fs/f2fs/node.c

   1 /*
   2  * fs/f2fs/node.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/mpage.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/swap.h>
  18
  19 #include "f2fs.h"
  20 #include "node.h"
  21 #include "segment.h"
  22 #include <trace/events/f2fs.h>
  23
  24 static struct kmem_cache *nat_entry_slab;
  25 static struct kmem_cache *free_nid_slab;
  26
  27 static void clear_node_page_dirty(struct page *page)
  28 {
  29         struct address_space *mapping = page->mapping;
  30         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
  31         unsigned int long flags;
  32
  33         if (PageDirty(page)) {
  34                 spin_lock_irqsave(&mapping->tree_lock, flags);
  35                 radix_tree_tag_clear(&mapping->page_tree,
  36                                 page_index(page),
  37                                 PAGECACHE_TAG_DIRTY);
  38                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  39
  40                 clear_page_dirty_for_io(page);
  41                 dec_page_count(sbi, F2FS_DIRTY_NODES);
  42         }
  43         ClearPageUptodate(page);
  44 }
  45
  46 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  47 {
  48         pgoff_t index = current_nat_addr(sbi, nid);
  49         return get_meta_page(sbi, index);
  50 }
  51
  52 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  53 {
  54         struct page *src_page;
  55         struct page *dst_page;
  56         pgoff_t src_off;
  57         pgoff_t dst_off;
  58         void *src_addr;
  59         void *dst_addr;
  60         struct f2fs_nm_info *nm_i = NM_I(sbi);
  61
  62         src_off = current_nat_addr(sbi, nid);
  63         dst_off = next_nat_addr(sbi, src_off);
  64
  65         /* get current nat block page with lock */
  66         src_page = get_meta_page(sbi, src_off);
  67
  68         /* Dirty src_page means that it is already the new target NAT page. */
  69         if (PageDirty(src_page))
  70                 return src_page;
  71
  72         dst_page = grab_meta_page(sbi, dst_off);
  73
  74         src_addr = page_address(src_page);
  75         dst_addr = page_address(dst_page);
  76         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
  77         set_page_dirty(dst_page);
  78         f2fs_put_page(src_page, 1);
  79
  80         set_to_next_nat(nm_i, nid);
  81
  82         return dst_page;
  83 }
  84
  85 /*
  86  * Readahead NAT pages
  87  */
  88 static void ra_nat_pages(struct f2fs_sb_info *sbi, int nid)
  89 {
  90         struct address_space *mapping = sbi->meta_inode->i_mapping;
  91         struct f2fs_nm_info *nm_i = NM_I(sbi);
  92         struct blk_plug plug;
  93         struct page *page;
  94         pgoff_t index;
  95         int i;
  96
  97         blk_start_plug(&plug);
  98
  99         for (i = 0; i < FREE_NID_PAGES; i++, nid += NAT_ENTRY_PER_BLOCK) {
 100                 if (nid >= nm_i->max_nid)
 101                         nid = 0;
 102                 index = current_nat_addr(sbi, nid);
 103
 104                 page = grab_cache_page(mapping, index);
 105                 if (!page)
 106                         continue;
 107                 if (PageUptodate(page)) {
 108                         f2fs_put_page(page, 1);
 109                         continue;
 110                 }
 111                 if (f2fs_readpage(sbi, page, index, READ))
 112                         continue;
 113
 114                 f2fs_put_page(page, 0);
 115         }
 116         blk_finish_plug(&plug);
 117 }
 118
 119 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 120 {
 121         return radix_tree_lookup(&nm_i->nat_root, n);
 122 }
 123
 124 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 125                 nid_t start, unsigned int nr, struct nat_entry **ep)
 126 {
 127         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 128 }
 129
 130 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 131 {
 132         list_del(&e->list);
 133         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 134         nm_i->nat_cnt--;
 135         kmem_cache_free(nat_entry_slab, e);
 136 }
 137
 138 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 139 {
 140         struct f2fs_nm_info *nm_i = NM_I(sbi);
 141         struct nat_entry *e;
 142         int is_cp = 1;
 143
 144         read_lock(&nm_i->nat_tree_lock);
 145         e = __lookup_nat_cache(nm_i, nid);
 146         if (e && !e->checkpointed)
 147                 is_cp = 0;
 148         read_unlock(&nm_i->nat_tree_lock);
 149         return is_cp;
 150 }
 151
 152 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 153 {
 154         struct nat_entry *new;
 155
 156         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 157         if (!new)
 158                 return NULL;
 159         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 160                 kmem_cache_free(nat_entry_slab, new);
 161                 return NULL;
 162         }
 163         memset(new, 0, sizeof(struct nat_entry));
 164         nat_set_nid(new, nid);
 165         list_add_tail(&new->list, &nm_i->nat_entries);
 166         nm_i->nat_cnt++;
 167         return new;
 168 }
 169
 170 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 171                                                 struct f2fs_nat_entry *ne)
 172 {
 173         struct nat_entry *e;
 174 retry:
 175         write_lock(&nm_i->nat_tree_lock);
 176         e = __lookup_nat_cache(nm_i, nid);
 177         if (!e) {
 178                 e = grab_nat_entry(nm_i, nid);
 179                 if (!e) {
 180                         write_unlock(&nm_i->nat_tree_lock);
 181                         goto retry;
 182                 }
 183                 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
 184                 nat_set_ino(e, le32_to_cpu(ne->ino));
 185                 nat_set_version(e, ne->version);
 186                 e->checkpointed = true;
 187         }
 188         write_unlock(&nm_i->nat_tree_lock);
 189 }
 190
 191 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 192                         block_t new_blkaddr)
 193 {
 194         struct f2fs_nm_info *nm_i = NM_I(sbi);
 195         struct nat_entry *e;
 196 retry:
 197         write_lock(&nm_i->nat_tree_lock);
 198         e = __lookup_nat_cache(nm_i, ni->nid);
 199         if (!e) {
 200                 e = grab_nat_entry(nm_i, ni->nid);
 201                 if (!e) {
 202                         write_unlock(&nm_i->nat_tree_lock);
 203                         goto retry;
 204                 }
 205                 e->ni = *ni;
 206                 e->checkpointed = true;
 207                 BUG_ON(ni->blk_addr == NEW_ADDR);
 208         } else if (new_blkaddr == NEW_ADDR) {
 209                 /*
 210                  * when nid is reallocated,
 211                  * previous nat entry can be remained in nat cache.
 212                  * So, reinitialize it with new information.
