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->nid_cnt[FREE_NID_LIST] *
  49                                 sizeof(struct free_nid)) >> 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         } else if (type == DIRTY_DENTS) {
  58                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  59                         return false;
  60                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  61                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  62         } else if (type == INO_ENTRIES) {
  63                 int i;
  64
  65                 for (i = 0; i <= UPDATE_INO; i++)
  66                         mem_size += (sbi->im[i].ino_num *
  67                                 sizeof(struct ino_entry)) >> PAGE_SHIFT;
  68                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  69         } else if (type == EXTENT_CACHE) {
  70                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  71                                 sizeof(struct extent_tree) +
  72                                 atomic_read(&sbi->total_ext_node) *
  73                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  74                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  75         } else {
  76                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
  77                         return true;
  78         }
  79         return res;
  80 }
  81
  82 static void clear_node_page_dirty(struct page *page)
  83 {
  84         struct address_space *mapping = page->mapping;
  85         unsigned int long flags;
  86
  87         if (PageDirty(page)) {
  88                 spin_lock_irqsave(&mapping->tree_lock, flags);
  89                 radix_tree_tag_clear(&mapping->page_tree,
  90                                 page_index(page),
  91                                 PAGECACHE_TAG_DIRTY);
  92                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  93
  94                 clear_page_dirty_for_io(page);
  95                 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
  96         }
  97         ClearPageUptodate(page);
  98 }
  99
 100 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 101 {
 102         pgoff_t index = current_nat_addr(sbi, nid);
 103         return get_meta_page(sbi, index);
 104 }
 105
 106 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 107 {
 108         struct page *src_page;
 109         struct page *dst_page;
 110         pgoff_t src_off;
 111         pgoff_t dst_off;
 112         void *src_addr;
 113         void *dst_addr;
 114         struct f2fs_nm_info *nm_i = NM_I(sbi);
 115
 116         src_off = current_nat_addr(sbi, nid);
 117         dst_off = next_nat_addr(sbi, src_off);
 118
 119         /* get current nat block page with lock */
 120         src_page = get_meta_page(sbi, src_off);
 121         dst_page = grab_meta_page(sbi, dst_off);
 122         f2fs_bug_on(sbi, PageDirty(src_page));
 123
 124         src_addr = page_address(src_page);
 125         dst_addr = page_address(dst_page);
 126         memcpy(dst_addr, src_addr, PAGE_SIZE);
 127         set_page_dirty(dst_page);
 128         f2fs_put_page(src_page, 1);
 129
 130         set_to_next_nat(nm_i, nid);
 131
 132         return dst_page;
 133 }
 134
 135 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 136 {
 137         return radix_tree_lookup(&nm_i->nat_root, n);
 138 }
 139
 140 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 141                 nid_t start, unsigned int nr, struct nat_entry **ep)
 142 {
 143         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 144 }
 145
 146 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 147 {
 148         list_del(&e->list);
 149         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 150         nm_i->nat_cnt--;
 151         kmem_cache_free(nat_entry_slab, e);
 152 }
 153
 154 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 155                                                 struct nat_entry *ne)
 156 {
 157         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 158         struct nat_entry_set *head;
 159
 160         if (get_nat_flag(ne, IS_DIRTY))
 161                 return;
 162
 163         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 164         if (!head) {
 165                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 166
 167                 INIT_LIST_HEAD(&head->entry_list);
 168                 INIT_LIST_HEAD(&head->set_list);
 169                 head->set = set;
 170                 head->entry_cnt = 0;
 171                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 172         }
 173         list_move_tail(&ne->list, &head->entry_list);
 174         nm_i->dirty_nat_cnt++;
 175         head->entry_cnt++;
 176         set_nat_flag(ne, IS_DIRTY, true);
 177 }
 178
 179 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 180                                                 struct nat_entry *ne)
 181 {
 182         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 183         struct nat_entry_set *head;
 184
 185         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 186         if (head) {
 187                 list_move_tail(&ne->list, &nm_i->nat_entries);
 188                 set_nat_flag(ne, IS_DIRTY, false);
 189                 head->entry_cnt--;
 190                 nm_i->dirty_nat_cnt--;
 191         }
 192 }
 193
 194 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 195                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 196 {
 197         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 198                                                         start, nr);
 199 }
 200
 201 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 202 {
 203         struct f2fs_nm_info *nm_i = NM_I(sbi);
 204         struct nat_entry *e;
 205         bool need = false;
 206
 207         down_read(&nm_i->nat_tree_lock);
 208         e = __lookup_nat_cache(nm_i, nid);
 209         if (e) {
 210                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 211                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 212                         need = true;
 213         }
 214         up_read(&nm_i->nat_tree_lock);
 215         return need;
 216 }
 217
 218 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 219 {
 220         struct f2fs_nm_info *nm_i = NM_I(sbi);
 221         struct nat_entry *e;
 222         bool is_cp = true;
 223
 224         down_read(&nm_i->nat_tree_lock);
 225         e = __lookup_nat_cache(nm_i, nid);
 226         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 227                 is_cp = false;
 228         up_read(&nm_i->nat_tree_lock);
 229         return is_cp;
 230 }
 231
 232 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 233 {
 234         struct f2fs_nm_info *nm_i = NM_I(sbi);
 235         struct nat_entry *e;
 236         bool need_update = true;
 237
 238         down_read(&nm_i->nat_tree_lock);
 239         e = __lookup_nat_cache(nm_i, ino);
 240         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 241                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 242                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 243                 need_update = false;
 244         up_read(&nm_i->nat_tree_lock);
 245         return need_update;
 246 }
 247
 248 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 249 {
 250         struct nat_entry *new;
 251
 252         new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
 253         f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
 254         memset(new, 0, sizeof(struct nat_entry));
 255         nat_set_nid(new, nid);
 256         nat_reset_flag(new);
 257         list_add_tail(&new->list, &nm_i->nat_entries);
 258         nm_i->nat_cnt++;
 259         return new;
 260 }
 261
 262 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 263                                                 struct f2fs_nat_entry *ne)
 264 {
 265         struct f2fs_nm_info *nm_i = NM_I(sbi);
 266         struct nat_entry *e;
 267
 268         e = __lookup_nat_cache(nm_i, nid);
 269         if (!e) {
 270                 e = grab_nat_entry(nm_i, nid);
 271                 node_info_from_raw_nat(&e->ni, ne);
 272         } else {
 273                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 274                                 nat_get_blkaddr(e) !=
 275                                         le32_to_cpu(ne->block_addr) ||
 276                                 nat_get_version(e) != ne->version);
 277         }
 278 }
 279
 280 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 281                         block_t new_blkaddr, bool fsync_done)
 282 {
 283         struct f2fs_nm_info *nm_i = NM_I(sbi);
 284         struct nat_entry *e;
 285
 286         down_write(&nm_i->nat_tree_lock);
 287         e = __lookup_nat_cache(nm_i, ni->nid);
 288         if (!e) {
 289                 e = grab_nat_entry(nm_i, ni->nid);
 290                 copy_node_info(&e->ni, ni);
 291                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 292         } else if (new_blkaddr == NEW_ADDR) {
 293                 /*
 294                  * when nid is reallocated,
 295                  * previous nat entry can be remained in nat cache.
 296                  * So, reinitialize it with new information.
