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 = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1222         if (!page)
1223                 goto iput_out;
1224
1225         if (!PageUptodate(page))
1226                 goto page_out;
1227
1228         if (!PageDirty(page))
1229                 goto page_out;
1230
1231         if (!clear_page_dirty_for_io(page))
1232                 goto page_out;
1233
1234         ret = f2fs_write_inline_data(inode, page);
1235         inode_dec_dirty_pages(inode);
1236         remove_dirty_inode(inode);
1237         if (ret)
1238                 set_page_dirty(page);
1239 page_out:
1240         f2fs_put_page(page, 1);
1241 iput_out:
1242         iput(inode);
1243 }
1244
1245 void move_node_page(struct page *node_page, int gc_type)
1246 {
1247         if (gc_type == FG_GC) {
1248                 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1249                 struct writeback_control wbc = {
1250                         .sync_mode = WB_SYNC_ALL,
1251                         .nr_to_write = 1,
1252                         .for_reclaim = 0,
1253                 };
1254
1255                 set_page_dirty(node_page);
1256                 f2fs_wait_on_page_writeback(node_page, NODE, true);
1257
1258                 f2fs_bug_on(sbi, PageWriteback(node_page));
1259                 if (!clear_page_dirty_for_io(node_page))
1260                         goto out_page;
1261
1262                 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1263                         unlock_page(node_page);
1264                 goto release_page;
1265         } else {
1266                 /* set page dirty and write it */
1267                 if (!PageWriteback(node_page))
1268                         set_page_dirty(node_page);
1269         }
1270 out_page:
1271         unlock_page(node_page);
1272 release_page:
1273         f2fs_put_page(node_page, 0);
1274 }
1275
1276 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1277 {
1278         pgoff_t index, end;
1279         struct pagevec pvec;
1280         struct page *last_page = NULL;
1281
1282         pagevec_init(&pvec, 0);
1283         index = 0;
1284         end = ULONG_MAX;
1285
1286         while (index <= end) {
1287                 int i, nr_pages;
1288                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1289                                 PAGECACHE_TAG_DIRTY,
1290                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1291                 if (nr_pages == 0)
1292                         break;
1293
1294                 for (i = 0; i < nr_pages; i++) {
1295                         struct page *page = pvec.pages[i];
1296
1297                         if (unlikely(f2fs_cp_error(sbi))) {
1298                                 f2fs_put_page(last_page, 0);
1299                                 pagevec_release(&pvec);
1300                                 return ERR_PTR(-EIO);
1301                         }
1302
1303                         if (!IS_DNODE(page) || !is_cold_node(page))
1304                                 continue;
1305                         if (ino_of_node(page) != ino)
1306                                 continue;
1307
1308                         lock_page(page);
1309
1310                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1311 continue_unlock:
1312                                 unlock_page(page);
1313                                 continue;
1314                         }
1315                         if (ino_of_node(page) != ino)
1316                                 goto continue_unlock;
1317
1318                         if (!PageDirty(page)) {
1319                                 /* someone wrote it for us */
1320                                 goto continue_unlock;
1321                         }
1322
1323                         if (last_page)
1324                                 f2fs_put_page(last_page, 0);
1325
1326                         get_page(page);
1327                         last_page = page;
1328                         unlock_page(page);
1329                 }
1330                 pagevec_release(&pvec);
1331                 cond_resched();
1332         }
1333         return last_page;
1334 }
1335
1336 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1337                                 struct writeback_control *wbc, bool do_balance,
1338                                 enum iostat_type io_type)
1339 {
1340         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1341         nid_t nid;
1342         struct node_info ni;
1343         struct f2fs_io_info fio = {
1344                 .sbi = sbi,
1345                 .ino = ino_of_node(page),
1346                 .type = NODE,
1347                 .op = REQ_OP_WRITE,
1348                 .op_flags = wbc_to_write_flags(wbc),
1349                 .page = page,
1350                 .encrypted_page = NULL,
1351                 .submitted = false,
1352                 .io_type = io_type,
1353         };
1354
1355         trace_f2fs_writepage(page, NODE);
1356
1357         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1358                 goto redirty_out;
1359         if (unlikely(f2fs_cp_error(sbi)))
1360                 goto redirty_out;
1361
1362         /* get old block addr of this node page */
1363         nid = nid_of_node(page);
1364         f2fs_bug_on(sbi, page->index != nid);
1365
1366         if (wbc->for_reclaim) {
1367                 if (!down_read_trylock(&sbi->node_write))
1368                         goto redirty_out;
1369         } else {
1370                 down_read(&sbi->node_write);
1371         }
1372
1373         get_node_info(sbi, nid, &ni);
1374
1375         /* This page is already truncated */
1376         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1377                 ClearPageUptodate(page);
1378                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1379                 up_read(&sbi->node_write);
1380                 unlock_page(page);
1381                 return 0;
1382         }
1383
1384         if (atomic && !test_opt(sbi, NOBARRIER))
1385                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1386
1387         set_page_writeback(page);
1388         fio.old_blkaddr = ni.blk_addr;
1389         write_node_page(nid, &fio);
1390         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1391         dec_page_count(sbi, F2FS_DIRTY_NODES);
1392         up_read(&sbi->node_write);
1393
1394         if (wbc->for_reclaim) {
1395                 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1396                                                 page->index, NODE);
1397                 submitted = NULL;
1398         }
1399
1400         unlock_page(page);
1401
1402         if (unlikely(f2fs_cp_error(sbi))) {
1403                 f2fs_submit_merged_write(sbi, NODE);
1404                 submitted = NULL;
1405         }
1406         if (submitted)
1407                 *submitted = fio.submitted;
1408
1409         if (do_balance)
1410                 f2fs_balance_fs(sbi, false);
1411         return 0;
1412
1413 redirty_out:
1414         redirty_page_for_writepage(wbc, page);
1415         return AOP_WRITEPAGE_ACTIVATE;
1416 }
1417
1418 static int f2fs_write_node_page(struct page *page,
1419                                 struct writeback_control *wbc)
1420 {
1421         return __write_node_page(page, false, NULL, wbc, false, FS_NODE_IO);
1422 }
1423
1424 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1425                         struct writeback_control *wbc, bool atomic)
1426 {
1427         pgoff_t index, end;
1428         pgoff_t last_idx = ULONG_MAX;
1429         struct pagevec pvec;
1430         int ret = 0;
1431         struct page *last_page = NULL;
1432         bool marked = false;
1433         nid_t ino = inode->i_ino;
1434
1435         if (atomic) {
1436                 last_page = last_fsync_dnode(sbi, ino);
1437                 if (IS_ERR_OR_NULL(last_page))
1438                         return PTR_ERR_OR_ZERO(last_page);
1439         }
