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
  23 static struct kmem_cache *nat_entry_slab;
  24 static struct kmem_cache *free_nid_slab;
  25
  26 static void clear_node_page_dirty(struct page *page)
  27 {
  28         struct address_space *mapping = page->mapping;
  29         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
  30         unsigned int long flags;
  31
  32         if (PageDirty(page)) {
  33                 spin_lock_irqsave(&mapping->tree_lock, flags);
  34                 radix_tree_tag_clear(&mapping->page_tree,
  35                                 page_index(page),
  36                                 PAGECACHE_TAG_DIRTY);
  37                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  38
  39                 clear_page_dirty_for_io(page);
  40                 dec_page_count(sbi, F2FS_DIRTY_NODES);
  41         }
  42         ClearPageUptodate(page);
  43 }
  44
  45 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  46 {
  47         pgoff_t index = current_nat_addr(sbi, nid);
  48         return get_meta_page(sbi, index);
  49 }
  50
  51 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  52 {
  53         struct page *src_page;
  54         struct page *dst_page;
  55         pgoff_t src_off;
  56         pgoff_t dst_off;
  57         void *src_addr;
  58         void *dst_addr;
  59         struct f2fs_nm_info *nm_i = NM_I(sbi);
  60
  61         src_off = current_nat_addr(sbi, nid);
  62         dst_off = next_nat_addr(sbi, src_off);
  63
  64         /* get current nat block page with lock */
  65         src_page = get_meta_page(sbi, src_off);
  66
  67         /* Dirty src_page means that it is already the new target NAT page. */
  68         if (PageDirty(src_page))
  69                 return src_page;
  70
  71         dst_page = grab_meta_page(sbi, dst_off);
  72
  73         src_addr = page_address(src_page);
  74         dst_addr = page_address(dst_page);
  75         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
  76         set_page_dirty(dst_page);
  77         f2fs_put_page(src_page, 1);
  78
  79         set_to_next_nat(nm_i, nid);
  80
  81         return dst_page;
  82 }
  83
  84 /*
  85  * Readahead NAT pages
  86  */
  87 static void ra_nat_pages(struct f2fs_sb_info *sbi, int nid)
  88 {
  89         struct address_space *mapping = sbi->meta_inode->i_mapping;
  90         struct f2fs_nm_info *nm_i = NM_I(sbi);
  91         struct page *page;
  92         pgoff_t index;
  93         int i;
  94
  95         for (i = 0; i < FREE_NID_PAGES; i++, nid += NAT_ENTRY_PER_BLOCK) {
  96                 if (nid >= nm_i->max_nid)
  97                         nid = 0;
  98                 index = current_nat_addr(sbi, nid);
  99
 100                 page = grab_cache_page(mapping, index);
 101                 if (!page)
 102                         continue;
 103                 if (f2fs_readpage(sbi, page, index, READ)) {
 104                         f2fs_put_page(page, 1);
 105                         continue;
 106                 }
 107                 f2fs_put_page(page, 0);
 108         }
 109 }
 110
 111 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 112 {
 113         return radix_tree_lookup(&nm_i->nat_root, n);
 114 }
 115
 116 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 117                 nid_t start, unsigned int nr, struct nat_entry **ep)
 118 {
 119         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 120 }
 121
 122 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 123 {
 124         list_del(&e->list);
 125         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 126         nm_i->nat_cnt--;
 127         kmem_cache_free(nat_entry_slab, e);
 128 }
 129
 130 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 131 {
 132         struct f2fs_nm_info *nm_i = NM_I(sbi);
 133         struct nat_entry *e;
 134         int is_cp = 1;
 135
 136         read_lock(&nm_i->nat_tree_lock);
 137         e = __lookup_nat_cache(nm_i, nid);
 138         if (e && !e->checkpointed)
 139                 is_cp = 0;
 140         read_unlock(&nm_i->nat_tree_lock);
 141         return is_cp;
 142 }
 143
 144 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 145 {
 146         struct nat_entry *new;
 147
 148         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 149         if (!new)
 150                 return NULL;
 151         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 152                 kmem_cache_free(nat_entry_slab, new);
 153                 return NULL;
 154         }
 155         memset(new, 0, sizeof(struct nat_entry));
 156         nat_set_nid(new, nid);
 157         list_add_tail(&new->list, &nm_i->nat_entries);
 158         nm_i->nat_cnt++;
 159         return new;
 160 }
 161
 162 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 163                                                 struct f2fs_nat_entry *ne)
 164 {
 165         struct nat_entry *e;
 166 retry:
 167         write_lock(&nm_i->nat_tree_lock);
 168         e = __lookup_nat_cache(nm_i, nid);
 169         if (!e) {
 170                 e = grab_nat_entry(nm_i, nid);
 171                 if (!e) {
 172                         write_unlock(&nm_i->nat_tree_lock);
 173                         goto retry;
 174                 }
 175                 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
 176                 nat_set_ino(e, le32_to_cpu(ne->ino));
 177                 nat_set_version(e, ne->version);
 178                 e->checkpointed = true;
 179         }
 180         write_unlock(&nm_i->nat_tree_lock);
 181 }
 182
 183 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 184                         block_t new_blkaddr)
 185 {
 186         struct f2fs_nm_info *nm_i = NM_I(sbi);
 187         struct nat_entry *e;
 188 retry:
 189         write_lock(&nm_i->nat_tree_lock);
 190         e = __lookup_nat_cache(nm_i, ni->nid);
 191         if (!e) {
 192                 e = grab_nat_entry(nm_i, ni->nid);
 193                 if (!e) {
 194                         write_unlock(&nm_i->nat_tree_lock);
 195                         goto retry;
 196                 }
 197                 e->ni = *ni;
 198                 e->checkpointed = true;
 199                 BUG_ON(ni->blk_addr == NEW_ADDR);
 200         } else if (new_blkaddr == NEW_ADDR) {
 201                 /*
 202                  * when nid is reallocated,
 203                  * previous nat entry can be remained in nat cache.
 204                  * So, reinitialize it with new information.
