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.h"
  23 #include <trace/events/f2fs.h>
  24
  25 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
  26
  27 static struct kmem_cache *nat_entry_slab;
  28 static struct kmem_cache *free_nid_slab;
  29 static struct kmem_cache *nat_entry_set_slab;
  30
  31 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
  32 {
  33         struct f2fs_nm_info *nm_i = NM_I(sbi);
  34         struct sysinfo val;
  35         unsigned long avail_ram;
  36         unsigned long mem_size = 0;
  37         bool res = false;
  38
  39         si_meminfo(&val);
  40
  41         /* only uses low memory */
  42         avail_ram = val.totalram - val.totalhigh;
  43
  44         /*
  45          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  46          */
  47         if (type == FREE_NIDS) {
  48                 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
  49                                                         PAGE_SHIFT;
  50                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  51         } else if (type == NAT_ENTRIES) {
  52                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
  53                                                         PAGE_SHIFT;
  54                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  55                 if (excess_cached_nats(sbi))
  56                         res = false;
  57                 if (nm_i->nat_cnt > DEF_NAT_CACHE_THRESHOLD)
  58                         res = false;
  59         } else if (type == DIRTY_DENTS) {
  60                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  61                         return false;
  62                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  63                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  64         } else if (type == INO_ENTRIES) {
  65                 int i;
  66
  67                 for (i = 0; i <= UPDATE_INO; i++)
  68                         mem_size += (sbi->im[i].ino_num *
  69                                 sizeof(struct ino_entry)) >> PAGE_SHIFT;
  70                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  71         } else if (type == EXTENT_CACHE) {
  72                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  73                                 sizeof(struct extent_tree) +
  74                                 atomic_read(&sbi->total_ext_node) *
  75                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  76                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  77         } else {
  78                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
  79                         return true;
  80         }
  81         return res;
  82 }
  83
  84 static void clear_node_page_dirty(struct page *page)
  85 {
  86         struct address_space *mapping = page->mapping;
  87         unsigned int long flags;
  88
  89         if (PageDirty(page)) {
  90                 spin_lock_irqsave(&mapping->tree_lock, flags);
  91                 radix_tree_tag_clear(&mapping->page_tree,
  92                                 page_index(page),
  93                                 PAGECACHE_TAG_DIRTY);
  94                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  95
  96                 clear_page_dirty_for_io(page);
  97                 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
  98         }
  99         ClearPageUptodate(page);
 100 }
 101
 102 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 103 {
 104         pgoff_t index = current_nat_addr(sbi, nid);
 105         return get_meta_page(sbi, index);
 106 }
 107
 108 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 109 {
 110         struct page *src_page;
 111         struct page *dst_page;
 112         pgoff_t src_off;
 113         pgoff_t dst_off;
 114         void *src_addr;
 115         void *dst_addr;
 116         struct f2fs_nm_info *nm_i = NM_I(sbi);
 117
 118         src_off = current_nat_addr(sbi, nid);
 119         dst_off = next_nat_addr(sbi, src_off);
 120
 121         /* get current nat block page with lock */
 122         src_page = get_meta_page(sbi, src_off);
 123         dst_page = grab_meta_page(sbi, dst_off);
 124         f2fs_bug_on(sbi, PageDirty(src_page));
 125
 126         src_addr = page_address(src_page);
 127         dst_addr = page_address(dst_page);
 128         memcpy(dst_addr, src_addr, PAGE_SIZE);
 129         set_page_dirty(dst_page);
 130         f2fs_put_page(src_page, 1);
 131
 132         set_to_next_nat(nm_i, nid);
 133
 134         return dst_page;
 135 }
 136
 137 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 138 {
 139         return radix_tree_lookup(&nm_i->nat_root, n);
 140 }
 141
 142 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 143                 nid_t start, unsigned int nr, struct nat_entry **ep)
 144 {
 145         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 146 }
 147
 148 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 149 {
 150         list_del(&e->list);
 151         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 152         nm_i->nat_cnt--;
 153         kmem_cache_free(nat_entry_slab, e);
 154 }
 155
 156 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 157                                                 struct nat_entry *ne)
 158 {
 159         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 160         struct nat_entry_set *head;
 161
 162         if (get_nat_flag(ne, IS_DIRTY))
 163                 return;
 164
 165         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 166         if (!head) {
 167                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 168
 169                 INIT_LIST_HEAD(&head->entry_list);
 170                 INIT_LIST_HEAD(&head->set_list);
 171                 head->set = set;
 172                 head->entry_cnt = 0;
 173                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 174         }
 175         list_move_tail(&ne->list, &head->entry_list);
 176         nm_i->dirty_nat_cnt++;
 177         head->entry_cnt++;
 178         set_nat_flag(ne, IS_DIRTY, true);
 179 }
 180
 181 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 182                                                 struct nat_entry *ne)
 183 {
 184         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 185         struct nat_entry_set *head;
 186
 187         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 188         if (head) {
 189                 list_move_tail(&ne->list, &nm_i->nat_entries);
 190                 set_nat_flag(ne, IS_DIRTY, false);
 191                 head->entry_cnt--;
 192                 nm_i->dirty_nat_cnt--;
 193         }
 194 }
 195
 196 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 197                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 198 {
 199         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 200                                                         start, nr);
 201 }
 202
 203 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 204 {
 205         struct f2fs_nm_info *nm_i = NM_I(sbi);
 206         struct nat_entry *e;
 207         bool need = false;
 208
 209         down_read(&nm_i->nat_tree_lock);
 210         e = __lookup_nat_cache(nm_i, nid);
 211         if (e) {
 212                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 213                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 214                         need = true;
 215         }
 216         up_read(&nm_i->nat_tree_lock);
 217         return need;
 218 }
 219
 220 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 221 {
 222         struct f2fs_nm_info *nm_i = NM_I(sbi);
 223         struct nat_entry *e;
 224         bool is_cp = true;
 225
 226         down_read(&nm_i->nat_tree_lock);
 227         e = __lookup_nat_cache(nm_i, nid);
 228         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 229                 is_cp = false;
 230         up_read(&nm_i->nat_tree_lock);
 231         return is_cp;
 232 }
 233
 234 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 235 {
 236         struct f2fs_nm_info *nm_i = NM_I(sbi);
 237         struct nat_entry *e;
 238         bool need_update = true;
 239
 240         down_read(&nm_i->nat_tree_lock);
 241         e = __lookup_nat_cache(nm_i, ino);
 242         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 243                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 244                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 245                 need_update = false;
 246         up_read(&nm_i->nat_tree_lock);
 247         return need_update;
 248 }
 249
 250 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 251 {
 252         struct nat_entry *new;
 253
 254         new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
 255         f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
 256         memset(new, 0, sizeof(struct nat_entry));
 257         nat_set_nid(new, nid);
 258         nat_reset_flag(new);
 259         list_add_tail(&new->list, &nm_i->nat_entries);
 260         nm_i->nat_cnt++;
 261         return new;
 262 }
 263
 264 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 265                                                 struct f2fs_nat_entry *ne)
 266 {
 267         struct f2fs_nm_info *nm_i = NM_I(sbi);
 268         struct nat_entry *e;
 269
 270         e = __lookup_nat_cache(nm_i, nid);
 271         if (!e) {
 272                 e = grab_nat_entry(nm_i, nid);
 273                 node_info_from_raw_nat(&e->ni, ne);
 274         } else {
 275                 f2fs_bug_on(sbi, nat_get_ino(e) != ne->ino ||
 276                                 nat_get_blkaddr(e) != ne->block_addr ||
 277                                 nat_get_version(e) != ne->version);
 278         }
 279 }
 280
 281 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 282                         block_t new_blkaddr, bool fsync_done)
 283 {
 284         struct f2fs_nm_info *nm_i = NM_I(sbi);
 285         struct nat_entry *e;
 286
 287         down_write(&nm_i->nat_tree_lock);
 288         e = __lookup_nat_cache(nm_i, ni->nid);
 289         if (!e) {
 290                 e = grab_nat_entry(nm_i, ni->nid);
 291                 copy_node_info(&e->ni, ni);
 292                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 293         } else if (new_blkaddr == NEW_ADDR) {
 294                 /*
 295                  * when nid is reallocated,
 296                  * previous nat entry can be remained in nat cache.
 297                  * So, reinitialize it with new information.
