fs/f2fs/node.c

   1 /*
   2  * fs/f2fs/node.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/mpage.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/swap.h>
  18
  19 #include "f2fs.h"
  20 #include "node.h"
  21 #include "segment.h"
  22 #include <trace/events/f2fs.h>
  23
  24 static struct kmem_cache *nat_entry_slab;
  25 static struct kmem_cache *free_nid_slab;
  26
  27 static void clear_node_page_dirty(struct page *page)
  28 {
  29         struct address_space *mapping = page->mapping;
  30         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
  31         unsigned int long flags;
  32
  33         if (PageDirty(page)) {
  34                 spin_lock_irqsave(&mapping->tree_lock, flags);
  35                 radix_tree_tag_clear(&mapping->page_tree,
  36                                 page_index(page),
  37                                 PAGECACHE_TAG_DIRTY);
  38                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  39
  40                 clear_page_dirty_for_io(page);
  41                 dec_page_count(sbi, F2FS_DIRTY_NODES);
  42         }
  43         ClearPageUptodate(page);
  44 }
  45
  46 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  47 {
  48         pgoff_t index = current_nat_addr(sbi, nid);
  49         return get_meta_page(sbi, index);
  50 }
  51
  52 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  53 {
  54         struct page *src_page;
  55         struct page *dst_page;
  56         pgoff_t src_off;
  57         pgoff_t dst_off;
  58         void *src_addr;
  59         void *dst_addr;
  60         struct f2fs_nm_info *nm_i = NM_I(sbi);
  61
  62         src_off = current_nat_addr(sbi, nid);
  63         dst_off = next_nat_addr(sbi, src_off);
  64
  65         /* get current nat block page with lock */
  66         src_page = get_meta_page(sbi, src_off);
  67
  68         /* Dirty src_page means that it is already the new target NAT page. */
  69         if (PageDirty(src_page))
  70                 return src_page;
  71
  72         dst_page = grab_meta_page(sbi, dst_off);
  73
  74         src_addr = page_address(src_page);
  75         dst_addr = page_address(dst_page);
  76         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
  77         set_page_dirty(dst_page);
  78         f2fs_put_page(src_page, 1);
  79
  80         set_to_next_nat(nm_i, nid);
  81
  82         return dst_page;
  83 }
  84
  85 /*
  86  * Readahead NAT pages
  87  */
  88 static void ra_nat_pages(struct f2fs_sb_info *sbi, int nid)
  89 {
  90         struct address_space *mapping = sbi->meta_inode->i_mapping;
  91         struct f2fs_nm_info *nm_i = NM_I(sbi);
  92         struct page *page;
  93         pgoff_t index;
  94         int i;
  95
  96         for (i = 0; i < FREE_NID_PAGES; i++, nid += NAT_ENTRY_PER_BLOCK) {
  97                 if (nid >= nm_i->max_nid)
  98                         nid = 0;
  99                 index = current_nat_addr(sbi, nid);
 100
 101                 page = grab_cache_page(mapping, index);
 102                 if (!page)
 103                         continue;
 104                 if (PageUptodate(page)) {
 105                         mark_page_accessed(page);
 106                         f2fs_put_page(page, 1);
 107                         continue;
 108                 }
 109                 f2fs_submit_page_mbio(sbi, page, index, META, READ);
 110                 mark_page_accessed(page);
 111                 f2fs_put_page(page, 0);
 112         }
 113         f2fs_submit_merged_bio(sbi, META, true, READ);
 114 }
 115
 116 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 117 {
 118         return radix_tree_lookup(&nm_i->nat_root, n);
 119 }
 120
 121 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 122                 nid_t start, unsigned int nr, struct nat_entry **ep)
 123 {
 124         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 125 }
 126
 127 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 128 {
 129         list_del(&e->list);
 130         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 131         nm_i->nat_cnt--;
 132         kmem_cache_free(nat_entry_slab, e);
 133 }
 134
 135 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 136 {
 137         struct f2fs_nm_info *nm_i = NM_I(sbi);
 138         struct nat_entry *e;
 139         int is_cp = 1;
 140
 141         read_lock(&nm_i->nat_tree_lock);
 142         e = __lookup_nat_cache(nm_i, nid);
 143         if (e && !e->checkpointed)
 144                 is_cp = 0;
 145         read_unlock(&nm_i->nat_tree_lock);
 146         return is_cp;
 147 }
 148
 149 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 150 {
 151         struct nat_entry *new;
 152
 153         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 154         if (!new)
 155                 return NULL;
 156         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 157                 kmem_cache_free(nat_entry_slab, new);
 158                 return NULL;
 159         }
 160         memset(new, 0, sizeof(struct nat_entry));
 161         nat_set_nid(new, nid);
 162         list_add_tail(&new->list, &nm_i->nat_entries);
 163         nm_i->nat_cnt++;
 164         return new;
 165 }
 166
 167 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 168                                                 struct f2fs_nat_entry *ne)
 169 {
 170         struct nat_entry *e;
 171 retry:
 172         write_lock(&nm_i->nat_tree_lock);
 173         e = __lookup_nat_cache(nm_i, nid);
 174         if (!e) {
 175                 e = grab_nat_entry(nm_i, nid);
 176                 if (!e) {
 177                         write_unlock(&nm_i->nat_tree_lock);
 178                         goto retry;
 179                 }
 180                 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
 181                 nat_set_ino(e, le32_to_cpu(ne->ino));
 182                 nat_set_version(e, ne->version);
 183                 e->checkpointed = true;
 184         }
 185         write_unlock(&nm_i->nat_tree_lock);
 186 }
 187
 188 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 189                         block_t new_blkaddr)
 190 {
 191         struct f2fs_nm_info *nm_i = NM_I(sbi);
 192         struct nat_entry *e;
 193 retry:
 194         write_lock(&nm_i->nat_tree_lock);
 195         e = __lookup_nat_cache(nm_i, ni->nid);
 196         if (!e) {
 197                 e = grab_nat_entry(nm_i, ni->nid);
 198                 if (!e) {
 199                         write_unlock(&nm_i->nat_tree_lock);
 200                         goto retry;
 201                 }
 202                 e->ni = *ni;
 203                 e->checkpointed = true;
 204                 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
 205         } else if (new_blkaddr == NEW_ADDR) {
 206                 /*
 207                  * when nid is reallocated,
 208                  * previous nat entry can be remained in nat cache.
