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 (PageUptodate(page)) {
 104                         f2fs_put_page(page, 1);
 105                         continue;
 106                 }
 107                 if (f2fs_readpage(sbi, page, index, READ))
 108                         continue;
 109
 110                 f2fs_put_page(page, 0);
 111         }
 112 }
 113
 114 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 115 {
 116         return radix_tree_lookup(&nm_i->nat_root, n);
 117 }
 118
 119 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 120                 nid_t start, unsigned int nr, struct nat_entry **ep)
 121 {
 122         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 123 }
 124
 125 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 126 {
 127         list_del(&e->list);
 128         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 129         nm_i->nat_cnt--;
 130         kmem_cache_free(nat_entry_slab, e);
 131 }
 132
 133 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 134 {
 135         struct f2fs_nm_info *nm_i = NM_I(sbi);
 136         struct nat_entry *e;
 137         int is_cp = 1;
 138
 139         read_lock(&nm_i->nat_tree_lock);
 140         e = __lookup_nat_cache(nm_i, nid);
 141         if (e && !e->checkpointed)
 142                 is_cp = 0;
 143         read_unlock(&nm_i->nat_tree_lock);
 144         return is_cp;
 145 }
 146
 147 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 148 {
 149         struct nat_entry *new;
 150
 151         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 152         if (!new)
 153                 return NULL;
 154         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 155                 kmem_cache_free(nat_entry_slab, new);
 156                 return NULL;
 157         }
 158         memset(new, 0, sizeof(struct nat_entry));
 159         nat_set_nid(new, nid);
 160         list_add_tail(&new->list, &nm_i->nat_entries);
 161         nm_i->nat_cnt++;
 162         return new;
 163 }
 164
 165 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 166                                                 struct f2fs_nat_entry *ne)
 167 {
 168         struct nat_entry *e;
 169 retry:
 170         write_lock(&nm_i->nat_tree_lock);
 171         e = __lookup_nat_cache(nm_i, nid);
 172         if (!e) {
 173                 e = grab_nat_entry(nm_i, nid);
 174                 if (!e) {
 175                         write_unlock(&nm_i->nat_tree_lock);
 176                         goto retry;
 177                 }
 178                 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
 179                 nat_set_ino(e, le32_to_cpu(ne->ino));
 180                 nat_set_version(e, ne->version);
 181                 e->checkpointed = true;
 182         }
 183         write_unlock(&nm_i->nat_tree_lock);
 184 }
 185
 186 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 187                         block_t new_blkaddr)
 188 {
 189         struct f2fs_nm_info *nm_i = NM_I(sbi);
 190         struct nat_entry *e;
 191 retry:
 192         write_lock(&nm_i->nat_tree_lock);
 193         e = __lookup_nat_cache(nm_i, ni->nid);
 194         if (!e) {
 195                 e = grab_nat_entry(nm_i, ni->nid);
 196                 if (!e) {
 197                         write_unlock(&nm_i->nat_tree_lock);
 198                         goto retry;
 199                 }
 200                 e->ni = *ni;
 201                 e->checkpointed = true;
 202                 BUG_ON(ni->blk_addr == NEW_ADDR);
 203         } else if (new_blkaddr == NEW_ADDR) {
 204                 /*
 205                  * when nid is reallocated,
 206                  * previous nat entry can be remained in nat cache.
 207                  * So, reinitialize it with new information.
 208                  */
 209                 e->ni = *ni;
 210                 BUG_ON(ni->blk_addr != NULL_ADDR);
 211         }
 212
 213         if (new_blkaddr == NEW_ADDR)
 214                 e->checkpointed = false;
 215
 216         /* sanity check */
 217         BUG_ON(nat_get_blkaddr(e) != ni->blk_addr);
 218         BUG_ON(nat_get_blkaddr(e) == NULL_ADDR &&
 219                         new_blkaddr == NULL_ADDR);
 220         BUG_ON(nat_get_blkaddr(e) == NEW_ADDR &&
 221                         new_blkaddr == NEW_ADDR);
 222         BUG_ON(nat_get_blkaddr(e) != NEW_ADDR &&
 223                         nat_get_blkaddr(e) != NULL_ADDR &&
 224                         new_blkaddr == NEW_ADDR);
 225
 226         /* increament version no as node is removed */
 227         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 228                 unsigned char version = nat_get_version(e);
 229                 nat_set_version(e, inc_node_version(version));
 230         }
 231
 232         /* change address */
 233         nat_set_blkaddr(e, new_blkaddr);
 234         __set_nat_cache_dirty(nm_i, e);
 235         write_unlock(&nm_i->nat_tree_lock);
 236 }
 237
 238 static int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 239 {
 240         struct f2fs_nm_info *nm_i = NM_I(sbi);
 241
 242         if (nm_i->nat_cnt < 2 * NM_WOUT_THRESHOLD)
 243                 return 0;
 244
 245         write_lock(&nm_i->nat_tree_lock);
 246         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 247                 struct nat_entry *ne;
 248                 ne = list_first_entry(&nm_i->nat_entries,
 249                                         struct nat_entry, list);
 250                 __del_from_nat_cache(nm_i, ne);
 251                 nr_shrink--;
 252         }
 253         write_unlock(&nm_i->nat_tree_lock);
 254         return nr_shrink;
 255 }
 256
 257 /*
 258  * This function returns always success
 259  */
 260 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 261 {
 262         struct f2fs_nm_info *nm_i = NM_I(sbi);
 263         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 264         struct f2fs_summary_block *sum = curseg->sum_blk;
 265         nid_t start_nid = START_NID(nid);
 266         struct f2fs_nat_block *nat_blk;
 267         struct page *page = NULL;
 268         struct f2fs_nat_entry ne;
 269         struct nat_entry *e;
 270         int i;
 271
 272         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 273         ni->nid = nid;
 274
 275         /* Check nat cache */
 276         read_lock(&nm_i->nat_tree_lock);
 277         e = __lookup_nat_cache(nm_i, nid);
 278         if (e) {
 279                 ni->ino = nat_get_ino(e);
 280                 ni->blk_addr = nat_get_blkaddr(e);
 281                 ni->version = nat_get_version(e);
 282         }
 283         read_unlock(&nm_i->nat_tree_lock);
 284         if (e)
 285                 return;
 286
 287         /* Check current segment summary */
 288         mutex_lock(&curseg->curseg_mutex);
 289         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 290         if (i >= 0) {
 291                 ne = nat_in_journal(sum, i);
 292                 node_info_from_raw_nat(ni, &ne);
 293         }
 294         mutex_unlock(&curseg->curseg_mutex);
 295         if (i >= 0)
 296                 goto cache;
 297
 298         /* Fill node_info from nat page */
 299         page = get_current_nat_page(sbi, start_nid);
 300         nat_blk = (struct f2fs_nat_block *)page_address(page);
 301         ne = nat_blk->entries[nid - start_nid];
 302         node_info_from_raw_nat(ni, &ne);
 303         f2fs_put_page(page, 1);
 304 cache:
 305         /* cache nat entry */
 306         cache_nat_entry(NM_I(sbi), nid, &ne);
 307 }
 308
 309 /*
 310  * The maximum depth is four.
