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 mode)
 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         if (level != 0)
 407                 nids[1] = get_nid(parent, offset[0], true);
 408         dn->inode_page = npage[0];
 409         dn->inode_page_locked = true;
 410
 411         /* get indirect or direct nodes */
 412         for (i = 1; i <= level; i++) {
 413                 bool done = false;
 414
 415                 if (!nids[i] && mode == ALLOC_NODE) {
 416                         mutex_lock_op(sbi, NODE_NEW);
 417
 418                         /* alloc new node */
 419                         if (!alloc_nid(sbi, &(nids[i]))) {
 420                                 mutex_unlock_op(sbi, NODE_NEW);
 421                                 err = -ENOSPC;
 422                                 goto release_pages;
 423                         }
 424
 425                         dn->nid = nids[i];
 426                         npage[i] = new_node_page(dn, noffset[i]);
 427                         if (IS_ERR(npage[i])) {
 428                                 alloc_nid_failed(sbi, nids[i]);
 429                                 mutex_unlock_op(sbi, NODE_NEW);
 430                                 err = PTR_ERR(npage[i]);
 431                                 goto release_pages;
 432                         }
 433
 434                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 435                         alloc_nid_done(sbi, nids[i]);
 436                         mutex_unlock_op(sbi, NODE_NEW);
 437                         done = true;
 438                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 439                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 440                         if (IS_ERR(npage[i])) {
 441                                 err = PTR_ERR(npage[i]);
 442                                 goto release_pages;
 443                         }
 444                         done = true;
 445                 }
 446                 if (i == 1) {
 447                         dn->inode_page_locked = false;
 448                         unlock_page(parent);
 449                 } else {
 450                         f2fs_put_page(parent, 1);
 451                 }
 452
 453                 if (!done) {
 454                         npage[i] = get_node_page(sbi, nids[i]);
 455                         if (IS_ERR(npage[i])) {
 456                                 err = PTR_ERR(npage[i]);
 457                                 f2fs_put_page(npage[0], 0);
 458                                 goto release_out;
 459                         }
 460                 }
 461                 if (i < level) {
 462                         parent = npage[i];
 463                         nids[i + 1] = get_nid(parent, offset[i], false);
 464                 }
 465         }
 466         dn->nid = nids[level];
 467         dn->ofs_in_node = offset[level];
 468         dn->node_page = npage[level];
 469         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 470         return 0;
 471
 472 release_pages:
 473         f2fs_put_page(parent, 1);
 474         if (i > 1)
 475                 f2fs_put_page(npage[0], 0);
 476 release_out:
 477         dn->inode_page = NULL;
 478         dn->node_page = NULL;
 479         return err;
 480 }
 481
 482 static void truncate_node(struct dnode_of_data *dn)
 483 {
 484         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 485         struct node_info ni;
 486
 487         get_node_info(sbi, dn->nid, &ni);
 488         if (dn->inode->i_blocks == 0) {
 489                 BUG_ON(ni.blk_addr != NULL_ADDR);
 490                 goto invalidate;
 491         }
 492         BUG_ON(ni.blk_addr == NULL_ADDR);
 493
 494         /* Deallocate node address */
 495         invalidate_blocks(sbi, ni.blk_addr);
 496         dec_valid_node_count(sbi, dn->inode, 1);
 497         set_node_addr(sbi, &ni, NULL_ADDR);
 498
 499         if (dn->nid == dn->inode->i_ino) {
 500                 remove_orphan_inode(sbi, dn->nid);
 501                 dec_valid_inode_count(sbi);
 502         } else {
 503                 sync_inode_page(dn);
 504         }
 505 invalidate:
 506         clear_node_page_dirty(dn->node_page);
 507         F2FS_SET_SB_DIRT(sbi);
 508
 509         f2fs_put_page(dn->node_page, 1);
 510         dn->node_page = NULL;
 511 }
 512
 513 static int truncate_dnode(struct dnode_of_data *dn)
 514 {
 515         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 516         struct page *page;
 517
 518         if (dn->nid == 0)
 519                 return 1;
 520
 521         /* get direct node */
 522         page = get_node_page(sbi, dn->nid);
 523         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 524                 return 1;
 525         else if (IS_ERR(page))
 526                 return PTR_ERR(page);
 527
 528         /* Make dnode_of_data for parameter */
 529         dn->node_page = page;
 530         dn->ofs_in_node = 0;
 531         truncate_data_blocks(dn);
 532         truncate_node(dn);
 533         return 1;
 534 }
 535
 536 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 537                                                 int ofs, int depth)
 538 {
 539         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 540         struct dnode_of_data rdn = *dn;
 541         struct page *page;
 542         struct f2fs_node *rn;
 543         nid_t child_nid;
 544         unsigned int child_nofs;
 545         int freed = 0;
 546         int i, ret;
 547
 548         if (dn->nid == 0)
 549                 return NIDS_PER_BLOCK + 1;
 550
 551         page = get_node_page(sbi, dn->nid);
 552         if (IS_ERR(page))
 553                 return PTR_ERR(page);
 554
 555         rn = (struct f2fs_node *)page_address(page);
 556         if (depth < 3) {
 557                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 558                         child_nid = le32_to_cpu(rn->in.nid[i]);
 559                         if (child_nid == 0)
 560                                 continue;
 561                         rdn.nid = child_nid;
 562                         ret = truncate_dnode(&rdn);