 209                  * So, reinitialize it with new information.
 210                  */
 211                 e->ni = *ni;
 212                 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
 213         }
 214
 215         if (new_blkaddr == NEW_ADDR)
 216                 e->checkpointed = false;
 217
 218         /* sanity check */
 219         f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
 220         f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
 221                         new_blkaddr == NULL_ADDR);
 222         f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
 223                         new_blkaddr == NEW_ADDR);
 224         f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
 225                         nat_get_blkaddr(e) != NULL_ADDR &&
 226                         new_blkaddr == NEW_ADDR);
 227
 228         /* increament version no as node is removed */
 229         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 230                 unsigned char version = nat_get_version(e);
 231                 nat_set_version(e, inc_node_version(version));
 232         }
 233
 234         /* change address */
 235         nat_set_blkaddr(e, new_blkaddr);
 236         __set_nat_cache_dirty(nm_i, e);
 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 (nm_i->nat_cnt <= NM_WOUT_THRESHOLD)
 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 mutex by calling mutex_lock_op() and
 392  * mutex_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);
 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         dn->node_page = NULL;
 517         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 518 }
 519
 520 static int truncate_dnode(struct dnode_of_data *dn)
 521 {
 522         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 523         struct page *page;
 524
 525         if (dn->nid == 0)
 526                 return 1;
 527
 528         /* get direct node */
 529         page = get_node_page(sbi, dn->nid);
 530         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 531                 return 1;
 532         else if (IS_ERR(page))
 533                 return PTR_ERR(page);
 534
 535         /* Make dnode_of_data for parameter */
 536         dn->node_page = page;
 537         dn->ofs_in_node = 0;
 538         truncate_data_blocks(dn);
 539         truncate_node(dn);
 540         return 1;
 541 }
 542
 543 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 544                                                 int ofs, int depth)
 545 {
 546         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 547         struct dnode_of_data rdn = *dn;
 548         struct page *page;
 549         struct f2fs_node *rn;
 550         nid_t child_nid;
 551         unsigned int child_nofs;
 552         int freed = 0;
 553         int i, ret;
 554
 555         if (dn->nid == 0)
 556                 return NIDS_PER_BLOCK + 1;
 557
 558         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 559
 560         page = get_node_page(sbi, dn->nid);
 561         if (IS_ERR(page)) {
 562                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 563                 return PTR_ERR(page);
 564         }
 565
 566         rn = F2FS_NODE(page);
 567         if (depth < 3) {
 568                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 569                         child_nid = le32_to_cpu(rn->in.nid[i]);
 570                         if (child_nid == 0)
 571                                 continue;
 572                         rdn.nid = child_nid;
 573                         ret = truncate_dnode(&rdn);
 574                         if (ret < 0)
 575                                 goto out_err;
 576                         set_nid(page, i, 0, false);
 577                 }
 578         } else {
 579                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 580                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 581                         child_nid = le32_to_cpu(rn->in.nid[i]);
 582                         if (child_nid == 0) {
 583                                 child_nofs += NIDS_PER_BLOCK + 1;
 584                                 continue;
 585                         }
 586                         rdn.nid = child_nid;
 587                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 588                         if (ret == (NIDS_PER_BLOCK + 1)) {
 589                                 set_nid(page, i, 0, false);
 590                                 child_nofs += ret;
 591                         } else if (ret < 0 && ret != -ENOENT) {
 592                                 goto out_err;
 593                         }
 594                 }
 595                 freed = child_nofs;
 596         }
 597
 598         if (!ofs) {
 599                 /* remove current indirect node */
 600                 dn->node_page = page;
 601                 truncate_node(dn);
 602                 freed++;
 603         } else {
 604                 f2fs_put_page(page, 1);
 605         }
 606         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 607         return freed;
 608
 609 out_err:
 610         f2fs_put_page(page, 1);
 611         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 612         return ret;
 613 }
 614
 615 static int truncate_partial_nodes(struct dnode_of_data *dn,
 616                         struct f2fs_inode *ri, int *offset, int depth)
 617 {
 618         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 619         struct page *pages[2];
 620         nid_t nid[3];
 621         nid_t child_nid;
 622         int err = 0;
 623         int i;
 624         int idx = depth - 2;
 625
 626         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 627         if (!nid[0])
 628                 return 0;
 629
 630         /* get indirect nodes in the path */
 631         for (i = 0; i < depth - 1; i++) {
 632                 /* refernece count'll be increased */
 633                 pages[i] = get_node_page(sbi, nid[i]);
 634                 if (IS_ERR(pages[i])) {
 635                         depth = i + 1;
 636                         err = PTR_ERR(pages[i]);
 637                         goto fail;
 638                 }
 639                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 640         }
 641
 642         /* free direct nodes linked to a partial indirect node */
 643         for (i = offset[depth - 1]; i < NIDS_PER_BLOCK; i++) {
 644                 child_nid = get_nid(pages[idx], i, false);
 645                 if (!child_nid)
 646                         continue;
 647                 dn->nid = child_nid;
 648                 err = truncate_dnode(dn);
 649                 if (err < 0)
 650                         goto fail;
 651                 set_nid(pages[idx], i, 0, false);
 652         }
 653
 654         if (offset[depth - 1] == 0) {
 655                 dn->node_page = pages[idx];
 656                 dn->nid = nid[idx];
 657                 truncate_node(dn);
 658         } else {
 659                 f2fs_put_page(pages[idx], 1);
 660         }
 661         offset[idx]++;
 662         offset[depth - 1] = 0;
 663 fail:
 664         for (i = depth - 3; i >= 0; i--)
 665                 f2fs_put_page(pages[i], 1);
 666
 667         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 668
 669         return err;
 670 }
 671
 672 /*
 673  * All the block addresses of data and nodes should be nullified.
