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