fs/f2fs/node.c

   1 /*
   2  * fs/f2fs/node.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/mpage.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/swap.h>
  18
  19 #include "f2fs.h"
  20 #include "node.h"
  21 #include "segment.h"
  22 #include "xattr.h"
  23 #include "trace.h"
  24 #include <trace/events/f2fs.h>
  25
  26 #define on_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
  27
  28 static struct kmem_cache *nat_entry_slab;
  29 static struct kmem_cache *free_nid_slab;
  30 static struct kmem_cache *nat_entry_set_slab;
  31
  32 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
  33 {
  34         struct f2fs_nm_info *nm_i = NM_I(sbi);
  35         struct sysinfo val;
  36         unsigned long avail_ram;
  37         unsigned long mem_size = 0;
  38         bool res = false;
  39
  40         si_meminfo(&val);
  41
  42         /* only uses low memory */
  43         avail_ram = val.totalram - val.totalhigh;
  44
  45         /*
  46          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  47          */
  48         if (type == FREE_NIDS) {
  49                 mem_size = (nm_i->nid_cnt[FREE_NID] *
  50                                 sizeof(struct free_nid)) >> PAGE_SHIFT;
  51                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  52         } else if (type == NAT_ENTRIES) {
  53                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
  54                                                         PAGE_SHIFT;
  55                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  56                 if (excess_cached_nats(sbi))
  57                         res = false;
  58         } else if (type == DIRTY_DENTS) {
  59                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  60                         return false;
  61                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  62                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  63         } else if (type == INO_ENTRIES) {
  64                 int i;
  65
  66                 for (i = 0; i < MAX_INO_ENTRY; i++)
  67                         mem_size += sbi->im[i].ino_num *
  68                                                 sizeof(struct ino_entry);
  69                 mem_size >>= PAGE_SHIFT;
  70                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  71         } else if (type == EXTENT_CACHE) {
  72                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  73                                 sizeof(struct extent_tree) +
  74                                 atomic_read(&sbi->total_ext_node) *
  75                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  76                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  77         } else if (type == INMEM_PAGES) {
  78                 /* it allows 20% / total_ram for inmemory pages */
  79                 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
  80                 res = mem_size < (val.totalram / 5);
  81         } else {
  82                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
  83                         return true;
  84         }
  85         return res;
  86 }
  87
  88 static void clear_node_page_dirty(struct page *page)
  89 {
  90         struct address_space *mapping = page->mapping;
  91         unsigned int long flags;
  92
  93         if (PageDirty(page)) {
  94                 spin_lock_irqsave(&mapping->tree_lock, flags);
  95                 radix_tree_tag_clear(&mapping->page_tree,
  96                                 page_index(page),
  97                                 PAGECACHE_TAG_DIRTY);
  98                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  99
 100                 clear_page_dirty_for_io(page);
 101                 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
 102         }
 103         ClearPageUptodate(page);
 104 }
 105
 106 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 107 {
 108         pgoff_t index = current_nat_addr(sbi, nid);
 109         return get_meta_page(sbi, index);
 110 }
 111
 112 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 113 {
 114         struct page *src_page;
 115         struct page *dst_page;
 116         pgoff_t src_off;
 117         pgoff_t dst_off;
 118         void *src_addr;
 119         void *dst_addr;
 120         struct f2fs_nm_info *nm_i = NM_I(sbi);
 121
 122         src_off = current_nat_addr(sbi, nid);
 123         dst_off = next_nat_addr(sbi, src_off);
 124
 125         /* get current nat block page with lock */
 126         src_page = get_meta_page(sbi, src_off);
 127         dst_page = grab_meta_page(sbi, dst_off);
 128         f2fs_bug_on(sbi, PageDirty(src_page));
 129
 130         src_addr = page_address(src_page);
 131         dst_addr = page_address(dst_page);
 132         memcpy(dst_addr, src_addr, PAGE_SIZE);
 133         set_page_dirty(dst_page);
 134         f2fs_put_page(src_page, 1);
 135
 136         set_to_next_nat(nm_i, nid);
 137
 138         return dst_page;
 139 }
 140
 141 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
 142 {
 143         struct nat_entry *new;
 144
 145         if (no_fail)
 146                 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
 147         else
 148                 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
 149         if (new) {
 150                 nat_set_nid(new, nid);
 151                 nat_reset_flag(new);
 152         }
 153         return new;
 154 }
 155
 156 static void __free_nat_entry(struct nat_entry *e)
 157 {
 158         kmem_cache_free(nat_entry_slab, e);
 159 }
 160
 161 /* must be locked by nat_tree_lock */
 162 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
 163         struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
 164 {
 165         if (no_fail)
 166                 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
 167         else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
 168                 return NULL;
 169
 170         if (raw_ne)
 171                 node_info_from_raw_nat(&ne->ni, raw_ne);
 172         list_add_tail(&ne->list, &nm_i->nat_entries);
 173         nm_i->nat_cnt++;
 174         return ne;
 175 }
 176
 177 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 178 {
 179         return radix_tree_lookup(&nm_i->nat_root, n);
 180 }
 181
 182 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 183                 nid_t start, unsigned int nr, struct nat_entry **ep)
 184 {
 185         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 186 }
 187
 188 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 189 {
 190         list_del(&e->list);
 191         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 192         nm_i->nat_cnt--;
 193         __free_nat_entry(e);
 194 }
 195
 196 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 197                                                 struct nat_entry *ne)
 198 {
 199         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 200         struct nat_entry_set *head;
 201
 202         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 203         if (!head) {
 204                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 205
 206                 INIT_LIST_HEAD(&head->entry_list);
 207                 INIT_LIST_HEAD(&head->set_list);
 208                 head->set = set;
 209                 head->entry_cnt = 0;
 210                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 211         }
 212
 213         if (get_nat_flag(ne, IS_DIRTY))
 214                 goto refresh_list;
 215
 216         nm_i->dirty_nat_cnt++;
 217         head->entry_cnt++;
 218         set_nat_flag(ne, IS_DIRTY, true);
 219 refresh_list:
 220         if (nat_get_blkaddr(ne) == NEW_ADDR)
 221                 list_del_init(&ne->list);
 222         else
 223                 list_move_tail(&ne->list, &head->entry_list);
 224 }
 225
 226 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 227                 struct nat_entry_set *set, struct nat_entry *ne)
 228 {
 229         list_move_tail(&ne->list, &nm_i->nat_entries);
 230         set_nat_flag(ne, IS_DIRTY, false);
 231         set->entry_cnt--;
 232         nm_i->dirty_nat_cnt--;
 233 }
 234
 235 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 236                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 237 {
 238         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 239                                                         start, nr);
 240 }
 241
 242 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 243 {
 244         struct f2fs_nm_info *nm_i = NM_I(sbi);
 245         struct nat_entry *e;
 246         bool need = false;
 247
 248         down_read(&nm_i->nat_tree_lock);
 249         e = __lookup_nat_cache(nm_i, nid);
 250         if (e) {
 251                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 252                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 253                         need = true;
 254         }
 255         up_read(&nm_i->nat_tree_lock);
 256         return need;
 257 }
 258
 259 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 260 {
 261         struct f2fs_nm_info *nm_i = NM_I(sbi);
 262         struct nat_entry *e;
 263         bool is_cp = true;
 264
 265         down_read(&nm_i->nat_tree_lock);
 266         e = __lookup_nat_cache(nm_i, nid);
 267         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 268                 is_cp = false;
 269         up_read(&nm_i->nat_tree_lock);
 270         return is_cp;
 271 }
 272
 273 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 274 {
 275         struct f2fs_nm_info *nm_i = NM_I(sbi);
 276         struct nat_entry *e;
 277         bool need_update = true;
 278
 279         down_read(&nm_i->nat_tree_lock);
 280         e = __lookup_nat_cache(nm_i, ino);
 281         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 282                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 283                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 284                 need_update = false;
 285         up_read(&nm_i->nat_tree_lock);
 286         return need_update;
 287 }
 288
 289 /* must be locked by nat_tree_lock */
 290 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 291                                                 struct f2fs_nat_entry *ne)
 292 {
 293         struct f2fs_nm_info *nm_i = NM_I(sbi);
 294         struct nat_entry *new, *e;
 295
 296         new = __alloc_nat_entry(nid, false);
 297         if (!new)
 298                 return;
 299
 300         down_write(&nm_i->nat_tree_lock);
 301         e = __lookup_nat_cache(nm_i, nid);
 302         if (!e)
 303                 e = __init_nat_entry(nm_i, new, ne, false);
 304         else
 305                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 306                                 nat_get_blkaddr(e) !=
 307                                         le32_to_cpu(ne->block_addr) ||
 308                                 nat_get_version(e) != ne->version);
 309         up_write(&nm_i->nat_tree_lock);
 310         if (e != new)
 311                 __free_nat_entry(new);
 312 }
 313
 314 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 315                         block_t new_blkaddr, bool fsync_done)
 316 {
 317         struct f2fs_nm_info *nm_i = NM_I(sbi);
 318         struct nat_entry *e;
 319         struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
 320
 321         down_write(&nm_i->nat_tree_lock);
 322         e = __lookup_nat_cache(nm_i, ni->nid);
 323         if (!e) {
 324                 e = __init_nat_entry(nm_i, new, NULL, true);
 325                 copy_node_info(&e->ni, ni);
 326                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 327         } else if (new_blkaddr == NEW_ADDR) {
 328                 /*
 329                  * when nid is reallocated,
 330                  * previous nat entry can be remained in nat cache.
 331                  * So, reinitialize it with new information.
