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 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
  25
  26 static struct kmem_cache *nat_entry_slab;
  27 static struct kmem_cache *free_nid_slab;
  28
  29 static inline bool available_free_memory(struct f2fs_nm_info *nm_i, int type)
  30 {
  31         struct sysinfo val;
  32         unsigned long mem_size = 0;
  33
  34         si_meminfo(&val);
  35         if (type == FREE_NIDS)
  36                 mem_size = nm_i->fcnt * sizeof(struct free_nid);
  37         else if (type == NAT_ENTRIES)
  38                 mem_size += nm_i->nat_cnt * sizeof(struct nat_entry);
  39         mem_size >>= 12;
  40
  41         /* give 50:50 memory for free nids and nat caches respectively */
  42         return (mem_size < ((val.totalram * nm_i->ram_thresh) >> 11));
  43 }
  44
  45 static void clear_node_page_dirty(struct page *page)
  46 {
  47         struct address_space *mapping = page->mapping;
  48         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
  49         unsigned int long flags;
  50
  51         if (PageDirty(page)) {
  52                 spin_lock_irqsave(&mapping->tree_lock, flags);
  53                 radix_tree_tag_clear(&mapping->page_tree,
  54                                 page_index(page),
  55                                 PAGECACHE_TAG_DIRTY);
  56                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  57
  58                 clear_page_dirty_for_io(page);
  59                 dec_page_count(sbi, F2FS_DIRTY_NODES);
  60         }
  61         ClearPageUptodate(page);
  62 }
  63
  64 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  65 {
  66         pgoff_t index = current_nat_addr(sbi, nid);
  67         return get_meta_page(sbi, index);
  68 }
  69
  70 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  71 {
  72         struct page *src_page;
  73         struct page *dst_page;
  74         pgoff_t src_off;
  75         pgoff_t dst_off;
  76         void *src_addr;
  77         void *dst_addr;
  78         struct f2fs_nm_info *nm_i = NM_I(sbi);
  79
  80         src_off = current_nat_addr(sbi, nid);
  81         dst_off = next_nat_addr(sbi, src_off);
  82
  83         /* get current nat block page with lock */
  84         src_page = get_meta_page(sbi, src_off);
  85
  86         /* Dirty src_page means that it is already the new target NAT page. */
  87         if (PageDirty(src_page))
  88                 return src_page;
  89
  90         dst_page = grab_meta_page(sbi, dst_off);
  91
  92         src_addr = page_address(src_page);
  93         dst_addr = page_address(dst_page);
  94         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
  95         set_page_dirty(dst_page);
  96         f2fs_put_page(src_page, 1);
  97
  98         set_to_next_nat(nm_i, nid);
  99
 100         return dst_page;
 101 }
 102
 103 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 104 {
 105         return radix_tree_lookup(&nm_i->nat_root, n);
 106 }
 107
 108 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 109                 nid_t start, unsigned int nr, struct nat_entry **ep)
 110 {
 111         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 112 }
 113
 114 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 115 {
 116         list_del(&e->list);
 117         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 118         nm_i->nat_cnt--;
 119         kmem_cache_free(nat_entry_slab, e);
 120 }
 121
 122 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 123 {
 124         struct f2fs_nm_info *nm_i = NM_I(sbi);
 125         struct nat_entry *e;
 126         int is_cp = 1;
 127
 128         read_lock(&nm_i->nat_tree_lock);
 129         e = __lookup_nat_cache(nm_i, nid);
 130         if (e && !e->checkpointed)
 131                 is_cp = 0;
 132         read_unlock(&nm_i->nat_tree_lock);
 133         return is_cp;
 134 }
 135
 136 bool fsync_mark_done(struct f2fs_sb_info *sbi, nid_t nid)
 137 {
 138         struct f2fs_nm_info *nm_i = NM_I(sbi);
 139         struct nat_entry *e;
 140         bool fsync_done = false;
 141
 142         read_lock(&nm_i->nat_tree_lock);
 143         e = __lookup_nat_cache(nm_i, nid);
 144         if (e)
 145                 fsync_done = e->fsync_done;
 146         read_unlock(&nm_i->nat_tree_lock);
 147         return fsync_done;
 148 }
 149
 150 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 151 {
 152         struct nat_entry *new;
 153
 154         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 155         if (!new)
 156                 return NULL;
 157         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 158                 kmem_cache_free(nat_entry_slab, new);
 159                 return NULL;
 160         }
 161         memset(new, 0, sizeof(struct nat_entry));
 162         nat_set_nid(new, nid);
 163         new->checkpointed = true;
 164         list_add_tail(&new->list, &nm_i->nat_entries);
 165         nm_i->nat_cnt++;
 166         return new;
 167 }
 168
 169 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 170                                                 struct f2fs_nat_entry *ne)
 171 {
 172         struct nat_entry *e;
 173 retry:
 174         write_lock(&nm_i->nat_tree_lock);
 175         e = __lookup_nat_cache(nm_i, nid);
 176         if (!e) {
 177                 e = grab_nat_entry(nm_i, nid);
 178                 if (!e) {
 179                         write_unlock(&nm_i->nat_tree_lock);
 180                         goto retry;
 181                 }
 182                 node_info_from_raw_nat(&e->ni, ne);
 183         }
 184         write_unlock(&nm_i->nat_tree_lock);
 185 }
 186
 187 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 188                         block_t new_blkaddr, bool fsync_done)
 189 {
 190         struct f2fs_nm_info *nm_i = NM_I(sbi);
 191         struct nat_entry *e;
 192 retry:
 193         write_lock(&nm_i->nat_tree_lock);
 194         e = __lookup_nat_cache(nm_i, ni->nid);
 195         if (!e) {
 196                 e = grab_nat_entry(nm_i, ni->nid);
 197                 if (!e) {
 198                         write_unlock(&nm_i->nat_tree_lock);
 199                         goto retry;
 200                 }
 201                 e->ni = *ni;
 202                 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
 203         } else if (new_blkaddr == NEW_ADDR) {
 204                 /*
 205                  * when nid is reallocated,
 206                  * previous nat entry can be remained in nat cache.
 207                  * So, reinitialize it with new information.
