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->nid_cnt[FREE_NID_LIST] *
  49                                 sizeof(struct free_nid)) >> 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                                                                 bool no_fail)
 250 {
 251         struct nat_entry *new;
 252
 253         if (no_fail) {
 254                 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
 255                 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
 256         } else {
 257                 new = kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
 258                 if (!new)
 259                         return NULL;
 260                 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 261                         kmem_cache_free(nat_entry_slab, new);
 262                         return NULL;
 263                 }
 264         }
 265
 266         memset(new, 0, sizeof(struct nat_entry));
 267         nat_set_nid(new, nid);
 268         nat_reset_flag(new);
 269         list_add_tail(&new->list, &nm_i->nat_entries);
 270         nm_i->nat_cnt++;
 271         return new;
 272 }
 273
 274 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 275                                                 struct f2fs_nat_entry *ne)
 276 {
 277         struct f2fs_nm_info *nm_i = NM_I(sbi);
 278         struct nat_entry *e;
 279
 280         e = __lookup_nat_cache(nm_i, nid);
 281         if (!e) {
 282                 e = grab_nat_entry(nm_i, nid, false);
 283                 if (e)
 284                         node_info_from_raw_nat(&e->ni, ne);
 285         } else {
 286                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 287                                 nat_get_blkaddr(e) !=
 288                                         le32_to_cpu(ne->block_addr) ||
 289                                 nat_get_version(e) != ne->version);
 290         }
 291 }
 292
 293 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 294                         block_t new_blkaddr, bool fsync_done)
 295 {
 296         struct f2fs_nm_info *nm_i = NM_I(sbi);
 297         struct nat_entry *e;
 298
 299         down_write(&nm_i->nat_tree_lock);
 300         e = __lookup_nat_cache(nm_i, ni->nid);
 301         if (!e) {
 302                 e = grab_nat_entry(nm_i, ni->nid, true);
 303                 copy_node_info(&e->ni, ni);
 304                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 305         } else if (new_blkaddr == NEW_ADDR) {
 306                 /*
 307                  * when nid is reallocated,
 308                  * previous nat entry can be remained in nat cache.
 309                  * So, reinitialize it with new information.
 310                  */
 311                 copy_node_info(&e->ni, ni);
 312                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 313         }
 314
 315         /* sanity check */
 316         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 317         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 318                         new_blkaddr == NULL_ADDR);
 319         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 320                         new_blkaddr == NEW_ADDR);
 321         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 322                         nat_get_blkaddr(e) != NULL_ADDR &&
 323                         new_blkaddr == NEW_ADDR);
 324
 325         /* increment version no as node is removed */
 326         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 327                 unsigned char version = nat_get_version(e);
 328                 nat_set_version(e, inc_node_version(version));
 329
 330                 /* in order to reuse the nid */
 331                 if (nm_i->next_scan_nid > ni->nid)
 332                         nm_i->next_scan_nid = ni->nid;
 333         }
 334
 335         /* change address */
 336         nat_set_blkaddr(e, new_blkaddr);
 337         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 338                 set_nat_flag(e, IS_CHECKPOINTED, false);
 339         __set_nat_cache_dirty(nm_i, e);
 340
 341         /* update fsync_mark if its inode nat entry is still alive */
 342         if (ni->nid != ni->ino)
 343                 e = __lookup_nat_cache(nm_i, ni->ino);
 344         if (e) {
 345                 if (fsync_done && ni->nid == ni->ino)
 346                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 347                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 348         }
 349         up_write(&nm_i->nat_tree_lock);
 350 }
 351
 352 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 353 {
 354         struct f2fs_nm_info *nm_i = NM_I(sbi);
 355         int nr = nr_shrink;
 356
 357         if (!down_write_trylock(&nm_i->nat_tree_lock))
 358                 return 0;
 359
 360         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 361                 struct nat_entry *ne;
 362                 ne = list_first_entry(&nm_i->nat_entries,
 363                                         struct nat_entry, list);
 364                 __del_from_nat_cache(nm_i, ne);
 365                 nr_shrink--;
 366         }
 367         up_write(&nm_i->nat_tree_lock);
 368         return nr - nr_shrink;
 369 }
 370
 371 /*
 372  * This function always returns success
 373  */
 374 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 375 {
 376         struct f2fs_nm_info *nm_i = NM_I(sbi);
 377         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 378         struct f2fs_journal *journal = curseg->journal;
 379         nid_t start_nid = START_NID(nid);
 380         struct f2fs_nat_block *nat_blk;
 381         struct page *page = NULL;
 382         struct f2fs_nat_entry ne;
 383         struct nat_entry *e;
 384         int i;
 385
 386         ni->nid = nid;
 387
 388         /* Check nat cache */
 389         down_read(&nm_i->nat_tree_lock);
 390         e = __lookup_nat_cache(nm_i, nid);
 391         if (e) {
 392                 ni->ino = nat_get_ino(e);
 393                 ni->blk_addr = nat_get_blkaddr(e);
 394                 ni->version = nat_get_version(e);
 395                 up_read(&nm_i->nat_tree_lock);
 396                 return;
 397         }
 398
 399         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 400
 401         /* Check current segment summary */
 402         down_read(&curseg->journal_rwsem);
 403         i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 404         if (i >= 0) {
 405                 ne = nat_in_journal(journal, i);
 406                 node_info_from_raw_nat(ni, &ne);
 407         }
 408         up_read(&curseg->journal_rwsem);
 409         if (i >= 0)
 410                 goto cache;
 411
 412         /* Fill node_info from nat page */
 413         page = get_current_nat_page(sbi, start_nid);
 414         nat_blk = (struct f2fs_nat_block *)page_address(page);
 415         ne = nat_blk->entries[nid - start_nid];
 416         node_info_from_raw_nat(ni, &ne);
 417         f2fs_put_page(page, 1);
 418 cache:
 419         up_read(&nm_i->nat_tree_lock);
 420         /* cache nat entry */
 421         down_write(&nm_i->nat_tree_lock);
 422         cache_nat_entry(sbi, nid, &ne);
 423         up_write(&nm_i->nat_tree_lock);
 424 }
 425
 426 /*
 427  * readahead MAX_RA_NODE number of node pages.
