git.rot13.org Git - linux/blob - 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 *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 142 {
 143         return radix_tree_lookup(&nm_i->nat_root, n);
 144 }
 145
 146 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 147                 nid_t start, unsigned int nr, struct nat_entry **ep)
 148 {
 149         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 150 }
 151
 152 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 153 {
 154         list_del(&e->list);
 155         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 156         nm_i->nat_cnt--;
 157         kmem_cache_free(nat_entry_slab, e);
 158 }
 159
 160 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 161                                                 struct nat_entry *ne)
 162 {
 163         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 164         struct nat_entry_set *head;
 165
 166         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 167         if (!head) {
 168                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 169
 170                 INIT_LIST_HEAD(&head->entry_list);
 171                 INIT_LIST_HEAD(&head->set_list);
 172                 head->set = set;
 173                 head->entry_cnt = 0;
 174                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 175         }
 176
 177         if (get_nat_flag(ne, IS_DIRTY))
 178                 goto refresh_list;
 179
 180         nm_i->dirty_nat_cnt++;
 181         head->entry_cnt++;
 182         set_nat_flag(ne, IS_DIRTY, true);
 183 refresh_list:
 184         if (nat_get_blkaddr(ne) == NEW_ADDR)
 185                 list_del_init(&ne->list);
 186         else
 187                 list_move_tail(&ne->list, &head->entry_list);
 188 }
 189
 190 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 191                 struct nat_entry_set *set, struct nat_entry *ne)
 192 {
 193         list_move_tail(&ne->list, &nm_i->nat_entries);
 194         set_nat_flag(ne, IS_DIRTY, false);
 195         set->entry_cnt--;
 196         nm_i->dirty_nat_cnt--;
 197 }
 198
 199 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 200                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 201 {
 202         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 203                                                         start, nr);
 204 }
 205
 206 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 207 {
 208         struct f2fs_nm_info *nm_i = NM_I(sbi);
 209         struct nat_entry *e;
 210         bool need = false;
 211
 212         down_read(&nm_i->nat_tree_lock);
 213         e = __lookup_nat_cache(nm_i, nid);
 214         if (e) {
 215                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 216                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 217                         need = true;
 218         }
 219         up_read(&nm_i->nat_tree_lock);
 220         return need;
 221 }
 222
 223 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 224 {
 225         struct f2fs_nm_info *nm_i = NM_I(sbi);
 226         struct nat_entry *e;
 227         bool is_cp = true;
 228
 229         down_read(&nm_i->nat_tree_lock);
 230         e = __lookup_nat_cache(nm_i, nid);
 231         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 232                 is_cp = false;
 233         up_read(&nm_i->nat_tree_lock);
 234         return is_cp;
 235 }
 236
 237 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 238 {
 239         struct f2fs_nm_info *nm_i = NM_I(sbi);
 240         struct nat_entry *e;
 241         bool need_update = true;
 242
 243         down_read(&nm_i->nat_tree_lock);
 244         e = __lookup_nat_cache(nm_i, ino);
 245         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 246                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 247                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 248                 need_update = false;
 249         up_read(&nm_i->nat_tree_lock);
 250         return need_update;
 251 }
 252
 253 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 254                                                                 bool no_fail)
 255 {
 256         struct nat_entry *new;
 257
 258         if (no_fail) {
 259                 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
 260                 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
 261         } else {
 262                 new = kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
 263                 if (!new)
 264                         return NULL;
 265                 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 266                         kmem_cache_free(nat_entry_slab, new);
 267                         return NULL;
 268                 }
 269         }
 270
 271         memset(new, 0, sizeof(struct nat_entry));
 272         nat_set_nid(new, nid);
 273         nat_reset_flag(new);
 274         list_add_tail(&new->list, &nm_i->nat_entries);
 275         nm_i->nat_cnt++;
 276         return new;
 277 }
 278
 279 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 280                                                 struct f2fs_nat_entry *ne)
 281 {
 282         struct f2fs_nm_info *nm_i = NM_I(sbi);
 283         struct nat_entry *e;
 284
 285         e = __lookup_nat_cache(nm_i, nid);
 286         if (!e) {
 287                 e = grab_nat_entry(nm_i, nid, false);
 288                 if (e)
 289                         node_info_from_raw_nat(&e->ni, ne);
 290         } else {
 291                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 292                                 nat_get_blkaddr(e) !=
 293                                         le32_to_cpu(ne->block_addr) ||
 294                                 nat_get_version(e) != ne->version);
 295         }
 296 }
 297
 298 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 299                         block_t new_blkaddr, bool fsync_done)
 300 {
 301         struct f2fs_nm_info *nm_i = NM_I(sbi);
 302         struct nat_entry *e;
 303
 304         down_write(&nm_i->nat_tree_lock);
 305         e = __lookup_nat_cache(nm_i, ni->nid);
 306         if (!e) {
 307                 e = grab_nat_entry(nm_i, ni->nid, true);
 308                 copy_node_info(&e->ni, ni);
 309                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 310         } else if (new_blkaddr == NEW_ADDR) {
 311                 /*
 312                  * when nid is reallocated,
 313                  * previous nat entry can be remained in nat cache.
 314                  * So, reinitialize it with new information.
 315                  */
 316                 copy_node_info(&e->ni, ni);
 317                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 318         }
 319
 320         /* sanity check */
 321         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 322         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 323                         new_blkaddr == NULL_ADDR);
 324         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 325                         new_blkaddr == NEW_ADDR);
 326         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 327                         nat_get_blkaddr(e) != NULL_ADDR &&
 328                         new_blkaddr == NEW_ADDR);
 329
 330         /* increment version no as node is removed */
 331         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 332                 unsigned char version = nat_get_version(e);
 333                 nat_set_version(e, inc_node_version(version));
 334         }
 335
 336         /* change address */
 337         nat_set_blkaddr(e, new_blkaddr);
 338         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 339                 set_nat_flag(e, IS_CHECKPOINTED, false);
 340         __set_nat_cache_dirty(nm_i, e);
 341
 342         /* update fsync_mark if its inode nat entry is still alive */
 343         if (ni->nid != ni->ino)
 344                 e = __lookup_nat_cache(nm_i, ni->ino);
 345         if (e) {
 346                 if (fsync_done && ni->nid == ni->ino)
 347                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 348                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 349         }
 350         up_write(&nm_i->nat_tree_lock);
 351 }
 352
 353 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 354 {
 355         struct f2fs_nm_info *nm_i = NM_I(sbi);
 356         int nr = nr_shrink;
 357
 358         if (!down_write_trylock(&nm_i->nat_tree_lock))
 359                 return 0;
 360
 361         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 362                 struct nat_entry *ne;
 363                 ne = list_first_entry(&nm_i->nat_entries,
 364                                         struct nat_entry, list);
 365                 __del_from_nat_cache(nm_i, ne);
 366                 nr_shrink--;
 367         }
 368         up_write(&nm_i->nat_tree_lock);
 369         return nr - nr_shrink;
 370 }
 371
 372 /*
 373  * This function always returns success
 374  */
 375 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 376 {
 377         struct f2fs_nm_info *nm_i = NM_I(sbi);
 378         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 379         struct f2fs_journal *journal = curseg->journal;
 380         nid_t start_nid = START_NID(nid);
 381         struct f2fs_nat_block *nat_blk;
 382         struct page *page = NULL;
 383         struct f2fs_nat_entry ne;
 384         struct nat_entry *e;
 385         pgoff_t index;
 386         int i;
 387
 388         ni->nid = nid;
 389
 390         /* Check nat cache */
 391         down_read(&nm_i->nat_tree_lock);
 392         e = __lookup_nat_cache(nm_i, nid);
 393         if (e) {
 394                 ni->ino = nat_get_ino(e);
 395                 ni->blk_addr = nat_get_blkaddr(e);
 396                 ni->version = nat_get_version(e);
 397                 up_read(&nm_i->nat_tree_lock);
 398                 return;
 399         }
 400
 401         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 402
 403         /* Check current segment summary */
 404         down_read(&curseg->journal_rwsem);
 405         i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 406         if (i >= 0) {
 407                 ne = nat_in_journal(journal, i);
 408                 node_info_from_raw_nat(ni, &ne);
 409         }
 410         up_read(&curseg->journal_rwsem);
 411         if (i >= 0) {
 412                 up_read(&nm_i->nat_tree_lock);
 413                 goto cache;
 414         }
 415
 416         /* Fill node_info from nat page */
 417         index = current_nat_addr(sbi, nid);
 418         up_read(&nm_i->nat_tree_lock);
 419
 420         page = get_meta_page(sbi, index);
 421         nat_blk = (struct f2fs_nat_block *)page_address(page);
 422         ne = nat_blk->entries[nid - start_nid];
 423         node_info_from_raw_nat(ni, &ne);
 424         f2fs_put_page(page, 1);
 425 cache:
 426         /* cache nat entry */
 427         down_write(&nm_i->nat_tree_lock);
 428         cache_nat_entry(sbi, nid, &ne);
 429         up_write(&nm_i->nat_tree_lock);
 430 }
 431
 432 /*
 433  * readahead MAX_RA_NODE number of node pages.
 434  */
 435 static void ra_node_pages(struct page *parent, int start, int n)
 436 {
 437         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 438         struct blk_plug plug;
 439         int i, end;
 440         nid_t nid;
 441
 442         blk_start_plug(&plug);
 443
 444         /* Then, try readahead for siblings of the desired node */
 445         end = start + n;
 446         end = min(end, NIDS_PER_BLOCK);
 447         for (i = start; i < end; i++) {
 448                 nid = get_nid(parent, i, false);
 449                 ra_node_page(sbi, nid);
 450         }
 451
 452         blk_finish_plug(&plug);
 453 }
 454
 455 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 456 {
 457         const long direct_index = ADDRS_PER_INODE(dn->inode);
 458         const long direct_blks = ADDRS_PER_BLOCK;
 459         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 460         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 461         int cur_level = dn->cur_level;
 462         int max_level = dn->max_level;
 463         pgoff_t base = 0;
 464
 465         if (!dn->max_level)
 466                 return pgofs + 1;
 467
 468         while (max_level-- > cur_level)
 469                 skipped_unit *= NIDS_PER_BLOCK;
 470
 471         switch (dn->max_level) {
 472         case 3:
 473                 base += 2 * indirect_blks;
 474         case 2:
 475                 base += 2 * direct_blks;
 476         case 1:
 477                 base += direct_index;
 478                 break;
 479         default:
 480                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 481         }
 482
 483         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 484 }
 485
 486 /*
 487  * The maximum depth is four.