 298                  */
 299                 copy_node_info(&e->ni, ni);
 300                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 301         }
 302
 303         /* sanity check */
 304         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 305         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 306                         new_blkaddr == NULL_ADDR);
 307         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 308                         new_blkaddr == NEW_ADDR);
 309         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 310                         nat_get_blkaddr(e) != NULL_ADDR &&
 311                         new_blkaddr == NEW_ADDR);
 312
 313         /* increment version no as node is removed */
 314         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 315                 unsigned char version = nat_get_version(e);
 316                 nat_set_version(e, inc_node_version(version));
 317
 318                 /* in order to reuse the nid */
 319                 if (nm_i->next_scan_nid > ni->nid)
 320                         nm_i->next_scan_nid = ni->nid;
 321         }
 322
 323         /* change address */
 324         nat_set_blkaddr(e, new_blkaddr);
 325         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 326                 set_nat_flag(e, IS_CHECKPOINTED, false);
 327         __set_nat_cache_dirty(nm_i, e);
 328
 329         /* update fsync_mark if its inode nat entry is still alive */
 330         if (ni->nid != ni->ino)
 331                 e = __lookup_nat_cache(nm_i, ni->ino);
 332         if (e) {
 333                 if (fsync_done && ni->nid == ni->ino)
 334                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 335                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 336         }
 337         up_write(&nm_i->nat_tree_lock);
 338 }
 339
 340 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 341 {
 342         struct f2fs_nm_info *nm_i = NM_I(sbi);
 343         int nr = nr_shrink;
 344
 345         if (!down_write_trylock(&nm_i->nat_tree_lock))
 346                 return 0;
 347
 348         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 349                 struct nat_entry *ne;
 350                 ne = list_first_entry(&nm_i->nat_entries,
 351                                         struct nat_entry, list);
 352                 __del_from_nat_cache(nm_i, ne);
 353                 nr_shrink--;
 354         }
 355         up_write(&nm_i->nat_tree_lock);
 356         return nr - nr_shrink;
 357 }
 358
 359 /*
 360  * This function always returns success
 361  */
 362 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 363 {
 364         struct f2fs_nm_info *nm_i = NM_I(sbi);
 365         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 366         struct f2fs_journal *journal = curseg->journal;
 367         nid_t start_nid = START_NID(nid);
 368         struct f2fs_nat_block *nat_blk;
 369         struct page *page = NULL;
 370         struct f2fs_nat_entry ne;
 371         struct nat_entry *e;
 372         int i;
 373
 374         ni->nid = nid;
 375
 376         /* Check nat cache */
 377         down_read(&nm_i->nat_tree_lock);
 378         e = __lookup_nat_cache(nm_i, nid);
 379         if (e) {
 380                 ni->ino = nat_get_ino(e);
 381                 ni->blk_addr = nat_get_blkaddr(e);
 382                 ni->version = nat_get_version(e);
 383                 up_read(&nm_i->nat_tree_lock);
 384                 return;
 385         }
 386
 387         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 388
 389         /* Check current segment summary */
 390         down_read(&curseg->journal_rwsem);
 391         i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 392         if (i >= 0) {
 393                 ne = nat_in_journal(journal, i);
 394                 node_info_from_raw_nat(ni, &ne);
 395         }
 396         up_read(&curseg->journal_rwsem);
 397         if (i >= 0)
 398                 goto cache;
 399
 400         /* Fill node_info from nat page */
 401         page = get_current_nat_page(sbi, start_nid);
 402         nat_blk = (struct f2fs_nat_block *)page_address(page);
 403         ne = nat_blk->entries[nid - start_nid];
 404         node_info_from_raw_nat(ni, &ne);
 405         f2fs_put_page(page, 1);
 406 cache:
 407         up_read(&nm_i->nat_tree_lock);
 408         /* cache nat entry */
 409         down_write(&nm_i->nat_tree_lock);
 410         cache_nat_entry(sbi, nid, &ne);
 411         up_write(&nm_i->nat_tree_lock);
 412 }
 413
 414 /*
 415  * readahead MAX_RA_NODE number of node pages.
 416  */
 417 static void ra_node_pages(struct page *parent, int start, int n)
 418 {
 419         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 420         struct blk_plug plug;
 421         int i, end;
 422         nid_t nid;
 423
 424         blk_start_plug(&plug);
 425
 426         /* Then, try readahead for siblings of the desired node */
 427         end = start + n;
 428         end = min(end, NIDS_PER_BLOCK);
 429         for (i = start; i < end; i++) {
 430                 nid = get_nid(parent, i, false);
 431                 ra_node_page(sbi, nid);
 432         }
 433
 434         blk_finish_plug(&plug);
 435 }
 436
 437 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 438 {
 439         const long direct_index = ADDRS_PER_INODE(dn->inode);
 440         const long direct_blks = ADDRS_PER_BLOCK;
 441         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 442         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 443         int cur_level = dn->cur_level;
 444         int max_level = dn->max_level;
 445         pgoff_t base = 0;
 446
 447         if (!dn->max_level)
 448                 return pgofs + 1;
 449
 450         while (max_level-- > cur_level)
 451                 skipped_unit *= NIDS_PER_BLOCK;
 452
 453         switch (dn->max_level) {
 454         case 3:
 455                 base += 2 * indirect_blks;
 456         case 2:
 457                 base += 2 * direct_blks;
 458         case 1:
 459                 base += direct_index;
 460                 break;
 461         default:
 462                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 463         }
 464
 465         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 466 }
 467
 468 /*
 469  * The maximum depth is four.
 470  * Offset[0] will have raw inode offset.
 471  */
 472 static int get_node_path(struct inode *inode, long block,
 473                                 int offset[4], unsigned int noffset[4])
 474 {
 475         const long direct_index = ADDRS_PER_INODE(inode);
 476         const long direct_blks = ADDRS_PER_BLOCK;
 477         const long dptrs_per_blk = NIDS_PER_BLOCK;
 478         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 479         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 480         int n = 0;
 481         int level = 0;
 482
 483         noffset[0] = 0;
 484
 485         if (block < direct_index) {
 486                 offset[n] = block;
 487                 goto got;
 488         }
 489         block -= direct_index;
 490         if (block < direct_blks) {
 491                 offset[n++] = NODE_DIR1_BLOCK;
 492                 noffset[n] = 1;
 493                 offset[n] = block;
 494                 level = 1;
 495                 goto got;
 496         }
 497         block -= direct_blks;
 498         if (block < direct_blks) {
 499                 offset[n++] = NODE_DIR2_BLOCK;
 500                 noffset[n] = 2;
 501                 offset[n] = block;
 502                 level = 1;
 503                 goto got;
 504         }
 505         block -= direct_blks;
 506         if (block < indirect_blks) {
 507                 offset[n++] = NODE_IND1_BLOCK;
 508                 noffset[n] = 3;
 509                 offset[n++] = block / direct_blks;
 510                 noffset[n] = 4 + offset[n - 1];
 511                 offset[n] = block % direct_blks;
 512                 level = 2;
 513                 goto got;
 514         }
 515         block -= indirect_blks;
 516         if (block < indirect_blks) {
 517                 offset[n++] = NODE_IND2_BLOCK;
 518                 noffset[n] = 4 + dptrs_per_blk;
 519                 offset[n++] = block / direct_blks;
 520                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 521                 offset[n] = block % direct_blks;
 522                 level = 2;
 523                 goto got;
 524         }
 525         block -= indirect_blks;
 526         if (block < dindirect_blks) {
 527                 offset[n++] = NODE_DIND_BLOCK;
 528                 noffset[n] = 5 + (dptrs_per_blk * 2);
 529                 offset[n++] = block / indirect_blks;
 530                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 531                               offset[n - 1] * (dptrs_per_blk + 1);
 532                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 533                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 534                               offset[n - 2] * (dptrs_per_blk + 1) +
 535                               offset[n - 1];
 536                 offset[n] = block % direct_blks;
 537                 level = 3;
 538                 goto got;
 539         } else {
 540                 BUG();
 541         }
 542 got:
 543         return level;
 544 }
 545
 546 /*
 547  * Caller should call f2fs_put_dnode(dn).
 548  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 549  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 550  * In the case of RDONLY_NODE, we don't need to care about mutex.
