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_SHIFT;
  50                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  51         } else if (type == NAT_ENTRIES) {
  52                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
  53                                                         PAGE_SHIFT;
  54                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  55                 if (excess_cached_nats(sbi))
  56                         res = false;
  57         } else if (type == DIRTY_DENTS) {
  58                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  59                         return false;
  60                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  61                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  62         } else if (type == INO_ENTRIES) {
  63                 int i;
  64
  65                 for (i = 0; i <= UPDATE_INO; i++)
  66                         mem_size += (sbi->im[i].ino_num *
  67                                 sizeof(struct ino_entry)) >> PAGE_SHIFT;
  68                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  69         } else if (type == EXTENT_CACHE) {
  70                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  71                                 sizeof(struct extent_tree) +
  72                                 atomic_read(&sbi->total_ext_node) *
  73                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  74                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  75         } else {
  76                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
  77                         return true;
  78         }
  79         return res;
  80 }
  81
  82 static void clear_node_page_dirty(struct page *page)
  83 {
  84         struct address_space *mapping = page->mapping;
  85         unsigned int long flags;
  86
  87         if (PageDirty(page)) {
  88                 spin_lock_irqsave(&mapping->tree_lock, flags);
  89                 radix_tree_tag_clear(&mapping->page_tree,
  90                                 page_index(page),
  91                                 PAGECACHE_TAG_DIRTY);
  92                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  93
  94                 clear_page_dirty_for_io(page);
  95                 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
  96         }
  97         ClearPageUptodate(page);
  98 }
  99
 100 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 101 {
 102         pgoff_t index = current_nat_addr(sbi, nid);
 103         return get_meta_page(sbi, index);
 104 }
 105
 106 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 107 {
 108         struct page *src_page;
 109         struct page *dst_page;
 110         pgoff_t src_off;
 111         pgoff_t dst_off;
 112         void *src_addr;
 113         void *dst_addr;
 114         struct f2fs_nm_info *nm_i = NM_I(sbi);
 115
 116         src_off = current_nat_addr(sbi, nid);
 117         dst_off = next_nat_addr(sbi, src_off);
 118
 119         /* get current nat block page with lock */
 120         src_page = get_meta_page(sbi, src_off);
 121         dst_page = grab_meta_page(sbi, dst_off);
 122         f2fs_bug_on(sbi, PageDirty(src_page));
 123
 124         src_addr = page_address(src_page);
 125         dst_addr = page_address(dst_page);
 126         memcpy(dst_addr, src_addr, PAGE_SIZE);
 127         set_page_dirty(dst_page);
 128         f2fs_put_page(src_page, 1);
 129
 130         set_to_next_nat(nm_i, nid);
 131
 132         return dst_page;
 133 }
 134
 135 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 136 {
 137         return radix_tree_lookup(&nm_i->nat_root, n);
 138 }
 139
 140 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 141                 nid_t start, unsigned int nr, struct nat_entry **ep)
 142 {
 143         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 144 }
 145
 146 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 147 {
 148         list_del(&e->list);
 149         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 150         nm_i->nat_cnt--;
 151         kmem_cache_free(nat_entry_slab, e);
 152 }
 153
 154 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 155                                                 struct nat_entry *ne)
 156 {
 157         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 158         struct nat_entry_set *head;
 159
 160         if (get_nat_flag(ne, IS_DIRTY))
 161                 return;
 162
 163         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 164         if (!head) {
 165                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 166
 167                 INIT_LIST_HEAD(&head->entry_list);
 168                 INIT_LIST_HEAD(&head->set_list);
 169                 head->set = set;
 170                 head->entry_cnt = 0;
 171                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 172         }
 173         list_move_tail(&ne->list, &head->entry_list);
 174         nm_i->dirty_nat_cnt++;
 175         head->entry_cnt++;
 176         set_nat_flag(ne, IS_DIRTY, true);
 177 }
 178
 179 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 180                                                 struct nat_entry *ne)
 181 {
 182         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 183         struct nat_entry_set *head;
 184
 185         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 186         if (head) {
 187                 list_move_tail(&ne->list, &nm_i->nat_entries);
 188                 set_nat_flag(ne, IS_DIRTY, false);
 189                 head->entry_cnt--;
 190                 nm_i->dirty_nat_cnt--;
 191         }
 192 }
 193
 194 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 195                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 196 {
 197         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 198                                                         start, nr);
 199 }
 200
 201 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 202 {
 203         struct f2fs_nm_info *nm_i = NM_I(sbi);
 204         struct nat_entry *e;
 205         bool need = false;
 206
 207         down_read(&nm_i->nat_tree_lock);
 208         e = __lookup_nat_cache(nm_i, nid);
 209         if (e) {
 210                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 211                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 212                         need = true;
 213         }
 214         up_read(&nm_i->nat_tree_lock);
 215         return need;
 216 }
 217
 218 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 219 {
 220         struct f2fs_nm_info *nm_i = NM_I(sbi);
 221         struct nat_entry *e;
 222         bool is_cp = true;
 223
 224         down_read(&nm_i->nat_tree_lock);
 225         e = __lookup_nat_cache(nm_i, nid);
 226         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 227                 is_cp = false;
 228         up_read(&nm_i->nat_tree_lock);
 229         return is_cp;
 230 }
 231
 232 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 233 {
 234         struct f2fs_nm_info *nm_i = NM_I(sbi);
 235         struct nat_entry *e;
 236         bool need_update = true;
 237
 238         down_read(&nm_i->nat_tree_lock);
 239         e = __lookup_nat_cache(nm_i, ino);
 240         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 241                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 242                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 243                 need_update = false;
 244         up_read(&nm_i->nat_tree_lock);
 245         return need_update;
 246 }
 247
 248 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 249 {
 250         struct nat_entry *new;
 251
 252         new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
 253         f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
 254         memset(new, 0, sizeof(struct nat_entry));
 255         nat_set_nid(new, nid);
 256         nat_reset_flag(new);
 257         list_add_tail(&new->list, &nm_i->nat_entries);
 258         nm_i->nat_cnt++;
 259         return new;
 260 }
 261
 262 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 263                                                 struct f2fs_nat_entry *ne)
 264 {
 265         struct f2fs_nm_info *nm_i = NM_I(sbi);
 266         struct nat_entry *e;
 267
 268         e = __lookup_nat_cache(nm_i, nid);
 269         if (!e) {
 270                 e = grab_nat_entry(nm_i, nid);
 271                 node_info_from_raw_nat(&e->ni, ne);
 272         } else {
 273                 f2fs_bug_on(sbi, nat_get_ino(e) != ne->ino ||
 274                                 nat_get_blkaddr(e) != ne->block_addr ||
 275                                 nat_get_version(e) != ne->version);
 276         }
 277 }
 278
 279 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 280                         block_t new_blkaddr, bool fsync_done)
 281 {
 282         struct f2fs_nm_info *nm_i = NM_I(sbi);
 283         struct nat_entry *e;
 284
 285         down_write(&nm_i->nat_tree_lock);
 286         e = __lookup_nat_cache(nm_i, ni->nid);
 287         if (!e) {
 288                 e = grab_nat_entry(nm_i, ni->nid);
 289                 copy_node_info(&e->ni, ni);
 290                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 291         } else if (new_blkaddr == NEW_ADDR) {
 292                 /*
 293                  * when nid is reallocated,
 294                  * previous nat entry can be remained in nat cache.
 295                  * So, reinitialize it with new information.
