fs/f2fs/node.c

   1 /*
   2  * fs/f2fs/node.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/mpage.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/swap.h>
  18
  19 #include "f2fs.h"
  20 #include "node.h"
  21 #include "segment.h"
  22 #include <trace/events/f2fs.h>
  23
  24 static struct kmem_cache *nat_entry_slab;
  25 static struct kmem_cache *free_nid_slab;
  26
  27 static void clear_node_page_dirty(struct page *page)
  28 {
  29         struct address_space *mapping = page->mapping;
  30         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
  31         unsigned int long flags;
  32
  33         if (PageDirty(page)) {
  34                 spin_lock_irqsave(&mapping->tree_lock, flags);
  35                 radix_tree_tag_clear(&mapping->page_tree,
  36                                 page_index(page),
  37                                 PAGECACHE_TAG_DIRTY);
  38                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  39
  40                 clear_page_dirty_for_io(page);
  41                 dec_page_count(sbi, F2FS_DIRTY_NODES);
  42         }
  43         ClearPageUptodate(page);
  44 }
  45
  46 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  47 {
  48         pgoff_t index = current_nat_addr(sbi, nid);
  49         return get_meta_page(sbi, index);
  50 }
  51
  52 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  53 {
  54         struct page *src_page;
  55         struct page *dst_page;
  56         pgoff_t src_off;
  57         pgoff_t dst_off;
  58         void *src_addr;
  59         void *dst_addr;
  60         struct f2fs_nm_info *nm_i = NM_I(sbi);
  61
  62         src_off = current_nat_addr(sbi, nid);
  63         dst_off = next_nat_addr(sbi, src_off);
  64
  65         /* get current nat block page with lock */
  66         src_page = get_meta_page(sbi, src_off);
  67
  68         /* Dirty src_page means that it is already the new target NAT page. */
  69         if (PageDirty(src_page))
  70                 return src_page;
  71
  72         dst_page = grab_meta_page(sbi, dst_off);
  73
  74         src_addr = page_address(src_page);
  75         dst_addr = page_address(dst_page);
  76         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
  77         set_page_dirty(dst_page);
  78         f2fs_put_page(src_page, 1);
  79
  80         set_to_next_nat(nm_i, nid);
  81
  82         return dst_page;
  83 }
  84
  85 /*
  86  * Readahead NAT pages
  87  */
  88 static void ra_nat_pages(struct f2fs_sb_info *sbi, int nid)
  89 {
  90         struct address_space *mapping = sbi->meta_inode->i_mapping;
  91         struct f2fs_nm_info *nm_i = NM_I(sbi);
  92         struct blk_plug plug;
  93         struct page *page;
  94         pgoff_t index;
  95         int i;
  96
  97         blk_start_plug(&plug);
  98
  99         for (i = 0; i < FREE_NID_PAGES; i++, nid += NAT_ENTRY_PER_BLOCK) {
 100                 if (nid >= nm_i->max_nid)
 101                         nid = 0;
 102                 index = current_nat_addr(sbi, nid);
 103
 104                 page = grab_cache_page(mapping, index);
 105                 if (!page)
 106                         continue;
 107                 if (PageUptodate(page)) {
 108                         f2fs_put_page(page, 1);
 109                         continue;
 110                 }
 111                 if (f2fs_readpage(sbi, page, index, READ))
 112                         continue;
 113
 114                 f2fs_put_page(page, 0);
 115         }
 116         blk_finish_plug(&plug);
 117 }
 118
 119 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 120 {
 121         return radix_tree_lookup(&nm_i->nat_root, n);
 122 }
 123
 124 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 125                 nid_t start, unsigned int nr, struct nat_entry **ep)
 126 {
 127         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 128 }
 129
 130 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 131 {
 132         list_del(&e->list);
 133         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 134         nm_i->nat_cnt--;
 135         kmem_cache_free(nat_entry_slab, e);
 136 }
 137
 138 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 139 {
 140         struct f2fs_nm_info *nm_i = NM_I(sbi);
 141         struct nat_entry *e;
 142         int is_cp = 1;
 143
 144         read_lock(&nm_i->nat_tree_lock);
 145         e = __lookup_nat_cache(nm_i, nid);
 146         if (e && !e->checkpointed)
 147                 is_cp = 0;
 148         read_unlock(&nm_i->nat_tree_lock);
 149         return is_cp;
 150 }
 151
 152 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 153 {
 154         struct nat_entry *new;
 155
 156         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 157         if (!new)
 158                 return NULL;
 159         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 160                 kmem_cache_free(nat_entry_slab, new);
 161                 return NULL;
 162         }
 163         memset(new, 0, sizeof(struct nat_entry));
 164         nat_set_nid(new, nid);
 165         list_add_tail(&new->list, &nm_i->nat_entries);
 166         nm_i->nat_cnt++;
 167         return new;
 168 }
 169
 170 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 171                                                 struct f2fs_nat_entry *ne)
 172 {
 173         struct nat_entry *e;
 174 retry:
 175         write_lock(&nm_i->nat_tree_lock);
 176         e = __lookup_nat_cache(nm_i, nid);
 177         if (!e) {
 178                 e = grab_nat_entry(nm_i, nid);
 179                 if (!e) {
 180                         write_unlock(&nm_i->nat_tree_lock);
 181                         goto retry;
 182                 }
 183                 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
 184                 nat_set_ino(e, le32_to_cpu(ne->ino));
 185                 nat_set_version(e, ne->version);
 186                 e->checkpointed = true;
 187         }
 188         write_unlock(&nm_i->nat_tree_lock);
 189 }
 190
 191 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 192                         block_t new_blkaddr)
 193 {
 194         struct f2fs_nm_info *nm_i = NM_I(sbi);
 195         struct nat_entry *e;
 196 retry:
 197         write_lock(&nm_i->nat_tree_lock);
 198         e = __lookup_nat_cache(nm_i, ni->nid);
 199         if (!e) {
 200                 e = grab_nat_entry(nm_i, ni->nid);
 201                 if (!e) {
 202                         write_unlock(&nm_i->nat_tree_lock);
 203                         goto retry;
 204                 }
 205                 e->ni = *ni;
 206                 e->checkpointed = true;
 207                 BUG_ON(ni->blk_addr == NEW_ADDR);
 208         } else if (new_blkaddr == NEW_ADDR) {
 209                 /*
 210                  * when nid is reallocated,
 211                  * previous nat entry can be remained in nat cache.
 212                  * So, reinitialize it with new information.