 674  */
 675 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 676 {
 677         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 678         struct address_space *node_mapping = sbi->node_inode->i_mapping;
 679         int err = 0, cont = 1;
 680         int level, offset[4], noffset[4];
 681         unsigned int nofs = 0;
 682         struct f2fs_node *rn;
 683         struct dnode_of_data dn;
 684         struct page *page;
 685
 686         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 687
 688         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 689 restart:
 690         page = get_node_page(sbi, inode->i_ino);
 691         if (IS_ERR(page)) {
 692                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 693                 return PTR_ERR(page);
 694         }
 695
 696         set_new_dnode(&dn, inode, page, NULL, 0);
 697         unlock_page(page);
 698
 699         rn = F2FS_NODE(page);
 700         switch (level) {
 701         case 0:
 702         case 1:
 703                 nofs = noffset[1];
 704                 break;
 705         case 2:
 706                 nofs = noffset[1];
 707                 if (!offset[level - 1])
 708                         goto skip_partial;
 709                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 710                 if (err < 0 && err != -ENOENT)
 711                         goto fail;
 712                 nofs += 1 + NIDS_PER_BLOCK;
 713                 break;
 714         case 3:
 715                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 716                 if (!offset[level - 1])
 717                         goto skip_partial;
 718                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
 719                 if (err < 0 && err != -ENOENT)
 720                         goto fail;
 721                 break;
 722         default:
 723                 BUG();
 724         }
 725
 726 skip_partial:
 727         while (cont) {
 728                 dn.nid = le32_to_cpu(rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]);
 729                 switch (offset[0]) {
 730                 case NODE_DIR1_BLOCK:
 731                 case NODE_DIR2_BLOCK:
 732                         err = truncate_dnode(&dn);
 733                         break;
 734
 735                 case NODE_IND1_BLOCK:
 736                 case NODE_IND2_BLOCK:
 737                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 738                         break;
 739
 740                 case NODE_DIND_BLOCK:
 741                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 742                         cont = 0;
 743                         break;
 744
 745                 default:
 746                         BUG();
 747                 }
 748                 if (err < 0 && err != -ENOENT)
 749                         goto fail;
 750                 if (offset[1] == 0 &&
 751                                 rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 752                         lock_page(page);
 753                         if (page->mapping != node_mapping) {
 754                                 f2fs_put_page(page, 1);
 755                                 goto restart;
 756                         }
 757                         wait_on_page_writeback(page);
 758                         rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 759                         set_page_dirty(page);
 760                         unlock_page(page);
 761                 }
 762                 offset[1] = 0;
 763                 offset[0]++;
 764                 nofs += err;
 765         }
 766 fail:
 767         f2fs_put_page(page, 0);
 768         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 769         return err > 0 ? 0 : err;
 770 }
 771
 772 int truncate_xattr_node(struct inode *inode, struct page *page)
 773 {
 774         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 775         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 776         struct dnode_of_data dn;
 777         struct page *npage;
 778
 779         if (!nid)
 780                 return 0;
 781
 782         npage = get_node_page(sbi, nid);
 783         if (IS_ERR(npage))
 784                 return PTR_ERR(npage);
 785
 786         F2FS_I(inode)->i_xattr_nid = 0;
 787
 788         /* need to do checkpoint during fsync */
 789         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 790
 791         set_new_dnode(&dn, inode, page, npage, nid);
 792
 793         if (page)
 794                 dn.inode_page_locked = true;
 795         truncate_node(&dn);
 796         return 0;
 797 }
 798
 799 /*
 800  * Caller should grab and release a mutex by calling mutex_lock_op() and
 801  * mutex_unlock_op().
 802  */
 803 void remove_inode_page(struct inode *inode)
 804 {
 805         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 806         struct page *page;
 807         nid_t ino = inode->i_ino;
 808         struct dnode_of_data dn;
 809
 810         page = get_node_page(sbi, ino);
 811         if (IS_ERR(page))
 812                 return;
 813
 814         if (truncate_xattr_node(inode, page)) {
 815                 f2fs_put_page(page, 1);
 816                 return;
 817         }
 818         /* 0 is possible, after f2fs_new_inode() is failed */
 819         f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
 820         set_new_dnode(&dn, inode, page, page, ino);
 821         truncate_node(&dn);
 822 }
 823
 824 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
 825 {
 826         struct dnode_of_data dn;
 827
 828         /* allocate inode page for new inode */
 829         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 830
 831         /* caller should f2fs_put_page(page, 1); */
 832         return new_node_page(&dn, 0, NULL);
 833 }
 834
 835 struct page *new_node_page(struct dnode_of_data *dn,
 836                                 unsigned int ofs, struct page *ipage)
 837 {
 838         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 839         struct address_space *mapping = sbi->node_inode->i_mapping;
 840         struct node_info old_ni, new_ni;
 841         struct page *page;
 842         int err;
 843
 844         if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
 845                 return ERR_PTR(-EPERM);
 846
 847         page = grab_cache_page(mapping, dn->nid);
 848         if (!page)
 849                 return ERR_PTR(-ENOMEM);
 850
 851         if (!inc_valid_node_count(sbi, dn->inode)) {
 852                 err = -ENOSPC;
 853                 goto fail;
 854         }
 855
 856         get_node_info(sbi, dn->nid, &old_ni);
 857
 858         /* Reinitialize old_ni with new node page */
 859         f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
 860         new_ni = old_ni;
 861         new_ni.ino = dn->inode->i_ino;
 862         set_node_addr(sbi, &new_ni, NEW_ADDR);
 863
 864         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 865         set_cold_node(dn->inode, page);
 866         SetPageUptodate(page);
 867         set_page_dirty(page);
 868
 869         if (ofs == XATTR_NODE_OFFSET)
 870                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 871
 872         dn->node_page = page;
 873         if (ipage)
 874                 update_inode(dn->inode, ipage);
 875         else
 876                 sync_inode_page(dn);
 877         if (ofs == 0)
 878                 inc_valid_inode_count(sbi);
 879
 880         return page;
 881
 882 fail:
 883         clear_node_page_dirty(page);
 884         f2fs_put_page(page, 1);
 885         return ERR_PTR(err);
 886 }
 887
 888 /*
 889  * Caller should do after getting the following values.
