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_f2fs_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 /*
  33  * Check whether the given nid is within node id range.
  34  */
  35 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
  36 {
  37         if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
  38                 set_sbi_flag(sbi, SBI_NEED_FSCK);
  39                 f2fs_msg(sbi->sb, KERN_WARNING,
  40                                 "%s: out-of-range nid=%x, run fsck to fix.",
  41                                 __func__, nid);
  42                 return -EINVAL;
  43         }
  44         return 0;
  45 }
  46
  47 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
  48 {
  49         struct f2fs_nm_info *nm_i = NM_I(sbi);
  50         struct sysinfo val;
  51         unsigned long avail_ram;
  52         unsigned long mem_size = 0;
  53         bool res = false;
  54
  55         si_meminfo(&val);
  56
  57         /* only uses low memory */
  58         avail_ram = val.totalram - val.totalhigh;
  59
  60         /*
  61          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  62          */
  63         if (type == FREE_NIDS) {
  64                 mem_size = (nm_i->nid_cnt[FREE_NID] *
  65                                 sizeof(struct free_nid)) >> PAGE_SHIFT;
  66                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  67         } else if (type == NAT_ENTRIES) {
  68                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
  69                                                         PAGE_SHIFT;
  70                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  71                 if (excess_cached_nats(sbi))
  72                         res = false;
  73         } else if (type == DIRTY_DENTS) {
  74                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  75                         return false;
  76                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  77                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  78         } else if (type == INO_ENTRIES) {
  79                 int i;
  80
  81                 for (i = 0; i < MAX_INO_ENTRY; i++)
  82                         mem_size += sbi->im[i].ino_num *
  83                                                 sizeof(struct ino_entry);
  84                 mem_size >>= PAGE_SHIFT;
  85                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  86         } else if (type == EXTENT_CACHE) {
  87                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  88                                 sizeof(struct extent_tree) +
  89                                 atomic_read(&sbi->total_ext_node) *
  90                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  91                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  92         } else if (type == INMEM_PAGES) {
  93                 /* it allows 20% / total_ram for inmemory pages */
  94                 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
  95                 res = mem_size < (val.totalram / 5);
  96         } else {
  97                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
  98                         return true;
  99         }
 100         return res;
 101 }
 102
 103 static void clear_node_page_dirty(struct page *page)
 104 {
 105         if (PageDirty(page)) {
 106                 f2fs_clear_radix_tree_dirty_tag(page);
 107                 clear_page_dirty_for_io(page);
 108                 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
 109         }
 110         ClearPageUptodate(page);
 111 }
 112
 113 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 114 {
 115         pgoff_t index = current_nat_addr(sbi, nid);
 116         return f2fs_get_meta_page(sbi, index);
 117 }
 118
 119 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 120 {
 121         struct page *src_page;
 122         struct page *dst_page;
 123         pgoff_t src_off;
 124         pgoff_t dst_off;
 125         void *src_addr;
 126         void *dst_addr;
 127         struct f2fs_nm_info *nm_i = NM_I(sbi);
 128
 129         src_off = current_nat_addr(sbi, nid);
 130         dst_off = next_nat_addr(sbi, src_off);
 131
 132         /* get current nat block page with lock */
 133         src_page = f2fs_get_meta_page(sbi, src_off);
 134         dst_page = f2fs_grab_meta_page(sbi, dst_off);
 135         f2fs_bug_on(sbi, PageDirty(src_page));
 136
 137         src_addr = page_address(src_page);
 138         dst_addr = page_address(dst_page);
 139         memcpy(dst_addr, src_addr, PAGE_SIZE);
 140         set_page_dirty(dst_page);
 141         f2fs_put_page(src_page, 1);
 142
 143         set_to_next_nat(nm_i, nid);
 144
 145         return dst_page;
 146 }
 147
 148 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
 149 {
 150         struct nat_entry *new;
 151
 152         if (no_fail)
 153                 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
 154         else
 155                 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
 156         if (new) {
 157                 nat_set_nid(new, nid);
 158                 nat_reset_flag(new);
 159         }
 160         return new;
 161 }
 162
 163 static void __free_nat_entry(struct nat_entry *e)
 164 {
 165         kmem_cache_free(nat_entry_slab, e);
 166 }
 167
 168 /* must be locked by nat_tree_lock */
 169 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
 170         struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
 171 {
 172         if (no_fail)
 173                 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
 174         else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
 175                 return NULL;
 176
 177         if (raw_ne)
 178                 node_info_from_raw_nat(&ne->ni, raw_ne);
 179         list_add_tail(&ne->list, &nm_i->nat_entries);
 180         nm_i->nat_cnt++;
 181         return ne;
 182 }
 183
 184 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 185 {
 186         return radix_tree_lookup(&nm_i->nat_root, n);
 187 }
 188
 189 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 190                 nid_t start, unsigned int nr, struct nat_entry **ep)
 191 {
 192         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 193 }
 194
 195 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 196 {
 197         list_del(&e->list);
 198         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 199         nm_i->nat_cnt--;
 200         __free_nat_entry(e);
 201 }
 202
 203 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
 204                                                         struct nat_entry *ne)
 205 {
 206         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 207         struct nat_entry_set *head;
 208
 209         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 210         if (!head) {
 211                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 212
 213                 INIT_LIST_HEAD(&head->entry_list);
 214                 INIT_LIST_HEAD(&head->set_list);
 215                 head->set = set;
 216                 head->entry_cnt = 0;
 217                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 218         }
 219         return head;
 220 }
 221
 222 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 223                                                 struct nat_entry *ne)
 224 {
 225         struct nat_entry_set *head;
 226         bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
 227
 228         if (!new_ne)
 229                 head = __grab_nat_entry_set(nm_i, ne);
 230
 231         /*
 232          * update entry_cnt in below condition:
 233          * 1. update NEW_ADDR to valid block address;
 234          * 2. update old block address to new one;
 235          */
 236         if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
 237                                 !get_nat_flag(ne, IS_DIRTY)))
 238                 head->entry_cnt++;
 239
 240         set_nat_flag(ne, IS_PREALLOC, new_ne);
 241
 242         if (get_nat_flag(ne, IS_DIRTY))
 243                 goto refresh_list;
 244
 245         nm_i->dirty_nat_cnt++;
 246         set_nat_flag(ne, IS_DIRTY, true);
 247 refresh_list:
 248         if (new_ne)
 249                 list_del_init(&ne->list);
 250         else
 251                 list_move_tail(&ne->list, &head->entry_list);
 252 }
 253
 254 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 255                 struct nat_entry_set *set, struct nat_entry *ne)
 256 {
 257         list_move_tail(&ne->list, &nm_i->nat_entries);
 258         set_nat_flag(ne, IS_DIRTY, false);
 259         set->entry_cnt--;
 260         nm_i->dirty_nat_cnt--;
 261 }
 262
 263 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 264                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 265 {
 266         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 267                                                         start, nr);
 268 }
 269
 270 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 271 {
 272         struct f2fs_nm_info *nm_i = NM_I(sbi);
 273         struct nat_entry *e;
 274         bool need = false;
 275
 276         down_read(&nm_i->nat_tree_lock);
 277         e = __lookup_nat_cache(nm_i, nid);
 278         if (e) {
 279                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 280                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 281                         need = true;
 282         }
 283         up_read(&nm_i->nat_tree_lock);
 284         return need;
 285 }
 286
 287 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 288 {
 289         struct f2fs_nm_info *nm_i = NM_I(sbi);
 290         struct nat_entry *e;
 291         bool is_cp = true;
 292
 293         down_read(&nm_i->nat_tree_lock);
 294         e = __lookup_nat_cache(nm_i, nid);
 295         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 296                 is_cp = false;
 297         up_read(&nm_i->nat_tree_lock);
 298         return is_cp;
 299 }
 300
 301 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 302 {
 303         struct f2fs_nm_info *nm_i = NM_I(sbi);
 304         struct nat_entry *e;
 305         bool need_update = true;
 306
 307         down_read(&nm_i->nat_tree_lock);
 308         e = __lookup_nat_cache(nm_i, ino);
 309         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 310                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 311                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 312                 need_update = false;
 313         up_read(&nm_i->nat_tree_lock);
 314         return need_update;
 315 }
 316
 317 /* must be locked by nat_tree_lock */
 318 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 319                                                 struct f2fs_nat_entry *ne)
 320 {
 321         struct f2fs_nm_info *nm_i = NM_I(sbi);
 322         struct nat_entry *new, *e;
 323
 324         new = __alloc_nat_entry(nid, false);
 325         if (!new)
 326                 return;
 327
 328         down_write(&nm_i->nat_tree_lock);
 329         e = __lookup_nat_cache(nm_i, nid);
 330         if (!e)
 331                 e = __init_nat_entry(nm_i, new, ne, false);
 332         else
 333                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 334                                 nat_get_blkaddr(e) !=
 335                                         le32_to_cpu(ne->block_addr) ||
 336                                 nat_get_version(e) != ne->version);
 337         up_write(&nm_i->nat_tree_lock);
 338         if (e != new)
 339                 __free_nat_entry(new);
 340 }
 341
 342 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 343                         block_t new_blkaddr, bool fsync_done)
 344 {
 345         struct f2fs_nm_info *nm_i = NM_I(sbi);
 346         struct nat_entry *e;
 347         struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
 348
 349         down_write(&nm_i->nat_tree_lock);
 350         e = __lookup_nat_cache(nm_i, ni->nid);
 351         if (!e) {
 352                 e = __init_nat_entry(nm_i, new, NULL, true);
 353                 copy_node_info(&e->ni, ni);
 354                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 355         } else if (new_blkaddr == NEW_ADDR) {
 356                 /*
 357                  * when nid is reallocated,
 358                  * previous nat entry can be remained in nat cache.
 359                  * So, reinitialize it with new information.