 213                  */
 214                 e->ni = *ni;
 215                 BUG_ON(ni->blk_addr != NULL_ADDR);
 216         }
 217
 218         if (new_blkaddr == NEW_ADDR)
 219                 e->checkpointed = false;
 220
 221         /* sanity check */
 222         BUG_ON(nat_get_blkaddr(e) != ni->blk_addr);
 223         BUG_ON(nat_get_blkaddr(e) == NULL_ADDR &&
 224                         new_blkaddr == NULL_ADDR);
 225         BUG_ON(nat_get_blkaddr(e) == NEW_ADDR &&
 226                         new_blkaddr == NEW_ADDR);
 227         BUG_ON(nat_get_blkaddr(e) != NEW_ADDR &&
 228                         nat_get_blkaddr(e) != NULL_ADDR &&
 229                         new_blkaddr == NEW_ADDR);
 230
 231         /* increament version no as node is removed */
 232         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 233                 unsigned char version = nat_get_version(e);
 234                 nat_set_version(e, inc_node_version(version));
 235         }
 236
 237         /* change address */
 238         nat_set_blkaddr(e, new_blkaddr);
 239         __set_nat_cache_dirty(nm_i, e);
 240         write_unlock(&nm_i->nat_tree_lock);
 241 }
 242
 243 static int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 244 {
 245         struct f2fs_nm_info *nm_i = NM_I(sbi);
 246
 247         if (nm_i->nat_cnt < 2 * NM_WOUT_THRESHOLD)
 248                 return 0;
 249
 250         write_lock(&nm_i->nat_tree_lock);
 251         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 252                 struct nat_entry *ne;
 253                 ne = list_first_entry(&nm_i->nat_entries,
 254                                         struct nat_entry, list);
 255                 __del_from_nat_cache(nm_i, ne);
 256                 nr_shrink--;
 257         }
 258         write_unlock(&nm_i->nat_tree_lock);
 259         return nr_shrink;
 260 }
 261
 262 /*
 263  * This function returns always success
 264  */
 265 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 266 {
 267         struct f2fs_nm_info *nm_i = NM_I(sbi);
 268         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 269         struct f2fs_summary_block *sum = curseg->sum_blk;
 270         nid_t start_nid = START_NID(nid);
 271         struct f2fs_nat_block *nat_blk;
 272         struct page *page = NULL;
 273         struct f2fs_nat_entry ne;
 274         struct nat_entry *e;
 275         int i;
 276
 277         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 278         ni->nid = nid;
 279
 280         /* Check nat cache */
 281         read_lock(&nm_i->nat_tree_lock);
 282         e = __lookup_nat_cache(nm_i, nid);
 283         if (e) {
 284                 ni->ino = nat_get_ino(e);
 285                 ni->blk_addr = nat_get_blkaddr(e);
 286                 ni->version = nat_get_version(e);
 287         }
 288         read_unlock(&nm_i->nat_tree_lock);
 289         if (e)
 290                 return;
 291
 292         /* Check current segment summary */
 293         mutex_lock(&curseg->curseg_mutex);
 294         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 295         if (i >= 0) {
 296                 ne = nat_in_journal(sum, i);
 297                 node_info_from_raw_nat(ni, &ne);
 298         }
 299         mutex_unlock(&curseg->curseg_mutex);
 300         if (i >= 0)
 301                 goto cache;
 302
 303         /* Fill node_info from nat page */
 304         page = get_current_nat_page(sbi, start_nid);
 305         nat_blk = (struct f2fs_nat_block *)page_address(page);
 306         ne = nat_blk->entries[nid - start_nid];
 307         node_info_from_raw_nat(ni, &ne);
 308         f2fs_put_page(page, 1);
 309 cache:
 310         /* cache nat entry */
 311         cache_nat_entry(NM_I(sbi), nid, &ne);
 312 }
 313
 314 /*
 315  * The maximum depth is four.
 316  * Offset[0] will have raw inode offset.
 317  */
 318 static int get_node_path(long block, int offset[4], unsigned int noffset[4])
 319 {
 320         const long direct_index = ADDRS_PER_INODE;
 321         const long direct_blks = ADDRS_PER_BLOCK;
 322         const long dptrs_per_blk = NIDS_PER_BLOCK;
 323         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 324         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 325         int n = 0;
 326         int level = 0;
 327
 328         noffset[0] = 0;
 329
 330         if (block < direct_index) {
 331                 offset[n] = block;
 332                 goto got;
 333         }
 334         block -= direct_index;
 335         if (block < direct_blks) {
 336                 offset[n++] = NODE_DIR1_BLOCK;
 337                 noffset[n] = 1;
 338                 offset[n] = block;
 339                 level = 1;
 340                 goto got;
 341         }
 342         block -= direct_blks;
 343         if (block < direct_blks) {
 344                 offset[n++] = NODE_DIR2_BLOCK;
 345                 noffset[n] = 2;
 346                 offset[n] = block;
 347                 level = 1;
 348                 goto got;
 349         }
 350         block -= direct_blks;
 351         if (block < indirect_blks) {
 352                 offset[n++] = NODE_IND1_BLOCK;
 353                 noffset[n] = 3;
 354                 offset[n++] = block / direct_blks;
 355                 noffset[n] = 4 + offset[n - 1];
 356                 offset[n] = block % direct_blks;
 357                 level = 2;
 358                 goto got;
 359         }
 360         block -= indirect_blks;
 361         if (block < indirect_blks) {
 362                 offset[n++] = NODE_IND2_BLOCK;
 363                 noffset[n] = 4 + dptrs_per_blk;
 364                 offset[n++] = block / direct_blks;
 365                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 366                 offset[n] = block % direct_blks;
 367                 level = 2;
 368                 goto got;
 369         }
 370         block -= indirect_blks;
 371         if (block < dindirect_blks) {
 372                 offset[n++] = NODE_DIND_BLOCK;
 373                 noffset[n] = 5 + (dptrs_per_blk * 2);
 374                 offset[n++] = block / indirect_blks;
 375                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 376                               offset[n - 1] * (dptrs_per_blk + 1);
 377                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 378                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 379                               offset[n - 2] * (dptrs_per_blk + 1) +
 380                               offset[n - 1];
 381                 offset[n] = block % direct_blks;
 382                 level = 3;
 383                 goto got;
 384         } else {
 385                 BUG();
 386         }
 387 got:
 388         return level;
 389 }
 390
 391 /*
 392  * Caller should call f2fs_put_dnode(dn).
 393  * Also, it should grab and release a mutex by calling mutex_lock_op() and
 394  * mutex_unlock_op() only if ro is not set RDONLY_NODE.
 395  * In the case of RDONLY_NODE, we don't need to care about mutex.