 428  */
 429 static void ra_node_pages(struct page *parent, int start, int n)
 430 {
 431         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 432         struct blk_plug plug;
 433         int i, end;
 434         nid_t nid;
 435
 436         blk_start_plug(&plug);
 437
 438         /* Then, try readahead for siblings of the desired node */
 439         end = start + n;
 440         end = min(end, NIDS_PER_BLOCK);
 441         for (i = start; i < end; i++) {
 442                 nid = get_nid(parent, i, false);
 443                 ra_node_page(sbi, nid);
 444         }
 445
 446         blk_finish_plug(&plug);
 447 }
 448
 449 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 450 {
 451         const long direct_index = ADDRS_PER_INODE(dn->inode);
 452         const long direct_blks = ADDRS_PER_BLOCK;
 453         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 454         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 455         int cur_level = dn->cur_level;
 456         int max_level = dn->max_level;
 457         pgoff_t base = 0;
 458
 459         if (!dn->max_level)
 460                 return pgofs + 1;
 461
 462         while (max_level-- > cur_level)
 463                 skipped_unit *= NIDS_PER_BLOCK;
 464
 465         switch (dn->max_level) {
 466         case 3:
 467                 base += 2 * indirect_blks;
 468         case 2:
 469                 base += 2 * direct_blks;
 470         case 1:
 471                 base += direct_index;
 472                 break;
 473         default:
 474                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 475         }
 476
 477         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 478 }
 479
 480 /*
 481  * The maximum depth is four.
 482  * Offset[0] will have raw inode offset.
 483  */
 484 static int get_node_path(struct inode *inode, long block,
 485                                 int offset[4], unsigned int noffset[4])
 486 {
 487         const long direct_index = ADDRS_PER_INODE(inode);
 488         const long direct_blks = ADDRS_PER_BLOCK;
 489         const long dptrs_per_blk = NIDS_PER_BLOCK;
 490         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 491         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 492         int n = 0;
 493         int level = 0;
 494
 495         noffset[0] = 0;
 496
 497         if (block < direct_index) {
 498                 offset[n] = block;
 499                 goto got;
 500         }
 501         block -= direct_index;
 502         if (block < direct_blks) {
 503                 offset[n++] = NODE_DIR1_BLOCK;
 504                 noffset[n] = 1;
 505                 offset[n] = block;
 506                 level = 1;
 507                 goto got;
 508         }
 509         block -= direct_blks;
 510         if (block < direct_blks) {
 511                 offset[n++] = NODE_DIR2_BLOCK;
 512                 noffset[n] = 2;
 513                 offset[n] = block;
 514                 level = 1;
 515                 goto got;
 516         }
 517         block -= direct_blks;
 518         if (block < indirect_blks) {
 519                 offset[n++] = NODE_IND1_BLOCK;
 520                 noffset[n] = 3;
 521                 offset[n++] = block / direct_blks;
 522                 noffset[n] = 4 + offset[n - 1];
 523                 offset[n] = block % direct_blks;
 524                 level = 2;
 525                 goto got;
 526         }
 527         block -= indirect_blks;
 528         if (block < indirect_blks) {
 529                 offset[n++] = NODE_IND2_BLOCK;
 530                 noffset[n] = 4 + dptrs_per_blk;
 531                 offset[n++] = block / direct_blks;
 532                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 533                 offset[n] = block % direct_blks;
 534                 level = 2;
 535                 goto got;
 536         }
 537         block -= indirect_blks;
 538         if (block < dindirect_blks) {
 539                 offset[n++] = NODE_DIND_BLOCK;
 540                 noffset[n] = 5 + (dptrs_per_blk * 2);
 541                 offset[n++] = block / indirect_blks;
 542                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 543                               offset[n - 1] * (dptrs_per_blk + 1);
 544                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 545                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 546                               offset[n - 2] * (dptrs_per_blk + 1) +
 547                               offset[n - 1];
 548                 offset[n] = block % direct_blks;
 549                 level = 3;
 550                 goto got;
 551         } else {
 552                 BUG();
 553         }
 554 got:
 555         return level;
 556 }
 557
 558 /*
 559  * Caller should call f2fs_put_dnode(dn).
 560  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 561  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 562  * In the case of RDONLY_NODE, we don't need to care about mutex.
 563  */
 564 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 565 {
 566         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 567         struct page *npage[4];
 568         struct page *parent = NULL;
 569         int offset[4];
 570         unsigned int noffset[4];
 571         nid_t nids[4];
 572         int level, i = 0;
 573         int err = 0;
 574
 575         level = get_node_path(dn->inode, index, offset, noffset);
 576
 577         nids[0] = dn->inode->i_ino;
 578         npage[0] = dn->inode_page;
 579
 580         if (!npage[0]) {
 581                 npage[0] = get_node_page(sbi, nids[0]);
 582                 if (IS_ERR(npage[0]))
 583                         return PTR_ERR(npage[0]);
 584         }
 585
 586         /* if inline_data is set, should not report any block indices */
 587         if (f2fs_has_inline_data(dn->inode) && index) {
 588                 err = -ENOENT;
 589                 f2fs_put_page(npage[0], 1);
 590                 goto release_out;
 591         }
 592
 593         parent = npage[0];
 594         if (level != 0)
 595                 nids[1] = get_nid(parent, offset[0], true);
 596         dn->inode_page = npage[0];
 597         dn->inode_page_locked = true;
 598
 599         /* get indirect or direct nodes */
 600         for (i = 1; i <= level; i++) {
 601                 bool done = false;
 602
 603                 if (!nids[i] && mode == ALLOC_NODE) {
 604                         /* alloc new node */
 605                         if (!alloc_nid(sbi, &(nids[i]))) {
 606                                 err = -ENOSPC;
 607                                 goto release_pages;
 608                         }
 609
 610                         dn->nid = nids[i];
 611                         npage[i] = new_node_page(dn, noffset[i], NULL);
 612                         if (IS_ERR(npage[i])) {
 613                                 alloc_nid_failed(sbi, nids[i]);
 614                                 err = PTR_ERR(npage[i]);
 615                                 goto release_pages;
 616                         }
 617
 618                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 619                         alloc_nid_done(sbi, nids[i]);
 620                         done = true;
 621                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 622                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 623                         if (IS_ERR(npage[i])) {
 624                                 err = PTR_ERR(npage[i]);
 625                                 goto release_pages;
 626                         }
 627                         done = true;
 628                 }
 629                 if (i == 1) {
 630                         dn->inode_page_locked = false;
 631                         unlock_page(parent);
 632                 } else {
 633                         f2fs_put_page(parent, 1);
 634                 }
 635
 636                 if (!done) {
 637                         npage[i] = get_node_page(sbi, nids[i]);
 638                         if (IS_ERR(npage[i])) {
 639                                 err = PTR_ERR(npage[i]);
 640                                 f2fs_put_page(npage[0], 0);