 213                  */
 214                 e->ni = *ni;
 215                 BUG_ON(ni->blk_addr != NULL_ADDR);
 216         }
 217
 218         if (new_blkaddr == NEW_ADDR)
 219                 e->checkpointed = false;
 220
 221         /* sanity check */
 222         BUG_ON(nat_get_blkaddr(e) != ni->blk_addr);
 223         BUG_ON(nat_get_blkaddr(e) == NULL_ADDR &&
 224                         new_blkaddr == NULL_ADDR);
 225         BUG_ON(nat_get_blkaddr(e) == NEW_ADDR &&
 226                         new_blkaddr == NEW_ADDR);
 227         BUG_ON(nat_get_blkaddr(e) != NEW_ADDR &&
 228                         nat_get_blkaddr(e) != NULL_ADDR &&
 229                         new_blkaddr == NEW_ADDR);
 230
 231         /* increament version no as node is removed */
 232         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 233                 unsigned char version = nat_get_version(e);
 234                 nat_set_version(e, inc_node_version(version));
 235         }
 236
 237         /* change address */
 238         nat_set_blkaddr(e, new_blkaddr);
 239         __set_nat_cache_dirty(nm_i, e);
 240         write_unlock(&nm_i->nat_tree_lock);
 241 }
 242
 243 static int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 244 {
 245         struct f2fs_nm_info *nm_i = NM_I(sbi);
 246
 247         if (nm_i->nat_cnt <= NM_WOUT_THRESHOLD)
 248                 return 0;
 249
 250         write_lock(&nm_i->nat_tree_lock);
 251         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 252                 struct nat_entry *ne;
 253                 ne = list_first_entry(&nm_i->nat_entries,
 254                                         struct nat_entry, list);
 255                 __del_from_nat_cache(nm_i, ne);
 256                 nr_shrink--;
 257         }
 258         write_unlock(&nm_i->nat_tree_lock);
 259         return nr_shrink;
 260 }
 261
 262 /*
 263  * This function returns always success
 264  */
 265 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 266 {
 267         struct f2fs_nm_info *nm_i = NM_I(sbi);
 268         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 269         struct f2fs_summary_block *sum = curseg->sum_blk;
 270         nid_t start_nid = START_NID(nid);
 271         struct f2fs_nat_block *nat_blk;
 272         struct page *page = NULL;
 273         struct f2fs_nat_entry ne;
 274         struct nat_entry *e;
 275         int i;
 276
 277         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 278         ni->nid = nid;
 279
 280         /* Check nat cache */
 281         read_lock(&nm_i->nat_tree_lock);
 282         e = __lookup_nat_cache(nm_i, nid);
 283         if (e) {
 284                 ni->ino = nat_get_ino(e);
 285                 ni->blk_addr = nat_get_blkaddr(e);
 286                 ni->version = nat_get_version(e);
 287         }
 288         read_unlock(&nm_i->nat_tree_lock);
 289         if (e)
 290                 return;
 291
 292         /* Check current segment summary */
 293         mutex_lock(&curseg->curseg_mutex);
 294         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 295         if (i >= 0) {
 296                 ne = nat_in_journal(sum, i);
 297                 node_info_from_raw_nat(ni, &ne);
 298         }
 299         mutex_unlock(&curseg->curseg_mutex);
 300         if (i >= 0)
 301                 goto cache;
 302
 303         /* Fill node_info from nat page */
 304         page = get_current_nat_page(sbi, start_nid);
 305         nat_blk = (struct f2fs_nat_block *)page_address(page);
 306         ne = nat_blk->entries[nid - start_nid];
 307         node_info_from_raw_nat(ni, &ne);
 308         f2fs_put_page(page, 1);
 309 cache:
 310         /* cache nat entry */
 311         cache_nat_entry(NM_I(sbi), nid, &ne);
 312 }
 313
 314 /*
 315  * The maximum depth is four.
 316  * Offset[0] will have raw inode offset.
 317  */
 318 static int get_node_path(struct f2fs_inode_info *fi, long block,
 319                                 int offset[4], unsigned int noffset[4])
 320 {
 321         const long direct_index = ADDRS_PER_INODE(fi);
 322         const long direct_blks = ADDRS_PER_BLOCK;
 323         const long dptrs_per_blk = NIDS_PER_BLOCK;
 324         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 325         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 326         int n = 0;
 327         int level = 0;
 328
 329         noffset[0] = 0;
 330
 331         if (block < direct_index) {
 332                 offset[n] = block;
 333                 goto got;
 334         }
 335         block -= direct_index;
 336         if (block < direct_blks) {
 337                 offset[n++] = NODE_DIR1_BLOCK;
 338                 noffset[n] = 1;
 339                 offset[n] = block;
 340                 level = 1;
 341                 goto got;
 342         }
 343         block -= direct_blks;
 344         if (block < direct_blks) {
 345                 offset[n++] = NODE_DIR2_BLOCK;
 346                 noffset[n] = 2;
 347                 offset[n] = block;
 348                 level = 1;
 349                 goto got;
 350         }
 351         block -= direct_blks;
 352         if (block < indirect_blks) {
 353                 offset[n++] = NODE_IND1_BLOCK;
 354                 noffset[n] = 3;
 355                 offset[n++] = block / direct_blks;
 356                 noffset[n] = 4 + offset[n - 1];
 357                 offset[n] = block % direct_blks;
 358                 level = 2;
 359                 goto got;
 360         }
 361         block -= indirect_blks;
 362         if (block < indirect_blks) {
 363                 offset[n++] = NODE_IND2_BLOCK;
 364                 noffset[n] = 4 + dptrs_per_blk;
 365                 offset[n++] = block / direct_blks;
 366                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 367                 offset[n] = block % direct_blks;
 368                 level = 2;
 369                 goto got;
 370         }
 371         block -= indirect_blks;
 372         if (block < dindirect_blks) {
 373                 offset[n++] = NODE_DIND_BLOCK;
 374                 noffset[n] = 5 + (dptrs_per_blk * 2);
 375                 offset[n++] = block / indirect_blks;
 376                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 377                               offset[n - 1] * (dptrs_per_blk + 1);
 378                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 379                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 380                               offset[n - 2] * (dptrs_per_blk + 1) +
 381                               offset[n - 1];
 382                 offset[n] = block % direct_blks;
 383                 level = 3;
 384                 goto got;
 385         } else {
 386                 BUG();
 387         }
 388 got:
 389         return level;
 390 }
 391
 392 /*
 393  * Caller should call f2fs_put_dnode(dn).
 394  * Also, it should grab and release a mutex by calling mutex_lock_op() and
 395  * mutex_unlock_op() only if ro is not set RDONLY_NODE.
 396  * In the case of RDONLY_NODE, we don't need to care about mutex.
