fs/f2fs/node.c

   1 /*
   2  * fs/f2fs/node.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/mpage.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/swap.h>
  18
  19 #include "f2fs.h"
  20 #include "node.h"
  21 #include "segment.h"
  22 #include "trace.h"
  23 #include <trace/events/f2fs.h>
  24
  25 #define on_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
  26
  27 static struct kmem_cache *nat_entry_slab;
  28 static struct kmem_cache *free_nid_slab;
  29 static struct kmem_cache *nat_entry_set_slab;
  30
  31 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
  32 {
  33         struct f2fs_nm_info *nm_i = NM_I(sbi);
  34         struct sysinfo val;
  35         unsigned long avail_ram;
  36         unsigned long mem_size = 0;
  37         bool res = false;
  38
  39         si_meminfo(&val);
  40
  41         /* only uses low memory */
  42         avail_ram = val.totalram - val.totalhigh;
  43
  44         /*
  45          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  46          */
  47         if (type == FREE_NIDS) {
  48                 mem_size = (nm_i->nid_cnt[FREE_NID_LIST] *
  49                                 sizeof(struct free_nid)) >> PAGE_SHIFT;
  50                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  51         } else if (type == NAT_ENTRIES) {
  52                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
  53                                                         PAGE_SHIFT;
  54                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  55                 if (excess_cached_nats(sbi))
  56                         res = false;
  57         } else if (type == DIRTY_DENTS) {
  58                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  59                         return false;
  60                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  61                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  62         } else if (type == INO_ENTRIES) {
  63                 int i;
  64
  65                 for (i = 0; i <= UPDATE_INO; i++)
  66                         mem_size += sbi->im[i].ino_num *
  67                                                 sizeof(struct ino_entry);
  68                 mem_size >>= PAGE_SHIFT;
  69                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  70         } else if (type == EXTENT_CACHE) {
  71                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  72                                 sizeof(struct extent_tree) +
  73                                 atomic_read(&sbi->total_ext_node) *
  74                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  75                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  76         } else {
  77                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
  78                         return true;
  79         }
  80         return res;
  81 }
  82
  83 static void clear_node_page_dirty(struct page *page)
  84 {
  85         struct address_space *mapping = page->mapping;
  86         unsigned int long flags;
  87
  88         if (PageDirty(page)) {
  89                 spin_lock_irqsave(&mapping->tree_lock, flags);
  90                 radix_tree_tag_clear(&mapping->page_tree,
  91                                 page_index(page),
  92                                 PAGECACHE_TAG_DIRTY);
  93                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  94
  95                 clear_page_dirty_for_io(page);
  96                 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
  97         }
  98         ClearPageUptodate(page);
  99 }
 100
 101 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 102 {
 103         pgoff_t index = current_nat_addr(sbi, nid);
 104         return get_meta_page(sbi, index);
 105 }
 106
 107 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 108 {
 109         struct page *src_page;
 110         struct page *dst_page;
 111         pgoff_t src_off;
 112         pgoff_t dst_off;
 113         void *src_addr;
 114         void *dst_addr;
 115         struct f2fs_nm_info *nm_i = NM_I(sbi);
 116
 117         src_off = current_nat_addr(sbi, nid);
 118         dst_off = next_nat_addr(sbi, src_off);
 119
 120         /* get current nat block page with lock */
 121         src_page = get_meta_page(sbi, src_off);
 122         dst_page = grab_meta_page(sbi, dst_off);
 123         f2fs_bug_on(sbi, PageDirty(src_page));
 124
 125         src_addr = page_address(src_page);
 126         dst_addr = page_address(dst_page);
 127         memcpy(dst_addr, src_addr, PAGE_SIZE);
 128         set_page_dirty(dst_page);
 129         f2fs_put_page(src_page, 1);
 130
 131         set_to_next_nat(nm_i, nid);
 132
 133         return dst_page;
 134 }
 135
 136 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 137 {
 138         return radix_tree_lookup(&nm_i->nat_root, n);
 139 }
 140
 141 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 142                 nid_t start, unsigned int nr, struct nat_entry **ep)
 143 {
 144         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 145 }
 146
 147 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 148 {
 149         list_del(&e->list);
 150         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 151         nm_i->nat_cnt--;
 152         kmem_cache_free(nat_entry_slab, e);
 153 }
 154
 155 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 156                                                 struct nat_entry *ne)
 157 {
 158         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 159         struct nat_entry_set *head;
 160
 161         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 162         if (!head) {
 163                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 164
 165                 INIT_LIST_HEAD(&head->entry_list);
 166                 INIT_LIST_HEAD(&head->set_list);
 167                 head->set = set;
 168                 head->entry_cnt = 0;
 169                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 170         }
 171
 172         if (get_nat_flag(ne, IS_DIRTY))
 173                 goto refresh_list;
 174
 175         nm_i->dirty_nat_cnt++;
 176         head->entry_cnt++;
 177         set_nat_flag(ne, IS_DIRTY, true);
 178 refresh_list:
 179         if (nat_get_blkaddr(ne) == NEW_ADDR)
 180                 list_del_init(&ne->list);
 181         else
 182                 list_move_tail(&ne->list, &head->entry_list);
 183 }
 184
 185 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 186                 struct nat_entry_set *set, struct nat_entry *ne)
 187 {
 188         list_move_tail(&ne->list, &nm_i->nat_entries);
 189         set_nat_flag(ne, IS_DIRTY, false);
 190         set->entry_cnt--;
 191         nm_i->dirty_nat_cnt--;
 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         /* update fsync_mark if its inode nat entry is still alive */
 342         if (ni->nid != ni->ino)
 343                 e = __lookup_nat_cache(nm_i, ni->ino);
 344         if (e) {
 345                 if (fsync_done && ni->nid == ni->ino)
 346                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 347                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 348         }
 349         up_write(&nm_i->nat_tree_lock);
 350 }
 351
 352 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 353 {
 354         struct f2fs_nm_info *nm_i = NM_I(sbi);
 355         int nr = nr_shrink;
 356
 357         if (!down_write_trylock(&nm_i->nat_tree_lock))
 358                 return 0;
 359
 360         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 361                 struct nat_entry *ne;
 362                 ne = list_first_entry(&nm_i->nat_entries,
 363                                         struct nat_entry, list);
 364                 __del_from_nat_cache(nm_i, ne);
 365                 nr_shrink--;
 366         }
 367         up_write(&nm_i->nat_tree_lock);
 368         return nr - nr_shrink;
 369 }
 370
 371 /*
 372  * This function always returns success
 373  */
 374 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 375 {
 376         struct f2fs_nm_info *nm_i = NM_I(sbi);
 377         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 378         struct f2fs_journal *journal = curseg->journal;
 379         nid_t start_nid = START_NID(nid);
 380         struct f2fs_nat_block *nat_blk;
 381         struct page *page = NULL;
 382         struct f2fs_nat_entry ne;
 383         struct nat_entry *e;
 384         pgoff_t index;
 385         int i;
 386
 387         ni->nid = nid;
 388
 389         /* Check nat cache */
 390         down_read(&nm_i->nat_tree_lock);
 391         e = __lookup_nat_cache(nm_i, nid);
 392         if (e) {
 393                 ni->ino = nat_get_ino(e);
 394                 ni->blk_addr = nat_get_blkaddr(e);
 395                 ni->version = nat_get_version(e);
 396                 up_read(&nm_i->nat_tree_lock);
 397                 return;
 398         }
 399
 400         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 401
 402         /* Check current segment summary */
 403         down_read(&curseg->journal_rwsem);
 404         i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 405         if (i >= 0) {
 406                 ne = nat_in_journal(journal, i);
 407                 node_info_from_raw_nat(ni, &ne);
 408         }
 409         up_read(&curseg->journal_rwsem);
 410         if (i >= 0) {
 411                 up_read(&nm_i->nat_tree_lock);
 412                 goto cache;
 413         }
 414
 415         /* Fill node_info from nat page */
 416         index = current_nat_addr(sbi, nid);
 417         up_read(&nm_i->nat_tree_lock);
 418
 419         page = get_meta_page(sbi, index);
 420         nat_blk = (struct f2fs_nat_block *)page_address(page);
 421         ne = nat_blk->entries[nid - start_nid];
 422         node_info_from_raw_nat(ni, &ne);
 423         f2fs_put_page(page, 1);
 424 cache:
 425         /* cache nat entry */
 426         down_write(&nm_i->nat_tree_lock);
 427         cache_nat_entry(sbi, nid, &ne);
 428         up_write(&nm_i->nat_tree_lock);
 429 }
 430
 431 /*
 432  * readahead MAX_RA_NODE number of node pages.
 433  */
 434 static void ra_node_pages(struct page *parent, int start, int n)
 435 {
 436         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 437         struct blk_plug plug;
 438         int i, end;
 439         nid_t nid;
 440
 441         blk_start_plug(&plug);
 442
 443         /* Then, try readahead for siblings of the desired node */
 444         end = start + n;
 445         end = min(end, NIDS_PER_BLOCK);
 446         for (i = start; i < end; i++) {
 447                 nid = get_nid(parent, i, false);
 448                 ra_node_page(sbi, nid);
 449         }
 450
 451         blk_finish_plug(&plug);
 452 }
 453
 454 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 455 {
 456         const long direct_index = ADDRS_PER_INODE(dn->inode);
 457         const long direct_blks = ADDRS_PER_BLOCK;
 458         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 459         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 460         int cur_level = dn->cur_level;
 461         int max_level = dn->max_level;
 462         pgoff_t base = 0;
 463
 464         if (!dn->max_level)
 465                 return pgofs + 1;
 466
 467         while (max_level-- > cur_level)
 468                 skipped_unit *= NIDS_PER_BLOCK;
 469
 470         switch (dn->max_level) {
 471         case 3:
 472                 base += 2 * indirect_blks;
 473         case 2:
 474                 base += 2 * direct_blks;
 475         case 1:
 476                 base += direct_index;
 477                 break;
 478         default:
 479                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 480         }
 481
 482         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 483 }
 484
 485 /*
 486  * The maximum depth is four.
 487  * Offset[0] will have raw inode offset.