 890  * 0: f2fs_put_page(page, 0)
 891  * LOCKED_PAGE: f2fs_put_page(page, 1)
 892  * error: nothing
 893  */
 894 static int read_node_page(struct page *page, int rw)
 895 {
 896         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 897         struct node_info ni;
 898
 899         get_node_info(sbi, page->index, &ni);
 900
 901         if (ni.blk_addr == NULL_ADDR) {
 902                 f2fs_put_page(page, 1);
 903                 return -ENOENT;
 904         }
 905
 906         if (PageUptodate(page))
 907                 return LOCKED_PAGE;
 908
 909         return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
 910 }
 911
 912 /*
 913  * Readahead a node page
 914  */
 915 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 916 {
 917         struct address_space *mapping = sbi->node_inode->i_mapping;
 918         struct page *apage;
 919         int err;
 920
 921         apage = find_get_page(mapping, nid);
 922         if (apage && PageUptodate(apage)) {
 923                 f2fs_put_page(apage, 0);
 924                 return;
 925         }
 926         f2fs_put_page(apage, 0);
 927
 928         apage = grab_cache_page(mapping, nid);
 929         if (!apage)
 930                 return;
 931
 932         err = read_node_page(apage, READA);
 933         if (err == 0)
 934                 f2fs_put_page(apage, 0);
 935         else if (err == LOCKED_PAGE)
 936                 f2fs_put_page(apage, 1);
 937 }
 938
 939 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 940 {
 941         struct address_space *mapping = sbi->node_inode->i_mapping;
 942         struct page *page;
 943         int err;
 944 repeat:
 945         page = grab_cache_page(mapping, nid);
 946         if (!page)
 947                 return ERR_PTR(-ENOMEM);
 948
 949         err = read_node_page(page, READ_SYNC);
 950         if (err < 0)
 951                 return ERR_PTR(err);
 952         else if (err == LOCKED_PAGE)
 953                 goto got_it;
 954
 955         lock_page(page);
 956         if (!PageUptodate(page)) {
 957                 f2fs_put_page(page, 1);
 958                 return ERR_PTR(-EIO);
 959         }
 960         if (page->mapping != mapping) {
 961                 f2fs_put_page(page, 1);
 962                 goto repeat;
 963         }
 964 got_it:
 965         f2fs_bug_on(nid != nid_of_node(page));
 966         mark_page_accessed(page);
 967         return page;
 968 }
 969
 970 /*
 971  * Return a locked page for the desired node page.
 972  * And, readahead MAX_RA_NODE number of node pages.
 973  */
 974 struct page *get_node_page_ra(struct page *parent, int start)
 975 {
 976         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 977         struct address_space *mapping = sbi->node_inode->i_mapping;
 978         struct blk_plug plug;
 979         struct page *page;
 980         int err, i, end;
 981         nid_t nid;
 982
 983         /* First, try getting the desired direct node. */
 984         nid = get_nid(parent, start, false);
 985         if (!nid)
 986                 return ERR_PTR(-ENOENT);
 987 repeat:
 988         page = grab_cache_page(mapping, nid);
 989         if (!page)
 990                 return ERR_PTR(-ENOMEM);
 991
 992         err = read_node_page(page, READ_SYNC);
 993         if (err < 0)
 994                 return ERR_PTR(err);
 995         else if (err == LOCKED_PAGE)
 996                 goto page_hit;
 997
 998         blk_start_plug(&plug);
 999
1000         /* Then, try readahead for siblings of the desired node */
1001         end = start + MAX_RA_NODE;
1002         end = min(end, NIDS_PER_BLOCK);
1003         for (i = start + 1; i < end; i++) {
1004                 nid = get_nid(parent, i, false);
1005                 if (!nid)
1006                         continue;
1007                 ra_node_page(sbi, nid);
1008         }
1009
1010         blk_finish_plug(&plug);
1011
1012         lock_page(page);
1013         if (page->mapping != mapping) {
1014                 f2fs_put_page(page, 1);
1015                 goto repeat;
1016         }
1017 page_hit:
1018         if (!PageUptodate(page)) {
1019                 f2fs_put_page(page, 1);
1020                 return ERR_PTR(-EIO);
1021         }
1022         mark_page_accessed(page);
1023         return page;
1024 }
1025
1026 void sync_inode_page(struct dnode_of_data *dn)
1027 {
1028         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1029                 update_inode(dn->inode, dn->node_page);
1030         } else if (dn->inode_page) {
1031                 if (!dn->inode_page_locked)
1032                         lock_page(dn->inode_page);
1033                 update_inode(dn->inode, dn->inode_page);
1034                 if (!dn->inode_page_locked)
1035                         unlock_page(dn->inode_page);
1036         } else {
1037                 update_inode_page(dn->inode);
1038         }
1039 }
1040
1041 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1042                                         struct writeback_control *wbc)
1043 {
1044         struct address_space *mapping = sbi->node_inode->i_mapping;
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, mapping, &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 != mapping)) {
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                         mapping->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, wbc->sync_mode == WB_SYNC_ALL,
1140                                                                         WRITE);
1141         return nwritten;
1142 }
1143
1144 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1145 {
1146         struct address_space *mapping = sbi->node_inode->i_mapping;
1147         pgoff_t index = 0, end = LONG_MAX;
1148         struct pagevec pvec;
1149         int nr_pages;
1150         int ret2 = 0, ret = 0;
1151
1152         pagevec_init(&pvec, 0);
1153         while ((index <= end) &&
1154                         (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1155                         PAGECACHE_TAG_WRITEBACK,
1156                         min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
1157                 unsigned i;
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 (page->index > end)
1164                                 continue;
1165
1166                         if (ino && ino_of_node(page) == ino) {
1167                                 wait_on_page_writeback(page);
1168                                 if (TestClearPageError(page))
1169                                         ret = -EIO;
1170                         }
1171                 }
1172                 pagevec_release(&pvec);
1173                 cond_resched();
1174         }
1175
1176         if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