 396  */
 397 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 398 {
 399         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 400         struct page *npage[4];
 401         struct page *parent;
 402         int offset[4];
 403         unsigned int noffset[4];
 404         nid_t nids[4];
 405         int level, i;
 406         int err = 0;
 407
 408         level = get_node_path(index, offset, noffset);
 409
 410         nids[0] = dn->inode->i_ino;
 411         npage[0] = get_node_page(sbi, nids[0]);
 412         if (IS_ERR(npage[0]))
 413                 return PTR_ERR(npage[0]);
 414
 415         parent = npage[0];
 416         if (level != 0)
 417                 nids[1] = get_nid(parent, offset[0], true);
 418         dn->inode_page = npage[0];
 419         dn->inode_page_locked = true;
 420
 421         /* get indirect or direct nodes */
 422         for (i = 1; i <= level; i++) {
 423                 bool done = false;
 424
 425                 if (!nids[i] && mode == ALLOC_NODE) {
 426                         /* alloc new node */
 427                         if (!alloc_nid(sbi, &(nids[i]))) {
 428                                 err = -ENOSPC;
 429                                 goto release_pages;
 430                         }
 431
 432                         dn->nid = nids[i];
 433                         npage[i] = new_node_page(dn, noffset[i]);
 434                         if (IS_ERR(npage[i])) {
 435                                 alloc_nid_failed(sbi, nids[i]);
 436                                 err = PTR_ERR(npage[i]);
 437                                 goto release_pages;
 438                         }
 439
 440                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 441                         alloc_nid_done(sbi, nids[i]);
 442                         done = true;
 443                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 444                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 445                         if (IS_ERR(npage[i])) {
 446                                 err = PTR_ERR(npage[i]);
 447                                 goto release_pages;
 448                         }
 449                         done = true;
 450                 }
 451                 if (i == 1) {
 452                         dn->inode_page_locked = false;
 453                         unlock_page(parent);
 454                 } else {
 455                         f2fs_put_page(parent, 1);
 456                 }
 457
 458                 if (!done) {
 459                         npage[i] = get_node_page(sbi, nids[i]);
 460                         if (IS_ERR(npage[i])) {
 461                                 err = PTR_ERR(npage[i]);
 462                                 f2fs_put_page(npage[0], 0);
 463                                 goto release_out;
 464                         }
 465                 }
 466                 if (i < level) {
 467                         parent = npage[i];
 468                         nids[i + 1] = get_nid(parent, offset[i], false);
 469                 }
 470         }
 471         dn->nid = nids[level];
 472         dn->ofs_in_node = offset[level];
 473         dn->node_page = npage[level];
 474         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 475         return 0;
 476
 477 release_pages:
 478         f2fs_put_page(parent, 1);
 479         if (i > 1)
 480                 f2fs_put_page(npage[0], 0);
 481 release_out:
 482         dn->inode_page = NULL;
 483         dn->node_page = NULL;
 484         return err;
 485 }
 486
 487 static void truncate_node(struct dnode_of_data *dn)
 488 {
 489         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 490         struct node_info ni;
 491
 492         get_node_info(sbi, dn->nid, &ni);
 493         if (dn->inode->i_blocks == 0) {
 494                 BUG_ON(ni.blk_addr != NULL_ADDR);
 495                 goto invalidate;
 496         }
 497         BUG_ON(ni.blk_addr == NULL_ADDR);
 498
 499         /* Deallocate node address */
 500         invalidate_blocks(sbi, ni.blk_addr);
 501         dec_valid_node_count(sbi, dn->inode, 1);
 502         set_node_addr(sbi, &ni, NULL_ADDR);
 503
 504         if (dn->nid == dn->inode->i_ino) {
 505                 remove_orphan_inode(sbi, dn->nid);
 506                 dec_valid_inode_count(sbi);
 507         } else {
 508                 sync_inode_page(dn);
 509         }
 510 invalidate:
 511         clear_node_page_dirty(dn->node_page);
 512         F2FS_SET_SB_DIRT(sbi);
 513
 514         f2fs_put_page(dn->node_page, 1);
 515         dn->node_page = NULL;
 516         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 517 }
 518
 519 static int truncate_dnode(struct dnode_of_data *dn)
 520 {
 521         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 522         struct page *page;
 523
 524         if (dn->nid == 0)
 525                 return 1;
 526
 527         /* get direct node */
 528         page = get_node_page(sbi, dn->nid);
 529         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 530                 return 1;
 531         else if (IS_ERR(page))
 532                 return PTR_ERR(page);
 533
 534         /* Make dnode_of_data for parameter */
 535         dn->node_page = page;
 536         dn->ofs_in_node = 0;
 537         truncate_data_blocks(dn);
 538         truncate_node(dn);
 539         return 1;
 540 }
 541
 542 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 543                                                 int ofs, int depth)
 544 {
 545         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 546         struct dnode_of_data rdn = *dn;
 547         struct page *page;
 548         struct f2fs_node *rn;
 549         nid_t child_nid;
 550         unsigned int child_nofs;
 551         int freed = 0;
 552         int i, ret;
 553
 554         if (dn->nid == 0)
 555                 return NIDS_PER_BLOCK + 1;
 556
 557         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 558
 559         page = get_node_page(sbi, dn->nid);
 560         if (IS_ERR(page)) {
 561                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 562                 return PTR_ERR(page);
 563         }
 564
 565         rn = (struct f2fs_node *)page_address(page);
 566         if (depth < 3) {
 567                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 568                         child_nid = le32_to_cpu(rn->in.nid[i]);
 569                         if (child_nid == 0)
 570                                 continue;
 571                         rdn.nid = child_nid;
 572                         ret = truncate_dnode(&rdn);
 573                         if (ret < 0)
 574                                 goto out_err;
 575                         set_nid(page, i, 0, false);
 576                 }
 577         } else {
 578                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 579                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 580                         child_nid = le32_to_cpu(rn->in.nid[i]);
 581                         if (child_nid == 0) {
 582                                 child_nofs += NIDS_PER_BLOCK + 1;
 583                                 continue;
 584                         }
 585                         rdn.nid = child_nid;
 586                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 587                         if (ret == (NIDS_PER_BLOCK + 1)) {
 588                                 set_nid(page, i, 0, false);
 589                                 child_nofs += ret;
 590                         } else if (ret < 0 && ret != -ENOENT) {
 591                                 goto out_err;
 592                         }
 593                 }
 594                 freed = child_nofs;
 595         }
 596
 597         if (!ofs) {
 598                 /* remove current indirect node */
 599                 dn->node_page = page;
 600                 truncate_node(dn);
 601                 freed++;
 602         } else {
 603                 f2fs_put_page(page, 1);
 604         }
 605         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 606         return freed;
 607
 608 out_err:
 609         f2fs_put_page(page, 1);
 610         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 611         return ret;
 612 }
 613
 614 static int truncate_partial_nodes(struct dnode_of_data *dn,
 615                         struct f2fs_inode *ri, int *offset, int depth)
 616 {
 617         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 618         struct page *pages[2];
 619         nid_t nid[3];
 620         nid_t child_nid;
 621         int err = 0;
 622         int i;
 623         int idx = depth - 2;
 624
 625         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 626         if (!nid[0])
 627                 return 0;
 628
 629         /* get indirect nodes in the path */
 630         for (i = 0; i < depth - 1; i++) {
 631                 /* refernece count'll be increased */
 632                 pages[i] = get_node_page(sbi, nid[i]);
 633                 if (IS_ERR(pages[i])) {
 634                         depth = i + 1;
 635                         err = PTR_ERR(pages[i]);
 636                         goto fail;
 637                 }
 638                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 639         }
 640
 641         /* free direct nodes linked to a partial indirect node */
 642         for (i = offset[depth - 1]; i < NIDS_PER_BLOCK; i++) {
 643                 child_nid = get_nid(pages[idx], i, false);
 644                 if (!child_nid)
 645                         continue;
 646                 dn->nid = child_nid;
 647                 err = truncate_dnode(dn);
 648                 if (err < 0)
 649                         goto fail;
 650                 set_nid(pages[idx], i, 0, false);
 651         }
 652
 653         if (offset[depth - 1] == 0) {
 654                 dn->node_page = pages[idx];
 655                 dn->nid = nid[idx];
 656                 truncate_node(dn);
 657         } else {
 658                 f2fs_put_page(pages[idx], 1);
 659         }
 660         offset[idx]++;
 661         offset[depth - 1] = 0;
 662 fail:
 663         for (i = depth - 3; i >= 0; i--)
 664                 f2fs_put_page(pages[i], 1);
 665
 666         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 667
 668         return err;
 669 }
 670
 671 /*
 672  * All the block addresses of data and nodes should be nullified.
