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 page *page;
  93         pgoff_t index;
  94         int i;
  95         struct f2fs_io_info fio = {
  96                 .type = META,
  97                 .rw = READ_SYNC,
  98                 .rw_flag = REQ_META | REQ_PRIO
  99         };
 100
 101
 102         for (i = 0; i < FREE_NID_PAGES; i++, nid += NAT_ENTRY_PER_BLOCK) {
 103                 if (unlikely(nid >= nm_i->max_nid))
 104                         nid = 0;
 105                 index = current_nat_addr(sbi, nid);
 106
 107                 page = grab_cache_page(mapping, index);
 108                 if (!page)
 109                         continue;
 110                 if (PageUptodate(page)) {
 111                         mark_page_accessed(page);
 112                         f2fs_put_page(page, 1);
 113                         continue;
 114                 }
 115                 f2fs_submit_page_mbio(sbi, page, index, &fio);
 116                 mark_page_accessed(page);
 117                 f2fs_put_page(page, 0);
 118         }
 119         f2fs_submit_merged_bio(sbi, META, READ);
 120 }
 121
 122 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 123 {
 124         return radix_tree_lookup(&nm_i->nat_root, n);
 125 }
 126
 127 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 128                 nid_t start, unsigned int nr, struct nat_entry **ep)
 129 {
 130         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 131 }
 132
 133 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 134 {
 135         list_del(&e->list);
 136         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 137         nm_i->nat_cnt--;
 138         kmem_cache_free(nat_entry_slab, e);
 139 }
 140
 141 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 142 {
 143         struct f2fs_nm_info *nm_i = NM_I(sbi);
 144         struct nat_entry *e;
 145         int is_cp = 1;
 146
 147         read_lock(&nm_i->nat_tree_lock);
 148         e = __lookup_nat_cache(nm_i, nid);
 149         if (e && !e->checkpointed)
 150                 is_cp = 0;
 151         read_unlock(&nm_i->nat_tree_lock);
 152         return is_cp;
 153 }
 154
 155 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 156 {
 157         struct nat_entry *new;
 158
 159         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 160         if (!new)
 161                 return NULL;
 162         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 163                 kmem_cache_free(nat_entry_slab, new);
 164                 return NULL;
 165         }
 166         memset(new, 0, sizeof(struct nat_entry));
 167         nat_set_nid(new, nid);
 168         list_add_tail(&new->list, &nm_i->nat_entries);
 169         nm_i->nat_cnt++;
 170         return new;
 171 }
 172
 173 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 174                                                 struct f2fs_nat_entry *ne)
 175 {
 176         struct nat_entry *e;
 177 retry:
 178         write_lock(&nm_i->nat_tree_lock);
 179         e = __lookup_nat_cache(nm_i, nid);
 180         if (!e) {
 181                 e = grab_nat_entry(nm_i, nid);
 182                 if (!e) {
 183                         write_unlock(&nm_i->nat_tree_lock);
 184                         goto retry;
 185                 }
 186                 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
 187                 nat_set_ino(e, le32_to_cpu(ne->ino));
 188                 nat_set_version(e, ne->version);
 189                 e->checkpointed = true;
 190         }
 191         write_unlock(&nm_i->nat_tree_lock);
 192 }
 193
 194 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 195                         block_t new_blkaddr)
 196 {
 197         struct f2fs_nm_info *nm_i = NM_I(sbi);
 198         struct nat_entry *e;
 199 retry:
 200         write_lock(&nm_i->nat_tree_lock);
 201         e = __lookup_nat_cache(nm_i, ni->nid);
 202         if (!e) {
 203                 e = grab_nat_entry(nm_i, ni->nid);
 204                 if (!e) {
 205                         write_unlock(&nm_i->nat_tree_lock);
 206                         goto retry;
 207                 }
 208                 e->ni = *ni;
 209                 e->checkpointed = true;
 210                 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
 211         } else if (new_blkaddr == NEW_ADDR) {
 212                 /*
 213                  * when nid is reallocated,
 214                  * previous nat entry can be remained in nat cache.
 215                  * So, reinitialize it with new information.
 216                  */
 217                 e->ni = *ni;
 218                 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
 219         }
 220
 221         if (new_blkaddr == NEW_ADDR)
 222                 e->checkpointed = false;
 223
 224         /* sanity check */
 225         f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
 226         f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
 227                         new_blkaddr == NULL_ADDR);
 228         f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
 229                         new_blkaddr == NEW_ADDR);
 230         f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
 231                         nat_get_blkaddr(e) != NULL_ADDR &&
 232                         new_blkaddr == NEW_ADDR);
 233
 234         /* increament version no as node is removed */
 235         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 236                 unsigned char version = nat_get_version(e);
 237                 nat_set_version(e, inc_node_version(version));
 238         }
 239
 240         /* change address */
 241         nat_set_blkaddr(e, new_blkaddr);
 242         __set_nat_cache_dirty(nm_i, e);
 243         write_unlock(&nm_i->nat_tree_lock);
 244 }
 245
 246 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 247 {
 248         struct f2fs_nm_info *nm_i = NM_I(sbi);
 249
 250         if (nm_i->nat_cnt <= NM_WOUT_THRESHOLD)
 251                 return 0;
 252
 253         write_lock(&nm_i->nat_tree_lock);
 254         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 255                 struct nat_entry *ne;
 256                 ne = list_first_entry(&nm_i->nat_entries,
 257                                         struct nat_entry, list);
 258                 __del_from_nat_cache(nm_i, ne);
 259                 nr_shrink--;
 260         }
 261         write_unlock(&nm_i->nat_tree_lock);
 262         return nr_shrink;
 263 }
 264
 265 /*
 266  * This function returns always success
 267  */
 268 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 269 {
 270         struct f2fs_nm_info *nm_i = NM_I(sbi);
 271         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 272         struct f2fs_summary_block *sum = curseg->sum_blk;
 273         nid_t start_nid = START_NID(nid);
 274         struct f2fs_nat_block *nat_blk;
 275         struct page *page = NULL;
 276         struct f2fs_nat_entry ne;
 277         struct nat_entry *e;
 278         int i;
 279
 280         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 281         ni->nid = nid;
 282
 283         /* Check nat cache */
 284         read_lock(&nm_i->nat_tree_lock);
 285         e = __lookup_nat_cache(nm_i, nid);
 286         if (e) {
 287                 ni->ino = nat_get_ino(e);
 288                 ni->blk_addr = nat_get_blkaddr(e);
 289                 ni->version = nat_get_version(e);
 290         }
 291         read_unlock(&nm_i->nat_tree_lock);
 292         if (e)
 293                 return;
 294
 295         /* Check current segment summary */
 296         mutex_lock(&curseg->curseg_mutex);
 297         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 298         if (i >= 0) {
 299                 ne = nat_in_journal(sum, i);
 300                 node_info_from_raw_nat(ni, &ne);
 301         }
 302         mutex_unlock(&curseg->curseg_mutex);
 303         if (i >= 0)
 304                 goto cache;
 305
 306         /* Fill node_info from nat page */
 307         page = get_current_nat_page(sbi, start_nid);
 308         nat_blk = (struct f2fs_nat_block *)page_address(page);
 309         ne = nat_blk->entries[nid - start_nid];
 310         node_info_from_raw_nat(ni, &ne);
 311         f2fs_put_page(page, 1);
 312 cache:
 313         /* cache nat entry */
 314         cache_nat_entry(NM_I(sbi), nid, &ne);
 315 }
 316
 317 /*
 318  * The maximum depth is four.