 208                  */
 209                 e->ni = *ni;
 210                 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
 211         }
 212
 213         /* sanity check */
 214         f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
 215         f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
 216                         new_blkaddr == NULL_ADDR);
 217         f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
 218                         new_blkaddr == NEW_ADDR);
 219         f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
 220                         nat_get_blkaddr(e) != NULL_ADDR &&
 221                         new_blkaddr == NEW_ADDR);
 222
 223         /* increament version no as node is removed */
 224         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 225                 unsigned char version = nat_get_version(e);
 226                 nat_set_version(e, inc_node_version(version));
 227         }
 228
 229         /* change address */
 230         nat_set_blkaddr(e, new_blkaddr);
 231         __set_nat_cache_dirty(nm_i, e);
 232
 233         /* update fsync_mark if its inode nat entry is still alive */
 234         e = __lookup_nat_cache(nm_i, ni->ino);
 235         if (e)
 236                 e->fsync_done = fsync_done;
 237         write_unlock(&nm_i->nat_tree_lock);
 238 }
 239
 240 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 241 {
 242         struct f2fs_nm_info *nm_i = NM_I(sbi);
 243
 244         if (available_free_memory(nm_i, NAT_ENTRIES))
 245                 return 0;
 246
 247         write_lock(&nm_i->nat_tree_lock);
 248         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 249                 struct nat_entry *ne;
 250                 ne = list_first_entry(&nm_i->nat_entries,
 251                                         struct nat_entry, list);
 252                 __del_from_nat_cache(nm_i, ne);
 253                 nr_shrink--;
 254         }
 255         write_unlock(&nm_i->nat_tree_lock);
 256         return nr_shrink;
 257 }
 258
 259 /*
 260  * This function returns always success
 261  */
 262 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 263 {
 264         struct f2fs_nm_info *nm_i = NM_I(sbi);
 265         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 266         struct f2fs_summary_block *sum = curseg->sum_blk;
 267         nid_t start_nid = START_NID(nid);
 268         struct f2fs_nat_block *nat_blk;
 269         struct page *page = NULL;
 270         struct f2fs_nat_entry ne;
 271         struct nat_entry *e;
 272         int i;
 273
 274         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 275         ni->nid = nid;
 276
 277         /* Check nat cache */
 278         read_lock(&nm_i->nat_tree_lock);
 279         e = __lookup_nat_cache(nm_i, nid);
 280         if (e) {
 281                 ni->ino = nat_get_ino(e);
 282                 ni->blk_addr = nat_get_blkaddr(e);
 283                 ni->version = nat_get_version(e);
 284         }
 285         read_unlock(&nm_i->nat_tree_lock);
 286         if (e)
 287                 return;
 288
 289         /* Check current segment summary */
 290         mutex_lock(&curseg->curseg_mutex);
 291         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 292         if (i >= 0) {
 293                 ne = nat_in_journal(sum, i);
 294                 node_info_from_raw_nat(ni, &ne);
 295         }
 296         mutex_unlock(&curseg->curseg_mutex);
 297         if (i >= 0)
 298                 goto cache;
 299
 300         /* Fill node_info from nat page */
 301         page = get_current_nat_page(sbi, start_nid);
 302         nat_blk = (struct f2fs_nat_block *)page_address(page);
 303         ne = nat_blk->entries[nid - start_nid];
 304         node_info_from_raw_nat(ni, &ne);
 305         f2fs_put_page(page, 1);
 306 cache:
 307         /* cache nat entry */
 308         cache_nat_entry(NM_I(sbi), nid, &ne);
 309 }
 310
 311 /*
 312  * The maximum depth is four.
 313  * Offset[0] will have raw inode offset.
 314  */
 315 static int get_node_path(struct f2fs_inode_info *fi, long block,
 316                                 int offset[4], unsigned int noffset[4])
 317 {
 318         const long direct_index = ADDRS_PER_INODE(fi);
 319         const long direct_blks = ADDRS_PER_BLOCK;
 320         const long dptrs_per_blk = NIDS_PER_BLOCK;
 321         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 322         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 323         int n = 0;
 324         int level = 0;
 325
 326         noffset[0] = 0;
 327
 328         if (block < direct_index) {
 329                 offset[n] = block;
 330                 goto got;
 331         }
 332         block -= direct_index;
 333         if (block < direct_blks) {
 334                 offset[n++] = NODE_DIR1_BLOCK;
 335                 noffset[n] = 1;
 336                 offset[n] = block;
 337                 level = 1;
 338                 goto got;
 339         }
 340         block -= direct_blks;
 341         if (block < direct_blks) {
 342                 offset[n++] = NODE_DIR2_BLOCK;
 343                 noffset[n] = 2;
 344                 offset[n] = block;
 345                 level = 1;
 346                 goto got;
 347         }
 348         block -= direct_blks;
 349         if (block < indirect_blks) {
 350                 offset[n++] = NODE_IND1_BLOCK;
 351                 noffset[n] = 3;
 352                 offset[n++] = block / direct_blks;
 353                 noffset[n] = 4 + offset[n - 1];
 354                 offset[n] = block % direct_blks;
 355                 level = 2;
 356                 goto got;
 357         }
 358         block -= indirect_blks;
 359         if (block < indirect_blks) {
 360                 offset[n++] = NODE_IND2_BLOCK;
 361                 noffset[n] = 4 + dptrs_per_blk;
 362                 offset[n++] = block / direct_blks;
 363                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 364                 offset[n] = block % direct_blks;
 365                 level = 2;
 366                 goto got;
 367         }
 368         block -= indirect_blks;
 369         if (block < dindirect_blks) {
 370                 offset[n++] = NODE_DIND_BLOCK;
 371                 noffset[n] = 5 + (dptrs_per_blk * 2);
 372                 offset[n++] = block / indirect_blks;
 373                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 374                               offset[n - 1] * (dptrs_per_blk + 1);
 375                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 376                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 377                               offset[n - 2] * (dptrs_per_blk + 1) +
 378                               offset[n - 1];
 379                 offset[n] = block % direct_blks;
 380                 level = 3;
 381                 goto got;
 382         } else {
 383                 BUG();
 384         }
 385 got:
 386         return level;
 387 }
 388
 389 /*
 390  * Caller should call f2fs_put_dnode(dn).
 391  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 392  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 393  * In the case of RDONLY_NODE, we don't need to care about mutex.
 394  */
 395 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 396 {
 397         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 398         struct page *npage[4];
 399         struct page *parent;
 400         int offset[4];
 401         unsigned int noffset[4];
 402         nid_t nids[4];
 403         int level, i;
 404         int err = 0;
 405
 406         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 407
 408         nids[0] = dn->inode->i_ino;
 409         npage[0] = dn->inode_page;
 410
 411         if (!npage[0]) {
 412                 npage[0] = get_node_page(sbi, nids[0]);
 413                 if (IS_ERR(npage[0]))
 414                         return PTR_ERR(npage[0]);
 415         }
 416         parent = npage[0];
 417         if (level != 0)
 418                 nids[1] = get_nid(parent, offset[0], true);
 419         dn->inode_page = npage[0];
 420         dn->inode_page_locked = true;
 421
 422         /* get indirect or direct nodes */
 423         for (i = 1; i <= level; i++) {
 424                 bool done = false;
 425
 426                 if (!nids[i] && mode == ALLOC_NODE) {
 427                         /* alloc new node */
 428                         if (!alloc_nid(sbi, &(nids[i]))) {
 429                                 err = -ENOSPC;
 430                                 goto release_pages;
 431                         }
 432
 433                         dn->nid = nids[i];
 434                         npage[i] = new_node_page(dn, noffset[i], NULL);
 435                         if (IS_ERR(npage[i])) {
 436                                 alloc_nid_failed(sbi, nids[i]);
 437                                 err = PTR_ERR(npage[i]);
 438                                 goto release_pages;