 488  * Offset[0] will have raw inode offset.
 489  */
 490 static int get_node_path(struct inode *inode, long block,
 491                                 int offset[4], unsigned int noffset[4])
 492 {
 493         const long direct_index = ADDRS_PER_INODE(inode);
 494         const long direct_blks = ADDRS_PER_BLOCK;
 495         const long dptrs_per_blk = NIDS_PER_BLOCK;
 496         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 497         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 498         int n = 0;
 499         int level = 0;
 500
 501         noffset[0] = 0;
 502
 503         if (block < direct_index) {
 504                 offset[n] = block;
 505                 goto got;
 506         }
 507         block -= direct_index;
 508         if (block < direct_blks) {
 509                 offset[n++] = NODE_DIR1_BLOCK;
 510                 noffset[n] = 1;
 511                 offset[n] = block;
 512                 level = 1;
 513                 goto got;
 514         }
 515         block -= direct_blks;
 516         if (block < direct_blks) {
 517                 offset[n++] = NODE_DIR2_BLOCK;
 518                 noffset[n] = 2;
 519                 offset[n] = block;
 520                 level = 1;
 521                 goto got;
 522         }
 523         block -= direct_blks;
 524         if (block < indirect_blks) {
 525                 offset[n++] = NODE_IND1_BLOCK;
 526                 noffset[n] = 3;
 527                 offset[n++] = block / direct_blks;
 528                 noffset[n] = 4 + offset[n - 1];
 529                 offset[n] = block % direct_blks;
 530                 level = 2;
 531                 goto got;
 532         }
 533         block -= indirect_blks;
 534         if (block < indirect_blks) {
 535                 offset[n++] = NODE_IND2_BLOCK;
 536                 noffset[n] = 4 + dptrs_per_blk;
 537                 offset[n++] = block / direct_blks;
 538                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 539                 offset[n] = block % direct_blks;
 540                 level = 2;
 541                 goto got;
 542         }
 543         block -= indirect_blks;
 544         if (block < dindirect_blks) {
 545                 offset[n++] = NODE_DIND_BLOCK;
 546                 noffset[n] = 5 + (dptrs_per_blk * 2);
 547                 offset[n++] = block / indirect_blks;
 548                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 549                               offset[n - 1] * (dptrs_per_blk + 1);
 550                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 551                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 552                               offset[n - 2] * (dptrs_per_blk + 1) +
 553                               offset[n - 1];
 554                 offset[n] = block % direct_blks;
 555                 level = 3;
 556                 goto got;
 557         } else {
 558                 return -E2BIG;
 559         }
 560 got:
 561         return level;
 562 }
 563
 564 /*
 565  * Caller should call f2fs_put_dnode(dn).
 566  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 567  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 568  * In the case of RDONLY_NODE, we don't need to care about mutex.
 569  */
 570 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 571 {
 572         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 573         struct page *npage[4];
 574         struct page *parent = NULL;
 575         int offset[4];
 576         unsigned int noffset[4];
 577         nid_t nids[4];
 578         int level, i = 0;
 579         int err = 0;
 580
 581         level = get_node_path(dn->inode, index, offset, noffset);
 582         if (level < 0)
 583                 return level;
 584
 585         nids[0] = dn->inode->i_ino;
 586         npage[0] = dn->inode_page;
 587
 588         if (!npage[0]) {
 589                 npage[0] = get_node_page(sbi, nids[0]);
 590                 if (IS_ERR(npage[0]))
 591                         return PTR_ERR(npage[0]);
 592         }
 593
 594         /* if inline_data is set, should not report any block indices */
 595         if (f2fs_has_inline_data(dn->inode) && index) {
 596                 err = -ENOENT;
 597                 f2fs_put_page(npage[0], 1);
 598                 goto release_out;
 599         }
 600
 601         parent = npage[0];
 602         if (level != 0)
 603                 nids[1] = get_nid(parent, offset[0], true);
 604         dn->inode_page = npage[0];
 605         dn->inode_page_locked = true;
 606
 607         /* get indirect or direct nodes */
 608         for (i = 1; i <= level; i++) {
 609                 bool done = false;
 610
 611                 if (!nids[i] && mode == ALLOC_NODE) {
 612                         /* alloc new node */
 613                         if (!alloc_nid(sbi, &(nids[i]))) {
 614                                 err = -ENOSPC;
 615                                 goto release_pages;
 616                         }
 617
 618                         dn->nid = nids[i];
 619                         npage[i] = new_node_page(dn, noffset[i]);
 620                         if (IS_ERR(npage[i])) {
 621                                 alloc_nid_failed(sbi, nids[i]);
 622                                 err = PTR_ERR(npage[i]);
 623                                 goto release_pages;
 624                         }
 625
 626                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 627                         alloc_nid_done(sbi, nids[i]);
 628                         done = true;
 629                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 630                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 631                         if (IS_ERR(npage[i])) {
 632                                 err = PTR_ERR(npage[i]);
 633                                 goto release_pages;
 634                         }
 635                         done = true;
 636                 }
 637                 if (i == 1) {
 638                         dn->inode_page_locked = false;
 639                         unlock_page(parent);
 640                 } else {
 641                         f2fs_put_page(parent, 1);
 642                 }
 643
 644                 if (!done) {
 645                         npage[i] = get_node_page(sbi, nids[i]);
 646                         if (IS_ERR(npage[i])) {
 647                                 err = PTR_ERR(npage[i]);
 648                                 f2fs_put_page(npage[0], 0);
 649                                 goto release_out;
 650                         }
 651                 }
 652                 if (i < level) {
 653                         parent = npage[i];
 654                         nids[i + 1] = get_nid(parent, offset[i], false);
 655                 }
 656         }
 657         dn->nid = nids[level];
 658         dn->ofs_in_node = offset[level];
 659         dn->node_page = npage[level];
 660         dn->data_blkaddr = datablock_addr(dn->inode,
 661                                 dn->node_page, dn->ofs_in_node);
 662         return 0;
 663
 664 release_pages:
 665         f2fs_put_page(parent, 1);
 666         if (i > 1)
 667                 f2fs_put_page(npage[0], 0);
 668 release_out:
 669         dn->inode_page = NULL;
 670         dn->node_page = NULL;
 671         if (err == -ENOENT) {
 672                 dn->cur_level = i;
 673                 dn->max_level = level;
 674                 dn->ofs_in_node = offset[level];
 675         }
 676         return err;
 677 }
 678
 679 static void truncate_node(struct dnode_of_data *dn)
 680 {
 681         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 682         struct node_info ni;
 683
 684         get_node_info(sbi, dn->nid, &ni);
 685         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 686
 687         /* Deallocate node address */
 688         invalidate_blocks(sbi, ni.blk_addr);
 689         dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
 690         set_node_addr(sbi, &ni, NULL_ADDR, false);
 691
 692         if (dn->nid == dn->inode->i_ino) {
 693                 remove_orphan_inode(sbi, dn->nid);
 694                 dec_valid_inode_count(sbi);
 695                 f2fs_inode_synced(dn->inode);
 696         }
 697
 698         clear_node_page_dirty(dn->node_page);
 699         set_sbi_flag(sbi, SBI_IS_DIRTY);
 700
 701         f2fs_put_page(dn->node_page, 1);
 702
 703         invalidate_mapping_pages(NODE_MAPPING(sbi),
 704                         dn->node_page->index, dn->node_page->index);
 705
 706         dn->node_page = NULL;
 707         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 708 }
 709
 710 static int truncate_dnode(struct dnode_of_data *dn)
 711 {
 712         struct page *page;
 713
 714         if (dn->nid == 0)
 715                 return 1;
 716
 717         /* get direct node */
 718         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 719         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 720                 return 1;
 721         else if (IS_ERR(page))
 722                 return PTR_ERR(page);
 723
 724         /* Make dnode_of_data for parameter */
 725         dn->node_page = page;
 726         dn->ofs_in_node = 0;
 727         truncate_data_blocks(dn);
 728         truncate_node(dn);
 729         return 1;
 730 }
 731
 732 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 733                                                 int ofs, int depth)
 734 {
 735         struct dnode_of_data rdn = *dn;
 736         struct page *page;
 737         struct f2fs_node *rn;
 738         nid_t child_nid;
 739         unsigned int child_nofs;
 740         int freed = 0;
 741         int i, ret;
 742
 743         if (dn->nid == 0)
 744                 return NIDS_PER_BLOCK + 1;
 745
 746         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 747
 748         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 749         if (IS_ERR(page)) {
 750                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 751                 return PTR_ERR(page);
 752         }
 753
 754         ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 755
 756         rn = F2FS_NODE(page);
 757         if (depth < 3) {
 758                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 759                         child_nid = le32_to_cpu(rn->in.nid[i]);
 760                         if (child_nid == 0)
 761                                 continue;
 762                         rdn.nid = child_nid;
 763                         ret = truncate_dnode(&rdn);
 764                         if (ret < 0)
 765                                 goto out_err;
 766                         if (set_nid(page, i, 0, false))
 767                                 dn->node_changed = true;
 768                 }
 769         } else {
 770                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 771                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 772                         child_nid = le32_to_cpu(rn->in.nid[i]);
 773                         if (child_nid == 0) {
 774                                 child_nofs += NIDS_PER_BLOCK + 1;
 775                                 continue;
 776                         }
 777                         rdn.nid = child_nid;
 778                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 779                         if (ret == (NIDS_PER_BLOCK + 1)) {
 780                                 if (set_nid(page, i, 0, false))
 781                                         dn->node_changed = true;
 782                                 child_nofs += ret;
 783                         } else if (ret < 0 && ret != -ENOENT) {
 784                                 goto out_err;
 785                         }
 786                 }
 787                 freed = child_nofs;
 788         }
 789
 790         if (!ofs) {
 791                 /* remove current indirect node */
 792                 dn->node_page = page;
 793                 truncate_node(dn);
 794                 freed++;
 795         } else {
 796                 f2fs_put_page(page, 1);
 797         }
 798         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 799         return freed;
 800
 801 out_err:
 802         f2fs_put_page(page, 1);
 803         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 804         return ret;
 805 }
 806
 807 static int truncate_partial_nodes(struct dnode_of_data *dn,
 808                         struct f2fs_inode *ri, int *offset, int depth)
 809 {
 810         struct page *pages[2];
 811         nid_t nid[3];
 812         nid_t child_nid;
 813         int err = 0;
 814         int i;
 815         int idx = depth - 2;
 816
 817         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 818         if (!nid[0])
 819                 return 0;
 820
 821         /* get indirect nodes in the path */
 822         for (i = 0; i < idx + 1; i++) {
 823                 /* reference count'll be increased */
 824                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 825                 if (IS_ERR(pages[i])) {
 826                         err = PTR_ERR(pages[i]);
 827                         idx = i - 1;
 828                         goto fail;
 829                 }
 830                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 831         }
 832
 833         ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 834
 835         /* free direct nodes linked to a partial indirect node */
 836         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 837                 child_nid = get_nid(pages[idx], i, false);
 838                 if (!child_nid)
 839                         continue;
 840                 dn->nid = child_nid;
 841                 err = truncate_dnode(dn);
 842                 if (err < 0)
 843                         goto fail;
 844                 if (set_nid(pages[idx], i, 0, false))
 845                         dn->node_changed = true;
 846         }
 847
 848         if (offset[idx + 1] == 0) {
 849                 dn->node_page = pages[idx];
 850                 dn->nid = nid[idx];
 851                 truncate_node(dn);
 852         } else {
 853                 f2fs_put_page(pages[idx], 1);
 854         }
 855         offset[idx]++;
 856         offset[idx + 1] = 0;
 857         idx--;
 858 fail:
 859         for (i = idx; i >= 0; i--)
 860                 f2fs_put_page(pages[i], 1);
 861
 862         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 863
 864         return err;
 865 }
 866
 867 /*
 868  * All the block addresses of data and nodes should be nullified.