1440 retry:
1441         pagevec_init(&pvec, 0);
1442         index = 0;
1443         end = ULONG_MAX;
1444
1445         while (index <= end) {
1446                 int i, nr_pages;
1447                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1448                                 PAGECACHE_TAG_DIRTY,
1449                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1450                 if (nr_pages == 0)
1451                         break;
1452
1453                 for (i = 0; i < nr_pages; i++) {
1454                         struct page *page = pvec.pages[i];
1455                         bool submitted = false;
1456
1457                         if (unlikely(f2fs_cp_error(sbi))) {
1458                                 f2fs_put_page(last_page, 0);
1459                                 pagevec_release(&pvec);
1460                                 ret = -EIO;
1461                                 goto out;
1462                         }
1463
1464                         if (!IS_DNODE(page) || !is_cold_node(page))
1465                                 continue;
1466                         if (ino_of_node(page) != ino)
1467                                 continue;
1468
1469                         lock_page(page);
1470
1471                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1472 continue_unlock:
1473                                 unlock_page(page);
1474                                 continue;
1475                         }
1476                         if (ino_of_node(page) != ino)
1477                                 goto continue_unlock;
1478
1479                         if (!PageDirty(page) && page != last_page) {
1480                                 /* someone wrote it for us */
1481                                 goto continue_unlock;
1482                         }
1483
1484                         f2fs_wait_on_page_writeback(page, NODE, true);
1485                         BUG_ON(PageWriteback(page));
1486
1487                         set_fsync_mark(page, 0);
1488                         set_dentry_mark(page, 0);
1489
1490                         if (!atomic || page == last_page) {
1491                                 set_fsync_mark(page, 1);
1492                                 if (IS_INODE(page)) {
1493                                         if (is_inode_flag_set(inode,
1494                                                                 FI_DIRTY_INODE))
1495                                                 update_inode(inode, page);
1496                                         set_dentry_mark(page,
1497                                                 need_dentry_mark(sbi, ino));
1498                                 }
1499                                 /*  may be written by other thread */
1500                                 if (!PageDirty(page))
1501                                         set_page_dirty(page);
1502                         }
1503
1504                         if (!clear_page_dirty_for_io(page))
1505                                 goto continue_unlock;
1506
1507                         ret = __write_node_page(page, atomic &&
1508                                                 page == last_page,
1509                                                 &submitted, wbc, true,
1510                                                 FS_NODE_IO);
1511                         if (ret) {
1512                                 unlock_page(page);
1513                                 f2fs_put_page(last_page, 0);
1514                                 break;
1515                         } else if (submitted) {
1516                                 last_idx = page->index;
1517                         }
1518
1519                         if (page == last_page) {
1520                                 f2fs_put_page(page, 0);
1521                                 marked = true;
1522                                 break;
1523                         }
1524                 }
1525                 pagevec_release(&pvec);
1526                 cond_resched();
1527
1528                 if (ret || marked)
1529                         break;
1530         }
1531         if (!ret && atomic && !marked) {
1532                 f2fs_msg(sbi->sb, KERN_DEBUG,
1533                         "Retry to write fsync mark: ino=%u, idx=%lx",
1534                                         ino, last_page->index);
1535                 lock_page(last_page);
1536                 f2fs_wait_on_page_writeback(last_page, NODE, true);
1537                 set_page_dirty(last_page);
1538                 unlock_page(last_page);
1539                 goto retry;
1540         }
1541 out:
1542         if (last_idx != ULONG_MAX)
1543                 f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1544         return ret ? -EIO: 0;
1545 }
1546
1547 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc,
1548                                 bool do_balance, enum iostat_type io_type)
1549 {
1550         pgoff_t index, end;
1551         struct pagevec pvec;
1552         int step = 0;
1553         int nwritten = 0;
1554         int ret = 0;
1555
1556         pagevec_init(&pvec, 0);
1557
1558 next_step:
1559         index = 0;
1560         end = ULONG_MAX;
1561
1562         while (index <= end) {
1563                 int i, nr_pages;
1564                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1565                                 PAGECACHE_TAG_DIRTY,
1566                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1567                 if (nr_pages == 0)
1568                         break;
1569
1570                 for (i = 0; i < nr_pages; i++) {
1571                         struct page *page = pvec.pages[i];
1572                         bool submitted = false;
1573
1574                         if (unlikely(f2fs_cp_error(sbi))) {
1575                                 pagevec_release(&pvec);
1576                                 ret = -EIO;
1577                                 goto out;
1578                         }
1579
1580                         /*
1581                          * flushing sequence with step:
1582                          * 0. indirect nodes
1583                          * 1. dentry dnodes
1584                          * 2. file dnodes
1585                          */
1586                         if (step == 0 && IS_DNODE(page))
1587                                 continue;
1588                         if (step == 1 && (!IS_DNODE(page) ||
1589                                                 is_cold_node(page)))
1590                                 continue;
1591                         if (step == 2 && (!IS_DNODE(page) ||
1592                                                 !is_cold_node(page)))
1593                                 continue;
1594 lock_node:
1595                         if (!trylock_page(page))
1596                                 continue;
1597
1598                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1599 continue_unlock:
1600                                 unlock_page(page);
1601                                 continue;
1602                         }
1603
1604                         if (!PageDirty(page)) {
1605                                 /* someone wrote it for us */
1606                                 goto continue_unlock;
1607                         }
1608
1609                         /* flush inline_data */
1610                         if (is_inline_node(page)) {
1611                                 clear_inline_node(page);
1612                                 unlock_page(page);
1613                                 flush_inline_data(sbi, ino_of_node(page));
1614                                 goto lock_node;
1615                         }
1616
1617                         f2fs_wait_on_page_writeback(page, NODE, true);
1618
1619                         BUG_ON(PageWriteback(page));
1620                         if (!clear_page_dirty_for_io(page))
1621                                 goto continue_unlock;
1622
1623                         set_fsync_mark(page, 0);