 563                         if (ret < 0)
 564                                 goto out_err;
 565                         set_nid(page, i, 0, false);
 566                 }
 567         } else {
 568                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 569                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 570                         child_nid = le32_to_cpu(rn->in.nid[i]);
 571                         if (child_nid == 0) {
 572                                 child_nofs += NIDS_PER_BLOCK + 1;
 573                                 continue;
 574                         }
 575                         rdn.nid = child_nid;
 576                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 577                         if (ret == (NIDS_PER_BLOCK + 1)) {
 578                                 set_nid(page, i, 0, false);
 579                                 child_nofs += ret;
 580                         } else if (ret < 0 && ret != -ENOENT) {
 581                                 goto out_err;
 582                         }
 583                 }
 584                 freed = child_nofs;
 585         }
 586
 587         if (!ofs) {
 588                 /* remove current indirect node */
 589                 dn->node_page = page;
 590                 truncate_node(dn);
 591                 freed++;
 592         } else {
 593                 f2fs_put_page(page, 1);
 594         }
 595         return freed;
 596
 597 out_err:
 598         f2fs_put_page(page, 1);
 599         return ret;
 600 }
 601
 602 static int truncate_partial_nodes(struct dnode_of_data *dn,
 603                         struct f2fs_inode *ri, int *offset, int depth)
 604 {
 605         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 606         struct page *pages[2];
 607         nid_t nid[3];
 608         nid_t child_nid;
 609         int err = 0;
 610         int i;
 611         int idx = depth - 2;
 612
 613         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 614         if (!nid[0])
 615                 return 0;
 616
 617         /* get indirect nodes in the path */
 618         for (i = 0; i < depth - 1; i++) {
 619                 /* refernece count'll be increased */
 620                 pages[i] = get_node_page(sbi, nid[i]);
 621                 if (IS_ERR(pages[i])) {
 622                         depth = i + 1;
 623                         err = PTR_ERR(pages[i]);
 624                         goto fail;
 625                 }
 626                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 627         }
 628
 629         /* free direct nodes linked to a partial indirect node */
 630         for (i = offset[depth - 1]; i < NIDS_PER_BLOCK; i++) {
 631                 child_nid = get_nid(pages[idx], i, false);
 632                 if (!child_nid)
 633                         continue;
 634                 dn->nid = child_nid;
 635                 err = truncate_dnode(dn);
 636                 if (err < 0)
 637                         goto fail;
 638                 set_nid(pages[idx], i, 0, false);
 639         }
 640
 641         if (offset[depth - 1] == 0) {
 642                 dn->node_page = pages[idx];
 643                 dn->nid = nid[idx];
 644                 truncate_node(dn);
 645         } else {
 646                 f2fs_put_page(pages[idx], 1);
 647         }
 648         offset[idx]++;
 649         offset[depth - 1] = 0;
 650 fail:
 651         for (i = depth - 3; i >= 0; i--)
 652                 f2fs_put_page(pages[i], 1);
 653         return err;
 654 }
 655
 656 /*
 657  * All the block addresses of data and nodes should be nullified.
 658  */
 659 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 660 {
 661         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 662         int err = 0, cont = 1;
 663         int level, offset[4], noffset[4];
 664         unsigned int nofs = 0;
 665         struct f2fs_node *rn;
 666         struct dnode_of_data dn;
 667         struct page *page;
 668
 669         level = get_node_path(from, offset, noffset);
 670
 671         page = get_node_page(sbi, inode->i_ino);
 672         if (IS_ERR(page))
 673                 return PTR_ERR(page);
 674
 675         set_new_dnode(&dn, inode, page, NULL, 0);
 676         unlock_page(page);
 677
 678         rn = page_address(page);
 679         switch (level) {
 680         case 0:
 681         case 1:
 682                 nofs = noffset[1];
 683                 break;
 684         case 2:
 685                 nofs = noffset[1];
 686                 if (!offset[level - 1])
 687                         goto skip_partial;
 688                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 689                 if (err < 0 && err != -ENOENT)
 690                         goto fail;
 691                 nofs += 1 + NIDS_PER_BLOCK;
 692                 break;
 693         case 3:
 694                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 695                 if (!offset[level - 1])
 696                         goto skip_partial;
 697                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 698                 if (err < 0 && err != -ENOENT)
 699                         goto fail;
 700                 break;
 701         default:
 702                 BUG();
 703         }
 704
 705 skip_partial:
 706         while (cont) {
 707                 dn.nid = le32_to_cpu(rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]);
 708                 switch (offset[0]) {
 709                 case NODE_DIR1_BLOCK:
 710                 case NODE_DIR2_BLOCK:
 711                         err = truncate_dnode(&dn);
 712                         break;
 713
 714                 case NODE_IND1_BLOCK:
 715                 case NODE_IND2_BLOCK:
 716                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 717                         break;
 718
 719                 case NODE_DIND_BLOCK:
 720                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 721                         cont = 0;
 722                         break;
 723
 724                 default:
 725                         BUG();
 726                 }
 727                 if (err < 0 && err != -ENOENT)
 728                         goto fail;
 729                 if (offset[1] == 0 &&
 730                                 rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 731                         lock_page(page);
 732                         wait_on_page_writeback(page);
 733                         rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 734                         set_page_dirty(page);
 735                         unlock_page(page);
 736                 }
 737                 offset[1] = 0;
 738                 offset[0]++;
 739                 nofs += err;
 740         }
 741 fail:
 742         f2fs_put_page(page, 0);
 743         return err > 0 ? 0 : err;
 744 }
 745
 746 int remove_inode_page(struct inode *inode)
 747 {
 748         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 749         struct page *page;
 750         nid_t ino = inode->i_ino;
 751         struct dnode_of_data dn;
 752
 753         mutex_lock_op(sbi, NODE_TRUNC);
 754         page = get_node_page(sbi, ino);
 755         if (IS_ERR(page)) {
 756                 mutex_unlock_op(sbi, NODE_TRUNC);
 757                 return PTR_ERR(page);
 758         }
 759
 760         if (F2FS_I(inode)->i_xattr_nid) {
 761                 nid_t nid = F2FS_I(inode)->i_xattr_nid;
 762                 struct page *npage = get_node_page(sbi, nid);
 763
 764                 if (IS_ERR(npage)) {
 765                         mutex_unlock_op(sbi, NODE_TRUNC);
 766                         return PTR_ERR(npage);
 767                 }
 768
 769                 F2FS_I(inode)->i_xattr_nid = 0;
 770                 set_new_dnode(&dn, inode, page, npage, nid);
 771                 dn.inode_page_locked = 1;
 772                 truncate_node(&dn);
 773         }
 774
 775         /* 0 is possible, after f2fs_new_inode() is failed */
 776         BUG_ON(inode->i_blocks != 0 && inode->i_blocks != 1);
 777         set_new_dnode(&dn, inode, page, page, ino);
 778         truncate_node(&dn);
 779
 780         mutex_unlock_op(sbi, NODE_TRUNC);
 781         return 0;
 782 }
 783
 784 int new_inode_page(struct inode *inode, const struct qstr *name)
 785 {
 786         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 787         struct page *page;
 788         struct dnode_of_data dn;
 789
 790         /* allocate inode page for new inode */
 791         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 792         mutex_lock_op(sbi, NODE_NEW);
 793         page = new_node_page(&dn, 0);
 794         init_dent_inode(name, page);
 795         mutex_unlock_op(sbi, NODE_NEW);
 796         if (IS_ERR(page))
 797                 return PTR_ERR(page);
 798         f2fs_put_page(page, 1);
 799         return 0;
 800 }
 801
 802 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
 803 {
 804         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 805         struct address_space *mapping = sbi->node_inode->i_mapping;
 806         struct node_info old_ni, new_ni;
 807         struct page *page;
 808         int err;
 809
 810         if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
 811                 return ERR_PTR(-EPERM);
 812
 813         page = grab_cache_page(mapping, dn->nid);
 814         if (!page)
 815                 return ERR_PTR(-ENOMEM);
 816
 817         get_node_info(sbi, dn->nid, &old_ni);
 818
 819         SetPageUptodate(page);
 820         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 821
 822         /* Reinitialize old_ni with new node page */
 823         BUG_ON(old_ni.blk_addr != NULL_ADDR);
 824         new_ni = old_ni;
 825         new_ni.ino = dn->inode->i_ino;
 826
 827         if (!inc_valid_node_count(sbi, dn->inode, 1)) {
 828                 err = -ENOSPC;
 829                 goto fail;
 830         }
 831         set_node_addr(sbi, &new_ni, NEW_ADDR);
 832         set_cold_node(dn->inode, page);
 833
 834         dn->node_page = page;
 835         sync_inode_page(dn);
 836         set_page_dirty(page);
 837         if (ofs == 0)
 838                 inc_valid_inode_count(sbi);
 839
 840         return page;
 841
 842 fail:
 843         clear_node_page_dirty(page);
 844         f2fs_put_page(page, 1);
 845         return ERR_PTR(err);
 846 }
 847
 848 static int read_node_page(struct page *page, int type)
 849 {
 850         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 851         struct node_info ni;
 852
 853         get_node_info(sbi, page->index, &ni);
 854
 855         if (ni.blk_addr == NULL_ADDR)
 856                 return -ENOENT;
 857         return f2fs_readpage(sbi, page, ni.blk_addr, type);
 858 }
 859
 860 /*
 861  * Readahead a node page
 862  */
 863 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 864 {
 865         struct address_space *mapping = sbi->node_inode->i_mapping;
 866         struct page *apage;
 867
 868         apage = find_get_page(mapping, nid);
 869         if (apage && PageUptodate(apage))
 870                 goto release_out;
 871         f2fs_put_page(apage, 0);
 872
 873         apage = grab_cache_page(mapping, nid);
 874         if (!apage)
 875                 return;
 876
 877         if (read_node_page(apage, READA))
 878                 unlock_page(apage);
 879
 880 release_out:
 881         f2fs_put_page(apage, 0);
 882         return;
 883 }
 884
 885 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 886 {
 887         int err;
 888         struct page *page;
 889         struct address_space *mapping = sbi->node_inode->i_mapping;
 890
 891         page = grab_cache_page(mapping, nid);
 892         if (!page)
 893                 return ERR_PTR(-ENOMEM);
 894
 895         err = read_node_page(page, READ_SYNC);
 896         if (err) {
 897                 f2fs_put_page(page, 1);
 898                 return ERR_PTR(err);
 899         }
 900
 901         BUG_ON(nid != nid_of_node(page));
 902         mark_page_accessed(page);
 903         return page;
 904 }
 905
 906 /*
 907  * Return a locked page for the desired node page.
 908  * And, readahead MAX_RA_NODE number of node pages.