 488  */
 489 static int get_node_path(struct inode *inode, long block,
 490                                 int offset[4], unsigned int noffset[4])
 491 {
 492         const long direct_index = ADDRS_PER_INODE(inode);
 493         const long direct_blks = ADDRS_PER_BLOCK;
 494         const long dptrs_per_blk = NIDS_PER_BLOCK;
 495         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 496         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 497         int n = 0;
 498         int level = 0;
 499
 500         noffset[0] = 0;
 501
 502         if (block < direct_index) {
 503                 offset[n] = block;
 504                 goto got;
 505         }
 506         block -= direct_index;
 507         if (block < direct_blks) {
 508                 offset[n++] = NODE_DIR1_BLOCK;
 509                 noffset[n] = 1;
 510                 offset[n] = block;
 511                 level = 1;
 512                 goto got;
 513         }
 514         block -= direct_blks;
 515         if (block < direct_blks) {
 516                 offset[n++] = NODE_DIR2_BLOCK;
 517                 noffset[n] = 2;
 518                 offset[n] = block;
 519                 level = 1;
 520                 goto got;
 521         }
 522         block -= direct_blks;
 523         if (block < indirect_blks) {
 524                 offset[n++] = NODE_IND1_BLOCK;
 525                 noffset[n] = 3;
 526                 offset[n++] = block / direct_blks;
 527                 noffset[n] = 4 + offset[n - 1];
 528                 offset[n] = block % direct_blks;
 529                 level = 2;
 530                 goto got;
 531         }
 532         block -= indirect_blks;
 533         if (block < indirect_blks) {
 534                 offset[n++] = NODE_IND2_BLOCK;
 535                 noffset[n] = 4 + dptrs_per_blk;
 536                 offset[n++] = block / direct_blks;
 537                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 538                 offset[n] = block % direct_blks;
 539                 level = 2;
 540                 goto got;
 541         }
 542         block -= indirect_blks;
 543         if (block < dindirect_blks) {
 544                 offset[n++] = NODE_DIND_BLOCK;
 545                 noffset[n] = 5 + (dptrs_per_blk * 2);
 546                 offset[n++] = block / indirect_blks;
 547                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 548                               offset[n - 1] * (dptrs_per_blk + 1);
 549                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 550                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 551                               offset[n - 2] * (dptrs_per_blk + 1) +
 552                               offset[n - 1];
 553                 offset[n] = block % direct_blks;
 554                 level = 3;
 555                 goto got;
 556         } else {
 557                 BUG();
 558         }
 559 got:
 560         return level;
 561 }
 562
 563 /*
 564  * Caller should call f2fs_put_dnode(dn).
 565  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 566  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 567  * In the case of RDONLY_NODE, we don't need to care about mutex.
 568  */
 569 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 570 {
 571         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 572         struct page *npage[4];
 573         struct page *parent = NULL;
 574         int offset[4];
 575         unsigned int noffset[4];
 576         nid_t nids[4];
 577         int level, i = 0;
 578         int err = 0;
 579
 580         level = get_node_path(dn->inode, index, offset, noffset);
 581
 582         nids[0] = dn->inode->i_ino;
 583         npage[0] = dn->inode_page;
 584
 585         if (!npage[0]) {
 586                 npage[0] = get_node_page(sbi, nids[0]);
 587                 if (IS_ERR(npage[0]))
 588                         return PTR_ERR(npage[0]);
 589         }
 590
 591         /* if inline_data is set, should not report any block indices */
 592         if (f2fs_has_inline_data(dn->inode) && index) {
 593                 err = -ENOENT;
 594                 f2fs_put_page(npage[0], 1);
 595                 goto release_out;
 596         }
 597
 598         parent = npage[0];
 599         if (level != 0)
 600                 nids[1] = get_nid(parent, offset[0], true);
 601         dn->inode_page = npage[0];
 602         dn->inode_page_locked = true;
 603
 604         /* get indirect or direct nodes */
 605         for (i = 1; i <= level; i++) {
 606                 bool done = false;
 607
 608                 if (!nids[i] && mode == ALLOC_NODE) {
 609                         /* alloc new node */
 610                         if (!alloc_nid(sbi, &(nids[i]))) {
 611                                 err = -ENOSPC;
 612                                 goto release_pages;
 613                         }
 614
 615                         dn->nid = nids[i];
 616                         npage[i] = new_node_page(dn, noffset[i], NULL);
 617                         if (IS_ERR(npage[i])) {
 618                                 alloc_nid_failed(sbi, nids[i]);
 619                                 err = PTR_ERR(npage[i]);
 620                                 goto release_pages;
 621                         }
 622
 623                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 624                         alloc_nid_done(sbi, nids[i]);
 625                         done = true;
 626                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 627                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 628                         if (IS_ERR(npage[i])) {
 629                                 err = PTR_ERR(npage[i]);
 630                                 goto release_pages;
 631                         }
 632                         done = true;
 633                 }
 634                 if (i == 1) {
 635                         dn->inode_page_locked = false;
 636                         unlock_page(parent);
 637                 } else {
 638                         f2fs_put_page(parent, 1);
 639                 }
 640
 641                 if (!done) {
 642                         npage[i] = get_node_page(sbi, nids[i]);
 643                         if (IS_ERR(npage[i])) {
 644                                 err = PTR_ERR(npage[i]);
 645                                 f2fs_put_page(npage[0], 0);
 646                                 goto release_out;
 647                         }
 648                 }
 649                 if (i < level) {
 650                         parent = npage[i];
 651                         nids[i + 1] = get_nid(parent, offset[i], false);
 652                 }
 653         }
 654         dn->nid = nids[level];
 655         dn->ofs_in_node = offset[level];
 656         dn->node_page = npage[level];
 657         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 658         return 0;
 659
 660 release_pages:
 661         f2fs_put_page(parent, 1);
 662         if (i > 1)
 663                 f2fs_put_page(npage[0], 0);
 664 release_out:
 665         dn->inode_page = NULL;
 666         dn->node_page = NULL;
 667         if (err == -ENOENT) {
 668                 dn->cur_level = i;
 669                 dn->max_level = level;
 670                 dn->ofs_in_node = offset[level];
 671         }
 672         return err;
 673 }
 674
 675 static void truncate_node(struct dnode_of_data *dn)
 676 {
 677         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 678         struct node_info ni;
 679
 680         get_node_info(sbi, dn->nid, &ni);
 681         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 682
 683         /* Deallocate node address */
 684         invalidate_blocks(sbi, ni.blk_addr);
 685         dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
 686         set_node_addr(sbi, &ni, NULL_ADDR, false);
 687
 688         if (dn->nid == dn->inode->i_ino) {
 689                 remove_orphan_inode(sbi, dn->nid);
 690                 dec_valid_inode_count(sbi);
 691                 f2fs_inode_synced(dn->inode);
 692         }
 693
 694         clear_node_page_dirty(dn->node_page);
 695         set_sbi_flag(sbi, SBI_IS_DIRTY);
 696
 697         f2fs_put_page(dn->node_page, 1);
 698
 699         invalidate_mapping_pages(NODE_MAPPING(sbi),
 700                         dn->node_page->index, dn->node_page->index);
 701
 702         dn->node_page = NULL;
 703         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 704 }
 705
 706 static int truncate_dnode(struct dnode_of_data *dn)
 707 {
 708         struct page *page;
 709
 710         if (dn->nid == 0)
 711                 return 1;
 712
 713         /* get direct node */
 714         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 715         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 716                 return 1;
 717         else if (IS_ERR(page))
 718                 return PTR_ERR(page);
 719
 720         /* Make dnode_of_data for parameter */
 721         dn->node_page = page;
 722         dn->ofs_in_node = 0;
 723         truncate_data_blocks(dn);
 724         truncate_node(dn);
 725         return 1;
 726 }
 727
 728 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 729                                                 int ofs, int depth)
 730 {
 731         struct dnode_of_data rdn = *dn;
 732         struct page *page;
 733         struct f2fs_node *rn;
 734         nid_t child_nid;
 735         unsigned int child_nofs;
 736         int freed = 0;
 737         int i, ret;
 738
 739         if (dn->nid == 0)
 740                 return NIDS_PER_BLOCK + 1;
 741
 742         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 743
 744         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 745         if (IS_ERR(page)) {
 746                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 747                 return PTR_ERR(page);
 748         }
 749
 750         ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 751
 752         rn = F2FS_NODE(page);
 753         if (depth < 3) {
 754                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 755                         child_nid = le32_to_cpu(rn->in.nid[i]);
 756                         if (child_nid == 0)
 757                                 continue;
 758                         rdn.nid = child_nid;
 759                         ret = truncate_dnode(&rdn);
 760                         if (ret < 0)
 761                                 goto out_err;
 762                         if (set_nid(page, i, 0, false))
 763                                 dn->node_changed = true;
 764                 }
 765         } else {
 766                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 767                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 768                         child_nid = le32_to_cpu(rn->in.nid[i]);
 769                         if (child_nid == 0) {
 770                                 child_nofs += NIDS_PER_BLOCK + 1;
 771                                 continue;
 772                         }
 773                         rdn.nid = child_nid;
 774                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 775                         if (ret == (NIDS_PER_BLOCK + 1)) {
 776                                 if (set_nid(page, i, 0, false))
 777                                         dn->node_changed = true;
 778                                 child_nofs += ret;
 779                         } else if (ret < 0 && ret != -ENOENT) {
 780                                 goto out_err;
 781                         }
 782                 }
 783                 freed = child_nofs;
 784         }
 785
 786         if (!ofs) {
 787                 /* remove current indirect node */
 788                 dn->node_page = page;
 789                 truncate_node(dn);
 790                 freed++;
 791         } else {
 792                 f2fs_put_page(page, 1);
 793         }
 794         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 795         return freed;
 796
 797 out_err:
 798         f2fs_put_page(page, 1);
 799         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 800         return ret;
 801 }
 802
 803 static int truncate_partial_nodes(struct dnode_of_data *dn,
 804                         struct f2fs_inode *ri, int *offset, int depth)
 805 {
 806         struct page *pages[2];
 807         nid_t nid[3];
 808         nid_t child_nid;
 809         int err = 0;
 810         int i;
 811         int idx = depth - 2;
 812
 813         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 814         if (!nid[0])
 815                 return 0;
 816
 817         /* get indirect nodes in the path */
 818         for (i = 0; i < idx + 1; i++) {
 819                 /* reference count'll be increased */
 820                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 821                 if (IS_ERR(pages[i])) {
 822                         err = PTR_ERR(pages[i]);
 823                         idx = i - 1;
 824                         goto fail;
 825                 }
 826                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 827         }
 828
 829         ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 830
 831         /* free direct nodes linked to a partial indirect node */
 832         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 833                 child_nid = get_nid(pages[idx], i, false);
 834                 if (!child_nid)
 835                         continue;
 836                 dn->nid = child_nid;
 837                 err = truncate_dnode(dn);
 838                 if (err < 0)
 839                         goto fail;
 840                 if (set_nid(pages[idx], i, 0, false))
 841                         dn->node_changed = true;
 842         }
 843
 844         if (offset[idx + 1] == 0) {
 845                 dn->node_page = pages[idx];
 846                 dn->nid = nid[idx];
 847                 truncate_node(dn);
 848         } else {
 849                 f2fs_put_page(pages[idx], 1);
 850         }
 851         offset[idx]++;
 852         offset[idx + 1] = 0;
 853         idx--;
 854 fail:
 855         for (i = idx; i >= 0; i--)
 856                 f2fs_put_page(pages[i], 1);
 857
 858         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 859
 860         return err;
 861 }
 862
 863 /*
 864  * All the block addresses of data and nodes should be nullified.