 296                  */
 297                 copy_node_info(&e->ni, ni);
 298                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 299         }
 300
 301         /* sanity check */
 302         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 303         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 304                         new_blkaddr == NULL_ADDR);
 305         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 306                         new_blkaddr == NEW_ADDR);
 307         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 308                         nat_get_blkaddr(e) != NULL_ADDR &&
 309                         new_blkaddr == NEW_ADDR);
 310
 311         /* increment version no as node is removed */
 312         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 313                 unsigned char version = nat_get_version(e);
 314                 nat_set_version(e, inc_node_version(version));
 315
 316                 /* in order to reuse the nid */
 317                 if (nm_i->next_scan_nid > ni->nid)
 318                         nm_i->next_scan_nid = ni->nid;
 319         }
 320
 321         /* change address */
 322         nat_set_blkaddr(e, new_blkaddr);
 323         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 324                 set_nat_flag(e, IS_CHECKPOINTED, false);
 325         __set_nat_cache_dirty(nm_i, e);
 326
 327         /* update fsync_mark if its inode nat entry is still alive */
 328         if (ni->nid != ni->ino)
 329                 e = __lookup_nat_cache(nm_i, ni->ino);
 330         if (e) {
 331                 if (fsync_done && ni->nid == ni->ino)
 332                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 333                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 334         }
 335         up_write(&nm_i->nat_tree_lock);
 336 }
 337
 338 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 339 {
 340         struct f2fs_nm_info *nm_i = NM_I(sbi);
 341         int nr = nr_shrink;
 342
 343         if (!down_write_trylock(&nm_i->nat_tree_lock))
 344                 return 0;
 345
 346         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 347                 struct nat_entry *ne;
 348                 ne = list_first_entry(&nm_i->nat_entries,
 349                                         struct nat_entry, list);
 350                 __del_from_nat_cache(nm_i, ne);
 351                 nr_shrink--;
 352         }
 353         up_write(&nm_i->nat_tree_lock);
 354         return nr - nr_shrink;
 355 }
 356
 357 /*
 358  * This function always returns success
 359  */
 360 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 361 {
 362         struct f2fs_nm_info *nm_i = NM_I(sbi);
 363         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 364         struct f2fs_journal *journal = curseg->journal;
 365         nid_t start_nid = START_NID(nid);
 366         struct f2fs_nat_block *nat_blk;
 367         struct page *page = NULL;
 368         struct f2fs_nat_entry ne;
 369         struct nat_entry *e;
 370         int i;
 371
 372         ni->nid = nid;
 373
 374         /* Check nat cache */
 375         down_read(&nm_i->nat_tree_lock);
 376         e = __lookup_nat_cache(nm_i, nid);
 377         if (e) {
 378                 ni->ino = nat_get_ino(e);
 379                 ni->blk_addr = nat_get_blkaddr(e);
 380                 ni->version = nat_get_version(e);
 381                 up_read(&nm_i->nat_tree_lock);
 382                 return;
 383         }
 384
 385         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 386
 387         /* Check current segment summary */
 388         down_read(&curseg->journal_rwsem);
 389         i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 390         if (i >= 0) {
 391                 ne = nat_in_journal(journal, i);
 392                 node_info_from_raw_nat(ni, &ne);
 393         }
 394         up_read(&curseg->journal_rwsem);
 395         if (i >= 0)
 396                 goto cache;
 397
 398         /* Fill node_info from nat page */
 399         page = get_current_nat_page(sbi, start_nid);
 400         nat_blk = (struct f2fs_nat_block *)page_address(page);
 401         ne = nat_blk->entries[nid - start_nid];
 402         node_info_from_raw_nat(ni, &ne);
 403         f2fs_put_page(page, 1);
 404 cache:
 405         up_read(&nm_i->nat_tree_lock);
 406         /* cache nat entry */
 407         down_write(&nm_i->nat_tree_lock);
 408         cache_nat_entry(sbi, nid, &ne);
 409         up_write(&nm_i->nat_tree_lock);
 410 }
 411
 412 /*
 413  * readahead MAX_RA_NODE number of node pages.
 414  */
 415 static void ra_node_pages(struct page *parent, int start, int n)
 416 {
 417         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 418         struct blk_plug plug;
 419         int i, end;
 420         nid_t nid;
 421
 422         blk_start_plug(&plug);
 423
 424         /* Then, try readahead for siblings of the desired node */
 425         end = start + n;
 426         end = min(end, NIDS_PER_BLOCK);
 427         for (i = start; i < end; i++) {
 428                 nid = get_nid(parent, i, false);
 429                 ra_node_page(sbi, nid);
 430         }
 431
 432         blk_finish_plug(&plug);
 433 }
 434
 435 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 436 {
 437         const long direct_index = ADDRS_PER_INODE(dn->inode);
 438         const long direct_blks = ADDRS_PER_BLOCK;
 439         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 440         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 441         int cur_level = dn->cur_level;
 442         int max_level = dn->max_level;
 443         pgoff_t base = 0;
 444
 445         if (!dn->max_level)
 446                 return pgofs + 1;
 447
 448         while (max_level-- > cur_level)
 449                 skipped_unit *= NIDS_PER_BLOCK;
 450
 451         switch (dn->max_level) {
 452         case 3:
 453                 base += 2 * indirect_blks;
 454         case 2:
 455                 base += 2 * direct_blks;
 456         case 1:
 457                 base += direct_index;
 458                 break;
 459         default:
 460                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 461         }
 462
 463         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 464 }
 465
 466 /*
 467  * The maximum depth is four.
 468  * Offset[0] will have raw inode offset.
 469  */
 470 static int get_node_path(struct inode *inode, long block,
 471                                 int offset[4], unsigned int noffset[4])
 472 {
 473         const long direct_index = ADDRS_PER_INODE(inode);
 474         const long direct_blks = ADDRS_PER_BLOCK;
 475         const long dptrs_per_blk = NIDS_PER_BLOCK;
 476         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 477         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 478         int n = 0;
 479         int level = 0;
 480
 481         noffset[0] = 0;
 482
 483         if (block < direct_index) {
 484                 offset[n] = block;
 485                 goto got;
 486         }
 487         block -= direct_index;
 488         if (block < direct_blks) {
 489                 offset[n++] = NODE_DIR1_BLOCK;
 490                 noffset[n] = 1;
 491                 offset[n] = block;
 492                 level = 1;
 493                 goto got;
 494         }
 495         block -= direct_blks;
 496         if (block < direct_blks) {
 497                 offset[n++] = NODE_DIR2_BLOCK;
 498                 noffset[n] = 2;
 499                 offset[n] = block;
 500                 level = 1;
 501                 goto got;
 502         }
 503         block -= direct_blks;
 504         if (block < indirect_blks) {
 505                 offset[n++] = NODE_IND1_BLOCK;
 506                 noffset[n] = 3;
 507                 offset[n++] = block / direct_blks;
 508                 noffset[n] = 4 + offset[n - 1];
 509                 offset[n] = block % direct_blks;
 510                 level = 2;
 511                 goto got;
 512         }
 513         block -= indirect_blks;
 514         if (block < indirect_blks) {
 515                 offset[n++] = NODE_IND2_BLOCK;
 516                 noffset[n] = 4 + dptrs_per_blk;
 517                 offset[n++] = block / direct_blks;
 518                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 519                 offset[n] = block % direct_blks;
 520                 level = 2;
 521                 goto got;
 522         }
 523         block -= indirect_blks;
 524         if (block < dindirect_blks) {
 525                 offset[n++] = NODE_DIND_BLOCK;
 526                 noffset[n] = 5 + (dptrs_per_blk * 2);
 527                 offset[n++] = block / indirect_blks;
 528                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 529                               offset[n - 1] * (dptrs_per_blk + 1);
 530                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 531                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 532                               offset[n - 2] * (dptrs_per_blk + 1) +
 533                               offset[n - 1];
 534                 offset[n] = block % direct_blks;
 535                 level = 3;
 536                 goto got;
 537         } else {
 538                 BUG();
 539         }
 540 got:
 541         return level;
 542 }
 543
 544 /*
 545  * Caller should call f2fs_put_dnode(dn).
 546  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 547  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 548  * In the case of RDONLY_NODE, we don't need to care about mutex.