 439                         }
 440
 441                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 442                         alloc_nid_done(sbi, nids[i]);
 443                         done = true;
 444                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 445                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 446                         if (IS_ERR(npage[i])) {
 447                                 err = PTR_ERR(npage[i]);
 448                                 goto release_pages;
 449                         }
 450                         done = true;
 451                 }
 452                 if (i == 1) {
 453                         dn->inode_page_locked = false;
 454                         unlock_page(parent);
 455                 } else {
 456                         f2fs_put_page(parent, 1);
 457                 }
 458
 459                 if (!done) {
 460                         npage[i] = get_node_page(sbi, nids[i]);
 461                         if (IS_ERR(npage[i])) {
 462                                 err = PTR_ERR(npage[i]);
 463                                 f2fs_put_page(npage[0], 0);
 464                                 goto release_out;
 465                         }
 466                 }
 467                 if (i < level) {
 468                         parent = npage[i];
 469                         nids[i + 1] = get_nid(parent, offset[i], false);
 470                 }
 471         }
 472         dn->nid = nids[level];
 473         dn->ofs_in_node = offset[level];
 474         dn->node_page = npage[level];
 475         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 476         return 0;
 477
 478 release_pages:
 479         f2fs_put_page(parent, 1);
 480         if (i > 1)
 481                 f2fs_put_page(npage[0], 0);
 482 release_out:
 483         dn->inode_page = NULL;
 484         dn->node_page = NULL;
 485         return err;
 486 }
 487
 488 static void truncate_node(struct dnode_of_data *dn)
 489 {
 490         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 491         struct node_info ni;
 492
 493         get_node_info(sbi, dn->nid, &ni);
 494         if (dn->inode->i_blocks == 0) {
 495                 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
 496                 goto invalidate;
 497         }
 498         f2fs_bug_on(ni.blk_addr == NULL_ADDR);
 499
 500         /* Deallocate node address */
 501         invalidate_blocks(sbi, ni.blk_addr);
 502         dec_valid_node_count(sbi, dn->inode);
 503         set_node_addr(sbi, &ni, NULL_ADDR, false);
 504
 505         if (dn->nid == dn->inode->i_ino) {
 506                 remove_orphan_inode(sbi, dn->nid);
 507                 dec_valid_inode_count(sbi);
 508         } else {
 509                 sync_inode_page(dn);
 510         }
 511 invalidate:
 512         clear_node_page_dirty(dn->node_page);
 513         F2FS_SET_SB_DIRT(sbi);
 514
 515         f2fs_put_page(dn->node_page, 1);
 516
 517         invalidate_mapping_pages(NODE_MAPPING(sbi),
 518                         dn->node_page->index, dn->node_page->index);
 519
 520         dn->node_page = NULL;
 521         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 522 }
 523
 524 static int truncate_dnode(struct dnode_of_data *dn)
 525 {
 526         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 527         struct page *page;
 528
 529         if (dn->nid == 0)
 530                 return 1;
 531
 532         /* get direct node */
 533         page = get_node_page(sbi, dn->nid);
 534         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 535                 return 1;
 536         else if (IS_ERR(page))
 537                 return PTR_ERR(page);
 538
 539         /* Make dnode_of_data for parameter */
 540         dn->node_page = page;
 541         dn->ofs_in_node = 0;
 542         truncate_data_blocks(dn);
 543         truncate_node(dn);
 544         return 1;
 545 }
 546
 547 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 548                                                 int ofs, int depth)
 549 {
 550         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 551         struct dnode_of_data rdn = *dn;
 552         struct page *page;
 553         struct f2fs_node *rn;
 554         nid_t child_nid;
 555         unsigned int child_nofs;
 556         int freed = 0;
 557         int i, ret;
 558
 559         if (dn->nid == 0)
 560                 return NIDS_PER_BLOCK + 1;
 561
 562         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 563
 564         page = get_node_page(sbi, dn->nid);
 565         if (IS_ERR(page)) {
 566                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 567                 return PTR_ERR(page);
 568         }
 569
 570         rn = F2FS_NODE(page);
 571         if (depth < 3) {
 572                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 573                         child_nid = le32_to_cpu(rn->in.nid[i]);
 574                         if (child_nid == 0)
 575                                 continue;
 576                         rdn.nid = child_nid;
 577                         ret = truncate_dnode(&rdn);
 578                         if (ret < 0)
 579                                 goto out_err;
 580                         set_nid(page, i, 0, false);
 581                 }
 582         } else {
 583                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 584                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 585                         child_nid = le32_to_cpu(rn->in.nid[i]);
 586                         if (child_nid == 0) {
 587                                 child_nofs += NIDS_PER_BLOCK + 1;
 588                                 continue;
 589                         }
 590                         rdn.nid = child_nid;
 591                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 592                         if (ret == (NIDS_PER_BLOCK + 1)) {
 593                                 set_nid(page, i, 0, false);
 594                                 child_nofs += ret;
 595                         } else if (ret < 0 && ret != -ENOENT) {
 596                                 goto out_err;
 597                         }
 598                 }
 599                 freed = child_nofs;
 600         }
 601
 602         if (!ofs) {
 603                 /* remove current indirect node */
 604                 dn->node_page = page;
 605                 truncate_node(dn);
 606                 freed++;
 607         } else {
 608                 f2fs_put_page(page, 1);
 609         }
 610         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 611         return freed;
 612
 613 out_err:
 614         f2fs_put_page(page, 1);
 615         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 616         return ret;
 617 }
 618
 619 static int truncate_partial_nodes(struct dnode_of_data *dn,
 620                         struct f2fs_inode *ri, int *offset, int depth)
 621 {
 622         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 623         struct page *pages[2];
 624         nid_t nid[3];
 625         nid_t child_nid;
 626         int err = 0;
 627         int i;
 628         int idx = depth - 2;
 629
 630         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 631         if (!nid[0])
 632                 return 0;
 633
 634         /* get indirect nodes in the path */
 635         for (i = 0; i < idx + 1; i++) {
 636                 /* refernece count'll be increased */
 637                 pages[i] = get_node_page(sbi, nid[i]);
 638                 if (IS_ERR(pages[i])) {
 639                         err = PTR_ERR(pages[i]);
 640                         idx = i - 1;
 641                         goto fail;
 642                 }
 643                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 644         }
 645
 646         /* free direct nodes linked to a partial indirect node */
 647         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 648                 child_nid = get_nid(pages[idx], i, false);
 649                 if (!child_nid)
 650                         continue;
 651                 dn->nid = child_nid;
 652                 err = truncate_dnode(dn);
 653                 if (err < 0)
 654                         goto fail;
 655                 set_nid(pages[idx], i, 0, false);
 656         }
 657
 658         if (offset[idx + 1] == 0) {
 659                 dn->node_page = pages[idx];
 660                 dn->nid = nid[idx];
 661                 truncate_node(dn);
 662         } else {
 663                 f2fs_put_page(pages[idx], 1);
 664         }
 665         offset[idx]++;
 666         offset[idx + 1] = 0;
 667         idx--;
 668 fail:
 669         for (i = idx; i >= 0; i--)
 670                 f2fs_put_page(pages[i], 1);
 671
 672         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 673
 674         return err;
 675 }
 676
 677 /*
 678  * All the block addresses of data and nodes should be nullified.
