fs/f2fs/node.c

   1 /*
   2  * fs/f2fs/node.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/mpage.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/swap.h>
  18
  19 #include "f2fs.h"
  20 #include "node.h"
  21 #include "segment.h"
  22 #include <trace/events/f2fs.h>
  23
  24 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
  25
  26 static struct kmem_cache *nat_entry_slab;
  27 static struct kmem_cache *free_nid_slab;
  28 static struct kmem_cache *nat_entry_set_slab;
  29
  30 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
  31 {
  32         struct f2fs_nm_info *nm_i = NM_I(sbi);
  33         struct sysinfo val;
  34         unsigned long avail_ram;
  35         unsigned long mem_size = 0;
  36         bool res = false;
  37
  38         si_meminfo(&val);
  39
  40         /* only uses low memory */
  41         avail_ram = val.totalram - val.totalhigh;
  42
  43         /* give 25%, 25%, 50%, 50% memory for each components respectively */
  44         if (type == FREE_NIDS) {
  45                 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
  46                                                         PAGE_CACHE_SHIFT;
  47                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  48         } else if (type == NAT_ENTRIES) {
  49                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
  50                                                         PAGE_CACHE_SHIFT;
  51                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  52         } else if (type == DIRTY_DENTS) {
  53                 if (sbi->sb->s_bdi->dirty_exceeded)
  54                         return false;
  55                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  56                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  57         } else if (type == INO_ENTRIES) {
  58                 int i;
  59
  60                 if (sbi->sb->s_bdi->dirty_exceeded)
  61                         return false;
  62                 for (i = 0; i <= UPDATE_INO; i++)
  63                         mem_size += (sbi->im[i].ino_num *
  64                                 sizeof(struct ino_entry)) >> PAGE_CACHE_SHIFT;
  65                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  66         }
  67         return res;
  68 }
  69
  70 static void clear_node_page_dirty(struct page *page)
  71 {
  72         struct address_space *mapping = page->mapping;
  73         unsigned int long flags;
  74
  75         if (PageDirty(page)) {
  76                 spin_lock_irqsave(&mapping->tree_lock, flags);
  77                 radix_tree_tag_clear(&mapping->page_tree,
  78                                 page_index(page),
  79                                 PAGECACHE_TAG_DIRTY);
  80                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  81
  82                 clear_page_dirty_for_io(page);
  83                 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
  84         }
  85         ClearPageUptodate(page);
  86 }
  87
  88 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  89 {
  90         pgoff_t index = current_nat_addr(sbi, nid);
  91         return get_meta_page(sbi, index);
  92 }
  93
  94 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  95 {
  96         struct page *src_page;
  97         struct page *dst_page;
  98         pgoff_t src_off;
  99         pgoff_t dst_off;
 100         void *src_addr;
 101         void *dst_addr;
 102         struct f2fs_nm_info *nm_i = NM_I(sbi);
 103
 104         src_off = current_nat_addr(sbi, nid);
 105         dst_off = next_nat_addr(sbi, src_off);
 106
 107         /* get current nat block page with lock */
 108         src_page = get_meta_page(sbi, src_off);
 109         dst_page = grab_meta_page(sbi, dst_off);
 110         f2fs_bug_on(sbi, PageDirty(src_page));
 111
 112         src_addr = page_address(src_page);
 113         dst_addr = page_address(dst_page);
 114         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
 115         set_page_dirty(dst_page);
 116         f2fs_put_page(src_page, 1);
 117
 118         set_to_next_nat(nm_i, nid);
 119
 120         return dst_page;
 121 }
 122
 123 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 124 {
 125         return radix_tree_lookup(&nm_i->nat_root, n);
 126 }
 127
 128 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 129                 nid_t start, unsigned int nr, struct nat_entry **ep)
 130 {
 131         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 132 }
 133
 134 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 135 {
 136         list_del(&e->list);
 137         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 138         nm_i->nat_cnt--;
 139         kmem_cache_free(nat_entry_slab, e);
 140 }
 141
 142 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 143                                                 struct nat_entry *ne)
 144 {
 145         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 146         struct nat_entry_set *head;
 147
 148         if (get_nat_flag(ne, IS_DIRTY))
 149                 return;
 150 retry:
 151         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 152         if (!head) {
 153                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_ATOMIC);
 154
 155                 INIT_LIST_HEAD(&head->entry_list);
 156                 INIT_LIST_HEAD(&head->set_list);
 157                 head->set = set;
 158                 head->entry_cnt = 0;
 159
 160                 if (radix_tree_insert(&nm_i->nat_set_root, set, head)) {
 161                         cond_resched();
 162                         goto retry;
 163                 }
 164         }
 165         list_move_tail(&ne->list, &head->entry_list);
 166         nm_i->dirty_nat_cnt++;
 167         head->entry_cnt++;
 168         set_nat_flag(ne, IS_DIRTY, true);
 169 }
 170
 171 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 172                                                 struct nat_entry *ne)
 173 {
 174         nid_t set = ne->ni.nid / NAT_ENTRY_PER_BLOCK;
 175         struct nat_entry_set *head;
 176
 177         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 178         if (head) {
 179                 list_move_tail(&ne->list, &nm_i->nat_entries);
 180                 set_nat_flag(ne, IS_DIRTY, false);
 181                 head->entry_cnt--;
 182                 nm_i->dirty_nat_cnt--;
 183         }
 184 }
 185
 186 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 187                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 188 {
 189         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 190                                                         start, nr);
 191 }
 192
 193 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 194 {
 195         struct f2fs_nm_info *nm_i = NM_I(sbi);
 196         struct nat_entry *e;
 197         bool is_cp = true;
 198
 199         read_lock(&nm_i->nat_tree_lock);
 200         e = __lookup_nat_cache(nm_i, nid);
 201         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 202                 is_cp = false;
 203         read_unlock(&nm_i->nat_tree_lock);
 204         return is_cp;
 205 }
 206
 207 bool has_fsynced_inode(struct f2fs_sb_info *sbi, nid_t ino)
 208 {
 209         struct f2fs_nm_info *nm_i = NM_I(sbi);
 210         struct nat_entry *e;
 211         bool fsynced = false;
 212
 213         read_lock(&nm_i->nat_tree_lock);
 214         e = __lookup_nat_cache(nm_i, ino);
 215         if (e && get_nat_flag(e, HAS_FSYNCED_INODE))
 216                 fsynced = true;
 217         read_unlock(&nm_i->nat_tree_lock);
 218         return fsynced;
 219 }
 220
 221 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 222 {
 223         struct f2fs_nm_info *nm_i = NM_I(sbi);
 224         struct nat_entry *e;
 225         bool need_update = true;
 226
 227         read_lock(&nm_i->nat_tree_lock);
 228         e = __lookup_nat_cache(nm_i, ino);
 229         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 230                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 231                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 232                 need_update = false;
 233         read_unlock(&nm_i->nat_tree_lock);
 234         return need_update;
 235 }
 236
 237 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 238 {
 239         struct nat_entry *new;
 240
 241         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 242         if (!new)
 243                 return NULL;
 244         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 245                 kmem_cache_free(nat_entry_slab, new);
 246                 return NULL;
 247         }
 248         memset(new, 0, sizeof(struct nat_entry));
 249         nat_set_nid(new, nid);
 250         nat_reset_flag(new);
 251         list_add_tail(&new->list, &nm_i->nat_entries);
 252         nm_i->nat_cnt++;
 253         return new;
 254 }
 255
 256 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 257                                                 struct f2fs_nat_entry *ne)
 258 {
 259         struct nat_entry *e;
 260 retry:
 261         write_lock(&nm_i->nat_tree_lock);
 262         e = __lookup_nat_cache(nm_i, nid);
 263         if (!e) {
 264                 e = grab_nat_entry(nm_i, nid);
 265                 if (!e) {
 266                         write_unlock(&nm_i->nat_tree_lock);
 267                         goto retry;
 268                 }
 269                 node_info_from_raw_nat(&e->ni, ne);
 270         }
 271         write_unlock(&nm_i->nat_tree_lock);
 272 }
 273
 274 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 275                         block_t new_blkaddr, bool fsync_done)
 276 {
 277         struct f2fs_nm_info *nm_i = NM_I(sbi);
 278         struct nat_entry *e;
 279 retry:
 280         write_lock(&nm_i->nat_tree_lock);
 281         e = __lookup_nat_cache(nm_i, ni->nid);
 282         if (!e) {
 283                 e = grab_nat_entry(nm_i, ni->nid);
 284                 if (!e) {
 285                         write_unlock(&nm_i->nat_tree_lock);
 286                         goto retry;
 287                 }
 288                 e->ni = *ni;
 289                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 290         } else if (new_blkaddr == NEW_ADDR) {
 291                 /*
 292                  * when nid is reallocated,
 293                  * previous nat entry can be remained in nat cache.