 397  */
 398 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 399 {
 400         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 401         struct page *npage[4];
 402         struct page *parent;
 403         int offset[4];
 404         unsigned int noffset[4];
 405         nid_t nids[4];
 406         int level, i;
 407         int err = 0;
 408
 409         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 410
 411         nids[0] = dn->inode->i_ino;
 412         npage[0] = dn->inode_page;
 413
 414         if (!npage[0]) {
 415                 npage[0] = get_node_page(sbi, nids[0]);
 416                 if (IS_ERR(npage[0]))
 417                         return PTR_ERR(npage[0]);
 418         }
 419         parent = npage[0];
 420         if (level != 0)
 421                 nids[1] = get_nid(parent, offset[0], true);
 422         dn->inode_page = npage[0];
 423         dn->inode_page_locked = true;
 424
 425         /* get indirect or direct nodes */
 426         for (i = 1; i <= level; i++) {
 427                 bool done = false;
 428
 429                 if (!nids[i] && mode == ALLOC_NODE) {
 430                         /* alloc new node */
 431                         if (!alloc_nid(sbi, &(nids[i]))) {
 432                                 err = -ENOSPC;
 433                                 goto release_pages;
 434                         }
 435
 436                         dn->nid = nids[i];
 437                         npage[i] = new_node_page(dn, noffset[i], NULL);
 438                         if (IS_ERR(npage[i])) {
 439                                 alloc_nid_failed(sbi, nids[i]);
 440                                 err = PTR_ERR(npage[i]);
 441                                 goto release_pages;
 442                         }
 443
 444                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 445                         alloc_nid_done(sbi, nids[i]);
 446                         done = true;
 447                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 448                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 449                         if (IS_ERR(npage[i])) {
 450                                 err = PTR_ERR(npage[i]);
 451                                 goto release_pages;
 452                         }
 453                         done = true;
 454                 }
 455                 if (i == 1) {
 456                         dn->inode_page_locked = false;
 457                         unlock_page(parent);
 458                 } else {
 459                         f2fs_put_page(parent, 1);
 460                 }
 461
 462                 if (!done) {
 463                         npage[i] = get_node_page(sbi, nids[i]);
 464                         if (IS_ERR(npage[i])) {
 465                                 err = PTR_ERR(npage[i]);
 466                                 f2fs_put_page(npage[0], 0);
 467                                 goto release_out;
 468                         }
 469                 }
 470                 if (i < level) {
 471                         parent = npage[i];
 472                         nids[i + 1] = get_nid(parent, offset[i], false);
 473                 }
 474         }
 475         dn->nid = nids[level];
 476         dn->ofs_in_node = offset[level];
 477         dn->node_page = npage[level];
 478         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 479         return 0;
 480
 481 release_pages:
 482         f2fs_put_page(parent, 1);
 483         if (i > 1)
 484                 f2fs_put_page(npage[0], 0);
 485 release_out:
 486         dn->inode_page = NULL;
 487         dn->node_page = NULL;
 488         return err;
 489 }
 490
 491 static void truncate_node(struct dnode_of_data *dn)
 492 {
 493         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 494         struct node_info ni;
 495
 496         get_node_info(sbi, dn->nid, &ni);
 497         if (dn->inode->i_blocks == 0) {
 498                 BUG_ON(ni.blk_addr != NULL_ADDR);
 499                 goto invalidate;
 500         }
 501         BUG_ON(ni.blk_addr == NULL_ADDR);
 502
 503         /* Deallocate node address */
 504         invalidate_blocks(sbi, ni.blk_addr);
 505         dec_valid_node_count(sbi, dn->inode, 1);
 506         set_node_addr(sbi, &ni, NULL_ADDR);
 507
 508         if (dn->nid == dn->inode->i_ino) {
 509                 remove_orphan_inode(sbi, dn->nid);
 510                 dec_valid_inode_count(sbi);
 511         } else {
 512                 sync_inode_page(dn);
 513         }
 514 invalidate:
 515         clear_node_page_dirty(dn->node_page);
 516         F2FS_SET_SB_DIRT(sbi);
 517
 518         f2fs_put_page(dn->node_page, 1);
 519         dn->node_page = NULL;
 520         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 521 }
 522
 523 static int truncate_dnode(struct dnode_of_data *dn)
 524 {
 525         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 526         struct page *page;
 527
 528         if (dn->nid == 0)
 529                 return 1;
 530
 531         /* get direct node */
 532         page = get_node_page(sbi, dn->nid);
 533         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 534                 return 1;
 535         else if (IS_ERR(page))
 536                 return PTR_ERR(page);
 537
 538         /* Make dnode_of_data for parameter */
 539         dn->node_page = page;
 540         dn->ofs_in_node = 0;
 541         truncate_data_blocks(dn);
 542         truncate_node(dn);
 543         return 1;
 544 }
 545
 546 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 547                                                 int ofs, int depth)
 548 {
 549         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 550         struct dnode_of_data rdn = *dn;
 551         struct page *page;
 552         struct f2fs_node *rn;
 553         nid_t child_nid;
 554         unsigned int child_nofs;
 555         int freed = 0;
 556         int i, ret;
 557
 558         if (dn->nid == 0)
 559                 return NIDS_PER_BLOCK + 1;
 560
 561         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 562
 563         page = get_node_page(sbi, dn->nid);
 564         if (IS_ERR(page)) {
 565                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 566                 return PTR_ERR(page);
 567         }
 568
 569         rn = F2FS_NODE(page);
 570         if (depth < 3) {
 571                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 572                         child_nid = le32_to_cpu(rn->in.nid[i]);
 573                         if (child_nid == 0)
 574                                 continue;
 575                         rdn.nid = child_nid;
 576                         ret = truncate_dnode(&rdn);
 577                         if (ret < 0)
 578                                 goto out_err;
 579                         set_nid(page, i, 0, false);
 580                 }
 581         } else {
 582                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 583                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 584                         child_nid = le32_to_cpu(rn->in.nid[i]);
 585                         if (child_nid == 0) {
 586                                 child_nofs += NIDS_PER_BLOCK + 1;
 587                                 continue;
 588                         }
 589                         rdn.nid = child_nid;
 590                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 591                         if (ret == (NIDS_PER_BLOCK + 1)) {
 592                                 set_nid(page, i, 0, false);
 593                                 child_nofs += ret;
 594                         } else if (ret < 0 && ret != -ENOENT) {
 595                                 goto out_err;
 596                         }
 597                 }
 598                 freed = child_nofs;
 599         }
 600
 601         if (!ofs) {
 602                 /* remove current indirect node */
 603                 dn->node_page = page;
 604                 truncate_node(dn);
 605                 freed++;
 606         } else {
 607                 f2fs_put_page(page, 1);
 608         }
 609         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 610         return freed;
 611
 612 out_err:
 613         f2fs_put_page(page, 1);
 614         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 615         return ret;
 616 }
 617
 618 static int truncate_partial_nodes(struct dnode_of_data *dn,
 619                         struct f2fs_inode *ri, int *offset, int depth)
 620 {
 621         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 622         struct page *pages[2];
 623         nid_t nid[3];
 624         nid_t child_nid;
 625         int err = 0;
 626         int i;
 627         int idx = depth - 2;
 628
 629         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 630         if (!nid[0])
 631                 return 0;
 632
 633         /* get indirect nodes in the path */
 634         for (i = 0; i < depth - 1; i++) {
 635                 /* refernece count'll be increased */
 636                 pages[i] = get_node_page(sbi, nid[i]);
 637                 if (IS_ERR(pages[i])) {
 638                         depth = i + 1;
 639                         err = PTR_ERR(pages[i]);
 640                         goto fail;
 641                 }
 642                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 643         }
 644
 645         /* free direct nodes linked to a partial indirect node */
 646         for (i = offset[depth - 1]; i < NIDS_PER_BLOCK; i++) {
 647                 child_nid = get_nid(pages[idx], i, false);
 648                 if (!child_nid)
 649                         continue;
 650                 dn->nid = child_nid;
 651                 err = truncate_dnode(dn);
 652                 if (err < 0)
 653                         goto fail;
 654                 set_nid(pages[idx], i, 0, false);
 655         }
 656
 657         if (offset[depth - 1] == 0) {
 658                 dn->node_page = pages[idx];
 659                 dn->nid = nid[idx];
 660                 truncate_node(dn);
 661         } else {
 662                 f2fs_put_page(pages[idx], 1);
 663         }
 664         offset[idx]++;
 665         offset[depth - 1] = 0;
 666 fail:
 667         for (i = depth - 3; i >= 0; i--)
 668                 f2fs_put_page(pages[i], 1);
 669
 670         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 671
 672         return err;
 673 }
 674
 675 /*
 676  * All the block addresses of data and nodes should be nullified.