 205                  */
 206                 e->ni = *ni;
 207                 BUG_ON(ni->blk_addr != NULL_ADDR);
 208         }
 209
 210         if (new_blkaddr == NEW_ADDR)
 211                 e->checkpointed = false;
 212
 213         /* sanity check */
 214         BUG_ON(nat_get_blkaddr(e) != ni->blk_addr);
 215         BUG_ON(nat_get_blkaddr(e) == NULL_ADDR &&
 216                         new_blkaddr == NULL_ADDR);
 217         BUG_ON(nat_get_blkaddr(e) == NEW_ADDR &&
 218                         new_blkaddr == NEW_ADDR);
 219         BUG_ON(nat_get_blkaddr(e) != NEW_ADDR &&
 220                         nat_get_blkaddr(e) != NULL_ADDR &&
 221                         new_blkaddr == NEW_ADDR);
 222
 223         /* increament version no as node is removed */
 224         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 225                 unsigned char version = nat_get_version(e);
 226                 nat_set_version(e, inc_node_version(version));
 227         }
 228
 229         /* change address */
 230         nat_set_blkaddr(e, new_blkaddr);
 231         __set_nat_cache_dirty(nm_i, e);
 232         write_unlock(&nm_i->nat_tree_lock);
 233 }
 234
 235 static int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 236 {
 237         struct f2fs_nm_info *nm_i = NM_I(sbi);
 238
 239         if (nm_i->nat_cnt < 2 * NM_WOUT_THRESHOLD)
 240                 return 0;
 241
 242         write_lock(&nm_i->nat_tree_lock);
 243         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 244                 struct nat_entry *ne;
 245                 ne = list_first_entry(&nm_i->nat_entries,
 246                                         struct nat_entry, list);
 247                 __del_from_nat_cache(nm_i, ne);
 248                 nr_shrink--;
 249         }
 250         write_unlock(&nm_i->nat_tree_lock);
 251         return nr_shrink;
 252 }
 253
 254 /*
 255  * This function returns always success
 256  */
 257 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 258 {
 259         struct f2fs_nm_info *nm_i = NM_I(sbi);
 260         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 261         struct f2fs_summary_block *sum = curseg->sum_blk;
 262         nid_t start_nid = START_NID(nid);
 263         struct f2fs_nat_block *nat_blk;
 264         struct page *page = NULL;
 265         struct f2fs_nat_entry ne;
 266         struct nat_entry *e;
 267         int i;
 268
 269         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 270         ni->nid = nid;
 271
 272         /* Check nat cache */
 273         read_lock(&nm_i->nat_tree_lock);
 274         e = __lookup_nat_cache(nm_i, nid);
 275         if (e) {
 276                 ni->ino = nat_get_ino(e);
 277                 ni->blk_addr = nat_get_blkaddr(e);
 278                 ni->version = nat_get_version(e);
 279         }
 280         read_unlock(&nm_i->nat_tree_lock);
 281         if (e)
 282                 return;
 283
 284         /* Check current segment summary */
 285         mutex_lock(&curseg->curseg_mutex);
 286         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 287         if (i >= 0) {
 288                 ne = nat_in_journal(sum, i);
 289                 node_info_from_raw_nat(ni, &ne);
 290         }
 291         mutex_unlock(&curseg->curseg_mutex);
 292         if (i >= 0)
 293                 goto cache;
 294
 295         /* Fill node_info from nat page */
 296         page = get_current_nat_page(sbi, start_nid);
 297         nat_blk = (struct f2fs_nat_block *)page_address(page);
 298         ne = nat_blk->entries[nid - start_nid];
 299         node_info_from_raw_nat(ni, &ne);
 300         f2fs_put_page(page, 1);
 301 cache:
 302         /* cache nat entry */
 303         cache_nat_entry(NM_I(sbi), nid, &ne);
 304 }
 305
 306 /*
 307  * The maximum depth is four.
 308  * Offset[0] will have raw inode offset.
 309  */
 310 static int get_node_path(long block, int offset[4], unsigned int noffset[4])
 311 {
 312         const long direct_index = ADDRS_PER_INODE;
 313         const long direct_blks = ADDRS_PER_BLOCK;
 314         const long dptrs_per_blk = NIDS_PER_BLOCK;
 315         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 316         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 317         int n = 0;
 318         int level = 0;
 319
 320         noffset[0] = 0;
 321
 322         if (block < direct_index) {
 323                 offset[n++] = block;
 324                 level = 0;
 325                 goto got;
 326         }
 327         block -= direct_index;
 328         if (block < direct_blks) {
 329                 offset[n++] = NODE_DIR1_BLOCK;
 330                 noffset[n] = 1;
 331                 offset[n++] = block;
 332                 level = 1;
 333                 goto got;
 334         }
 335         block -= direct_blks;
 336         if (block < direct_blks) {
 337                 offset[n++] = NODE_DIR2_BLOCK;
 338                 noffset[n] = 2;
 339                 offset[n++] = block;
 340                 level = 1;
 341                 goto got;
 342         }
 343         block -= direct_blks;
 344         if (block < indirect_blks) {
 345                 offset[n++] = NODE_IND1_BLOCK;
 346                 noffset[n] = 3;
 347                 offset[n++] = block / direct_blks;
 348                 noffset[n] = 4 + offset[n - 1];
 349                 offset[n++] = block % direct_blks;
 350                 level = 2;
 351                 goto got;
 352         }
 353         block -= indirect_blks;
 354         if (block < indirect_blks) {
 355                 offset[n++] = NODE_IND2_BLOCK;
 356                 noffset[n] = 4 + dptrs_per_blk;
 357                 offset[n++] = block / direct_blks;
 358                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 359                 offset[n++] = block % direct_blks;
 360                 level = 2;
 361                 goto got;
 362         }
 363         block -= indirect_blks;
 364         if (block < dindirect_blks) {
 365                 offset[n++] = NODE_DIND_BLOCK;
 366                 noffset[n] = 5 + (dptrs_per_blk * 2);
 367                 offset[n++] = block / indirect_blks;
 368                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 369                               offset[n - 1] * (dptrs_per_blk + 1);
 370                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 371                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 372                               offset[n - 2] * (dptrs_per_blk + 1) +
 373                               offset[n - 1];
 374                 offset[n++] = block % direct_blks;
 375                 level = 3;
 376                 goto got;
 377         } else {
 378                 BUG();
 379         }
 380 got:
 381         return level;
 382 }
 383
 384 /*
 385  * Caller should call f2fs_put_dnode(dn).