 297                  */
 298                 copy_node_info(&e->ni, ni);
 299                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 300         }
 301
 302         /* sanity check */
 303         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 304         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 305                         new_blkaddr == NULL_ADDR);
 306         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 307                         new_blkaddr == NEW_ADDR);
 308         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 309                         nat_get_blkaddr(e) != NULL_ADDR &&
 310                         new_blkaddr == NEW_ADDR);
 311
 312         /* increment version no as node is removed */
 313         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 314                 unsigned char version = nat_get_version(e);
 315                 nat_set_version(e, inc_node_version(version));
 316
 317                 /* in order to reuse the nid */
 318                 if (nm_i->next_scan_nid > ni->nid)
 319                         nm_i->next_scan_nid = ni->nid;
 320         }
 321
 322         /* change address */
 323         nat_set_blkaddr(e, new_blkaddr);
 324         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 325                 set_nat_flag(e, IS_CHECKPOINTED, false);
 326         __set_nat_cache_dirty(nm_i, e);
 327
 328         /* update fsync_mark if its inode nat entry is still alive */
 329         if (ni->nid != ni->ino)
 330                 e = __lookup_nat_cache(nm_i, ni->ino);
 331         if (e) {
 332                 if (fsync_done && ni->nid == ni->ino)
 333                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 334                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 335         }
 336         up_write(&nm_i->nat_tree_lock);
 337 }
 338
 339 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 340 {
 341         struct f2fs_nm_info *nm_i = NM_I(sbi);
 342         int nr = nr_shrink;
 343
 344         if (!down_write_trylock(&nm_i->nat_tree_lock))
 345                 return 0;
 346
 347         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 348                 struct nat_entry *ne;
 349                 ne = list_first_entry(&nm_i->nat_entries,
 350                                         struct nat_entry, list);
 351                 __del_from_nat_cache(nm_i, ne);
 352                 nr_shrink--;
 353         }
 354         up_write(&nm_i->nat_tree_lock);
 355         return nr - nr_shrink;
 356 }
 357
 358 /*
 359  * This function always returns success
 360  */
 361 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 362 {
 363         struct f2fs_nm_info *nm_i = NM_I(sbi);
 364         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 365         struct f2fs_journal *journal = curseg->journal;
 366         nid_t start_nid = START_NID(nid);
 367         struct f2fs_nat_block *nat_blk;
 368         struct page *page = NULL;
 369         struct f2fs_nat_entry ne;
 370         struct nat_entry *e;
 371         int i;
 372
 373         ni->nid = nid;
 374
 375         /* Check nat cache */
 376         down_read(&nm_i->nat_tree_lock);
 377         e = __lookup_nat_cache(nm_i, nid);
 378         if (e) {
 379                 ni->ino = nat_get_ino(e);
 380                 ni->blk_addr = nat_get_blkaddr(e);
 381                 ni->version = nat_get_version(e);
 382                 up_read(&nm_i->nat_tree_lock);
 383                 return;
 384         }
 385
 386         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 387
 388         /* Check current segment summary */
 389         down_read(&curseg->journal_rwsem);
 390         i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 391         if (i >= 0) {
 392                 ne = nat_in_journal(journal, i);
 393                 node_info_from_raw_nat(ni, &ne);
 394         }
 395         up_read(&curseg->journal_rwsem);
 396         if (i >= 0)
 397                 goto cache;
 398
 399         /* Fill node_info from nat page */
 400         page = get_current_nat_page(sbi, start_nid);
 401         nat_blk = (struct f2fs_nat_block *)page_address(page);
 402         ne = nat_blk->entries[nid - start_nid];
 403         node_info_from_raw_nat(ni, &ne);
 404         f2fs_put_page(page, 1);
 405 cache:
 406         up_read(&nm_i->nat_tree_lock);
 407         /* cache nat entry */
 408         down_write(&nm_i->nat_tree_lock);
 409         cache_nat_entry(sbi, nid, &ne);
 410         up_write(&nm_i->nat_tree_lock);
 411 }
 412
 413 /*
 414  * readahead MAX_RA_NODE number of node pages.
 415  */
 416 static void ra_node_pages(struct page *parent, int start, int n)
 417 {
 418         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 419         struct blk_plug plug;
 420         int i, end;
 421         nid_t nid;
 422
 423         blk_start_plug(&plug);
 424
 425         /* Then, try readahead for siblings of the desired node */
 426         end = start + n;
 427         end = min(end, NIDS_PER_BLOCK);
 428         for (i = start; i < end; i++) {
 429                 nid = get_nid(parent, i, false);
 430                 ra_node_page(sbi, nid);
 431         }
 432
 433         blk_finish_plug(&plug);
 434 }
 435
 436 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 437 {
 438         const long direct_index = ADDRS_PER_INODE(dn->inode);
 439         const long direct_blks = ADDRS_PER_BLOCK;
 440         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 441         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 442         int cur_level = dn->cur_level;
 443         int max_level = dn->max_level;
 444         pgoff_t base = 0;
 445
 446         if (!dn->max_level)
 447                 return pgofs + 1;
 448
 449         while (max_level-- > cur_level)
 450                 skipped_unit *= NIDS_PER_BLOCK;
 451
 452         switch (dn->max_level) {
 453         case 3:
 454                 base += 2 * indirect_blks;
 455         case 2:
 456                 base += 2 * direct_blks;
 457         case 1:
 458                 base += direct_index;
 459                 break;
 460         default:
 461                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 462         }
 463
 464         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 465 }
 466
 467 /*
 468  * The maximum depth is four.
 469  * Offset[0] will have raw inode offset.
 470  */
 471 static int get_node_path(struct inode *inode, long block,
 472                                 int offset[4], unsigned int noffset[4])
 473 {
 474         const long direct_index = ADDRS_PER_INODE(inode);
 475         const long direct_blks = ADDRS_PER_BLOCK;
 476         const long dptrs_per_blk = NIDS_PER_BLOCK;
 477         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 478         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 479         int n = 0;
 480         int level = 0;
 481
 482         noffset[0] = 0;
 483
 484         if (block < direct_index) {
 485                 offset[n] = block;
 486                 goto got;
 487         }
 488         block -= direct_index;
 489         if (block < direct_blks) {
 490                 offset[n++] = NODE_DIR1_BLOCK;
 491                 noffset[n] = 1;
 492                 offset[n] = block;
 493                 level = 1;
 494                 goto got;
 495         }
 496         block -= direct_blks;
 497         if (block < direct_blks) {
 498                 offset[n++] = NODE_DIR2_BLOCK;
 499                 noffset[n] = 2;
 500                 offset[n] = block;
 501                 level = 1;
 502                 goto got;
 503         }
 504         block -= direct_blks;
 505         if (block < indirect_blks) {
 506                 offset[n++] = NODE_IND1_BLOCK;
 507                 noffset[n] = 3;
 508                 offset[n++] = block / direct_blks;
 509                 noffset[n] = 4 + offset[n - 1];
 510                 offset[n] = block % direct_blks;
 511                 level = 2;
 512                 goto got;
 513         }
 514         block -= indirect_blks;
 515         if (block < indirect_blks) {
 516                 offset[n++] = NODE_IND2_BLOCK;
 517                 noffset[n] = 4 + dptrs_per_blk;
 518                 offset[n++] = block / direct_blks;
 519                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 520                 offset[n] = block % direct_blks;
 521                 level = 2;
 522                 goto got;
 523         }
 524         block -= indirect_blks;
 525         if (block < dindirect_blks) {
 526                 offset[n++] = NODE_DIND_BLOCK;
 527                 noffset[n] = 5 + (dptrs_per_blk * 2);
 528                 offset[n++] = block / indirect_blks;
 529                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 530                               offset[n - 1] * (dptrs_per_blk + 1);
 531                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 532                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 533                               offset[n - 2] * (dptrs_per_blk + 1) +
 534                               offset[n - 1];
 535                 offset[n] = block % direct_blks;
 536                 level = 3;
 537                 goto got;
 538         } else {
 539                 BUG();
 540         }
 541 got:
 542         return level;
 543 }
 544
 545 /*
 546  * Caller should call f2fs_put_dnode(dn).
 547  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 548  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 549  * In the case of RDONLY_NODE, we don't need to care about mutex.