 865  */
 866 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 867 {
 868         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 869         int err = 0, cont = 1;
 870         int level, offset[4], noffset[4];
 871         unsigned int nofs = 0;
 872         struct f2fs_inode *ri;
 873         struct dnode_of_data dn;
 874         struct page *page;
 875
 876         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 877
 878         level = get_node_path(inode, from, offset, noffset);
 879
 880         page = get_node_page(sbi, inode->i_ino);
 881         if (IS_ERR(page)) {
 882                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 883                 return PTR_ERR(page);
 884         }
 885
 886         set_new_dnode(&dn, inode, page, NULL, 0);
 887         unlock_page(page);
 888
 889         ri = F2FS_INODE(page);
 890         switch (level) {
 891         case 0:
 892         case 1:
 893                 nofs = noffset[1];
 894                 break;
 895         case 2:
 896                 nofs = noffset[1];
 897                 if (!offset[level - 1])
 898                         goto skip_partial;
 899                 err = truncate_partial_nodes(&dn, ri, offset, level);
 900                 if (err < 0 && err != -ENOENT)
 901                         goto fail;
 902                 nofs += 1 + NIDS_PER_BLOCK;
 903                 break;
 904         case 3:
 905                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 906                 if (!offset[level - 1])
 907                         goto skip_partial;
 908                 err = truncate_partial_nodes(&dn, ri, offset, level);
 909                 if (err < 0 && err != -ENOENT)
 910                         goto fail;
 911                 break;
 912         default:
 913                 BUG();
 914         }
 915
 916 skip_partial:
 917         while (cont) {
 918                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 919                 switch (offset[0]) {
 920                 case NODE_DIR1_BLOCK:
 921                 case NODE_DIR2_BLOCK:
 922                         err = truncate_dnode(&dn);
 923                         break;
 924
 925                 case NODE_IND1_BLOCK:
 926                 case NODE_IND2_BLOCK:
 927                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 928                         break;
 929
 930                 case NODE_DIND_BLOCK:
 931                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 932                         cont = 0;
 933                         break;
 934
 935                 default:
 936                         BUG();
 937                 }
 938                 if (err < 0 && err != -ENOENT)
 939                         goto fail;
 940                 if (offset[1] == 0 &&
 941                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 942                         lock_page(page);
 943                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
 944                         f2fs_wait_on_page_writeback(page, NODE, true);
 945                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 946                         set_page_dirty(page);
 947                         unlock_page(page);
 948                 }
 949                 offset[1] = 0;
 950                 offset[0]++;
 951                 nofs += err;
 952         }
 953 fail:
 954         f2fs_put_page(page, 0);
 955         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 956         return err > 0 ? 0 : err;
 957 }
 958
 959 int truncate_xattr_node(struct inode *inode, struct page *page)
 960 {
 961         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 962         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 963         struct dnode_of_data dn;
 964         struct page *npage;
 965
 966         if (!nid)
 967                 return 0;
 968
 969         npage = get_node_page(sbi, nid);
 970         if (IS_ERR(npage))
 971                 return PTR_ERR(npage);
 972
 973         f2fs_i_xnid_write(inode, 0);
 974
 975         set_new_dnode(&dn, inode, page, npage, nid);
 976
 977         if (page)
 978                 dn.inode_page_locked = true;
 979         truncate_node(&dn);
 980         return 0;
 981 }
 982
 983 /*
 984  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 985  * f2fs_unlock_op().
 986  */
 987 int remove_inode_page(struct inode *inode)
 988 {
 989         struct dnode_of_data dn;
 990         int err;
 991
 992         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 993         err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
 994         if (err)
 995                 return err;
 996
 997         err = truncate_xattr_node(inode, dn.inode_page);
 998         if (err) {
 999                 f2fs_put_dnode(&dn);
1000                 return err;
1001         }
1002
1003         /* remove potential inline_data blocks */
1004         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1005                                 S_ISLNK(inode->i_mode))
1006                 truncate_data_blocks_range(&dn, 1);
1007
1008         /* 0 is possible, after f2fs_new_inode() has failed */
1009         f2fs_bug_on(F2FS_I_SB(inode),
1010                         inode->i_blocks != 0 && inode->i_blocks != 8);
1011
1012         /* will put inode & node pages */
1013         truncate_node(&dn);
1014         return 0;
1015 }
1016
1017 struct page *new_inode_page(struct inode *inode)
1018 {
1019         struct dnode_of_data dn;
1020
1021         /* allocate inode page for new inode */
1022         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1023
1024         /* caller should f2fs_put_page(page, 1); */
1025         return new_node_page(&dn, 0, NULL);
1026 }
1027
1028 struct page *new_node_page(struct dnode_of_data *dn,
1029                                 unsigned int ofs, struct page *ipage)
1030 {
1031         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1032         struct node_info new_ni;
1033         struct page *page;
1034         int err;
1035
1036         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1037                 return ERR_PTR(-EPERM);
1038
1039         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1040         if (!page)
1041                 return ERR_PTR(-ENOMEM);
1042
1043         if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1044                 goto fail;
1045
1046 #ifdef CONFIG_F2FS_CHECK_FS
1047         get_node_info(sbi, dn->nid, &new_ni);
1048         f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1049 #endif
1050         new_ni.nid = dn->nid;
1051         new_ni.ino = dn->inode->i_ino;
1052         new_ni.blk_addr = NULL_ADDR;
1053         new_ni.flag = 0;
1054         new_ni.version = 0;
1055         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1056
1057         f2fs_wait_on_page_writeback(page, NODE, true);
1058         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1059         set_cold_node(dn->inode, page);
1060         if (!PageUptodate(page))
1061                 SetPageUptodate(page);
1062         if (set_page_dirty(page))
1063                 dn->node_changed = true;
1064
1065         if (f2fs_has_xattr_block(ofs))
1066                 f2fs_i_xnid_write(dn->inode, dn->nid);
1067
1068         if (ofs == 0)
1069                 inc_valid_inode_count(sbi);
1070         return page;
1071
1072 fail:
1073         clear_node_page_dirty(page);
1074         f2fs_put_page(page, 1);
1075         return ERR_PTR(err);
1076 }
1077
1078 /*
1079  * Caller should do after getting the following values.
1080  * 0: f2fs_put_page(page, 0)
1081  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1082  */
1083 static int read_node_page(struct page *page, int op_flags)
1084 {
1085         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1086         struct node_info ni;
1087         struct f2fs_io_info fio = {
1088                 .sbi = sbi,
1089                 .type = NODE,
1090                 .op = REQ_OP_READ,
1091                 .op_flags = op_flags,
1092                 .page = page,
1093                 .encrypted_page = NULL,
1094         };
1095
1096         if (PageUptodate(page))
1097                 return LOCKED_PAGE;
1098
1099         get_node_info(sbi, page->index, &ni);
1100
1101         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1102                 ClearPageUptodate(page);
1103                 return -ENOENT;
1104         }
1105
1106         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1107         return f2fs_submit_page_bio(&fio);
1108 }
1109
1110 /*
1111  * Readahead a node page
1112  */
1113 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1114 {
1115         struct page *apage;
1116         int err;
1117
1118         if (!nid)
1119                 return;
1120         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1121
1122         rcu_read_lock();
1123         apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1124         rcu_read_unlock();
1125         if (apage)
1126                 return;
1127
1128         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1129         if (!apage)
1130                 return;
1131
1132         err = read_node_page(apage, REQ_RAHEAD);
1133         f2fs_put_page(apage, err ? 1 : 0);
1134 }
1135
1136 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1137                                         struct page *parent, int start)
1138 {
1139         struct page *page;
1140         int err;
1141
1142         if (!nid)
1143                 return ERR_PTR(-ENOENT);
1144         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1145 repeat:
1146         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1147         if (!page)
1148                 return ERR_PTR(-ENOMEM);
1149
1150         err = read_node_page(page, 0);
1151         if (err < 0) {
1152                 f2fs_put_page(page, 1);
1153                 return ERR_PTR(err);
1154         } else if (err == LOCKED_PAGE) {
1155                 err = 0;
1156                 goto page_hit;
1157         }
1158
1159         if (parent)
1160                 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1161
1162         lock_page(page);
1163
1164         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1165                 f2fs_put_page(page, 1);
1166                 goto repeat;
1167         }
1168
1169         if (unlikely(!PageUptodate(page))) {
1170                 err = -EIO;
1171                 goto out_err;
1172         }
1173 page_hit:
1174         if(unlikely(nid != nid_of_node(page))) {
1175                 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1176                         "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1177                         nid, nid_of_node(page), ino_of_node(page),
1178                         ofs_of_node(page), cpver_of_node(page),
1179                         next_blkaddr_of_node(page));
1180                 ClearPageUptodate(page);
1181                 err = -EINVAL;
1182 out_err:
1183                 f2fs_put_page(page, 1);
1184                 return ERR_PTR(err);
1185         }
1186         return page;
1187 }
1188
1189 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1190 {
1191         return __get_node_page(sbi, nid, NULL, 0);
1192 }
1193
1194 struct page *get_node_page_ra(struct page *parent, int start)
1195 {
1196         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1197         nid_t nid = get_nid(parent, start, false);
1198