 332                  */
 333                 copy_node_info(&e->ni, ni);
 334                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 335         }
 336         /* let's free early to reduce memory consumption */
 337         if (e != new)
 338                 __free_nat_entry(new);
 339
 340         /* sanity check */
 341         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 342         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 343                         new_blkaddr == NULL_ADDR);
 344         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 345                         new_blkaddr == NEW_ADDR);
 346         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 347                         nat_get_blkaddr(e) != NULL_ADDR &&
 348                         new_blkaddr == NEW_ADDR);
 349
 350         /* increment version no as node is removed */
 351         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 352                 unsigned char version = nat_get_version(e);
 353                 nat_set_version(e, inc_node_version(version));
 354         }
 355
 356         /* change address */
 357         nat_set_blkaddr(e, new_blkaddr);
 358         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 359                 set_nat_flag(e, IS_CHECKPOINTED, false);
 360         __set_nat_cache_dirty(nm_i, e);
 361
 362         /* update fsync_mark if its inode nat entry is still alive */
 363         if (ni->nid != ni->ino)
 364                 e = __lookup_nat_cache(nm_i, ni->ino);
 365         if (e) {
 366                 if (fsync_done && ni->nid == ni->ino)
 367                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 368                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 369         }
 370         up_write(&nm_i->nat_tree_lock);
 371 }
 372
 373 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 374 {
 375         struct f2fs_nm_info *nm_i = NM_I(sbi);
 376         int nr = nr_shrink;
 377
 378         if (!down_write_trylock(&nm_i->nat_tree_lock))
 379                 return 0;
 380
 381         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 382                 struct nat_entry *ne;
 383                 ne = list_first_entry(&nm_i->nat_entries,
 384                                         struct nat_entry, list);
 385                 __del_from_nat_cache(nm_i, ne);
 386                 nr_shrink--;
 387         }
 388         up_write(&nm_i->nat_tree_lock);
 389         return nr - nr_shrink;
 390 }
 391
 392 /*
 393  * This function always returns success
 394  */
 395 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 396 {
 397         struct f2fs_nm_info *nm_i = NM_I(sbi);
 398         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 399         struct f2fs_journal *journal = curseg->journal;
 400         nid_t start_nid = START_NID(nid);
 401         struct f2fs_nat_block *nat_blk;
 402         struct page *page = NULL;
 403         struct f2fs_nat_entry ne;
 404         struct nat_entry *e;
 405         pgoff_t index;
 406         int i;
 407
 408         ni->nid = nid;
 409
 410         /* Check nat cache */
 411         down_read(&nm_i->nat_tree_lock);
 412         e = __lookup_nat_cache(nm_i, nid);
 413         if (e) {
 414                 ni->ino = nat_get_ino(e);
 415                 ni->blk_addr = nat_get_blkaddr(e);
 416                 ni->version = nat_get_version(e);
 417                 up_read(&nm_i->nat_tree_lock);
 418                 return;
 419         }
 420
 421         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 422
 423         /* Check current segment summary */
 424         down_read(&curseg->journal_rwsem);
 425         i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 426         if (i >= 0) {
 427                 ne = nat_in_journal(journal, i);
 428                 node_info_from_raw_nat(ni, &ne);
 429         }
 430         up_read(&curseg->journal_rwsem);
 431         if (i >= 0) {
 432                 up_read(&nm_i->nat_tree_lock);
 433                 goto cache;
 434         }
 435
 436         /* Fill node_info from nat page */
 437         index = current_nat_addr(sbi, nid);
 438         up_read(&nm_i->nat_tree_lock);
 439
 440         page = get_meta_page(sbi, index);
 441         nat_blk = (struct f2fs_nat_block *)page_address(page);
 442         ne = nat_blk->entries[nid - start_nid];
 443         node_info_from_raw_nat(ni, &ne);
 444         f2fs_put_page(page, 1);
 445 cache:
 446         /* cache nat entry */
 447         cache_nat_entry(sbi, nid, &ne);
 448 }
 449
 450 /*
 451  * readahead MAX_RA_NODE number of node pages.
 452  */
 453 static void ra_node_pages(struct page *parent, int start, int n)
 454 {
 455         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 456         struct blk_plug plug;
 457         int i, end;
 458         nid_t nid;
 459
 460         blk_start_plug(&plug);
 461
 462         /* Then, try readahead for siblings of the desired node */
 463         end = start + n;
 464         end = min(end, NIDS_PER_BLOCK);
 465         for (i = start; i < end; i++) {
 466                 nid = get_nid(parent, i, false);
 467                 ra_node_page(sbi, nid);
 468         }
 469
 470         blk_finish_plug(&plug);
 471 }
 472
 473 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 474 {
 475         const long direct_index = ADDRS_PER_INODE(dn->inode);
 476         const long direct_blks = ADDRS_PER_BLOCK;
 477         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 478         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 479         int cur_level = dn->cur_level;
 480         int max_level = dn->max_level;
 481         pgoff_t base = 0;
 482
 483         if (!dn->max_level)
 484                 return pgofs + 1;
 485
 486         while (max_level-- > cur_level)
 487                 skipped_unit *= NIDS_PER_BLOCK;
 488
 489         switch (dn->max_level) {
 490         case 3:
 491                 base += 2 * indirect_blks;
 492         case 2:
 493                 base += 2 * direct_blks;
 494         case 1:
 495                 base += direct_index;
 496                 break;
 497         default:
 498                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 499         }
 500
 501         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 502 }
 503
 504 /*
 505  * The maximum depth is four.
 506  * Offset[0] will have raw inode offset.
 507  */
 508 static int get_node_path(struct inode *inode, long block,
 509                                 int offset[4], unsigned int noffset[4])
 510 {
 511         const long direct_index = ADDRS_PER_INODE(inode);
 512         const long direct_blks = ADDRS_PER_BLOCK;
 513         const long dptrs_per_blk = NIDS_PER_BLOCK;
 514         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 515         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 516         int n = 0;
 517         int level = 0;
 518
 519         noffset[0] = 0;
 520
 521         if (block < direct_index) {
 522                 offset[n] = block;
 523                 goto got;
 524         }
 525         block -= direct_index;
 526         if (block < direct_blks) {
 527                 offset[n++] = NODE_DIR1_BLOCK;
 528                 noffset[n] = 1;
 529                 offset[n] = block;
 530                 level = 1;
 531                 goto got;
 532         }
 533         block -= direct_blks;
 534         if (block < direct_blks) {
 535                 offset[n++] = NODE_DIR2_BLOCK;
 536                 noffset[n] = 2;
 537                 offset[n] = block;
 538                 level = 1;
 539                 goto got;
 540         }
 541         block -= direct_blks;
 542         if (block < indirect_blks) {
 543                 offset[n++] = NODE_IND1_BLOCK;
 544                 noffset[n] = 3;
 545                 offset[n++] = block / direct_blks;
 546                 noffset[n] = 4 + offset[n - 1];
 547                 offset[n] = block % direct_blks;
 548                 level = 2;
 549                 goto got;
 550         }
 551         block -= indirect_blks;
 552         if (block < indirect_blks) {
 553                 offset[n++] = NODE_IND2_BLOCK;
 554                 noffset[n] = 4 + dptrs_per_blk;
 555                 offset[n++] = block / direct_blks;
 556                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 557                 offset[n] = block % direct_blks;
 558                 level = 2;
 559                 goto got;
 560         }
 561         block -= indirect_blks;
 562         if (block < dindirect_blks) {
 563                 offset[n++] = NODE_DIND_BLOCK;
 564                 noffset[n] = 5 + (dptrs_per_blk * 2);
 565                 offset[n++] = block / indirect_blks;
 566                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 567                               offset[n - 1] * (dptrs_per_blk + 1);
 568                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 569                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 570                               offset[n - 2] * (dptrs_per_blk + 1) +
 571                               offset[n - 1];
 572                 offset[n] = block % direct_blks;
 573                 level = 3;
 574                 goto got;
 575         } else {
 576                 return -E2BIG;
 577         }
 578 got:
 579         return level;
 580 }
 581
 582 /*
 583  * Caller should call f2fs_put_dnode(dn).
 584  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 585  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 586  * In the case of RDONLY_NODE, we don't need to care about mutex.
 587  */
 588 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 589 {
 590         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 591         struct page *npage[4];
 592         struct page *parent = NULL;
 593         int offset[4];
 594         unsigned int noffset[4];
 595         nid_t nids[4];
 596         int level, i = 0;
 597         int err = 0;
 598
 599         level = get_node_path(dn->inode, index, offset, noffset);
 600         if (level < 0)
 601                 return level;
 602
 603         nids[0] = dn->inode->i_ino;
 604         npage[0] = dn->inode_page;
 605
 606         if (!npage[0]) {
 607                 npage[0] = get_node_page(sbi, nids[0]);
 608                 if (IS_ERR(npage[0]))
 609                         return PTR_ERR(npage[0]);
 610         }
 611
 612         /* if inline_data is set, should not report any block indices */
 613         if (f2fs_has_inline_data(dn->inode) && index) {
 614                 err = -ENOENT;
 615                 f2fs_put_page(npage[0], 1);
 616                 goto release_out;
 617         }
 618
 619         parent = npage[0];
 620         if (level != 0)
 621                 nids[1] = get_nid(parent, offset[0], true);
 622         dn->inode_page = npage[0];
 623         dn->inode_page_locked = true;
 624
 625         /* get indirect or direct nodes */
 626         for (i = 1; i <= level; i++) {
 627                 bool done = false;
 628
 629                 if (!nids[i] && mode == ALLOC_NODE) {
 630                         /* alloc new node */
 631                         if (!alloc_nid(sbi, &(nids[i]))) {
 632                                 err = -ENOSPC;
 633                                 goto release_pages;
 634                         }
 635
 636                         dn->nid = nids[i];
 637                         npage[i] = new_node_page(dn, noffset[i]);
 638                         if (IS_ERR(npage[i])) {
 639                                 alloc_nid_failed(sbi, nids[i]);
 640                                 err = PTR_ERR(npage[i]);
 641                                 goto release_pages;
 642                         }
 643
 644                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 645                         alloc_nid_done(sbi, nids[i]);
 646                         done = true;
 647                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 648                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 649                         if (IS_ERR(npage[i])) {
 650                                 err = PTR_ERR(npage[i]);
 651                                 goto release_pages;
 652                         }
 653                         done = true;
 654                 }
 655                 if (i == 1) {
 656                         dn->inode_page_locked = false;
 657                         unlock_page(parent);
 658                 } else {
 659                         f2fs_put_page(parent, 1);
 660                 }
 661
 662                 if (!done) {
 663                         npage[i] = get_node_page(sbi, nids[i]);
 664                         if (IS_ERR(npage[i])) {
 665                                 err = PTR_ERR(npage[i]);
 666                                 f2fs_put_page(npage[0], 0);
 667                                 goto release_out;
 668                         }
 669                 }
 670                 if (i < level) {
 671                         parent = npage[i];
 672                         nids[i + 1] = get_nid(parent, offset[i], false);
 673                 }
 674         }
 675         dn->nid = nids[level];
 676         dn->ofs_in_node = offset[level];
 677         dn->node_page = npage[level];
 678         dn->data_blkaddr = datablock_addr(dn->inode,
 679                                 dn->node_page, dn->ofs_in_node);
 680         return 0;
 681
 682 release_pages:
 683         f2fs_put_page(parent, 1);
 684         if (i > 1)
 685                 f2fs_put_page(npage[0], 0);
 686 release_out:
 687         dn->inode_page = NULL;
 688         dn->node_page = NULL;
 689         if (err == -ENOENT) {
 690                 dn->cur_level = i;
 691                 dn->max_level = level;
 692                 dn->ofs_in_node = offset[level];
 693         }
 694         return err;
 695 }
 696
 697 static void truncate_node(struct dnode_of_data *dn)
 698 {
 699         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 700         struct node_info ni;
 701
 702         get_node_info(sbi, dn->nid, &ni);
 703
 704         /* Deallocate node address */
