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 <= 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         struct address_space *node_mapping = sbi->node_inode->i_mapping;
 678         int err = 0, cont = 1;
 679         int level, offset[4], noffset[4];
 680         unsigned int nofs = 0;
 681         struct f2fs_node *rn;
 682         struct dnode_of_data dn;
 683         struct page *page;
 684
 685         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 686
 687         level = get_node_path(from, offset, noffset);
 688 restart:
 689         page = get_node_page(sbi, inode->i_ino);
 690         if (IS_ERR(page)) {
 691                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 692                 return PTR_ERR(page);
 693         }
 694
 695         set_new_dnode(&dn, inode, page, NULL, 0);
 696         unlock_page(page);
 697
 698         rn = page_address(page);
 699         switch (level) {
 700         case 0:
 701         case 1:
 702                 nofs = noffset[1];
 703                 break;
 704         case 2:
 705                 nofs = noffset[1];
 706                 if (!offset[level - 1])
 707                         goto skip_partial;
 708                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 709                 if (err < 0 && err != -ENOENT)
 710                         goto fail;
 711                 nofs += 1 + NIDS_PER_BLOCK;
 712                 break;
 713         case 3:
 714                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 715                 if (!offset[level - 1])
 716                         goto skip_partial;
 717                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 718                 if (err < 0 && err != -ENOENT)
 719                         goto fail;
 720                 break;
 721         default:
 722                 BUG();
 723         }
 724
 725 skip_partial:
 726         while (cont) {
 727                 dn.nid = le32_to_cpu(rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]);
 728                 switch (offset[0]) {
 729                 case NODE_DIR1_BLOCK:
 730                 case NODE_DIR2_BLOCK:
 731                         err = truncate_dnode(&dn);
 732                         break;
 733
 734                 case NODE_IND1_BLOCK:
 735                 case NODE_IND2_BLOCK:
 736                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 737                         break;
 738
 739                 case NODE_DIND_BLOCK:
 740                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 741                         cont = 0;
 742                         break;
 743
 744                 default:
 745                         BUG();
 746                 }
 747                 if (err < 0 && err != -ENOENT)
 748                         goto fail;
 749                 if (offset[1] == 0 &&
 750                                 rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 751                         lock_page(page);
 752                         if (page->mapping != node_mapping) {
 753                                 f2fs_put_page(page, 1);
 754                                 goto restart;
 755                         }
 756                         wait_on_page_writeback(page);
 757                         rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 758                         set_page_dirty(page);
 759                         unlock_page(page);
 760                 }
 761                 offset[1] = 0;
 762                 offset[0]++;
 763                 nofs += err;
 764         }
 765 fail:
 766         f2fs_put_page(page, 0);
 767         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 768         return err > 0 ? 0 : err;
 769 }
 770
 771 /*
 772  * Caller should grab and release a mutex by calling mutex_lock_op() and
 773  * mutex_unlock_op().
 774  */
 775 int remove_inode_page(struct inode *inode)
 776 {
 777         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 778         struct page *page;
 779         nid_t ino = inode->i_ino;
 780         struct dnode_of_data dn;
 781
 782         page = get_node_page(sbi, ino);
 783         if (IS_ERR(page))
 784                 return PTR_ERR(page);
 785
 786         if (F2FS_I(inode)->i_xattr_nid) {
 787                 nid_t nid = F2FS_I(inode)->i_xattr_nid;
 788                 struct page *npage = get_node_page(sbi, nid);
 789
 790                 if (IS_ERR(npage))
 791                         return PTR_ERR(npage);
 792
 793                 F2FS_I(inode)->i_xattr_nid = 0;
 794                 set_new_dnode(&dn, inode, page, npage, nid);
 795                 dn.inode_page_locked = 1;
 796                 truncate_node(&dn);
 797         }
 798
 799         /* 0 is possible, after f2fs_new_inode() is failed */