 549  */
 550 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 551 {
 552         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 553         struct page *npage[4];
 554         struct page *parent = NULL;
 555         int offset[4];
 556         unsigned int noffset[4];
 557         nid_t nids[4];
 558         int level, i = 0;
 559         int err = 0;
 560
 561         level = get_node_path(dn->inode, index, offset, noffset);
 562
 563         nids[0] = dn->inode->i_ino;
 564         npage[0] = dn->inode_page;
 565
 566         if (!npage[0]) {
 567                 npage[0] = get_node_page(sbi, nids[0]);
 568                 if (IS_ERR(npage[0]))
 569                         return PTR_ERR(npage[0]);
 570         }
 571
 572         /* if inline_data is set, should not report any block indices */
 573         if (f2fs_has_inline_data(dn->inode) && index) {
 574                 err = -ENOENT;
 575                 f2fs_put_page(npage[0], 1);
 576                 goto release_out;
 577         }
 578
 579         parent = npage[0];
 580         if (level != 0)
 581                 nids[1] = get_nid(parent, offset[0], true);
 582         dn->inode_page = npage[0];
 583         dn->inode_page_locked = true;
 584
 585         /* get indirect or direct nodes */
 586         for (i = 1; i <= level; i++) {
 587                 bool done = false;
 588
 589                 if (!nids[i] && mode == ALLOC_NODE) {
 590                         /* alloc new node */
 591                         if (!alloc_nid(sbi, &(nids[i]))) {
 592                                 err = -ENOSPC;
 593                                 goto release_pages;
 594                         }
 595
 596                         dn->nid = nids[i];
 597                         npage[i] = new_node_page(dn, noffset[i], NULL);
 598                         if (IS_ERR(npage[i])) {
 599                                 alloc_nid_failed(sbi, nids[i]);
 600                                 err = PTR_ERR(npage[i]);
 601                                 goto release_pages;
 602                         }
 603
 604                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 605                         alloc_nid_done(sbi, nids[i]);
 606                         done = true;
 607                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 608                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 609                         if (IS_ERR(npage[i])) {
 610                                 err = PTR_ERR(npage[i]);
 611                                 goto release_pages;
 612                         }
 613                         done = true;
 614                 }
 615                 if (i == 1) {
 616                         dn->inode_page_locked = false;
 617                         unlock_page(parent);
 618                 } else {
 619                         f2fs_put_page(parent, 1);
 620                 }
 621
 622                 if (!done) {
 623                         npage[i] = get_node_page(sbi, nids[i]);
 624                         if (IS_ERR(npage[i])) {
 625                                 err = PTR_ERR(npage[i]);
 626                                 f2fs_put_page(npage[0], 0);
 627                                 goto release_out;
 628                         }
 629                 }
 630                 if (i < level) {
 631                         parent = npage[i];
 632                         nids[i + 1] = get_nid(parent, offset[i], false);
 633                 }
 634         }
 635         dn->nid = nids[level];
 636         dn->ofs_in_node = offset[level];
 637         dn->node_page = npage[level];
 638         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 639         return 0;
 640
 641 release_pages:
 642         f2fs_put_page(parent, 1);
 643         if (i > 1)
 644                 f2fs_put_page(npage[0], 0);
 645 release_out:
 646         dn->inode_page = NULL;
 647         dn->node_page = NULL;
 648         if (err == -ENOENT) {
 649                 dn->cur_level = i;
 650                 dn->max_level = level;
 651                 dn->ofs_in_node = offset[level];
 652         }
 653         return err;
 654 }
 655
 656 static void truncate_node(struct dnode_of_data *dn)
 657 {
 658         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 659         struct node_info ni;
 660
 661         get_node_info(sbi, dn->nid, &ni);
 662         if (dn->inode->i_blocks == 0) {
 663                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 664                 goto invalidate;
 665         }
 666         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 667
 668         /* Deallocate node address */
 669         invalidate_blocks(sbi, ni.blk_addr);
 670         dec_valid_node_count(sbi, dn->inode);
 671         set_node_addr(sbi, &ni, NULL_ADDR, false);
 672
 673         if (dn->nid == dn->inode->i_ino) {
 674                 remove_orphan_inode(sbi, dn->nid);
 675                 dec_valid_inode_count(sbi);
 676                 f2fs_inode_synced(dn->inode);
 677         }
 678 invalidate:
 679         clear_node_page_dirty(dn->node_page);
 680         set_sbi_flag(sbi, SBI_IS_DIRTY);
 681
 682         f2fs_put_page(dn->node_page, 1);
 683
 684         invalidate_mapping_pages(NODE_MAPPING(sbi),
 685                         dn->node_page->index, dn->node_page->index);
 686
 687         dn->node_page = NULL;
 688         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 689 }
 690
 691 static int truncate_dnode(struct dnode_of_data *dn)
 692 {
 693         struct page *page;
 694
 695         if (dn->nid == 0)
 696                 return 1;
 697
 698         /* get direct node */
 699         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 700         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 701                 return 1;
 702         else if (IS_ERR(page))
 703                 return PTR_ERR(page);
 704
 705         /* Make dnode_of_data for parameter */
 706         dn->node_page = page;
 707         dn->ofs_in_node = 0;
 708         truncate_data_blocks(dn);
 709         truncate_node(dn);
 710         return 1;
 711 }
 712
 713 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 714                                                 int ofs, int depth)
 715 {
 716         struct dnode_of_data rdn = *dn;
 717         struct page *page;
 718         struct f2fs_node *rn;
 719         nid_t child_nid;
 720         unsigned int child_nofs;
 721         int freed = 0;
 722         int i, ret;
 723
 724         if (dn->nid == 0)
 725                 return NIDS_PER_BLOCK + 1;
 726
 727         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 728
 729         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 730         if (IS_ERR(page)) {
 731                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 732                 return PTR_ERR(page);
 733         }
 734
 735         ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 736
 737         rn = F2FS_NODE(page);
 738         if (depth < 3) {
 739                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 740                         child_nid = le32_to_cpu(rn->in.nid[i]);
 741                         if (child_nid == 0)
 742                                 continue;
 743                         rdn.nid = child_nid;
 744                         ret = truncate_dnode(&rdn);
 745                         if (ret < 0)
 746                                 goto out_err;
 747                         if (set_nid(page, i, 0, false))
 748                                 dn->node_changed = true;
 749                 }
 750         } else {
 751                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 752                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 753                         child_nid = le32_to_cpu(rn->in.nid[i]);
 754                         if (child_nid == 0) {
 755                                 child_nofs += NIDS_PER_BLOCK + 1;
 756                                 continue;
 757                         }
 758                         rdn.nid = child_nid;
 759                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 760                         if (ret == (NIDS_PER_BLOCK + 1)) {
 761                                 if (set_nid(page, i, 0, false))
 762                                         dn->node_changed = true;
 763                                 child_nofs += ret;
 764                         } else if (ret < 0 && ret != -ENOENT) {
 765                                 goto out_err;
 766                         }
 767                 }
 768                 freed = child_nofs;
 769         }
 770
 771         if (!ofs) {
 772                 /* remove current indirect node */
 773                 dn->node_page = page;
 774                 truncate_node(dn);
 775                 freed++;
 776         } else {
 777                 f2fs_put_page(page, 1);
 778         }
 779         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 780         return freed;
 781
 782 out_err:
 783         f2fs_put_page(page, 1);
 784         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 785         return ret;
 786 }
 787
 788 static int truncate_partial_nodes(struct dnode_of_data *dn,
 789                         struct f2fs_inode *ri, int *offset, int depth)
 790 {
 791         struct page *pages[2];
 792         nid_t nid[3];
 793         nid_t child_nid;
 794         int err = 0;
 795         int i;
 796         int idx = depth - 2;
 797
 798         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 799         if (!nid[0])
 800                 return 0;
 801
 802         /* get indirect nodes in the path */
 803         for (i = 0; i < idx + 1; i++) {
 804                 /* reference count'll be increased */
 805                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 806                 if (IS_ERR(pages[i])) {
 807                         err = PTR_ERR(pages[i]);
 808                         idx = i - 1;
 809                         goto fail;
 810                 }
 811                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 812         }
 813
 814         ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 815
 816         /* free direct nodes linked to a partial indirect node */
 817         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 818                 child_nid = get_nid(pages[idx], i, false);
 819                 if (!child_nid)
 820                         continue;
 821                 dn->nid = child_nid;
 822                 err = truncate_dnode(dn);
 823                 if (err < 0)
 824                         goto fail;
 825                 if (set_nid(pages[idx], i, 0, false))
 826                         dn->node_changed = true;
 827         }
 828
 829         if (offset[idx + 1] == 0) {
 830                 dn->node_page = pages[idx];
 831                 dn->nid = nid[idx];
 832                 truncate_node(dn);
 833         } else {
 834                 f2fs_put_page(pages[idx], 1);
 835         }
 836         offset[idx]++;
 837         offset[idx + 1] = 0;
 838         idx--;
 839 fail:
 840         for (i = idx; i >= 0; i--)
 841                 f2fs_put_page(pages[i], 1);
 842
 843         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 844
 845         return err;
 846 }
 847
 848 /*
 849  * All the block addresses of data and nodes should be nullified.
 850  */
 851 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 852 {
 853         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 854         int err = 0, cont = 1;
 855         int level, offset[4], noffset[4];
 856         unsigned int nofs = 0;
 857         struct f2fs_inode *ri;
 858         struct dnode_of_data dn;
 859         struct page *page;
 860
 861         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 862
 863         level = get_node_path(inode, from, offset, noffset);
 864
 865         page = get_node_page(sbi, inode->i_ino);
 866         if (IS_ERR(page)) {
 867                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 868                 return PTR_ERR(page);
 869         }
 870
 871         set_new_dnode(&dn, inode, page, NULL, 0);
 872         unlock_page(page);
 873
 874         ri = F2FS_INODE(page);
 875         switch (level) {
 876         case 0:
 877         case 1:
 878                 nofs = noffset[1];
 879                 break;
 880         case 2:
 881                 nofs = noffset[1];
 882                 if (!offset[level - 1])
 883                         goto skip_partial;
 884                 err = truncate_partial_nodes(&dn, ri, offset, level);
 885                 if (err < 0 && err != -ENOENT)
 886                         goto fail;
 887                 nofs += 1 + NIDS_PER_BLOCK;
 888                 break;
 889         case 3:
 890                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 891                 if (!offset[level - 1])
 892                         goto skip_partial;
 893                 err = truncate_partial_nodes(&dn, ri, offset, level);
 894                 if (err < 0 && err != -ENOENT)
 895                         goto fail;
 896                 break;
 897         default:
 898                 BUG();
 899         }
 900
 901 skip_partial:
 902         while (cont) {
 903                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 904                 switch (offset[0]) {
 905                 case NODE_DIR1_BLOCK:
 906                 case NODE_DIR2_BLOCK:
 907                         err = truncate_dnode(&dn);
 908                         break;
 909
 910                 case NODE_IND1_BLOCK:
 911                 case NODE_IND2_BLOCK:
 912                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 913                         break;
 914
 915                 case NODE_DIND_BLOCK:
 916                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 917                         cont = 0;
 918                         break;
 919
 920                 default:
 921                         BUG();
 922                 }
 923                 if (err < 0 && err != -ENOENT)
 924                         goto fail;
 925                 if (offset[1] == 0 &&
 926                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 927                         lock_page(page);
 928                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
 929                         f2fs_wait_on_page_writeback(page, NODE, true);
 930                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 931                         set_page_dirty(page);
 932                         unlock_page(page);
 933                 }
 934                 offset[1] = 0;
 935                 offset[0]++;
 936                 nofs += err;
 937         }
 938 fail:
 939         f2fs_put_page(page, 0);
 940         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 941         return err > 0 ? 0 : err;
 942 }
 943
 944 int truncate_xattr_node(struct inode *inode, struct page *page)
 945 {
 946         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 947         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 948         struct dnode_of_data dn;
 949         struct page *npage;
 950
 951         if (!nid)
 952                 return 0;
 953
 954         npage = get_node_page(sbi, nid);
 955         if (IS_ERR(npage))
 956                 return PTR_ERR(npage);
 957
 958         f2fs_i_xnid_write(inode, 0);
 959
 960         /* need to do checkpoint during fsync */
 961         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 962
 963         set_new_dnode(&dn, inode, page, npage, nid);
 964
 965         if (page)
 966                 dn.inode_page_locked = true;
 967         truncate_node(&dn);
 968         return 0;
 969 }
 970
 971 /*
 972  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 973  * f2fs_unlock_op().