 294                  * So, reinitialize it with new information.
 295                  */
 296                 e->ni = *ni;
 297                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 298         }
 299
 300         /* sanity check */
 301         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 302         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 303                         new_blkaddr == NULL_ADDR);
 304         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 305                         new_blkaddr == NEW_ADDR);
 306         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 307                         nat_get_blkaddr(e) != NULL_ADDR &&
 308                         new_blkaddr == NEW_ADDR);
 309
 310         /* increment version no as node is removed */
 311         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 312                 unsigned char version = nat_get_version(e);
 313                 nat_set_version(e, inc_node_version(version));
 314         }
 315
 316         /* change address */
 317         nat_set_blkaddr(e, new_blkaddr);
 318         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 319                 set_nat_flag(e, IS_CHECKPOINTED, false);
 320         __set_nat_cache_dirty(nm_i, e);
 321
 322         /* update fsync_mark if its inode nat entry is still alive */
 323         e = __lookup_nat_cache(nm_i, ni->ino);
 324         if (e) {
 325                 if (fsync_done && ni->nid == ni->ino)
 326                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 327                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 328         }
 329         write_unlock(&nm_i->nat_tree_lock);
 330 }
 331
 332 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 333 {
 334         struct f2fs_nm_info *nm_i = NM_I(sbi);
 335
 336         if (available_free_memory(sbi, NAT_ENTRIES))
 337                 return 0;
 338
 339         write_lock(&nm_i->nat_tree_lock);
 340         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 341                 struct nat_entry *ne;
 342                 ne = list_first_entry(&nm_i->nat_entries,
 343                                         struct nat_entry, list);
 344                 __del_from_nat_cache(nm_i, ne);
 345                 nr_shrink--;
 346         }
 347         write_unlock(&nm_i->nat_tree_lock);
 348         return nr_shrink;
 349 }
 350
 351 /*
 352  * This function always returns success
 353  */
 354 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 355 {
 356         struct f2fs_nm_info *nm_i = NM_I(sbi);
 357         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 358         struct f2fs_summary_block *sum = curseg->sum_blk;
 359         nid_t start_nid = START_NID(nid);
 360         struct f2fs_nat_block *nat_blk;
 361         struct page *page = NULL;
 362         struct f2fs_nat_entry ne;
 363         struct nat_entry *e;
 364         int i;
 365
 366         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 367         ni->nid = nid;
 368
 369         /* Check nat cache */
 370         read_lock(&nm_i->nat_tree_lock);
 371         e = __lookup_nat_cache(nm_i, nid);
 372         if (e) {
 373                 ni->ino = nat_get_ino(e);
 374                 ni->blk_addr = nat_get_blkaddr(e);
 375                 ni->version = nat_get_version(e);
 376         }
 377         read_unlock(&nm_i->nat_tree_lock);
 378         if (e)
 379                 return;
 380
 381         /* Check current segment summary */
 382         mutex_lock(&curseg->curseg_mutex);
 383         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 384         if (i >= 0) {
 385                 ne = nat_in_journal(sum, i);
 386                 node_info_from_raw_nat(ni, &ne);
 387         }
 388         mutex_unlock(&curseg->curseg_mutex);
 389         if (i >= 0)
 390                 goto cache;
 391
 392         /* Fill node_info from nat page */
 393         page = get_current_nat_page(sbi, start_nid);
 394         nat_blk = (struct f2fs_nat_block *)page_address(page);
 395         ne = nat_blk->entries[nid - start_nid];
 396         node_info_from_raw_nat(ni, &ne);
 397         f2fs_put_page(page, 1);
 398 cache:
 399         /* cache nat entry */
 400         cache_nat_entry(NM_I(sbi), nid, &ne);
 401 }
 402
 403 /*
 404  * The maximum depth is four.
 405  * Offset[0] will have raw inode offset.
 406  */
 407 static int get_node_path(struct f2fs_inode_info *fi, long block,
 408                                 int offset[4], unsigned int noffset[4])
 409 {
 410         const long direct_index = ADDRS_PER_INODE(fi);
 411         const long direct_blks = ADDRS_PER_BLOCK;
 412         const long dptrs_per_blk = NIDS_PER_BLOCK;
 413         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 414         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 415         int n = 0;
 416         int level = 0;
 417
 418         noffset[0] = 0;
 419
 420         if (block < direct_index) {
 421                 offset[n] = block;
 422                 goto got;
 423         }
 424         block -= direct_index;
 425         if (block < direct_blks) {
 426                 offset[n++] = NODE_DIR1_BLOCK;
 427                 noffset[n] = 1;
 428                 offset[n] = block;
 429                 level = 1;
 430                 goto got;
 431         }
 432         block -= direct_blks;
 433         if (block < direct_blks) {
 434                 offset[n++] = NODE_DIR2_BLOCK;
 435                 noffset[n] = 2;
 436                 offset[n] = block;
 437                 level = 1;
 438                 goto got;
 439         }
 440         block -= direct_blks;
 441         if (block < indirect_blks) {
 442                 offset[n++] = NODE_IND1_BLOCK;
 443                 noffset[n] = 3;
 444                 offset[n++] = block / direct_blks;
 445                 noffset[n] = 4 + offset[n - 1];
 446                 offset[n] = block % direct_blks;
 447                 level = 2;
 448                 goto got;
 449         }
 450         block -= indirect_blks;
 451         if (block < indirect_blks) {
 452                 offset[n++] = NODE_IND2_BLOCK;
 453                 noffset[n] = 4 + dptrs_per_blk;
 454                 offset[n++] = block / direct_blks;
 455                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 456                 offset[n] = block % direct_blks;
 457                 level = 2;
 458                 goto got;
 459         }
 460         block -= indirect_blks;
 461         if (block < dindirect_blks) {
 462                 offset[n++] = NODE_DIND_BLOCK;
 463                 noffset[n] = 5 + (dptrs_per_blk * 2);
 464                 offset[n++] = block / indirect_blks;
 465                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 466                               offset[n - 1] * (dptrs_per_blk + 1);
 467                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 468                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 469                               offset[n - 2] * (dptrs_per_blk + 1) +
 470                               offset[n - 1];
 471                 offset[n] = block % direct_blks;
 472                 level = 3;
 473                 goto got;
 474         } else {
 475                 BUG();
 476         }
 477 got:
 478         return level;
 479 }
 480
 481 /*
 482  * Caller should call f2fs_put_dnode(dn).
 483  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 484  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 485  * In the case of RDONLY_NODE, we don't need to care about mutex.
