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                 f2fs_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                 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
 216         }
 217
 218         if (new_blkaddr == NEW_ADDR)
 219                 e->checkpointed = false;
 220
 221         /* sanity check */
 222         f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
 223         f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
 224                         new_blkaddr == NULL_ADDR);
 225         f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
 226                         new_blkaddr == NEW_ADDR);
 227         f2fs_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 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                 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
 499                 goto invalidate;
 500         }
 501         f2fs_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);
 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 void 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
 813         page = get_node_page(sbi, ino);
 814         if (IS_ERR(page))
 815                 return;
 816
 817         if (truncate_xattr_node(inode, page)) {
 818                 f2fs_put_page(page, 1);
 819                 return;
 820         }
 821         /* 0 is possible, after f2fs_new_inode() is failed */
 822         f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
 823         set_new_dnode(&dn, inode, page, page, ino);
 824         truncate_node(&dn);
 825 }
 826
 827 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
 828 {
 829         struct dnode_of_data dn;
 830
 831         /* allocate inode page for new inode */
 832         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 833
 834         /* caller should f2fs_put_page(page, 1); */
 835         return new_node_page(&dn, 0, NULL);
 836 }
 837
 838 struct page *new_node_page(struct dnode_of_data *dn,
 839                                 unsigned int ofs, struct page *ipage)
 840 {
 841         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 842         struct address_space *mapping = sbi->node_inode->i_mapping;
 843         struct node_info old_ni, new_ni;
 844         struct page *page;
 845         int err;
 846
 847         if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
 848                 return ERR_PTR(-EPERM);
 849
 850         page = grab_cache_page(mapping, dn->nid);
 851         if (!page)
 852                 return ERR_PTR(-ENOMEM);
 853
 854         if (!inc_valid_node_count(sbi, dn->inode)) {
 855                 err = -ENOSPC;
 856                 goto fail;
 857         }
 858
 859         get_node_info(sbi, dn->nid, &old_ni);
 860
 861         /* Reinitialize old_ni with new node page */
 862         f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
 863         new_ni = old_ni;
 864         new_ni.ino = dn->inode->i_ino;
 865         set_node_addr(sbi, &new_ni, NEW_ADDR);
 866
 867         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 868         set_cold_node(dn->inode, page);
 869         SetPageUptodate(page);
 870         set_page_dirty(page);
 871
 872         if (ofs == XATTR_NODE_OFFSET)
 873                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 874
 875         dn->node_page = page;
 876         if (ipage)
 877                 update_inode(dn->inode, ipage);
 878         else
 879                 sync_inode_page(dn);
 880         if (ofs == 0)
 881                 inc_valid_inode_count(sbi);
 882
 883         return page;
 884
 885 fail:
 886         clear_node_page_dirty(page);
 887         f2fs_put_page(page, 1);
 888         return ERR_PTR(err);
 889 }
 890
 891 /*
 892  * Caller should do after getting the following values.
 893  * 0: f2fs_put_page(page, 0)
 894  * LOCKED_PAGE: f2fs_put_page(page, 1)
 895  * error: nothing
 896  */
 897 static int read_node_page(struct page *page, int type)
 898 {
 899         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 900         struct node_info ni;
 901
 902         get_node_info(sbi, page->index, &ni);
 903
 904         if (ni.blk_addr == NULL_ADDR) {
 905                 f2fs_put_page(page, 1);
 906                 return -ENOENT;
 907         }
 908
 909         if (PageUptodate(page))
 910                 return LOCKED_PAGE;
 911
 912         return f2fs_readpage(sbi, page, ni.blk_addr, type);
 913 }
 914
 915 /*
 916  * Readahead a node page
 917  */
 918 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 919 {
 920         struct address_space *mapping = sbi->node_inode->i_mapping;
 921         struct page *apage;
 922         int err;
 923
 924         apage = find_get_page(mapping, nid);
 925         if (apage && PageUptodate(apage)) {
 926                 f2fs_put_page(apage, 0);
 927                 return;
 928         }
 929         f2fs_put_page(apage, 0);
 930
 931         apage = grab_cache_page(mapping, nid);
 932         if (!apage)
 933                 return;
 934
 935         err = read_node_page(apage, READA);
 936         if (err == 0)
 937                 f2fs_put_page(apage, 0);
 938         else if (err == LOCKED_PAGE)
 939                 f2fs_put_page(apage, 1);
 940 }
 941
 942 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 943 {
 944         struct address_space *mapping = sbi->node_inode->i_mapping;
 945         struct page *page;
 946         int err;
 947 repeat:
 948         page = grab_cache_page(mapping, nid);
 949         if (!page)
 950                 return ERR_PTR(-ENOMEM);
 951
 952         err = read_node_page(page, READ_SYNC);
 953         if (err < 0)
 954                 return ERR_PTR(err);
 955         else if (err == LOCKED_PAGE)
 956                 goto got_it;
 957
 958         lock_page(page);
 959         if (!PageUptodate(page)) {
 960                 f2fs_put_page(page, 1);
 961                 return ERR_PTR(-EIO);
 962         }
 963         if (page->mapping != mapping) {
 964                 f2fs_put_page(page, 1);
 965                 goto repeat;
 966         }
 967 got_it:
 968         f2fs_bug_on(nid != nid_of_node(page));
 969         mark_page_accessed(page);
 970         return page;
 971 }
 972
 973 /*
 974  * Return a locked page for the desired node page.