 550  */
 551 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 552 {
 553         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 554         struct page *npage[4];
 555         struct page *parent = NULL;
 556         int offset[4];
 557         unsigned int noffset[4];
 558         nid_t nids[4];
 559         int level, i = 0;
 560         int err = 0;
 561
 562         level = get_node_path(dn->inode, index, offset, noffset);
 563
 564         nids[0] = dn->inode->i_ino;
 565         npage[0] = dn->inode_page;
 566
 567         if (!npage[0]) {
 568                 npage[0] = get_node_page(sbi, nids[0]);
 569                 if (IS_ERR(npage[0]))
 570                         return PTR_ERR(npage[0]);
 571         }
 572
 573         /* if inline_data is set, should not report any block indices */
 574         if (f2fs_has_inline_data(dn->inode) && index) {
 575                 err = -ENOENT;
 576                 f2fs_put_page(npage[0], 1);
 577                 goto release_out;
 578         }
 579
 580         parent = npage[0];
 581         if (level != 0)
 582                 nids[1] = get_nid(parent, offset[0], true);
 583         dn->inode_page = npage[0];
 584         dn->inode_page_locked = true;
 585
 586         /* get indirect or direct nodes */
 587         for (i = 1; i <= level; i++) {
 588                 bool done = false;
 589
 590                 if (!nids[i] && mode == ALLOC_NODE) {
 591                         /* alloc new node */
 592                         if (!alloc_nid(sbi, &(nids[i]))) {
 593                                 err = -ENOSPC;
 594                                 goto release_pages;
 595                         }
 596
 597                         dn->nid = nids[i];
 598                         npage[i] = new_node_page(dn, noffset[i], NULL);
 599                         if (IS_ERR(npage[i])) {
 600                                 alloc_nid_failed(sbi, nids[i]);
 601                                 err = PTR_ERR(npage[i]);
 602                                 goto release_pages;
 603                         }
 604
 605                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 606                         alloc_nid_done(sbi, nids[i]);
 607                         done = true;
 608                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 609                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 610                         if (IS_ERR(npage[i])) {
 611                                 err = PTR_ERR(npage[i]);
 612                                 goto release_pages;
 613                         }
 614                         done = true;
 615                 }
 616                 if (i == 1) {
 617                         dn->inode_page_locked = false;
 618                         unlock_page(parent);
 619                 } else {
 620                         f2fs_put_page(parent, 1);
 621                 }
 622
 623                 if (!done) {
 624                         npage[i] = get_node_page(sbi, nids[i]);
 625                         if (IS_ERR(npage[i])) {
 626                                 err = PTR_ERR(npage[i]);
 627                                 f2fs_put_page(npage[0], 0);
 628                                 goto release_out;
 629                         }
 630                 }
 631                 if (i < level) {
 632                         parent = npage[i];
 633                         nids[i + 1] = get_nid(parent, offset[i], false);
 634                 }
 635         }
 636         dn->nid = nids[level];
 637         dn->ofs_in_node = offset[level];
 638         dn->node_page = npage[level];
 639         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 640         return 0;
 641
 642 release_pages:
 643         f2fs_put_page(parent, 1);
 644         if (i > 1)
 645                 f2fs_put_page(npage[0], 0);
 646 release_out:
 647         dn->inode_page = NULL;
 648         dn->node_page = NULL;
 649         if (err == -ENOENT) {
 650                 dn->cur_level = i;
 651                 dn->max_level = level;
 652                 dn->ofs_in_node = offset[level];
 653         }
 654         return err;
 655 }
 656
 657 static void truncate_node(struct dnode_of_data *dn)
 658 {
 659         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 660         struct node_info ni;
 661
 662         get_node_info(sbi, dn->nid, &ni);
 663         if (dn->inode->i_blocks == 0) {
 664                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 665                 goto invalidate;
 666         }
 667         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 668
 669         /* Deallocate node address */
 670         invalidate_blocks(sbi, ni.blk_addr);
 671         dec_valid_node_count(sbi, dn->inode);
 672         set_node_addr(sbi, &ni, NULL_ADDR, false);
 673
 674         if (dn->nid == dn->inode->i_ino) {
 675                 remove_orphan_inode(sbi, dn->nid);
 676                 dec_valid_inode_count(sbi);
 677                 f2fs_inode_synced(dn->inode);
 678         }
 679 invalidate:
 680         clear_node_page_dirty(dn->node_page);
 681         set_sbi_flag(sbi, SBI_IS_DIRTY);
 682
 683         f2fs_put_page(dn->node_page, 1);
 684
 685         invalidate_mapping_pages(NODE_MAPPING(sbi),
 686                         dn->node_page->index, dn->node_page->index);
 687
 688         dn->node_page = NULL;
 689         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 690 }
 691
 692 static int truncate_dnode(struct dnode_of_data *dn)
 693 {
 694         struct page *page;
 695
 696         if (dn->nid == 0)
 697                 return 1;
 698
 699         /* get direct node */
 700         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 701         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 702                 return 1;
 703         else if (IS_ERR(page))
 704                 return PTR_ERR(page);
 705
 706         /* Make dnode_of_data for parameter */
 707         dn->node_page = page;
 708         dn->ofs_in_node = 0;
 709         truncate_data_blocks(dn);
 710         truncate_node(dn);
 711         return 1;
 712 }
 713
 714 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 715                                                 int ofs, int depth)
 716 {
 717         struct dnode_of_data rdn = *dn;
 718         struct page *page;
 719         struct f2fs_node *rn;
 720         nid_t child_nid;
 721         unsigned int child_nofs;
 722         int freed = 0;
 723         int i, ret;
 724
 725         if (dn->nid == 0)
 726                 return NIDS_PER_BLOCK + 1;
 727
 728         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 729
 730         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 731         if (IS_ERR(page)) {
 732                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 733                 return PTR_ERR(page);
 734         }
 735
 736         ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 737
 738         rn = F2FS_NODE(page);
 739         if (depth < 3) {
 740                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 741                         child_nid = le32_to_cpu(rn->in.nid[i]);
 742                         if (child_nid == 0)
 743                                 continue;
 744                         rdn.nid = child_nid;
 745                         ret = truncate_dnode(&rdn);
 746                         if (ret < 0)
 747                                 goto out_err;
 748                         if (set_nid(page, i, 0, false))
 749                                 dn->node_changed = true;
 750                 }
 751         } else {
 752                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 753                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 754                         child_nid = le32_to_cpu(rn->in.nid[i]);
 755                         if (child_nid == 0) {
 756                                 child_nofs += NIDS_PER_BLOCK + 1;
 757                                 continue;
 758                         }
 759                         rdn.nid = child_nid;
 760                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 761                         if (ret == (NIDS_PER_BLOCK + 1)) {
 762                                 if (set_nid(page, i, 0, false))
 763                                         dn->node_changed = true;
 764                                 child_nofs += ret;
 765                         } else if (ret < 0 && ret != -ENOENT) {
 766                                 goto out_err;
 767                         }
 768                 }
 769                 freed = child_nofs;
 770         }
 771
 772         if (!ofs) {
 773                 /* remove current indirect node */
 774                 dn->node_page = page;
 775                 truncate_node(dn);
 776                 freed++;
 777         } else {
 778                 f2fs_put_page(page, 1);
 779         }
 780         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 781         return freed;
 782
 783 out_err:
 784         f2fs_put_page(page, 1);
 785         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 786         return ret;
 787 }
 788
 789 static int truncate_partial_nodes(struct dnode_of_data *dn,
 790                         struct f2fs_inode *ri, int *offset, int depth)
 791 {
 792         struct page *pages[2];
 793         nid_t nid[3];
 794         nid_t child_nid;
 795         int err = 0;
 796         int i;
 797         int idx = depth - 2;
 798
 799         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 800         if (!nid[0])
 801                 return 0;
 802
 803         /* get indirect nodes in the path */
 804         for (i = 0; i < idx + 1; i++) {
 805                 /* reference count'll be increased */
 806                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 807                 if (IS_ERR(pages[i])) {
 808                         err = PTR_ERR(pages[i]);
 809                         idx = i - 1;
 810                         goto fail;
 811                 }
 812                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 813         }
 814
 815         ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 816
 817         /* free direct nodes linked to a partial indirect node */
 818         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 819                 child_nid = get_nid(pages[idx], i, false);
 820                 if (!child_nid)
 821                         continue;
 822                 dn->nid = child_nid;
 823                 err = truncate_dnode(dn);
 824                 if (err < 0)
 825                         goto fail;
 826                 if (set_nid(pages[idx], i, 0, false))
 827                         dn->node_changed = true;
 828         }
 829
 830         if (offset[idx + 1] == 0) {
 831                 dn->node_page = pages[idx];
 832                 dn->nid = nid[idx];
 833                 truncate_node(dn);
 834         } else {
 835                 f2fs_put_page(pages[idx], 1);
 836         }
 837         offset[idx]++;
 838         offset[idx + 1] = 0;
 839         idx--;
 840 fail:
 841         for (i = idx; i >= 0; i--)
 842                 f2fs_put_page(pages[i], 1);
 843
 844         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 845
 846         return err;
 847 }
 848
 849 /*
 850  * All the block addresses of data and nodes should be nullified.
 851  */
 852 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 853 {
 854         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 855         int err = 0, cont = 1;
 856         int level, offset[4], noffset[4];
 857         unsigned int nofs = 0;
 858         struct f2fs_inode *ri;
 859         struct dnode_of_data dn;
 860         struct page *page;
 861
 862         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 863
 864         level = get_node_path(inode, from, offset, noffset);
 865
 866         page = get_node_page(sbi, inode->i_ino);
 867         if (IS_ERR(page)) {
 868                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 869                 return PTR_ERR(page);
 870         }
 871
 872         set_new_dnode(&dn, inode, page, NULL, 0);
 873         unlock_page(page);
 874
 875         ri = F2FS_INODE(page);
 876         switch (level) {
 877         case 0:
 878         case 1:
 879                 nofs = noffset[1];
 880                 break;
 881         case 2:
 882                 nofs = noffset[1];
 883                 if (!offset[level - 1])
 884                         goto skip_partial;
 885                 err = truncate_partial_nodes(&dn, ri, offset, level);
 886                 if (err < 0 && err != -ENOENT)
 887                         goto fail;
 888                 nofs += 1 + NIDS_PER_BLOCK;
 889                 break;
 890         case 3:
 891                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 892                 if (!offset[level - 1])
 893                         goto skip_partial;
 894                 err = truncate_partial_nodes(&dn, ri, offset, level);
 895                 if (err < 0 && err != -ENOENT)
 896                         goto fail;
 897                 break;
 898         default:
 899                 BUG();
 900         }
 901
 902 skip_partial:
 903         while (cont) {
 904                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 905                 switch (offset[0]) {
 906                 case NODE_DIR1_BLOCK:
 907                 case NODE_DIR2_BLOCK:
 908                         err = truncate_dnode(&dn);
 909                         break;
 910
 911                 case NODE_IND1_BLOCK:
 912                 case NODE_IND2_BLOCK:
 913                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 914                         break;
 915
 916                 case NODE_DIND_BLOCK:
 917                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 918                         cont = 0;
 919                         break;
 920
 921                 default:
 922                         BUG();
 923                 }
 924                 if (err < 0 && err != -ENOENT)
 925                         goto fail;
 926                 if (offset[1] == 0 &&
 927                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 928                         lock_page(page);
 929                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
 930                         f2fs_wait_on_page_writeback(page, NODE, true);
 931                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 932                         set_page_dirty(page);
 933                         unlock_page(page);
 934                 }
 935                 offset[1] = 0;
 936                 offset[0]++;
 937                 nofs += err;
 938         }
 939 fail:
 940         f2fs_put_page(page, 0);
 941         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 942         return err > 0 ? 0 : err;
 943 }
 944
 945 int truncate_xattr_node(struct inode *inode, struct page *page)
 946 {
 947         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 948         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 949         struct dnode_of_data dn;
 950         struct page *npage;
 951
 952         if (!nid)
 953                 return 0;
 954
 955         npage = get_node_page(sbi, nid);
 956         if (IS_ERR(npage))
 957                 return PTR_ERR(npage);
 958
 959         f2fs_i_xnid_write(inode, 0);
 960
 961         /* need to do checkpoint during fsync */
 962         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 963
 964         set_new_dnode(&dn, inode, page, npage, nid);
 965
 966         if (page)
 967                 dn.inode_page_locked = true;
 968         truncate_node(&dn);
 969         return 0;
 970 }
 971
 972 /*
 973  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 974  * f2fs_unlock_op().