 319  * Offset[0] will have raw inode offset.
 320  */
 321 static int get_node_path(struct f2fs_inode_info *fi, long block,
 322                                 int offset[4], unsigned int noffset[4])
 323 {
 324         const long direct_index = ADDRS_PER_INODE(fi);
 325         const long direct_blks = ADDRS_PER_BLOCK;
 326         const long dptrs_per_blk = NIDS_PER_BLOCK;
 327         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 328         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 329         int n = 0;
 330         int level = 0;
 331
 332         noffset[0] = 0;
 333
 334         if (block < direct_index) {
 335                 offset[n] = block;
 336                 goto got;
 337         }
 338         block -= direct_index;
 339         if (block < direct_blks) {
 340                 offset[n++] = NODE_DIR1_BLOCK;
 341                 noffset[n] = 1;
 342                 offset[n] = block;
 343                 level = 1;
 344                 goto got;
 345         }
 346         block -= direct_blks;
 347         if (block < direct_blks) {
 348                 offset[n++] = NODE_DIR2_BLOCK;
 349                 noffset[n] = 2;
 350                 offset[n] = block;
 351                 level = 1;
 352                 goto got;
 353         }
 354         block -= direct_blks;
 355         if (block < indirect_blks) {
 356                 offset[n++] = NODE_IND1_BLOCK;
 357                 noffset[n] = 3;
 358                 offset[n++] = block / direct_blks;
 359                 noffset[n] = 4 + offset[n - 1];
 360                 offset[n] = block % direct_blks;
 361                 level = 2;
 362                 goto got;
 363         }
 364         block -= indirect_blks;
 365         if (block < indirect_blks) {
 366                 offset[n++] = NODE_IND2_BLOCK;
 367                 noffset[n] = 4 + dptrs_per_blk;
 368                 offset[n++] = block / direct_blks;
 369                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 370                 offset[n] = block % direct_blks;
 371                 level = 2;
 372                 goto got;
 373         }
 374         block -= indirect_blks;
 375         if (block < dindirect_blks) {
 376                 offset[n++] = NODE_DIND_BLOCK;
 377                 noffset[n] = 5 + (dptrs_per_blk * 2);
 378                 offset[n++] = block / indirect_blks;
 379                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 380                               offset[n - 1] * (dptrs_per_blk + 1);
 381                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 382                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 383                               offset[n - 2] * (dptrs_per_blk + 1) +
 384                               offset[n - 1];
 385                 offset[n] = block % direct_blks;
 386                 level = 3;
 387                 goto got;
 388         } else {
 389                 BUG();
 390         }
 391 got:
 392         return level;
 393 }
 394
 395 /*
 396  * Caller should call f2fs_put_dnode(dn).
 397  * Also, it should grab and release a mutex by calling mutex_lock_op() and
 398  * mutex_unlock_op() only if ro is not set RDONLY_NODE.
 399  * In the case of RDONLY_NODE, we don't need to care about mutex.
 400  */
 401 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 402 {
 403         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 404         struct page *npage[4];
 405         struct page *parent;
 406         int offset[4];
 407         unsigned int noffset[4];
 408         nid_t nids[4];
 409         int level, i;
 410         int err = 0;
 411
 412         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 413
 414         nids[0] = dn->inode->i_ino;
 415         npage[0] = dn->inode_page;
 416
 417         if (!npage[0]) {
 418                 npage[0] = get_node_page(sbi, nids[0]);
 419                 if (IS_ERR(npage[0]))
 420                         return PTR_ERR(npage[0]);
 421         }
 422         parent = npage[0];
 423         if (level != 0)
 424                 nids[1] = get_nid(parent, offset[0], true);
 425         dn->inode_page = npage[0];
 426         dn->inode_page_locked = true;
 427
 428         /* get indirect or direct nodes */
 429         for (i = 1; i <= level; i++) {
 430                 bool done = false;
 431
 432                 if (!nids[i] && mode == ALLOC_NODE) {
 433                         /* alloc new node */
 434                         if (!alloc_nid(sbi, &(nids[i]))) {
 435                                 err = -ENOSPC;
 436                                 goto release_pages;
 437                         }
 438
 439                         dn->nid = nids[i];
 440                         npage[i] = new_node_page(dn, noffset[i], NULL);
 441                         if (IS_ERR(npage[i])) {
 442                                 alloc_nid_failed(sbi, nids[i]);
 443                                 err = PTR_ERR(npage[i]);
 444                                 goto release_pages;
 445                         }
 446
 447                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 448                         alloc_nid_done(sbi, nids[i]);
 449                         done = true;
 450                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 451                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 452                         if (IS_ERR(npage[i])) {
 453                                 err = PTR_ERR(npage[i]);
 454                                 goto release_pages;
 455                         }
 456                         done = true;
 457                 }
 458                 if (i == 1) {
 459                         dn->inode_page_locked = false;
 460                         unlock_page(parent);
 461                 } else {
 462                         f2fs_put_page(parent, 1);
 463                 }
 464
 465                 if (!done) {
 466                         npage[i] = get_node_page(sbi, nids[i]);
 467                         if (IS_ERR(npage[i])) {
 468                                 err = PTR_ERR(npage[i]);
 469                                 f2fs_put_page(npage[0], 0);
 470                                 goto release_out;
 471                         }
 472                 }
 473                 if (i < level) {
 474                         parent = npage[i];
 475                         nids[i + 1] = get_nid(parent, offset[i], false);
 476                 }
 477         }
 478         dn->nid = nids[level];
 479         dn->ofs_in_node = offset[level];
 480         dn->node_page = npage[level];
 481         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 482         return 0;
 483
 484 release_pages:
 485         f2fs_put_page(parent, 1);
 486         if (i > 1)
 487                 f2fs_put_page(npage[0], 0);
 488 release_out:
 489         dn->inode_page = NULL;
 490         dn->node_page = NULL;
 491         return err;
 492 }
 493
 494 static void truncate_node(struct dnode_of_data *dn)
 495 {
