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
 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                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 160         }
 161         list_move_tail(&ne->list, &head->entry_list);
 162         nm_i->dirty_nat_cnt++;
 163         head->entry_cnt++;
 164         set_nat_flag(ne, IS_DIRTY, true);
 165 }
 166
 167 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 168                                                 struct nat_entry *ne)
 169 {
 170         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 171         struct nat_entry_set *head;
 172
 173         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 174         if (head) {
 175                 list_move_tail(&ne->list, &nm_i->nat_entries);
 176                 set_nat_flag(ne, IS_DIRTY, false);
 177                 head->entry_cnt--;
 178                 nm_i->dirty_nat_cnt--;
 179         }
 180 }
 181
 182 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 183                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 184 {
 185         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 186                                                         start, nr);
 187 }
 188
 189 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 190 {
 191         struct f2fs_nm_info *nm_i = NM_I(sbi);
 192         struct nat_entry *e;
 193         bool is_cp = true;
 194
 195         down_read(&nm_i->nat_tree_lock);
 196         e = __lookup_nat_cache(nm_i, nid);
 197         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 198                 is_cp = false;
 199         up_read(&nm_i->nat_tree_lock);
 200         return is_cp;
 201 }
 202
 203 bool has_fsynced_inode(struct f2fs_sb_info *sbi, nid_t ino)
 204 {
 205         struct f2fs_nm_info *nm_i = NM_I(sbi);
 206         struct nat_entry *e;
 207         bool fsynced = false;
 208
 209         down_read(&nm_i->nat_tree_lock);
 210         e = __lookup_nat_cache(nm_i, ino);
 211         if (e && get_nat_flag(e, HAS_FSYNCED_INODE))
 212                 fsynced = true;
 213         up_read(&nm_i->nat_tree_lock);
 214         return fsynced;
 215 }
 216
 217 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 218 {
 219         struct f2fs_nm_info *nm_i = NM_I(sbi);
 220         struct nat_entry *e;
 221         bool need_update = true;
 222
 223         down_read(&nm_i->nat_tree_lock);
 224         e = __lookup_nat_cache(nm_i, ino);
 225         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 226                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 227                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 228                 need_update = false;
 229         up_read(&nm_i->nat_tree_lock);
 230         return need_update;
 231 }
 232
 233 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 234 {
 235         struct nat_entry *new;
 236
 237         new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 238         f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
 239         memset(new, 0, sizeof(struct nat_entry));
 240         nat_set_nid(new, nid);
 241         nat_reset_flag(new);
 242         list_add_tail(&new->list, &nm_i->nat_entries);
 243         nm_i->nat_cnt++;
 244         return new;
 245 }
 246
 247 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 248                                                 struct f2fs_nat_entry *ne)
 249 {
 250         struct nat_entry *e;
 251
 252         down_write(&nm_i->nat_tree_lock);
 253         e = __lookup_nat_cache(nm_i, nid);
 254         if (!e) {
 255                 e = grab_nat_entry(nm_i, nid);
 256                 node_info_from_raw_nat(&e->ni, ne);
 257         }
 258         up_write(&nm_i->nat_tree_lock);
 259 }
 260
 261 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 262                         block_t new_blkaddr, bool fsync_done)
 263 {
 264         struct f2fs_nm_info *nm_i = NM_I(sbi);
 265         struct nat_entry *e;
 266
 267         down_write(&nm_i->nat_tree_lock);
 268         e = __lookup_nat_cache(nm_i, ni->nid);
 269         if (!e) {
 270                 e = grab_nat_entry(nm_i, ni->nid);
 271                 e->ni = *ni;
 272                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 273         } else if (new_blkaddr == NEW_ADDR) {
 274                 /*
 275                  * when nid is reallocated,
 276                  * previous nat entry can be remained in nat cache.
 277                  * So, reinitialize it with new information.
 278                  */
 279                 e->ni = *ni;
 280                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 281         }
 282
 283         /* sanity check */
 284         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 285         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 286                         new_blkaddr == NULL_ADDR);
 287         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 288                         new_blkaddr == NEW_ADDR);
 289         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 290                         nat_get_blkaddr(e) != NULL_ADDR &&
 291                         new_blkaddr == NEW_ADDR);
 292
 293         /* increment version no as node is removed */
 294         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 295                 unsigned char version = nat_get_version(e);
 296                 nat_set_version(e, inc_node_version(version));
 297         }
 298
 299         /* change address */
 300         nat_set_blkaddr(e, new_blkaddr);
 301         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 302                 set_nat_flag(e, IS_CHECKPOINTED, false);
 303         __set_nat_cache_dirty(nm_i, e);
 304
 305         /* update fsync_mark if its inode nat entry is still alive */
 306         e = __lookup_nat_cache(nm_i, ni->ino);
 307         if (e) {
 308                 if (fsync_done && ni->nid == ni->ino)
 309                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 310                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 311         }
 312         up_write(&nm_i->nat_tree_lock);
 313 }
 314
 315 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 316 {
 317         struct f2fs_nm_info *nm_i = NM_I(sbi);
 318
 319         if (available_free_memory(sbi, NAT_ENTRIES))
 320                 return 0;
 321
 322         down_write(&nm_i->nat_tree_lock);
 323         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 324                 struct nat_entry *ne;
 325                 ne = list_first_entry(&nm_i->nat_entries,
 326                                         struct nat_entry, list);
 327                 __del_from_nat_cache(nm_i, ne);
 328                 nr_shrink--;
 329         }
 330         up_write(&nm_i->nat_tree_lock);
 331         return nr_shrink;
 332 }
 333
 334 /*
 335  * This function always returns success
 336  */
 337 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 338 {
 339         struct f2fs_nm_info *nm_i = NM_I(sbi);
 340         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 341         struct f2fs_summary_block *sum = curseg->sum_blk;
 342         nid_t start_nid = START_NID(nid);
 343         struct f2fs_nat_block *nat_blk;
 344         struct page *page = NULL;
 345         struct f2fs_nat_entry ne;
 346         struct nat_entry *e;
 347         int i;
 348
 349         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 350         ni->nid = nid;
 351
 352         /* Check nat cache */
 353         down_read(&nm_i->nat_tree_lock);
 354         e = __lookup_nat_cache(nm_i, nid);
 355         if (e) {
 356                 ni->ino = nat_get_ino(e);
 357                 ni->blk_addr = nat_get_blkaddr(e);
 358                 ni->version = nat_get_version(e);
 359         }
 360         up_read(&nm_i->nat_tree_lock);
 361         if (e)
 362                 return;
 363
 364         /* Check current segment summary */
 365         mutex_lock(&curseg->curseg_mutex);
 366         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 367         if (i >= 0) {
 368                 ne = nat_in_journal(sum, i);
 369                 node_info_from_raw_nat(ni, &ne);
 370         }
 371         mutex_unlock(&curseg->curseg_mutex);
 372         if (i >= 0)
 373                 goto cache;
 374
 375         /* Fill node_info from nat page */
 376         page = get_current_nat_page(sbi, start_nid);
 377         nat_blk = (struct f2fs_nat_block *)page_address(page);
 378         ne = nat_blk->entries[nid - start_nid];
 379         node_info_from_raw_nat(ni, &ne);
 380         f2fs_put_page(page, 1);
 381 cache:
 382         /* cache nat entry */
 383         cache_nat_entry(NM_I(sbi), nid, &ne);
 384 }
 385
 386 /*
 387  * The maximum depth is four.
 388  * Offset[0] will have raw inode offset.