 673  */
 674 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 675 {
 676         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 677         int err = 0, cont = 1;
 678         int level, offset[4], noffset[4];
 679         unsigned int nofs = 0;
 680         struct f2fs_node *rn;
 681         struct dnode_of_data dn;
 682         struct page *page;
 683
 684         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 685
 686         level = get_node_path(from, offset, noffset);
 687
 688         page = get_node_page(sbi, inode->i_ino);
 689         if (IS_ERR(page)) {
 690                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 691                 return PTR_ERR(page);
 692         }
 693
 694         set_new_dnode(&dn, inode, page, NULL, 0);
 695         unlock_page(page);
 696
 697         rn = page_address(page);
 698         switch (level) {
 699         case 0:
 700         case 1:
 701                 nofs = noffset[1];
 702                 break;
 703         case 2:
 704                 nofs = noffset[1];
 705                 if (!offset[level - 1])
 706                         goto skip_partial;
 707                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 708                 if (err < 0 && err != -ENOENT)
 709                         goto fail;
 710                 nofs += 1 + NIDS_PER_BLOCK;
 711                 break;
 712         case 3:
 713                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 714                 if (!offset[level - 1])
 715                         goto skip_partial;
 716                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 717                 if (err < 0 && err != -ENOENT)
 718                         goto fail;
 719                 break;
 720         default:
 721                 BUG();
 722         }
 723
 724 skip_partial:
 725         while (cont) {
 726                 dn.nid = le32_to_cpu(rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]);
 727                 switch (offset[0]) {
 728                 case NODE_DIR1_BLOCK:
 729                 case NODE_DIR2_BLOCK:
 730                         err = truncate_dnode(&dn);
 731                         break;
 732
 733                 case NODE_IND1_BLOCK:
 734                 case NODE_IND2_BLOCK:
 735                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 736                         break;
 737
 738                 case NODE_DIND_BLOCK:
 739                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 740                         cont = 0;
 741                         break;
 742
 743                 default:
 744                         BUG();
 745                 }
 746                 if (err < 0 && err != -ENOENT)
 747                         goto fail;
 748                 if (offset[1] == 0 &&
 749                                 rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 750                         lock_page(page);
 751                         wait_on_page_writeback(page);
 752                         rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 753                         set_page_dirty(page);
 754                         unlock_page(page);
 755                 }
 756                 offset[1] = 0;
 757                 offset[0]++;
 758                 nofs += err;
 759         }
 760 fail:
 761         f2fs_put_page(page, 0);
 762         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 763         return err > 0 ? 0 : err;
 764 }
 765
 766 /*
 767  * Caller should grab and release a mutex by calling mutex_lock_op() and
 768  * mutex_unlock_op().
 769  */
 770 int remove_inode_page(struct inode *inode)
 771 {
 772         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 773         struct page *page;
 774         nid_t ino = inode->i_ino;
 775         struct dnode_of_data dn;
 776
 777         page = get_node_page(sbi, ino);
 778         if (IS_ERR(page))
 779                 return PTR_ERR(page);
 780
 781         if (F2FS_I(inode)->i_xattr_nid) {
 782                 nid_t nid = F2FS_I(inode)->i_xattr_nid;
 783                 struct page *npage = get_node_page(sbi, nid);
 784
 785                 if (IS_ERR(npage))
 786                         return PTR_ERR(npage);
 787
 788                 F2FS_I(inode)->i_xattr_nid = 0;
 789                 set_new_dnode(&dn, inode, page, npage, nid);
 790                 dn.inode_page_locked = 1;
 791                 truncate_node(&dn);
 792         }
 793
 794         /* 0 is possible, after f2fs_new_inode() is failed */
 795         BUG_ON(inode->i_blocks != 0 && inode->i_blocks != 1);
 796         set_new_dnode(&dn, inode, page, page, ino);
 797         truncate_node(&dn);
 798         return 0;
 799 }
 800
 801 int new_inode_page(struct inode *inode, const struct qstr *name)
 802 {
 803         struct page *page;
 804         struct dnode_of_data dn;
 805
 806         /* allocate inode page for new inode */
 807         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 808         page = new_node_page(&dn, 0);
 809         init_dent_inode(name, page);
 810         if (IS_ERR(page))
 811                 return PTR_ERR(page);
 812         f2fs_put_page(page, 1);
 813         return 0;
 814 }
 815
 816 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
 817 {
 818         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 819         struct address_space *mapping = sbi->node_inode->i_mapping;
 820         struct node_info old_ni, new_ni;
 821         struct page *page;
 822         int err;
 823
 824         if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
 825                 return ERR_PTR(-EPERM);
 826
 827         page = grab_cache_page(mapping, dn->nid);
 828         if (!page)
 829                 return ERR_PTR(-ENOMEM);
 830
 831         get_node_info(sbi, dn->nid, &old_ni);
 832
 833         SetPageUptodate(page);
 834         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 835
 836         /* Reinitialize old_ni with new node page */
 837         BUG_ON(old_ni.blk_addr != NULL_ADDR);
 838         new_ni = old_ni;
 839         new_ni.ino = dn->inode->i_ino;
 840
 841         if (!inc_valid_node_count(sbi, dn->inode, 1)) {
 842                 err = -ENOSPC;
 843                 goto fail;
 844         }
 845         set_node_addr(sbi, &new_ni, NEW_ADDR);
 846         set_cold_node(dn->inode, page);
 847
 848         dn->node_page = page;
 849         sync_inode_page(dn);
 850         set_page_dirty(page);
 851         if (ofs == 0)
 852                 inc_valid_inode_count(sbi);
 853
 854         return page;
 855
 856 fail:
 857         clear_node_page_dirty(page);
 858         f2fs_put_page(page, 1);
 859         return ERR_PTR(err);
 860 }
 861
 862 /*
 863  * Caller should do after getting the following values.
 864  * 0: f2fs_put_page(page, 0)
 865  * LOCKED_PAGE: f2fs_put_page(page, 1)
 866  * error: nothing
 867  */
 868 static int read_node_page(struct page *page, int type)
 869 {
 870         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 871         struct node_info ni;
 872
 873         get_node_info(sbi, page->index, &ni);
 874
 875         if (ni.blk_addr == NULL_ADDR) {
 876                 f2fs_put_page(page, 1);
 877                 return -ENOENT;
 878         }
 879
 880         if (PageUptodate(page))
 881                 return LOCKED_PAGE;
 882
 883         return f2fs_readpage(sbi, page, ni.blk_addr, type);
 884 }
 885
 886 /*
 887  * Readahead a node page
 888  */
 889 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 890 {
 891         struct address_space *mapping = sbi->node_inode->i_mapping;
 892         struct page *apage;
 893         int err;
 894
 895         apage = find_get_page(mapping, nid);
 896         if (apage && PageUptodate(apage)) {
 897                 f2fs_put_page(apage, 0);
 898                 return;
 899         }
 900         f2fs_put_page(apage, 0);
 901
 902         apage = grab_cache_page(mapping, nid);
 903         if (!apage)
 904                 return;
 905
 906         err = read_node_page(apage, READA);
 907         if (err == 0)
 908                 f2fs_put_page(apage, 0);
 909         else if (err == LOCKED_PAGE)
 910                 f2fs_put_page(apage, 1);
 911         return;
 912 }
 913
 914 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 915 {
 916         struct address_space *mapping = sbi->node_inode->i_mapping;
 917         struct page *page;
 918         int err;
 919
 920         page = grab_cache_page(mapping, nid);
 921         if (!page)
 922                 return ERR_PTR(-ENOMEM);
 923
 924         err = read_node_page(page, READ_SYNC);
 925         if (err < 0)
 926                 return ERR_PTR(err);
 927         else if (err == LOCKED_PAGE)
 928                 goto got_it;
 929
 930         lock_page(page);
 931         if (!PageUptodate(page)) {
 932                 f2fs_put_page(page, 1);
 933                 return ERR_PTR(-EIO);
 934         }
 935 got_it:
 936         BUG_ON(nid != nid_of_node(page));
 937         mark_page_accessed(page);
 938         return page;
 939 }
 940
 941 /*
 942  * Return a locked page for the desired node page.