 975  */
 976 int remove_inode_page(struct inode *inode)
 977 {
 978         struct dnode_of_data dn;
 979         int err;
 980
 981         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 982         err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
 983         if (err)
 984                 return err;
 985
 986         err = truncate_xattr_node(inode, dn.inode_page);
 987         if (err) {
 988                 f2fs_put_dnode(&dn);
 989                 return err;
 990         }
 991
 992         /* remove potential inline_data blocks */
 993         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 994                                 S_ISLNK(inode->i_mode))
 995                 truncate_data_blocks_range(&dn, 1);
 996
 997         /* 0 is possible, after f2fs_new_inode() has failed */
 998         f2fs_bug_on(F2FS_I_SB(inode),
 999                         inode->i_blocks != 0 && inode->i_blocks != 1);
1000
1001         /* will put inode & node pages */
1002         truncate_node(&dn);
1003         return 0;
1004 }
1005
1006 struct page *new_inode_page(struct inode *inode)
1007 {
1008         struct dnode_of_data dn;
1009
1010         /* allocate inode page for new inode */
1011         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1012
1013         /* caller should f2fs_put_page(page, 1); */
1014         return new_node_page(&dn, 0, NULL);
1015 }
1016
1017 struct page *new_node_page(struct dnode_of_data *dn,
1018                                 unsigned int ofs, struct page *ipage)
1019 {
1020         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1021         struct node_info old_ni, new_ni;
1022         struct page *page;
1023         int err;
1024
1025         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1026                 return ERR_PTR(-EPERM);
1027
1028         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1029         if (!page)
1030                 return ERR_PTR(-ENOMEM);
1031
1032         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1033                 err = -ENOSPC;
1034                 goto fail;
1035         }
1036
1037         get_node_info(sbi, dn->nid, &old_ni);
1038
1039         /* Reinitialize old_ni with new node page */
1040         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1041         new_ni = old_ni;
1042         new_ni.ino = dn->inode->i_ino;
1043         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1044
1045         f2fs_wait_on_page_writeback(page, NODE, true);
1046         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1047         set_cold_node(dn->inode, page);
1048         if (!PageUptodate(page))
1049                 SetPageUptodate(page);
1050         if (set_page_dirty(page))
1051                 dn->node_changed = true;
1052
1053         if (f2fs_has_xattr_block(ofs))
1054                 f2fs_i_xnid_write(dn->inode, dn->nid);
1055
1056         if (ofs == 0)
1057                 inc_valid_inode_count(sbi);
1058         return page;
1059
1060 fail:
1061         clear_node_page_dirty(page);
1062         f2fs_put_page(page, 1);
1063         return ERR_PTR(err);
1064 }
1065
1066 /*
1067  * Caller should do after getting the following values.
1068  * 0: f2fs_put_page(page, 0)
1069  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1070  */
1071 static int read_node_page(struct page *page, int op_flags)
1072 {
1073         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1074         struct node_info ni;
1075         struct f2fs_io_info fio = {
1076                 .sbi = sbi,
1077                 .type = NODE,
1078                 .op = REQ_OP_READ,
1079                 .op_flags = op_flags,
1080                 .page = page,
1081                 .encrypted_page = NULL,
1082         };
1083
1084         if (PageUptodate(page))
1085                 return LOCKED_PAGE;
1086
1087         get_node_info(sbi, page->index, &ni);
1088
1089         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1090                 ClearPageUptodate(page);
1091                 return -ENOENT;
1092         }
1093
1094         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1095         return f2fs_submit_page_bio(&fio);
1096 }
1097
1098 /*
1099  * Readahead a node page
1100  */
1101 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1102 {
1103         struct page *apage;
1104         int err;
1105
1106         if (!nid)
1107                 return;
1108         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1109
1110         rcu_read_lock();
1111         apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1112         rcu_read_unlock();
1113         if (apage)
1114                 return;
1115
1116         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1117         if (!apage)
1118                 return;
1119
1120         err = read_node_page(apage, REQ_RAHEAD);
1121         f2fs_put_page(apage, err ? 1 : 0);
1122 }
1123
1124 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1125                                         struct page *parent, int start)
1126 {
1127         struct page *page;
1128         int err;
1129
1130         if (!nid)
1131                 return ERR_PTR(-ENOENT);
1132         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1133 repeat:
1134         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1135         if (!page)
1136                 return ERR_PTR(-ENOMEM);
1137
1138         err = read_node_page(page, 0);
1139         if (err < 0) {
1140                 f2fs_put_page(page, 1);
1141                 return ERR_PTR(err);
1142         } else if (err == LOCKED_PAGE) {
1143                 goto page_hit;
1144         }
1145
1146         if (parent)
1147                 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1148
1149         lock_page(page);
1150
1151         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1152                 f2fs_put_page(page, 1);
1153                 goto repeat;
1154         }
1155
1156         if (unlikely(!PageUptodate(page)))
1157                 goto out_err;
1158 page_hit:
1159         if(unlikely(nid != nid_of_node(page))) {
1160                 f2fs_bug_on(sbi, 1);
1161                 ClearPageUptodate(page);
1162 out_err:
1163                 f2fs_put_page(page, 1);
1164                 return ERR_PTR(-EIO);
1165         }
1166         return page;
1167 }
1168
1169 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1170 {
1171         return __get_node_page(sbi, nid, NULL, 0);
1172 }
1173
1174 struct page *get_node_page_ra(struct page *parent, int start)
1175 {
1176         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1177         nid_t nid = get_nid(parent, start, false);
1178
1179         return __get_node_page(sbi, nid, parent, start);
1180 }
1181
1182 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1183 {
1184         struct inode *inode;
1185         struct page *page;
1186         int ret;
1187
1188         /* should flush inline_data before evict_inode */
1189         inode = ilookup(sbi->sb, ino);
1190         if (!inode)
1191                 return;
1192
1193         page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1194         if (!page)
1195                 goto iput_out;
1196
1197         if (!PageUptodate(page))
1198                 goto page_out;
1199
1200         if (!PageDirty(page))
1201                 goto page_out;
1202
1203         if (!clear_page_dirty_for_io(page))
1204                 goto page_out;
1205
1206         ret = f2fs_write_inline_data(inode, page);
1207         inode_dec_dirty_pages(inode);
1208         remove_dirty_inode(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         int nwritten = 0;
1318
1319         if (atomic) {
1320                 last_page = last_fsync_dnode(sbi, ino);
1321                 if (IS_ERR_OR_NULL(last_page))
1322                         return PTR_ERR_OR_ZERO(last_page);
1323         }
1324 retry:
1325         pagevec_init(&pvec, 0);
1326         index = 0;
1327         end = ULONG_MAX;
1328
1329         while (index <= end) {
1330                 int i, nr_pages;
1331                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1332                                 PAGECACHE_TAG_DIRTY,
1333                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1334                 if (nr_pages == 0)
1335                         break;
1336
1337                 for (i = 0; i < nr_pages; i++) {
1338                         struct page *page = pvec.pages[i];
1339
1340                         if (unlikely(f2fs_cp_error(sbi))) {
1341                                 f2fs_put_page(last_page, 0);
1342                                 pagevec_release(&pvec);
1343                                 ret = -EIO;
1344                                 goto out;
1345                         }
1346
1347                         if (!IS_DNODE(page) || !is_cold_node(page))
1348                                 continue;
1349                         if (ino_of_node(page) != ino)
1350                                 continue;
1351
1352                         lock_page(page);
1353
1354                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1355 continue_unlock:
1356                                 unlock_page(page);
1357                                 continue;
1358                         }
1359                         if (ino_of_node(page) != ino)
1360                                 goto continue_unlock;
1361
1362                         if (!PageDirty(page) && page != last_page) {
1363                                 /* someone wrote it for us */
1364                                 goto continue_unlock;
1365                         }
1366
1367                         f2fs_wait_on_page_writeback(page, NODE, true);
1368                         BUG_ON(PageWriteback(page));
1369
1370                         if (!atomic || page == last_page) {
1371                                 set_fsync_mark(page, 1);
1372                                 if (IS_INODE(page)) {
1373                                         if (is_inode_flag_set(inode,
1374                                                                 FI_DIRTY_INODE))
1375                                                 update_inode(inode, page);
1376                                         set_dentry_mark(page,
1377                                                 need_dentry_mark(sbi, ino));