 389  */
 390 static int get_node_path(struct f2fs_inode_info *fi, long block,
 391                                 int offset[4], unsigned int noffset[4])
 392 {
 393         const long direct_index = ADDRS_PER_INODE(fi);
 394         const long direct_blks = ADDRS_PER_BLOCK;
 395         const long dptrs_per_blk = NIDS_PER_BLOCK;
 396         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 397         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 398         int n = 0;
 399         int level = 0;
 400
 401         noffset[0] = 0;
 402
 403         if (block < direct_index) {
 404                 offset[n] = block;
 405                 goto got;
 406         }
 407         block -= direct_index;
 408         if (block < direct_blks) {
 409                 offset[n++] = NODE_DIR1_BLOCK;
 410                 noffset[n] = 1;
 411                 offset[n] = block;
 412                 level = 1;
 413                 goto got;
 414         }
 415         block -= direct_blks;
 416         if (block < direct_blks) {
 417                 offset[n++] = NODE_DIR2_BLOCK;
 418                 noffset[n] = 2;
 419                 offset[n] = block;
 420                 level = 1;
 421                 goto got;
 422         }
 423         block -= direct_blks;
 424         if (block < indirect_blks) {
 425                 offset[n++] = NODE_IND1_BLOCK;
 426                 noffset[n] = 3;
 427                 offset[n++] = block / direct_blks;
 428                 noffset[n] = 4 + offset[n - 1];
 429                 offset[n] = block % direct_blks;
 430                 level = 2;
 431                 goto got;
 432         }
 433         block -= indirect_blks;
 434         if (block < indirect_blks) {
 435                 offset[n++] = NODE_IND2_BLOCK;
 436                 noffset[n] = 4 + dptrs_per_blk;
 437                 offset[n++] = block / direct_blks;
 438                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 439                 offset[n] = block % direct_blks;
 440                 level = 2;
 441                 goto got;
 442         }
 443         block -= indirect_blks;
 444         if (block < dindirect_blks) {
 445                 offset[n++] = NODE_DIND_BLOCK;
 446                 noffset[n] = 5 + (dptrs_per_blk * 2);
 447                 offset[n++] = block / indirect_blks;
 448                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 449                               offset[n - 1] * (dptrs_per_blk + 1);
 450                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 451                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 452                               offset[n - 2] * (dptrs_per_blk + 1) +
 453                               offset[n - 1];
 454                 offset[n] = block % direct_blks;
 455                 level = 3;
 456                 goto got;
 457         } else {
 458                 BUG();
 459         }
 460 got:
 461         return level;
 462 }
 463
 464 /*
 465  * Caller should call f2fs_put_dnode(dn).
 466  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 467  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 468  * In the case of RDONLY_NODE, we don't need to care about mutex.
 469  */
 470 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 471 {
 472         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 473         struct page *npage[4];
 474         struct page *parent;
 475         int offset[4];
 476         unsigned int noffset[4];
 477         nid_t nids[4];
 478         int level, i;
 479         int err = 0;
 480
 481         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 482
 483         nids[0] = dn->inode->i_ino;
 484         npage[0] = dn->inode_page;
 485
 486         if (!npage[0]) {
 487                 npage[0] = get_node_page(sbi, nids[0]);
 488                 if (IS_ERR(npage[0]))
 489                         return PTR_ERR(npage[0]);
 490         }
 491         parent = npage[0];
 492         if (level != 0)
 493                 nids[1] = get_nid(parent, offset[0], true);
 494         dn->inode_page = npage[0];
 495         dn->inode_page_locked = true;
 496
 497         /* get indirect or direct nodes */
 498         for (i = 1; i <= level; i++) {
 499                 bool done = false;
 500
 501                 if (!nids[i] && mode == ALLOC_NODE) {
 502                         /* alloc new node */
 503                         if (!alloc_nid(sbi, &(nids[i]))) {
 504                                 err = -ENOSPC;
 505                                 goto release_pages;
 506                         }
 507
 508                         dn->nid = nids[i];
 509                         npage[i] = new_node_page(dn, noffset[i], NULL);
 510                         if (IS_ERR(npage[i])) {
 511                                 alloc_nid_failed(sbi, nids[i]);
 512                                 err = PTR_ERR(npage[i]);
 513                                 goto release_pages;
 514                         }
 515
 516                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 517                         alloc_nid_done(sbi, nids[i]);
 518                         done = true;
 519                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 520                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 521                         if (IS_ERR(npage[i])) {
 522                                 err = PTR_ERR(npage[i]);
 523                                 goto release_pages;
 524                         }
 525                         done = true;
 526                 }
 527                 if (i == 1) {
 528                         dn->inode_page_locked = false;
 529                         unlock_page(parent);
 530                 } else {
 531                         f2fs_put_page(parent, 1);
 532                 }
 533
 534                 if (!done) {
 535                         npage[i] = get_node_page(sbi, nids[i]);
 536                         if (IS_ERR(npage[i])) {
 537                                 err = PTR_ERR(npage[i]);
 538                                 f2fs_put_page(npage[0], 0);
 539                                 goto release_out;
 540                         }
 541                 }
 542                 if (i < level) {
 543                         parent = npage[i];
 544                         nids[i + 1] = get_nid(parent, offset[i], false);
 545                 }
 546         }
 547         dn->nid = nids[level];
 548         dn->ofs_in_node = offset[level];
 549         dn->node_page = npage[level];
 550         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 551         return 0;
 552
 553 release_pages:
 554         f2fs_put_page(parent, 1);
 555         if (i > 1)
 556                 f2fs_put_page(npage[0], 0);
 557 release_out:
 558         dn->inode_page = NULL;
 559         dn->node_page = NULL;
 560         return err;
 561 }
 562
 563 static void truncate_node(struct dnode_of_data *dn)
 564 {
 565         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 566         struct node_info ni;
 567
 568         get_node_info(sbi, dn->nid, &ni);
 569         if (dn->inode->i_blocks == 0) {
 570                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 571                 goto invalidate;
 572         }
 573         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 574
 575         /* Deallocate node address */
 576         invalidate_blocks(sbi, ni.blk_addr);
 577         dec_valid_node_count(sbi, dn->inode);
 578         set_node_addr(sbi, &ni, NULL_ADDR, false);
 579
 580         if (dn->nid == dn->inode->i_ino) {
 581                 remove_orphan_inode(sbi, dn->nid);
 582                 dec_valid_inode_count(sbi);
 583         } else {
 584                 sync_inode_page(dn);
 585         }
 586 invalidate:
 587         clear_node_page_dirty(dn->node_page);
 588         F2FS_SET_SB_DIRT(sbi);
 589
 590         f2fs_put_page(dn->node_page, 1);
 591
 592         invalidate_mapping_pages(NODE_MAPPING(sbi),
 593                         dn->node_page->index, dn->node_page->index);
 594
 595         dn->node_page = NULL;
 596         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 597 }
 598
 599 static int truncate_dnode(struct dnode_of_data *dn)
 600 {
 601         struct page *page;
 602
 603         if (dn->nid == 0)
 604                 return 1;
 605
 606         /* get direct node */
 607         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 608         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 609                 return 1;
 610         else if (IS_ERR(page))
 611                 return PTR_ERR(page);
 612
 613         /* Make dnode_of_data for parameter */
 614         dn->node_page = page;
 615         dn->ofs_in_node = 0;
 616         truncate_data_blocks(dn);
 617         truncate_node(dn);
 618         return 1;
 619 }
 620
 621 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 622                                                 int ofs, int depth)
 623 {
 624         struct dnode_of_data rdn = *dn;
 625         struct page *page;
 626         struct f2fs_node *rn;
 627         nid_t child_nid;
 628         unsigned int child_nofs;
 629         int freed = 0;