 641                                 goto release_out;
 642                         }
 643                 }
 644                 if (i < level) {
 645                         parent = npage[i];
 646                         nids[i + 1] = get_nid(parent, offset[i], false);
 647                 }
 648         }
 649         dn->nid = nids[level];
 650         dn->ofs_in_node = offset[level];
 651         dn->node_page = npage[level];
 652         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 653         return 0;
 654
 655 release_pages:
 656         f2fs_put_page(parent, 1);
 657         if (i > 1)
 658                 f2fs_put_page(npage[0], 0);
 659 release_out:
 660         dn->inode_page = NULL;
 661         dn->node_page = NULL;
 662         if (err == -ENOENT) {
 663                 dn->cur_level = i;
 664                 dn->max_level = level;
 665                 dn->ofs_in_node = offset[level];
 666         }
 667         return err;
 668 }
 669
 670 static void truncate_node(struct dnode_of_data *dn)
 671 {
 672         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 673         struct node_info ni;
 674
 675         get_node_info(sbi, dn->nid, &ni);
 676         if (dn->inode->i_blocks == 0) {
 677                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 678                 goto invalidate;
 679         }
 680         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 681
 682         /* Deallocate node address */
 683         invalidate_blocks(sbi, ni.blk_addr);
 684         dec_valid_node_count(sbi, dn->inode);
 685         set_node_addr(sbi, &ni, NULL_ADDR, false);
 686
 687         if (dn->nid == dn->inode->i_ino) {
 688                 remove_orphan_inode(sbi, dn->nid);
 689                 dec_valid_inode_count(sbi);
 690                 f2fs_inode_synced(dn->inode);
 691         }
 692 invalidate:
 693         clear_node_page_dirty(dn->node_page);
 694         set_sbi_flag(sbi, SBI_IS_DIRTY);
 695
 696         f2fs_put_page(dn->node_page, 1);
 697
 698         invalidate_mapping_pages(NODE_MAPPING(sbi),
 699                         dn->node_page->index, dn->node_page->index);
 700
 701         dn->node_page = NULL;
 702         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 703 }
 704
 705 static int truncate_dnode(struct dnode_of_data *dn)
 706 {
 707         struct page *page;
 708
 709         if (dn->nid == 0)
 710                 return 1;
 711
 712         /* get direct node */
 713         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 714         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 715                 return 1;
 716         else if (IS_ERR(page))
 717                 return PTR_ERR(page);
 718
 719         /* Make dnode_of_data for parameter */
 720         dn->node_page = page;
 721         dn->ofs_in_node = 0;
 722         truncate_data_blocks(dn);
 723         truncate_node(dn);
 724         return 1;
 725 }
 726
 727 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 728                                                 int ofs, int depth)
 729 {
 730         struct dnode_of_data rdn = *dn;
 731         struct page *page;
 732         struct f2fs_node *rn;
 733         nid_t child_nid;
 734         unsigned int child_nofs;
 735         int freed = 0;
 736         int i, ret;
 737
 738         if (dn->nid == 0)
 739                 return NIDS_PER_BLOCK + 1;
 740
 741         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 742
 743         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 744         if (IS_ERR(page)) {
 745                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 746                 return PTR_ERR(page);
 747         }
 748
 749         ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 750
 751         rn = F2FS_NODE(page);
 752         if (depth < 3) {
 753                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 754                         child_nid = le32_to_cpu(rn->in.nid[i]);
 755                         if (child_nid == 0)
 756                                 continue;
 757                         rdn.nid = child_nid;
 758                         ret = truncate_dnode(&rdn);
 759                         if (ret < 0)
 760                                 goto out_err;
 761                         if (set_nid(page, i, 0, false))
 762                                 dn->node_changed = true;
 763                 }
 764         } else {
 765                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 766                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 767                         child_nid = le32_to_cpu(rn->in.nid[i]);
 768                         if (child_nid == 0) {
 769                                 child_nofs += NIDS_PER_BLOCK + 1;
 770                                 continue;
 771                         }
 772                         rdn.nid = child_nid;
 773                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 774                         if (ret == (NIDS_PER_BLOCK + 1)) {
 775                                 if (set_nid(page, i, 0, false))
 776                                         dn->node_changed = true;
 777                                 child_nofs += ret;
 778                         } else if (ret < 0 && ret != -ENOENT) {
 779                                 goto out_err;
 780                         }
 781                 }
 782                 freed = child_nofs;
 783         }
 784
 785         if (!ofs) {
 786                 /* remove current indirect node */
 787                 dn->node_page = page;
 788                 truncate_node(dn);
 789                 freed++;
 790         } else {
 791                 f2fs_put_page(page, 1);
 792         }
 793         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 794         return freed;
 795
 796 out_err:
 797         f2fs_put_page(page, 1);
 798         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 799         return ret;
 800 }
 801
 802 static int truncate_partial_nodes(struct dnode_of_data *dn,
 803                         struct f2fs_inode *ri, int *offset, int depth)
 804 {
 805         struct page *pages[2];
 806         nid_t nid[3];
 807         nid_t child_nid;
 808         int err = 0;
 809         int i;
 810         int idx = depth - 2;
 811
 812         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 813         if (!nid[0])
 814                 return 0;
 815
 816         /* get indirect nodes in the path */
 817         for (i = 0; i < idx + 1; i++) {
 818                 /* reference count'll be increased */
 819                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 820                 if (IS_ERR(pages[i])) {
 821                         err = PTR_ERR(pages[i]);
 822                         idx = i - 1;
 823                         goto fail;
 824                 }
 825                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 826         }
 827
 828         ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 829
 830         /* free direct nodes linked to a partial indirect node */
 831         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 832                 child_nid = get_nid(pages[idx], i, false);
 833                 if (!child_nid)
 834                         continue;
 835                 dn->nid = child_nid;
 836                 err = truncate_dnode(dn);
 837                 if (err < 0)
 838                         goto fail;
 839                 if (set_nid(pages[idx], i, 0, false))
 840                         dn->node_changed = true;
 841         }
 842
 843         if (offset[idx + 1] == 0) {
 844                 dn->node_page = pages[idx];
 845                 dn->nid = nid[idx];
 846                 truncate_node(dn);
 847         } else {
 848                 f2fs_put_page(pages[idx], 1);
 849         }
 850         offset[idx]++;
 851         offset[idx + 1] = 0;
 852         idx--;
 853 fail:
 854         for (i = idx; i >= 0; i--)
 855                 f2fs_put_page(pages[i], 1);
 856
 857         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 858
 859         return err;
 860 }
 861
 862 /*
 863  * All the block addresses of data and nodes should be nullified.