 705         invalidate_blocks(sbi, ni.blk_addr);
 706         dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
 707         set_node_addr(sbi, &ni, NULL_ADDR, false);
 708
 709         if (dn->nid == dn->inode->i_ino) {
 710                 remove_orphan_inode(sbi, dn->nid);
 711                 dec_valid_inode_count(sbi);
 712                 f2fs_inode_synced(dn->inode);
 713         }
 714
 715         clear_node_page_dirty(dn->node_page);
 716         set_sbi_flag(sbi, SBI_IS_DIRTY);
 717
 718         f2fs_put_page(dn->node_page, 1);
 719
 720         invalidate_mapping_pages(NODE_MAPPING(sbi),
 721                         dn->node_page->index, dn->node_page->index);
 722
 723         dn->node_page = NULL;
 724         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 725 }
 726
 727 static int truncate_dnode(struct dnode_of_data *dn)
 728 {
 729         struct page *page;
 730
 731         if (dn->nid == 0)
 732                 return 1;
 733
 734         /* get direct node */
 735         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 736         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 737                 return 1;
 738         else if (IS_ERR(page))
 739                 return PTR_ERR(page);
 740
 741         /* Make dnode_of_data for parameter */
 742         dn->node_page = page;
 743         dn->ofs_in_node = 0;
 744         truncate_data_blocks(dn);
 745         truncate_node(dn);
 746         return 1;
 747 }
 748
 749 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 750                                                 int ofs, int depth)
 751 {
 752         struct dnode_of_data rdn = *dn;
 753         struct page *page;
 754         struct f2fs_node *rn;
 755         nid_t child_nid;
 756         unsigned int child_nofs;
 757         int freed = 0;
 758         int i, ret;
 759
 760         if (dn->nid == 0)
 761                 return NIDS_PER_BLOCK + 1;
 762
 763         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 764
 765         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 766         if (IS_ERR(page)) {
 767                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 768                 return PTR_ERR(page);
 769         }
 770
 771         ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 772
 773         rn = F2FS_NODE(page);
 774         if (depth < 3) {
 775                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 776                         child_nid = le32_to_cpu(rn->in.nid[i]);
 777                         if (child_nid == 0)
 778                                 continue;
 779                         rdn.nid = child_nid;
 780                         ret = truncate_dnode(&rdn);
 781                         if (ret < 0)
 782                                 goto out_err;
 783                         if (set_nid(page, i, 0, false))
 784                                 dn->node_changed = true;
 785                 }
 786         } else {
 787                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 788                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 789                         child_nid = le32_to_cpu(rn->in.nid[i]);
 790                         if (child_nid == 0) {
 791                                 child_nofs += NIDS_PER_BLOCK + 1;
 792                                 continue;
 793                         }
 794                         rdn.nid = child_nid;
 795                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 796                         if (ret == (NIDS_PER_BLOCK + 1)) {
 797                                 if (set_nid(page, i, 0, false))
 798                                         dn->node_changed = true;
 799                                 child_nofs += ret;
 800                         } else if (ret < 0 && ret != -ENOENT) {
 801                                 goto out_err;
 802                         }
 803                 }
 804                 freed = child_nofs;
 805         }
 806
 807         if (!ofs) {
 808                 /* remove current indirect node */
 809                 dn->node_page = page;
 810                 truncate_node(dn);
 811                 freed++;
 812         } else {
 813                 f2fs_put_page(page, 1);
 814         }
 815         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 816         return freed;
 817
 818 out_err:
 819         f2fs_put_page(page, 1);
 820         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 821         return ret;
 822 }
 823
 824 static int truncate_partial_nodes(struct dnode_of_data *dn,
 825                         struct f2fs_inode *ri, int *offset, int depth)
 826 {
 827         struct page *pages[2];
 828         nid_t nid[3];
 829         nid_t child_nid;
 830         int err = 0;
 831         int i;
 832         int idx = depth - 2;
 833
 834         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 835         if (!nid[0])
 836                 return 0;
 837
 838         /* get indirect nodes in the path */
 839         for (i = 0; i < idx + 1; i++) {
 840                 /* reference count'll be increased */
 841                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 842                 if (IS_ERR(pages[i])) {
 843                         err = PTR_ERR(pages[i]);
 844                         idx = i - 1;
 845                         goto fail;
 846                 }
 847                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 848         }
 849
 850         ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 851
 852         /* free direct nodes linked to a partial indirect node */
 853         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 854                 child_nid = get_nid(pages[idx], i, false);
 855                 if (!child_nid)
 856                         continue;
 857                 dn->nid = child_nid;
 858                 err = truncate_dnode(dn);
 859                 if (err < 0)
 860                         goto fail;
 861                 if (set_nid(pages[idx], i, 0, false))
 862                         dn->node_changed = true;
 863         }
 864
 865         if (offset[idx + 1] == 0) {
 866                 dn->node_page = pages[idx];
 867                 dn->nid = nid[idx];
 868                 truncate_node(dn);
 869         } else {
 870                 f2fs_put_page(pages[idx], 1);
 871         }
 872         offset[idx]++;
 873         offset[idx + 1] = 0;
 874         idx--;
 875 fail:
 876         for (i = idx; i >= 0; i--)
 877                 f2fs_put_page(pages[i], 1);
 878
 879         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 880
 881         return err;
 882 }
 883
 884 /*
 885  * All the block addresses of data and nodes should be nullified.
 886  */
 887 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 888 {
 889         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 890         int err = 0, cont = 1;
 891         int level, offset[4], noffset[4];
 892         unsigned int nofs = 0;
 893         struct f2fs_inode *ri;
 894         struct dnode_of_data dn;
 895         struct page *page;
 896
 897         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 898
 899         level = get_node_path(inode, from, offset, noffset);
 900         if (level < 0)
 901                 return level;
 902
 903         page = get_node_page(sbi, inode->i_ino);
 904         if (IS_ERR(page)) {
 905                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 906                 return PTR_ERR(page);
 907         }
 908
 909         set_new_dnode(&dn, inode, page, NULL, 0);
 910         unlock_page(page);
 911
 912         ri = F2FS_INODE(page);
 913         switch (level) {
 914         case 0:
 915         case 1:
 916                 nofs = noffset[1];
 917                 break;
 918         case 2:
 919                 nofs = noffset[1];
 920                 if (!offset[level - 1])
 921                         goto skip_partial;
 922                 err = truncate_partial_nodes(&dn, ri, offset, level);
 923                 if (err < 0 && err != -ENOENT)
 924                         goto fail;
 925                 nofs += 1 + NIDS_PER_BLOCK;
 926                 break;
 927         case 3:
 928                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 929                 if (!offset[level - 1])
 930                         goto skip_partial;
 931                 err = truncate_partial_nodes(&dn, ri, offset, level);
 932                 if (err < 0 && err != -ENOENT)
 933                         goto fail;
 934                 break;
 935         default:
 936                 BUG();
 937         }
 938
 939 skip_partial:
 940         while (cont) {
 941                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 942                 switch (offset[0]) {
 943                 case NODE_DIR1_BLOCK:
 944                 case NODE_DIR2_BLOCK:
 945                         err = truncate_dnode(&dn);
 946                         break;
 947
 948                 case NODE_IND1_BLOCK:
 949                 case NODE_IND2_BLOCK:
 950                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 951                         break;
 952
 953                 case NODE_DIND_BLOCK:
 954                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 955                         cont = 0;
 956                         break;
 957
 958                 default:
 959                         BUG();
 960                 }
 961                 if (err < 0 && err != -ENOENT)
 962                         goto fail;
 963                 if (offset[1] == 0 &&
 964                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 965                         lock_page(page);
 966                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
 967                         f2fs_wait_on_page_writeback(page, NODE, true);
 968                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 969                         set_page_dirty(page);
 970                         unlock_page(page);
 971                 }
 972                 offset[1] = 0;
 973                 offset[0]++;
 974                 nofs += err;
 975         }
 976 fail:
 977         f2fs_put_page(page, 0);
 978         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 979         return err > 0 ? 0 : err;
 980 }
 981
 982 /* caller must lock inode page */
 983 int truncate_xattr_node(struct inode *inode)
 984 {
 985         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 986         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 987         struct dnode_of_data dn;
 988         struct page *npage;
 989
 990         if (!nid)
 991                 return 0;
 992
 993         npage = get_node_page(sbi, nid);
 994         if (IS_ERR(npage))
 995                 return PTR_ERR(npage);
 996
 997         f2fs_i_xnid_write(inode, 0);
 998
 999         set_new_dnode(&dn, inode, NULL, npage, nid);
1000         truncate_node(&dn);
1001         return 0;
1002 }
1003
1004 /*
1005  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1006  * f2fs_unlock_op().
1007  */
1008 int remove_inode_page(struct inode *inode)
1009 {
1010         struct dnode_of_data dn;
1011         int err;
1012
1013         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1014         err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1015         if (err)
1016                 return err;
1017
1018         err = truncate_xattr_node(inode);
1019         if (err) {
1020                 f2fs_put_dnode(&dn);
1021                 return err;
1022         }
1023
1024         /* remove potential inline_data blocks */
1025         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1026                                 S_ISLNK(inode->i_mode))
1027                 truncate_data_blocks_range(&dn, 1);
1028
1029         /* 0 is possible, after f2fs_new_inode() has failed */
1030         f2fs_bug_on(F2FS_I_SB(inode),
1031                         inode->i_blocks != 0 && inode->i_blocks != 8);
1032
1033         /* will put inode & node pages */
1034         truncate_node(&dn);
1035         return 0;
1036 }
1037
1038 struct page *new_inode_page(struct inode *inode)
1039 {
1040         struct dnode_of_data dn;
1041
1042         /* allocate inode page for new inode */
1043         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1044
1045         /* caller should f2fs_put_page(page, 1); */
1046         return new_node_page(&dn, 0);
1047 }
1048
1049 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1050 {
1051         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1052         struct node_info new_ni;
1053         struct page *page;
1054         int err;
1055
1056         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1057                 return ERR_PTR(-EPERM);
1058
1059         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1060         if (!page)
1061                 return ERR_PTR(-ENOMEM);
1062
1063         if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1064                 goto fail;
1065
1066 #ifdef CONFIG_F2FS_CHECK_FS
1067         get_node_info(sbi, dn->nid, &new_ni);
1068         f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1069 #endif
1070         new_ni.nid = dn->nid;
1071         new_ni.ino = dn->inode->i_ino;
1072         new_ni.blk_addr = NULL_ADDR;
1073         new_ni.flag = 0;
1074         new_ni.version = 0;
1075         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1076
1077         f2fs_wait_on_page_writeback(page, NODE, true);
1078         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1079         set_cold_node(dn->inode, page);
1080         if (!PageUptodate(page))
1081                 SetPageUptodate(page);
1082         if (set_page_dirty(page))
1083                 dn->node_changed = true;
1084
1085         if (f2fs_has_xattr_block(ofs))
1086                 f2fs_i_xnid_write(dn->inode, dn->nid);
1087
1088         if (ofs == 0)
1089                 inc_valid_inode_count(sbi);
1090         return page;
1091
1092 fail:
1093         clear_node_page_dirty(page);
1094         f2fs_put_page(page, 1);
1095         return ERR_PTR(err);
1096 }
1097
1098 /*
1099  * Caller should do after getting the following values.
