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 = META_MAPPING(sbi);
  91         struct f2fs_nm_info *nm_i = NM_I(sbi);
  92         struct page *page;
  93         pgoff_t index;
  94         int i;
  95         struct f2fs_io_info fio = {
  96                 .type = META,
  97                 .rw = READ_SYNC | REQ_META | REQ_PRIO
  98         };
  99
 100
 101         for (i = 0; i < FREE_NID_PAGES; i++, nid += NAT_ENTRY_PER_BLOCK) {
 102                 if (unlikely(nid >= nm_i->max_nid))
 103                         nid = 0;
 104                 index = current_nat_addr(sbi, nid);
 105
 106                 page = grab_cache_page(mapping, index);
 107                 if (!page)
 108                         continue;
 109                 if (PageUptodate(page)) {
 110                         mark_page_accessed(page);
 111                         f2fs_put_page(page, 1);
 112                         continue;
 113                 }
 114                 f2fs_submit_page_mbio(sbi, page, index, &fio);
 115                 mark_page_accessed(page);
 116                 f2fs_put_page(page, 0);
 117         }
 118         f2fs_submit_merged_bio(sbi, META, READ);
 119 }
 120
 121 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 122 {
 123         return radix_tree_lookup(&nm_i->nat_root, n);
 124 }
 125
 126 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 127                 nid_t start, unsigned int nr, struct nat_entry **ep)
 128 {
 129         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 130 }
 131
 132 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 133 {
 134         list_del(&e->list);
 135         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 136         nm_i->nat_cnt--;
 137         kmem_cache_free(nat_entry_slab, e);
 138 }
 139
 140 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 141 {
 142         struct f2fs_nm_info *nm_i = NM_I(sbi);
 143         struct nat_entry *e;
 144         int is_cp = 1;
 145
 146         read_lock(&nm_i->nat_tree_lock);
 147         e = __lookup_nat_cache(nm_i, nid);
 148         if (e && !e->checkpointed)
 149                 is_cp = 0;
 150         read_unlock(&nm_i->nat_tree_lock);
 151         return is_cp;
 152 }
 153
 154 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 155 {
 156         struct nat_entry *new;
 157
 158         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 159         if (!new)
 160                 return NULL;
 161         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 162                 kmem_cache_free(nat_entry_slab, new);
 163                 return NULL;
 164         }
 165         memset(new, 0, sizeof(struct nat_entry));
 166         nat_set_nid(new, nid);
 167         list_add_tail(&new->list, &nm_i->nat_entries);
 168         nm_i->nat_cnt++;
 169         return new;
 170 }
 171
 172 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 173                                                 struct f2fs_nat_entry *ne)
 174 {
 175         struct nat_entry *e;
 176 retry:
 177         write_lock(&nm_i->nat_tree_lock);
 178         e = __lookup_nat_cache(nm_i, nid);
 179         if (!e) {
 180                 e = grab_nat_entry(nm_i, nid);
 181                 if (!e) {
 182                         write_unlock(&nm_i->nat_tree_lock);
 183                         goto retry;
 184                 }
 185                 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
 186                 nat_set_ino(e, le32_to_cpu(ne->ino));
 187                 nat_set_version(e, ne->version);
 188                 e->checkpointed = true;
 189         }
 190         write_unlock(&nm_i->nat_tree_lock);
 191 }
 192
 193 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 194                         block_t new_blkaddr)
 195 {
 196         struct f2fs_nm_info *nm_i = NM_I(sbi);
 197         struct nat_entry *e;
 198 retry:
 199         write_lock(&nm_i->nat_tree_lock);
 200         e = __lookup_nat_cache(nm_i, ni->nid);
 201         if (!e) {
 202                 e = grab_nat_entry(nm_i, ni->nid);
 203                 if (!e) {
 204                         write_unlock(&nm_i->nat_tree_lock);
 205                         goto retry;
 206                 }
 207                 e->ni = *ni;
 208                 e->checkpointed = true;
 209                 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
 210         } else if (new_blkaddr == NEW_ADDR) {
 211                 /*
 212                  * when nid is reallocated,
 213                  * previous nat entry can be remained in nat cache.
 214                  * So, reinitialize it with new information.
 215                  */
 216                 e->ni = *ni;
 217                 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
 218         }
 219
 220         if (new_blkaddr == NEW_ADDR)
 221                 e->checkpointed = false;
 222
 223         /* sanity check */
 224         f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
 225         f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
 226                         new_blkaddr == NULL_ADDR);
 227         f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
 228                         new_blkaddr == NEW_ADDR);
 229         f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
 230                         nat_get_blkaddr(e) != NULL_ADDR &&
 231                         new_blkaddr == NEW_ADDR);
 232
 233         /* increament version no as node is removed */
 234         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 235                 unsigned char version = nat_get_version(e);
 236                 nat_set_version(e, inc_node_version(version));
 237         }
 238
 239         /* change address */
 240         nat_set_blkaddr(e, new_blkaddr);
 241         __set_nat_cache_dirty(nm_i, e);
 242         write_unlock(&nm_i->nat_tree_lock);
 243 }
 244
 245 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 246 {
 247         struct f2fs_nm_info *nm_i = NM_I(sbi);
 248
 249         if (nm_i->nat_cnt <= NM_WOUT_THRESHOLD)
 250                 return 0;
 251
 252         write_lock(&nm_i->nat_tree_lock);
 253         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 254                 struct nat_entry *ne;
 255                 ne = list_first_entry(&nm_i->nat_entries,
 256                                         struct nat_entry, list);
 257                 __del_from_nat_cache(nm_i, ne);
 258                 nr_shrink--;
 259         }
 260         write_unlock(&nm_i->nat_tree_lock);
 261         return nr_shrink;
 262 }
 263
 264 /*
 265  * This function returns always success
 266  */
 267 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 268 {
 269         struct f2fs_nm_info *nm_i = NM_I(sbi);
 270         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 271         struct f2fs_summary_block *sum = curseg->sum_blk;
 272         nid_t start_nid = START_NID(nid);
 273         struct f2fs_nat_block *nat_blk;
 274         struct page *page = NULL;
 275         struct f2fs_nat_entry ne;
 276         struct nat_entry *e;
 277         int i;
 278
 279         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 280         ni->nid = nid;
 281
 282         /* Check nat cache */
 283         read_lock(&nm_i->nat_tree_lock);
 284         e = __lookup_nat_cache(nm_i, nid);
 285         if (e) {
 286                 ni->ino = nat_get_ino(e);
 287                 ni->blk_addr = nat_get_blkaddr(e);
 288                 ni->version = nat_get_version(e);
 289         }
 290         read_unlock(&nm_i->nat_tree_lock);
 291         if (e)
 292                 return;
 293
 294         /* Check current segment summary */
 295         mutex_lock(&curseg->curseg_mutex);
 296         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 297         if (i >= 0) {
 298                 ne = nat_in_journal(sum, i);
 299                 node_info_from_raw_nat(ni, &ne);
 300         }
 301         mutex_unlock(&curseg->curseg_mutex);
 302         if (i >= 0)
 303                 goto cache;
 304
 305         /* Fill node_info from nat page */
 306         page = get_current_nat_page(sbi, start_nid);
 307         nat_blk = (struct f2fs_nat_block *)page_address(page);
 308         ne = nat_blk->entries[nid - start_nid];
 309         node_info_from_raw_nat(ni, &ne);
 310         f2fs_put_page(page, 1);
 311 cache:
 312         /* cache nat entry */
 313         cache_nat_entry(NM_I(sbi), nid, &ne);
 314 }
 315
 316 /*
 317  * The maximum depth is four.