 679  */
 680 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 681 {
 682         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 683         int err = 0, cont = 1;
 684         int level, offset[4], noffset[4];
 685         unsigned int nofs = 0;
 686         struct f2fs_inode *ri;
 687         struct dnode_of_data dn;
 688         struct page *page;
 689
 690         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 691
 692         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 693 restart:
 694         page = get_node_page(sbi, inode->i_ino);
 695         if (IS_ERR(page)) {
 696                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 697                 return PTR_ERR(page);
 698         }
 699
 700         set_new_dnode(&dn, inode, page, NULL, 0);
 701         unlock_page(page);
 702
 703         ri = F2FS_INODE(page);
 704         switch (level) {
 705         case 0:
 706         case 1:
 707                 nofs = noffset[1];
 708                 break;
 709         case 2:
 710                 nofs = noffset[1];
 711                 if (!offset[level - 1])
 712                         goto skip_partial;
 713                 err = truncate_partial_nodes(&dn, ri, offset, level);
 714                 if (err < 0 && err != -ENOENT)
 715                         goto fail;
 716                 nofs += 1 + NIDS_PER_BLOCK;
 717                 break;
 718         case 3:
 719                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 720                 if (!offset[level - 1])
 721                         goto skip_partial;
 722                 err = truncate_partial_nodes(&dn, ri, offset, level);
 723                 if (err < 0 && err != -ENOENT)
 724                         goto fail;
 725                 break;
 726         default:
 727                 BUG();
 728         }
 729
 730 skip_partial:
 731         while (cont) {
 732                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 733                 switch (offset[0]) {
 734                 case NODE_DIR1_BLOCK:
 735                 case NODE_DIR2_BLOCK:
 736                         err = truncate_dnode(&dn);
 737                         break;
 738
 739                 case NODE_IND1_BLOCK:
 740                 case NODE_IND2_BLOCK:
 741                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 742                         break;
 743
 744                 case NODE_DIND_BLOCK:
 745                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 746                         cont = 0;
 747                         break;
 748
 749                 default:
 750                         BUG();
 751                 }
 752                 if (err < 0 && err != -ENOENT)
 753                         goto fail;
 754                 if (offset[1] == 0 &&
 755                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 756                         lock_page(page);
 757                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 758                                 f2fs_put_page(page, 1);
 759                                 goto restart;
 760                         }
 761                         f2fs_wait_on_page_writeback(page, NODE);
 762                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 763                         set_page_dirty(page);
 764                         unlock_page(page);
 765                 }
 766                 offset[1] = 0;
 767                 offset[0]++;
 768                 nofs += err;
 769         }
 770 fail:
 771         f2fs_put_page(page, 0);
 772         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 773         return err > 0 ? 0 : err;
 774 }
 775
 776 int truncate_xattr_node(struct inode *inode, struct page *page)
 777 {
 778         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 779         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 780         struct dnode_of_data dn;
 781         struct page *npage;
 782
 783         if (!nid)
 784                 return 0;
 785
 786         npage = get_node_page(sbi, nid);
 787         if (IS_ERR(npage))
 788                 return PTR_ERR(npage);
 789
 790         F2FS_I(inode)->i_xattr_nid = 0;
 791
 792         /* need to do checkpoint during fsync */
 793         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 794
 795         set_new_dnode(&dn, inode, page, npage, nid);
 796
 797         if (page)
 798                 dn.inode_page_locked = true;
 799         truncate_node(&dn);
 800         return 0;
 801 }
 802
 803 /*
 804  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 805  * f2fs_unlock_op().
 806  */
 807 void remove_inode_page(struct inode *inode)
 808 {
 809         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 810         struct page *page;
 811         nid_t ino = inode->i_ino;
 812         struct dnode_of_data dn;
 813
 814         page = get_node_page(sbi, ino);
 815         if (IS_ERR(page))
 816                 return;
 817
 818         if (truncate_xattr_node(inode, page)) {
 819                 f2fs_put_page(page, 1);
 820                 return;
 821         }
 822         /* 0 is possible, after f2fs_new_inode() is failed */
 823         f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
 824         set_new_dnode(&dn, inode, page, page, ino);
 825         truncate_node(&dn);
 826 }
 827
 828 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
 829 {
 830         struct dnode_of_data dn;
 831
 832         /* allocate inode page for new inode */
 833         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 834
 835         /* caller should f2fs_put_page(page, 1); */
 836         return new_node_page(&dn, 0, NULL);
 837 }
 838
 839 struct page *new_node_page(struct dnode_of_data *dn,
 840                                 unsigned int ofs, struct page *ipage)
 841 {
 842         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 843         struct node_info old_ni, new_ni;
 844         struct page *page;
 845         int err;
 846
 847         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 848                 return ERR_PTR(-EPERM);
 849
 850         page = grab_cache_page_write_begin(NODE_MAPPING(sbi),
 851                                         dn->nid, AOP_FLAG_NOFS);
 852         if (!page)
 853                 return ERR_PTR(-ENOMEM);
 854
 855         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 856                 err = -ENOSPC;
 857                 goto fail;
 858         }
 859
 860         get_node_info(sbi, dn->nid, &old_ni);
 861
 862         /* Reinitialize old_ni with new node page */
 863         f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
 864         new_ni = old_ni;
 865         new_ni.ino = dn->inode->i_ino;
 866         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
 867
 868         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 869         set_cold_node(dn->inode, page);
 870         SetPageUptodate(page);
 871         set_page_dirty(page);
 872
 873         if (f2fs_has_xattr_block(ofs))
 874                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 875
 876         dn->node_page = page;
 877         if (ipage)
 878                 update_inode(dn->inode, ipage);
 879         else
 880                 sync_inode_page(dn);
 881         if (ofs == 0)
 882                 inc_valid_inode_count(sbi);
 883
 884         return page;
 885
 886 fail:
 887         clear_node_page_dirty(page);
 888         f2fs_put_page(page, 1);
 889         return ERR_PTR(err);
 890 }
 891
 892 /*
 893  * Caller should do after getting the following values.
 894  * 0: f2fs_put_page(page, 0)
 895  * LOCKED_PAGE: f2fs_put_page(page, 1)
 896  * error: nothing
 897  */
 898 static int read_node_page(struct page *page, int rw)
 899 {
 900         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 901         struct node_info ni;
 902
 903         get_node_info(sbi, page->index, &ni);
 904
 905         if (unlikely(ni.blk_addr == NULL_ADDR)) {
 906                 f2fs_put_page(page, 1);
 907                 return -ENOENT;
 908         }
 909
 910         if (PageUptodate(page))
 911                 return LOCKED_PAGE;
 912
 913         return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
 914 }
 915
 916 /*
 917  * Readahead a node page
 918  */
 919 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 920 {
 921         struct page *apage;
 922         int err;
 923
 924         apage = find_get_page(NODE_MAPPING(sbi), nid);
 925         if (apage && PageUptodate(apage)) {
 926                 f2fs_put_page(apage, 0);
 927                 return;
 928         }
 929         f2fs_put_page(apage, 0);
 930
 931         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
 932         if (!apage)
 933                 return;
 934
 935         err = read_node_page(apage, READA);
 936         if (err == 0)
 937                 f2fs_put_page(apage, 0);
 938         else if (err == LOCKED_PAGE)
 939                 f2fs_put_page(apage, 1);
 940 }
 941
 942 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 943 {
 944         struct page *page;
 945         int err;
 946 repeat:
 947         page = grab_cache_page_write_begin(NODE_MAPPING(sbi),
 948                                         nid, AOP_FLAG_NOFS);
 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 (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
 960                 f2fs_put_page(page, 1);
 961                 return ERR_PTR(-EIO);
 962         }
 963         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 964                 f2fs_put_page(page, 1);
 965                 goto repeat;
 966         }
 967 got_it:
 968         mark_page_accessed(page);
 969         return page;
 970 }
 971
 972 /*
 973  * Return a locked page for the desired node page.
 974  * And, readahead MAX_RA_NODE number of node pages.
