fs/f2fs/node.c

   1 /*
   2  * fs/f2fs/node.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/mpage.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/swap.h>
  18
  19 #include "f2fs.h"
  20 #include "node.h"
  21 #include "segment.h"
  22 #include "trace.h"
  23 #include <trace/events/f2fs.h>
  24
  25 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
  26
  27 static struct kmem_cache *nat_entry_slab;
  28 static struct kmem_cache *free_nid_slab;
  29 static struct kmem_cache *nat_entry_set_slab;
  30
  31 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
  32 {
  33         struct f2fs_nm_info *nm_i = NM_I(sbi);
  34         struct sysinfo val;
  35         unsigned long avail_ram;
  36         unsigned long mem_size = 0;
  37         bool res = false;
  38
  39         si_meminfo(&val);
  40
  41         /* only uses low memory */
  42         avail_ram = val.totalram - val.totalhigh;
  43
  44         /*
  45          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  46          */
  47         if (type == FREE_NIDS) {
  48                 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
  49                                                         PAGE_CACHE_SHIFT;
  50                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  51         } else if (type == NAT_ENTRIES) {
  52                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
  53                                                         PAGE_CACHE_SHIFT;
  54                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  55         } else if (type == DIRTY_DENTS) {
  56                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  57                         return false;
  58                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  59                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  60         } else if (type == INO_ENTRIES) {
  61                 int i;
  62
  63                 for (i = 0; i <= UPDATE_INO; i++)
  64                         mem_size += (sbi->im[i].ino_num *
  65                                 sizeof(struct ino_entry)) >> PAGE_CACHE_SHIFT;
  66                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  67         } else if (type == EXTENT_CACHE) {
  68                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  69                                 sizeof(struct extent_tree) +
  70                                 atomic_read(&sbi->total_ext_node) *
  71                                 sizeof(struct extent_node)) >> PAGE_CACHE_SHIFT;
  72                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  73         } else {
  74                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
  75                         return true;
  76         }
  77         return res;
  78 }
  79
  80 static void clear_node_page_dirty(struct page *page)
  81 {
  82         struct address_space *mapping = page->mapping;
  83         unsigned int long flags;
  84
  85         if (PageDirty(page)) {
  86                 spin_lock_irqsave(&mapping->tree_lock, flags);
  87                 radix_tree_tag_clear(&mapping->page_tree,
  88                                 page_index(page),
  89                                 PAGECACHE_TAG_DIRTY);
  90                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  91
  92                 clear_page_dirty_for_io(page);
  93                 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
  94         }
  95         ClearPageUptodate(page);
  96 }
  97
  98 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  99 {
 100         pgoff_t index = current_nat_addr(sbi, nid);
 101         return get_meta_page(sbi, index);
 102 }
 103
 104 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 105 {
 106         struct page *src_page;
 107         struct page *dst_page;
 108         pgoff_t src_off;
 109         pgoff_t dst_off;
 110         void *src_addr;
 111         void *dst_addr;
 112         struct f2fs_nm_info *nm_i = NM_I(sbi);
 113
 114         src_off = current_nat_addr(sbi, nid);
 115         dst_off = next_nat_addr(sbi, src_off);
 116
 117         /* get current nat block page with lock */
 118         src_page = get_meta_page(sbi, src_off);
 119         dst_page = grab_meta_page(sbi, dst_off);
 120         f2fs_bug_on(sbi, PageDirty(src_page));
 121
 122         src_addr = page_address(src_page);
 123         dst_addr = page_address(dst_page);
 124         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
 125         set_page_dirty(dst_page);
 126         f2fs_put_page(src_page, 1);
 127
 128         set_to_next_nat(nm_i, nid);
 129
 130         return dst_page;
 131 }
 132
 133 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 134 {
 135         return radix_tree_lookup(&nm_i->nat_root, n);
 136 }
 137
 138 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 139                 nid_t start, unsigned int nr, struct nat_entry **ep)
 140 {
 141         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 142 }
 143
 144 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 145 {
 146         list_del(&e->list);
 147         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 148         nm_i->nat_cnt--;
 149         kmem_cache_free(nat_entry_slab, e);
 150 }
 151
 152 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 153                                                 struct nat_entry *ne)
 154 {
 155         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 156         struct nat_entry_set *head;
 157
 158         if (get_nat_flag(ne, IS_DIRTY))
 159                 return;
 160
 161         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 162         if (!head) {
 163                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 164
 165                 INIT_LIST_HEAD(&head->entry_list);
 166                 INIT_LIST_HEAD(&head->set_list);
 167                 head->set = set;
 168                 head->entry_cnt = 0;
 169                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 170         }
 171         list_move_tail(&ne->list, &head->entry_list);
 172         nm_i->dirty_nat_cnt++;
 173         head->entry_cnt++;
 174         set_nat_flag(ne, IS_DIRTY, true);
 175 }
 176
 177 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 178                                                 struct nat_entry *ne)
 179 {
 180         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 181         struct nat_entry_set *head;
 182
 183         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 184         if (head) {
 185                 list_move_tail(&ne->list, &nm_i->nat_entries);
 186                 set_nat_flag(ne, IS_DIRTY, false);
 187                 head->entry_cnt--;
 188                 nm_i->dirty_nat_cnt--;
 189         }
 190 }
 191
 192 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 193                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 194 {
 195         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 196                                                         start, nr);
 197 }
 198
 199 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 200 {
 201         struct f2fs_nm_info *nm_i = NM_I(sbi);
 202         struct nat_entry *e;
 203         bool need = false;
 204
 205         down_read(&nm_i->nat_tree_lock);
 206         e = __lookup_nat_cache(nm_i, nid);
 207         if (e) {
 208                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 209                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 210                         need = true;
 211         }
 212         up_read(&nm_i->nat_tree_lock);
 213         return need;
 214 }
 215
 216 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 217 {
 218         struct f2fs_nm_info *nm_i = NM_I(sbi);
 219         struct nat_entry *e;
 220         bool is_cp = true;
 221
 222         down_read(&nm_i->nat_tree_lock);
 223         e = __lookup_nat_cache(nm_i, nid);
 224         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 225                 is_cp = false;
 226         up_read(&nm_i->nat_tree_lock);
 227         return is_cp;
 228 }
 229
 230 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 231 {
 232         struct f2fs_nm_info *nm_i = NM_I(sbi);
 233         struct nat_entry *e;
 234         bool need_update = true;
 235
 236         down_read(&nm_i->nat_tree_lock);
 237         e = __lookup_nat_cache(nm_i, ino);
 238         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 239                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 240                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 241                 need_update = false;
 242         up_read(&nm_i->nat_tree_lock);
 243         return need_update;
 244 }
 245
 246 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 247 {
 248         struct nat_entry *new;
 249
 250         new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
 251         f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
 252         memset(new, 0, sizeof(struct nat_entry));
 253         nat_set_nid(new, nid);
 254         nat_reset_flag(new);
 255         list_add_tail(&new->list, &nm_i->nat_entries);
 256         nm_i->nat_cnt++;
 257         return new;
 258 }
 259
 260 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 261                                                 struct f2fs_nat_entry *ne)
 262 {
 263         struct nat_entry *e;
 264
 265         e = __lookup_nat_cache(nm_i, nid);
 266         if (!e) {
 267                 e = grab_nat_entry(nm_i, nid);
 268                 node_info_from_raw_nat(&e->ni, ne);
 269         }
 270 }
 271
 272 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 273                         block_t new_blkaddr, bool fsync_done)
 274 {
 275         struct f2fs_nm_info *nm_i = NM_I(sbi);
 276         struct nat_entry *e;
 277
 278         down_write(&nm_i->nat_tree_lock);
 279         e = __lookup_nat_cache(nm_i, ni->nid);
 280         if (!e) {
 281                 e = grab_nat_entry(nm_i, ni->nid);
 282                 copy_node_info(&e->ni, ni);
 283                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 284         } else if (new_blkaddr == NEW_ADDR) {
 285                 /*
 286                  * when nid is reallocated,
 287                  * previous nat entry can be remained in nat cache.
 288                  * So, reinitialize it with new information.