1378                                 }
1379                                 /*  may be written by other thread */
1380                                 if (!PageDirty(page))
1381                                         set_page_dirty(page);
1382                         }
1383
1384                         if (!clear_page_dirty_for_io(page))
1385                                 goto continue_unlock;
1386
1387                         ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1388                         if (ret) {
1389                                 unlock_page(page);
1390                                 f2fs_put_page(last_page, 0);
1391                                 break;
1392                         } else {
1393                                 nwritten++;
1394                         }
1395
1396                         if (page == last_page) {
1397                                 f2fs_put_page(page, 0);
1398                                 marked = true;
1399                                 break;
1400                         }
1401                 }
1402                 pagevec_release(&pvec);
1403                 cond_resched();
1404
1405                 if (ret || marked)
1406                         break;
1407         }
1408         if (!ret && atomic && !marked) {
1409                 f2fs_msg(sbi->sb, KERN_DEBUG,
1410                         "Retry to write fsync mark: ino=%u, idx=%lx",
1411                                         ino, last_page->index);
1412                 lock_page(last_page);
1413                 f2fs_wait_on_page_writeback(last_page, NODE, true);
1414                 set_page_dirty(last_page);
1415                 unlock_page(last_page);
1416                 goto retry;
1417         }
1418 out:
1419         if (nwritten)
1420                 f2fs_submit_merged_bio_cond(sbi, NULL, NULL, ino, NODE, WRITE);
1421         return ret ? -EIO: 0;
1422 }
1423
1424 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1425 {
1426         pgoff_t index, end;
1427         struct pagevec pvec;
1428         int step = 0;
1429         int nwritten = 0;
1430         int ret = 0;
1431
1432         pagevec_init(&pvec, 0);
1433
1434 next_step:
1435         index = 0;
1436         end = ULONG_MAX;
1437
1438         while (index <= end) {
1439                 int i, nr_pages;
1440                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1441                                 PAGECACHE_TAG_DIRTY,
1442                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1443                 if (nr_pages == 0)
1444                         break;
1445
1446                 for (i = 0; i < nr_pages; i++) {
1447                         struct page *page = pvec.pages[i];
1448
1449                         if (unlikely(f2fs_cp_error(sbi))) {
1450                                 pagevec_release(&pvec);
1451                                 ret = -EIO;
1452                                 goto out;
1453                         }
1454
1455                         /*
1456                          * flushing sequence with step:
1457                          * 0. indirect nodes
1458                          * 1. dentry dnodes
1459                          * 2. file dnodes
1460                          */
1461                         if (step == 0 && IS_DNODE(page))
1462                                 continue;
1463                         if (step == 1 && (!IS_DNODE(page) ||
1464                                                 is_cold_node(page)))
1465                                 continue;
1466                         if (step == 2 && (!IS_DNODE(page) ||
1467                                                 !is_cold_node(page)))
1468                                 continue;
1469 lock_node:
1470                         if (!trylock_page(page))
1471                                 continue;
1472
1473                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1474 continue_unlock:
1475                                 unlock_page(page);
1476                                 continue;
1477                         }
1478
1479                         if (!PageDirty(page)) {
1480                                 /* someone wrote it for us */
1481                                 goto continue_unlock;
1482                         }
1483
1484                         /* flush inline_data */
1485                         if (is_inline_node(page)) {
1486                                 clear_inline_node(page);
1487                                 unlock_page(page);
1488                                 flush_inline_data(sbi, ino_of_node(page));
1489                                 goto lock_node;
1490                         }
1491
1492                         f2fs_wait_on_page_writeback(page, NODE, true);
1493
1494                         BUG_ON(PageWriteback(page));
1495                         if (!clear_page_dirty_for_io(page))
1496                                 goto continue_unlock;
1497
1498                         set_fsync_mark(page, 0);
1499                         set_dentry_mark(page, 0);
1500
1501                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1502                                 unlock_page(page);
1503                         else
1504                                 nwritten++;
1505
1506                         if (--wbc->nr_to_write == 0)
1507                                 break;
1508                 }
1509                 pagevec_release(&pvec);
1510                 cond_resched();
1511
1512                 if (wbc->nr_to_write == 0) {
1513                         step = 2;
1514                         break;
1515                 }
1516         }
1517
1518         if (step < 2) {
1519                 step++;
1520                 goto next_step;
1521         }
1522 out:
1523         if (nwritten)
1524                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1525         return ret;
1526 }
1527
1528 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1529 {
1530         pgoff_t index = 0, end = ULONG_MAX;
1531         struct pagevec pvec;
1532         int ret2, ret = 0;
1533
1534         pagevec_init(&pvec, 0);
1535
1536         while (index <= end) {
1537                 int i, nr_pages;
1538                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1539                                 PAGECACHE_TAG_WRITEBACK,
1540                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1541                 if (nr_pages == 0)
1542                         break;
1543
1544                 for (i = 0; i < nr_pages; i++) {
1545                         struct page *page = pvec.pages[i];
1546
1547                         /* until radix tree lookup accepts end_index */
1548                         if (unlikely(page->index > end))
1549                                 continue;
1550
1551                         if (ino && ino_of_node(page) == ino) {
1552                                 f2fs_wait_on_page_writeback(page, NODE, true);
1553                                 if (TestClearPageError(page))
1554                                         ret = -EIO;
1555                         }
1556                 }
1557                 pagevec_release(&pvec);
1558                 cond_resched();
1559         }
1560
1561         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1562         if (!ret)
1563                 ret = ret2;
1564         return ret;
1565 }
1566
1567 static int f2fs_write_node_page(struct page *page,
1568                                 struct writeback_control *wbc)
1569 {
1570         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1571         nid_t nid;
1572         struct node_info ni;
1573         struct f2fs_io_info fio = {
1574                 .sbi = sbi,
1575                 .type = NODE,
1576                 .op = REQ_OP_WRITE,
1577                 .op_flags = wbc_to_write_flags(wbc),
1578                 .page = page,
1579                 .encrypted_page = NULL,
1580         };
1581
1582         trace_f2fs_writepage(page, NODE);
1583
1584         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1585                 goto redirty_out;
1586         if (unlikely(f2fs_cp_error(sbi)))
1587                 goto redirty_out;
1588
1589         /* get old block addr of this node page */
1590         nid = nid_of_node(page);
1591         f2fs_bug_on(sbi, page->index != nid);
1592
1593         if (wbc->for_reclaim) {
1594                 if (!down_read_trylock(&sbi->node_write))
1595                         goto redirty_out;
1596         } else {
1597                 down_read(&sbi->node_write);
1598         }
1599
1600         get_node_info(sbi, nid, &ni);
1601
1602         /* This page is already truncated */
1603         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1604                 ClearPageUptodate(page);
1605                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1606                 up_read(&sbi->node_write);
1607                 unlock_page(page);
1608                 return 0;
1609         }
1610
1611         set_page_writeback(page);
1612         fio.old_blkaddr = ni.blk_addr;
1613         write_node_page(nid, &fio);
1614         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1615         dec_page_count(sbi, F2FS_DIRTY_NODES);
1616         up_read(&sbi->node_write);
1617
1618         if (wbc->for_reclaim)
1619                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1620
1621         unlock_page(page);
1622
1623         if (unlikely(f2fs_cp_error(sbi)))
1624                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1625
1626         return 0;
1627
1628 redirty_out:
1629         redirty_page_for_writepage(wbc, page);
1630         return AOP_WRITEPAGE_ACTIVATE;
1631 }
1632
1633 static int f2fs_write_node_pages(struct address_space *mapping,
1634                             struct writeback_control *wbc)
1635 {
1636         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1637         struct blk_plug plug;
1638         long diff;
1639
1640         /* balancing f2fs's metadata in background */
1641         f2fs_balance_fs_bg(sbi);
1642
1643         /* collect a number of dirty node pages and write together */
1644         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1645                 goto skip_write;
1646
1647         trace_f2fs_writepages(mapping->host, wbc, NODE);
1648
1649         diff = nr_pages_to_write(sbi, NODE, wbc);