1100  * 0: f2fs_put_page(page, 0)
1101  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1102  */
1103 static int read_node_page(struct page *page, int op_flags)
1104 {
1105         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1106         struct node_info ni;
1107         struct f2fs_io_info fio = {
1108                 .sbi = sbi,
1109                 .type = NODE,
1110                 .op = REQ_OP_READ,
1111                 .op_flags = op_flags,
1112                 .page = page,
1113                 .encrypted_page = NULL,
1114         };
1115
1116         if (PageUptodate(page))
1117                 return LOCKED_PAGE;
1118
1119         get_node_info(sbi, page->index, &ni);
1120
1121         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1122                 ClearPageUptodate(page);
1123                 return -ENOENT;
1124         }
1125
1126         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1127         return f2fs_submit_page_bio(&fio);
1128 }
1129
1130 /*
1131  * Readahead a node page
1132  */
1133 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1134 {
1135         struct page *apage;
1136         int err;
1137
1138         if (!nid)
1139                 return;
1140         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1141
1142         rcu_read_lock();
1143         apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1144         rcu_read_unlock();
1145         if (apage)
1146                 return;
1147
1148         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1149         if (!apage)
1150                 return;
1151
1152         err = read_node_page(apage, REQ_RAHEAD);
1153         f2fs_put_page(apage, err ? 1 : 0);
1154 }
1155
1156 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1157                                         struct page *parent, int start)
1158 {
1159         struct page *page;
1160         int err;
1161
1162         if (!nid)
1163                 return ERR_PTR(-ENOENT);
1164         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1165 repeat:
1166         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1167         if (!page)
1168                 return ERR_PTR(-ENOMEM);
1169
1170         err = read_node_page(page, 0);
1171         if (err < 0) {
1172                 f2fs_put_page(page, 1);
1173                 return ERR_PTR(err);
1174         } else if (err == LOCKED_PAGE) {
1175                 err = 0;
1176                 goto page_hit;
1177         }
1178
1179         if (parent)
1180                 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1181
1182         lock_page(page);
1183
1184         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1185                 f2fs_put_page(page, 1);
1186                 goto repeat;
1187         }
1188
1189         if (unlikely(!PageUptodate(page))) {
1190                 err = -EIO;
1191                 goto out_err;
1192         }
1193
1194         if (!f2fs_inode_chksum_verify(sbi, page)) {
1195                 err = -EBADMSG;
1196                 goto out_err;
1197         }
1198 page_hit:
1199         if(unlikely(nid != nid_of_node(page))) {
1200                 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1201                         "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1202                         nid, nid_of_node(page), ino_of_node(page),
1203                         ofs_of_node(page), cpver_of_node(page),
1204                         next_blkaddr_of_node(page));
1205                 err = -EINVAL;
1206 out_err:
1207                 ClearPageUptodate(page);
1208                 f2fs_put_page(page, 1);
1209                 return ERR_PTR(err);
1210         }
1211         return page;
1212 }
1213
1214 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1215 {
1216         return __get_node_page(sbi, nid, NULL, 0);
1217 }
1218
1219 struct page *get_node_page_ra(struct page *parent, int start)
1220 {
1221         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1222         nid_t nid = get_nid(parent, start, false);
1223
1224         return __get_node_page(sbi, nid, parent, start);
1225 }
1226
1227 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1228 {
1229         struct inode *inode;
1230         struct page *page;
1231         int ret;
1232
1233         /* should flush inline_data before evict_inode */
1234         inode = ilookup(sbi->sb, ino);
1235         if (!inode)
1236                 return;
1237
1238         page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1239                                         FGP_LOCK|FGP_NOWAIT, 0);
1240         if (!page)
1241                 goto iput_out;
1242
1243         if (!PageUptodate(page))
1244                 goto page_out;
1245
1246         if (!PageDirty(page))
1247                 goto page_out;
1248
1249         if (!clear_page_dirty_for_io(page))
1250                 goto page_out;
1251
1252         ret = f2fs_write_inline_data(inode, page);
1253         inode_dec_dirty_pages(inode);
1254         remove_dirty_inode(inode);
1255         if (ret)
1256                 set_page_dirty(page);
1257 page_out:
1258         f2fs_put_page(page, 1);
1259 iput_out:
1260         iput(inode);
1261 }
1262
1263 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1264 {
1265         pgoff_t index;
1266         struct pagevec pvec;
1267         struct page *last_page = NULL;
1268         int nr_pages;
1269
1270         pagevec_init(&pvec);
1271         index = 0;
1272
1273         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1274                                 PAGECACHE_TAG_DIRTY))) {
1275                 int i;
1276
1277                 for (i = 0; i < nr_pages; i++) {
1278                         struct page *page = pvec.pages[i];
1279
1280                         if (unlikely(f2fs_cp_error(sbi))) {
1281                                 f2fs_put_page(last_page, 0);
1282                                 pagevec_release(&pvec);
1283                                 return ERR_PTR(-EIO);
1284                         }
1285
1286                         if (!IS_DNODE(page) || !is_cold_node(page))
1287                                 continue;
1288                         if (ino_of_node(page) != ino)
1289                                 continue;
1290
1291                         lock_page(page);
1292
1293                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1294 continue_unlock:
1295                                 unlock_page(page);
1296                                 continue;
1297                         }
1298                         if (ino_of_node(page) != ino)
1299                                 goto continue_unlock;
1300
1301                         if (!PageDirty(page)) {
1302                                 /* someone wrote it for us */
1303                                 goto continue_unlock;
1304                         }
1305
1306                         if (last_page)
1307                                 f2fs_put_page(last_page, 0);
1308
1309                         get_page(page);
1310                         last_page = page;
1311                         unlock_page(page);
1312                 }
1313                 pagevec_release(&pvec);
1314                 cond_resched();
1315         }
1316         return last_page;
1317 }
1318
1319 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1320                                 struct writeback_control *wbc, bool do_balance,
1321                                 enum iostat_type io_type)
1322 {
1323         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1324         nid_t nid;
1325         struct node_info ni;
1326         struct f2fs_io_info fio = {
1327                 .sbi = sbi,
1328                 .ino = ino_of_node(page),
1329                 .type = NODE,
1330                 .op = REQ_OP_WRITE,
1331                 .op_flags = wbc_to_write_flags(wbc),
1332                 .page = page,
1333                 .encrypted_page = NULL,
1334                 .submitted = false,
1335                 .io_type = io_type,
1336                 .io_wbc = wbc,
1337         };
1338
1339         trace_f2fs_writepage(page, NODE);
1340
1341         if (unlikely(f2fs_cp_error(sbi))) {
1342                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1343                 unlock_page(page);
1344                 return 0;
1345         }
1346
1347         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1348                 goto redirty_out;
1349
1350         /* get old block addr of this node page */
1351         nid = nid_of_node(page);
1352         f2fs_bug_on(sbi, page->index != nid);
1353
1354         if (wbc->for_reclaim) {
1355                 if (!down_read_trylock(&sbi->node_write))
1356                         goto redirty_out;
1357         } else {
1358                 down_read(&sbi->node_write);
1359         }
1360
1361         get_node_info(sbi, nid, &ni);
1362
1363         /* This page is already truncated */
1364         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1365                 ClearPageUptodate(page);
1366                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1367                 up_read(&sbi->node_write);
1368                 unlock_page(page);
1369                 return 0;
1370         }
1371
1372         if (atomic && !test_opt(sbi, NOBARRIER))
1373                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1374
1375         set_page_writeback(page);
1376         fio.old_blkaddr = ni.blk_addr;
1377         write_node_page(nid, &fio);
1378         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1379         dec_page_count(sbi, F2FS_DIRTY_NODES);
1380         up_read(&sbi->node_write);
1381
1382         if (wbc->for_reclaim) {
1383                 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1384                                                 page->index, NODE);
1385                 submitted = NULL;
1386         }
1387
1388         unlock_page(page);
1389
1390         if (unlikely(f2fs_cp_error(sbi))) {
1391                 f2fs_submit_merged_write(sbi, NODE);
1392                 submitted = NULL;
1393         }
1394         if (submitted)
1395                 *submitted = fio.submitted;
1396
1397         if (do_balance)
1398                 f2fs_balance_fs(sbi, false);
1399         return 0;
1400
1401 redirty_out:
1402         redirty_page_for_writepage(wbc, page);
1403         return AOP_WRITEPAGE_ACTIVATE;
1404 }
1405
1406 void move_node_page(struct page *node_page, int gc_type)
1407 {
1408         if (gc_type == FG_GC) {
1409                 struct writeback_control wbc = {
1410                         .sync_mode = WB_SYNC_ALL,
1411                         .nr_to_write = 1,
1412                         .for_reclaim = 0,
1413                 };
1414
1415                 set_page_dirty(node_page);
1416                 f2fs_wait_on_page_writeback(node_page, NODE, true);
1417
1418                 f2fs_bug_on(F2FS_P_SB(node_page), PageWriteback(node_page));
1419                 if (!clear_page_dirty_for_io(node_page))
1420                         goto out_page;
1421
1422                 if (__write_node_page(node_page, false, NULL,
1423                                         &wbc, false, FS_GC_NODE_IO))
1424                         unlock_page(node_page);
1425                 goto release_page;
1426         } else {
1427                 /* set page dirty and write it */
1428                 if (!PageWriteback(node_page))
1429                         set_page_dirty(node_page);
1430         }
1431 out_page:
1432         unlock_page(node_page);
1433 release_page:
1434         f2fs_put_page(node_page, 0);
1435 }
1436
1437 static int f2fs_write_node_page(struct page *page,
1438                                 struct writeback_control *wbc)
1439 {
1440         return __write_node_page(page, false, NULL, wbc, false, FS_NODE_IO);
1441 }
1442
1443 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1444                         struct writeback_control *wbc, bool atomic)
1445 {
1446         pgoff_t index;
1447         pgoff_t last_idx = ULONG_MAX;
1448         struct pagevec pvec;
1449         int ret = 0;
1450         struct page *last_page = NULL;
1451         bool marked = false;
1452         nid_t ino = inode->i_ino;
1453         int nr_pages;
1454
1455         if (atomic) {
1456                 last_page = last_fsync_dnode(sbi, ino);
1457                 if (IS_ERR_OR_NULL(last_page))
1458                         return PTR_ERR_OR_ZERO(last_page);
1459         }
1460 retry:
1461         pagevec_init(&pvec);
1462         index = 0;
1463
1464         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1465                                 PAGECACHE_TAG_DIRTY))) {
1466                 int i;
1467
1468                 for (i = 0; i < nr_pages; i++) {
1469                         struct page *page = pvec.pages[i];
1470                         bool submitted = false;