 869  */
 870 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 871 {
 872         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 873         int err = 0, cont = 1;
 874         int level, offset[4], noffset[4];
 875         unsigned int nofs = 0;
 876         struct f2fs_inode *ri;
 877         struct dnode_of_data dn;
 878         struct page *page;
 879
 880         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 881
 882         level = get_node_path(inode, from, offset, noffset);
 883         if (level < 0)
 884                 return level;
 885
 886         page = get_node_page(sbi, inode->i_ino);
 887         if (IS_ERR(page)) {
 888                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 889                 return PTR_ERR(page);
 890         }
 891
 892         set_new_dnode(&dn, inode, page, NULL, 0);
 893         unlock_page(page);
 894
 895         ri = F2FS_INODE(page);
 896         switch (level) {
 897         case 0:
 898         case 1:
 899                 nofs = noffset[1];
 900                 break;
 901         case 2:
 902                 nofs = noffset[1];
 903                 if (!offset[level - 1])
 904                         goto skip_partial;
 905                 err = truncate_partial_nodes(&dn, ri, offset, level);
 906                 if (err < 0 && err != -ENOENT)
 907                         goto fail;
 908                 nofs += 1 + NIDS_PER_BLOCK;
 909                 break;
 910         case 3:
 911                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 912                 if (!offset[level - 1])
 913                         goto skip_partial;
 914                 err = truncate_partial_nodes(&dn, ri, offset, level);
 915                 if (err < 0 && err != -ENOENT)
 916                         goto fail;
 917                 break;
 918         default:
 919                 BUG();
 920         }
 921
 922 skip_partial:
 923         while (cont) {
 924                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 925                 switch (offset[0]) {
 926                 case NODE_DIR1_BLOCK:
 927                 case NODE_DIR2_BLOCK:
 928                         err = truncate_dnode(&dn);
 929                         break;
 930
 931                 case NODE_IND1_BLOCK:
 932                 case NODE_IND2_BLOCK:
 933                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 934                         break;
 935
 936                 case NODE_DIND_BLOCK:
 937                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 938                         cont = 0;
 939                         break;
 940
 941                 default:
 942                         BUG();
 943                 }
 944                 if (err < 0 && err != -ENOENT)
 945                         goto fail;
 946                 if (offset[1] == 0 &&
 947                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 948                         lock_page(page);
 949                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
 950                         f2fs_wait_on_page_writeback(page, NODE, true);
 951                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 952                         set_page_dirty(page);
 953                         unlock_page(page);
 954                 }
 955                 offset[1] = 0;
 956                 offset[0]++;
 957                 nofs += err;
 958         }
 959 fail:
 960         f2fs_put_page(page, 0);
 961         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 962         return err > 0 ? 0 : err;
 963 }
 964
 965 /* caller must lock inode page */
 966 int truncate_xattr_node(struct inode *inode)
 967 {
 968         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 969         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 970         struct dnode_of_data dn;
 971         struct page *npage;
 972
 973         if (!nid)
 974                 return 0;
 975
 976         npage = get_node_page(sbi, nid);
 977         if (IS_ERR(npage))
 978                 return PTR_ERR(npage);
 979
 980         f2fs_i_xnid_write(inode, 0);
 981
 982         set_new_dnode(&dn, inode, NULL, npage, nid);
 983         truncate_node(&dn);
 984         return 0;
 985 }
 986
 987 /*
 988  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 989  * f2fs_unlock_op().
 990  */
 991 int remove_inode_page(struct inode *inode)
 992 {
 993         struct dnode_of_data dn;
 994         int err;
 995
 996         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 997         err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
 998         if (err)
 999                 return err;
1000
1001         err = truncate_xattr_node(inode);
1002         if (err) {
1003                 f2fs_put_dnode(&dn);
1004                 return err;
1005         }
1006
1007         /* remove potential inline_data blocks */
1008         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1009                                 S_ISLNK(inode->i_mode))
1010                 truncate_data_blocks_range(&dn, 1);
1011
1012         /* 0 is possible, after f2fs_new_inode() has failed */
1013         f2fs_bug_on(F2FS_I_SB(inode),
1014                         inode->i_blocks != 0 && inode->i_blocks != 8);
1015
1016         /* will put inode & node pages */
1017         truncate_node(&dn);
1018         return 0;
1019 }
1020
1021 struct page *new_inode_page(struct inode *inode)
1022 {
1023         struct dnode_of_data dn;
1024
1025         /* allocate inode page for new inode */
1026         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1027
1028         /* caller should f2fs_put_page(page, 1); */
1029         return new_node_page(&dn, 0);
1030 }
1031
1032 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1033 {
1034         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1035         struct node_info new_ni;
1036         struct page *page;
1037         int err;
1038
1039         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1040                 return ERR_PTR(-EPERM);
1041
1042         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1043         if (!page)
1044                 return ERR_PTR(-ENOMEM);
1045
1046         if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1047                 goto fail;
1048
1049 #ifdef CONFIG_F2FS_CHECK_FS
1050         get_node_info(sbi, dn->nid, &new_ni);
1051         f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1052 #endif
1053         new_ni.nid = dn->nid;
1054         new_ni.ino = dn->inode->i_ino;
1055         new_ni.blk_addr = NULL_ADDR;
1056         new_ni.flag = 0;
1057         new_ni.version = 0;
1058         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1059
1060         f2fs_wait_on_page_writeback(page, NODE, true);
1061         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1062         set_cold_node(dn->inode, page);
1063         if (!PageUptodate(page))
1064                 SetPageUptodate(page);
1065         if (set_page_dirty(page))
1066                 dn->node_changed = true;
1067
1068         if (f2fs_has_xattr_block(ofs))
1069                 f2fs_i_xnid_write(dn->inode, dn->nid);
1070
1071         if (ofs == 0)
1072                 inc_valid_inode_count(sbi);
1073         return page;
1074
1075 fail:
1076         clear_node_page_dirty(page);
1077         f2fs_put_page(page, 1);
1078         return ERR_PTR(err);
1079 }
1080
1081 /*
1082  * Caller should do after getting the following values.