 975  * And, readahead MAX_RA_NODE number of node pages.
 976  */
 977 struct page *get_node_page_ra(struct page *parent, int start)
 978 {
 979         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 980         struct address_space *mapping = sbi->node_inode->i_mapping;
 981         struct blk_plug plug;
 982         struct page *page;
 983         int err, i, end;
 984         nid_t nid;
 985
 986         /* First, try getting the desired direct node. */
 987         nid = get_nid(parent, start, false);
 988         if (!nid)
 989                 return ERR_PTR(-ENOENT);
 990 repeat:
 991         page = grab_cache_page(mapping, nid);
 992         if (!page)
 993                 return ERR_PTR(-ENOMEM);
 994
 995         err = read_node_page(page, READ_SYNC);
 996         if (err < 0)
 997                 return ERR_PTR(err);
 998         else if (err == LOCKED_PAGE)
 999                 goto page_hit;
1000
1001         blk_start_plug(&plug);
1002
1003         /* Then, try readahead for siblings of the desired node */
1004         end = start + MAX_RA_NODE;
1005         end = min(end, NIDS_PER_BLOCK);
1006         for (i = start + 1; i < end; i++) {
1007                 nid = get_nid(parent, i, false);
1008                 if (!nid)
1009                         continue;
1010                 ra_node_page(sbi, nid);
1011         }
1012
1013         blk_finish_plug(&plug);
1014
1015         lock_page(page);
1016         if (page->mapping != mapping) {
1017                 f2fs_put_page(page, 1);
1018                 goto repeat;
1019         }
1020 page_hit:
1021         if (!PageUptodate(page)) {
1022                 f2fs_put_page(page, 1);
1023                 return ERR_PTR(-EIO);
1024         }
1025         mark_page_accessed(page);
1026         return page;
1027 }
1028
1029 void sync_inode_page(struct dnode_of_data *dn)
1030 {
1031         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1032                 update_inode(dn->inode, dn->node_page);
1033         } else if (dn->inode_page) {
1034                 if (!dn->inode_page_locked)
1035                         lock_page(dn->inode_page);
1036                 update_inode(dn->inode, dn->inode_page);
1037                 if (!dn->inode_page_locked)
1038                         unlock_page(dn->inode_page);
1039         } else {
1040                 update_inode_page(dn->inode);
1041         }
1042 }
1043
1044 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1045                                         struct writeback_control *wbc)
1046 {
1047         struct address_space *mapping = sbi->node_inode->i_mapping;
1048         pgoff_t index, end;
1049         struct pagevec pvec;
1050         int step = ino ? 2 : 0;
1051         int nwritten = 0, wrote = 0;
1052
1053         pagevec_init(&pvec, 0);
1054
1055 next_step:
1056         index = 0;
1057         end = LONG_MAX;
1058
1059         while (index <= end) {
1060                 int i, nr_pages;
1061                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1062                                 PAGECACHE_TAG_DIRTY,
1063                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1064                 if (nr_pages == 0)
1065                         break;
1066
1067                 for (i = 0; i < nr_pages; i++) {
1068                         struct page *page = pvec.pages[i];
1069
1070                         /*
1071                          * flushing sequence with step:
1072                          * 0. indirect nodes
1073                          * 1. dentry dnodes
1074                          * 2. file dnodes
1075                          */
1076                         if (step == 0 && IS_DNODE(page))
1077                                 continue;
1078                         if (step == 1 && (!IS_DNODE(page) ||
1079                                                 is_cold_node(page)))
1080                                 continue;
1081                         if (step == 2 && (!IS_DNODE(page) ||
1082                                                 !is_cold_node(page)))
1083                                 continue;
1084
1085                         /*
1086                          * If an fsync mode,
1087                          * we should not skip writing node pages.