 386  */
 387 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int ro)
 388 {
 389         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 390         struct page *npage[4];
 391         struct page *parent;
 392         int offset[4];
 393         unsigned int noffset[4];
 394         nid_t nids[4];
 395         int level, i;
 396         int err = 0;
 397
 398         level = get_node_path(index, offset, noffset);
 399
 400         nids[0] = dn->inode->i_ino;
 401         npage[0] = get_node_page(sbi, nids[0]);
 402         if (IS_ERR(npage[0]))
 403                 return PTR_ERR(npage[0]);
 404
 405         parent = npage[0];
 406         nids[1] = get_nid(parent, offset[0], true);
 407         dn->inode_page = npage[0];
 408         dn->inode_page_locked = true;
 409
 410         /* get indirect or direct nodes */
 411         for (i = 1; i <= level; i++) {
 412                 bool done = false;
 413
 414                 if (!nids[i] && !ro) {
 415                         mutex_lock_op(sbi, NODE_NEW);
 416
 417                         /* alloc new node */
 418                         if (!alloc_nid(sbi, &(nids[i]))) {
 419                                 mutex_unlock_op(sbi, NODE_NEW);
 420                                 err = -ENOSPC;
 421                                 goto release_pages;
 422                         }
 423
 424                         dn->nid = nids[i];
 425                         npage[i] = new_node_page(dn, noffset[i]);
 426                         if (IS_ERR(npage[i])) {
 427                                 alloc_nid_failed(sbi, nids[i]);
 428                                 mutex_unlock_op(sbi, NODE_NEW);
 429                                 err = PTR_ERR(npage[i]);
 430                                 goto release_pages;
 431                         }
 432
 433                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 434                         alloc_nid_done(sbi, nids[i]);
 435                         mutex_unlock_op(sbi, NODE_NEW);
 436                         done = true;
 437                 } else if (ro && i == level && level > 1) {
 438                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 439                         if (IS_ERR(npage[i])) {
 440                                 err = PTR_ERR(npage[i]);
 441                                 goto release_pages;
 442                         }
 443                         done = true;
 444                 }
 445                 if (i == 1) {
 446                         dn->inode_page_locked = false;
 447                         unlock_page(parent);
 448                 } else {
 449                         f2fs_put_page(parent, 1);
 450                 }
 451
 452                 if (!done) {
 453                         npage[i] = get_node_page(sbi, nids[i]);
 454                         if (IS_ERR(npage[i])) {
 455                                 err = PTR_ERR(npage[i]);
 456                                 f2fs_put_page(npage[0], 0);
 457                                 goto release_out;
 458                         }
 459                 }
 460                 if (i < level) {
 461                         parent = npage[i];
 462                         nids[i + 1] = get_nid(parent, offset[i], false);
 463                 }
 464         }
 465         dn->nid = nids[level];
 466         dn->ofs_in_node = offset[level];
 467         dn->node_page = npage[level];
 468         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 469         return 0;
 470
 471 release_pages:
 472         f2fs_put_page(parent, 1);
 473         if (i > 1)
 474                 f2fs_put_page(npage[0], 0);
 475 release_out:
 476         dn->inode_page = NULL;
 477         dn->node_page = NULL;
 478         return err;
 479 }
 480
 481 static void truncate_node(struct dnode_of_data *dn)
 482 {
 483         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 484         struct node_info ni;
 485
 486         get_node_info(sbi, dn->nid, &ni);
 487         if (dn->inode->i_blocks == 0) {
 488                 BUG_ON(ni.blk_addr != NULL_ADDR);
 489                 goto invalidate;
 490         }
 491         BUG_ON(ni.blk_addr == NULL_ADDR);
 492
 493         /* Deallocate node address */
 494         invalidate_blocks(sbi, ni.blk_addr);
 495         dec_valid_node_count(sbi, dn->inode, 1);
 496         set_node_addr(sbi, &ni, NULL_ADDR);
 497
 498         if (dn->nid == dn->inode->i_ino) {
 499                 remove_orphan_inode(sbi, dn->nid);
 500                 dec_valid_inode_count(sbi);
 501         } else {
 502                 sync_inode_page(dn);
 503         }
 504 invalidate:
 505         clear_node_page_dirty(dn->node_page);
 506         F2FS_SET_SB_DIRT(sbi);
 507
 508         f2fs_put_page(dn->node_page, 1);
 509         dn->node_page = NULL;
 510 }
 511
 512 static int truncate_dnode(struct dnode_of_data *dn)
 513 {
 514         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 515         struct page *page;
 516
 517         if (dn->nid == 0)
 518                 return 1;
 519
 520         /* get direct node */
 521         page = get_node_page(sbi, dn->nid);
 522         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 523                 return 1;
 524         else if (IS_ERR(page))
 525                 return PTR_ERR(page);
 526
 527         /* Make dnode_of_data for parameter */
 528         dn->node_page = page;
 529         dn->ofs_in_node = 0;
 530         truncate_data_blocks(dn);
 531         truncate_node(dn);
 532         return 1;
 533 }
 534
 535 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 536                                                 int ofs, int depth)
 537 {
 538         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 539         struct dnode_of_data rdn = *dn;
 540         struct page *page;
 541         struct f2fs_node *rn;
 542         nid_t child_nid;
 543         unsigned int child_nofs;
 544         int freed = 0;
 545         int i, ret;
 546
 547         if (dn->nid == 0)
 548                 return NIDS_PER_BLOCK + 1;
 549
 550         page = get_node_page(sbi, dn->nid);
 551         if (IS_ERR(page))
 552                 return PTR_ERR(page);
 553
 554         rn = (struct f2fs_node *)page_address(page);
 555         if (depth < 3) {
 556                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 557                         child_nid = le32_to_cpu(rn->in.nid[i]);
 558                         if (child_nid == 0)
 559                                 continue;
 560                         rdn.nid = child_nid;
 561                         ret = truncate_dnode(&rdn);
 562                         if (ret < 0)
 563                                 goto out_err;
 564                         set_nid(page, i, 0, false);
 565                 }
 566         } else {
 567                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 568                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 569                         child_nid = le32_to_cpu(rn->in.nid[i]);
 570                         if (child_nid == 0) {
 571                                 child_nofs += NIDS_PER_BLOCK + 1;
 572                                 continue;
 573                         }
 574                         rdn.nid = child_nid;
 575                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 576                         if (ret == (NIDS_PER_BLOCK + 1)) {
 577                                 set_nid(page, i, 0, false);
 578                                 child_nofs += ret;
 579                         } else if (ret < 0 && ret != -ENOENT) {
 580                                 goto out_err;
 581                         }
 582                 }
 583                 freed = child_nofs;
 584         }
 585
 586         if (!ofs) {
 587                 /* remove current indirect node */
 588                 dn->node_page = page;
 589                 truncate_node(dn);
 590                 freed++;
 591         } else {
 592                 f2fs_put_page(page, 1);
 593         }
 594         return freed;
 595
 596 out_err:
 597         f2fs_put_page(page, 1);
 598         return ret;
 599 }
 600
 601 static int truncate_partial_nodes(struct dnode_of_data *dn,
 602                         struct f2fs_inode *ri, int *offset, int depth)
 603 {
 604         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 605         struct page *pages[2];
 606         nid_t nid[3];
 607         nid_t child_nid;
 608         int err = 0;
 609         int i;
 610         int idx = depth - 2;
 611
 612         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 613         if (!nid[0])
 614                 return 0;
 615
 616         /* get indirect nodes in the path */
 617         for (i = 0; i < depth - 1; i++) {
 618                 /* refernece count'll be increased */
 619                 pages[i] = get_node_page(sbi, nid[i]);
 620                 if (IS_ERR(pages[i])) {
 621                         depth = i + 1;
 622                         err = PTR_ERR(pages[i]);
 623                         goto fail;
 624                 }
 625                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 626         }
 627
 628         /* free direct nodes linked to a partial indirect node */
 629         for (i = offset[depth - 1]; i < NIDS_PER_BLOCK; i++) {
 630                 child_nid = get_nid(pages[idx], i, false);
 631                 if (!child_nid)
 632                         continue;
 633                 dn->nid = child_nid;
 634                 err = truncate_dnode(dn);
 635                 if (err < 0)
 636                         goto fail;
 637                 set_nid(pages[idx], i, 0, false);
 638         }
 639
 640         if (offset[depth - 1] == 0) {
 641                 dn->node_page = pages[idx];
 642                 dn->nid = nid[idx];
 643                 truncate_node(dn);
 644         } else {
 645                 f2fs_put_page(pages[idx], 1);
 646         }
 647         offset[idx]++;
 648         offset[depth - 1] = 0;
 649 fail:
 650         for (i = depth - 3; i >= 0; i--)
 651                 f2fs_put_page(pages[i], 1);
 652         return err;
 653 }
 654
 655 /*
 656  * All the block addresses of data and nodes should be nullified.