 909  */
 910 struct page *get_node_page_ra(struct page *parent, int start)
 911 {
 912         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 913         struct address_space *mapping = sbi->node_inode->i_mapping;
 914         int i, end;
 915         int err = 0;
 916         nid_t nid;
 917         struct page *page;
 918
 919         /* First, try getting the desired direct node. */
 920         nid = get_nid(parent, start, false);
 921         if (!nid)
 922                 return ERR_PTR(-ENOENT);
 923
 924         page = find_get_page(mapping, nid);
 925         if (page && PageUptodate(page))
 926                 goto page_hit;
 927         f2fs_put_page(page, 0);
 928
 929 repeat:
 930         page = grab_cache_page(mapping, nid);
 931         if (!page)
 932                 return ERR_PTR(-ENOMEM);
 933
 934         err = read_node_page(page, READ_SYNC);
 935         if (err) {
 936                 f2fs_put_page(page, 1);
 937                 return ERR_PTR(err);
 938         }
 939
 940         /* Then, try readahead for siblings of the desired node */
 941         end = start + MAX_RA_NODE;
 942         end = min(end, NIDS_PER_BLOCK);
 943         for (i = start + 1; i < end; i++) {
 944                 nid = get_nid(parent, i, false);
 945                 if (!nid)
 946                         continue;
 947                 ra_node_page(sbi, nid);
 948         }
 949
 950 page_hit:
 951         lock_page(page);
 952         if (PageError(page)) {
 953                 f2fs_put_page(page, 1);
 954                 return ERR_PTR(-EIO);
 955         }
 956
 957         /* Has the page been truncated? */
 958         if (page->mapping != mapping) {
 959                 f2fs_put_page(page, 1);
 960                 goto repeat;
 961         }
 962         return page;
 963 }
 964
 965 void sync_inode_page(struct dnode_of_data *dn)
 966 {
 967         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
 968                 update_inode(dn->inode, dn->node_page);
 969         } else if (dn->inode_page) {
 970                 if (!dn->inode_page_locked)
 971                         lock_page(dn->inode_page);
 972                 update_inode(dn->inode, dn->inode_page);
 973                 if (!dn->inode_page_locked)
 974                         unlock_page(dn->inode_page);
 975         } else {
 976                 f2fs_write_inode(dn->inode, NULL);
 977         }
 978 }
 979
 980 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
 981                                         struct writeback_control *wbc)
 982 {
 983         struct address_space *mapping = sbi->node_inode->i_mapping;
 984         pgoff_t index, end;
 985         struct pagevec pvec;
 986         int step = ino ? 2 : 0;
 987         int nwritten = 0, wrote = 0;
 988
 989         pagevec_init(&pvec, 0);
 990
 991 next_step:
 992         index = 0;
 993         end = LONG_MAX;
 994
 995         while (index <= end) {
 996                 int i, nr_pages;
 997                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 998                                 PAGECACHE_TAG_DIRTY,
 999                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1000                 if (nr_pages == 0)
1001                         break;
1002
1003                 for (i = 0; i < nr_pages; i++) {
1004                         struct page *page = pvec.pages[i];
1005
1006                         /*
1007                          * flushing sequence with step:
1008                          * 0. indirect nodes
1009                          * 1. dentry dnodes
1010                          * 2. file dnodes
1011                          */
1012                         if (step == 0 && IS_DNODE(page))
1013                                 continue;
1014                         if (step == 1 && (!IS_DNODE(page) ||
1015                                                 is_cold_node(page)))
1016                                 continue;
1017                         if (step == 2 && (!IS_DNODE(page) ||
1018                                                 !is_cold_node(page)))
1019                                 continue;
1020
1021                         /*
1022                          * If an fsync mode,
1023                          * we should not skip writing node pages.
1024                          */
1025                         if (ino && ino_of_node(page) == ino)
1026                                 lock_page(page);
1027                         else if (!trylock_page(page))
1028                                 continue;
1029
1030                         if (unlikely(page->mapping != mapping)) {
1031 continue_unlock:
1032                                 unlock_page(page);
1033                                 continue;
1034                         }
1035                         if (ino && ino_of_node(page) != ino)
1036                                 goto continue_unlock;
1037
1038                         if (!PageDirty(page)) {
1039                                 /* someone wrote it for us */
1040                                 goto continue_unlock;
1041                         }
1042
1043                         if (!clear_page_dirty_for_io(page))
1044                                 goto continue_unlock;
1045
1046                         /* called by fsync() */
1047                         if (ino && IS_DNODE(page)) {
1048                                 int mark = !is_checkpointed_node(sbi, ino);
1049                                 set_fsync_mark(page, 1);
1050                                 if (IS_INODE(page))
1051                                         set_dentry_mark(page, mark);
1052                                 nwritten++;
1053                         } else {
1054                                 set_fsync_mark(page, 0);
1055                                 set_dentry_mark(page, 0);
1056                         }
1057                         mapping->a_ops->writepage(page, wbc);
1058                         wrote++;
1059
1060                         if (--wbc->nr_to_write == 0)
1061                                 break;
1062                 }
1063                 pagevec_release(&pvec);
1064                 cond_resched();
1065
1066                 if (wbc->nr_to_write == 0) {
1067                         step = 2;
1068                         break;
1069                 }
1070         }
1071
1072         if (step < 2) {
1073                 step++;
1074                 goto next_step;
1075         }
1076
1077         if (wrote)
1078                 f2fs_submit_bio(sbi, NODE, wbc->sync_mode == WB_SYNC_ALL);
1079
1080         return nwritten;
1081 }
1082
1083 static int f2fs_write_node_page(struct page *page,
1084                                 struct writeback_control *wbc)
1085 {
1086         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1087         nid_t nid;
1088         block_t new_addr;
1089         struct node_info ni;
1090
1091         if (wbc->for_reclaim) {
1092                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1093                 wbc->pages_skipped++;
1094                 set_page_dirty(page);
1095                 return AOP_WRITEPAGE_ACTIVATE;
1096         }
1097
1098         wait_on_page_writeback(page);
1099
1100         mutex_lock_op(sbi, NODE_WRITE);
1101
1102         /* get old block addr of this node page */
1103         nid = nid_of_node(page);
1104         BUG_ON(page->index != nid);
1105
1106         get_node_info(sbi, nid, &ni);
1107
1108         /* This page is already truncated */
1109         if (ni.blk_addr == NULL_ADDR)
1110                 goto out;
1111
1112         set_page_writeback(page);
1113
1114         /* insert node offset */
1115         write_node_page(sbi, page, nid, ni.blk_addr, &new_addr);
1116         set_node_addr(sbi, &ni, new_addr);
1117 out:
1118         dec_page_count(sbi, F2FS_DIRTY_NODES);
1119         mutex_unlock_op(sbi, NODE_WRITE);
1120         unlock_page(page);
1121         return 0;
1122 }
1123
1124 /*
1125  * It is very important to gather dirty pages and write at once, so that we can
1126  * submit a big bio without interfering other data writes.