 943  * And, readahead MAX_RA_NODE number of node pages.
 944  */
 945 struct page *get_node_page_ra(struct page *parent, int start)
 946 {
 947         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 948         struct address_space *mapping = sbi->node_inode->i_mapping;
 949         struct blk_plug plug;
 950         struct page *page;
 951         int err, i, end;
 952         nid_t nid;
 953
 954         /* First, try getting the desired direct node. */
 955         nid = get_nid(parent, start, false);
 956         if (!nid)
 957                 return ERR_PTR(-ENOENT);
 958
 959         page = grab_cache_page(mapping, nid);
 960         if (!page)
 961                 return ERR_PTR(-ENOMEM);
 962
 963         err = read_node_page(page, READ_SYNC);
 964         if (err < 0)
 965                 return ERR_PTR(err);
 966         else if (err == LOCKED_PAGE)
 967                 goto page_hit;
 968
 969         blk_start_plug(&plug);
 970
 971         /* Then, try readahead for siblings of the desired node */
 972         end = start + MAX_RA_NODE;
 973         end = min(end, NIDS_PER_BLOCK);
 974         for (i = start + 1; i < end; i++) {
 975                 nid = get_nid(parent, i, false);
 976                 if (!nid)
 977                         continue;
 978                 ra_node_page(sbi, nid);
 979         }
 980
 981         blk_finish_plug(&plug);
 982
 983         lock_page(page);
 984
 985 page_hit:
 986         if (!PageUptodate(page)) {
 987                 f2fs_put_page(page, 1);
 988                 return ERR_PTR(-EIO);
 989         }
 990         mark_page_accessed(page);
 991         return page;
 992 }
 993
 994 void sync_inode_page(struct dnode_of_data *dn)
 995 {
 996         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
 997                 update_inode(dn->inode, dn->node_page);
 998         } else if (dn->inode_page) {
 999                 if (!dn->inode_page_locked)
1000                         lock_page(dn->inode_page);
1001                 update_inode(dn->inode, dn->inode_page);
1002                 if (!dn->inode_page_locked)
1003                         unlock_page(dn->inode_page);
1004         } else {
1005                 update_inode_page(dn->inode);
1006         }
1007 }
1008
1009 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1010                                         struct writeback_control *wbc)
1011 {
1012         struct address_space *mapping = sbi->node_inode->i_mapping;
1013         pgoff_t index, end;
1014         struct pagevec pvec;
1015         int step = ino ? 2 : 0;
1016         int nwritten = 0, wrote = 0;
1017
1018         pagevec_init(&pvec, 0);
1019
1020 next_step:
1021         index = 0;
1022         end = LONG_MAX;
1023
1024         while (index <= end) {
1025                 int i, nr_pages;
1026                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1027                                 PAGECACHE_TAG_DIRTY,
1028                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1029                 if (nr_pages == 0)
1030                         break;
1031
1032                 for (i = 0; i < nr_pages; i++) {
1033                         struct page *page = pvec.pages[i];
1034
1035                         /*
1036                          * flushing sequence with step:
1037                          * 0. indirect nodes
1038                          * 1. dentry dnodes
1039                          * 2. file dnodes
1040                          */
1041                         if (step == 0 && IS_DNODE(page))
1042                                 continue;
1043                         if (step == 1 && (!IS_DNODE(page) ||
1044                                                 is_cold_node(page)))
1045                                 continue;
1046                         if (step == 2 && (!IS_DNODE(page) ||
1047                                                 !is_cold_node(page)))
1048                                 continue;
1049
1050                         /*
1051                          * If an fsync mode,
1052                          * we should not skip writing node pages.
1053                          */
1054                         if (ino && ino_of_node(page) == ino)
1055                                 lock_page(page);
1056                         else if (!trylock_page(page))
1057                                 continue;
1058
1059                         if (unlikely(page->mapping != mapping)) {
1060 continue_unlock:
1061                                 unlock_page(page);
1062                                 continue;
1063                         }
1064                         if (ino && ino_of_node(page) != ino)
1065                                 goto continue_unlock;
1066
1067                         if (!PageDirty(page)) {
1068                                 /* someone wrote it for us */
1069                                 goto continue_unlock;
1070                         }
1071
1072                         if (!clear_page_dirty_for_io(page))
1073                                 goto continue_unlock;
1074
1075                         /* called by fsync() */
1076                         if (ino && IS_DNODE(page)) {
1077                                 int mark = !is_checkpointed_node(sbi, ino);
1078                                 set_fsync_mark(page, 1);
1079                                 if (IS_INODE(page))
1080                                         set_dentry_mark(page, mark);
1081                                 nwritten++;
1082                         } else {
1083                                 set_fsync_mark(page, 0);
1084                                 set_dentry_mark(page, 0);
1085                         }
1086                         mapping->a_ops->writepage(page, wbc);
1087                         wrote++;
1088
1089                         if (--wbc->nr_to_write == 0)
1090                                 break;
1091                 }
1092                 pagevec_release(&pvec);
1093                 cond_resched();
1094
1095                 if (wbc->nr_to_write == 0) {
1096                         step = 2;
1097                         break;
1098                 }
1099         }
1100
1101         if (step < 2) {
1102                 step++;
1103                 goto next_step;
1104         }
1105
1106         if (wrote)
1107                 f2fs_submit_bio(sbi, NODE, wbc->sync_mode == WB_SYNC_ALL);
1108
1109         return nwritten;
1110 }
1111
1112 static int f2fs_write_node_page(struct page *page,
1113                                 struct writeback_control *wbc)
1114 {
1115         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1116         nid_t nid;
1117         block_t new_addr;
1118         struct node_info ni;
1119
1120         wait_on_page_writeback(page);
1121
1122         /* get old block addr of this node page */
1123         nid = nid_of_node(page);
1124         BUG_ON(page->index != nid);
1125
1126         get_node_info(sbi, nid, &ni);
1127
1128         /* This page is already truncated */
1129         if (ni.blk_addr == NULL_ADDR) {
1130                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1131                 unlock_page(page);
1132                 return 0;
1133         }
1134
1135         if (wbc->for_reclaim) {
1136                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1137                 wbc->pages_skipped++;
1138                 set_page_dirty(page);
1139                 return AOP_WRITEPAGE_ACTIVATE;
1140         }
1141
1142         mutex_lock(&sbi->node_write);
1143         set_page_writeback(page);
1144         write_node_page(sbi, page, nid, ni.blk_addr, &new_addr);
1145         set_node_addr(sbi, &ni, new_addr);
1146         dec_page_count(sbi, F2FS_DIRTY_NODES);
1147         mutex_unlock(&sbi->node_write);
1148         unlock_page(page);
1149         return 0;
1150 }
1151
1152 /*
1153  * It is very important to gather dirty pages and write at once, so that we can
1154  * submit a big bio without interfering other data writes.