 864  */
 865 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 866 {
 867         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 868         int err = 0, cont = 1;
 869         int level, offset[4], noffset[4];
 870         unsigned int nofs = 0;
 871         struct f2fs_inode *ri;
 872         struct dnode_of_data dn;
 873         struct page *page;
 874
 875         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 876
 877         level = get_node_path(inode, from, offset, noffset);
 878
 879         page = get_node_page(sbi, inode->i_ino);
 880         if (IS_ERR(page)) {
 881                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 882                 return PTR_ERR(page);
 883         }
 884
 885         set_new_dnode(&dn, inode, page, NULL, 0);
 886         unlock_page(page);
 887
 888         ri = F2FS_INODE(page);
 889         switch (level) {
 890         case 0:
 891         case 1:
 892                 nofs = noffset[1];
 893                 break;
 894         case 2:
 895                 nofs = noffset[1];
 896                 if (!offset[level - 1])
 897                         goto skip_partial;
 898                 err = truncate_partial_nodes(&dn, ri, offset, level);
 899                 if (err < 0 && err != -ENOENT)
 900                         goto fail;
 901                 nofs += 1 + NIDS_PER_BLOCK;
 902                 break;
 903         case 3:
 904                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 905                 if (!offset[level - 1])
 906                         goto skip_partial;
 907                 err = truncate_partial_nodes(&dn, ri, offset, level);
 908                 if (err < 0 && err != -ENOENT)
 909                         goto fail;
 910                 break;
 911         default:
 912                 BUG();
 913         }
 914
 915 skip_partial:
 916         while (cont) {
 917                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 918                 switch (offset[0]) {
 919                 case NODE_DIR1_BLOCK:
 920                 case NODE_DIR2_BLOCK:
 921                         err = truncate_dnode(&dn);
 922                         break;
 923
 924                 case NODE_IND1_BLOCK:
 925                 case NODE_IND2_BLOCK:
 926                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 927                         break;
 928
 929                 case NODE_DIND_BLOCK:
 930                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 931                         cont = 0;
 932                         break;
 933
 934                 default:
 935                         BUG();
 936                 }
 937                 if (err < 0 && err != -ENOENT)
 938                         goto fail;
 939                 if (offset[1] == 0 &&
 940                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 941                         lock_page(page);
 942                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
 943                         f2fs_wait_on_page_writeback(page, NODE, true);
 944                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 945                         set_page_dirty(page);
 946                         unlock_page(page);
 947                 }
 948                 offset[1] = 0;
 949                 offset[0]++;
 950                 nofs += err;
 951         }
 952 fail:
 953         f2fs_put_page(page, 0);
 954         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 955         return err > 0 ? 0 : err;
 956 }
 957
 958 int truncate_xattr_node(struct inode *inode, struct page *page)
 959 {
 960         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 961         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 962         struct dnode_of_data dn;
 963         struct page *npage;
 964
 965         if (!nid)
 966                 return 0;
 967
 968         npage = get_node_page(sbi, nid);
 969         if (IS_ERR(npage))
 970                 return PTR_ERR(npage);
 971
 972         f2fs_i_xnid_write(inode, 0);
 973
 974         /* need to do checkpoint during fsync */
 975         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 976
 977         set_new_dnode(&dn, inode, page, npage, nid);
 978
 979         if (page)
 980                 dn.inode_page_locked = true;
 981         truncate_node(&dn);
 982         return 0;
 983 }
 984
 985 /*
 986  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 987  * f2fs_unlock_op().
 988  */
 989 int remove_inode_page(struct inode *inode)
 990 {
 991         struct dnode_of_data dn;
 992         int err;
 993
 994         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 995         err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
 996         if (err)
 997                 return err;
 998
 999         err = truncate_xattr_node(inode, dn.inode_page);
1000         if (err) {
1001                 f2fs_put_dnode(&dn);
1002                 return err;
1003         }
1004
1005         /* remove potential inline_data blocks */
1006         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1007                                 S_ISLNK(inode->i_mode))
1008                 truncate_data_blocks_range(&dn, 1);
1009
1010         /* 0 is possible, after f2fs_new_inode() has failed */
1011         f2fs_bug_on(F2FS_I_SB(inode),
1012                         inode->i_blocks != 0 && inode->i_blocks != 1);
1013
1014         /* will put inode & node pages */
1015         truncate_node(&dn);
1016         return 0;
1017 }
1018
1019 struct page *new_inode_page(struct inode *inode)
1020 {
1021         struct dnode_of_data dn;
1022
1023         /* allocate inode page for new inode */
1024         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1025
1026         /* caller should f2fs_put_page(page, 1); */
1027         return new_node_page(&dn, 0, NULL);
1028 }
1029
1030 struct page *new_node_page(struct dnode_of_data *dn,
1031                                 unsigned int ofs, struct page *ipage)
1032 {
1033         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1034         struct node_info old_ni, new_ni;
1035         struct page *page;
1036         int err;
1037
1038         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1039                 return ERR_PTR(-EPERM);
1040
1041         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1042         if (!page)
1043                 return ERR_PTR(-ENOMEM);
1044
1045         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1046                 err = -ENOSPC;
1047                 goto fail;
1048         }
1049
1050         get_node_info(sbi, dn->nid, &old_ni);
1051
1052         /* Reinitialize old_ni with new node page */
1053         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1054         new_ni = old_ni;
1055         new_ni.ino = dn->inode->i_ino;
1056         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1057
1058         f2fs_wait_on_page_writeback(page, NODE, true);
1059         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1060         set_cold_node(dn->inode, page);
1061         if (!PageUptodate(page))
1062                 SetPageUptodate(page);
1063         if (set_page_dirty(page))
1064                 dn->node_changed = true;
1065
1066         if (f2fs_has_xattr_block(ofs))
1067                 f2fs_i_xnid_write(dn->inode, dn->nid);
1068
1069         if (ofs == 0)
1070                 inc_valid_inode_count(sbi);
1071         return page;
1072
1073 fail:
1074         clear_node_page_dirty(page);
1075         f2fs_put_page(page, 1);
1076         return ERR_PTR(err);
1077 }
1078
1079 /*
1080  * Caller should do after getting the following values.