 311  * Offset[0] will have raw inode offset.
 312  */
 313 static int get_node_path(long block, int offset[4], unsigned int noffset[4])
 314 {
 315         const long direct_index = ADDRS_PER_INODE;
 316         const long direct_blks = ADDRS_PER_BLOCK;
 317         const long dptrs_per_blk = NIDS_PER_BLOCK;
 318         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 319         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 320         int n = 0;
 321         int level = 0;
 322
 323         noffset[0] = 0;
 324
 325         if (block < direct_index) {
 326                 offset[n] = block;
 327                 goto got;
 328         }
 329         block -= direct_index;
 330         if (block < direct_blks) {
 331                 offset[n++] = NODE_DIR1_BLOCK;
 332                 noffset[n] = 1;
 333                 offset[n] = block;
 334                 level = 1;
 335                 goto got;
 336         }
 337         block -= direct_blks;
 338         if (block < direct_blks) {
 339                 offset[n++] = NODE_DIR2_BLOCK;
 340                 noffset[n] = 2;
 341                 offset[n] = block;
 342                 level = 1;
 343                 goto got;
 344         }
 345         block -= direct_blks;
 346         if (block < indirect_blks) {
 347                 offset[n++] = NODE_IND1_BLOCK;
 348                 noffset[n] = 3;
 349                 offset[n++] = block / direct_blks;
 350                 noffset[n] = 4 + offset[n - 1];
 351                 offset[n] = block % direct_blks;
 352                 level = 2;
 353                 goto got;
 354         }
 355         block -= indirect_blks;
 356         if (block < indirect_blks) {
 357                 offset[n++] = NODE_IND2_BLOCK;
 358                 noffset[n] = 4 + dptrs_per_blk;
 359                 offset[n++] = block / direct_blks;
 360                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 361                 offset[n] = block % direct_blks;
 362                 level = 2;
 363                 goto got;
 364         }
 365         block -= indirect_blks;
 366         if (block < dindirect_blks) {
 367                 offset[n++] = NODE_DIND_BLOCK;
 368                 noffset[n] = 5 + (dptrs_per_blk * 2);
 369                 offset[n++] = block / indirect_blks;
 370                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 371                               offset[n - 1] * (dptrs_per_blk + 1);
 372                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 373                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 374                               offset[n - 2] * (dptrs_per_blk + 1) +
 375                               offset[n - 1];
 376                 offset[n] = block % direct_blks;
 377                 level = 3;
 378                 goto got;
 379         } else {
 380                 BUG();
 381         }
 382 got:
 383         return level;
 384 }
 385
 386 /*
 387  * Caller should call f2fs_put_dnode(dn).
 388  */
 389 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 390 {
 391         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 392         struct page *npage[4];
 393         struct page *parent;
 394         int offset[4];
 395         unsigned int noffset[4];
 396         nid_t nids[4];
 397         int level, i;
 398         int err = 0;
 399
 400         level = get_node_path(index, offset, noffset);
 401
 402         nids[0] = dn->inode->i_ino;
 403         npage[0] = get_node_page(sbi, nids[0]);
 404         if (IS_ERR(npage[0]))
 405                 return PTR_ERR(npage[0]);
 406
 407         parent = npage[0];
 408         if (level != 0)
 409                 nids[1] = get_nid(parent, offset[0], true);
 410         dn->inode_page = npage[0];
 411         dn->inode_page_locked = true;
 412
 413         /* get indirect or direct nodes */
 414         for (i = 1; i <= level; i++) {
 415                 bool done = false;
 416
 417                 if (!nids[i] && mode == ALLOC_NODE) {
 418                         mutex_lock_op(sbi, NODE_NEW);
 419
 420                         /* alloc new node */
 421                         if (!alloc_nid(sbi, &(nids[i]))) {
 422                                 mutex_unlock_op(sbi, NODE_NEW);
 423                                 err = -ENOSPC;
 424                                 goto release_pages;
 425                         }
 426
 427                         dn->nid = nids[i];
 428                         npage[i] = new_node_page(dn, noffset[i]);
 429                         if (IS_ERR(npage[i])) {
 430                                 alloc_nid_failed(sbi, nids[i]);
 431                                 mutex_unlock_op(sbi, NODE_NEW);
 432                                 err = PTR_ERR(npage[i]);
 433                                 goto release_pages;
 434                         }
 435
 436                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 437                         alloc_nid_done(sbi, nids[i]);
 438                         mutex_unlock_op(sbi, NODE_NEW);
 439                         done = true;
 440                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 441                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 442                         if (IS_ERR(npage[i])) {
 443                                 err = PTR_ERR(npage[i]);
 444                                 goto release_pages;
 445                         }
 446                         done = true;
 447                 }
 448                 if (i == 1) {
 449                         dn->inode_page_locked = false;
 450                         unlock_page(parent);
 451                 } else {
 452                         f2fs_put_page(parent, 1);
 453                 }
 454
 455                 if (!done) {
 456                         npage[i] = get_node_page(sbi, nids[i]);
 457                         if (IS_ERR(npage[i])) {
 458                                 err = PTR_ERR(npage[i]);
 459                                 f2fs_put_page(npage[0], 0);
 460                                 goto release_out;
 461                         }
 462                 }
 463                 if (i < level) {
 464                         parent = npage[i];
 465                         nids[i + 1] = get_nid(parent, offset[i], false);
 466                 }
 467         }
 468         dn->nid = nids[level];
 469         dn->ofs_in_node = offset[level];
 470         dn->node_page = npage[level];
 471         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 472         return 0;
 473
 474 release_pages:
 475         f2fs_put_page(parent, 1);
 476         if (i > 1)