 657  */
 658 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 659 {
 660         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 661         int err = 0, cont = 1;
 662         int level, offset[4], noffset[4];
 663         unsigned int nofs = 0;
 664         struct f2fs_node *rn;
 665         struct dnode_of_data dn;
 666         struct page *page;
 667
 668         level = get_node_path(from, offset, noffset);
 669
 670         page = get_node_page(sbi, inode->i_ino);
 671         if (IS_ERR(page))
 672                 return PTR_ERR(page);
 673
 674         set_new_dnode(&dn, inode, page, NULL, 0);
 675         unlock_page(page);
 676
 677         rn = page_address(page);
 678         switch (level) {
 679         case 0:
 680         case 1:
 681                 nofs = noffset[1];
 682                 break;
 683         case 2:
 684                 nofs = noffset[1];
 685                 if (!offset[level - 1])
 686                         goto skip_partial;
 687                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 688                 if (err < 0 && err != -ENOENT)
 689                         goto fail;
 690                 nofs += 1 + NIDS_PER_BLOCK;
 691                 break;
 692         case 3:
 693                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 694                 if (!offset[level - 1])
 695                         goto skip_partial;
 696                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 697                 if (err < 0 && err != -ENOENT)
 698                         goto fail;
 699                 break;
 700         default:
 701                 BUG();
 702         }
 703
 704 skip_partial:
 705         while (cont) {
 706                 dn.nid = le32_to_cpu(rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]);
 707                 switch (offset[0]) {
 708                 case NODE_DIR1_BLOCK:
 709                 case NODE_DIR2_BLOCK:
 710                         err = truncate_dnode(&dn);
 711                         break;
 712
 713                 case NODE_IND1_BLOCK:
 714                 case NODE_IND2_BLOCK:
 715                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 716                         break;
 717
 718                 case NODE_DIND_BLOCK:
 719                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 720                         cont = 0;
 721                         break;
 722
 723                 default:
 724                         BUG();
 725                 }
 726                 if (err < 0 && err != -ENOENT)
 727                         goto fail;
 728                 if (offset[1] == 0 &&
 729                                 rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 730                         lock_page(page);
 731                         wait_on_page_writeback(page);
 732                         rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 733                         set_page_dirty(page);
 734                         unlock_page(page);
 735                 }
 736                 offset[1] = 0;
 737                 offset[0]++;
 738                 nofs += err;
 739         }
 740 fail:
 741         f2fs_put_page(page, 0);
 742         return err > 0 ? 0 : err;
 743 }
 744
 745 int remove_inode_page(struct inode *inode)
 746 {
 747         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 748         struct page *page;
 749         nid_t ino = inode->i_ino;
 750         struct dnode_of_data dn;
 751
 752         mutex_lock_op(sbi, NODE_TRUNC);
 753         page = get_node_page(sbi, ino);
 754         if (IS_ERR(page)) {
 755                 mutex_unlock_op(sbi, NODE_TRUNC);
 756                 return PTR_ERR(page);
 757         }
 758
 759         if (F2FS_I(inode)->i_xattr_nid) {
 760                 nid_t nid = F2FS_I(inode)->i_xattr_nid;
 761                 struct page *npage = get_node_page(sbi, nid);
 762
 763                 if (IS_ERR(npage)) {
 764                         mutex_unlock_op(sbi, NODE_TRUNC);
 765                         return PTR_ERR(npage);
 766                 }
 767
 768                 F2FS_I(inode)->i_xattr_nid = 0;
 769                 set_new_dnode(&dn, inode, page, npage, nid);
 770                 dn.inode_page_locked = 1;
 771                 truncate_node(&dn);
 772         }
 773
 774         /* 0 is possible, after f2fs_new_inode() is failed */
 775         BUG_ON(inode->i_blocks != 0 && inode->i_blocks != 1);
 776         set_new_dnode(&dn, inode, page, page, ino);
 777         truncate_node(&dn);
 778
 779         mutex_unlock_op(sbi, NODE_TRUNC);
 780         return 0;
 781 }
 782
 783 int new_inode_page(struct inode *inode, const struct qstr *name)
 784 {
 785         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 786         struct page *page;
 787         struct dnode_of_data dn;
 788
 789         /* allocate inode page for new inode */
 790         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 791         mutex_lock_op(sbi, NODE_NEW);
 792         page = new_node_page(&dn, 0);
 793         init_dent_inode(name, page);
 794         mutex_unlock_op(sbi, NODE_NEW);
 795         if (IS_ERR(page))
 796                 return PTR_ERR(page);
 797         f2fs_put_page(page, 1);
 798         return 0;
 799 }
 800
 801 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
 802 {
 803         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 804         struct address_space *mapping = sbi->node_inode->i_mapping;
 805         struct node_info old_ni, new_ni;
 806         struct page *page;
 807         int err;
 808
 809         if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
 810                 return ERR_PTR(-EPERM);
 811
 812         page = grab_cache_page(mapping, dn->nid);
 813         if (!page)
 814                 return ERR_PTR(-ENOMEM);
 815
 816         get_node_info(sbi, dn->nid, &old_ni);
 817
 818         SetPageUptodate(page);
 819         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 820
 821         /* Reinitialize old_ni with new node page */
 822         BUG_ON(old_ni.blk_addr != NULL_ADDR);
 823         new_ni = old_ni;
 824         new_ni.ino = dn->inode->i_ino;
 825
 826         if (!inc_valid_node_count(sbi, dn->inode, 1)) {
 827                 err = -ENOSPC;
 828                 goto fail;
 829         }
 830         set_node_addr(sbi, &new_ni, NEW_ADDR);
 831         set_cold_node(dn->inode, page);
 832
 833         dn->node_page = page;
 834         sync_inode_page(dn);
 835         set_page_dirty(page);
 836         if (ofs == 0)
 837                 inc_valid_inode_count(sbi);
 838
 839         return page;
 840
 841 fail:
 842         clear_node_page_dirty(page);
 843         f2fs_put_page(page, 1);
 844         return ERR_PTR(err);
 845 }
 846
 847 static int read_node_page(struct page *page, int type)
 848 {
 849         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 850         struct node_info ni;
 851
 852         get_node_info(sbi, page->index, &ni);
 853
 854         if (ni.blk_addr == NULL_ADDR)
 855                 return -ENOENT;
 856         return f2fs_readpage(sbi, page, ni.blk_addr, type);
 857 }
 858
 859 /*
 860  * Readahead a node page
 861  */
 862 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 863 {
 864         struct address_space *mapping = sbi->node_inode->i_mapping;
 865         struct page *apage;
 866
 867         apage = find_get_page(mapping, nid);
 868         if (apage && PageUptodate(apage))
 869                 goto release_out;
 870         f2fs_put_page(apage, 0);
 871
 872         apage = grab_cache_page(mapping, nid);
 873         if (!apage)
 874                 return;
 875
 876         if (read_node_page(apage, READA))
 877                 unlock_page(apage);
 878
 879 release_out:
 880         f2fs_put_page(apage, 0);
 881         return;
 882 }
 883
 884 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 885 {
 886         int err;
 887         struct page *page;
 888         struct address_space *mapping = sbi->node_inode->i_mapping;
 889
 890         page = grab_cache_page(mapping, nid);
 891         if (!page)
 892                 return ERR_PTR(-ENOMEM);
 893
 894         err = read_node_page(page, READ_SYNC);
 895         if (err) {
 896                 f2fs_put_page(page, 1);
 897                 return ERR_PTR(err);
 898         }
 899
 900         BUG_ON(nid != nid_of_node(page));
 901         mark_page_accessed(page);
 902         return page;
 903 }
 904
 905 /*
 906  * Return a locked page for the desired node page.