1127  * Be default, 512 pages (2MB), a segment size, is quite reasonable.
1128  */
1129 #define COLLECT_DIRTY_NODES     512
1130 static int f2fs_write_node_pages(struct address_space *mapping,
1131                             struct writeback_control *wbc)
1132 {
1133         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1134         struct block_device *bdev = sbi->sb->s_bdev;
1135         long nr_to_write = wbc->nr_to_write;
1136
1137         /* First check balancing cached NAT entries */
1138         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK)) {
1139                 write_checkpoint(sbi, false);
1140                 return 0;
1141         }
1142
1143         /* collect a number of dirty node pages and write together */
1144         if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1145                 return 0;
1146
1147         /* if mounting is failed, skip writing node pages */
1148         wbc->nr_to_write = bio_get_nr_vecs(bdev);
1149         sync_node_pages(sbi, 0, wbc);
1150         wbc->nr_to_write = nr_to_write -
1151                 (bio_get_nr_vecs(bdev) - wbc->nr_to_write);
1152         return 0;
1153 }
1154
1155 static int f2fs_set_node_page_dirty(struct page *page)
1156 {
1157         struct address_space *mapping = page->mapping;
1158         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1159
1160         SetPageUptodate(page);
1161         if (!PageDirty(page)) {
1162                 __set_page_dirty_nobuffers(page);
1163                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1164                 SetPagePrivate(page);
1165                 return 1;
1166         }
1167         return 0;
1168 }
1169
1170 static void f2fs_invalidate_node_page(struct page *page, unsigned long offset)
1171 {
1172         struct inode *inode = page->mapping->host;
1173         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1174         if (PageDirty(page))
1175                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1176         ClearPagePrivate(page);
1177 }
1178
1179 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1180 {
1181         ClearPagePrivate(page);
1182         return 0;
1183 }
1184
1185 /*
1186  * Structure of the f2fs node operations
1187  */
1188 const struct address_space_operations f2fs_node_aops = {
1189         .writepage      = f2fs_write_node_page,
1190         .writepages     = f2fs_write_node_pages,
1191         .set_page_dirty = f2fs_set_node_page_dirty,
1192         .invalidatepage = f2fs_invalidate_node_page,
1193         .releasepage    = f2fs_release_node_page,
1194 };
1195
1196 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1197 {
1198         struct list_head *this;
1199         struct free_nid *i = NULL;
1200         list_for_each(this, head) {
1201                 i = list_entry(this, struct free_nid, list);
1202                 if (i->nid == n)
1203                         break;
1204                 i = NULL;
1205         }
1206         return i;
1207 }
1208
1209 static void __del_from_free_nid_list(struct free_nid *i)
1210 {
1211         list_del(&i->list);
1212         kmem_cache_free(free_nid_slab, i);
1213 }
1214
1215 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1216 {
1217         struct free_nid *i;
1218
1219         if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1220                 return 0;
1221 retry:
1222         i = kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1223         if (!i) {
1224                 cond_resched();
1225                 goto retry;
1226         }
1227         i->nid = nid;
1228         i->state = NID_NEW;
1229
1230         spin_lock(&nm_i->free_nid_list_lock);
1231         if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1232                 spin_unlock(&nm_i->free_nid_list_lock);
1233                 kmem_cache_free(free_nid_slab, i);
1234                 return 0;
1235         }
1236         list_add_tail(&i->list, &nm_i->free_nid_list);
1237         nm_i->fcnt++;
1238         spin_unlock(&nm_i->free_nid_list_lock);
1239         return 1;
1240 }
1241
1242 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1243 {
1244         struct free_nid *i;
1245         spin_lock(&nm_i->free_nid_list_lock);
1246         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1247         if (i && i->state == NID_NEW) {
1248                 __del_from_free_nid_list(i);
1249                 nm_i->fcnt--;
1250         }
1251         spin_unlock(&nm_i->free_nid_list_lock);
1252 }
1253
1254 static int scan_nat_page(struct f2fs_nm_info *nm_i,
1255                         struct page *nat_page, nid_t start_nid)
1256 {
1257         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1258         block_t blk_addr;
1259         int fcnt = 0;
1260         int i;
1261
1262         /* 0 nid should not be used */
1263         if (start_nid == 0)
1264                 ++start_nid;
1265
1266         i = start_nid % NAT_ENTRY_PER_BLOCK;
1267