 477                 f2fs_put_page(npage[0], 0);
 478 release_out:
 479         dn->inode_page = NULL;
 480         dn->node_page = NULL;
 481         return err;
 482 }
 483
 484 static void truncate_node(struct dnode_of_data *dn)
 485 {
 486         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 487         struct node_info ni;
 488
 489         get_node_info(sbi, dn->nid, &ni);
 490         if (dn->inode->i_blocks == 0) {
 491                 BUG_ON(ni.blk_addr != NULL_ADDR);
 492                 goto invalidate;
 493         }
 494         BUG_ON(ni.blk_addr == NULL_ADDR);
 495
 496         /* Deallocate node address */
 497         invalidate_blocks(sbi, ni.blk_addr);
 498         dec_valid_node_count(sbi, dn->inode, 1);
 499         set_node_addr(sbi, &ni, NULL_ADDR);
 500
 501         if (dn->nid == dn->inode->i_ino) {
 502                 remove_orphan_inode(sbi, dn->nid);
 503                 dec_valid_inode_count(sbi);
 504         } else {
 505                 sync_inode_page(dn);
 506         }
 507 invalidate:
 508         clear_node_page_dirty(dn->node_page);
 509         F2FS_SET_SB_DIRT(sbi);
 510
 511         f2fs_put_page(dn->node_page, 1);
 512         dn->node_page = NULL;
 513 }
 514
 515 static int truncate_dnode(struct dnode_of_data *dn)
 516 {
 517         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 518         struct page *page;
 519
 520         if (dn->nid == 0)
 521                 return 1;
 522
 523         /* get direct node */
 524         page = get_node_page(sbi, dn->nid);
 525         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 526                 return 1;
 527         else if (IS_ERR(page))
 528                 return PTR_ERR(page);
 529
 530         /* Make dnode_of_data for parameter */
 531         dn->node_page = page;
 532         dn->ofs_in_node = 0;
 533         truncate_data_blocks(dn);
 534         truncate_node(dn);
 535         return 1;
 536 }
 537
 538 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 539                                                 int ofs, int depth)
 540 {
 541         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 542         struct dnode_of_data rdn = *dn;
 543         struct page *page;
 544         struct f2fs_node *rn;
 545         nid_t child_nid;
 546         unsigned int child_nofs;
 547         int freed = 0;
 548         int i, ret;
 549
 550         if (dn->nid == 0)
 551                 return NIDS_PER_BLOCK + 1;
 552
 553         page = get_node_page(sbi, dn->nid);
 554         if (IS_ERR(page))
 555                 return PTR_ERR(page);
 556
 557         rn = (struct f2fs_node *)page_address(page);
 558         if (depth < 3) {
 559                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 560                         child_nid = le32_to_cpu(rn->in.nid[i]);
 561                         if (child_nid == 0)
 562                                 continue;
 563                         rdn.nid = child_nid;
 564                         ret = truncate_dnode(&rdn);
 565                         if (ret < 0)
 566                                 goto out_err;
 567                         set_nid(page, i, 0, false);
 568                 }
 569         } else {
 570                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 571                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 572                         child_nid = le32_to_cpu(rn->in.nid[i]);
 573                         if (child_nid == 0) {
 574                                 child_nofs += NIDS_PER_BLOCK + 1;
 575                                 continue;
 576                         }
 577                         rdn.nid = child_nid;
 578                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 579                         if (ret == (NIDS_PER_BLOCK + 1)) {
 580                                 set_nid(page, i, 0, false);
 581                                 child_nofs += ret;
 582                         } else if (ret < 0 && ret != -ENOENT) {
 583                                 goto out_err;
 584                         }
 585                 }
 586                 freed = child_nofs;
 587         }
 588
 589         if (!ofs) {
 590                 /* remove current indirect node */
 591                 dn->node_page = page;
 592                 truncate_node(dn);
 593                 freed++;
 594         } else {
 595                 f2fs_put_page(page, 1);
 596         }
 597         return freed;
 598
 599 out_err:
 600         f2fs_put_page(page, 1);
 601         return ret;
 602 }
 603
 604 static int truncate_partial_nodes(struct dnode_of_data *dn,
 605                         struct f2fs_inode *ri, int *offset, int depth)
 606 {
 607         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 608         struct page *pages[2];
 609         nid_t nid[3];
 610         nid_t child_nid;
 611         int err = 0;
 612         int i;
 613         int idx = depth - 2;
 614
 615         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 616         if (!nid[0])
 617                 return 0;
 618
 619         /* get indirect nodes in the path */
 620         for (i = 0; i < depth - 1; i++) {
 621                 /* refernece count'll be increased */
 622                 pages[i] = get_node_page(sbi, nid[i]);
 623                 if (IS_ERR(pages[i])) {
 624                         depth = i + 1;
 625                         err = PTR_ERR(pages[i]);
 626                         goto fail;
 627                 }
 628                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 629         }
 630
 631         /* free direct nodes linked to a partial indirect node */
 632         for (i = offset[depth - 1]; i < NIDS_PER_BLOCK; i++) {
 633                 child_nid = get_nid(pages[idx], i, false);
 634                 if (!child_nid)
 635                         continue;
 636                 dn->nid = child_nid;
 637                 err = truncate_dnode(dn);
 638                 if (err < 0)
 639                         goto fail;
 640                 set_nid(pages[idx], i, 0, false);
 641         }
 642
 643         if (offset[depth - 1] == 0) {
 644                 dn->node_page = pages[idx];
 645                 dn->nid = nid[idx];
 646                 truncate_node(dn);
 647         } else {
 648                 f2fs_put_page(pages[idx], 1);
 649         }
 650         offset[idx]++;
 651         offset[depth - 1] = 0;
 652 fail:
 653         for (i = depth - 3; i >= 0; i--)
 654                 f2fs_put_page(pages[i], 1);
 655         return err;
 656 }
 657
 658 /*
 659  * All the block addresses of data and nodes should be nullified.