1650         wbc->sync_mode = WB_SYNC_NONE;
1651         blk_start_plug(&plug);
1652         sync_node_pages(sbi, wbc);
1653         blk_finish_plug(&plug);
1654         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1655         return 0;
1656
1657 skip_write:
1658         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1659         trace_f2fs_writepages(mapping->host, wbc, NODE);
1660         return 0;
1661 }
1662
1663 static int f2fs_set_node_page_dirty(struct page *page)
1664 {
1665         trace_f2fs_set_page_dirty(page, NODE);
1666
1667         if (!PageUptodate(page))
1668                 SetPageUptodate(page);
1669         if (!PageDirty(page)) {
1670                 f2fs_set_page_dirty_nobuffers(page);
1671                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1672                 SetPagePrivate(page);
1673                 f2fs_trace_pid(page);
1674                 return 1;
1675         }
1676         return 0;
1677 }
1678
1679 /*
1680  * Structure of the f2fs node operations
1681  */
1682 const struct address_space_operations f2fs_node_aops = {
1683         .writepage      = f2fs_write_node_page,
1684         .writepages     = f2fs_write_node_pages,
1685         .set_page_dirty = f2fs_set_node_page_dirty,
1686         .invalidatepage = f2fs_invalidate_page,
1687         .releasepage    = f2fs_release_page,
1688 #ifdef CONFIG_MIGRATION
1689         .migratepage    = f2fs_migrate_page,
1690 #endif
1691 };
1692
1693 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1694                                                 nid_t n)
1695 {
1696         return radix_tree_lookup(&nm_i->free_nid_root, n);
1697 }
1698
1699 static int __insert_nid_to_list(struct f2fs_sb_info *sbi,
1700                         struct free_nid *i, enum nid_list list, bool new)
1701 {
1702         struct f2fs_nm_info *nm_i = NM_I(sbi);
1703
1704         if (new) {
1705                 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1706                 if (err)
1707                         return err;
1708         }
1709
1710         f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1711                                                 i->state != NID_ALLOC);
1712         nm_i->nid_cnt[list]++;
1713         list_add_tail(&i->list, &nm_i->nid_list[list]);
1714         return 0;
1715 }
1716
1717 static void __remove_nid_from_list(struct f2fs_sb_info *sbi,
1718                         struct free_nid *i, enum nid_list list, bool reuse)
1719 {
1720         struct f2fs_nm_info *nm_i = NM_I(sbi);
1721
1722         f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1723                                                 i->state != NID_ALLOC);
1724         nm_i->nid_cnt[list]--;
1725         list_del(&i->list);
1726         if (!reuse)
1727                 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1728 }
1729
1730 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1731 {
1732         struct f2fs_nm_info *nm_i = NM_I(sbi);
1733         struct free_nid *i;
1734         struct nat_entry *ne;
1735         int err;
1736
1737         /* 0 nid should not be used */
1738         if (unlikely(nid == 0))
1739                 return 0;
1740
1741         if (build) {
1742                 /* do not add allocated nids */
1743                 ne = __lookup_nat_cache(nm_i, nid);
1744                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1745                                 nat_get_blkaddr(ne) != NULL_ADDR))
1746                         return 0;
1747         }
1748
1749         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1750         i->nid = nid;
1751         i->state = NID_NEW;
1752
1753         if (radix_tree_preload(GFP_NOFS)) {
1754                 kmem_cache_free(free_nid_slab, i);
1755                 return 0;
1756         }
1757
1758         spin_lock(&nm_i->nid_list_lock);
1759         err = __insert_nid_to_list(sbi, i, FREE_NID_LIST, true);
1760         spin_unlock(&nm_i->nid_list_lock);
1761         radix_tree_preload_end();
1762         if (err) {
1763                 kmem_cache_free(free_nid_slab, i);
1764                 return 0;
1765         }
1766         return 1;
1767 }
1768
1769 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1770 {
1771         struct f2fs_nm_info *nm_i = NM_I(sbi);
1772         struct free_nid *i;
1773         bool need_free = false;
1774
1775         spin_lock(&nm_i->nid_list_lock);
1776         i = __lookup_free_nid_list(nm_i, nid);
1777         if (i && i->state == NID_NEW) {
1778                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
1779                 need_free = true;
1780         }
1781         spin_unlock(&nm_i->nid_list_lock);
1782
1783         if (need_free)
1784                 kmem_cache_free(free_nid_slab, i);
1785 }
1786
1787 static void scan_nat_page(struct f2fs_sb_info *sbi,
1788                         struct page *nat_page, nid_t start_nid)
1789 {
1790         struct f2fs_nm_info *nm_i = NM_I(sbi);
1791         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1792         block_t blk_addr;
1793         int i;
1794
1795         i = start_nid % NAT_ENTRY_PER_BLOCK;
1796
1797         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1798
1799                 if (unlikely(start_nid >= nm_i->max_nid))
1800                         break;
1801
1802                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1803                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1804                 if (blk_addr == NULL_ADDR)
1805                         add_free_nid(sbi, start_nid, true);
1806         }
1807 }
1808
1809 static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync)
1810 {
1811         struct f2fs_nm_info *nm_i = NM_I(sbi);
1812         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1813         struct f2fs_journal *journal = curseg->journal;
1814         int i = 0;
1815         nid_t nid = nm_i->next_scan_nid;
1816
1817         /* Enough entries */
1818         if (nm_i->nid_cnt[FREE_NID_LIST] >= NAT_ENTRY_PER_BLOCK)
1819                 return;
1820
1821         if (!sync && !available_free_memory(sbi, FREE_NIDS))
1822                 return;
1823
1824         /* readahead nat pages to be scanned */
1825         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1826                                                         META_NAT, true);
1827
1828         down_read(&nm_i->nat_tree_lock);
1829
1830         while (1) {
1831                 struct page *page = get_current_nat_page(sbi, nid);
1832
1833                 scan_nat_page(sbi, page, nid);
1834                 f2fs_put_page(page, 1);
1835
1836                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1837                 if (unlikely(nid >= nm_i->max_nid))
1838                         nid = 0;
1839
1840                 if (++i >= FREE_NID_PAGES)
1841                         break;
1842         }
1843
1844         /* go to the next free nat pages to find free nids abundantly */
1845         nm_i->next_scan_nid = nid;
1846
1847         /* find free nids from current sum_pages */
1848         down_read(&curseg->journal_rwsem);
1849         for (i = 0; i < nats_in_cursum(journal); i++) {
1850                 block_t addr;
1851
1852                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1853                 nid = le32_to_cpu(nid_in_journal(journal, i));
1854                 if (addr == NULL_ADDR)
1855                         add_free_nid(sbi, nid, true);
1856                 else
1857                         remove_free_nid(sbi, nid);
1858         }
1859         up_read(&curseg->journal_rwsem);
1860         up_read(&nm_i->nat_tree_lock);
1861
1862         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1863                                         nm_i->ra_nid_pages, META_NAT, false);
1864 }
1865
1866 void build_free_nids(struct f2fs_sb_info *sbi, bool sync)
1867 {
1868         mutex_lock(&NM_I(sbi)->build_lock);
1869         __build_free_nids(sbi, sync);
1870         mutex_unlock(&NM_I(sbi)->build_lock);
1871 }
1872
1873 /*
1874  * If this function returns success, caller can obtain a new nid
1875  * from second parameter of this function.
1876  * The returned nid could be used ino as well as nid when inode is created.
1877  */
1878 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1879 {
1880         struct f2fs_nm_info *nm_i = NM_I(sbi);
1881         struct free_nid *i = NULL;
1882 retry:
1883 #ifdef CONFIG_F2FS_FAULT_INJECTION
1884         if (time_to_inject(sbi, FAULT_ALLOC_NID))
1885                 return false;
1886 #endif
1887         spin_lock(&nm_i->nid_list_lock);
1888
1889         if (unlikely(nm_i->available_nids == 0)) {
1890                 spin_unlock(&nm_i->nid_list_lock);
1891                 return false;
1892         }
1893
1894         /* We should not use stale free nids created by build_free_nids */
1895         if (nm_i->nid_cnt[FREE_NID_LIST] && !on_build_free_nids(nm_i)) {
1896                 f2fs_bug_on(sbi, list_empty(&nm_i->nid_list[FREE_NID_LIST]));
1897                 i = list_first_entry(&nm_i->nid_list[FREE_NID_LIST],
1898                                         struct free_nid, list);
1899                 *nid = i->nid;
1900
1901                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, true);
1902                 i->state = NID_ALLOC;
1903                 __insert_nid_to_list(sbi, i, ALLOC_NID_LIST, false);
1904                 nm_i->available_nids--;
1905                 spin_unlock(&nm_i->nid_list_lock);
1906                 return true;
1907         }
1908         spin_unlock(&nm_i->nid_list_lock);
1909
1910         /* Let's scan nat pages and its caches to get free nids */
1911         build_free_nids(sbi, true);
1912         goto retry;
1913 }
1914
1915 /*
1916  * alloc_nid() should be called prior to this function.
1917  */
1918 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1919 {
1920         struct f2fs_nm_info *nm_i = NM_I(sbi);
1921         struct free_nid *i;
1922
1923         spin_lock(&nm_i->nid_list_lock);
1924         i = __lookup_free_nid_list(nm_i, nid);
1925         f2fs_bug_on(sbi, !i);
1926         __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
1927         spin_unlock(&nm_i->nid_list_lock);
1928
1929         kmem_cache_free(free_nid_slab, i);
1930 }
1931
1932 /*
1933  * alloc_nid() should be called prior to this function.