1088                          */
1089                         if (ino && ino_of_node(page) == ino)
1090                                 lock_page(page);
1091                         else if (!trylock_page(page))
1092                                 continue;
1093
1094                         if (unlikely(page->mapping != mapping)) {
1095 continue_unlock:
1096                                 unlock_page(page);
1097                                 continue;
1098                         }
1099                         if (ino && ino_of_node(page) != ino)
1100                                 goto continue_unlock;
1101
1102                         if (!PageDirty(page)) {
1103                                 /* someone wrote it for us */
1104                                 goto continue_unlock;
1105                         }
1106
1107                         if (!clear_page_dirty_for_io(page))
1108                                 goto continue_unlock;
1109
1110                         /* called by fsync() */
1111                         if (ino && IS_DNODE(page)) {
1112                                 int mark = !is_checkpointed_node(sbi, ino);
1113                                 set_fsync_mark(page, 1);
1114                                 if (IS_INODE(page))
1115                                         set_dentry_mark(page, mark);
1116                                 nwritten++;
1117                         } else {
1118                                 set_fsync_mark(page, 0);
1119                                 set_dentry_mark(page, 0);
1120                         }
1121                         mapping->a_ops->writepage(page, wbc);
1122                         wrote++;
1123
1124                         if (--wbc->nr_to_write == 0)
1125                                 break;
1126                 }
1127                 pagevec_release(&pvec);
1128                 cond_resched();
1129
1130                 if (wbc->nr_to_write == 0) {
1131                         step = 2;
1132                         break;
1133                 }
1134         }
1135
1136         if (step < 2) {
1137                 step++;
1138                 goto next_step;
1139         }
1140
1141         if (wrote)
1142                 f2fs_submit_bio(sbi, NODE, wbc->sync_mode == WB_SYNC_ALL);
1143
1144         return nwritten;
1145 }
1146
1147 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1148 {
1149         struct address_space *mapping = sbi->node_inode->i_mapping;
1150         pgoff_t index = 0, end = LONG_MAX;
1151         struct pagevec pvec;
1152         int nr_pages;
1153         int ret2 = 0, ret = 0;
1154
1155         pagevec_init(&pvec, 0);
1156         while ((index <= end) &&
1157                         (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1158                         PAGECACHE_TAG_WRITEBACK,
1159                         min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
1160                 unsigned i;
1161
1162                 for (i = 0; i < nr_pages; i++) {
1163                         struct page *page = pvec.pages[i];
1164
1165                         /* until radix tree lookup accepts end_index */
1166                         if (page->index > end)
1167                                 continue;
1168
1169                         if (ino && ino_of_node(page) == ino) {
1170                                 wait_on_page_writeback(page);
1171                                 if (TestClearPageError(page))
1172                                         ret = -EIO;
1173                         }
1174                 }
1175                 pagevec_release(&pvec);
1176                 cond_resched();
1177         }
1178
1179         if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
1180                 ret2 = -ENOSPC;
1181         if (test_and_clear_bit(AS_EIO, &mapping->flags))
1182                 ret2 = -EIO;
1183         if (!ret)
1184                 ret = ret2;
1185         return ret;
1186 }
1187
1188 static int f2fs_write_node_page(struct page *page,
1189                                 struct writeback_control *wbc)
1190 {
1191         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1192         nid_t nid;
1193         block_t new_addr;
1194         struct node_info ni;
1195
1196         if (sbi->por_doing)
1197                 goto redirty_out;
1198
1199         wait_on_page_writeback(page);
1200
1201         /* get old block addr of this node page */
1202         nid = nid_of_node(page);
1203         f2fs_bug_on(page->index != nid);
1204
1205         get_node_info(sbi, nid, &ni);
1206
1207         /* This page is already truncated */
1208         if (ni.blk_addr == NULL_ADDR) {
1209                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1210                 unlock_page(page);
1211                 return 0;
1212         }
1213
1214         if (wbc->for_reclaim)
1215                 goto redirty_out;
1216
1217         mutex_lock(&sbi->node_write);
1218         set_page_writeback(page);
1219         write_node_page(sbi, page, nid, ni.blk_addr, &new_addr);
1220         set_node_addr(sbi, &ni, new_addr);
1221         dec_page_count(sbi, F2FS_DIRTY_NODES);
1222         mutex_unlock(&sbi->node_write);
1223         unlock_page(page);
1224         return 0;
1225
1226 redirty_out:
1227         dec_page_count(sbi, F2FS_DIRTY_NODES);
1228         wbc->pages_skipped++;
1229         set_page_dirty(page);
1230         return AOP_WRITEPAGE_ACTIVATE;
1231 }
1232
1233 /*
1234  * It is very important to gather dirty pages and write at once, so that we can
1235  * submit a big bio without interfering other data writes.