 318  * Offset[0] will have raw inode offset.
 319  */
 320 static int get_node_path(struct f2fs_inode_info *fi, long block,
 321                                 int offset[4], unsigned int noffset[4])
 322 {
 323         const long direct_index = ADDRS_PER_INODE(fi);
 324         const long direct_blks = ADDRS_PER_BLOCK;
 325         const long dptrs_per_blk = NIDS_PER_BLOCK;
 326         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 327         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 328         int n = 0;
 329         int level = 0;
 330
 331         noffset[0] = 0;
 332
 333         if (block < direct_index) {
 334                 offset[n] = block;
 335                 goto got;
 336         }
 337         block -= direct_index;
 338         if (block < direct_blks) {
 339                 offset[n++] = NODE_DIR1_BLOCK;
 340                 noffset[n] = 1;
 341                 offset[n] = block;
 342                 level = 1;
 343                 goto got;
 344         }
 345         block -= direct_blks;
 346         if (block < direct_blks) {
 347                 offset[n++] = NODE_DIR2_BLOCK;
 348                 noffset[n] = 2;
 349                 offset[n] = block;
 350                 level = 1;
 351                 goto got;
 352         }
 353         block -= direct_blks;
 354         if (block < indirect_blks) {
 355                 offset[n++] = NODE_IND1_BLOCK;
 356                 noffset[n] = 3;
 357                 offset[n++] = block / direct_blks;
 358                 noffset[n] = 4 + offset[n - 1];
 359                 offset[n] = block % direct_blks;
 360                 level = 2;
 361                 goto got;
 362         }
 363         block -= indirect_blks;
 364         if (block < indirect_blks) {
 365                 offset[n++] = NODE_IND2_BLOCK;
 366                 noffset[n] = 4 + dptrs_per_blk;
 367                 offset[n++] = block / direct_blks;
 368                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 369                 offset[n] = block % direct_blks;
 370                 level = 2;
 371                 goto got;
 372         }
 373         block -= indirect_blks;
 374         if (block < dindirect_blks) {
 375                 offset[n++] = NODE_DIND_BLOCK;
 376                 noffset[n] = 5 + (dptrs_per_blk * 2);
 377                 offset[n++] = block / indirect_blks;
 378                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 379                               offset[n - 1] * (dptrs_per_blk + 1);
 380                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 381                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 382                               offset[n - 2] * (dptrs_per_blk + 1) +
 383                               offset[n - 1];
 384                 offset[n] = block % direct_blks;
 385                 level = 3;
 386                 goto got;
 387         } else {
 388                 BUG();
 389         }
 390 got:
 391         return level;
 392 }
 393
 394 /*
 395  * Caller should call f2fs_put_dnode(dn).
 396  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 397  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 398  * In the case of RDONLY_NODE, we don't need to care about mutex.
 399  */
 400 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 401 {
 402         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 403         struct page *npage[4];
 404         struct page *parent;
 405         int offset[4];
 406         unsigned int noffset[4];
 407         nid_t nids[4];
 408         int level, i;
 409         int err = 0;
 410
 411         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 412
 413         nids[0] = dn->inode->i_ino;
 414         npage[0] = dn->inode_page;
 415
 416         if (!npage[0]) {
 417                 npage[0] = get_node_page(sbi, nids[0]);
 418                 if (IS_ERR(npage[0]))
 419                         return PTR_ERR(npage[0]);
 420         }
 421         parent = npage[0];
 422         if (level != 0)
 423                 nids[1] = get_nid(parent, offset[0], true);
 424         dn->inode_page = npage[0];
 425         dn->inode_page_locked = true;
 426
 427         /* get indirect or direct nodes */
 428         for (i = 1; i <= level; i++) {
 429                 bool done = false;
 430
 431                 if (!nids[i] && mode == ALLOC_NODE) {
 432                         /* alloc new node */
 433                         if (!alloc_nid(sbi, &(nids[i]))) {
 434                                 err = -ENOSPC;
 435                                 goto release_pages;
 436                         }
 437
 438                         dn->nid = nids[i];
 439                         npage[i] = new_node_page(dn, noffset[i], NULL);
 440                         if (IS_ERR(npage[i])) {
 441                                 alloc_nid_failed(sbi, nids[i]);
 442                                 err = PTR_ERR(npage[i]);
 443                                 goto release_pages;
 444                         }
 445
 446                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 447                         alloc_nid_done(sbi, nids[i]);
 448                         done = true;
 449                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 450                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 451                         if (IS_ERR(npage[i])) {
 452                                 err = PTR_ERR(npage[i]);
 453                                 goto release_pages;
 454                         }
 455                         done = true;
 456                 }
 457                 if (i == 1) {
 458                         dn->inode_page_locked = false;
 459                         unlock_page(parent);
 460                 } else {
 461                         f2fs_put_page(parent, 1);
 462                 }
 463
 464                 if (!done) {
 465                         npage[i] = get_node_page(sbi, nids[i]);
 466                         if (IS_ERR(npage[i])) {
 467                                 err = PTR_ERR(npage[i]);
 468                                 f2fs_put_page(npage[0], 0);
 469                                 goto release_out;
 470                         }
 471                 }
 472                 if (i < level) {
 473                         parent = npage[i];
 474                         nids[i + 1] = get_nid(parent, offset[i], false);
 475                 }
 476         }
 477         dn->nid = nids[level];
 478         dn->ofs_in_node = offset[level];
 479         dn->node_page = npage[level];
 480         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 481         return 0;
 482
 483 release_pages:
 484         f2fs_put_page(parent, 1);
 485         if (i > 1)
 486                 f2fs_put_page(npage[0], 0);
 487 release_out:
 488         dn->inode_page = NULL;
 489         dn->node_page = NULL;
 490         return err;
 491 }
 492
 493 static void truncate_node(struct dnode_of_data *dn)
 494 {
 495         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 496         struct node_info ni;
 497
 498         get_node_info(sbi, dn->nid, &ni);
 499         if (dn->inode->i_blocks == 0) {
 500                 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
 501                 goto invalidate;
 502         }
 503         f2fs_bug_on(ni.blk_addr == NULL_ADDR);
 504
 505         /* Deallocate node address */
 506         invalidate_blocks(sbi, ni.blk_addr);
 507         dec_valid_node_count(sbi, dn->inode);
 508         set_node_addr(sbi, &ni, NULL_ADDR);
 509
 510         if (dn->nid == dn->inode->i_ino) {
 511                 remove_orphan_inode(sbi, dn->nid);
 512                 dec_valid_inode_count(sbi);
 513         } else {
 514                 sync_inode_page(dn);
 515         }
 516 invalidate:
 517         clear_node_page_dirty(dn->node_page);
 518         F2FS_SET_SB_DIRT(sbi);
 519
 520         f2fs_put_page(dn->node_page, 1);
 521         dn->node_page = NULL;
 522         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 523 }
 524
 525 static int truncate_dnode(struct dnode_of_data *dn)
 526 {
 527         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 528         struct page *page;
 529
 530         if (dn->nid == 0)
 531                 return 1;
 532
 533         /* get direct node */
 534         page = get_node_page(sbi, dn->nid);
 535         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 536                 return 1;
 537         else if (IS_ERR(page))
 538                 return PTR_ERR(page);
 539
 540         /* Make dnode_of_data for parameter */
 541         dn->node_page = page;
 542         dn->ofs_in_node = 0;
 543         truncate_data_blocks(dn);
 544         truncate_node(dn);
 545         return 1;
 546 }
 547
 548 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 549                                                 int ofs, int depth)
 550 {
 551         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 552         struct dnode_of_data rdn = *dn;
 553         struct page *page;
 554         struct f2fs_node *rn;
 555         nid_t child_nid;
 556         unsigned int child_nofs;
 557         int freed = 0;
 558         int i, ret;