 660  */
 661 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 662 {
 663         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 664         int err = 0, cont = 1;
 665         int level, offset[4], noffset[4];
 666         unsigned int nofs = 0;
 667         struct f2fs_node *rn;
 668         struct dnode_of_data dn;
 669         struct page *page;
 670
 671         level = get_node_path(from, offset, noffset);
 672
 673         page = get_node_page(sbi, inode->i_ino);
 674         if (IS_ERR(page))
 675                 return PTR_ERR(page);
 676
 677         set_new_dnode(&dn, inode, page, NULL, 0);
 678         unlock_page(page);
 679
 680         rn = page_address(page);
 681         switch (level) {
 682         case 0:
 683         case 1:
 684                 nofs = noffset[1];
 685                 break;
 686         case 2:
 687                 nofs = noffset[1];
 688                 if (!offset[level - 1])
 689                         goto skip_partial;
 690                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 691                 if (err < 0 && err != -ENOENT)
 692                         goto fail;
 693                 nofs += 1 + NIDS_PER_BLOCK;
 694                 break;
 695         case 3:
 696                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 697                 if (!offset[level - 1])
 698                         goto skip_partial;
 699                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 700                 if (err < 0 && err != -ENOENT)
 701                         goto fail;
 702                 break;
 703         default:
 704                 BUG();
 705         }
 706
 707 skip_partial:
 708         while (cont) {
 709                 dn.nid = le32_to_cpu(rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]);
 710                 switch (offset[0]) {
 711                 case NODE_DIR1_BLOCK:
 712                 case NODE_DIR2_BLOCK:
 713                         err = truncate_dnode(&dn);
 714                         break;
 715
 716                 case NODE_IND1_BLOCK:
 717                 case NODE_IND2_BLOCK:
 718                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 719                         break;
 720
 721                 case NODE_DIND_BLOCK:
 722                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 723                         cont = 0;
 724                         break;
 725
 726                 default:
 727                         BUG();
 728                 }
 729                 if (err < 0 && err != -ENOENT)
 730                         goto fail;
 731                 if (offset[1] == 0 &&
 732                                 rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 733                         lock_page(page);
 734                         wait_on_page_writeback(page);
 735                         rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 736                         set_page_dirty(page);
 737                         unlock_page(page);
 738                 }
 739                 offset[1] = 0;
 740                 offset[0]++;
 741                 nofs += err;
 742         }
 743 fail:
 744         f2fs_put_page(page, 0);
 745         return err > 0 ? 0 : err;
 746 }
 747
 748 int remove_inode_page(struct inode *inode)
 749 {
 750         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 751         struct page *page;
 752         nid_t ino = inode->i_ino;
 753         struct dnode_of_data dn;
 754
 755         mutex_lock_op(sbi, NODE_TRUNC);
 756         page = get_node_page(sbi, ino);
 757         if (IS_ERR(page)) {
 758                 mutex_unlock_op(sbi, NODE_TRUNC);
 759                 return PTR_ERR(page);
 760         }
 761
 762         if (F2FS_I(inode)->i_xattr_nid) {
 763                 nid_t nid = F2FS_I(inode)->i_xattr_nid;
 764                 struct page *npage = get_node_page(sbi, nid);
 765
 766                 if (IS_ERR(npage)) {
 767                         mutex_unlock_op(sbi, NODE_TRUNC);
 768                         return PTR_ERR(npage);
 769                 }
 770
 771                 F2FS_I(inode)->i_xattr_nid = 0;
 772                 set_new_dnode(&dn, inode, page, npage, nid);
 773                 dn.inode_page_locked = 1;
 774                 truncate_node(&dn);
 775         }
 776
 777         /* 0 is possible, after f2fs_new_inode() is failed */
 778         BUG_ON(inode->i_blocks != 0 && inode->i_blocks != 1);
 779         set_new_dnode(&dn, inode, page, page, ino);
 780         truncate_node(&dn);
 781
 782         mutex_unlock_op(sbi, NODE_TRUNC);
 783         return 0;
 784 }
 785
 786 int new_inode_page(struct inode *inode, const struct qstr *name)
 787 {
 788         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 789         struct page *page;
 790         struct dnode_of_data dn;
 791
 792         /* allocate inode page for new inode */
 793         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 794         mutex_lock_op(sbi, NODE_NEW);
 795         page = new_node_page(&dn, 0);
 796         init_dent_inode(name, page);
 797         mutex_unlock_op(sbi, NODE_NEW);
 798         if (IS_ERR(page))
 799                 return PTR_ERR(page);
 800         f2fs_put_page(page, 1);
 801         return 0;
 802 }
 803
 804 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
 805 {
 806         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 807         struct address_space *mapping = sbi->node_inode->i_mapping;
 808         struct node_info old_ni, new_ni;
 809         struct page *page;
 810         int err;
 811
 812         if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
 813                 return ERR_PTR(-EPERM);
 814
 815         page = grab_cache_page(mapping, dn->nid);
 816         if (!page)
 817                 return ERR_PTR(-ENOMEM);
 818
 819         get_node_info(sbi, dn->nid, &old_ni);
 820
 821         SetPageUptodate(page);
 822         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 823
 824         /* Reinitialize old_ni with new node page */
 825         BUG_ON(old_ni.blk_addr != NULL_ADDR);
 826         new_ni = old_ni;
 827         new_ni.ino = dn->inode->i_ino;
 828
 829         if (!inc_valid_node_count(sbi, dn->inode, 1)) {
 830                 err = -ENOSPC;
 831                 goto fail;
 832         }
 833         set_node_addr(sbi, &new_ni, NEW_ADDR);
 834         set_cold_node(dn->inode, page);
 835
 836         dn->node_page = page;
 837         sync_inode_page(dn);
 838         set_page_dirty(page);
 839         if (ofs == 0)
 840                 inc_valid_inode_count(sbi);
 841
 842         return page;
 843
 844 fail:
 845         clear_node_page_dirty(page);
 846         f2fs_put_page(page, 1);
 847         return ERR_PTR(err);
 848 }
 849
 850 /*
 851  * Caller should do after getting the following values.
 852  * 0: f2fs_put_page(page, 0)
 853  * LOCKED_PAGE: f2fs_put_page(page, 1)
 854  * error: nothing
 855  */
 856 static int read_node_page(struct page *page, int type)
 857 {
 858         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 859         struct node_info ni;
 860
 861         get_node_info(sbi, page->index, &ni);
 862
 863         if (ni.blk_addr == NULL_ADDR) {
 864                 f2fs_put_page(page, 1);
 865                 return -ENOENT;
 866         }
 867
 868         if (PageUptodate(page))
 869                 return LOCKED_PAGE;
 870
 871         return f2fs_readpage(sbi, page, ni.blk_addr, type);
 872 }
 873
 874 /*
 875  * Readahead a node page
 876  */
 877 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 878 {
 879         struct address_space *mapping = sbi->node_inode->i_mapping;
 880         struct page *apage;
 881         int err;
 882
 883         apage = find_get_page(mapping, nid);
 884         if (apage && PageUptodate(apage)) {
 885                 f2fs_put_page(apage, 0);
 886                 return;
 887         }
 888         f2fs_put_page(apage, 0);
 889
 890         apage = grab_cache_page(mapping, nid);
 891         if (!apage)
 892                 return;
 893
 894         err = read_node_page(apage, READA);
 895         if (err == 0)
 896                 f2fs_put_page(apage, 0);
 897         else if (err == LOCKED_PAGE)
 898                 f2fs_put_page(apage, 1);
 899         return;
 900 }
 901
 902 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 903 {
 904         struct address_space *mapping = sbi->node_inode->i_mapping;
 905         struct page *page;
 906         int err;
 907
 908         page = grab_cache_page(mapping, nid);
 909         if (!page)
 910                 return ERR_PTR(-ENOMEM);
 911
 912         err = read_node_page(page, READ_SYNC);
 913         if (err < 0)
 914                 return ERR_PTR(err);
 915         else if (err == LOCKED_PAGE)
 916                 goto got_it;
 917
 918         lock_page(page);
 919         if (!PageUptodate(page)) {
 920                 f2fs_put_page(page, 1);
 921                 return ERR_PTR(-EIO);
 922         }
 923 got_it:
 924         BUG_ON(nid != nid_of_node(page));
 925         mark_page_accessed(page);
 926         return page;
 927 }
 928
 929 /*
 930  * Return a locked page for the desired node page.
 931  * And, readahead MAX_RA_NODE number of node pages.