 907  * And, readahead MAX_RA_NODE number of node pages.
 908  */
 909 struct page *get_node_page_ra(struct page *parent, int start)
 910 {
 911         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 912         struct address_space *mapping = sbi->node_inode->i_mapping;
 913         int i, end;
 914         int err = 0;
 915         nid_t nid;
 916         struct page *page;
 917
 918         /* First, try getting the desired direct node. */
 919         nid = get_nid(parent, start, false);
 920         if (!nid)
 921                 return ERR_PTR(-ENOENT);
 922
 923         page = find_get_page(mapping, nid);
 924         if (page && PageUptodate(page))
 925                 goto page_hit;
 926         f2fs_put_page(page, 0);
 927
 928 repeat:
 929         page = grab_cache_page(mapping, nid);
 930         if (!page)
 931                 return ERR_PTR(-ENOMEM);
 932
 933         err = read_node_page(page, READA);
 934         if (err) {
 935                 f2fs_put_page(page, 1);
 936                 return ERR_PTR(err);
 937         }
 938
 939         /* Then, try readahead for siblings of the desired node */
 940         end = start + MAX_RA_NODE;
 941         end = min(end, NIDS_PER_BLOCK);
 942         for (i = start + 1; i < end; i++) {
 943                 nid = get_nid(parent, i, false);
 944                 if (!nid)
 945                         continue;
 946                 ra_node_page(sbi, nid);
 947         }
 948
 949 page_hit:
 950         lock_page(page);
 951         if (PageError(page)) {
 952                 f2fs_put_page(page, 1);
 953                 return ERR_PTR(-EIO);
 954         }
 955
 956         /* Has the page been truncated? */
 957         if (page->mapping != mapping) {
 958                 f2fs_put_page(page, 1);
 959                 goto repeat;
 960         }
 961         return page;
 962 }
 963
 964 void sync_inode_page(struct dnode_of_data *dn)
 965 {
 966         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
 967                 update_inode(dn->inode, dn->node_page);
 968         } else if (dn->inode_page) {
 969                 if (!dn->inode_page_locked)
 970                         lock_page(dn->inode_page);
 971                 update_inode(dn->inode, dn->inode_page);
 972                 if (!dn->inode_page_locked)
 973                         unlock_page(dn->inode_page);
 974         } else {
 975                 f2fs_write_inode(dn->inode, NULL);
 976         }
 977 }
 978
 979 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
 980                                         struct writeback_control *wbc)
 981 {
 982         struct address_space *mapping = sbi->node_inode->i_mapping;
 983         pgoff_t index, end;
 984         struct pagevec pvec;
 985         int step = ino ? 2 : 0;
 986         int nwritten = 0, wrote = 0;
 987
 988         pagevec_init(&pvec, 0);
 989
 990 next_step:
 991         index = 0;
 992         end = LONG_MAX;
 993
 994         while (index <= end) {
 995                 int i, nr_pages;
 996                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 997                                 PAGECACHE_TAG_DIRTY,
 998                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 999                 if (nr_pages == 0)
1000                         break;
1001
1002                 for (i = 0; i < nr_pages; i++) {
1003                         struct page *page = pvec.pages[i];
1004
1005                         /*
1006                          * flushing sequence with step:
1007                          * 0. indirect nodes
1008                          * 1. dentry dnodes
1009                          * 2. file dnodes
1010                          */
1011                         if (step == 0 && IS_DNODE(page))
1012                                 continue;
1013                         if (step == 1 && (!IS_DNODE(page) ||
1014                                                 is_cold_node(page)))
1015                                 continue;
1016                         if (step == 2 && (!IS_DNODE(page) ||
1017                                                 !is_cold_node(page)))
1018                                 continue;
1019
1020                         /*
1021                          * If an fsync mode,
1022                          * we should not skip writing node pages.
1023                          */
1024                         if (ino && ino_of_node(page) == ino)
1025                                 lock_page(page);
1026                         else if (!trylock_page(page))
1027                                 continue;
1028
1029                         if (unlikely(page->mapping != mapping)) {
1030 continue_unlock:
1031                                 unlock_page(page);
1032                                 continue;
1033                         }
1034                         if (ino && ino_of_node(page) != ino)
1035                                 goto continue_unlock;
1036
1037                         if (!PageDirty(page)) {
1038                                 /* someone wrote it for us */
1039                                 goto continue_unlock;
1040                         }
1041
1042                         if (!clear_page_dirty_for_io(page))
1043                                 goto continue_unlock;
1044
1045                         /* called by fsync() */
1046                         if (ino && IS_DNODE(page)) {
1047                                 int mark = !is_checkpointed_node(sbi, ino);
1048                                 set_fsync_mark(page, 1);
1049                                 if (IS_INODE(page))
1050                                         set_dentry_mark(page, mark);
1051                                 nwritten++;
1052                         } else {
1053                                 set_fsync_mark(page, 0);
1054                                 set_dentry_mark(page, 0);
1055                         }
1056                         mapping->a_ops->writepage(page, wbc);
1057                         wrote++;
1058
1059                         if (--wbc->nr_to_write == 0)
1060                                 break;
1061                 }
1062                 pagevec_release(&pvec);
1063                 cond_resched();
1064
1065                 if (wbc->nr_to_write == 0) {
1066                         step = 2;
1067                         break;
1068                 }
1069         }
1070
1071         if (step < 2) {
1072                 step++;
1073                 goto next_step;
1074         }
1075
1076         if (wrote)
1077                 f2fs_submit_bio(sbi, NODE, wbc->sync_mode == WB_SYNC_ALL);
1078
1079         return nwritten;
1080 }
1081
1082 static int f2fs_write_node_page(struct page *page,
1083                                 struct writeback_control *wbc)
1084 {
1085         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1086         nid_t nid;
1087         block_t new_addr;
1088         struct node_info ni;
1089
1090         if (wbc->for_reclaim) {
1091                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1092                 wbc->pages_skipped++;
1093                 set_page_dirty(page);
1094                 return AOP_WRITEPAGE_ACTIVATE;
1095         }
1096
1097         wait_on_page_writeback(page);
1098
1099         mutex_lock_op(sbi, NODE_WRITE);
1100
1101         /* get old block addr of this node page */
1102         nid = nid_of_node(page);
1103         BUG_ON(page->index != nid);
1104
1105         get_node_info(sbi, nid, &ni);
1106
1107         /* This page is already truncated */
1108         if (ni.blk_addr == NULL_ADDR)
1109                 return 0;
1110
1111         set_page_writeback(page);
1112
1113         /* insert node offset */
1114         write_node_page(sbi, page, nid, ni.blk_addr, &new_addr);
1115         set_node_addr(sbi, &ni, new_addr);
1116         dec_page_count(sbi, F2FS_DIRTY_NODES);
1117
1118         mutex_unlock_op(sbi, NODE_WRITE);
1119         unlock_page(page);
1120         return 0;
1121 }
1122
1123 /*
1124  * It is very important to gather dirty pages and write at once, so that we can
1125  * submit a big bio without interfering other data writes.