 289                  */
 290                 copy_node_info(&e->ni, ni);
 291                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 292         }
 293
 294         /* sanity check */
 295         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 296         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 297                         new_blkaddr == NULL_ADDR);
 298         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 299                         new_blkaddr == NEW_ADDR);
 300         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 301                         nat_get_blkaddr(e) != NULL_ADDR &&
 302                         new_blkaddr == NEW_ADDR);
 303
 304         /* increment version no as node is removed */
 305         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 306                 unsigned char version = nat_get_version(e);
 307                 nat_set_version(e, inc_node_version(version));
 308
 309                 /* in order to reuse the nid */
 310                 if (nm_i->next_scan_nid > ni->nid)
 311                         nm_i->next_scan_nid = ni->nid;
 312         }
 313
 314         /* change address */
 315         nat_set_blkaddr(e, new_blkaddr);
 316         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 317                 set_nat_flag(e, IS_CHECKPOINTED, false);
 318         __set_nat_cache_dirty(nm_i, e);
 319
 320         /* update fsync_mark if its inode nat entry is still alive */
 321         if (ni->nid != ni->ino)
 322                 e = __lookup_nat_cache(nm_i, ni->ino);
 323         if (e) {
 324                 if (fsync_done && ni->nid == ni->ino)
 325                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 326                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 327         }
 328         up_write(&nm_i->nat_tree_lock);
 329 }
 330
 331 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 332 {
 333         struct f2fs_nm_info *nm_i = NM_I(sbi);
 334         int nr = nr_shrink;
 335
 336         if (!down_write_trylock(&nm_i->nat_tree_lock))
 337                 return 0;
 338
 339         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 340                 struct nat_entry *ne;
 341                 ne = list_first_entry(&nm_i->nat_entries,
 342                                         struct nat_entry, list);
 343                 __del_from_nat_cache(nm_i, ne);
 344                 nr_shrink--;
 345         }
 346         up_write(&nm_i->nat_tree_lock);
 347         return nr - nr_shrink;
 348 }
 349
 350 /*
 351  * This function always returns success
 352  */
 353 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 354 {
 355         struct f2fs_nm_info *nm_i = NM_I(sbi);
 356         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 357         struct f2fs_summary_block *sum = curseg->sum_blk;
 358         nid_t start_nid = START_NID(nid);
 359         struct f2fs_nat_block *nat_blk;
 360         struct page *page = NULL;
 361         struct f2fs_nat_entry ne;
 362         struct nat_entry *e;
 363         int i;
 364
 365         ni->nid = nid;
 366
 367         /* Check nat cache */
 368         down_read(&nm_i->nat_tree_lock);
 369         e = __lookup_nat_cache(nm_i, nid);
 370         if (e) {
 371                 ni->ino = nat_get_ino(e);
 372                 ni->blk_addr = nat_get_blkaddr(e);
 373                 ni->version = nat_get_version(e);
 374         }
 375         up_read(&nm_i->nat_tree_lock);
 376         if (e)
 377                 return;
 378
 379         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 380
 381         down_write(&nm_i->nat_tree_lock);
 382
 383         /* Check current segment summary */
 384         mutex_lock(&curseg->curseg_mutex);
 385         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 386         if (i >= 0) {
 387                 ne = nat_in_journal(sum, i);
 388                 node_info_from_raw_nat(ni, &ne);
 389         }
 390         mutex_unlock(&curseg->curseg_mutex);
 391         if (i >= 0)
 392                 goto cache;
 393
 394         /* Fill node_info from nat page */
 395         page = get_current_nat_page(sbi, start_nid);
 396         nat_blk = (struct f2fs_nat_block *)page_address(page);
 397         ne = nat_blk->entries[nid - start_nid];
 398         node_info_from_raw_nat(ni, &ne);
 399         f2fs_put_page(page, 1);
 400 cache:
 401         /* cache nat entry */
 402         cache_nat_entry(NM_I(sbi), nid, &ne);
 403         up_write(&nm_i->nat_tree_lock);
 404 }
 405
 406 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 407 {
 408         const long direct_index = ADDRS_PER_INODE(dn->inode);
 409         const long direct_blks = ADDRS_PER_BLOCK;
 410         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 411         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 412         int cur_level = dn->cur_level;
 413         int max_level = dn->max_level;
 414         pgoff_t base = 0;
 415
 416         if (!dn->max_level)
 417                 return pgofs + 1;
 418
 419         while (max_level-- > cur_level)
 420                 skipped_unit *= NIDS_PER_BLOCK;
 421
 422         switch (dn->max_level) {
 423         case 3:
 424                 base += 2 * indirect_blks;
 425         case 2:
 426                 base += 2 * direct_blks;
 427         case 1:
 428                 base += direct_index;
 429                 break;
 430         default:
 431                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 432         }
 433
 434         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 435 }
 436
 437 /*
 438  * The maximum depth is four.
 439  * Offset[0] will have raw inode offset.
 440  */
 441 static int get_node_path(struct inode *inode, long block,
 442                                 int offset[4], unsigned int noffset[4])
 443 {
 444         const long direct_index = ADDRS_PER_INODE(inode);
 445         const long direct_blks = ADDRS_PER_BLOCK;
 446         const long dptrs_per_blk = NIDS_PER_BLOCK;
 447         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 448         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 449         int n = 0;
 450         int level = 0;
 451
 452         noffset[0] = 0;
 453
 454         if (block < direct_index) {
 455                 offset[n] = block;
 456                 goto got;
 457         }
 458         block -= direct_index;
 459         if (block < direct_blks) {
 460                 offset[n++] = NODE_DIR1_BLOCK;
 461                 noffset[n] = 1;
 462                 offset[n] = block;
 463                 level = 1;
 464                 goto got;
 465         }
 466         block -= direct_blks;
 467         if (block < direct_blks) {
 468                 offset[n++] = NODE_DIR2_BLOCK;
 469                 noffset[n] = 2;
 470                 offset[n] = block;
 471                 level = 1;
 472                 goto got;
 473         }
 474         block -= direct_blks;
 475         if (block < indirect_blks) {
 476                 offset[n++] = NODE_IND1_BLOCK;
 477                 noffset[n] = 3;
 478                 offset[n++] = block / direct_blks;
 479                 noffset[n] = 4 + offset[n - 1];
 480                 offset[n] = block % direct_blks;
 481                 level = 2;
 482                 goto got;
 483         }
 484         block -= indirect_blks;
 485         if (block < indirect_blks) {
 486                 offset[n++] = NODE_IND2_BLOCK;
 487                 noffset[n] = 4 + dptrs_per_blk;
 488                 offset[n++] = block / direct_blks;
 489                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 490                 offset[n] = block % direct_blks;
 491                 level = 2;
 492                 goto got;
 493         }
 494         block -= indirect_blks;
 495         if (block < dindirect_blks) {
 496                 offset[n++] = NODE_DIND_BLOCK;
 497                 noffset[n] = 5 + (dptrs_per_blk * 2);
 498                 offset[n++] = block / indirect_blks;
 499                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 500                               offset[n - 1] * (dptrs_per_blk + 1);
 501                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 502                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 503                               offset[n - 2] * (dptrs_per_blk + 1) +
 504                               offset[n - 1];
 505                 offset[n] = block % direct_blks;
 506                 level = 3;
 507                 goto got;
 508         } else {
 509                 BUG();
 510         }
 511 got:
 512         return level;
 513 }
 514
 515 /*
 516  * Caller should call f2fs_put_dnode(dn).
 517  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 518  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 519  * In the case of RDONLY_NODE, we don't need to care about mutex.