1934  */
1935 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1936 {
1937         struct f2fs_nm_info *nm_i = NM_I(sbi);
1938         struct free_nid *i;
1939         bool need_free = false;
1940
1941         if (!nid)
1942                 return;
1943
1944         spin_lock(&nm_i->nid_list_lock);
1945         i = __lookup_free_nid_list(nm_i, nid);
1946         f2fs_bug_on(sbi, !i);
1947
1948         if (!available_free_memory(sbi, FREE_NIDS)) {
1949                 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
1950                 need_free = true;
1951         } else {
1952                 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, true);
1953                 i->state = NID_NEW;
1954                 __insert_nid_to_list(sbi, i, FREE_NID_LIST, false);
1955         }
1956
1957         nm_i->available_nids++;
1958
1959         spin_unlock(&nm_i->nid_list_lock);
1960
1961         if (need_free)
1962                 kmem_cache_free(free_nid_slab, i);
1963 }
1964
1965 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1966 {
1967         struct f2fs_nm_info *nm_i = NM_I(sbi);
1968         struct free_nid *i, *next;
1969         int nr = nr_shrink;
1970
1971         if (nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
1972                 return 0;
1973
1974         if (!mutex_trylock(&nm_i->build_lock))
1975                 return 0;
1976
1977         spin_lock(&nm_i->nid_list_lock);
1978         list_for_each_entry_safe(i, next, &nm_i->nid_list[FREE_NID_LIST],
1979                                                                         list) {
1980                 if (nr_shrink <= 0 ||
1981                                 nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
1982                         break;
1983
1984                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
1985                 kmem_cache_free(free_nid_slab, i);
1986                 nr_shrink--;
1987         }
1988         spin_unlock(&nm_i->nid_list_lock);
1989         mutex_unlock(&nm_i->build_lock);
1990
1991         return nr - nr_shrink;
1992 }
1993
1994 void recover_inline_xattr(struct inode *inode, struct page *page)
1995 {
1996         void *src_addr, *dst_addr;
1997         size_t inline_size;
1998         struct page *ipage;
1999         struct f2fs_inode *ri;
2000
2001         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2002         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2003
2004         ri = F2FS_INODE(page);
2005         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2006                 clear_inode_flag(inode, FI_INLINE_XATTR);
2007                 goto update_inode;
2008         }
2009
2010         dst_addr = inline_xattr_addr(ipage);
2011         src_addr = inline_xattr_addr(page);
2012         inline_size = inline_xattr_size(inode);
2013
2014         f2fs_wait_on_page_writeback(ipage, NODE, true);
2015         memcpy(dst_addr, src_addr, inline_size);
2016 update_inode:
2017         update_inode(inode, ipage);
2018         f2fs_put_page(ipage, 1);
2019 }
2020
2021 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2022 {
2023         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2024         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2025         nid_t new_xnid = nid_of_node(page);
2026         struct node_info ni;
2027
2028         /* 1: invalidate the previous xattr nid */
2029         if (!prev_xnid)
2030                 goto recover_xnid;
2031
2032         /* Deallocate node address */
2033         get_node_info(sbi, prev_xnid, &ni);
2034         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2035         invalidate_blocks(sbi, ni.blk_addr);
2036         dec_valid_node_count(sbi, inode);
2037         set_node_addr(sbi, &ni, NULL_ADDR, false);
2038
2039 recover_xnid:
2040         /* 2: allocate new xattr nid */
2041         if (unlikely(!inc_valid_node_count(sbi, inode)))
2042                 f2fs_bug_on(sbi, 1);
2043
2044         remove_free_nid(sbi, new_xnid);
2045         get_node_info(sbi, new_xnid, &ni);
2046         ni.ino = inode->i_ino;
2047         set_node_addr(sbi, &ni, NEW_ADDR, false);
2048         f2fs_i_xnid_write(inode, new_xnid);
2049
2050         /* 3: update xattr blkaddr */
2051         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
2052         set_node_addr(sbi, &ni, blkaddr, false);
2053 }
2054
2055 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2056 {
2057         struct f2fs_inode *src, *dst;
2058         nid_t ino = ino_of_node(page);
2059         struct node_info old_ni, new_ni;
2060         struct page *ipage;
2061
2062         get_node_info(sbi, ino, &old_ni);
2063
2064         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2065                 return -EINVAL;
2066 retry:
2067         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2068         if (!ipage) {
2069                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2070                 goto retry;
2071         }
2072
2073         /* Should not use this inode from free nid list */
2074         remove_free_nid(sbi, ino);
2075
2076         if (!PageUptodate(ipage))
2077                 SetPageUptodate(ipage);
2078         fill_node_footer(ipage, ino, ino, 0, true);
2079
2080         src = F2FS_INODE(page);
2081         dst = F2FS_INODE(ipage);
2082
2083         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2084         dst->i_size = 0;
2085         dst->i_blocks = cpu_to_le64(1);
2086         dst->i_links = cpu_to_le32(1);
2087         dst->i_xattr_nid = 0;
2088         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2089
2090         new_ni = old_ni;
2091         new_ni.ino = ino;
2092
2093         if (unlikely(!inc_valid_node_count(sbi, NULL)))
2094                 WARN_ON(1);
2095         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2096         inc_valid_inode_count(sbi);
2097         set_page_dirty(ipage);
2098         f2fs_put_page(ipage, 1);
2099         return 0;
2100 }
2101
2102 int restore_node_summary(struct f2fs_sb_info *sbi,
2103                         unsigned int segno, struct f2fs_summary_block *sum)
2104 {
2105         struct f2fs_node *rn;
2106         struct f2fs_summary *sum_entry;
2107         block_t addr;
2108         int i, idx, last_offset, nrpages;
2109
2110         /* scan the node segment */
2111         last_offset = sbi->blocks_per_seg;
2112         addr = START_BLOCK(sbi, segno);
2113         sum_entry = &sum->entries[0];
2114
2115         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2116                 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2117
2118                 /* readahead node pages */
2119                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2120
2121                 for (idx = addr; idx < addr + nrpages; idx++) {
2122                         struct page *page = get_tmp_page(sbi, idx);
2123
2124                         rn = F2FS_NODE(page);
2125                         sum_entry->nid = rn->footer.nid;
2126                         sum_entry->version = 0;
2127                         sum_entry->ofs_in_node = 0;
2128                         sum_entry++;
2129                         f2fs_put_page(page, 1);
2130                 }
2131
2132                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2133                                                         addr + nrpages);
2134         }
2135         return 0;
2136 }
2137
2138 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2139 {
2140         struct f2fs_nm_info *nm_i = NM_I(sbi);
2141         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2142         struct f2fs_journal *journal = curseg->journal;
2143         int i;
2144
2145         down_write(&curseg->journal_rwsem);
2146         for (i = 0; i < nats_in_cursum(journal); i++) {
2147                 struct nat_entry *ne;
2148                 struct f2fs_nat_entry raw_ne;
2149                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2150
2151                 raw_ne = nat_in_journal(journal, i);
2152
2153                 ne = __lookup_nat_cache(nm_i, nid);
2154                 if (!ne) {
2155                         ne = grab_nat_entry(nm_i, nid);
2156                         node_info_from_raw_nat(&ne->ni, &raw_ne);
2157                 }
2158
2159                 /*
2160                  * if a free nat in journal has not been used after last
2161                  * checkpoint, we should remove it from available nids,
2162                  * since later we will add it again.
2163                  */
2164                 if (!get_nat_flag(ne, IS_DIRTY) &&
2165                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2166                         spin_lock(&nm_i->nid_list_lock);
2167                         nm_i->available_nids--;
2168                         spin_unlock(&nm_i->nid_list_lock);
2169                 }
2170
2171                 __set_nat_cache_dirty(nm_i, ne);
2172         }
2173         update_nats_in_cursum(journal, -i);
2174         up_write(&curseg->journal_rwsem);
2175 }
2176
2177 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2178                                                 struct list_head *head, int max)
2179 {
2180         struct nat_entry_set *cur;
2181
2182         if (nes->entry_cnt >= max)
2183                 goto add_out;
2184
2185         list_for_each_entry(cur, head, set_list) {
2186                 if (cur->entry_cnt >= nes->entry_cnt) {
2187                         list_add(&nes->set_list, cur->set_list.prev);
2188                         return;
2189                 }
2190         }
2191 add_out:
2192         list_add_tail(&nes->set_list, head);
2193 }
2194
2195 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2196                                         struct nat_entry_set *set)
2197 {
2198         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2199         struct f2fs_journal *journal = curseg->journal;
2200         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2201         bool to_journal = true;
2202         struct f2fs_nat_block *nat_blk;
2203         struct nat_entry *ne, *cur;
2204         struct page *page = NULL;
2205
2206         /*
2207          * there are two steps to flush nat entries:
2208          * #1, flush nat entries to journal in current hot data summary block.