1177                 ret2 = -ENOSPC;
1178         if (test_and_clear_bit(AS_EIO, &mapping->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
1193         if (sbi->por_doing)
1194                 goto redirty_out;
1195
1196         wait_on_page_writeback(page);
1197
1198         /* get old block addr of this node page */
1199         nid = nid_of_node(page);
1200         f2fs_bug_on(page->index != nid);
1201
1202         get_node_info(sbi, nid, &ni);
1203
1204         /* This page is already truncated */
1205         if (ni.blk_addr == NULL_ADDR) {
1206                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1207                 unlock_page(page);
1208                 return 0;
1209         }
1210
1211         if (wbc->for_reclaim)
1212                 goto redirty_out;
1213
1214         mutex_lock(&sbi->node_write);
1215         set_page_writeback(page);
1216         write_node_page(sbi, page, nid, ni.blk_addr, &new_addr);
1217         set_node_addr(sbi, &ni, new_addr);
1218         dec_page_count(sbi, F2FS_DIRTY_NODES);
1219         mutex_unlock(&sbi->node_write);
1220         unlock_page(page);
1221         return 0;
1222
1223 redirty_out:
1224         dec_page_count(sbi, F2FS_DIRTY_NODES);
1225         wbc->pages_skipped++;
1226         set_page_dirty(page);
1227         return AOP_WRITEPAGE_ACTIVATE;
1228 }
1229
1230 /*
1231  * It is very important to gather dirty pages and write at once, so that we can
1232  * submit a big bio without interfering other data writes.
1233  * Be default, 512 pages (2MB) * 3 node types, is more reasonable.
1234  */
1235 #define COLLECT_DIRTY_NODES     1536
1236 static int f2fs_write_node_pages(struct address_space *mapping,
1237                             struct writeback_control *wbc)
1238 {
1239         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1240         long nr_to_write = wbc->nr_to_write;
1241
1242         /* balancing f2fs's metadata in background */
1243         f2fs_balance_fs_bg(sbi);
1244
1245         /* collect a number of dirty node pages and write together */
1246         if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1247                 return 0;
1248
1249         /* if mounting is failed, skip writing node pages */
1250         wbc->nr_to_write = 3 * max_hw_blocks(sbi);
1251         sync_node_pages(sbi, 0, wbc);
1252         wbc->nr_to_write = nr_to_write - (3 * max_hw_blocks(sbi) -
1253                                                 wbc->nr_to_write);
1254         return 0;
1255 }
1256
1257 static int f2fs_set_node_page_dirty(struct page *page)
1258 {
1259         struct address_space *mapping = page->mapping;
1260         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1261
1262         trace_f2fs_set_page_dirty(page, NODE);
1263
1264         SetPageUptodate(page);
1265         if (!PageDirty(page)) {
1266                 __set_page_dirty_nobuffers(page);
1267                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1268                 SetPagePrivate(page);
1269                 return 1;
1270         }
1271         return 0;
1272 }
1273
1274 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1275                                       unsigned int length)
1276 {
1277         struct inode *inode = page->mapping->host;
1278         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1279         if (PageDirty(page))
1280                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1281         ClearPagePrivate(page);
1282 }
1283
1284 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1285 {
1286         ClearPagePrivate(page);
1287         return 1;
1288 }
1289
1290 /*
1291  * Structure of the f2fs node operations
1292  */
1293 const struct address_space_operations f2fs_node_aops = {
1294         .writepage      = f2fs_write_node_page,
1295         .writepages     = f2fs_write_node_pages,
1296         .set_page_dirty = f2fs_set_node_page_dirty,
1297         .invalidatepage = f2fs_invalidate_node_page,
1298         .releasepage    = f2fs_release_node_page,
1299 };
1300
1301 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1302 {
1303         struct list_head *this;
1304         struct free_nid *i;
1305         list_for_each(this, head) {
1306                 i = list_entry(this, struct free_nid, list);
1307                 if (i->nid == n)
1308                         return i;
1309         }
1310         return NULL;
1311 }
1312
1313 static void __del_from_free_nid_list(struct free_nid *i)
1314 {
1315         list_del(&i->list);
1316         kmem_cache_free(free_nid_slab, i);
1317 }
1318
1319 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build)
1320 {
1321         struct free_nid *i;
1322         struct nat_entry *ne;
1323         bool allocated = false;
1324
1325         if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1326                 return -1;
1327
1328         /* 0 nid should not be used */
1329         if (nid == 0)
1330                 return 0;
1331
1332         if (build) {
1333                 /* do not add allocated nids */
1334                 read_lock(&nm_i->nat_tree_lock);
1335                 ne = __lookup_nat_cache(nm_i, nid);
1336                 if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1337                         allocated = true;
1338                 read_unlock(&nm_i->nat_tree_lock);
1339                 if (allocated)
1340                         return 0;
1341         }
1342
1343         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1344         i->nid = nid;
1345         i->state = NID_NEW;
1346
1347         spin_lock(&nm_i->free_nid_list_lock);
1348         if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1349                 spin_unlock(&nm_i->free_nid_list_lock);
1350                 kmem_cache_free(free_nid_slab, i);
1351                 return 0;
1352         }
1353         list_add_tail(&i->list, &nm_i->free_nid_list);
1354         nm_i->fcnt++;
1355         spin_unlock(&nm_i->free_nid_list_lock);
1356         return 1;
1357 }
1358
1359 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1360 {
1361         struct free_nid *i;
1362         spin_lock(&nm_i->free_nid_list_lock);
1363         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1364         if (i && i->state == NID_NEW) {
1365                 __del_from_free_nid_list(i);
1366                 nm_i->fcnt--;
1367         }
1368         spin_unlock(&nm_i->free_nid_list_lock);
1369 }
1370
1371 static void scan_nat_page(struct f2fs_nm_info *nm_i,
1372                         struct page *nat_page, nid_t start_nid)
1373 {
1374         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1375         block_t blk_addr;
1376         int i;
1377
1378         i = start_nid % NAT_ENTRY_PER_BLOCK;
1379