1236  * Be default, 512 pages (2MB) * 3 node types, is more reasonable.
1237  */
1238 #define COLLECT_DIRTY_NODES     1536
1239 static int f2fs_write_node_pages(struct address_space *mapping,
1240                             struct writeback_control *wbc)
1241 {
1242         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1243         long nr_to_write = wbc->nr_to_write;
1244
1245         /* balancing f2fs's metadata in background */
1246         f2fs_balance_fs_bg(sbi);
1247
1248         /* collect a number of dirty node pages and write together */
1249         if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1250                 return 0;
1251
1252         /* if mounting is failed, skip writing node pages */
1253         wbc->nr_to_write = 3 * max_hw_blocks(sbi);
1254         sync_node_pages(sbi, 0, wbc);
1255         wbc->nr_to_write = nr_to_write - (3 * max_hw_blocks(sbi) -
1256                                                 wbc->nr_to_write);
1257         return 0;
1258 }
1259
1260 static int f2fs_set_node_page_dirty(struct page *page)
1261 {
1262         struct address_space *mapping = page->mapping;
1263         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1264
1265         trace_f2fs_set_page_dirty(page, NODE);
1266
1267         SetPageUptodate(page);
1268         if (!PageDirty(page)) {
1269                 __set_page_dirty_nobuffers(page);
1270                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1271                 SetPagePrivate(page);
1272                 return 1;
1273         }
1274         return 0;
1275 }
1276
1277 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1278                                       unsigned int length)
1279 {
1280         struct inode *inode = page->mapping->host;
1281         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1282         if (PageDirty(page))
1283                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1284         ClearPagePrivate(page);
1285 }
1286
1287 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1288 {
1289         ClearPagePrivate(page);
1290         return 1;
1291 }
1292
1293 /*
1294  * Structure of the f2fs node operations
1295  */
1296 const struct address_space_operations f2fs_node_aops = {
1297         .writepage      = f2fs_write_node_page,
1298         .writepages     = f2fs_write_node_pages,
1299         .set_page_dirty = f2fs_set_node_page_dirty,
1300         .invalidatepage = f2fs_invalidate_node_page,
1301         .releasepage    = f2fs_release_node_page,
1302 };
1303
1304 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1305 {
1306         struct list_head *this;
1307         struct free_nid *i;
1308         list_for_each(this, head) {
1309                 i = list_entry(this, struct free_nid, list);
1310                 if (i->nid == n)
1311                         return i;
1312         }
1313         return NULL;
1314 }
1315
1316 static void __del_from_free_nid_list(struct free_nid *i)
1317 {
1318         list_del(&i->list);
1319         kmem_cache_free(free_nid_slab, i);
1320 }
1321
1322 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build)
1323 {
1324         struct free_nid *i;
1325         struct nat_entry *ne;
1326         bool allocated = false;
1327
1328         if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1329                 return -1;
1330
1331         /* 0 nid should not be used */
1332         if (nid == 0)
1333                 return 0;
1334
1335         if (build) {
1336                 /* do not add allocated nids */
1337                 read_lock(&nm_i->nat_tree_lock);
1338                 ne = __lookup_nat_cache(nm_i, nid);
1339                 if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1340                         allocated = true;
1341                 read_unlock(&nm_i->nat_tree_lock);
1342                 if (allocated)
1343                         return 0;
1344         }
1345
1346         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1347         i->nid = nid;
1348         i->state = NID_NEW;
1349
1350         spin_lock(&nm_i->free_nid_list_lock);
1351         if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1352                 spin_unlock(&nm_i->free_nid_list_lock);
1353                 kmem_cache_free(free_nid_slab, i);
1354                 return 0;
1355         }
1356         list_add_tail(&i->list, &nm_i->free_nid_list);
1357         nm_i->fcnt++;
1358         spin_unlock(&nm_i->free_nid_list_lock);
1359         return 1;
1360 }
1361
1362 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1363 {
1364         struct free_nid *i;