 630         int i, ret;
 631
 632         if (dn->nid == 0)
 633                 return NIDS_PER_BLOCK + 1;
 634
 635         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 636
 637         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 638         if (IS_ERR(page)) {
 639                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 640                 return PTR_ERR(page);
 641         }
 642
 643         rn = F2FS_NODE(page);
 644         if (depth < 3) {
 645                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 646                         child_nid = le32_to_cpu(rn->in.nid[i]);
 647                         if (child_nid == 0)
 648                                 continue;
 649                         rdn.nid = child_nid;
 650                         ret = truncate_dnode(&rdn);
 651                         if (ret < 0)
 652                                 goto out_err;
 653                         set_nid(page, i, 0, false);
 654                 }
 655         } else {
 656                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 657                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 658                         child_nid = le32_to_cpu(rn->in.nid[i]);
 659                         if (child_nid == 0) {
 660                                 child_nofs += NIDS_PER_BLOCK + 1;
 661                                 continue;
 662                         }
 663                         rdn.nid = child_nid;
 664                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 665                         if (ret == (NIDS_PER_BLOCK + 1)) {
 666                                 set_nid(page, i, 0, false);
 667                                 child_nofs += ret;
 668                         } else if (ret < 0 && ret != -ENOENT) {
 669                                 goto out_err;
 670                         }
 671                 }
 672                 freed = child_nofs;
 673         }
 674
 675         if (!ofs) {
 676                 /* remove current indirect node */
 677                 dn->node_page = page;
 678                 truncate_node(dn);
 679                 freed++;
 680         } else {
 681                 f2fs_put_page(page, 1);
 682         }
 683         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 684         return freed;
 685
 686 out_err:
 687         f2fs_put_page(page, 1);
 688         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 689         return ret;
 690 }
 691
 692 static int truncate_partial_nodes(struct dnode_of_data *dn,
 693                         struct f2fs_inode *ri, int *offset, int depth)
 694 {
 695         struct page *pages[2];
 696         nid_t nid[3];
 697         nid_t child_nid;
 698         int err = 0;
 699         int i;
 700         int idx = depth - 2;
 701
 702         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 703         if (!nid[0])
 704                 return 0;
 705
 706         /* get indirect nodes in the path */
 707         for (i = 0; i < idx + 1; i++) {
 708                 /* reference count'll be increased */
 709                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 710                 if (IS_ERR(pages[i])) {
 711                         err = PTR_ERR(pages[i]);
 712                         idx = i - 1;
 713                         goto fail;
 714                 }
 715                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 716         }
 717
 718         /* free direct nodes linked to a partial indirect node */
 719         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 720                 child_nid = get_nid(pages[idx], i, false);
 721                 if (!child_nid)
 722                         continue;
 723                 dn->nid = child_nid;
 724                 err = truncate_dnode(dn);
 725                 if (err < 0)
 726                         goto fail;
 727                 set_nid(pages[idx], i, 0, false);
 728         }
 729
 730         if (offset[idx + 1] == 0) {
 731                 dn->node_page = pages[idx];
 732                 dn->nid = nid[idx];
 733                 truncate_node(dn);
 734         } else {
 735                 f2fs_put_page(pages[idx], 1);
 736         }
 737         offset[idx]++;
 738         offset[idx + 1] = 0;
 739         idx--;
 740 fail:
 741         for (i = idx; i >= 0; i--)
 742                 f2fs_put_page(pages[i], 1);
 743
 744         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 745
 746         return err;
 747 }
 748
 749 /*
 750  * All the block addresses of data and nodes should be nullified.
 751  */
 752 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 753 {
 754         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 755         int err = 0, cont = 1;
 756         int level, offset[4], noffset[4];
 757         unsigned int nofs = 0;
 758         struct f2fs_inode *ri;
 759         struct dnode_of_data dn;
 760         struct page *page;
 761
 762         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 763
 764         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 765 restart:
 766         page = get_node_page(sbi, inode->i_ino);
 767         if (IS_ERR(page)) {
 768                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 769                 return PTR_ERR(page);
 770         }
 771
 772         set_new_dnode(&dn, inode, page, NULL, 0);
 773         unlock_page(page);
 774
 775         ri = F2FS_INODE(page);
 776         switch (level) {
 777         case 0:
 778         case 1:
 779                 nofs = noffset[1];
 780                 break;
 781         case 2:
 782                 nofs = noffset[1];
 783                 if (!offset[level - 1])
 784                         goto skip_partial;
 785                 err = truncate_partial_nodes(&dn, ri, offset, level);
 786                 if (err < 0 && err != -ENOENT)
 787                         goto fail;
 788                 nofs += 1 + NIDS_PER_BLOCK;
 789                 break;
 790         case 3:
 791                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 792                 if (!offset[level - 1])
 793                         goto skip_partial;
 794                 err = truncate_partial_nodes(&dn, ri, offset, level);
 795                 if (err < 0 && err != -ENOENT)
 796                         goto fail;
 797                 break;
 798         default:
 799                 BUG();
 800         }
 801
 802 skip_partial:
 803         while (cont) {
 804                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 805                 switch (offset[0]) {
 806                 case NODE_DIR1_BLOCK:
 807                 case NODE_DIR2_BLOCK:
 808                         err = truncate_dnode(&dn);
 809                         break;
 810
 811                 case NODE_IND1_BLOCK:
 812                 case NODE_IND2_BLOCK:
 813                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 814                         break;
 815
 816                 case NODE_DIND_BLOCK:
 817                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 818                         cont = 0;
 819                         break;
 820
 821                 default:
 822                         BUG();
 823                 }
 824                 if (err < 0 && err != -ENOENT)
 825                         goto fail;
 826                 if (offset[1] == 0 &&
 827                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 828                         lock_page(page);
 829                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 830                                 f2fs_put_page(page, 1);
 831                                 goto restart;
 832                         }
 833                         f2fs_wait_on_page_writeback(page, NODE);
 834                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 835                         set_page_dirty(page);
 836                         unlock_page(page);
 837                 }
 838                 offset[1] = 0;
 839                 offset[0]++;
 840                 nofs += err;
 841         }
 842 fail:
 843         f2fs_put_page(page, 0);
 844         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 845         return err > 0 ? 0 : err;
 846 }
 847
 848 int truncate_xattr_node(struct inode *inode, struct page *page)
 849 {
 850         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 851         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 852         struct dnode_of_data dn;
 853         struct page *npage;
 854
 855         if (!nid)
 856                 return 0;
 857
 858         npage = get_node_page(sbi, nid);
 859         if (IS_ERR(npage))
 860                 return PTR_ERR(npage);
 861
 862         F2FS_I(inode)->i_xattr_nid = 0;
 863
 864         /* need to do checkpoint during fsync */
 865         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 866
 867         set_new_dnode(&dn, inode, page, npage, nid);
 868
 869         if (page)
 870                 dn.inode_page_locked = true;
 871         truncate_node(&dn);
 872         return 0;
 873 }
 874
 875 /*
 876  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 877  * f2fs_unlock_op().