1380         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1381
1382                 if (start_nid >= nm_i->max_nid)
1383                         break;
1384
1385                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1386                 f2fs_bug_on(blk_addr == NEW_ADDR);
1387                 if (blk_addr == NULL_ADDR) {
1388                         if (add_free_nid(nm_i, start_nid, true) < 0)
1389                                 break;
1390                 }
1391         }
1392 }
1393
1394 static void build_free_nids(struct f2fs_sb_info *sbi)
1395 {
1396         struct f2fs_nm_info *nm_i = NM_I(sbi);
1397         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1398         struct f2fs_summary_block *sum = curseg->sum_blk;
1399         int i = 0;
1400         nid_t nid = nm_i->next_scan_nid;
1401
1402         /* Enough entries */
1403         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1404                 return;
1405
1406         /* readahead nat pages to be scanned */
1407         ra_nat_pages(sbi, nid);
1408
1409         while (1) {
1410                 struct page *page = get_current_nat_page(sbi, nid);
1411
1412                 scan_nat_page(nm_i, page, nid);
1413                 f2fs_put_page(page, 1);
1414
1415                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1416                 if (nid >= nm_i->max_nid)
1417                         nid = 0;
1418
1419                 if (i++ == FREE_NID_PAGES)
1420                         break;
1421         }
1422
1423         /* go to the next free nat pages to find free nids abundantly */
1424         nm_i->next_scan_nid = nid;
1425
1426         /* find free nids from current sum_pages */
1427         mutex_lock(&curseg->curseg_mutex);
1428         for (i = 0; i < nats_in_cursum(sum); i++) {
1429                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1430                 nid = le32_to_cpu(nid_in_journal(sum, i));
1431                 if (addr == NULL_ADDR)
1432                         add_free_nid(nm_i, nid, true);
1433                 else
1434                         remove_free_nid(nm_i, nid);
1435         }
1436         mutex_unlock(&curseg->curseg_mutex);
1437 }
1438
1439 /*
1440  * If this function returns success, caller can obtain a new nid
1441  * from second parameter of this function.
1442  * The returned nid could be used ino as well as nid when inode is created.
1443  */
1444 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1445 {
1446         struct f2fs_nm_info *nm_i = NM_I(sbi);
1447         struct free_nid *i = NULL;
1448         struct list_head *this;
1449 retry:
1450         if (sbi->total_valid_node_count + 1 >= nm_i->max_nid)
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 && !sbi->on_build_free_nids) {
1457                 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1458                 list_for_each(this, &nm_i->free_nid_list) {
1459                         i = list_entry(this, struct free_nid, list);
1460                         if (i->state == NID_NEW)
1461                                 break;
1462                 }
1463
1464                 f2fs_bug_on(i->state != NID_NEW);
1465                 *nid = i->nid;
1466                 i->state = NID_ALLOC;
1467                 nm_i->fcnt--;
1468                 spin_unlock(&nm_i->free_nid_list_lock);
1469                 return true;
1470         }
1471         spin_unlock(&nm_i->free_nid_list_lock);
1472
1473         /* Let's scan nat pages and its caches to get free nids */
1474         mutex_lock(&nm_i->build_lock);
1475         sbi->on_build_free_nids = true;
1476         build_free_nids(sbi);
1477         sbi->on_build_free_nids = false;
1478         mutex_unlock(&nm_i->build_lock);
1479         goto retry;
1480 }
1481
1482 /*
1483  * alloc_nid() should be called prior to this function.
1484  */
1485 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1486 {
1487         struct f2fs_nm_info *nm_i = NM_I(sbi);
1488         struct free_nid *i;
1489
1490         spin_lock(&nm_i->free_nid_list_lock);
1491         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1492         f2fs_bug_on(!i || i->state != NID_ALLOC);
1493         __del_from_free_nid_list(i);
1494         spin_unlock(&nm_i->free_nid_list_lock);
1495 }
1496
1497 /*
1498  * alloc_nid() should be called prior to this function.
1499  */
1500 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1501 {
1502         struct f2fs_nm_info *nm_i = NM_I(sbi);
1503         struct free_nid *i;
1504
1505         if (!nid)
1506                 return;
1507
1508         spin_lock(&nm_i->free_nid_list_lock);
1509         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1510         f2fs_bug_on(!i || i->state != NID_ALLOC);
1511         if (nm_i->fcnt > 2 * MAX_FREE_NIDS) {
1512                 __del_from_free_nid_list(i);
1513         } else {
1514                 i->state = NID_NEW;
1515                 nm_i->fcnt++;
1516         }
1517         spin_unlock(&nm_i->free_nid_list_lock);
1518 }
1519
1520 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1521                 struct f2fs_summary *sum, struct node_info *ni,
1522                 block_t new_blkaddr)
1523 {
1524         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1525         set_node_addr(sbi, ni, new_blkaddr);
1526         clear_node_page_dirty(page);
1527 }
1528
1529 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1530 {
1531         struct address_space *mapping = sbi->node_inode->i_mapping;
1532         struct f2fs_node *src, *dst;
1533         nid_t ino = ino_of_node(page);
1534         struct node_info old_ni, new_ni;
1535         struct page *ipage;
1536
1537         ipage = grab_cache_page(mapping, ino);
1538         if (!ipage)
1539                 return -ENOMEM;
1540
1541         /* Should not use this inode  from free nid list */
1542         remove_free_nid(NM_I(sbi), ino);
1543
1544         get_node_info(sbi, ino, &old_ni);
1545         SetPageUptodate(ipage);
1546         fill_node_footer(ipage, ino, ino, 0, true);
1547
1548         src = F2FS_NODE(page);
1549         dst = F2FS_NODE(ipage);
1550
1551         memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i);
1552         dst->i.i_size = 0;
1553         dst->i.i_blocks = cpu_to_le64(1);
1554         dst->i.i_links = cpu_to_le32(1);
1555         dst->i.i_xattr_nid = 0;
1556
1557         new_ni = old_ni;
1558         new_ni.ino = ino;
1559
1560         if (!inc_valid_node_count(sbi, NULL))
1561                 WARN_ON(1);
1562         set_node_addr(sbi, &new_ni, NEW_ADDR);
1563         inc_valid_inode_count(sbi);
1564         f2fs_put_page(ipage, 1);
1565         return 0;
1566 }
1567
1568 /*
1569  * ra_sum_pages() merge contiguous pages into one bio and submit.