 975  */
 976 struct page *get_node_page_ra(struct page *parent, int start)
 977 {
 978         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 979         struct blk_plug plug;
 980         struct page *page;
 981         int err, i, end;
 982         nid_t nid;
 983
 984         /* First, try getting the desired direct node. */
 985         nid = get_nid(parent, start, false);
 986         if (!nid)
 987                 return ERR_PTR(-ENOENT);
 988 repeat:
 989         page = grab_cache_page(NODE_MAPPING(sbi), nid);
 990         if (!page)
 991                 return ERR_PTR(-ENOMEM);
 992
 993         err = read_node_page(page, READ_SYNC);
 994         if (err < 0)
 995                 return ERR_PTR(err);
 996         else if (err == LOCKED_PAGE)
 997                 goto page_hit;
 998
 999         blk_start_plug(&plug);
1000
1001         /* Then, try readahead for siblings of the desired node */
1002         end = start + MAX_RA_NODE;
1003         end = min(end, NIDS_PER_BLOCK);
1004         for (i = start + 1; i < end; i++) {
1005                 nid = get_nid(parent, i, false);
1006                 if (!nid)
1007                         continue;
1008                 ra_node_page(sbi, nid);
1009         }
1010
1011         blk_finish_plug(&plug);
1012
1013         lock_page(page);
1014         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1015                 f2fs_put_page(page, 1);
1016                 goto repeat;
1017         }
1018 page_hit:
1019         if (unlikely(!PageUptodate(page))) {
1020                 f2fs_put_page(page, 1);
1021                 return ERR_PTR(-EIO);
1022         }
1023         mark_page_accessed(page);
1024         return page;
1025 }
1026
1027 void sync_inode_page(struct dnode_of_data *dn)
1028 {
1029         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1030                 update_inode(dn->inode, dn->node_page);
1031         } else if (dn->inode_page) {
1032                 if (!dn->inode_page_locked)
1033                         lock_page(dn->inode_page);
1034                 update_inode(dn->inode, dn->inode_page);
1035                 if (!dn->inode_page_locked)
1036                         unlock_page(dn->inode_page);
1037         } else {
1038                 update_inode_page(dn->inode);
1039         }
1040 }
1041
1042 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1043                                         struct writeback_control *wbc)
1044 {
1045         pgoff_t index, end;
1046         struct pagevec pvec;
1047         int step = ino ? 2 : 0;
1048         int nwritten = 0, wrote = 0;
1049
1050         pagevec_init(&pvec, 0);
1051
1052 next_step:
1053         index = 0;
1054         end = LONG_MAX;
1055
1056         while (index <= end) {
1057                 int i, nr_pages;
1058                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1059                                 PAGECACHE_TAG_DIRTY,
1060                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1061                 if (nr_pages == 0)
1062                         break;
1063
1064                 for (i = 0; i < nr_pages; i++) {
1065                         struct page *page = pvec.pages[i];
1066
1067                         /*
1068                          * flushing sequence with step:
1069                          * 0. indirect nodes
1070                          * 1. dentry dnodes
1071                          * 2. file dnodes
1072                          */
1073                         if (step == 0 && IS_DNODE(page))
1074                                 continue;
1075                         if (step == 1 && (!IS_DNODE(page) ||
1076                                                 is_cold_node(page)))
1077                                 continue;
1078                         if (step == 2 && (!IS_DNODE(page) ||
1079                                                 !is_cold_node(page)))
1080                                 continue;
1081
1082                         /*
1083                          * If an fsync mode,
1084                          * we should not skip writing node pages.
1085                          */
1086                         if (ino && ino_of_node(page) == ino)
1087                                 lock_page(page);
1088                         else if (!trylock_page(page))
1089                                 continue;
1090
1091                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1092 continue_unlock:
1093                                 unlock_page(page);
1094                                 continue;
1095                         }
1096                         if (ino && ino_of_node(page) != ino)
1097                                 goto continue_unlock;
1098
1099                         if (!PageDirty(page)) {
1100                                 /* someone wrote it for us */
1101                                 goto continue_unlock;
1102                         }
1103
1104                         if (!clear_page_dirty_for_io(page))
1105                                 goto continue_unlock;
1106
1107                         /* called by fsync() */
1108                         if (ino && IS_DNODE(page)) {
1109                                 int mark = !is_checkpointed_node(sbi, ino);
1110                                 set_fsync_mark(page, 1);
1111                                 if (IS_INODE(page))
1112                                         set_dentry_mark(page, mark);
1113                                 nwritten++;
1114                         } else {
1115                                 set_fsync_mark(page, 0);
1116                                 set_dentry_mark(page, 0);
1117                         }
1118                         NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1119                         wrote++;
1120
1121                         if (--wbc->nr_to_write == 0)
1122                                 break;
1123                 }
1124                 pagevec_release(&pvec);
1125                 cond_resched();
1126
1127                 if (wbc->nr_to_write == 0) {
1128                         step = 2;
1129                         break;
1130                 }
1131         }
1132
1133         if (step < 2) {
1134                 step++;
1135                 goto next_step;
1136         }
1137
1138         if (wrote)
1139                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1140         return nwritten;
1141 }
1142
1143 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1144 {
1145         pgoff_t index = 0, end = LONG_MAX;
1146         struct pagevec pvec;
1147         int ret2 = 0, ret = 0;
1148
1149         pagevec_init(&pvec, 0);
1150
1151         while (index <= end) {
1152                 int i, nr_pages;
1153                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1154                                 PAGECACHE_TAG_WRITEBACK,
1155                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1156                 if (nr_pages == 0)
1157                         break;
1158
1159                 for (i = 0; i < nr_pages; i++) {
1160                         struct page *page = pvec.pages[i];
1161
1162                         /* until radix tree lookup accepts end_index */
1163                         if (unlikely(page->index > end))
1164                                 continue;
1165
1166                         if (ino && ino_of_node(page) == ino) {
1167                                 f2fs_wait_on_page_writeback(page, NODE);
1168                                 if (TestClearPageError(page))
1169                                         ret = -EIO;
1170                         }
1171                 }
1172                 pagevec_release(&pvec);
1173                 cond_resched();
1174         }
1175
1176         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1177                 ret2 = -ENOSPC;
1178         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1179                 ret2 = -EIO;
1180         if (!ret)
1181                 ret = ret2;
1182         return ret;
1183 }
1184
1185 static int f2fs_write_node_page(struct page *page,
1186                                 struct writeback_control *wbc)
1187 {
1188         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1189         nid_t nid;
1190         block_t new_addr;
1191         struct node_info ni;
1192         struct f2fs_io_info fio = {
1193                 .type = NODE,
1194                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1195         };
1196
1197         if (unlikely(sbi->por_doing))
1198                 goto redirty_out;
1199
1200         f2fs_wait_on_page_writeback(page, NODE);
1201
1202         /* get old block addr of this node page */
1203         nid = nid_of_node(page);
1204         f2fs_bug_on(page->index != nid);
1205
1206         get_node_info(sbi, nid, &ni);
1207
1208         /* This page is already truncated */
1209         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1210                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1211                 unlock_page(page);
1212                 return 0;
1213         }
1214
1215         if (wbc->for_reclaim)
1216                 goto redirty_out;
1217
1218         mutex_lock(&sbi->node_write);
1219         set_page_writeback(page);
1220         write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1221         set_node_addr(sbi, &ni, new_addr, is_fsync_dnode(page));
1222         dec_page_count(sbi, F2FS_DIRTY_NODES);
1223         mutex_unlock(&sbi->node_write);
1224         unlock_page(page);
1225         return 0;
1226
1227 redirty_out:
1228         redirty_page_for_writepage(wbc, page);