 677  */
 678 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 679 {
 680         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 681         struct address_space *node_mapping = sbi->node_inode->i_mapping;
 682         int err = 0, cont = 1;
 683         int level, offset[4], noffset[4];
 684         unsigned int nofs = 0;
 685         struct f2fs_node *rn;
 686         struct dnode_of_data dn;
 687         struct page *page;
 688
 689         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 690
 691         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 692 restart:
 693         page = get_node_page(sbi, inode->i_ino);
 694         if (IS_ERR(page)) {
 695                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 696                 return PTR_ERR(page);
 697         }
 698
 699         set_new_dnode(&dn, inode, page, NULL, 0);
 700         unlock_page(page);
 701
 702         rn = F2FS_NODE(page);
 703         switch (level) {
 704         case 0:
 705         case 1:
 706                 nofs = noffset[1];
 707                 break;
 708         case 2:
 709                 nofs = noffset[1];
 710                 if (!offset[level - 1])
 711                         goto skip_partial;
 712                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 713                 if (err < 0 && err != -ENOENT)
 714                         goto fail;
 715                 nofs += 1 + NIDS_PER_BLOCK;
 716                 break;
 717         case 3:
 718                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 719                 if (!offset[level - 1])
 720                         goto skip_partial;
 721                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 722                 if (err < 0 && err != -ENOENT)
 723                         goto fail;
 724                 break;
 725         default:
 726                 BUG();
 727         }
 728
 729 skip_partial:
 730         while (cont) {
 731                 dn.nid = le32_to_cpu(rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]);
 732                 switch (offset[0]) {
 733                 case NODE_DIR1_BLOCK:
 734                 case NODE_DIR2_BLOCK:
 735                         err = truncate_dnode(&dn);
 736                         break;
 737
 738                 case NODE_IND1_BLOCK:
 739                 case NODE_IND2_BLOCK:
 740                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 741                         break;
 742
 743                 case NODE_DIND_BLOCK:
 744                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 745                         cont = 0;
 746                         break;
 747
 748                 default:
 749                         BUG();
 750                 }
 751                 if (err < 0 && err != -ENOENT)
 752                         goto fail;
 753                 if (offset[1] == 0 &&
 754                                 rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 755                         lock_page(page);
 756                         if (page->mapping != node_mapping) {
 757                                 f2fs_put_page(page, 1);
 758                                 goto restart;
 759                         }
 760                         wait_on_page_writeback(page);
 761                         rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 762                         set_page_dirty(page);
 763                         unlock_page(page);
 764                 }
 765                 offset[1] = 0;
 766                 offset[0]++;
 767                 nofs += err;
 768         }
 769 fail:
 770         f2fs_put_page(page, 0);
 771         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 772         return err > 0 ? 0 : err;
 773 }
 774
 775 int truncate_xattr_node(struct inode *inode, struct page *page)
 776 {
 777         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 778         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 779         struct dnode_of_data dn;
 780         struct page *npage;
 781
 782         if (!nid)
 783                 return 0;
 784
 785         npage = get_node_page(sbi, nid);
 786         if (IS_ERR(npage))
 787                 return PTR_ERR(npage);
 788
 789         F2FS_I(inode)->i_xattr_nid = 0;
 790
 791         /* need to do checkpoint during fsync */
 792         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 793
 794         set_new_dnode(&dn, inode, page, npage, nid);
 795
 796         if (page)
 797                 dn.inode_page_locked = 1;
 798         truncate_node(&dn);
 799         return 0;
 800 }
 801
 802 /*
 803  * Caller should grab and release a mutex by calling mutex_lock_op() and
 804  * mutex_unlock_op().
 805  */
 806 int remove_inode_page(struct inode *inode)
 807 {
 808         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 809         struct page *page;
 810         nid_t ino = inode->i_ino;
 811         struct dnode_of_data dn;
 812         int err;
 813
 814         page = get_node_page(sbi, ino);
 815         if (IS_ERR(page))
 816                 return PTR_ERR(page);
 817
 818         err = truncate_xattr_node(inode, page);
 819         if (err) {
 820                 f2fs_put_page(page, 1);
 821                 return err;
 822         }
 823
 824         /* 0 is possible, after f2fs_new_inode() is failed */
 825         BUG_ON(inode->i_blocks != 0 && inode->i_blocks != 1);
 826         set_new_dnode(&dn, inode, page, page, ino);
 827         truncate_node(&dn);
 828         return 0;
 829 }
 830
 831 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
 832 {
 833         struct dnode_of_data dn;
 834
 835         /* allocate inode page for new inode */
 836         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 837
 838         /* caller should f2fs_put_page(page, 1); */
 839         return new_node_page(&dn, 0, NULL);
 840 }
 841
 842 struct page *new_node_page(struct dnode_of_data *dn,
 843                                 unsigned int ofs, struct page *ipage)
 844 {
 845         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 846         struct address_space *mapping = sbi->node_inode->i_mapping;
 847         struct node_info old_ni, new_ni;
 848         struct page *page;
 849         int err;
 850
 851         if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
 852                 return ERR_PTR(-EPERM);
 853
 854         page = grab_cache_page(mapping, dn->nid);
 855         if (!page)
 856                 return ERR_PTR(-ENOMEM);
 857
 858         if (!inc_valid_node_count(sbi, dn->inode, 1)) {
 859                 err = -ENOSPC;
 860                 goto fail;
 861         }
 862
 863         get_node_info(sbi, dn->nid, &old_ni);
 864
 865         /* Reinitialize old_ni with new node page */
 866         BUG_ON(old_ni.blk_addr != NULL_ADDR);
 867         new_ni = old_ni;
 868         new_ni.ino = dn->inode->i_ino;
 869         set_node_addr(sbi, &new_ni, NEW_ADDR);
 870
 871         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 872         set_cold_node(dn->inode, page);
 873         SetPageUptodate(page);
 874         set_page_dirty(page);
 875
 876         if (ofs == XATTR_NODE_OFFSET)
 877                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 878
 879         dn->node_page = page;
 880         if (ipage)
 881                 update_inode(dn->inode, ipage);
 882         else
 883                 sync_inode_page(dn);
 884         if (ofs == 0)
 885                 inc_valid_inode_count(sbi);
 886
 887         return page;
 888
 889 fail:
 890         clear_node_page_dirty(page);
 891         f2fs_put_page(page, 1);
 892         return ERR_PTR(err);
 893 }
 894
 895 /*
 896  * Caller should do after getting the following values.
