write(&nm_i->nat_tree_lock);
2545         /* Allow dirty nats by node block allocation in write_begin */
2546 }
2547
2548 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2549 {
2550         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2551         struct f2fs_nm_info *nm_i = NM_I(sbi);
2552         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2553         unsigned int i;
2554         __u64 cp_ver = cur_cp_version(ckpt);
2555         block_t nat_bits_addr;
2556
2557         if (!enabled_nat_bits(sbi, NULL))
2558                 return 0;
2559
2560         nm_i->nat_bits_blocks = F2FS_BYTES_TO_BLK((nat_bits_bytes << 1) + 8 +
2561                                                 F2FS_BLKSIZE - 1);
2562         nm_i->nat_bits = kzalloc(nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS,
2563                                                 GFP_KERNEL);
2564         if (!nm_i->nat_bits)
2565                 return -ENOMEM;
2566
2567         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2568                                                 nm_i->nat_bits_blocks;
2569         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2570                 struct page *page = get_meta_page(sbi, nat_bits_addr++);
2571
2572                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2573                                         page_address(page), F2FS_BLKSIZE);
2574                 f2fs_put_page(page, 1);
2575         }
2576
2577         cp_ver |= (cur_cp_crc(ckpt) << 32);
2578         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2579                 disable_nat_bits(sbi, true);
2580                 return 0;
2581         }
2582
2583         nm_i->full_nat_bits = nm_i->nat_bits + 8;
2584         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2585
2586         f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2587         return 0;
2588 }
2589
2590 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2591 {
2592         struct f2fs_nm_info *nm_i = NM_I(sbi);
2593         unsigned int i = 0;
2594         nid_t nid, last_nid;
2595
2596         if (!enabled_nat_bits(sbi, NULL))
2597                 return;
2598
2599         for (i = 0; i < nm_i->nat_blocks; i++) {
2600                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2601                 if (i >= nm_i->nat_blocks)
2602                         break;
2603
2604                 __set_bit_le(i, nm_i->nat_block_bitmap);
2605
2606                 nid = i * NAT_ENTRY_PER_BLOCK;
2607                 last_nid = (i + 1) * NAT_ENTRY_PER_BLOCK;
2608
2609                 spin_lock(&NM_I(sbi)->nid_list_lock);
2610                 for (; nid < last_nid; nid++)
2611                         update_free_nid_bitmap(sbi, nid, true, true);
2612                 spin_unlock(&NM_I(sbi)->nid_list_lock);
2613         }
2614
2615         for (i = 0; i < nm_i->nat_blocks; i++) {
2616                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2617                 if (i >= nm_i->nat_blocks)
2618                         break;
2619
2620                 __set_bit_le(i, nm_i->nat_block_bitmap);
2621         }
2622 }
2623
2624 static int init_node_manager(struct f2fs_sb_info *sbi)
2625 {
2626         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2627         struct f2fs_nm_info *nm_i = NM_I(sbi);
2628         unsigned char *version_bitmap;
2629         unsigned int nat_segs;
2630         int err;
2631
2632         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2633
2634         /* segment_count_nat includes pair segment so divide to 2. */
2635         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2636         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2637         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2638
2639         /* not used nids: 0, node, meta, (and root counted as valid node) */
2640         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2641                                                         F2FS_RESERVED_NODE_NUM;
2642         nm_i->nid_cnt[FREE_NID_LIST] = 0;
2643         nm_i->nid_cnt[ALLOC_NID_LIST] = 0;
2644         nm_i->nat_cnt = 0;
2645         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2646         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2647         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2648
2649         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2650         INIT_LIST_HEAD(&nm_i->nid_list[FREE_NID_LIST]);
2651         INIT_LIST_HEAD(&nm_i->nid_list[ALLOC_NID_LIST]);
2652         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2653         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2654         INIT_LIST_HEAD(&nm_i->nat_entries);
2655
2656         mutex_init(&nm_i->build_lock);
2657         spin_lock_init(&nm_i->nid_list_lock);
2658         init_rwsem(&nm_i->nat_tree_lock);
2659
2660         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2661         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2662         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2663         if (!version_bitmap)
2664                 return -EFAULT;
2665
2666         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2667                                         GFP_KERNEL);
2668         if (!nm_i->nat_bitmap)
2669                 return -ENOMEM;
2670
2671         err = __get_nat_bitmaps(sbi);
2672         if (err)
2673                 return err;
2674
2675 #ifdef CONFIG_F2FS_CHECK_FS