1570  * these pre-readed pages are linked in pages list.
1571  */
1572 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
1573                                 int start, int nrpages)
1574 {
1575         struct page *page;
1576         int page_idx = start;
1577
1578         for (; page_idx < start + nrpages; page_idx++) {
1579                 /* alloc temporal page for read node summary info*/
1580                 page = alloc_page(GFP_F2FS_ZERO);
1581                 if (!page) {
1582                         struct page *tmp;
1583                         list_for_each_entry_safe(page, tmp, pages, lru) {
1584                                 list_del(&page->lru);
1585                                 unlock_page(page);
1586                                 __free_pages(page, 0);
1587                         }
1588                         return -ENOMEM;
1589                 }
1590
1591                 lock_page(page);
1592                 page->index = page_idx;
1593                 list_add_tail(&page->lru, pages);
1594         }
1595
1596         list_for_each_entry(page, pages, lru)
1597                 f2fs_submit_page_mbio(sbi, page, page->index, META, READ);
1598
1599         f2fs_submit_merged_bio(sbi, META, true, READ);
1600         return 0;
1601 }
1602
1603 int restore_node_summary(struct f2fs_sb_info *sbi,
1604                         unsigned int segno, struct f2fs_summary_block *sum)
1605 {
1606         struct f2fs_node *rn;
1607         struct f2fs_summary *sum_entry;
1608         struct page *page, *tmp;
1609         block_t addr;
1610         int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1611         int i, last_offset, nrpages, err = 0;
1612         LIST_HEAD(page_list);
1613
1614         /* scan the node segment */
1615         last_offset = sbi->blocks_per_seg;
1616         addr = START_BLOCK(sbi, segno);
1617         sum_entry = &sum->entries[0];
1618
1619         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1620                 nrpages = min(last_offset - i, bio_blocks);
1621
1622                 /* read ahead node pages */
1623                 err = ra_sum_pages(sbi, &page_list, addr, nrpages);
1624                 if (err)
1625                         return err;
1626
1627                 list_for_each_entry_safe(page, tmp, &page_list, lru) {
1628
1629                         lock_page(page);
1630                         if(PageUptodate(page)) {
1631                                 rn = F2FS_NODE(page);
1632                                 sum_entry->nid = rn->footer.nid;
1633                                 sum_entry->version = 0;
1634                                 sum_entry->ofs_in_node = 0;
1635                                 sum_entry++;
1636                         } else {
1637                                 err = -EIO;
1638                         }
1639
1640                         list_del(&page->lru);
1641                         unlock_page(page);
1642                         __free_pages(page, 0);
1643                 }
1644         }
1645         return err;
1646 }
1647
1648 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1649 {
1650         struct f2fs_nm_info *nm_i = NM_I(sbi);
1651         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1652         struct f2fs_summary_block *sum = curseg->sum_blk;
1653         int i;
1654
1655         mutex_lock(&curseg->curseg_mutex);
1656
1657         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1658                 mutex_unlock(&curseg->curseg_mutex);
1659                 return false;
1660         }
1661
1662         for (i = 0; i < nats_in_cursum(sum); i++) {
1663                 struct nat_entry *ne;
1664                 struct f2fs_nat_entry raw_ne;
1665                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1666
1667                 raw_ne = nat_in_journal(sum, i);
1668 retry:
1669                 write_lock(&nm_i->nat_tree_lock);
1670                 ne = __lookup_nat_cache(nm_i, nid);
1671                 if (ne) {
1672                         __set_nat_cache_dirty(nm_i, ne);
1673                         write_unlock(&nm_i->nat_tree_lock);
1674                         continue;
1675                 }
1676                 ne = grab_nat_entry(nm_i, nid);
1677                 if (!ne) {
1678                         write_unlock(&nm_i->nat_tree_lock);
1679                         goto retry;
1680                 }
1681                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1682                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1683                 nat_set_version(ne, raw_ne.version);
1684                 __set_nat_cache_dirty(nm_i, ne);
1685                 write_unlock(&nm_i->nat_tree_lock);
1686         }
1687         update_nats_in_cursum(sum, -i);
1688         mutex_unlock(&curseg->curseg_mutex);
1689         return true;
1690 }
1691
1692 /*
1693  * This function is called during the checkpointing process.