 878  */
 879 void remove_inode_page(struct inode *inode)
 880 {
 881         struct dnode_of_data dn;
 882
 883         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 884         if (get_dnode_of_data(&dn, 0, LOOKUP_NODE))
 885                 return;
 886
 887         if (truncate_xattr_node(inode, dn.inode_page)) {
 888                 f2fs_put_dnode(&dn);
 889                 return;
 890         }
 891
 892         /* remove potential inline_data blocks */
 893         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 894                                 S_ISLNK(inode->i_mode))
 895                 truncate_data_blocks_range(&dn, 1);
 896
 897         /* 0 is possible, after f2fs_new_inode() has failed */
 898         f2fs_bug_on(F2FS_I_SB(inode),
 899                         inode->i_blocks != 0 && inode->i_blocks != 1);
 900
 901         /* will put inode & node pages */
 902         truncate_node(&dn);
 903 }
 904
 905 struct page *new_inode_page(struct inode *inode)
 906 {
 907         struct dnode_of_data dn;
 908
 909         /* allocate inode page for new inode */
 910         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 911
 912         /* caller should f2fs_put_page(page, 1); */
 913         return new_node_page(&dn, 0, NULL);
 914 }
 915
 916 struct page *new_node_page(struct dnode_of_data *dn,
 917                                 unsigned int ofs, struct page *ipage)
 918 {
 919         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 920         struct node_info old_ni, new_ni;
 921         struct page *page;
 922         int err;
 923
 924         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 925                 return ERR_PTR(-EPERM);
 926
 927         page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
 928         if (!page)
 929                 return ERR_PTR(-ENOMEM);
 930
 931         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 932                 err = -ENOSPC;
 933                 goto fail;
 934         }
 935
 936         get_node_info(sbi, dn->nid, &old_ni);
 937
 938         /* Reinitialize old_ni with new node page */
 939         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
 940         new_ni = old_ni;
 941         new_ni.ino = dn->inode->i_ino;
 942         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
 943
 944         f2fs_wait_on_page_writeback(page, NODE);
 945         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 946         set_cold_node(dn->inode, page);
 947         SetPageUptodate(page);
 948         set_page_dirty(page);
 949
 950         if (f2fs_has_xattr_block(ofs))
 951                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 952
 953         dn->node_page = page;
 954         if (ipage)
 955                 update_inode(dn->inode, ipage);
 956         else
 957                 sync_inode_page(dn);
 958         if (ofs == 0)
 959                 inc_valid_inode_count(sbi);
 960
 961         return page;
 962
 963 fail:
 964         clear_node_page_dirty(page);
 965         f2fs_put_page(page, 1);
 966         return ERR_PTR(err);
 967 }
 968
 969 /*
 970  * Caller should do after getting the following values.
 971  * 0: f2fs_put_page(page, 0)
 972  * LOCKED_PAGE: f2fs_put_page(page, 1)
 973  * error: nothing
 974  */
 975 static int read_node_page(struct page *page, int rw)
 976 {
 977         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 978         struct node_info ni;
 979
 980         get_node_info(sbi, page->index, &ni);
 981
 982         if (unlikely(ni.blk_addr == NULL_ADDR)) {
 983                 f2fs_put_page(page, 1);
 984                 return -ENOENT;
 985         }
 986
 987         if (PageUptodate(page))
 988                 return LOCKED_PAGE;
 989
 990         return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
 991 }
 992
 993 /*
 994  * Readahead a node page
 995  */
 996 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 997 {
 998         struct page *apage;
 999         int err;
1000
1001         apage = find_get_page(NODE_MAPPING(sbi), nid);
1002         if (apage && PageUptodate(apage)) {
1003                 f2fs_put_page(apage, 0);
1004                 return;
1005         }
1006         f2fs_put_page(apage, 0);
1007
1008         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1009         if (!apage)
1010                 return;
1011
1012         err = read_node_page(apage, READA);
1013         if (err == 0)
1014                 f2fs_put_page(apage, 0);
1015         else if (err == LOCKED_PAGE)
1016                 f2fs_put_page(apage, 1);
1017 }
1018
1019 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1020 {
1021         struct page *page;
1022         int err;
1023 repeat:
1024         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1025         if (!page)
1026                 return ERR_PTR(-ENOMEM);
1027
1028         err = read_node_page(page, READ_SYNC);
1029         if (err < 0)
1030                 return ERR_PTR(err);
1031         else if (err == LOCKED_PAGE)
1032                 goto got_it;
1033
1034         lock_page(page);
1035         if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
1036                 f2fs_put_page(page, 1);
1037                 return ERR_PTR(-EIO);
1038         }
1039         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1040                 f2fs_put_page(page, 1);
1041                 goto repeat;
1042         }
1043 got_it:
1044         return page;
1045 }
1046
1047 /*
1048  * Return a locked page for the desired node page.
1049  * And, readahead MAX_RA_NODE number of node pages.
1050  */
1051 struct page *get_node_page_ra(struct page *parent, int start)
1052 {
1053         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1054         struct blk_plug plug;
1055         struct page *page;
1056         int err, i, end;
1057         nid_t nid;
1058
1059         /* First, try getting the desired direct node. */
1060         nid = get_nid(parent, start, false);
1061         if (!nid)
1062                 return ERR_PTR(-ENOENT);
1063 repeat:
1064         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1065         if (!page)
1066                 return ERR_PTR(-ENOMEM);
1067
1068         err = read_node_page(page, READ_SYNC);
1069         if (err < 0)
1070                 return ERR_PTR(err);
1071         else if (err == LOCKED_PAGE)
1072                 goto page_hit;
1073
1074         blk_start_plug(&plug);
1075
1076         /* Then, try readahead for siblings of the desired node */
1077         end = start + MAX_RA_NODE;
1078         end = min(end, NIDS_PER_BLOCK);
1079         for (i = start + 1; i < end; i++) {
1080                 nid = get_nid(parent, i, false);
1081                 if (!nid)
1082                         continue;
1083                 ra_node_page(sbi, nid);
1084         }
1085
1086         blk_finish_plug(&plug);
1087
1088         lock_page(page);
1089         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1090                 f2fs_put_page(page, 1);
1091                 goto repeat;
1092         }
1093 page_hit:
1094         if (unlikely(!PageUptodate(page))) {
1095                 f2fs_put_page(page, 1);
1096                 return ERR_PTR(-EIO);
1097         }
1098         return page;
1099 }
1100
1101 void sync_inode_page(struct dnode_of_data *dn)
1102 {
1103         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1104                 update_inode(dn->inode, dn->node_page);
1105         } else if (dn->inode_page) {
1106                 if (!dn->inode_page_locked)
1107                         lock_page(dn->inode_page);
1108                 update_inode(dn->inode, dn->inode_page);
1109                 if (!dn->inode_page_locked)
1110                         unlock_page(dn->inode_page);
1111         } else {
1112                 update_inode_page(dn->inode);
1113         }
1114 }
1115
1116 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1117                                         struct writeback_control *wbc)
1118 {
1119         pgoff_t index, end;
1120         struct pagevec pvec;
1121         int step = ino ? 2 : 0;
1122         int nwritten = 0, wrote = 0;
1123
1124         pagevec_init(&pvec, 0);
1125
1126 next_step:
1127         index = 0;
1128         end = LONG_MAX;
1129
1130         while (index <= end) {
1131                 int i, nr_pages;
1132                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1133                                 PAGECACHE_TAG_DIRTY,
1134                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1135                 if (nr_pages == 0)
1136                         break;
1137
1138                 for (i = 0; i < nr_pages; i++) {
1139                         struct page *page = pvec.pages[i];
1140
1141                         /*
1142                          * flushing sequence with step:
1143                          * 0. indirect nodes
1144                          * 1. dentry dnodes
1145                          * 2. file dnodes
1146                          */
1147                         if (step == 0 && IS_DNODE(page))
1148                                 continue;
1149                         if (step == 1 && (!IS_DNODE(page) ||
1150                                                 is_cold_node(page)))
1151                                 continue;
1152                         if (step == 2 && (!IS_DNODE(page) ||
1153                                                 !is_cold_node(page)))
1154                                 continue;
1155
1156                         /*
1157                          * If an fsync mode,
1158                          * we should not skip writing node pages.