2676         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2677                                         GFP_KERNEL);
2678         if (!nm_i->nat_bitmap_mir)
2679                 return -ENOMEM;
2680 #endif
2681
2682         return 0;
2683 }
2684
2685 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2686 {
2687         struct f2fs_nm_info *nm_i = NM_I(sbi);
2688
2689         nm_i->free_nid_bitmap = kvzalloc(nm_i->nat_blocks *
2690                                         NAT_ENTRY_BITMAP_SIZE, GFP_KERNEL);
2691         if (!nm_i->free_nid_bitmap)
2692                 return -ENOMEM;
2693
2694         nm_i->nat_block_bitmap = kvzalloc(nm_i->nat_blocks / 8,
2695                                                                 GFP_KERNEL);
2696         if (!nm_i->nat_block_bitmap)
2697                 return -ENOMEM;
2698
2699         nm_i->free_nid_count = kvzalloc(nm_i->nat_blocks *
2700                                         sizeof(unsigned short), GFP_KERNEL);
2701         if (!nm_i->free_nid_count)
2702                 return -ENOMEM;
2703         return 0;
2704 }
2705
2706 int build_node_manager(struct f2fs_sb_info *sbi)
2707 {
2708         int err;
2709
2710         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2711         if (!sbi->nm_info)
2712                 return -ENOMEM;
2713
2714         err = init_node_manager(sbi);
2715         if (err)
2716                 return err;
2717
2718         err = init_free_nid_cache(sbi);
2719         if (err)
2720                 return err;
2721
2722         /* load free nid status from nat_bits table */
2723         load_free_nid_bitmap(sbi);
2724
2725         build_free_nids(sbi, true, true);
2726         return 0;
2727 }
2728
2729 void destroy_node_manager(struct f2fs_sb_info *sbi)
2730 {
2731         struct f2fs_nm_info *nm_i = NM_I(sbi);
2732         struct free_nid *i, *next_i;
2733         struct nat_entry *natvec[NATVEC_SIZE];
2734         struct nat_entry_set *setvec[SETVEC_SIZE];
2735         nid_t nid = 0;
2736         unsigned int found;
2737
2738         if (!nm_i)
2739                 return;
2740
2741         /* destroy free nid list */
2742         spin_lock(&nm_i->nid_list_lock);
2743         list_for_each_entry_safe(i, next_i, &nm_i->nid_list[FREE_NID_LIST],
2744                                                                         list) {
2745                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2746                 spin_unlock(&nm_i->nid_list_lock);
2747                 kmem_cache_free(free_nid_slab, i);
2748                 spin_lock(&nm_i->nid_list_lock);
2749         }
2750         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID_LIST]);
2751         f2fs_bug_on(sbi, nm_i->nid_cnt[ALLOC_NID_LIST]);
2752         f2fs_bug_on(sbi, !list_empty(&nm_i->nid_list[ALLOC_NID_LIST]));
2753         spin_unlock(&nm_i->nid_list_lock);
2754
2755         /* destroy nat cache */
2756         down_write(&nm_i->nat_tree_lock);
2757         while ((found = __gang_lookup_nat_cache(nm_i,
2758                                         nid, NATVEC_SIZE, natvec))) {
2759                 unsigned idx;
2760
2761                 nid = nat_get_nid(natvec[found - 1]) + 1;
2762                 for (idx = 0; idx < found; idx++)
2763                         __del_from_nat_cache(nm_i, natvec[idx]);
2764         }
2765         f2fs_bug_on(sbi, nm_i->nat_cnt);
2766
2767         /* destroy nat set cache */
2768         nid = 0;
2769         while ((found = __gang_lookup_nat_set(nm_i,
2770                                         nid, SETVEC_SIZE, setvec))) {
2771                 unsigned idx;
2772
2773                 nid = setvec[found - 1]->set + 1;
2774                 for (idx = 0; idx < found; idx++) {
2775                         /* entry_cnt is not zero, when cp_error was occurred */
2776                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2777                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2778                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2779                 }
2780         }
2781         up_write(&nm_i->nat_tree_lock);
2782
2783         kvfree(nm_i->nat_block_bitmap);
2784         kvfree(nm_i->free_nid_bitmap);
2785         kvfree(nm_i->free_nid_count);
2786
2787         kfree(nm_i->nat_bitmap);
2788         kfree(nm_i->nat_bits);
2789 #ifdef CONFIG_F2FS_CHECK_FS
2790         kfree(nm_i->nat_bitmap_mir);
2791 #endif
2792         sbi->nm_info = NULL;
2793         kfree(nm_i);
2794 }
2795
2796 int __init create_node_manager_caches(void)
2797 {
2798         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2799                         sizeof(struct nat_entry));
2800         if (!nat_entry_slab)
2801                 goto fail;
2802
2803         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2804                         sizeof(struct free_nid));
2805         if (!free_nid_slab)
2806                 goto destroy_nat_entry;
2807
2808         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2809                         sizeof(struct nat_entry_set));
2810         if (!nat_entry_set_slab)
2811                 goto destroy_free_nid;
2812         return 0;
2813
2814 destroy_free_nid:
2815         kmem_cache_destroy(free_nid_slab);
2816 destroy_nat_entry:
2817         kmem_cache_destroy(nat_entry_slab);
2818 fail:
2819         return -ENOMEM;
2820 }
2821
2822 void destroy_node_manager_caches(void)
2823 {
2824         kmem_cache_destroy(nat_entry_set_slab);
2825         kmem_cache_destroy(free_nid_slab);
2826         kmem_cache_destroy(nat_entry_slab);
2827 }