1471
1472                         if (unlikely(f2fs_cp_error(sbi))) {
1473                                 f2fs_put_page(last_page, 0);
1474                                 pagevec_release(&pvec);
1475                                 ret = -EIO;
1476                                 goto out;
1477                         }
1478
1479                         if (!IS_DNODE(page) || !is_cold_node(page))
1480                                 continue;
1481                         if (ino_of_node(page) != ino)
1482                                 continue;
1483
1484                         lock_page(page);
1485
1486                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1487 continue_unlock:
1488                                 unlock_page(page);
1489                                 continue;
1490                         }
1491                         if (ino_of_node(page) != ino)
1492                                 goto continue_unlock;
1493
1494                         if (!PageDirty(page) && page != last_page) {
1495                                 /* someone wrote it for us */
1496                                 goto continue_unlock;
1497                         }
1498
1499                         f2fs_wait_on_page_writeback(page, NODE, true);
1500                         BUG_ON(PageWriteback(page));
1501
1502                         set_fsync_mark(page, 0);
1503                         set_dentry_mark(page, 0);
1504
1505                         if (!atomic || page == last_page) {
1506                                 set_fsync_mark(page, 1);
1507                                 if (IS_INODE(page)) {
1508                                         if (is_inode_flag_set(inode,
1509                                                                 FI_DIRTY_INODE))
1510                                                 update_inode(inode, page);
1511                                         set_dentry_mark(page,
1512                                                 need_dentry_mark(sbi, ino));
1513                                 }
1514                                 /*  may be written by other thread */
1515                                 if (!PageDirty(page))
1516                                         set_page_dirty(page);
1517                         }
1518
1519                         if (!clear_page_dirty_for_io(page))
1520                                 goto continue_unlock;
1521
1522                         ret = __write_node_page(page, atomic &&
1523                                                 page == last_page,
1524                                                 &submitted, wbc, true,
1525                                                 FS_NODE_IO);
1526                         if (ret) {
1527                                 unlock_page(page);
1528                                 f2fs_put_page(last_page, 0);
1529                                 break;
1530                         } else if (submitted) {
1531                                 last_idx = page->index;
1532                         }
1533
1534                         if (page == last_page) {
1535                                 f2fs_put_page(page, 0);
1536                                 marked = true;
1537                                 break;
1538                         }
1539                 }
1540                 pagevec_release(&pvec);
1541                 cond_resched();
1542
1543                 if (ret || marked)
1544                         break;
1545         }
1546         if (!ret && atomic && !marked) {
1547                 f2fs_msg(sbi->sb, KERN_DEBUG,
1548                         "Retry to write fsync mark: ino=%u, idx=%lx",
1549                                         ino, last_page->index);
1550                 lock_page(last_page);
1551                 f2fs_wait_on_page_writeback(last_page, NODE, true);
1552                 set_page_dirty(last_page);
1553                 unlock_page(last_page);
1554                 goto retry;
1555         }
1556 out:
1557         if (last_idx != ULONG_MAX)
1558                 f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1559         return ret ? -EIO: 0;
1560 }
1561
1562 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc,
1563                                 bool do_balance, enum iostat_type io_type)
1564 {
1565         pgoff_t index;
1566         struct pagevec pvec;
1567         int step = 0;
1568         int nwritten = 0;
1569         int ret = 0;
1570         int nr_pages;
1571
1572         pagevec_init(&pvec);
1573
1574 next_step:
1575         index = 0;
1576
1577         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1578                                 PAGECACHE_TAG_DIRTY))) {
1579                 int i;
1580
1581                 for (i = 0; i < nr_pages; i++) {
1582                         struct page *page = pvec.pages[i];
1583                         bool submitted = false;
1584
1585                         /*
1586                          * flushing sequence with step:
1587                          * 0. indirect nodes
1588                          * 1. dentry dnodes
1589                          * 2. file dnodes
1590                          */
1591                         if (step == 0 && IS_DNODE(page))
1592                                 continue;
1593                         if (step == 1 && (!IS_DNODE(page) ||
1594                                                 is_cold_node(page)))
1595                                 continue;
1596                         if (step == 2 && (!IS_DNODE(page) ||
1597                                                 !is_cold_node(page)))
1598                                 continue;
1599 lock_node:
1600                         if (!trylock_page(page))
1601                                 continue;
1602
1603                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1604 continue_unlock:
1605                                 unlock_page(page);
1606                                 continue;
1607                         }
1608
1609                         if (!PageDirty(page)) {
1610                                 /* someone wrote it for us */
1611                                 goto continue_unlock;
1612                         }
1613
1614                         /* flush inline_data */
1615                         if (is_inline_node(page)) {
1616                                 clear_inline_node(page);
1617                                 unlock_page(page);
1618                                 flush_inline_data(sbi, ino_of_node(page));
1619                                 goto lock_node;
1620                         }
1621
1622                         f2fs_wait_on_page_writeback(page, NODE, true);
1623
1624                         BUG_ON(PageWriteback(page));
1625                         if (!clear_page_dirty_for_io(page))
1626                                 goto continue_unlock;
1627
1628                         set_fsync_mark(page, 0);
1629                         set_dentry_mark(page, 0);
1630
1631                         ret = __write_node_page(page, false, &submitted,
1632                                                 wbc, do_balance, io_type);
1633                         if (ret)
1634                                 unlock_page(page);
1635                         else if (submitted)
1636                                 nwritten++;
1637
1638                         if (--wbc->nr_to_write == 0)
1639                                 break;
1640                 }
1641                 pagevec_release(&pvec);
1642                 cond_resched();
1643
1644                 if (wbc->nr_to_write == 0) {
1645                         step = 2;
1646                         break;
1647                 }
1648         }
1649
1650         if (step < 2) {
1651                 step++;
1652                 goto next_step;
1653         }
1654
1655         if (nwritten)
1656                 f2fs_submit_merged_write(sbi, NODE);
1657
1658         if (unlikely(f2fs_cp_error(sbi)))
1659                 return -EIO;
1660         return ret;
1661 }
1662
1663 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1664 {
1665         pgoff_t index = 0;
1666         struct pagevec pvec;
1667         int ret2, ret = 0;
1668         int nr_pages;
1669
1670         pagevec_init(&pvec);
1671
1672         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1673                                 PAGECACHE_TAG_WRITEBACK))) {
1674                 int i;
1675
1676                 for (i = 0; i < nr_pages; i++) {
1677                         struct page *page = pvec.pages[i];
1678
1679                         if (ino && ino_of_node(page) == ino) {
1680                                 f2fs_wait_on_page_writeback(page, NODE, true);
1681                                 if (TestClearPageError(page))
1682                                         ret = -EIO;
1683                         }
1684                 }
1685                 pagevec_release(&pvec);
1686                 cond_resched();
1687         }
1688
1689         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1690         if (!ret)
1691                 ret = ret2;
1692         return ret;
1693 }
1694
1695 static int f2fs_write_node_pages(struct address_space *mapping,
1696                             struct writeback_control *wbc)
1697 {
1698         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1699         struct blk_plug plug;
1700         long diff;
1701
1702         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1703                 goto skip_write;
1704
1705         /* balancing f2fs's metadata in background */
1706         f2fs_balance_fs_bg(sbi);
1707
1708         /* collect a number of dirty node pages and write together */
1709         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1710                 goto skip_write;
1711
1712         trace_f2fs_writepages(mapping->host, wbc, NODE);
1713
1714         diff = nr_pages_to_write(sbi, NODE, wbc);
1715         wbc->sync_mode = WB_SYNC_NONE;
1716         blk_start_plug(&plug);
1717         sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1718         blk_finish_plug(&plug);
1719         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1720         return 0;
1721
1722 skip_write:
1723         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1724         trace_f2fs_writepages(mapping->host, wbc, NODE);
1725         return 0;
1726 }
1727
1728 static int f2fs_set_node_page_dirty(struct page *page)
1729 {
1730         trace_f2fs_set_page_dirty(page, NODE);
1731
1732         if (!PageUptodate(page))
1733                 SetPageUptodate(page);
1734         if (!PageDirty(page)) {
1735                 f2fs_set_page_dirty_nobuffers(page);
1736                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1737                 SetPagePrivate(page);
1738                 f2fs_trace_pid(page);
1739                 return 1;
1740         }
1741         return 0;
1742 }
1743
1744 /*
1745  * Structure of the f2fs node operations
1746  */
1747 const struct address_space_operations f2fs_node_aops = {
1748         .writepage      = f2fs_write_node_page,
1749         .writepages     = f2fs_write_node_pages,
1750         .set_page_dirty = f2fs_set_node_page_dirty,
1751         .invalidatepage = f2fs_invalidate_page,
1752         .releasepage    = f2fs_release_page,
1753 #ifdef CONFIG_MIGRATION
1754         .migratepage    = f2fs_migrate_page,
1755 #endif
1756 };
1757
1758 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1759                                                 nid_t n)
1760 {
1761         return radix_tree_lookup(&nm_i->free_nid_root, n);
1762 }
1763
1764 static int __insert_free_nid(struct f2fs_sb_info *sbi,
1765                         struct free_nid *i, enum nid_state state)
1766 {
1767         struct f2fs_nm_info *nm_i = NM_I(sbi);
1768
1769         int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1770         if (err)
1771                 return err;
1772
1773         f2fs_bug_on(sbi, state != i->state);
1774         nm_i->nid_cnt[state]++;
1775         if (state == FREE_NID)
1776                 list_add_tail(&i->list, &nm_i->free_nid_list);
1777         return 0;
1778 }
1779
1780 static void __remove_free_nid(struct f2fs_sb_info *sbi,
1781                         struct free_nid *i, enum nid_state state)
1782 {
1783         struct f2fs_nm_info *nm_i = NM_I(sbi);
1784
1785         f2fs_bug_on(sbi, state != i->state);
1786         nm_i->nid_cnt[state]--;
1787         if (state == FREE_NID)
1788                 list_del(&i->list);
1789         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1790 }
1791
1792 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
1793                         enum nid_state org_state, enum nid_state dst_state)
1794 {
1795         struct f2fs_nm_info *nm_i = NM_I(sbi);
1796
1797         f2fs_bug_on(sbi, org_state != i->state);
1798         i->state = dst_state;