1159                          */
1160                         if (ino && ino_of_node(page) == ino)
1161                                 lock_page(page);
1162                         else if (!trylock_page(page))
1163                                 continue;
1164
1165                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1166 continue_unlock:
1167                                 unlock_page(page);
1168                                 continue;
1169                         }
1170                         if (ino && ino_of_node(page) != ino)
1171                                 goto continue_unlock;
1172
1173                         if (!PageDirty(page)) {
1174                                 /* someone wrote it for us */
1175                                 goto continue_unlock;
1176                         }
1177
1178                         if (!clear_page_dirty_for_io(page))
1179                                 goto continue_unlock;
1180
1181                         /* called by fsync() */
1182                         if (ino && IS_DNODE(page)) {
1183                                 set_fsync_mark(page, 1);
1184                                 if (IS_INODE(page)) {
1185                                         if (!is_checkpointed_node(sbi, ino) &&
1186                                                 !has_fsynced_inode(sbi, ino))
1187                                                 set_dentry_mark(page, 1);
1188                                         else
1189                                                 set_dentry_mark(page, 0);
1190                                 }
1191                                 nwritten++;
1192                         } else {
1193                                 set_fsync_mark(page, 0);
1194                                 set_dentry_mark(page, 0);
1195                         }
1196
1197                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1198                                 unlock_page(page);
1199                         else
1200                                 wrote++;
1201
1202                         if (--wbc->nr_to_write == 0)
1203                                 break;
1204                 }
1205                 pagevec_release(&pvec);
1206                 cond_resched();
1207
1208                 if (wbc->nr_to_write == 0) {
1209                         step = 2;
1210                         break;
1211                 }
1212         }
1213
1214         if (step < 2) {
1215                 step++;
1216                 goto next_step;
1217         }
1218
1219         if (wrote)
1220                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1221         return nwritten;
1222 }
1223
1224 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1225 {
1226         pgoff_t index = 0, end = LONG_MAX;
1227         struct pagevec pvec;
1228         int ret2 = 0, ret = 0;
1229
1230         pagevec_init(&pvec, 0);
1231
1232         while (index <= end) {
1233                 int i, nr_pages;
1234                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1235                                 PAGECACHE_TAG_WRITEBACK,
1236                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1237                 if (nr_pages == 0)
1238                         break;
1239
1240                 for (i = 0; i < nr_pages; i++) {
1241                         struct page *page = pvec.pages[i];
1242
1243                         /* until radix tree lookup accepts end_index */
1244                         if (unlikely(page->index > end))
1245                                 continue;
1246
1247                         if (ino && ino_of_node(page) == ino) {
1248                                 f2fs_wait_on_page_writeback(page, NODE);
1249                                 if (TestClearPageError(page))
1250                                         ret = -EIO;
1251                         }
1252                 }
1253                 pagevec_release(&pvec);
1254                 cond_resched();
1255         }
1256
1257         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1258                 ret2 = -ENOSPC;
1259         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1260                 ret2 = -EIO;
1261         if (!ret)
1262                 ret = ret2;
1263         return ret;
1264 }
1265
1266 static int f2fs_write_node_page(struct page *page,
1267                                 struct writeback_control *wbc)
1268 {
1269         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1270         nid_t nid;
1271         block_t new_addr;
1272         struct node_info ni;
1273         struct f2fs_io_info fio = {
1274                 .type = NODE,
1275                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1276         };
1277
1278         trace_f2fs_writepage(page, NODE);
1279
1280         if (unlikely(sbi->por_doing))
1281                 goto redirty_out;
1282         if (unlikely(f2fs_cp_error(sbi)))
1283                 goto redirty_out;
1284
1285         f2fs_wait_on_page_writeback(page, NODE);
1286
1287         /* get old block addr of this node page */
1288         nid = nid_of_node(page);
1289         f2fs_bug_on(sbi, page->index != nid);
1290
1291         get_node_info(sbi, nid, &ni);
1292
1293         /* This page is already truncated */
1294         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1295                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1296                 unlock_page(page);
1297                 return 0;
1298         }
1299
1300         if (wbc->for_reclaim) {
1301                 if (!down_read_trylock(&sbi->node_write))
1302                         goto redirty_out;
1303         } else {
1304                 down_read(&sbi->node_write);
1305         }
1306         set_page_writeback(page);
1307         write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1308         set_node_addr(sbi, &ni, new_addr, is_fsync_dnode(page));
1309         dec_page_count(sbi, F2FS_DIRTY_NODES);
1310         up_read(&sbi->node_write);
1311         unlock_page(page);
1312
1313         if (wbc->for_reclaim)
1314                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1315
1316         return 0;
1317
1318 redirty_out:
1319         redirty_page_for_writepage(wbc, page);
1320         return AOP_WRITEPAGE_ACTIVATE;
1321 }
1322
1323 static int f2fs_write_node_pages(struct address_space *mapping,
1324                             struct writeback_control *wbc)
1325 {
1326         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1327         long diff;
1328
1329         trace_f2fs_writepages(mapping->host, wbc, NODE);
1330
1331         /* balancing f2fs's metadata in background */
1332         f2fs_balance_fs_bg(sbi);
1333
1334         /* collect a number of dirty node pages and write together */
1335         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1336                 goto skip_write;
1337
1338         diff = nr_pages_to_write(sbi, NODE, wbc);
1339         wbc->sync_mode = WB_SYNC_NONE;
1340         sync_node_pages(sbi, 0, wbc);
1341         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1342         return 0;
1343
1344 skip_write:
1345         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1346         return 0;
1347 }
1348
1349 static int f2fs_set_node_page_dirty(struct page *page)
1350 {
1351         trace_f2fs_set_page_dirty(page, NODE);
1352
1353         SetPageUptodate(page);
1354         if (!PageDirty(page)) {
1355                 __set_page_dirty_nobuffers(page);
1356                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1357                 SetPagePrivate(page);
1358                 return 1;
1359         }
1360         return 0;
1361 }
1362
1363 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1364                                       unsigned int length)
1365 {
1366         struct inode *inode = page->mapping->host;
1367         if (PageDirty(page))
1368                 dec_page_count(F2FS_I_SB(inode), F2FS_DIRTY_NODES);
1369         ClearPagePrivate(page);
1370 }
1371
1372 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1373 {