1624                         set_dentry_mark(page, 0);
1625
1626                         ret = __write_node_page(page, false, &submitted,
1627                                                 wbc, do_balance, io_type);
1628                         if (ret)
1629                                 unlock_page(page);
1630                         else if (submitted)
1631                                 nwritten++;
1632
1633                         if (--wbc->nr_to_write == 0)
1634                                 break;
1635                 }
1636                 pagevec_release(&pvec);
1637                 cond_resched();
1638
1639                 if (wbc->nr_to_write == 0) {
1640                         step = 2;
1641                         break;
1642                 }
1643         }
1644
1645         if (step < 2) {
1646                 step++;
1647                 goto next_step;
1648         }
1649 out:
1650         if (nwritten)
1651                 f2fs_submit_merged_write(sbi, NODE);
1652         return ret;
1653 }
1654
1655 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1656 {
1657         pgoff_t index = 0, end = ULONG_MAX;
1658         struct pagevec pvec;
1659         int ret2, ret = 0;
1660
1661         pagevec_init(&pvec, 0);
1662
1663         while (index <= end) {
1664                 int i, nr_pages;
1665                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1666                                 PAGECACHE_TAG_WRITEBACK,
1667                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1668                 if (nr_pages == 0)
1669                         break;
1670
1671                 for (i = 0; i < nr_pages; i++) {
1672                         struct page *page = pvec.pages[i];
1673
1674                         /* until radix tree lookup accepts end_index */
1675                         if (unlikely(page->index > end))
1676                                 continue;
1677
1678                         if (ino && ino_of_node(page) == ino) {
1679                                 f2fs_wait_on_page_writeback(page, NODE, true);
1680                                 if (TestClearPageError(page))
1681                                         ret = -EIO;
1682                         }
1683                 }
1684                 pagevec_release(&pvec);
1685                 cond_resched();
1686         }
1687
1688         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1689         if (!ret)
1690                 ret = ret2;
1691         return ret;
1692 }
1693
1694 static int f2fs_write_node_pages(struct address_space *mapping,
1695                             struct writeback_control *wbc)
1696 {
1697         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1698         struct blk_plug plug;
1699         long diff;
1700
1701         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1702                 goto skip_write;
1703
1704         /* balancing f2fs's metadata in background */
1705         f2fs_balance_fs_bg(sbi);
1706
1707         /* collect a number of dirty node pages and write together */
1708         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1709                 goto skip_write;
1710
1711         trace_f2fs_writepages(mapping->host, wbc, NODE);
1712
1713         diff = nr_pages_to_write(sbi, NODE, wbc);
1714         wbc->sync_mode = WB_SYNC_NONE;
1715         blk_start_plug(&plug);
1716         sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1717         blk_finish_plug(&plug);
1718         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1719         return 0;
1720
1721 skip_write:
1722         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1723         trace_f2fs_writepages(mapping->host, wbc, NODE);
1724         return 0;
1725 }
1726
1727 static int f2fs_set_node_page_dirty(struct page *page)
1728 {
1729         trace_f2fs_set_page_dirty(page, NODE);
1730
1731         if (!PageUptodate(page))
1732                 SetPageUptodate(page);
1733         if (!PageDirty(page)) {
1734                 f2fs_set_page_dirty_nobuffers(page);
1735                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1736                 SetPagePrivate(page);
1737                 f2fs_trace_pid(page);
1738                 return 1;
1739         }
1740         return 0;
1741 }
1742
1743 /*
1744  * Structure of the f2fs node operations
1745  */
1746 const struct address_space_operations f2fs_node_aops = {
1747         .writepage      = f2fs_write_node_page,
1748         .writepages     = f2fs_write_node_pages,
1749         .set_page_dirty = f2fs_set_node_page_dirty,
1750         .invalidatepage = f2fs_invalidate_page,
1751         .releasepage    = f2fs_release_page,
1752 #ifdef CONFIG_MIGRATION
1753         .migratepage    = f2fs_migrate_page,
1754 #endif
1755 };
1756
1757 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1758                                                 nid_t n)
1759 {
1760         return radix_tree_lookup(&nm_i->free_nid_root, n);
1761 }
1762
1763 static int __insert_free_nid(struct f2fs_sb_info *sbi,
1764                         struct free_nid *i, enum nid_state state, bool new)
1765 {
1766         struct f2fs_nm_info *nm_i = NM_I(sbi);
1767
1768         if (new) {
1769                 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1770                 if (err)
1771                         return err;
1772         }
1773
1774         f2fs_bug_on(sbi, state != i->state);
1775         nm_i->nid_cnt[state]++;
1776         if (state == FREE_NID)
1777                 list_add_tail(&i->list, &nm_i->free_nid_list);
1778         return 0;
1779 }
1780
1781 static void __remove_free_nid(struct f2fs_sb_info *sbi,
1782                         struct free_nid *i, enum nid_state state, bool reuse)
1783 {
1784         struct f2fs_nm_info *nm_i = NM_I(sbi);
1785
1786         f2fs_bug_on(sbi, state != i->state);
1787         nm_i->nid_cnt[state]--;
1788         if (state == FREE_NID)
1789                 list_del(&i->list);
1790         if (!reuse)
1791                 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1792 }
1793
1794 /* return if the nid is recognized as free */
1795 static bool add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1796 {
1797         struct f2fs_nm_info *nm_i = NM_I(sbi);
1798         struct free_nid *i, *e;
1799         struct nat_entry *ne;
1800         int err = -EINVAL;
1801         bool ret = false;
1802
1803         /* 0 nid should not be used */
1804         if (unlikely(nid == 0))
1805                 return false;
1806
1807         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1808         i->nid = nid;
1809         i->state = FREE_NID;
1810
1811         if (radix_tree_preload(GFP_NOFS))
1812                 goto err;
1813
1814         spin_lock(&nm_i->nid_list_lock);
1815
1816         if (build) {
1817                 /*
1818                  *   Thread A             Thread B
1819                  *  - f2fs_create
1820                  *   - f2fs_new_inode
1821                  *    - alloc_nid
1822                  *     - __insert_nid_to_list(PREALLOC_NID)
1823                  *                     - f2fs_balance_fs_bg
1824                  *                      - build_free_nids
1825                  *                       - __build_free_nids
1826                  *                        - scan_nat_page
1827                  *                         - add_free_nid
1828                  *                          - __lookup_nat_cache
1829                  *  - f2fs_add_link
1830                  *   - init_inode_metadata
1831                  *    - new_inode_page
1832                  *     - new_node_page
1833                  *      - set_node_addr
1834                  *  - alloc_nid_done
1835                  *   - __remove_nid_from_list(PREALLOC_NID)
1836                  *                         - __insert_nid_to_list(FREE_NID)
1837                  */
1838                 ne = __lookup_nat_cache(nm_i, nid);
1839                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1840                                 nat_get_blkaddr(ne) != NULL_ADDR))