 520  */
 521 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 522 {
 523         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 524         struct page *npage[4];
 525         struct page *parent = NULL;
 526         int offset[4];
 527         unsigned int noffset[4];
 528         nid_t nids[4];
 529         int level, i = 0;
 530         int err = 0;
 531
 532         level = get_node_path(dn->inode, index, offset, noffset);
 533
 534         nids[0] = dn->inode->i_ino;
 535         npage[0] = dn->inode_page;
 536
 537         if (!npage[0]) {
 538                 npage[0] = get_node_page(sbi, nids[0]);
 539                 if (IS_ERR(npage[0]))
 540                         return PTR_ERR(npage[0]);
 541         }
 542
 543         /* if inline_data is set, should not report any block indices */
 544         if (f2fs_has_inline_data(dn->inode) && index) {
 545                 err = -ENOENT;
 546                 f2fs_put_page(npage[0], 1);
 547                 goto release_out;
 548         }
 549
 550         parent = npage[0];
 551         if (level != 0)
 552                 nids[1] = get_nid(parent, offset[0], true);
 553         dn->inode_page = npage[0];
 554         dn->inode_page_locked = true;
 555
 556         /* get indirect or direct nodes */
 557         for (i = 1; i <= level; i++) {
 558                 bool done = false;
 559
 560                 if (!nids[i] && mode == ALLOC_NODE) {
 561                         /* alloc new node */
 562                         if (!alloc_nid(sbi, &(nids[i]))) {
 563                                 err = -ENOSPC;
 564                                 goto release_pages;
 565                         }
 566
 567                         dn->nid = nids[i];
 568                         npage[i] = new_node_page(dn, noffset[i], NULL);
 569                         if (IS_ERR(npage[i])) {
 570                                 alloc_nid_failed(sbi, nids[i]);
 571                                 err = PTR_ERR(npage[i]);
 572                                 goto release_pages;
 573                         }
 574
 575                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 576                         alloc_nid_done(sbi, nids[i]);
 577                         done = true;
 578                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 579                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 580                         if (IS_ERR(npage[i])) {
 581                                 err = PTR_ERR(npage[i]);
 582                                 goto release_pages;
 583                         }
 584                         done = true;
 585                 }
 586                 if (i == 1) {
 587                         dn->inode_page_locked = false;
 588                         unlock_page(parent);
 589                 } else {
 590                         f2fs_put_page(parent, 1);
 591                 }
 592
 593                 if (!done) {
 594                         npage[i] = get_node_page(sbi, nids[i]);
 595                         if (IS_ERR(npage[i])) {
 596                                 err = PTR_ERR(npage[i]);
 597                                 f2fs_put_page(npage[0], 0);
 598                                 goto release_out;
 599                         }
 600                 }
 601                 if (i < level) {
 602                         parent = npage[i];
 603                         nids[i + 1] = get_nid(parent, offset[i], false);
 604                 }
 605         }
 606         dn->nid = nids[level];
 607         dn->ofs_in_node = offset[level];
 608         dn->node_page = npage[level];
 609         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 610         return 0;
 611
 612 release_pages:
 613         f2fs_put_page(parent, 1);
 614         if (i > 1)
 615                 f2fs_put_page(npage[0], 0);
 616 release_out:
 617         dn->inode_page = NULL;
 618         dn->node_page = NULL;
 619         if (err == -ENOENT) {
 620                 dn->cur_level = i;
 621                 dn->max_level = level;
 622         }
 623         return err;
 624 }
 625
 626 static void truncate_node(struct dnode_of_data *dn)
 627 {
 628         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 629         struct node_info ni;
 630
 631         get_node_info(sbi, dn->nid, &ni);
 632         if (dn->inode->i_blocks == 0) {
 633                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 634                 goto invalidate;
 635         }
 636         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 637
 638         /* Deallocate node address */
 639         invalidate_blocks(sbi, ni.blk_addr);
 640         dec_valid_node_count(sbi, dn->inode);
 641         set_node_addr(sbi, &ni, NULL_ADDR, false);
 642
 643         if (dn->nid == dn->inode->i_ino) {
 644                 remove_orphan_inode(sbi, dn->nid);
 645                 dec_valid_inode_count(sbi);
 646         } else {
 647                 sync_inode_page(dn);
 648         }
 649 invalidate:
 650         clear_node_page_dirty(dn->node_page);
 651         set_sbi_flag(sbi, SBI_IS_DIRTY);
 652
 653         f2fs_put_page(dn->node_page, 1);
 654
 655         invalidate_mapping_pages(NODE_MAPPING(sbi),
 656                         dn->node_page->index, dn->node_page->index);
 657
 658         dn->node_page = NULL;
 659         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 660 }
 661
 662 static int truncate_dnode(struct dnode_of_data *dn)
 663 {
 664         struct page *page;
 665
 666         if (dn->nid == 0)
 667                 return 1;
 668
 669         /* get direct node */
 670         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 671         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 672                 return 1;
 673         else if (IS_ERR(page))
 674                 return PTR_ERR(page);
 675
 676         /* Make dnode_of_data for parameter */
 677         dn->node_page = page;
 678         dn->ofs_in_node = 0;
 679         truncate_data_blocks(dn);
 680         truncate_node(dn);
 681         return 1;
 682 }
 683
 684 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 685                                                 int ofs, int depth)
 686 {
 687         struct dnode_of_data rdn = *dn;
 688         struct page *page;
 689         struct f2fs_node *rn;
 690         nid_t child_nid;
 691         unsigned int child_nofs;
 692         int freed = 0;
 693         int i, ret;
 694
 695         if (dn->nid == 0)
 696                 return NIDS_PER_BLOCK + 1;
 697
 698         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 699
 700         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 701         if (IS_ERR(page)) {
 702                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 703                 return PTR_ERR(page);
 704         }
 705
 706         rn = F2FS_NODE(page);
 707         if (depth < 3) {
 708                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 709                         child_nid = le32_to_cpu(rn->in.nid[i]);
 710                         if (child_nid == 0)
 711                                 continue;
 712                         rdn.nid = child_nid;
 713                         ret = truncate_dnode(&rdn);
 714                         if (ret < 0)
 715                                 goto out_err;
 716                         if (set_nid(page, i, 0, false))
 717                                 dn->node_changed = true;
 718                 }
 719         } else {
 720                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 721                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 722                         child_nid = le32_to_cpu(rn->in.nid[i]);
 723                         if (child_nid == 0) {
 724                                 child_nofs += NIDS_PER_BLOCK + 1;
 725                                 continue;
 726                         }
 727                         rdn.nid = child_nid;
 728                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 729                         if (ret == (NIDS_PER_BLOCK + 1)) {
 730                                 if (set_nid(page, i, 0, false))
 731                                         dn->node_changed = true;
 732                                 child_nofs += ret;
 733                         } else if (ret < 0 && ret != -ENOENT) {
 734                                 goto out_err;
 735                         }
 736                 }
 737                 freed = child_nofs;
 738         }
 739
 740         if (!ofs) {
 741                 /* remove current indirect node */
 742                 dn->node_page = page;
 743                 truncate_node(dn);
 744                 freed++;
 745         } else {
 746                 f2fs_put_page(page, 1);
 747         }
 748         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 749         return freed;
 750
 751 out_err:
 752         f2fs_put_page(page, 1);
 753         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 754         return ret;
 755 }
 756
 757 static int truncate_partial_nodes(struct dnode_of_data *dn,
 758                         struct f2fs_inode *ri, int *offset, int depth)
 759 {
 760         struct page *pages[2];
 761         nid_t nid[3];
 762         nid_t child_nid;
 763         int err = 0;
 764         int i;
 765         int idx = depth - 2;
 766
 767         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 768         if (!nid[0])
 769                 return 0;
 770
 771         /* get indirect nodes in the path */
 772         for (i = 0; i < idx + 1; i++) {
 773                 /* reference count'll be increased */
 774                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 775                 if (IS_ERR(pages[i])) {
 776                         err = PTR_ERR(pages[i]);
 777                         idx = i - 1;
 778                         goto fail;
 779                 }
 780                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 781         }
 782
 783         /* free direct nodes linked to a partial indirect node */
 784         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 785                 child_nid = get_nid(pages[idx], i, false);
 786                 if (!child_nid)
 787                         continue;
 788                 dn->nid = child_nid;
 789                 err = truncate_dnode(dn);
 790                 if (err < 0)
 791                         goto fail;
 792                 if (set_nid(pages[idx], i, 0, false))
 793                         dn->node_changed = true;
 794         }
 795
 796         if (offset[idx + 1] == 0) {
 797                 dn->node_page = pages[idx];
 798                 dn->nid = nid[idx];
 799                 truncate_node(dn);
 800         } else {
 801                 f2fs_put_page(pages[idx], 1);
 802         }
 803         offset[idx]++;
 804         offset[idx + 1] = 0;
 805         idx--;
 806 fail:
 807         for (i = idx; i >= 0; i--)
 808                 f2fs_put_page(pages[i], 1);
 809
 810         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 811
 812         return err;
 813 }
 814
 815 /*
 816  * All the block addresses of data and nodes should be nullified.