1268         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1269                 blk_addr  = le32_to_cpu(nat_blk->entries[i].block_addr);
1270                 BUG_ON(blk_addr == NEW_ADDR);
1271                 if (blk_addr == NULL_ADDR)
1272                         fcnt += add_free_nid(nm_i, start_nid);
1273         }
1274         return fcnt;
1275 }
1276
1277 static void build_free_nids(struct f2fs_sb_info *sbi)
1278 {
1279         struct free_nid *fnid, *next_fnid;
1280         struct f2fs_nm_info *nm_i = NM_I(sbi);
1281         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1282         struct f2fs_summary_block *sum = curseg->sum_blk;
1283         nid_t nid = 0;
1284         bool is_cycled = false;
1285         int fcnt = 0;
1286         int i;
1287
1288         nid = nm_i->next_scan_nid;
1289         nm_i->init_scan_nid = nid;
1290
1291         ra_nat_pages(sbi, nid);
1292
1293         while (1) {
1294                 struct page *page = get_current_nat_page(sbi, nid);
1295
1296                 fcnt += scan_nat_page(nm_i, page, nid);
1297                 f2fs_put_page(page, 1);
1298
1299                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1300
1301                 if (nid >= nm_i->max_nid) {
1302                         nid = 0;
1303                         is_cycled = true;
1304                 }
1305                 if (fcnt > MAX_FREE_NIDS)
1306                         break;
1307                 if (is_cycled && nm_i->init_scan_nid <= nid)
1308                         break;
1309         }
1310
1311         nm_i->next_scan_nid = nid;
1312
1313         /* find free nids from current sum_pages */
1314         mutex_lock(&curseg->curseg_mutex);
1315         for (i = 0; i < nats_in_cursum(sum); i++) {
1316                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1317                 nid = le32_to_cpu(nid_in_journal(sum, i));
1318                 if (addr == NULL_ADDR)
1319                         add_free_nid(nm_i, nid);
1320                 else
1321                         remove_free_nid(nm_i, nid);
1322         }
1323         mutex_unlock(&curseg->curseg_mutex);
1324
1325         /* remove the free nids from current allocated nids */
1326         list_for_each_entry_safe(fnid, next_fnid, &nm_i->free_nid_list, list) {
1327                 struct nat_entry *ne;
1328
1329                 read_lock(&nm_i->nat_tree_lock);
1330                 ne = __lookup_nat_cache(nm_i, fnid->nid);
1331                 if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1332                         remove_free_nid(nm_i, fnid->nid);
1333                 read_unlock(&nm_i->nat_tree_lock);
1334         }
1335 }
1336
1337 /*
1338  * If this function returns success, caller can obtain a new nid
1339  * from second parameter of this function.
1340  * The returned nid could be used ino as well as nid when inode is created.
1341  */
1342 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1343 {
1344         struct f2fs_nm_info *nm_i = NM_I(sbi);
1345         struct free_nid *i = NULL;
1346         struct list_head *this;
1347 retry:
1348         mutex_lock(&nm_i->build_lock);
1349         if (!nm_i->fcnt) {
1350                 /* scan NAT in order to build free nid list */
1351                 build_free_nids(sbi);
1352                 if (!nm_i->fcnt) {
1353                         mutex_unlock(&nm_i->build_lock);
1354                         return false;
1355                 }
1356         }
1357         mutex_unlock(&nm_i->build_lock);
1358
1359         /*
1360          * We check fcnt again since previous check is racy as
1361          * we didn't hold free_nid_list_lock. So other thread
1362          * could consume all of free nids.
1363          */
1364         spin_lock(&nm_i->free_nid_list_lock);
1365         if (!nm_i->fcnt) {
1366                 spin_unlock(&nm_i->free_nid_list_lock);
1367                 goto retry;
1368         }
1369
1370         BUG_ON(list_empty(&nm_i->free_nid_list));
1371         list_for_each(this, &nm_i->free_nid_list) {
1372                 i = list_entry(this, struct free_nid, list);
1373                 if (i->state == NID_NEW)
1374                         break;
1375         }
1376
1377         BUG_ON(i->state != NID_NEW);
1378         *nid = i->nid;
1379         i->state = NID_ALLOC;
1380         nm_i->fcnt--;
1381         spin_unlock(&nm_i->free_nid_list_lock);
1382         return true;
1383 }
1384
1385 /*
1386  * alloc_nid() should be called prior to this function.
1387  */
1388 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1389 {
1390         struct f2fs_nm_info *nm_i = NM_I(sbi);
1391         struct free_nid *i;
1392
1393         spin_lock(&nm_i->free_nid_list_lock);
1394         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1395         if (i) {
1396                 BUG_ON(i->state != NID_ALLOC);
1397                 __del_from_free_nid_list(i);
1398         }
1399         spin_unlock(&nm_i->free_nid_list_lock);
1400 }
1401
1402 /*
1403  * alloc_nid() should be called prior to this function.