1374         ClearPagePrivate(page);
1375         return 1;
1376 }
1377
1378 /*
1379  * Structure of the f2fs node operations
1380  */
1381 const struct address_space_operations f2fs_node_aops = {
1382         .writepage      = f2fs_write_node_page,
1383         .writepages     = f2fs_write_node_pages,
1384         .set_page_dirty = f2fs_set_node_page_dirty,
1385         .invalidatepage = f2fs_invalidate_node_page,
1386         .releasepage    = f2fs_release_node_page,
1387 };
1388
1389 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1390                                                 nid_t n)
1391 {
1392         return radix_tree_lookup(&nm_i->free_nid_root, n);
1393 }
1394
1395 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1396                                                 struct free_nid *i)
1397 {
1398         list_del(&i->list);
1399         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1400 }
1401
1402 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1403 {
1404         struct f2fs_nm_info *nm_i = NM_I(sbi);
1405         struct free_nid *i;
1406         struct nat_entry *ne;
1407         bool allocated = false;
1408
1409         if (!available_free_memory(sbi, FREE_NIDS))
1410                 return -1;
1411
1412         /* 0 nid should not be used */
1413         if (unlikely(nid == 0))
1414                 return 0;
1415
1416         if (build) {
1417                 /* do not add allocated nids */
1418                 down_read(&nm_i->nat_tree_lock);
1419                 ne = __lookup_nat_cache(nm_i, nid);
1420                 if (ne &&
1421                         (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1422                                 nat_get_blkaddr(ne) != NULL_ADDR))
1423                         allocated = true;
1424                 up_read(&nm_i->nat_tree_lock);
1425                 if (allocated)
1426                         return 0;
1427         }
1428
1429         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1430         i->nid = nid;
1431         i->state = NID_NEW;
1432
1433         if (radix_tree_preload(GFP_NOFS)) {
1434                 kmem_cache_free(free_nid_slab, i);
1435                 return 0;
1436         }
1437
1438         spin_lock(&nm_i->free_nid_list_lock);
1439         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1440                 spin_unlock(&nm_i->free_nid_list_lock);
1441                 radix_tree_preload_end();
1442                 kmem_cache_free(free_nid_slab, i);
1443                 return 0;
1444         }
1445         list_add_tail(&i->list, &nm_i->free_nid_list);
1446         nm_i->fcnt++;
1447         spin_unlock(&nm_i->free_nid_list_lock);
1448         radix_tree_preload_end();
1449         return 1;
1450 }
1451
1452 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1453 {
1454         struct free_nid *i;
1455         bool need_free = false;
1456
1457         spin_lock(&nm_i->free_nid_list_lock);
1458         i = __lookup_free_nid_list(nm_i, nid);
1459         if (i && i->state == NID_NEW) {
1460                 __del_from_free_nid_list(nm_i, i);
1461                 nm_i->fcnt--;
1462                 need_free = true;
1463         }
1464         spin_unlock(&nm_i->free_nid_list_lock);
1465
1466         if (need_free)
1467                 kmem_cache_free(free_nid_slab, i);
1468 }
1469
1470 static void scan_nat_page(struct f2fs_sb_info *sbi,
1471                         struct page *nat_page, nid_t start_nid)
1472 {
1473         struct f2fs_nm_info *nm_i = NM_I(sbi);
1474         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1475         block_t blk_addr;
1476         int i;
1477
1478         i = start_nid % NAT_ENTRY_PER_BLOCK;
1479
1480         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1481
1482                 if (unlikely(start_nid >= nm_i->max_nid))
1483                         break;
1484
1485                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1486                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1487                 if (blk_addr == NULL_ADDR) {
1488                         if (add_free_nid(sbi, start_nid, true) < 0)
1489                                 break;
1490                 }
1491         }
1492 }
1493
1494 static void build_free_nids(struct f2fs_sb_info *sbi)
1495 {
1496         struct f2fs_nm_info *nm_i = NM_I(sbi);
1497         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1498         struct f2fs_summary_block *sum = curseg->sum_blk;
1499         int i = 0;
1500         nid_t nid = nm_i->next_scan_nid;
1501
1502         /* Enough entries */
1503         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1504                 return;
1505
1506         /* readahead nat pages to be scanned */
1507         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1508
1509         while (1) {
1510                 struct page *page = get_current_nat_page(sbi, nid);
1511
1512                 scan_nat_page(sbi, page, nid);
1513                 f2fs_put_page(page, 1);
1514
1515                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1516                 if (unlikely(nid >= nm_i->max_nid))
1517                         nid = 0;
1518
1519                 if (i++ == FREE_NID_PAGES)
1520                         break;
1521         }
1522
1523         /* go to the next free nat pages to find free nids abundantly */
1524         nm_i->next_scan_nid = nid;
1525
1526         /* find free nids from current sum_pages */
1527         mutex_lock(&curseg->curseg_mutex);
1528         for (i = 0; i < nats_in_cursum(sum); i++) {
1529                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1530                 nid = le32_to_cpu(nid_in_journal(sum, i));
1531                 if (addr == NULL_ADDR)
1532                         add_free_nid(sbi, nid, true);
1533                 else
1534                         remove_free_nid(nm_i, nid);
1535         }
1536         mutex_unlock(&curseg->curseg_mutex);
1537 }
1538
1539 /*
1540  * If this function returns success, caller can obtain a new nid
1541  * from second parameter of this function.
1542  * The returned nid could be used ino as well as nid when inode is created.
1543  */
1544 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1545 {
1546         struct f2fs_nm_info *nm_i = NM_I(sbi);
1547         struct free_nid *i = NULL;
1548 retry:
1549         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1550                 return false;
1551
1552         spin_lock(&nm_i->free_nid_list_lock);
1553
1554         /* We should not use stale free nids created by build_free_nids */
1555         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1556                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1557                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1558                         if (i->state == NID_NEW)
1559                                 break;
1560
1561                 f2fs_bug_on(sbi, i->state != NID_NEW);
1562                 *nid = i->nid;
1563                 i->state = NID_ALLOC;
1564                 nm_i->fcnt--;
1565                 spin_unlock(&nm_i->free_nid_list_lock);
1566                 return true;
1567         }
1568         spin_unlock(&nm_i->free_nid_list_lock);
1569
1570         /* Let's scan nat pages and its caches to get free nids */
1571         mutex_lock(&nm_i->build_lock);
1572         build_free_nids(sbi);
1573         mutex_unlock(&nm_i->build_lock);
1574         goto retry;
1575 }
1576
1577 /*
1578  * alloc_nid() should be called prior to this function.
1579  */
1580 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1581 {
1582         struct f2fs_nm_info *nm_i = NM_I(sbi);
1583         struct free_nid *i;
1584
1585         spin_lock(&nm_i->free_nid_list_lock);
1586         i = __lookup_free_nid_list(nm_i, nid);
1587         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1588         __del_from_free_nid_list(nm_i, i);
1589         spin_unlock(&nm_i->free_nid_list_lock);
1590
1591         kmem_cache_free(free_nid_slab, i);
1592 }
1593
1594 /*
1595  * alloc_nid() should be called prior to this function.