1365         spin_lock(&nm_i->free_nid_list_lock);
1366         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1367         if (i && i->state == NID_NEW) {
1368                 __del_from_free_nid_list(i);
1369                 nm_i->fcnt--;
1370         }
1371         spin_unlock(&nm_i->free_nid_list_lock);
1372 }
1373
1374 static void scan_nat_page(struct f2fs_nm_info *nm_i,
1375                         struct page *nat_page, nid_t start_nid)
1376 {
1377         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1378         block_t blk_addr;
1379         int i;
1380
1381         i = start_nid % NAT_ENTRY_PER_BLOCK;
1382
1383         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1384
1385                 if (start_nid >= nm_i->max_nid)
1386                         break;
1387
1388                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1389                 f2fs_bug_on(blk_addr == NEW_ADDR);
1390                 if (blk_addr == NULL_ADDR) {
1391                         if (add_free_nid(nm_i, start_nid, true) < 0)
1392                                 break;
1393                 }
1394         }
1395 }
1396
1397 static void build_free_nids(struct f2fs_sb_info *sbi)
1398 {
1399         struct f2fs_nm_info *nm_i = NM_I(sbi);
1400         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1401         struct f2fs_summary_block *sum = curseg->sum_blk;
1402         int i = 0;
1403         nid_t nid = nm_i->next_scan_nid;
1404
1405         /* Enough entries */
1406         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1407                 return;
1408
1409         /* readahead nat pages to be scanned */
1410         ra_nat_pages(sbi, nid);
1411
1412         while (1) {
1413                 struct page *page = get_current_nat_page(sbi, nid);
1414
1415                 scan_nat_page(nm_i, page, nid);
1416                 f2fs_put_page(page, 1);
1417
1418                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1419                 if (nid >= nm_i->max_nid)
1420                         nid = 0;
1421
1422                 if (i++ == FREE_NID_PAGES)
1423                         break;
1424         }
1425
1426         /* go to the next free nat pages to find free nids abundantly */
1427         nm_i->next_scan_nid = nid;
1428
1429         /* find free nids from current sum_pages */
1430         mutex_lock(&curseg->curseg_mutex);
1431         for (i = 0; i < nats_in_cursum(sum); i++) {
1432                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1433                 nid = le32_to_cpu(nid_in_journal(sum, i));
1434                 if (addr == NULL_ADDR)
1435                         add_free_nid(nm_i, nid, true);
1436                 else
1437                         remove_free_nid(nm_i, nid);
1438         }
1439         mutex_unlock(&curseg->curseg_mutex);
1440 }
1441
1442 /*
1443  * If this function returns success, caller can obtain a new nid
1444  * from second parameter of this function.
1445  * The returned nid could be used ino as well as nid when inode is created.
1446  */
1447 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1448 {
1449         struct f2fs_nm_info *nm_i = NM_I(sbi);
1450         struct free_nid *i = NULL;
1451         struct list_head *this;
1452 retry:
1453         if (sbi->total_valid_node_count + 1 >= nm_i->max_nid)
1454                 return false;
1455
1456         spin_lock(&nm_i->free_nid_list_lock);
1457
1458         /* We should not use stale free nids created by build_free_nids */
1459         if (nm_i->fcnt && !sbi->on_build_free_nids) {
1460                 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1461                 list_for_each(this, &nm_i->free_nid_list) {
1462                         i = list_entry(this, struct free_nid, list);
1463                         if (i->state == NID_NEW)
1464                                 break;
1465                 }
1466
1467                 f2fs_bug_on(i->state != NID_NEW);
1468                 *nid = i->nid;
1469                 i->state = NID_ALLOC;
1470                 nm_i->fcnt--;
1471                 spin_unlock(&nm_i->free_nid_list_lock);
1472                 return true;
1473         }
1474         spin_unlock(&nm_i->free_nid_list_lock);
1475
1476         /* Let's scan nat pages and its caches to get free nids */
1477         mutex_lock(&nm_i->build_lock);
1478         sbi->on_build_free_nids = true;
1479         build_free_nids(sbi);
1480         sbi->on_build_free_nids = false;
1481         mutex_unlock(&nm_i->build_lock);
1482         goto retry;
1483 }
1484
1485 /*
1486  * alloc_nid() should be called prior to this function.