 817  */
 818 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 819 {
 820         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 821         int err = 0, cont = 1;
 822         int level, offset[4], noffset[4];
 823         unsigned int nofs = 0;
 824         struct f2fs_inode *ri;
 825         struct dnode_of_data dn;
 826         struct page *page;
 827
 828         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 829
 830         level = get_node_path(inode, from, offset, noffset);
 831 restart:
 832         page = get_node_page(sbi, inode->i_ino);
 833         if (IS_ERR(page)) {
 834                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 835                 return PTR_ERR(page);
 836         }
 837
 838         set_new_dnode(&dn, inode, page, NULL, 0);
 839         unlock_page(page);
 840
 841         ri = F2FS_INODE(page);
 842         switch (level) {
 843         case 0:
 844         case 1:
 845                 nofs = noffset[1];
 846                 break;
 847         case 2:
 848                 nofs = noffset[1];
 849                 if (!offset[level - 1])
 850                         goto skip_partial;
 851                 err = truncate_partial_nodes(&dn, ri, offset, level);
 852                 if (err < 0 && err != -ENOENT)
 853                         goto fail;
 854                 nofs += 1 + NIDS_PER_BLOCK;
 855                 break;
 856         case 3:
 857                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 858                 if (!offset[level - 1])
 859                         goto skip_partial;
 860                 err = truncate_partial_nodes(&dn, ri, offset, level);
 861                 if (err < 0 && err != -ENOENT)
 862                         goto fail;
 863                 break;
 864         default:
 865                 BUG();
 866         }
 867
 868 skip_partial:
 869         while (cont) {
 870                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 871                 switch (offset[0]) {
 872                 case NODE_DIR1_BLOCK:
 873                 case NODE_DIR2_BLOCK:
 874                         err = truncate_dnode(&dn);
 875                         break;
 876
 877                 case NODE_IND1_BLOCK:
 878                 case NODE_IND2_BLOCK:
 879                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 880                         break;
 881
 882                 case NODE_DIND_BLOCK:
 883                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 884                         cont = 0;
 885                         break;
 886
 887                 default:
 888                         BUG();
 889                 }
 890                 if (err < 0 && err != -ENOENT)
 891                         goto fail;
 892                 if (offset[1] == 0 &&
 893                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 894                         lock_page(page);
 895                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 896                                 f2fs_put_page(page, 1);
 897                                 goto restart;
 898                         }
 899                         f2fs_wait_on_page_writeback(page, NODE, true);
 900                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 901                         set_page_dirty(page);
 902                         unlock_page(page);
 903                 }
 904                 offset[1] = 0;
 905                 offset[0]++;
 906                 nofs += err;
 907         }
 908 fail:
 909         f2fs_put_page(page, 0);
 910         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 911         return err > 0 ? 0 : err;
 912 }
 913
 914 int truncate_xattr_node(struct inode *inode, struct page *page)
 915 {
 916         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 917         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 918         struct dnode_of_data dn;
 919         struct page *npage;
 920
 921         if (!nid)
 922                 return 0;
 923
 924         npage = get_node_page(sbi, nid);
 925         if (IS_ERR(npage))
 926                 return PTR_ERR(npage);
 927
 928         F2FS_I(inode)->i_xattr_nid = 0;
 929
 930         /* need to do checkpoint during fsync */
 931         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 932
 933         set_new_dnode(&dn, inode, page, npage, nid);
 934
 935         if (page)
 936                 dn.inode_page_locked = true;
 937         truncate_node(&dn);
 938         return 0;
 939 }
 940
 941 /*
 942  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 943  * f2fs_unlock_op().
 944  */
 945 int remove_inode_page(struct inode *inode)
 946 {
 947         struct dnode_of_data dn;
 948         int err;
 949
 950         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 951         err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
 952         if (err)
 953                 return err;
 954
 955         err = truncate_xattr_node(inode, dn.inode_page);
 956         if (err) {
 957                 f2fs_put_dnode(&dn);
 958                 return err;
 959         }
 960
 961         /* remove potential inline_data blocks */
 962         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 963                                 S_ISLNK(inode->i_mode))
 964                 truncate_data_blocks_range(&dn, 1);
 965
 966         /* 0 is possible, after f2fs_new_inode() has failed */
 967         f2fs_bug_on(F2FS_I_SB(inode),
 968                         inode->i_blocks != 0 && inode->i_blocks != 1);
 969
 970         /* will put inode & node pages */
 971         truncate_node(&dn);
 972         return 0;
 973 }
 974
 975 struct page *new_inode_page(struct inode *inode)
 976 {
 977         struct dnode_of_data dn;
 978
 979         /* allocate inode page for new inode */
 980         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 981
 982         /* caller should f2fs_put_page(page, 1); */
 983         return new_node_page(&dn, 0, NULL);
 984 }
 985
 986 struct page *new_node_page(struct dnode_of_data *dn,
 987                                 unsigned int ofs, struct page *ipage)
 988 {
 989         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 990         struct node_info old_ni, new_ni;
 991         struct page *page;
 992         int err;
 993
 994         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 995                 return ERR_PTR(-EPERM);
 996
 997         page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
 998         if (!page)
 999                 return ERR_PTR(-ENOMEM);
1000
1001         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1002                 err = -ENOSPC;
1003                 goto fail;
1004         }
1005
1006         get_node_info(sbi, dn->nid, &old_ni);
1007
1008         /* Reinitialize old_ni with new node page */
1009         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1010         new_ni = old_ni;
1011         new_ni.ino = dn->inode->i_ino;
1012         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1013
1014         f2fs_wait_on_page_writeback(page, NODE, true);
1015         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1016         set_cold_node(dn->inode, page);
1017         SetPageUptodate(page);
1018         if (set_page_dirty(page))
1019                 dn->node_changed = true;
1020
1021         if (f2fs_has_xattr_block(ofs))
1022                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
1023
1024         dn->node_page = page;
1025         if (ipage)
1026                 update_inode(dn->inode, ipage);
1027         else
1028                 sync_inode_page(dn);
1029         if (ofs == 0)
1030                 inc_valid_inode_count(sbi);
1031
1032         return page;
1033
1034 fail:
1035         clear_node_page_dirty(page);
1036         f2fs_put_page(page, 1);
1037         return ERR_PTR(err);
1038 }
1039
1040 /*
1041  * Caller should do after getting the following values.
1042  * 0: f2fs_put_page(page, 0)
1043  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1044  */
1045 static int read_node_page(struct page *page, int rw)
1046 {
1047         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1048         struct node_info ni;
1049         struct f2fs_io_info fio = {
1050                 .sbi = sbi,
1051                 .type = NODE,
1052                 .rw = rw,
1053                 .page = page,
1054                 .encrypted_page = NULL,
1055         };
1056
1057         get_node_info(sbi, page->index, &ni);
1058
1059         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1060                 ClearPageUptodate(page);
1061                 return -ENOENT;
1062         }
1063
1064         if (PageUptodate(page))
1065                 return LOCKED_PAGE;
1066
1067         fio.blk_addr = ni.blk_addr;
1068         return f2fs_submit_page_bio(&fio);
1069 }
1070
1071 /*
1072  * Readahead a node page
1073  */
1074 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1075 {
1076         struct page *apage;
1077         int err;
1078
1079         if (!nid)
1080                 return;
1081         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1082
1083         apage = find_get_page(NODE_MAPPING(sbi), nid);
1084         if (apage && PageUptodate(apage)) {
1085                 f2fs_put_page(apage, 0);
1086                 return;
1087         }
1088         f2fs_put_page(apage, 0);
1089
1090         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1091         if (!apage)
1092                 return;
1093
1094         err = read_node_page(apage, READA);
1095         f2fs_put_page(apage, err ? 1 : 0);
1096 }
1097
1098 /*
1099  * readahead MAX_RA_NODE number of node pages.