 496         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 497         struct node_info ni;
 498
 499         get_node_info(sbi, dn->nid, &ni);
 500         if (dn->inode->i_blocks == 0) {
 501                 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
 502                 goto invalidate;
 503         }
 504         f2fs_bug_on(ni.blk_addr == NULL_ADDR);
 505
 506         /* Deallocate node address */
 507         invalidate_blocks(sbi, ni.blk_addr);
 508         dec_valid_node_count(sbi, dn->inode);
 509         set_node_addr(sbi, &ni, NULL_ADDR);
 510
 511         if (dn->nid == dn->inode->i_ino) {
 512                 remove_orphan_inode(sbi, dn->nid);
 513                 dec_valid_inode_count(sbi);
 514         } else {
 515                 sync_inode_page(dn);
 516         }
 517 invalidate:
 518         clear_node_page_dirty(dn->node_page);
 519         F2FS_SET_SB_DIRT(sbi);
 520
 521         f2fs_put_page(dn->node_page, 1);
 522         dn->node_page = NULL;
 523         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 524 }
 525
 526 static int truncate_dnode(struct dnode_of_data *dn)
 527 {
 528         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 529         struct page *page;
 530
 531         if (dn->nid == 0)
 532                 return 1;
 533
 534         /* get direct node */
 535         page = get_node_page(sbi, dn->nid);
 536         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 537                 return 1;
 538         else if (IS_ERR(page))
 539                 return PTR_ERR(page);
 540
 541         /* Make dnode_of_data for parameter */
 542         dn->node_page = page;
 543         dn->ofs_in_node = 0;
 544         truncate_data_blocks(dn);
 545         truncate_node(dn);
 546         return 1;
 547 }
 548
 549 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 550                                                 int ofs, int depth)
 551 {
 552         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 553         struct dnode_of_data rdn = *dn;
 554         struct page *page;
 555         struct f2fs_node *rn;
 556         nid_t child_nid;
 557         unsigned int child_nofs;
 558         int freed = 0;
 559         int i, ret;
 560
 561         if (dn->nid == 0)
 562                 return NIDS_PER_BLOCK + 1;
 563
 564         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 565
 566         page = get_node_page(sbi, dn->nid);
 567         if (IS_ERR(page)) {
 568                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 569                 return PTR_ERR(page);
 570         }
 571
 572         rn = F2FS_NODE(page);
 573         if (depth < 3) {
 574                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 575                         child_nid = le32_to_cpu(rn->in.nid[i]);
 576                         if (child_nid == 0)
 577                                 continue;
 578                         rdn.nid = child_nid;
 579                         ret = truncate_dnode(&rdn);
 580                         if (ret < 0)
 581                                 goto out_err;
 582                         set_nid(page, i, 0, false);
 583                 }
 584         } else {
 585                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 586                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 587                         child_nid = le32_to_cpu(rn->in.nid[i]);
 588                         if (child_nid == 0) {
 589                                 child_nofs += NIDS_PER_BLOCK + 1;
 590                                 continue;
 591                         }
 592                         rdn.nid = child_nid;
 593                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 594                         if (ret == (NIDS_PER_BLOCK + 1)) {
 595                                 set_nid(page, i, 0, false);
 596                                 child_nofs += ret;
 597                         } else if (ret < 0 && ret != -ENOENT) {
 598                                 goto out_err;
 599                         }
 600                 }
 601                 freed = child_nofs;
 602         }
 603
 604         if (!ofs) {
 605                 /* remove current indirect node */
 606                 dn->node_page = page;
 607                 truncate_node(dn);
 608                 freed++;
 609         } else {
 610                 f2fs_put_page(page, 1);
 611         }
 612         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 613         return freed;
 614
 615 out_err:
 616         f2fs_put_page(page, 1);
 617         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 618         return ret;
 619 }
 620
 621 static int truncate_partial_nodes(struct dnode_of_data *dn,
 622                         struct f2fs_inode *ri, int *offset, int depth)
 623 {
 624         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 625         struct page *pages[2];
 626         nid_t nid[3];
 627         nid_t child_nid;
 628         int err = 0;
 629         int i;
 630         int idx = depth - 2;
 631
 632         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 633         if (!nid[0])
 634                 return 0;
 635
 636         /* get indirect nodes in the path */
 637         for (i = 0; i < depth - 1; i++) {
 638                 /* refernece count'll be increased */
 639                 pages[i] = get_node_page(sbi, nid[i]);
 640                 if (IS_ERR(pages[i])) {
 641                         depth = i + 1;
 642                         err = PTR_ERR(pages[i]);
 643                         goto fail;
 644                 }
 645                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 646         }
 647
 648         /* free direct nodes linked to a partial indirect node */
 649         for (i = offset[depth - 1]; i < NIDS_PER_BLOCK; i++) {
 650                 child_nid = get_nid(pages[idx], i, false);
 651                 if (!child_nid)
 652                         continue;
 653                 dn->nid = child_nid;
 654                 err = truncate_dnode(dn);
 655                 if (err < 0)
 656                         goto fail;
 657                 set_nid(pages[idx], i, 0, false);
 658         }
 659
 660         if (offset[depth - 1] == 0) {
 661                 dn->node_page = pages[idx];
 662                 dn->nid = nid[idx];
 663                 truncate_node(dn);
 664         } else {
 665                 f2fs_put_page(pages[idx], 1);
 666         }
 667         offset[idx]++;
 668         offset[depth - 1] = 0;
 669 fail:
 670         for (i = depth - 3; 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         struct address_space *node_mapping = sbi->node_inode->i_mapping;
 685         int err = 0, cont = 1;
 686         int level, offset[4], noffset[4];
 687         unsigned int nofs = 0;
 688         struct f2fs_node *rn;
 689         struct dnode_of_data dn;
 690         struct page *page;
 691
 692         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 693
 694         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 695 restart:
 696         page = get_node_page(sbi, inode->i_ino);