1841                         goto err_out;
1842
1843                 e = __lookup_free_nid_list(nm_i, nid);
1844                 if (e) {
1845                         if (e->state == FREE_NID)
1846                                 ret = true;
1847                         goto err_out;
1848                 }
1849         }
1850         ret = true;
1851         err = __insert_free_nid(sbi, i, FREE_NID, true);
1852 err_out:
1853         spin_unlock(&nm_i->nid_list_lock);
1854         radix_tree_preload_end();
1855 err:
1856         if (err)
1857                 kmem_cache_free(free_nid_slab, i);
1858         return ret;
1859 }
1860
1861 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1862 {
1863         struct f2fs_nm_info *nm_i = NM_I(sbi);
1864         struct free_nid *i;
1865         bool need_free = false;
1866
1867         spin_lock(&nm_i->nid_list_lock);
1868         i = __lookup_free_nid_list(nm_i, nid);
1869         if (i && i->state == FREE_NID) {
1870                 __remove_free_nid(sbi, i, FREE_NID, false);
1871                 need_free = true;
1872         }
1873         spin_unlock(&nm_i->nid_list_lock);
1874
1875         if (need_free)
1876                 kmem_cache_free(free_nid_slab, i);
1877 }
1878
1879 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
1880                                                         bool set, bool build)
1881 {
1882         struct f2fs_nm_info *nm_i = NM_I(sbi);
1883         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
1884         unsigned int nid_ofs = nid - START_NID(nid);
1885
1886         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1887                 return;
1888
1889         if (set)
1890                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1891         else
1892                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1893
1894         if (set)
1895                 nm_i->free_nid_count[nat_ofs]++;
1896         else if (!build)
1897                 nm_i->free_nid_count[nat_ofs]--;
1898 }
1899
1900 static void scan_nat_page(struct f2fs_sb_info *sbi,
1901                         struct page *nat_page, nid_t start_nid)
1902 {
1903         struct f2fs_nm_info *nm_i = NM_I(sbi);
1904         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1905         block_t blk_addr;
1906         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
1907         int i;
1908
1909         if (test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1910                 return;
1911
1912         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
1913
1914         i = start_nid % NAT_ENTRY_PER_BLOCK;
1915
1916         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1917                 bool freed = false;
1918
1919                 if (unlikely(start_nid >= nm_i->max_nid))
1920                         break;
1921
1922                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1923                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1924                 if (blk_addr == NULL_ADDR)
1925                         freed = add_free_nid(sbi, start_nid, true);
1926                 spin_lock(&NM_I(sbi)->nid_list_lock);
1927                 update_free_nid_bitmap(sbi, start_nid, freed, true);
1928                 spin_unlock(&NM_I(sbi)->nid_list_lock);
1929         }
1930 }
1931
1932 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
1933 {
1934         struct f2fs_nm_info *nm_i = NM_I(sbi);
1935         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1936         struct f2fs_journal *journal = curseg->journal;
1937         unsigned int i, idx;
1938
1939         down_read(&nm_i->nat_tree_lock);
1940
1941         for (i = 0; i < nm_i->nat_blocks; i++) {
1942                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
1943                         continue;
1944                 if (!nm_i->free_nid_count[i])
1945                         continue;
1946                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
1947                         nid_t nid;
1948
1949                         if (!test_bit_le(idx, nm_i->free_nid_bitmap[i]))
1950                                 continue;
1951
1952                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
1953                         add_free_nid(sbi, nid, true);
1954
1955                         if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
1956                                 goto out;
1957                 }
1958         }
1959 out:
1960         down_read(&curseg->journal_rwsem);
1961         for (i = 0; i < nats_in_cursum(journal); i++) {
1962                 block_t addr;
1963                 nid_t nid;
1964
1965                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1966                 nid = le32_to_cpu(nid_in_journal(journal, i));
1967                 if (addr == NULL_ADDR)
1968                         add_free_nid(sbi, nid, true);
1969                 else
1970                         remove_free_nid(sbi, nid);
1971         }
1972         up_read(&curseg->journal_rwsem);
1973         up_read(&nm_i->nat_tree_lock);
1974 }
1975
1976 static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
1977 {
1978         struct f2fs_nm_info *nm_i = NM_I(sbi);
1979         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1980         struct f2fs_journal *journal = curseg->journal;
1981         int i = 0;
1982         nid_t nid = nm_i->next_scan_nid;
1983
1984         if (unlikely(nid >= nm_i->max_nid))
1985                 nid = 0;
1986
1987         /* Enough entries */
1988         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
1989                 return;
1990
1991         if (!sync && !available_free_memory(sbi, FREE_NIDS))
1992                 return;
1993
1994         if (!mount) {
1995                 /* try to find free nids in free_nid_bitmap */
1996                 scan_free_nid_bits(sbi);
1997
1998                 if (nm_i->nid_cnt[FREE_NID])
1999                         return;
2000         }
2001
2002         /* readahead nat pages to be scanned */
2003         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2004                                                         META_NAT, true);
2005
2006         down_read(&nm_i->nat_tree_lock);
2007
2008         while (1) {
2009                 struct page *page = get_current_nat_page(sbi, nid);
2010
2011                 scan_nat_page(sbi, page, nid);
2012                 f2fs_put_page(page, 1);
2013
2014                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2015                 if (unlikely(nid >= nm_i->max_nid))
2016                         nid = 0;
2017
2018                 if (++i >= FREE_NID_PAGES)
2019                         break;
2020         }
2021
2022         /* go to the next free nat pages to find free nids abundantly */
2023         nm_i->next_scan_nid = nid;
2024
2025         /* find free nids from current sum_pages */
2026         down_read(&curseg->journal_rwsem);
2027         for (i = 0; i < nats_in_cursum(journal); i++) {
2028                 block_t addr;
2029
2030                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2031                 nid = le32_to_cpu(nid_in_journal(journal, i));
2032                 if (addr == NULL_ADDR)
2033                         add_free_nid(sbi, nid, true);
2034                 else
2035                         remove_free_nid(sbi, nid);
2036         }
2037         up_read(&curseg->journal_rwsem);
2038         up_read(&nm_i->nat_tree_lock);
2039
2040         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2041                                         nm_i->ra_nid_pages, META_NAT, false);
2042 }
2043
2044 void build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2045 {
2046         mutex_lock(&NM_I(sbi)->build_lock);
2047         __build_free_nids(sbi, sync, mount);
2048         mutex_unlock(&NM_I(sbi)->build_lock);
2049 }
2050
2051 /*
2052  * If this function returns success, caller can obtain a new nid
2053  * from second parameter of this function.