1100  */
1101 void ra_node_pages(struct page *parent, int start)
1102 {
1103         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1104         struct blk_plug plug;
1105         int i, end;
1106         nid_t nid;
1107
1108         blk_start_plug(&plug);
1109
1110         /* Then, try readahead for siblings of the desired node */
1111         end = start + MAX_RA_NODE;
1112         end = min(end, NIDS_PER_BLOCK);
1113         for (i = start; i < end; i++) {
1114                 nid = get_nid(parent, i, false);
1115                 ra_node_page(sbi, nid);
1116         }
1117
1118         blk_finish_plug(&plug);
1119 }
1120
1121 struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1122                                         struct page *parent, int start)
1123 {
1124         struct page *page;
1125         int err;
1126
1127         if (!nid)
1128                 return ERR_PTR(-ENOENT);
1129         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1130 repeat:
1131         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1132         if (!page)
1133                 return ERR_PTR(-ENOMEM);
1134
1135         err = read_node_page(page, READ_SYNC);
1136         if (err < 0) {
1137                 f2fs_put_page(page, 1);
1138                 return ERR_PTR(err);
1139         } else if (err == LOCKED_PAGE) {
1140                 goto page_hit;
1141         }
1142
1143         if (parent)
1144                 ra_node_pages(parent, start + 1);
1145
1146         lock_page(page);
1147
1148         if (unlikely(!PageUptodate(page))) {
1149                 f2fs_put_page(page, 1);
1150                 return ERR_PTR(-EIO);
1151         }
1152         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1153                 f2fs_put_page(page, 1);
1154                 goto repeat;
1155         }
1156 page_hit:
1157         f2fs_bug_on(sbi, nid != nid_of_node(page));
1158         return page;
1159 }
1160
1161 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1162 {
1163         return __get_node_page(sbi, nid, NULL, 0);
1164 }
1165
1166 struct page *get_node_page_ra(struct page *parent, int start)
1167 {
1168         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1169         nid_t nid = get_nid(parent, start, false);
1170
1171         return __get_node_page(sbi, nid, parent, start);
1172 }
1173
1174 void sync_inode_page(struct dnode_of_data *dn)
1175 {
1176         int ret = 0;
1177
1178         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1179                 ret = update_inode(dn->inode, dn->node_page);
1180         } else if (dn->inode_page) {
1181                 if (!dn->inode_page_locked)
1182                         lock_page(dn->inode_page);
1183                 ret = update_inode(dn->inode, dn->inode_page);
1184                 if (!dn->inode_page_locked)
1185                         unlock_page(dn->inode_page);
1186         } else {
1187                 ret = update_inode_page(dn->inode);
1188         }
1189         dn->node_changed = ret ? true: false;
1190 }
1191
1192 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1193 {
1194         struct inode *inode;
1195         struct page *page;
1196
1197         /* should flush inline_data before evict_inode */
1198         inode = ilookup(sbi->sb, ino);
1199         if (!inode)
1200                 return;
1201
1202         page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1203         if (!page)
1204                 goto iput_out;
1205
1206         if (!PageUptodate(page))
1207                 goto page_out;
1208
1209         if (!PageDirty(page))
1210                 goto page_out;
1211
1212         if (!clear_page_dirty_for_io(page))
1213                 goto page_out;
1214
1215         if (!f2fs_write_inline_data(inode, page))
1216                 inode_dec_dirty_pages(inode);
1217         else
1218                 set_page_dirty(page);
1219 page_out:
1220         f2fs_put_page(page, 1);
1221 iput_out:
1222         iput(inode);
1223 }
1224
1225 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1226                                         struct writeback_control *wbc)
1227 {
1228         pgoff_t index, end;
1229         struct pagevec pvec;
1230         int step = ino ? 2 : 0;
1231         int nwritten = 0, wrote = 0;
1232
1233         pagevec_init(&pvec, 0);
1234
1235 next_step:
1236         index = 0;
1237         end = LONG_MAX;
1238
1239         while (index <= end) {
1240                 int i, nr_pages;
1241                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1242                                 PAGECACHE_TAG_DIRTY,
1243                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1244                 if (nr_pages == 0)
1245                         break;
1246
1247                 for (i = 0; i < nr_pages; i++) {
1248                         struct page *page = pvec.pages[i];
1249
1250                         if (unlikely(f2fs_cp_error(sbi))) {
1251                                 pagevec_release(&pvec);
1252                                 return -EIO;
1253                         }
1254
1255                         /*
1256                          * flushing sequence with step:
1257                          * 0. indirect nodes
1258                          * 1. dentry dnodes
1259                          * 2. file dnodes
1260                          */
1261                         if (step == 0 && IS_DNODE(page))
1262                                 continue;
1263                         if (step == 1 && (!IS_DNODE(page) ||
1264                                                 is_cold_node(page)))
1265                                 continue;
1266                         if (step == 2 && (!IS_DNODE(page) ||
1267                                                 !is_cold_node(page)))
1268                                 continue;
1269
1270                         /*
1271                          * If an fsync mode,
1272                          * we should not skip writing node pages.
1273                          */
1274                         if (ino && ino_of_node(page) == ino)
1275                                 lock_page(page);
1276                         else if (!trylock_page(page))
1277                                 continue;
1278
1279                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1280 continue_unlock:
1281                                 unlock_page(page);
1282                                 continue;
1283                         }
1284                         if (ino && ino_of_node(page) != ino)
1285                                 goto continue_unlock;
1286
1287                         if (!PageDirty(page)) {
1288                                 /* someone wrote it for us */
1289                                 goto continue_unlock;
1290                         }
1291
1292                         /* flush inline_data */
1293                         if (!ino && is_inline_node(page)) {
1294                                 clear_inline_node(page);
1295                                 unlock_page(page);
1296                                 flush_inline_data(sbi, ino_of_node(page));
1297                                 continue;
1298                         }
1299
1300                         f2fs_wait_on_page_writeback(page, NODE, true);
1301
1302                         BUG_ON(PageWriteback(page));
1303                         if (!clear_page_dirty_for_io(page))
1304                                 goto continue_unlock;
1305
1306                         /* called by fsync() */
1307                         if (ino && IS_DNODE(page)) {
1308                                 set_fsync_mark(page, 1);
1309                                 if (IS_INODE(page))
1310                                         set_dentry_mark(page,
1311                                                 need_dentry_mark(sbi, ino));
1312                                 nwritten++;
1313                         } else {
1314                                 set_fsync_mark(page, 0);
1315                                 set_dentry_mark(page, 0);
1316                         }
1317
1318                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1319                                 unlock_page(page);
1320                         else
1321                                 wrote++;
1322
1323                         if (--wbc->nr_to_write == 0)
1324                                 break;
1325                 }
1326                 pagevec_release(&pvec);
1327                 cond_resched();
1328
1329                 if (wbc->nr_to_write == 0) {
1330                         step = 2;
1331                         break;
1332                 }
1333         }
1334
1335         if (step < 2) {
1336                 step++;
1337                 goto next_step;
1338         }
1339
1340         if (wrote) {
1341                 if (ino)
1342                         f2fs_submit_merged_bio_cond(sbi, NULL, NULL,
1343                                                         ino, NODE, WRITE);
1344                 else
1345                         f2fs_submit_merged_bio(sbi, NODE, WRITE);
1346         }
1347         return nwritten;
1348 }
1349
1350 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1351 {
1352         pgoff_t index = 0, end = LONG_MAX;
1353         struct pagevec pvec;
1354         int ret2 = 0, ret = 0;
1355
1356         pagevec_init(&pvec, 0);
1357
1358         while (index <= end) {
1359                 int i, nr_pages;
1360                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1361                                 PAGECACHE_TAG_WRITEBACK,
1362                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1363                 if (nr_pages == 0)
1364                         break;
1365
1366                 for (i = 0; i < nr_pages; i++) {
1367                         struct page *page = pvec.pages[i];
1368
1369                         /* until radix tree lookup accepts end_index */
1370                         if (unlikely(page->index > end))
1371                                 continue;
1372
1373                         if (ino && ino_of_node(page) == ino) {
1374                                 f2fs_wait_on_page_writeback(page, NODE, true);
1375                                 if (TestClearPageError(page))
1376                                         ret = -EIO;
1377                         }
1378                 }
1379                 pagevec_release(&pvec);
1380                 cond_resched();
1381         }
1382
1383         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1384                 ret2 = -ENOSPC;
1385         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1386                 ret2 = -EIO;
1387         if (!ret)
1388                 ret = ret2;
1389         return ret;
1390 }
1391
1392 static int f2fs_write_node_page(struct page *page,
1393                                 struct writeback_control *wbc)
1394 {
1395         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1396         nid_t nid;
1397         struct node_info ni;
1398         struct f2fs_io_info fio = {
1399                 .sbi = sbi,