 932  */
 933 struct page *get_node_page_ra(struct page *parent, int start)
 934 {
 935         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 936         struct address_space *mapping = sbi->node_inode->i_mapping;
 937         struct page *page;
 938         int err, i, end;
 939         nid_t nid;
 940
 941         /* First, try getting the desired direct node. */
 942         nid = get_nid(parent, start, false);
 943         if (!nid)
 944                 return ERR_PTR(-ENOENT);
 945
 946         page = grab_cache_page(mapping, nid);
 947         if (!page)
 948                 return ERR_PTR(-ENOMEM);
 949
 950         err = read_node_page(page, READ_SYNC);
 951         if (err < 0)
 952                 return ERR_PTR(err);
 953         else if (err == LOCKED_PAGE)
 954                 goto page_hit;
 955
 956         /* Then, try readahead for siblings of the desired node */
 957         end = start + MAX_RA_NODE;
 958         end = min(end, NIDS_PER_BLOCK);
 959         for (i = start + 1; i < end; i++) {
 960                 nid = get_nid(parent, i, false);
 961                 if (!nid)
 962                         continue;
 963                 ra_node_page(sbi, nid);
 964         }
 965
 966         lock_page(page);
 967
 968 page_hit:
 969         if (!PageUptodate(page)) {
 970                 f2fs_put_page(page, 1);
 971                 return ERR_PTR(-EIO);
 972         }
 973         mark_page_accessed(page);
 974         return page;
 975 }
 976
 977 void sync_inode_page(struct dnode_of_data *dn)
 978 {
 979         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
 980                 update_inode(dn->inode, dn->node_page);
 981         } else if (dn->inode_page) {
 982                 if (!dn->inode_page_locked)
 983                         lock_page(dn->inode_page);
 984                 update_inode(dn->inode, dn->inode_page);
 985                 if (!dn->inode_page_locked)
 986                         unlock_page(dn->inode_page);
 987         } else {
 988                 f2fs_write_inode(dn->inode, NULL);
 989         }
 990 }
 991
 992 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
 993                                         struct writeback_control *wbc)
 994 {
 995         struct address_space *mapping = sbi->node_inode->i_mapping;
 996         pgoff_t index, end;
 997         struct pagevec pvec;
 998         int step = ino ? 2 : 0;
 999         int nwritten = 0, wrote = 0;
1000
1001         pagevec_init(&pvec, 0);
1002
1003 next_step:
1004         index = 0;
1005         end = LONG_MAX;
1006
1007         while (index <= end) {
1008                 int i, nr_pages;
1009                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1010                                 PAGECACHE_TAG_DIRTY,
1011                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1012                 if (nr_pages == 0)
1013                         break;
1014
1015                 for (i = 0; i < nr_pages; i++) {
1016                         struct page *page = pvec.pages[i];
1017
1018                         /*
1019                          * flushing sequence with step:
1020                          * 0. indirect nodes
1021                          * 1. dentry dnodes
1022                          * 2. file dnodes
1023                          */
1024                         if (step == 0 && IS_DNODE(page))
1025                                 continue;
1026                         if (step == 1 && (!IS_DNODE(page) ||
1027                                                 is_cold_node(page)))
1028                                 continue;
1029                         if (step == 2 && (!IS_DNODE(page) ||
1030                                                 !is_cold_node(page)))
1031                                 continue;
1032
1033                         /*
1034                          * If an fsync mode,
1035                          * we should not skip writing node pages.
1036                          */
1037                         if (ino && ino_of_node(page) == ino)
1038                                 lock_page(page);
1039                         else if (!trylock_page(page))
1040                                 continue;
1041
1042                         if (unlikely(page->mapping != mapping)) {
1043 continue_unlock:
1044                                 unlock_page(page);
1045                                 continue;
1046                         }
1047                         if (ino && ino_of_node(page) != ino)
1048                                 goto continue_unlock;
1049
1050                         if (!PageDirty(page)) {
1051                                 /* someone wrote it for us */
1052                                 goto continue_unlock;
1053                         }
1054
1055                         if (!clear_page_dirty_for_io(page))
1056                                 goto continue_unlock;
1057
1058                         /* called by fsync() */
1059                         if (ino && IS_DNODE(page)) {
1060                                 int mark = !is_checkpointed_node(sbi, ino);
1061                                 set_fsync_mark(page, 1);
1062                                 if (IS_INODE(page))
1063                                         set_dentry_mark(page, mark);
1064                                 nwritten++;
1065                         } else {
1066                                 set_fsync_mark(page, 0);
1067                                 set_dentry_mark(page, 0);
1068                         }
1069                         mapping->a_ops->writepage(page, wbc);
1070                         wrote++;
1071
1072                         if (--wbc->nr_to_write == 0)
1073                                 break;
1074                 }
1075                 pagevec_release(&pvec);
1076                 cond_resched();
1077
1078                 if (wbc->nr_to_write == 0) {
1079                         step = 2;
1080                         break;
1081                 }
1082         }
1083
1084         if (step < 2) {
1085                 step++;
1086                 goto next_step;
1087         }
1088
1089         if (wrote)
1090                 f2fs_submit_bio(sbi, NODE, wbc->sync_mode == WB_SYNC_ALL);
1091
1092         return nwritten;
1093 }
1094
1095 static int f2fs_write_node_page(struct page *page,
1096                                 struct writeback_control *wbc)
1097 {
1098         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1099         nid_t nid;
1100         block_t new_addr;
1101         struct node_info ni;
1102
1103         wait_on_page_writeback(page);
1104
1105         mutex_lock_op(sbi, NODE_WRITE);
1106
1107         /* get old block addr of this node page */
1108         nid = nid_of_node(page);
1109         BUG_ON(page->index != nid);
1110
1111         get_node_info(sbi, nid, &ni);
1112
1113         /* This page is already truncated */
1114         if (ni.blk_addr == NULL_ADDR)
1115                 goto out;
1116
1117         if (wbc->for_reclaim) {
1118                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1119                 wbc->pages_skipped++;
1120                 set_page_dirty(page);
1121                 mutex_unlock_op(sbi, NODE_WRITE);
1122                 return AOP_WRITEPAGE_ACTIVATE;
1123         }
1124
1125         set_page_writeback(page);
1126
1127         /* insert node offset */
1128         write_node_page(sbi, page, nid, ni.blk_addr, &new_addr);
1129         set_node_addr(sbi, &ni, new_addr);
1130 out:
1131         dec_page_count(sbi, F2FS_DIRTY_NODES);
1132         mutex_unlock_op(sbi, NODE_WRITE);
1133         unlock_page(page);
1134         return 0;
1135 }
1136
1137 /*
1138  * It is very important to gather dirty pages and write at once, so that we can
1139  * submit a big bio without interfering other data writes.
1140  * Be default, 512 pages (2MB), a segment size, is quite reasonable.