1083  * 0: f2fs_put_page(page, 0)
1084  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1085  */
1086 static int read_node_page(struct page *page, int op_flags)
1087 {
1088         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1089         struct node_info ni;
1090         struct f2fs_io_info fio = {
1091                 .sbi = sbi,
1092                 .type = NODE,
1093                 .op = REQ_OP_READ,
1094                 .op_flags = op_flags,
1095                 .page = page,
1096                 .encrypted_page = NULL,
1097         };
1098
1099         if (PageUptodate(page))
1100                 return LOCKED_PAGE;
1101
1102         get_node_info(sbi, page->index, &ni);
1103
1104         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1105                 ClearPageUptodate(page);
1106                 return -ENOENT;
1107         }
1108
1109         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1110         return f2fs_submit_page_bio(&fio);
1111 }
1112
1113 /*
1114  * Readahead a node page
1115  */
1116 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1117 {
1118         struct page *apage;
1119         int err;
1120
1121         if (!nid)
1122                 return;
1123         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1124
1125         rcu_read_lock();
1126         apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1127         rcu_read_unlock();
1128         if (apage)
1129                 return;
1130
1131         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1132         if (!apage)
1133                 return;
1134
1135         err = read_node_page(apage, REQ_RAHEAD);
1136         f2fs_put_page(apage, err ? 1 : 0);
1137 }
1138
1139 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1140                                         struct page *parent, int start)
1141 {
1142         struct page *page;
1143         int err;
1144
1145         if (!nid)
1146                 return ERR_PTR(-ENOENT);
1147         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1148 repeat:
1149         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1150         if (!page)
1151                 return ERR_PTR(-ENOMEM);
1152
1153         err = read_node_page(page, 0);
1154         if (err < 0) {
1155                 f2fs_put_page(page, 1);
1156                 return ERR_PTR(err);
1157         } else if (err == LOCKED_PAGE) {
1158                 err = 0;
1159                 goto page_hit;
1160         }
1161
1162         if (parent)
1163                 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1164
1165         lock_page(page);
1166
1167         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1168                 f2fs_put_page(page, 1);
1169                 goto repeat;
1170         }
1171
1172         if (unlikely(!PageUptodate(page))) {
1173                 err = -EIO;
1174                 goto out_err;
1175         }
1176
1177         if (!f2fs_inode_chksum_verify(sbi, page)) {
1178                 err = -EBADMSG;
1179                 goto out_err;
1180         }
1181 page_hit:
1182         if(unlikely(nid != nid_of_node(page))) {
1183                 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1184                         "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1185                         nid, nid_of_node(page), ino_of_node(page),
1186                         ofs_of_node(page), cpver_of_node(page),
1187                         next_blkaddr_of_node(page));
1188                 err = -EINVAL;
1189 out_err:
1190                 ClearPageUptodate(page);
1191                 f2fs_put_page(page, 1);
1192                 return ERR_PTR(err);
1193         }
1194         return page;
1195 }
1196
1197 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1198 {
1199         return __get_node_page(sbi, nid, NULL, 0);
1200 }
1201
1202 struct page *get_node_page_ra(struct page *parent, int start)
1203 {
1204         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1205         nid_t nid = get_nid(parent, start, false);
1206
1207         return __get_node_page(sbi, nid, parent, start);
1208 }
1209
1210 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1211 {
1212         struct inode *inode;
1213         struct page *page;
1214         int ret;
1215
1216         /* should flush inline_data before evict_inode */
1217         inode = ilookup(sbi->sb, ino);
1218         if (!inode)
1219                 return;
1220
1221         page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1222                                         FGP_LOCK|FGP_NOWAIT, 0);
1223         if (!page)
1224                 goto iput_out;
1225
1226         if (!PageUptodate(page))
1227                 goto page_out;
1228
1229         if (!PageDirty(page))
1230                 goto page_out;
1231
1232         if (!clear_page_dirty_for_io(page))
1233                 goto page_out;
1234
1235         ret = f2fs_write_inline_data(inode, page);
1236         inode_dec_dirty_pages(inode);
1237         remove_dirty_inode(inode);
1238         if (ret)
1239                 set_page_dirty(page);
1240 page_out:
1241         f2fs_put_page(page, 1);
1242 iput_out:
1243         iput(inode);
1244 }
1245
1246 void move_node_page(struct page *node_page, int gc_type)
1247 {
1248         if (gc_type == FG_GC) {
1249                 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1250                 struct writeback_control wbc = {
1251                         .sync_mode = WB_SYNC_ALL,
1252                         .nr_to_write = 1,
1253                         .for_reclaim = 0,
1254                 };
1255
1256                 set_page_dirty(node_page);
1257                 f2fs_wait_on_page_writeback(node_page, NODE, true);
1258
1259                 f2fs_bug_on(sbi, PageWriteback(node_page));
1260                 if (!clear_page_dirty_for_io(node_page))
1261                         goto out_page;
1262
1263                 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1264                         unlock_page(node_page);
1265                 goto release_page;
1266         } else {
1267                 /* set page dirty and write it */
1268                 if (!PageWriteback(node_page))
1269                         set_page_dirty(node_page);
1270         }
1271 out_page:
1272         unlock_page(node_page);
1273 release_page:
1274         f2fs_put_page(node_page, 0);
1275 }
1276
1277 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1278 {
1279         pgoff_t index, end;
1280         struct pagevec pvec;
1281         struct page *last_page = NULL;
1282
1283         pagevec_init(&pvec, 0);
1284         index = 0;
1285         end = ULONG_MAX;
1286
1287         while (index <= end) {
1288                 int i, nr_pages;
1289                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1290                                 PAGECACHE_TAG_DIRTY,
1291                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1292                 if (nr_pages == 0)
1293                         break;
1294
1295                 for (i = 0; i < nr_pages; i++) {
1296                         struct page *page = pvec.pages[i];
1297
1298                         if (unlikely(f2fs_cp_error(sbi))) {
1299                                 f2fs_put_page(last_page, 0);
1300                                 pagevec_release(&pvec);
1301                                 return ERR_PTR(-EIO);
1302                         }
1303
1304                         if (!IS_DNODE(page) || !is_cold_node(page))
1305                                 continue;
1306                         if (ino_of_node(page) != ino)
1307                                 continue;
1308
1309                         lock_page(page);
1310
1311                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1312 continue_unlock:
1313                                 unlock_page(page);
1314                                 continue;
1315                         }
1316                         if (ino_of_node(page) != ino)
1317                                 goto continue_unlock;
1318
1319                         if (!PageDirty(page)) {
1320                                 /* someone wrote it for us */
1321                                 goto continue_unlock;
1322                         }
1323
1324                         if (last_page)
1325                                 f2fs_put_page(last_page, 0);
1326
1327                         get_page(page);
1328                         last_page = page;
1329                         unlock_page(page);
1330                 }
1331                 pagevec_release(&pvec);
1332                 cond_resched();
1333         }
1334         return last_page;
1335 }
1336
1337 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1338                                 struct writeback_control *wbc, bool do_balance,
1339                                 enum iostat_type io_type)
1340 {
1341         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1342         nid_t nid;
1343         struct node_info ni;
1344         struct f2fs_io_info fio = {
1345                 .sbi = sbi,
1346                 .ino = ino_of_node(page),
1347                 .type = NODE,
1348                 .op = REQ_OP_WRITE,
1349                 .op_flags = wbc_to_write_flags(wbc),
1350                 .page = page,
1351                 .encrypted_page = NULL,
1352                 .submitted = false,
1353                 .io_type = io_type,
1354         };
1355
1356         trace_f2fs_writepage(page, NODE);
1357
1358         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1359                 goto redirty_out;
1360         if (unlikely(f2fs_cp_error(sbi)))
1361                 goto redirty_out;
1362
1363         /* get old block addr of this node page */
1364         nid = nid_of_node(page);
1365         f2fs_bug_on(sbi, page->index != nid);
1366
1367         if (wbc->for_reclaim) {
1368                 if (!down_read_trylock(&sbi->node_write))
1369                         goto redirty_out;
1370         } else {
1371                 down_read(&sbi->node_write);
1372         }
1373
1374         get_node_info(sbi, nid, &ni);
1375
1376         /* This page is already truncated */
1377         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1378                 ClearPageUptodate(page);
1379                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1380                 up_read(&sbi->node_write);
1381                 unlock_page(page);
1382                 return 0;
1383         }
1384
1385         if (atomic && !test_opt(sbi, NOBARRIER))
1386                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1387
1388         set_page_writeback(page);
1389         fio.old_blkaddr = ni.blk_addr;
1390         write_node_page(nid, &fio);
1391         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1392         dec_page_count(sbi, F2FS_DIRTY_NODES);
1393         up_read(&sbi->node_write);
1394
1395         if (wbc->for_reclaim) {
1396                 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1397                                                 page->index, NODE);
1398                 submitted = NULL;
1399         }
1400
1401         unlock_page(page);
1402
1403         if (unlikely(f2fs_cp_error(sbi))) {
1404                 f2fs_submit_merged_write(sbi, NODE);
1405                 submitted = NULL;
1406         }
1407         if (submitted)
1408                 *submitted = fio.submitted;
1409
1410         if (do_balance)
1411                 f2fs_balance_fs(sbi, false);
1412         return 0;
1413
1414 redirty_out:
1415         redirty_page_for_writepage(wbc, page);
1416         return AOP_WRITEPAGE_ACTIVATE;
1417 }
1418
1419 static int f2fs_write_node_page(struct page *page,
1420                                 struct writeback_control *wbc)
1421 {
1422         return __write_node_page(page, false, NULL, wbc, false, FS_NODE_IO);
1423 }
1424
1425 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1426                         struct writeback_control *wbc, bool atomic)
1427 {
1428         pgoff_t index, end;
1429         pgoff_t last_idx = ULONG_MAX;
1430         struct pagevec pvec;
1431         int ret = 0;
1432         struct page *last_page = NULL;
1433         bool marked = false;
1434         nid_t ino = inode->i_ino;
1435
1436         if (atomic) {
1437                 last_page = last_fsync_dnode(sbi, ino);
1438                 if (IS_ERR_OR_NULL(last_page))
1439                         return PTR_ERR_OR_ZERO(last_page);
1440         }
1441 retry:
1442         pagevec_init(&pvec, 0);
1443         index = 0;
1444         end = ULONG_MAX;
1445
1446         while (index <= end) {
1447                 int i, nr_pages;
1448                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1449                                 PAGECACHE_TAG_DIRTY,
1450                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1451                 if (nr_pages == 0)
1452                         break;
1453