 486  */
 487 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 488 {
 489         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 490         struct page *npage[4];
 491         struct page *parent;
 492         int offset[4];
 493         unsigned int noffset[4];
 494         nid_t nids[4];
 495         int level, i;
 496         int err = 0;
 497
 498         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 499
 500         nids[0] = dn->inode->i_ino;
 501         npage[0] = dn->inode_page;
 502
 503         if (!npage[0]) {
 504                 npage[0] = get_node_page(sbi, nids[0]);
 505                 if (IS_ERR(npage[0]))
 506                         return PTR_ERR(npage[0]);
 507         }
 508         parent = npage[0];
 509         if (level != 0)
 510                 nids[1] = get_nid(parent, offset[0], true);
 511         dn->inode_page = npage[0];
 512         dn->inode_page_locked = true;
 513
 514         /* get indirect or direct nodes */
 515         for (i = 1; i <= level; i++) {
 516                 bool done = false;
 517
 518                 if (!nids[i] && mode == ALLOC_NODE) {
 519                         /* alloc new node */
 520                         if (!alloc_nid(sbi, &(nids[i]))) {
 521                                 err = -ENOSPC;
 522                                 goto release_pages;
 523                         }
 524
 525                         dn->nid = nids[i];
 526                         npage[i] = new_node_page(dn, noffset[i], NULL);
 527                         if (IS_ERR(npage[i])) {
 528                                 alloc_nid_failed(sbi, nids[i]);
 529                                 err = PTR_ERR(npage[i]);
 530                                 goto release_pages;
 531                         }
 532
 533                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 534                         alloc_nid_done(sbi, nids[i]);
 535                         done = true;
 536                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 537                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 538                         if (IS_ERR(npage[i])) {
 539                                 err = PTR_ERR(npage[i]);
 540                                 goto release_pages;
 541                         }
 542                         done = true;
 543                 }
 544                 if (i == 1) {
 545                         dn->inode_page_locked = false;
 546                         unlock_page(parent);
 547                 } else {
 548                         f2fs_put_page(parent, 1);
 549                 }
 550
 551                 if (!done) {
 552                         npage[i] = get_node_page(sbi, nids[i]);
 553                         if (IS_ERR(npage[i])) {
 554                                 err = PTR_ERR(npage[i]);
 555                                 f2fs_put_page(npage[0], 0);
 556                                 goto release_out;
 557                         }
 558                 }
 559                 if (i < level) {
 560                         parent = npage[i];
 561                         nids[i + 1] = get_nid(parent, offset[i], false);
 562                 }
 563         }
 564         dn->nid = nids[level];
 565         dn->ofs_in_node = offset[level];
 566         dn->node_page = npage[level];
 567         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 568         return 0;
 569
 570 release_pages:
 571         f2fs_put_page(parent, 1);
 572         if (i > 1)
 573                 f2fs_put_page(npage[0], 0);
 574 release_out:
 575         dn->inode_page = NULL;
 576         dn->node_page = NULL;
 577         return err;
 578 }
 579
 580 static void truncate_node(struct dnode_of_data *dn)
 581 {
 582         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 583         struct node_info ni;
 584
 585         get_node_info(sbi, dn->nid, &ni);
 586         if (dn->inode->i_blocks == 0) {
 587                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 588                 goto invalidate;
 589         }
 590         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 591
 592         /* Deallocate node address */
 593         invalidate_blocks(sbi, ni.blk_addr);
 594         dec_valid_node_count(sbi, dn->inode);
 595         set_node_addr(sbi, &ni, NULL_ADDR, false);
 596
 597         if (dn->nid == dn->inode->i_ino) {
 598                 remove_orphan_inode(sbi, dn->nid);
 599                 dec_valid_inode_count(sbi);
 600         } else {
 601                 sync_inode_page(dn);
 602         }
 603 invalidate:
 604         clear_node_page_dirty(dn->node_page);
 605         F2FS_SET_SB_DIRT(sbi);
 606
 607         f2fs_put_page(dn->node_page, 1);
 608
 609         invalidate_mapping_pages(NODE_MAPPING(sbi),
 610                         dn->node_page->index, dn->node_page->index);
 611
 612         dn->node_page = NULL;
 613         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 614 }
 615
 616 static int truncate_dnode(struct dnode_of_data *dn)
 617 {
 618         struct page *page;
 619
 620         if (dn->nid == 0)
 621                 return 1;
 622
 623         /* get direct node */
 624         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 625         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 626                 return 1;
 627         else if (IS_ERR(page))
 628                 return PTR_ERR(page);
 629
 630         /* Make dnode_of_data for parameter */
 631         dn->node_page = page;
 632         dn->ofs_in_node = 0;
 633         truncate_data_blocks(dn);
 634         truncate_node(dn);
 635         return 1;
 636 }
 637
 638 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 639                                                 int ofs, int depth)
 640 {
 641         struct dnode_of_data rdn = *dn;
 642         struct page *page;
 643         struct f2fs_node *rn;
 644         nid_t child_nid;
 645         unsigned int child_nofs;
 646         int freed = 0;
 647         int i, ret;
 648
 649         if (dn->nid == 0)
 650                 return NIDS_PER_BLOCK + 1;
 651
 652         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 653
 654         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 655         if (IS_ERR(page)) {
 656                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 657                 return PTR_ERR(page);
 658         }
 659
 660         rn = F2FS_NODE(page);
 661         if (depth < 3) {
 662                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 663                         child_nid = le32_to_cpu(rn->in.nid[i]);
 664                         if (child_nid == 0)
 665                                 continue;
 666                         rdn.nid = child_nid;
 667                         ret = truncate_dnode(&rdn);
 668                         if (ret < 0)
 669                                 goto out_err;
 670                         set_nid(page, i, 0, false);
 671                 }
 672         } else {
 673                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 674                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 675                         child_nid = le32_to_cpu(rn->in.nid[i]);
 676                         if (child_nid == 0) {
 677                                 child_nofs += NIDS_PER_BLOCK + 1;
 678                                 continue;
 679                         }
 680                         rdn.nid = child_nid;
 681                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 682                         if (ret == (NIDS_PER_BLOCK + 1)) {
 683                                 set_nid(page, i, 0, false);
 684                                 child_nofs += ret;
 685                         } else if (ret < 0 && ret != -ENOENT) {
 686                                 goto out_err;
 687                         }
 688                 }
 689                 freed = child_nofs;
 690         }
 691
 692         if (!ofs) {
 693                 /* remove current indirect node */
 694                 dn->node_page = page;
 695                 truncate_node(dn);
 696                 freed++;
 697         } else {
 698                 f2fs_put_page(page, 1);
 699         }
 700         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 701         return freed;
 702
 703 out_err:
 704         f2fs_put_page(page, 1);
 705         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 706         return ret;
 707 }
 708
 709 static int truncate_partial_nodes(struct dnode_of_data *dn,
 710                         struct f2fs_inode *ri, int *offset, int depth)
 711 {
 712         struct page *pages[2];
 713         nid_t nid[3];
 714         nid_t child_nid;
 715         int err = 0;
 716         int i;
 717         int idx = depth - 2;
 718
 719         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 720         if (!nid[0])
 721                 return 0;
 722
 723         /* get indirect nodes in the path */
 724         for (i = 0; i < idx + 1; i++) {
 725                 /* reference count'll be increased */
 726                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 727                 if (IS_ERR(pages[i])) {
 728                         err = PTR_ERR(pages[i]);
 729                         idx = i - 1;
 730                         goto fail;
 731                 }
 732                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 733         }
 734
 735         /* free direct nodes linked to a partial indirect node */
 736         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 737                 child_nid = get_nid(pages[idx], i, false);
 738                 if (!child_nid)
 739                         continue;
 740                 dn->nid = child_nid;
 741                 err = truncate_dnode(dn);
 742                 if (err < 0)
 743                         goto fail;
 744                 set_nid(pages[idx], i, 0, false);
 745         }
 746
 747         if (offset[idx + 1] == 0) {
 748                 dn->node_page = pages[idx];
 749                 dn->nid = nid[idx];
 750                 truncate_node(dn);
 751         } else {
 752                 f2fs_put_page(pages[idx], 1);
 753         }
 754         offset[idx]++;
 755         offset[idx + 1] = 0;
 756         idx--;
 757 fail:
 758         for (i = idx; i >= 0; i--)
 759                 f2fs_put_page(pages[i], 1);
 760
 761         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 762
 763         return err;
 764 }
 765
 766 /*
 767  * All the block addresses of data and nodes should be nullified.