2054  * The returned nid could be used ino as well as nid when inode is created.
2055  */
2056 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2057 {
2058         struct f2fs_nm_info *nm_i = NM_I(sbi);
2059         struct free_nid *i = NULL;
2060 retry:
2061 #ifdef CONFIG_F2FS_FAULT_INJECTION
2062         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2063                 f2fs_show_injection_info(FAULT_ALLOC_NID);
2064                 return false;
2065         }
2066 #endif
2067         spin_lock(&nm_i->nid_list_lock);
2068
2069         if (unlikely(nm_i->available_nids == 0)) {
2070                 spin_unlock(&nm_i->nid_list_lock);
2071                 return false;
2072         }
2073
2074         /* We should not use stale free nids created by build_free_nids */
2075         if (nm_i->nid_cnt[FREE_NID] && !on_build_free_nids(nm_i)) {
2076                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2077                 i = list_first_entry(&nm_i->free_nid_list,
2078                                         struct free_nid, list);
2079                 *nid = i->nid;
2080
2081                 __remove_free_nid(sbi, i, FREE_NID, true);
2082                 i->state = PREALLOC_NID;
2083                 __insert_free_nid(sbi, i, PREALLOC_NID, false);
2084                 nm_i->available_nids--;
2085
2086                 update_free_nid_bitmap(sbi, *nid, false, false);
2087
2088                 spin_unlock(&nm_i->nid_list_lock);
2089                 return true;
2090         }
2091         spin_unlock(&nm_i->nid_list_lock);
2092
2093         /* Let's scan nat pages and its caches to get free nids */
2094         build_free_nids(sbi, true, false);
2095         goto retry;
2096 }
2097
2098 /*
2099  * alloc_nid() should be called prior to this function.
2100  */
2101 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2102 {
2103         struct f2fs_nm_info *nm_i = NM_I(sbi);
2104         struct free_nid *i;
2105
2106         spin_lock(&nm_i->nid_list_lock);
2107         i = __lookup_free_nid_list(nm_i, nid);
2108         f2fs_bug_on(sbi, !i);
2109         __remove_free_nid(sbi, i, PREALLOC_NID, false);
2110         spin_unlock(&nm_i->nid_list_lock);
2111
2112         kmem_cache_free(free_nid_slab, i);
2113 }
2114
2115 /*
2116  * alloc_nid() should be called prior to this function.
2117  */
2118 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2119 {
2120         struct f2fs_nm_info *nm_i = NM_I(sbi);
2121         struct free_nid *i;
2122         bool need_free = false;
2123
2124         if (!nid)
2125                 return;
2126
2127         spin_lock(&nm_i->nid_list_lock);
2128         i = __lookup_free_nid_list(nm_i, nid);
2129         f2fs_bug_on(sbi, !i);
2130
2131         if (!available_free_memory(sbi, FREE_NIDS)) {
2132                 __remove_free_nid(sbi, i, PREALLOC_NID, false);
2133                 need_free = true;
2134         } else {
2135                 __remove_free_nid(sbi, i, PREALLOC_NID, true);
2136                 i->state = FREE_NID;
2137                 __insert_free_nid(sbi, i, FREE_NID, false);
2138         }
2139
2140         nm_i->available_nids++;
2141
2142         update_free_nid_bitmap(sbi, nid, true, false);
2143
2144         spin_unlock(&nm_i->nid_list_lock);
2145
2146         if (need_free)
2147                 kmem_cache_free(free_nid_slab, i);
2148 }
2149
2150 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2151 {
2152         struct f2fs_nm_info *nm_i = NM_I(sbi);
2153         struct free_nid *i, *next;
2154         int nr = nr_shrink;
2155
2156         if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2157                 return 0;
2158
2159         if (!mutex_trylock(&nm_i->build_lock))
2160                 return 0;
2161
2162         spin_lock(&nm_i->nid_list_lock);
2163         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2164                 if (nr_shrink <= 0 ||
2165                                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2166                         break;
2167
2168                 __remove_free_nid(sbi, i, FREE_NID, false);
2169                 kmem_cache_free(free_nid_slab, i);
2170                 nr_shrink--;
2171         }
2172         spin_unlock(&nm_i->nid_list_lock);
2173         mutex_unlock(&nm_i->build_lock);
2174
2175         return nr - nr_shrink;
2176 }
2177
2178 void recover_inline_xattr(struct inode *inode, struct page *page)
2179 {
2180         void *src_addr, *dst_addr;
2181         size_t inline_size;
2182         struct page *ipage;
2183         struct f2fs_inode *ri;
2184
2185         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2186         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2187
2188         ri = F2FS_INODE(page);
2189         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2190                 clear_inode_flag(inode, FI_INLINE_XATTR);
2191                 goto update_inode;
2192         }
2193
2194         dst_addr = inline_xattr_addr(ipage);
2195         src_addr = inline_xattr_addr(page);
2196         inline_size = inline_xattr_size(inode);
2197
2198         f2fs_wait_on_page_writeback(ipage, NODE, true);
2199         memcpy(dst_addr, src_addr, inline_size);
2200 update_inode:
2201         update_inode(inode, ipage);
2202         f2fs_put_page(ipage, 1);
2203 }
2204
2205 int recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2206 {
2207         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2208         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2209         nid_t new_xnid;
2210         struct dnode_of_data dn;
2211         struct node_info ni;
2212         struct page *xpage;
2213
2214         if (!prev_xnid)
2215                 goto recover_xnid;
2216
2217         /* 1: invalidate the previous xattr nid */
2218         get_node_info(sbi, prev_xnid, &ni);
2219         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2220         invalidate_blocks(sbi, ni.blk_addr);
2221         dec_valid_node_count(sbi, inode, false);
2222         set_node_addr(sbi, &ni, NULL_ADDR, false);
2223
2224 recover_xnid:
2225         /* 2: update xattr nid in inode */
2226         if (!alloc_nid(sbi, &new_xnid))
2227                 return -ENOSPC;
2228
2229         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2230         xpage = new_node_page(&dn, XATTR_NODE_OFFSET);
2231         if (IS_ERR(xpage)) {
2232                 alloc_nid_failed(sbi, new_xnid);
2233                 return PTR_ERR(xpage);
2234         }
2235
2236         alloc_nid_done(sbi, new_xnid);
2237         update_inode_page(inode);
2238
2239         /* 3: update and set xattr node page dirty */
2240         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2241
2242         set_page_dirty(xpage);
2243         f2fs_put_page(xpage, 1);
2244
2245         return 0;
2246 }
2247
2248 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2249 {
2250         struct f2fs_inode *src, *dst;
2251         nid_t ino = ino_of_node(page);
2252         struct node_info old_ni, new_ni;
2253         struct page *ipage;
2254
2255         get_node_info(sbi, ino, &old_ni);
2256
2257         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2258                 return -EINVAL;
2259 retry:
2260         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2261         if (!ipage) {
2262                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2263                 goto retry;
2264         }
2265
2266         /* Should not use this inode from free nid list */
2267         remove_free_nid(sbi, ino);
2268
2269         if (!PageUptodate(ipage))
2270                 SetPageUptodate(ipage);
2271         fill_node_footer(ipage, ino, ino, 0, true);
2272
2273         src = F2FS_INODE(page);
2274         dst = F2FS_INODE(ipage);
2275
2276         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2277         dst->i_size = 0;
2278         dst->i_blocks = cpu_to_le64(1);
2279         dst->i_links = cpu_to_le32(1);
2280         dst->i_xattr_nid = 0;
2281         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2282         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2283                 dst->i_extra_isize = src->i_extra_isize;
2284                 if (f2fs_sb_has_project_quota(sbi->sb) &&
2285                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2286                                                                 i_projid))
2287                         dst->i_projid = src->i_projid;
2288         }
2289
2290         new_ni = old_ni;
2291         new_ni.ino = ino;
2292
2293         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2294                 WARN_ON(1);
2295         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2296         inc_valid_inode_count(sbi);
2297         set_page_dirty(ipage);
2298         f2fs_put_page(ipage, 1);
2299         return 0;
2300 }
2301
2302 int restore_node_summary(struct f2fs_sb_info *sbi,
2303                         unsigned int segno, struct f2fs_summary_block *sum)
2304 {
2305         struct f2fs_node *rn;
2306         struct f2fs_summary *sum_entry;
2307         block_t addr;
2308         int i, idx, last_offset, nrpages;
2309
2310         /* scan the node segment */
2311         last_offset = sbi->blocks_per_seg;
2312         addr = START_BLOCK(sbi, segno);
2313         sum_entry = &sum->entries[0];
2314
2315         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2316                 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2317
2318                 /* readahead node pages */
2319                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2320
2321                 for (idx = addr; idx < addr + nrpages; idx++) {
2322                         struct page *page = get_tmp_page(sbi, idx);
2323
2324                         rn = F2FS_NODE(page);
2325                         sum_entry->nid = rn->footer.nid;
2326                         sum_entry->version = 0;
2327                         sum_entry->ofs_in_node = 0;
2328                         sum_entry++;
2329                         f2fs_put_page(page, 1);
2330                 }
2331
2332                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2333                                                         addr + nrpages);
2334         }
2335         return 0;
2336 }
2337
2338 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2339 {
2340         struct f2fs_nm_info *nm_i = NM_I(sbi);
2341         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2342         struct f2fs_journal *journal = curseg->journal;
2343         int i;
2344
2345         down_write(&curseg->journal_rwsem);
2346         for (i = 0; i < nats_in_cursum(journal); i++) {
2347                 struct nat_entry *ne;
2348                 struct f2fs_nat_entry raw_ne;
2349                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2350
2351                 raw_ne = nat_in_journal(journal, i);
2352
2353                 ne = __lookup_nat_cache(nm_i, nid);
2354                 if (!ne) {
2355                         ne = grab_nat_entry(nm_i, nid, true);
2356                         node_info_from_raw_nat(&ne->ni, &raw_ne);
2357                 }
2358
2359                 /*
2360                  * if a free nat in journal has not been used after last
2361                  * checkpoint, we should remove it from available nids,
2362                  * since later we will add it again.
2363                  */
2364                 if (!get_nat_flag(ne, IS_DIRTY) &&
2365                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2366                         spin_lock(&nm_i->nid_list_lock);
2367                         nm_i->available_nids--;
2368                         spin_unlock(&nm_i->nid_list_lock);
2369                 }
2370
2371                 __set_nat_cache_dirty(nm_i, ne);
2372         }
2373         update_nats_in_cursum(journal, -i);
2374         up_write(&curseg->journal_rwsem);
2375 }
2376
2377 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2378                                                 struct list_head *head, int max)
2379 {
2380         struct nat_entry_set *cur;
2381
2382         if (nes->entry_cnt >= max)
2383                 goto add_out;
2384
2385         list_for_each_entry(cur, head, set_list) {
2386                 if (cur->entry_cnt >= nes->entry_cnt) {
2387                         list_add(&nes->set_list, cur->set_list.prev);
2388                         return;
2389                 }
2390         }
2391 add_out:
2392         list_add_tail(&nes->set_list, head);
2393 }
2394
2395 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2396                                                 struct page *page)
2397 {
2398         struct f2fs_nm_info *nm_i = NM_I(sbi);
2399         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2400         struct f2fs_nat_block *nat_blk = page_address(page);
2401         int valid = 0;
2402         int i;
2403
2404         if (!enabled_nat_bits(sbi, NULL))
2405                 return;
2406
2407         for (i = 0; i < NAT_ENTRY_PER_BLOCK; i++) {
2408                 if (start_nid == 0 && i == 0)
2409                         valid++;
2410                 if (nat_blk->entries[i].block_addr)
2411                         valid++;
2412         }
2413         if (valid == 0) {
2414                 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2415                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2416                 return;
2417         }
2418
2419         __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2420         if (valid == NAT_ENTRY_PER_BLOCK)
2421                 __set_bit_le(nat_index, nm_i->full_nat_bits);
2422         else
2423                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2424 }
2425
2426 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2427                 struct nat_entry_set *set, struct cp_control *cpc)
2428 {
2429         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2430         struct f2fs_journal *journal = curseg->journal;
2431         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2432         bool to_journal = true;
2433         struct f2fs_nat_block *nat_blk;
2434         struct nat_entry *ne, *cur;
2435         struct page *page = NULL;
2436
2437         /*
2438          * there are two steps to flush nat entries:
2439          * #1, flush nat entries to journal in current hot data summary block.
2440          * #2, flush nat entries to nat page.