 360                  */
 361                 copy_node_info(&e->ni, ni);
 362                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 363         }
 364         /* let's free early to reduce memory consumption */
 365         if (e != new)
 366                 __free_nat_entry(new);
 367
 368         /* sanity check */
 369         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 370         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 371                         new_blkaddr == NULL_ADDR);
 372         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 373                         new_blkaddr == NEW_ADDR);
 374         f2fs_bug_on(sbi, is_valid_blkaddr(nat_get_blkaddr(e)) &&
 375                         new_blkaddr == NEW_ADDR);
 376
 377         /* increment version no as node is removed */
 378         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 379                 unsigned char version = nat_get_version(e);
 380                 nat_set_version(e, inc_node_version(version));
 381         }
 382
 383         /* change address */
 384         nat_set_blkaddr(e, new_blkaddr);
 385         if (!is_valid_blkaddr(new_blkaddr))
 386                 set_nat_flag(e, IS_CHECKPOINTED, false);
 387         __set_nat_cache_dirty(nm_i, e);
 388
 389         /* update fsync_mark if its inode nat entry is still alive */
 390         if (ni->nid != ni->ino)
 391                 e = __lookup_nat_cache(nm_i, ni->ino);
 392         if (e) {
 393                 if (fsync_done && ni->nid == ni->ino)
 394                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 395                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 396         }
 397         up_write(&nm_i->nat_tree_lock);
 398 }
 399
 400 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 401 {
 402         struct f2fs_nm_info *nm_i = NM_I(sbi);
 403         int nr = nr_shrink;
 404
 405         if (!down_write_trylock(&nm_i->nat_tree_lock))
 406                 return 0;
 407
 408         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 409                 struct nat_entry *ne;
 410                 ne = list_first_entry(&nm_i->nat_entries,
 411                                         struct nat_entry, list);
 412                 __del_from_nat_cache(nm_i, ne);
 413                 nr_shrink--;
 414         }
 415         up_write(&nm_i->nat_tree_lock);
 416         return nr - nr_shrink;
 417 }
 418
 419 /*
 420  * This function always returns success
 421  */
 422 void f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
 423                                                 struct node_info *ni)
 424 {
 425         struct f2fs_nm_info *nm_i = NM_I(sbi);
 426         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 427         struct f2fs_journal *journal = curseg->journal;
 428         nid_t start_nid = START_NID(nid);
 429         struct f2fs_nat_block *nat_blk;
 430         struct page *page = NULL;
 431         struct f2fs_nat_entry ne;
 432         struct nat_entry *e;
 433         pgoff_t index;
 434         int i;
 435
 436         ni->nid = nid;
 437
 438         /* Check nat cache */
 439         down_read(&nm_i->nat_tree_lock);
 440         e = __lookup_nat_cache(nm_i, nid);
 441         if (e) {
 442                 ni->ino = nat_get_ino(e);
 443                 ni->blk_addr = nat_get_blkaddr(e);
 444                 ni->version = nat_get_version(e);
 445                 up_read(&nm_i->nat_tree_lock);
 446                 return;
 447         }
 448
 449         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 450
 451         /* Check current segment summary */
 452         down_read(&curseg->journal_rwsem);
 453         i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 454         if (i >= 0) {
 455                 ne = nat_in_journal(journal, i);
 456                 node_info_from_raw_nat(ni, &ne);
 457         }
 458         up_read(&curseg->journal_rwsem);
 459         if (i >= 0) {
 460                 up_read(&nm_i->nat_tree_lock);
 461                 goto cache;
 462         }
 463
 464         /* Fill node_info from nat page */
 465         index = current_nat_addr(sbi, nid);
 466         up_read(&nm_i->nat_tree_lock);
 467
 468         page = f2fs_get_meta_page(sbi, index);
 469         nat_blk = (struct f2fs_nat_block *)page_address(page);
 470         ne = nat_blk->entries[nid - start_nid];
 471         node_info_from_raw_nat(ni, &ne);
 472         f2fs_put_page(page, 1);
 473 cache:
 474         /* cache nat entry */
 475         cache_nat_entry(sbi, nid, &ne);
 476 }
 477
 478 /*
 479  * readahead MAX_RA_NODE number of node pages.
 480  */
 481 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
 482 {
 483         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 484         struct blk_plug plug;
 485         int i, end;
 486         nid_t nid;
 487
 488         blk_start_plug(&plug);
 489
 490         /* Then, try readahead for siblings of the desired node */
 491         end = start + n;
 492         end = min(end, NIDS_PER_BLOCK);
 493         for (i = start; i < end; i++) {
 494                 nid = get_nid(parent, i, false);
 495                 f2fs_ra_node_page(sbi, nid);
 496         }
 497
 498         blk_finish_plug(&plug);
 499 }
 500
 501 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 502 {
 503         const long direct_index = ADDRS_PER_INODE(dn->inode);
 504         const long direct_blks = ADDRS_PER_BLOCK;
 505         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 506         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 507         int cur_level = dn->cur_level;
 508         int max_level = dn->max_level;
 509         pgoff_t base = 0;
 510
 511         if (!dn->max_level)
 512                 return pgofs + 1;
 513
 514         while (max_level-- > cur_level)
 515                 skipped_unit *= NIDS_PER_BLOCK;
 516
 517         switch (dn->max_level) {
 518         case 3:
 519                 base += 2 * indirect_blks;
 520         case 2:
 521                 base += 2 * direct_blks;
 522         case 1:
 523                 base += direct_index;
 524                 break;
 525         default:
 526                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 527         }
 528
 529         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 530 }
 531
 532 /*
 533  * The maximum depth is four.
 534  * Offset[0] will have raw inode offset.
 535  */
 536 static int get_node_path(struct inode *inode, long block,
 537                                 int offset[4], unsigned int noffset[4])
 538 {
 539         const long direct_index = ADDRS_PER_INODE(inode);
 540         const long direct_blks = ADDRS_PER_BLOCK;
 541         const long dptrs_per_blk = NIDS_PER_BLOCK;
 542         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 543         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 544         int n = 0;
 545         int level = 0;
 546
 547         noffset[0] = 0;
 548
 549         if (block < direct_index) {
 550                 offset[n] = block;
 551                 goto got;
 552         }
 553         block -= direct_index;
 554         if (block < direct_blks) {
 555                 offset[n++] = NODE_DIR1_BLOCK;
 556                 noffset[n] = 1;
 557                 offset[n] = block;
 558                 level = 1;
 559                 goto got;
 560         }
 561         block -= direct_blks;
 562         if (block < direct_blks) {
 563                 offset[n++] = NODE_DIR2_BLOCK;
 564                 noffset[n] = 2;
 565                 offset[n] = block;
 566                 level = 1;
 567                 goto got;
 568         }
 569         block -= direct_blks;
 570         if (block < indirect_blks) {
 571                 offset[n++] = NODE_IND1_BLOCK;
 572                 noffset[n] = 3;
 573                 offset[n++] = block / direct_blks;
 574                 noffset[n] = 4 + offset[n - 1];
 575                 offset[n] = block % direct_blks;
 576                 level = 2;
 577                 goto got;
 578         }
 579         block -= indirect_blks;
 580         if (block < indirect_blks) {
 581                 offset[n++] = NODE_IND2_BLOCK;
 582                 noffset[n] = 4 + dptrs_per_blk;
 583                 offset[n++] = block / direct_blks;
 584                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 585                 offset[n] = block % direct_blks;
 586                 level = 2;
 587                 goto got;
 588         }
 589         block -= indirect_blks;
 590         if (block < dindirect_blks) {
 591                 offset[n++] = NODE_DIND_BLOCK;
 592                 noffset[n] = 5 + (dptrs_per_blk * 2);
 593                 offset[n++] = block / indirect_blks;
 594                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 595                               offset[n - 1] * (dptrs_per_blk + 1);
 596                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 597                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 598                               offset[n - 2] * (dptrs_per_blk + 1) +
 599                               offset[n - 1];
 600                 offset[n] = block % direct_blks;
 601                 level = 3;
 602                 goto got;
 603         } else {
 604                 return -E2BIG;
 605         }
 606 got:
 607         return level;
 608 }
 609
 610 /*
 611  * Caller should call f2fs_put_dnode(dn).
 612  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 613  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 614  * In the case of RDONLY_NODE, we don't need to care about mutex.
 615  */
 616 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 617 {
 618         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 619         struct page *npage[4];
 620         struct page *parent = NULL;
 621         int offset[4];
 622         unsigned int noffset[4];
 623         nid_t nids[4];
 624         int level, i = 0;
 625         int err = 0;
 626
 627         level = get_node_path(dn->inode, index, offset, noffset);
 628         if (level < 0)
 629                 return level;
 630
 631         nids[0] = dn->inode->i_ino;
 632         npage[0] = dn->inode_page;
 633
 634         if (!npage[0]) {
 635                 npage[0] = f2fs_get_node_page(sbi, nids[0]);
 636                 if (IS_ERR(npage[0]))
 637                         return PTR_ERR(npage[0]);
 638         }
 639
 640         /* if inline_data is set, should not report any block indices */
 641         if (f2fs_has_inline_data(dn->inode) && index) {
 642                 err = -ENOENT;
 643                 f2fs_put_page(npage[0], 1);
 644                 goto release_out;
 645         }
 646
 647         parent = npage[0];
 648         if (level != 0)
 649                 nids[1] = get_nid(parent, offset[0], true);
 650         dn->inode_page = npage[0];
 651         dn->inode_page_locked = true;
 652
 653         /* get indirect or direct nodes */
 654         for (i = 1; i <= level; i++) {
 655                 bool done = false;
 656
 657                 if (!nids[i] && mode == ALLOC_NODE) {
 658                         /* alloc new node */
 659                         if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
 660                                 err = -ENOSPC;
 661                                 goto release_pages;
 662                         }
 663
 664                         dn->nid = nids[i];
 665                         npage[i] = f2fs_new_node_page(dn, noffset[i]);
 666                         if (IS_ERR(npage[i])) {
 667                                 f2fs_alloc_nid_failed(sbi, nids[i]);
 668                                 err = PTR_ERR(npage[i]);
 669                                 goto release_pages;
 670                         }
 671
 672                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 673                         f2fs_alloc_nid_done(sbi, nids[i]);
 674                         done = true;
 675                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 676                         npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
 677                         if (IS_ERR(npage[i])) {
 678                                 err = PTR_ERR(npage[i]);
 679                                 goto release_pages;
 680                         }
 681                         done = true;
 682                 }
 683                 if (i == 1) {
 684                         dn->inode_page_locked = false;
 685                         unlock_page(parent);
 686                 } else {
 687                         f2fs_put_page(parent, 1);
 688                 }
 689
 690                 if (!done) {
 691                         npage[i] = f2fs_get_node_page(sbi, nids[i]);
 692                         if (IS_ERR(npage[i])) {
 693                                 err = PTR_ERR(npage[i]);
 694                                 f2fs_put_page(npage[0], 0);
 695                                 goto release_out;
 696                         }
 697                 }
 698                 if (i < level) {
 699                         parent = npage[i];
 700                         nids[i + 1] = get_nid(parent, offset[i], false);
 701                 }
 702         }
 703         dn->nid = nids[level];
 704         dn->ofs_in_node = offset[level];
 705         dn->node_page = npage[level];
 706         dn->data_blkaddr = datablock_addr(dn->inode,
 707                                 dn->node_page, dn->ofs_in_node);
 708         return 0;
 709
 710 release_pages:
 711         f2fs_put_page(parent, 1);
 712         if (i > 1)
 713                 f2fs_put_page(npage[0], 0);
 714 release_out:
 715         dn->inode_page = NULL;
 716         dn->node_page = NULL;
 717         if (err == -ENOENT) {
 718                 dn->cur_level = i;
 719                 dn->max_level = level;
 720                 dn->ofs_in_node = offset[level];
 721         }
 722         return err;
 723 }
 724
 725 static void truncate_node(struct dnode_of_data *dn)
 726 {
 727         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 728         struct node_info ni;
 729
 730         f2fs_get_node_info(sbi, dn->nid, &ni);
 731
 732         /* Deallocate node address */