 800         BUG_ON(inode->i_blocks != 0 && inode->i_blocks != 1);
 801         set_new_dnode(&dn, inode, page, page, ino);
 802         truncate_node(&dn);
 803         return 0;
 804 }
 805
 806 int new_inode_page(struct inode *inode, const struct qstr *name)
 807 {
 808         struct page *page;
 809         struct dnode_of_data dn;
 810
 811         /* allocate inode page for new inode */
 812         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 813         page = new_node_page(&dn, 0);
 814         init_dent_inode(name, page);
 815         if (IS_ERR(page))
 816                 return PTR_ERR(page);
 817         f2fs_put_page(page, 1);
 818         return 0;
 819 }
 820
 821 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
 822 {
 823         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 824         struct address_space *mapping = sbi->node_inode->i_mapping;
 825         struct node_info old_ni, new_ni;
 826         struct page *page;
 827         int err;
 828
 829         if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
 830                 return ERR_PTR(-EPERM);
 831
 832         page = grab_cache_page(mapping, dn->nid);
 833         if (!page)
 834                 return ERR_PTR(-ENOMEM);
 835
 836         get_node_info(sbi, dn->nid, &old_ni);
 837
 838         SetPageUptodate(page);
 839         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 840
 841         /* Reinitialize old_ni with new node page */
 842         BUG_ON(old_ni.blk_addr != NULL_ADDR);
 843         new_ni = old_ni;
 844         new_ni.ino = dn->inode->i_ino;
 845
 846         if (!inc_valid_node_count(sbi, dn->inode, 1)) {
 847                 err = -ENOSPC;
 848                 goto fail;
 849         }
 850         set_node_addr(sbi, &new_ni, NEW_ADDR);
 851         set_cold_node(dn->inode, page);
 852
 853         dn->node_page = page;
 854         sync_inode_page(dn);
 855         set_page_dirty(page);
 856         if (ofs == 0)
 857                 inc_valid_inode_count(sbi);
 858
 859         return page;
 860
 861 fail:
 862         clear_node_page_dirty(page);
 863         f2fs_put_page(page, 1);
 864         return ERR_PTR(err);
 865 }
 866
 867 /*
 868  * Caller should do after getting the following values.
 869  * 0: f2fs_put_page(page, 0)
 870  * LOCKED_PAGE: f2fs_put_page(page, 1)
 871  * error: nothing
 872  */
 873 static int read_node_page(struct page *page, int type)
 874 {
 875         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 876         struct node_info ni;
 877
 878         get_node_info(sbi, page->index, &ni);
 879
 880         if (ni.blk_addr == NULL_ADDR) {
 881                 f2fs_put_page(page, 1);
 882                 return -ENOENT;
 883         }
 884
 885         if (PageUptodate(page))
 886                 return LOCKED_PAGE;
 887
 888         return f2fs_readpage(sbi, page, ni.blk_addr, type);
 889 }
 890
 891 /*
 892  * Readahead a node page
 893  */
 894 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 895 {
 896         struct address_space *mapping = sbi->node_inode->i_mapping;
 897         struct page *apage;
 898         int err;
 899
 900         apage = find_get_page(mapping, nid);
 901         if (apage && PageUptodate(apage)) {
 902                 f2fs_put_page(apage, 0);
 903                 return;
 904         }
 905         f2fs_put_page(apage, 0);
 906
 907         apage = grab_cache_page(mapping, nid);
 908         if (!apage)
 909                 return;
 910
 911         err = read_node_page(apage, READA);
 912         if (err == 0)
 913                 f2fs_put_page(apage, 0);
 914         else if (err == LOCKED_PAGE)
 915                 f2fs_put_page(apage, 1);
 916         return;
 917 }
 918
 919 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 920 {
 921         struct address_space *mapping = sbi->node_inode->i_mapping;
 922         struct page *page;
 923         int err;
 924 repeat:
 925         page = grab_cache_page(mapping, nid);
 926         if (!page)
 927                 return ERR_PTR(-ENOMEM);
 928
 929         err = read_node_page(page, READ_SYNC);
 930         if (err < 0)
 931                 return ERR_PTR(err);
 932         else if (err == LOCKED_PAGE)
 933                 goto got_it;
 934
 935         lock_page(page);
 936         if (!PageUptodate(page)) {
 937                 f2fs_put_page(page, 1);
 938                 return ERR_PTR(-EIO);
 939         }
 940         if (page->mapping != mapping) {
 941                 f2fs_put_page(page, 1);
 942                 goto repeat;
 943         }
 944 got_it:
 945         BUG_ON(nid != nid_of_node(page));
 946         mark_page_accessed(page);
 947         return page;
 948 }
 949
 950 /*
 951  * Return a locked page for the desired node page.
 952  * And, readahead MAX_RA_NODE number of node pages.