1199         return __get_node_page(sbi, nid, parent, start);
1200 }
1201
1202 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1203 {
1204         struct inode *inode;
1205         struct page *page;
1206         int ret;
1207
1208         /* should flush inline_data before evict_inode */
1209         inode = ilookup(sbi->sb, ino);
1210         if (!inode)
1211                 return;
1212
1213         page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1214         if (!page)
1215                 goto iput_out;
1216
1217         if (!PageUptodate(page))
1218                 goto page_out;
1219
1220         if (!PageDirty(page))
1221                 goto page_out;
1222
1223         if (!clear_page_dirty_for_io(page))
1224                 goto page_out;
1225
1226         ret = f2fs_write_inline_data(inode, page);
1227         inode_dec_dirty_pages(inode);
1228         remove_dirty_inode(inode);
1229         if (ret)
1230                 set_page_dirty(page);
1231 page_out:
1232         f2fs_put_page(page, 1);
1233 iput_out:
1234         iput(inode);
1235 }
1236
1237 void move_node_page(struct page *node_page, int gc_type)
1238 {
1239         if (gc_type == FG_GC) {
1240                 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1241                 struct writeback_control wbc = {
1242                         .sync_mode = WB_SYNC_ALL,
1243                         .nr_to_write = 1,
1244                         .for_reclaim = 0,
1245                 };
1246
1247                 set_page_dirty(node_page);
1248                 f2fs_wait_on_page_writeback(node_page, NODE, true);
1249
1250                 f2fs_bug_on(sbi, PageWriteback(node_page));
1251                 if (!clear_page_dirty_for_io(node_page))
1252                         goto out_page;
1253
1254                 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1255                         unlock_page(node_page);
1256                 goto release_page;
1257         } else {
1258                 /* set page dirty and write it */
1259                 if (!PageWriteback(node_page))
1260                         set_page_dirty(node_page);
1261         }
1262 out_page:
1263         unlock_page(node_page);
1264 release_page:
1265         f2fs_put_page(node_page, 0);
1266 }
1267
1268 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1269 {
1270         pgoff_t index, end;
1271         struct pagevec pvec;
1272         struct page *last_page = NULL;
1273
1274         pagevec_init(&pvec, 0);
1275         index = 0;
1276         end = ULONG_MAX;
1277
1278         while (index <= end) {
1279                 int i, nr_pages;
1280                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1281                                 PAGECACHE_TAG_DIRTY,
1282                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1283                 if (nr_pages == 0)
1284                         break;
1285
1286                 for (i = 0; i < nr_pages; i++) {
1287                         struct page *page = pvec.pages[i];
1288
1289                         if (unlikely(f2fs_cp_error(sbi))) {
1290                                 f2fs_put_page(last_page, 0);
1291                                 pagevec_release(&pvec);
1292                                 return ERR_PTR(-EIO);
1293                         }
1294
1295                         if (!IS_DNODE(page) || !is_cold_node(page))
1296                                 continue;
1297                         if (ino_of_node(page) != ino)
1298                                 continue;
1299
1300                         lock_page(page);
1301
1302                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1303 continue_unlock:
1304                                 unlock_page(page);
1305                                 continue;
1306                         }
1307                         if (ino_of_node(page) != ino)
1308                                 goto continue_unlock;
1309
1310                         if (!PageDirty(page)) {
1311                                 /* someone wrote it for us */
1312                                 goto continue_unlock;
1313                         }
1314
1315                         if (last_page)
1316                                 f2fs_put_page(last_page, 0);
1317
1318                         get_page(page);
1319                         last_page = page;
1320                         unlock_page(page);
1321                 }
1322                 pagevec_release(&pvec);
1323                 cond_resched();
1324         }
1325         return last_page;
1326 }
1327
1328 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1329                                 struct writeback_control *wbc)
1330 {
1331         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1332         nid_t nid;
1333         struct node_info ni;
1334         struct f2fs_io_info fio = {
1335                 .sbi = sbi,
1336                 .type = NODE,
1337                 .op = REQ_OP_WRITE,
1338                 .op_flags = wbc_to_write_flags(wbc),
1339                 .page = page,
1340                 .encrypted_page = NULL,
1341                 .submitted = false,
1342         };
1343
1344         trace_f2fs_writepage(page, NODE);
1345
1346         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1347                 goto redirty_out;
1348         if (unlikely(f2fs_cp_error(sbi)))
1349                 goto redirty_out;
1350
1351         /* get old block addr of this node page */
1352         nid = nid_of_node(page);
1353         f2fs_bug_on(sbi, page->index != nid);
1354
1355         if (wbc->for_reclaim) {
1356                 if (!down_read_trylock(&sbi->node_write))
1357                         goto redirty_out;
1358         } else {
1359                 down_read(&sbi->node_write);
1360         }
1361
1362         get_node_info(sbi, nid, &ni);
1363
1364         /* This page is already truncated */
1365         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1366                 ClearPageUptodate(page);
1367                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1368                 up_read(&sbi->node_write);
1369                 unlock_page(page);
1370                 return 0;
1371         }
1372
1373         if (atomic && !test_opt(sbi, NOBARRIER))
1374                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1375
1376         set_page_writeback(page);
1377         fio.old_blkaddr = ni.blk_addr;
1378         write_node_page(nid, &fio);
1379         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1380         dec_page_count(sbi, F2FS_DIRTY_NODES);
1381         up_read(&sbi->node_write);
1382
1383         if (wbc->for_reclaim) {
1384                 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1385                                                 page->index, NODE);
1386                 submitted = NULL;
1387         }
1388
1389         unlock_page(page);
1390
1391         if (unlikely(f2fs_cp_error(sbi))) {
1392                 f2fs_submit_merged_write(sbi, NODE);
1393                 submitted = NULL;
1394         }
1395         if (submitted)
1396                 *submitted = fio.submitted;
1397
1398         return 0;
1399
1400 redirty_out:
1401         redirty_page_for_writepage(wbc, page);
1402         return AOP_WRITEPAGE_ACTIVATE;
1403 }
1404
1405 static int f2fs_write_node_page(struct page *page,
1406                                 struct writeback_control *wbc)
1407 {
1408         return __write_node_page(page, false, NULL, wbc);
1409 }
1410
1411 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1412                         struct writeback_control *wbc, bool atomic)
1413 {
1414         pgoff_t index, end;
1415         pgoff_t last_idx = ULONG_MAX;
1416         struct pagevec pvec;
1417         int ret = 0;
1418         struct page *last_page = NULL;
1419         bool marked = false;
1420         nid_t ino = inode->i_ino;
1421
1422         if (atomic) {
1423                 last_page = last_fsync_dnode(sbi, ino);
1424                 if (IS_ERR_OR_NULL(last_page))
1425                         return PTR_ERR_OR_ZERO(last_page);
1426         }
1427 retry:
1428         pagevec_init(&pvec, 0);
1429         index = 0;
1430         end = ULONG_MAX;
1431
1432         while (index <= end) {
1433                 int i, nr_pages;
1434                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1435                                 PAGECACHE_TAG_DIRTY,
1436                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1437                 if (nr_pages == 0)
1438                         break;
1439
1440                 for (i = 0; i < nr_pages; i++) {
1441                         struct page *page = pvec.pages[i];
1442                         bool submitted = false;
1443
1444                         if (unlikely(f2fs_cp_error(sbi))) {
1445                                 f2fs_put_page(last_page, 0);
1446                                 pagevec_release(&pvec);
1447                                 ret = -EIO;
1448                                 goto out;
1449                         }
1450
1451                         if (!IS_DNODE(page) || !is_cold_node(page))
1452                                 continue;
1453                         if (ino_of_node(page) != ino)
1454                                 continue;
1455
1456                         lock_page(page);
1457
1458                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1459 continue_unlock:
1460                                 unlock_page(page);
1461                                 continue;
1462                         }
1463                         if (ino_of_node(page) != ino)
1464                                 goto continue_unlock;
1465
1466                         if (!PageDirty(page) && page != last_page) {
1467                                 /* someone wrote it for us */
1468                                 goto continue_unlock;
1469                         }
1470
1471                         f2fs_wait_on_page_writeback(page, NODE, true);
1472                         BUG_ON(PageWriteback(page));
1473
1474                         set_fsync_mark(page, 0);
1475                         set_dentry_mark(page, 0);
1476
1477                         if (!atomic || page == last_page) {
1478                                 set_fsync_mark(page, 1);
1479                                 if (IS_INODE(page)) {
1480                                         if (is_inode_flag_set(inode,
1481                                                                 FI_DIRTY_INODE))
1482                                                 update_inode(inode, page);
1483                                         set_dentry_mark(page,
1484                                                 need_dentry_mark(sbi, ino));
1485                                 }
1486                                 /*  may be written by other thread */
1487                                 if (!PageDirty(page))
1488                                         set_page_dirty(page);
1489                         }
1490
1491                         if (!clear_page_dirty_for_io(page))
1492                                 goto continue_unlock;
1493
1494                         ret = __write_node_page(page, atomic &&
1495                                                 page == last_page,
1496                                                 &submitted, wbc);
1497                         if (ret) {
1498                                 unlock_page(page);