 733         f2fs_invalidate_blocks(sbi, ni.blk_addr);
 734         dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
 735         set_node_addr(sbi, &ni, NULL_ADDR, false);
 736
 737         if (dn->nid == dn->inode->i_ino) {
 738                 f2fs_remove_orphan_inode(sbi, dn->nid);
 739                 dec_valid_inode_count(sbi);
 740                 f2fs_inode_synced(dn->inode);
 741         }
 742
 743         clear_node_page_dirty(dn->node_page);
 744         set_sbi_flag(sbi, SBI_IS_DIRTY);
 745
 746         f2fs_put_page(dn->node_page, 1);
 747
 748         invalidate_mapping_pages(NODE_MAPPING(sbi),
 749                         dn->node_page->index, dn->node_page->index);
 750
 751         dn->node_page = NULL;
 752         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 753 }
 754
 755 static int truncate_dnode(struct dnode_of_data *dn)
 756 {
 757         struct page *page;
 758
 759         if (dn->nid == 0)
 760                 return 1;
 761
 762         /* get direct node */
 763         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 764         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 765                 return 1;
 766         else if (IS_ERR(page))
 767                 return PTR_ERR(page);
 768
 769         /* Make dnode_of_data for parameter */
 770         dn->node_page = page;
 771         dn->ofs_in_node = 0;
 772         f2fs_truncate_data_blocks(dn);
 773         truncate_node(dn);
 774         return 1;
 775 }
 776
 777 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 778                                                 int ofs, int depth)
 779 {
 780         struct dnode_of_data rdn = *dn;
 781         struct page *page;
 782         struct f2fs_node *rn;
 783         nid_t child_nid;
 784         unsigned int child_nofs;
 785         int freed = 0;
 786         int i, ret;
 787
 788         if (dn->nid == 0)
 789                 return NIDS_PER_BLOCK + 1;
 790
 791         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 792
 793         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 794         if (IS_ERR(page)) {
 795                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 796                 return PTR_ERR(page);
 797         }
 798
 799         f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 800
 801         rn = F2FS_NODE(page);
 802         if (depth < 3) {
 803                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 804                         child_nid = le32_to_cpu(rn->in.nid[i]);
 805                         if (child_nid == 0)
 806                                 continue;
 807                         rdn.nid = child_nid;
 808                         ret = truncate_dnode(&rdn);
 809                         if (ret < 0)
 810                                 goto out_err;
 811                         if (set_nid(page, i, 0, false))
 812                                 dn->node_changed = true;
 813                 }
 814         } else {
 815                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 816                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 817                         child_nid = le32_to_cpu(rn->in.nid[i]);
 818                         if (child_nid == 0) {
 819                                 child_nofs += NIDS_PER_BLOCK + 1;
 820                                 continue;
 821                         }
 822                         rdn.nid = child_nid;
 823                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 824                         if (ret == (NIDS_PER_BLOCK + 1)) {
 825                                 if (set_nid(page, i, 0, false))
 826                                         dn->node_changed = true;
 827                                 child_nofs += ret;
 828                         } else if (ret < 0 && ret != -ENOENT) {
 829                                 goto out_err;
 830                         }
 831                 }
 832                 freed = child_nofs;
 833         }
 834
 835         if (!ofs) {
 836                 /* remove current indirect node */
 837                 dn->node_page = page;
 838                 truncate_node(dn);
 839                 freed++;
 840         } else {
 841                 f2fs_put_page(page, 1);
 842         }
 843         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 844         return freed;
 845
 846 out_err:
 847         f2fs_put_page(page, 1);
 848         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 849         return ret;
 850 }
 851
 852 static int truncate_partial_nodes(struct dnode_of_data *dn,
 853                         struct f2fs_inode *ri, int *offset, int depth)
 854 {
 855         struct page *pages[2];
 856         nid_t nid[3];
 857         nid_t child_nid;
 858         int err = 0;
 859         int i;
 860         int idx = depth - 2;
 861
 862         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 863         if (!nid[0])
 864                 return 0;
 865
 866         /* get indirect nodes in the path */
 867         for (i = 0; i < idx + 1; i++) {
 868                 /* reference count'll be increased */
 869                 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 870                 if (IS_ERR(pages[i])) {
 871                         err = PTR_ERR(pages[i]);
 872                         idx = i - 1;
 873                         goto fail;
 874                 }
 875                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 876         }
 877
 878         f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 879
 880         /* free direct nodes linked to a partial indirect node */
 881         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 882                 child_nid = get_nid(pages[idx], i, false);
 883                 if (!child_nid)
 884                         continue;
 885                 dn->nid = child_nid;
 886                 err = truncate_dnode(dn);
 887                 if (err < 0)
 888                         goto fail;
 889                 if (set_nid(pages[idx], i, 0, false))
 890                         dn->node_changed = true;
 891         }
 892
 893         if (offset[idx + 1] == 0) {
 894                 dn->node_page = pages[idx];
 895                 dn->nid = nid[idx];
 896                 truncate_node(dn);
 897         } else {
 898                 f2fs_put_page(pages[idx], 1);
 899         }
 900         offset[idx]++;
 901         offset[idx + 1] = 0;
 902         idx--;
 903 fail:
 904         for (i = idx; i >= 0; i--)
 905                 f2fs_put_page(pages[i], 1);
 906
 907         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 908
 909         return err;
 910 }
 911
 912 /*
 913  * All the block addresses of data and nodes should be nullified.
 914  */
 915 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
 916 {
 917         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 918         int err = 0, cont = 1;
 919         int level, offset[4], noffset[4];
 920         unsigned int nofs = 0;
 921         struct f2fs_inode *ri;
 922         struct dnode_of_data dn;
 923         struct page *page;
 924
 925         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 926
 927         level = get_node_path(inode, from, offset, noffset);
 928         if (level < 0)
 929                 return level;
 930
 931         page = f2fs_get_node_page(sbi, inode->i_ino);
 932         if (IS_ERR(page)) {
 933                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 934                 return PTR_ERR(page);
 935         }
 936
 937         set_new_dnode(&dn, inode, page, NULL, 0);
 938         unlock_page(page);
 939
 940         ri = F2FS_INODE(page);
 941         switch (level) {
 942         case 0:
 943         case 1:
 944                 nofs = noffset[1];
 945                 break;
 946         case 2:
 947                 nofs = noffset[1];
 948                 if (!offset[level - 1])
 949                         goto skip_partial;
 950                 err = truncate_partial_nodes(&dn, ri, offset, level);
 951                 if (err < 0 && err != -ENOENT)
 952                         goto fail;
 953                 nofs += 1 + NIDS_PER_BLOCK;
 954                 break;
 955         case 3:
 956                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 957                 if (!offset[level - 1])
 958                         goto skip_partial;
 959                 err = truncate_partial_nodes(&dn, ri, offset, level);
 960                 if (err < 0 && err != -ENOENT)
 961                         goto fail;
 962                 break;
 963         default:
 964                 BUG();
 965         }
 966
 967 skip_partial:
 968         while (cont) {
 969                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 970                 switch (offset[0]) {
 971                 case NODE_DIR1_BLOCK:
 972                 case NODE_DIR2_BLOCK:
 973                         err = truncate_dnode(&dn);
 974                         break;
 975
 976                 case NODE_IND1_BLOCK:
 977                 case NODE_IND2_BLOCK:
 978                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 979                         break;
 980
 981                 case NODE_DIND_BLOCK:
 982                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 983                         cont = 0;
 984                         break;
 985
 986                 default:
 987                         BUG();
 988                 }
 989                 if (err < 0 && err != -ENOENT)
 990                         goto fail;
 991                 if (offset[1] == 0 &&
 992                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 993                         lock_page(page);
 994                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
 995                         f2fs_wait_on_page_writeback(page, NODE, true);
 996                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 997                         set_page_dirty(page);
 998                         unlock_page(page);
 999                 }
1000                 offset[1] = 0;
1001                 offset[0]++;
1002                 nofs += err;
1003         }
1004 fail:
1005         f2fs_put_page(page, 0);
1006         trace_f2fs_truncate_inode_blocks_exit(inode, err);
1007         return err > 0 ? 0 : err;
1008 }
1009
1010 /* caller must lock inode page */
1011 int f2fs_truncate_xattr_node(struct inode *inode)
1012 {
1013         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1014         nid_t nid = F2FS_I(inode)->i_xattr_nid;
1015         struct dnode_of_data dn;
1016         struct page *npage;
1017
1018         if (!nid)
1019                 return 0;
1020
1021         npage = f2fs_get_node_page(sbi, nid);
1022         if (IS_ERR(npage))
1023                 return PTR_ERR(npage);
1024
1025         f2fs_i_xnid_write(inode, 0);
1026
1027         set_new_dnode(&dn, inode, NULL, npage, nid);
1028         truncate_node(&dn);
1029         return 0;
1030 }
1031
1032 /*
1033  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1034  * f2fs_unlock_op().
1035  */
1036 int f2fs_remove_inode_page(struct inode *inode)
1037 {
1038         struct dnode_of_data dn;
1039         int err;
1040
1041         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1042         err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1043         if (err)
1044                 return err;
1045
1046         err = f2fs_truncate_xattr_node(inode);
1047         if (err) {
1048                 f2fs_put_dnode(&dn);
1049                 return err;
1050         }
1051
1052         /* remove potential inline_data blocks */
1053         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1054                                 S_ISLNK(inode->i_mode))
1055                 f2fs_truncate_data_blocks_range(&dn, 1);
1056
1057         /* 0 is possible, after f2fs_new_inode() has failed */
1058         f2fs_bug_on(F2FS_I_SB(inode),
1059                         inode->i_blocks != 0 && inode->i_blocks != 8);
1060
1061         /* will put inode & node pages */
1062         truncate_node(&dn);
1063         return 0;
1064 }
1065
1066 struct page *f2fs_new_inode_page(struct inode *inode)
1067 {
1068         struct dnode_of_data dn;
1069
1070         /* allocate inode page for new inode */
1071         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1072
1073         /* caller should f2fs_put_page(page, 1); */
1074         return f2fs_new_node_page(&dn, 0);
1075 }
1076
1077 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1078 {
1079         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1080         struct node_info new_ni;
1081         struct page *page;
1082         int err;
1083
1084         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1085                 return ERR_PTR(-EPERM);
1086
1087         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1088         if (!page)
1089                 return ERR_PTR(-ENOMEM);
1090
1091         if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1092                 goto fail;
1093
1094 #ifdef CONFIG_F2FS_CHECK_FS
1095         f2fs_get_node_info(sbi, dn->nid, &new_ni);
1096         f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1097 #endif
1098         new_ni.nid = dn->nid;
1099         new_ni.ino = dn->inode->i_ino;
1100         new_ni.blk_addr = NULL_ADDR;
1101         new_ni.flag = 0;
1102         new_ni.version = 0;
1103         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1104
1105         f2fs_wait_on_page_writeback(page, NODE, true);
1106         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1107         set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1108         if (!PageUptodate(page))
1109                 SetPageUptodate(page);
1110         if (set_page_dirty(page))
1111                 dn->node_changed = true;
1112
1113         if (f2fs_has_xattr_block(ofs))
1114                 f2fs_i_xnid_write(dn->inode, dn->nid);
1115
1116         if (ofs == 0)
1117                 inc_valid_inode_count(sbi);
1118         return page;
1119
1120 fail:
1121         clear_node_page_dirty(page);
1122         f2fs_put_page(page, 1);
1123         return ERR_PTR(err);
1124 }
1125
1126 /*
1127  * Caller should do after getting the following values.