 897  * 0: f2fs_put_page(page, 0)
 898  * LOCKED_PAGE: f2fs_put_page(page, 1)
 899  * error: nothing
 900  */
 901 static int read_node_page(struct page *page, int type)
 902 {
 903         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 904         struct node_info ni;
 905
 906         get_node_info(sbi, page->index, &ni);
 907
 908         if (ni.blk_addr == NULL_ADDR) {
 909                 f2fs_put_page(page, 1);
 910                 return -ENOENT;
 911         }
 912
 913         if (PageUptodate(page))
 914                 return LOCKED_PAGE;
 915
 916         return f2fs_readpage(sbi, page, ni.blk_addr, type);
 917 }
 918
 919 /*
 920  * Readahead a node page
 921  */
 922 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 923 {
 924         struct address_space *mapping = sbi->node_inode->i_mapping;
 925         struct page *apage;
 926         int err;
 927
 928         apage = find_get_page(mapping, nid);
 929         if (apage && PageUptodate(apage)) {
 930                 f2fs_put_page(apage, 0);
 931                 return;
 932         }
 933         f2fs_put_page(apage, 0);
 934
 935         apage = grab_cache_page(mapping, nid);
 936         if (!apage)
 937                 return;
 938
 939         err = read_node_page(apage, READA);
 940         if (err == 0)
 941                 f2fs_put_page(apage, 0);
 942         else if (err == LOCKED_PAGE)
 943                 f2fs_put_page(apage, 1);
 944 }
 945
 946 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 947 {
 948         struct address_space *mapping = sbi->node_inode->i_mapping;
 949         struct page *page;
 950         int err;
 951 repeat:
 952         page = grab_cache_page(mapping, nid);
 953         if (!page)
 954                 return ERR_PTR(-ENOMEM);
 955
 956         err = read_node_page(page, READ_SYNC);
 957         if (err < 0)
 958                 return ERR_PTR(err);
 959         else if (err == LOCKED_PAGE)
 960                 goto got_it;
 961
 962         lock_page(page);
 963         if (!PageUptodate(page)) {
 964                 f2fs_put_page(page, 1);
 965                 return ERR_PTR(-EIO);
 966         }
 967         if (page->mapping != mapping) {
 968                 f2fs_put_page(page, 1);
 969                 goto repeat;
 970         }
 971 got_it:
 972         BUG_ON(nid != nid_of_node(page));
 973         mark_page_accessed(page);
 974         return page;
 975 }
 976
 977 /*
 978  * Return a locked page for the desired node page.
 979  * And, readahead MAX_RA_NODE number of node pages.
 980  */
 981 struct page *get_node_page_ra(struct page *parent, int start)
 982 {
 983         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 984         struct address_space *mapping = sbi->node_inode->i_mapping;
 985         struct blk_plug plug;
 986         struct page *page;
 987         int err, i, end;
 988         nid_t nid;
 989
 990         /* First, try getting the desired direct node. */
 991         nid = get_nid(parent, start, false);
 992         if (!nid)
 993                 return ERR_PTR(-ENOENT);
 994 repeat:
 995         page = grab_cache_page(mapping, nid);
 996         if (!page)
 997                 return ERR_PTR(-ENOMEM);
 998
 999         err = read_node_page(page, READ_SYNC);
1000         if (err < 0)
1001                 return ERR_PTR(err);
1002         else if (err == LOCKED_PAGE)
1003                 goto page_hit;
1004
1005         blk_start_plug(&plug);
1006
1007         /* Then, try readahead for siblings of the desired node */
1008         end = start + MAX_RA_NODE;
1009         end = min(end, NIDS_PER_BLOCK);
1010         for (i = start + 1; i < end; i++) {
1011                 nid = get_nid(parent, i, false);
1012                 if (!nid)
1013                         continue;
1014                 ra_node_page(sbi, nid);
1015         }
1016
1017         blk_finish_plug(&plug);
1018
1019         lock_page(page);
1020         if (page->mapping != mapping) {
1021                 f2fs_put_page(page, 1);
1022                 goto repeat;
1023         }
1024 page_hit:
1025         if (!PageUptodate(page)) {
1026                 f2fs_put_page(page, 1);
1027                 return ERR_PTR(-EIO);
1028         }
1029         mark_page_accessed(page);
1030         return page;
1031 }
1032
1033 void sync_inode_page(struct dnode_of_data *dn)
1034 {
1035         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1036                 update_inode(dn->inode, dn->node_page);
1037         } else if (dn->inode_page) {
1038                 if (!dn->inode_page_locked)
1039                         lock_page(dn->inode_page);
1040                 update_inode(dn->inode, dn->inode_page);
1041                 if (!dn->inode_page_locked)
1042                         unlock_page(dn->inode_page);
1043         } else {
1044                 update_inode_page(dn->inode);
1045         }
1046 }
1047
1048 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1049                                         struct writeback_control *wbc)
1050 {
1051         struct address_space *mapping = sbi->node_inode->i_mapping;
1052         pgoff_t index, end;
1053         struct pagevec pvec;
1054         int step = ino ? 2 : 0;
1055         int nwritten = 0, wrote = 0;
1056
1057         pagevec_init(&pvec, 0);
1058
1059 next_step:
1060         index = 0;
1061         end = LONG_MAX;
1062
1063         while (index <= end) {
1064                 int i, nr_pages;
1065                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1066                                 PAGECACHE_TAG_DIRTY,
1067                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1068                 if (nr_pages == 0)
1069                         break;
1070
1071                 for (i = 0; i < nr_pages; i++) {
1072                         struct page *page = pvec.pages[i];
1073
1074                         /*
1075                          * flushing sequence with step:
1076                          * 0. indirect nodes
1077                          * 1. dentry dnodes
1078                          * 2. file dnodes
1079                          */
1080                         if (step == 0 && IS_DNODE(page))
1081                                 continue;
1082                         if (step == 1 && (!IS_DNODE(page) ||
1083                                                 is_cold_node(page)))
1084                                 continue;
1085                         if (step == 2 && (!IS_DNODE(page) ||
1086                                                 !is_cold_node(page)))
1087                                 continue;
1088
1089                         /*
1090                          * If an fsync mode,
1091                          * we should not skip writing node pages.
1092                          */
1093                         if (ino && ino_of_node(page) == ino)
1094                                 lock_page(page);
1095                         else if (!trylock_page(page))
1096                                 continue;
1097
1098                         if (unlikely(page->mapping != mapping)) {
1099 continue_unlock:
1100                                 unlock_page(page);
1101                                 continue;
1102                         }
1103                         if (ino && ino_of_node(page) != ino)
1104                                 goto continue_unlock;
1105
1106                         if (!PageDirty(page)) {
1107                                 /* someone wrote it for us */
1108                                 goto continue_unlock;
1109                         }
1110
1111                         if (!clear_page_dirty_for_io(page))
1112                                 goto continue_unlock;
1113
1114                         /* called by fsync() */
1115                         if (ino && IS_DNODE(page)) {
1116                                 int mark = !is_checkpointed_node(sbi, ino);
1117                                 set_fsync_mark(page, 1);
1118                                 if (IS_INODE(page))
1119                                         set_dentry_mark(page, mark);
1120                                 nwritten++;
1121                         } else {
1122                                 set_fsync_mark(page, 0);
1123                                 set_dentry_mark(page, 0);
1124                         }
1125                         mapping->a_ops->writepage(page, wbc);
1126                         wrote++;
1127
1128                         if (--wbc->nr_to_write == 0)
1129                                 break;
1130                 }
1131                 pagevec_release(&pvec);
1132                 cond_resched();
1133
1134                 if (wbc->nr_to_write == 0) {
1135                         step = 2;
1136                         break;
1137                 }
1138         }
1139
1140         if (step < 2) {
1141                 step++;
1142                 goto next_step;
1143         }
1144
1145         if (wrote)
1146                 f2fs_submit_bio(sbi, NODE, wbc->sync_mode == WB_SYNC_ALL);
1147
1148         return nwritten;
1149 }
1150
1151 static int f2fs_write_node_page(struct page *page,
1152                                 struct writeback_control *wbc)
1153 {
1154         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1155         nid_t nid;
1156         block_t new_addr;
1157         struct node_info ni;
1158
1159         if (sbi->por_doing)
1160                 goto redirty_out;
1161
1162         wait_on_page_writeback(page);
1163
1164         /* get old block addr of this node page */
1165         nid = nid_of_node(page);
1166         BUG_ON(page->index != nid);
1167
1168         get_node_info(sbi, nid, &ni);
1169
1170         /* This page is already truncated */
1171         if (ni.blk_addr == NULL_ADDR) {
1172                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1173                 unlock_page(page);
1174                 return 0;
1175         }
1176
1177         if (wbc->for_reclaim)
1178                 goto redirty_out;
1179
1180         mutex_lock(&sbi->node_write);
1181         set_page_writeback(page);
1182         write_node_page(sbi, page, nid, ni.blk_addr, &new_addr);
1183         set_node_addr(sbi, &ni, new_addr);
1184         dec_page_count(sbi, F2FS_DIRTY_NODES);
1185         mutex_unlock(&sbi->node_write);
1186         unlock_page(page);
1187         return 0;
1188
1189 redirty_out:
1190         dec_page_count(sbi, F2FS_DIRTY_NODES);
1191         wbc->pages_skipped++;
1192         set_page_dirty(page);
1193         return AOP_WRITEPAGE_ACTIVATE;
1194 }
1195
1196 /*
1197  * It is very important to gather dirty pages and write at once, so that we can
1198  * submit a big bio without interfering other data writes.