 974  */
 975 int remove_inode_page(struct inode *inode)
 976 {
 977         struct dnode_of_data dn;
 978         int err;
 979
 980         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 981         err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
 982         if (err)
 983                 return err;
 984
 985         err = truncate_xattr_node(inode, dn.inode_page);
 986         if (err) {
 987                 f2fs_put_dnode(&dn);
 988                 return err;
 989         }
 990
 991         /* remove potential inline_data blocks */
 992         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 993                                 S_ISLNK(inode->i_mode))
 994                 truncate_data_blocks_range(&dn, 1);
 995
 996         /* 0 is possible, after f2fs_new_inode() has failed */
 997         f2fs_bug_on(F2FS_I_SB(inode),
 998                         inode->i_blocks != 0 && inode->i_blocks != 1);
 999
1000         /* will put inode & node pages */
1001         truncate_node(&dn);
1002         return 0;
1003 }
1004
1005 struct page *new_inode_page(struct inode *inode)
1006 {
1007         struct dnode_of_data dn;
1008
1009         /* allocate inode page for new inode */
1010         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1011
1012         /* caller should f2fs_put_page(page, 1); */
1013         return new_node_page(&dn, 0, NULL);
1014 }
1015
1016 struct page *new_node_page(struct dnode_of_data *dn,
1017                                 unsigned int ofs, struct page *ipage)
1018 {
1019         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1020         struct node_info old_ni, new_ni;
1021         struct page *page;
1022         int err;
1023
1024         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1025                 return ERR_PTR(-EPERM);
1026
1027         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1028         if (!page)
1029                 return ERR_PTR(-ENOMEM);
1030
1031         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1032                 err = -ENOSPC;
1033                 goto fail;
1034         }
1035
1036         get_node_info(sbi, dn->nid, &old_ni);
1037
1038         /* Reinitialize old_ni with new node page */
1039         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1040         new_ni = old_ni;
1041         new_ni.ino = dn->inode->i_ino;
1042         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1043
1044         f2fs_wait_on_page_writeback(page, NODE, true);
1045         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1046         set_cold_node(dn->inode, page);
1047         if (!PageUptodate(page))
1048                 SetPageUptodate(page);
1049         if (set_page_dirty(page))
1050                 dn->node_changed = true;
1051
1052         if (f2fs_has_xattr_block(ofs))
1053                 f2fs_i_xnid_write(dn->inode, dn->nid);
1054
1055         if (ofs == 0)
1056                 inc_valid_inode_count(sbi);
1057         return page;
1058
1059 fail:
1060         clear_node_page_dirty(page);
1061         f2fs_put_page(page, 1);
1062         return ERR_PTR(err);
1063 }
1064
1065 /*
1066  * Caller should do after getting the following values.
1067  * 0: f2fs_put_page(page, 0)
1068  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1069  */
1070 static int read_node_page(struct page *page, int op_flags)
1071 {
1072         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1073         struct node_info ni;
1074         struct f2fs_io_info fio = {
1075                 .sbi = sbi,
1076                 .type = NODE,
1077                 .op = REQ_OP_READ,
1078                 .op_flags = op_flags,
1079                 .page = page,
1080                 .encrypted_page = NULL,
1081         };
1082
1083         if (PageUptodate(page))
1084                 return LOCKED_PAGE;
1085
1086         get_node_info(sbi, page->index, &ni);
1087
1088         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1089                 ClearPageUptodate(page);
1090                 return -ENOENT;
1091         }
1092
1093         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1094         return f2fs_submit_page_bio(&fio);
1095 }
1096
1097 /*
1098  * Readahead a node page
1099  */
1100 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1101 {
1102         struct page *apage;
1103         int err;
1104
1105         if (!nid)
1106                 return;
1107         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1108
1109         rcu_read_lock();
1110         apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1111         rcu_read_unlock();
1112         if (apage)
1113                 return;
1114
1115         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1116         if (!apage)
1117                 return;
1118
1119         err = read_node_page(apage, REQ_RAHEAD);
1120         f2fs_put_page(apage, err ? 1 : 0);
1121 }
1122
1123 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1124                                         struct page *parent, int start)
1125 {
1126         struct page *page;
1127         int err;
1128
1129         if (!nid)
1130                 return ERR_PTR(-ENOENT);
1131         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1132 repeat:
1133         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1134         if (!page)
1135                 return ERR_PTR(-ENOMEM);
1136
1137         err = read_node_page(page, READ_SYNC);
1138         if (err < 0) {
1139                 f2fs_put_page(page, 1);
1140                 return ERR_PTR(err);
1141         } else if (err == LOCKED_PAGE) {
1142                 goto page_hit;
1143         }
1144
1145         if (parent)
1146                 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1147
1148         lock_page(page);
1149
1150         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1151                 f2fs_put_page(page, 1);
1152                 goto repeat;
1153         }
1154
1155         if (unlikely(!PageUptodate(page)))
1156                 goto out_err;
1157 page_hit:
1158         if(unlikely(nid != nid_of_node(page))) {
1159                 f2fs_bug_on(sbi, 1);
1160                 ClearPageUptodate(page);
1161 out_err:
1162                 f2fs_put_page(page, 1);
1163                 return ERR_PTR(-EIO);
1164         }
1165         return page;
1166 }
1167
1168 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1169 {
1170         return __get_node_page(sbi, nid, NULL, 0);
1171 }
1172
1173 struct page *get_node_page_ra(struct page *parent, int start)
1174 {
1175         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1176         nid_t nid = get_nid(parent, start, false);
1177
1178         return __get_node_page(sbi, nid, parent, start);
1179 }
1180
1181 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1182 {
1183         struct inode *inode;
1184         struct page *page;
1185         int ret;
1186
1187         /* should flush inline_data before evict_inode */
1188         inode = ilookup(sbi->sb, ino);
1189         if (!inode)
1190                 return;
1191
1192         page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1193         if (!page)
1194                 goto iput_out;
1195
1196         if (!PageUptodate(page))
1197                 goto page_out;
1198
1199         if (!PageDirty(page))
1200                 goto page_out;
1201
1202         if (!clear_page_dirty_for_io(page))
1203                 goto page_out;
1204
1205         ret = f2fs_write_inline_data(inode, page);
1206         inode_dec_dirty_pages(inode);
1207         if (ret)
1208                 set_page_dirty(page);
1209 page_out:
1210         f2fs_put_page(page, 1);
1211 iput_out:
1212         iput(inode);
1213 }
1214
1215 void move_node_page(struct page *node_page, int gc_type)
1216 {
1217         if (gc_type == FG_GC) {
1218                 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1219                 struct writeback_control wbc = {
1220                         .sync_mode = WB_SYNC_ALL,
1221                         .nr_to_write = 1,
1222                         .for_reclaim = 0,
1223                 };
1224
1225                 set_page_dirty(node_page);
1226                 f2fs_wait_on_page_writeback(node_page, NODE, true);
1227
1228                 f2fs_bug_on(sbi, PageWriteback(node_page));
1229                 if (!clear_page_dirty_for_io(node_page))
1230                         goto out_page;
1231
1232                 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1233                         unlock_page(node_page);
1234                 goto release_page;
1235         } else {
1236                 /* set page dirty and write it */
1237                 if (!PageWriteback(node_page))
1238                         set_page_dirty(node_page);
1239         }
1240 out_page:
1241         unlock_page(node_page);
1242 release_page:
1243         f2fs_put_page(node_page, 0);
1244 }
1245
1246 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1247 {
1248         pgoff_t index, end;
1249         struct pagevec pvec;
1250         struct page *last_page = NULL;
1251
1252         pagevec_init(&pvec, 0);
1253         index = 0;
1254         end = ULONG_MAX;
1255
1256         while (index <= end) {
1257                 int i, nr_pages;
1258                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1259                                 PAGECACHE_TAG_DIRTY,
1260                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1261                 if (nr_pages == 0)
1262                         break;
1263
1264                 for (i = 0; i < nr_pages; i++) {
1265                         struct page *page = pvec.pages[i];
1266
1267                         if (unlikely(f2fs_cp_error(sbi))) {
1268                                 f2fs_put_page(last_page, 0);
1269                                 pagevec_release(&pvec);
1270                                 return ERR_PTR(-EIO);
1271                         }
1272
1273                         if (!IS_DNODE(page) || !is_cold_node(page))
1274                                 continue;
1275                         if (ino_of_node(page) != ino)
1276                                 continue;
1277
1278                         lock_page(page);
1279
1280                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1281 continue_unlock:
1282                                 unlock_page(page);
1283                                 continue;
1284                         }