 953  */
 954 struct page *get_node_page_ra(struct page *parent, int start)
 955 {
 956         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 957         struct address_space *mapping = sbi->node_inode->i_mapping;
 958         struct blk_plug plug;
 959         struct page *page;
 960         int err, i, end;
 961         nid_t nid;
 962
 963         /* First, try getting the desired direct node. */
 964         nid = get_nid(parent, start, false);
 965         if (!nid)
 966                 return ERR_PTR(-ENOENT);
 967 repeat:
 968         page = grab_cache_page(mapping, nid);
 969         if (!page)
 970                 return ERR_PTR(-ENOMEM);
 971
 972         err = read_node_page(page, READ_SYNC);
 973         if (err < 0)
 974                 return ERR_PTR(err);
 975         else if (err == LOCKED_PAGE)
 976                 goto page_hit;
 977
 978         blk_start_plug(&plug);
 979
 980         /* Then, try readahead for siblings of the desired node */
 981         end = start + MAX_RA_NODE;
 982         end = min(end, NIDS_PER_BLOCK);
 983         for (i = start + 1; i < end; i++) {
 984                 nid = get_nid(parent, i, false);
 985                 if (!nid)
 986                         continue;
 987                 ra_node_page(sbi, nid);
 988         }
 989
 990         blk_finish_plug(&plug);
 991
 992         lock_page(page);
 993         if (page->mapping != mapping) {
 994                 f2fs_put_page(page, 1);
 995                 goto repeat;
 996         }
 997 page_hit:
 998         if (!PageUptodate(page)) {
 999                 f2fs_put_page(page, 1);
1000                 return ERR_PTR(-EIO);
1001         }
1002         mark_page_accessed(page);
1003         return page;
1004 }
1005
1006 void sync_inode_page(struct dnode_of_data *dn)
1007 {
1008         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1009                 update_inode(dn->inode, dn->node_page);
1010         } else if (dn->inode_page) {
1011                 if (!dn->inode_page_locked)
1012                         lock_page(dn->inode_page);
1013                 update_inode(dn->inode, dn->inode_page);
1014                 if (!dn->inode_page_locked)
1015                         unlock_page(dn->inode_page);
1016         } else {
1017                 update_inode_page(dn->inode);
1018         }
1019 }
1020
1021 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1022                                         struct writeback_control *wbc)
1023 {
1024         struct address_space *mapping = sbi->node_inode->i_mapping;
1025         pgoff_t index, end;
1026         struct pagevec pvec;
1027         int step = ino ? 2 : 0;
1028         int nwritten = 0, wrote = 0;
1029
1030         pagevec_init(&pvec, 0);
1031
1032 next_step:
1033         index = 0;
1034         end = LONG_MAX;
1035
1036         while (index <= end) {
1037                 int i, nr_pages;
1038                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1039                                 PAGECACHE_TAG_DIRTY,
1040                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1041                 if (nr_pages == 0)
1042                         break;
1043
1044                 for (i = 0; i < nr_pages; i++) {
1045                         struct page *page = pvec.pages[i];
1046
1047                         /*
1048                          * flushing sequence with step:
1049                          * 0. indirect nodes
1050                          * 1. dentry dnodes
1051                          * 2. file dnodes
1052                          */
1053                         if (step == 0 && IS_DNODE(page))
1054                                 continue;
1055                         if (step == 1 && (!IS_DNODE(page) ||
1056                                                 is_cold_node(page)))
1057                                 continue;
1058                         if (step == 2 && (!IS_DNODE(page) ||
1059                                                 !is_cold_node(page)))
1060                                 continue;
1061
1062                         /*
1063                          * If an fsync mode,
1064                          * we should not skip writing node pages.
1065                          */
1066                         if (ino && ino_of_node(page) == ino)
1067                                 lock_page(page);
1068                         else if (!trylock_page(page))
1069                                 continue;
1070
1071                         if (unlikely(page->mapping != mapping)) {
1072 continue_unlock:
1073                                 unlock_page(page);
1074                                 continue;
1075                         }
1076                         if (ino && ino_of_node(page) != ino)
1077                                 goto continue_unlock;
1078
1079                         if (!PageDirty(page)) {
1080                                 /* someone wrote it for us */
1081                                 goto continue_unlock;
1082                         }
1083
1084                         if (!clear_page_dirty_for_io(page))
1085                                 goto continue_unlock;
1086
1087                         /* called by fsync() */
1088                         if (ino && IS_DNODE(page)) {