1596  */
1597 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1598 {
1599         struct f2fs_nm_info *nm_i = NM_I(sbi);
1600         struct free_nid *i;
1601         bool need_free = false;
1602
1603         if (!nid)
1604                 return;
1605
1606         spin_lock(&nm_i->free_nid_list_lock);
1607         i = __lookup_free_nid_list(nm_i, nid);
1608         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1609         if (!available_free_memory(sbi, FREE_NIDS)) {
1610                 __del_from_free_nid_list(nm_i, i);
1611                 need_free = true;
1612         } else {
1613                 i->state = NID_NEW;
1614                 nm_i->fcnt++;
1615         }
1616         spin_unlock(&nm_i->free_nid_list_lock);
1617
1618         if (need_free)
1619                 kmem_cache_free(free_nid_slab, i);
1620 }
1621
1622 void recover_inline_xattr(struct inode *inode, struct page *page)
1623 {
1624         void *src_addr, *dst_addr;
1625         size_t inline_size;
1626         struct page *ipage;
1627         struct f2fs_inode *ri;
1628
1629         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1630         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1631
1632         ri = F2FS_INODE(page);
1633         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1634                 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1635                 goto update_inode;
1636         }
1637
1638         dst_addr = inline_xattr_addr(ipage);
1639         src_addr = inline_xattr_addr(page);
1640         inline_size = inline_xattr_size(inode);
1641
1642         f2fs_wait_on_page_writeback(ipage, NODE);
1643         memcpy(dst_addr, src_addr, inline_size);
1644 update_inode:
1645         update_inode(inode, ipage);
1646         f2fs_put_page(ipage, 1);
1647 }
1648
1649 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1650 {
1651         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1652         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1653         nid_t new_xnid = nid_of_node(page);
1654         struct node_info ni;
1655
1656         /* 1: invalidate the previous xattr nid */
1657         if (!prev_xnid)
1658                 goto recover_xnid;
1659
1660         /* Deallocate node address */
1661         get_node_info(sbi, prev_xnid, &ni);
1662         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1663         invalidate_blocks(sbi, ni.blk_addr);
1664         dec_valid_node_count(sbi, inode);
1665         set_node_addr(sbi, &ni, NULL_ADDR, false);
1666
1667 recover_xnid:
1668         /* 2: allocate new xattr nid */
1669         if (unlikely(!inc_valid_node_count(sbi, inode)))
1670                 f2fs_bug_on(sbi, 1);
1671
1672         remove_free_nid(NM_I(sbi), new_xnid);
1673         get_node_info(sbi, new_xnid, &ni);
1674         ni.ino = inode->i_ino;
1675         set_node_addr(sbi, &ni, NEW_ADDR, false);
1676         F2FS_I(inode)->i_xattr_nid = new_xnid;
1677
1678         /* 3: update xattr blkaddr */
1679         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1680         set_node_addr(sbi, &ni, blkaddr, false);
1681
1682         update_inode_page(inode);
1683 }
1684
1685 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1686 {
1687         struct f2fs_inode *src, *dst;
1688         nid_t ino = ino_of_node(page);
1689         struct node_info old_ni, new_ni;
1690         struct page *ipage;
1691
1692         get_node_info(sbi, ino, &old_ni);
1693
1694         if (unlikely(old_ni.blk_addr != NULL_ADDR))
1695                 return -EINVAL;
1696
1697         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1698         if (!ipage)
1699                 return -ENOMEM;
1700
1701         /* Should not use this inode from free nid list */
1702         remove_free_nid(NM_I(sbi), ino);
1703
1704         SetPageUptodate(ipage);
1705         fill_node_footer(ipage, ino, ino, 0, true);
1706
1707         src = F2FS_INODE(page);
1708         dst = F2FS_INODE(ipage);
1709
1710         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1711         dst->i_size = 0;
1712         dst->i_blocks = cpu_to_le64(1);
1713         dst->i_links = cpu_to_le32(1);
1714         dst->i_xattr_nid = 0;
1715         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1716
1717         new_ni = old_ni;
1718         new_ni.ino = ino;
1719
1720         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1721                 WARN_ON(1);
1722         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1723         inc_valid_inode_count(sbi);
1724         set_page_dirty(ipage);
1725         f2fs_put_page(ipage, 1);
1726         return 0;
1727 }
1728
1729 /*
1730  * ra_sum_pages() merge contiguous pages into one bio and submit.
1731  * these pre-read pages are allocated in bd_inode's mapping tree.
1732  */
1733 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct page **pages,
1734                                 int start, int nrpages)
1735 {
1736         struct inode *inode = sbi->sb->s_bdev->bd_inode;
1737         struct address_space *mapping = inode->i_mapping;
1738         int i, page_idx = start;
1739         struct f2fs_io_info fio = {
1740                 .type = META,
1741                 .rw = READ_SYNC | REQ_META | REQ_PRIO
1742         };
1743
1744         for (i = 0; page_idx < start + nrpages; page_idx++, i++) {
1745                 /* alloc page in bd_inode for reading node summary info */
1746                 pages[i] = grab_cache_page(mapping, page_idx);
1747                 if (!pages[i])
1748                         break;
1749                 f2fs_submit_page_mbio(sbi, pages[i], page_idx, &fio);
1750         }
1751
1752         f2fs_submit_merged_bio(sbi, META, READ);
1753         return i;
1754 }
1755
1756 int restore_node_summary(struct f2fs_sb_info *sbi,
1757                         unsigned int segno, struct f2fs_summary_block *sum)
1758 {
1759         struct f2fs_node *rn;
1760         struct f2fs_summary *sum_entry;
1761         struct inode *inode = sbi->sb->s_bdev->bd_inode;
1762         block_t addr;
1763         int bio_blocks = MAX_BIO_BLOCKS(sbi);
1764         struct page *pages[bio_blocks];
1765         int i, idx, last_offset, nrpages, err = 0;
1766
1767         /* scan the node segment */
1768         last_offset = sbi->blocks_per_seg;
1769         addr = START_BLOCK(sbi, segno);
1770         sum_entry = &sum->entries[0];
1771
1772         for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) {
1773                 nrpages = min(last_offset - i, bio_blocks);
1774
1775                 /* readahead node pages */
1776                 nrpages = ra_sum_pages(sbi, pages, addr, nrpages);
1777                 if (!nrpages)
1778                         return -ENOMEM;
1779
1780                 for (idx = 0; idx < nrpages; idx++) {
1781                         if (err)
1782                                 goto skip;
1783
1784                         lock_page(pages[idx]);
1785                         if (unlikely(!PageUptodate(pages[idx]))) {
1786                                 err = -EIO;
1787                         } else {
1788                                 rn = F2FS_NODE(pages[idx]);
1789                                 sum_entry->nid = rn->footer.nid;
1790                                 sum_entry->version = 0;
1791                                 sum_entry->ofs_in_node = 0;
1792                                 sum_entry++;
1793                         }
1794                         unlock_page(pages[idx]);
1795 skip:
1796                         page_cache_release(pages[idx]);
1797                 }
1798
1799                 invalidate_mapping_pages(inode->i_mapping, addr,
1800                                                         addr + nrpages);
1801         }
1802         return err;
1803 }
1804
1805 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1806 {
1807         struct f2fs_nm_info *nm_i = NM_I(sbi);
1808         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1809         struct f2fs_summary_block *sum = curseg->sum_blk;
1810         int i;
1811
1812         mutex_lock(&curseg->curseg_mutex);
1813         for (i = 0; i < nats_in_cursum(sum); i++) {
1814                 struct nat_entry *ne;
1815                 struct f2fs_nat_entry raw_ne;
1816                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1817
1818                 raw_ne = nat_in_journal(sum, i);
1819
1820                 down_write(&nm_i->nat_tree_lock);
1821                 ne = __lookup_nat_cache(nm_i, nid);
1822                 if (!ne) {
1823                         ne = grab_nat_entry(nm_i, nid);
1824                         node_info_from_raw_nat(&ne->ni, &raw_ne);
1825                 }
1826                 __set_nat_cache_dirty(nm_i, ne);
1827                 up_write(&nm_i->nat_tree_lock);
1828         }
1829         update_nats_in_cursum(sum, -i);
1830         mutex_unlock(&curseg->curseg_mutex);
1831 }
1832
1833 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1834                                                 struct list_head *head, int max)
1835 {
1836         struct nat_entry_set *cur;
1837
1838         if (nes->entry_cnt >= max)
1839                 goto add_out;
1840
1841         list_for_each_entry(cur, head, set_list) {
1842                 if (cur->entry_cnt >= nes->entry_cnt) {
1843                         list_add(&nes->set_list, cur->set_list.prev);
1844                         return;
1845                 }
1846         }
1847 add_out:
1848         list_add_tail(&nes->set_list, head);
1849 }
1850
1851 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1852                                         struct nat_entry_set *set)
1853 {
1854         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1855         struct f2fs_summary_block *sum = curseg->sum_blk;
1856         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1857         bool to_journal = true;
1858         struct f2fs_nat_block *nat_blk;
1859         struct nat_entry *ne, *cur;
1860         struct page *page = NULL;
1861
1862         /*
1863          * there are two steps to flush nat entries:
1864          * #1, flush nat entries to journal in current hot data summary block.
1865          * #2, flush nat entries to nat page.