1404  */
1405 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1406 {
1407         alloc_nid_done(sbi, nid);
1408         add_free_nid(NM_I(sbi), nid);
1409 }
1410
1411 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1412                 struct f2fs_summary *sum, struct node_info *ni,
1413                 block_t new_blkaddr)
1414 {
1415         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1416         set_node_addr(sbi, ni, new_blkaddr);
1417         clear_node_page_dirty(page);
1418 }
1419
1420 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1421 {
1422         struct address_space *mapping = sbi->node_inode->i_mapping;
1423         struct f2fs_node *src, *dst;
1424         nid_t ino = ino_of_node(page);
1425         struct node_info old_ni, new_ni;
1426         struct page *ipage;
1427
1428         ipage = grab_cache_page(mapping, ino);
1429         if (!ipage)
1430                 return -ENOMEM;
1431
1432         /* Should not use this inode  from free nid list */
1433         remove_free_nid(NM_I(sbi), ino);
1434
1435         get_node_info(sbi, ino, &old_ni);
1436         SetPageUptodate(ipage);
1437         fill_node_footer(ipage, ino, ino, 0, true);
1438
1439         src = (struct f2fs_node *)page_address(page);
1440         dst = (struct f2fs_node *)page_address(ipage);
1441
1442         memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i);
1443         dst->i.i_size = 0;
1444         dst->i.i_blocks = cpu_to_le64(1);
1445         dst->i.i_links = cpu_to_le32(1);
1446         dst->i.i_xattr_nid = 0;
1447
1448         new_ni = old_ni;
1449         new_ni.ino = ino;
1450
1451         set_node_addr(sbi, &new_ni, NEW_ADDR);
1452         inc_valid_inode_count(sbi);
1453
1454         f2fs_put_page(ipage, 1);
1455         return 0;
1456 }
1457
1458 int restore_node_summary(struct f2fs_sb_info *sbi,
1459                         unsigned int segno, struct f2fs_summary_block *sum)
1460 {
1461         struct f2fs_node *rn;
1462         struct f2fs_summary *sum_entry;
1463         struct page *page;
1464         block_t addr;
1465         int i, last_offset;
1466
1467         /* alloc temporal page for read node */
1468         page = alloc_page(GFP_NOFS | __GFP_ZERO);
1469         if (IS_ERR(page))
1470                 return PTR_ERR(page);
1471         lock_page(page);
1472
1473         /* scan the node segment */
1474         last_offset = sbi->blocks_per_seg;
1475         addr = START_BLOCK(sbi, segno);
1476         sum_entry = &sum->entries[0];
1477
1478         for (i = 0; i < last_offset; i++, sum_entry++) {
1479                 if (f2fs_readpage(sbi, page, addr, READ_SYNC))
1480                         goto out;
1481
1482                 rn = (struct f2fs_node *)page_address(page);
1483                 sum_entry->nid = rn->footer.nid;
1484                 sum_entry->version = 0;
1485                 sum_entry->ofs_in_node = 0;
1486                 addr++;
1487
1488                 /*
1489                  * In order to read next node page,
1490                  * we must clear PageUptodate flag.
1491                  */
1492                 ClearPageUptodate(page);
1493         }
1494 out:
1495         unlock_page(page);
1496         __free_pages(page, 0);
1497         return 0;
1498 }
1499
1500 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1501 {
1502         struct f2fs_nm_info *nm_i = NM_I(sbi);
1503         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1504         struct f2fs_summary_block *sum = curseg->sum_blk;
1505         int i;
1506
1507         mutex_lock(&curseg->curseg_mutex);
1508
1509         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1510                 mutex_unlock(&curseg->curseg_mutex);
1511                 return false;
1512         }
1513
1514         for (i = 0; i < nats_in_cursum(sum); i++) {
1515                 struct nat_entry *ne;
1516                 struct f2fs_nat_entry raw_ne;
1517                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1518
1519                 raw_ne = nat_in_journal(sum, i);
1520 retry:
1521                 write_lock(&nm_i->nat_tree_lock);
1522                 ne = __lookup_nat_cache(nm_i, nid);
1523                 if (ne) {
1524                         __set_nat_cache_dirty(nm_i, ne);
1525                         write_unlock(&nm_i->nat_tree_lock);
1526                         continue;
1527                 }
1528                 ne = grab_nat_entry(nm_i, nid);
1529                 if (!ne) {
1530                         write_unlock(&nm_i->nat_tree_lock);
1531                         goto retry;
1532                 }
1533                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1534                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1535                 nat_set_version(ne, raw_ne.version);
1536                 __set_nat_cache_dirty(nm_i, ne);
1537                 write_unlock(&nm_i->nat_tree_lock);
1538         }
1539         update_nats_in_cursum(sum, -i);
1540         mutex_unlock(&curseg->curseg_mutex);
1541         return true;
1542 }
1543
1544 /*
1545  * This function is called during the checkpointing process.