1089                                 int mark = !is_checkpointed_node(sbi, ino);
1090                                 set_fsync_mark(page, 1);
1091                                 if (IS_INODE(page))
1092                                         set_dentry_mark(page, mark);
1093                                 nwritten++;
1094                         } else {
1095                                 set_fsync_mark(page, 0);
1096                                 set_dentry_mark(page, 0);
1097                         }
1098                         mapping->a_ops->writepage(page, wbc);
1099                         wrote++;
1100
1101                         if (--wbc->nr_to_write == 0)
1102                                 break;
1103                 }
1104                 pagevec_release(&pvec);
1105                 cond_resched();
1106
1107                 if (wbc->nr_to_write == 0) {
1108                         step = 2;
1109                         break;
1110                 }
1111         }
1112
1113         if (step < 2) {
1114                 step++;
1115                 goto next_step;
1116         }
1117
1118         if (wrote)
1119                 f2fs_submit_bio(sbi, NODE, wbc->sync_mode == WB_SYNC_ALL);
1120
1121         return nwritten;
1122 }
1123
1124 static int f2fs_write_node_page(struct page *page,
1125                                 struct writeback_control *wbc)
1126 {
1127         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1128         nid_t nid;
1129         block_t new_addr;
1130         struct node_info ni;
1131
1132         wait_on_page_writeback(page);
1133
1134         /* get old block addr of this node page */
1135         nid = nid_of_node(page);
1136         BUG_ON(page->index != nid);
1137
1138         get_node_info(sbi, nid, &ni);
1139
1140         /* This page is already truncated */
1141         if (ni.blk_addr == NULL_ADDR) {
1142                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1143                 unlock_page(page);
1144                 return 0;
1145         }
1146
1147         if (wbc->for_reclaim) {
1148                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1149                 wbc->pages_skipped++;
1150                 set_page_dirty(page);
1151                 return AOP_WRITEPAGE_ACTIVATE;
1152         }
1153
1154         mutex_lock(&sbi->node_write);
1155         set_page_writeback(page);
1156         write_node_page(sbi, page, nid, ni.blk_addr, &new_addr);
1157         set_node_addr(sbi, &ni, new_addr);
1158         dec_page_count(sbi, F2FS_DIRTY_NODES);
1159         mutex_unlock(&sbi->node_write);
1160         unlock_page(page);
1161         return 0;
1162 }
1163
1164 /*
1165  * It is very important to gather dirty pages and write at once, so that we can
1166  * submit a big bio without interfering other data writes.
1167  * Be default, 512 pages (2MB), a segment size, is quite reasonable.
1168  */
1169 #define COLLECT_DIRTY_NODES     512
1170 static int f2fs_write_node_pages(struct address_space *mapping,
1171                             struct writeback_control *wbc)
1172 {
1173         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1174         long nr_to_write = wbc->nr_to_write;
1175
1176         /* First check balancing cached NAT entries */
1177         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK)) {
1178                 f2fs_sync_fs(sbi->sb, true);
1179                 return 0;
1180         }
1181
1182         /* collect a number of dirty node pages and write together */
1183         if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1184                 return 0;
1185
1186         /* if mounting is failed, skip writing node pages */
1187         wbc->nr_to_write = max_hw_blocks(sbi);
1188         sync_node_pages(sbi, 0, wbc);
1189         wbc->nr_to_write = nr_to_write - (max_hw_blocks(sbi) - wbc->nr_to_write);
1190         return 0;
1191 }
1192
1193 static int f2fs_set_node_page_dirty(struct page *page)
1194 {
1195         struct address_space *mapping = page->mapping;
1196         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1197
1198         SetPageUptodate(page);
1199         if (!PageDirty(page)) {
1200                 __set_page_dirty_nobuffers(page);
1201                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1202                 SetPagePrivate(page);
1203                 return 1;
1204         }
1205         return 0;
1206 }
1207
1208 static void f2fs_invalidate_node_page(struct page *page, unsigned long offset)
1209 {
1210         struct inode *inode = page->mapping->host;
1211         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1212         if (PageDirty(page))
1213                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1214         ClearPagePrivate(page);
1215 }
1216
1217 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1218 {
1219         ClearPagePrivate(page);
1220         return 1;
1221 }
1222
1223 /*
1224  * Structure of the f2fs node operations
1225  */
1226 const struct address_space_operations f2fs_node_aops = {
1227         .writepage      = f2fs_write_node_page,
1228         .writepages     = f2fs_write_node_pages,
1229         .set_page_dirty = f2fs_set_node_page_dirty,
1230         .invalidatepage = f2fs_invalidate_node_page,