1081  * 0: f2fs_put_page(page, 0)
1082  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1083  */
1084 static int read_node_page(struct page *page, int op_flags)
1085 {
1086         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1087         struct node_info ni;
1088         struct f2fs_io_info fio = {
1089                 .sbi = sbi,
1090                 .type = NODE,
1091                 .op = REQ_OP_READ,
1092                 .op_flags = op_flags,
1093                 .page = page,
1094                 .encrypted_page = NULL,
1095         };
1096
1097         if (PageUptodate(page))
1098                 return LOCKED_PAGE;
1099
1100         get_node_info(sbi, page->index, &ni);
1101
1102         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1103                 ClearPageUptodate(page);
1104                 return -ENOENT;
1105         }
1106
1107         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1108         return f2fs_submit_page_bio(&fio);
1109 }
1110
1111 /*
1112  * Readahead a node page
1113  */
1114 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1115 {
1116         struct page *apage;
1117         int err;
1118
1119         if (!nid)
1120                 return;
1121         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1122
1123         rcu_read_lock();
1124         apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1125         rcu_read_unlock();
1126         if (apage)
1127                 return;
1128
1129         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1130         if (!apage)
1131                 return;
1132
1133         err = read_node_page(apage, REQ_RAHEAD);
1134         f2fs_put_page(apage, err ? 1 : 0);
1135 }
1136
1137 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1138                                         struct page *parent, int start)
1139 {
1140         struct page *page;
1141         int err;
1142
1143         if (!nid)
1144                 return ERR_PTR(-ENOENT);
1145         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1146 repeat:
1147         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1148         if (!page)
1149                 return ERR_PTR(-ENOMEM);
1150
1151         err = read_node_page(page, 0);
1152         if (err < 0) {
1153                 f2fs_put_page(page, 1);
1154                 return ERR_PTR(err);
1155         } else if (err == LOCKED_PAGE) {
1156                 goto page_hit;
1157         }
1158
1159         if (parent)
1160                 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1161
1162         lock_page(page);
1163
1164         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1165                 f2fs_put_page(page, 1);
1166                 goto repeat;
1167         }
1168
1169         if (unlikely(!PageUptodate(page)))
1170                 goto out_err;
1171 page_hit:
1172         if(unlikely(nid != nid_of_node(page))) {
1173                 f2fs_bug_on(sbi, 1);
1174                 ClearPageUptodate(page);
1175 out_err:
1176                 f2fs_put_page(page, 1);
1177                 return ERR_PTR(-EIO);
1178         }
1179         return page;
1180 }
1181
1182 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1183 {
1184         return __get_node_page(sbi, nid, NULL, 0);
1185 }
1186
1187 struct page *get_node_page_ra(struct page *parent, int start)
1188 {
1189         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1190         nid_t nid = get_nid(parent, start, false);
1191
1192         return __get_node_page(sbi, nid, parent, start);
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         int ret;
1200
1201         /* should flush inline_data before evict_inode */
1202         inode = ilookup(sbi->sb, ino);
1203         if (!inode)
1204                 return;
1205
1206         page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1207         if (!page)
1208                 goto iput_out;
1209
1210         if (!PageUptodate(page))
1211                 goto page_out;
1212
1213         if (!PageDirty(page))
1214                 goto page_out;
1215
1216         if (!clear_page_dirty_for_io(page))
1217                 goto page_out;
1218
1219         ret = f2fs_write_inline_data(inode, page);
1220         inode_dec_dirty_pages(inode);
1221         remove_dirty_inode(inode);
1222         if (ret)
1223                 set_page_dirty(page);
1224 page_out:
1225         f2fs_put_page(page, 1);
1226 iput_out:
1227         iput(inode);
1228 }
1229
1230 void move_node_page(struct page *node_page, int gc_type)
1231 {
1232         if (gc_type == FG_GC) {
1233                 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1234                 struct writeback_control wbc = {
1235                         .sync_mode = WB_SYNC_ALL,
1236                         .nr_to_write = 1,
1237                         .for_reclaim = 0,
1238                 };
1239
1240                 set_page_dirty(node_page);
1241                 f2fs_wait_on_page_writeback(node_page, NODE, true);
1242
1243                 f2fs_bug_on(sbi, PageWriteback(node_page));
1244                 if (!clear_page_dirty_for_io(node_page))
1245                         goto out_page;
1246
1247                 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1248                         unlock_page(node_page);
1249                 goto release_page;
1250         } else {
1251                 /* set page dirty and write it */
1252                 if (!PageWriteback(node_page))
1253                         set_page_dirty(node_page);
1254         }
1255 out_page:
1256         unlock_page(node_page);
1257 release_page:
1258         f2fs_put_page(node_page, 0);
1259 }
1260
1261 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1262 {
1263         pgoff_t index, end;
1264         struct pagevec pvec;
1265         struct page *last_page = NULL;
1266
1267         pagevec_init(&pvec, 0);
1268         index = 0;
1269         end = ULONG_MAX;
1270
1271         while (index <= end) {
1272                 int i, nr_pages;
1273                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1274                                 PAGECACHE_TAG_DIRTY,
1275                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1276                 if (nr_pages == 0)
1277                         break;
1278
1279                 for (i = 0; i < nr_pages; i++) {
1280                         struct page *page = pvec.pages[i];
1281
1282                         if (unlikely(f2fs_cp_error(sbi))) {
1283                                 f2fs_put_page(last_page, 0);
1284                                 pagevec_release(&pvec);
1285                                 return ERR_PTR(-EIO);
1286                         }
1287
1288                         if (!IS_DNODE(page) || !is_cold_node(page))
1289                                 continue;
1290                         if (ino_of_node(page) != ino)
1291                                 continue;
1292
1293                         lock_page(page);
1294
1295                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1296 continue_unlock:
1297                                 unlock_page(page);
1298                                 continue;
1299                         }
1300                         if (ino_of_node(page) != ino)
1301                                 goto continue_unlock;
1302
1303                         if (!PageDirty(page)) {
1304                                 /* someone wrote it for us */
1305                                 goto continue_unlock;
1306                         }
1307
1308                         if (last_page)
1309                                 f2fs_put_page(last_page, 0);
1310
1311                         get_page(page);
1312                         last_page = page;
1313                         unlock_page(page);
1314                 }
1315                 pagevec_release(&pvec);
1316                 cond_resched();
1317         }
1318         return last_page;
1319 }
1320
1321 static int __write_node_page(struct page *page, bool atomic,
1322                                 struct writeback_control *wbc)
1323 {
1324         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1325         nid_t nid;
1326         struct node_info ni;
1327         struct f2fs_io_info fio = {
1328                 .sbi = sbi,
1329                 .type = NODE,
1330                 .op = REQ_OP_WRITE,
1331                 .op_flags = wbc_to_write_flags(wbc),
1332                 .page = page,
1333                 .encrypted_page = NULL,
1334         };
1335
1336         trace_f2fs_writepage(page, NODE);