1454                 for (i = 0; i < nr_pages; i++) {
1455                         struct page *page = pvec.pages[i];
1456                         bool submitted = false;
1457
1458                         if (unlikely(f2fs_cp_error(sbi))) {
1459                                 f2fs_put_page(last_page, 0);
1460                                 pagevec_release(&pvec);
1461                                 ret = -EIO;
1462                                 goto out;
1463                         }
1464
1465                         if (!IS_DNODE(page) || !is_cold_node(page))
1466                                 continue;
1467                         if (ino_of_node(page) != ino)
1468                                 continue;
1469
1470                         lock_page(page);
1471
1472                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1473 continue_unlock:
1474                                 unlock_page(page);
1475                                 continue;
1476                         }
1477                         if (ino_of_node(page) != ino)
1478                                 goto continue_unlock;
1479
1480                         if (!PageDirty(page) && page != last_page) {
1481                                 /* someone wrote it for us */
1482                                 goto continue_unlock;
1483                         }
1484
1485                         f2fs_wait_on_page_writeback(page, NODE, true);
1486                         BUG_ON(PageWriteback(page));
1487
1488                         set_fsync_mark(page, 0);
1489                         set_dentry_mark(page, 0);
1490
1491                         if (!atomic || page == last_page) {
1492                                 set_fsync_mark(page, 1);
1493                                 if (IS_INODE(page)) {
1494                                         if (is_inode_flag_set(inode,
1495                                                                 FI_DIRTY_INODE))
1496                                                 update_inode(inode, page);
1497                                         set_dentry_mark(page,
1498                                                 need_dentry_mark(sbi, ino));
1499                                 }
1500                                 /*  may be written by other thread */
1501                                 if (!PageDirty(page))
1502                                         set_page_dirty(page);
1503                         }
1504
1505                         if (!clear_page_dirty_for_io(page))
1506                                 goto continue_unlock;
1507
1508                         ret = __write_node_page(page, atomic &&
1509                                                 page == last_page,
1510                                                 &submitted, wbc, true,
1511                                                 FS_NODE_IO);
1512                         if (ret) {
1513                                 unlock_page(page);
1514                                 f2fs_put_page(last_page, 0);
1515                                 break;
1516                         } else if (submitted) {
1517                                 last_idx = page->index;
1518                         }
1519
1520                         if (page == last_page) {
1521                                 f2fs_put_page(page, 0);
1522                                 marked = true;
1523                                 break;
1524                         }
1525                 }
1526                 pagevec_release(&pvec);
1527                 cond_resched();
1528
1529                 if (ret || marked)
1530                         break;
1531         }
1532         if (!ret && atomic && !marked) {
1533                 f2fs_msg(sbi->sb, KERN_DEBUG,
1534                         "Retry to write fsync mark: ino=%u, idx=%lx",
1535                                         ino, last_page->index);
1536                 lock_page(last_page);
1537                 f2fs_wait_on_page_writeback(last_page, NODE, true);
1538                 set_page_dirty(last_page);
1539                 unlock_page(last_page);
1540                 goto retry;
1541         }
1542 out:
1543         if (last_idx != ULONG_MAX)
1544                 f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1545         return ret ? -EIO: 0;
1546 }
1547
1548 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc,
1549                                 bool do_balance, enum iostat_type io_type)
1550 {
1551         pgoff_t index, end;
1552         struct pagevec pvec;
1553         int step = 0;
1554         int nwritten = 0;
1555         int ret = 0;
1556
1557         pagevec_init(&pvec, 0);
1558
1559 next_step:
1560         index = 0;
1561         end = ULONG_MAX;
1562
1563         while (index <= end) {
1564                 int i, nr_pages;
1565                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1566                                 PAGECACHE_TAG_DIRTY,
1567                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1568                 if (nr_pages == 0)
1569                         break;
1570
1571                 for (i = 0; i < nr_pages; i++) {
1572                         struct page *page = pvec.pages[i];
1573                         bool submitted = false;
1574
1575                         if (unlikely(f2fs_cp_error(sbi))) {
1576                                 pagevec_release(&pvec);
1577                                 ret = -EIO;
1578                                 goto out;
1579                         }
1580
1581                         /*
1582                          * flushing sequence with step:
1583                          * 0. indirect nodes
1584                          * 1. dentry dnodes
1585                          * 2. file dnodes
1586                          */
1587                         if (step == 0 && IS_DNODE(page))
1588                                 continue;
1589                         if (step == 1 && (!IS_DNODE(page) ||
1590                                                 is_cold_node(page)))
1591                                 continue;
1592                         if (step == 2 && (!IS_DNODE(page) ||
1593                                                 !is_cold_node(page)))
1594                                 continue;
1595 lock_node:
1596                         if (!trylock_page(page))
1597                                 continue;
1598
1599                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1600 continue_unlock:
1601                                 unlock_page(page);
1602                                 continue;
1603                         }
1604
1605                         if (!PageDirty(page)) {
1606                                 /* someone wrote it for us */
1607                                 goto continue_unlock;
1608                         }
1609
1610                         /* flush inline_data */
1611                         if (is_inline_node(page)) {
1612                                 clear_inline_node(page);
1613                                 unlock_page(page);
1614                                 flush_inline_data(sbi, ino_of_node(page));
1615                                 goto lock_node;
1616                         }
1617
1618                         f2fs_wait_on_page_writeback(page, NODE, true);
1619
1620                         BUG_ON(PageWriteback(page));
1621                         if (!clear_page_dirty_for_io(page))
1622                                 goto continue_unlock;
1623
1624                         set_fsync_mark(page, 0);
1625                         set_dentry_mark(page, 0);
1626
1627                         ret = __write_node_page(page, false, &submitted,
1628                                                 wbc, do_balance, io_type);
1629                         if (ret)
1630                                 unlock_page(page);
1631                         else if (submitted)
1632                                 nwritten++;
1633
1634                         if (--wbc->nr_to_write == 0)
1635                                 break;
1636                 }
1637                 pagevec_release(&pvec);
1638                 cond_resched();
1639
1640                 if (wbc->nr_to_write == 0) {
1641                         step = 2;
1642                         break;
1643                 }
1644         }
1645
1646         if (step < 2) {
1647                 step++;
1648                 goto next_step;
1649         }
1650 out:
1651         if (nwritten)
1652                 f2fs_submit_merged_write(sbi, NODE);
1653         return ret;
1654 }
1655
1656 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1657 {
1658         pgoff_t index = 0, end = ULONG_MAX;
1659         struct pagevec pvec;
1660         int ret2, ret = 0;
1661
1662         pagevec_init(&pvec, 0);
1663
1664         while (index <= end) {
1665                 int i, nr_pages;
1666                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1667                                 PAGECACHE_TAG_WRITEBACK,
1668                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1669                 if (nr_pages == 0)
1670                         break;
1671
1672                 for (i = 0; i < nr_pages; i++) {
1673                         struct page *page = pvec.pages[i];
1674
1675                         /* until radix tree lookup accepts end_index */
1676                         if (unlikely(page->index > end))
1677                                 continue;
1678
1679                         if (ino && ino_of_node(page) == ino) {
1680                                 f2fs_wait_on_page_writeback(page, NODE, true);
1681                                 if (TestClearPageError(page))
1682                                         ret = -EIO;
1683                         }
1684                 }
1685                 pagevec_release(&pvec);
1686                 cond_resched();
1687         }
1688
1689         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1690         if (!ret)
1691                 ret = ret2;
1692         return ret;
1693 }
1694
1695 static int f2fs_write_node_pages(struct address_space *mapping,
1696                             struct writeback_control *wbc)
1697 {
1698         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1699         struct blk_plug plug;
1700         long diff;
1701
1702         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1703                 goto skip_write;
1704
1705         /* balancing f2fs's metadata in background */
1706         f2fs_balance_fs_bg(sbi);
1707
1708         /* collect a number of dirty node pages and write together */
1709         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1710                 goto skip_write;
1711
1712         trace_f2fs_writepages(mapping->host, wbc, NODE);
1713
1714         diff = nr_pages_to_write(sbi, NODE, wbc);
1715         wbc->sync_mode = WB_SYNC_NONE;
1716         blk_start_plug(&plug);
1717         sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1718         blk_finish_plug(&plug);
1719         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1720         return 0;
1721
1722 skip_write:
1723         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1724         trace_f2fs_writepages(mapping->host, wbc, NODE);
1725         return 0;
1726 }
1727
1728 static int f2fs_set_node_page_dirty(struct page *page)
1729 {
1730         trace_f2fs_set_page_dirty(page, NODE);
1731
1732         if (!PageUptodate(page))
1733                 SetPageUptodate(page);
1734         if (!PageDirty(page)) {
1735                 f2fs_set_page_dirty_nobuffers(page);
1736                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1737                 SetPagePrivate(page);
1738                 f2fs_trace_pid(page);
1739                 return 1;
1740         }
1741         return 0;
1742 }
1743
1744 /*
1745  * Structure of the f2fs node operations
1746  */
1747 const struct address_space_operations f2fs_node_aops = {
1748         .writepage      = f2fs_write_node_page,
1749         .writepages     = f2fs_write_node_pages,
1750         .set_page_dirty = f2fs_set_node_page_dirty,
1751         .invalidatepage = f2fs_invalidate_page,
1752         .releasepage    = f2fs_release_page,
1753 #ifdef CONFIG_MIGRATION
1754         .migratepage    = f2fs_migrate_page,
1755 #endif
1756 };
1757
1758 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1759                                                 nid_t n)
1760 {
1761         return radix_tree_lookup(&nm_i->free_nid_root, n);
1762 }
1763
1764 static int __insert_free_nid(struct f2fs_sb_info *sbi,
1765                         struct free_nid *i, enum nid_state state, bool new)
1766 {
1767         struct f2fs_nm_info *nm_i = NM_I(sbi);
1768
1769         if (new) {
1770                 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1771                 if (err)
1772                         return err;
1773         }
1774
1775         f2fs_bug_on(sbi, state != i->state);
