fs/f2fs/node.c

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