 697         if (IS_ERR(page)) {
 698                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 699                 return PTR_ERR(page);
 700         }
 701
 702         set_new_dnode(&dn, inode, page, NULL, 0);
 703         unlock_page(page);
 704
 705         rn = F2FS_NODE(page);
 706         switch (level) {
 707         case 0:
 708         case 1:
 709                 nofs = noffset[1];
 710                 break;
 711         case 2:
 712                 nofs = noffset[1];
 713                 if (!offset[level - 1])
 714                         goto skip_partial;
 715                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 716                 if (err < 0 && err != -ENOENT)
 717                         goto fail;
 718                 nofs += 1 + NIDS_PER_BLOCK;
 719                 break;
 720         case 3:
 721                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 722                 if (!offset[level - 1])
 723                         goto skip_partial;
 724                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 725                 if (err < 0 && err != -ENOENT)
 726                         goto fail;
 727                 break;
 728         default:
 729                 BUG();
 730         }
 731
 732 skip_partial:
 733         while (cont) {
 734                 dn.nid = le32_to_cpu(rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]);
 735                 switch (offset[0]) {
 736                 case NODE_DIR1_BLOCK:
 737                 case NODE_DIR2_BLOCK:
 738                         err = truncate_dnode(&dn);
 739                         break;
 740
 741                 case NODE_IND1_BLOCK:
 742                 case NODE_IND2_BLOCK:
 743                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 744                         break;
 745
 746                 case NODE_DIND_BLOCK:
 747                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 748                         cont = 0;
 749                         break;
 750
 751                 default:
 752                         BUG();
 753                 }
 754                 if (err < 0 && err != -ENOENT)
 755                         goto fail;
 756                 if (offset[1] == 0 &&
 757                                 rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 758                         lock_page(page);
 759                         if (unlikely(page->mapping != node_mapping)) {
 760                                 f2fs_put_page(page, 1);
 761                                 goto restart;
 762                         }
 763                         wait_on_page_writeback(page);
 764                         rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 765                         set_page_dirty(page);
 766                         unlock_page(page);
 767                 }
 768                 offset[1] = 0;
 769                 offset[0]++;
 770                 nofs += err;
 771         }
 772 fail:
 773         f2fs_put_page(page, 0);
 774         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 775         return err > 0 ? 0 : err;
 776 }
 777
 778 int truncate_xattr_node(struct inode *inode, struct page *page)
 779 {
 780         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 781         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 782         struct dnode_of_data dn;
 783         struct page *npage;
 784
 785         if (!nid)
 786                 return 0;
 787
 788         npage = get_node_page(sbi, nid);
 789         if (IS_ERR(npage))
 790                 return PTR_ERR(npage);
 791
 792         F2FS_I(inode)->i_xattr_nid = 0;
 793
 794         /* need to do checkpoint during fsync */
 795         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 796
 797         set_new_dnode(&dn, inode, page, npage, nid);
 798
 799         if (page)
 800                 dn.inode_page_locked = true;
 801         truncate_node(&dn);
 802         return 0;
 803 }
 804
 805 /*
 806  * Caller should grab and release a mutex by calling mutex_lock_op() and
 807  * mutex_unlock_op().
 808  */
 809 void remove_inode_page(struct inode *inode)
 810 {
 811         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 812         struct page *page;
 813         nid_t ino = inode->i_ino;
 814         struct dnode_of_data dn;
 815
 816         page = get_node_page(sbi, ino);
 817         if (IS_ERR(page))
 818                 return;
 819
 820         if (truncate_xattr_node(inode, page)) {
 821                 f2fs_put_page(page, 1);
 822                 return;
 823         }
 824         /* 0 is possible, after f2fs_new_inode() is failed */
 825         f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
 826         set_new_dnode(&dn, inode, page, page, ino);
 827         truncate_node(&dn);
 828 }
 829
 830 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
 831 {
 832         struct dnode_of_data dn;
 833
 834         /* allocate inode page for new inode */
 835         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 836
 837         /* caller should f2fs_put_page(page, 1); */
 838         return new_node_page(&dn, 0, NULL);
 839 }
 840
 841 struct page *new_node_page(struct dnode_of_data *dn,
 842                                 unsigned int ofs, struct page *ipage)
 843 {
 844         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 845         struct address_space *mapping = sbi->node_inode->i_mapping;
 846         struct node_info old_ni, new_ni;
 847         struct page *page;
 848         int err;
 849
 850         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 851                 return ERR_PTR(-EPERM);
 852
 853         page = grab_cache_page(mapping, dn->nid);
 854         if (!page)
 855                 return ERR_PTR(-ENOMEM);
 856
 857         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 858                 err = -ENOSPC;
 859                 goto fail;
 860         }
 861
 862         get_node_info(sbi, dn->nid, &old_ni);
 863
 864         /* Reinitialize old_ni with new node page */
 865         f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
 866         new_ni = old_ni;
 867         new_ni.ino = dn->inode->i_ino;
 868         set_node_addr(sbi, &new_ni, NEW_ADDR);
 869
 870         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 871         set_cold_node(dn->inode, page);
 872         SetPageUptodate(page);
 873         set_page_dirty(page);
 874
 875         if (ofs == XATTR_NODE_OFFSET)
 876                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 877
 878         dn->node_page = page;
 879         if (ipage)
 880                 update_inode(dn->inode, ipage);
 881         else
 882                 sync_inode_page(dn);
 883         if (ofs == 0)
 884                 inc_valid_inode_count(sbi);
 885
 886         return page;
 887
 888 fail:
 889         clear_node_page_dirty(page);
 890         f2fs_put_page(page, 1);
 891         return ERR_PTR(err);
 892 }
 893
 894 /*
 895  * Caller should do after getting the following values.