 559
 560         if (dn->nid == 0)
 561                 return NIDS_PER_BLOCK + 1;
 562
 563         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 564
 565         page = get_node_page(sbi, dn->nid);
 566         if (IS_ERR(page)) {
 567                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 568                 return PTR_ERR(page);
 569         }
 570
 571         rn = F2FS_NODE(page);
 572         if (depth < 3) {
 573                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 574                         child_nid = le32_to_cpu(rn->in.nid[i]);
 575                         if (child_nid == 0)
 576                                 continue;
 577                         rdn.nid = child_nid;
 578                         ret = truncate_dnode(&rdn);
 579                         if (ret < 0)
 580                                 goto out_err;
 581                         set_nid(page, i, 0, false);
 582                 }
 583         } else {
 584                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 585                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 586                         child_nid = le32_to_cpu(rn->in.nid[i]);
 587                         if (child_nid == 0) {
 588                                 child_nofs += NIDS_PER_BLOCK + 1;
 589                                 continue;
 590                         }
 591                         rdn.nid = child_nid;
 592                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 593                         if (ret == (NIDS_PER_BLOCK + 1)) {
 594                                 set_nid(page, i, 0, false);
 595                                 child_nofs += ret;
 596                         } else if (ret < 0 && ret != -ENOENT) {
 597                                 goto out_err;
 598                         }
 599                 }
 600                 freed = child_nofs;
 601         }
 602
 603         if (!ofs) {
 604                 /* remove current indirect node */
 605                 dn->node_page = page;
 606                 truncate_node(dn);
 607                 freed++;
 608         } else {
 609                 f2fs_put_page(page, 1);
 610         }
 611         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 612         return freed;
 613
 614 out_err:
 615         f2fs_put_page(page, 1);
 616         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 617         return ret;
 618 }
 619
 620 static int truncate_partial_nodes(struct dnode_of_data *dn,
 621                         struct f2fs_inode *ri, int *offset, int depth)
 622 {
 623         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 624         struct page *pages[2];
 625         nid_t nid[3];
 626         nid_t child_nid;
 627         int err = 0;
 628         int i;
 629         int idx = depth - 2;
 630
 631         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 632         if (!nid[0])
 633                 return 0;
 634
 635         /* get indirect nodes in the path */
 636         for (i = 0; i < idx + 1; i++) {
 637                 /* refernece count'll be increased */
 638                 pages[i] = get_node_page(sbi, nid[i]);
 639                 if (IS_ERR(pages[i])) {
 640                         err = PTR_ERR(pages[i]);
 641                         idx = i - 1;
 642                         goto fail;
 643                 }
 644                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 645         }
 646
 647         /* free direct nodes linked to a partial indirect node */
 648         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 649                 child_nid = get_nid(pages[idx], i, false);
 650                 if (!child_nid)
 651                         continue;
 652                 dn->nid = child_nid;
 653                 err = truncate_dnode(dn);
 654                 if (err < 0)
 655                         goto fail;
 656                 set_nid(pages[idx], i, 0, false);
 657         }
 658
 659         if (offset[idx + 1] == 0) {
 660                 dn->node_page = pages[idx];
 661                 dn->nid = nid[idx];
 662                 truncate_node(dn);
 663         } else {
 664                 f2fs_put_page(pages[idx], 1);
 665         }
 666         offset[idx]++;
 667         offset[idx + 1] = 0;
 668         idx--;
 669 fail:
 670         for (i = idx; i >= 0; i--)
 671                 f2fs_put_page(pages[i], 1);
 672
 673         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 674
 675         return err;
 676 }
 677
 678 /*
 679  * All the block addresses of data and nodes should be nullified.
 680  */
 681 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 682 {
 683         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 684         int err = 0, cont = 1;
 685         int level, offset[4], noffset[4];
 686         unsigned int nofs = 0;
 687         struct f2fs_inode *ri;
 688         struct dnode_of_data dn;
 689         struct page *page;
 690
 691         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 692
 693         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 694 restart:
 695         page = get_node_page(sbi, inode->i_ino);
 696         if (IS_ERR(page)) {
 697                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 698                 return PTR_ERR(page);
 699         }
 700
 701         set_new_dnode(&dn, inode, page, NULL, 0);
 702         unlock_page(page);
 703
 704         ri = F2FS_INODE(page);
 705         switch (level) {
 706         case 0:
 707         case 1:
 708                 nofs = noffset[1];
 709                 break;
 710         case 2:
 711                 nofs = noffset[1];
 712                 if (!offset[level - 1])
 713                         goto skip_partial;
 714                 err = truncate_partial_nodes(&dn, ri, offset, level);
 715                 if (err < 0 && err != -ENOENT)
 716                         goto fail;
 717                 nofs += 1 + NIDS_PER_BLOCK;
 718                 break;
 719         case 3:
 720                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 721                 if (!offset[level - 1])
 722                         goto skip_partial;
 723                 err = truncate_partial_nodes(&dn, ri, offset, level);
 724                 if (err < 0 && err != -ENOENT)
 725                         goto fail;
 726                 break;
 727         default:
 728                 BUG();
 729         }
 730
 731 skip_partial:
 732         while (cont) {
 733                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 734                 switch (offset[0]) {
 735                 case NODE_DIR1_BLOCK:
 736                 case NODE_DIR2_BLOCK:
 737                         err = truncate_dnode(&dn);
 738                         break;
 739
 740                 case NODE_IND1_BLOCK:
 741                 case NODE_IND2_BLOCK:
 742                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 743                         break;
 744
 745                 case NODE_DIND_BLOCK:
 746                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 747                         cont = 0;
 748                         break;
 749
 750                 default:
 751                         BUG();
 752                 }
 753                 if (err < 0 && err != -ENOENT)
 754                         goto fail;
 755                 if (offset[1] == 0 &&
 756                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 757                         lock_page(page);
 758                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 759                                 f2fs_put_page(page, 1);
 760                                 goto restart;
 761                         }
 762                         wait_on_page_writeback(page);
 763                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 764                         set_page_dirty(page);
 765                         unlock_page(page);
 766                 }
 767                 offset[1] = 0;
 768                 offset[0]++;
 769                 nofs += err;
 770         }
 771 fail:
 772         f2fs_put_page(page, 0);
 773         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 774         return err > 0 ? 0 : err;
 775 }
 776
 777 int truncate_xattr_node(struct inode *inode, struct page *page)
 778 {
 779         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 780         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 781         struct dnode_of_data dn;
 782         struct page *npage;
 783
 784         if (!nid)
 785                 return 0;
 786
 787         npage = get_node_page(sbi, nid);
 788         if (IS_ERR(npage))
 789                 return PTR_ERR(npage);
 790
 791         F2FS_I(inode)->i_xattr_nid = 0;
 792
 793         /* need to do checkpoint during fsync */
 794         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 795
 796         set_new_dnode(&dn, inode, page, npage, nid);
 797
 798         if (page)
 799                 dn.inode_page_locked = true;
 800         truncate_node(&dn);
 801         return 0;
 802 }
 803
 804 /*
 805  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 806  * f2fs_unlock_op().