1128  * 0: f2fs_put_page(page, 0)
1129  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1130  */
1131 static int read_node_page(struct page *page, int op_flags)
1132 {
1133         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1134         struct node_info ni;
1135         struct f2fs_io_info fio = {
1136                 .sbi = sbi,
1137                 .type = NODE,
1138                 .op = REQ_OP_READ,
1139                 .op_flags = op_flags,
1140                 .page = page,
1141                 .encrypted_page = NULL,
1142         };
1143
1144         if (PageUptodate(page))
1145                 return LOCKED_PAGE;
1146
1147         f2fs_get_node_info(sbi, page->index, &ni);
1148
1149         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1150                 ClearPageUptodate(page);
1151                 return -ENOENT;
1152         }
1153
1154         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1155         return f2fs_submit_page_bio(&fio);
1156 }
1157
1158 /*
1159  * Readahead a node page
1160  */
1161 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1162 {
1163         struct page *apage;
1164         int err;
1165
1166         if (!nid)
1167                 return;
1168         if (f2fs_check_nid_range(sbi, nid))
1169                 return;
1170
1171         rcu_read_lock();
1172         apage = radix_tree_lookup(&NODE_MAPPING(sbi)->i_pages, nid);
1173         rcu_read_unlock();
1174         if (apage)
1175                 return;
1176
1177         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1178         if (!apage)
1179                 return;
1180
1181         err = read_node_page(apage, REQ_RAHEAD);
1182         f2fs_put_page(apage, err ? 1 : 0);
1183 }
1184
1185 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1186                                         struct page *parent, int start)
1187 {
1188         struct page *page;
1189         int err;
1190
1191         if (!nid)
1192                 return ERR_PTR(-ENOENT);
1193         if (f2fs_check_nid_range(sbi, nid))
1194                 return ERR_PTR(-EINVAL);
1195 repeat:
1196         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1197         if (!page)
1198                 return ERR_PTR(-ENOMEM);
1199
1200         err = read_node_page(page, 0);
1201         if (err < 0) {
1202                 f2fs_put_page(page, 1);
1203                 return ERR_PTR(err);
1204         } else if (err == LOCKED_PAGE) {
1205                 err = 0;
1206                 goto page_hit;
1207         }
1208
1209         if (parent)
1210                 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1211
1212         lock_page(page);
1213
1214         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1215                 f2fs_put_page(page, 1);
1216                 goto repeat;
1217         }
1218
1219         if (unlikely(!PageUptodate(page))) {
1220                 err = -EIO;
1221                 goto out_err;
1222         }
1223
1224         if (!f2fs_inode_chksum_verify(sbi, page)) {
1225                 err = -EBADMSG;
1226                 goto out_err;
1227         }
1228 page_hit:
1229         if(unlikely(nid != nid_of_node(page))) {
1230                 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1231                         "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1232                         nid, nid_of_node(page), ino_of_node(page),
1233                         ofs_of_node(page), cpver_of_node(page),
1234                         next_blkaddr_of_node(page));
1235                 err = -EINVAL;
1236 out_err:
1237                 ClearPageUptodate(page);
1238                 f2fs_put_page(page, 1);
1239                 return ERR_PTR(err);
1240         }
1241         return page;
1242 }
1243
1244 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1245 {
1246         return __get_node_page(sbi, nid, NULL, 0);
1247 }
1248
1249 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1250 {
1251         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1252         nid_t nid = get_nid(parent, start, false);
1253
1254         return __get_node_page(sbi, nid, parent, start);
1255 }
1256
1257 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1258 {
1259         struct inode *inode;
1260         struct page *page;
1261         int ret;
1262
1263         /* should flush inline_data before evict_inode */
1264         inode = ilookup(sbi->sb, ino);
1265         if (!inode)
1266                 return;
1267
1268         page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1269                                         FGP_LOCK|FGP_NOWAIT, 0);
1270         if (!page)
1271                 goto iput_out;
1272
1273         if (!PageUptodate(page))
1274                 goto page_out;
1275
1276         if (!PageDirty(page))
1277                 goto page_out;
1278
1279         if (!clear_page_dirty_for_io(page))
1280                 goto page_out;
1281
1282         ret = f2fs_write_inline_data(inode, page);
1283         inode_dec_dirty_pages(inode);
1284         f2fs_remove_dirty_inode(inode);
1285         if (ret)
1286                 set_page_dirty(page);
1287 page_out:
1288         f2fs_put_page(page, 1);
1289 iput_out:
1290         iput(inode);
1291 }
1292
1293 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1294 {
1295         pgoff_t index;
1296         struct pagevec pvec;
1297         struct page *last_page = NULL;
1298         int nr_pages;
1299
1300         pagevec_init(&pvec);
1301         index = 0;
1302
1303         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1304                                 PAGECACHE_TAG_DIRTY))) {
1305                 int i;
1306
1307                 for (i = 0; i < nr_pages; i++) {
1308                         struct page *page = pvec.pages[i];
1309
1310                         if (unlikely(f2fs_cp_error(sbi))) {
1311                                 f2fs_put_page(last_page, 0);
1312                                 pagevec_release(&pvec);
1313                                 return ERR_PTR(-EIO);
1314                         }
1315
1316                         if (!IS_DNODE(page) || !is_cold_node(page))
1317                                 continue;
1318                         if (ino_of_node(page) != ino)
1319                                 continue;
1320
1321                         lock_page(page);
1322
1323                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1324 continue_unlock:
1325                                 unlock_page(page);
1326                                 continue;
1327                         }
1328                         if (ino_of_node(page) != ino)
1329                                 goto continue_unlock;
1330
1331                         if (!PageDirty(page)) {
1332                                 /* someone wrote it for us */
1333                                 goto continue_unlock;
1334                         }
1335
1336                         if (last_page)
1337                                 f2fs_put_page(last_page, 0);
1338
1339                         get_page(page);
1340                         last_page = page;
1341                         unlock_page(page);
1342                 }
1343                 pagevec_release(&pvec);
1344                 cond_resched();
1345         }
1346         return last_page;
1347 }
1348
1349 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1350                                 struct writeback_control *wbc, bool do_balance,
1351                                 enum iostat_type io_type)
1352 {
1353         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1354         nid_t nid;
1355         struct node_info ni;
1356         struct f2fs_io_info fio = {
1357                 .sbi = sbi,
1358                 .ino = ino_of_node(page),
1359                 .type = NODE,
1360                 .op = REQ_OP_WRITE,
1361                 .op_flags = wbc_to_write_flags(wbc),
1362                 .page = page,
1363                 .encrypted_page = NULL,
1364                 .submitted = false,
1365                 .io_type = io_type,
1366                 .io_wbc = wbc,
1367         };
1368
1369         trace_f2fs_writepage(page, NODE);
1370
1371         if (unlikely(f2fs_cp_error(sbi)))
1372                 goto redirty_out;
1373
1374         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1375                 goto redirty_out;
1376
1377         /* get old block addr of this node page */
1378         nid = nid_of_node(page);
1379         f2fs_bug_on(sbi, page->index != nid);
1380
1381         if (wbc->for_reclaim) {
1382                 if (!down_read_trylock(&sbi->node_write))
1383                         goto redirty_out;
1384         } else {
1385                 down_read(&sbi->node_write);
1386         }
1387
1388         f2fs_get_node_info(sbi, nid, &ni);
1389
1390         /* This page is already truncated */
1391         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1392                 ClearPageUptodate(page);
1393                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1394                 up_read(&sbi->node_write);
1395                 unlock_page(page);
1396                 return 0;
1397         }
1398
1399         if (atomic && !test_opt(sbi, NOBARRIER))
1400                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1401
1402         set_page_writeback(page);
1403         ClearPageError(page);
1404         fio.old_blkaddr = ni.blk_addr;
1405         f2fs_do_write_node_page(nid, &fio);
1406         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1407         dec_page_count(sbi, F2FS_DIRTY_NODES);
1408         up_read(&sbi->node_write);
1409
1410         if (wbc->for_reclaim) {
1411                 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1412                                                 page->index, NODE);
1413                 submitted = NULL;
1414         }
1415
1416         unlock_page(page);
1417
1418         if (unlikely(f2fs_cp_error(sbi))) {
1419                 f2fs_submit_merged_write(sbi, NODE);
1420                 submitted = NULL;
1421         }
1422         if (submitted)
1423                 *submitted = fio.submitted;
1424
1425         if (do_balance)
1426                 f2fs_balance_fs(sbi, false);
1427         return 0;
1428
1429 redirty_out:
1430         redirty_page_for_writepage(wbc, page);
1431         return AOP_WRITEPAGE_ACTIVATE;
1432 }
1433
1434 void f2fs_move_node_page(struct page *node_page, int gc_type)
1435 {
1436         if (gc_type == FG_GC) {
1437                 struct writeback_control wbc = {
1438                         .sync_mode = WB_SYNC_ALL,
1439                         .nr_to_write = 1,
1440                         .for_reclaim = 0,
1441                 };
1442
1443                 set_page_dirty(node_page);
1444                 f2fs_wait_on_page_writeback(node_page, NODE, true);
1445
1446                 f2fs_bug_on(F2FS_P_SB(node_page), PageWriteback(node_page));
1447                 if (!clear_page_dirty_for_io(node_page))
1448                         goto out_page;
1449
1450                 if (__write_node_page(node_page, false, NULL,
1451                                         &wbc, false, FS_GC_NODE_IO))
1452                         unlock_page(node_page);
1453                 goto release_page;
1454         } else {
1455                 /* set page dirty and write it */
1456                 if (!PageWriteback(node_page))
1457                         set_page_dirty(node_page);
1458         }
1459 out_page:
1460         unlock_page(node_page);
1461 release_page:
1462         f2fs_put_page(node_page, 0);
1463 }
1464
1465 static int f2fs_write_node_page(struct page *page,
1466                                 struct writeback_control *wbc)
1467 {
1468         return __write_node_page(page, false, NULL, wbc, false, FS_NODE_IO);
1469 }
1470
1471 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1472                         struct writeback_control *wbc, bool atomic)
1473 {
1474         pgoff_t index;
1475         pgoff_t last_idx = ULONG_MAX;
1476         struct pagevec pvec;
1477         int ret = 0;
1478         struct page *last_page = NULL;
1479         bool marked = false;
1480         nid_t ino = inode->i_ino;
1481         int nr_pages;
1482
1483         if (atomic) {
1484                 last_page = last_fsync_dnode(sbi, ino);
1485                 if (IS_ERR_OR_NULL(last_page))
1486                         return PTR_ERR_OR_ZERO(last_page);
1487         }
1488 retry:
1489         pagevec_init(&pvec);
1490         index = 0;
1491
1492         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1493                                 PAGECACHE_TAG_DIRTY))) {
1494                 int i;
1495
1496                 for (i = 0; i < nr_pages; i++) {
1497                         struct page *page = pvec.pages[i];
1498                         bool submitted = false;
1499
1500                         if (unlikely(f2fs_cp_error(sbi))) {
1501                                 f2fs_put_page(last_page, 0);
1502                                 pagevec_release(&pvec);
1503                                 ret = -EIO;
1504                                 goto out;