1141  */
1142 #define COLLECT_DIRTY_NODES     512
1143 static int f2fs_write_node_pages(struct address_space *mapping,
1144                             struct writeback_control *wbc)
1145 {
1146         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1147         struct block_device *bdev = sbi->sb->s_bdev;
1148         long nr_to_write = wbc->nr_to_write;
1149
1150         /* First check balancing cached NAT entries */
1151         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK)) {
1152                 f2fs_sync_fs(sbi->sb, true);
1153                 return 0;
1154         }
1155
1156         /* collect a number of dirty node pages and write together */
1157         if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1158                 return 0;
1159
1160         /* if mounting is failed, skip writing node pages */
1161         wbc->nr_to_write = bio_get_nr_vecs(bdev);
1162         sync_node_pages(sbi, 0, wbc);
1163         wbc->nr_to_write = nr_to_write -
1164                 (bio_get_nr_vecs(bdev) - wbc->nr_to_write);
1165         return 0;
1166 }
1167
1168 static int f2fs_set_node_page_dirty(struct page *page)
1169 {
1170         struct address_space *mapping = page->mapping;
1171         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1172
1173         SetPageUptodate(page);
1174         if (!PageDirty(page)) {
1175                 __set_page_dirty_nobuffers(page);
1176                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1177                 SetPagePrivate(page);
1178                 return 1;
1179         }
1180         return 0;
1181 }
1182
1183 static void f2fs_invalidate_node_page(struct page *page, unsigned long offset)
1184 {
1185         struct inode *inode = page->mapping->host;
1186         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1187         if (PageDirty(page))
1188                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1189         ClearPagePrivate(page);
1190 }
1191
1192 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1193 {
1194         ClearPagePrivate(page);
1195         return 1;
1196 }
1197
1198 /*
1199  * Structure of the f2fs node operations
1200  */
1201 const struct address_space_operations f2fs_node_aops = {
1202         .writepage      = f2fs_write_node_page,
1203         .writepages     = f2fs_write_node_pages,
1204         .set_page_dirty = f2fs_set_node_page_dirty,
1205         .invalidatepage = f2fs_invalidate_node_page,
1206         .releasepage    = f2fs_release_node_page,
1207 };
1208
1209 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1210 {
1211         struct list_head *this;
1212         struct free_nid *i;
1213         list_for_each(this, head) {
1214                 i = list_entry(this, struct free_nid, list);
1215                 if (i->nid == n)
1216                         return i;
1217         }
1218         return NULL;
1219 }
1220
1221 static void __del_from_free_nid_list(struct free_nid *i)
1222 {
1223         list_del(&i->list);
1224         kmem_cache_free(free_nid_slab, i);
1225 }
1226
1227 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1228 {
1229         struct free_nid *i;
1230
1231         if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1232                 return 0;
1233 retry:
1234         i = kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1235         if (!i) {
1236                 cond_resched();
1237                 goto retry;
1238         }
1239         i->nid = nid;
1240         i->state = NID_NEW;
1241
1242         spin_lock(&nm_i->free_nid_list_lock);
1243         if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1244                 spin_unlock(&nm_i->free_nid_list_lock);
1245                 kmem_cache_free(free_nid_slab, i);
1246                 return 0;
1247         }
1248         list_add_tail(&i->list, &nm_i->free_nid_list);
1249         nm_i->fcnt++;
1250         spin_unlock(&nm_i->free_nid_list_lock);
1251         return 1;
1252 }
1253
1254 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1255 {
1256         struct free_nid *i;
1257         spin_lock(&nm_i->free_nid_list_lock);
1258         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1259         if (i && i->state == NID_NEW) {
1260                 __del_from_free_nid_list(i);
1261                 nm_i->fcnt--;
1262         }
1263         spin_unlock(&nm_i->free_nid_list_lock);
1264 }
1265
1266 static int scan_nat_page(struct f2fs_nm_info *nm_i,
1267                         struct page *nat_page, nid_t start_nid)
1268 {
1269         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1270         block_t blk_addr;
1271         int fcnt = 0;
1272         int i;
1273
1274         /* 0 nid should not be used */
1275         if (start_nid == 0)
1276                 ++start_nid;
1277
1278         i = start_nid % NAT_ENTRY_PER_BLOCK;
1279
1280         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1281                 if (start_nid >= nm_i->max_nid)
1282                         break;
1283                 blk_addr  = le32_to_cpu(nat_blk->entries[i].block_addr);
1284                 BUG_ON(blk_addr == NEW_ADDR);
1285                 if (blk_addr == NULL_ADDR)
1286                         fcnt += add_free_nid(nm_i, start_nid);
1287         }
1288         return fcnt;
1289 }
1290
1291 static void build_free_nids(struct f2fs_sb_info *sbi)
1292 {
1293         struct free_nid *fnid, *next_fnid;
1294         struct f2fs_nm_info *nm_i = NM_I(sbi);
1295         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1296         struct f2fs_summary_block *sum = curseg->sum_blk;
1297         nid_t nid = 0;
1298         bool is_cycled = false;
1299         int fcnt = 0;
1300         int i;
1301
1302         nid = nm_i->next_scan_nid;
1303         nm_i->init_scan_nid = nid;
1304
1305         ra_nat_pages(sbi, nid);
1306
1307         while (1) {
1308                 struct page *page = get_current_nat_page(sbi, nid);
1309
1310                 fcnt += scan_nat_page(nm_i, page, nid);
1311                 f2fs_put_page(page, 1);
1312
1313                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1314
1315                 if (nid >= nm_i->max_nid) {
1316                         nid = 0;
1317                         is_cycled = true;
1318                 }
1319                 if (fcnt > MAX_FREE_NIDS)
1320                         break;
1321                 if (is_cycled && nm_i->init_scan_nid <= nid)
1322                         break;
1323         }
1324
1325         /* go to the next nat page in order to reuse free nids first */
1326         nm_i->next_scan_nid = nm_i->init_scan_nid + NAT_ENTRY_PER_BLOCK;
1327
1328         /* find free nids from current sum_pages */
1329         mutex_lock(&curseg->curseg_mutex);
1330         for (i = 0; i < nats_in_cursum(sum); i++) {
1331                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1332                 nid = le32_to_cpu(nid_in_journal(sum, i));
1333                 if (addr == NULL_ADDR)
1334                         add_free_nid(nm_i, nid);
1335                 else
1336                         remove_free_nid(nm_i, nid);
1337         }
1338         mutex_unlock(&curseg->curseg_mutex);
1339
1340         /* remove the free nids from current allocated nids */
1341         list_for_each_entry_safe(fnid, next_fnid, &nm_i->free_nid_list, list) {
1342                 struct nat_entry *ne;
1343
1344                 read_lock(&nm_i->nat_tree_lock);
1345                 ne = __lookup_nat_cache(nm_i, fnid->nid);
1346                 if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1347                         remove_free_nid(nm_i, fnid->nid);
1348                 read_unlock(&nm_i->nat_tree_lock);
1349         }
1350 }
1351
1352 /*
1353  * If this function returns success, caller can obtain a new nid
1354  * from second parameter of this function.
1355  * The returned nid could be used ino as well as nid when inode is created.