 807  */
 808 void remove_inode_page(struct inode *inode)
 809 {
 810         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 811         struct page *page;
 812         nid_t ino = inode->i_ino;
 813         struct dnode_of_data dn;
 814
 815         page = get_node_page(sbi, ino);
 816         if (IS_ERR(page))
 817                 return;
 818
 819         if (truncate_xattr_node(inode, page)) {
 820                 f2fs_put_page(page, 1);
 821                 return;
 822         }
 823         /* 0 is possible, after f2fs_new_inode() is failed */
 824         f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
 825         set_new_dnode(&dn, inode, page, page, ino);
 826         truncate_node(&dn);
 827 }
 828
 829 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
 830 {
 831         struct dnode_of_data dn;
 832
 833         /* allocate inode page for new inode */
 834         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 835
 836         /* caller should f2fs_put_page(page, 1); */
 837         return new_node_page(&dn, 0, NULL);
 838 }
 839
 840 struct page *new_node_page(struct dnode_of_data *dn,
 841                                 unsigned int ofs, struct page *ipage)
 842 {
 843         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 844         struct node_info old_ni, new_ni;
 845         struct page *page;
 846         int err;
 847
 848         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 849                 return ERR_PTR(-EPERM);
 850
 851         page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
 852         if (!page)
 853                 return ERR_PTR(-ENOMEM);
 854
 855         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 856                 err = -ENOSPC;
 857                 goto fail;
 858         }
 859
 860         get_node_info(sbi, dn->nid, &old_ni);
 861
 862         /* Reinitialize old_ni with new node page */
 863         f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
 864         new_ni = old_ni;
 865         new_ni.ino = dn->inode->i_ino;
 866         set_node_addr(sbi, &new_ni, NEW_ADDR);
 867
 868         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 869         set_cold_node(dn->inode, page);
 870         SetPageUptodate(page);
 871         set_page_dirty(page);
 872
 873         if (ofs == XATTR_NODE_OFFSET)
 874                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 875
 876         dn->node_page = page;
 877         if (ipage)
 878                 update_inode(dn->inode, ipage);
 879         else
 880                 sync_inode_page(dn);
 881         if (ofs == 0)
 882                 inc_valid_inode_count(sbi);
 883
 884         return page;
 885
 886 fail:
 887         clear_node_page_dirty(page);
 888         f2fs_put_page(page, 1);
 889         return ERR_PTR(err);
 890 }
 891
 892 /*
 893  * Caller should do after getting the following values.
 894  * 0: f2fs_put_page(page, 0)
 895  * LOCKED_PAGE: f2fs_put_page(page, 1)
 896  * error: nothing
 897  */
 898 static int read_node_page(struct page *page, int rw)
 899 {
 900         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 901         struct node_info ni;
 902
 903         get_node_info(sbi, page->index, &ni);
 904
 905         if (unlikely(ni.blk_addr == NULL_ADDR)) {
 906                 f2fs_put_page(page, 1);
 907                 return -ENOENT;
 908         }
 909
 910         if (PageUptodate(page))
 911                 return LOCKED_PAGE;
 912
 913         return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
 914 }
 915
 916 /*
 917  * Readahead a node page
 918  */
 919 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 920 {
 921         struct page *apage;
 922         int err;
 923
 924         apage = find_get_page(NODE_MAPPING(sbi), nid);
 925         if (apage && PageUptodate(apage)) {
 926                 f2fs_put_page(apage, 0);
 927                 return;
 928         }
 929         f2fs_put_page(apage, 0);
 930
 931         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
 932         if (!apage)
 933                 return;
 934
 935         err = read_node_page(apage, READA);
 936         if (err == 0)
 937                 f2fs_put_page(apage, 0);
 938         else if (err == LOCKED_PAGE)
 939                 f2fs_put_page(apage, 1);
 940 }
 941
 942 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 943 {
 944         struct page *page;
 945         int err;
 946 repeat:
 947         page = grab_cache_page(NODE_MAPPING(sbi), nid);
 948         if (!page)
 949                 return ERR_PTR(-ENOMEM);
 950
 951         err = read_node_page(page, READ_SYNC);
 952         if (err < 0)
 953                 return ERR_PTR(err);
 954         else if (err == LOCKED_PAGE)
 955                 goto got_it;
 956
 957         lock_page(page);
 958         if (unlikely(!PageUptodate(page))) {
 959                 f2fs_put_page(page, 1);
 960                 return ERR_PTR(-EIO);
 961         }
 962         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 963                 f2fs_put_page(page, 1);
 964                 goto repeat;
 965         }
 966 got_it:
 967         f2fs_bug_on(nid != nid_of_node(page));
 968         mark_page_accessed(page);
 969         return page;
 970 }
 971
 972 /*
 973  * Return a locked page for the desired node page.
 974  * And, readahead MAX_RA_NODE number of node pages.