1487  */
1488 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1489 {
1490         struct f2fs_nm_info *nm_i = NM_I(sbi);
1491         struct free_nid *i;
1492
1493         spin_lock(&nm_i->free_nid_list_lock);
1494         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1495         f2fs_bug_on(!i || i->state != NID_ALLOC);
1496         __del_from_free_nid_list(i);
1497         spin_unlock(&nm_i->free_nid_list_lock);
1498 }
1499
1500 /*
1501  * alloc_nid() should be called prior to this function.
1502  */
1503 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1504 {
1505         struct f2fs_nm_info *nm_i = NM_I(sbi);
1506         struct free_nid *i;
1507
1508         if (!nid)
1509                 return;
1510
1511         spin_lock(&nm_i->free_nid_list_lock);
1512         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1513         f2fs_bug_on(!i || i->state != NID_ALLOC);
1514         if (nm_i->fcnt > 2 * MAX_FREE_NIDS) {
1515                 __del_from_free_nid_list(i);
1516         } else {
1517                 i->state = NID_NEW;
1518                 nm_i->fcnt++;
1519         }
1520         spin_unlock(&nm_i->free_nid_list_lock);
1521 }
1522
1523 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1524                 struct f2fs_summary *sum, struct node_info *ni,
1525                 block_t new_blkaddr)
1526 {
1527         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1528         set_node_addr(sbi, ni, new_blkaddr);
1529         clear_node_page_dirty(page);
1530 }
1531
1532 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1533 {
1534         struct address_space *mapping = sbi->node_inode->i_mapping;
1535         struct f2fs_node *src, *dst;
1536         nid_t ino = ino_of_node(page);
1537         struct node_info old_ni, new_ni;
1538         struct page *ipage;
1539
1540         ipage = grab_cache_page(mapping, ino);
1541         if (!ipage)
1542                 return -ENOMEM;
1543
1544         /* Should not use this inode  from free nid list */
1545         remove_free_nid(NM_I(sbi), ino);
1546
1547         get_node_info(sbi, ino, &old_ni);
1548         SetPageUptodate(ipage);
1549         fill_node_footer(ipage, ino, ino, 0, true);
1550
1551         src = F2FS_NODE(page);
1552         dst = F2FS_NODE(ipage);
1553
1554         memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i);
1555         dst->i.i_size = 0;
1556         dst->i.i_blocks = cpu_to_le64(1);
1557         dst->i.i_links = cpu_to_le32(1);
1558         dst->i.i_xattr_nid = 0;
1559
1560         new_ni = old_ni;
1561         new_ni.ino = ino;
1562
1563         if (!inc_valid_node_count(sbi, NULL))
1564                 WARN_ON(1);
1565         set_node_addr(sbi, &new_ni, NEW_ADDR);
1566         inc_valid_inode_count(sbi);
1567         f2fs_put_page(ipage, 1);
1568         return 0;
1569 }
1570
1571 int restore_node_summary(struct f2fs_sb_info *sbi,
1572                         unsigned int segno, struct f2fs_summary_block *sum)
1573 {
1574         struct f2fs_node *rn;
1575         struct f2fs_summary *sum_entry;
1576         struct page *page;
1577         block_t addr;
1578         int i, last_offset;
1579
1580         /* alloc temporal page for read node */
1581         page = alloc_page(GFP_NOFS | __GFP_ZERO);
1582         if (!page)
1583                 return -ENOMEM;
1584         lock_page(page);
1585
1586         /* scan the node segment */
1587         last_offset = sbi->blocks_per_seg;
1588         addr = START_BLOCK(sbi, segno);
1589         sum_entry = &sum->entries[0];
1590
1591         for (i = 0; i < last_offset; i++, sum_entry++) {
1592                 /*
1593                  * In order to read next node page,
1594                  * we must clear PageUptodate flag.