1546  */
1547 void flush_nat_entries(struct f2fs_sb_info *sbi)
1548 {
1549         struct f2fs_nm_info *nm_i = NM_I(sbi);
1550         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1551         struct f2fs_summary_block *sum = curseg->sum_blk;
1552         struct list_head *cur, *n;
1553         struct page *page = NULL;
1554         struct f2fs_nat_block *nat_blk = NULL;
1555         nid_t start_nid = 0, end_nid = 0;
1556         bool flushed;
1557
1558         flushed = flush_nats_in_journal(sbi);
1559
1560         if (!flushed)
1561                 mutex_lock(&curseg->curseg_mutex);
1562
1563         /* 1) flush dirty nat caches */
1564         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1565                 struct nat_entry *ne;
1566                 nid_t nid;
1567                 struct f2fs_nat_entry raw_ne;
1568                 int offset = -1;
1569                 block_t new_blkaddr;
1570
1571                 ne = list_entry(cur, struct nat_entry, list);
1572                 nid = nat_get_nid(ne);
1573
1574                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1575                         continue;
1576                 if (flushed)
1577                         goto to_nat_page;
1578
1579                 /* if there is room for nat enries in curseg->sumpage */
1580                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1581                 if (offset >= 0) {
1582                         raw_ne = nat_in_journal(sum, offset);
1583                         goto flush_now;
1584                 }
1585 to_nat_page:
1586                 if (!page || (start_nid > nid || nid > end_nid)) {
1587                         if (page) {
1588                                 f2fs_put_page(page, 1);
1589                                 page = NULL;
1590                         }
1591                         start_nid = START_NID(nid);
1592                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1593
1594                         /*
1595                          * get nat block with dirty flag, increased reference
1596                          * count, mapped and lock
1597                          */
1598                         page = get_next_nat_page(sbi, start_nid);
1599                         nat_blk = page_address(page);
1600                 }
1601
1602                 BUG_ON(!nat_blk);
1603                 raw_ne = nat_blk->entries[nid - start_nid];
1604 flush_now:
1605                 new_blkaddr = nat_get_blkaddr(ne);
1606
1607                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1608                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1609                 raw_ne.version = nat_get_version(ne);
1610
1611                 if (offset < 0) {
1612                         nat_blk->entries[nid - start_nid] = raw_ne;
1613                 } else {
1614                         nat_in_journal(sum, offset) = raw_ne;
1615                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1616                 }
1617
1618                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
1619                         write_lock(&nm_i->nat_tree_lock);
1620                         __del_from_nat_cache(nm_i, ne);
1621                         write_unlock(&nm_i->nat_tree_lock);
1622
1623                         /* We can reuse this freed nid at this point */
1624                         add_free_nid(NM_I(sbi), nid);
1625                 } else {
1626                         write_lock(&nm_i->nat_tree_lock);
1627                         __clear_nat_cache_dirty(nm_i, ne);
1628                         ne->checkpointed = true;
1629                         write_unlock(&nm_i->nat_tree_lock);
1630                 }
1631         }
1632         if (!flushed)
1633                 mutex_unlock(&curseg->curseg_mutex);
1634         f2fs_put_page(page, 1);
1635
1636         /* 2) shrink nat caches if necessary */
1637         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1638 }
1639
1640 static int init_node_manager(struct f2fs_sb_info *sbi)
1641 {
1642         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1643         struct f2fs_nm_info *nm_i = NM_I(sbi);
1644         unsigned char *version_bitmap;
1645         unsigned int nat_segs, nat_blocks;
1646
1647         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1648
1649         /* segment_count_nat includes pair segment so divide to 2. */
1650         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1651         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1652         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1653         nm_i->fcnt = 0;
1654         nm_i->nat_cnt = 0;
1655
1656         INIT_LIST_HEAD(&nm_i->free_nid_list);
1657         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1658         INIT_LIST_HEAD(&nm_i->nat_entries);
1659         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1660
1661         mutex_init(&nm_i->build_lock);
1662         spin_lock_init(&nm_i->free_nid_list_lock);
1663         rwlock_init(&nm_i->nat_tree_lock);
1664
1665         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1666         nm_i->init_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1667         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1668
1669         nm_i->nat_bitmap = kzalloc(nm_i->bitmap_size, GFP_KERNEL);
1670         if (!nm_i->nat_bitmap)
1671                 return -ENOMEM;
1672         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1673         if (!version_bitmap)
1674                 return -EFAULT;
1675
1676         /* copy version bitmap */
1677         memcpy(nm_i->nat_bitmap, version_bitmap, nm_i->bitmap_size);
1678         return 0;
1679 }
1680
1681 int build_node_manager(struct f2fs_sb_info *sbi)
1682 {
1683         int err;
1684
1685         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1686         if (!sbi->nm_info)
1687                 return -ENOMEM;
1688
1689         err = init_node_manager(sbi);
1690         if (err)
1691                 return err;
1692
1693         build_free_nids(sbi);
1694         return 0;
1695 }
1696
1697 void destroy_node_manager(struct f2fs_sb_info *sbi)
1698 {
1699         struct f2fs_nm_info *nm_i = NM_I(sbi);
1700         struct free_nid *i, *next_i;
1701         struct nat_entry *natvec[NATVEC_SIZE];
1702         nid_t nid = 0;
1703         unsigned int found;
1704
1705         if (!nm_i)
1706                 return;
1707
1708         /* destroy free nid list */
1709         spin_lock(&nm_i->free_nid_list_lock);
1710         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1711                 BUG_ON(i->state == NID_ALLOC);
1712                 __del_from_free_nid_list(i);
1713                 nm_i->fcnt--;
1714         }
1715         BUG_ON(nm_i->fcnt);
1716         spin_unlock(&nm_i->free_nid_list_lock);
1717
1718         /* destroy nat cache */
1719         write_lock(&nm_i->nat_tree_lock);
1720         while ((found = __gang_lookup_nat_cache(nm_i,
1721                                         nid, NATVEC_SIZE, natvec))) {
1722                 unsigned idx;
1723                 for (idx = 0; idx < found; idx++) {
1724                         struct nat_entry *e = natvec[idx];
1725                         nid = nat_get_nid(e) + 1;
1726                         __del_from_nat_cache(nm_i, e);
1727                 }
1728         }
1729         BUG_ON(nm_i->nat_cnt);
1730         write_unlock(&nm_i->nat_tree_lock);
1731
1732         kfree(nm_i->nat_bitmap);
1733         sbi->nm_info = NULL;
1734         kfree(nm_i);
1735 }
1736
1737 int __init create_node_manager_caches(void)
1738 {
1739         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1740                         sizeof(struct nat_entry), NULL);
1741         if (!nat_entry_slab)
1742                 return -ENOMEM;
1743
1744         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1745                         sizeof(struct free_nid), NULL);
1746         if (!free_nid_slab) {
1747                 kmem_cache_destroy(nat_entry_slab);
1748                 return -ENOMEM;
1749         }
1750         return 0;
1751 }
1752
1753 void destroy_node_manager_caches(void)
1754 {
1755         kmem_cache_destroy(free_nid_slab);
1756         kmem_cache_destroy(nat_entry_slab);
1757 }