1231         .releasepage    = f2fs_release_node_page,
1232 };
1233
1234 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1235 {
1236         struct list_head *this;
1237         struct free_nid *i;
1238         list_for_each(this, head) {
1239                 i = list_entry(this, struct free_nid, list);
1240                 if (i->nid == n)
1241                         return i;
1242         }
1243         return NULL;
1244 }
1245
1246 static void __del_from_free_nid_list(struct free_nid *i)
1247 {
1248         list_del(&i->list);
1249         kmem_cache_free(free_nid_slab, i);
1250 }
1251
1252 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build)
1253 {
1254         struct free_nid *i;
1255         struct nat_entry *ne;
1256         bool allocated = false;
1257
1258         if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1259                 return -1;
1260
1261         /* 0 nid should not be used */
1262         if (nid == 0)
1263                 return 0;
1264
1265         if (!build)
1266                 goto retry;
1267
1268         /* do not add allocated nids */
1269         read_lock(&nm_i->nat_tree_lock);
1270         ne = __lookup_nat_cache(nm_i, nid);
1271         if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1272                 allocated = true;
1273         read_unlock(&nm_i->nat_tree_lock);
1274         if (allocated)
1275                 return 0;
1276 retry:
1277         i = kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1278         if (!i) {
1279                 cond_resched();
1280                 goto retry;
1281         }
1282         i->nid = nid;
1283         i->state = NID_NEW;
1284
1285         spin_lock(&nm_i->free_nid_list_lock);
1286         if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1287                 spin_unlock(&nm_i->free_nid_list_lock);
1288                 kmem_cache_free(free_nid_slab, i);
1289                 return 0;
1290         }
1291         list_add_tail(&i->list, &nm_i->free_nid_list);
1292         nm_i->fcnt++;
1293         spin_unlock(&nm_i->free_nid_list_lock);
1294         return 1;
1295 }
1296
1297 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1298 {
1299         struct free_nid *i;
1300         spin_lock(&nm_i->free_nid_list_lock);
1301         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1302         if (i && i->state == NID_NEW) {
1303                 __del_from_free_nid_list(i);
1304                 nm_i->fcnt--;
1305         }
1306         spin_unlock(&nm_i->free_nid_list_lock);
1307 }
1308
1309 static void scan_nat_page(struct f2fs_nm_info *nm_i,
1310                         struct page *nat_page, nid_t start_nid)
1311 {
1312         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1313         block_t blk_addr;
1314         int i;
1315
1316         i = start_nid % NAT_ENTRY_PER_BLOCK;
1317
1318         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1319
1320                 if (start_nid >= nm_i->max_nid)
1321                         break;
1322
1323                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1324                 BUG_ON(blk_addr == NEW_ADDR);
1325                 if (blk_addr == NULL_ADDR) {
1326                         if (add_free_nid(nm_i, start_nid, true) < 0)
1327                                 break;
1328                 }
1329         }
1330 }
1331
1332 static void build_free_nids(struct f2fs_sb_info *sbi)
1333 {
1334         struct f2fs_nm_info *nm_i = NM_I(sbi);
1335         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1336         struct f2fs_summary_block *sum = curseg->sum_blk;
1337         int i = 0;
1338         nid_t nid = nm_i->next_scan_nid;
1339
1340         /* Enough entries */
1341         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1342                 return;
1343
1344         /* readahead nat pages to be scanned */
1345         ra_nat_pages(sbi, nid);
1346
1347         while (1) {
1348                 struct page *page = get_current_nat_page(sbi, nid);
1349
1350                 scan_nat_page(nm_i, page, nid);
1351                 f2fs_put_page(page, 1);
1352
1353                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1354                 if (nid >= nm_i->max_nid)
1355                         nid = 0;
1356
1357                 if (i++ == FREE_NID_PAGES)
1358                         break;
1359         }
1360
1361         /* go to the next free nat pages to find free nids abundantly */
1362         nm_i->next_scan_nid = nid;
1363
1364         /* find free nids from current sum_pages */
1365         mutex_lock(&curseg->curseg_mutex);
1366         for (i = 0; i < nats_in_cursum(sum); i++) {
1367                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1368                 nid = le32_to_cpu(nid_in_journal(sum, i));
1369                 if (addr == NULL_ADDR)
1370                         add_free_nid(nm_i, nid, true);
1371                 else
1372                         remove_free_nid(nm_i, nid);
1373         }
1374         mutex_unlock(&curseg->curseg_mutex);
1375 }
1376
1377 /*
1378  * If this function returns success, caller can obtain a new nid
1379  * from second parameter of this function.
1380  * The returned nid could be used ino as well as nid when inode is created.