1799         nm_i->nid_cnt[org_state]--;
1800         nm_i->nid_cnt[dst_state]++;
1801
1802         switch (dst_state) {
1803         case PREALLOC_NID:
1804                 list_del(&i->list);
1805                 break;
1806         case FREE_NID:
1807                 list_add_tail(&i->list, &nm_i->free_nid_list);
1808                 break;
1809         default:
1810                 BUG_ON(1);
1811         }
1812 }
1813
1814 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
1815                                                         bool set, bool build)
1816 {
1817         struct f2fs_nm_info *nm_i = NM_I(sbi);
1818         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
1819         unsigned int nid_ofs = nid - START_NID(nid);
1820
1821         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1822                 return;
1823
1824         if (set) {
1825                 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
1826                         return;
1827                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1828                 nm_i->free_nid_count[nat_ofs]++;
1829         } else {
1830                 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
1831                         return;
1832                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1833                 if (!build)
1834                         nm_i->free_nid_count[nat_ofs]--;
1835         }
1836 }
1837
1838 /* return if the nid is recognized as free */
1839 static bool add_free_nid(struct f2fs_sb_info *sbi,
1840                                 nid_t nid, bool build, bool update)
1841 {
1842         struct f2fs_nm_info *nm_i = NM_I(sbi);
1843         struct free_nid *i, *e;
1844         struct nat_entry *ne;
1845         int err = -EINVAL;
1846         bool ret = false;
1847
1848         /* 0 nid should not be used */
1849         if (unlikely(nid == 0))
1850                 return false;
1851
1852         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1853         i->nid = nid;
1854         i->state = FREE_NID;
1855
1856         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
1857
1858         spin_lock(&nm_i->nid_list_lock);
1859
1860         if (build) {
1861                 /*
1862                  *   Thread A             Thread B
1863                  *  - f2fs_create
1864                  *   - f2fs_new_inode
1865                  *    - alloc_nid
1866                  *     - __insert_nid_to_list(PREALLOC_NID)
1867                  *                     - f2fs_balance_fs_bg
1868                  *                      - build_free_nids
1869                  *                       - __build_free_nids
1870                  *                        - scan_nat_page
1871                  *                         - add_free_nid
1872                  *                          - __lookup_nat_cache
1873                  *  - f2fs_add_link
1874                  *   - init_inode_metadata
1875                  *    - new_inode_page
1876                  *     - new_node_page
1877                  *      - set_node_addr
1878                  *  - alloc_nid_done
1879                  *   - __remove_nid_from_list(PREALLOC_NID)
1880                  *                         - __insert_nid_to_list(FREE_NID)
1881                  */
1882                 ne = __lookup_nat_cache(nm_i, nid);
1883                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1884                                 nat_get_blkaddr(ne) != NULL_ADDR))
1885                         goto err_out;
1886
1887                 e = __lookup_free_nid_list(nm_i, nid);
1888                 if (e) {
1889                         if (e->state == FREE_NID)
1890                                 ret = true;
1891                         goto err_out;
1892                 }
1893         }
1894         ret = true;
1895         err = __insert_free_nid(sbi, i, FREE_NID);
1896 err_out:
1897         if (update) {
1898                 update_free_nid_bitmap(sbi, nid, ret, build);
1899                 if (!build)
1900                         nm_i->available_nids++;
1901         }
1902         spin_unlock(&nm_i->nid_list_lock);
1903         radix_tree_preload_end();
1904
1905         if (err)
1906                 kmem_cache_free(free_nid_slab, i);
1907         return ret;
1908 }
1909
1910 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1911 {
1912         struct f2fs_nm_info *nm_i = NM_I(sbi);
1913         struct free_nid *i;
1914         bool need_free = false;
1915
1916         spin_lock(&nm_i->nid_list_lock);
1917         i = __lookup_free_nid_list(nm_i, nid);
1918         if (i && i->state == FREE_NID) {
1919                 __remove_free_nid(sbi, i, FREE_NID);
1920                 need_free = true;
1921         }
1922         spin_unlock(&nm_i->nid_list_lock);
1923
1924         if (need_free)
1925                 kmem_cache_free(free_nid_slab, i);
1926 }
1927
1928 static void scan_nat_page(struct f2fs_sb_info *sbi,
1929                         struct page *nat_page, nid_t start_nid)
1930 {
1931         struct f2fs_nm_info *nm_i = NM_I(sbi);
1932         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1933         block_t blk_addr;
1934         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
1935         int i;
1936
1937         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
1938
1939         i = start_nid % NAT_ENTRY_PER_BLOCK;
1940
1941         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1942                 if (unlikely(start_nid >= nm_i->max_nid))
1943                         break;
1944
1945                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1946                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1947                 if (blk_addr == NULL_ADDR) {
1948                         add_free_nid(sbi, start_nid, true, true);
1949                 } else {
1950                         spin_lock(&NM_I(sbi)->nid_list_lock);
1951                         update_free_nid_bitmap(sbi, start_nid, false, true);
1952                         spin_unlock(&NM_I(sbi)->nid_list_lock);
1953                 }
1954         }
1955 }
1956
1957 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
1958 {
1959         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1960         struct f2fs_journal *journal = curseg->journal;
1961         int i;
1962
1963         down_read(&curseg->journal_rwsem);
1964         for (i = 0; i < nats_in_cursum(journal); i++) {
1965                 block_t addr;
1966                 nid_t nid;
1967
1968                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1969                 nid = le32_to_cpu(nid_in_journal(journal, i));
1970                 if (addr == NULL_ADDR)
1971                         add_free_nid(sbi, nid, true, false);
1972                 else
1973                         remove_free_nid(sbi, nid);
1974         }
1975         up_read(&curseg->journal_rwsem);
1976 }
1977
1978 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
1979 {
1980         struct f2fs_nm_info *nm_i = NM_I(sbi);
1981         unsigned int i, idx;
1982         nid_t nid;
1983
1984         down_read(&nm_i->nat_tree_lock);
1985
1986         for (i = 0; i < nm_i->nat_blocks; i++) {
1987                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
1988                         continue;
1989                 if (!nm_i->free_nid_count[i])
1990                         continue;
1991                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
1992                         idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
1993                                                 NAT_ENTRY_PER_BLOCK, idx);
1994                         if (idx >= NAT_ENTRY_PER_BLOCK)
1995                                 break;
1996
1997                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
1998                         add_free_nid(sbi, nid, true, false);
1999
2000                         if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2001                                 goto out;
2002                 }
2003         }
2004 out:
2005         scan_curseg_cache(sbi);
2006
2007         up_read(&nm_i->nat_tree_lock);
2008 }
2009
2010 static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2011 {
2012         struct f2fs_nm_info *nm_i = NM_I(sbi);
2013         int i = 0;
2014         nid_t nid = nm_i->next_scan_nid;
2015
2016         if (unlikely(nid >= nm_i->max_nid))
2017                 nid = 0;
2018
2019         /* Enough entries */
2020         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2021                 return;
2022
2023         if (!sync && !available_free_memory(sbi, FREE_NIDS))
2024                 return;
2025
2026         if (!mount) {
2027                 /* try to find free nids in free_nid_bitmap */
2028                 scan_free_nid_bits(sbi);
2029
2030                 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2031                         return;
2032         }
2033
2034         /* readahead nat pages to be scanned */
2035         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2036                                                         META_NAT, true);
2037
2038         down_read(&nm_i->nat_tree_lock);
2039
2040         while (1) {
2041                 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2042                                                 nm_i->nat_block_bitmap)) {
2043                         struct page *page = get_current_nat_page(sbi, nid);
2044
2045                         scan_nat_page(sbi, page, nid);
2046                         f2fs_put_page(page, 1);
2047                 }
2048
2049                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2050                 if (unlikely(nid >= nm_i->max_nid))
2051                         nid = 0;
2052
2053                 if (++i >= FREE_NID_PAGES)
2054                         break;
2055         }
2056
2057         /* go to the next free nat pages to find free nids abundantly */
2058         nm_i->next_scan_nid = nid;
2059
2060         /* find free nids from current sum_pages */
2061         scan_curseg_cache(sbi);
2062
2063         up_read(&nm_i->nat_tree_lock);
2064
2065         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2066                                         nm_i->ra_nid_pages, META_NAT, false);
2067 }
2068
2069 void build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2070 {
2071         mutex_lock(&NM_I(sbi)->build_lock);
2072         __build_free_nids(sbi, sync, mount);
2073         mutex_unlock(&NM_I(sbi)->build_lock);
2074 }
2075
2076 /*
2077  * If this function returns success, caller can obtain a new nid
2078  * from second parameter of this function.
2079  * The returned nid could be used ino as well as nid when inode is created.
2080  */
2081 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2082 {
2083         struct f2fs_nm_info *nm_i = NM_I(sbi);
2084         struct free_nid *i = NULL;
2085 retry:
2086 #ifdef CONFIG_F2FS_FAULT_INJECTION
2087         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2088                 f2fs_show_injection_info(FAULT_ALLOC_NID);
2089                 return false;
2090         }
2091 #endif
2092         spin_lock(&nm_i->nid_list_lock);
2093
2094         if (unlikely(nm_i->available_nids == 0)) {
2095                 spin_unlock(&nm_i->nid_list_lock);
2096                 return false;
2097         }
2098
2099         /* We should not use stale free nids created by build_free_nids */
2100         if (nm_i->nid_cnt[FREE_NID] && !on_build_free_nids(nm_i)) {
2101                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2102                 i = list_first_entry(&nm_i->free_nid_list,
2103                                         struct free_nid, list);
2104                 *nid = i->nid;
2105
2106                 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2107                 nm_i->available_nids--;
2108
2109                 update_free_nid_bitmap(sbi, *nid, false, false);
2110
2111                 spin_unlock(&nm_i->nid_list_lock);
2112                 return true;
2113         }
2114         spin_unlock(&nm_i->nid_list_lock);
2115
2116         /* Let's scan nat pages and its caches to get free nids */
2117         build_free_nids(sbi, true, false);
2118         goto retry;
2119 }
2120
2121 /*
2122  * alloc_nid() should be called prior to this function.
2123  */
2124 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2125 {
2126         struct f2fs_nm_info *nm_i = NM_I(sbi);
2127         struct free_nid *i;
2128
2129         spin_lock(&nm_i->nid_list_lock);
2130         i = __lookup_free_nid_list(nm_i, nid);
2131         f2fs_bug_on(sbi, !i);
2132         __remove_free_nid(sbi, i, PREALLOC_NID);
2133         spin_unlock(&nm_i->nid_list_lock);
2134
2135         kmem_cache_free(free_nid_slab, i);
2136 }
2137
2138 /*
2139  * alloc_nid() should be called prior to this function.