1499                                 f2fs_put_page(last_page, 0);
1500                                 break;
1501                         } else if (submitted) {
1502                                 last_idx = page->index;
1503                         }
1504
1505                         if (page == last_page) {
1506                                 f2fs_put_page(page, 0);
1507                                 marked = true;
1508                                 break;
1509                         }
1510                 }
1511                 pagevec_release(&pvec);
1512                 cond_resched();
1513
1514                 if (ret || marked)
1515                         break;
1516         }
1517         if (!ret && atomic && !marked) {
1518                 f2fs_msg(sbi->sb, KERN_DEBUG,
1519                         "Retry to write fsync mark: ino=%u, idx=%lx",
1520                                         ino, last_page->index);
1521                 lock_page(last_page);
1522                 f2fs_wait_on_page_writeback(last_page, NODE, true);
1523                 set_page_dirty(last_page);
1524                 unlock_page(last_page);
1525                 goto retry;
1526         }
1527 out:
1528         if (last_idx != ULONG_MAX)
1529                 f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1530         return ret ? -EIO: 0;
1531 }
1532
1533 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1534 {
1535         pgoff_t index, end;
1536         struct pagevec pvec;
1537         int step = 0;
1538         int nwritten = 0;
1539         int ret = 0;
1540
1541         pagevec_init(&pvec, 0);
1542
1543 next_step:
1544         index = 0;
1545         end = ULONG_MAX;
1546
1547         while (index <= end) {
1548                 int i, nr_pages;
1549                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1550                                 PAGECACHE_TAG_DIRTY,
1551                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1552                 if (nr_pages == 0)
1553                         break;
1554
1555                 for (i = 0; i < nr_pages; i++) {
1556                         struct page *page = pvec.pages[i];
1557                         bool submitted = false;
1558
1559                         if (unlikely(f2fs_cp_error(sbi))) {
1560                                 pagevec_release(&pvec);
1561                                 ret = -EIO;
1562                                 goto out;
1563                         }
1564
1565                         /*
1566                          * flushing sequence with step:
1567                          * 0. indirect nodes
1568                          * 1. dentry dnodes
1569                          * 2. file dnodes
1570                          */
1571                         if (step == 0 && IS_DNODE(page))
1572                                 continue;
1573                         if (step == 1 && (!IS_DNODE(page) ||
1574                                                 is_cold_node(page)))
1575                                 continue;
1576                         if (step == 2 && (!IS_DNODE(page) ||
1577                                                 !is_cold_node(page)))
1578                                 continue;
1579 lock_node:
1580                         if (!trylock_page(page))
1581                                 continue;
1582
1583                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1584 continue_unlock:
1585                                 unlock_page(page);
1586                                 continue;
1587                         }
1588
1589                         if (!PageDirty(page)) {
1590                                 /* someone wrote it for us */
1591                                 goto continue_unlock;
1592                         }
1593
1594                         /* flush inline_data */
1595                         if (is_inline_node(page)) {
1596                                 clear_inline_node(page);
1597                                 unlock_page(page);
1598                                 flush_inline_data(sbi, ino_of_node(page));
1599                                 goto lock_node;
1600                         }
1601
1602                         f2fs_wait_on_page_writeback(page, NODE, true);
1603
1604                         BUG_ON(PageWriteback(page));
1605                         if (!clear_page_dirty_for_io(page))
1606                                 goto continue_unlock;
1607
1608                         set_fsync_mark(page, 0);
1609                         set_dentry_mark(page, 0);
1610
1611                         ret = __write_node_page(page, false, &submitted, wbc);
1612                         if (ret)
1613                                 unlock_page(page);
1614                         else if (submitted)
1615                                 nwritten++;
1616
1617                         if (--wbc->nr_to_write == 0)
1618                                 break;
1619                 }
1620                 pagevec_release(&pvec);
1621                 cond_resched();
1622
1623                 if (wbc->nr_to_write == 0) {
1624                         step = 2;
1625                         break;
1626                 }
1627         }
1628
1629         if (step < 2) {
1630                 step++;
1631                 goto next_step;
1632         }
1633 out:
1634         if (nwritten)
1635                 f2fs_submit_merged_write(sbi, NODE);
1636         return ret;
1637 }
1638
1639 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1640 {
1641         pgoff_t index = 0, end = ULONG_MAX;
1642         struct pagevec pvec;
1643         int ret2, ret = 0;
1644
1645         pagevec_init(&pvec, 0);
1646
1647         while (index <= end) {
1648                 int i, nr_pages;
1649                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1650                                 PAGECACHE_TAG_WRITEBACK,
1651                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1652                 if (nr_pages == 0)
1653                         break;
1654
1655                 for (i = 0; i < nr_pages; i++) {
1656                         struct page *page = pvec.pages[i];
1657
1658                         /* until radix tree lookup accepts end_index */
1659                         if (unlikely(page->index > end))
1660                                 continue;
1661
1662                         if (ino && ino_of_node(page) == ino) {
1663                                 f2fs_wait_on_page_writeback(page, NODE, true);
1664                                 if (TestClearPageError(page))
1665                                         ret = -EIO;
1666                         }
1667                 }
1668                 pagevec_release(&pvec);
1669                 cond_resched();
1670         }
1671
1672         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1673         if (!ret)
1674                 ret = ret2;
1675         return ret;
1676 }
1677
1678 static int f2fs_write_node_pages(struct address_space *mapping,
1679                             struct writeback_control *wbc)
1680 {
1681         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1682         struct blk_plug plug;
1683         long diff;
1684
1685         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1686                 goto skip_write;
1687
1688         /* balancing f2fs's metadata in background */
1689         f2fs_balance_fs_bg(sbi);
1690
1691         /* collect a number of dirty node pages and write together */
1692         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1693                 goto skip_write;
1694
1695         trace_f2fs_writepages(mapping->host, wbc, NODE);
1696
1697         diff = nr_pages_to_write(sbi, NODE, wbc);
1698         wbc->sync_mode = WB_SYNC_NONE;
1699         blk_start_plug(&plug);
1700         sync_node_pages(sbi, wbc);
1701         blk_finish_plug(&plug);
1702         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1703         return 0;
1704
1705 skip_write:
1706         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1707         trace_f2fs_writepages(mapping->host, wbc, NODE);
1708         return 0;
1709 }
1710
1711 static int f2fs_set_node_page_dirty(struct page *page)
1712 {
1713         trace_f2fs_set_page_dirty(page, NODE);
1714
1715         if (!PageUptodate(page))
1716                 SetPageUptodate(page);
1717         if (!PageDirty(page)) {
1718                 f2fs_set_page_dirty_nobuffers(page);
1719                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1720                 SetPagePrivate(page);
1721                 f2fs_trace_pid(page);
1722                 return 1;
1723         }
1724         return 0;
1725 }
1726
1727 /*
1728  * Structure of the f2fs node operations
1729  */
1730 const struct address_space_operations f2fs_node_aops = {
1731         .writepage      = f2fs_write_node_page,
1732         .writepages     = f2fs_write_node_pages,
1733         .set_page_dirty = f2fs_set_node_page_dirty,
1734         .invalidatepage = f2fs_invalidate_page,
1735         .releasepage    = f2fs_release_page,
1736 #ifdef CONFIG_MIGRATION
1737         .migratepage    = f2fs_migrate_page,
1738 #endif
1739 };
1740
1741 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1742                                                 nid_t n)
1743 {
1744         return radix_tree_lookup(&nm_i->free_nid_root, n);
1745 }
1746
1747 static int __insert_nid_to_list(struct f2fs_sb_info *sbi,
1748                         struct free_nid *i, enum nid_list list, bool new)
1749 {
1750         struct f2fs_nm_info *nm_i = NM_I(sbi);
1751
1752         if (new) {
1753                 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1754                 if (err)
1755                         return err;
1756         }
1757
1758         f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1759                                                 i->state != NID_ALLOC);
1760         nm_i->nid_cnt[list]++;
1761         list_add_tail(&i->list, &nm_i->nid_list[list]);
1762         return 0;
1763 }
1764
1765 static void __remove_nid_from_list(struct f2fs_sb_info *sbi,
1766                         struct free_nid *i, enum nid_list list, bool reuse)
1767 {
1768         struct f2fs_nm_info *nm_i = NM_I(sbi);
1769
1770         f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1771                                                 i->state != NID_ALLOC);
1772         nm_i->nid_cnt[list]--;
1773         list_del(&i->list);
1774         if (!reuse)
1775                 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1776 }
1777
1778 /* return if the nid is recognized as free */
1779 static bool add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1780 {
1781         struct f2fs_nm_info *nm_i = NM_I(sbi);
1782         struct free_nid *i, *e;
1783         struct nat_entry *ne;
1784         int err = -EINVAL;
1785         bool ret = false;
1786
1787         /* 0 nid should not be used */
1788         if (unlikely(nid == 0))
1789                 return false;
1790
1791         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1792         i->nid = nid;
1793         i->state = NID_NEW;
1794
1795         if (radix_tree_preload(GFP_NOFS))
1796                 goto err;
1797
1798         spin_lock(&nm_i->nid_list_lock);
1799
1800         if (build) {
1801                 /*
1802                  *   Thread A             Thread B
1803                  *  - f2fs_create
1804                  *   - f2fs_new_inode
1805                  *    - alloc_nid