1505                         }
1506
1507                         if (!IS_DNODE(page) || !is_cold_node(page))
1508                                 continue;
1509                         if (ino_of_node(page) != ino)
1510                                 continue;
1511
1512                         lock_page(page);
1513
1514                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1515 continue_unlock:
1516                                 unlock_page(page);
1517                                 continue;
1518                         }
1519                         if (ino_of_node(page) != ino)
1520                                 goto continue_unlock;
1521
1522                         if (!PageDirty(page) && page != last_page) {
1523                                 /* someone wrote it for us */
1524                                 goto continue_unlock;
1525                         }
1526
1527                         f2fs_wait_on_page_writeback(page, NODE, true);
1528                         BUG_ON(PageWriteback(page));
1529
1530                         set_fsync_mark(page, 0);
1531                         set_dentry_mark(page, 0);
1532
1533                         if (!atomic || page == last_page) {
1534                                 set_fsync_mark(page, 1);
1535                                 if (IS_INODE(page)) {
1536                                         if (is_inode_flag_set(inode,
1537                                                                 FI_DIRTY_INODE))
1538                                                 f2fs_update_inode(inode, page);
1539                                         set_dentry_mark(page,
1540                                                 f2fs_need_dentry_mark(sbi, ino));
1541                                 }
1542                                 /*  may be written by other thread */
1543                                 if (!PageDirty(page))
1544                                         set_page_dirty(page);
1545                         }
1546
1547                         if (!clear_page_dirty_for_io(page))
1548                                 goto continue_unlock;
1549
1550                         ret = __write_node_page(page, atomic &&
1551                                                 page == last_page,
1552                                                 &submitted, wbc, true,
1553                                                 FS_NODE_IO);
1554                         if (ret) {
1555                                 unlock_page(page);
1556                                 f2fs_put_page(last_page, 0);
1557                                 break;
1558                         } else if (submitted) {
1559                                 last_idx = page->index;
1560                         }
1561
1562                         if (page == last_page) {
1563                                 f2fs_put_page(page, 0);
1564                                 marked = true;
1565                                 break;
1566                         }
1567                 }
1568                 pagevec_release(&pvec);
1569                 cond_resched();
1570
1571                 if (ret || marked)
1572                         break;
1573         }
1574         if (!ret && atomic && !marked) {
1575                 f2fs_msg(sbi->sb, KERN_DEBUG,
1576                         "Retry to write fsync mark: ino=%u, idx=%lx",
1577                                         ino, last_page->index);
1578                 lock_page(last_page);
1579                 f2fs_wait_on_page_writeback(last_page, NODE, true);
1580                 set_page_dirty(last_page);
1581                 unlock_page(last_page);
1582                 goto retry;
1583         }
1584 out:
1585         if (last_idx != ULONG_MAX)
1586                 f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1587         return ret ? -EIO: 0;
1588 }
1589
1590 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1591                                 struct writeback_control *wbc,
1592                                 bool do_balance, enum iostat_type io_type)
1593 {
1594         pgoff_t index;
1595         struct pagevec pvec;
1596         int step = 0;
1597         int nwritten = 0;
1598         int ret = 0;
1599         int nr_pages, done = 0;
1600
1601         pagevec_init(&pvec);
1602
1603 next_step:
1604         index = 0;
1605
1606         while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1607                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1608                 int i;
1609
1610                 for (i = 0; i < nr_pages; i++) {
1611                         struct page *page = pvec.pages[i];
1612                         bool submitted = false;
1613
1614                         /* give a priority to WB_SYNC threads */
1615                         if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1616                                         wbc->sync_mode == WB_SYNC_NONE) {
1617                                 done = 1;
1618                                 break;
1619                         }
1620
1621                         /*
1622                          * flushing sequence with step:
1623                          * 0. indirect nodes
1624                          * 1. dentry dnodes
1625                          * 2. file dnodes
1626                          */
1627                         if (step == 0 && IS_DNODE(page))
1628                                 continue;
1629                         if (step == 1 && (!IS_DNODE(page) ||
1630                                                 is_cold_node(page)))
1631                                 continue;
1632                         if (step == 2 && (!IS_DNODE(page) ||
1633                                                 !is_cold_node(page)))
1634                                 continue;
1635 lock_node:
1636                         if (!trylock_page(page))
1637                                 continue;
1638
1639                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1640 continue_unlock:
1641                                 unlock_page(page);
1642                                 continue;
1643                         }
1644
1645                         if (!PageDirty(page)) {
1646                                 /* someone wrote it for us */
1647                                 goto continue_unlock;
1648                         }
1649
1650                         /* flush inline_data */
1651                         if (is_inline_node(page)) {
1652                                 clear_inline_node(page);
1653                                 unlock_page(page);
1654                                 flush_inline_data(sbi, ino_of_node(page));
1655                                 goto lock_node;
1656                         }
1657
1658                         f2fs_wait_on_page_writeback(page, NODE, true);
1659
1660                         BUG_ON(PageWriteback(page));
1661                         if (!clear_page_dirty_for_io(page))
1662                                 goto continue_unlock;
1663
1664                         set_fsync_mark(page, 0);
1665                         set_dentry_mark(page, 0);
1666
1667                         ret = __write_node_page(page, false, &submitted,
1668                                                 wbc, do_balance, io_type);
1669                         if (ret)
1670                                 unlock_page(page);
1671                         else if (submitted)
1672                                 nwritten++;
1673
1674                         if (--wbc->nr_to_write == 0)
1675                                 break;
1676                 }
1677                 pagevec_release(&pvec);
1678                 cond_resched();
1679
1680                 if (wbc->nr_to_write == 0) {
1681                         step = 2;
1682                         break;
1683                 }
1684         }
1685
1686         if (step < 2) {
1687                 step++;
1688                 goto next_step;
1689         }
1690
1691         if (nwritten)
1692                 f2fs_submit_merged_write(sbi, NODE);
1693
1694         if (unlikely(f2fs_cp_error(sbi)))
1695                 return -EIO;
1696         return ret;
1697 }
1698
1699 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1700 {
1701         pgoff_t index = 0;
1702         struct pagevec pvec;
1703         int ret2, ret = 0;
1704         int nr_pages;
1705
1706         pagevec_init(&pvec);
1707
1708         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1709                                 PAGECACHE_TAG_WRITEBACK))) {
1710                 int i;
1711
1712                 for (i = 0; i < nr_pages; i++) {
1713                         struct page *page = pvec.pages[i];
1714
1715                         if (ino && ino_of_node(page) == ino) {
1716                                 f2fs_wait_on_page_writeback(page, NODE, true);
1717                                 if (TestClearPageError(page))
1718                                         ret = -EIO;
1719                         }
1720                 }
1721                 pagevec_release(&pvec);
1722                 cond_resched();
1723         }
1724
1725         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1726         if (!ret)
1727                 ret = ret2;
1728         return ret;
1729 }
1730
1731 static int f2fs_write_node_pages(struct address_space *mapping,
1732                             struct writeback_control *wbc)
1733 {
1734         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1735         struct blk_plug plug;
1736         long diff;
1737
1738         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1739                 goto skip_write;
1740
1741         /* balancing f2fs's metadata in background */
1742         f2fs_balance_fs_bg(sbi);
1743
1744         /* collect a number of dirty node pages and write together */
1745         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1746                 goto skip_write;
1747
1748         if (wbc->sync_mode == WB_SYNC_ALL)
1749                 atomic_inc(&sbi->wb_sync_req[NODE]);
1750         else if (atomic_read(&sbi->wb_sync_req[NODE]))
1751                 goto skip_write;
1752
1753         trace_f2fs_writepages(mapping->host, wbc, NODE);
1754
1755         diff = nr_pages_to_write(sbi, NODE, wbc);
1756         blk_start_plug(&plug);
1757         f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1758         blk_finish_plug(&plug);
1759         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1760
1761         if (wbc->sync_mode == WB_SYNC_ALL)
1762                 atomic_dec(&sbi->wb_sync_req[NODE]);
1763         return 0;
1764
1765 skip_write:
1766         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1767         trace_f2fs_writepages(mapping->host, wbc, NODE);
1768         return 0;
1769 }
1770
1771 static int f2fs_set_node_page_dirty(struct page *page)
1772 {
1773         trace_f2fs_set_page_dirty(page, NODE);
1774
1775         if (!PageUptodate(page))
1776                 SetPageUptodate(page);
1777         if (!PageDirty(page)) {
1778                 __set_page_dirty_nobuffers(page);
1779                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1780                 SetPagePrivate(page);
1781                 f2fs_trace_pid(page);
1782                 return 1;
1783         }
1784         return 0;
1785 }
1786
1787 /*
1788  * Structure of the f2fs node operations
1789  */
1790 const struct address_space_operations f2fs_node_aops = {
1791         .writepage      = f2fs_write_node_page,
1792         .writepages     = f2fs_write_node_pages,
1793         .set_page_dirty = f2fs_set_node_page_dirty,
1794         .invalidatepage = f2fs_invalidate_page,
1795         .releasepage    = f2fs_release_page,
1796 #ifdef CONFIG_MIGRATION
1797         .migratepage    = f2fs_migrate_page,
1798 #endif
1799 };
1800
1801 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1802                                                 nid_t n)
1803 {
1804         return radix_tree_lookup(&nm_i->free_nid_root, n);
1805 }
1806
1807 static int __insert_free_nid(struct f2fs_sb_info *sbi,
1808                         struct free_nid *i, enum nid_state state)
1809 {
1810         struct f2fs_nm_info *nm_i = NM_I(sbi);
1811
1812         int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1813         if (err)
1814                 return err;
1815
1816         f2fs_bug_on(sbi, state != i->state);
1817         nm_i->nid_cnt[state]++;
1818         if (state == FREE_NID)
1819                 list_add_tail(&i->list, &nm_i->free_nid_list);
1820         return 0;
1821 }
1822
1823 static void __remove_free_nid(struct f2fs_sb_info *sbi,
1824                         struct free_nid *i, enum nid_state state)
1825 {
1826         struct f2fs_nm_info *nm_i = NM_I(sbi);
1827
1828         f2fs_bug_on(sbi, state != i->state);
1829         nm_i->nid_cnt[state]--;
1830         if (state == FREE_NID)
1831                 list_del(&i->list);
1832         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1833 }
1834
1835 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
1836                         enum nid_state org_state, enum nid_state dst_state)
1837 {