1126  * Be default, 512 pages (2MB), a segment size, is quite reasonable.
1127  */
1128 #define COLLECT_DIRTY_NODES     512
1129 static int f2fs_write_node_pages(struct address_space *mapping,
1130                             struct writeback_control *wbc)
1131 {
1132         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1133         struct block_device *bdev = sbi->sb->s_bdev;
1134         long nr_to_write = wbc->nr_to_write;
1135
1136         /* First check balancing cached NAT entries */
1137         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK)) {
1138                 write_checkpoint(sbi, false);
1139                 return 0;
1140         }
1141
1142         /* collect a number of dirty node pages and write together */
1143         if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1144                 return 0;
1145
1146         /* if mounting is failed, skip writing node pages */
1147         wbc->nr_to_write = bio_get_nr_vecs(bdev);
1148         sync_node_pages(sbi, 0, wbc);
1149         wbc->nr_to_write = nr_to_write -
1150                 (bio_get_nr_vecs(bdev) - wbc->nr_to_write);
1151         return 0;
1152 }
1153
1154 static int f2fs_set_node_page_dirty(struct page *page)
1155 {
1156         struct address_space *mapping = page->mapping;
1157         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1158
1159         SetPageUptodate(page);
1160         if (!PageDirty(page)) {
1161                 __set_page_dirty_nobuffers(page);
1162                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1163                 SetPagePrivate(page);
1164                 return 1;
1165         }
1166         return 0;
1167 }
1168
1169 static void f2fs_invalidate_node_page(struct page *page, unsigned long offset)
1170 {
1171         struct inode *inode = page->mapping->host;
1172         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1173         if (PageDirty(page))
1174                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1175         ClearPagePrivate(page);
1176 }
1177
1178 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1179 {
1180         ClearPagePrivate(page);
1181         return 0;
1182 }
1183
1184 /*
1185  * Structure of the f2fs node operations
1186  */
1187 const struct address_space_operations f2fs_node_aops = {
1188         .writepage      = f2fs_write_node_page,
1189         .writepages     = f2fs_write_node_pages,
1190         .set_page_dirty = f2fs_set_node_page_dirty,
1191         .invalidatepage = f2fs_invalidate_node_page,
1192         .releasepage    = f2fs_release_node_page,
1193 };
1194
1195 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1196 {
1197         struct list_head *this;
1198         struct free_nid *i = NULL;
1199         list_for_each(this, head) {
1200                 i = list_entry(this, struct free_nid, list);
1201                 if (i->nid == n)
1202                         break;
1203                 i = NULL;
1204         }
1205         return i;
1206 }
1207
1208 static void __del_from_free_nid_list(struct free_nid *i)
1209 {
1210         list_del(&i->list);
1211         kmem_cache_free(free_nid_slab, i);
1212 }
1213
1214 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1215 {
1216         struct free_nid *i;
1217
1218         if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1219                 return 0;
1220 retry:
1221         i = kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1222         if (!i) {
1223                 cond_resched();
1224                 goto retry;
1225         }
1226         i->nid = nid;
1227         i->state = NID_NEW;
1228
1229         spin_lock(&nm_i->free_nid_list_lock);
1230         if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1231                 spin_unlock(&nm_i->free_nid_list_lock);
1232                 kmem_cache_free(free_nid_slab, i);
1233                 return 0;
1234         }
1235         list_add_tail(&i->list, &nm_i->free_nid_list);
1236         nm_i->fcnt++;
1237         spin_unlock(&nm_i->free_nid_list_lock);
1238         return 1;
1239 }
1240
1241 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1242 {
1243         struct free_nid *i;
1244         spin_lock(&nm_i->free_nid_list_lock);
1245         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1246         if (i && i->state == NID_NEW) {
1247                 __del_from_free_nid_list(i);
1248                 nm_i->fcnt--;
1249         }
1250         spin_unlock(&nm_i->free_nid_list_lock);
1251 }
1252
1253 static int scan_nat_page(struct f2fs_nm_info *nm_i,
1254                         struct page *nat_page, nid_t start_nid)
1255 {
1256         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1257         block_t blk_addr;
1258         int fcnt = 0;
1259         int i;
1260
1261         /* 0 nid should not be used */
1262         if (start_nid == 0)
1263                 ++start_nid;
1264
1265         i = start_nid % NAT_ENTRY_PER_BLOCK;
1266
1267         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1268                 blk_addr  = le32_to_cpu(nat_blk->entries[i].block_addr);
1269                 BUG_ON(blk_addr == NEW_ADDR);
1270                 if (blk_addr == NULL_ADDR)
1271                         fcnt += add_free_nid(nm_i, start_nid);
1272         }
1273         return fcnt;
1274 }
1275
1276 static void build_free_nids(struct f2fs_sb_info *sbi)
1277 {
1278         struct free_nid *fnid, *next_fnid;
1279         struct f2fs_nm_info *nm_i = NM_I(sbi);
1280         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1281         struct f2fs_summary_block *sum = curseg->sum_blk;
1282         nid_t nid = 0;
1283         bool is_cycled = false;
1284         int fcnt = 0;
1285         int i;
1286
1287         nid = nm_i->next_scan_nid;
1288         nm_i->init_scan_nid = nid;
1289
1290         ra_nat_pages(sbi, nid);
1291
1292         while (1) {
1293                 struct page *page = get_current_nat_page(sbi, nid);
1294
1295                 fcnt += scan_nat_page(nm_i, page, nid);
1296                 f2fs_put_page(page, 1);
1297
1298                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1299
1300                 if (nid >= nm_i->max_nid) {
1301                         nid = 0;
1302                         is_cycled = true;
1303                 }
1304                 if (fcnt > MAX_FREE_NIDS)
1305                         break;
1306                 if (is_cycled && nm_i->init_scan_nid <= nid)
1307                         break;
1308         }
1309
1310         nm_i->next_scan_nid = nid;
1311
1312         /* find free nids from current sum_pages */
1313         mutex_lock(&curseg->curseg_mutex);
1314         for (i = 0; i < nats_in_cursum(sum); i++) {
1315                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1316                 nid = le32_to_cpu(nid_in_journal(sum, i));
1317                 if (addr == NULL_ADDR)
1318                         add_free_nid(nm_i, nid);
1319                 else
1320                         remove_free_nid(nm_i, nid);
1321         }
1322         mutex_unlock(&curseg->curseg_mutex);
1323
1324         /* remove the free nids from current allocated nids */
1325         list_for_each_entry_safe(fnid, next_fnid, &nm_i->free_nid_list, list) {
1326                 struct nat_entry *ne;
1327
1328                 read_lock(&nm_i->nat_tree_lock);
1329                 ne = __lookup_nat_cache(nm_i, fnid->nid);
1330                 if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1331                         remove_free_nid(nm_i, fnid->nid);
1332                 read_unlock(&nm_i->nat_tree_lock);
1333         }
1334 }
1335
1336 /*
1337  * If this function returns success, caller can obtain a new nid
1338  * from second parameter of this function.