 551  */
 552 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 553 {
 554         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 555         struct page *npage[4];
 556         struct page *parent = NULL;
 557         int offset[4];
 558         unsigned int noffset[4];
 559         nid_t nids[4];
 560         int level, i = 0;
 561         int err = 0;
 562
 563         level = get_node_path(dn->inode, index, offset, noffset);
 564
 565         nids[0] = dn->inode->i_ino;
 566         npage[0] = dn->inode_page;
 567
 568         if (!npage[0]) {
 569                 npage[0] = get_node_page(sbi, nids[0]);
 570                 if (IS_ERR(npage[0]))
 571                         return PTR_ERR(npage[0]);
 572         }
 573
 574         /* if inline_data is set, should not report any block indices */
 575         if (f2fs_has_inline_data(dn->inode) && index) {
 576                 err = -ENOENT;
 577                 f2fs_put_page(npage[0], 1);
 578                 goto release_out;
 579         }
 580
 581         parent = npage[0];
 582         if (level != 0)
 583                 nids[1] = get_nid(parent, offset[0], true);
 584         dn->inode_page = npage[0];
 585         dn->inode_page_locked = true;
 586
 587         /* get indirect or direct nodes */
 588         for (i = 1; i <= level; i++) {
 589                 bool done = false;
 590
 591                 if (!nids[i] && mode == ALLOC_NODE) {
 592                         /* alloc new node */
 593                         if (!alloc_nid(sbi, &(nids[i]))) {
 594                                 err = -ENOSPC;
 595                                 goto release_pages;
 596                         }
 597
 598                         dn->nid = nids[i];
 599                         npage[i] = new_node_page(dn, noffset[i], NULL);
 600                         if (IS_ERR(npage[i])) {
 601                                 alloc_nid_failed(sbi, nids[i]);
 602                                 err = PTR_ERR(npage[i]);
 603                                 goto release_pages;
 604                         }
 605
 606                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 607                         alloc_nid_done(sbi, nids[i]);
 608                         done = true;
 609                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 610                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 611                         if (IS_ERR(npage[i])) {
 612                                 err = PTR_ERR(npage[i]);
 613                                 goto release_pages;
 614                         }
 615                         done = true;
 616                 }
 617                 if (i == 1) {
 618                         dn->inode_page_locked = false;
 619                         unlock_page(parent);
 620                 } else {
 621                         f2fs_put_page(parent, 1);
 622                 }
 623
 624                 if (!done) {
 625                         npage[i] = get_node_page(sbi, nids[i]);
 626                         if (IS_ERR(npage[i])) {
 627                                 err = PTR_ERR(npage[i]);
 628                                 f2fs_put_page(npage[0], 0);
 629                                 goto release_out;
 630                         }
 631                 }
 632                 if (i < level) {
 633                         parent = npage[i];
 634                         nids[i + 1] = get_nid(parent, offset[i], false);
 635                 }
 636         }
 637         dn->nid = nids[level];
 638         dn->ofs_in_node = offset[level];
 639         dn->node_page = npage[level];
 640         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 641         return 0;
 642
 643 release_pages:
 644         f2fs_put_page(parent, 1);
 645         if (i > 1)
 646                 f2fs_put_page(npage[0], 0);
 647 release_out:
 648         dn->inode_page = NULL;
 649         dn->node_page = NULL;
 650         if (err == -ENOENT) {
 651                 dn->cur_level = i;
 652                 dn->max_level = level;
 653                 dn->ofs_in_node = offset[level];
 654         }
 655         return err;
 656 }
 657
 658 static void truncate_node(struct dnode_of_data *dn)
 659 {
 660         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 661         struct node_info ni;
 662
 663         get_node_info(sbi, dn->nid, &ni);
 664         if (dn->inode->i_blocks == 0) {
 665                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 666                 goto invalidate;
 667         }
 668         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 669
 670         /* Deallocate node address */
 671         invalidate_blocks(sbi, ni.blk_addr);
 672         dec_valid_node_count(sbi, dn->inode);
 673         set_node_addr(sbi, &ni, NULL_ADDR, false);
 674
 675         if (dn->nid == dn->inode->i_ino) {
 676                 remove_orphan_inode(sbi, dn->nid);
 677                 dec_valid_inode_count(sbi);
 678                 f2fs_inode_synced(dn->inode);
 679         }
 680 invalidate:
 681         clear_node_page_dirty(dn->node_page);
 682         set_sbi_flag(sbi, SBI_IS_DIRTY);
 683
 684         f2fs_put_page(dn->node_page, 1);
 685
 686         invalidate_mapping_pages(NODE_MAPPING(sbi),
 687                         dn->node_page->index, dn->node_page->index);
 688
 689         dn->node_page = NULL;
 690         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 691 }
 692
 693 static int truncate_dnode(struct dnode_of_data *dn)
 694 {
 695         struct page *page;
 696
 697         if (dn->nid == 0)
 698                 return 1;
 699
 700         /* get direct node */
 701         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 702         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 703                 return 1;
 704         else if (IS_ERR(page))
 705                 return PTR_ERR(page);
 706
 707         /* Make dnode_of_data for parameter */
 708         dn->node_page = page;
 709         dn->ofs_in_node = 0;
 710         truncate_data_blocks(dn);
 711         truncate_node(dn);
 712         return 1;
 713 }
 714
 715 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 716                                                 int ofs, int depth)
 717 {
 718         struct dnode_of_data rdn = *dn;
 719         struct page *page;
 720         struct f2fs_node *rn;
 721         nid_t child_nid;
 722         unsigned int child_nofs;
 723         int freed = 0;
 724         int i, ret;
 725
 726         if (dn->nid == 0)
 727                 return NIDS_PER_BLOCK + 1;
 728
 729         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 730
 731         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 732         if (IS_ERR(page)) {
 733                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 734                 return PTR_ERR(page);
 735         }
 736
 737         ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 738
 739         rn = F2FS_NODE(page);
 740         if (depth < 3) {
 741                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 742                         child_nid = le32_to_cpu(rn->in.nid[i]);
 743                         if (child_nid == 0)
 744                                 continue;
 745                         rdn.nid = child_nid;
 746                         ret = truncate_dnode(&rdn);
 747                         if (ret < 0)
 748                                 goto out_err;
 749                         if (set_nid(page, i, 0, false))
 750                                 dn->node_changed = true;
 751                 }
 752         } else {
 753                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 754                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 755                         child_nid = le32_to_cpu(rn->in.nid[i]);
 756                         if (child_nid == 0) {
 757                                 child_nofs += NIDS_PER_BLOCK + 1;
 758                                 continue;
 759                         }
 760                         rdn.nid = child_nid;
 761                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 762                         if (ret == (NIDS_PER_BLOCK + 1)) {
 763                                 if (set_nid(page, i, 0, false))
 764                                         dn->node_changed = true;
 765                                 child_nofs += ret;
 766                         } else if (ret < 0 && ret != -ENOENT) {
 767                                 goto out_err;
 768                         }
 769                 }
 770                 freed = child_nofs;
 771         }
 772
 773         if (!ofs) {
 774                 /* remove current indirect node */
 775                 dn->node_page = page;
 776                 truncate_node(dn);
 777                 freed++;
 778         } else {
 779                 f2fs_put_page(page, 1);
 780         }
 781         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 782         return freed;
 783
 784 out_err:
 785         f2fs_put_page(page, 1);
 786         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 787         return ret;
 788 }
 789
 790 static int truncate_partial_nodes(struct dnode_of_data *dn,
 791                         struct f2fs_inode *ri, int *offset, int depth)
 792 {
 793         struct page *pages[2];
 794         nid_t nid[3];
 795         nid_t child_nid;
 796         int err = 0;
 797         int i;
 798         int idx = depth - 2;
 799
 800         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 801         if (!nid[0])
 802                 return 0;
 803
 804         /* get indirect nodes in the path */
 805         for (i = 0; i < idx + 1; i++) {
 806                 /* reference count'll be increased */
 807                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 808                 if (IS_ERR(pages[i])) {
 809                         err = PTR_ERR(pages[i]);
 810                         idx = i - 1;
 811                         goto fail;
 812                 }
 813                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 814         }
 815
 816         ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 817
 818         /* free direct nodes linked to a partial indirect node */
 819         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 820                 child_nid = get_nid(pages[idx], i, false);
 821                 if (!child_nid)
 822                         continue;
 823                 dn->nid = child_nid;
 824                 err = truncate_dnode(dn);
 825                 if (err < 0)
 826                         goto fail;
 827                 if (set_nid(pages[idx], i, 0, false))
 828                         dn->node_changed = true;
 829         }
 830
 831         if (offset[idx + 1] == 0) {
 832                 dn->node_page = pages[idx];
 833                 dn->nid = nid[idx];
 834                 truncate_node(dn);
 835         } else {
 836                 f2fs_put_page(pages[idx], 1);
 837         }
 838         offset[idx]++;
 839         offset[idx + 1] = 0;
 840         idx--;
 841 fail:
 842         for (i = idx; i >= 0; i--)
 843                 f2fs_put_page(pages[i], 1);
 844
 845         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 846
 847         return err;
 848 }
 849
 850 /*
 851  * All the block addresses of data and nodes should be nullified.
 852  */
 853 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 854 {
 855         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 856         int err = 0, cont = 1;
 857         int level, offset[4], noffset[4];
 858         unsigned int nofs = 0;
 859         struct f2fs_inode *ri;
 860         struct dnode_of_data dn;
 861         struct page *page;
 862
 863         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 864
 865         level = get_node_path(inode, from, offset, noffset);
 866
 867         page = get_node_page(sbi, inode->i_ino);
 868         if (IS_ERR(page)) {
 869                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 870                 return PTR_ERR(page);
 871         }
 872
 873         set_new_dnode(&dn, inode, page, NULL, 0);
 874         unlock_page(page);
 875
 876         ri = F2FS_INODE(page);
 877         switch (level) {
 878         case 0:
 879         case 1:
 880                 nofs = noffset[1];
 881                 break;
 882         case 2:
 883                 nofs = noffset[1];
 884                 if (!offset[level - 1])
 885                         goto skip_partial;
 886                 err = truncate_partial_nodes(&dn, ri, offset, level);
 887                 if (err < 0 && err != -ENOENT)
 888                         goto fail;
 889                 nofs += 1 + NIDS_PER_BLOCK;
 890                 break;
 891         case 3:
 892                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 893                 if (!offset[level - 1])
 894                         goto skip_partial;
 895                 err = truncate_partial_nodes(&dn, ri, offset, level);
 896                 if (err < 0 && err != -ENOENT)
 897                         goto fail;
 898                 break;
 899         default:
 900                 BUG();
 901         }
 902
 903 skip_partial:
 904         while (cont) {
 905                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 906                 switch (offset[0]) {
 907                 case NODE_DIR1_BLOCK:
 908                 case NODE_DIR2_BLOCK:
 909                         err = truncate_dnode(&dn);
 910                         break;
 911
 912                 case NODE_IND1_BLOCK:
 913                 case NODE_IND2_BLOCK:
 914                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 915                         break;
 916
 917                 case NODE_DIND_BLOCK:
 918                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 919                         cont = 0;
 920                         break;
 921
 922                 default:
 923                         BUG();
 924                 }
 925                 if (err < 0 && err != -ENOENT)
 926                         goto fail;
 927                 if (offset[1] == 0 &&
 928                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 929                         lock_page(page);
 930                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
 931                         f2fs_wait_on_page_writeback(page, NODE, true);
 932                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 933                         set_page_dirty(page);
 934                         unlock_page(page);
 935                 }
 936                 offset[1] = 0;
 937                 offset[0]++;
 938                 nofs += err;
 939         }
 940 fail:
 941         f2fs_put_page(page, 0);
 942         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 943         return err > 0 ? 0 : err;
 944 }
 945
 946 int truncate_xattr_node(struct inode *inode, struct page *page)
 947 {
 948         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 949         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 950         struct dnode_of_data dn;
 951         struct page *npage;
 952
 953         if (!nid)
 954                 return 0;
 955
 956         npage = get_node_page(sbi, nid);
 957         if (IS_ERR(npage))
 958                 return PTR_ERR(npage);
 959
 960         f2fs_i_xnid_write(inode, 0);
 961
 962         /* need to do checkpoint during fsync */
 963         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 964
 965         set_new_dnode(&dn, inode, page, npage, nid);
 966
 967         if (page)
 968                 dn.inode_page_locked = true;
 969         truncate_node(&dn);
 970         return 0;
 971 }
 972
 973 /*
 974  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 975  * f2fs_unlock_op().