1838         struct f2fs_nm_info *nm_i = NM_I(sbi);
1839
1840         f2fs_bug_on(sbi, org_state != i->state);
1841         i->state = dst_state;
1842         nm_i->nid_cnt[org_state]--;
1843         nm_i->nid_cnt[dst_state]++;
1844
1845         switch (dst_state) {
1846         case PREALLOC_NID:
1847                 list_del(&i->list);
1848                 break;
1849         case FREE_NID:
1850                 list_add_tail(&i->list, &nm_i->free_nid_list);
1851                 break;
1852         default:
1853                 BUG_ON(1);
1854         }
1855 }
1856
1857 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
1858                                                         bool set, bool build)
1859 {
1860         struct f2fs_nm_info *nm_i = NM_I(sbi);
1861         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
1862         unsigned int nid_ofs = nid - START_NID(nid);
1863
1864         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1865                 return;
1866
1867         if (set) {
1868                 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
1869                         return;
1870                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1871                 nm_i->free_nid_count[nat_ofs]++;
1872         } else {
1873                 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
1874                         return;
1875                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1876                 if (!build)
1877                         nm_i->free_nid_count[nat_ofs]--;
1878         }
1879 }
1880
1881 /* return if the nid is recognized as free */
1882 static bool add_free_nid(struct f2fs_sb_info *sbi,
1883                                 nid_t nid, bool build, bool update)
1884 {
1885         struct f2fs_nm_info *nm_i = NM_I(sbi);
1886         struct free_nid *i, *e;
1887         struct nat_entry *ne;
1888         int err = -EINVAL;
1889         bool ret = false;
1890
1891         /* 0 nid should not be used */
1892         if (unlikely(nid == 0))
1893                 return false;
1894
1895         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1896         i->nid = nid;
1897         i->state = FREE_NID;
1898
1899         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
1900
1901         spin_lock(&nm_i->nid_list_lock);
1902
1903         if (build) {
1904                 /*
1905                  *   Thread A             Thread B
1906                  *  - f2fs_create
1907                  *   - f2fs_new_inode
1908                  *    - f2fs_alloc_nid
1909                  *     - __insert_nid_to_list(PREALLOC_NID)
1910                  *                     - f2fs_balance_fs_bg
1911                  *                      - f2fs_build_free_nids
1912                  *                       - __f2fs_build_free_nids
1913                  *                        - scan_nat_page
1914                  *                         - add_free_nid
1915                  *                          - __lookup_nat_cache
1916                  *  - f2fs_add_link
1917                  *   - f2fs_init_inode_metadata
1918                  *    - f2fs_new_inode_page
1919                  *     - f2fs_new_node_page
1920                  *      - set_node_addr
1921                  *  - f2fs_alloc_nid_done
1922                  *   - __remove_nid_from_list(PREALLOC_NID)
1923                  *                         - __insert_nid_to_list(FREE_NID)
1924                  */
1925                 ne = __lookup_nat_cache(nm_i, nid);
1926                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1927                                 nat_get_blkaddr(ne) != NULL_ADDR))
1928                         goto err_out;
1929
1930                 e = __lookup_free_nid_list(nm_i, nid);
1931                 if (e) {
1932                         if (e->state == FREE_NID)
1933                                 ret = true;
1934                         goto err_out;
1935                 }
1936         }
1937         ret = true;
1938         err = __insert_free_nid(sbi, i, FREE_NID);
1939 err_out:
1940         if (update) {
1941                 update_free_nid_bitmap(sbi, nid, ret, build);
1942                 if (!build)
1943                         nm_i->available_nids++;
1944         }
1945         spin_unlock(&nm_i->nid_list_lock);
1946         radix_tree_preload_end();
1947
1948         if (err)
1949                 kmem_cache_free(free_nid_slab, i);
1950         return ret;
1951 }
1952
1953 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1954 {
1955         struct f2fs_nm_info *nm_i = NM_I(sbi);
1956         struct free_nid *i;
1957         bool need_free = false;
1958
1959         spin_lock(&nm_i->nid_list_lock);
1960         i = __lookup_free_nid_list(nm_i, nid);
1961         if (i && i->state == FREE_NID) {
1962                 __remove_free_nid(sbi, i, FREE_NID);
1963                 need_free = true;
1964         }
1965         spin_unlock(&nm_i->nid_list_lock);
1966
1967         if (need_free)
1968                 kmem_cache_free(free_nid_slab, i);
1969 }
1970
1971 static void scan_nat_page(struct f2fs_sb_info *sbi,
1972                         struct page *nat_page, nid_t start_nid)
1973 {
1974         struct f2fs_nm_info *nm_i = NM_I(sbi);
1975         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1976         block_t blk_addr;
1977         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
1978         int i;
1979
1980         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
1981
1982         i = start_nid % NAT_ENTRY_PER_BLOCK;
1983
1984         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1985                 if (unlikely(start_nid >= nm_i->max_nid))
1986                         break;
1987
1988                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1989                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1990                 if (blk_addr == NULL_ADDR) {
1991                         add_free_nid(sbi, start_nid, true, true);
1992                 } else {
1993                         spin_lock(&NM_I(sbi)->nid_list_lock);
1994                         update_free_nid_bitmap(sbi, start_nid, false, true);
1995                         spin_unlock(&NM_I(sbi)->nid_list_lock);
1996                 }
1997         }
1998 }
1999
2000 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2001 {
2002         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2003         struct f2fs_journal *journal = curseg->journal;
2004         int i;
2005
2006         down_read(&curseg->journal_rwsem);
2007         for (i = 0; i < nats_in_cursum(journal); i++) {
2008                 block_t addr;
2009                 nid_t nid;
2010
2011                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2012                 nid = le32_to_cpu(nid_in_journal(journal, i));
2013                 if (addr == NULL_ADDR)
2014                         add_free_nid(sbi, nid, true, false);
2015                 else
2016                         remove_free_nid(sbi, nid);
2017         }
2018         up_read(&curseg->journal_rwsem);
2019 }
2020
2021 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2022 {
2023         struct f2fs_nm_info *nm_i = NM_I(sbi);
2024         unsigned int i, idx;
2025         nid_t nid;
2026
2027         down_read(&nm_i->nat_tree_lock);
2028
2029         for (i = 0; i < nm_i->nat_blocks; i++) {
2030                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2031                         continue;
2032                 if (!nm_i->free_nid_count[i])
2033                         continue;
2034                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2035                         idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2036                                                 NAT_ENTRY_PER_BLOCK, idx);
2037                         if (idx >= NAT_ENTRY_PER_BLOCK)
2038                                 break;
2039
2040                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
2041                         add_free_nid(sbi, nid, true, false);
2042
2043                         if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2044                                 goto out;
2045                 }
2046         }
2047 out:
2048         scan_curseg_cache(sbi);
2049
2050         up_read(&nm_i->nat_tree_lock);
2051 }
2052
2053 static void __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2054                                                 bool sync, bool mount)
2055 {
2056         struct f2fs_nm_info *nm_i = NM_I(sbi);
2057         int i = 0;
2058         nid_t nid = nm_i->next_scan_nid;
2059
2060         if (unlikely(nid >= nm_i->max_nid))
2061                 nid = 0;
2062
2063         /* Enough entries */
2064         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2065                 return;
2066
2067         if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2068                 return;
2069
2070         if (!mount) {
2071                 /* try to find free nids in free_nid_bitmap */
2072                 scan_free_nid_bits(sbi);
2073
2074                 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2075                         return;
2076         }
2077
2078         /* readahead nat pages to be scanned */
2079         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2080                                                         META_NAT, true);
2081
2082         down_read(&nm_i->nat_tree_lock);
2083
2084         while (1) {
2085                 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2086                                                 nm_i->nat_block_bitmap)) {
2087                         struct page *page = get_current_nat_page(sbi, nid);
2088
2089                         scan_nat_page(sbi, page, nid);
2090                         f2fs_put_page(page, 1);
2091                 }
2092
2093                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2094                 if (unlikely(nid >= nm_i->max_nid))
2095                         nid = 0;
2096
2097                 if (++i >= FREE_NID_PAGES)
2098                         break;
2099         }
2100
2101         /* go to the next free nat pages to find free nids abundantly */
2102         nm_i->next_scan_nid = nid;
2103
2104         /* find free nids from current sum_pages */
2105         scan_curseg_cache(sbi);
2106
2107         up_read(&nm_i->nat_tree_lock);
2108
2109         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2110                                         nm_i->ra_nid_pages, META_NAT, false);
2111 }
2112
2113 void f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2114 {
2115         mutex_lock(&NM_I(sbi)->build_lock);
2116         __f2fs_build_free_nids(sbi, sync, mount);
2117         mutex_unlock(&NM_I(sbi)->build_lock);
2118 }
2119
2120 /*
2121  * If this function returns success, caller can obtain a new nid
2122  * from second parameter of this function.
2123  * The returned nid could be used ino as well as nid when inode is created.
2124  */
2125 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2126 {
2127         struct f2fs_nm_info *nm_i = NM_I(sbi);
2128         struct free_nid *i = NULL;
2129 retry:
2130 #ifdef CONFIG_F2FS_FAULT_INJECTION
2131         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2132                 f2fs_show_injection_info(FAULT_ALLOC_NID);
2133                 return false;
2134         }
2135 #endif
2136         spin_lock(&nm_i->nid_list_lock);
2137
2138         if (unlikely(nm_i->available_nids == 0)) {
2139                 spin_unlock(&nm_i->nid_list_lock);
2140                 return false;
2141         }
2142
2143         /* We should not use stale free nids created by f2fs_build_free_nids */
2144         if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2145                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2146                 i = list_first_entry(&nm_i->free_nid_list,
2147                                         struct free_nid, list);
2148                 *nid = i->nid;
2149
2150                 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2151                 nm_i->available_nids--;
2152
2153                 update_free_nid_bitmap(sbi, *nid, false, false);
2154
2155                 spin_unlock(&nm_i->nid_list_lock);
2156                 return true;
2157         }
2158         spin_unlock(&nm_i->nid_list_lock);
2159
2160         /* Let's scan nat pages and its caches to get free nids */
2161         f2fs_build_free_nids(sbi, true, false);
2162         goto retry;
2163 }
2164
2165 /*
2166  * f2fs_alloc_nid() should be called prior to this function.
2167  */
2168 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2169 {
2170         struct f2fs_nm_info *nm_i = NM_I(sbi);
2171         struct free_nid *i;
2172
2173         spin_lock(&nm_i->nid_list_lock);
2174         i = __lookup_free_nid_list(nm_i, nid);
2175         f2fs_bug_on(sbi, !i);
2176         __remove_free_nid(sbi, i, PREALLOC_NID);
2177         spin_unlock(&nm_i->nid_list_lock);
2178
2179         kmem_cache_free(free_nid_slab, i);
2180 }
2181
2182 /*
2183  * f2fs_alloc_nid() should be called prior to this function.