1339  * The returned nid could be used ino as well as nid when inode is created.
1340  */
1341 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1342 {
1343         struct f2fs_nm_info *nm_i = NM_I(sbi);
1344         struct free_nid *i = NULL;
1345         struct list_head *this;
1346 retry:
1347         mutex_lock(&nm_i->build_lock);
1348         if (!nm_i->fcnt) {
1349                 /* scan NAT in order to build free nid list */
1350                 build_free_nids(sbi);
1351                 if (!nm_i->fcnt) {
1352                         mutex_unlock(&nm_i->build_lock);
1353                         return false;
1354                 }
1355         }
1356         mutex_unlock(&nm_i->build_lock);
1357
1358         /*
1359          * We check fcnt again since previous check is racy as
1360          * we didn't hold free_nid_list_lock. So other thread
1361          * could consume all of free nids.
1362          */
1363         spin_lock(&nm_i->free_nid_list_lock);
1364         if (!nm_i->fcnt) {
1365                 spin_unlock(&nm_i->free_nid_list_lock);
1366                 goto retry;
1367         }
1368
1369         BUG_ON(list_empty(&nm_i->free_nid_list));
1370         list_for_each(this, &nm_i->free_nid_list) {
1371                 i = list_entry(this, struct free_nid, list);
1372                 if (i->state == NID_NEW)
1373                         break;
1374         }
1375
1376         BUG_ON(i->state != NID_NEW);
1377         *nid = i->nid;
1378         i->state = NID_ALLOC;
1379         nm_i->fcnt--;
1380         spin_unlock(&nm_i->free_nid_list_lock);
1381         return true;
1382 }
1383
1384 /*
1385  * alloc_nid() should be called prior to this function.
1386  */
1387 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1388 {
1389         struct f2fs_nm_info *nm_i = NM_I(sbi);
1390         struct free_nid *i;
1391
1392         spin_lock(&nm_i->free_nid_list_lock);
1393         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1394         if (i) {
1395                 BUG_ON(i->state != NID_ALLOC);
1396                 __del_from_free_nid_list(i);
1397         }
1398         spin_unlock(&nm_i->free_nid_list_lock);
1399 }
1400
1401 /*
1402  * alloc_nid() should be called prior to this function.
1403  */
1404 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1405 {
1406         alloc_nid_done(sbi, nid);
1407         add_free_nid(NM_I(sbi), nid);
1408 }
1409
1410 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1411                 struct f2fs_summary *sum, struct node_info *ni,
1412                 block_t new_blkaddr)
1413 {
1414         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1415         set_node_addr(sbi, ni, new_blkaddr);
1416         clear_node_page_dirty(page);
1417 }
1418
1419 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1420 {
1421         struct address_space *mapping = sbi->node_inode->i_mapping;
1422         struct f2fs_node *src, *dst;
1423         nid_t ino = ino_of_node(page);
1424         struct node_info old_ni, new_ni;
1425         struct page *ipage;
1426
1427         ipage = grab_cache_page(mapping, ino);
1428         if (!ipage)
1429                 return -ENOMEM;
1430
1431         /* Should not use this inode  from free nid list */
1432         remove_free_nid(NM_I(sbi), ino);
1433
1434         get_node_info(sbi, ino, &old_ni);
1435         SetPageUptodate(ipage);
1436         fill_node_footer(ipage, ino, ino, 0, true);
1437
1438         src = (struct f2fs_node *)page_address(page);
1439         dst = (struct f2fs_node *)page_address(ipage);
1440
1441         memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i);
1442         dst->i.i_size = 0;
1443         dst->i.i_blocks = cpu_to_le64(1);
1444         dst->i.i_links = cpu_to_le32(1);
1445         dst->i.i_xattr_nid = 0;
1446
1447         new_ni = old_ni;
1448         new_ni.ino = ino;
1449
1450         set_node_addr(sbi, &new_ni, NEW_ADDR);
1451         inc_valid_inode_count(sbi);
1452
1453         f2fs_put_page(ipage, 1);
1454         return 0;
1455 }
1456
1457 int restore_node_summary(struct f2fs_sb_info *sbi,
1458                         unsigned int segno, struct f2fs_summary_block *sum)
1459 {
1460         struct f2fs_node *rn;
1461         struct f2fs_summary *sum_entry;
1462         struct page *page;
1463         block_t addr;
1464         int i, last_offset;
1465
1466         /* alloc temporal page for read node */
1467         page = alloc_page(GFP_NOFS | __GFP_ZERO);
1468         if (IS_ERR(page))
1469                 return PTR_ERR(page);
1470         lock_page(page);
1471
1472         /* scan the node segment */
1473         last_offset = sbi->blocks_per_seg;
1474         addr = START_BLOCK(sbi, segno);
1475         sum_entry = &sum->entries[0];
1476
1477         for (i = 0; i < last_offset; i++, sum_entry++) {
1478                 if (f2fs_readpage(sbi, page, addr, READ_SYNC))
1479                         goto out;
1480
1481                 rn = (struct f2fs_node *)page_address(page);
1482                 sum_entry->nid = rn->footer.nid;
1483                 sum_entry->version = 0;
1484                 sum_entry->ofs_in_node = 0;
1485                 addr++;
1486
1487                 /*
1488                  * In order to read next node page,
1489                  * we must clear PageUptodate flag.
1490                  */
1491                 ClearPageUptodate(page);
1492         }
1493 out:
1494         unlock_page(page);
1495         __free_pages(page, 0);
1496         return 0;
1497 }
1498
1499 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1500 {
1501         struct f2fs_nm_info *nm_i = NM_I(sbi);
1502         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1503         struct f2fs_summary_block *sum = curseg->sum_blk;
1504         int i;
1505
1506         mutex_lock(&curseg->curseg_mutex);
1507
1508         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1509                 mutex_unlock(&curseg->curseg_mutex);
1510                 return false;
1511         }
1512
1513         for (i = 0; i < nats_in_cursum(sum); i++) {
1514                 struct nat_entry *ne;
1515                 struct f2fs_nat_entry raw_ne;
1516                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1517
1518                 raw_ne = nat_in_journal(sum, i);
1519 retry:
1520                 write_lock(&nm_i->nat_tree_lock);
1521                 ne = __lookup_nat_cache(nm_i, nid);
1522                 if (ne) {
1523                         __set_nat_cache_dirty(nm_i, ne);
1524                         write_unlock(&nm_i->nat_tree_lock);
1525                         continue;
1526                 }
1527                 ne = grab_nat_entry(nm_i, nid);
1528                 if (!ne) {
1529                         write_unlock(&nm_i->nat_tree_lock);
1530                         goto retry;
1531                 }
1532                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1533                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1534                 nat_set_version(ne, raw_ne.version);
1535                 __set_nat_cache_dirty(nm_i, ne);
1536                 write_unlock(&nm_i->nat_tree_lock);
1537         }
1538         update_nats_in_cursum(sum, -i);
1539         mutex_unlock(&curseg->curseg_mutex);
1540         return true;
1541 }
1542
1543 /*
1544  * This function is called during the checkpointing process.