1806                  *     - __insert_nid_to_list(ALLOC_NID_LIST)
1807                  *                     - f2fs_balance_fs_bg
1808                  *                      - build_free_nids
1809                  *                       - __build_free_nids
1810                  *                        - scan_nat_page
1811                  *                         - add_free_nid
1812                  *                          - __lookup_nat_cache
1813                  *  - f2fs_add_link
1814                  *   - init_inode_metadata
1815                  *    - new_inode_page
1816                  *     - new_node_page
1817                  *      - set_node_addr
1818                  *  - alloc_nid_done
1819                  *   - __remove_nid_from_list(ALLOC_NID_LIST)
1820                  *                         - __insert_nid_to_list(FREE_NID_LIST)
1821                  */
1822                 ne = __lookup_nat_cache(nm_i, nid);
1823                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1824                                 nat_get_blkaddr(ne) != NULL_ADDR))
1825                         goto err_out;
1826
1827                 e = __lookup_free_nid_list(nm_i, nid);
1828                 if (e) {
1829                         if (e->state == NID_NEW)
1830                                 ret = true;
1831                         goto err_out;
1832                 }
1833         }
1834         ret = true;
1835         err = __insert_nid_to_list(sbi, i, FREE_NID_LIST, true);
1836 err_out:
1837         spin_unlock(&nm_i->nid_list_lock);
1838         radix_tree_preload_end();
1839 err:
1840         if (err)
1841                 kmem_cache_free(free_nid_slab, i);
1842         return ret;
1843 }
1844
1845 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1846 {
1847         struct f2fs_nm_info *nm_i = NM_I(sbi);
1848         struct free_nid *i;
1849         bool need_free = false;
1850
1851         spin_lock(&nm_i->nid_list_lock);
1852         i = __lookup_free_nid_list(nm_i, nid);
1853         if (i && i->state == NID_NEW) {
1854                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
1855                 need_free = true;
1856         }
1857         spin_unlock(&nm_i->nid_list_lock);
1858
1859         if (need_free)
1860                 kmem_cache_free(free_nid_slab, i);
1861 }
1862
1863 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
1864                                                         bool set, bool build)
1865 {
1866         struct f2fs_nm_info *nm_i = NM_I(sbi);
1867         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
1868         unsigned int nid_ofs = nid - START_NID(nid);
1869
1870         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1871                 return;
1872
1873         if (set)
1874                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1875         else
1876                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1877
1878         if (set)
1879                 nm_i->free_nid_count[nat_ofs]++;
1880         else if (!build)
1881                 nm_i->free_nid_count[nat_ofs]--;
1882 }
1883
1884 static void scan_nat_page(struct f2fs_sb_info *sbi,
1885                         struct page *nat_page, nid_t start_nid)
1886 {
1887         struct f2fs_nm_info *nm_i = NM_I(sbi);
1888         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1889         block_t blk_addr;
1890         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
1891         int i;
1892
1893         if (test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1894                 return;
1895
1896         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
1897
1898         i = start_nid % NAT_ENTRY_PER_BLOCK;
1899
1900         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1901                 bool freed = false;
1902
1903                 if (unlikely(start_nid >= nm_i->max_nid))
1904                         break;
1905
1906                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1907                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1908                 if (blk_addr == NULL_ADDR)
1909                         freed = add_free_nid(sbi, start_nid, true);
1910                 spin_lock(&NM_I(sbi)->nid_list_lock);
1911                 update_free_nid_bitmap(sbi, start_nid, freed, true);
1912                 spin_unlock(&NM_I(sbi)->nid_list_lock);
1913         }
1914 }
1915
1916 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
1917 {
1918         struct f2fs_nm_info *nm_i = NM_I(sbi);
1919         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1920         struct f2fs_journal *journal = curseg->journal;
1921         unsigned int i, idx;
1922
1923         down_read(&nm_i->nat_tree_lock);
1924
1925         for (i = 0; i < nm_i->nat_blocks; i++) {
1926                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
1927                         continue;
1928                 if (!nm_i->free_nid_count[i])
1929                         continue;
1930                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
1931                         nid_t nid;
1932
1933                         if (!test_bit_le(idx, nm_i->free_nid_bitmap[i]))
1934                                 continue;
1935
1936                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
1937                         add_free_nid(sbi, nid, true);
1938
1939                         if (nm_i->nid_cnt[FREE_NID_LIST] >= MAX_FREE_NIDS)
1940                                 goto out;
1941                 }
1942         }
1943 out:
1944         down_read(&curseg->journal_rwsem);
1945         for (i = 0; i < nats_in_cursum(journal); i++) {
1946                 block_t addr;
1947                 nid_t nid;
1948
1949                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1950                 nid = le32_to_cpu(nid_in_journal(journal, i));
1951                 if (addr == NULL_ADDR)
1952                         add_free_nid(sbi, nid, true);
1953                 else
1954                         remove_free_nid(sbi, nid);
1955         }
1956         up_read(&curseg->journal_rwsem);
1957         up_read(&nm_i->nat_tree_lock);
1958 }
1959
1960 static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
1961 {
1962         struct f2fs_nm_info *nm_i = NM_I(sbi);
1963         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1964         struct f2fs_journal *journal = curseg->journal;
1965         int i = 0;
1966         nid_t nid = nm_i->next_scan_nid;
1967
1968         if (unlikely(nid >= nm_i->max_nid))
1969                 nid = 0;
1970
1971         /* Enough entries */
1972         if (nm_i->nid_cnt[FREE_NID_LIST] >= NAT_ENTRY_PER_BLOCK)
1973                 return;
1974
1975         if (!sync && !available_free_memory(sbi, FREE_NIDS))
1976                 return;
1977
1978         if (!mount) {
1979                 /* try to find free nids in free_nid_bitmap */
1980                 scan_free_nid_bits(sbi);
1981
1982                 if (nm_i->nid_cnt[FREE_NID_LIST])
1983                         return;
1984         }
1985
1986         /* readahead nat pages to be scanned */
1987         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1988                                                         META_NAT, true);
1989
1990         down_read(&nm_i->nat_tree_lock);
1991
1992         while (1) {
1993                 struct page *page = get_current_nat_page(sbi, nid);
1994
1995                 scan_nat_page(sbi, page, nid);
1996                 f2fs_put_page(page, 1);
1997
1998                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1999                 if (unlikely(nid >= nm_i->max_nid))
2000                         nid = 0;
2001
2002                 if (++i >= FREE_NID_PAGES)
2003                         break;
2004         }
2005
2006         /* go to the next free nat pages to find free nids abundantly */
2007         nm_i->next_scan_nid = nid;
2008
2009         /* find free nids from current sum_pages */
2010         down_read(&curseg->journal_rwsem);
2011         for (i = 0; i < nats_in_cursum(journal); i++) {
2012                 block_t addr;
2013
2014                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2015                 nid = le32_to_cpu(nid_in_journal(journal, i));
2016                 if (addr == NULL_ADDR)
2017                         add_free_nid(sbi, nid, true);
2018                 else
2019                         remove_free_nid(sbi, nid);
2020         }
2021         up_read(&curseg->journal_rwsem);
2022         up_read(&nm_i->nat_tree_lock);
2023
2024         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2025                                         nm_i->ra_nid_pages, META_NAT, false);
2026 }
2027
2028 void build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2029 {
2030         mutex_lock(&NM_I(sbi)->build_lock);
2031         __build_free_nids(sbi, sync, mount);
2032         mutex_unlock(&NM_I(sbi)->build_lock);
2033 }
2034
2035 /*
2036  * If this function returns success, caller can obtain a new nid
2037  * from second parameter of this function.
2038  * The returned nid could be used ino as well as nid when inode is created.
2039  */
2040 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2041 {
2042         struct f2fs_nm_info *nm_i = NM_I(sbi);
2043         struct free_nid *i = NULL;
2044 retry:
2045 #ifdef CONFIG_F2FS_FAULT_INJECTION
2046         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2047                 f2fs_show_injection_info(FAULT_ALLOC_NID);
2048                 return false;
2049         }
2050 #endif
2051         spin_lock(&nm_i->nid_list_lock);
2052
2053         if (unlikely(nm_i->available_nids == 0)) {
2054                 spin_unlock(&nm_i->nid_list_lock);
2055                 return false;
2056         }
2057
2058         /* We should not use stale free nids created by build_free_nids */
2059         if (nm_i->nid_cnt[FREE_NID_LIST] && !on_build_free_nids(nm_i)) {
2060                 f2fs_bug_on(sbi, list_empty(&nm_i->nid_list[FREE_NID_LIST]));
2061                 i = list_first_entry(&nm_i->nid_list[FREE_NID_LIST],
2062                                         struct free_nid, list);
2063                 *nid = i->nid;
2064
2065                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, true);
2066                 i->state = NID_ALLOC;
2067                 __insert_nid_to_list(sbi, i, ALLOC_NID_LIST, false);
2068                 nm_i->available_nids--;
2069
2070                 update_free_nid_bitmap(sbi, *nid, false, false);
2071
2072                 spin_unlock(&nm_i->nid_list_lock);
2073                 return true;
2074         }
2075         spin_unlock(&nm_i->nid_list_lock);
2076
2077         /* Let's scan nat pages and its caches to get free nids */
2078         build_free_nids(sbi, true, false);
2079         goto retry;
2080 }
2081
2082 /*
2083  * alloc_nid() should be called prior to this function.
2084  */
2085 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2086 {
2087         struct f2fs_nm_info *nm_i = NM_I(sbi);
2088         struct free_nid *i;
2089
2090         spin_lock(&nm_i->nid_list_lock);
2091         i = __lookup_free_nid_list(nm_i, nid);
2092         f2fs_bug_on(sbi, !i);
2093         __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
2094         spin_unlock(&nm_i->nid_list_lock);
2095
2096         kmem_cache_free(free_nid_slab, i);
2097 }
2098
2099 /*
2100  * alloc_nid() should be called prior to this function.