 896  * 0: f2fs_put_page(page, 0)
 897  * LOCKED_PAGE: f2fs_put_page(page, 1)
 898  * error: nothing
 899  */
 900 static int read_node_page(struct page *page, int rw)
 901 {
 902         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 903         struct node_info ni;
 904
 905         get_node_info(sbi, page->index, &ni);
 906
 907         if (unlikely(ni.blk_addr == NULL_ADDR)) {
 908                 f2fs_put_page(page, 1);
 909                 return -ENOENT;
 910         }
 911
 912         if (PageUptodate(page))
 913                 return LOCKED_PAGE;
 914
 915         return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
 916 }
 917
 918 /*
 919  * Readahead a node page
 920  */
 921 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 922 {
 923         struct address_space *mapping = sbi->node_inode->i_mapping;
 924         struct page *apage;
 925         int err;
 926
 927         apage = find_get_page(mapping, nid);
 928         if (apage && PageUptodate(apage)) {
 929                 f2fs_put_page(apage, 0);
 930                 return;
 931         }
 932         f2fs_put_page(apage, 0);
 933
 934         apage = grab_cache_page(mapping, nid);
 935         if (!apage)
 936                 return;
 937
 938         err = read_node_page(apage, READA);
 939         if (err == 0)
 940                 f2fs_put_page(apage, 0);
 941         else if (err == LOCKED_PAGE)
 942                 f2fs_put_page(apage, 1);
 943 }
 944
 945 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 946 {
 947         struct address_space *mapping = sbi->node_inode->i_mapping;
 948         struct page *page;
 949         int err;
 950 repeat:
 951         page = grab_cache_page(mapping, nid);
 952         if (!page)
 953                 return ERR_PTR(-ENOMEM);
 954
 955         err = read_node_page(page, READ_SYNC);
 956         if (err < 0)
 957                 return ERR_PTR(err);
 958         else if (err == LOCKED_PAGE)
 959                 goto got_it;
 960
 961         lock_page(page);
 962         if (unlikely(!PageUptodate(page))) {
 963                 f2fs_put_page(page, 1);
 964                 return ERR_PTR(-EIO);
 965         }
 966         if (unlikely(page->mapping != mapping)) {
 967                 f2fs_put_page(page, 1);
 968                 goto repeat;
 969         }
 970 got_it:
 971         f2fs_bug_on(nid != nid_of_node(page));
 972         mark_page_accessed(page);
 973         return page;
 974 }
 975
 976 /*
 977  * Return a locked page for the desired node page.
 978  * And, readahead MAX_RA_NODE number of node pages.
 979  */
 980 struct page *get_node_page_ra(struct page *parent, int start)
 981 {
 982         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 983         struct address_space *mapping = sbi->node_inode->i_mapping;
 984         struct blk_plug plug;
 985         struct page *page;
 986         int err, i, end;
 987         nid_t nid;
 988
 989         /* First, try getting the desired direct node. */
 990         nid = get_nid(parent, start, false);
 991         if (!nid)
 992                 return ERR_PTR(-ENOENT);
 993 repeat:
 994         page = grab_cache_page(mapping, nid);
 995         if (!page)
 996                 return ERR_PTR(-ENOMEM);
 997
 998         err = read_node_page(page, READ_SYNC);
 999         if (err < 0)
1000                 return ERR_PTR(err);
1001         else if (err == LOCKED_PAGE)
1002                 goto page_hit;
1003
1004         blk_start_plug(&plug);
1005
1006         /* Then, try readahead for siblings of the desired node */
1007         end = start + MAX_RA_NODE;
1008         end = min(end, NIDS_PER_BLOCK);
1009         for (i = start + 1; i < end; i++) {
1010                 nid = get_nid(parent, i, false);
1011                 if (!nid)
1012                         continue;
1013                 ra_node_page(sbi, nid);
1014         }
1015
1016         blk_finish_plug(&plug);
1017
1018         lock_page(page);
1019         if (unlikely(page->mapping != mapping)) {
1020                 f2fs_put_page(page, 1);
1021                 goto repeat;
1022         }
1023 page_hit:
1024         if (unlikely(!PageUptodate(page))) {
1025                 f2fs_put_page(page, 1);
1026                 return ERR_PTR(-EIO);
1027         }
1028         mark_page_accessed(page);
1029         return page;
1030 }
1031
1032 void sync_inode_page(struct dnode_of_data *dn)
1033 {
1034         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1035                 update_inode(dn->inode, dn->node_page);
1036         } else if (dn->inode_page) {
1037                 if (!dn->inode_page_locked)
1038                         lock_page(dn->inode_page);
1039                 update_inode(dn->inode, dn->inode_page);
1040                 if (!dn->inode_page_locked)
1041                         unlock_page(dn->inode_page);
1042         } else {
1043                 update_inode_page(dn->inode);
1044         }
1045 }
1046
1047 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1048                                         struct writeback_control *wbc)
1049 {
1050         struct address_space *mapping = sbi->node_inode->i_mapping;
1051         pgoff_t index, end;
1052         struct pagevec pvec;
1053         int step = ino ? 2 : 0;
1054         int nwritten = 0, wrote = 0;
1055
1056         pagevec_init(&pvec, 0);
1057
1058 next_step:
1059         index = 0;
1060         end = LONG_MAX;
1061
1062         while (index <= end) {
1063                 int i, nr_pages;
1064                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1065                                 PAGECACHE_TAG_DIRTY,
1066                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1067                 if (nr_pages == 0)
1068                         break;
1069
1070                 for (i = 0; i < nr_pages; i++) {
1071                         struct page *page = pvec.pages[i];
1072
1073                         /*
1074                          * flushing sequence with step:
1075                          * 0. indirect nodes
1076                          * 1. dentry dnodes
1077                          * 2. file dnodes
1078                          */
1079                         if (step == 0 && IS_DNODE(page))
1080                                 continue;
1081                         if (step == 1 && (!IS_DNODE(page) ||
1082                                                 is_cold_node(page)))
1083                                 continue;
1084                         if (step == 2 && (!IS_DNODE(page) ||
1085                                                 !is_cold_node(page)))
1086                                 continue;
1087
1088                         /*
1089                          * If an fsync mode,
1090                          * we should not skip writing node pages.