1155  * Be default, 512 pages (2MB), a segment size, is quite reasonable.
1156  */
1157 #define COLLECT_DIRTY_NODES     512
1158 static int f2fs_write_node_pages(struct address_space *mapping,
1159                             struct writeback_control *wbc)
1160 {
1161         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1162         struct block_device *bdev = sbi->sb->s_bdev;
1163         long nr_to_write = wbc->nr_to_write;
1164
1165         /* First check balancing cached NAT entries */
1166         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK)) {
1167                 f2fs_sync_fs(sbi->sb, true);
1168                 return 0;
1169         }
1170
1171         /* collect a number of dirty node pages and write together */
1172         if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1173                 return 0;
1174
1175         /* if mounting is failed, skip writing node pages */
1176         wbc->nr_to_write = bio_get_nr_vecs(bdev);
1177         sync_node_pages(sbi, 0, wbc);
1178         wbc->nr_to_write = nr_to_write -
1179                 (bio_get_nr_vecs(bdev) - wbc->nr_to_write);
1180         return 0;
1181 }
1182
1183 static int f2fs_set_node_page_dirty(struct page *page)
1184 {
1185         struct address_space *mapping = page->mapping;
1186         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1187
1188         SetPageUptodate(page);
1189         if (!PageDirty(page)) {
1190                 __set_page_dirty_nobuffers(page);
1191                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1192                 SetPagePrivate(page);
1193                 return 1;
1194         }
1195         return 0;
1196 }
1197
1198 static void f2fs_invalidate_node_page(struct page *page, unsigned long offset)
1199 {
1200         struct inode *inode = page->mapping->host;
1201         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1202         if (PageDirty(page))
1203                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1204         ClearPagePrivate(page);
1205 }
1206
1207 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1208 {
1209         ClearPagePrivate(page);
1210         return 1;
1211 }
1212
1213 /*
1214  * Structure of the f2fs node operations
1215  */
1216 const struct address_space_operations f2fs_node_aops = {
1217         .writepage      = f2fs_write_node_page,
1218         .writepages     = f2fs_write_node_pages,
1219         .set_page_dirty = f2fs_set_node_page_dirty,
1220         .invalidatepage = f2fs_invalidate_node_page,
1221         .releasepage    = f2fs_release_node_page,
1222 };
1223
1224 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1225 {
1226         struct list_head *this;
1227         struct free_nid *i;
1228         list_for_each(this, head) {
1229                 i = list_entry(this, struct free_nid, list);
1230                 if (i->nid == n)
1231                         return i;
1232         }
1233         return NULL;
1234 }
1235
1236 static void __del_from_free_nid_list(struct free_nid *i)
1237 {
1238         list_del(&i->list);
1239         kmem_cache_free(free_nid_slab, i);
1240 }
1241
1242 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1243 {
1244         struct free_nid *i;
1245
1246         if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1247                 return 0;
1248
1249         /* 0 nid should not be used */
1250         if (nid == 0)
1251                 return 0;
1252 retry:
1253         i = kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1254         if (!i) {
1255                 cond_resched();
1256                 goto retry;
1257         }
1258         i->nid = nid;
1259         i->state = NID_NEW;
1260
1261         spin_lock(&nm_i->free_nid_list_lock);
1262         if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1263                 spin_unlock(&nm_i->free_nid_list_lock);
1264                 kmem_cache_free(free_nid_slab, i);
1265                 return 0;
1266         }
1267         list_add_tail(&i->list, &nm_i->free_nid_list);
1268         nm_i->fcnt++;
1269         spin_unlock(&nm_i->free_nid_list_lock);
1270         return 1;
1271 }
1272
1273 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1274 {
1275         struct free_nid *i;
1276         spin_lock(&nm_i->free_nid_list_lock);
1277         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1278         if (i && i->state == NID_NEW) {
1279                 __del_from_free_nid_list(i);
1280                 nm_i->fcnt--;
1281         }
1282         spin_unlock(&nm_i->free_nid_list_lock);
1283 }
1284
1285 static int scan_nat_page(struct f2fs_nm_info *nm_i,
1286                         struct page *nat_page, nid_t start_nid)
1287 {
1288         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1289         block_t blk_addr;
1290         int fcnt = 0;
1291         int i;
1292
1293         i = start_nid % NAT_ENTRY_PER_BLOCK;
1294
1295         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1296                 if (start_nid >= nm_i->max_nid)
1297                         break;
1298                 blk_addr  = le32_to_cpu(nat_blk->entries[i].block_addr);
1299                 BUG_ON(blk_addr == NEW_ADDR);
1300                 if (blk_addr == NULL_ADDR)
1301                         fcnt += add_free_nid(nm_i, start_nid);
1302         }
1303         return fcnt;
1304 }
1305
1306 static void build_free_nids(struct f2fs_sb_info *sbi)
1307 {
1308         struct free_nid *fnid, *next_fnid;
1309         struct f2fs_nm_info *nm_i = NM_I(sbi);
1310         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1311         struct f2fs_summary_block *sum = curseg->sum_blk;
1312         nid_t nid = 0;
1313         bool is_cycled = false;
1314         int fcnt = 0;
1315         int i;
1316
1317         nid = nm_i->next_scan_nid;
1318         nm_i->init_scan_nid = nid;
1319
1320         ra_nat_pages(sbi, nid);
1321
1322         while (1) {
1323                 struct page *page = get_current_nat_page(sbi, nid);
1324
1325                 fcnt += scan_nat_page(nm_i, page, nid);
1326                 f2fs_put_page(page, 1);
1327
1328                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1329
1330                 if (nid >= nm_i->max_nid) {
1331                         nid = 0;
1332                         is_cycled = true;
1333                 }
1334                 if (fcnt > MAX_FREE_NIDS)
1335                         break;
1336                 if (is_cycled && nm_i->init_scan_nid <= nid)
1337                         break;
1338         }
1339
1340         /* go to the next nat page in order to reuse free nids first */
1341         nm_i->next_scan_nid = nm_i->init_scan_nid + NAT_ENTRY_PER_BLOCK;
1342
1343         /* find free nids from current sum_pages */
1344         mutex_lock(&curseg->curseg_mutex);
1345         for (i = 0; i < nats_in_cursum(sum); i++) {
1346                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1347                 nid = le32_to_cpu(nid_in_journal(sum, i));
1348                 if (addr == NULL_ADDR)
1349                         add_free_nid(nm_i, nid);
1350                 else
1351                         remove_free_nid(nm_i, nid);
1352         }
1353         mutex_unlock(&curseg->curseg_mutex);
1354
1355         /* remove the free nids from current allocated nids */
1356         list_for_each_entry_safe(fnid, next_fnid, &nm_i->free_nid_list, list) {
1357                 struct nat_entry *ne;
1358
1359                 read_lock(&nm_i->nat_tree_lock);
1360                 ne = __lookup_nat_cache(nm_i, fnid->nid);
1361                 if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1362                         remove_free_nid(nm_i, fnid->nid);
1363                 read_unlock(&nm_i->nat_tree_lock);
1364         }
1365 }
1366
1367 /*
1368  * If this function returns success, caller can obtain a new nid
1369  * from second parameter of this function.