1866          */
1867         if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1868                 to_journal = false;
1869
1870         if (to_journal) {
1871                 mutex_lock(&curseg->curseg_mutex);
1872         } else {
1873                 page = get_next_nat_page(sbi, start_nid);
1874                 nat_blk = page_address(page);
1875                 f2fs_bug_on(sbi, !nat_blk);
1876         }
1877
1878         /* flush dirty nats in nat entry set */
1879         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1880                 struct f2fs_nat_entry *raw_ne;
1881                 nid_t nid = nat_get_nid(ne);
1882                 int offset;
1883
1884                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1885                         continue;
1886
1887                 if (to_journal) {
1888                         offset = lookup_journal_in_cursum(sum,
1889                                                         NAT_JOURNAL, nid, 1);
1890                         f2fs_bug_on(sbi, offset < 0);
1891                         raw_ne = &nat_in_journal(sum, offset);
1892                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1893                 } else {
1894                         raw_ne = &nat_blk->entries[nid - start_nid];
1895                 }
1896                 raw_nat_from_node_info(raw_ne, &ne->ni);
1897
1898                 down_write(&NM_I(sbi)->nat_tree_lock);
1899                 nat_reset_flag(ne);
1900                 __clear_nat_cache_dirty(NM_I(sbi), ne);
1901                 up_write(&NM_I(sbi)->nat_tree_lock);
1902
1903                 if (nat_get_blkaddr(ne) == NULL_ADDR)
1904                         add_free_nid(sbi, nid, false);
1905         }
1906
1907         if (to_journal)
1908                 mutex_unlock(&curseg->curseg_mutex);
1909         else
1910                 f2fs_put_page(page, 1);
1911
1912         f2fs_bug_on(sbi, set->entry_cnt);
1913
1914         radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1915         kmem_cache_free(nat_entry_set_slab, set);
1916 }
1917
1918 /*
1919  * This function is called during the checkpointing process.
1920  */
1921 void flush_nat_entries(struct f2fs_sb_info *sbi)
1922 {
1923         struct f2fs_nm_info *nm_i = NM_I(sbi);
1924         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1925         struct f2fs_summary_block *sum = curseg->sum_blk;
1926         struct nat_entry_set *setvec[NATVEC_SIZE];
1927         struct nat_entry_set *set, *tmp;
1928         unsigned int found;
1929         nid_t set_idx = 0;
1930         LIST_HEAD(sets);
1931
1932         if (!nm_i->dirty_nat_cnt)
1933                 return;
1934         /*
1935          * if there are no enough space in journal to store dirty nat
1936          * entries, remove all entries from journal and merge them
1937          * into nat entry set.
1938          */
1939         if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1940                 remove_nats_in_journal(sbi);
1941
1942         while ((found = __gang_lookup_nat_set(nm_i,
1943                                         set_idx, NATVEC_SIZE, setvec))) {
1944                 unsigned idx;
1945                 set_idx = setvec[found - 1]->set + 1;
1946                 for (idx = 0; idx < found; idx++)
1947                         __adjust_nat_entry_set(setvec[idx], &sets,
1948                                                         MAX_NAT_JENTRIES(sum));
1949         }
1950
1951         /* flush dirty nats in nat entry set */
1952         list_for_each_entry_safe(set, tmp, &sets, set_list)
1953                 __flush_nat_entry_set(sbi, set);
1954
1955         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1956 }
1957
1958 static int init_node_manager(struct f2fs_sb_info *sbi)
1959 {
1960         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1961         struct f2fs_nm_info *nm_i = NM_I(sbi);
1962         unsigned char *version_bitmap;
1963         unsigned int nat_segs, nat_blocks;
1964
1965         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1966
1967         /* segment_count_nat includes pair segment so divide to 2. */
1968         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1969         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1970
1971         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1972
1973         /* not used nids: 0, node, meta, (and root counted as valid node) */
1974         nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
1975         nm_i->fcnt = 0;
1976         nm_i->nat_cnt = 0;
1977         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1978
1979         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1980         INIT_LIST_HEAD(&nm_i->free_nid_list);
1981         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
1982         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
1983         INIT_LIST_HEAD(&nm_i->nat_entries);
1984
1985         mutex_init(&nm_i->build_lock);
1986         spin_lock_init(&nm_i->free_nid_list_lock);
1987         init_rwsem(&nm_i->nat_tree_lock);
1988
1989         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1990         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1991         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1992         if (!version_bitmap)
1993                 return -EFAULT;
1994
1995         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1996                                         GFP_KERNEL);
1997         if (!nm_i->nat_bitmap)
1998                 return -ENOMEM;
1999         return 0;
2000 }
2001
2002 int build_node_manager(struct f2fs_sb_info *sbi)
2003 {
2004         int err;
2005
2006         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2007         if (!sbi->nm_info)
2008                 return -ENOMEM;
2009
2010         err = init_node_manager(sbi);
2011         if (err)
2012                 return err;
2013
2014         build_free_nids(sbi);
2015         return 0;
2016 }
2017
2018 void destroy_node_manager(struct f2fs_sb_info *sbi)
2019 {
2020         struct f2fs_nm_info *nm_i = NM_I(sbi);
2021         struct free_nid *i, *next_i;
2022         struct nat_entry *natvec[NATVEC_SIZE];
2023         nid_t nid = 0;
2024         unsigned int found;
2025
2026         if (!nm_i)
2027                 return;
2028
2029         /* destroy free nid list */
2030         spin_lock(&nm_i->free_nid_list_lock);
2031         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2032                 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2033                 __del_from_free_nid_list(nm_i, i);
2034                 nm_i->fcnt--;
2035                 spin_unlock(&nm_i->free_nid_list_lock);
2036                 kmem_cache_free(free_nid_slab, i);
2037                 spin_lock(&nm_i->free_nid_list_lock);
2038         }
2039         f2fs_bug_on(sbi, nm_i->fcnt);
2040         spin_unlock(&nm_i->free_nid_list_lock);
2041
2042         /* destroy nat cache */
2043         down_write(&nm_i->nat_tree_lock);
2044         while ((found = __gang_lookup_nat_cache(nm_i,
2045                                         nid, NATVEC_SIZE, natvec))) {
2046                 unsigned idx;
2047                 nid = nat_get_nid(natvec[found - 1]) + 1;
2048                 for (idx = 0; idx < found; idx++)
2049                         __del_from_nat_cache(nm_i, natvec[idx]);
2050         }
2051         f2fs_bug_on(sbi, nm_i->nat_cnt);
2052         up_write(&nm_i->nat_tree_lock);
2053
2054         kfree(nm_i->nat_bitmap);
2055         sbi->nm_info = NULL;
2056         kfree(nm_i);
2057 }
2058
2059 int __init create_node_manager_caches(void)
2060 {
2061         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2062                         sizeof(struct nat_entry));
2063         if (!nat_entry_slab)
2064                 goto fail;
2065
2066         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2067                         sizeof(struct free_nid));
2068         if (!free_nid_slab)
2069                 goto destroy_nat_entry;
2070
2071         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2072                         sizeof(struct nat_entry_set));
2073         if (!nat_entry_set_slab)
2074                 goto destroy_free_nid;
2075         return 0;
2076
2077 destroy_free_nid:
2078         kmem_cache_destroy(free_nid_slab);
2079 destroy_nat_entry:
2080         kmem_cache_destroy(nat_entry_slab);
2081 fail:
2082         return -ENOMEM;
2083 }
2084
2085 void destroy_node_manager_caches(void)
2086 {
2087         kmem_cache_destroy(nat_entry_set_slab);
2088         kmem_cache_destroy(free_nid_slab);
2089         kmem_cache_destroy(nat_entry_slab);
2090 }