1381  */
1382 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1383 {
1384         struct f2fs_nm_info *nm_i = NM_I(sbi);
1385         struct free_nid *i = NULL;
1386         struct list_head *this;
1387 retry:
1388         if (sbi->total_valid_node_count + 1 >= nm_i->max_nid)
1389                 return false;
1390
1391         spin_lock(&nm_i->free_nid_list_lock);
1392
1393         /* We should not use stale free nids created by build_free_nids */
1394         if (nm_i->fcnt && !sbi->on_build_free_nids) {
1395                 BUG_ON(list_empty(&nm_i->free_nid_list));
1396                 list_for_each(this, &nm_i->free_nid_list) {
1397                         i = list_entry(this, struct free_nid, list);
1398                         if (i->state == NID_NEW)
1399                                 break;
1400                 }
1401
1402                 BUG_ON(i->state != NID_NEW);
1403                 *nid = i->nid;
1404                 i->state = NID_ALLOC;
1405                 nm_i->fcnt--;
1406                 spin_unlock(&nm_i->free_nid_list_lock);
1407                 return true;
1408         }
1409         spin_unlock(&nm_i->free_nid_list_lock);
1410
1411         /* Let's scan nat pages and its caches to get free nids */
1412         mutex_lock(&nm_i->build_lock);
1413         sbi->on_build_free_nids = 1;
1414         build_free_nids(sbi);
1415         sbi->on_build_free_nids = 0;
1416         mutex_unlock(&nm_i->build_lock);
1417         goto retry;
1418 }
1419
1420 /*
1421  * alloc_nid() should be called prior to this function.
1422  */
1423 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1424 {
1425         struct f2fs_nm_info *nm_i = NM_I(sbi);
1426         struct free_nid *i;
1427
1428         spin_lock(&nm_i->free_nid_list_lock);
1429         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1430         BUG_ON(!i || i->state != NID_ALLOC);
1431         __del_from_free_nid_list(i);
1432         spin_unlock(&nm_i->free_nid_list_lock);
1433 }
1434
1435 /*
1436  * alloc_nid() should be called prior to this function.
1437  */
1438 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1439 {
1440         struct f2fs_nm_info *nm_i = NM_I(sbi);
1441         struct free_nid *i;
1442
1443         spin_lock(&nm_i->free_nid_list_lock);
1444         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1445         BUG_ON(!i || i->state != NID_ALLOC);
1446         if (nm_i->fcnt > 2 * MAX_FREE_NIDS) {
1447                 __del_from_free_nid_list(i);
1448         } else {
1449                 i->state = NID_NEW;
1450                 nm_i->fcnt++;
1451         }
1452         spin_unlock(&nm_i->free_nid_list_lock);
1453 }
1454
1455 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1456                 struct f2fs_summary *sum, struct node_info *ni,
1457                 block_t new_blkaddr)
1458 {
1459         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1460         set_node_addr(sbi, ni, new_blkaddr);
1461         clear_node_page_dirty(page);
1462 }
1463
1464 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1465 {
1466         struct address_space *mapping = sbi->node_inode->i_mapping;
1467         struct f2fs_node *src, *dst;
1468         nid_t ino = ino_of_node(page);
1469         struct node_info old_ni, new_ni;
1470         struct page *ipage;
1471
1472         ipage = grab_cache_page(mapping, ino);
1473         if (!ipage)
1474                 return -ENOMEM;
1475
1476         /* Should not use this inode  from free nid list */
1477         remove_free_nid(NM_I(sbi), ino);
1478
1479         get_node_info(sbi, ino, &old_ni);
1480         SetPageUptodate(ipage);
1481         fill_node_footer(ipage, ino, ino, 0, true);
1482
1483         src = (struct f2fs_node *)page_address(page);
1484         dst = (struct f2fs_node *)page_address(ipage);
1485
1486         memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i);
1487         dst->i.i_size = 0;
1488         dst->i.i_blocks = cpu_to_le64(1);
1489         dst->i.i_links = cpu_to_le32(1);
1490         dst->i.i_xattr_nid = 0;
1491
1492         new_ni = old_ni;
1493         new_ni.ino = ino;
1494
1495         set_node_addr(sbi, &new_ni, NEW_ADDR);
1496         inc_valid_inode_count(sbi);
1497
1498         f2fs_put_page(ipage, 1);
1499         return 0;
1500 }
1501
1502 int restore_node_summary(struct f2fs_sb_info *sbi,
1503                         unsigned int segno, struct f2fs_summary_block *sum)
1504 {
1505         struct f2fs_node *rn;
1506         struct f2fs_summary *sum_entry;
1507         struct page *page;
1508         block_t addr;
1509         int i, last_offset;
1510
1511         /* alloc temporal page for read node */
1512         page = alloc_page(GFP_NOFS | __GFP_ZERO);
1513         if (IS_ERR(page))
1514                 return PTR_ERR(page);
1515         lock_page(page);
1516
1517         /* scan the node segment */
1518         last_offset = sbi->blocks_per_seg;
1519         addr = START_BLOCK(sbi, segno);
1520         sum_entry = &sum->entries[0];
1521
1522         for (i = 0; i < last_offset; i++, sum_entry++) {
1523                 /*
1524                  * In order to read next node page,
1525                  * we must clear PageUptodate flag.
