fs/f2fs/node.c

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