1356  */
1357 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1358 {
1359         struct f2fs_nm_info *nm_i = NM_I(sbi);
1360         struct free_nid *i = NULL;
1361         struct list_head *this;
1362 retry:
1363         mutex_lock(&nm_i->build_lock);
1364         if (!nm_i->fcnt) {
1365                 /* scan NAT in order to build free nid list */
1366                 build_free_nids(sbi);
1367                 if (!nm_i->fcnt) {
1368                         mutex_unlock(&nm_i->build_lock);
1369                         return false;
1370                 }
1371         }
1372         mutex_unlock(&nm_i->build_lock);
1373
1374         /*
1375          * We check fcnt again since previous check is racy as
1376          * we didn't hold free_nid_list_lock. So other thread
1377          * could consume all of free nids.
1378          */
1379         spin_lock(&nm_i->free_nid_list_lock);
1380         if (!nm_i->fcnt) {
1381                 spin_unlock(&nm_i->free_nid_list_lock);
1382                 goto retry;
1383         }
1384
1385         BUG_ON(list_empty(&nm_i->free_nid_list));
1386         list_for_each(this, &nm_i->free_nid_list) {
1387                 i = list_entry(this, struct free_nid, list);
1388                 if (i->state == NID_NEW)
1389                         break;
1390         }
1391
1392         BUG_ON(i->state != NID_NEW);
1393         *nid = i->nid;
1394         i->state = NID_ALLOC;
1395         nm_i->fcnt--;
1396         spin_unlock(&nm_i->free_nid_list_lock);
1397         return true;
1398 }
1399
1400 /*
1401  * alloc_nid() should be called prior to this function.
1402  */
1403 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1404 {
1405         struct f2fs_nm_info *nm_i = NM_I(sbi);
1406         struct free_nid *i;
1407
1408         spin_lock(&nm_i->free_nid_list_lock);
1409         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1410         if (i) {
1411                 BUG_ON(i->state != NID_ALLOC);
1412                 __del_from_free_nid_list(i);
1413         }
1414         spin_unlock(&nm_i->free_nid_list_lock);
1415 }
1416
1417 /*
1418  * alloc_nid() should be called prior to this function.
1419  */
1420 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1421 {
1422         alloc_nid_done(sbi, nid);
1423         add_free_nid(NM_I(sbi), nid);
1424 }
1425
1426 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1427                 struct f2fs_summary *sum, struct node_info *ni,
1428                 block_t new_blkaddr)
1429 {
1430         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1431         set_node_addr(sbi, ni, new_blkaddr);
1432         clear_node_page_dirty(page);
1433 }
1434
1435 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1436 {
1437         struct address_space *mapping = sbi->node_inode->i_mapping;
1438         struct f2fs_node *src, *dst;
1439         nid_t ino = ino_of_node(page);
1440         struct node_info old_ni, new_ni;
1441         struct page *ipage;
1442
1443         ipage = grab_cache_page(mapping, ino);
1444         if (!ipage)
1445                 return -ENOMEM;
1446
1447         /* Should not use this inode  from free nid list */
1448         remove_free_nid(NM_I(sbi), ino);
1449
1450         get_node_info(sbi, ino, &old_ni);
1451         SetPageUptodate(ipage);
1452         fill_node_footer(ipage, ino, ino, 0, true);
1453
1454         src = (struct f2fs_node *)page_address(page);
1455         dst = (struct f2fs_node *)page_address(ipage);
1456
1457         memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i);
1458         dst->i.i_size = 0;
1459         dst->i.i_blocks = cpu_to_le64(1);
1460         dst->i.i_links = cpu_to_le32(1);
1461         dst->i.i_xattr_nid = 0;
1462
1463         new_ni = old_ni;
1464         new_ni.ino = ino;
1465
1466         set_node_addr(sbi, &new_ni, NEW_ADDR);
1467         inc_valid_inode_count(sbi);
1468
1469         f2fs_put_page(ipage, 1);
1470         return 0;
1471 }
1472
1473 int restore_node_summary(struct f2fs_sb_info *sbi,
1474                         unsigned int segno, struct f2fs_summary_block *sum)
1475 {
1476         struct f2fs_node *rn;
1477         struct f2fs_summary *sum_entry;
1478         struct page *page;
1479         block_t addr;
1480         int i, last_offset;
1481
1482         /* alloc temporal page for read node */
1483         page = alloc_page(GFP_NOFS | __GFP_ZERO);
1484         if (IS_ERR(page))
1485                 return PTR_ERR(page);
1486         lock_page(page);
1487
1488         /* scan the node segment */
1489         last_offset = sbi->blocks_per_seg;
1490         addr = START_BLOCK(sbi, segno);
1491         sum_entry = &sum->entries[0];
1492
1493         for (i = 0; i < last_offset; i++, sum_entry++) {
1494                 /*
1495                  * In order to read next node page,
1496                  * we must clear PageUptodate flag.
1497                  */
1498                 ClearPageUptodate(page);
1499
1500                 if (f2fs_readpage(sbi, page, addr, READ_SYNC))
1501                         goto out;
1502
1503                 lock_page(page);
1504                 rn = (struct f2fs_node *)page_address(page);
1505                 sum_entry->nid = rn->footer.nid;
1506                 sum_entry->version = 0;
1507                 sum_entry->ofs_in_node = 0;
1508                 addr++;
1509         }
1510         unlock_page(page);
1511 out:
1512         __free_pages(page, 0);
1513         return 0;
1514 }
1515
1516 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1517 {
1518         struct f2fs_nm_info *nm_i = NM_I(sbi);
1519         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1520         struct f2fs_summary_block *sum = curseg->sum_blk;
1521         int i;
1522
1523         mutex_lock(&curseg->curseg_mutex);
1524
1525         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1526                 mutex_unlock(&curseg->curseg_mutex);
1527                 return false;
1528         }
1529
1530         for (i = 0; i < nats_in_cursum(sum); i++) {
1531                 struct nat_entry *ne;
1532                 struct f2fs_nat_entry raw_ne;
1533                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1534
1535                 raw_ne = nat_in_journal(sum, i);
1536 retry:
1537                 write_lock(&nm_i->nat_tree_lock);
1538                 ne = __lookup_nat_cache(nm_i, nid);
1539                 if (ne) {
1540                         __set_nat_cache_dirty(nm_i, ne);
1541                         write_unlock(&nm_i->nat_tree_lock);
1542                         continue;
1543                 }
1544                 ne = grab_nat_entry(nm_i, nid);
1545                 if (!ne) {
1546                         write_unlock(&nm_i->nat_tree_lock);
1547                         goto retry;
1548                 }
1549                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1550                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1551                 nat_set_version(ne, raw_ne.version);
1552                 __set_nat_cache_dirty(nm_i, ne);
1553                 write_unlock(&nm_i->nat_tree_lock);
1554         }
1555         update_nats_in_cursum(sum, -i);
1556         mutex_unlock(&curseg->curseg_mutex);
1557         return true;
1558 }
1559
1560 /*
1561  * This function is called during the checkpointing process.