 768  */
 769 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 770 {
 771         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 772         int err = 0, cont = 1;
 773         int level, offset[4], noffset[4];
 774         unsigned int nofs = 0;
 775         struct f2fs_inode *ri;
 776         struct dnode_of_data dn;
 777         struct page *page;
 778
 779         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 780
 781         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 782 restart:
 783         page = get_node_page(sbi, inode->i_ino);
 784         if (IS_ERR(page)) {
 785                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 786                 return PTR_ERR(page);
 787         }
 788
 789         set_new_dnode(&dn, inode, page, NULL, 0);
 790         unlock_page(page);
 791
 792         ri = F2FS_INODE(page);
 793         switch (level) {
 794         case 0:
 795         case 1:
 796                 nofs = noffset[1];
 797                 break;
 798         case 2:
 799                 nofs = noffset[1];
 800                 if (!offset[level - 1])
 801                         goto skip_partial;
 802                 err = truncate_partial_nodes(&dn, ri, offset, level);
 803                 if (err < 0 && err != -ENOENT)
 804                         goto fail;
 805                 nofs += 1 + NIDS_PER_BLOCK;
 806                 break;
 807         case 3:
 808                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 809                 if (!offset[level - 1])
 810                         goto skip_partial;
 811                 err = truncate_partial_nodes(&dn, ri, offset, level);
 812                 if (err < 0 && err != -ENOENT)
 813                         goto fail;
 814                 break;
 815         default:
 816                 BUG();
 817         }
 818
 819 skip_partial:
 820         while (cont) {
 821                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 822                 switch (offset[0]) {
 823                 case NODE_DIR1_BLOCK:
 824                 case NODE_DIR2_BLOCK:
 825                         err = truncate_dnode(&dn);
 826                         break;
 827
 828                 case NODE_IND1_BLOCK:
 829                 case NODE_IND2_BLOCK:
 830                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 831                         break;
 832
 833                 case NODE_DIND_BLOCK:
 834                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 835                         cont = 0;
 836                         break;
 837
 838                 default:
 839                         BUG();
 840                 }
 841                 if (err < 0 && err != -ENOENT)
 842                         goto fail;
 843                 if (offset[1] == 0 &&
 844                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 845                         lock_page(page);
 846                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 847                                 f2fs_put_page(page, 1);
 848                                 goto restart;
 849                         }
 850                         f2fs_wait_on_page_writeback(page, NODE);
 851                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 852                         set_page_dirty(page);
 853                         unlock_page(page);
 854                 }
 855                 offset[1] = 0;
 856                 offset[0]++;
 857                 nofs += err;
 858         }
 859 fail:
 860         f2fs_put_page(page, 0);
 861         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 862         return err > 0 ? 0 : err;
 863 }
 864
 865 int truncate_xattr_node(struct inode *inode, struct page *page)
 866 {
 867         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 868         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 869         struct dnode_of_data dn;
 870         struct page *npage;
 871
 872         if (!nid)
 873                 return 0;
 874
 875         npage = get_node_page(sbi, nid);
 876         if (IS_ERR(npage))
 877                 return PTR_ERR(npage);
 878
 879         F2FS_I(inode)->i_xattr_nid = 0;
 880
 881         /* need to do checkpoint during fsync */
 882         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 883
 884         set_new_dnode(&dn, inode, page, npage, nid);
 885
 886         if (page)
 887                 dn.inode_page_locked = true;
 888         truncate_node(&dn);
 889         return 0;
 890 }
 891
 892 /*
 893  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 894  * f2fs_unlock_op().
 895  */
 896 void remove_inode_page(struct inode *inode)
 897 {
 898         struct dnode_of_data dn;
 899
 900         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 901         if (get_dnode_of_data(&dn, 0, LOOKUP_NODE))
 902                 return;
 903
 904         if (truncate_xattr_node(inode, dn.inode_page)) {
 905                 f2fs_put_dnode(&dn);
 906                 return;
 907         }
 908
 909         /* remove potential inline_data blocks */
 910         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 911                                 S_ISLNK(inode->i_mode))
 912                 truncate_data_blocks_range(&dn, 1);
 913
 914         /* 0 is possible, after f2fs_new_inode() has failed */
 915         f2fs_bug_on(F2FS_I_SB(inode),
 916                         inode->i_blocks != 0 && inode->i_blocks != 1);
 917
 918         /* will put inode & node pages */
 919         truncate_node(&dn);
 920 }
 921
 922 struct page *new_inode_page(struct inode *inode)
 923 {
 924         struct dnode_of_data dn;
 925
 926         /* allocate inode page for new inode */
 927         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 928
 929         /* caller should f2fs_put_page(page, 1); */
 930         return new_node_page(&dn, 0, NULL);
 931 }
 932
 933 struct page *new_node_page(struct dnode_of_data *dn,
 934                                 unsigned int ofs, struct page *ipage)
 935 {
 936         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 937         struct node_info old_ni, new_ni;
 938         struct page *page;
 939         int err;
 940
 941         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 942                 return ERR_PTR(-EPERM);
 943
 944         page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
 945         if (!page)
 946                 return ERR_PTR(-ENOMEM);
 947
 948         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 949                 err = -ENOSPC;
 950                 goto fail;
 951         }
 952
 953         get_node_info(sbi, dn->nid, &old_ni);
 954
 955         /* Reinitialize old_ni with new node page */
 956         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
 957         new_ni = old_ni;
 958         new_ni.ino = dn->inode->i_ino;
 959         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
 960
 961         f2fs_wait_on_page_writeback(page, NODE);
 962         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 963         set_cold_node(dn->inode, page);
 964         SetPageUptodate(page);
 965         set_page_dirty(page);
 966
 967         if (f2fs_has_xattr_block(ofs))
 968                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 969
 970         dn->node_page = page;
 971         if (ipage)
 972                 update_inode(dn->inode, ipage);
 973         else
 974                 sync_inode_page(dn);
 975         if (ofs == 0)
 976                 inc_valid_inode_count(sbi);
 977
 978         return page;
 979
 980 fail:
 981         clear_node_page_dirty(page);
 982         f2fs_put_page(page, 1);
 983         return ERR_PTR(err);
 984 }
 985
 986 /*
 987  * Caller should do after getting the following values.
 988  * 0: f2fs_put_page(page, 0)
 989  * LOCKED_PAGE: f2fs_put_page(page, 1)
 990  * error: nothing
 991  */
 992 static int read_node_page(struct page *page, int rw)
 993 {
 994         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 995         struct node_info ni;
 996
 997         get_node_info(sbi, page->index, &ni);
 998
 999         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1000                 f2fs_put_page(page, 1);
1001                 return -ENOENT;
1002         }
1003
1004         if (PageUptodate(page))
1005                 return LOCKED_PAGE;
1006
1007         return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
1008 }
1009
1010 /*
1011  * Readahead a node page
1012  */
1013 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1014 {
1015         struct page *apage;
1016         int err;
1017
1018         apage = find_get_page(NODE_MAPPING(sbi), nid);
1019         if (apage && PageUptodate(apage)) {
1020                 f2fs_put_page(apage, 0);
1021                 return;
1022         }
1023         f2fs_put_page(apage, 0);
1024
1025         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1026         if (!apage)
1027                 return;
1028
1029         err = read_node_page(apage, READA);
1030         if (err == 0)
1031                 f2fs_put_page(apage, 0);
1032         else if (err == LOCKED_PAGE)
1033                 f2fs_put_page(apage, 1);
1034 }
1035
1036 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1037 {
1038         struct page *page;
1039         int err;
1040 repeat:
1041         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1042         if (!page)
1043                 return ERR_PTR(-ENOMEM);
1044
1045         err = read_node_page(page, READ_SYNC);
1046         if (err < 0)
1047                 return ERR_PTR(err);
1048         else if (err == LOCKED_PAGE)
1049                 goto got_it;
1050
1051         lock_page(page);
1052         if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
1053                 f2fs_put_page(page, 1);
1054                 return ERR_PTR(-EIO);
1055         }
1056         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1057                 f2fs_put_page(page, 1);
1058                 goto repeat;
1059         }
1060 got_it:
1061         return page;
1062 }
1063
1064 /*
1065  * Return a locked page for the desired node page.
1066  * And, readahead MAX_RA_NODE number of node pages.