 976  */
 977 int remove_inode_page(struct inode *inode)
 978 {
 979         struct dnode_of_data dn;
 980         int err;
 981
 982         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 983         err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
 984         if (err)
 985                 return err;
 986
 987         err = truncate_xattr_node(inode, dn.inode_page);
 988         if (err) {
 989                 f2fs_put_dnode(&dn);
 990                 return err;
 991         }
 992
 993         /* remove potential inline_data blocks */
 994         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 995                                 S_ISLNK(inode->i_mode))
 996                 truncate_data_blocks_range(&dn, 1);
 997
 998         /* 0 is possible, after f2fs_new_inode() has failed */
 999         f2fs_bug_on(F2FS_I_SB(inode),
1000                         inode->i_blocks != 0 && inode->i_blocks != 1);
1001
1002         /* will put inode & node pages */
1003         truncate_node(&dn);
1004         return 0;
1005 }
1006
1007 struct page *new_inode_page(struct inode *inode)
1008 {
1009         struct dnode_of_data dn;
1010
1011         /* allocate inode page for new inode */
1012         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1013
1014         /* caller should f2fs_put_page(page, 1); */
1015         return new_node_page(&dn, 0, NULL);
1016 }
1017
1018 struct page *new_node_page(struct dnode_of_data *dn,
1019                                 unsigned int ofs, struct page *ipage)
1020 {
1021         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1022         struct node_info old_ni, new_ni;
1023         struct page *page;
1024         int err;
1025
1026         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1027                 return ERR_PTR(-EPERM);
1028
1029         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1030         if (!page)
1031                 return ERR_PTR(-ENOMEM);
1032
1033         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1034                 err = -ENOSPC;
1035                 goto fail;
1036         }
1037
1038         get_node_info(sbi, dn->nid, &old_ni);
1039
1040         /* Reinitialize old_ni with new node page */
1041         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1042         new_ni = old_ni;
1043         new_ni.ino = dn->inode->i_ino;
1044         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1045
1046         f2fs_wait_on_page_writeback(page, NODE, true);
1047         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1048         set_cold_node(dn->inode, page);
1049         if (!PageUptodate(page))
1050                 SetPageUptodate(page);
1051         if (set_page_dirty(page))
1052                 dn->node_changed = true;
1053
1054         if (f2fs_has_xattr_block(ofs))
1055                 f2fs_i_xnid_write(dn->inode, dn->nid);
1056
1057         if (ofs == 0)
1058                 inc_valid_inode_count(sbi);
1059         return page;
1060
1061 fail:
1062         clear_node_page_dirty(page);
1063         f2fs_put_page(page, 1);
1064         return ERR_PTR(err);
1065 }
1066
1067 /*
1068  * Caller should do after getting the following values.
1069  * 0: f2fs_put_page(page, 0)
1070  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1071  */
1072 static int read_node_page(struct page *page, int op_flags)
1073 {
1074         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1075         struct node_info ni;
1076         struct f2fs_io_info fio = {
1077                 .sbi = sbi,
1078                 .type = NODE,
1079                 .op = REQ_OP_READ,
1080                 .op_flags = op_flags,
1081                 .page = page,
1082                 .encrypted_page = NULL,
1083         };
1084
1085         if (PageUptodate(page))
1086                 return LOCKED_PAGE;
1087
1088         get_node_info(sbi, page->index, &ni);
1089
1090         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1091                 ClearPageUptodate(page);
1092                 return -ENOENT;
1093         }
1094
1095         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1096         return f2fs_submit_page_bio(&fio);
1097 }
1098
1099 /*
1100  * Readahead a node page
1101  */
1102 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1103 {
1104         struct page *apage;
1105         int err;
1106
1107         if (!nid)
1108                 return;
1109         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1110
1111         rcu_read_lock();
1112         apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1113         rcu_read_unlock();
1114         if (apage)
1115                 return;
1116
1117         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1118         if (!apage)
1119                 return;
1120
1121         err = read_node_page(apage, REQ_RAHEAD);
1122         f2fs_put_page(apage, err ? 1 : 0);
1123 }
1124
1125 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1126                                         struct page *parent, int start)
1127 {
1128         struct page *page;
1129         int err;
1130
1131         if (!nid)
1132                 return ERR_PTR(-ENOENT);
1133         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1134 repeat:
1135         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1136         if (!page)
1137                 return ERR_PTR(-ENOMEM);
1138
1139         err = read_node_page(page, READ_SYNC);
1140         if (err < 0) {
1141                 f2fs_put_page(page, 1);
1142                 return ERR_PTR(err);
1143         } else if (err == LOCKED_PAGE) {
1144                 goto page_hit;
1145         }
1146
1147         if (parent)
1148                 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1149
1150         lock_page(page);
1151
1152         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1153                 f2fs_put_page(page, 1);
1154                 goto repeat;
1155         }
1156
1157         if (unlikely(!PageUptodate(page)))
1158                 goto out_err;
1159 page_hit:
1160         if(unlikely(nid != nid_of_node(page))) {
1161                 f2fs_bug_on(sbi, 1);
1162                 ClearPageUptodate(page);
1163 out_err:
1164                 f2fs_put_page(page, 1);
1165                 return ERR_PTR(-EIO);
1166         }
1167         return page;
1168 }
1169
1170 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1171 {
1172         return __get_node_page(sbi, nid, NULL, 0);
1173 }
1174
1175 struct page *get_node_page_ra(struct page *parent, int start)
1176 {
1177         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1178         nid_t nid = get_nid(parent, start, false);
1179
1180         return __get_node_page(sbi, nid, parent, start);
1181 }
1182
1183 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1184 {
1185         struct inode *inode;
1186         struct page *page;
1187         int ret;
1188
1189         /* should flush inline_data before evict_inode */
1190         inode = ilookup(sbi->sb, ino);
1191         if (!inode)
1192                 return;
1193
1194         page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1195         if (!page)
1196                 goto iput_out;
1197
1198         if (!PageUptodate(page))
1199                 goto page_out;
1200
1201         if (!PageDirty(page))
1202                 goto page_out;
1203
1204         if (!clear_page_dirty_for_io(page))
1205                 goto page_out;
1206
1207         ret = f2fs_write_inline_data(inode, page);
1208         inode_dec_dirty_pages(inode);
1209         if (ret)
1210                 set_page_dirty(page);
1211 page_out:
1212         f2fs_put_page(page, 1);
1213 iput_out:
1214         iput(inode);
1215 }
1216
1217 void move_node_page(struct page *node_page, int gc_type)
1218 {
1219         if (gc_type == FG_GC) {
1220                 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1221                 struct writeback_control wbc = {
1222                         .sync_mode = WB_SYNC_ALL,
1223                         .nr_to_write = 1,
1224                         .for_reclaim = 0,
1225                 };
1226
1227                 set_page_dirty(node_page);
1228                 f2fs_wait_on_page_writeback(node_page, NODE, true);
1229
1230                 f2fs_bug_on(sbi, PageWriteback(node_page));
1231                 if (!clear_page_dirty_for_io(node_page))
1232                         goto out_page;
1233
1234                 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1235                         unlock_page(node_page);
1236                 goto release_page;
1237         } else {
1238                 /* set page dirty and write it */
1239                 if (!PageWriteback(node_page))
1240                         set_page_dirty(node_page);
1241         }
1242 out_page:
1243         unlock_page(node_page);
1244 release_page:
1245         f2fs_put_page(node_page, 0);
1246 }
1247
1248 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1249 {
1250         pgoff_t index, end;
1251         struct pagevec pvec;
1252         struct page *last_page = NULL;
1253
1254         pagevec_init(&pvec, 0);
1255         index = 0;
1256         end = ULONG_MAX;
1257
1258         while (index <= end) {
1259                 int i, nr_pages;
1260                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1261                                 PAGECACHE_TAG_DIRTY,
1262                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1263                 if (nr_pages == 0)
1264                         break;
1265
1266                 for (i = 0; i < nr_pages; i++) {
1267                         struct page *page = pvec.pages[i];
1268
1269                         if (unlikely(f2fs_cp_error(sbi))) {
1270                                 f2fs_put_page(last_page, 0);
1271                                 pagevec_release(&pvec);
1272                                 return ERR_PTR(-EIO);
1273                         }
1274
1275                         if (!IS_DNODE(page) || !is_cold_node(page))
1276                                 continue;
1277                         if (ino_of_node(page) != ino)
1278                                 continue;
1279
1280                         lock_page(page);
1281
1282                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1283 continue_unlock:
1284                                 unlock_page(page);
1285                                 continue;
1286                         }