2140  */
2141 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2142 {
2143         struct f2fs_nm_info *nm_i = NM_I(sbi);
2144         struct free_nid *i;
2145         bool need_free = false;
2146
2147         if (!nid)
2148                 return;
2149
2150         spin_lock(&nm_i->nid_list_lock);
2151         i = __lookup_free_nid_list(nm_i, nid);
2152         f2fs_bug_on(sbi, !i);
2153
2154         if (!available_free_memory(sbi, FREE_NIDS)) {
2155                 __remove_free_nid(sbi, i, PREALLOC_NID);
2156                 need_free = true;
2157         } else {
2158                 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2159         }
2160
2161         nm_i->available_nids++;
2162
2163         update_free_nid_bitmap(sbi, nid, true, false);
2164
2165         spin_unlock(&nm_i->nid_list_lock);
2166
2167         if (need_free)
2168                 kmem_cache_free(free_nid_slab, i);
2169 }
2170
2171 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2172 {
2173         struct f2fs_nm_info *nm_i = NM_I(sbi);
2174         struct free_nid *i, *next;
2175         int nr = nr_shrink;
2176
2177         if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2178                 return 0;
2179
2180         if (!mutex_trylock(&nm_i->build_lock))
2181                 return 0;
2182
2183         spin_lock(&nm_i->nid_list_lock);
2184         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2185                 if (nr_shrink <= 0 ||
2186                                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2187                         break;
2188
2189                 __remove_free_nid(sbi, i, FREE_NID);
2190                 kmem_cache_free(free_nid_slab, i);
2191                 nr_shrink--;
2192         }
2193         spin_unlock(&nm_i->nid_list_lock);
2194         mutex_unlock(&nm_i->build_lock);
2195
2196         return nr - nr_shrink;
2197 }
2198
2199 void recover_inline_xattr(struct inode *inode, struct page *page)
2200 {
2201         void *src_addr, *dst_addr;
2202         size_t inline_size;
2203         struct page *ipage;
2204         struct f2fs_inode *ri;
2205
2206         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2207         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2208
2209         ri = F2FS_INODE(page);
2210         if (ri->i_inline & F2FS_INLINE_XATTR) {
2211                 set_inode_flag(inode, FI_INLINE_XATTR);
2212         } else {
2213                 clear_inode_flag(inode, FI_INLINE_XATTR);
2214                 goto update_inode;
2215         }
2216
2217         dst_addr = inline_xattr_addr(inode, ipage);
2218         src_addr = inline_xattr_addr(inode, page);
2219         inline_size = inline_xattr_size(inode);
2220
2221         f2fs_wait_on_page_writeback(ipage, NODE, true);
2222         memcpy(dst_addr, src_addr, inline_size);
2223 update_inode:
2224         update_inode(inode, ipage);
2225         f2fs_put_page(ipage, 1);
2226 }
2227
2228 int recover_xattr_data(struct inode *inode, struct page *page)
2229 {
2230         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2231         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2232         nid_t new_xnid;
2233         struct dnode_of_data dn;
2234         struct node_info ni;
2235         struct page *xpage;
2236
2237         if (!prev_xnid)
2238                 goto recover_xnid;
2239
2240         /* 1: invalidate the previous xattr nid */
2241         get_node_info(sbi, prev_xnid, &ni);
2242         invalidate_blocks(sbi, ni.blk_addr);
2243         dec_valid_node_count(sbi, inode, false);
2244         set_node_addr(sbi, &ni, NULL_ADDR, false);
2245
2246 recover_xnid:
2247         /* 2: update xattr nid in inode */
2248         if (!alloc_nid(sbi, &new_xnid))
2249                 return -ENOSPC;
2250
2251         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2252         xpage = new_node_page(&dn, XATTR_NODE_OFFSET);
2253         if (IS_ERR(xpage)) {
2254                 alloc_nid_failed(sbi, new_xnid);
2255                 return PTR_ERR(xpage);
2256         }
2257
2258         alloc_nid_done(sbi, new_xnid);
2259         update_inode_page(inode);
2260
2261         /* 3: update and set xattr node page dirty */
2262         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2263
2264         set_page_dirty(xpage);
2265         f2fs_put_page(xpage, 1);
2266
2267         return 0;
2268 }
2269
2270 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2271 {
2272         struct f2fs_inode *src, *dst;
2273         nid_t ino = ino_of_node(page);
2274         struct node_info old_ni, new_ni;
2275         struct page *ipage;
2276
2277         get_node_info(sbi, ino, &old_ni);
2278
2279         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2280                 return -EINVAL;
2281 retry:
2282         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2283         if (!ipage) {
2284                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2285                 goto retry;
2286         }
2287
2288         /* Should not use this inode from free nid list */
2289         remove_free_nid(sbi, ino);
2290
2291         if (!PageUptodate(ipage))
2292                 SetPageUptodate(ipage);
2293         fill_node_footer(ipage, ino, ino, 0, true);
2294
2295         src = F2FS_INODE(page);
2296         dst = F2FS_INODE(ipage);
2297
2298         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2299         dst->i_size = 0;
2300         dst->i_blocks = cpu_to_le64(1);
2301         dst->i_links = cpu_to_le32(1);
2302         dst->i_xattr_nid = 0;
2303         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2304         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2305                 dst->i_extra_isize = src->i_extra_isize;
2306
2307                 if (f2fs_sb_has_flexible_inline_xattr(sbi->sb) &&
2308                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2309                                                         i_inline_xattr_size))
2310                         dst->i_inline_xattr_size = src->i_inline_xattr_size;
2311
2312                 if (f2fs_sb_has_project_quota(sbi->sb) &&
2313                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2314                                                                 i_projid))
2315                         dst->i_projid = src->i_projid;
2316         }
2317
2318         new_ni = old_ni;
2319         new_ni.ino = ino;
2320
2321         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2322                 WARN_ON(1);
2323         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2324         inc_valid_inode_count(sbi);
2325         set_page_dirty(ipage);
2326         f2fs_put_page(ipage, 1);
2327         return 0;
2328 }
2329
2330 void restore_node_summary(struct f2fs_sb_info *sbi,
2331                         unsigned int segno, struct f2fs_summary_block *sum)
2332 {
2333         struct f2fs_node *rn;
2334         struct f2fs_summary *sum_entry;
2335         block_t addr;
2336         int i, idx, last_offset, nrpages;
2337
2338         /* scan the node segment */
2339         last_offset = sbi->blocks_per_seg;
2340         addr = START_BLOCK(sbi, segno);
2341         sum_entry = &sum->entries[0];
2342
2343         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2344                 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2345
2346                 /* readahead node pages */
2347                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2348
2349                 for (idx = addr; idx < addr + nrpages; idx++) {
2350                         struct page *page = get_tmp_page(sbi, idx);
2351
2352                         rn = F2FS_NODE(page);
2353                         sum_entry->nid = rn->footer.nid;
2354                         sum_entry->version = 0;
2355                         sum_entry->ofs_in_node = 0;
2356                         sum_entry++;
2357                         f2fs_put_page(page, 1);
2358                 }
2359
2360                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2361                                                         addr + nrpages);
2362         }
2363 }
2364
2365 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2366 {
2367         struct f2fs_nm_info *nm_i = NM_I(sbi);
2368         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2369         struct f2fs_journal *journal = curseg->journal;
2370         int i;
2371
2372         down_write(&curseg->journal_rwsem);
2373         for (i = 0; i < nats_in_cursum(journal); i++) {
2374                 struct nat_entry *ne;
2375                 struct f2fs_nat_entry raw_ne;
2376                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2377
2378                 raw_ne = nat_in_journal(journal, i);
2379
2380                 ne = __lookup_nat_cache(nm_i, nid);
2381                 if (!ne) {
2382                         ne = __alloc_nat_entry(nid, true);
2383                         __init_nat_entry(nm_i, ne, &raw_ne, true);
2384                 }
2385
2386                 /*
2387                  * if a free nat in journal has not been used after last
2388                  * checkpoint, we should remove it from available nids,
2389                  * since later we will add it again.
2390                  */
2391                 if (!get_nat_flag(ne, IS_DIRTY) &&
2392                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2393                         spin_lock(&nm_i->nid_list_lock);
2394                         nm_i->available_nids--;
2395                         spin_unlock(&nm_i->nid_list_lock);
2396                 }
2397
2398                 __set_nat_cache_dirty(nm_i, ne);
2399         }
2400         update_nats_in_cursum(journal, -i);
2401         up_write(&curseg->journal_rwsem);
2402 }
2403
2404 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2405                                                 struct list_head *head, int max)
2406 {
2407         struct nat_entry_set *cur;
2408
2409         if (nes->entry_cnt >= max)
2410                 goto add_out;
2411
2412         list_for_each_entry(cur, head, set_list) {
2413                 if (cur->entry_cnt >= nes->entry_cnt) {
2414                         list_add(&nes->set_list, cur->set_list.prev);
2415                         return;
2416                 }
2417         }
2418 add_out:
2419         list_add_tail(&nes->set_list, head);
2420 }
2421
2422 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2423                                                 struct page *page)
2424 {
2425         struct f2fs_nm_info *nm_i = NM_I(sbi);
2426         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2427         struct f2fs_nat_block *nat_blk = page_address(page);
2428         int valid = 0;
2429         int i = 0;
2430
2431         if (!enabled_nat_bits(sbi, NULL))
2432                 return;
2433
2434         if (nat_index == 0) {
2435                 valid = 1;
2436                 i = 1;
2437         }
2438         for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2439                 if (nat_blk->entries[i].block_addr != NULL_ADDR)
2440                         valid++;
2441         }
2442         if (valid == 0) {
2443                 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2444                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2445                 return;
2446         }
2447
2448         __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2449         if (valid == NAT_ENTRY_PER_BLOCK)
2450                 __set_bit_le(nat_index, nm_i->full_nat_bits);
2451         else
2452                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2453 }
2454
2455 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2456                 struct nat_entry_set *set, struct cp_control *cpc)
2457 {
2458         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2459         struct f2fs_journal *journal = curseg->journal;
2460         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2461         bool to_journal = true;
2462         struct f2fs_nat_block *nat_blk;
2463         struct nat_entry *ne, *cur;
2464         struct page *page = NULL;
2465
2466         /*
2467          * there are two steps to flush nat entries:
2468          * #1, flush nat entries to journal in current hot data summary block.
2469          * #2, flush nat entries to nat page.
2470          */
2471         if (enabled_nat_bits(sbi, cpc) ||
2472                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2473                 to_journal = false;
2474
2475         if (to_journal) {
2476                 down_write(&curseg->journal_rwsem);
2477         } else {
2478                 page = get_next_nat_page(sbi, start_nid);
2479                 nat_blk = page_address(page);
2480                 f2fs_bug_on(sbi, !nat_blk);
2481         }
2482
2483         /* flush dirty nats in nat entry set */
2484         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2485                 struct f2fs_nat_entry *raw_ne;
2486                 nid_t nid = nat_get_nid(ne);
2487                 int offset;
2488
2489                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2490
2491                 if (to_journal) {
2492                         offset = lookup_journal_in_cursum(journal,
2493                                                         NAT_JOURNAL, nid, 1);
2494                         f2fs_bug_on(sbi, offset < 0);
2495                         raw_ne = &nat_in_journal(journal, offset);
2496                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2497                 } else {
2498                         raw_ne = &nat_blk->entries[nid - start_nid];
2499                 }
2500                 raw_nat_from_node_info(raw_ne, &ne->ni);
2501                 nat_reset_flag(ne);
2502                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2503                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2504                         add_free_nid(sbi, nid, false, true);
2505                 } else {
2506                         spin_lock(&NM_I(sbi)->nid_list_lock);
2507                         update_free_nid_bitmap(sbi, nid, false, false);
2508                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2509                 }
2510         }
2511
2512         if (to_journal) {
2513                 up_write(&curseg->journal_rwsem);
2514         } else {
2515                 __update_nat_bits(sbi, start_nid, page);
2516                 f2fs_put_page(page, 1);
2517         }
2518
2519         /* Allow dirty nats by node block allocation in write_begin */
2520         if (!set->entry_cnt) {
2521                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2522                 kmem_cache_free(nat_entry_set_slab, set);
2523         }
2524 }
2525
2526 /*
2527  * This function is called during the checkpointing process.