1562  */
1563 void flush_nat_entries(struct f2fs_sb_info *sbi)
1564 {
1565         struct f2fs_nm_info *nm_i = NM_I(sbi);
1566         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1567         struct f2fs_summary_block *sum = curseg->sum_blk;
1568         struct list_head *cur, *n;
1569         struct page *page = NULL;
1570         struct f2fs_nat_block *nat_blk = NULL;
1571         nid_t start_nid = 0, end_nid = 0;
1572         bool flushed;
1573
1574         flushed = flush_nats_in_journal(sbi);
1575
1576         if (!flushed)
1577                 mutex_lock(&curseg->curseg_mutex);
1578
1579         /* 1) flush dirty nat caches */
1580         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1581                 struct nat_entry *ne;
1582                 nid_t nid;
1583                 struct f2fs_nat_entry raw_ne;
1584                 int offset = -1;
1585                 block_t new_blkaddr;
1586
1587                 ne = list_entry(cur, struct nat_entry, list);
1588                 nid = nat_get_nid(ne);
1589
1590                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1591                         continue;
1592                 if (flushed)
1593                         goto to_nat_page;
1594
1595                 /* if there is room for nat enries in curseg->sumpage */
1596                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1597                 if (offset >= 0) {
1598                         raw_ne = nat_in_journal(sum, offset);
1599                         goto flush_now;
1600                 }
1601 to_nat_page:
1602                 if (!page || (start_nid > nid || nid > end_nid)) {
1603                         if (page) {
1604                                 f2fs_put_page(page, 1);
1605                                 page = NULL;
1606                         }
1607                         start_nid = START_NID(nid);
1608                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1609
1610                         /*
1611                          * get nat block with dirty flag, increased reference
1612                          * count, mapped and lock
1613                          */
1614                         page = get_next_nat_page(sbi, start_nid);
1615                         nat_blk = page_address(page);
1616                 }
1617
1618                 BUG_ON(!nat_blk);
1619                 raw_ne = nat_blk->entries[nid - start_nid];
1620 flush_now:
1621                 new_blkaddr = nat_get_blkaddr(ne);
1622
1623                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1624                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1625                 raw_ne.version = nat_get_version(ne);
1626
1627                 if (offset < 0) {
1628                         nat_blk->entries[nid - start_nid] = raw_ne;
1629                 } else {
1630                         nat_in_journal(sum, offset) = raw_ne;
1631                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1632                 }
1633
1634                 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1635                                         !add_free_nid(NM_I(sbi), nid)) {
1636                         write_lock(&nm_i->nat_tree_lock);
1637                         __del_from_nat_cache(nm_i, ne);
1638                         write_unlock(&nm_i->nat_tree_lock);
1639                 } else {
1640                         write_lock(&nm_i->nat_tree_lock);
1641                         __clear_nat_cache_dirty(nm_i, ne);
1642                         ne->checkpointed = true;
1643                         write_unlock(&nm_i->nat_tree_lock);
1644                 }
1645         }
1646         if (!flushed)
1647                 mutex_unlock(&curseg->curseg_mutex);
1648         f2fs_put_page(page, 1);
1649
1650         /* 2) shrink nat caches if necessary */
1651         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1652 }
1653
1654 static int init_node_manager(struct f2fs_sb_info *sbi)
1655 {
1656         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1657         struct f2fs_nm_info *nm_i = NM_I(sbi);
1658         unsigned char *version_bitmap;
1659         unsigned int nat_segs, nat_blocks;
1660
1661         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1662
1663         /* segment_count_nat includes pair segment so divide to 2. */
1664         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1665         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1666         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1667         nm_i->fcnt = 0;
1668         nm_i->nat_cnt = 0;
1669
1670         INIT_LIST_HEAD(&nm_i->free_nid_list);
1671         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1672         INIT_LIST_HEAD(&nm_i->nat_entries);
1673         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1674
1675         mutex_init(&nm_i->build_lock);
1676         spin_lock_init(&nm_i->free_nid_list_lock);
1677         rwlock_init(&nm_i->nat_tree_lock);
1678
1679         nm_i->init_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1680         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1681         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1682         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1683         if (!version_bitmap)
1684                 return -EFAULT;
1685
1686         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1687                                         GFP_KERNEL);
1688         if (!nm_i->nat_bitmap)
1689                 return -ENOMEM;
1690         return 0;
1691 }
1692
1693 int build_node_manager(struct f2fs_sb_info *sbi)
1694 {
1695         int err;
1696
1697         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1698         if (!sbi->nm_info)
1699                 return -ENOMEM;
1700
1701         err = init_node_manager(sbi);
1702         if (err)
1703                 return err;
1704
1705         build_free_nids(sbi);
1706         return 0;
1707 }
1708
1709 void destroy_node_manager(struct f2fs_sb_info *sbi)
1710 {
1711         struct f2fs_nm_info *nm_i = NM_I(sbi);
1712         struct free_nid *i, *next_i;
1713         struct nat_entry *natvec[NATVEC_SIZE];
1714         nid_t nid = 0;
1715         unsigned int found;
1716
1717         if (!nm_i)
1718                 return;
1719
1720         /* destroy free nid list */
1721         spin_lock(&nm_i->free_nid_list_lock);
1722         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1723                 BUG_ON(i->state == NID_ALLOC);
1724                 __del_from_free_nid_list(i);
1725                 nm_i->fcnt--;
1726         }
1727         BUG_ON(nm_i->fcnt);
1728         spin_unlock(&nm_i->free_nid_list_lock);
1729
1730         /* destroy nat cache */
1731         write_lock(&nm_i->nat_tree_lock);
1732         while ((found = __gang_lookup_nat_cache(nm_i,
1733                                         nid, NATVEC_SIZE, natvec))) {
1734                 unsigned idx;
1735                 for (idx = 0; idx < found; idx++) {
1736                         struct nat_entry *e = natvec[idx];
1737                         nid = nat_get_nid(e) + 1;
1738                         __del_from_nat_cache(nm_i, e);
1739                 }
1740         }
1741         BUG_ON(nm_i->nat_cnt);
1742         write_unlock(&nm_i->nat_tree_lock);
1743
1744         kfree(nm_i->nat_bitmap);
1745         sbi->nm_info = NULL;
1746         kfree(nm_i);
1747 }
1748
1749 int __init create_node_manager_caches(void)
1750 {
1751         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1752                         sizeof(struct nat_entry), NULL);
1753         if (!nat_entry_slab)
1754                 return -ENOMEM;
1755
1756         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1757                         sizeof(struct free_nid), NULL);
1758         if (!free_nid_slab) {
1759                 kmem_cache_destroy(nat_entry_slab);
1760                 return -ENOMEM;
1761         }
1762         return 0;
1763 }
1764
1765 void destroy_node_manager_caches(void)
1766 {
1767         kmem_cache_destroy(free_nid_slab);
1768         kmem_cache_destroy(nat_entry_slab);
1769 }