1287                         if (ino_of_node(page) != ino)
1288                                 goto continue_unlock;
1289
1290                         if (!PageDirty(page)) {
1291                                 /* someone wrote it for us */
1292                                 goto continue_unlock;
1293                         }
1294
1295                         if (last_page)
1296                                 f2fs_put_page(last_page, 0);
1297
1298                         get_page(page);
1299                         last_page = page;
1300                         unlock_page(page);
1301                 }
1302                 pagevec_release(&pvec);
1303                 cond_resched();
1304         }
1305         return last_page;
1306 }
1307
1308 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1309                         struct writeback_control *wbc, bool atomic)
1310 {
1311         pgoff_t index, end;
1312         struct pagevec pvec;
1313         int ret = 0;
1314         struct page *last_page = NULL;
1315         bool marked = false;
1316         nid_t ino = inode->i_ino;
1317
1318         if (atomic) {
1319                 last_page = last_fsync_dnode(sbi, ino);
1320                 if (IS_ERR_OR_NULL(last_page))
1321                         return PTR_ERR_OR_ZERO(last_page);
1322         }
1323 retry:
1324         pagevec_init(&pvec, 0);
1325         index = 0;
1326         end = ULONG_MAX;
1327
1328         while (index <= end) {
1329                 int i, nr_pages;
1330                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1331                                 PAGECACHE_TAG_DIRTY,
1332                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1333                 if (nr_pages == 0)
1334                         break;
1335
1336                 for (i = 0; i < nr_pages; i++) {
1337                         struct page *page = pvec.pages[i];
1338
1339                         if (unlikely(f2fs_cp_error(sbi))) {
1340                                 f2fs_put_page(last_page, 0);
1341                                 pagevec_release(&pvec);
1342                                 return -EIO;
1343                         }
1344
1345                         if (!IS_DNODE(page) || !is_cold_node(page))
1346                                 continue;
1347                         if (ino_of_node(page) != ino)
1348                                 continue;
1349
1350                         lock_page(page);
1351
1352                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1353 continue_unlock:
1354                                 unlock_page(page);
1355                                 continue;
1356                         }
1357                         if (ino_of_node(page) != ino)
1358                                 goto continue_unlock;
1359
1360                         if (!PageDirty(page) && page != last_page) {
1361                                 /* someone wrote it for us */
1362                                 goto continue_unlock;
1363                         }
1364
1365                         f2fs_wait_on_page_writeback(page, NODE, true);
1366                         BUG_ON(PageWriteback(page));
1367
1368                         if (!atomic || page == last_page) {
1369                                 set_fsync_mark(page, 1);
1370                                 if (IS_INODE(page)) {
1371                                         if (is_inode_flag_set(inode,
1372                                                                 FI_DIRTY_INODE))
1373                                                 update_inode(inode, page);
1374                                         set_dentry_mark(page,
1375                                                 need_dentry_mark(sbi, ino));
1376                                 }
1377                                 /*  may be written by other thread */
1378                                 if (!PageDirty(page))
1379                                         set_page_dirty(page);
1380                         }
1381
1382                         if (!clear_page_dirty_for_io(page))
1383                                 goto continue_unlock;
1384
1385                         ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1386                         if (ret) {
1387                                 unlock_page(page);
1388                                 f2fs_put_page(last_page, 0);
1389                                 break;
1390                         }
1391                         if (page == last_page) {
1392                                 f2fs_put_page(page, 0);
1393                                 marked = true;
1394                                 break;
1395                         }
1396                 }
1397                 pagevec_release(&pvec);
1398                 cond_resched();
1399
1400                 if (ret || marked)
1401                         break;
1402         }
1403         if (!ret && atomic && !marked) {
1404                 f2fs_msg(sbi->sb, KERN_DEBUG,
1405                         "Retry to write fsync mark: ino=%u, idx=%lx",
1406                                         ino, last_page->index);
1407                 lock_page(last_page);
1408                 set_page_dirty(last_page);
1409                 unlock_page(last_page);
1410                 goto retry;
1411         }
1412         return ret ? -EIO: 0;
1413 }
1414
1415 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1416 {
1417         pgoff_t index, end;
1418         struct pagevec pvec;
1419         int step = 0;
1420         int nwritten = 0;
1421
1422         pagevec_init(&pvec, 0);
1423
1424 next_step:
1425         index = 0;
1426         end = ULONG_MAX;
1427
1428         while (index <= end) {
1429                 int i, nr_pages;
1430                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1431                                 PAGECACHE_TAG_DIRTY,
1432                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1433                 if (nr_pages == 0)
1434                         break;
1435
1436                 for (i = 0; i < nr_pages; i++) {
1437                         struct page *page = pvec.pages[i];
1438
1439                         if (unlikely(f2fs_cp_error(sbi))) {
1440                                 pagevec_release(&pvec);
1441                                 return -EIO;
1442                         }
1443
1444                         /*
1445                          * flushing sequence with step:
1446                          * 0. indirect nodes
1447                          * 1. dentry dnodes
1448                          * 2. file dnodes
1449                          */
1450                         if (step == 0 && IS_DNODE(page))
1451                                 continue;
1452                         if (step == 1 && (!IS_DNODE(page) ||
1453                                                 is_cold_node(page)))
1454                                 continue;
1455                         if (step == 2 && (!IS_DNODE(page) ||
1456                                                 !is_cold_node(page)))
1457                                 continue;
1458 lock_node:
1459                         if (!trylock_page(page))
1460                                 continue;
1461
1462                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1463 continue_unlock:
1464                                 unlock_page(page);
1465                                 continue;
1466                         }
1467
1468                         if (!PageDirty(page)) {
1469                                 /* someone wrote it for us */
1470                                 goto continue_unlock;
1471                         }
1472
1473                         /* flush inline_data */
1474                         if (is_inline_node(page)) {
1475                                 clear_inline_node(page);
1476                                 unlock_page(page);
1477                                 flush_inline_data(sbi, ino_of_node(page));
1478                                 goto lock_node;
1479                         }
1480
1481                         f2fs_wait_on_page_writeback(page, NODE, true);
1482
1483                         BUG_ON(PageWriteback(page));
1484                         if (!clear_page_dirty_for_io(page))
1485                                 goto continue_unlock;
1486
1487                         set_fsync_mark(page, 0);
1488                         set_dentry_mark(page, 0);
1489
1490                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1491                                 unlock_page(page);
1492
1493                         if (--wbc->nr_to_write == 0)
1494                                 break;
1495                 }
1496                 pagevec_release(&pvec);
1497                 cond_resched();
1498
1499                 if (wbc->nr_to_write == 0) {
1500                         step = 2;
1501                         break;
1502                 }
1503         }
1504
1505         if (step < 2) {
1506                 step++;
1507                 goto next_step;
1508         }
1509         return nwritten;
1510 }
1511
1512 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1513 {
1514         pgoff_t index = 0, end = ULONG_MAX;
1515         struct pagevec pvec;
1516         int ret2 = 0, ret = 0;
1517
1518         pagevec_init(&pvec, 0);
1519
1520         while (index <= end) {
1521                 int i, nr_pages;
1522                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1523                                 PAGECACHE_TAG_WRITEBACK,
1524                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1525                 if (nr_pages == 0)
1526                         break;
1527
1528                 for (i = 0; i < nr_pages; i++) {
1529                         struct page *page = pvec.pages[i];
1530
1531                         /* until radix tree lookup accepts end_index */
1532                         if (unlikely(page->index > end))
1533                                 continue;
1534
1535                         if (ino && ino_of_node(page) == ino) {
1536                                 f2fs_wait_on_page_writeback(page, NODE, true);
1537                                 if (TestClearPageError(page))
1538                                         ret = -EIO;
1539                         }
1540                 }
1541                 pagevec_release(&pvec);
1542                 cond_resched();
1543         }