1400                 .type = NODE,
1401                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1402                 .page = page,
1403                 .encrypted_page = NULL,
1404         };
1405
1406         trace_f2fs_writepage(page, NODE);
1407
1408         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1409                 goto redirty_out;
1410         if (unlikely(f2fs_cp_error(sbi)))
1411                 goto redirty_out;
1412
1413         /* get old block addr of this node page */
1414         nid = nid_of_node(page);
1415         f2fs_bug_on(sbi, page->index != nid);
1416
1417         if (wbc->for_reclaim) {
1418                 if (!down_read_trylock(&sbi->node_write))
1419                         goto redirty_out;
1420         } else {
1421                 down_read(&sbi->node_write);
1422         }
1423
1424         get_node_info(sbi, nid, &ni);
1425
1426         /* This page is already truncated */
1427         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1428                 ClearPageUptodate(page);
1429                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1430                 up_read(&sbi->node_write);
1431                 unlock_page(page);
1432                 return 0;
1433         }
1434
1435         set_page_writeback(page);
1436         fio.blk_addr = ni.blk_addr;
1437         write_node_page(nid, &fio);
1438         set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1439         dec_page_count(sbi, F2FS_DIRTY_NODES);
1440         up_read(&sbi->node_write);
1441
1442         if (wbc->for_reclaim)
1443                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1444
1445         unlock_page(page);
1446
1447         if (unlikely(f2fs_cp_error(sbi)))
1448                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1449
1450         return 0;
1451
1452 redirty_out:
1453         redirty_page_for_writepage(wbc, page);
1454         return AOP_WRITEPAGE_ACTIVATE;
1455 }
1456
1457 static int f2fs_write_node_pages(struct address_space *mapping,
1458                             struct writeback_control *wbc)
1459 {
1460         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1461         long diff;
1462
1463         /* balancing f2fs's metadata in background */
1464         f2fs_balance_fs_bg(sbi);
1465
1466         /* collect a number of dirty node pages and write together */
1467         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1468                 goto skip_write;
1469
1470         trace_f2fs_writepages(mapping->host, wbc, NODE);
1471
1472         diff = nr_pages_to_write(sbi, NODE, wbc);
1473         wbc->sync_mode = WB_SYNC_NONE;
1474         sync_node_pages(sbi, 0, wbc);
1475         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1476         return 0;
1477
1478 skip_write:
1479         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1480         trace_f2fs_writepages(mapping->host, wbc, NODE);
1481         return 0;
1482 }
1483
1484 static int f2fs_set_node_page_dirty(struct page *page)
1485 {
1486         trace_f2fs_set_page_dirty(page, NODE);
1487
1488         SetPageUptodate(page);
1489         if (!PageDirty(page)) {
1490                 __set_page_dirty_nobuffers(page);
1491                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1492                 SetPagePrivate(page);
1493                 f2fs_trace_pid(page);
1494                 return 1;
1495         }
1496         return 0;
1497 }
1498
1499 /*
1500  * Structure of the f2fs node operations
1501  */
1502 const struct address_space_operations f2fs_node_aops = {
1503         .writepage      = f2fs_write_node_page,
1504         .writepages     = f2fs_write_node_pages,
1505         .set_page_dirty = f2fs_set_node_page_dirty,
1506         .invalidatepage = f2fs_invalidate_page,
1507         .releasepage    = f2fs_release_page,
1508 };
1509
1510 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1511                                                 nid_t n)
1512 {
1513         return radix_tree_lookup(&nm_i->free_nid_root, n);
1514 }
1515
1516 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1517                                                 struct free_nid *i)
1518 {
1519         list_del(&i->list);
1520         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1521 }
1522
1523 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1524 {
1525         struct f2fs_nm_info *nm_i = NM_I(sbi);
1526         struct free_nid *i;
1527         struct nat_entry *ne;
1528         bool allocated = false;
1529
1530         if (!available_free_memory(sbi, FREE_NIDS))
1531                 return -1;
1532
1533         /* 0 nid should not be used */
1534         if (unlikely(nid == 0))
1535                 return 0;
1536
1537         if (build) {
1538                 /* do not add allocated nids */
1539                 ne = __lookup_nat_cache(nm_i, nid);
1540                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1541                                 nat_get_blkaddr(ne) != NULL_ADDR))
1542                         allocated = true;
1543                 if (allocated)
1544                         return 0;
1545         }
1546
1547         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1548         i->nid = nid;
1549         i->state = NID_NEW;
1550
1551         if (radix_tree_preload(GFP_NOFS)) {
1552                 kmem_cache_free(free_nid_slab, i);
1553                 return 0;
1554         }
1555
1556         spin_lock(&nm_i->free_nid_list_lock);
1557         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1558                 spin_unlock(&nm_i->free_nid_list_lock);
1559                 radix_tree_preload_end();
1560                 kmem_cache_free(free_nid_slab, i);
1561                 return 0;
1562         }
1563         list_add_tail(&i->list, &nm_i->free_nid_list);
1564         nm_i->fcnt++;
1565         spin_unlock(&nm_i->free_nid_list_lock);
1566         radix_tree_preload_end();
1567         return 1;
1568 }
1569
1570 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1571 {
1572         struct free_nid *i;
1573         bool need_free = false;
1574
1575         spin_lock(&nm_i->free_nid_list_lock);
1576         i = __lookup_free_nid_list(nm_i, nid);
1577         if (i && i->state == NID_NEW) {
1578                 __del_from_free_nid_list(nm_i, i);
1579                 nm_i->fcnt--;
1580                 need_free = true;
1581         }
1582         spin_unlock(&nm_i->free_nid_list_lock);
1583
1584         if (need_free)
1585                 kmem_cache_free(free_nid_slab, i);
1586 }
1587
1588 static void scan_nat_page(struct f2fs_sb_info *sbi,
1589                         struct page *nat_page, nid_t start_nid)
1590 {
1591         struct f2fs_nm_info *nm_i = NM_I(sbi);
1592         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1593         block_t blk_addr;
1594         int i;
1595
1596         i = start_nid % NAT_ENTRY_PER_BLOCK;
1597
1598         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1599
1600                 if (unlikely(start_nid >= nm_i->max_nid))
1601                         break;
1602
1603                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1604                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1605                 if (blk_addr == NULL_ADDR) {
1606                         if (add_free_nid(sbi, start_nid, true) < 0)
1607                                 break;
1608                 }
1609         }
1610 }
1611
1612 static void build_free_nids(struct f2fs_sb_info *sbi)
1613 {
1614         struct f2fs_nm_info *nm_i = NM_I(sbi);
1615         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1616         struct f2fs_summary_block *sum = curseg->sum_blk;
1617         int i = 0;
1618         nid_t nid = nm_i->next_scan_nid;
1619
1620         /* Enough entries */
1621         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1622                 return;
1623
1624         /* readahead nat pages to be scanned */
1625         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1626                                                         META_NAT, true);
1627
1628         down_read(&nm_i->nat_tree_lock);
1629
1630         while (1) {
1631                 struct page *page = get_current_nat_page(sbi, nid);
1632
1633                 scan_nat_page(sbi, page, nid);
1634                 f2fs_put_page(page, 1);
1635
1636                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1637                 if (unlikely(nid >= nm_i->max_nid))
1638                         nid = 0;
1639
1640                 if (++i >= FREE_NID_PAGES)
1641                         break;
1642         }
1643
1644         /* go to the next free nat pages to find free nids abundantly */
1645         nm_i->next_scan_nid = nid;
1646
1647         /* find free nids from current sum_pages */
1648         mutex_lock(&curseg->curseg_mutex);
1649         for (i = 0; i < nats_in_cursum(sum); i++) {
1650                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1651                 nid = le32_to_cpu(nid_in_journal(sum, i));
1652                 if (addr == NULL_ADDR)
1653                         add_free_nid(sbi, nid, true);
1654                 else
1655                         remove_free_nid(nm_i, nid);
1656         }
1657         mutex_unlock(&curseg->curseg_mutex);
1658         up_read(&nm_i->nat_tree_lock);
1659
1660         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1661                                         nm_i->ra_nid_pages, META_NAT, false);
1662 }
1663
1664 /*
1665  * If this function returns success, caller can obtain a new nid
1666  * from second parameter of this function.
1667  * The returned nid could be used ino as well as nid when inode is created.
1668  */
1669 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1670 {
1671         struct f2fs_nm_info *nm_i = NM_I(sbi);
1672         struct free_nid *i = NULL;
1673 retry:
1674         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1675                 return false;
1676
1677         spin_lock(&nm_i->free_nid_list_lock);
1678
1679         /* We should not use stale free nids created by build_free_nids */
1680         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1681                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1682                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1683                         if (i->state == NID_NEW)
1684                                 break;
1685
1686                 f2fs_bug_on(sbi, i->state != NID_NEW);
1687                 *nid = i->nid;
1688                 i->state = NID_ALLOC;
1689                 nm_i->fcnt--;
1690                 spin_unlock(&nm_i->free_nid_list_lock);
1691                 return true;
1692         }
1693         spin_unlock(&nm_i->free_nid_list_lock);
1694
1695         /* Let's scan nat pages and its caches to get free nids */
1696         mutex_lock(&nm_i->build_lock);
1697         build_free_nids(sbi);
1698         mutex_unlock(&nm_i->build_lock);
1699         goto retry;
1700 }
1701
1702 /*
1703  * alloc_nid() should be called prior to this function.