1337
1338         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1339                 goto redirty_out;
1340         if (unlikely(f2fs_cp_error(sbi)))
1341                 goto redirty_out;
1342
1343         /* get old block addr of this node page */
1344         nid = nid_of_node(page);
1345         f2fs_bug_on(sbi, page->index != nid);
1346
1347         if (wbc->for_reclaim) {
1348                 if (!down_read_trylock(&sbi->node_write))
1349                         goto redirty_out;
1350         } else {
1351                 down_read(&sbi->node_write);
1352         }
1353
1354         get_node_info(sbi, nid, &ni);
1355
1356         /* This page is already truncated */
1357         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1358                 ClearPageUptodate(page);
1359                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1360                 up_read(&sbi->node_write);
1361                 unlock_page(page);
1362                 return 0;
1363         }
1364
1365         if (atomic && !test_opt(sbi, NOBARRIER))
1366                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1367
1368         set_page_writeback(page);
1369         fio.old_blkaddr = ni.blk_addr;
1370         write_node_page(nid, &fio);
1371         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1372         dec_page_count(sbi, F2FS_DIRTY_NODES);
1373         up_read(&sbi->node_write);
1374
1375         if (wbc->for_reclaim)
1376                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1377
1378         unlock_page(page);
1379
1380         if (unlikely(f2fs_cp_error(sbi)))
1381                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1382
1383         return 0;
1384
1385 redirty_out:
1386         redirty_page_for_writepage(wbc, page);
1387         return AOP_WRITEPAGE_ACTIVATE;
1388 }
1389
1390 static int f2fs_write_node_page(struct page *page,
1391                                 struct writeback_control *wbc)
1392 {
1393         return __write_node_page(page, false, wbc);
1394 }
1395
1396 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1397                         struct writeback_control *wbc, bool atomic)
1398 {
1399         pgoff_t index, end;
1400         struct pagevec pvec;
1401         int ret = 0;
1402         struct page *last_page = NULL;
1403         bool marked = false;
1404         nid_t ino = inode->i_ino;
1405         int nwritten = 0;
1406
1407         if (atomic) {
1408                 last_page = last_fsync_dnode(sbi, ino);
1409                 if (IS_ERR_OR_NULL(last_page))
1410                         return PTR_ERR_OR_ZERO(last_page);
1411         }
1412 retry:
1413         pagevec_init(&pvec, 0);
1414         index = 0;
1415         end = ULONG_MAX;
1416
1417         while (index <= end) {
1418                 int i, nr_pages;
1419                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1420                                 PAGECACHE_TAG_DIRTY,
1421                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1422                 if (nr_pages == 0)
1423                         break;
1424
1425                 for (i = 0; i < nr_pages; i++) {
1426                         struct page *page = pvec.pages[i];
1427
1428                         if (unlikely(f2fs_cp_error(sbi))) {
1429                                 f2fs_put_page(last_page, 0);
1430                                 pagevec_release(&pvec);
1431                                 ret = -EIO;
1432                                 goto out;
1433                         }
1434
1435                         if (!IS_DNODE(page) || !is_cold_node(page))
1436                                 continue;
1437                         if (ino_of_node(page) != ino)
1438                                 continue;
1439
1440                         lock_page(page);
1441
1442                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1443 continue_unlock:
1444                                 unlock_page(page);
1445                                 continue;
1446                         }
1447                         if (ino_of_node(page) != ino)
1448                                 goto continue_unlock;
1449
1450                         if (!PageDirty(page) && page != last_page) {
1451                                 /* someone wrote it for us */
1452                                 goto continue_unlock;
1453                         }
1454
1455                         f2fs_wait_on_page_writeback(page, NODE, true);
1456                         BUG_ON(PageWriteback(page));
1457
1458                         if (!atomic || page == last_page) {
1459                                 set_fsync_mark(page, 1);
1460                                 if (IS_INODE(page)) {
1461                                         if (is_inode_flag_set(inode,
1462                                                                 FI_DIRTY_INODE))
1463                                                 update_inode(inode, page);
1464                                         set_dentry_mark(page,
1465                                                 need_dentry_mark(sbi, ino));
1466                                 }
1467                                 /*  may be written by other thread */
1468                                 if (!PageDirty(page))
1469                                         set_page_dirty(page);
1470                         }
1471
1472                         if (!clear_page_dirty_for_io(page))
1473                                 goto continue_unlock;
1474
1475                         ret = __write_node_page(page, atomic &&
1476                                                 page == last_page, wbc);
1477                         if (ret) {
1478                                 unlock_page(page);
1479                                 f2fs_put_page(last_page, 0);
1480                                 break;
1481                         } else {
1482                                 nwritten++;
1483                         }
1484
1485                         if (page == last_page) {
1486                                 f2fs_put_page(page, 0);
1487                                 marked = true;
1488                                 break;
1489                         }
1490                 }
1491                 pagevec_release(&pvec);
1492                 cond_resched();
1493
1494                 if (ret || marked)
1495                         break;
1496         }
1497         if (!ret && atomic && !marked) {
1498                 f2fs_msg(sbi->sb, KERN_DEBUG,
1499                         "Retry to write fsync mark: ino=%u, idx=%lx",
1500                                         ino, last_page->index);
1501                 lock_page(last_page);
1502                 f2fs_wait_on_page_writeback(last_page, NODE, true);
1503                 set_page_dirty(last_page);
1504                 unlock_page(last_page);
1505                 goto retry;
1506         }
1507 out:
1508         if (nwritten)
1509                 f2fs_submit_merged_bio_cond(sbi, NULL, NULL, ino, NODE, WRITE);
1510         return ret ? -EIO: 0;
1511 }
1512
1513 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1514 {
1515         pgoff_t index, end;
1516         struct pagevec pvec;
1517         int step = 0;
1518         int nwritten = 0;
1519         int ret = 0;
1520
1521         pagevec_init(&pvec, 0);
1522
1523 next_step:
1524         index = 0;
1525         end = ULONG_MAX;
1526
1527         while (index <= end) {
1528                 int i, nr_pages;
1529                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1530                                 PAGECACHE_TAG_DIRTY,
1531                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1532                 if (nr_pages == 0)
1533                         break;
1534
1535                 for (i = 0; i < nr_pages; i++) {
1536                         struct page *page = pvec.pages[i];
1537
1538                         if (unlikely(f2fs_cp_error(sbi))) {
1539                                 pagevec_release(&pvec);
1540                                 ret = -EIO;
1541                                 goto out;
1542                         }
1543
1544                         /*
1545                          * flushing sequence with step:
1546                          * 0. indirect nodes
1547                          * 1. dentry dnodes
1548                          * 2. file dnodes
1549                          */
1550                         if (step == 0 && IS_DNODE(page))
1551                                 continue;