1545  */
1546 void flush_nat_entries(struct f2fs_sb_info *sbi)
1547 {
1548         struct f2fs_nm_info *nm_i = NM_I(sbi);
1549         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1550         struct f2fs_summary_block *sum = curseg->sum_blk;
1551         struct list_head *cur, *n;
1552         struct page *page = NULL;
1553         struct f2fs_nat_block *nat_blk = NULL;
1554         nid_t start_nid = 0, end_nid = 0;
1555         bool flushed;
1556
1557         flushed = flush_nats_in_journal(sbi);
1558
1559         if (!flushed)
1560                 mutex_lock(&curseg->curseg_mutex);
1561
1562         /* 1) flush dirty nat caches */
1563         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1564                 struct nat_entry *ne;
1565                 nid_t nid;
1566                 struct f2fs_nat_entry raw_ne;
1567                 int offset = -1;
1568                 block_t new_blkaddr;
1569
1570                 ne = list_entry(cur, struct nat_entry, list);
1571                 nid = nat_get_nid(ne);
1572
1573                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1574                         continue;
1575                 if (flushed)
1576                         goto to_nat_page;
1577
1578                 /* if there is room for nat enries in curseg->sumpage */
1579                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1580                 if (offset >= 0) {
1581                         raw_ne = nat_in_journal(sum, offset);
1582                         goto flush_now;
1583                 }
1584 to_nat_page:
1585                 if (!page || (start_nid > nid || nid > end_nid)) {
1586                         if (page) {
1587                                 f2fs_put_page(page, 1);
1588                                 page = NULL;
1589                         }
1590                         start_nid = START_NID(nid);
1591                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1592
1593                         /*
1594                          * get nat block with dirty flag, increased reference
1595                          * count, mapped and lock
1596                          */
1597                         page = get_next_nat_page(sbi, start_nid);
1598                         nat_blk = page_address(page);
1599                 }
1600
1601                 BUG_ON(!nat_blk);
1602                 raw_ne = nat_blk->entries[nid - start_nid];
1603 flush_now:
1604                 new_blkaddr = nat_get_blkaddr(ne);
1605
1606                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1607                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1608                 raw_ne.version = nat_get_version(ne);
1609
1610                 if (offset < 0) {
1611                         nat_blk->entries[nid - start_nid] = raw_ne;
1612                 } else {
1613                         nat_in_journal(sum, offset) = raw_ne;
1614                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1615                 }
1616
1617                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
1618                         write_lock(&nm_i->nat_tree_lock);
1619                         __del_from_nat_cache(nm_i, ne);
1620                         write_unlock(&nm_i->nat_tree_lock);
1621
1622                         /* We can reuse this freed nid at this point */
1623                         add_free_nid(NM_I(sbi), nid);
1624                 } else {
1625                         write_lock(&nm_i->nat_tree_lock);
1626                         __clear_nat_cache_dirty(nm_i, ne);
1627                         ne->checkpointed = true;
1628                         write_unlock(&nm_i->nat_tree_lock);
1629                 }
1630         }
1631         if (!flushed)
1632                 mutex_unlock(&curseg->curseg_mutex);
1633         f2fs_put_page(page, 1);
1634
1635         /* 2) shrink nat caches if necessary */
1636         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1637 }
1638
1639 static int init_node_manager(struct f2fs_sb_info *sbi)
1640 {
1641         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1642         struct f2fs_nm_info *nm_i = NM_I(sbi);
1643         unsigned char *version_bitmap;
1644         unsigned int nat_segs, nat_blocks;
1645
1646         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1647
1648         /* segment_count_nat includes pair segment so divide to 2. */
1649         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1650         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1651         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1652         nm_i->fcnt = 0;
1653         nm_i->nat_cnt = 0;
1654
1655         INIT_LIST_HEAD(&nm_i->free_nid_list);
1656         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1657         INIT_LIST_HEAD(&nm_i->nat_entries);
1658         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1659
1660         mutex_init(&nm_i->build_lock);
1661         spin_lock_init(&nm_i->free_nid_list_lock);
1662         rwlock_init(&nm_i->nat_tree_lock);
1663
1664         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1665         nm_i->init_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1666         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1667
1668         nm_i->nat_bitmap = kzalloc(nm_i->bitmap_size, GFP_KERNEL);
1669         if (!nm_i->nat_bitmap)
1670                 return -ENOMEM;
1671         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1672         if (!version_bitmap)
1673                 return -EFAULT;
1674
1675         /* copy version bitmap */
1676         memcpy(nm_i->nat_bitmap, version_bitmap, nm_i->bitmap_size);
1677         return 0;
1678 }
1679
1680 int build_node_manager(struct f2fs_sb_info *sbi)
1681 {
1682         int err;
1683
1684         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1685         if (!sbi->nm_info)
1686                 return -ENOMEM;
1687
1688         err = init_node_manager(sbi);
1689         if (err)
1690                 return err;
1691
1692         build_free_nids(sbi);
1693         return 0;
1694 }
1695
1696 void destroy_node_manager(struct f2fs_sb_info *sbi)
1697 {
1698         struct f2fs_nm_info *nm_i = NM_I(sbi);
1699         struct free_nid *i, *next_i;
1700         struct nat_entry *natvec[NATVEC_SIZE];
1701         nid_t nid = 0;
1702         unsigned int found;
1703
1704         if (!nm_i)
1705                 return;
1706
1707         /* destroy free nid list */
1708         spin_lock(&nm_i->free_nid_list_lock);
1709         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1710                 BUG_ON(i->state == NID_ALLOC);
1711                 __del_from_free_nid_list(i);
1712                 nm_i->fcnt--;
1713         }
1714         BUG_ON(nm_i->fcnt);
1715         spin_unlock(&nm_i->free_nid_list_lock);
1716
1717         /* destroy nat cache */
1718         write_lock(&nm_i->nat_tree_lock);
1719         while ((found = __gang_lookup_nat_cache(nm_i,
1720                                         nid, NATVEC_SIZE, natvec))) {
1721                 unsigned idx;
1722                 for (idx = 0; idx < found; idx++) {
1723                         struct nat_entry *e = natvec[idx];
1724                         nid = nat_get_nid(e) + 1;
1725                         __del_from_nat_cache(nm_i, e);
1726                 }
1727         }
1728         BUG_ON(nm_i->nat_cnt);
1729         write_unlock(&nm_i->nat_tree_lock);
1730
1731         kfree(nm_i->nat_bitmap);
1732         sbi->nm_info = NULL;
1733         kfree(nm_i);
1734 }
1735
1736 int __init create_node_manager_caches(void)
1737 {
1738         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1739                         sizeof(struct nat_entry), NULL);
1740         if (!nat_entry_slab)
1741                 return -ENOMEM;
1742
1743         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1744                         sizeof(struct free_nid), NULL);
1745         if (!free_nid_slab) {
1746                 kmem_cache_destroy(nat_entry_slab);
1747                 return -ENOMEM;
1748         }
1749         return 0;
1750 }
1751
1752 void destroy_node_manager_caches(void)
1753 {
1754         kmem_cache_destroy(free_nid_slab);
1755         kmem_cache_destroy(nat_entry_slab);
1756 }