1544
1545         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1546                 ret2 = -ENOSPC;
1547         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1548                 ret2 = -EIO;
1549         if (!ret)
1550                 ret = ret2;
1551         return ret;
1552 }
1553
1554 static int f2fs_write_node_page(struct page *page,
1555                                 struct writeback_control *wbc)
1556 {
1557         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1558         nid_t nid;
1559         struct node_info ni;
1560         struct f2fs_io_info fio = {
1561                 .sbi = sbi,
1562                 .type = NODE,
1563                 .op = REQ_OP_WRITE,
1564                 .op_flags = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0,
1565                 .page = page,
1566                 .encrypted_page = NULL,
1567         };
1568
1569         trace_f2fs_writepage(page, NODE);
1570
1571         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1572                 goto redirty_out;
1573         if (unlikely(f2fs_cp_error(sbi)))
1574                 goto redirty_out;
1575
1576         /* get old block addr of this node page */
1577         nid = nid_of_node(page);
1578         f2fs_bug_on(sbi, page->index != nid);
1579
1580         if (wbc->for_reclaim) {
1581                 if (!down_read_trylock(&sbi->node_write))
1582                         goto redirty_out;
1583         } else {
1584                 down_read(&sbi->node_write);
1585         }
1586
1587         get_node_info(sbi, nid, &ni);
1588
1589         /* This page is already truncated */
1590         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1591                 ClearPageUptodate(page);
1592                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1593                 up_read(&sbi->node_write);
1594                 unlock_page(page);
1595                 return 0;
1596         }
1597
1598         set_page_writeback(page);
1599         fio.old_blkaddr = ni.blk_addr;
1600         write_node_page(nid, &fio);
1601         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1602         dec_page_count(sbi, F2FS_DIRTY_NODES);
1603         up_read(&sbi->node_write);
1604
1605         if (wbc->for_reclaim)
1606                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1607
1608         unlock_page(page);
1609
1610         if (unlikely(f2fs_cp_error(sbi)))
1611                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1612
1613         return 0;
1614
1615 redirty_out:
1616         redirty_page_for_writepage(wbc, page);
1617         return AOP_WRITEPAGE_ACTIVATE;
1618 }
1619
1620 static int f2fs_write_node_pages(struct address_space *mapping,
1621                             struct writeback_control *wbc)
1622 {
1623         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1624         struct blk_plug plug;
1625         long diff;
1626
1627         /* balancing f2fs's metadata in background */
1628         f2fs_balance_fs_bg(sbi);
1629
1630         /* collect a number of dirty node pages and write together */
1631         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1632                 goto skip_write;
1633
1634         trace_f2fs_writepages(mapping->host, wbc, NODE);
1635
1636         diff = nr_pages_to_write(sbi, NODE, wbc);
1637         wbc->sync_mode = WB_SYNC_NONE;
1638         blk_start_plug(&plug);
1639         sync_node_pages(sbi, wbc);
1640         blk_finish_plug(&plug);
1641         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1642         return 0;
1643
1644 skip_write:
1645         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1646         trace_f2fs_writepages(mapping->host, wbc, NODE);
1647         return 0;
1648 }
1649
1650 static int f2fs_set_node_page_dirty(struct page *page)
1651 {
1652         trace_f2fs_set_page_dirty(page, NODE);
1653
1654         if (!PageUptodate(page))
1655                 SetPageUptodate(page);
1656         if (!PageDirty(page)) {
1657                 f2fs_set_page_dirty_nobuffers(page);
1658                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1659                 SetPagePrivate(page);
1660                 f2fs_trace_pid(page);
1661                 return 1;
1662         }
1663         return 0;
1664 }
1665
1666 /*
1667  * Structure of the f2fs node operations
1668  */
1669 const struct address_space_operations f2fs_node_aops = {
1670         .writepage      = f2fs_write_node_page,
1671         .writepages     = f2fs_write_node_pages,
1672         .set_page_dirty = f2fs_set_node_page_dirty,
1673         .invalidatepage = f2fs_invalidate_page,
1674         .releasepage    = f2fs_release_page,
1675 };
1676
1677 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1678                                                 nid_t n)
1679 {
1680         return radix_tree_lookup(&nm_i->free_nid_root, n);
1681 }
1682
1683 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1684                                                 struct free_nid *i)
1685 {
1686         list_del(&i->list);
1687         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1688 }
1689
1690 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1691 {
1692         struct f2fs_nm_info *nm_i = NM_I(sbi);
1693         struct free_nid *i;
1694         struct nat_entry *ne;
1695
1696         if (!available_free_memory(sbi, FREE_NIDS))
1697                 return -1;
1698
1699         /* 0 nid should not be used */
1700         if (unlikely(nid == 0))
1701                 return 0;
1702
1703         if (build) {
1704                 /* do not add allocated nids */
1705                 ne = __lookup_nat_cache(nm_i, nid);
1706                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1707                                 nat_get_blkaddr(ne) != NULL_ADDR))
1708                         return 0;
1709         }
1710
1711         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1712         i->nid = nid;
1713         i->state = NID_NEW;
1714
1715         if (radix_tree_preload(GFP_NOFS)) {
1716                 kmem_cache_free(free_nid_slab, i);
1717                 return 0;
1718         }
1719
1720         spin_lock(&nm_i->free_nid_list_lock);
1721         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1722                 spin_unlock(&nm_i->free_nid_list_lock);
1723                 radix_tree_preload_end();
1724                 kmem_cache_free(free_nid_slab, i);
1725                 return 0;
1726         }
1727         list_add_tail(&i->list, &nm_i->free_nid_list);
1728         nm_i->fcnt++;
1729         spin_unlock(&nm_i->free_nid_list_lock);
1730         radix_tree_preload_end();
1731         return 1;
1732 }
1733
1734 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1735 {
1736         struct free_nid *i;
1737         bool need_free = false;
1738
1739         spin_lock(&nm_i->free_nid_list_lock);
1740         i = __lookup_free_nid_list(nm_i, nid);
1741         if (i && i->state == NID_NEW) {
1742                 __del_from_free_nid_list(nm_i, i);
1743                 nm_i->fcnt--;
1744                 need_free = true;
1745         }
1746         spin_unlock(&nm_i->free_nid_list_lock);
1747
1748         if (need_free)
1749                 kmem_cache_free(free_nid_slab, i);
1750 }
1751
1752 static void scan_nat_page(struct f2fs_sb_info *sbi,
1753                         struct page *nat_page, nid_t start_nid)
1754 {
1755         struct f2fs_nm_info *nm_i = NM_I(sbi);
1756         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1757         block_t blk_addr;
1758         int i;
1759
1760         i = start_nid % NAT_ENTRY_PER_BLOCK;
1761
1762         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1763
1764                 if (unlikely(start_nid >= nm_i->max_nid))
1765                         break;
1766
1767                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1768                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1769                 if (blk_addr == NULL_ADDR) {
1770                         if (add_free_nid(sbi, start_nid, true) < 0)
1771                                 break;
1772                 }
1773         }
1774 }
1775
1776 void build_free_nids(struct f2fs_sb_info *sbi)
1777 {
1778         struct f2fs_nm_info *nm_i = NM_I(sbi);
1779         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1780         struct f2fs_journal *journal = curseg->journal;
1781         int i = 0;
1782         nid_t nid = nm_i->next_scan_nid;
1783
1784         /* Enough entries */
1785         if (nm_i->fcnt >= NAT_ENTRY_PER_BLOCK)
1786                 return;
1787
1788         /* readahead nat pages to be scanned */
1789         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1790                                                         META_NAT, true);
1791
1792         down_read(&nm_i->nat_tree_lock);
1793
1794         while (1) {
1795                 struct page *page = get_current_nat_page(sbi, nid);
1796
1797                 scan_nat_page(sbi, page, nid);
1798                 f2fs_put_page(page, 1);
1799
1800                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1801                 if (unlikely(nid >= nm_i->max_nid))
1802                         nid = 0;
1803
1804                 if (++i >= FREE_NID_PAGES)
1805                         break;
1806         }
1807
1808         /* go to the next free nat pages to find free nids abundantly */
1809         nm_i->next_scan_nid = nid;
1810
1811         /* find free nids from current sum_pages */
1812         down_read(&curseg->journal_rwsem);
1813         for (i = 0; i < nats_in_cursum(journal); i++) {
1814                 block_t addr;
1815
1816                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1817                 nid = le32_to_cpu(nid_in_journal(journal, i));
1818                 if (addr == NULL_ADDR)
1819                         add_free_nid(sbi, nid, true);
1820                 else
1821                         remove_free_nid(nm_i, nid);
1822         }
1823         up_read(&curseg->journal_rwsem);
1824         up_read(&nm_i->nat_tree_lock);
1825
1826         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1827                                         nm_i->ra_nid_pages, META_NAT, false);
1828 }
1829
1830 /*
1831  * If this function returns success, caller can obtain a new nid
1832  * from second parameter of this function.
1833  * The returned nid could be used ino as well as nid when inode is created.