1285                         if (ino_of_node(page) != ino)
1286                                 goto continue_unlock;
1287
1288                         if (!PageDirty(page)) {
1289                                 /* someone wrote it for us */
1290                                 goto continue_unlock;
1291                         }
1292
1293                         if (last_page)
1294                                 f2fs_put_page(last_page, 0);
1295
1296                         get_page(page);
1297                         last_page = page;
1298                         unlock_page(page);
1299                 }
1300                 pagevec_release(&pvec);
1301                 cond_resched();
1302         }
1303         return last_page;
1304 }
1305
1306 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1307                         struct writeback_control *wbc, bool atomic)
1308 {
1309         pgoff_t index, end;
1310         struct pagevec pvec;
1311         int ret = 0;
1312         struct page *last_page = NULL;
1313         bool marked = false;
1314         nid_t ino = inode->i_ino;
1315
1316         if (atomic) {
1317                 last_page = last_fsync_dnode(sbi, ino);
1318                 if (IS_ERR_OR_NULL(last_page))
1319                         return PTR_ERR_OR_ZERO(last_page);
1320         }
1321 retry:
1322         pagevec_init(&pvec, 0);
1323         index = 0;
1324         end = ULONG_MAX;
1325
1326         while (index <= end) {
1327                 int i, nr_pages;
1328                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1329                                 PAGECACHE_TAG_DIRTY,
1330                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1331                 if (nr_pages == 0)
1332                         break;
1333
1334                 for (i = 0; i < nr_pages; i++) {
1335                         struct page *page = pvec.pages[i];
1336
1337                         if (unlikely(f2fs_cp_error(sbi))) {
1338                                 f2fs_put_page(last_page, 0);
1339                                 pagevec_release(&pvec);
1340                                 return -EIO;
1341                         }
1342
1343                         if (!IS_DNODE(page) || !is_cold_node(page))
1344                                 continue;
1345                         if (ino_of_node(page) != ino)
1346                                 continue;
1347
1348                         lock_page(page);
1349
1350                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1351 continue_unlock:
1352                                 unlock_page(page);
1353                                 continue;
1354                         }
1355                         if (ino_of_node(page) != ino)
1356                                 goto continue_unlock;
1357
1358                         if (!PageDirty(page) && page != last_page) {
1359                                 /* someone wrote it for us */
1360                                 goto continue_unlock;
1361                         }
1362
1363                         f2fs_wait_on_page_writeback(page, NODE, true);
1364                         BUG_ON(PageWriteback(page));
1365
1366                         if (!atomic || page == last_page) {
1367                                 set_fsync_mark(page, 1);
1368                                 if (IS_INODE(page)) {
1369                                         if (is_inode_flag_set(inode,
1370                                                                 FI_DIRTY_INODE))
1371                                                 update_inode(inode, page);
1372                                         set_dentry_mark(page,
1373                                                 need_dentry_mark(sbi, ino));
1374                                 }
1375                                 /*  may be written by other thread */
1376                                 if (!PageDirty(page))
1377                                         set_page_dirty(page);
1378                         }
1379
1380                         if (!clear_page_dirty_for_io(page))
1381                                 goto continue_unlock;
1382
1383                         ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1384                         if (ret) {
1385                                 unlock_page(page);
1386                                 f2fs_put_page(last_page, 0);
1387                                 break;
1388                         }
1389                         if (page == last_page) {
1390                                 f2fs_put_page(page, 0);
1391                                 marked = true;
1392                                 break;
1393                         }
1394                 }
1395                 pagevec_release(&pvec);
1396                 cond_resched();
1397
1398                 if (ret || marked)
1399                         break;
1400         }
1401         if (!ret && atomic && !marked) {
1402                 f2fs_msg(sbi->sb, KERN_DEBUG,
1403                         "Retry to write fsync mark: ino=%u, idx=%lx",
1404                                         ino, last_page->index);
1405                 lock_page(last_page);
1406                 set_page_dirty(last_page);
1407                 unlock_page(last_page);
1408                 goto retry;
1409         }
1410         return ret ? -EIO: 0;
1411 }
1412
1413 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1414 {
1415         pgoff_t index, end;
1416         struct pagevec pvec;
1417         int step = 0;
1418         int nwritten = 0;
1419
1420         pagevec_init(&pvec, 0);
1421
1422 next_step:
1423         index = 0;
1424         end = ULONG_MAX;
1425
1426         while (index <= end) {
1427                 int i, nr_pages;
1428                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1429                                 PAGECACHE_TAG_DIRTY,
1430                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1431                 if (nr_pages == 0)
1432                         break;
1433
1434                 for (i = 0; i < nr_pages; i++) {
1435                         struct page *page = pvec.pages[i];
1436
1437                         if (unlikely(f2fs_cp_error(sbi))) {
1438                                 pagevec_release(&pvec);
1439                                 return -EIO;
1440                         }
1441
1442                         /*
1443                          * flushing sequence with step:
1444                          * 0. indirect nodes
1445                          * 1. dentry dnodes
1446                          * 2. file dnodes
1447                          */
1448                         if (step == 0 && IS_DNODE(page))
1449                                 continue;
1450                         if (step == 1 && (!IS_DNODE(page) ||
1451                                                 is_cold_node(page)))
1452                                 continue;
1453                         if (step == 2 && (!IS_DNODE(page) ||
1454                                                 !is_cold_node(page)))
1455                                 continue;
1456 lock_node:
1457                         if (!trylock_page(page))
1458                                 continue;
1459
1460                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1461 continue_unlock:
1462                                 unlock_page(page);
1463                                 continue;
1464                         }
1465
1466                         if (!PageDirty(page)) {
1467                                 /* someone wrote it for us */
1468                                 goto continue_unlock;
1469                         }
1470
1471                         /* flush inline_data */
1472                         if (is_inline_node(page)) {
1473                                 clear_inline_node(page);
1474                                 unlock_page(page);
1475                                 flush_inline_data(sbi, ino_of_node(page));
1476                                 goto lock_node;
1477                         }
1478
1479                         f2fs_wait_on_page_writeback(page, NODE, true);
1480
1481                         BUG_ON(PageWriteback(page));
1482                         if (!clear_page_dirty_for_io(page))
1483                                 goto continue_unlock;
1484
1485                         set_fsync_mark(page, 0);
1486                         set_dentry_mark(page, 0);
1487
1488                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1489                                 unlock_page(page);
1490
1491                         if (--wbc->nr_to_write == 0)
1492                                 break;
1493                 }
1494                 pagevec_release(&pvec);
1495                 cond_resched();
1496
1497                 if (wbc->nr_to_write == 0) {
1498                         step = 2;
1499                         break;
1500                 }
1501         }
1502
1503         if (step < 2) {
1504                 step++;
1505                 goto next_step;
1506         }
1507         return nwritten;
1508 }
1509
1510 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1511 {
1512         pgoff_t index = 0, end = ULONG_MAX;
1513         struct pagevec pvec;
1514         int ret2 = 0, ret = 0;
1515
1516         pagevec_init(&pvec, 0);
1517
1518         while (index <= end) {
1519                 int i, nr_pages;
1520                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1521                                 PAGECACHE_TAG_WRITEBACK,
1522                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1523                 if (nr_pages == 0)
1524                         break;
1525
1526                 for (i = 0; i < nr_pages; i++) {
1527                         struct page *page = pvec.pages[i];
1528
1529                         /* until radix tree lookup accepts end_index */
1530                         if (unlikely(page->index > end))
1531                                 continue;
1532
1533                         if (ino && ino_of_node(page) == ino) {
1534                                 f2fs_wait_on_page_writeback(page, NODE, true);
1535                                 if (TestClearPageError(page))
1536                                         ret = -EIO;
1537                         }
1538                 }
1539                 pagevec_release(&pvec);
1540                 cond_resched();
1541         }
1542
1543         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1544                 ret2 = -ENOSPC;