2209          * #2, flush nat entries to nat page.
2210          */
2211         if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2212                 to_journal = false;
2213
2214         if (to_journal) {
2215                 down_write(&curseg->journal_rwsem);
2216         } else {
2217                 page = get_next_nat_page(sbi, start_nid);
2218                 nat_blk = page_address(page);
2219                 f2fs_bug_on(sbi, !nat_blk);
2220         }
2221
2222         /* flush dirty nats in nat entry set */
2223         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2224                 struct f2fs_nat_entry *raw_ne;
2225                 nid_t nid = nat_get_nid(ne);
2226                 int offset;
2227
2228                 if (nat_get_blkaddr(ne) == NEW_ADDR)
2229                         continue;
2230
2231                 if (to_journal) {
2232                         offset = lookup_journal_in_cursum(journal,
2233                                                         NAT_JOURNAL, nid, 1);
2234                         f2fs_bug_on(sbi, offset < 0);
2235                         raw_ne = &nat_in_journal(journal, offset);
2236                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2237                 } else {
2238                         raw_ne = &nat_blk->entries[nid - start_nid];
2239                 }
2240                 raw_nat_from_node_info(raw_ne, &ne->ni);
2241                 nat_reset_flag(ne);
2242                 __clear_nat_cache_dirty(NM_I(sbi), ne);
2243                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2244                         add_free_nid(sbi, nid, false);
2245                         spin_lock(&NM_I(sbi)->nid_list_lock);
2246                         NM_I(sbi)->available_nids++;
2247                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2248                 }
2249         }
2250
2251         if (to_journal)
2252                 up_write(&curseg->journal_rwsem);
2253         else
2254                 f2fs_put_page(page, 1);
2255
2256         f2fs_bug_on(sbi, set->entry_cnt);
2257
2258         radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2259         kmem_cache_free(nat_entry_set_slab, set);
2260 }
2261
2262 /*
2263  * This function is called during the checkpointing process.
2264  */
2265 void flush_nat_entries(struct f2fs_sb_info *sbi)
2266 {
2267         struct f2fs_nm_info *nm_i = NM_I(sbi);
2268         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2269         struct f2fs_journal *journal = curseg->journal;
2270         struct nat_entry_set *setvec[SETVEC_SIZE];
2271         struct nat_entry_set *set, *tmp;
2272         unsigned int found;
2273         nid_t set_idx = 0;
2274         LIST_HEAD(sets);
2275
2276         if (!nm_i->dirty_nat_cnt)
2277                 return;
2278
2279         down_write(&nm_i->nat_tree_lock);
2280
2281         /*
2282          * if there are no enough space in journal to store dirty nat
2283          * entries, remove all entries from journal and merge them
2284          * into nat entry set.
2285          */
2286         if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2287                 remove_nats_in_journal(sbi);
2288
2289         while ((found = __gang_lookup_nat_set(nm_i,
2290                                         set_idx, SETVEC_SIZE, setvec))) {
2291                 unsigned idx;
2292                 set_idx = setvec[found - 1]->set + 1;
2293                 for (idx = 0; idx < found; idx++)
2294                         __adjust_nat_entry_set(setvec[idx], &sets,
2295                                                 MAX_NAT_JENTRIES(journal));
2296         }
2297
2298         /* flush dirty nats in nat entry set */
2299         list_for_each_entry_safe(set, tmp, &sets, set_list)
2300                 __flush_nat_entry_set(sbi, set);
2301
2302         up_write(&nm_i->nat_tree_lock);
2303
2304         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2305 }
2306
2307 static int init_node_manager(struct f2fs_sb_info *sbi)
2308 {
2309         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2310         struct f2fs_nm_info *nm_i = NM_I(sbi);
2311         unsigned char *version_bitmap;
2312         unsigned int nat_segs, nat_blocks;
2313
2314         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2315
2316         /* segment_count_nat includes pair segment so divide to 2. */
2317         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2318         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2319
2320         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2321
2322         /* not used nids: 0, node, meta, (and root counted as valid node) */
2323         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2324                                                         F2FS_RESERVED_NODE_NUM;
2325         nm_i->nid_cnt[FREE_NID_LIST] = 0;
2326         nm_i->nid_cnt[ALLOC_NID_LIST] = 0;
2327         nm_i->nat_cnt = 0;
2328         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2329         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2330         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2331
2332         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2333         INIT_LIST_HEAD(&nm_i->nid_list[FREE_NID_LIST]);
2334         INIT_LIST_HEAD(&nm_i->nid_list[ALLOC_NID_LIST]);
2335         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2336         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2337         INIT_LIST_HEAD(&nm_i->nat_entries);
2338
2339         mutex_init(&nm_i->build_lock);
2340         spin_lock_init(&nm_i->nid_list_lock);
2341         init_rwsem(&nm_i->nat_tree_lock);
2342
2343         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2344         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2345         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2346         if (!version_bitmap)
2347                 return -EFAULT;
2348
2349         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2350                                         GFP_KERNEL);
2351         if (!nm_i->nat_bitmap)
2352                 return -ENOMEM;
2353         return 0;
2354 }
2355
2356 int build_node_manager(struct f2fs_sb_info *sbi)
2357 {
2358         int err;
2359
2360         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2361         if (!sbi->nm_info)
2362                 return -ENOMEM;
2363
2364         err = init_node_manager(sbi);
2365         if (err)
2366                 return err;
2367
2368         build_free_nids(sbi, true);
2369         return 0;
2370 }
2371
2372 void destroy_node_manager(struct f2fs_sb_info *sbi)
2373 {
2374         struct f2fs_nm_info *nm_i = NM_I(sbi);
2375         struct free_nid *i, *next_i;
2376         struct nat_entry *natvec[NATVEC_SIZE];
2377         struct nat_entry_set *setvec[SETVEC_SIZE];
2378         nid_t nid = 0;
2379         unsigned int found;
2380
2381         if (!nm_i)
2382                 return;
2383
2384         /* destroy free nid list */
2385         spin_lock(&nm_i->nid_list_lock);
2386         list_for_each_entry_safe(i, next_i, &nm_i->nid_list[FREE_NID_LIST],
2387                                                                         list) {
2388                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2389                 spin_unlock(&nm_i->nid_list_lock);
2390                 kmem_cache_free(free_nid_slab, i);
2391                 spin_lock(&nm_i->nid_list_lock);
2392         }
2393         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID_LIST]);
2394         f2fs_bug_on(sbi, nm_i->nid_cnt[ALLOC_NID_LIST]);
2395         f2fs_bug_on(sbi, !list_empty(&nm_i->nid_list[ALLOC_NID_LIST]));
2396         spin_unlock(&nm_i->nid_list_lock);
2397
2398         /* destroy nat cache */
2399         down_write(&nm_i->nat_tree_lock);
2400         while ((found = __gang_lookup_nat_cache(nm_i,
2401                                         nid, NATVEC_SIZE, natvec))) {
2402                 unsigned idx;
2403
2404                 nid = nat_get_nid(natvec[found - 1]) + 1;
2405                 for (idx = 0; idx < found; idx++)
2406                         __del_from_nat_cache(nm_i, natvec[idx]);
2407         }
2408         f2fs_bug_on(sbi, nm_i->nat_cnt);
2409
2410         /* destroy nat set cache */
2411         nid = 0;
2412         while ((found = __gang_lookup_nat_set(nm_i,
2413                                         nid, SETVEC_SIZE, setvec))) {
2414                 unsigned idx;
2415
2416                 nid = setvec[found - 1]->set + 1;
2417                 for (idx = 0; idx < found; idx++) {
2418                         /* entry_cnt is not zero, when cp_error was occurred */
2419                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2420                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2421                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2422                 }
2423         }
2424         up_write(&nm_i->nat_tree_lock);
2425
2426         kfree(nm_i->nat_bitmap);
2427         sbi->nm_info = NULL;
2428         kfree(nm_i);
2429 }
2430
2431 int __init create_node_manager_caches(void)
2432 {
2433         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2434                         sizeof(struct nat_entry));
2435         if (!nat_entry_slab)
2436                 goto fail;
2437
2438         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2439                         sizeof(struct free_nid));
2440         if (!free_nid_slab)
2441                 goto destroy_nat_entry;
2442
2443         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2444                         sizeof(struct nat_entry_set));
2445         if (!nat_entry_set_slab)
2446                 goto destroy_free_nid;
2447         return 0;
2448
2449 destroy_free_nid:
2450         kmem_cache_destroy(free_nid_slab);
2451 destroy_nat_entry:
2452         kmem_cache_destroy(nat_entry_slab);
2453 fail:
2454         return -ENOMEM;
2455 }
2456
2457 void destroy_node_manager_caches(void)
2458 {
2459         kmem_cache_destroy(nat_entry_set_slab);
2460         kmem_cache_destroy(free_nid_slab);
2461         kmem_cache_destroy(nat_entry_slab);
2462 }