1834  */
1835 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1836 {
1837         struct f2fs_nm_info *nm_i = NM_I(sbi);
1838         struct free_nid *i = NULL;
1839 retry:
1840 #ifdef CONFIG_F2FS_FAULT_INJECTION
1841         if (time_to_inject(FAULT_ALLOC_NID))
1842                 return false;
1843 #endif
1844         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1845                 return false;
1846
1847         spin_lock(&nm_i->free_nid_list_lock);
1848
1849         /* We should not use stale free nids created by build_free_nids */
1850         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1851                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1852                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1853                         if (i->state == NID_NEW)
1854                                 break;
1855
1856                 f2fs_bug_on(sbi, i->state != NID_NEW);
1857                 *nid = i->nid;
1858                 i->state = NID_ALLOC;
1859                 nm_i->fcnt--;
1860                 spin_unlock(&nm_i->free_nid_list_lock);
1861                 return true;
1862         }
1863         spin_unlock(&nm_i->free_nid_list_lock);
1864
1865         /* Let's scan nat pages and its caches to get free nids */
1866         mutex_lock(&nm_i->build_lock);
1867         build_free_nids(sbi);
1868         mutex_unlock(&nm_i->build_lock);
1869         goto retry;
1870 }
1871
1872 /*
1873  * alloc_nid() should be called prior to this function.
1874  */
1875 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1876 {
1877         struct f2fs_nm_info *nm_i = NM_I(sbi);
1878         struct free_nid *i;
1879
1880         spin_lock(&nm_i->free_nid_list_lock);
1881         i = __lookup_free_nid_list(nm_i, nid);
1882         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1883         __del_from_free_nid_list(nm_i, i);
1884         spin_unlock(&nm_i->free_nid_list_lock);
1885
1886         kmem_cache_free(free_nid_slab, i);
1887 }
1888
1889 /*
1890  * alloc_nid() should be called prior to this function.
1891  */
1892 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1893 {
1894         struct f2fs_nm_info *nm_i = NM_I(sbi);
1895         struct free_nid *i;
1896         bool need_free = false;
1897
1898         if (!nid)
1899                 return;
1900
1901         spin_lock(&nm_i->free_nid_list_lock);
1902         i = __lookup_free_nid_list(nm_i, nid);
1903         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1904         if (!available_free_memory(sbi, FREE_NIDS)) {
1905                 __del_from_free_nid_list(nm_i, i);
1906                 need_free = true;
1907         } else {
1908                 i->state = NID_NEW;
1909                 nm_i->fcnt++;
1910         }
1911         spin_unlock(&nm_i->free_nid_list_lock);
1912
1913         if (need_free)
1914                 kmem_cache_free(free_nid_slab, i);
1915 }
1916
1917 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1918 {
1919         struct f2fs_nm_info *nm_i = NM_I(sbi);
1920         struct free_nid *i, *next;
1921         int nr = nr_shrink;
1922
1923         if (nm_i->fcnt <= MAX_FREE_NIDS)
1924                 return 0;
1925
1926         if (!mutex_trylock(&nm_i->build_lock))
1927                 return 0;
1928
1929         spin_lock(&nm_i->free_nid_list_lock);
1930         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1931                 if (nr_shrink <= 0 || nm_i->fcnt <= MAX_FREE_NIDS)
1932                         break;
1933                 if (i->state == NID_ALLOC)
1934                         continue;
1935                 __del_from_free_nid_list(nm_i, i);
1936                 kmem_cache_free(free_nid_slab, i);
1937                 nm_i->fcnt--;
1938                 nr_shrink--;
1939         }
1940         spin_unlock(&nm_i->free_nid_list_lock);
1941         mutex_unlock(&nm_i->build_lock);
1942
1943         return nr - nr_shrink;
1944 }
1945
1946 void recover_inline_xattr(struct inode *inode, struct page *page)
1947 {
1948         void *src_addr, *dst_addr;
1949         size_t inline_size;
1950         struct page *ipage;
1951         struct f2fs_inode *ri;
1952
1953         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1954         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1955
1956         ri = F2FS_INODE(page);
1957         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1958                 clear_inode_flag(inode, FI_INLINE_XATTR);
1959                 goto update_inode;
1960         }
1961
1962         dst_addr = inline_xattr_addr(ipage);
1963         src_addr = inline_xattr_addr(page);
1964         inline_size = inline_xattr_size(inode);
1965
1966         f2fs_wait_on_page_writeback(ipage, NODE, true);
1967         memcpy(dst_addr, src_addr, inline_size);
1968 update_inode:
1969         update_inode(inode, ipage);
1970         f2fs_put_page(ipage, 1);
1971 }
1972
1973 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1974 {
1975         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1976         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1977         nid_t new_xnid = nid_of_node(page);
1978         struct node_info ni;
1979
1980         /* 1: invalidate the previous xattr nid */
1981         if (!prev_xnid)
1982                 goto recover_xnid;
1983
1984         /* Deallocate node address */
1985         get_node_info(sbi, prev_xnid, &ni);
1986         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1987         invalidate_blocks(sbi, ni.blk_addr);
1988         dec_valid_node_count(sbi, inode);
1989         set_node_addr(sbi, &ni, NULL_ADDR, false);
1990
1991 recover_xnid:
1992         /* 2: allocate new xattr nid */
1993         if (unlikely(!inc_valid_node_count(sbi, inode)))
1994                 f2fs_bug_on(sbi, 1);
1995
1996         remove_free_nid(NM_I(sbi), new_xnid);
1997         get_node_info(sbi, new_xnid, &ni);
1998         ni.ino = inode->i_ino;
1999         set_node_addr(sbi, &ni, NEW_ADDR, false);
2000         f2fs_i_xnid_write(inode, new_xnid);
2001
2002         /* 3: update xattr blkaddr */
2003         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
2004         set_node_addr(sbi, &ni, blkaddr, false);
2005 }
2006
2007 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2008 {
2009         struct f2fs_inode *src, *dst;
2010         nid_t ino = ino_of_node(page);
2011         struct node_info old_ni, new_ni;
2012         struct page *ipage;
2013
2014         get_node_info(sbi, ino, &old_ni);
2015
2016         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2017                 return -EINVAL;
2018
2019         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2020         if (!ipage)
2021                 return -ENOMEM;
2022
2023         /* Should not use this inode from free nid list */
2024         remove_free_nid(NM_I(sbi), ino);
2025
2026         if (!PageUptodate(ipage))
2027                 SetPageUptodate(ipage);
2028         fill_node_footer(ipage, ino, ino, 0, true);
2029
2030         src = F2FS_INODE(page);
2031         dst = F2FS_INODE(ipage);
2032
2033         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2034         dst->i_size = 0;
2035         dst->i_blocks = cpu_to_le64(1);
2036         dst->i_links = cpu_to_le32(1);
2037         dst->i_xattr_nid = 0;
2038         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2039
2040         new_ni = old_ni;
2041         new_ni.ino = ino;
2042
2043         if (unlikely(!inc_valid_node_count(sbi, NULL)))
2044                 WARN_ON(1);
2045         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2046         inc_valid_inode_count(sbi);
2047         set_page_dirty(ipage);
2048         f2fs_put_page(ipage, 1);
2049         return 0;
2050 }
2051
2052 int restore_node_summary(struct f2fs_sb_info *sbi,
2053                         unsigned int segno, struct f2fs_summary_block *sum)
2054 {
2055         struct f2fs_node *rn;
2056         struct f2fs_summary *sum_entry;
2057         block_t addr;
2058         int bio_blocks = MAX_BIO_BLOCKS(sbi);
2059         int i, idx, last_offset, nrpages;
2060
2061         /* scan the node segment */
2062         last_offset = sbi->blocks_per_seg;
2063         addr = START_BLOCK(sbi, segno);
2064         sum_entry = &sum->entries[0];
2065
2066         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2067                 nrpages = min(last_offset - i, bio_blocks);
2068
2069                 /* readahead node pages */
2070                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2071
2072                 for (idx = addr; idx < addr + nrpages; idx++) {
2073                         struct page *page = get_tmp_page(sbi, idx);
2074
2075                         rn = F2FS_NODE(page);
2076                         sum_entry->nid = rn->footer.nid;
2077                         sum_entry->version = 0;
2078                         sum_entry->ofs_in_node = 0;
2079                         sum_entry++;
2080                         f2fs_put_page(page, 1);
2081                 }
2082
2083                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2084                                                         addr + nrpages);
2085         }
2086         return 0;
2087 }
2088
2089 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2090 {
2091         struct f2fs_nm_info *nm_i = NM_I(sbi);
2092         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2093         struct f2fs_journal *journal = curseg->journal;
2094         int i;
2095
2096         down_write(&curseg->journal_rwsem);
2097         for (i = 0; i < nats_in_cursum(journal); i++) {
2098                 struct nat_entry *ne;
2099                 struct f2fs_nat_entry raw_ne;
2100                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2101
2102                 raw_ne = nat_in_journal(journal, i);
2103
2104                 ne = __lookup_nat_cache(nm_i, nid);
2105                 if (!ne) {
2106                         ne = grab_nat_entry(nm_i, nid);
2107                         node_info_from_raw_nat(&ne->ni, &raw_ne);
2108                 }
2109                 __set_nat_cache_dirty(nm_i, ne);
2110         }
2111         update_nats_in_cursum(journal, -i);
2112         up_write(&curseg->journal_rwsem);
2113 }
2114
2115 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2116                                                 struct list_head *head, int max)
2117 {
2118         struct nat_entry_set *cur;
2119
2120         if (nes->entry_cnt >= max)
2121                 goto add_out;
2122
2123         list_for_each_entry(cur, head, set_list) {
2124                 if (cur->entry_cnt >= nes->entry_cnt) {
2125                         list_add(&nes->set_list, cur->set_list.prev);
2126                         return;
2127                 }
2128         }
2129 add_out:
2130         list_add_tail(&nes->set_list, head);
2131 }
2132
2133 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2134                                         struct nat_entry_set *set)
2135 {
2136         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2137         struct f2fs_journal *journal = curseg->journal;
2138         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2139         bool to_journal = true;
2140         struct f2fs_nat_block *nat_blk;
2141         struct nat_entry *ne, *cur;
2142         struct page *page = NULL;
2143
2144         /*
2145          * there are two steps to flush nat entries:
2146          * #1, flush nat entries to journal in current hot data summary block.