1545         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1546                 ret2 = -EIO;
1547         if (!ret)
1548                 ret = ret2;
1549         return ret;
1550 }
1551
1552 static int f2fs_write_node_page(struct page *page,
1553                                 struct writeback_control *wbc)
1554 {
1555         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1556         nid_t nid;
1557         struct node_info ni;
1558         struct f2fs_io_info fio = {
1559                 .sbi = sbi,
1560                 .type = NODE,
1561                 .op = REQ_OP_WRITE,
1562                 .op_flags = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0,
1563                 .page = page,
1564                 .encrypted_page = NULL,
1565         };
1566
1567         trace_f2fs_writepage(page, NODE);
1568
1569         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1570                 goto redirty_out;
1571         if (unlikely(f2fs_cp_error(sbi)))
1572                 goto redirty_out;
1573
1574         /* get old block addr of this node page */
1575         nid = nid_of_node(page);
1576         f2fs_bug_on(sbi, page->index != nid);
1577
1578         if (wbc->for_reclaim) {
1579                 if (!down_read_trylock(&sbi->node_write))
1580                         goto redirty_out;
1581         } else {
1582                 down_read(&sbi->node_write);
1583         }
1584
1585         get_node_info(sbi, nid, &ni);
1586
1587         /* This page is already truncated */
1588         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1589                 ClearPageUptodate(page);
1590                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1591                 up_read(&sbi->node_write);
1592                 unlock_page(page);
1593                 return 0;
1594         }
1595
1596         set_page_writeback(page);
1597         fio.old_blkaddr = ni.blk_addr;
1598         write_node_page(nid, &fio);
1599         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1600         dec_page_count(sbi, F2FS_DIRTY_NODES);
1601         up_read(&sbi->node_write);
1602
1603         if (wbc->for_reclaim)
1604                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1605
1606         unlock_page(page);
1607
1608         if (unlikely(f2fs_cp_error(sbi)))
1609                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1610
1611         return 0;
1612
1613 redirty_out:
1614         redirty_page_for_writepage(wbc, page);
1615         return AOP_WRITEPAGE_ACTIVATE;
1616 }
1617
1618 static int f2fs_write_node_pages(struct address_space *mapping,
1619                             struct writeback_control *wbc)
1620 {
1621         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1622         struct blk_plug plug;
1623         long diff;
1624
1625         /* balancing f2fs's metadata in background */
1626         f2fs_balance_fs_bg(sbi);
1627
1628         /* collect a number of dirty node pages and write together */
1629         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1630                 goto skip_write;
1631
1632         trace_f2fs_writepages(mapping->host, wbc, NODE);
1633
1634         diff = nr_pages_to_write(sbi, NODE, wbc);
1635         wbc->sync_mode = WB_SYNC_NONE;
1636         blk_start_plug(&plug);
1637         sync_node_pages(sbi, wbc);
1638         blk_finish_plug(&plug);
1639         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1640         return 0;
1641
1642 skip_write:
1643         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1644         trace_f2fs_writepages(mapping->host, wbc, NODE);
1645         return 0;
1646 }
1647
1648 static int f2fs_set_node_page_dirty(struct page *page)
1649 {
1650         trace_f2fs_set_page_dirty(page, NODE);
1651
1652         if (!PageUptodate(page))
1653                 SetPageUptodate(page);
1654         if (!PageDirty(page)) {
1655                 f2fs_set_page_dirty_nobuffers(page);
1656                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1657                 SetPagePrivate(page);
1658                 f2fs_trace_pid(page);
1659                 return 1;
1660         }
1661         return 0;
1662 }
1663
1664 /*
1665  * Structure of the f2fs node operations
1666  */
1667 const struct address_space_operations f2fs_node_aops = {
1668         .writepage      = f2fs_write_node_page,
1669         .writepages     = f2fs_write_node_pages,
1670         .set_page_dirty = f2fs_set_node_page_dirty,
1671         .invalidatepage = f2fs_invalidate_page,
1672         .releasepage    = f2fs_release_page,
1673 #ifdef CONFIG_MIGRATION
1674         .migratepage    = f2fs_migrate_page,
1675 #endif
1676 };
1677
1678 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1679                                                 nid_t n)
1680 {
1681         return radix_tree_lookup(&nm_i->free_nid_root, n);
1682 }
1683
1684 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1685                                                 struct free_nid *i)
1686 {
1687         list_del(&i->list);
1688         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1689 }
1690
1691 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1692 {
1693         struct f2fs_nm_info *nm_i = NM_I(sbi);
1694         struct free_nid *i;
1695         struct nat_entry *ne;
1696
1697         if (!available_free_memory(sbi, FREE_NIDS))
1698                 return -1;
1699
1700         /* 0 nid should not be used */
1701         if (unlikely(nid == 0))
1702                 return 0;
1703
1704         if (build) {
1705                 /* do not add allocated nids */
1706                 ne = __lookup_nat_cache(nm_i, nid);
1707                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1708                                 nat_get_blkaddr(ne) != NULL_ADDR))
1709                         return 0;
1710         }
1711
1712         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1713         i->nid = nid;
1714         i->state = NID_NEW;
1715
1716         if (radix_tree_preload(GFP_NOFS)) {
1717                 kmem_cache_free(free_nid_slab, i);
1718                 return 0;
1719         }
1720
1721         spin_lock(&nm_i->free_nid_list_lock);
1722         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1723                 spin_unlock(&nm_i->free_nid_list_lock);
1724                 radix_tree_preload_end();
1725                 kmem_cache_free(free_nid_slab, i);
1726                 return 0;
1727         }
1728         list_add_tail(&i->list, &nm_i->free_nid_list);
1729         nm_i->fcnt++;
1730         spin_unlock(&nm_i->free_nid_list_lock);
1731         radix_tree_preload_end();
1732         return 1;
1733 }
1734
1735 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1736 {
1737         struct free_nid *i;
1738         bool need_free = false;
1739
1740         spin_lock(&nm_i->free_nid_list_lock);
1741         i = __lookup_free_nid_list(nm_i, nid);
1742         if (i && i->state == NID_NEW) {
1743                 __del_from_free_nid_list(nm_i, i);
1744                 nm_i->fcnt--;
1745                 need_free = true;
1746         }
1747         spin_unlock(&nm_i->free_nid_list_lock);
1748
1749         if (need_free)
1750                 kmem_cache_free(free_nid_slab, i);
1751 }
1752
1753 static void scan_nat_page(struct f2fs_sb_info *sbi,
1754                         struct page *nat_page, nid_t start_nid)
1755 {
1756         struct f2fs_nm_info *nm_i = NM_I(sbi);
1757         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1758         block_t blk_addr;
1759         int i;
1760
1761         i = start_nid % NAT_ENTRY_PER_BLOCK;
1762
1763         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1764
1765                 if (unlikely(start_nid >= nm_i->max_nid))
1766                         break;
1767
1768                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1769                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1770                 if (blk_addr == NULL_ADDR) {
1771                         if (add_free_nid(sbi, start_nid, true) < 0)
1772                                 break;
1773                 }
1774         }
1775 }
1776
1777 void build_free_nids(struct f2fs_sb_info *sbi)
1778 {
1779         struct f2fs_nm_info *nm_i = NM_I(sbi);
1780         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1781         struct f2fs_journal *journal = curseg->journal;
1782         int i = 0;
1783         nid_t nid = nm_i->next_scan_nid;
1784
1785         /* Enough entries */
1786         if (nm_i->fcnt >= NAT_ENTRY_PER_BLOCK)
1787                 return;
1788
1789         /* readahead nat pages to be scanned */
1790         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1791                                                         META_NAT, true);
1792
1793         down_read(&nm_i->nat_tree_lock);
1794
1795         while (1) {
1796                 struct page *page = get_current_nat_page(sbi, nid);
1797
1798                 scan_nat_page(sbi, page, nid);
1799                 f2fs_put_page(page, 1);
1800
1801                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1802                 if (unlikely(nid >= nm_i->max_nid))
1803                         nid = 0;
1804
1805                 if (++i >= FREE_NID_PAGES)
1806                         break;
1807         }
1808
1809         /* go to the next free nat pages to find free nids abundantly */
1810         nm_i->next_scan_nid = nid;
1811
1812         /* find free nids from current sum_pages */
1813         down_read(&curseg->journal_rwsem);
1814         for (i = 0; i < nats_in_cursum(journal); i++) {
1815                 block_t addr;
1816
1817                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1818                 nid = le32_to_cpu(nid_in_journal(journal, i));
1819                 if (addr == NULL_ADDR)
1820                         add_free_nid(sbi, nid, true);
1821                 else
1822                         remove_free_nid(nm_i, nid);
1823         }
1824         up_read(&curseg->journal_rwsem);
1825         up_read(&nm_i->nat_tree_lock);
1826
1827         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1828                                         nm_i->ra_nid_pages, META_NAT, false);
1829 }
1830
1831 /*
1832  * If this function returns success, caller can obtain a new nid
1833  * from second parameter of this function.
1834  * The returned nid could be used ino as well as nid when inode is created.