1199  * Be default, 512 pages (2MB) * 3 node types, is more reasonable.
1200  */
1201 #define COLLECT_DIRTY_NODES     1536
1202 static int f2fs_write_node_pages(struct address_space *mapping,
1203                             struct writeback_control *wbc)
1204 {
1205         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1206         long nr_to_write = wbc->nr_to_write;
1207
1208         /* First check balancing cached NAT entries */
1209         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
1210                                 excess_prefree_segs(sbi)) {
1211                 f2fs_sync_fs(sbi->sb, true);
1212                 return 0;
1213         }
1214
1215         /* collect a number of dirty node pages and write together */
1216         if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1217                 return 0;
1218
1219         /* if mounting is failed, skip writing node pages */
1220         wbc->nr_to_write = 3 * max_hw_blocks(sbi);
1221         sync_node_pages(sbi, 0, wbc);
1222         wbc->nr_to_write = nr_to_write - (3 * max_hw_blocks(sbi) -
1223                                                 wbc->nr_to_write);
1224         return 0;
1225 }
1226
1227 static int f2fs_set_node_page_dirty(struct page *page)
1228 {
1229         struct address_space *mapping = page->mapping;
1230         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1231
1232         SetPageUptodate(page);
1233         if (!PageDirty(page)) {
1234                 __set_page_dirty_nobuffers(page);
1235                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1236                 SetPagePrivate(page);
1237                 return 1;
1238         }
1239         return 0;
1240 }
1241
1242 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1243                                       unsigned int length)
1244 {
1245         struct inode *inode = page->mapping->host;
1246         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1247         if (PageDirty(page))
1248                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1249         ClearPagePrivate(page);
1250 }
1251
1252 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1253 {
1254         ClearPagePrivate(page);
1255         return 1;
1256 }
1257
1258 /*
1259  * Structure of the f2fs node operations
1260  */
1261 const struct address_space_operations f2fs_node_aops = {
1262         .writepage      = f2fs_write_node_page,
1263         .writepages     = f2fs_write_node_pages,
1264         .set_page_dirty = f2fs_set_node_page_dirty,
1265         .invalidatepage = f2fs_invalidate_node_page,
1266         .releasepage    = f2fs_release_node_page,
1267 };
1268
1269 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1270 {
1271         struct list_head *this;
1272         struct free_nid *i;
1273         list_for_each(this, head) {
1274                 i = list_entry(this, struct free_nid, list);
1275                 if (i->nid == n)
1276                         return i;
1277         }
1278         return NULL;
1279 }
1280
1281 static void __del_from_free_nid_list(struct free_nid *i)
1282 {
1283         list_del(&i->list);
1284         kmem_cache_free(free_nid_slab, i);
1285 }
1286
1287 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build)
1288 {
1289         struct free_nid *i;
1290         struct nat_entry *ne;
1291         bool allocated = false;
1292
1293         if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1294                 return -1;
1295
1296         /* 0 nid should not be used */
1297         if (nid == 0)
1298                 return 0;
1299
1300         if (build) {
1301                 /* do not add allocated nids */
1302                 read_lock(&nm_i->nat_tree_lock);
1303                 ne = __lookup_nat_cache(nm_i, nid);
1304                 if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1305                         allocated = true;
1306                 read_unlock(&nm_i->nat_tree_lock);
1307                 if (allocated)
1308                         return 0;
1309         }
1310
1311         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1312         i->nid = nid;
1313         i->state = NID_NEW;
1314
1315         spin_lock(&nm_i->free_nid_list_lock);
1316         if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1317                 spin_unlock(&nm_i->free_nid_list_lock);
1318                 kmem_cache_free(free_nid_slab, i);
1319                 return 0;
1320         }
1321         list_add_tail(&i->list, &nm_i->free_nid_list);
1322         nm_i->fcnt++;
1323         spin_unlock(&nm_i->free_nid_list_lock);
1324         return 1;
1325 }
1326
1327 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1328 {
1329         struct free_nid *i;
1330         spin_lock(&nm_i->free_nid_list_lock);
1331         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1332         if (i && i->state == NID_NEW) {
1333                 __del_from_free_nid_list(i);
1334                 nm_i->fcnt--;
1335         }
1336         spin_unlock(&nm_i->free_nid_list_lock);
1337 }
1338
1339 static void scan_nat_page(struct f2fs_nm_info *nm_i,
1340                         struct page *nat_page, nid_t start_nid)
1341 {
1342         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1343         block_t blk_addr;
1344         int i;
1345
1346         i = start_nid % NAT_ENTRY_PER_BLOCK;
1347
1348         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1349
1350                 if (start_nid >= nm_i->max_nid)
1351                         break;
1352
1353                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1354                 BUG_ON(blk_addr == NEW_ADDR);
1355                 if (blk_addr == NULL_ADDR) {
1356                         if (add_free_nid(nm_i, start_nid, true) < 0)
1357                                 break;
1358                 }
1359         }
1360 }
1361
1362 static void build_free_nids(struct f2fs_sb_info *sbi)
1363 {
1364         struct f2fs_nm_info *nm_i = NM_I(sbi);
1365         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1366         struct f2fs_summary_block *sum = curseg->sum_blk;
1367         int i = 0;
1368         nid_t nid = nm_i->next_scan_nid;
1369
1370         /* Enough entries */
1371         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1372                 return;
1373
1374         /* readahead nat pages to be scanned */
1375         ra_nat_pages(sbi, nid);
1376
1377         while (1) {
1378                 struct page *page = get_current_nat_page(sbi, nid);
1379
1380                 scan_nat_page(nm_i, page, nid);
1381                 f2fs_put_page(page, 1);
1382
1383                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1384                 if (nid >= nm_i->max_nid)
1385                         nid = 0;
1386
1387                 if (i++ == FREE_NID_PAGES)
1388                         break;
1389         }
1390
1391         /* go to the next free nat pages to find free nids abundantly */
1392         nm_i->next_scan_nid = nid;
1393
1394         /* find free nids from current sum_pages */
1395         mutex_lock(&curseg->curseg_mutex);
1396         for (i = 0; i < nats_in_cursum(sum); i++) {
1397                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1398                 nid = le32_to_cpu(nid_in_journal(sum, i));
1399                 if (addr == NULL_ADDR)
1400                         add_free_nid(nm_i, nid, true);
1401                 else
1402                         remove_free_nid(nm_i, nid);
1403         }
1404         mutex_unlock(&curseg->curseg_mutex);
1405 }
1406
1407 /*
1408  * If this function returns success, caller can obtain a new nid
1409  * from second parameter of this function.