 975  */
 976 struct page *get_node_page_ra(struct page *parent, int start)
 977 {
 978         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 979         struct blk_plug plug;
 980         struct page *page;
 981         int err, i, end;
 982         nid_t nid;
 983
 984         /* First, try getting the desired direct node. */
 985         nid = get_nid(parent, start, false);
 986         if (!nid)
 987                 return ERR_PTR(-ENOENT);
 988 repeat:
 989         page = grab_cache_page(NODE_MAPPING(sbi), nid);
 990         if (!page)
 991                 return ERR_PTR(-ENOMEM);
 992
 993         err = read_node_page(page, READ_SYNC);
 994         if (err < 0)
 995                 return ERR_PTR(err);
 996         else if (err == LOCKED_PAGE)
 997                 goto page_hit;
 998
 999         blk_start_plug(&plug);
1000
1001         /* Then, try readahead for siblings of the desired node */
1002         end = start + MAX_RA_NODE;
1003         end = min(end, NIDS_PER_BLOCK);
1004         for (i = start + 1; i < end; i++) {
1005                 nid = get_nid(parent, i, false);
1006                 if (!nid)
1007                         continue;
1008                 ra_node_page(sbi, nid);
1009         }
1010
1011         blk_finish_plug(&plug);
1012
1013         lock_page(page);
1014         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1015                 f2fs_put_page(page, 1);
1016                 goto repeat;
1017         }
1018 page_hit:
1019         if (unlikely(!PageUptodate(page))) {
1020                 f2fs_put_page(page, 1);
1021                 return ERR_PTR(-EIO);
1022         }
1023         mark_page_accessed(page);
1024         return page;
1025 }
1026
1027 void sync_inode_page(struct dnode_of_data *dn)
1028 {
1029         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1030                 update_inode(dn->inode, dn->node_page);
1031         } else if (dn->inode_page) {
1032                 if (!dn->inode_page_locked)
1033                         lock_page(dn->inode_page);
1034                 update_inode(dn->inode, dn->inode_page);
1035                 if (!dn->inode_page_locked)
1036                         unlock_page(dn->inode_page);
1037         } else {
1038                 update_inode_page(dn->inode);
1039         }
1040 }
1041
1042 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1043                                         struct writeback_control *wbc)
1044 {
1045         pgoff_t index, end;
1046         struct pagevec pvec;
1047         int step = ino ? 2 : 0;
1048         int nwritten = 0, wrote = 0;
1049
1050         pagevec_init(&pvec, 0);
1051
1052 next_step:
1053         index = 0;
1054         end = LONG_MAX;
1055
1056         while (index <= end) {
1057                 int i, nr_pages;
1058                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1059                                 PAGECACHE_TAG_DIRTY,
1060                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1061                 if (nr_pages == 0)
1062                         break;
1063
1064                 for (i = 0; i < nr_pages; i++) {
1065                         struct page *page = pvec.pages[i];
1066
1067                         /*
1068                          * flushing sequence with step:
1069                          * 0. indirect nodes
1070                          * 1. dentry dnodes
1071                          * 2. file dnodes
1072                          */
1073                         if (step == 0 && IS_DNODE(page))
1074                                 continue;
1075                         if (step == 1 && (!IS_DNODE(page) ||
1076                                                 is_cold_node(page)))
1077                                 continue;
1078                         if (step == 2 && (!IS_DNODE(page) ||
1079                                                 !is_cold_node(page)))
1080                                 continue;
1081
1082                         /*
1083                          * If an fsync mode,
1084                          * we should not skip writing node pages.
1085                          */
1086                         if (ino && ino_of_node(page) == ino)
1087                                 lock_page(page);
1088                         else if (!trylock_page(page))
1089                                 continue;
1090
1091                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1092 continue_unlock:
1093                                 unlock_page(page);
1094                                 continue;
1095                         }
1096                         if (ino && ino_of_node(page) != ino)
1097                                 goto continue_unlock;
1098
1099                         if (!PageDirty(page)) {
1100                                 /* someone wrote it for us */
1101                                 goto continue_unlock;
1102                         }
1103
1104                         if (!clear_page_dirty_for_io(page))
1105                                 goto continue_unlock;
1106
1107                         /* called by fsync() */
1108                         if (ino && IS_DNODE(page)) {
1109                                 int mark = !is_checkpointed_node(sbi, ino);
1110                                 set_fsync_mark(page, 1);
1111                                 if (IS_INODE(page))
1112                                         set_dentry_mark(page, mark);
1113                                 nwritten++;
1114                         } else {
1115                                 set_fsync_mark(page, 0);
1116                                 set_dentry_mark(page, 0);
1117                         }
1118                         NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1119                         wrote++;
1120
1121                         if (--wbc->nr_to_write == 0)
1122                                 break;
1123                 }
1124                 pagevec_release(&pvec);
1125                 cond_resched();
1126
1127                 if (wbc->nr_to_write == 0) {
1128                         step = 2;
1129                         break;
1130                 }
1131         }
1132
1133         if (step < 2) {
1134                 step++;
1135                 goto next_step;
1136         }
1137
1138         if (wrote)
1139                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1140         return nwritten;
1141 }
1142
1143 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1144 {
1145         pgoff_t index = 0, end = LONG_MAX;
1146         struct pagevec pvec;
1147         int ret2 = 0, ret = 0;
1148
1149         pagevec_init(&pvec, 0);
1150
1151         while (index <= end) {
1152                 int i, nr_pages;
1153                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1154                                 PAGECACHE_TAG_WRITEBACK,
1155                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1156                 if (nr_pages == 0)
1157                         break;
1158
1159                 for (i = 0; i < nr_pages; i++) {
1160                         struct page *page = pvec.pages[i];
1161
1162                         /* until radix tree lookup accepts end_index */
1163                         if (unlikely(page->index > end))
1164                                 continue;
1165
1166                         if (ino && ino_of_node(page) == ino) {
1167                                 wait_on_page_writeback(page);
1168                                 if (TestClearPageError(page))
1169                                         ret = -EIO;
1170                         }
1171                 }
1172                 pagevec_release(&pvec);
1173                 cond_resched();
1174         }
1175
1176         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1177                 ret2 = -ENOSPC;
1178         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1179                 ret2 = -EIO;
1180         if (!ret)
1181                 ret = ret2;
1182         return ret;
1183 }
1184
1185 static int f2fs_write_node_page(struct page *page,
1186                                 struct writeback_control *wbc)
1187 {
1188         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1189         nid_t nid;
1190         block_t new_addr;
1191         struct node_info ni;
1192         struct f2fs_io_info fio = {
1193                 .type = NODE,
1194                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1195         };
1196
1197         if (unlikely(sbi->por_doing))
1198                 goto redirty_out;
1199
1200         wait_on_page_writeback(page);
1201
1202         /* get old block addr of this node page */
1203         nid = nid_of_node(page);
1204         f2fs_bug_on(page->index != nid);
1205
1206         get_node_info(sbi, nid, &ni);
1207
1208         /* This page is already truncated */
1209         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1210                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1211                 unlock_page(page);
1212                 return 0;
1213         }
1214
1215         if (wbc->for_reclaim)
1216                 goto redirty_out;
1217
1218         mutex_lock(&sbi->node_write);
1219         set_page_writeback(page);
1220         write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1221         set_node_addr(sbi, &ni, new_addr);
1222         dec_page_count(sbi, F2FS_DIRTY_NODES);
1223         mutex_unlock(&sbi->node_write);
1224         unlock_page(page);
1225         return 0;
1226
1227 redirty_out:
1228         dec_page_count(sbi, F2FS_DIRTY_NODES);
1229         wbc->pages_skipped++;
1230         set_page_dirty(page);
1231         return AOP_WRITEPAGE_ACTIVATE;
1232 }
1233
1234 /*
1235  * It is very important to gather dirty pages and write at once, so that we can
1236  * submit a big bio without interfering other data writes.
1237  * Be default, 512 pages (2MB) * 3 node types, is more reasonable.