1229         return AOP_WRITEPAGE_ACTIVATE;
1230 }
1231
1232 static int f2fs_write_node_pages(struct address_space *mapping,
1233                             struct writeback_control *wbc)
1234 {
1235         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1236         long diff;
1237
1238         /* balancing f2fs's metadata in background */
1239         f2fs_balance_fs_bg(sbi);
1240
1241         /* collect a number of dirty node pages and write together */
1242         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1243                 goto skip_write;
1244
1245         diff = nr_pages_to_write(sbi, NODE, wbc);
1246         wbc->sync_mode = WB_SYNC_NONE;
1247         sync_node_pages(sbi, 0, wbc);
1248         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1249         return 0;
1250
1251 skip_write:
1252         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1253         return 0;
1254 }
1255
1256 static int f2fs_set_node_page_dirty(struct page *page)
1257 {
1258         struct address_space *mapping = page->mapping;
1259         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1260
1261         trace_f2fs_set_page_dirty(page, NODE);
1262
1263         SetPageUptodate(page);
1264         if (!PageDirty(page)) {
1265                 __set_page_dirty_nobuffers(page);
1266                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1267                 SetPagePrivate(page);
1268                 return 1;
1269         }
1270         return 0;
1271 }
1272
1273 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1274                                       unsigned int length)
1275 {
1276         struct inode *inode = page->mapping->host;
1277         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1278         if (PageDirty(page))
1279                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1280         ClearPagePrivate(page);
1281 }
1282
1283 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1284 {
1285         ClearPagePrivate(page);
1286         return 1;
1287 }
1288
1289 /*
1290  * Structure of the f2fs node operations
1291  */
1292 const struct address_space_operations f2fs_node_aops = {
1293         .writepage      = f2fs_write_node_page,
1294         .writepages     = f2fs_write_node_pages,
1295         .set_page_dirty = f2fs_set_node_page_dirty,
1296         .invalidatepage = f2fs_invalidate_node_page,
1297         .releasepage    = f2fs_release_node_page,
1298 };
1299
1300 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1301                                                 nid_t n)
1302 {
1303         return radix_tree_lookup(&nm_i->free_nid_root, n);
1304 }
1305
1306 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1307                                                 struct free_nid *i)
1308 {
1309         list_del(&i->list);
1310         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1311 }
1312
1313 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build)
1314 {
1315         struct free_nid *i;
1316         struct nat_entry *ne;
1317         bool allocated = false;
1318
1319         if (!available_free_memory(nm_i, FREE_NIDS))
1320                 return -1;
1321
1322         /* 0 nid should not be used */
1323         if (unlikely(nid == 0))
1324                 return 0;
1325
1326         if (build) {
1327                 /* do not add allocated nids */
1328                 read_lock(&nm_i->nat_tree_lock);
1329                 ne = __lookup_nat_cache(nm_i, nid);
1330                 if (ne &&
1331                         (!ne->checkpointed || nat_get_blkaddr(ne) != NULL_ADDR))
1332                         allocated = true;
1333                 read_unlock(&nm_i->nat_tree_lock);
1334                 if (allocated)
1335                         return 0;
1336         }
1337
1338         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1339         i->nid = nid;
1340         i->state = NID_NEW;
1341
1342         spin_lock(&nm_i->free_nid_list_lock);
1343         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1344                 spin_unlock(&nm_i->free_nid_list_lock);
1345                 kmem_cache_free(free_nid_slab, i);
1346                 return 0;
1347         }
1348         list_add_tail(&i->list, &nm_i->free_nid_list);
1349         nm_i->fcnt++;
1350         spin_unlock(&nm_i->free_nid_list_lock);
1351         return 1;
1352 }
1353
1354 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1355 {
1356         struct free_nid *i;
1357         bool need_free = false;
1358
1359         spin_lock(&nm_i->free_nid_list_lock);
1360         i = __lookup_free_nid_list(nm_i, nid);
1361         if (i && i->state == NID_NEW) {
1362                 __del_from_free_nid_list(nm_i, i);
1363                 nm_i->fcnt--;
1364                 need_free = true;
1365         }
1366         spin_unlock(&nm_i->free_nid_list_lock);
1367
1368         if (need_free)
1369                 kmem_cache_free(free_nid_slab, i);
1370 }
1371
1372 static void scan_nat_page(struct f2fs_nm_info *nm_i,
1373                         struct page *nat_page, nid_t start_nid)
1374 {
1375         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1376         block_t blk_addr;
1377         int i;
1378
1379         i = start_nid % NAT_ENTRY_PER_BLOCK;
1380
1381         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1382
1383                 if (unlikely(start_nid >= nm_i->max_nid))
1384                         break;
1385
1386                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1387                 f2fs_bug_on(blk_addr == NEW_ADDR);
1388                 if (blk_addr == NULL_ADDR) {
1389                         if (add_free_nid(nm_i, start_nid, true) < 0)
1390                                 break;
1391                 }
1392         }
1393 }
1394
1395 static void build_free_nids(struct f2fs_sb_info *sbi)
1396 {
1397         struct f2fs_nm_info *nm_i = NM_I(sbi);
1398         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1399         struct f2fs_summary_block *sum = curseg->sum_blk;
1400         int i = 0;
1401         nid_t nid = nm_i->next_scan_nid;
1402
1403         /* Enough entries */
1404         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1405                 return;
1406
1407         /* readahead nat pages to be scanned */
1408         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1409
1410         while (1) {
1411                 struct page *page = get_current_nat_page(sbi, nid);
1412
1413                 scan_nat_page(nm_i, page, nid);
1414                 f2fs_put_page(page, 1);
1415
1416                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1417                 if (unlikely(nid >= nm_i->max_nid))
1418                         nid = 0;
1419
1420                 if (i++ == FREE_NID_PAGES)
1421                         break;
1422         }
1423
1424         /* go to the next free nat pages to find free nids abundantly */
1425         nm_i->next_scan_nid = nid;
1426
1427         /* find free nids from current sum_pages */
1428         mutex_lock(&curseg->curseg_mutex);
1429         for (i = 0; i < nats_in_cursum(sum); i++) {
1430                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1431                 nid = le32_to_cpu(nid_in_journal(sum, i));
1432                 if (addr == NULL_ADDR)
1433                         add_free_nid(nm_i, nid, true);
1434                 else
1435                         remove_free_nid(nm_i, nid);
1436         }
1437         mutex_unlock(&curseg->curseg_mutex);
1438 }
1439
1440 /*
1441  * If this function returns success, caller can obtain a new nid
1442  * from second parameter of this function.
1443  * The returned nid could be used ino as well as nid when inode is created.
1444  */
1445 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1446 {
1447         struct f2fs_nm_info *nm_i = NM_I(sbi);
1448         struct free_nid *i = NULL;
1449 retry:
1450         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1451                 return false;
1452
1453         spin_lock(&nm_i->free_nid_list_lock);
1454
1455         /* We should not use stale free nids created by build_free_nids */
1456         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1457                 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1458                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1459                         if (i->state == NID_NEW)
1460                                 break;
1461
1462                 f2fs_bug_on(i->state != NID_NEW);
1463                 *nid = i->nid;
1464                 i->state = NID_ALLOC;
1465                 nm_i->fcnt--;
1466                 spin_unlock(&nm_i->free_nid_list_lock);
1467                 return true;
1468         }
1469         spin_unlock(&nm_i->free_nid_list_lock);
1470
1471         /* Let's scan nat pages and its caches to get free nids */
1472         mutex_lock(&nm_i->build_lock);
1473         build_free_nids(sbi);
1474         mutex_unlock(&nm_i->build_lock);
1475         goto retry;
1476 }
1477
1478 /*
1479  * alloc_nid() should be called prior to this function.
1480  */
1481 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1482 {
1483         struct f2fs_nm_info *nm_i = NM_I(sbi);
1484         struct free_nid *i;
1485
1486         spin_lock(&nm_i->free_nid_list_lock);
1487         i = __lookup_free_nid_list(nm_i, nid);
1488         f2fs_bug_on(!i || i->state != NID_ALLOC);
1489         __del_from_free_nid_list(nm_i, i);
1490         spin_unlock(&nm_i->free_nid_list_lock);
1491
1492         kmem_cache_free(free_nid_slab, i);
1493 }
1494
1495 /*
1496  * alloc_nid() should be called prior to this function.