1410  * The returned nid could be used ino as well as nid when inode is created.
1411  */
1412 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1413 {
1414         struct f2fs_nm_info *nm_i = NM_I(sbi);
1415         struct free_nid *i = NULL;
1416         struct list_head *this;
1417 retry:
1418         if (sbi->total_valid_node_count + 1 >= nm_i->max_nid)
1419                 return false;
1420
1421         spin_lock(&nm_i->free_nid_list_lock);
1422
1423         /* We should not use stale free nids created by build_free_nids */
1424         if (nm_i->fcnt && !sbi->on_build_free_nids) {
1425                 BUG_ON(list_empty(&nm_i->free_nid_list));
1426                 list_for_each(this, &nm_i->free_nid_list) {
1427                         i = list_entry(this, struct free_nid, list);
1428                         if (i->state == NID_NEW)
1429                                 break;
1430                 }
1431
1432                 BUG_ON(i->state != NID_NEW);
1433                 *nid = i->nid;
1434                 i->state = NID_ALLOC;
1435                 nm_i->fcnt--;
1436                 spin_unlock(&nm_i->free_nid_list_lock);
1437                 return true;
1438         }
1439         spin_unlock(&nm_i->free_nid_list_lock);
1440
1441         /* Let's scan nat pages and its caches to get free nids */
1442         mutex_lock(&nm_i->build_lock);
1443         sbi->on_build_free_nids = true;
1444         build_free_nids(sbi);
1445         sbi->on_build_free_nids = false;
1446         mutex_unlock(&nm_i->build_lock);
1447         goto retry;
1448 }
1449
1450 /*
1451  * alloc_nid() should be called prior to this function.
1452  */
1453 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1454 {
1455         struct f2fs_nm_info *nm_i = NM_I(sbi);
1456         struct free_nid *i;
1457
1458         spin_lock(&nm_i->free_nid_list_lock);
1459         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1460         BUG_ON(!i || i->state != NID_ALLOC);
1461         __del_from_free_nid_list(i);
1462         spin_unlock(&nm_i->free_nid_list_lock);
1463 }
1464
1465 /*
1466  * alloc_nid() should be called prior to this function.
1467  */
1468 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1469 {
1470         struct f2fs_nm_info *nm_i = NM_I(sbi);
1471         struct free_nid *i;
1472
1473         if (!nid)
1474                 return;
1475
1476         spin_lock(&nm_i->free_nid_list_lock);
1477         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1478         BUG_ON(!i || i->state != NID_ALLOC);
1479         if (nm_i->fcnt > 2 * MAX_FREE_NIDS) {
1480                 __del_from_free_nid_list(i);
1481         } else {
1482                 i->state = NID_NEW;
1483                 nm_i->fcnt++;
1484         }
1485         spin_unlock(&nm_i->free_nid_list_lock);
1486 }
1487
1488 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1489                 struct f2fs_summary *sum, struct node_info *ni,
1490                 block_t new_blkaddr)
1491 {
1492         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1493         set_node_addr(sbi, ni, new_blkaddr);
1494         clear_node_page_dirty(page);
1495 }
1496
1497 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1498 {
1499         struct address_space *mapping = sbi->node_inode->i_mapping;
1500         struct f2fs_node *src, *dst;
1501         nid_t ino = ino_of_node(page);
1502         struct node_info old_ni, new_ni;
1503         struct page *ipage;
1504
1505         ipage = grab_cache_page(mapping, ino);
1506         if (!ipage)
1507                 return -ENOMEM;
1508
1509         /* Should not use this inode  from free nid list */
1510         remove_free_nid(NM_I(sbi), ino);
1511
1512         get_node_info(sbi, ino, &old_ni);
1513         SetPageUptodate(ipage);
1514         fill_node_footer(ipage, ino, ino, 0, true);
1515
1516         src = F2FS_NODE(page);
1517         dst = F2FS_NODE(ipage);
1518
1519         memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i);
1520         dst->i.i_size = 0;
1521         dst->i.i_blocks = cpu_to_le64(1);
1522         dst->i.i_links = cpu_to_le32(1);
1523         dst->i.i_xattr_nid = 0;
1524
1525         new_ni = old_ni;
1526         new_ni.ino = ino;
1527
1528         if (!inc_valid_node_count(sbi, NULL, 1))
1529                 WARN_ON(1);
1530         set_node_addr(sbi, &new_ni, NEW_ADDR);
1531         inc_valid_inode_count(sbi);
1532         f2fs_put_page(ipage, 1);
1533         return 0;
1534 }
1535
1536 int restore_node_summary(struct f2fs_sb_info *sbi,
1537                         unsigned int segno, struct f2fs_summary_block *sum)
1538 {
1539         struct f2fs_node *rn;
1540         struct f2fs_summary *sum_entry;
1541         struct page *page;
1542         block_t addr;
1543         int i, last_offset;
1544
1545         /* alloc temporal page for read node */
1546         page = alloc_page(GFP_NOFS | __GFP_ZERO);
1547         if (!page)
1548                 return -ENOMEM;
1549         lock_page(page);
1550
1551         /* scan the node segment */
1552         last_offset = sbi->blocks_per_seg;
1553         addr = START_BLOCK(sbi, segno);
1554         sum_entry = &sum->entries[0];
1555
1556         for (i = 0; i < last_offset; i++, sum_entry++) {
1557                 /*
1558                  * In order to read next node page,
1559                  * we must clear PageUptodate flag.
1560                  */
1561                 ClearPageUptodate(page);
1562
1563                 if (f2fs_readpage(sbi, page, addr, READ_SYNC))
1564                         goto out;
1565
1566                 lock_page(page);
1567                 rn = F2FS_NODE(page);
1568                 sum_entry->nid = rn->footer.nid;
1569                 sum_entry->version = 0;
1570                 sum_entry->ofs_in_node = 0;
1571                 addr++;
1572         }
1573         unlock_page(page);
1574 out:
1575         __free_pages(page, 0);
1576         return 0;
1577 }
1578
1579 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1580 {
1581         struct f2fs_nm_info *nm_i = NM_I(sbi);
1582         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1583         struct f2fs_summary_block *sum = curseg->sum_blk;
1584         int i;
1585
1586         mutex_lock(&curseg->curseg_mutex);
1587
1588         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1589                 mutex_unlock(&curseg->curseg_mutex);
1590                 return false;
1591         }
1592
1593         for (i = 0; i < nats_in_cursum(sum); i++) {
1594                 struct nat_entry *ne;
1595                 struct f2fs_nat_entry raw_ne;
1596                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1597
1598                 raw_ne = nat_in_journal(sum, i);
1599 retry:
1600                 write_lock(&nm_i->nat_tree_lock);
1601                 ne = __lookup_nat_cache(nm_i, nid);
1602                 if (ne) {
1603                         __set_nat_cache_dirty(nm_i, ne);
1604                         write_unlock(&nm_i->nat_tree_lock);
1605                         continue;
1606                 }
1607                 ne = grab_nat_entry(nm_i, nid);
1608                 if (!ne) {
1609                         write_unlock(&nm_i->nat_tree_lock);
1610                         goto retry;
1611                 }
1612                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1613                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1614                 nat_set_version(ne, raw_ne.version);
1615                 __set_nat_cache_dirty(nm_i, ne);
1616                 write_unlock(&nm_i->nat_tree_lock);
1617         }
1618         update_nats_in_cursum(sum, -i);
1619         mutex_unlock(&curseg->curseg_mutex);
1620         return true;
1621 }
1622
1623 /*
1624  * This function is called during the checkpointing process.