1595                  */
1596                 ClearPageUptodate(page);
1597
1598                 if (f2fs_readpage(sbi, page, addr, READ_SYNC))
1599                         goto out;
1600
1601                 lock_page(page);
1602                 rn = F2FS_NODE(page);
1603                 sum_entry->nid = rn->footer.nid;
1604                 sum_entry->version = 0;
1605                 sum_entry->ofs_in_node = 0;
1606                 addr++;
1607         }
1608         unlock_page(page);
1609 out:
1610         __free_pages(page, 0);
1611         return 0;
1612 }
1613
1614 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1615 {
1616         struct f2fs_nm_info *nm_i = NM_I(sbi);
1617         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1618         struct f2fs_summary_block *sum = curseg->sum_blk;
1619         int i;
1620
1621         mutex_lock(&curseg->curseg_mutex);
1622
1623         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1624                 mutex_unlock(&curseg->curseg_mutex);
1625                 return false;
1626         }
1627
1628         for (i = 0; i < nats_in_cursum(sum); i++) {
1629                 struct nat_entry *ne;
1630                 struct f2fs_nat_entry raw_ne;
1631                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1632
1633                 raw_ne = nat_in_journal(sum, i);
1634 retry:
1635                 write_lock(&nm_i->nat_tree_lock);
1636                 ne = __lookup_nat_cache(nm_i, nid);
1637                 if (ne) {
1638                         __set_nat_cache_dirty(nm_i, ne);
1639                         write_unlock(&nm_i->nat_tree_lock);
1640                         continue;
1641                 }
1642                 ne = grab_nat_entry(nm_i, nid);
1643                 if (!ne) {
1644                         write_unlock(&nm_i->nat_tree_lock);
1645                         goto retry;
1646                 }
1647                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1648                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1649                 nat_set_version(ne, raw_ne.version);
1650                 __set_nat_cache_dirty(nm_i, ne);
1651                 write_unlock(&nm_i->nat_tree_lock);
1652         }
1653         update_nats_in_cursum(sum, -i);
1654         mutex_unlock(&curseg->curseg_mutex);
1655         return true;
1656 }
1657
1658 /*
1659  * This function is called during the checkpointing process.
1660  */
1661 void flush_nat_entries(struct f2fs_sb_info *sbi)
1662 {
1663         struct f2fs_nm_info *nm_i = NM_I(sbi);
1664         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1665         struct f2fs_summary_block *sum = curseg->sum_blk;
1666         struct list_head *cur, *n;
1667         struct page *page = NULL;
1668         struct f2fs_nat_block *nat_blk = NULL;
1669         nid_t start_nid = 0, end_nid = 0;
1670         bool flushed;
1671
1672         flushed = flush_nats_in_journal(sbi);
1673
1674         if (!flushed)
1675                 mutex_lock(&curseg->curseg_mutex);
1676
1677         /* 1) flush dirty nat caches */
1678         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1679                 struct nat_entry *ne;
1680                 nid_t nid;
1681                 struct f2fs_nat_entry raw_ne;
1682                 int offset = -1;
1683                 block_t new_blkaddr;
1684
1685                 ne = list_entry(cur, struct nat_entry, list);
1686                 nid = nat_get_nid(ne);
1687
1688                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1689                         continue;
1690                 if (flushed)
1691                         goto to_nat_page;
1692
1693                 /* if there is room for nat enries in curseg->sumpage */
1694                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1695                 if (offset >= 0) {
1696                         raw_ne = nat_in_journal(sum, offset);
1697                         goto flush_now;
1698                 }
1699 to_nat_page:
1700                 if (!page || (start_nid > nid || nid > end_nid)) {
1701                         if (page) {
1702                                 f2fs_put_page(page, 1);
1703                                 page = NULL;
1704                         }
1705                         start_nid = START_NID(nid);
1706                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1707
1708                         /*
1709                          * get nat block with dirty flag, increased reference
1710                          * count, mapped and lock
1711                          */
1712                         page = get_next_nat_page(sbi, start_nid);
1713                         nat_blk = page_address(page);
1714                 }
1715
1716                 f2fs_bug_on(!nat_blk);
1717                 raw_ne = nat_blk->entries[nid - start_nid];
1718 flush_now:
1719                 new_blkaddr = nat_get_blkaddr(ne);
1720
1721                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1722                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1723                 raw_ne.version = nat_get_version(ne);
1724
1725                 if (offset < 0) {
1726                         nat_blk->entries[nid - start_nid] = raw_ne;