1067  */
1068 struct page *get_node_page_ra(struct page *parent, int start)
1069 {
1070         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1071         struct blk_plug plug;
1072         struct page *page;
1073         int err, i, end;
1074         nid_t nid;
1075
1076         /* First, try getting the desired direct node. */
1077         nid = get_nid(parent, start, false);
1078         if (!nid)
1079                 return ERR_PTR(-ENOENT);
1080 repeat:
1081         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1082         if (!page)
1083                 return ERR_PTR(-ENOMEM);
1084
1085         err = read_node_page(page, READ_SYNC);
1086         if (err < 0)
1087                 return ERR_PTR(err);
1088         else if (err == LOCKED_PAGE)
1089                 goto page_hit;
1090
1091         blk_start_plug(&plug);
1092
1093         /* Then, try readahead for siblings of the desired node */
1094         end = start + MAX_RA_NODE;
1095         end = min(end, NIDS_PER_BLOCK);
1096         for (i = start + 1; i < end; i++) {
1097                 nid = get_nid(parent, i, false);
1098                 if (!nid)
1099                         continue;
1100                 ra_node_page(sbi, nid);
1101         }
1102
1103         blk_finish_plug(&plug);
1104
1105         lock_page(page);
1106         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1107                 f2fs_put_page(page, 1);
1108                 goto repeat;
1109         }
1110 page_hit:
1111         if (unlikely(!PageUptodate(page))) {
1112                 f2fs_put_page(page, 1);
1113                 return ERR_PTR(-EIO);
1114         }
1115         return page;
1116 }
1117
1118 void sync_inode_page(struct dnode_of_data *dn)
1119 {
1120         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1121                 update_inode(dn->inode, dn->node_page);
1122         } else if (dn->inode_page) {
1123                 if (!dn->inode_page_locked)
1124                         lock_page(dn->inode_page);
1125                 update_inode(dn->inode, dn->inode_page);
1126                 if (!dn->inode_page_locked)
1127                         unlock_page(dn->inode_page);
1128         } else {
1129                 update_inode_page(dn->inode);
1130         }
1131 }
1132
1133 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1134                                         struct writeback_control *wbc)
1135 {
1136         pgoff_t index, end;
1137         struct pagevec pvec;
1138         int step = ino ? 2 : 0;
1139         int nwritten = 0, wrote = 0;
1140
1141         pagevec_init(&pvec, 0);
1142
1143 next_step:
1144         index = 0;
1145         end = LONG_MAX;
1146
1147         while (index <= end) {
1148                 int i, nr_pages;
1149                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1150                                 PAGECACHE_TAG_DIRTY,
1151                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1152                 if (nr_pages == 0)
1153                         break;
1154
1155                 for (i = 0; i < nr_pages; i++) {
1156                         struct page *page = pvec.pages[i];
1157
1158                         /*
1159                          * flushing sequence with step:
1160                          * 0. indirect nodes
1161                          * 1. dentry dnodes
1162                          * 2. file dnodes
1163                          */
1164                         if (step == 0 && IS_DNODE(page))
1165                                 continue;
1166                         if (step == 1 && (!IS_DNODE(page) ||
1167                                                 is_cold_node(page)))
1168                                 continue;
1169                         if (step == 2 && (!IS_DNODE(page) ||
1170                                                 !is_cold_node(page)))
1171                                 continue;
1172
1173                         /*
1174                          * If an fsync mode,
1175                          * we should not skip writing node pages.
1176                          */
1177                         if (ino && ino_of_node(page) == ino)
1178                                 lock_page(page);
1179                         else if (!trylock_page(page))
1180                                 continue;
1181
1182                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1183 continue_unlock:
1184                                 unlock_page(page);
1185                                 continue;
1186                         }
1187                         if (ino && ino_of_node(page) != ino)
1188                                 goto continue_unlock;
1189
1190                         if (!PageDirty(page)) {
1191                                 /* someone wrote it for us */
1192                                 goto continue_unlock;
1193                         }
1194
1195                         if (!clear_page_dirty_for_io(page))
1196                                 goto continue_unlock;
1197
1198                         /* called by fsync() */
1199                         if (ino && IS_DNODE(page)) {
1200                                 set_fsync_mark(page, 1);
1201                                 if (IS_INODE(page)) {
1202                                         if (!is_checkpointed_node(sbi, ino) &&
1203                                                 !has_fsynced_inode(sbi, ino))
1204                                                 set_dentry_mark(page, 1);
1205                                         else
1206                                                 set_dentry_mark(page, 0);
1207                                 }
1208                                 nwritten++;
1209                         } else {
1210                                 set_fsync_mark(page, 0);
1211                                 set_dentry_mark(page, 0);
1212                         }
1213
1214                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1215                                 unlock_page(page);
1216                         else
1217                                 wrote++;
1218
1219                         if (--wbc->nr_to_write == 0)
1220                                 break;
1221                 }
1222                 pagevec_release(&pvec);
1223                 cond_resched();
1224
1225                 if (wbc->nr_to_write == 0) {
1226                         step = 2;
1227                         break;
1228                 }
1229         }
1230
1231         if (step < 2) {
1232                 step++;
1233                 goto next_step;
1234         }
1235
1236         if (wrote)
1237                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1238         return nwritten;
1239 }
1240
1241 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1242 {
1243         pgoff_t index = 0, end = LONG_MAX;
1244         struct pagevec pvec;
1245         int ret2 = 0, ret = 0;
1246
1247         pagevec_init(&pvec, 0);
1248
1249         while (index <= end) {
1250                 int i, nr_pages;
1251                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1252                                 PAGECACHE_TAG_WRITEBACK,
1253                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1254                 if (nr_pages == 0)
1255                         break;
1256
1257                 for (i = 0; i < nr_pages; i++) {
1258                         struct page *page = pvec.pages[i];
1259
1260                         /* until radix tree lookup accepts end_index */
1261                         if (unlikely(page->index > end))
1262                                 continue;
1263
1264                         if (ino && ino_of_node(page) == ino) {
1265                                 f2fs_wait_on_page_writeback(page, NODE);
1266                                 if (TestClearPageError(page))
1267                                         ret = -EIO;
1268                         }
1269                 }
1270                 pagevec_release(&pvec);
1271                 cond_resched();
1272         }
1273
1274         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1275                 ret2 = -ENOSPC;
1276         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1277                 ret2 = -EIO;
1278         if (!ret)
1279                 ret = ret2;
1280         return ret;
1281 }
1282
1283 static int f2fs_write_node_page(struct page *page,
1284                                 struct writeback_control *wbc)
1285 {
1286         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1287         nid_t nid;
1288         block_t new_addr;
1289         struct node_info ni;
1290         struct f2fs_io_info fio = {
1291                 .type = NODE,
1292                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1293         };
1294
1295         trace_f2fs_writepage(page, NODE);
1296
1297         if (unlikely(sbi->por_doing))
1298                 goto redirty_out;
1299         if (unlikely(f2fs_cp_error(sbi)))
1300                 goto redirty_out;
1301
1302         f2fs_wait_on_page_writeback(page, NODE);
1303
1304         /* get old block addr of this node page */
1305         nid = nid_of_node(page);
1306         f2fs_bug_on(sbi, page->index != nid);
1307
1308         get_node_info(sbi, nid, &ni);
1309
1310         /* This page is already truncated */
1311         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1312                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1313                 unlock_page(page);
1314                 return 0;
1315         }
1316
1317         if (wbc->for_reclaim) {
1318                 if (!down_read_trylock(&sbi->node_write))
1319                         goto redirty_out;
1320         } else {
1321                 down_read(&sbi->node_write);
1322         }
1323         set_page_writeback(page);
1324         write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1325         set_node_addr(sbi, &ni, new_addr, is_fsync_dnode(page));
1326         dec_page_count(sbi, F2FS_DIRTY_NODES);