2101  */
2102 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2103 {
2104         struct f2fs_nm_info *nm_i = NM_I(sbi);
2105         struct free_nid *i;
2106         bool need_free = false;
2107
2108         if (!nid)
2109                 return;
2110
2111         spin_lock(&nm_i->nid_list_lock);
2112         i = __lookup_free_nid_list(nm_i, nid);
2113         f2fs_bug_on(sbi, !i);
2114
2115         if (!available_free_memory(sbi, FREE_NIDS)) {
2116                 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
2117                 need_free = true;
2118         } else {
2119                 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, true);
2120                 i->state = NID_NEW;
2121                 __insert_nid_to_list(sbi, i, FREE_NID_LIST, false);
2122         }
2123
2124         nm_i->available_nids++;
2125
2126         update_free_nid_bitmap(sbi, nid, true, false);
2127
2128         spin_unlock(&nm_i->nid_list_lock);
2129
2130         if (need_free)
2131                 kmem_cache_free(free_nid_slab, i);
2132 }
2133
2134 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2135 {
2136         struct f2fs_nm_info *nm_i = NM_I(sbi);
2137         struct free_nid *i, *next;
2138         int nr = nr_shrink;
2139
2140         if (nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
2141                 return 0;
2142
2143         if (!mutex_trylock(&nm_i->build_lock))
2144                 return 0;
2145
2146         spin_lock(&nm_i->nid_list_lock);
2147         list_for_each_entry_safe(i, next, &nm_i->nid_list[FREE_NID_LIST],
2148                                                                         list) {
2149                 if (nr_shrink <= 0 ||
2150                                 nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
2151                         break;
2152
2153                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2154                 kmem_cache_free(free_nid_slab, i);
2155                 nr_shrink--;
2156         }
2157         spin_unlock(&nm_i->nid_list_lock);
2158         mutex_unlock(&nm_i->build_lock);
2159
2160         return nr - nr_shrink;
2161 }
2162
2163 void recover_inline_xattr(struct inode *inode, struct page *page)
2164 {
2165         void *src_addr, *dst_addr;
2166         size_t inline_size;
2167         struct page *ipage;
2168         struct f2fs_inode *ri;
2169
2170         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2171         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2172
2173         ri = F2FS_INODE(page);
2174         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2175                 clear_inode_flag(inode, FI_INLINE_XATTR);
2176                 goto update_inode;
2177         }
2178
2179         dst_addr = inline_xattr_addr(ipage);
2180         src_addr = inline_xattr_addr(page);
2181         inline_size = inline_xattr_size(inode);
2182
2183         f2fs_wait_on_page_writeback(ipage, NODE, true);
2184         memcpy(dst_addr, src_addr, inline_size);
2185 update_inode:
2186         update_inode(inode, ipage);
2187         f2fs_put_page(ipage, 1);
2188 }
2189
2190 int recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2191 {
2192         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2193         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2194         nid_t new_xnid = nid_of_node(page);
2195         struct node_info ni;
2196         struct page *xpage;
2197
2198         if (!prev_xnid)
2199                 goto recover_xnid;
2200
2201         /* 1: invalidate the previous xattr nid */
2202         get_node_info(sbi, prev_xnid, &ni);
2203         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2204         invalidate_blocks(sbi, ni.blk_addr);
2205         dec_valid_node_count(sbi, inode, false);
2206         set_node_addr(sbi, &ni, NULL_ADDR, false);
2207
2208 recover_xnid:
2209         /* 2: update xattr nid in inode */
2210         remove_free_nid(sbi, new_xnid);
2211         f2fs_i_xnid_write(inode, new_xnid);
2212         if (unlikely(inc_valid_node_count(sbi, inode, false)))
2213                 f2fs_bug_on(sbi, 1);
2214         update_inode_page(inode);
2215
2216         /* 3: update and set xattr node page dirty */
2217         xpage = grab_cache_page(NODE_MAPPING(sbi), new_xnid);
2218         if (!xpage)
2219                 return -ENOMEM;
2220
2221         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), PAGE_SIZE);
2222
2223         get_node_info(sbi, new_xnid, &ni);
2224         ni.ino = inode->i_ino;
2225         set_node_addr(sbi, &ni, NEW_ADDR, false);
2226         set_page_dirty(xpage);
2227         f2fs_put_page(xpage, 1);
2228
2229         return 0;
2230 }
2231
2232 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2233 {
2234         struct f2fs_inode *src, *dst;
2235         nid_t ino = ino_of_node(page);
2236         struct node_info old_ni, new_ni;
2237         struct page *ipage;
2238
2239         get_node_info(sbi, ino, &old_ni);
2240
2241         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2242                 return -EINVAL;
2243 retry:
2244         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2245         if (!ipage) {
2246                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2247                 goto retry;
2248         }
2249
2250         /* Should not use this inode from free nid list */
2251         remove_free_nid(sbi, ino);
2252
2253         if (!PageUptodate(ipage))
2254                 SetPageUptodate(ipage);
2255         fill_node_footer(ipage, ino, ino, 0, true);
2256
2257         src = F2FS_INODE(page);
2258         dst = F2FS_INODE(ipage);
2259
2260         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2261         dst->i_size = 0;
2262         dst->i_blocks = cpu_to_le64(1);
2263         dst->i_links = cpu_to_le32(1);
2264         dst->i_xattr_nid = 0;
2265         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2266
2267         new_ni = old_ni;
2268         new_ni.ino = ino;
2269
2270         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2271                 WARN_ON(1);
2272         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2273         inc_valid_inode_count(sbi);
2274         set_page_dirty(ipage);
2275         f2fs_put_page(ipage, 1);
2276         return 0;
2277 }
2278
2279 int restore_node_summary(struct f2fs_sb_info *sbi,
2280                         unsigned int segno, struct f2fs_summary_block *sum)
2281 {
2282         struct f2fs_node *rn;
2283         struct f2fs_summary *sum_entry;
2284         block_t addr;
2285         int i, idx, last_offset, nrpages;
2286
2287         /* scan the node segment */
2288         last_offset = sbi->blocks_per_seg;
2289         addr = START_BLOCK(sbi, segno);
2290         sum_entry = &sum->entries[0];
2291
2292         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2293                 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2294
2295                 /* readahead node pages */
2296                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2297
2298                 for (idx = addr; idx < addr + nrpages; idx++) {
2299                         struct page *page = get_tmp_page(sbi, idx);
2300
2301                         rn = F2FS_NODE(page);
2302                         sum_entry->nid = rn->footer.nid;
2303                         sum_entry->version = 0;
2304                         sum_entry->ofs_in_node = 0;
2305                         sum_entry++;
2306                         f2fs_put_page(page, 1);
2307                 }
2308
2309                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2310                                                         addr + nrpages);
2311         }
2312         return 0;
2313 }
2314
2315 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2316 {
2317         struct f2fs_nm_info *nm_i = NM_I(sbi);
2318         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2319         struct f2fs_journal *journal = curseg->journal;
2320         int i;
2321
2322         down_write(&curseg->journal_rwsem);
2323         for (i = 0; i < nats_in_cursum(journal); i++) {
2324                 struct nat_entry *ne;
2325                 struct f2fs_nat_entry raw_ne;
2326                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2327
2328                 raw_ne = nat_in_journal(journal, i);
2329
2330                 ne = __lookup_nat_cache(nm_i, nid);
2331                 if (!ne) {
2332                         ne = grab_nat_entry(nm_i, nid, true);
2333                         node_info_from_raw_nat(&ne->ni, &raw_ne);
2334                 }
2335
2336                 /*
2337                  * if a free nat in journal has not been used after last
2338                  * checkpoint, we should remove it from available nids,
2339                  * since later we will add it again.
2340                  */
2341                 if (!get_nat_flag(ne, IS_DIRTY) &&
2342                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2343                         spin_lock(&nm_i->nid_list_lock);
2344                         nm_i->available_nids--;
2345                         spin_unlock(&nm_i->nid_list_lock);
2346                 }
2347
2348                 __set_nat_cache_dirty(nm_i, ne);
2349         }
2350         update_nats_in_cursum(journal, -i);
2351         up_write(&curseg->journal_rwsem);
2352 }
2353
2354 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2355                                                 struct list_head *head, int max)
2356 {
2357         struct nat_entry_set *cur;
2358
2359         if (nes->entry_cnt >= max)
2360                 goto add_out;
2361
2362         list_for_each_entry(cur, head, set_list) {
2363                 if (cur->entry_cnt >= nes->entry_cnt) {
2364                         list_add(&nes->set_list, cur->set_list.prev);
2365                         return;
2366                 }
2367         }
2368 add_out:
2369         list_add_tail(&nes->set_list, head);
2370 }
2371
2372 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2373                                                 struct page *page)
2374 {
2375         struct f2fs_nm_info *nm_i = NM_I(sbi);
2376         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2377         struct f2fs_nat_block *nat_blk = page_address(page);
2378         int valid = 0;
2379         int i;
2380
2381         if (!enabled_nat_bits(sbi, NULL))
2382                 return;
2383
2384         for (i = 0; i < NAT_ENTRY_PER_BLOCK; i++) {
2385                 if (start_nid == 0 && i == 0)
2386                         valid++;
2387                 if (nat_blk->entries[i].block_addr)
2388                         valid++;
2389         }
2390         if (valid == 0) {
2391                 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2392                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2393                 return;
2394         }
2395
2396         __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2397         if (valid == NAT_ENTRY_PER_BLOCK)
2398                 __set_bit_le(nat_index, nm_i->full_nat_bits);
2399         else
2400                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2401 }
2402
2403 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2404                 struct nat_entry_set *set, struct cp_control *cpc)
2405 {
2406         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2407         struct f2fs_journal *journal = curseg->journal;
2408         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2409         bool to_journal = true;
2410         struct f2fs_nat_block *nat_blk;
2411         struct nat_entry *ne, *cur;
2412         struct page *page = NULL;
2413
2414         /*
2415          * there are two steps to flush nat entries:
2416          * #1, flush nat entries to journal in current hot data summary block.
2417          * #2, flush nat entries to nat page.
2418          */
2419         if (enabled_nat_bits(sbi, cpc) ||
2420                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2421                 to_journal = false;
2422
2423         if (to_journal) {
2424                 down_write(&curseg->journal_rwsem);
2425         } else {
2426                 page = get_next_nat_page(sbi, start_nid);
2427                 nat_blk = page_address(page);
2428                 f2fs_bug_on(sbi, !nat_blk);
2429         }
2430
2431         /* flush dirty nats in nat entry set */
2432         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2433                 struct f2fs_nat_entry *raw_ne;
2434                 nid_t nid = nat_get_nid(ne);
2435                 int offset;
2436
2437                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2438
2439                 if (to_journal) {
2440                         offset = lookup_journal_in_cursum(journal,
2441                                                         NAT_JOURNAL, nid, 1);
2442                         f2fs_bug_on(sbi, offset < 0);
2443                         raw_ne = &nat_in_journal(journal, offset);
2444                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2445                 } else {
2446                         raw_ne = &nat_blk->entries[nid - start_nid];
2447                 }
2448                 raw_nat_from_node_info(raw_ne, &ne->ni);
2449                 nat_reset_flag(ne);
2450                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2451                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2452                         add_free_nid(sbi, nid, false);
2453                         spin_lock(&NM_I(sbi)->nid_list_lock);
2454                         NM_I(sbi)->available_nids++;
2455                         update_free_nid_bitmap(sbi, nid, true, false);
2456                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2457                 } else {
2458                         spin_lock(&NM_I(sbi)->nid_list_lock);
2459                         update_free_nid_bitmap(sbi, nid, false, false);
2460                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2461                 }
2462         }
2463
2464         if (to_journal) {
2465                 up_write(&curseg->journal_rwsem);
2466         } else {
2467                 __update_nat_bits(sbi, start_nid, page);
2468                 f2fs_put_page(page, 1);
2469         }
2470
2471         /* Allow dirty nats by node block allocation in write_begin */
2472         if (!set->entry_cnt) {
2473                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2474                 kmem_cache_free(nat_entry_set_slab, set);
2475         }
2476 }
2477
2478 /*
2479  * This function is called during the checkpointing process.
2480  */
2481 void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2482 {
2483         struct f2fs_nm_info *nm_i = NM_I(sbi);
2484         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2485         struct f2fs_journal *journal = curseg->journal;
2486         struct nat_entry_set *setvec[SETVEC_SIZE];
2487         struct nat_entry_set *set, *tmp;
2488         unsigned int found;
2489         nid_t set_idx = 0;
2490         LIST_HEAD(sets);
2491
2492         if (!nm_i->dirty_nat_cnt)
2493                 return;
2494
2495         down_write(&nm_i->nat_tree_lock);
2496
2497         /*
2498          * if there are no enough space in journal to store dirty nat
2499          * entries, remove all entries from journal and merge them
2500          * into nat entry set.