2441          */
2442         if (enabled_nat_bits(sbi, cpc) ||
2443                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2444                 to_journal = false;
2445
2446         if (to_journal) {
2447                 down_write(&curseg->journal_rwsem);
2448         } else {
2449                 page = get_next_nat_page(sbi, start_nid);
2450                 nat_blk = page_address(page);
2451                 f2fs_bug_on(sbi, !nat_blk);
2452         }
2453
2454         /* flush dirty nats in nat entry set */
2455         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2456                 struct f2fs_nat_entry *raw_ne;
2457                 nid_t nid = nat_get_nid(ne);
2458                 int offset;
2459
2460                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2461
2462                 if (to_journal) {
2463                         offset = lookup_journal_in_cursum(journal,
2464                                                         NAT_JOURNAL, nid, 1);
2465                         f2fs_bug_on(sbi, offset < 0);
2466                         raw_ne = &nat_in_journal(journal, offset);
2467                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2468                 } else {
2469                         raw_ne = &nat_blk->entries[nid - start_nid];
2470                 }
2471                 raw_nat_from_node_info(raw_ne, &ne->ni);
2472                 nat_reset_flag(ne);
2473                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2474                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2475                         add_free_nid(sbi, nid, false);
2476                         spin_lock(&NM_I(sbi)->nid_list_lock);
2477                         NM_I(sbi)->available_nids++;
2478                         update_free_nid_bitmap(sbi, nid, true, false);
2479                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2480                 } else {
2481                         spin_lock(&NM_I(sbi)->nid_list_lock);
2482                         update_free_nid_bitmap(sbi, nid, false, false);
2483                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2484                 }
2485         }
2486
2487         if (to_journal) {
2488                 up_write(&curseg->journal_rwsem);
2489         } else {
2490                 __update_nat_bits(sbi, start_nid, page);
2491                 f2fs_put_page(page, 1);
2492         }
2493
2494         /* Allow dirty nats by node block allocation in write_begin */
2495         if (!set->entry_cnt) {
2496                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2497                 kmem_cache_free(nat_entry_set_slab, set);
2498         }
2499 }
2500
2501 /*
2502  * This function is called during the checkpointing process.
2503  */
2504 void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2505 {
2506         struct f2fs_nm_info *nm_i = NM_I(sbi);
2507         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2508         struct f2fs_journal *journal = curseg->journal;
2509         struct nat_entry_set *setvec[SETVEC_SIZE];
2510         struct nat_entry_set *set, *tmp;
2511         unsigned int found;
2512         nid_t set_idx = 0;
2513         LIST_HEAD(sets);
2514
2515         if (!nm_i->dirty_nat_cnt)
2516                 return;
2517
2518         down_write(&nm_i->nat_tree_lock);
2519
2520         /*
2521          * if there are no enough space in journal to store dirty nat
2522          * entries, remove all entries from journal and merge them
2523          * into nat entry set.
2524          */
2525         if (enabled_nat_bits(sbi, cpc) ||
2526                 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2527                 remove_nats_in_journal(sbi);
2528
2529         while ((found = __gang_lookup_nat_set(nm_i,
2530                                         set_idx, SETVEC_SIZE, setvec))) {
2531                 unsigned idx;
2532                 set_idx = setvec[found - 1]->set + 1;
2533                 for (idx = 0; idx < found; idx++)
2534                         __adjust_nat_entry_set(setvec[idx], &sets,
2535                                                 MAX_NAT_JENTRIES(journal));
2536         }
2537
2538         /* flush dirty nats in nat entry set */
2539         list_for_each_entry_safe(set, tmp, &sets, set_list)
2540                 __flush_nat_entry_set(sbi, set, cpc);
2541
2542         up_write(&nm_i->nat_tree_lock);
2543         /* Allow dirty nats by node block allocation in write_begin */
2544 }
2545
2546 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2547 {
2548         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2549         struct f2fs_nm_info *nm_i = NM_I(sbi);
2550         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2551         unsigned int i;
2552         __u64 cp_ver = cur_cp_version(ckpt);
2553         block_t nat_bits_addr;
2554
2555         if (!enabled_nat_bits(sbi, NULL))
2556                 return 0;
2557
2558         nm_i->nat_bits_blocks = F2FS_BYTES_TO_BLK((nat_bits_bytes << 1) + 8 +
2559                                                 F2FS_BLKSIZE - 1);
2560         nm_i->nat_bits = kzalloc(nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS,
2561                                                 GFP_KERNEL);
2562         if (!nm_i->nat_bits)
2563                 return -ENOMEM;
2564
2565         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2566                                                 nm_i->nat_bits_blocks;
2567         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2568                 struct page *page = get_meta_page(sbi, nat_bits_addr++);
2569
2570                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2571                                         page_address(page), F2FS_BLKSIZE);
2572                 f2fs_put_page(page, 1);
2573         }
2574
2575         cp_ver |= (cur_cp_crc(ckpt) << 32);
2576         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2577                 disable_nat_bits(sbi, true);
2578                 return 0;
2579         }
2580
2581         nm_i->full_nat_bits = nm_i->nat_bits + 8;
2582         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2583
2584         f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2585         return 0;
2586 }
2587
2588 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2589 {
2590         struct f2fs_nm_info *nm_i = NM_I(sbi);
2591         unsigned int i = 0;
2592         nid_t nid, last_nid;
2593
2594         if (!enabled_nat_bits(sbi, NULL))
2595                 return;
2596
2597         for (i = 0; i < nm_i->nat_blocks; i++) {
2598                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2599                 if (i >= nm_i->nat_blocks)
2600                         break;
2601
2602                 __set_bit_le(i, nm_i->nat_block_bitmap);
2603
2604                 nid = i * NAT_ENTRY_PER_BLOCK;
2605                 last_nid = (i + 1) * NAT_ENTRY_PER_BLOCK;
2606
2607                 spin_lock(&NM_I(sbi)->nid_list_lock);
2608                 for (; nid < last_nid; nid++)
2609                         update_free_nid_bitmap(sbi, nid, true, true);
2610                 spin_unlock(&NM_I(sbi)->nid_list_lock);
2611         }
2612
2613         for (i = 0; i < nm_i->nat_blocks; i++) {
2614                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2615                 if (i >= nm_i->nat_blocks)
2616                         break;
2617
2618                 __set_bit_le(i, nm_i->nat_block_bitmap);
2619         }
2620 }
2621
2622 static int init_node_manager(struct f2fs_sb_info *sbi)
2623 {