1238  */
1239 #define COLLECT_DIRTY_NODES     1536
1240 static int f2fs_write_node_pages(struct address_space *mapping,
1241                             struct writeback_control *wbc)
1242 {
1243         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1244         long nr_to_write = wbc->nr_to_write;
1245
1246         /* balancing f2fs's metadata in background */
1247         f2fs_balance_fs_bg(sbi);
1248
1249         /* collect a number of dirty node pages and write together */
1250         if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1251                 return 0;
1252
1253         /* if mounting is failed, skip writing node pages */
1254         wbc->nr_to_write = 3 * max_hw_blocks(sbi);
1255         wbc->sync_mode = WB_SYNC_NONE;
1256         sync_node_pages(sbi, 0, wbc);
1257         wbc->nr_to_write = nr_to_write - (3 * max_hw_blocks(sbi) -
1258                                                 wbc->nr_to_write);
1259         return 0;
1260 }
1261
1262 static int f2fs_set_node_page_dirty(struct page *page)
1263 {
1264         struct address_space *mapping = page->mapping;
1265         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1266
1267         trace_f2fs_set_page_dirty(page, NODE);
1268
1269         SetPageUptodate(page);
1270         if (!PageDirty(page)) {
1271                 __set_page_dirty_nobuffers(page);
1272                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1273                 SetPagePrivate(page);
1274                 return 1;
1275         }
1276         return 0;
1277 }
1278
1279 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1280                                       unsigned int length)
1281 {
1282         struct inode *inode = page->mapping->host;
1283         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1284         if (PageDirty(page))
1285                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1286         ClearPagePrivate(page);
1287 }
1288
1289 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1290 {
1291         ClearPagePrivate(page);
1292         return 1;
1293 }
1294
1295 /*
1296  * Structure of the f2fs node operations
1297  */
1298 const struct address_space_operations f2fs_node_aops = {
1299         .writepage      = f2fs_write_node_page,
1300         .writepages     = f2fs_write_node_pages,
1301         .set_page_dirty = f2fs_set_node_page_dirty,
1302         .invalidatepage = f2fs_invalidate_node_page,
1303         .releasepage    = f2fs_release_node_page,
1304 };
1305
1306 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1307 {
1308         struct list_head *this;
1309         struct free_nid *i;
1310         list_for_each(this, head) {
1311                 i = list_entry(this, struct free_nid, list);
1312                 if (i->nid == n)
1313                         return i;
1314         }
1315         return NULL;
1316 }
1317
1318 static void __del_from_free_nid_list(struct free_nid *i)
1319 {
1320         list_del(&i->list);
1321         kmem_cache_free(free_nid_slab, i);
1322 }
1323
1324 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build)
1325 {
1326         struct free_nid *i;
1327         struct nat_entry *ne;
1328         bool allocated = false;
1329
1330         if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1331                 return -1;
1332
1333         /* 0 nid should not be used */
1334         if (unlikely(nid == 0))
1335                 return 0;
1336
1337         if (build) {
1338                 /* do not add allocated nids */
1339                 read_lock(&nm_i->nat_tree_lock);
1340                 ne = __lookup_nat_cache(nm_i, nid);
1341                 if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1342                         allocated = true;
1343                 read_unlock(&nm_i->nat_tree_lock);
1344                 if (allocated)
1345                         return 0;
1346         }
1347
1348         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1349         i->nid = nid;
1350         i->state = NID_NEW;
1351
1352         spin_lock(&nm_i->free_nid_list_lock);
1353         if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1354                 spin_unlock(&nm_i->free_nid_list_lock);
1355                 kmem_cache_free(free_nid_slab, i);
1356                 return 0;
1357         }
1358         list_add_tail(&i->list, &nm_i->free_nid_list);
1359         nm_i->fcnt++;
1360         spin_unlock(&nm_i->free_nid_list_lock);
1361         return 1;
1362 }
1363
1364 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1365 {
1366         struct free_nid *i;
1367         spin_lock(&nm_i->free_nid_list_lock);
1368         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1369         if (i && i->state == NID_NEW) {
1370                 __del_from_free_nid_list(i);
1371                 nm_i->fcnt--;
1372         }
1373         spin_unlock(&nm_i->free_nid_list_lock);
1374 }
1375
1376 static void scan_nat_page(struct f2fs_nm_info *nm_i,
1377                         struct page *nat_page, nid_t start_nid)
1378 {
1379         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1380         block_t blk_addr;
1381         int i;
1382
1383         i = start_nid % NAT_ENTRY_PER_BLOCK;
1384
1385         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1386
1387                 if (unlikely(start_nid >= nm_i->max_nid))
1388                         break;
1389
1390                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1391                 f2fs_bug_on(blk_addr == NEW_ADDR);
1392                 if (blk_addr == NULL_ADDR) {
1393                         if (add_free_nid(nm_i, start_nid, true) < 0)
1394                                 break;
1395                 }
1396         }
1397 }
1398
1399 static void build_free_nids(struct f2fs_sb_info *sbi)
1400 {
1401         struct f2fs_nm_info *nm_i = NM_I(sbi);
1402         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1403         struct f2fs_summary_block *sum = curseg->sum_blk;
1404         int i = 0;
1405         nid_t nid = nm_i->next_scan_nid;
1406
1407         /* Enough entries */
1408         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1409                 return;
1410
1411         /* readahead nat pages to be scanned */
1412         ra_nat_pages(sbi, nid);
1413
1414         while (1) {
1415                 struct page *page = get_current_nat_page(sbi, nid);
1416
1417                 scan_nat_page(nm_i, page, nid);
1418                 f2fs_put_page(page, 1);
1419
1420                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1421                 if (unlikely(nid >= nm_i->max_nid))
1422                         nid = 0;
1423
1424                 if (i++ == FREE_NID_PAGES)
1425                         break;
1426         }
1427
1428         /* go to the next free nat pages to find free nids abundantly */
1429         nm_i->next_scan_nid = nid;
1430
1431         /* find free nids from current sum_pages */
1432         mutex_lock(&curseg->curseg_mutex);
1433         for (i = 0; i < nats_in_cursum(sum); i++) {
1434                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1435                 nid = le32_to_cpu(nid_in_journal(sum, i));
1436                 if (addr == NULL_ADDR)
1437                         add_free_nid(nm_i, nid, true);
1438                 else
1439                         remove_free_nid(nm_i, nid);
1440         }
1441         mutex_unlock(&curseg->curseg_mutex);
1442 }
1443
1444 /*
1445  * If this function returns success, caller can obtain a new nid
1446  * from second parameter of this function.
1447  * The returned nid could be used ino as well as nid when inode is created.
1448  */
1449 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1450 {
1451         struct f2fs_nm_info *nm_i = NM_I(sbi);
1452         struct free_nid *i = NULL;
1453         struct list_head *this;
1454 retry:
1455         if (unlikely(sbi->total_valid_node_count + 1 >= nm_i->max_nid))
1456                 return false;
1457
1458         spin_lock(&nm_i->free_nid_list_lock);
1459
1460         /* We should not use stale free nids created by build_free_nids */
1461         if (nm_i->fcnt && !sbi->on_build_free_nids) {
1462                 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1463                 list_for_each(this, &nm_i->free_nid_list) {
1464                         i = list_entry(this, struct free_nid, list);
1465                         if (i->state == NID_NEW)
1466                                 break;
1467                 }
1468
1469                 f2fs_bug_on(i->state != NID_NEW);
1470                 *nid = i->nid;
1471                 i->state = NID_ALLOC;
1472                 nm_i->fcnt--;
1473                 spin_unlock(&nm_i->free_nid_list_lock);
1474                 return true;
1475         }
1476         spin_unlock(&nm_i->free_nid_list_lock);
1477
1478         /* Let's scan nat pages and its caches to get free nids */
1479         mutex_lock(&nm_i->build_lock);
1480         sbi->on_build_free_nids = true;
1481         build_free_nids(sbi);
1482         sbi->on_build_free_nids = false;
1483         mutex_unlock(&nm_i->build_lock);
1484         goto retry;
1485 }
1486
1487 /*
1488  * alloc_nid() should be called prior to this function.