1497  */
1498 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1499 {
1500         struct f2fs_nm_info *nm_i = NM_I(sbi);
1501         struct free_nid *i;
1502         bool need_free = false;
1503
1504         if (!nid)
1505                 return;
1506
1507         spin_lock(&nm_i->free_nid_list_lock);
1508         i = __lookup_free_nid_list(nm_i, nid);
1509         f2fs_bug_on(!i || i->state != NID_ALLOC);
1510         if (!available_free_memory(nm_i, FREE_NIDS)) {
1511                 __del_from_free_nid_list(nm_i, i);
1512                 need_free = true;
1513         } else {
1514                 i->state = NID_NEW;
1515                 nm_i->fcnt++;
1516         }
1517         spin_unlock(&nm_i->free_nid_list_lock);
1518
1519         if (need_free)
1520                 kmem_cache_free(free_nid_slab, i);
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, false);
1529         clear_node_page_dirty(page);
1530 }
1531
1532 static void recover_inline_xattr(struct inode *inode, struct page *page)
1533 {
1534         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1535         void *src_addr, *dst_addr;
1536         size_t inline_size;
1537         struct page *ipage;
1538         struct f2fs_inode *ri;
1539
1540         if (!f2fs_has_inline_xattr(inode))
1541                 return;
1542
1543         if (!IS_INODE(page))
1544                 return;
1545
1546         ri = F2FS_INODE(page);
1547         if (!(ri->i_inline & F2FS_INLINE_XATTR))
1548                 return;
1549
1550         ipage = get_node_page(sbi, inode->i_ino);
1551         f2fs_bug_on(IS_ERR(ipage));
1552
1553         dst_addr = inline_xattr_addr(ipage);
1554         src_addr = inline_xattr_addr(page);
1555         inline_size = inline_xattr_size(inode);
1556
1557         memcpy(dst_addr, src_addr, inline_size);
1558
1559         update_inode(inode, ipage);
1560         f2fs_put_page(ipage, 1);
1561 }
1562
1563 bool recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1564 {
1565         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1566         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1567         nid_t new_xnid = nid_of_node(page);
1568         struct node_info ni;
1569
1570         recover_inline_xattr(inode, page);
1571
1572         if (!f2fs_has_xattr_block(ofs_of_node(page)))
1573                 return false;
1574
1575         /* 1: invalidate the previous xattr nid */
1576         if (!prev_xnid)
1577                 goto recover_xnid;
1578
1579         /* Deallocate node address */
1580         get_node_info(sbi, prev_xnid, &ni);
1581         f2fs_bug_on(ni.blk_addr == NULL_ADDR);
1582         invalidate_blocks(sbi, ni.blk_addr);
1583         dec_valid_node_count(sbi, inode);
1584         set_node_addr(sbi, &ni, NULL_ADDR, false);
1585
1586 recover_xnid:
1587         /* 2: allocate new xattr nid */
1588         if (unlikely(!inc_valid_node_count(sbi, inode)))
1589                 f2fs_bug_on(1);
1590
1591         remove_free_nid(NM_I(sbi), new_xnid);
1592         get_node_info(sbi, new_xnid, &ni);
1593         ni.ino = inode->i_ino;
1594         set_node_addr(sbi, &ni, NEW_ADDR, false);
1595         F2FS_I(inode)->i_xattr_nid = new_xnid;
1596
1597         /* 3: update xattr blkaddr */
1598         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1599         set_node_addr(sbi, &ni, blkaddr, false);
1600
1601         update_inode_page(inode);
1602         return true;
1603 }
1604
1605 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1606 {
1607         struct f2fs_inode *src, *dst;
1608         nid_t ino = ino_of_node(page);
1609         struct node_info old_ni, new_ni;
1610         struct page *ipage;
1611
1612         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1613         if (!ipage)
1614                 return -ENOMEM;
1615
1616         /* Should not use this inode  from free nid list */
1617         remove_free_nid(NM_I(sbi), ino);
1618
1619         get_node_info(sbi, ino, &old_ni);
1620         SetPageUptodate(ipage);
1621         fill_node_footer(ipage, ino, ino, 0, true);
1622
1623         src = F2FS_INODE(page);
1624         dst = F2FS_INODE(ipage);
1625
1626         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1627         dst->i_size = 0;
1628         dst->i_blocks = cpu_to_le64(1);
1629         dst->i_links = cpu_to_le32(1);
1630         dst->i_xattr_nid = 0;
1631
1632         new_ni = old_ni;
1633         new_ni.ino = ino;
1634
1635         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1636                 WARN_ON(1);
1637         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1638         inc_valid_inode_count(sbi);
1639         f2fs_put_page(ipage, 1);
1640         return 0;
1641 }
1642
1643 /*
1644  * ra_sum_pages() merge contiguous pages into one bio and submit.
1645  * these pre-readed pages are linked in pages list.
1646  */
1647 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
1648                                 int start, int nrpages)
1649 {
1650         struct page *page;
1651         int page_idx = start;
1652         struct f2fs_io_info fio = {
1653                 .type = META,
1654                 .rw = READ_SYNC | REQ_META | REQ_PRIO
1655         };
1656
1657         for (; page_idx < start + nrpages; page_idx++) {
1658                 /* alloc temporal page for read node summary info*/
1659                 page = alloc_page(GFP_F2FS_ZERO);
1660                 if (!page)
1661                         break;
1662
1663                 lock_page(page);
1664                 page->index = page_idx;
1665                 list_add_tail(&page->lru, pages);
1666         }
1667
1668         list_for_each_entry(page, pages, lru)
1669                 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1670
1671         f2fs_submit_merged_bio(sbi, META, READ);
1672
1673         return page_idx - start;
1674 }
1675
1676 int restore_node_summary(struct f2fs_sb_info *sbi,
1677                         unsigned int segno, struct f2fs_summary_block *sum)
1678 {
1679         struct f2fs_node *rn;
1680         struct f2fs_summary *sum_entry;
1681         struct page *page, *tmp;
1682         block_t addr;
1683         int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1684         int i, last_offset, nrpages, err = 0;
1685         LIST_HEAD(page_list);
1686
1687         /* scan the node segment */
1688         last_offset = sbi->blocks_per_seg;
1689         addr = START_BLOCK(sbi, segno);
1690         sum_entry = &sum->entries[0];
1691
1692         for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) {
1693                 nrpages = min(last_offset - i, bio_blocks);
1694
1695                 /* read ahead node pages */
1696                 nrpages = ra_sum_pages(sbi, &page_list, addr, nrpages);
1697                 if (!nrpages)
1698                         return -ENOMEM;
1699
1700                 list_for_each_entry_safe(page, tmp, &page_list, lru) {
1701                         if (err)
1702                                 goto skip;
1703
1704                         lock_page(page);
1705                         if (unlikely(!PageUptodate(page))) {
1706                                 err = -EIO;
1707                         } else {
1708                                 rn = F2FS_NODE(page);
1709                                 sum_entry->nid = rn->footer.nid;
1710                                 sum_entry->version = 0;
1711                                 sum_entry->ofs_in_node = 0;
1712                                 sum_entry++;
1713                         }
1714                         unlock_page(page);
1715 skip:
1716                         list_del(&page->lru);
1717                         __free_pages(page, 0);
1718                 }
1719         }
1720         return err;
1721 }
1722
1723 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1724 {
1725         struct f2fs_nm_info *nm_i = NM_I(sbi);
1726         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1727         struct f2fs_summary_block *sum = curseg->sum_blk;
1728         int i;
1729
1730         mutex_lock(&curseg->curseg_mutex);
1731
1732         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1733                 mutex_unlock(&curseg->curseg_mutex);
1734                 return false;
1735         }
1736
1737         for (i = 0; i < nats_in_cursum(sum); i++) {
1738                 struct nat_entry *ne;
1739                 struct f2fs_nat_entry raw_ne;
1740                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1741
1742                 raw_ne = nat_in_journal(sum, i);
1743 retry:
1744                 write_lock(&nm_i->nat_tree_lock);
1745                 ne = __lookup_nat_cache(nm_i, nid);
1746                 if (ne) {
1747                         __set_nat_cache_dirty(nm_i, ne);
1748                         write_unlock(&nm_i->nat_tree_lock);
1749                         continue;
1750                 }
1751                 ne = grab_nat_entry(nm_i, nid);
1752                 if (!ne) {
1753                         write_unlock(&nm_i->nat_tree_lock);
1754                         goto retry;
1755                 }
1756                 node_info_from_raw_nat(&ne->ni, &raw_ne);
1757                 __set_nat_cache_dirty(nm_i, ne);
1758                 write_unlock(&nm_i->nat_tree_lock);
1759         }
1760         update_nats_in_cursum(sum, -i);
1761         mutex_unlock(&curseg->curseg_mutex);
1762         return true;
1763 }
1764
1765 /*
1766  * This function is called during the checkpointing process.