1526                  */
1527                 ClearPageUptodate(page);
1528
1529                 if (f2fs_readpage(sbi, page, addr, READ_SYNC))
1530                         goto out;
1531
1532                 lock_page(page);
1533                 rn = (struct f2fs_node *)page_address(page);
1534                 sum_entry->nid = rn->footer.nid;
1535                 sum_entry->version = 0;
1536                 sum_entry->ofs_in_node = 0;
1537                 addr++;
1538         }
1539         unlock_page(page);
1540 out:
1541         __free_pages(page, 0);
1542         return 0;
1543 }
1544
1545 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1546 {
1547         struct f2fs_nm_info *nm_i = NM_I(sbi);
1548         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1549         struct f2fs_summary_block *sum = curseg->sum_blk;
1550         int i;
1551
1552         mutex_lock(&curseg->curseg_mutex);
1553
1554         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1555                 mutex_unlock(&curseg->curseg_mutex);
1556                 return false;
1557         }
1558
1559         for (i = 0; i < nats_in_cursum(sum); i++) {
1560                 struct nat_entry *ne;
1561                 struct f2fs_nat_entry raw_ne;
1562                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1563
1564                 raw_ne = nat_in_journal(sum, i);
1565 retry:
1566                 write_lock(&nm_i->nat_tree_lock);
1567                 ne = __lookup_nat_cache(nm_i, nid);
1568                 if (ne) {
1569                         __set_nat_cache_dirty(nm_i, ne);
1570                         write_unlock(&nm_i->nat_tree_lock);
1571                         continue;
1572                 }
1573                 ne = grab_nat_entry(nm_i, nid);
1574                 if (!ne) {
1575                         write_unlock(&nm_i->nat_tree_lock);
1576                         goto retry;
1577                 }
1578                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1579                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1580                 nat_set_version(ne, raw_ne.version);
1581                 __set_nat_cache_dirty(nm_i, ne);
1582                 write_unlock(&nm_i->nat_tree_lock);
1583         }
1584         update_nats_in_cursum(sum, -i);
1585         mutex_unlock(&curseg->curseg_mutex);
1586         return true;
1587 }
1588
1589 /*
1590  * This function is called during the checkpointing process.
1591  */
1592 void flush_nat_entries(struct f2fs_sb_info *sbi)
1593 {
1594         struct f2fs_nm_info *nm_i = NM_I(sbi);
1595         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1596         struct f2fs_summary_block *sum = curseg->sum_blk;
1597         struct list_head *cur, *n;
1598         struct page *page = NULL;
1599         struct f2fs_nat_block *nat_blk = NULL;
1600         nid_t start_nid = 0, end_nid = 0;
1601         bool flushed;
1602
1603         flushed = flush_nats_in_journal(sbi);
1604
1605         if (!flushed)
1606                 mutex_lock(&curseg->curseg_mutex);
1607
1608         /* 1) flush dirty nat caches */
1609         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1610                 struct nat_entry *ne;
1611                 nid_t nid;
1612                 struct f2fs_nat_entry raw_ne;
1613                 int offset = -1;
1614                 block_t new_blkaddr;
1615
1616                 ne = list_entry(cur, struct nat_entry, list);
1617                 nid = nat_get_nid(ne);
1618
1619                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1620                         continue;
1621                 if (flushed)
1622                         goto to_nat_page;
1623
1624                 /* if there is room for nat enries in curseg->sumpage */
1625                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1626                 if (offset >= 0) {
1627                         raw_ne = nat_in_journal(sum, offset);
1628                         goto flush_now;
1629                 }
1630 to_nat_page:
1631                 if (!page || (start_nid > nid || nid > end_nid)) {
1632                         if (page) {
1633                                 f2fs_put_page(page, 1);
1634                                 page = NULL;
1635                         }
1636                         start_nid = START_NID(nid);
1637                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1638
1639                         /*
1640                          * get nat block with dirty flag, increased reference
1641                          * count, mapped and lock
1642                          */
1643                         page = get_next_nat_page(sbi, start_nid);
1644                         nat_blk = page_address(page);
1645                 }
1646
1647                 BUG_ON(!nat_blk);
1648                 raw_ne = nat_blk->entries[nid - start_nid];
1649 flush_now:
1650                 new_blkaddr = nat_get_blkaddr(ne);
1651
1652                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1653                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1654                 raw_ne.version = nat_get_version(ne);
1655
1656                 if (offset < 0) {
1657                         nat_blk->entries[nid - start_nid] = raw_ne;