2624         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2625         struct f2fs_nm_info *nm_i = NM_I(sbi);
2626         unsigned char *version_bitmap;
2627         unsigned int nat_segs;
2628         int err;
2629
2630         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2631
2632         /* segment_count_nat includes pair segment so divide to 2. */
2633         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2634         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2635         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2636
2637         /* not used nids: 0, node, meta, (and root counted as valid node) */
2638         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2639                                                         F2FS_RESERVED_NODE_NUM;
2640         nm_i->nid_cnt[FREE_NID] = 0;
2641         nm_i->nid_cnt[PREALLOC_NID] = 0;
2642         nm_i->nat_cnt = 0;
2643         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2644         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2645         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2646
2647         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2648         INIT_LIST_HEAD(&nm_i->free_nid_list);
2649         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2650         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2651         INIT_LIST_HEAD(&nm_i->nat_entries);
2652
2653         mutex_init(&nm_i->build_lock);
2654         spin_lock_init(&nm_i->nid_list_lock);
2655         init_rwsem(&nm_i->nat_tree_lock);
2656
2657         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2658         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2659         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2660         if (!version_bitmap)
2661                 return -EFAULT;
2662
2663         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2664                                         GFP_KERNEL);
2665         if (!nm_i->nat_bitmap)
2666                 return -ENOMEM;
2667
2668         err = __get_nat_bitmaps(sbi);
2669         if (err)
2670                 return err;
2671
2672 #ifdef CONFIG_F2FS_CHECK_FS
2673         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2674                                         GFP_KERNEL);
2675         if (!nm_i->nat_bitmap_mir)
2676                 return -ENOMEM;
2677 #endif
2678
2679         return 0;
2680 }
2681
2682 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2683 {
2684         struct f2fs_nm_info *nm_i = NM_I(sbi);
2685
2686         nm_i->free_nid_bitmap = kvzalloc(nm_i->nat_blocks *
2687                                         NAT_ENTRY_BITMAP_SIZE, GFP_KERNEL);
2688         if (!nm_i->free_nid_bitmap)
2689                 return -ENOMEM;
2690
2691         nm_i->nat_block_bitmap = kvzalloc(nm_i->nat_blocks / 8,
2692                                                                 GFP_KERNEL);
2693         if (!nm_i->nat_block_bitmap)
2694                 return -ENOMEM;
2695
2696         nm_i->free_nid_count = kvzalloc(nm_i->nat_blocks *
2697                                         sizeof(unsigned short), GFP_KERNEL);
2698         if (!nm_i->free_nid_count)
2699                 return -ENOMEM;
2700         return 0;
2701 }
2702
2703 int build_node_manager(struct f2fs_sb_info *sbi)
2704 {
2705         int err;
2706
2707         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2708         if (!sbi->nm_info)
2709                 return -ENOMEM;
2710
2711         err = init_node_manager(sbi);
2712         if (err)
2713                 return err;
2714
2715         err = init_free_nid_cache(sbi);
2716         if (err)
2717                 return err;
2718
2719         /* load free nid status from nat_bits table */
2720         load_free_nid_bitmap(sbi);
2721
2722         build_free_nids(sbi, true, true);
2723         return 0;
2724 }
2725
2726 void destroy_node_manager(struct f2fs_sb_info *sbi)
2727 {
2728         struct f2fs_nm_info *nm_i = NM_I(sbi);
2729         struct free_nid *i, *next_i;
2730         struct nat_entry *natvec[NATVEC_SIZE];
2731         struct nat_entry_set *setvec[SETVEC_SIZE];
2732         nid_t nid = 0;
2733         unsigned int found;
2734
2735         if (!nm_i)
2736                 return;
2737
2738         /* destroy free nid list */
2739         spin_lock(&nm_i->nid_list_lock);
2740         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2741                 __remove_free_nid(sbi, i, FREE_NID, false);
2742                 spin_unlock(&nm_i->nid_list_lock);
2743                 kmem_cache_free(free_nid_slab, i);
2744                 spin_lock(&nm_i->nid_list_lock);
2745         }
2746         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
2747         f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
2748         f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
2749         spin_unlock(&nm_i->nid_list_lock);
2750
2751         /* destroy nat cache */
2752         down_write(&nm_i->nat_tree_lock);
2753         while ((found = __gang_lookup_nat_cache(nm_i,
2754                                         nid, NATVEC_SIZE, natvec))) {
2755                 unsigned idx;
2756
2757                 nid = nat_get_nid(natvec[found - 1]) + 1;
2758                 for (idx = 0; idx < found; idx++)
2759                         __del_from_nat_cache(nm_i, natvec[idx]);
2760         }
2761         f2fs_bug_on(sbi, nm_i->nat_cnt);
2762
2763         /* destroy nat set cache */
2764         nid = 0;
2765         while ((found = __gang_lookup_nat_set(nm_i,
2766                                         nid, SETVEC_SIZE, setvec))) {
2767                 unsigned idx;
2768
2769                 nid = setvec[found - 1]->set + 1;
2770                 for (idx = 0; idx < found; idx++) {
2771                         /* entry_cnt is not zero, when cp_error was occurred */
2772                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2773                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2774                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2775                 }
2776         }
2777         up_write(&nm_i->nat_tree_lock);
2778
2779         kvfree(nm_i->nat_block_bitmap);
2780         kvfree(nm_i->free_nid_bitmap);
2781         kvfree(nm_i->free_nid_count);
2782
2783         kfree(nm_i->nat_bitmap);
2784         kfree(nm_i->nat_bits);
2785 #ifdef CONFIG_F2FS_CHECK_FS
2786         kfree(nm_i->nat_bitmap_mir);
2787 #endif
2788         sbi->nm_info = NULL;
2789         kfree(nm_i);
2790 }
2791
2792 int __init create_node_manager_caches(void)
2793 {
2794         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2795                         sizeof(struct nat_entry));
2796         if (!nat_entry_slab)
2797                 goto fail;
2798
2799         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2800                         sizeof(struct free_nid));
2801         if (!free_nid_slab)
2802                 goto destroy_nat_entry;
2803
2804         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2805                         sizeof(struct nat_entry_set));
2806         if (!nat_entry_set_slab)
2807                 goto destroy_free_nid;
2808         return 0;
2809
2810 destroy_free_nid:
2811         kmem_cache_destroy(free_nid_slab);
2812 destroy_nat_entry:
2813         kmem_cache_destroy(nat_entry_slab);
2814 fail:
2815         return -ENOMEM;
2816 }
2817
2818 void destroy_node_manager_caches(void)
2819 {
2820         kmem_cache_destroy(nat_entry_set_slab);
2821         kmem_cache_destroy(free_nid_slab);
2822         kmem_cache_destroy(nat_entry_slab);
2823 }