1552                         if (step == 1 && (!IS_DNODE(page) ||
1553                                                 is_cold_node(page)))
1554                                 continue;
1555                         if (step == 2 && (!IS_DNODE(page) ||
1556                                                 !is_cold_node(page)))
1557                                 continue;
1558 lock_node:
1559                         if (!trylock_page(page))
1560                                 continue;
1561
1562                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1563 continue_unlock:
1564                                 unlock_page(page);
1565                                 continue;
1566                         }
1567
1568                         if (!PageDirty(page)) {
1569                                 /* someone wrote it for us */
1570                                 goto continue_unlock;
1571                         }
1572
1573                         /* flush inline_data */
1574                         if (is_inline_node(page)) {
1575                                 clear_inline_node(page);
1576                                 unlock_page(page);
1577                                 flush_inline_data(sbi, ino_of_node(page));
1578                                 goto lock_node;
1579                         }
1580
1581                         f2fs_wait_on_page_writeback(page, NODE, true);
1582
1583                         BUG_ON(PageWriteback(page));
1584                         if (!clear_page_dirty_for_io(page))
1585                                 goto continue_unlock;
1586
1587                         set_fsync_mark(page, 0);
1588                         set_dentry_mark(page, 0);
1589
1590                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1591                                 unlock_page(page);
1592                         else
1593                                 nwritten++;
1594
1595                         if (--wbc->nr_to_write == 0)
1596                                 break;
1597                 }
1598                 pagevec_release(&pvec);
1599                 cond_resched();
1600
1601                 if (wbc->nr_to_write == 0) {
1602                         step = 2;
1603                         break;
1604                 }
1605         }
1606
1607         if (step < 2) {
1608                 step++;
1609                 goto next_step;
1610         }
1611 out:
1612         if (nwritten)
1613                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1614         return ret;
1615 }
1616
1617 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1618 {
1619         pgoff_t index = 0, end = ULONG_MAX;
1620         struct pagevec pvec;
1621         int ret2, ret = 0;
1622
1623         pagevec_init(&pvec, 0);
1624
1625         while (index <= end) {
1626                 int i, nr_pages;
1627                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1628                                 PAGECACHE_TAG_WRITEBACK,
1629                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1630                 if (nr_pages == 0)
1631                         break;
1632
1633                 for (i = 0; i < nr_pages; i++) {
1634                         struct page *page = pvec.pages[i];
1635
1636                         /* until radix tree lookup accepts end_index */
1637                         if (unlikely(page->index > end))
1638                                 continue;
1639
1640                         if (ino && ino_of_node(page) == ino) {
1641                                 f2fs_wait_on_page_writeback(page, NODE, true);
1642                                 if (TestClearPageError(page))
1643                                         ret = -EIO;
1644                         }
1645                 }
1646                 pagevec_release(&pvec);
1647                 cond_resched();
1648         }
1649
1650         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1651         if (!ret)
1652                 ret = ret2;
1653         return ret;
1654 }
1655
1656 static int f2fs_write_node_pages(struct address_space *mapping,
1657                             struct writeback_control *wbc)
1658 {
1659         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1660         struct blk_plug plug;
1661         long diff;
1662
1663         /* balancing f2fs's metadata in background */
1664         f2fs_balance_fs_bg(sbi);
1665
1666         /* collect a number of dirty node pages and write together */
1667         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1668                 goto skip_write;
1669
1670         trace_f2fs_writepages(mapping->host, wbc, NODE);
1671
1672         diff = nr_pages_to_write(sbi, NODE, wbc);
1673         wbc->sync_mode = WB_SYNC_NONE;
1674         blk_start_plug(&plug);
1675         sync_node_pages(sbi, wbc);
1676         blk_finish_plug(&plug);
1677         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1678         return 0;
1679
1680 skip_write:
1681         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1682         trace_f2fs_writepages(mapping->host, wbc, NODE);
1683         return 0;
1684 }
1685
1686 static int f2fs_set_node_page_dirty(struct page *page)
1687 {
1688         trace_f2fs_set_page_dirty(page, NODE);
1689
1690         if (!PageUptodate(page))
1691                 SetPageUptodate(page);
1692         if (!PageDirty(page)) {
1693                 f2fs_set_page_dirty_nobuffers(page);
1694                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1695                 SetPagePrivate(page);
1696                 f2fs_trace_pid(page);
1697                 return 1;
1698         }
1699         return 0;
1700 }
1701
1702 /*
1703  * Structure of the f2fs node operations
1704  */
1705 const struct address_space_operations f2fs_node_aops = {
1706         .writepage      = f2fs_write_node_page,
1707         .writepages     = f2fs_write_node_pages,
1708         .set_page_dirty = f2fs_set_node_page_dirty,
1709         .invalidatepage = f2fs_invalidate_page,
1710         .releasepage    = f2fs_release_page,
1711 #ifdef CONFIG_MIGRATION
1712         .migratepage    = f2fs_migrate_page,
1713 #endif
1714 };
1715
1716 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1717                                                 nid_t n)
1718 {
1719         return radix_tree_lookup(&nm_i->free_nid_root, n);
1720 }
1721
1722 static int __insert_nid_to_list(struct f2fs_sb_info *sbi,
1723                         struct free_nid *i, enum nid_list list, bool new)
1724 {
1725         struct f2fs_nm_info *nm_i = NM_I(sbi);
1726
1727         if (new) {
1728                 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1729                 if (err)
1730                         return err;
1731         }
1732
1733         f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1734                                                 i->state != NID_ALLOC);
1735         nm_i->nid_cnt[list]++;
1736         list_add_tail(&i->list, &nm_i->nid_list[list]);
1737         return 0;
1738 }
1739
1740 static void __remove_nid_from_list(struct f2fs_sb_info *sbi,
1741                         struct free_nid *i, enum nid_list list, bool reuse)
1742 {
1743         struct f2fs_nm_info *nm_i = NM_I(sbi);
1744
1745         f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1746                                                 i->state != NID_ALLOC);
1747         nm_i->nid_cnt[list]--;
1748         list_del(&i->list);
1749         if (!reuse)
1750                 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1751 }
1752
1753 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1754 {
1755         struct f2fs_nm_info *nm_i = NM_I(sbi);
1756         struct free_nid *i;
1757         struct nat_entry *ne;
1758         int err;
1759
1760         /* 0 nid should not be used */
1761         if (unlikely(nid == 0))
1762                 return 0;
1763
1764         if (build) {
1765                 /* do not add allocated nids */
1766                 ne = __lookup_nat_cache(nm_i, nid);
1767                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1768                                 nat_get_blkaddr(ne) != NULL_ADDR))
1769                         return 0;
1770         }
1771
1772         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1773         i->nid = nid;
1774         i->state = NID_NEW;
1775
1776         if (radix_tree_preload(GFP_NOFS)) {
1777                 kmem_cache_free(free_nid_slab, i);
1778                 return 0;
1779         }
1780
1781         spin_lock(&nm_i->nid_list_lock);