1727                 } else {
1728                         nat_in_journal(sum, offset) = raw_ne;
1729                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1730                 }
1731
1732                 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1733                                 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1734                         write_lock(&nm_i->nat_tree_lock);
1735                         __del_from_nat_cache(nm_i, ne);
1736                         write_unlock(&nm_i->nat_tree_lock);
1737                 } else {
1738                         write_lock(&nm_i->nat_tree_lock);
1739                         __clear_nat_cache_dirty(nm_i, ne);
1740                         ne->checkpointed = true;
1741                         write_unlock(&nm_i->nat_tree_lock);
1742                 }
1743         }
1744         if (!flushed)
1745                 mutex_unlock(&curseg->curseg_mutex);
1746         f2fs_put_page(page, 1);
1747
1748         /* 2) shrink nat caches if necessary */
1749         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1750 }
1751
1752 static int init_node_manager(struct f2fs_sb_info *sbi)
1753 {
1754         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1755         struct f2fs_nm_info *nm_i = NM_I(sbi);
1756         unsigned char *version_bitmap;
1757         unsigned int nat_segs, nat_blocks;
1758
1759         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1760
1761         /* segment_count_nat includes pair segment so divide to 2. */
1762         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1763         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1764         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1765         nm_i->fcnt = 0;
1766         nm_i->nat_cnt = 0;
1767
1768         INIT_LIST_HEAD(&nm_i->free_nid_list);
1769         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1770         INIT_LIST_HEAD(&nm_i->nat_entries);
1771         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1772
1773         mutex_init(&nm_i->build_lock);
1774         spin_lock_init(&nm_i->free_nid_list_lock);
1775         rwlock_init(&nm_i->nat_tree_lock);
1776
1777         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1778         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1779         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1780         if (!version_bitmap)
1781                 return -EFAULT;
1782
1783         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1784                                         GFP_KERNEL);
1785         if (!nm_i->nat_bitmap)
1786                 return -ENOMEM;
1787         return 0;
1788 }
1789
1790 int build_node_manager(struct f2fs_sb_info *sbi)
1791 {
1792         int err;
1793
1794         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1795         if (!sbi->nm_info)
1796                 return -ENOMEM;
1797
1798         err = init_node_manager(sbi);
1799         if (err)
1800                 return err;
1801
1802         build_free_nids(sbi);
1803         return 0;
1804 }
1805
1806 void destroy_node_manager(struct f2fs_sb_info *sbi)
1807 {
1808         struct f2fs_nm_info *nm_i = NM_I(sbi);
1809         struct free_nid *i, *next_i;
1810         struct nat_entry *natvec[NATVEC_SIZE];
1811         nid_t nid = 0;
1812         unsigned int found;
1813
1814         if (!nm_i)
1815                 return;
1816
1817         /* destroy free nid list */
1818         spin_lock(&nm_i->free_nid_list_lock);
1819         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1820                 f2fs_bug_on(i->state == NID_ALLOC);
1821                 __del_from_free_nid_list(i);
1822                 nm_i->fcnt--;
1823         }
1824         f2fs_bug_on(nm_i->fcnt);
1825         spin_unlock(&nm_i->free_nid_list_lock);
1826
1827         /* destroy nat cache */
1828         write_lock(&nm_i->nat_tree_lock);
1829         while ((found = __gang_lookup_nat_cache(nm_i,
1830                                         nid, NATVEC_SIZE, natvec))) {
1831                 unsigned idx;
1832                 for (idx = 0; idx < found; idx++) {
1833                         struct nat_entry *e = natvec[idx];
1834                         nid = nat_get_nid(e) + 1;
1835                         __del_from_nat_cache(nm_i, e);
1836                 }
1837         }
1838         f2fs_bug_on(nm_i->nat_cnt);
1839         write_unlock(&nm_i->nat_tree_lock);
1840
1841         kfree(nm_i->nat_bitmap);
1842         sbi->nm_info = NULL;
1843         kfree(nm_i);
1844 }
1845
1846 int __init create_node_manager_caches(void)
1847 {
1848         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1849                         sizeof(struct nat_entry), NULL);
1850         if (!nat_entry_slab)
1851                 return -ENOMEM;
1852
1853         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1854                         sizeof(struct free_nid), NULL);
1855         if (!free_nid_slab) {
1856                 kmem_cache_destroy(nat_entry_slab);
1857                 return -ENOMEM;
1858         }
1859         return 0;
1860 }
1861
1862 void destroy_node_manager_caches(void)
1863 {
1864         kmem_cache_destroy(free_nid_slab);
1865         kmem_cache_destroy(nat_entry_slab);
1866 }