2528  */
2529 void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2530 {
2531         struct f2fs_nm_info *nm_i = NM_I(sbi);
2532         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2533         struct f2fs_journal *journal = curseg->journal;
2534         struct nat_entry_set *setvec[SETVEC_SIZE];
2535         struct nat_entry_set *set, *tmp;
2536         unsigned int found;
2537         nid_t set_idx = 0;
2538         LIST_HEAD(sets);
2539
2540         if (!nm_i->dirty_nat_cnt)
2541                 return;
2542
2543         down_write(&nm_i->nat_tree_lock);
2544
2545         /*
2546          * if there are no enough space in journal to store dirty nat
2547          * entries, remove all entries from journal and merge them
2548          * into nat entry set.
2549          */
2550         if (enabled_nat_bits(sbi, cpc) ||
2551                 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2552                 remove_nats_in_journal(sbi);
2553
2554         while ((found = __gang_lookup_nat_set(nm_i,
2555                                         set_idx, SETVEC_SIZE, setvec))) {
2556                 unsigned idx;
2557                 set_idx = setvec[found - 1]->set + 1;
2558                 for (idx = 0; idx < found; idx++)
2559                         __adjust_nat_entry_set(setvec[idx], &sets,
2560                                                 MAX_NAT_JENTRIES(journal));
2561         }
2562
2563         /* flush dirty nats in nat entry set */
2564         list_for_each_entry_safe(set, tmp, &sets, set_list)
2565                 __flush_nat_entry_set(sbi, set, cpc);
2566
2567         up_write(&nm_i->nat_tree_lock);
2568         /* Allow dirty nats by node block allocation in write_begin */
2569 }
2570
2571 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2572 {
2573         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2574         struct f2fs_nm_info *nm_i = NM_I(sbi);
2575         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2576         unsigned int i;
2577         __u64 cp_ver = cur_cp_version(ckpt);
2578         block_t nat_bits_addr;
2579
2580         if (!enabled_nat_bits(sbi, NULL))
2581                 return 0;
2582
2583         nm_i->nat_bits_blocks = F2FS_BYTES_TO_BLK((nat_bits_bytes << 1) + 8 +
2584                                                 F2FS_BLKSIZE - 1);
2585         nm_i->nat_bits = f2fs_kzalloc(sbi,
2586                         nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2587         if (!nm_i->nat_bits)
2588                 return -ENOMEM;
2589
2590         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2591                                                 nm_i->nat_bits_blocks;
2592         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2593                 struct page *page = get_meta_page(sbi, nat_bits_addr++);
2594
2595                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2596                                         page_address(page), F2FS_BLKSIZE);
2597                 f2fs_put_page(page, 1);
2598         }
2599
2600         cp_ver |= (cur_cp_crc(ckpt) << 32);
2601         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2602                 disable_nat_bits(sbi, true);
2603                 return 0;
2604         }
2605
2606         nm_i->full_nat_bits = nm_i->nat_bits + 8;
2607         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2608
2609         f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2610         return 0;
2611 }
2612
2613 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2614 {
2615         struct f2fs_nm_info *nm_i = NM_I(sbi);
2616         unsigned int i = 0;
2617         nid_t nid, last_nid;
2618
2619         if (!enabled_nat_bits(sbi, NULL))
2620                 return;
2621
2622         for (i = 0; i < nm_i->nat_blocks; i++) {
2623                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2624                 if (i >= nm_i->nat_blocks)
2625                         break;
2626
2627                 __set_bit_le(i, nm_i->nat_block_bitmap);
2628
2629                 nid = i * NAT_ENTRY_PER_BLOCK;
2630                 last_nid = nid + NAT_ENTRY_PER_BLOCK;
2631
2632                 spin_lock(&NM_I(sbi)->nid_list_lock);
2633                 for (; nid < last_nid; nid++)
2634                         update_free_nid_bitmap(sbi, nid, true, true);
2635                 spin_unlock(&NM_I(sbi)->nid_list_lock);
2636         }
2637
2638         for (i = 0; i < nm_i->nat_blocks; i++) {
2639                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2640                 if (i >= nm_i->nat_blocks)
2641                         break;
2642
2643                 __set_bit_le(i, nm_i->nat_block_bitmap);
2644         }
2645 }
2646
2647 static int init_node_manager(struct f2fs_sb_info *sbi)
2648 {
2649         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2650         struct f2fs_nm_info *nm_i = NM_I(sbi);
2651         unsigned char *version_bitmap;
2652         unsigned int nat_segs;
2653         int err;
2654
2655         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2656
2657         /* segment_count_nat includes pair segment so divide to 2. */
2658         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2659         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2660         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2661
2662         /* not used nids: 0, node, meta, (and root counted as valid node) */
2663         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2664                                 sbi->nquota_files - F2FS_RESERVED_NODE_NUM;
2665         nm_i->nid_cnt[FREE_NID] = 0;
2666         nm_i->nid_cnt[PREALLOC_NID] = 0;
2667         nm_i->nat_cnt = 0;
2668         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2669         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2670         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2671
2672         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2673         INIT_LIST_HEAD(&nm_i->free_nid_list);
2674         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2675         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2676         INIT_LIST_HEAD(&nm_i->nat_entries);
2677
2678         mutex_init(&nm_i->build_lock);
2679         spin_lock_init(&nm_i->nid_list_lock);
2680         init_rwsem(&nm_i->nat_tree_lock);
2681
2682         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2683         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2684         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2685         if (!version_bitmap)
2686                 return -EFAULT;
2687
2688         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2689                                         GFP_KERNEL);
2690         if (!nm_i->nat_bitmap)
2691                 return -ENOMEM;
2692
2693         err = __get_nat_bitmaps(sbi);
2694         if (err)
2695                 return err;
2696
2697 #ifdef CONFIG_F2FS_CHECK_FS
2698         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2699                                         GFP_KERNEL);
2700         if (!nm_i->nat_bitmap_mir)
2701                 return -ENOMEM;
2702 #endif
2703
2704         return 0;
2705 }
2706
2707 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2708 {
2709         struct f2fs_nm_info *nm_i = NM_I(sbi);
2710
2711         nm_i->free_nid_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks *
2712                                         NAT_ENTRY_BITMAP_SIZE, GFP_KERNEL);
2713         if (!nm_i->free_nid_bitmap)
2714                 return -ENOMEM;
2715
2716         nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
2717                                                                 GFP_KERNEL);
2718         if (!nm_i->nat_block_bitmap)
2719                 return -ENOMEM;
2720
2721         nm_i->free_nid_count = f2fs_kvzalloc(sbi, nm_i->nat_blocks *
2722                                         sizeof(unsigned short), GFP_KERNEL);
2723         if (!nm_i->free_nid_count)
2724                 return -ENOMEM;
2725         return 0;
2726 }
2727
2728 int build_node_manager(struct f2fs_sb_info *sbi)
2729 {
2730         int err;
2731
2732         sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
2733                                                         GFP_KERNEL);
2734         if (!sbi->nm_info)
2735                 return -ENOMEM;
2736
2737         err = init_node_manager(sbi);
2738         if (err)
2739                 return err;
2740
2741         err = init_free_nid_cache(sbi);
2742         if (err)
2743                 return err;
2744
2745         /* load free nid status from nat_bits table */
2746         load_free_nid_bitmap(sbi);
2747
2748         build_free_nids(sbi, true, true);
2749         return 0;
2750 }
2751
2752 void destroy_node_manager(struct f2fs_sb_info *sbi)
2753 {
2754         struct f2fs_nm_info *nm_i = NM_I(sbi);
2755         struct free_nid *i, *next_i;
2756         struct nat_entry *natvec[NATVEC_SIZE];
2757         struct nat_entry_set *setvec[SETVEC_SIZE];
2758         nid_t nid = 0;
2759         unsigned int found;
2760
2761         if (!nm_i)
2762                 return;
2763
2764         /* destroy free nid list */
2765         spin_lock(&nm_i->nid_list_lock);
2766         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2767                 __remove_free_nid(sbi, i, FREE_NID);
2768                 spin_unlock(&nm_i->nid_list_lock);
2769                 kmem_cache_free(free_nid_slab, i);
2770                 spin_lock(&nm_i->nid_list_lock);
2771         }
2772         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
2773         f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
2774         f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
2775         spin_unlock(&nm_i->nid_list_lock);
2776
2777         /* destroy nat cache */
2778         down_write(&nm_i->nat_tree_lock);
2779         while ((found = __gang_lookup_nat_cache(nm_i,
2780                                         nid, NATVEC_SIZE, natvec))) {
2781                 unsigned idx;
2782
2783                 nid = nat_get_nid(natvec[found - 1]) + 1;
2784                 for (idx = 0; idx < found; idx++)
2785                         __del_from_nat_cache(nm_i, natvec[idx]);
2786         }
2787         f2fs_bug_on(sbi, nm_i->nat_cnt);
2788
2789         /* destroy nat set cache */
2790         nid = 0;
2791         while ((found = __gang_lookup_nat_set(nm_i,
2792                                         nid, SETVEC_SIZE, setvec))) {
2793                 unsigned idx;
2794
2795                 nid = setvec[found - 1]->set + 1;
2796                 for (idx = 0; idx < found; idx++) {
2797                         /* entry_cnt is not zero, when cp_error was occurred */
2798                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2799                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2800                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2801                 }
2802         }
2803         up_write(&nm_i->nat_tree_lock);
2804
2805         kvfree(nm_i->nat_block_bitmap);
2806         kvfree(nm_i->free_nid_bitmap);
2807         kvfree(nm_i->free_nid_count);
2808
2809         kfree(nm_i->nat_bitmap);
2810         kfree(nm_i->nat_bits);
2811 #ifdef CONFIG_F2FS_CHECK_FS
2812         kfree(nm_i->nat_bitmap_mir);
2813 #endif
2814         sbi->nm_info = NULL;
2815         kfree(nm_i);
2816 }
2817
2818 int __init create_node_manager_caches(void)
2819 {
2820         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2821                         sizeof(struct nat_entry));
2822         if (!nat_entry_slab)
2823                 goto fail;
2824
2825         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2826                         sizeof(struct free_nid));
2827         if (!free_nid_slab)
2828                 goto destroy_nat_entry;
2829
2830         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2831                         sizeof(struct nat_entry_set));
2832         if (!nat_entry_set_slab)
2833                 goto destroy_free_nid;
2834         return 0;
2835
2836 destroy_free_nid:
2837         kmem_cache_destroy(free_nid_slab);
2838 destroy_nat_entry:
2839         kmem_cache_destroy(nat_entry_slab);
2840 fail:
2841         return -ENOMEM;
2842 }
2843
2844 void destroy_node_manager_caches(void)
2845 {
2846         kmem_cache_destroy(nat_entry_set_slab);
2847         kmem_cache_destroy(free_nid_slab);
2848         kmem_cache_destroy(nat_entry_slab);
2849 }