1776         nm_i->nid_cnt[state]++;
1777         if (state == FREE_NID)
1778                 list_add_tail(&i->list, &nm_i->free_nid_list);
1779         return 0;
1780 }
1781
1782 static void __remove_free_nid(struct f2fs_sb_info *sbi,
1783                         struct free_nid *i, enum nid_state state, bool reuse)
1784 {
1785         struct f2fs_nm_info *nm_i = NM_I(sbi);
1786
1787         f2fs_bug_on(sbi, state != i->state);
1788         nm_i->nid_cnt[state]--;
1789         if (state == FREE_NID)
1790                 list_del(&i->list);
1791         if (!reuse)
1792                 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1793 }
1794
1795 /* return if the nid is recognized as free */
1796 static bool add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1797 {
1798         struct f2fs_nm_info *nm_i = NM_I(sbi);
1799         struct free_nid *i, *e;
1800         struct nat_entry *ne;
1801         int err = -EINVAL;
1802         bool ret = false;
1803
1804         /* 0 nid should not be used */
1805         if (unlikely(nid == 0))
1806                 return false;
1807
1808         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1809         i->nid = nid;
1810         i->state = FREE_NID;
1811
1812         if (radix_tree_preload(GFP_NOFS))
1813                 goto err;
1814
1815         spin_lock(&nm_i->nid_list_lock);
1816
1817         if (build) {
1818                 /*
1819                  *   Thread A             Thread B
1820                  *  - f2fs_create
1821                  *   - f2fs_new_inode
1822                  *    - alloc_nid
1823                  *     - __insert_nid_to_list(PREALLOC_NID)
1824                  *                     - f2fs_balance_fs_bg
1825                  *                      - build_free_nids
1826                  *                       - __build_free_nids
1827                  *                        - scan_nat_page
1828                  *                         - add_free_nid
1829                  *                          - __lookup_nat_cache
1830                  *  - f2fs_add_link
1831                  *   - init_inode_metadata
1832                  *    - new_inode_page
1833                  *     - new_node_page
1834                  *      - set_node_addr
1835                  *  - alloc_nid_done
1836                  *   - __remove_nid_from_list(PREALLOC_NID)
1837                  *                         - __insert_nid_to_list(FREE_NID)
1838                  */
1839                 ne = __lookup_nat_cache(nm_i, nid);
1840                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1841                                 nat_get_blkaddr(ne) != NULL_ADDR))
1842                         goto err_out;
1843
1844                 e = __lookup_free_nid_list(nm_i, nid);
1845                 if (e) {
1846                         if (e->state == FREE_NID)
1847                                 ret = true;
1848                         goto err_out;
1849                 }
1850         }
1851         ret = true;
1852         err = __insert_free_nid(sbi, i, FREE_NID, true);
1853 err_out:
1854         spin_unlock(&nm_i->nid_list_lock);
1855         radix_tree_preload_end();
1856 err:
1857         if (err)
1858                 kmem_cache_free(free_nid_slab, i);
1859         return ret;
1860 }
1861
1862 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1863 {
1864         struct f2fs_nm_info *nm_i = NM_I(sbi);
1865         struct free_nid *i;
1866         bool need_free = false;
1867
1868         spin_lock(&nm_i->nid_list_lock);
1869         i = __lookup_free_nid_list(nm_i, nid);
1870         if (i && i->state == FREE_NID) {
1871                 __remove_free_nid(sbi, i, FREE_NID, false);
1872                 need_free = true;
1873         }
1874         spin_unlock(&nm_i->nid_list_lock);
1875
1876         if (need_free)
1877                 kmem_cache_free(free_nid_slab, i);
1878 }
1879
1880 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
1881                                                         bool set, bool build)
1882 {
1883         struct f2fs_nm_info *nm_i = NM_I(sbi);
1884         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
1885         unsigned int nid_ofs = nid - START_NID(nid);
1886
1887         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1888                 return;
1889
1890         if (set)
1891                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1892         else
1893                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1894
1895         if (set)
1896                 nm_i->free_nid_count[nat_ofs]++;
1897         else if (!build)
1898                 nm_i->free_nid_count[nat_ofs]--;
1899 }
1900
1901 static void scan_nat_page(struct f2fs_sb_info *sbi,
1902                         struct page *nat_page, nid_t start_nid)
1903 {
1904         struct f2fs_nm_info *nm_i = NM_I(sbi);
1905         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1906         block_t blk_addr;
1907         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
1908         int i;
1909
1910         if (test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1911                 return;
1912
1913         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
1914
1915         i = start_nid % NAT_ENTRY_PER_BLOCK;
1916
1917         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1918                 bool freed = false;
1919
1920                 if (unlikely(start_nid >= nm_i->max_nid))
1921                         break;
1922
1923                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1924                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1925                 if (blk_addr == NULL_ADDR)
1926                         freed = add_free_nid(sbi, start_nid, true);
1927                 spin_lock(&NM_I(sbi)->nid_list_lock);
1928                 update_free_nid_bitmap(sbi, start_nid, freed, true);
1929                 spin_unlock(&NM_I(sbi)->nid_list_lock);
1930         }
1931 }
1932
1933 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
1934 {
1935         struct f2fs_nm_info *nm_i = NM_I(sbi);
1936         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1937         struct f2fs_journal *journal = curseg->journal;
1938         unsigned int i, idx;
1939
1940         down_read(&nm_i->nat_tree_lock);
1941
1942         for (i = 0; i < nm_i->nat_blocks; i++) {
1943                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
1944                         continue;
1945                 if (!nm_i->free_nid_count[i])
1946                         continue;
1947                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
1948                         nid_t nid;
1949
1950                         if (!test_bit_le(idx, nm_i->free_nid_bitmap[i]))
1951                                 continue;
1952
1953                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
1954                         add_free_nid(sbi, nid, true);
1955
1956                         if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
1957                                 goto out;
1958                 }
1959         }
1960 out:
1961         down_read(&curseg->journal_rwsem);
1962         for (i = 0; i < nats_in_cursum(journal); i++) {
1963                 block_t addr;
1964                 nid_t nid;
1965
1966                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1967                 nid = le32_to_cpu(nid_in_journal(journal, i));
1968                 if (addr == NULL_ADDR)
1969                         add_free_nid(sbi, nid, true);
1970                 else
1971                         remove_free_nid(sbi, nid);
1972         }
1973         up_read(&curseg->journal_rwsem);
1974         up_read(&nm_i->nat_tree_lock);
1975 }
1976
1977 static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
1978 {
1979         struct f2fs_nm_info *nm_i = NM_I(sbi);
1980         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1981         struct f2fs_journal *journal = curseg->journal;
1982         int i = 0;
1983         nid_t nid = nm_i->next_scan_nid;
1984
1985         if (unlikely(nid >= nm_i->max_nid))
1986                 nid = 0;
1987
1988         /* Enough entries */
1989         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
1990                 return;
1991
1992         if (!sync && !available_free_memory(sbi, FREE_NIDS))
1993                 return;
1994
1995         if (!mount) {
1996                 /* try to find free nids in free_nid_bitmap */
1997                 scan_free_nid_bits(sbi);
1998
1999                 if (nm_i->nid_cnt[FREE_NID])
2000                         return;
2001         }
2002
2003         /* readahead nat pages to be scanned */
2004         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2005                                                         META_NAT, true);
2006
2007         down_read(&nm_i->nat_tree_lock);
2008
2009         while (1) {
2010                 struct page *page = get_current_nat_page(sbi, nid);
2011
2012                 scan_nat_page(sbi, page, nid);
2013                 f2fs_put_page(page, 1);
2014
2015                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2016                 if (unlikely(nid >= nm_i->max_nid))
2017                         nid = 0;
2018
2019                 if (++i >= FREE_NID_PAGES)
2020                         break;
2021         }
2022
2023         /* go to the next free nat pages to find free nids abundantly */
2024         nm_i->next_scan_nid = nid;
2025
2026         /* find free nids from current sum_pages */
2027         down_read(&curseg->journal_rwsem);
2028         for (i = 0; i < nats_in_cursum(journal); i++) {
2029                 block_t addr;
2030
2031                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2032                 nid = le32_to_cpu(nid_in_journal(journal, i));
2033                 if (addr == NULL_ADDR)
2034                         add_free_nid(sbi, nid, true);
2035                 else
2036                         remove_free_nid(sbi, nid);
2037         }
2038         up_read(&curseg->journal_rwsem);
2039         up_read(&nm_i->nat_tree_lock);
2040
2041         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2042                                         nm_i->ra_nid_pages, META_NAT, false);
2043 }
2044
2045 void build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2046 {
2047         mutex_lock(&NM_I(sbi)->build_lock);
2048         __build_free_nids(sbi, sync, mount);
2049         mutex_unlock(&NM_I(sbi)->build_lock);
2050 }
2051
2052 /*
2053  * If this function returns success, caller can obtain a new nid
2054  * from second parameter of this function.
2055  * The returned nid could be used ino as well as nid when inode is created.
2056  */
2057 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2058 {
2059         struct f2fs_nm_info *nm_i = NM_I(sbi);
2060         struct free_nid *i = NULL;
2061 retry:
2062 #ifdef CONFIG_F2FS_FAULT_INJECTION
2063         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2064                 f2fs_show_injection_info(FAULT_ALLOC_NID);
2065                 return false;
2066         }
2067 #endif
2068         spin_lock(&nm_i->nid_list_lock);
2069
2070         if (unlikely(nm_i->available_nids == 0)) {
2071                 spin_unlock(&nm_i->nid_list_lock);
2072                 return false;
2073         }
2074
2075         /* We should not use stale free nids created by build_free_nids */
2076         if (nm_i->nid_cnt[FREE_NID] && !on_build_free_nids(nm_i)) {
2077                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2078                 i = list_first_entry(&nm_i->free_nid_list,
2079                                         struct free_nid, list);
2080                 *nid = i->nid;
2081
2082                 __remove_free_nid(sbi, i, FREE_NID, true);
2083                 i->state = PREALLOC_NID;
2084                 __insert_free_nid(sbi, i, PREALLOC_NID, false);
2085                 nm_i->available_nids--;
2086
2087                 update_free_nid_bitmap(sbi, *nid, false, false);
2088
2089                 spin_unlock(&nm_i->nid_list_lock);
2090                 return true;
2091         }
2092         spin_unlock(&nm_i->nid_list_lock);
2093
2094         /* Let's scan nat pages and its caches to get free nids */
2095         build_free_nids(sbi, true, false);
2096         goto retry;
2097 }
2098
2099 /*
2100  * alloc_nid() should be called prior to this function.
2101  */
2102 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2103 {
2104         struct f2fs_nm_info *nm_i = NM_I(sbi);
2105         struct free_nid *i;
2106
2107         spin_lock(&nm_i->nid_list_lock);
2108         i = __lookup_free_nid_list(nm_i, nid);
2109         f2fs_bug_on(sbi, !i);
2110         __remove_free_nid(sbi, i, PREALLOC_NID, false);
2111         spin_unlock(&nm_i->nid_list_lock);
2112
2113         kmem_cache_free(free_nid_slab, i);
2114 }
2115
2116 /*
2117  * alloc_nid() should be called prior to this function.