1835  */
1836 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1837 {
1838         struct f2fs_nm_info *nm_i = NM_I(sbi);
1839         struct free_nid *i = NULL;
1840 retry:
1841 #ifdef CONFIG_F2FS_FAULT_INJECTION
1842         if (time_to_inject(sbi, FAULT_ALLOC_NID))
1843                 return false;
1844 #endif
1845         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1846                 return false;
1847
1848         spin_lock(&nm_i->free_nid_list_lock);
1849
1850         /* We should not use stale free nids created by build_free_nids */
1851         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1852                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1853                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1854                         if (i->state == NID_NEW)
1855                                 break;
1856
1857                 f2fs_bug_on(sbi, i->state != NID_NEW);
1858                 *nid = i->nid;
1859                 i->state = NID_ALLOC;
1860                 nm_i->fcnt--;
1861                 spin_unlock(&nm_i->free_nid_list_lock);
1862                 return true;
1863         }
1864         spin_unlock(&nm_i->free_nid_list_lock);
1865
1866         /* Let's scan nat pages and its caches to get free nids */
1867         mutex_lock(&nm_i->build_lock);
1868         build_free_nids(sbi);
1869         mutex_unlock(&nm_i->build_lock);
1870         goto retry;
1871 }
1872
1873 /*
1874  * alloc_nid() should be called prior to this function.
1875  */
1876 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1877 {
1878         struct f2fs_nm_info *nm_i = NM_I(sbi);
1879         struct free_nid *i;
1880
1881         spin_lock(&nm_i->free_nid_list_lock);
1882         i = __lookup_free_nid_list(nm_i, nid);
1883         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1884         __del_from_free_nid_list(nm_i, i);
1885         spin_unlock(&nm_i->free_nid_list_lock);
1886
1887         kmem_cache_free(free_nid_slab, i);
1888 }
1889
1890 /*
1891  * alloc_nid() should be called prior to this function.
1892  */
1893 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1894 {
1895         struct f2fs_nm_info *nm_i = NM_I(sbi);
1896         struct free_nid *i;
1897         bool need_free = false;
1898
1899         if (!nid)
1900                 return;
1901
1902         spin_lock(&nm_i->free_nid_list_lock);
1903         i = __lookup_free_nid_list(nm_i, nid);
1904         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1905         if (!available_free_memory(sbi, FREE_NIDS)) {
1906                 __del_from_free_nid_list(nm_i, i);
1907                 need_free = true;
1908         } else {
1909                 i->state = NID_NEW;
1910                 nm_i->fcnt++;
1911         }
1912         spin_unlock(&nm_i->free_nid_list_lock);
1913
1914         if (need_free)
1915                 kmem_cache_free(free_nid_slab, i);
1916 }
1917
1918 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1919 {
1920         struct f2fs_nm_info *nm_i = NM_I(sbi);
1921         struct free_nid *i, *next;
1922         int nr = nr_shrink;
1923
1924         if (nm_i->fcnt <= MAX_FREE_NIDS)
1925                 return 0;
1926
1927         if (!mutex_trylock(&nm_i->build_lock))
1928                 return 0;
1929
1930         spin_lock(&nm_i->free_nid_list_lock);
1931         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1932                 if (nr_shrink <= 0 || nm_i->fcnt <= MAX_FREE_NIDS)
1933                         break;
1934                 if (i->state == NID_ALLOC)
1935                         continue;
1936                 __del_from_free_nid_list(nm_i, i);
1937                 kmem_cache_free(free_nid_slab, i);
1938                 nm_i->fcnt--;
1939                 nr_shrink--;
1940         }
1941         spin_unlock(&nm_i->free_nid_list_lock);
1942         mutex_unlock(&nm_i->build_lock);
1943
1944         return nr - nr_shrink;
1945 }
1946
1947 void recover_inline_xattr(struct inode *inode, struct page *page)
1948 {
1949         void *src_addr, *dst_addr;
1950         size_t inline_size;
1951         struct page *ipage;
1952         struct f2fs_inode *ri;
1953
1954         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1955         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1956
1957         ri = F2FS_INODE(page);
1958         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1959                 clear_inode_flag(inode, FI_INLINE_XATTR);
1960                 goto update_inode;
1961         }
1962
1963         dst_addr = inline_xattr_addr(ipage);
1964         src_addr = inline_xattr_addr(page);
1965         inline_size = inline_xattr_size(inode);
1966
1967         f2fs_wait_on_page_writeback(ipage, NODE, true);
1968         memcpy(dst_addr, src_addr, inline_size);
1969 update_inode:
1970         update_inode(inode, ipage);
1971         f2fs_put_page(ipage, 1);
1972 }
1973
1974 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1975 {
1976         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1977         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1978         nid_t new_xnid = nid_of_node(page);
1979         struct node_info ni;
1980
1981         /* 1: invalidate the previous xattr nid */
1982         if (!prev_xnid)
1983                 goto recover_xnid;
1984
1985         /* Deallocate node address */
1986         get_node_info(sbi, prev_xnid, &ni);
1987         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1988         invalidate_blocks(sbi, ni.blk_addr);
1989         dec_valid_node_count(sbi, inode);
1990         set_node_addr(sbi, &ni, NULL_ADDR, false);
1991
1992 recover_xnid:
1993         /* 2: allocate new xattr nid */
1994         if (unlikely(!inc_valid_node_count(sbi, inode)))
1995                 f2fs_bug_on(sbi, 1);
1996
1997         remove_free_nid(NM_I(sbi), new_xnid);
1998         get_node_info(sbi, new_xnid, &ni);
1999         ni.ino = inode->i_ino;
2000         set_node_addr(sbi, &ni, NEW_ADDR, false);
2001         f2fs_i_xnid_write(inode, new_xnid);
2002
2003         /* 3: update xattr blkaddr */
2004         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
2005         set_node_addr(sbi, &ni, blkaddr, false);
2006 }
2007
2008 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2009 {
2010         struct f2fs_inode *src, *dst;
2011         nid_t ino = ino_of_node(page);
2012         struct node_info old_ni, new_ni;
2013         struct page *ipage;
2014
2015         get_node_info(sbi, ino, &old_ni);
2016
2017         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2018                 return -EINVAL;
2019 retry:
2020         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2021         if (!ipage) {
2022                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2023                 goto retry;
2024         }
2025
2026         /* Should not use this inode from free nid list */
2027         remove_free_nid(NM_I(sbi), ino);
2028
2029         if (!PageUptodate(ipage))
2030                 SetPageUptodate(ipage);
2031         fill_node_footer(ipage, ino, ino, 0, true);
2032
2033         src = F2FS_INODE(page);
2034         dst = F2FS_INODE(ipage);
2035
2036         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2037         dst->i_size = 0;
2038         dst->i_blocks = cpu_to_le64(1);
2039         dst->i_links = cpu_to_le32(1);
2040         dst->i_xattr_nid = 0;
2041         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2042
2043         new_ni = old_ni;
2044         new_ni.ino = ino;
2045
2046         if (unlikely(!inc_valid_node_count(sbi, NULL)))
2047                 WARN_ON(1);
2048         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2049         inc_valid_inode_count(sbi);
2050         set_page_dirty(ipage);
2051         f2fs_put_page(ipage, 1);
2052         return 0;
2053 }
2054
2055 int restore_node_summary(struct f2fs_sb_info *sbi,
2056                         unsigned int segno, struct f2fs_summary_block *sum)
2057 {
2058         struct f2fs_node *rn;
2059         struct f2fs_summary *sum_entry;
2060         block_t addr;
2061         int bio_blocks = MAX_BIO_BLOCKS(sbi);
2062         int i, idx, last_offset, nrpages;
2063
2064         /* scan the node segment */
2065         last_offset = sbi->blocks_per_seg;
2066         addr = START_BLOCK(sbi, segno);
2067         sum_entry = &sum->entries[0];
2068
2069         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2070                 nrpages = min(last_offset - i, bio_blocks);
2071
2072                 /* readahead node pages */
2073                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2074
2075                 for (idx = addr; idx < addr + nrpages; idx++) {
2076                         struct page *page = get_tmp_page(sbi, idx);
2077
2078                         rn = F2FS_NODE(page);
2079                         sum_entry->nid = rn->footer.nid;
2080                         sum_entry->version = 0;
2081                         sum_entry->ofs_in_node = 0;
2082                         sum_entry++;
2083                         f2fs_put_page(page, 1);
2084                 }
2085
2086                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2087                                                         addr + nrpages);
2088         }
2089         return 0;
2090 }
2091
2092 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2093 {
2094         struct f2fs_nm_info *nm_i = NM_I(sbi);
2095         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2096         struct f2fs_journal *journal = curseg->journal;
2097         int i;
2098
2099         down_write(&curseg->journal_rwsem);
2100         for (i = 0; i < nats_in_cursum(journal); i++) {
2101                 struct nat_entry *ne;
2102                 struct f2fs_nat_entry raw_ne;
2103                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2104
2105                 raw_ne = nat_in_journal(journal, i);
2106
2107                 ne = __lookup_nat_cache(nm_i, nid);
2108                 if (!ne) {
2109                         ne = grab_nat_entry(nm_i, nid);
2110                         node_info_from_raw_nat(&ne->ni, &raw_ne);
2111                 }
2112                 __set_nat_cache_dirty(nm_i, ne);
2113         }
2114         update_nats_in_cursum(journal, -i);
2115         up_write(&curseg->journal_rwsem);
2116 }
2117
2118 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2119                                                 struct list_head *head, int max)
2120 {
2121         struct nat_entry_set *cur;
2122
2123         if (nes->entry_cnt >= max)
2124                 goto add_out;
2125
2126         list_for_each_entry(cur, head, set_list) {
2127                 if (cur->entry_cnt >= nes->entry_cnt) {
2128                         list_add(&nes->set_list, cur->set_list.prev);
2129                         return;
2130                 }
2131         }
2132 add_out:
2133         list_add_tail(&nes->set_list, head);
2134 }
2135
2136 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2137                                         struct nat_entry_set *set)
2138 {
2139         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2140         struct f2fs_journal *journal = curseg->journal;
2141         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2142         bool to_journal = true;
2143         struct f2fs_nat_block *nat_blk;
2144         struct nat_entry *ne, *cur;
2145         struct page *page = NULL;
2146
2147         /*
2148          * there are two steps to flush nat entries:
2149          * #1, flush nat entries to journal in current hot data summary block.