1625  */
1626 void flush_nat_entries(struct f2fs_sb_info *sbi)
1627 {
1628         struct f2fs_nm_info *nm_i = NM_I(sbi);
1629         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1630         struct f2fs_summary_block *sum = curseg->sum_blk;
1631         struct list_head *cur, *n;
1632         struct page *page = NULL;
1633         struct f2fs_nat_block *nat_blk = NULL;
1634         nid_t start_nid = 0, end_nid = 0;
1635         bool flushed;
1636
1637         flushed = flush_nats_in_journal(sbi);
1638
1639         if (!flushed)
1640                 mutex_lock(&curseg->curseg_mutex);
1641
1642         /* 1) flush dirty nat caches */
1643         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1644                 struct nat_entry *ne;
1645                 nid_t nid;
1646                 struct f2fs_nat_entry raw_ne;
1647                 int offset = -1;
1648                 block_t new_blkaddr;
1649
1650                 ne = list_entry(cur, struct nat_entry, list);
1651                 nid = nat_get_nid(ne);
1652
1653                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1654                         continue;
1655                 if (flushed)
1656                         goto to_nat_page;
1657
1658                 /* if there is room for nat enries in curseg->sumpage */
1659                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1660                 if (offset >= 0) {
1661                         raw_ne = nat_in_journal(sum, offset);
1662                         goto flush_now;
1663                 }
1664 to_nat_page:
1665                 if (!page || (start_nid > nid || nid > end_nid)) {
1666                         if (page) {
1667                                 f2fs_put_page(page, 1);
1668                                 page = NULL;
1669                         }
1670                         start_nid = START_NID(nid);
1671                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1672
1673                         /*
1674                          * get nat block with dirty flag, increased reference
1675                          * count, mapped and lock
1676                          */
1677                         page = get_next_nat_page(sbi, start_nid);
1678                         nat_blk = page_address(page);
1679                 }
1680
1681                 BUG_ON(!nat_blk);
1682                 raw_ne = nat_blk->entries[nid - start_nid];
1683 flush_now:
1684                 new_blkaddr = nat_get_blkaddr(ne);
1685
1686                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1687                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1688                 raw_ne.version = nat_get_version(ne);
1689
1690                 if (offset < 0) {
1691                         nat_blk->entries[nid - start_nid] = raw_ne;
1692                 } else {
1693                         nat_in_journal(sum, offset) = raw_ne;
1694                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1695                 }
1696
1697                 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1698                                 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1699                         write_lock(&nm_i->nat_tree_lock);
1700                         __del_from_nat_cache(nm_i, ne);
1701                         write_unlock(&nm_i->nat_tree_lock);
1702                 } else {
1703                         write_lock(&nm_i->nat_tree_lock);
1704                         __clear_nat_cache_dirty(nm_i, ne);
1705                         ne->checkpointed = true;
1706                         write_unlock(&nm_i->nat_tree_lock);
1707                 }
1708         }
1709         if (!flushed)
1710                 mutex_unlock(&curseg->curseg_mutex);
1711         f2fs_put_page(page, 1);
1712
1713         /* 2) shrink nat caches if necessary */
1714         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1715 }
1716
1717 static int init_node_manager(struct f2fs_sb_info *sbi)
1718 {
1719         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1720         struct f2fs_nm_info *nm_i = NM_I(sbi);
1721         unsigned char *version_bitmap;
1722         unsigned int nat_segs, nat_blocks;
1723
1724         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1725
1726         /* segment_count_nat includes pair segment so divide to 2. */
1727         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1728         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1729         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1730         nm_i->fcnt = 0;
1731         nm_i->nat_cnt = 0;
1732
1733         INIT_LIST_HEAD(&nm_i->free_nid_list);
1734         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1735         INIT_LIST_HEAD(&nm_i->nat_entries);
1736         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1737
1738         mutex_init(&nm_i->build_lock);
1739         spin_lock_init(&nm_i->free_nid_list_lock);
1740         rwlock_init(&nm_i->nat_tree_lock);
1741
1742         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1743         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1744         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1745         if (!version_bitmap)
1746                 return -EFAULT;
1747
1748         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1749                                         GFP_KERNEL);
1750         if (!nm_i->nat_bitmap)
1751                 return -ENOMEM;
1752         return 0;
1753 }
1754
1755 int build_node_manager(struct f2fs_sb_info *sbi)
1756 {
1757         int err;
1758
1759         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1760         if (!sbi->nm_info)
1761                 return -ENOMEM;
1762
1763         err = init_node_manager(sbi);
1764         if (err)
1765                 return err;
1766
1767         build_free_nids(sbi);
1768         return 0;
1769 }
1770
1771 void destroy_node_manager(struct f2fs_sb_info *sbi)
1772 {
1773         struct f2fs_nm_info *nm_i = NM_I(sbi);
1774         struct free_nid *i, *next_i;
1775         struct nat_entry *natvec[NATVEC_SIZE];
1776         nid_t nid = 0;
1777         unsigned int found;
1778
1779         if (!nm_i)
1780                 return;
1781
1782         /* destroy free nid list */
1783         spin_lock(&nm_i->free_nid_list_lock);
1784         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1785                 BUG_ON(i->state == NID_ALLOC);
1786                 __del_from_free_nid_list(i);
1787                 nm_i->fcnt--;
1788         }
1789         BUG_ON(nm_i->fcnt);
1790         spin_unlock(&nm_i->free_nid_list_lock);
1791
1792         /* destroy nat cache */
1793         write_lock(&nm_i->nat_tree_lock);
1794         while ((found = __gang_lookup_nat_cache(nm_i,
1795                                         nid, NATVEC_SIZE, natvec))) {
1796                 unsigned idx;
1797                 for (idx = 0; idx < found; idx++) {
1798                         struct nat_entry *e = natvec[idx];
1799                         nid = nat_get_nid(e) + 1;
1800                         __del_from_nat_cache(nm_i, e);
1801                 }
1802         }
1803         BUG_ON(nm_i->nat_cnt);
1804         write_unlock(&nm_i->nat_tree_lock);
1805
1806         kfree(nm_i->nat_bitmap);
1807         sbi->nm_info = NULL;
1808         kfree(nm_i);
1809 }
1810
1811 int __init create_node_manager_caches(void)
1812 {
1813         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1814                         sizeof(struct nat_entry), NULL);
1815         if (!nat_entry_slab)
1816                 return -ENOMEM;
1817
1818         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1819                         sizeof(struct free_nid), NULL);
1820         if (!free_nid_slab) {
1821                 kmem_cache_destroy(nat_entry_slab);
1822                 return -ENOMEM;
1823         }
1824         return 0;
1825 }
1826
1827 void destroy_node_manager_caches(void)
1828 {
1829         kmem_cache_destroy(free_nid_slab);
1830         kmem_cache_destroy(nat_entry_slab);
1831 }