1091                          */
1092                         if (ino && ino_of_node(page) == ino)
1093                                 lock_page(page);
1094                         else if (!trylock_page(page))
1095                                 continue;
1096
1097                         if (unlikely(page->mapping != mapping)) {
1098 continue_unlock:
1099                                 unlock_page(page);
1100                                 continue;
1101                         }
1102                         if (ino && ino_of_node(page) != ino)
1103                                 goto continue_unlock;
1104
1105                         if (!PageDirty(page)) {
1106                                 /* someone wrote it for us */
1107                                 goto continue_unlock;
1108                         }
1109
1110                         if (!clear_page_dirty_for_io(page))
1111                                 goto continue_unlock;
1112
1113                         /* called by fsync() */
1114                         if (ino && IS_DNODE(page)) {
1115                                 int mark = !is_checkpointed_node(sbi, ino);
1116                                 set_fsync_mark(page, 1);
1117                                 if (IS_INODE(page))
1118                                         set_dentry_mark(page, mark);
1119                                 nwritten++;
1120                         } else {
1121                                 set_fsync_mark(page, 0);
1122                                 set_dentry_mark(page, 0);
1123                         }
1124                         mapping->a_ops->writepage(page, wbc);
1125                         wrote++;
1126
1127                         if (--wbc->nr_to_write == 0)
1128                                 break;
1129                 }
1130                 pagevec_release(&pvec);
1131                 cond_resched();
1132
1133                 if (wbc->nr_to_write == 0) {
1134                         step = 2;
1135                         break;
1136                 }
1137         }
1138
1139         if (step < 2) {
1140                 step++;
1141                 goto next_step;
1142         }
1143
1144         if (wrote)
1145                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1146         return nwritten;
1147 }
1148
1149 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1150 {
1151         struct address_space *mapping = sbi->node_inode->i_mapping;
1152         pgoff_t index = 0, end = LONG_MAX;
1153         struct pagevec pvec;
1154         int nr_pages;
1155         int ret2 = 0, ret = 0;
1156
1157         pagevec_init(&pvec, 0);
1158         while ((index <= end) &&
1159                         (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1160                         PAGECACHE_TAG_WRITEBACK,
1161                         min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
1162                 unsigned i;
1163
1164                 for (i = 0; i < nr_pages; i++) {
1165                         struct page *page = pvec.pages[i];
1166
1167                         /* until radix tree lookup accepts end_index */
1168                         if (unlikely(page->index > end))
1169                                 continue;
1170
1171                         if (ino && ino_of_node(page) == ino) {
1172                                 wait_on_page_writeback(page);
1173                                 if (TestClearPageError(page))
1174                                         ret = -EIO;
1175                         }
1176                 }
1177                 pagevec_release(&pvec);
1178                 cond_resched();
1179         }
1180
1181         if (unlikely(test_and_clear_bit(AS_ENOSPC, &mapping->flags)))
1182                 ret2 = -ENOSPC;
1183         if (unlikely(test_and_clear_bit(AS_EIO, &mapping->flags)))
1184                 ret2 = -EIO;
1185         if (!ret)
1186                 ret = ret2;
1187         return ret;
1188 }
1189
1190 static int f2fs_write_node_page(struct page *page,
1191                                 struct writeback_control *wbc)
1192 {
1193         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1194         nid_t nid;
1195         block_t new_addr;
1196         struct node_info ni;
1197
1198         if (unlikely(sbi->por_doing))
1199                 goto redirty_out;
1200
1201         wait_on_page_writeback(page);
1202
1203         /* get old block addr of this node page */
1204         nid = nid_of_node(page);
1205         f2fs_bug_on(page->index != nid);
1206
1207         get_node_info(sbi, nid, &ni);
1208
1209         /* This page is already truncated */
1210         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1211                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1212                 unlock_page(page);
1213                 return 0;
1214         }
1215
1216         if (wbc->for_reclaim)
1217                 goto redirty_out;
1218
1219         mutex_lock(&sbi->node_write);
1220         set_page_writeback(page);
1221         write_node_page(sbi, page, nid, ni.blk_addr, &new_addr);
1222         set_node_addr(sbi, &ni, new_addr);
1223         dec_page_count(sbi, F2FS_DIRTY_NODES);
1224         mutex_unlock(&sbi->node_write);
1225         unlock_page(page);
1226         return 0;
1227
1228 redirty_out:
1229         dec_page_count(sbi, F2FS_DIRTY_NODES);
1230         wbc->pages_skipped++;
1231         set_page_dirty(page);
1232         return AOP_WRITEPAGE_ACTIVATE;
1233 }
1234
1235 /*
1236  * It is very important to gather dirty pages and write at once, so that we can
1237  * submit a big bio without interfering other data writes.
1238  * Be default, 512 pages (2MB) * 3 node types, is more reasonable.