1489  */
1490 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1491 {
1492         struct f2fs_nm_info *nm_i = NM_I(sbi);
1493         struct free_nid *i;
1494
1495         spin_lock(&nm_i->free_nid_list_lock);
1496         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1497         f2fs_bug_on(!i || i->state != NID_ALLOC);
1498         __del_from_free_nid_list(i);
1499         spin_unlock(&nm_i->free_nid_list_lock);
1500 }
1501
1502 /*
1503  * alloc_nid() should be called prior to this function.
1504  */
1505 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1506 {
1507         struct f2fs_nm_info *nm_i = NM_I(sbi);
1508         struct free_nid *i;
1509
1510         if (!nid)
1511                 return;
1512
1513         spin_lock(&nm_i->free_nid_list_lock);
1514         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1515         f2fs_bug_on(!i || i->state != NID_ALLOC);
1516         if (nm_i->fcnt > 2 * MAX_FREE_NIDS) {
1517                 __del_from_free_nid_list(i);
1518         } else {
1519                 i->state = NID_NEW;
1520                 nm_i->fcnt++;
1521         }
1522         spin_unlock(&nm_i->free_nid_list_lock);
1523 }
1524
1525 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1526                 struct f2fs_summary *sum, struct node_info *ni,
1527                 block_t new_blkaddr)
1528 {
1529         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1530         set_node_addr(sbi, ni, new_blkaddr);
1531         clear_node_page_dirty(page);
1532 }
1533
1534 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1535 {
1536         struct f2fs_inode *src, *dst;
1537         nid_t ino = ino_of_node(page);
1538         struct node_info old_ni, new_ni;
1539         struct page *ipage;
1540
1541         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1542         if (!ipage)
1543                 return -ENOMEM;
1544
1545         /* Should not use this inode  from free nid list */
1546         remove_free_nid(NM_I(sbi), ino);
1547
1548         get_node_info(sbi, ino, &old_ni);
1549         SetPageUptodate(ipage);
1550         fill_node_footer(ipage, ino, ino, 0, true);
1551
1552         src = F2FS_INODE(page);
1553         dst = F2FS_INODE(ipage);
1554
1555         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1556         dst->i_size = 0;
1557         dst->i_blocks = cpu_to_le64(1);
1558         dst->i_links = cpu_to_le32(1);
1559         dst->i_xattr_nid = 0;
1560
1561         new_ni = old_ni;
1562         new_ni.ino = ino;
1563
1564         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1565                 WARN_ON(1);
1566         set_node_addr(sbi, &new_ni, NEW_ADDR);
1567         inc_valid_inode_count(sbi);
1568         f2fs_put_page(ipage, 1);
1569         return 0;
1570 }
1571
1572 /*
1573  * ra_sum_pages() merge contiguous pages into one bio and submit.
1574  * these pre-readed pages are linked in pages list.
1575  */
1576 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
1577                                 int start, int nrpages)
1578 {
1579         struct page *page;
1580         int page_idx = start;
1581         struct f2fs_io_info fio = {
1582                 .type = META,
1583                 .rw = READ_SYNC | REQ_META | REQ_PRIO
1584         };
1585
1586         for (; page_idx < start + nrpages; page_idx++) {
1587                 /* alloc temporal page for read node summary info*/
1588                 page = alloc_page(GFP_F2FS_ZERO);
1589                 if (!page) {
1590                         struct page *tmp;
1591                         list_for_each_entry_safe(page, tmp, pages, lru) {
1592                                 list_del(&page->lru);
1593                                 unlock_page(page);
1594                                 __free_pages(page, 0);
1595                         }
1596                         return -ENOMEM;
1597                 }
1598
1599                 lock_page(page);
1600                 page->index = page_idx;
1601                 list_add_tail(&page->lru, pages);
1602         }
1603
1604         list_for_each_entry(page, pages, lru)
1605                 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1606
1607         f2fs_submit_merged_bio(sbi, META, READ);
1608         return 0;
1609 }
1610
1611 int restore_node_summary(struct f2fs_sb_info *sbi,
1612                         unsigned int segno, struct f2fs_summary_block *sum)
1613 {
1614         struct f2fs_node *rn;
1615         struct f2fs_summary *sum_entry;
1616         struct page *page, *tmp;
1617         block_t addr;
1618         int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1619         int i, last_offset, nrpages, err = 0;
1620         LIST_HEAD(page_list);
1621
1622         /* scan the node segment */
1623         last_offset = sbi->blocks_per_seg;
1624         addr = START_BLOCK(sbi, segno);
1625         sum_entry = &sum->entries[0];
1626
1627         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1628                 nrpages = min(last_offset - i, bio_blocks);
1629
1630                 /* read ahead node pages */
1631                 err = ra_sum_pages(sbi, &page_list, addr, nrpages);
1632                 if (err)
1633                         return err;
1634
1635                 list_for_each_entry_safe(page, tmp, &page_list, lru) {
1636
1637                         lock_page(page);
1638                         if (unlikely(!PageUptodate(page))) {
1639                                 err = -EIO;
1640                         } else {
1641                                 rn = F2FS_NODE(page);
1642                                 sum_entry->nid = rn->footer.nid;
1643                                 sum_entry->version = 0;
1644                                 sum_entry->ofs_in_node = 0;
1645                                 sum_entry++;
1646                         }
1647
1648                         list_del(&page->lru);
1649                         unlock_page(page);
1650                         __free_pages(page, 0);
1651                 }
1652         }
1653         return err;
1654 }
1655
1656 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1657 {
1658         struct f2fs_nm_info *nm_i = NM_I(sbi);
1659         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1660         struct f2fs_summary_block *sum = curseg->sum_blk;
1661         int i;
1662
1663         mutex_lock(&curseg->curseg_mutex);
1664
1665         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1666                 mutex_unlock(&curseg->curseg_mutex);
1667                 return false;
1668         }
1669
1670         for (i = 0; i < nats_in_cursum(sum); i++) {
1671                 struct nat_entry *ne;
1672                 struct f2fs_nat_entry raw_ne;
1673                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1674
1675                 raw_ne = nat_in_journal(sum, i);
1676 retry:
1677                 write_lock(&nm_i->nat_tree_lock);
1678                 ne = __lookup_nat_cache(nm_i, nid);
1679                 if (ne) {
1680                         __set_nat_cache_dirty(nm_i, ne);
1681                         write_unlock(&nm_i->nat_tree_lock);
1682                         continue;
1683                 }
1684                 ne = grab_nat_entry(nm_i, nid);
1685                 if (!ne) {
1686                         write_unlock(&nm_i->nat_tree_lock);
1687                         goto retry;
1688                 }
1689                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1690                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1691                 nat_set_version(ne, raw_ne.version);
1692                 __set_nat_cache_dirty(nm_i, ne);
1693                 write_unlock(&nm_i->nat_tree_lock);
1694         }
1695         update_nats_in_cursum(sum, -i);
1696         mutex_unlock(&curseg->curseg_mutex);
1697         return true;
1698 }
1699
1700 /*
1701  * This function is called during the checkpointing process.