1704  */
1705 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1706 {
1707         struct f2fs_nm_info *nm_i = NM_I(sbi);
1708         struct free_nid *i;
1709
1710         spin_lock(&nm_i->free_nid_list_lock);
1711         i = __lookup_free_nid_list(nm_i, nid);
1712         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1713         __del_from_free_nid_list(nm_i, i);
1714         spin_unlock(&nm_i->free_nid_list_lock);
1715
1716         kmem_cache_free(free_nid_slab, i);
1717 }
1718
1719 /*
1720  * alloc_nid() should be called prior to this function.
1721  */
1722 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1723 {
1724         struct f2fs_nm_info *nm_i = NM_I(sbi);
1725         struct free_nid *i;
1726         bool need_free = false;
1727
1728         if (!nid)
1729                 return;
1730
1731         spin_lock(&nm_i->free_nid_list_lock);
1732         i = __lookup_free_nid_list(nm_i, nid);
1733         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1734         if (!available_free_memory(sbi, FREE_NIDS)) {
1735                 __del_from_free_nid_list(nm_i, i);
1736                 need_free = true;
1737         } else {
1738                 i->state = NID_NEW;
1739                 nm_i->fcnt++;
1740         }
1741         spin_unlock(&nm_i->free_nid_list_lock);
1742
1743         if (need_free)
1744                 kmem_cache_free(free_nid_slab, i);
1745 }
1746
1747 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1748 {
1749         struct f2fs_nm_info *nm_i = NM_I(sbi);
1750         struct free_nid *i, *next;
1751         int nr = nr_shrink;
1752
1753         if (!mutex_trylock(&nm_i->build_lock))
1754                 return 0;
1755
1756         spin_lock(&nm_i->free_nid_list_lock);
1757         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1758                 if (nr_shrink <= 0 || nm_i->fcnt <= NAT_ENTRY_PER_BLOCK)
1759                         break;
1760                 if (i->state == NID_ALLOC)
1761                         continue;
1762                 __del_from_free_nid_list(nm_i, i);
1763                 kmem_cache_free(free_nid_slab, i);
1764                 nm_i->fcnt--;
1765                 nr_shrink--;
1766         }
1767         spin_unlock(&nm_i->free_nid_list_lock);
1768         mutex_unlock(&nm_i->build_lock);
1769
1770         return nr - nr_shrink;
1771 }
1772
1773 void recover_inline_xattr(struct inode *inode, struct page *page)
1774 {
1775         void *src_addr, *dst_addr;
1776         size_t inline_size;
1777         struct page *ipage;
1778         struct f2fs_inode *ri;
1779
1780         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1781         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1782
1783         ri = F2FS_INODE(page);
1784         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1785                 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1786                 goto update_inode;
1787         }
1788
1789         dst_addr = inline_xattr_addr(ipage);
1790         src_addr = inline_xattr_addr(page);
1791         inline_size = inline_xattr_size(inode);
1792
1793         f2fs_wait_on_page_writeback(ipage, NODE, true);
1794         memcpy(dst_addr, src_addr, inline_size);
1795 update_inode:
1796         update_inode(inode, ipage);
1797         f2fs_put_page(ipage, 1);
1798 }
1799
1800 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1801 {
1802         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1803         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1804         nid_t new_xnid = nid_of_node(page);
1805         struct node_info ni;
1806
1807         /* 1: invalidate the previous xattr nid */
1808         if (!prev_xnid)
1809                 goto recover_xnid;
1810
1811         /* Deallocate node address */
1812         get_node_info(sbi, prev_xnid, &ni);
1813         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1814         invalidate_blocks(sbi, ni.blk_addr);
1815         dec_valid_node_count(sbi, inode);
1816         set_node_addr(sbi, &ni, NULL_ADDR, false);
1817
1818 recover_xnid:
1819         /* 2: allocate new xattr nid */
1820         if (unlikely(!inc_valid_node_count(sbi, inode)))
1821                 f2fs_bug_on(sbi, 1);
1822
1823         remove_free_nid(NM_I(sbi), new_xnid);
1824         get_node_info(sbi, new_xnid, &ni);
1825         ni.ino = inode->i_ino;
1826         set_node_addr(sbi, &ni, NEW_ADDR, false);
1827         F2FS_I(inode)->i_xattr_nid = new_xnid;
1828
1829         /* 3: update xattr blkaddr */
1830         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1831         set_node_addr(sbi, &ni, blkaddr, false);
1832
1833         update_inode_page(inode);
1834 }
1835
1836 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1837 {
1838         struct f2fs_inode *src, *dst;
1839         nid_t ino = ino_of_node(page);
1840         struct node_info old_ni, new_ni;
1841         struct page *ipage;
1842
1843         get_node_info(sbi, ino, &old_ni);
1844
1845         if (unlikely(old_ni.blk_addr != NULL_ADDR))
1846                 return -EINVAL;
1847
1848         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1849         if (!ipage)
1850                 return -ENOMEM;
1851
1852         /* Should not use this inode from free nid list */
1853         remove_free_nid(NM_I(sbi), ino);
1854
1855         SetPageUptodate(ipage);
1856         fill_node_footer(ipage, ino, ino, 0, true);
1857
1858         src = F2FS_INODE(page);
1859         dst = F2FS_INODE(ipage);
1860
1861         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1862         dst->i_size = 0;
1863         dst->i_blocks = cpu_to_le64(1);
1864         dst->i_links = cpu_to_le32(1);
1865         dst->i_xattr_nid = 0;
1866         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1867
1868         new_ni = old_ni;
1869         new_ni.ino = ino;
1870
1871         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1872                 WARN_ON(1);
1873         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1874         inc_valid_inode_count(sbi);
1875         set_page_dirty(ipage);
1876         f2fs_put_page(ipage, 1);
1877         return 0;
1878 }
1879
1880 int restore_node_summary(struct f2fs_sb_info *sbi,
1881                         unsigned int segno, struct f2fs_summary_block *sum)
1882 {
1883         struct f2fs_node *rn;
1884         struct f2fs_summary *sum_entry;
1885         block_t addr;
1886         int bio_blocks = MAX_BIO_BLOCKS(sbi);
1887         int i, idx, last_offset, nrpages;
1888
1889         /* scan the node segment */
1890         last_offset = sbi->blocks_per_seg;
1891         addr = START_BLOCK(sbi, segno);
1892         sum_entry = &sum->entries[0];
1893
1894         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1895                 nrpages = min(last_offset - i, bio_blocks);
1896
1897                 /* readahead node pages */
1898                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
1899
1900                 for (idx = addr; idx < addr + nrpages; idx++) {
1901                         struct page *page = get_tmp_page(sbi, idx);
1902
1903                         rn = F2FS_NODE(page);
1904                         sum_entry->nid = rn->footer.nid;
1905                         sum_entry->version = 0;
1906                         sum_entry->ofs_in_node = 0;
1907                         sum_entry++;
1908                         f2fs_put_page(page, 1);
1909                 }
1910
1911                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
1912                                                         addr + nrpages);
1913         }
1914         return 0;
1915 }
1916
1917 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1918 {
1919         struct f2fs_nm_info *nm_i = NM_I(sbi);
1920         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1921         struct f2fs_summary_block *sum = curseg->sum_blk;
1922         int i;
1923
1924         mutex_lock(&curseg->curseg_mutex);
1925         for (i = 0; i < nats_in_cursum(sum); i++) {
1926                 struct nat_entry *ne;
1927                 struct f2fs_nat_entry raw_ne;
1928                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1929
1930                 raw_ne = nat_in_journal(sum, i);
1931
1932                 ne = __lookup_nat_cache(nm_i, nid);
1933                 if (!ne) {
1934                         ne = grab_nat_entry(nm_i, nid);
1935                         node_info_from_raw_nat(&ne->ni, &raw_ne);
1936                 }
1937                 __set_nat_cache_dirty(nm_i, ne);
1938         }
1939         update_nats_in_cursum(sum, -i);
1940         mutex_unlock(&curseg->curseg_mutex);
1941 }
1942
1943 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1944                                                 struct list_head *head, int max)
1945 {
1946         struct nat_entry_set *cur;
1947
1948         if (nes->entry_cnt >= max)
1949                 goto add_out;
1950
1951         list_for_each_entry(cur, head, set_list) {
1952                 if (cur->entry_cnt >= nes->entry_cnt) {
1953                         list_add(&nes->set_list, cur->set_list.prev);
1954                         return;
1955                 }
1956         }
1957 add_out:
1958         list_add_tail(&nes->set_list, head);
1959 }
1960
1961 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1962                                         struct nat_entry_set *set)
1963 {
1964         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1965         struct f2fs_summary_block *sum = curseg->sum_blk;
1966         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1967         bool to_journal = true;
1968         struct f2fs_nat_block *nat_blk;
1969         struct nat_entry *ne, *cur;
1970         struct page *page = NULL;
1971
1972         /*
1973          * there are two steps to flush nat entries:
1974          * #1, flush nat entries to journal in current hot data summary block.