1239  */
1240 #define COLLECT_DIRTY_NODES     1536
1241 static int f2fs_write_node_pages(struct address_space *mapping,
1242                             struct writeback_control *wbc)
1243 {
1244         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1245         long nr_to_write = wbc->nr_to_write;
1246
1247         /* balancing f2fs's metadata in background */
1248         f2fs_balance_fs_bg(sbi);
1249
1250         /* collect a number of dirty node pages and write together */
1251         if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1252                 return 0;
1253
1254         /* if mounting is failed, skip writing node pages */
1255         wbc->nr_to_write = 3 * max_hw_blocks(sbi);
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 address_space *mapping = sbi->node_inode->i_mapping;
1537         struct f2fs_node *src, *dst;
1538         nid_t ino = ino_of_node(page);
1539         struct node_info old_ni, new_ni;
1540         struct page *ipage;
1541
1542         ipage = grab_cache_page(mapping, ino);
1543         if (!ipage)
1544                 return -ENOMEM;
1545
1546         /* Should not use this inode  from free nid list */
1547         remove_free_nid(NM_I(sbi), ino);
1548
1549         get_node_info(sbi, ino, &old_ni);
1550         SetPageUptodate(ipage);
1551         fill_node_footer(ipage, ino, ino, 0, true);
1552
1553         src = F2FS_NODE(page);
1554         dst = F2FS_NODE(ipage);
1555
1556         memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i);
1557         dst->i.i_size = 0;
1558         dst->i.i_blocks = cpu_to_le64(1);
1559         dst->i.i_links = cpu_to_le32(1);
1560         dst->i.i_xattr_nid = 0;
1561
1562         new_ni = old_ni;
1563         new_ni.ino = ino;
1564
1565         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1566                 WARN_ON(1);
1567         set_node_addr(sbi, &new_ni, NEW_ADDR);
1568         inc_valid_inode_count(sbi);
1569         f2fs_put_page(ipage, 1);
1570         return 0;
1571 }
1572
1573 /*
1574  * ra_sum_pages() merge contiguous pages into one bio and submit.
1575  * these pre-readed pages are linked in pages list.
1576  */
1577 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
1578                                 int start, int nrpages)
1579 {
1580         struct page *page;
1581         int page_idx = start;
1582         struct f2fs_io_info fio = {
1583                 .type = META,
1584                 .rw = READ_SYNC,
1585                 .rw_flag = REQ_META | REQ_PRIO
1586         };
1587
1588         for (; page_idx < start + nrpages; page_idx++) {
1589                 /* alloc temporal page for read node summary info*/
1590                 page = alloc_page(GFP_F2FS_ZERO);
1591                 if (!page) {
1592                         struct page *tmp;
1593                         list_for_each_entry_safe(page, tmp, pages, lru) {
1594                                 list_del(&page->lru);
1595                                 unlock_page(page);
1596                                 __free_pages(page, 0);
1597                         }
1598                         return -ENOMEM;
1599                 }
1600
1601                 lock_page(page);
1602                 page->index = page_idx;
1603                 list_add_tail(&page->lru, pages);
1604         }
1605
1606         list_for_each_entry(page, pages, lru)
1607                 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1608
1609         f2fs_submit_merged_bio(sbi, META, READ);
1610         return 0;
1611 }
1612
1613 int restore_node_summary(struct f2fs_sb_info *sbi,
1614                         unsigned int segno, struct f2fs_summary_block *sum)
1615 {
1616         struct f2fs_node *rn;
1617         struct f2fs_summary *sum_entry;
1618         struct page *page, *tmp;
1619         block_t addr;
1620         int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1621         int i, last_offset, nrpages, err = 0;
1622         LIST_HEAD(page_list);
1623
1624         /* scan the node segment */
1625         last_offset = sbi->blocks_per_seg;
1626         addr = START_BLOCK(sbi, segno);
1627         sum_entry = &sum->entries[0];
1628
1629         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1630                 nrpages = min(last_offset - i, bio_blocks);
1631
1632                 /* read ahead node pages */
1633                 err = ra_sum_pages(sbi, &page_list, addr, nrpages);
1634                 if (err)
1635                         return err;
1636
1637                 list_for_each_entry_safe(page, tmp, &page_list, lru) {
1638
1639                         lock_page(page);
1640                         if (unlikely(!PageUptodate(page))) {
1641                                 err = -EIO;
1642                         } else {
1643                                 rn = F2FS_NODE(page);
1644                                 sum_entry->nid = rn->footer.nid;
1645                                 sum_entry->version = 0;
1646                                 sum_entry->ofs_in_node = 0;
1647                                 sum_entry++;
1648                         }
1649
1650                         list_del(&page->lru);
1651                         unlock_page(page);
1652                         __free_pages(page, 0);
1653                 }
1654         }
1655         return err;
1656 }
1657
1658 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1659 {
1660         struct f2fs_nm_info *nm_i = NM_I(sbi);
1661         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1662         struct f2fs_summary_block *sum = curseg->sum_blk;
1663         int i;
1664
1665         mutex_lock(&curseg->curseg_mutex);
1666
1667         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1668                 mutex_unlock(&curseg->curseg_mutex);
1669                 return false;
1670         }
1671
1672         for (i = 0; i < nats_in_cursum(sum); i++) {
1673                 struct nat_entry *ne;
1674                 struct f2fs_nat_entry raw_ne;
1675                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1676
1677                 raw_ne = nat_in_journal(sum, i);
1678 retry:
1679                 write_lock(&nm_i->nat_tree_lock);
1680                 ne = __lookup_nat_cache(nm_i, nid);
1681                 if (ne) {
1682                         __set_nat_cache_dirty(nm_i, ne);
1683                         write_unlock(&nm_i->nat_tree_lock);
1684                         continue;
1685                 }
1686                 ne = grab_nat_entry(nm_i, nid);
1687                 if (!ne) {
1688                         write_unlock(&nm_i->nat_tree_lock);
1689                         goto retry;
1690                 }
1691                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1692                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1693                 nat_set_version(ne, raw_ne.version);
1694                 __set_nat_cache_dirty(nm_i, ne);
1695                 write_unlock(&nm_i->nat_tree_lock);
1696         }
1697         update_nats_in_cursum(sum, -i);
1698         mutex_unlock(&curseg->curseg_mutex);
1699         return true;
1700 }
1701
1702 /*
1703  * This function is called during the checkpointing process.