2184  */
2185 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2186 {
2187         struct f2fs_nm_info *nm_i = NM_I(sbi);
2188         struct free_nid *i;
2189         bool need_free = false;
2190
2191         if (!nid)
2192                 return;
2193
2194         spin_lock(&nm_i->nid_list_lock);
2195         i = __lookup_free_nid_list(nm_i, nid);
2196         f2fs_bug_on(sbi, !i);
2197
2198         if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2199                 __remove_free_nid(sbi, i, PREALLOC_NID);
2200                 need_free = true;
2201         } else {
2202                 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2203         }
2204
2205         nm_i->available_nids++;
2206
2207         update_free_nid_bitmap(sbi, nid, true, false);
2208
2209         spin_unlock(&nm_i->nid_list_lock);
2210
2211         if (need_free)
2212                 kmem_cache_free(free_nid_slab, i);
2213 }
2214
2215 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2216 {
2217         struct f2fs_nm_info *nm_i = NM_I(sbi);
2218         struct free_nid *i, *next;
2219         int nr = nr_shrink;
2220
2221         if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2222                 return 0;
2223
2224         if (!mutex_trylock(&nm_i->build_lock))
2225                 return 0;
2226
2227         spin_lock(&nm_i->nid_list_lock);
2228         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2229                 if (nr_shrink <= 0 ||
2230                                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2231                         break;
2232
2233                 __remove_free_nid(sbi, i, FREE_NID);
2234                 kmem_cache_free(free_nid_slab, i);
2235                 nr_shrink--;
2236         }
2237         spin_unlock(&nm_i->nid_list_lock);
2238         mutex_unlock(&nm_i->build_lock);
2239
2240         return nr - nr_shrink;
2241 }
2242
2243 void f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2244 {
2245         void *src_addr, *dst_addr;
2246         size_t inline_size;
2247         struct page *ipage;
2248         struct f2fs_inode *ri;
2249
2250         ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2251         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2252
2253         ri = F2FS_INODE(page);
2254         if (ri->i_inline & F2FS_INLINE_XATTR) {
2255                 set_inode_flag(inode, FI_INLINE_XATTR);
2256         } else {
2257                 clear_inode_flag(inode, FI_INLINE_XATTR);
2258                 goto update_inode;
2259         }
2260
2261         dst_addr = inline_xattr_addr(inode, ipage);
2262         src_addr = inline_xattr_addr(inode, page);
2263         inline_size = inline_xattr_size(inode);
2264
2265         f2fs_wait_on_page_writeback(ipage, NODE, true);
2266         memcpy(dst_addr, src_addr, inline_size);
2267 update_inode:
2268         f2fs_update_inode(inode, ipage);
2269         f2fs_put_page(ipage, 1);
2270 }
2271
2272 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2273 {
2274         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2275         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2276         nid_t new_xnid;
2277         struct dnode_of_data dn;
2278         struct node_info ni;
2279         struct page *xpage;
2280
2281         if (!prev_xnid)
2282                 goto recover_xnid;
2283
2284         /* 1: invalidate the previous xattr nid */
2285         f2fs_get_node_info(sbi, prev_xnid, &ni);
2286         f2fs_invalidate_blocks(sbi, ni.blk_addr);
2287         dec_valid_node_count(sbi, inode, false);
2288         set_node_addr(sbi, &ni, NULL_ADDR, false);
2289
2290 recover_xnid:
2291         /* 2: update xattr nid in inode */
2292         if (!f2fs_alloc_nid(sbi, &new_xnid))
2293                 return -ENOSPC;
2294
2295         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2296         xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2297         if (IS_ERR(xpage)) {
2298                 f2fs_alloc_nid_failed(sbi, new_xnid);
2299                 return PTR_ERR(xpage);
2300         }
2301
2302         f2fs_alloc_nid_done(sbi, new_xnid);
2303         f2fs_update_inode_page(inode);
2304
2305         /* 3: update and set xattr node page dirty */
2306         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2307
2308         set_page_dirty(xpage);
2309         f2fs_put_page(xpage, 1);
2310
2311         return 0;
2312 }
2313
2314 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2315 {
2316         struct f2fs_inode *src, *dst;
2317         nid_t ino = ino_of_node(page);
2318         struct node_info old_ni, new_ni;
2319         struct page *ipage;
2320
2321         f2fs_get_node_info(sbi, ino, &old_ni);
2322
2323         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2324                 return -EINVAL;
2325 retry:
2326         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2327         if (!ipage) {
2328                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2329                 goto retry;
2330         }
2331
2332         /* Should not use this inode from free nid list */
2333         remove_free_nid(sbi, ino);
2334
2335         if (!PageUptodate(ipage))
2336                 SetPageUptodate(ipage);
2337         fill_node_footer(ipage, ino, ino, 0, true);
2338         set_cold_node(page, false);
2339
2340         src = F2FS_INODE(page);
2341         dst = F2FS_INODE(ipage);
2342
2343         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2344         dst->i_size = 0;
2345         dst->i_blocks = cpu_to_le64(1);
2346         dst->i_links = cpu_to_le32(1);
2347         dst->i_xattr_nid = 0;
2348         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2349         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2350                 dst->i_extra_isize = src->i_extra_isize;
2351
2352                 if (f2fs_sb_has_flexible_inline_xattr(sbi->sb) &&
2353                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2354                                                         i_inline_xattr_size))
2355                         dst->i_inline_xattr_size = src->i_inline_xattr_size;
2356
2357                 if (f2fs_sb_has_project_quota(sbi->sb) &&
2358                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2359                                                                 i_projid))
2360                         dst->i_projid = src->i_projid;
2361         }
2362
2363         new_ni = old_ni;
2364         new_ni.ino = ino;
2365
2366         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2367                 WARN_ON(1);
2368         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2369         inc_valid_inode_count(sbi);
2370         set_page_dirty(ipage);
2371         f2fs_put_page(ipage, 1);
2372         return 0;
2373 }
2374
2375 void f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2376                         unsigned int segno, struct f2fs_summary_block *sum)
2377 {
2378         struct f2fs_node *rn;
2379         struct f2fs_summary *sum_entry;
2380         block_t addr;
2381         int i, idx, last_offset, nrpages;
2382
2383         /* scan the node segment */
2384         last_offset = sbi->blocks_per_seg;
2385         addr = START_BLOCK(sbi, segno);
2386         sum_entry = &sum->entries[0];
2387
2388         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2389                 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2390
2391                 /* readahead node pages */
2392                 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2393
2394                 for (idx = addr; idx < addr + nrpages; idx++) {
2395                         struct page *page = f2fs_get_tmp_page(sbi, idx);
2396
2397                         rn = F2FS_NODE(page);
2398                         sum_entry->nid = rn->footer.nid;
2399                         sum_entry->version = 0;
2400                         sum_entry->ofs_in_node = 0;
2401                         sum_entry++;
2402                         f2fs_put_page(page, 1);
2403                 }
2404
2405                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2406                                                         addr + nrpages);
2407         }
2408 }
2409
2410 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2411 {
2412         struct f2fs_nm_info *nm_i = NM_I(sbi);
2413         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2414         struct f2fs_journal *journal = curseg->journal;
2415         int i;
2416
2417         down_write(&curseg->journal_rwsem);
2418         for (i = 0; i < nats_in_cursum(journal); i++) {
2419                 struct nat_entry *ne;
2420                 struct f2fs_nat_entry raw_ne;
2421                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2422
2423                 raw_ne = nat_in_journal(journal, i);
2424
2425                 ne = __lookup_nat_cache(nm_i, nid);
2426                 if (!ne) {
2427                         ne = __alloc_nat_entry(nid, true);
2428                         __init_nat_entry(nm_i, ne, &raw_ne, true);
2429                 }
2430
2431                 /*
2432                  * if a free nat in journal has not been used after last
2433                  * checkpoint, we should remove it from available nids,
2434                  * since later we will add it again.
2435                  */
2436                 if (!get_nat_flag(ne, IS_DIRTY) &&
2437                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2438                         spin_lock(&nm_i->nid_list_lock);
2439                         nm_i->available_nids--;
2440                         spin_unlock(&nm_i->nid_list_lock);
2441                 }
2442
2443                 __set_nat_cache_dirty(nm_i, ne);
2444         }
2445         update_nats_in_cursum(journal, -i);
2446         up_write(&curseg->journal_rwsem);
2447 }
2448
2449 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2450                                                 struct list_head *head, int max)
2451 {
2452         struct nat_entry_set *cur;
2453
2454         if (nes->entry_cnt >= max)
2455                 goto add_out;
2456
2457         list_for_each_entry(cur, head, set_list) {
2458                 if (cur->entry_cnt >= nes->entry_cnt) {
2459                         list_add(&nes->set_list, cur->set_list.prev);
2460                         return;
2461                 }
2462         }
2463 add_out:
2464         list_add_tail(&nes->set_list, head);
2465 }
2466
2467 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2468                                                 struct page *page)
2469 {
2470         struct f2fs_nm_info *nm_i = NM_I(sbi);
2471         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2472         struct f2fs_nat_block *nat_blk = page_address(page);
2473         int valid = 0;
2474         int i = 0;
2475
2476         if (!enabled_nat_bits(sbi, NULL))
2477                 return;
2478
2479         if (nat_index == 0) {
2480                 valid = 1;
2481                 i = 1;
2482         }
2483         for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2484                 if (nat_blk->entries[i].block_addr != NULL_ADDR)
2485                         valid++;
2486         }
2487         if (valid == 0) {
2488                 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2489                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2490                 return;
2491         }
2492
2493         __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2494         if (valid == NAT_ENTRY_PER_BLOCK)
2495                 __set_bit_le(nat_index, nm_i->full_nat_bits);
2496         else
2497                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2498 }
2499
2500 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2501                 struct nat_entry_set *set, struct cp_control *cpc)
2502 {
2503         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2504         struct f2fs_journal *journal = curseg->journal;
2505         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2506         bool to_journal = true;
2507         struct f2fs_nat_block *nat_blk;
2508         struct nat_entry *ne, *cur;
2509         struct page *page = NULL;
2510
2511         /*
2512          * there are two steps to flush nat entries:
2513          * #1, flush nat entries to journal in current hot data summary block.
2514          * #2, flush nat entries to nat page.
2515          */
2516         if (enabled_nat_bits(sbi, cpc) ||
2517                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2518                 to_journal = false;
2519
2520         if (to_journal) {
2521                 down_write(&curseg->journal_rwsem);
2522         } else {
2523                 page = get_next_nat_page(sbi, start_nid);
2524                 nat_blk = page_address(page);
2525                 f2fs_bug_on(sbi, !nat_blk);
2526         }
2527
2528         /* flush dirty nats in nat entry set */
2529         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2530                 struct f2fs_nat_entry *raw_ne;
2531                 nid_t nid = nat_get_nid(ne);
2532                 int offset;
2533
2534                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2535
2536                 if (to_journal) {
2537                         offset = f2fs_lookup_journal_in_cursum(journal,
2538                                                         NAT_JOURNAL, nid, 1);
2539                         f2fs_bug_on(sbi, offset < 0);
2540                         raw_ne = &nat_in_journal(journal, offset);
2541                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2542                 } else {
2543                         raw_ne = &nat_blk->entries[nid - start_nid];
2544                 }
2545                 raw_nat_from_node_info(raw_ne, &ne->ni);
2546                 nat_reset_flag(ne);
2547                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2548                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2549                         add_free_nid(sbi, nid, false, true);
2550                 } else {
2551                         spin_lock(&NM_I(sbi)->nid_list_lock);
2552                         update_free_nid_bitmap(sbi, nid, false, false);
2553                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2554                 }
2555         }
2556
2557         if (to_journal) {
2558                 up_write(&curseg->journal_rwsem);
2559         } else {
2560                 __update_nat_bits(sbi, start_nid, page);
2561                 f2fs_put_page(page, 1);
2562         }
2563
2564         /* Allow dirty nats by node block allocation in write_begin */
2565         if (!set->entry_cnt) {
2566                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2567                 kmem_cache_free(nat_entry_set_slab, set);
2568         }
2569 }
2570
2571 /*
2572  * This function is called during the checkpointing process.