1975          * #2, flush nat entries to nat page.
1976          */
1977         if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1978                 to_journal = false;
1979
1980         if (to_journal) {
1981                 mutex_lock(&curseg->curseg_mutex);
1982         } else {
1983                 page = get_next_nat_page(sbi, start_nid);
1984                 nat_blk = page_address(page);
1985                 f2fs_bug_on(sbi, !nat_blk);
1986         }
1987
1988         /* flush dirty nats in nat entry set */
1989         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1990                 struct f2fs_nat_entry *raw_ne;
1991                 nid_t nid = nat_get_nid(ne);
1992                 int offset;
1993
1994                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1995                         continue;
1996
1997                 if (to_journal) {
1998                         offset = lookup_journal_in_cursum(sum,
1999                                                         NAT_JOURNAL, nid, 1);
2000                         f2fs_bug_on(sbi, offset < 0);
2001                         raw_ne = &nat_in_journal(sum, offset);
2002                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
2003                 } else {
2004                         raw_ne = &nat_blk->entries[nid - start_nid];
2005                 }
2006                 raw_nat_from_node_info(raw_ne, &ne->ni);
2007                 nat_reset_flag(ne);
2008                 __clear_nat_cache_dirty(NM_I(sbi), ne);
2009                 if (nat_get_blkaddr(ne) == NULL_ADDR)
2010                         add_free_nid(sbi, nid, false);
2011         }
2012
2013         if (to_journal)
2014                 mutex_unlock(&curseg->curseg_mutex);
2015         else
2016                 f2fs_put_page(page, 1);
2017
2018         f2fs_bug_on(sbi, set->entry_cnt);
2019
2020         radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2021         kmem_cache_free(nat_entry_set_slab, set);
2022 }
2023
2024 /*
2025  * This function is called during the checkpointing process.
2026  */
2027 void flush_nat_entries(struct f2fs_sb_info *sbi)
2028 {
2029         struct f2fs_nm_info *nm_i = NM_I(sbi);
2030         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2031         struct f2fs_summary_block *sum = curseg->sum_blk;
2032         struct nat_entry_set *setvec[SETVEC_SIZE];
2033         struct nat_entry_set *set, *tmp;
2034         unsigned int found;
2035         nid_t set_idx = 0;
2036         LIST_HEAD(sets);
2037
2038         if (!nm_i->dirty_nat_cnt)
2039                 return;
2040
2041         down_write(&nm_i->nat_tree_lock);
2042
2043         /*
2044          * if there are no enough space in journal to store dirty nat
2045          * entries, remove all entries from journal and merge them
2046          * into nat entry set.
2047          */
2048         if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2049                 remove_nats_in_journal(sbi);
2050
2051         while ((found = __gang_lookup_nat_set(nm_i,
2052                                         set_idx, SETVEC_SIZE, setvec))) {
2053                 unsigned idx;
2054                 set_idx = setvec[found - 1]->set + 1;
2055                 for (idx = 0; idx < found; idx++)
2056                         __adjust_nat_entry_set(setvec[idx], &sets,
2057                                                         MAX_NAT_JENTRIES(sum));
2058         }
2059
2060         /* flush dirty nats in nat entry set */
2061         list_for_each_entry_safe(set, tmp, &sets, set_list)
2062                 __flush_nat_entry_set(sbi, set);
2063
2064         up_write(&nm_i->nat_tree_lock);
2065
2066         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2067 }
2068
2069 static int init_node_manager(struct f2fs_sb_info *sbi)
2070 {
2071         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2072         struct f2fs_nm_info *nm_i = NM_I(sbi);
2073         unsigned char *version_bitmap;
2074         unsigned int nat_segs, nat_blocks;
2075
2076         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2077
2078         /* segment_count_nat includes pair segment so divide to 2. */
2079         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2080         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2081
2082         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2083
2084         /* not used nids: 0, node, meta, (and root counted as valid node) */
2085         nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2086         nm_i->fcnt = 0;
2087         nm_i->nat_cnt = 0;
2088         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2089         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2090         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2091
2092         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2093         INIT_LIST_HEAD(&nm_i->free_nid_list);
2094         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2095         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2096         INIT_LIST_HEAD(&nm_i->nat_entries);
2097
2098         mutex_init(&nm_i->build_lock);
2099         spin_lock_init(&nm_i->free_nid_list_lock);
2100         init_rwsem(&nm_i->nat_tree_lock);
2101
2102         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2103         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2104         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2105         if (!version_bitmap)
2106                 return -EFAULT;
2107
2108         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2109                                         GFP_KERNEL);
2110         if (!nm_i->nat_bitmap)
2111                 return -ENOMEM;
2112         return 0;
2113 }
2114
2115 int build_node_manager(struct f2fs_sb_info *sbi)
2116 {
2117         int err;
2118
2119         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2120         if (!sbi->nm_info)
2121                 return -ENOMEM;
2122
2123         err = init_node_manager(sbi);
2124         if (err)
2125                 return err;
2126
2127         build_free_nids(sbi);
2128         return 0;
2129 }
2130
2131 void destroy_node_manager(struct f2fs_sb_info *sbi)
2132 {
2133         struct f2fs_nm_info *nm_i = NM_I(sbi);
2134         struct free_nid *i, *next_i;
2135         struct nat_entry *natvec[NATVEC_SIZE];
2136         struct nat_entry_set *setvec[SETVEC_SIZE];
2137         nid_t nid = 0;
2138         unsigned int found;
2139
2140         if (!nm_i)
2141                 return;
2142
2143         /* destroy free nid list */
2144         spin_lock(&nm_i->free_nid_list_lock);
2145         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2146                 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2147                 __del_from_free_nid_list(nm_i, i);
2148                 nm_i->fcnt--;
2149                 spin_unlock(&nm_i->free_nid_list_lock);
2150                 kmem_cache_free(free_nid_slab, i);
2151                 spin_lock(&nm_i->free_nid_list_lock);
2152         }
2153         f2fs_bug_on(sbi, nm_i->fcnt);
2154         spin_unlock(&nm_i->free_nid_list_lock);
2155
2156         /* destroy nat cache */
2157         down_write(&nm_i->nat_tree_lock);
2158         while ((found = __gang_lookup_nat_cache(nm_i,
2159                                         nid, NATVEC_SIZE, natvec))) {
2160                 unsigned idx;
2161
2162                 nid = nat_get_nid(natvec[found - 1]) + 1;
2163                 for (idx = 0; idx < found; idx++)
2164                         __del_from_nat_cache(nm_i, natvec[idx]);
2165         }
2166         f2fs_bug_on(sbi, nm_i->nat_cnt);
2167
2168         /* destroy nat set cache */
2169         nid = 0;
2170         while ((found = __gang_lookup_nat_set(nm_i,
2171                                         nid, SETVEC_SIZE, setvec))) {
2172                 unsigned idx;
2173
2174                 nid = setvec[found - 1]->set + 1;
2175                 for (idx = 0; idx < found; idx++) {
2176                         /* entry_cnt is not zero, when cp_error was occurred */
2177                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2178                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2179                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2180                 }
2181         }
2182         up_write(&nm_i->nat_tree_lock);
2183
2184         kfree(nm_i->nat_bitmap);
2185         sbi->nm_info = NULL;
2186         kfree(nm_i);
2187 }
2188
2189 int __init create_node_manager_caches(void)
2190 {
2191         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2192                         sizeof(struct nat_entry));
2193         if (!nat_entry_slab)
2194                 goto fail;
2195
2196         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2197                         sizeof(struct free_nid));
2198         if (!free_nid_slab)
2199                 goto destroy_nat_entry;
2200
2201         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2202                         sizeof(struct nat_entry_set));
2203         if (!nat_entry_set_slab)
2204                 goto destroy_free_nid;
2205         return 0;
2206
2207 destroy_free_nid:
2208         kmem_cache_destroy(free_nid_slab);
2209 destroy_nat_entry:
2210         kmem_cache_destroy(nat_entry_slab);
2211 fail:
2212         return -ENOMEM;
2213 }
2214
2215 void destroy_node_manager_caches(void)
2216 {
2217         kmem_cache_destroy(nat_entry_set_slab);
2218         kmem_cache_destroy(free_nid_slab);
2219         kmem_cache_destroy(nat_entry_slab);
2220 }