1658                 } else {
1659                         nat_in_journal(sum, offset) = raw_ne;
1660                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1661                 }
1662
1663                 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1664                                 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1665                         write_lock(&nm_i->nat_tree_lock);
1666                         __del_from_nat_cache(nm_i, ne);
1667                         write_unlock(&nm_i->nat_tree_lock);
1668                 } else {
1669                         write_lock(&nm_i->nat_tree_lock);
1670                         __clear_nat_cache_dirty(nm_i, ne);
1671                         ne->checkpointed = true;
1672                         write_unlock(&nm_i->nat_tree_lock);
1673                 }
1674         }
1675         if (!flushed)
1676                 mutex_unlock(&curseg->curseg_mutex);
1677         f2fs_put_page(page, 1);
1678
1679         /* 2) shrink nat caches if necessary */
1680         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1681 }
1682
1683 static int init_node_manager(struct f2fs_sb_info *sbi)
1684 {
1685         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1686         struct f2fs_nm_info *nm_i = NM_I(sbi);
1687         unsigned char *version_bitmap;
1688         unsigned int nat_segs, nat_blocks;
1689
1690         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1691
1692         /* segment_count_nat includes pair segment so divide to 2. */
1693         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1694         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1695         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1696         nm_i->fcnt = 0;
1697         nm_i->nat_cnt = 0;
1698
1699         INIT_LIST_HEAD(&nm_i->free_nid_list);
1700         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1701         INIT_LIST_HEAD(&nm_i->nat_entries);
1702         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1703
1704         mutex_init(&nm_i->build_lock);
1705         spin_lock_init(&nm_i->free_nid_list_lock);
1706         rwlock_init(&nm_i->nat_tree_lock);
1707
1708         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1709         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1710         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1711         if (!version_bitmap)
1712                 return -EFAULT;
1713
1714         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1715                                         GFP_KERNEL);
1716         if (!nm_i->nat_bitmap)
1717                 return -ENOMEM;
1718         return 0;
1719 }
1720
1721 int build_node_manager(struct f2fs_sb_info *sbi)
1722 {
1723         int err;
1724
1725         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1726         if (!sbi->nm_info)
1727                 return -ENOMEM;
1728
1729         err = init_node_manager(sbi);
1730         if (err)
1731                 return err;
1732
1733         build_free_nids(sbi);
1734         return 0;
1735 }
1736
1737 void destroy_node_manager(struct f2fs_sb_info *sbi)
1738 {
1739         struct f2fs_nm_info *nm_i = NM_I(sbi);
1740         struct free_nid *i, *next_i;
1741         struct nat_entry *natvec[NATVEC_SIZE];
1742         nid_t nid = 0;
1743         unsigned int found;
1744
1745         if (!nm_i)
1746                 return;
1747
1748         /* destroy free nid list */
1749         spin_lock(&nm_i->free_nid_list_lock);
1750         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1751                 BUG_ON(i->state == NID_ALLOC);
1752                 __del_from_free_nid_list(i);
1753                 nm_i->fcnt--;
1754         }
1755         BUG_ON(nm_i->fcnt);
1756         spin_unlock(&nm_i->free_nid_list_lock);
1757
1758         /* destroy nat cache */
1759         write_lock(&nm_i->nat_tree_lock);
1760         while ((found = __gang_lookup_nat_cache(nm_i,
1761                                         nid, NATVEC_SIZE, natvec))) {
1762                 unsigned idx;
1763                 for (idx = 0; idx < found; idx++) {
1764                         struct nat_entry *e = natvec[idx];
1765                         nid = nat_get_nid(e) + 1;
1766                         __del_from_nat_cache(nm_i, e);
1767                 }
1768         }
1769         BUG_ON(nm_i->nat_cnt);
1770         write_unlock(&nm_i->nat_tree_lock);
1771
1772         kfree(nm_i->nat_bitmap);
1773         sbi->nm_info = NULL;
1774         kfree(nm_i);
1775 }
1776
1777 int __init create_node_manager_caches(void)
1778 {
1779         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1780                         sizeof(struct nat_entry), NULL);
1781         if (!nat_entry_slab)
1782                 return -ENOMEM;
1783
1784         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1785                         sizeof(struct free_nid), NULL);
1786         if (!free_nid_slab) {
1787                 kmem_cache_destroy(nat_entry_slab);
1788                 return -ENOMEM;
1789         }
1790         return 0;
1791 }
1792
1793 void destroy_node_manager_caches(void)
1794 {
1795         kmem_cache_destroy(free_nid_slab);
1796         kmem_cache_destroy(nat_entry_slab);
1797 }