1704  */
1705 void flush_nat_entries(struct f2fs_sb_info *sbi)
1706 {
1707         struct f2fs_nm_info *nm_i = NM_I(sbi);
1708         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1709         struct f2fs_summary_block *sum = curseg->sum_blk;
1710         struct list_head *cur, *n;
1711         struct page *page = NULL;
1712         struct f2fs_nat_block *nat_blk = NULL;
1713         nid_t start_nid = 0, end_nid = 0;
1714         bool flushed;
1715
1716         flushed = flush_nats_in_journal(sbi);
1717
1718         if (!flushed)
1719                 mutex_lock(&curseg->curseg_mutex);
1720
1721         /* 1) flush dirty nat caches */
1722         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1723                 struct nat_entry *ne;
1724                 nid_t nid;
1725                 struct f2fs_nat_entry raw_ne;
1726                 int offset = -1;
1727                 block_t new_blkaddr;
1728
1729                 ne = list_entry(cur, struct nat_entry, list);
1730                 nid = nat_get_nid(ne);
1731
1732                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1733                         continue;
1734                 if (flushed)
1735                         goto to_nat_page;
1736
1737                 /* if there is room for nat enries in curseg->sumpage */
1738                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1739                 if (offset >= 0) {
1740                         raw_ne = nat_in_journal(sum, offset);
1741                         goto flush_now;
1742                 }
1743 to_nat_page:
1744                 if (!page || (start_nid > nid || nid > end_nid)) {
1745                         if (page) {
1746                                 f2fs_put_page(page, 1);
1747                                 page = NULL;
1748                         }
1749                         start_nid = START_NID(nid);
1750                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1751
1752                         /*
1753                          * get nat block with dirty flag, increased reference
1754                          * count, mapped and lock
1755                          */
1756                         page = get_next_nat_page(sbi, start_nid);
1757                         nat_blk = page_address(page);
1758                 }
1759
1760                 f2fs_bug_on(!nat_blk);
1761                 raw_ne = nat_blk->entries[nid - start_nid];
1762 flush_now:
1763                 new_blkaddr = nat_get_blkaddr(ne);
1764
1765                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1766                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1767                 raw_ne.version = nat_get_version(ne);
1768
1769                 if (offset < 0) {
1770                         nat_blk->entries[nid - start_nid] = raw_ne;
1771                 } else {
1772                         nat_in_journal(sum, offset) = raw_ne;
1773                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1774                 }
1775
1776                 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1777                                 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1778                         write_lock(&nm_i->nat_tree_lock);
1779                         __del_from_nat_cache(nm_i, ne);
1780                         write_unlock(&nm_i->nat_tree_lock);
1781                 } else {
1782                         write_lock(&nm_i->nat_tree_lock);
1783                         __clear_nat_cache_dirty(nm_i, ne);
1784                         ne->checkpointed = true;
1785                         write_unlock(&nm_i->nat_tree_lock);
1786                 }
1787         }
1788         if (!flushed)
1789                 mutex_unlock(&curseg->curseg_mutex);
1790         f2fs_put_page(page, 1);
1791
1792         /* 2) shrink nat caches if necessary */
1793         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1794 }
1795
1796 static int init_node_manager(struct f2fs_sb_info *sbi)
1797 {
1798         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1799         struct f2fs_nm_info *nm_i = NM_I(sbi);
1800         unsigned char *version_bitmap;
1801         unsigned int nat_segs, nat_blocks;
1802
1803         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1804
1805         /* segment_count_nat includes pair segment so divide to 2. */
1806         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1807         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1808         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1809         nm_i->fcnt = 0;
1810         nm_i->nat_cnt = 0;
1811
1812         INIT_LIST_HEAD(&nm_i->free_nid_list);
1813         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1814         INIT_LIST_HEAD(&nm_i->nat_entries);
1815         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1816
1817         mutex_init(&nm_i->build_lock);
1818         spin_lock_init(&nm_i->free_nid_list_lock);
1819         rwlock_init(&nm_i->nat_tree_lock);
1820
1821         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1822         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1823         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1824         if (!version_bitmap)
1825                 return -EFAULT;
1826
1827         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1828                                         GFP_KERNEL);
1829         if (!nm_i->nat_bitmap)
1830                 return -ENOMEM;
1831         return 0;
1832 }
1833
1834 int build_node_manager(struct f2fs_sb_info *sbi)
1835 {
1836         int err;
1837
1838         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1839         if (!sbi->nm_info)
1840                 return -ENOMEM;
1841
1842         err = init_node_manager(sbi);
1843         if (err)
1844                 return err;
1845
1846         build_free_nids(sbi);
1847         return 0;
1848 }
1849
1850 void destroy_node_manager(struct f2fs_sb_info *sbi)
1851 {
1852         struct f2fs_nm_info *nm_i = NM_I(sbi);
1853         struct free_nid *i, *next_i;
1854         struct nat_entry *natvec[NATVEC_SIZE];
1855         nid_t nid = 0;
1856         unsigned int found;
1857
1858         if (!nm_i)
1859                 return;
1860
1861         /* destroy free nid list */
1862         spin_lock(&nm_i->free_nid_list_lock);
1863         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1864                 f2fs_bug_on(i->state == NID_ALLOC);
1865                 __del_from_free_nid_list(i);
1866                 nm_i->fcnt--;
1867         }
1868         f2fs_bug_on(nm_i->fcnt);
1869         spin_unlock(&nm_i->free_nid_list_lock);
1870
1871         /* destroy nat cache */
1872         write_lock(&nm_i->nat_tree_lock);
1873         while ((found = __gang_lookup_nat_cache(nm_i,
1874                                         nid, NATVEC_SIZE, natvec))) {
1875                 unsigned idx;
1876                 for (idx = 0; idx < found; idx++) {
1877                         struct nat_entry *e = natvec[idx];
1878                         nid = nat_get_nid(e) + 1;
1879                         __del_from_nat_cache(nm_i, e);
1880                 }
1881         }
1882         f2fs_bug_on(nm_i->nat_cnt);
1883         write_unlock(&nm_i->nat_tree_lock);
1884
1885         kfree(nm_i->nat_bitmap);
1886         sbi->nm_info = NULL;
1887         kfree(nm_i);
1888 }
1889
1890 int __init create_node_manager_caches(void)
1891 {
1892         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1893                         sizeof(struct nat_entry), NULL);
1894         if (!nat_entry_slab)
1895                 return -ENOMEM;
1896
1897         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1898                         sizeof(struct free_nid), NULL);
1899         if (!free_nid_slab) {
1900                 kmem_cache_destroy(nat_entry_slab);
1901                 return -ENOMEM;
1902         }
1903         return 0;
1904 }
1905
1906 void destroy_node_manager_caches(void)
1907 {
1908         kmem_cache_destroy(free_nid_slab);
1909         kmem_cache_destroy(nat_entry_slab);
1910 }