2501          */
2502         if (enabled_nat_bits(sbi, cpc) ||
2503                 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2504                 remove_nats_in_journal(sbi);
2505
2506         while ((found = __gang_lookup_nat_set(nm_i,
2507                                         set_idx, SETVEC_SIZE, setvec))) {
2508                 unsigned idx;
2509                 set_idx = setvec[found - 1]->set + 1;
2510                 for (idx = 0; idx < found; idx++)
2511                         __adjust_nat_entry_set(setvec[idx], &sets,
2512                                                 MAX_NAT_JENTRIES(journal));
2513         }
2514
2515         /* flush dirty nats in nat entry set */
2516         list_for_each_entry_safe(set, tmp, &sets, set_list)
2517                 __flush_nat_entry_set(sbi, set, cpc);
2518
2519         up_write(&nm_i->nat_tree_lock);
2520         /* Allow dirty nats by node block allocation in write_begin */
2521 }
2522
2523 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2524 {
2525         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2526         struct f2fs_nm_info *nm_i = NM_I(sbi);
2527         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2528         unsigned int i;
2529         __u64 cp_ver = cur_cp_version(ckpt);
2530         block_t nat_bits_addr;
2531
2532         if (!enabled_nat_bits(sbi, NULL))
2533                 return 0;
2534
2535         nm_i->nat_bits_blocks = F2FS_BYTES_TO_BLK((nat_bits_bytes << 1) + 8 +
2536                                                 F2FS_BLKSIZE - 1);
2537         nm_i->nat_bits = kzalloc(nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS,
2538                                                 GFP_KERNEL);
2539         if (!nm_i->nat_bits)
2540                 return -ENOMEM;
2541
2542         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2543                                                 nm_i->nat_bits_blocks;
2544         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2545                 struct page *page = get_meta_page(sbi, nat_bits_addr++);
2546
2547                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2548                                         page_address(page), F2FS_BLKSIZE);
2549                 f2fs_put_page(page, 1);
2550         }
2551
2552         cp_ver |= (cur_cp_crc(ckpt) << 32);
2553         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2554                 disable_nat_bits(sbi, true);
2555                 return 0;
2556         }
2557
2558         nm_i->full_nat_bits = nm_i->nat_bits + 8;
2559         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2560
2561         f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2562         return 0;
2563 }
2564
2565 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2566 {
2567         struct f2fs_nm_info *nm_i = NM_I(sbi);
2568         unsigned int i = 0;
2569         nid_t nid, last_nid;
2570
2571         if (!enabled_nat_bits(sbi, NULL))
2572                 return;
2573
2574         for (i = 0; i < nm_i->nat_blocks; i++) {
2575                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2576                 if (i >= nm_i->nat_blocks)
2577                         break;
2578
2579                 __set_bit_le(i, nm_i->nat_block_bitmap);
2580
2581                 nid = i * NAT_ENTRY_PER_BLOCK;
2582                 last_nid = (i + 1) * NAT_ENTRY_PER_BLOCK;
2583
2584                 spin_lock(&NM_I(sbi)->nid_list_lock);
2585                 for (; nid < last_nid; nid++)
2586                         update_free_nid_bitmap(sbi, nid, true, true);
2587                 spin_unlock(&NM_I(sbi)->nid_list_lock);
2588         }
2589
2590         for (i = 0; i < nm_i->nat_blocks; i++) {
2591                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2592                 if (i >= nm_i->nat_blocks)
2593                         break;
2594
2595                 __set_bit_le(i, nm_i->nat_block_bitmap);
2596         }
2597 }
2598
2599 static int init_node_manager(struct f2fs_sb_info *sbi)
2600 {
2601         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2602         struct f2fs_nm_info *nm_i = NM_I(sbi);
2603         unsigned char *version_bitmap;
2604         unsigned int nat_segs;
2605         int err;
2606
2607         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2608
2609         /* segment_count_nat includes pair segment so divide to 2. */
2610         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2611         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2612         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2613
2614         /* not used nids: 0, node, meta, (and root counted as valid node) */
2615         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2616                                                         F2FS_RESERVED_NODE_NUM;
2617         nm_i->nid_cnt[FREE_NID_LIST] = 0;
2618         nm_i->nid_cnt[ALLOC_NID_LIST] = 0;
2619         nm_i->nat_cnt = 0;
2620         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2621         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2622         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2623
2624         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2625         INIT_LIST_HEAD(&nm_i->nid_list[FREE_NID_LIST]);
2626         INIT_LIST_HEAD(&nm_i->nid_list[ALLOC_NID_LIST]);
2627         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2628         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2629         INIT_LIST_HEAD(&nm_i->nat_entries);
2630
2631         mutex_init(&nm_i->build_lock);
2632         spin_lock_init(&nm_i->nid_list_lock);
2633         init_rwsem(&nm_i->nat_tree_lock);
2634
2635         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2636         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2637         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2638         if (!version_bitmap)
2639                 return -EFAULT;
2640
2641         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2642                                         GFP_KERNEL);
2643         if (!nm_i->nat_bitmap)
2644                 return -ENOMEM;
2645
2646         err = __get_nat_bitmaps(sbi);
2647         if (err)
2648                 return err;
2649
2650 #ifdef CONFIG_F2FS_CHECK_FS
2651         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2652                                         GFP_KERNEL);
2653         if (!nm_i->nat_bitmap_mir)
2654                 return -ENOMEM;
2655 #endif
2656
2657         return 0;
2658 }
2659
2660 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2661 {
2662         struct f2fs_nm_info *nm_i = NM_I(sbi);
2663
2664         nm_i->free_nid_bitmap = f2fs_kvzalloc(nm_i->nat_blocks *
2665                                         NAT_ENTRY_BITMAP_SIZE, GFP_KERNEL);
2666         if (!nm_i->free_nid_bitmap)
2667                 return -ENOMEM;
2668
2669         nm_i->nat_block_bitmap = f2fs_kvzalloc(nm_i->nat_blocks / 8,
2670                                                                 GFP_KERNEL);
2671         if (!nm_i->nat_block_bitmap)
2672                 return -ENOMEM;
2673
2674         nm_i->free_nid_count = f2fs_kvzalloc(nm_i->nat_blocks *
2675                                         sizeof(unsigned short), GFP_KERNEL);
2676         if (!nm_i->free_nid_count)
2677                 return -ENOMEM;
2678         return 0;
2679 }
2680
2681 int build_node_manager(struct f2fs_sb_info *sbi)
2682 {
2683         int err;
2684
2685         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2686         if (!sbi->nm_info)
2687                 return -ENOMEM;
2688
2689         err = init_node_manager(sbi);
2690         if (err)
2691                 return err;
2692
2693         err = init_free_nid_cache(sbi);
2694         if (err)
2695                 return err;
2696
2697         /* load free nid status from nat_bits table */
2698         load_free_nid_bitmap(sbi);
2699
2700         build_free_nids(sbi, true, true);
2701         return 0;
2702 }
2703
2704 void destroy_node_manager(struct f2fs_sb_info *sbi)
2705 {
2706         struct f2fs_nm_info *nm_i = NM_I(sbi);
2707         struct free_nid *i, *next_i;
2708         struct nat_entry *natvec[NATVEC_SIZE];
2709         struct nat_entry_set *setvec[SETVEC_SIZE];
2710         nid_t nid = 0;
2711         unsigned int found;
2712
2713         if (!nm_i)
2714                 return;
2715
2716         /* destroy free nid list */
2717         spin_lock(&nm_i->nid_list_lock);
2718         list_for_each_entry_safe(i, next_i, &nm_i->nid_list[FREE_NID_LIST],
2719                                                                         list) {
2720                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2721                 spin_unlock(&nm_i->nid_list_lock);
2722                 kmem_cache_free(free_nid_slab, i);
2723                 spin_lock(&nm_i->nid_list_lock);
2724         }
2725         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID_LIST]);
2726         f2fs_bug_on(sbi, nm_i->nid_cnt[ALLOC_NID_LIST]);
2727         f2fs_bug_on(sbi, !list_empty(&nm_i->nid_list[ALLOC_NID_LIST]));
2728         spin_unlock(&nm_i->nid_list_lock);
2729
2730         /* destroy nat cache */
2731         down_write(&nm_i->nat_tree_lock);
2732         while ((found = __gang_lookup_nat_cache(nm_i,
2733                                         nid, NATVEC_SIZE, natvec))) {
2734                 unsigned idx;
2735
2736                 nid = nat_get_nid(natvec[found - 1]) + 1;
2737                 for (idx = 0; idx < found; idx++)
2738                         __del_from_nat_cache(nm_i, natvec[idx]);
2739         }
2740         f2fs_bug_on(sbi, nm_i->nat_cnt);
2741
2742         /* destroy nat set cache */
2743         nid = 0;
2744         while ((found = __gang_lookup_nat_set(nm_i,
2745                                         nid, SETVEC_SIZE, setvec))) {
2746                 unsigned idx;
2747
2748                 nid = setvec[found - 1]->set + 1;
2749                 for (idx = 0; idx < found; idx++) {
2750                         /* entry_cnt is not zero, when cp_error was occurred */
2751                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2752                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2753                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2754                 }
2755         }
2756         up_write(&nm_i->nat_tree_lock);
2757
2758         kvfree(nm_i->nat_block_bitmap);
2759         kvfree(nm_i->free_nid_bitmap);
2760         kvfree(nm_i->free_nid_count);
2761
2762         kfree(nm_i->nat_bitmap);
2763         kfree(nm_i->nat_bits);
2764 #ifdef CONFIG_F2FS_CHECK_FS
2765         kfree(nm_i->nat_bitmap_mir);
2766 #endif
2767         sbi->nm_info = NULL;
2768         kfree(nm_i);
2769 }
2770
2771 int __init create_node_manager_caches(void)
2772 {
2773         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2774                         sizeof(struct nat_entry));
2775         if (!nat_entry_slab)
2776                 goto fail;
2777
2778         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2779                         sizeof(struct free_nid));
2780         if (!free_nid_slab)
2781                 goto destroy_nat_entry;
2782
2783         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2784                         sizeof(struct nat_entry_set));
2785         if (!nat_entry_set_slab)
2786                 goto destroy_free_nid;
2787         return 0;
2788
2789 destroy_free_nid:
2790         kmem_cache_destroy(free_nid_slab);
2791 destroy_nat_entry:
2792         kmem_cache_destroy(nat_entry_slab);
2793 fail:
2794         return -ENOMEM;
2795 }
2796
2797 void destroy_node_manager_caches(void)
2798 {
2799         kmem_cache_destroy(nat_entry_set_slab);
2800         kmem_cache_destroy(free_nid_slab);
2801         kmem_cache_destroy(nat_entry_slab);
2802 }