2147          * #2, flush nat entries to nat page.
2148          */
2149         if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2150                 to_journal = false;
2151
2152         if (to_journal) {
2153                 down_write(&curseg->journal_rwsem);
2154         } else {
2155                 page = get_next_nat_page(sbi, start_nid);
2156                 nat_blk = page_address(page);
2157                 f2fs_bug_on(sbi, !nat_blk);
2158         }
2159
2160         /* flush dirty nats in nat entry set */
2161         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2162                 struct f2fs_nat_entry *raw_ne;
2163                 nid_t nid = nat_get_nid(ne);
2164                 int offset;
2165
2166                 if (nat_get_blkaddr(ne) == NEW_ADDR)
2167                         continue;
2168
2169                 if (to_journal) {
2170                         offset = lookup_journal_in_cursum(journal,
2171                                                         NAT_JOURNAL, nid, 1);
2172                         f2fs_bug_on(sbi, offset < 0);
2173                         raw_ne = &nat_in_journal(journal, offset);
2174                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2175                 } else {
2176                         raw_ne = &nat_blk->entries[nid - start_nid];
2177                 }
2178                 raw_nat_from_node_info(raw_ne, &ne->ni);
2179                 nat_reset_flag(ne);
2180                 __clear_nat_cache_dirty(NM_I(sbi), ne);
2181                 if (nat_get_blkaddr(ne) == NULL_ADDR)
2182                         add_free_nid(sbi, nid, false);
2183         }
2184
2185         if (to_journal)
2186                 up_write(&curseg->journal_rwsem);
2187         else
2188                 f2fs_put_page(page, 1);
2189
2190         f2fs_bug_on(sbi, set->entry_cnt);
2191
2192         radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2193         kmem_cache_free(nat_entry_set_slab, set);
2194 }
2195
2196 /*
2197  * This function is called during the checkpointing process.
2198  */
2199 void flush_nat_entries(struct f2fs_sb_info *sbi)
2200 {
2201         struct f2fs_nm_info *nm_i = NM_I(sbi);
2202         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2203         struct f2fs_journal *journal = curseg->journal;
2204         struct nat_entry_set *setvec[SETVEC_SIZE];
2205         struct nat_entry_set *set, *tmp;
2206         unsigned int found;
2207         nid_t set_idx = 0;
2208         LIST_HEAD(sets);
2209
2210         if (!nm_i->dirty_nat_cnt)
2211                 return;
2212
2213         down_write(&nm_i->nat_tree_lock);
2214
2215         /*
2216          * if there are no enough space in journal to store dirty nat
2217          * entries, remove all entries from journal and merge them
2218          * into nat entry set.
2219          */
2220         if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2221                 remove_nats_in_journal(sbi);
2222
2223         while ((found = __gang_lookup_nat_set(nm_i,
2224                                         set_idx, SETVEC_SIZE, setvec))) {
2225                 unsigned idx;
2226                 set_idx = setvec[found - 1]->set + 1;
2227                 for (idx = 0; idx < found; idx++)
2228                         __adjust_nat_entry_set(setvec[idx], &sets,
2229                                                 MAX_NAT_JENTRIES(journal));
2230         }
2231
2232         /* flush dirty nats in nat entry set */
2233         list_for_each_entry_safe(set, tmp, &sets, set_list)
2234                 __flush_nat_entry_set(sbi, set);
2235
2236         up_write(&nm_i->nat_tree_lock);
2237
2238         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2239 }
2240
2241 static int init_node_manager(struct f2fs_sb_info *sbi)
2242 {
2243         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2244         struct f2fs_nm_info *nm_i = NM_I(sbi);
2245         unsigned char *version_bitmap;
2246         unsigned int nat_segs, nat_blocks;
2247
2248         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2249
2250         /* segment_count_nat includes pair segment so divide to 2. */
2251         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2252         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2253
2254         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2255
2256         /* not used nids: 0, node, meta, (and root counted as valid node) */
2257         nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2258         nm_i->fcnt = 0;
2259         nm_i->nat_cnt = 0;
2260         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2261         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2262         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2263
2264         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2265         INIT_LIST_HEAD(&nm_i->free_nid_list);
2266         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2267         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2268         INIT_LIST_HEAD(&nm_i->nat_entries);
2269
2270         mutex_init(&nm_i->build_lock);
2271         spin_lock_init(&nm_i->free_nid_list_lock);
2272         init_rwsem(&nm_i->nat_tree_lock);
2273
2274         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2275         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2276         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2277         if (!version_bitmap)
2278                 return -EFAULT;
2279
2280         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2281                                         GFP_KERNEL);
2282         if (!nm_i->nat_bitmap)
2283                 return -ENOMEM;
2284         return 0;
2285 }
2286
2287 int build_node_manager(struct f2fs_sb_info *sbi)
2288 {
2289         int err;
2290
2291         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2292         if (!sbi->nm_info)
2293                 return -ENOMEM;
2294
2295         err = init_node_manager(sbi);
2296         if (err)
2297                 return err;
2298
2299         build_free_nids(sbi);
2300         return 0;
2301 }
2302
2303 void destroy_node_manager(struct f2fs_sb_info *sbi)
2304 {
2305         struct f2fs_nm_info *nm_i = NM_I(sbi);
2306         struct free_nid *i, *next_i;
2307         struct nat_entry *natvec[NATVEC_SIZE];
2308         struct nat_entry_set *setvec[SETVEC_SIZE];
2309         nid_t nid = 0;
2310         unsigned int found;
2311
2312         if (!nm_i)
2313                 return;
2314
2315         /* destroy free nid list */
2316         spin_lock(&nm_i->free_nid_list_lock);
2317         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2318                 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2319                 __del_from_free_nid_list(nm_i, i);
2320                 nm_i->fcnt--;
2321                 spin_unlock(&nm_i->free_nid_list_lock);
2322                 kmem_cache_free(free_nid_slab, i);
2323                 spin_lock(&nm_i->free_nid_list_lock);
2324         }
2325         f2fs_bug_on(sbi, nm_i->fcnt);
2326         spin_unlock(&nm_i->free_nid_list_lock);
2327
2328         /* destroy nat cache */
2329         down_write(&nm_i->nat_tree_lock);
2330         while ((found = __gang_lookup_nat_cache(nm_i,
2331                                         nid, NATVEC_SIZE, natvec))) {
2332                 unsigned idx;
2333
2334                 nid = nat_get_nid(natvec[found - 1]) + 1;
2335                 for (idx = 0; idx < found; idx++)
2336                         __del_from_nat_cache(nm_i, natvec[idx]);
2337         }
2338         f2fs_bug_on(sbi, nm_i->nat_cnt);
2339
2340         /* destroy nat set cache */
2341         nid = 0;
2342         while ((found = __gang_lookup_nat_set(nm_i,
2343                                         nid, SETVEC_SIZE, setvec))) {
2344                 unsigned idx;
2345
2346                 nid = setvec[found - 1]->set + 1;
2347                 for (idx = 0; idx < found; idx++) {
2348                         /* entry_cnt is not zero, when cp_error was occurred */
2349                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2350                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2351                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2352                 }
2353         }
2354         up_write(&nm_i->nat_tree_lock);
2355
2356         kfree(nm_i->nat_bitmap);
2357         sbi->nm_info = NULL;
2358         kfree(nm_i);
2359 }
2360
2361 int __init create_node_manager_caches(void)
2362 {
2363         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2364                         sizeof(struct nat_entry));
2365         if (!nat_entry_slab)
2366                 goto fail;
2367
2368         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2369                         sizeof(struct free_nid));
2370         if (!free_nid_slab)
2371                 goto destroy_nat_entry;
2372
2373         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2374                         sizeof(struct nat_entry_set));
2375         if (!nat_entry_set_slab)
2376                 goto destroy_free_nid;
2377         return 0;
2378
2379 destroy_free_nid:
2380         kmem_cache_destroy(free_nid_slab);
2381 destroy_nat_entry:
2382         kmem_cache_destroy(nat_entry_slab);
2383 fail:
2384         return -ENOMEM;
2385 }
2386
2387 void destroy_node_manager_caches(void)
2388 {
2389         kmem_cache_destroy(nat_entry_set_slab);
2390         kmem_cache_destroy(free_nid_slab);
2391         kmem_cache_destroy(nat_entry_slab);
2392 }