1370  * The returned nid could be used ino as well as nid when inode is created.
1371  */
1372 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1373 {
1374         struct f2fs_nm_info *nm_i = NM_I(sbi);
1375         struct free_nid *i = NULL;
1376         struct list_head *this;
1377 retry:
1378         mutex_lock(&nm_i->build_lock);
1379         if (!nm_i->fcnt) {
1380                 /* scan NAT in order to build free nid list */
1381                 build_free_nids(sbi);
1382                 if (!nm_i->fcnt) {
1383                         mutex_unlock(&nm_i->build_lock);
1384                         return false;
1385                 }
1386         }
1387         mutex_unlock(&nm_i->build_lock);
1388
1389         /*
1390          * We check fcnt again since previous check is racy as
1391          * we didn't hold free_nid_list_lock. So other thread
1392          * could consume all of free nids.
1393          */
1394         spin_lock(&nm_i->free_nid_list_lock);
1395         if (!nm_i->fcnt) {
1396                 spin_unlock(&nm_i->free_nid_list_lock);
1397                 goto retry;
1398         }
1399
1400         BUG_ON(list_empty(&nm_i->free_nid_list));
1401         list_for_each(this, &nm_i->free_nid_list) {
1402                 i = list_entry(this, struct free_nid, list);
1403                 if (i->state == NID_NEW)
1404                         break;
1405         }
1406
1407         BUG_ON(i->state != NID_NEW);
1408         *nid = i->nid;
1409         i->state = NID_ALLOC;
1410         nm_i->fcnt--;
1411         spin_unlock(&nm_i->free_nid_list_lock);
1412         return true;
1413 }
1414
1415 /*
1416  * alloc_nid() should be called prior to this function.
1417  */
1418 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1419 {
1420         struct f2fs_nm_info *nm_i = NM_I(sbi);
1421         struct free_nid *i;
1422
1423         spin_lock(&nm_i->free_nid_list_lock);
1424         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1425         BUG_ON(!i || i->state != NID_ALLOC);
1426         __del_from_free_nid_list(i);
1427         spin_unlock(&nm_i->free_nid_list_lock);
1428 }
1429
1430 /*
1431  * alloc_nid() should be called prior to this function.
1432  */
1433 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1434 {
1435         struct f2fs_nm_info *nm_i = NM_I(sbi);
1436         struct free_nid *i;
1437
1438         spin_lock(&nm_i->free_nid_list_lock);
1439         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1440         BUG_ON(!i || i->state != NID_ALLOC);
1441         i->state = NID_NEW;
1442         nm_i->fcnt++;
1443         spin_unlock(&nm_i->free_nid_list_lock);
1444 }
1445
1446 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1447                 struct f2fs_summary *sum, struct node_info *ni,
1448                 block_t new_blkaddr)
1449 {
1450         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1451         set_node_addr(sbi, ni, new_blkaddr);
1452         clear_node_page_dirty(page);
1453 }
1454
1455 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1456 {
1457         struct address_space *mapping = sbi->node_inode->i_mapping;
1458         struct f2fs_node *src, *dst;
1459         nid_t ino = ino_of_node(page);
1460         struct node_info old_ni, new_ni;
1461         struct page *ipage;
1462
1463         ipage = grab_cache_page(mapping, ino);
1464         if (!ipage)
1465                 return -ENOMEM;
1466
1467         /* Should not use this inode  from free nid list */
1468         remove_free_nid(NM_I(sbi), ino);
1469
1470         get_node_info(sbi, ino, &old_ni);
1471         SetPageUptodate(ipage);
1472         fill_node_footer(ipage, ino, ino, 0, true);
1473
1474         src = (struct f2fs_node *)page_address(page);
1475         dst = (struct f2fs_node *)page_address(ipage);
1476
1477         memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i);
1478         dst->i.i_size = 0;
1479         dst->i.i_blocks = cpu_to_le64(1);
1480         dst->i.i_links = cpu_to_le32(1);
1481         dst->i.i_xattr_nid = 0;
1482
1483         new_ni = old_ni;
1484         new_ni.ino = ino;
1485
1486         set_node_addr(sbi, &new_ni, NEW_ADDR);
1487         inc_valid_inode_count(sbi);
1488
1489         f2fs_put_page(ipage, 1);
1490         return 0;
1491 }
1492
1493 int restore_node_summary(struct f2fs_sb_info *sbi,
1494                         unsigned int segno, struct f2fs_summary_block *sum)
1495 {
1496         struct f2fs_node *rn;
1497         struct f2fs_summary *sum_entry;
1498         struct page *page;
1499         block_t addr;
1500         int i, last_offset;
1501
1502         /* alloc temporal page for read node */
1503         page = alloc_page(GFP_NOFS | __GFP_ZERO);
1504         if (IS_ERR(page))
1505                 return PTR_ERR(page);
1506         lock_page(page);
1507
1508         /* scan the node segment */
1509         last_offset = sbi->blocks_per_seg;
1510         addr = START_BLOCK(sbi, segno);
1511         sum_entry = &sum->entries[0];
1512
1513         for (i = 0; i < last_offset; i++, sum_entry++) {
1514                 /*
1515                  * In order to read next node page,
1516                  * we must clear PageUptodate flag.
1517                  */
1518                 ClearPageUptodate(page);
1519
1520                 if (f2fs_readpage(sbi, page, addr, READ_SYNC))
1521                         goto out;
1522
1523                 lock_page(page);
1524                 rn = (struct f2fs_node *)page_address(page);
1525                 sum_entry->nid = rn->footer.nid;
1526                 sum_entry->version = 0;
1527                 sum_entry->ofs_in_node = 0;
1528                 addr++;
1529         }
1530         unlock_page(page);
1531 out:
1532         __free_pages(page, 0);
1533         return 0;
1534 }
1535
1536 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1537 {
1538         struct f2fs_nm_info *nm_i = NM_I(sbi);
1539         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1540         struct f2fs_summary_block *sum = curseg->sum_blk;
1541         int i;
1542
1543         mutex_lock(&curseg->curseg_mutex);
1544
1545         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1546                 mutex_unlock(&curseg->curseg_mutex);
1547                 return false;
1548         }
1549
1550         for (i = 0; i < nats_in_cursum(sum); i++) {
1551                 struct nat_entry *ne;
1552                 struct f2fs_nat_entry raw_ne;
1553                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1554
1555                 raw_ne = nat_in_journal(sum, i);
1556 retry:
1557                 write_lock(&nm_i->nat_tree_lock);
1558                 ne = __lookup_nat_cache(nm_i, nid);
1559                 if (ne) {
1560                         __set_nat_cache_dirty(nm_i, ne);
1561                         write_unlock(&nm_i->nat_tree_lock);
1562                         continue;
1563                 }
1564                 ne = grab_nat_entry(nm_i, nid);
1565                 if (!ne) {
1566                         write_unlock(&nm_i->nat_tree_lock);
1567                         goto retry;
1568                 }
1569                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1570                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1571                 nat_set_version(ne, raw_ne.version);
1572                 __set_nat_cache_dirty(nm_i, ne);
1573                 write_unlock(&nm_i->nat_tree_lock);
1574         }
1575         update_nats_in_cursum(sum, -i);
1576         mutex_unlock(&curseg->curseg_mutex);
1577         return true;
1578 }
1579
1580 /*
1581  * This function is called during the checkpointing process.