2118  */
2119 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2120 {
2121         struct f2fs_nm_info *nm_i = NM_I(sbi);
2122         struct free_nid *i;
2123         bool need_free = false;
2124
2125         if (!nid)
2126                 return;
2127
2128         spin_lock(&nm_i->nid_list_lock);
2129         i = __lookup_free_nid_list(nm_i, nid);
2130         f2fs_bug_on(sbi, !i);
2131
2132         if (!available_free_memory(sbi, FREE_NIDS)) {
2133                 __remove_free_nid(sbi, i, PREALLOC_NID, false);
2134                 need_free = true;
2135         } else {
2136                 __remove_free_nid(sbi, i, PREALLOC_NID, true);
2137                 i->state = FREE_NID;
2138                 __insert_free_nid(sbi, i, FREE_NID, false);
2139         }
2140
2141         nm_i->available_nids++;
2142
2143         update_free_nid_bitmap(sbi, nid, true, false);
2144
2145         spin_unlock(&nm_i->nid_list_lock);
2146
2147         if (need_free)
2148                 kmem_cache_free(free_nid_slab, i);
2149 }
2150
2151 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2152 {
2153         struct f2fs_nm_info *nm_i = NM_I(sbi);
2154         struct free_nid *i, *next;
2155         int nr = nr_shrink;
2156
2157         if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2158                 return 0;
2159
2160         if (!mutex_trylock(&nm_i->build_lock))
2161                 return 0;
2162
2163         spin_lock(&nm_i->nid_list_lock);
2164         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2165                 if (nr_shrink <= 0 ||
2166                                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2167                         break;
2168
2169                 __remove_free_nid(sbi, i, FREE_NID, false);
2170                 kmem_cache_free(free_nid_slab, i);
2171                 nr_shrink--;
2172         }
2173         spin_unlock(&nm_i->nid_list_lock);
2174         mutex_unlock(&nm_i->build_lock);
2175
2176         return nr - nr_shrink;
2177 }
2178
2179 void recover_inline_xattr(struct inode *inode, struct page *page)
2180 {
2181         void *src_addr, *dst_addr;
2182         size_t inline_size;
2183         struct page *ipage;
2184         struct f2fs_inode *ri;
2185
2186         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2187         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2188
2189         ri = F2FS_INODE(page);
2190         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2191                 clear_inode_flag(inode, FI_INLINE_XATTR);
2192                 goto update_inode;
2193         }
2194
2195         dst_addr = inline_xattr_addr(inode, ipage);
2196         src_addr = inline_xattr_addr(inode, page);
2197         inline_size = inline_xattr_size(inode);
2198
2199         f2fs_wait_on_page_writeback(ipage, NODE, true);
2200         memcpy(dst_addr, src_addr, inline_size);
2201 update_inode:
2202         update_inode(inode, ipage);
2203         f2fs_put_page(ipage, 1);
2204 }
2205
2206 int recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2207 {
2208         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2209         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2210         nid_t new_xnid;
2211         struct dnode_of_data dn;
2212         struct node_info ni;
2213         struct page *xpage;
2214
2215         if (!prev_xnid)
2216                 goto recover_xnid;
2217
2218         /* 1: invalidate the previous xattr nid */
2219         get_node_info(sbi, prev_xnid, &ni);
2220         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2221         invalidate_blocks(sbi, ni.blk_addr);
2222         dec_valid_node_count(sbi, inode, false);
2223         set_node_addr(sbi, &ni, NULL_ADDR, false);
2224
2225 recover_xnid:
2226         /* 2: update xattr nid in inode */
2227         if (!alloc_nid(sbi, &new_xnid))
2228                 return -ENOSPC;
2229
2230         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2231         xpage = new_node_page(&dn, XATTR_NODE_OFFSET);
2232         if (IS_ERR(xpage)) {
2233                 alloc_nid_failed(sbi, new_xnid);
2234                 return PTR_ERR(xpage);
2235         }
2236
2237         alloc_nid_done(sbi, new_xnid);
2238         update_inode_page(inode);
2239
2240         /* 3: update and set xattr node page dirty */
2241         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2242
2243         set_page_dirty(xpage);
2244         f2fs_put_page(xpage, 1);
2245
2246         return 0;
2247 }
2248
2249 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2250 {
2251         struct f2fs_inode *src, *dst;
2252         nid_t ino = ino_of_node(page);
2253         struct node_info old_ni, new_ni;
2254         struct page *ipage;
2255
2256         get_node_info(sbi, ino, &old_ni);
2257
2258         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2259                 return -EINVAL;
2260 retry:
2261         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2262         if (!ipage) {
2263                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2264                 goto retry;
2265         }
2266
2267         /* Should not use this inode from free nid list */
2268         remove_free_nid(sbi, ino);
2269
2270         if (!PageUptodate(ipage))
2271                 SetPageUptodate(ipage);
2272         fill_node_footer(ipage, ino, ino, 0, true);
2273
2274         src = F2FS_INODE(page);
2275         dst = F2FS_INODE(ipage);
2276
2277         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2278         dst->i_size = 0;
2279         dst->i_blocks = cpu_to_le64(1);
2280         dst->i_links = cpu_to_le32(1);
2281         dst->i_xattr_nid = 0;
2282         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2283         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2284                 dst->i_extra_isize = src->i_extra_isize;
2285
2286                 if (f2fs_sb_has_flexible_inline_xattr(sbi->sb) &&
2287                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2288                                                         i_inline_xattr_size))
2289                         dst->i_inline_xattr_size = src->i_inline_xattr_size;
2290
2291                 if (f2fs_sb_has_project_quota(sbi->sb) &&
2292                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2293                                                                 i_projid))
2294                         dst->i_projid = src->i_projid;
2295         }
2296
2297         new_ni = old_ni;
2298         new_ni.ino = ino;
2299
2300         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2301                 WARN_ON(1);
2302         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2303         inc_valid_inode_count(sbi);
2304         set_page_dirty(ipage);
2305         f2fs_put_page(ipage, 1);
2306         return 0;
2307 }
2308
2309 int restore_node_summary(struct f2fs_sb_info *sbi,
2310                         unsigned int segno, struct f2fs_summary_block *sum)
2311 {
2312         struct f2fs_node *rn;
2313         struct f2fs_summary *sum_entry;
2314         block_t addr;
2315         int i, idx, last_offset, nrpages;
2316
2317         /* scan the node segment */
2318         last_offset = sbi->blocks_per_seg;
2319         addr = START_BLOCK(sbi, segno);
2320         sum_entry = &sum->entries[0];
2321
2322         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2323                 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2324
2325                 /* readahead node pages */
2326                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2327
2328                 for (idx = addr; idx < addr + nrpages; idx++) {
2329                         struct page *page = get_tmp_page(sbi, idx);
2330
2331                         rn = F2FS_NODE(page);
2332                         sum_entry->nid = rn->footer.nid;
2333                         sum_entry->version = 0;
2334                         sum_entry->ofs_in_node = 0;
2335                         sum_entry++;
2336                         f2fs_put_page(page, 1);
2337                 }
2338
2339                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2340                                                         addr + nrpages);
2341         }
2342         return 0;
2343 }
2344
2345 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2346 {
2347         struct f2fs_nm_info *nm_i = NM_I(sbi);
2348         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2349         struct f2fs_journal *journal = curseg->journal;
2350         int i;
2351
2352         down_write(&curseg->journal_rwsem);
2353         for (i = 0; i < nats_in_cursum(journal); i++) {
2354                 struct nat_entry *ne;
2355                 struct f2fs_nat_entry raw_ne;
2356                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2357
2358                 raw_ne = nat_in_journal(journal, i);
2359
2360                 ne = __lookup_nat_cache(nm_i, nid);
2361                 if (!ne) {
2362                         ne = grab_nat_entry(nm_i, nid, true);
2363                         node_info_from_raw_nat(&ne->ni, &raw_ne);
2364                 }
2365
2366                 /*
2367                  * if a free nat in journal has not been used after last
2368                  * checkpoint, we should remove it from available nids,
2369                  * since later we will add it again.
2370                  */
2371                 if (!get_nat_flag(ne, IS_DIRTY) &&
2372                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2373                         spin_lock(&nm_i->nid_list_lock);
2374                         nm_i->available_nids--;
2375                         spin_unlock(&nm_i->nid_list_lock);
2376                 }
2377
2378                 __set_nat_cache_dirty(nm_i, ne);
2379         }
2380         update_nats_in_cursum(journal, -i);
2381         up_write(&curseg->journal_rwsem);
2382 }
2383
2384 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2385                                                 struct list_head *head, int max)
2386 {
2387         struct nat_entry_set *cur;
2388
2389         if (nes->entry_cnt >= max)
2390                 goto add_out;
2391
2392         list_for_each_entry(cur, head, set_list) {
2393                 if (cur->entry_cnt >= nes->entry_cnt) {
2394                         list_add(&nes->set_list, cur->set_list.prev);
2395                         return;
2396                 }
2397         }
2398 add_out:
2399         list_add_tail(&nes->set_list, head);
2400 }
2401
2402 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2403                                                 struct page *page)
2404 {
2405         struct f2fs_nm_info *nm_i = NM_I(sbi);
2406         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2407         struct f2fs_nat_block *nat_blk = page_address(page);
2408         int valid = 0;
2409         int i;
2410
2411         if (!enabled_nat_bits(sbi, NULL))
2412                 return;
2413
2414         for (i = 0; i < NAT_ENTRY_PER_BLOCK; i++) {
2415                 if (start_nid == 0 && i == 0)
2416                         valid++;
2417                 if (nat_blk->entries[i].block_addr)
2418                         valid++;
2419         }
2420         if (valid == 0) {
2421                 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2422                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2423                 return;
2424         }
2425
2426         __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2427         if (valid == NAT_ENTRY_PER_BLOCK)
2428                 __set_bit_le(nat_index, nm_i->full_nat_bits);
2429         else
2430                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2431 }
2432
2433 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2434                 struct nat_entry_set *set, struct cp_control *cpc)
2435 {
2436         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2437         struct f2fs_journal *journal = curseg->journal;
2438         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2439         bool to_journal = true;
2440         struct f2fs_nat_block *nat_blk;
2441         struct nat_entry *ne, *cur;
2442         struct page *page = NULL;
2443
2444         /*
2445          * there are two steps to flush nat entries:
2446          * #1, flush nat entries to journal in current hot data summary block.
2447          * #2, flush nat entries to nat page.
2448          */
2449         if (enabled_nat_bits(sbi, cpc) ||
2450                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2451                 to_journal = false;
2452
2453         if (to_journal) {
2454                 down_write(&curseg->journal_rwsem);
2455         } else {
2456                 page = get_next_nat_page(sbi, start_nid);
2457                 nat_blk = page_address(page);
2458                 f2fs_bug_on(sbi, !nat_blk);
2459         }
2460
2461         /* flush dirty nats in nat entry set */
2462         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2463                 struct f2fs_nat_entry *raw_ne;
2464                 nid_t nid = nat_get_nid(ne);
2465                 int offset;
2466
2467                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2468
2469                 if (to_journal) {
2470                         offset = lookup_journal_in_cursum(journal,
2471                                                         NAT_JOURNAL, nid, 1);
2472                         f2fs_bug_on(sbi, offset < 0);
2473                         raw_ne = &nat_in_journal(journal, offset);
2474                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2475                 } else {
2476                         raw_ne = &nat_blk->entries[nid - start_nid];
2477                 }
2478                 raw_nat_from_node_info(raw_ne, &ne->ni);
2479                 nat_reset_flag(ne);
2480                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2481                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2482                         add_free_nid(sbi, nid, false);
2483                         spin_lock(&NM_I(sbi)->nid_list_lock);
2484                         NM_I(sbi)->available_nids++;
2485                         update_free_nid_bitmap(sbi, nid, true, false);
2486                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2487                 } else {
2488                         spin_lock(&NM_I(sbi)->nid_list_lock);
2489                         update_free_nid_bitmap(sbi, nid, false, false);
2490                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2491                 }
2492         }
2493
2494         if (to_journal) {
2495                 up_write(&curseg->journal_rwsem);
2496         } else {
2497                 __update_nat_bits(sbi, start_nid, page);
2498                 f2fs_put_page(page, 1);
2499         }
2500
2501         /* Allow dirty nats by node block allocation in write_begin */
2502         if (!set->entry_cnt) {
2503                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2504                 kmem_cache_free(nat_entry_set_slab, set);
2505         }
2506 }
2507
2508 /*
2509  * This function is called during the checkpointing process.
2510  */
2511 void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2512 {
2513         struct f2fs_nm_info *nm_i = NM_I(sbi);
2514         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2515         struct f2fs_journal *journal = curseg->journal;
2516         struct nat_entry_set *setvec[SETVEC_SIZE];
2517         struct nat_entry_set *set, *tmp;
2518         unsigned int found;
2519         nid_t set_idx = 0;
2520         LIST_HEAD(sets);
2521
2522         if (!nm_i->dirty_nat_cnt)
2523                 return;
2524
2525         down_write(&nm_i->nat_tree_lock);
2526
2527         /*
2528          * if there are no enough space in journal to store dirty nat
2529          * entries, remove all entries from journal and merge them
2530          * into nat entry set.