1327         up_read(&sbi->node_write);
1328         unlock_page(page);
1329
1330         if (wbc->for_reclaim)
1331                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1332
1333         return 0;
1334
1335 redirty_out:
1336         redirty_page_for_writepage(wbc, page);
1337         return AOP_WRITEPAGE_ACTIVATE;
1338 }
1339
1340 static int f2fs_write_node_pages(struct address_space *mapping,
1341                             struct writeback_control *wbc)
1342 {
1343         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1344         long diff;
1345
1346         trace_f2fs_writepages(mapping->host, wbc, NODE);
1347
1348         /* balancing f2fs's metadata in background */
1349         f2fs_balance_fs_bg(sbi);
1350
1351         /* collect a number of dirty node pages and write together */
1352         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1353                 goto skip_write;
1354
1355         diff = nr_pages_to_write(sbi, NODE, wbc);
1356         wbc->sync_mode = WB_SYNC_NONE;
1357         sync_node_pages(sbi, 0, wbc);
1358         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1359         return 0;
1360
1361 skip_write:
1362         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1363         return 0;
1364 }
1365
1366 static int f2fs_set_node_page_dirty(struct page *page)
1367 {
1368         trace_f2fs_set_page_dirty(page, NODE);
1369
1370         SetPageUptodate(page);
1371         if (!PageDirty(page)) {
1372                 __set_page_dirty_nobuffers(page);
1373                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1374                 SetPagePrivate(page);
1375                 return 1;
1376         }
1377         return 0;
1378 }
1379
1380 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1381                                       unsigned int length)
1382 {
1383         struct inode *inode = page->mapping->host;
1384         if (PageDirty(page))
1385                 dec_page_count(F2FS_I_SB(inode), F2FS_DIRTY_NODES);
1386         ClearPagePrivate(page);
1387 }
1388
1389 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1390 {
1391         ClearPagePrivate(page);
1392         return 1;
1393 }
1394
1395 /*
1396  * Structure of the f2fs node operations
1397  */
1398 const struct address_space_operations f2fs_node_aops = {
1399         .writepage      = f2fs_write_node_page,
1400         .writepages     = f2fs_write_node_pages,
1401         .set_page_dirty = f2fs_set_node_page_dirty,
1402         .invalidatepage = f2fs_invalidate_node_page,
1403         .releasepage    = f2fs_release_node_page,
1404 };
1405
1406 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1407                                                 nid_t n)
1408 {
1409         return radix_tree_lookup(&nm_i->free_nid_root, n);
1410 }
1411
1412 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1413                                                 struct free_nid *i)
1414 {
1415         list_del(&i->list);
1416         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1417 }
1418
1419 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1420 {
1421         struct f2fs_nm_info *nm_i = NM_I(sbi);
1422         struct free_nid *i;
1423         struct nat_entry *ne;
1424         bool allocated = false;
1425
1426         if (!available_free_memory(sbi, FREE_NIDS))
1427                 return -1;
1428
1429         /* 0 nid should not be used */
1430         if (unlikely(nid == 0))
1431                 return 0;
1432
1433         if (build) {
1434                 /* do not add allocated nids */
1435                 read_lock(&nm_i->nat_tree_lock);
1436                 ne = __lookup_nat_cache(nm_i, nid);
1437                 if (ne &&
1438                         (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1439                                 nat_get_blkaddr(ne) != NULL_ADDR))
1440                         allocated = true;
1441                 read_unlock(&nm_i->nat_tree_lock);
1442                 if (allocated)
1443                         return 0;
1444         }
1445
1446         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1447         i->nid = nid;
1448         i->state = NID_NEW;
1449
1450         spin_lock(&nm_i->free_nid_list_lock);
1451         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1452                 spin_unlock(&nm_i->free_nid_list_lock);
1453                 kmem_cache_free(free_nid_slab, i);
1454                 return 0;
1455         }
1456         list_add_tail(&i->list, &nm_i->free_nid_list);
1457         nm_i->fcnt++;
1458         spin_unlock(&nm_i->free_nid_list_lock);
1459         return 1;
1460 }
1461
1462 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1463 {
1464         struct free_nid *i;
1465         bool need_free = false;
1466
1467         spin_lock(&nm_i->free_nid_list_lock);
1468         i = __lookup_free_nid_list(nm_i, nid);
1469         if (i && i->state == NID_NEW) {
1470                 __del_from_free_nid_list(nm_i, i);
1471                 nm_i->fcnt--;
1472                 need_free = true;
1473         }
1474         spin_unlock(&nm_i->free_nid_list_lock);
1475
1476         if (need_free)
1477                 kmem_cache_free(free_nid_slab, i);
1478 }
1479
1480 static void scan_nat_page(struct f2fs_sb_info *sbi,
1481                         struct page *nat_page, nid_t start_nid)
1482 {
1483         struct f2fs_nm_info *nm_i = NM_I(sbi);
1484         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1485         block_t blk_addr;
1486         int i;
1487
1488         i = start_nid % NAT_ENTRY_PER_BLOCK;
1489
1490         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1491
1492                 if (unlikely(start_nid >= nm_i->max_nid))
1493                         break;
1494
1495                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1496                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1497                 if (blk_addr == NULL_ADDR) {
1498                         if (add_free_nid(sbi, start_nid, true) < 0)
1499                                 break;
1500                 }
1501         }
1502 }
1503
1504 static void build_free_nids(struct f2fs_sb_info *sbi)
1505 {
1506         struct f2fs_nm_info *nm_i = NM_I(sbi);
1507         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1508         struct f2fs_summary_block *sum = curseg->sum_blk;
1509         int i = 0;
1510         nid_t nid = nm_i->next_scan_nid;
1511
1512         /* Enough entries */
1513         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1514                 return;
1515
1516         /* readahead nat pages to be scanned */
1517         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1518
1519         while (1) {
1520                 struct page *page = get_current_nat_page(sbi, nid);
1521
1522                 scan_nat_page(sbi, page, nid);
1523                 f2fs_put_page(page, 1);
1524
1525                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1526                 if (unlikely(nid >= nm_i->max_nid))
1527                         nid = 0;
1528
1529                 if (i++ == FREE_NID_PAGES)
1530                         break;
1531         }
1532
1533         /* go to the next free nat pages to find free nids abundantly */
1534         nm_i->next_scan_nid = nid;
1535
1536         /* find free nids from current sum_pages */
1537         mutex_lock(&curseg->curseg_mutex);
1538         for (i = 0; i < nats_in_cursum(sum); i++) {
1539                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1540                 nid = le32_to_cpu(nid_in_journal(sum, i));
1541                 if (addr == NULL_ADDR)
1542                         add_free_nid(sbi, nid, true);
1543                 else
1544                         remove_free_nid(nm_i, nid);
1545         }
1546         mutex_unlock(&curseg->curseg_mutex);
1547 }
1548
1549 /*
1550  * If this function returns success, caller can obtain a new nid
1551  * from second parameter of this function.
1552  * The returned nid could be used ino as well as nid when inode is created.
1553  */
1554 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1555 {
1556         struct f2fs_nm_info *nm_i = NM_I(sbi);
1557         struct free_nid *i = NULL;
1558 retry:
1559         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1560                 return false;
1561
1562         spin_lock(&nm_i->free_nid_list_lock);
1563
1564         /* We should not use stale free nids created by build_free_nids */
1565         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1566                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1567                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1568                         if (i->state == NID_NEW)
1569                                 break;
1570
1571                 f2fs_bug_on(sbi, i->state != NID_NEW);
1572                 *nid = i->nid;
1573                 i->state = NID_ALLOC;
1574                 nm_i->fcnt--;
1575                 spin_unlock(&nm_i->free_nid_list_lock);
1576                 return true;
1577         }
1578         spin_unlock(&nm_i->free_nid_list_lock);
1579
1580         /* Let's scan nat pages and its caches to get free nids */
1581         mutex_lock(&nm_i->build_lock);
1582         build_free_nids(sbi);
1583         mutex_unlock(&nm_i->build_lock);
1584         goto retry;
1585 }
1586
1587 /*
1588  * alloc_nid() should be called prior to this function.
1589  */
1590 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1591 {
1592         struct f2fs_nm_info *nm_i = NM_I(sbi);
1593         struct free_nid *i;
1594
1595         spin_lock(&nm_i->free_nid_list_lock);
1596         i = __lookup_free_nid_list(nm_i, nid);
1597         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1598         __del_from_free_nid_list(nm_i, i);
1599         spin_unlock(&nm_i->free_nid_list_lock);
1600
1601         kmem_cache_free(free_nid_slab, i);
1602 }
1603
1604 /*
1605  * alloc_nid() should be called prior to this function.