2150          * #2, flush nat entries to nat page.
2151          */
2152         if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2153                 to_journal = false;
2154
2155         if (to_journal) {
2156                 down_write(&curseg->journal_rwsem);
2157         } else {
2158                 page = get_next_nat_page(sbi, start_nid);
2159                 nat_blk = page_address(page);
2160                 f2fs_bug_on(sbi, !nat_blk);
2161         }
2162
2163         /* flush dirty nats in nat entry set */
2164         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2165                 struct f2fs_nat_entry *raw_ne;
2166                 nid_t nid = nat_get_nid(ne);
2167                 int offset;
2168
2169                 if (nat_get_blkaddr(ne) == NEW_ADDR)
2170                         continue;
2171
2172                 if (to_journal) {
2173                         offset = lookup_journal_in_cursum(journal,
2174                                                         NAT_JOURNAL, nid, 1);
2175                         f2fs_bug_on(sbi, offset < 0);
2176                         raw_ne = &nat_in_journal(journal, offset);
2177                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2178                 } else {
2179                         raw_ne = &nat_blk->entries[nid - start_nid];
2180                 }
2181                 raw_nat_from_node_info(raw_ne, &ne->ni);
2182                 nat_reset_flag(ne);
2183                 __clear_nat_cache_dirty(NM_I(sbi), ne);
2184                 if (nat_get_blkaddr(ne) == NULL_ADDR)
2185                         add_free_nid(sbi, nid, false);
2186         }
2187
2188         if (to_journal)
2189                 up_write(&curseg->journal_rwsem);
2190         else
2191                 f2fs_put_page(page, 1);
2192
2193         f2fs_bug_on(sbi, set->entry_cnt);
2194
2195         radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2196         kmem_cache_free(nat_entry_set_slab, set);
2197 }
2198
2199 /*
2200  * This function is called during the checkpointing process.
2201  */
2202 void flush_nat_entries(struct f2fs_sb_info *sbi)
2203 {
2204         struct f2fs_nm_info *nm_i = NM_I(sbi);
2205         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2206         struct f2fs_journal *journal = curseg->journal;
2207         struct nat_entry_set *setvec[SETVEC_SIZE];
2208         struct nat_entry_set *set, *tmp;
2209         unsigned int found;
2210         nid_t set_idx = 0;
2211         LIST_HEAD(sets);
2212
2213         if (!nm_i->dirty_nat_cnt)
2214                 return;
2215
2216         down_write(&nm_i->nat_tree_lock);
2217
2218         /*
2219          * if there are no enough space in journal to store dirty nat
2220          * entries, remove all entries from journal and merge them
2221          * into nat entry set.
2222          */
2223         if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2224                 remove_nats_in_journal(sbi);
2225
2226         while ((found = __gang_lookup_nat_set(nm_i,
2227                                         set_idx, SETVEC_SIZE, setvec))) {
2228                 unsigned idx;
2229                 set_idx = setvec[found - 1]->set + 1;
2230                 for (idx = 0; idx < found; idx++)
2231                         __adjust_nat_entry_set(setvec[idx], &sets,
2232                                                 MAX_NAT_JENTRIES(journal));
2233         }
2234
2235         /* flush dirty nats in nat entry set */
2236         list_for_each_entry_safe(set, tmp, &sets, set_list)
2237                 __flush_nat_entry_set(sbi, set);
2238
2239         up_write(&nm_i->nat_tree_lock);
2240
2241         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2242 }
2243
2244 static int init_node_manager(struct f2fs_sb_info *sbi)
2245 {
2246         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2247         struct f2fs_nm_info *nm_i = NM_I(sbi);
2248         unsigned char *version_bitmap;
2249         unsigned int nat_segs, nat_blocks;
2250
2251         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2252
2253         /* segment_count_nat includes pair segment so divide to 2. */
2254         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2255         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2256
2257         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2258
2259         /* not used nids: 0, node, meta, (and root counted as valid node) */
2260         nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2261         nm_i->fcnt = 0;
2262         nm_i->nat_cnt = 0;
2263         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2264         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2265         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2266
2267         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2268         INIT_LIST_HEAD(&nm_i->free_nid_list);
2269         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2270         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2271         INIT_LIST_HEAD(&nm_i->nat_entries);
2272
2273         mutex_init(&nm_i->build_lock);
2274         spin_lock_init(&nm_i->free_nid_list_lock);
2275         init_rwsem(&nm_i->nat_tree_lock);
2276
2277         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2278         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2279         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2280         if (!version_bitmap)
2281                 return -EFAULT;
2282
2283         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2284                                         GFP_KERNEL);
2285         if (!nm_i->nat_bitmap)
2286                 return -ENOMEM;
2287         return 0;
2288 }
2289
2290 int build_node_manager(struct f2fs_sb_info *sbi)
2291 {
2292         int err;
2293
2294         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2295         if (!sbi->nm_info)
2296                 return -ENOMEM;
2297
2298         err = init_node_manager(sbi);
2299         if (err)
2300                 return err;
2301
2302         build_free_nids(sbi);
2303         return 0;
2304 }
2305
2306 void destroy_node_manager(struct f2fs_sb_info *sbi)
2307 {
2308         struct f2fs_nm_info *nm_i = NM_I(sbi);
2309         struct free_nid *i, *next_i;
2310         struct nat_entry *natvec[NATVEC_SIZE];
2311         struct nat_entry_set *setvec[SETVEC_SIZE];
2312         nid_t nid = 0;
2313         unsigned int found;
2314
2315         if (!nm_i)
2316                 return;
2317
2318         /* destroy free nid list */
2319         spin_lock(&nm_i->free_nid_list_lock);
2320         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2321                 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2322                 __del_from_free_nid_list(nm_i, i);
2323                 nm_i->fcnt--;
2324                 spin_unlock(&nm_i->free_nid_list_lock);
2325                 kmem_cache_free(free_nid_slab, i);
2326                 spin_lock(&nm_i->free_nid_list_lock);
2327         }
2328         f2fs_bug_on(sbi, nm_i->fcnt);
2329         spin_unlock(&nm_i->free_nid_list_lock);
2330
2331         /* destroy nat cache */
2332         down_write(&nm_i->nat_tree_lock);
2333         while ((found = __gang_lookup_nat_cache(nm_i,
2334                                         nid, NATVEC_SIZE, natvec))) {
2335                 unsigned idx;
2336
2337                 nid = nat_get_nid(natvec[found - 1]) + 1;
2338                 for (idx = 0; idx < found; idx++)
2339                         __del_from_nat_cache(nm_i, natvec[idx]);
2340         }
2341         f2fs_bug_on(sbi, nm_i->nat_cnt);
2342
2343         /* destroy nat set cache */
2344         nid = 0;
2345         while ((found = __gang_lookup_nat_set(nm_i,
2346                                         nid, SETVEC_SIZE, setvec))) {
2347                 unsigned idx;
2348
2349                 nid = setvec[found - 1]->set + 1;
2350                 for (idx = 0; idx < found; idx++) {
2351                         /* entry_cnt is not zero, when cp_error was occurred */
2352                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2353                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2354                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2355                 }
2356         }
2357         up_write(&nm_i->nat_tree_lock);
2358
2359         kfree(nm_i->nat_bitmap);
2360         sbi->nm_info = NULL;
2361         kfree(nm_i);
2362 }
2363
2364 int __init create_node_manager_caches(void)
2365 {
2366         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2367                         sizeof(struct nat_entry));
2368         if (!nat_entry_slab)
2369                 goto fail;
2370
2371         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2372                         sizeof(struct free_nid));
2373         if (!free_nid_slab)
2374                 goto destroy_nat_entry;
2375
2376         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2377                         sizeof(struct nat_entry_set));
2378         if (!nat_entry_set_slab)
2379                 goto destroy_free_nid;
2380         return 0;
2381
2382 destroy_free_nid:
2383         kmem_cache_destroy(free_nid_slab);
2384 destroy_nat_entry:
2385         kmem_cache_destroy(nat_entry_slab);
2386 fail:
2387         return -ENOMEM;
2388 }
2389
2390 void destroy_node_manager_caches(void)
2391 {
2392         kmem_cache_destroy(nat_entry_set_slab);
2393         kmem_cache_destroy(free_nid_slab);
2394         kmem_cache_destroy(nat_entry_slab);
2395 }