2531          */
2532         if (enabled_nat_bits(sbi, cpc) ||
2533                 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2534                 remove_nats_in_journal(sbi);
2535
2536         while ((found = __gang_lookup_nat_set(nm_i,
2537                                         set_idx, SETVEC_SIZE, setvec))) {
2538                 unsigned idx;
2539                 set_idx = setvec[found - 1]->set + 1;
2540                 for (idx = 0; idx < found; idx++)
2541                         __adjust_nat_entry_set(setvec[idx], &sets,
2542                                                 MAX_NAT_JENTRIES(journal));
2543         }
2544
2545         /* flush dirty nats in nat entry set */
2546         list_for_each_entry_safe(set, tmp, &sets, set_list)
2547                 __flush_nat_entry_set(sbi, set, cpc);
2548
2549         up_write(&nm_i->nat_tree_lock);
2550         /* Allow dirty nats by node block allocation in write_begin */
2551 }
2552
2553 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2554 {
2555         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2556         struct f2fs_nm_info *nm_i = NM_I(sbi);
2557         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2558         unsigned int i;
2559         __u64 cp_ver = cur_cp_version(ckpt);
2560         block_t nat_bits_addr;
2561
2562         if (!enabled_nat_bits(sbi, NULL))
2563                 return 0;
2564
2565         nm_i->nat_bits_blocks = F2FS_BYTES_TO_BLK((nat_bits_bytes << 1) + 8 +
2566                                                 F2FS_BLKSIZE - 1);
2567         nm_i->nat_bits = kzalloc(nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS,
2568                                                 GFP_KERNEL);
2569         if (!nm_i->nat_bits)
2570                 return -ENOMEM;
2571
2572         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2573                                                 nm_i->nat_bits_blocks;
2574         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2575                 struct page *page = get_meta_page(sbi, nat_bits_addr++);
2576
2577                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2578                                         page_address(page), F2FS_BLKSIZE);
2579                 f2fs_put_page(page, 1);
2580         }
2581
2582         cp_ver |= (cur_cp_crc(ckpt) << 32);
2583         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2584                 disable_nat_bits(sbi, true);
2585                 return 0;
2586         }
2587
2588         nm_i->full_nat_bits = nm_i->nat_bits + 8;
2589         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2590
2591         f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2592         return 0;
2593 }
2594
2595 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2596 {
2597         struct f2fs_nm_info *nm_i = NM_I(sbi);
2598         unsigned int i = 0;
2599         nid_t nid, last_nid;
2600
2601         if (!enabled_nat_bits(sbi, NULL))
2602                 return;
2603
2604         for (i = 0; i < nm_i->nat_blocks; i++) {
2605                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2606                 if (i >= nm_i->nat_blocks)
2607                         break;
2608
2609                 __set_bit_le(i, nm_i->nat_block_bitmap);
2610
2611                 nid = i * NAT_ENTRY_PER_BLOCK;
2612                 last_nid = (i + 1) * NAT_ENTRY_PER_BLOCK;
2613
2614                 spin_lock(&NM_I(sbi)->nid_list_lock);
2615                 for (; nid < last_nid; nid++)
2616                         update_free_nid_bitmap(sbi, nid, true, true);
2617                 spin_unlock(&NM_I(sbi)->nid_list_lock);
2618         }
2619
2620         for (i = 0; i < nm_i->nat_blocks; i++) {
2621                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2622                 if (i >= nm_i->nat_blocks)
2623                         break;
2624
2625                 __set_bit_le(i, nm_i->nat_block_bitmap);
2626         }
2627 }
2628
2629 static int init_node_manager(struct f2fs_sb_info *sbi)
2630 {
2631         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2632         struct f2fs_nm_info *nm_i = NM_I(sbi);
2633         unsigned char *version_bitmap;
2634         unsigned int nat_segs;
2635         int err;
2636
2637         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2638
2639         /* segment_count_nat includes pair segment so divide to 2. */
2640         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2641         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2642         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2643
2644         /* not used nids: 0, node, meta, (and root counted as valid node) */
2645         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2646                                                         F2FS_RESERVED_NODE_NUM;
2647         nm_i->nid_cnt[FREE_NID] = 0;
2648         nm_i->nid_cnt[PREALLOC_NID] = 0;
2649         nm_i->nat_cnt = 0;
2650         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2651         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2652         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2653
2654         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2655         INIT_LIST_HEAD(&nm_i->free_nid_list);
2656         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2657         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2658         INIT_LIST_HEAD(&nm_i->nat_entries);
2659
2660         mutex_init(&nm_i->build_lock);
2661         spin_lock_init(&nm_i->nid_list_lock);
2662         init_rwsem(&nm_i->nat_tree_lock);
2663
2664         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2665         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2666         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2667         if (!version_bitmap)
2668                 return -EFAULT;
2669
2670         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2671                                         GFP_KERNEL);
2672         if (!nm_i->nat_bitmap)
2673                 return -ENOMEM;
2674
2675         err = __get_nat_bitmaps(sbi);
2676         if (err)
2677                 return err;
2678
2679 #ifdef CONFIG_F2FS_CHECK_FS
2680         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2681                                         GFP_KERNEL);
2682         if (!nm_i->nat_bitmap_mir)
2683                 return -ENOMEM;
2684 #endif
2685
2686         return 0;
2687 }
2688
2689 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2690 {
2691         struct f2fs_nm_info *nm_i = NM_I(sbi);
2692
2693         nm_i->free_nid_bitmap = kvzalloc(nm_i->nat_blocks *
2694                                         NAT_ENTRY_BITMAP_SIZE, GFP_KERNEL);
2695         if (!nm_i->free_nid_bitmap)
2696                 return -ENOMEM;
2697
2698         nm_i->nat_block_bitmap = kvzalloc(nm_i->nat_blocks / 8,
2699                                                                 GFP_KERNEL);
2700         if (!nm_i->nat_block_bitmap)
2701                 return -ENOMEM;
2702
2703         nm_i->free_nid_count = kvzalloc(nm_i->nat_blocks *
2704                                         sizeof(unsigned short), GFP_KERNEL);
2705         if (!nm_i->free_nid_count)
2706                 return -ENOMEM;
2707         return 0;
2708 }
2709
2710 int build_node_manager(struct f2fs_sb_info *sbi)
2711 {
2712         int err;
2713
2714         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2715         if (!sbi->nm_info)
2716                 return -ENOMEM;
2717
2718         err = init_node_manager(sbi);
2719         if (err)
2720                 return err;
2721
2722         err = init_free_nid_cache(sbi);
2723         if (err)
2724                 return err;
2725
2726         /* load free nid status from nat_bits table */
2727         load_free_nid_bitmap(sbi);
2728
2729         build_free_nids(sbi, true, true);
2730         return 0;
2731 }
2732
2733 void destroy_node_manager(struct f2fs_sb_info *sbi)
2734 {
2735         struct f2fs_nm_info *nm_i = NM_I(sbi);
2736         struct free_nid *i, *next_i;
2737         struct nat_entry *natvec[NATVEC_SIZE];
2738         struct nat_entry_set *setvec[SETVEC_SIZE];
2739         nid_t nid = 0;
2740         unsigned int found;
2741
2742         if (!nm_i)
2743                 return;
2744
2745         /* destroy free nid list */
2746         spin_lock(&nm_i->nid_list_lock);
2747         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2748                 __remove_free_nid(sbi, i, FREE_NID, false);
2749                 spin_unlock(&nm_i->nid_list_lock);
2750                 kmem_cache_free(free_nid_slab, i);
2751                 spin_lock(&nm_i->nid_list_lock);
2752         }
2753         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
2754         f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
2755         f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
2756         spin_unlock(&nm_i->nid_list_lock);
2757
2758         /* destroy nat cache */
2759         down_write(&nm_i->nat_tree_lock);
2760         while ((found = __gang_lookup_nat_cache(nm_i,
2761                                         nid, NATVEC_SIZE, natvec))) {
2762                 unsigned idx;
2763
2764                 nid = nat_get_nid(natvec[found - 1]) + 1;
2765                 for (idx = 0; idx < found; idx++)
2766                         __del_from_nat_cache(nm_i, natvec[idx]);
2767         }
2768         f2fs_bug_on(sbi, nm_i->nat_cnt);
2769
2770         /* destroy nat set cache */
2771         nid = 0;
2772         while ((found = __gang_lookup_nat_set(nm_i,
2773                                         nid, SETVEC_SIZE, setvec))) {
2774                 unsigned idx;
2775
2776                 nid = setvec[found - 1]->set + 1;
2777                 for (idx = 0; idx < found; idx++) {
2778                         /* entry_cnt is not zero, when cp_error was occurred */
2779                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2780                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2781                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2782                 }
2783         }
2784         up_write(&nm_i->nat_tree_lock);
2785
2786         kvfree(nm_i->nat_block_bitmap);
2787         kvfree(nm_i->free_nid_bitmap);
2788         kvfree(nm_i->free_nid_count);
2789
2790         kfree(nm_i->nat_bitmap);
2791         kfree(nm_i->nat_bits);
2792 #ifdef CONFIG_F2FS_CHECK_FS
2793         kfree(nm_i->nat_bitmap_mir);
2794 #endif
2795         sbi->nm_info = NULL;
2796         kfree(nm_i);
2797 }
2798
2799 int __init create_node_manager_caches(void)
2800 {
2801         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2802                         sizeof(struct nat_entry));
2803         if (!nat_entry_slab)
2804                 goto fail;
2805
2806         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2807                         sizeof(struct free_nid));
2808         if (!free_nid_slab)
2809                 goto destroy_nat_entry;
2810
2811         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2812                         sizeof(struct nat_entry_set));
2813         if (!nat_entry_set_slab)
2814                 goto destroy_free_nid;
2815         return 0;
2816
2817 destroy_free_nid:
2818         kmem_cache_destroy(free_nid_slab);
2819 destroy_nat_entry:
2820         kmem_cache_destroy(nat_entry_slab);
2821 fail:
2822         return -ENOMEM;
2823 }
2824
2825 void destroy_node_manager_caches(void)
2826 {
2827         kmem_cache_destroy(nat_entry_set_slab);
2828         kmem_cache_destroy(free_nid_slab);
2829         kmem_cache_destroy(nat_entry_slab);
2830 }