2573  */
2574 void f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2575 {
2576         struct f2fs_nm_info *nm_i = NM_I(sbi);
2577         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2578         struct f2fs_journal *journal = curseg->journal;
2579         struct nat_entry_set *setvec[SETVEC_SIZE];
2580         struct nat_entry_set *set, *tmp;
2581         unsigned int found;
2582         nid_t set_idx = 0;
2583         LIST_HEAD(sets);
2584
2585         if (!nm_i->dirty_nat_cnt)
2586                 return;
2587
2588         down_write(&nm_i->nat_tree_lock);
2589
2590         /*
2591          * if there are no enough space in journal to store dirty nat
2592          * entries, remove all entries from journal and merge them
2593          * into nat entry set.
2594          */
2595         if (enabled_nat_bits(sbi, cpc) ||
2596                 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2597                 remove_nats_in_journal(sbi);
2598
2599         while ((found = __gang_lookup_nat_set(nm_i,
2600                                         set_idx, SETVEC_SIZE, setvec))) {
2601                 unsigned idx;
2602                 set_idx = setvec[found - 1]->set + 1;
2603                 for (idx = 0; idx < found; idx++)
2604                         __adjust_nat_entry_set(setvec[idx], &sets,
2605                                                 MAX_NAT_JENTRIES(journal));
2606         }
2607
2608         /* flush dirty nats in nat entry set */
2609         list_for_each_entry_safe(set, tmp, &sets, set_list)
2610                 __flush_nat_entry_set(sbi, set, cpc);
2611
2612         up_write(&nm_i->nat_tree_lock);
2613         /* Allow dirty nats by node block allocation in write_begin */
2614 }
2615
2616 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2617 {
2618         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2619         struct f2fs_nm_info *nm_i = NM_I(sbi);
2620         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2621         unsigned int i;
2622         __u64 cp_ver = cur_cp_version(ckpt);
2623         block_t nat_bits_addr;
2624
2625         if (!enabled_nat_bits(sbi, NULL))
2626                 return 0;
2627
2628         nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2629         nm_i->nat_bits = f2fs_kzalloc(sbi,
2630                         nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2631         if (!nm_i->nat_bits)
2632                 return -ENOMEM;
2633
2634         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2635                                                 nm_i->nat_bits_blocks;
2636         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2637                 struct page *page = f2fs_get_meta_page(sbi, nat_bits_addr++);
2638
2639                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2640                                         page_address(page), F2FS_BLKSIZE);
2641                 f2fs_put_page(page, 1);
2642         }
2643
2644         cp_ver |= (cur_cp_crc(ckpt) << 32);
2645         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2646                 disable_nat_bits(sbi, true);
2647                 return 0;
2648         }
2649
2650         nm_i->full_nat_bits = nm_i->nat_bits + 8;
2651         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2652
2653         f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2654         return 0;
2655 }
2656
2657 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2658 {
2659         struct f2fs_nm_info *nm_i = NM_I(sbi);
2660         unsigned int i = 0;
2661         nid_t nid, last_nid;
2662
2663         if (!enabled_nat_bits(sbi, NULL))
2664                 return;
2665
2666         for (i = 0; i < nm_i->nat_blocks; i++) {
2667                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2668                 if (i >= nm_i->nat_blocks)
2669                         break;
2670
2671                 __set_bit_le(i, nm_i->nat_block_bitmap);
2672
2673                 nid = i * NAT_ENTRY_PER_BLOCK;
2674                 last_nid = nid + NAT_ENTRY_PER_BLOCK;
2675
2676                 spin_lock(&NM_I(sbi)->nid_list_lock);
2677                 for (; nid < last_nid; nid++)
2678                         update_free_nid_bitmap(sbi, nid, true, true);
2679                 spin_unlock(&NM_I(sbi)->nid_list_lock);
2680         }
2681
2682         for (i = 0; i < nm_i->nat_blocks; i++) {
2683                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2684                 if (i >= nm_i->nat_blocks)
2685                         break;
2686
2687                 __set_bit_le(i, nm_i->nat_block_bitmap);
2688         }
2689 }
2690
2691 static int init_node_manager(struct f2fs_sb_info *sbi)
2692 {
2693         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2694         struct f2fs_nm_info *nm_i = NM_I(sbi);
2695         unsigned char *version_bitmap;
2696         unsigned int nat_segs;
2697         int err;
2698
2699         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2700
2701         /* segment_count_nat includes pair segment so divide to 2. */
2702         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2703         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2704         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2705
2706         /* not used nids: 0, node, meta, (and root counted as valid node) */
2707         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2708                                 sbi->nquota_files - F2FS_RESERVED_NODE_NUM;
2709         nm_i->nid_cnt[FREE_NID] = 0;
2710         nm_i->nid_cnt[PREALLOC_NID] = 0;
2711         nm_i->nat_cnt = 0;
2712         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2713         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2714         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2715
2716         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2717         INIT_LIST_HEAD(&nm_i->free_nid_list);
2718         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2719         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2720         INIT_LIST_HEAD(&nm_i->nat_entries);
2721
2722         mutex_init(&nm_i->build_lock);
2723         spin_lock_init(&nm_i->nid_list_lock);
2724         init_rwsem(&nm_i->nat_tree_lock);
2725
2726         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2727         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2728         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2729         if (!version_bitmap)
2730                 return -EFAULT;
2731
2732         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2733                                         GFP_KERNEL);
2734         if (!nm_i->nat_bitmap)
2735                 return -ENOMEM;
2736
2737         err = __get_nat_bitmaps(sbi);
2738         if (err)
2739                 return err;
2740
2741 #ifdef CONFIG_F2FS_CHECK_FS
2742         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2743                                         GFP_KERNEL);
2744         if (!nm_i->nat_bitmap_mir)
2745                 return -ENOMEM;
2746 #endif
2747
2748         return 0;
2749 }
2750
2751 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2752 {
2753         struct f2fs_nm_info *nm_i = NM_I(sbi);
2754         int i;
2755
2756         nm_i->free_nid_bitmap =
2757                 f2fs_kzalloc(sbi, array_size(sizeof(unsigned char *),
2758                                              nm_i->nat_blocks),
2759                              GFP_KERNEL);
2760         if (!nm_i->free_nid_bitmap)
2761                 return -ENOMEM;
2762
2763         for (i = 0; i < nm_i->nat_blocks; i++) {
2764                 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
2765                                 NAT_ENTRY_BITMAP_SIZE_ALIGNED, GFP_KERNEL);
2766                 if (!nm_i->free_nid_bitmap)
2767                         return -ENOMEM;
2768         }
2769
2770         nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
2771                                                                 GFP_KERNEL);
2772         if (!nm_i->nat_block_bitmap)
2773                 return -ENOMEM;
2774
2775         nm_i->free_nid_count =
2776                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
2777                                               nm_i->nat_blocks),
2778                               GFP_KERNEL);
2779         if (!nm_i->free_nid_count)
2780                 return -ENOMEM;
2781         return 0;
2782 }
2783
2784 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
2785 {
2786         int err;
2787
2788         sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
2789                                                         GFP_KERNEL);
2790         if (!sbi->nm_info)
2791                 return -ENOMEM;
2792
2793         err = init_node_manager(sbi);
2794         if (err)
2795                 return err;
2796
2797         err = init_free_nid_cache(sbi);
2798         if (err)
2799                 return err;
2800
2801         /* load free nid status from nat_bits table */
2802         load_free_nid_bitmap(sbi);
2803
2804         f2fs_build_free_nids(sbi, true, true);
2805         return 0;
2806 }
2807
2808 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
2809 {
2810         struct f2fs_nm_info *nm_i = NM_I(sbi);
2811         struct free_nid *i, *next_i;
2812         struct nat_entry *natvec[NATVEC_SIZE];
2813         struct nat_entry_set *setvec[SETVEC_SIZE];
2814         nid_t nid = 0;
2815         unsigned int found;
2816
2817         if (!nm_i)
2818                 return;
2819
2820         /* destroy free nid list */
2821         spin_lock(&nm_i->nid_list_lock);
2822         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2823                 __remove_free_nid(sbi, i, FREE_NID);
2824                 spin_unlock(&nm_i->nid_list_lock);
2825                 kmem_cache_free(free_nid_slab, i);
2826                 spin_lock(&nm_i->nid_list_lock);
2827         }
2828         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
2829         f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
2830         f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
2831         spin_unlock(&nm_i->nid_list_lock);
2832
2833         /* destroy nat cache */
2834         down_write(&nm_i->nat_tree_lock);
2835         while ((found = __gang_lookup_nat_cache(nm_i,
2836                                         nid, NATVEC_SIZE, natvec))) {
2837                 unsigned idx;
2838
2839                 nid = nat_get_nid(natvec[found - 1]) + 1;
2840                 for (idx = 0; idx < found; idx++)
2841                         __del_from_nat_cache(nm_i, natvec[idx]);
2842         }
2843         f2fs_bug_on(sbi, nm_i->nat_cnt);
2844
2845         /* destroy nat set cache */
2846         nid = 0;
2847         while ((found = __gang_lookup_nat_set(nm_i,
2848                                         nid, SETVEC_SIZE, setvec))) {
2849                 unsigned idx;
2850
2851                 nid = setvec[found - 1]->set + 1;
2852                 for (idx = 0; idx < found; idx++) {
2853                         /* entry_cnt is not zero, when cp_error was occurred */
2854                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2855                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2856                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2857                 }
2858         }
2859         up_write(&nm_i->nat_tree_lock);
2860
2861         kvfree(nm_i->nat_block_bitmap);
2862         if (nm_i->free_nid_bitmap) {
2863                 int i;
2864
2865                 for (i = 0; i < nm_i->nat_blocks; i++)
2866                         kvfree(nm_i->free_nid_bitmap[i]);
2867                 kfree(nm_i->free_nid_bitmap);
2868         }
2869         kvfree(nm_i->free_nid_count);
2870
2871         kfree(nm_i->nat_bitmap);
2872         kfree(nm_i->nat_bits);
2873 #ifdef CONFIG_F2FS_CHECK_FS
2874         kfree(nm_i->nat_bitmap_mir);
2875 #endif
2876         sbi->nm_info = NULL;
2877         kfree(nm_i);
2878 }
2879
2880 int __init f2fs_create_node_manager_caches(void)
2881 {
2882         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2883                         sizeof(struct nat_entry));
2884         if (!nat_entry_slab)
2885                 goto fail;
2886
2887         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2888                         sizeof(struct free_nid));
2889         if (!free_nid_slab)
2890                 goto destroy_nat_entry;
2891
2892         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2893                         sizeof(struct nat_entry_set));
2894         if (!nat_entry_set_slab)
2895                 goto destroy_free_nid;
2896         return 0;
2897
2898 destroy_free_nid:
2899         kmem_cache_destroy(free_nid_slab);
2900 destroy_nat_entry:
2901         kmem_cache_destroy(nat_entry_slab);
2902 fail:
2903         return -ENOMEM;
2904 }
2905
2906 void f2fs_destroy_node_manager_caches(void)
2907 {
2908         kmem_cache_destroy(nat_entry_set_slab);
2909         kmem_cache_destroy(free_nid_slab);
2910         kmem_cache_destroy(nat_entry_slab);
2911 }