1767  */
1768 void flush_nat_entries(struct f2fs_sb_info *sbi)
1769 {
1770         struct f2fs_nm_info *nm_i = NM_I(sbi);
1771         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1772         struct f2fs_summary_block *sum = curseg->sum_blk;
1773         struct nat_entry *ne, *cur;
1774         struct page *page = NULL;
1775         struct f2fs_nat_block *nat_blk = NULL;
1776         nid_t start_nid = 0, end_nid = 0;
1777         bool flushed;
1778
1779         flushed = flush_nats_in_journal(sbi);
1780
1781         if (!flushed)
1782                 mutex_lock(&curseg->curseg_mutex);
1783
1784         /* 1) flush dirty nat caches */
1785         list_for_each_entry_safe(ne, cur, &nm_i->dirty_nat_entries, list) {
1786                 nid_t nid;
1787                 struct f2fs_nat_entry raw_ne;
1788                 int offset = -1;
1789
1790                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1791                         continue;
1792
1793                 nid = nat_get_nid(ne);
1794
1795                 if (flushed)
1796                         goto to_nat_page;
1797
1798                 /* if there is room for nat enries in curseg->sumpage */
1799                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1800                 if (offset >= 0) {
1801                         raw_ne = nat_in_journal(sum, offset);
1802                         goto flush_now;
1803                 }
1804 to_nat_page:
1805                 if (!page || (start_nid > nid || nid > end_nid)) {
1806                         if (page) {
1807                                 f2fs_put_page(page, 1);
1808                                 page = NULL;
1809                         }
1810                         start_nid = START_NID(nid);
1811                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1812
1813                         /*
1814                          * get nat block with dirty flag, increased reference
1815                          * count, mapped and lock
1816                          */
1817                         page = get_next_nat_page(sbi, start_nid);
1818                         nat_blk = page_address(page);
1819                 }
1820
1821                 f2fs_bug_on(!nat_blk);
1822                 raw_ne = nat_blk->entries[nid - start_nid];
1823 flush_now:
1824                 raw_nat_from_node_info(&raw_ne, &ne->ni);
1825
1826                 if (offset < 0) {
1827                         nat_blk->entries[nid - start_nid] = raw_ne;
1828                 } else {
1829                         nat_in_journal(sum, offset) = raw_ne;
1830                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1831                 }
1832
1833                 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1834                                 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1835                         write_lock(&nm_i->nat_tree_lock);
1836                         __del_from_nat_cache(nm_i, ne);
1837                         write_unlock(&nm_i->nat_tree_lock);
1838                 } else {
1839                         write_lock(&nm_i->nat_tree_lock);
1840                         __clear_nat_cache_dirty(nm_i, ne);
1841                         write_unlock(&nm_i->nat_tree_lock);
1842                 }
1843         }
1844         if (!flushed)
1845                 mutex_unlock(&curseg->curseg_mutex);
1846         f2fs_put_page(page, 1);
1847 }
1848
1849 static int init_node_manager(struct f2fs_sb_info *sbi)
1850 {
1851         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1852         struct f2fs_nm_info *nm_i = NM_I(sbi);
1853         unsigned char *version_bitmap;
1854         unsigned int nat_segs, nat_blocks;
1855
1856         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1857
1858         /* segment_count_nat includes pair segment so divide to 2. */
1859         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1860         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1861
1862         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1863
1864         /* not used nids: 0, node, meta, (and root counted as valid node) */
1865         nm_i->available_nids = nm_i->max_nid - 3;
1866         nm_i->fcnt = 0;
1867         nm_i->nat_cnt = 0;
1868         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1869
1870         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1871         INIT_LIST_HEAD(&nm_i->free_nid_list);
1872         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1873         INIT_LIST_HEAD(&nm_i->nat_entries);
1874         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1875
1876         mutex_init(&nm_i->build_lock);
1877         spin_lock_init(&nm_i->free_nid_list_lock);
1878         rwlock_init(&nm_i->nat_tree_lock);
1879
1880         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1881         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1882         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1883         if (!version_bitmap)
1884                 return -EFAULT;
1885
1886         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1887                                         GFP_KERNEL);
1888         if (!nm_i->nat_bitmap)
1889                 return -ENOMEM;
1890         return 0;
1891 }
1892
1893 int build_node_manager(struct f2fs_sb_info *sbi)
1894 {
1895         int err;
1896
1897         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1898         if (!sbi->nm_info)
1899                 return -ENOMEM;
1900
1901         err = init_node_manager(sbi);
1902         if (err)
1903                 return err;
1904
1905         build_free_nids(sbi);
1906         return 0;
1907 }
1908
1909 void destroy_node_manager(struct f2fs_sb_info *sbi)
1910 {
1911         struct f2fs_nm_info *nm_i = NM_I(sbi);
1912         struct free_nid *i, *next_i;
1913         struct nat_entry *natvec[NATVEC_SIZE];
1914         nid_t nid = 0;
1915         unsigned int found;
1916
1917         if (!nm_i)
1918                 return;
1919
1920         /* destroy free nid list */
1921         spin_lock(&nm_i->free_nid_list_lock);
1922         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1923                 f2fs_bug_on(i->state == NID_ALLOC);
1924                 __del_from_free_nid_list(nm_i, i);
1925                 nm_i->fcnt--;
1926                 spin_unlock(&nm_i->free_nid_list_lock);
1927                 kmem_cache_free(free_nid_slab, i);
1928                 spin_lock(&nm_i->free_nid_list_lock);
1929         }
1930         f2fs_bug_on(nm_i->fcnt);
1931         spin_unlock(&nm_i->free_nid_list_lock);
1932
1933         /* destroy nat cache */
1934         write_lock(&nm_i->nat_tree_lock);
1935         while ((found = __gang_lookup_nat_cache(nm_i,
1936                                         nid, NATVEC_SIZE, natvec))) {
1937                 unsigned idx;
1938                 nid = nat_get_nid(natvec[found - 1]) + 1;
1939                 for (idx = 0; idx < found; idx++)
1940                         __del_from_nat_cache(nm_i, natvec[idx]);
1941         }
1942         f2fs_bug_on(nm_i->nat_cnt);
1943         write_unlock(&nm_i->nat_tree_lock);
1944
1945         kfree(nm_i->nat_bitmap);
1946         sbi->nm_info = NULL;
1947         kfree(nm_i);
1948 }
1949
1950 int __init create_node_manager_caches(void)
1951 {
1952         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1953                         sizeof(struct nat_entry));
1954         if (!nat_entry_slab)
1955                 return -ENOMEM;
1956
1957         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1958                         sizeof(struct free_nid));
1959         if (!free_nid_slab) {
1960                 kmem_cache_destroy(nat_entry_slab);
1961                 return -ENOMEM;
1962         }
1963         return 0;
1964 }
1965
1966 void destroy_node_manager_caches(void)
1967 {
1968         kmem_cache_destroy(free_nid_slab);
1969         kmem_cache_destroy(nat_entry_slab);
1970 }