1694  */
1695 void flush_nat_entries(struct f2fs_sb_info *sbi)
1696 {
1697         struct f2fs_nm_info *nm_i = NM_I(sbi);
1698         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1699         struct f2fs_summary_block *sum = curseg->sum_blk;
1700         struct list_head *cur, *n;
1701         struct page *page = NULL;
1702         struct f2fs_nat_block *nat_blk = NULL;
1703         nid_t start_nid = 0, end_nid = 0;
1704         bool flushed;
1705
1706         flushed = flush_nats_in_journal(sbi);
1707
1708         if (!flushed)
1709                 mutex_lock(&curseg->curseg_mutex);
1710
1711         /* 1) flush dirty nat caches */
1712         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1713                 struct nat_entry *ne;
1714                 nid_t nid;
1715                 struct f2fs_nat_entry raw_ne;
1716                 int offset = -1;
1717                 block_t new_blkaddr;
1718
1719                 ne = list_entry(cur, struct nat_entry, list);
1720                 nid = nat_get_nid(ne);
1721
1722                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1723                         continue;
1724                 if (flushed)
1725                         goto to_nat_page;
1726
1727                 /* if there is room for nat enries in curseg->sumpage */
1728                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1729                 if (offset >= 0) {
1730                         raw_ne = nat_in_journal(sum, offset);
1731                         goto flush_now;
1732                 }
1733 to_nat_page:
1734                 if (!page || (start_nid > nid || nid > end_nid)) {
1735                         if (page) {
1736                                 f2fs_put_page(page, 1);
1737                                 page = NULL;
1738                         }
1739                         start_nid = START_NID(nid);
1740                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1741
1742                         /*
1743                          * get nat block with dirty flag, increased reference
1744                          * count, mapped and lock
1745                          */
1746                         page = get_next_nat_page(sbi, start_nid);
1747                         nat_blk = page_address(page);
1748                 }
1749
1750                 f2fs_bug_on(!nat_blk);
1751                 raw_ne = nat_blk->entries[nid - start_nid];
1752 flush_now:
1753                 new_blkaddr = nat_get_blkaddr(ne);
1754
1755                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1756                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1757                 raw_ne.version = nat_get_version(ne);
1758
1759                 if (offset < 0) {
1760                         nat_blk->entries[nid - start_nid] = raw_ne;
1761                 } else {
1762                         nat_in_journal(sum, offset) = raw_ne;
1763                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1764                 }
1765
1766                 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1767                                 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1768                         write_lock(&nm_i->nat_tree_lock);
1769                         __del_from_nat_cache(nm_i, ne);
1770                         write_unlock(&nm_i->nat_tree_lock);
1771                 } else {
1772                         write_lock(&nm_i->nat_tree_lock);
1773                         __clear_nat_cache_dirty(nm_i, ne);
1774                         ne->checkpointed = true;
1775                         write_unlock(&nm_i->nat_tree_lock);
1776                 }
1777         }
1778         if (!flushed)
1779                 mutex_unlock(&curseg->curseg_mutex);
1780         f2fs_put_page(page, 1);
1781
1782         /* 2) shrink nat caches if necessary */
1783         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1784 }
1785
1786 static int init_node_manager(struct f2fs_sb_info *sbi)
1787 {
1788         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1789         struct f2fs_nm_info *nm_i = NM_I(sbi);
1790         unsigned char *version_bitmap;
1791         unsigned int nat_segs, nat_blocks;
1792
1793         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1794
1795         /* segment_count_nat includes pair segment so divide to 2. */
1796         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1797         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1798         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1799         nm_i->fcnt = 0;
1800         nm_i->nat_cnt = 0;
1801
1802         INIT_LIST_HEAD(&nm_i->free_nid_list);
1803         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1804         INIT_LIST_HEAD(&nm_i->nat_entries);
1805         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1806
1807         mutex_init(&nm_i->build_lock);
1808         spin_lock_init(&nm_i->free_nid_list_lock);
1809         rwlock_init(&nm_i->nat_tree_lock);
1810
1811         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1812         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1813         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1814         if (!version_bitmap)
1815                 return -EFAULT;
1816
1817         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1818                                         GFP_KERNEL);
1819         if (!nm_i->nat_bitmap)
1820                 return -ENOMEM;
1821         return 0;
1822 }
1823
1824 int build_node_manager(struct f2fs_sb_info *sbi)
1825 {
1826         int err;
1827
1828         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1829         if (!sbi->nm_info)
1830                 return -ENOMEM;
1831
1832         err = init_node_manager(sbi);
1833         if (err)
1834                 return err;
1835
1836         build_free_nids(sbi);
1837         return 0;
1838 }
1839
1840 void destroy_node_manager(struct f2fs_sb_info *sbi)
1841 {
1842         struct f2fs_nm_info *nm_i = NM_I(sbi);
1843         struct free_nid *i, *next_i;
1844         struct nat_entry *natvec[NATVEC_SIZE];
1845         nid_t nid = 0;
1846         unsigned int found;
1847
1848         if (!nm_i)
1849                 return;
1850
1851         /* destroy free nid list */
1852         spin_lock(&nm_i->free_nid_list_lock);
1853         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1854                 f2fs_bug_on(i->state == NID_ALLOC);
1855                 __del_from_free_nid_list(i);
1856                 nm_i->fcnt--;
1857         }
1858         f2fs_bug_on(nm_i->fcnt);
1859         spin_unlock(&nm_i->free_nid_list_lock);
1860
1861         /* destroy nat cache */
1862         write_lock(&nm_i->nat_tree_lock);
1863         while ((found = __gang_lookup_nat_cache(nm_i,
1864                                         nid, NATVEC_SIZE, natvec))) {
1865                 unsigned idx;
1866                 for (idx = 0; idx < found; idx++) {
1867                         struct nat_entry *e = natvec[idx];
1868                         nid = nat_get_nid(e) + 1;
1869                         __del_from_nat_cache(nm_i, e);
1870                 }
1871         }
1872         f2fs_bug_on(nm_i->nat_cnt);
1873         write_unlock(&nm_i->nat_tree_lock);
1874
1875         kfree(nm_i->nat_bitmap);
1876         sbi->nm_info = NULL;
1877         kfree(nm_i);
1878 }
1879
1880 int __init create_node_manager_caches(void)
1881 {
1882         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1883                         sizeof(struct nat_entry), NULL);
1884         if (!nat_entry_slab)
1885                 return -ENOMEM;
1886
1887         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1888                         sizeof(struct free_nid), NULL);
1889         if (!free_nid_slab) {
1890                 kmem_cache_destroy(nat_entry_slab);
1891                 return -ENOMEM;
1892         }
1893         return 0;
1894 }
1895
1896 void destroy_node_manager_caches(void)
1897 {
1898         kmem_cache_destroy(free_nid_slab);
1899         kmem_cache_destroy(nat_entry_slab);
1900 }