1702  */
1703 void flush_nat_entries(struct f2fs_sb_info *sbi)
1704 {
1705         struct f2fs_nm_info *nm_i = NM_I(sbi);
1706         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1707         struct f2fs_summary_block *sum = curseg->sum_blk;
1708         struct list_head *cur, *n;
1709         struct page *page = NULL;
1710         struct f2fs_nat_block *nat_blk = NULL;
1711         nid_t start_nid = 0, end_nid = 0;
1712         bool flushed;
1713
1714         flushed = flush_nats_in_journal(sbi);
1715
1716         if (!flushed)
1717                 mutex_lock(&curseg->curseg_mutex);
1718
1719         /* 1) flush dirty nat caches */
1720         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1721                 struct nat_entry *ne;
1722                 nid_t nid;
1723                 struct f2fs_nat_entry raw_ne;
1724                 int offset = -1;
1725                 block_t new_blkaddr;
1726
1727                 ne = list_entry(cur, struct nat_entry, list);
1728                 nid = nat_get_nid(ne);
1729
1730                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1731                         continue;
1732                 if (flushed)
1733                         goto to_nat_page;
1734
1735                 /* if there is room for nat enries in curseg->sumpage */
1736                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1737                 if (offset >= 0) {
1738                         raw_ne = nat_in_journal(sum, offset);
1739                         goto flush_now;
1740                 }
1741 to_nat_page:
1742                 if (!page || (start_nid > nid || nid > end_nid)) {
1743                         if (page) {
1744                                 f2fs_put_page(page, 1);
1745                                 page = NULL;
1746                         }
1747                         start_nid = START_NID(nid);
1748                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1749
1750                         /*
1751                          * get nat block with dirty flag, increased reference
1752                          * count, mapped and lock
1753                          */
1754                         page = get_next_nat_page(sbi, start_nid);
1755                         nat_blk = page_address(page);
1756                 }
1757
1758                 f2fs_bug_on(!nat_blk);
1759                 raw_ne = nat_blk->entries[nid - start_nid];
1760 flush_now:
1761                 new_blkaddr = nat_get_blkaddr(ne);
1762
1763                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1764                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1765                 raw_ne.version = nat_get_version(ne);
1766
1767                 if (offset < 0) {
1768                         nat_blk->entries[nid - start_nid] = raw_ne;
1769                 } else {
1770                         nat_in_journal(sum, offset) = raw_ne;
1771                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1772                 }
1773
1774                 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1775                                 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1776                         write_lock(&nm_i->nat_tree_lock);
1777                         __del_from_nat_cache(nm_i, ne);
1778                         write_unlock(&nm_i->nat_tree_lock);
1779                 } else {
1780                         write_lock(&nm_i->nat_tree_lock);
1781                         __clear_nat_cache_dirty(nm_i, ne);
1782                         ne->checkpointed = true;
1783                         write_unlock(&nm_i->nat_tree_lock);
1784                 }
1785         }
1786         if (!flushed)
1787                 mutex_unlock(&curseg->curseg_mutex);
1788         f2fs_put_page(page, 1);
1789
1790         /* 2) shrink nat caches if necessary */
1791         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1792 }
1793
1794 static int init_node_manager(struct f2fs_sb_info *sbi)
1795 {
1796         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1797         struct f2fs_nm_info *nm_i = NM_I(sbi);
1798         unsigned char *version_bitmap;
1799         unsigned int nat_segs, nat_blocks;
1800
1801         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1802
1803         /* segment_count_nat includes pair segment so divide to 2. */
1804         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1805         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1806         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1807         nm_i->fcnt = 0;
1808         nm_i->nat_cnt = 0;
1809
1810         INIT_LIST_HEAD(&nm_i->free_nid_list);
1811         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1812         INIT_LIST_HEAD(&nm_i->nat_entries);
1813         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1814
1815         mutex_init(&nm_i->build_lock);
1816         spin_lock_init(&nm_i->free_nid_list_lock);
1817         rwlock_init(&nm_i->nat_tree_lock);
1818
1819         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1820         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1821         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1822         if (!version_bitmap)
1823                 return -EFAULT;
1824
1825         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1826                                         GFP_KERNEL);
1827         if (!nm_i->nat_bitmap)
1828                 return -ENOMEM;
1829         return 0;
1830 }
1831
1832 int build_node_manager(struct f2fs_sb_info *sbi)
1833 {
1834         int err;
1835
1836         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1837         if (!sbi->nm_info)
1838                 return -ENOMEM;
1839
1840         err = init_node_manager(sbi);
1841         if (err)
1842                 return err;
1843
1844         build_free_nids(sbi);
1845         return 0;
1846 }
1847
1848 void destroy_node_manager(struct f2fs_sb_info *sbi)
1849 {
1850         struct f2fs_nm_info *nm_i = NM_I(sbi);
1851         struct free_nid *i, *next_i;
1852         struct nat_entry *natvec[NATVEC_SIZE];
1853         nid_t nid = 0;
1854         unsigned int found;
1855
1856         if (!nm_i)
1857                 return;
1858
1859         /* destroy free nid list */
1860         spin_lock(&nm_i->free_nid_list_lock);
1861         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1862                 f2fs_bug_on(i->state == NID_ALLOC);
1863                 __del_from_free_nid_list(i);
1864                 nm_i->fcnt--;
1865         }
1866         f2fs_bug_on(nm_i->fcnt);
1867         spin_unlock(&nm_i->free_nid_list_lock);
1868
1869         /* destroy nat cache */
1870         write_lock(&nm_i->nat_tree_lock);
1871         while ((found = __gang_lookup_nat_cache(nm_i,
1872                                         nid, NATVEC_SIZE, natvec))) {
1873                 unsigned idx;
1874                 for (idx = 0; idx < found; idx++) {
1875                         struct nat_entry *e = natvec[idx];
1876                         nid = nat_get_nid(e) + 1;
1877                         __del_from_nat_cache(nm_i, e);
1878                 }
1879         }
1880         f2fs_bug_on(nm_i->nat_cnt);
1881         write_unlock(&nm_i->nat_tree_lock);
1882
1883         kfree(nm_i->nat_bitmap);
1884         sbi->nm_info = NULL;
1885         kfree(nm_i);
1886 }
1887
1888 int __init create_node_manager_caches(void)
1889 {
1890         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1891                         sizeof(struct nat_entry), NULL);
1892         if (!nat_entry_slab)
1893                 return -ENOMEM;
1894
1895         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1896                         sizeof(struct free_nid), NULL);
1897         if (!free_nid_slab) {
1898                 kmem_cache_destroy(nat_entry_slab);
1899                 return -ENOMEM;
1900         }
1901         return 0;
1902 }
1903
1904 void destroy_node_manager_caches(void)
1905 {
1906         kmem_cache_destroy(free_nid_slab);
1907         kmem_cache_destroy(nat_entry_slab);
1908 }