1606  */
1607 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1608 {
1609         struct f2fs_nm_info *nm_i = NM_I(sbi);
1610         struct free_nid *i;
1611         bool need_free = false;
1612
1613         if (!nid)
1614                 return;
1615
1616         spin_lock(&nm_i->free_nid_list_lock);
1617         i = __lookup_free_nid_list(nm_i, nid);
1618         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1619         if (!available_free_memory(sbi, FREE_NIDS)) {
1620                 __del_from_free_nid_list(nm_i, i);
1621                 need_free = true;
1622         } else {
1623                 i->state = NID_NEW;
1624                 nm_i->fcnt++;
1625         }
1626         spin_unlock(&nm_i->free_nid_list_lock);
1627
1628         if (need_free)
1629                 kmem_cache_free(free_nid_slab, i);
1630 }
1631
1632 void recover_inline_xattr(struct inode *inode, struct page *page)
1633 {
1634         void *src_addr, *dst_addr;
1635         size_t inline_size;
1636         struct page *ipage;
1637         struct f2fs_inode *ri;
1638
1639         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1640         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1641
1642         ri = F2FS_INODE(page);
1643         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1644                 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1645                 goto update_inode;
1646         }
1647
1648         dst_addr = inline_xattr_addr(ipage);
1649         src_addr = inline_xattr_addr(page);
1650         inline_size = inline_xattr_size(inode);
1651
1652         f2fs_wait_on_page_writeback(ipage, NODE);
1653         memcpy(dst_addr, src_addr, inline_size);
1654 update_inode:
1655         update_inode(inode, ipage);
1656         f2fs_put_page(ipage, 1);
1657 }
1658
1659 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1660 {
1661         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1662         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1663         nid_t new_xnid = nid_of_node(page);
1664         struct node_info ni;
1665
1666         /* 1: invalidate the previous xattr nid */
1667         if (!prev_xnid)
1668                 goto recover_xnid;
1669
1670         /* Deallocate node address */
1671         get_node_info(sbi, prev_xnid, &ni);
1672         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1673         invalidate_blocks(sbi, ni.blk_addr);
1674         dec_valid_node_count(sbi, inode);
1675         set_node_addr(sbi, &ni, NULL_ADDR, false);
1676
1677 recover_xnid:
1678         /* 2: allocate new xattr nid */
1679         if (unlikely(!inc_valid_node_count(sbi, inode)))
1680                 f2fs_bug_on(sbi, 1);
1681
1682         remove_free_nid(NM_I(sbi), new_xnid);
1683         get_node_info(sbi, new_xnid, &ni);
1684         ni.ino = inode->i_ino;
1685         set_node_addr(sbi, &ni, NEW_ADDR, false);
1686         F2FS_I(inode)->i_xattr_nid = new_xnid;
1687
1688         /* 3: update xattr blkaddr */
1689         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1690         set_node_addr(sbi, &ni, blkaddr, false);
1691
1692         update_inode_page(inode);
1693 }
1694
1695 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1696 {
1697         struct f2fs_inode *src, *dst;
1698         nid_t ino = ino_of_node(page);
1699         struct node_info old_ni, new_ni;
1700         struct page *ipage;
1701
1702         get_node_info(sbi, ino, &old_ni);
1703
1704         if (unlikely(old_ni.blk_addr != NULL_ADDR))
1705                 return -EINVAL;
1706
1707         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1708         if (!ipage)
1709                 return -ENOMEM;
1710
1711         /* Should not use this inode from free nid list */
1712         remove_free_nid(NM_I(sbi), ino);
1713
1714         SetPageUptodate(ipage);
1715         fill_node_footer(ipage, ino, ino, 0, true);
1716
1717         src = F2FS_INODE(page);
1718         dst = F2FS_INODE(ipage);
1719
1720         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1721         dst->i_size = 0;
1722         dst->i_blocks = cpu_to_le64(1);
1723         dst->i_links = cpu_to_le32(1);
1724         dst->i_xattr_nid = 0;
1725         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1726
1727         new_ni = old_ni;
1728         new_ni.ino = ino;
1729
1730         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1731                 WARN_ON(1);
1732         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1733         inc_valid_inode_count(sbi);
1734         set_page_dirty(ipage);
1735         f2fs_put_page(ipage, 1);
1736         return 0;
1737 }
1738
1739 /*
1740  * ra_sum_pages() merge contiguous pages into one bio and submit.
1741  * these pre-read pages are allocated in bd_inode's mapping tree.
1742  */
1743 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct page **pages,
1744                                 int start, int nrpages)
1745 {
1746         struct inode *inode = sbi->sb->s_bdev->bd_inode;
1747         struct address_space *mapping = inode->i_mapping;
1748         int i, page_idx = start;
1749         struct f2fs_io_info fio = {
1750                 .type = META,
1751                 .rw = READ_SYNC | REQ_META | REQ_PRIO
1752         };
1753
1754         for (i = 0; page_idx < start + nrpages; page_idx++, i++) {
1755                 /* alloc page in bd_inode for reading node summary info */
1756                 pages[i] = grab_cache_page(mapping, page_idx);
1757                 if (!pages[i])
1758                         break;
1759                 f2fs_submit_page_mbio(sbi, pages[i], page_idx, &fio);
1760         }
1761
1762         f2fs_submit_merged_bio(sbi, META, READ);
1763         return i;
1764 }
1765
1766 int restore_node_summary(struct f2fs_sb_info *sbi,
1767                         unsigned int segno, struct f2fs_summary_block *sum)
1768 {
1769         struct f2fs_node *rn;
1770         struct f2fs_summary *sum_entry;
1771         struct inode *inode = sbi->sb->s_bdev->bd_inode;
1772         block_t addr;
1773         int bio_blocks = MAX_BIO_BLOCKS(sbi);
1774         struct page *pages[bio_blocks];
1775         int i, idx, last_offset, nrpages, err = 0;
1776
1777         /* scan the node segment */
1778         last_offset = sbi->blocks_per_seg;
1779         addr = START_BLOCK(sbi, segno);
1780         sum_entry = &sum->entries[0];
1781
1782         for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) {
1783                 nrpages = min(last_offset - i, bio_blocks);
1784
1785                 /* readahead node pages */
1786                 nrpages = ra_sum_pages(sbi, pages, addr, nrpages);
1787                 if (!nrpages)
1788                         return -ENOMEM;
1789
1790                 for (idx = 0; idx < nrpages; idx++) {
1791                         if (err)
1792                                 goto skip;
1793
1794                         lock_page(pages[idx]);
1795                         if (unlikely(!PageUptodate(pages[idx]))) {
1796                                 err = -EIO;
1797                         } else {
1798                                 rn = F2FS_NODE(pages[idx]);
1799                                 sum_entry->nid = rn->footer.nid;
1800                                 sum_entry->version = 0;
1801                                 sum_entry->ofs_in_node = 0;
1802                                 sum_entry++;
1803                         }
1804                         unlock_page(pages[idx]);
1805 skip:
1806                         page_cache_release(pages[idx]);
1807                 }
1808
1809                 invalidate_mapping_pages(inode->i_mapping, addr,
1810                                                         addr + nrpages);
1811         }
1812         return err;
1813 }
1814
1815 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1816 {
1817         struct f2fs_nm_info *nm_i = NM_I(sbi);
1818         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1819         struct f2fs_summary_block *sum = curseg->sum_blk;
1820         int i;
1821
1822         mutex_lock(&curseg->curseg_mutex);
1823         for (i = 0; i < nats_in_cursum(sum); i++) {
1824                 struct nat_entry *ne;
1825                 struct f2fs_nat_entry raw_ne;
1826                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1827
1828                 raw_ne = nat_in_journal(sum, i);
1829 retry:
1830                 write_lock(&nm_i->nat_tree_lock);
1831                 ne = __lookup_nat_cache(nm_i, nid);
1832                 if (ne)
1833                         goto found;
1834
1835                 ne = grab_nat_entry(nm_i, nid);
1836                 if (!ne) {
1837                         write_unlock(&nm_i->nat_tree_lock);
1838                         goto retry;
1839                 }
1840                 node_info_from_raw_nat(&ne->ni, &raw_ne);
1841 found:
1842                 __set_nat_cache_dirty(nm_i, ne);
1843                 write_unlock(&nm_i->nat_tree_lock);
1844         }
1845         update_nats_in_cursum(sum, -i);
1846         mutex_unlock(&curseg->curseg_mutex);
1847 }
1848
1849 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1850                                                 struct list_head *head, int max)
1851 {
1852         struct nat_entry_set *cur;
1853
1854         if (nes->entry_cnt >= max)
1855                 goto add_out;
1856
1857         list_for_each_entry(cur, head, set_list) {
1858                 if (cur->entry_cnt >= nes->entry_cnt) {
1859                         list_add(&nes->set_list, cur->set_list.prev);
1860                         return;
1861                 }
1862         }
1863 add_out:
1864         list_add_tail(&nes->set_list, head);
1865 }
1866
1867 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1868                                         struct nat_entry_set *set)
1869 {
1870         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1871         struct f2fs_summary_block *sum = curseg->sum_blk;
1872         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1873         bool to_journal = true;
1874         struct f2fs_nat_block *nat_blk;
1875         struct nat_entry *ne, *cur;
1876         struct page *page = NULL;
1877
1878         /*
1879          * there are two steps to flush nat entries:
1880          * #1, flush nat entries to journal in current hot data summary block.