1582  */
1583 void flush_nat_entries(struct f2fs_sb_info *sbi)
1584 {
1585         struct f2fs_nm_info *nm_i = NM_I(sbi);
1586         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1587         struct f2fs_summary_block *sum = curseg->sum_blk;
1588         struct list_head *cur, *n;
1589         struct page *page = NULL;
1590         struct f2fs_nat_block *nat_blk = NULL;
1591         nid_t start_nid = 0, end_nid = 0;
1592         bool flushed;
1593
1594         flushed = flush_nats_in_journal(sbi);
1595
1596         if (!flushed)
1597                 mutex_lock(&curseg->curseg_mutex);
1598
1599         /* 1) flush dirty nat caches */
1600         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1601                 struct nat_entry *ne;
1602                 nid_t nid;
1603                 struct f2fs_nat_entry raw_ne;
1604                 int offset = -1;
1605                 block_t new_blkaddr;
1606
1607                 ne = list_entry(cur, struct nat_entry, list);
1608                 nid = nat_get_nid(ne);
1609
1610                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1611                         continue;
1612                 if (flushed)
1613                         goto to_nat_page;
1614
1615                 /* if there is room for nat enries in curseg->sumpage */
1616                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1617                 if (offset >= 0) {
1618                         raw_ne = nat_in_journal(sum, offset);
1619                         goto flush_now;
1620                 }
1621 to_nat_page:
1622                 if (!page || (start_nid > nid || nid > end_nid)) {
1623                         if (page) {
1624                                 f2fs_put_page(page, 1);
1625                                 page = NULL;
1626                         }
1627                         start_nid = START_NID(nid);
1628                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1629
1630                         /*
1631                          * get nat block with dirty flag, increased reference
1632                          * count, mapped and lock
1633                          */
1634                         page = get_next_nat_page(sbi, start_nid);
1635                         nat_blk = page_address(page);
1636                 }
1637
1638                 BUG_ON(!nat_blk);
1639                 raw_ne = nat_blk->entries[nid - start_nid];
1640 flush_now:
1641                 new_blkaddr = nat_get_blkaddr(ne);
1642
1643                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1644                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1645                 raw_ne.version = nat_get_version(ne);
1646
1647                 if (offset < 0) {
1648                         nat_blk->entries[nid - start_nid] = raw_ne;
1649                 } else {
1650                         nat_in_journal(sum, offset) = raw_ne;
1651                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1652                 }
1653
1654                 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1655                                         !add_free_nid(NM_I(sbi), nid)) {
1656                         write_lock(&nm_i->nat_tree_lock);
1657                         __del_from_nat_cache(nm_i, ne);
1658                         write_unlock(&nm_i->nat_tree_lock);
1659                 } else {
1660                         write_lock(&nm_i->nat_tree_lock);
1661                         __clear_nat_cache_dirty(nm_i, ne);
1662                         ne->checkpointed = true;
1663                         write_unlock(&nm_i->nat_tree_lock);
1664                 }
1665         }
1666         if (!flushed)
1667                 mutex_unlock(&curseg->curseg_mutex);
1668         f2fs_put_page(page, 1);
1669
1670         /* 2) shrink nat caches if necessary */
1671         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1672 }
1673
1674 static int init_node_manager(struct f2fs_sb_info *sbi)
1675 {
1676         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1677         struct f2fs_nm_info *nm_i = NM_I(sbi);
1678         unsigned char *version_bitmap;
1679         unsigned int nat_segs, nat_blocks;
1680
1681         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1682
1683         /* segment_count_nat includes pair segment so divide to 2. */
1684         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1685         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1686         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1687         nm_i->fcnt = 0;
1688         nm_i->nat_cnt = 0;
1689
1690         INIT_LIST_HEAD(&nm_i->free_nid_list);
1691         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1692         INIT_LIST_HEAD(&nm_i->nat_entries);
1693         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1694
1695         mutex_init(&nm_i->build_lock);
1696         spin_lock_init(&nm_i->free_nid_list_lock);
1697         rwlock_init(&nm_i->nat_tree_lock);
1698
1699         nm_i->init_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1700         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1701         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1702         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1703         if (!version_bitmap)
1704                 return -EFAULT;
1705
1706         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1707                                         GFP_KERNEL);
1708         if (!nm_i->nat_bitmap)
1709                 return -ENOMEM;
1710         return 0;
1711 }
1712
1713 int build_node_manager(struct f2fs_sb_info *sbi)
1714 {
1715         int err;
1716
1717         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1718         if (!sbi->nm_info)
1719                 return -ENOMEM;
1720
1721         err = init_node_manager(sbi);
1722         if (err)
1723                 return err;
1724
1725         build_free_nids(sbi);
1726         return 0;
1727 }
1728
1729 void destroy_node_manager(struct f2fs_sb_info *sbi)
1730 {
1731         struct f2fs_nm_info *nm_i = NM_I(sbi);
1732         struct free_nid *i, *next_i;
1733         struct nat_entry *natvec[NATVEC_SIZE];
1734         nid_t nid = 0;
1735         unsigned int found;
1736
1737         if (!nm_i)
1738                 return;
1739
1740         /* destroy free nid list */
1741         spin_lock(&nm_i->free_nid_list_lock);
1742         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1743                 BUG_ON(i->state == NID_ALLOC);
1744                 __del_from_free_nid_list(i);
1745                 nm_i->fcnt--;
1746         }
1747         BUG_ON(nm_i->fcnt);
1748         spin_unlock(&nm_i->free_nid_list_lock);
1749
1750         /* destroy nat cache */
1751         write_lock(&nm_i->nat_tree_lock);
1752         while ((found = __gang_lookup_nat_cache(nm_i,
1753                                         nid, NATVEC_SIZE, natvec))) {
1754                 unsigned idx;
1755                 for (idx = 0; idx < found; idx++) {
1756                         struct nat_entry *e = natvec[idx];
1757                         nid = nat_get_nid(e) + 1;
1758                         __del_from_nat_cache(nm_i, e);
1759                 }
1760         }
1761         BUG_ON(nm_i->nat_cnt);
1762         write_unlock(&nm_i->nat_tree_lock);
1763
1764         kfree(nm_i->nat_bitmap);
1765         sbi->nm_info = NULL;
1766         kfree(nm_i);
1767 }
1768
1769 int __init create_node_manager_caches(void)
1770 {
1771         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1772                         sizeof(struct nat_entry), NULL);
1773         if (!nat_entry_slab)
1774                 return -ENOMEM;
1775
1776         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1777                         sizeof(struct free_nid), NULL);
1778         if (!free_nid_slab) {
1779                 kmem_cache_destroy(nat_entry_slab);
1780                 return -ENOMEM;
1781         }
1782         return 0;
1783 }
1784
1785 void destroy_node_manager_caches(void)
1786 {
1787         kmem_cache_destroy(free_nid_slab);
1788         kmem_cache_destroy(nat_entry_slab);
1789 }