1881          * #2, flush nat entries to nat page.
1882          */
1883         if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1884                 to_journal = false;
1885
1886         if (to_journal) {
1887                 mutex_lock(&curseg->curseg_mutex);
1888         } else {
1889                 page = get_next_nat_page(sbi, start_nid);
1890                 nat_blk = page_address(page);
1891                 f2fs_bug_on(sbi, !nat_blk);
1892         }
1893
1894         /* flush dirty nats in nat entry set */
1895         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1896                 struct f2fs_nat_entry *raw_ne;
1897                 nid_t nid = nat_get_nid(ne);
1898                 int offset;
1899
1900                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1901                         continue;
1902
1903                 if (to_journal) {
1904                         offset = lookup_journal_in_cursum(sum,
1905                                                         NAT_JOURNAL, nid, 1);
1906                         f2fs_bug_on(sbi, offset < 0);
1907                         raw_ne = &nat_in_journal(sum, offset);
1908                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1909                 } else {
1910                         raw_ne = &nat_blk->entries[nid - start_nid];
1911                 }
1912                 raw_nat_from_node_info(raw_ne, &ne->ni);
1913
1914                 write_lock(&NM_I(sbi)->nat_tree_lock);
1915                 nat_reset_flag(ne);
1916                 __clear_nat_cache_dirty(NM_I(sbi), ne);
1917                 write_unlock(&NM_I(sbi)->nat_tree_lock);
1918
1919                 if (nat_get_blkaddr(ne) == NULL_ADDR)
1920                         add_free_nid(sbi, nid, false);
1921         }
1922
1923         if (to_journal)
1924                 mutex_unlock(&curseg->curseg_mutex);
1925         else
1926                 f2fs_put_page(page, 1);
1927
1928         if (!set->entry_cnt) {
1929                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1930                 kmem_cache_free(nat_entry_set_slab, set);
1931         }
1932 }
1933
1934 /*
1935  * This function is called during the checkpointing process.
1936  */
1937 void flush_nat_entries(struct f2fs_sb_info *sbi)
1938 {
1939         struct f2fs_nm_info *nm_i = NM_I(sbi);
1940         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1941         struct f2fs_summary_block *sum = curseg->sum_blk;
1942         struct nat_entry_set *setvec[NATVEC_SIZE];
1943         struct nat_entry_set *set, *tmp;
1944         unsigned int found;
1945         nid_t set_idx = 0;
1946         LIST_HEAD(sets);
1947
1948         /*
1949          * if there are no enough space in journal to store dirty nat
1950          * entries, remove all entries from journal and merge them
1951          * into nat entry set.
1952          */
1953         if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1954                 remove_nats_in_journal(sbi);
1955
1956         if (!nm_i->dirty_nat_cnt)
1957                 return;
1958
1959         while ((found = __gang_lookup_nat_set(nm_i,
1960                                         set_idx, NATVEC_SIZE, setvec))) {
1961                 unsigned idx;
1962                 set_idx = setvec[found - 1]->set + 1;
1963                 for (idx = 0; idx < found; idx++)
1964                         __adjust_nat_entry_set(setvec[idx], &sets,
1965                                                         MAX_NAT_JENTRIES(sum));
1966         }
1967
1968         /* flush dirty nats in nat entry set */
1969         list_for_each_entry_safe(set, tmp, &sets, set_list)
1970                 __flush_nat_entry_set(sbi, set);
1971
1972         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1973 }
1974
1975 static int init_node_manager(struct f2fs_sb_info *sbi)
1976 {
1977         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1978         struct f2fs_nm_info *nm_i = NM_I(sbi);
1979         unsigned char *version_bitmap;
1980         unsigned int nat_segs, nat_blocks;
1981
1982         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1983
1984         /* segment_count_nat includes pair segment so divide to 2. */
1985         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1986         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1987
1988         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1989
1990         /* not used nids: 0, node, meta, (and root counted as valid node) */
1991         nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
1992         nm_i->fcnt = 0;
1993         nm_i->nat_cnt = 0;
1994         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1995
1996         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1997         INIT_LIST_HEAD(&nm_i->free_nid_list);
1998         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1999         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_ATOMIC);
2000         INIT_LIST_HEAD(&nm_i->nat_entries);
2001
2002         mutex_init(&nm_i->build_lock);
2003         spin_lock_init(&nm_i->free_nid_list_lock);
2004         rwlock_init(&nm_i->nat_tree_lock);
2005
2006         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2007         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2008         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2009         if (!version_bitmap)
2010                 return -EFAULT;
2011
2012         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2013                                         GFP_KERNEL);
2014         if (!nm_i->nat_bitmap)
2015                 return -ENOMEM;
2016         return 0;
2017 }
2018
2019 int build_node_manager(struct f2fs_sb_info *sbi)
2020 {
2021         int err;
2022
2023         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2024         if (!sbi->nm_info)
2025                 return -ENOMEM;
2026
2027         err = init_node_manager(sbi);
2028         if (err)
2029                 return err;
2030
2031         build_free_nids(sbi);
2032         return 0;
2033 }
2034
2035 void destroy_node_manager(struct f2fs_sb_info *sbi)
2036 {
2037         struct f2fs_nm_info *nm_i = NM_I(sbi);
2038         struct free_nid *i, *next_i;
2039         struct nat_entry *natvec[NATVEC_SIZE];
2040         nid_t nid = 0;
2041         unsigned int found;
2042
2043         if (!nm_i)
2044                 return;
2045
2046         /* destroy free nid list */
2047         spin_lock(&nm_i->free_nid_list_lock);
2048         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2049                 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2050                 __del_from_free_nid_list(nm_i, i);
2051                 nm_i->fcnt--;
2052                 spin_unlock(&nm_i->free_nid_list_lock);
2053                 kmem_cache_free(free_nid_slab, i);
2054                 spin_lock(&nm_i->free_nid_list_lock);
2055         }
2056         f2fs_bug_on(sbi, nm_i->fcnt);
2057         spin_unlock(&nm_i->free_nid_list_lock);
2058
2059         /* destroy nat cache */
2060         write_lock(&nm_i->nat_tree_lock);
2061         while ((found = __gang_lookup_nat_cache(nm_i,
2062                                         nid, NATVEC_SIZE, natvec))) {
2063                 unsigned idx;
2064                 nid = nat_get_nid(natvec[found - 1]) + 1;
2065                 for (idx = 0; idx < found; idx++)
2066                         __del_from_nat_cache(nm_i, natvec[idx]);
2067         }
2068         f2fs_bug_on(sbi, nm_i->nat_cnt);
2069         write_unlock(&nm_i->nat_tree_lock);
2070
2071         kfree(nm_i->nat_bitmap);
2072         sbi->nm_info = NULL;
2073         kfree(nm_i);
2074 }
2075
2076 int __init create_node_manager_caches(void)
2077 {
2078         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2079                         sizeof(struct nat_entry));
2080         if (!nat_entry_slab)
2081                 goto fail;
2082
2083         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2084                         sizeof(struct free_nid));
2085         if (!free_nid_slab)
2086                 goto destroy_nat_entry;
2087
2088         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2089                         sizeof(struct nat_entry_set));
2090         if (!nat_entry_set_slab)
2091                 goto destroy_free_nid;
2092         return 0;
2093
2094 destroy_free_nid:
2095         kmem_cache_destroy(free_nid_slab);
2096 destroy_nat_entry:
2097         kmem_cache_destroy(nat_entry_slab);
2098 fail:
2099         return -ENOMEM;
2100 }
2101
2102 void destroy_node_manager_caches(void)
2103 {
2104         kmem_cache_destroy(nat_entry_set_slab);
2105         kmem_cache_destroy(free_nid_slab);
2106         kmem_cache_destroy(nat_entry_slab);
2107 }