1782         err = __insert_nid_to_list(sbi, i, FREE_NID_LIST, true);
1783         spin_unlock(&nm_i->nid_list_lock);
1784         radix_tree_preload_end();
1785         if (err) {
1786                 kmem_cache_free(free_nid_slab, i);
1787                 return 0;
1788         }
1789         return 1;
1790 }
1791
1792 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1793 {
1794         struct f2fs_nm_info *nm_i = NM_I(sbi);
1795         struct free_nid *i;
1796         bool need_free = false;
1797
1798         spin_lock(&nm_i->nid_list_lock);
1799         i = __lookup_free_nid_list(nm_i, nid);
1800         if (i && i->state == NID_NEW) {
1801                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
1802                 need_free = true;
1803         }
1804         spin_unlock(&nm_i->nid_list_lock);
1805
1806         if (need_free)
1807                 kmem_cache_free(free_nid_slab, i);
1808 }
1809
1810 static void scan_nat_page(struct f2fs_sb_info *sbi,
1811                         struct page *nat_page, nid_t start_nid)
1812 {
1813         struct f2fs_nm_info *nm_i = NM_I(sbi);
1814         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1815         block_t blk_addr;
1816         int i;
1817
1818         i = start_nid % NAT_ENTRY_PER_BLOCK;
1819
1820         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1821
1822                 if (unlikely(start_nid >= nm_i->max_nid))
1823                         break;
1824
1825                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1826                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1827                 if (blk_addr == NULL_ADDR)
1828                         add_free_nid(sbi, start_nid, true);
1829         }
1830 }
1831
1832 static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync)
1833 {
1834         struct f2fs_nm_info *nm_i = NM_I(sbi);
1835         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1836         struct f2fs_journal *journal = curseg->journal;
1837         int i = 0;
1838         nid_t nid = nm_i->next_scan_nid;
1839
1840         /* Enough entries */
1841         if (nm_i->nid_cnt[FREE_NID_LIST] >= NAT_ENTRY_PER_BLOCK)
1842                 return;
1843
1844         if (!sync && !available_free_memory(sbi, FREE_NIDS))
1845                 return;
1846
1847         /* readahead nat pages to be scanned */
1848         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1849                                                         META_NAT, true);
1850
1851         down_read(&nm_i->nat_tree_lock);
1852
1853         while (1) {
1854                 struct page *page = get_current_nat_page(sbi, nid);
1855
1856                 scan_nat_page(sbi, page, nid);
1857                 f2fs_put_page(page, 1);
1858
1859                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1860                 if (unlikely(nid >= nm_i->max_nid))
1861                         nid = 0;
1862
1863                 if (++i >= FREE_NID_PAGES)
1864                         break;
1865         }
1866
1867         /* go to the next free nat pages to find free nids abundantly */
1868         nm_i->next_scan_nid = nid;
1869
1870         /* find free nids from current sum_pages */
1871         down_read(&curseg->journal_rwsem);
1872         for (i = 0; i < nats_in_cursum(journal); i++) {
1873                 block_t addr;
1874
1875                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1876                 nid = le32_to_cpu(nid_in_journal(journal, i));
1877                 if (addr == NULL_ADDR)
1878                         add_free_nid(sbi, nid, true);
1879                 else
1880                         remove_free_nid(sbi, nid);
1881         }
1882         up_read(&curseg->journal_rwsem);
1883         up_read(&nm_i->nat_tree_lock);
1884
1885         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1886                                         nm_i->ra_nid_pages, META_NAT, false);
1887 }
1888
1889 void build_free_nids(struct f2fs_sb_info *sbi, bool sync)
1890 {
1891         mutex_lock(&NM_I(sbi)->build_lock);
1892         __build_free_nids(sbi, sync);
1893         mutex_unlock(&NM_I(sbi)->build_lock);
1894 }
1895
1896 /*
1897  * If this function returns success, caller can obtain a new nid
1898  * from second parameter of this function.
1899  * The returned nid could be used ino as well as nid when inode is created.
1900  */
1901 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1902 {
1903         struct f2fs_nm_info *nm_i = NM_I(sbi);
1904         struct free_nid *i = NULL;
1905 retry:
1906 #ifdef CONFIG_F2FS_FAULT_INJECTION
1907         if (time_to_inject(sbi, FAULT_ALLOC_NID))
1908                 return false;
1909 #endif
1910         spin_lock(&nm_i->nid_list_lock);
1911
1912         if (unlikely(nm_i->available_nids == 0)) {
1913                 spin_unlock(&nm_i->nid_list_lock);
1914                 return false;
1915         }
1916
1917         /* We should not use stale free nids created by build_free_nids */
1918         if (nm_i->nid_cnt[FREE_NID_LIST] && !on_build_free_nids(nm_i)) {
1919                 f2fs_bug_on(sbi, list_empty(&nm_i->nid_list[FREE_NID_LIST]));
1920                 i = list_first_entry(&nm_i->nid_list[FREE_NID_LIST],
1921                                         struct free_nid, list);
1922                 *nid = i->nid;
1923
1924                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, true);
1925                 i->state = NID_ALLOC;
1926                 __insert_nid_to_list(sbi, i, ALLOC_NID_LIST, false);
1927                 nm_i->available_nids--;
1928                 spin_unlock(&nm_i->nid_list_lock);
1929                 return true;
1930         }
1931         spin_unlock(&nm_i->nid_list_lock);
1932
1933         /* Let's scan nat pages and its caches to get free nids */
1934         build_free_nids(sbi, true);
1935         goto retry;
1936 }
1937
1938 /*
1939  * alloc_nid() should be called prior to this function.
1940  */
1941 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1942 {
1943         struct f2fs_nm_info *nm_i = NM_I(sbi);
1944         struct free_nid *i;
1945
1946         spin_lock(&nm_i->nid_list_lock);
1947         i = __lookup_free_nid_list(nm_i, nid);
1948         f2fs_bug_on(sbi, !i);
1949         __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
1950         spin_unlock(&nm_i->nid_list_lock);
1951
1952         kmem_cache_free(free_nid_slab, i);
1953 }
1954
1955 /*
1956  * alloc_nid() should be called prior to this function.
1957  */
1958 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1959 {
1960         struct f2fs_nm_info *nm_i = NM_I(sbi);
1961         struct free_nid *i;
1962         bool need_free = false;
1963
1964         if (!nid)
1965                 return;
1966
1967         spin_lock(&nm_i->nid_list_lock);
1968         i = __lookup_free_nid_list(nm_i, nid);
1969         f2fs_bug_on(sbi, !i);
1970
1971         if (!available_free_memory(sbi, FREE_NIDS)) {
1972                 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
1973                 need_free = true;
1974         } else {
1975                 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, true);
1976                 i->state = NID_NEW;
1977                 __insert_nid_to_list(sbi, i, FREE_NID_LIST, false);
1978         }
1979
1980         nm_i->available_nids++;
1981
1982         spin_unlock(&nm_i->nid_list_lock);
1983
1984         if (need_free)
1985                 kmem_cache_free(free_nid_slab, i);
1986 }
1987
1988 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1989 {
1990         struct f2fs_nm_info *nm_i = NM_I(sbi);
1991         struct free_nid *i, *next;
1992         int nr = nr_shrink;
1993
1994         if (nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
1995                 return 0;
1996
1997         if (!mutex_trylock(&nm_i->build_lock))
1998                 return 0;
1999
2000         spin_lock(&nm_i->nid_list_lock);
2001         list_for_each_entry_safe(i, next, &nm_i->nid_list[FREE_NID_LIST],
2002                                                                         list) {
2003                 if (nr_shrink <= 0 ||
2004                                 nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
2005                         break;
2006
2007                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2008                 kmem_cache_free(free_nid_slab, i);
2009                 nr_shrink--;
2010         }
2011         spin_unlock(&nm_i->nid_list_lock);
2012         mutex_unlock(&nm_i->build_lock);
2013
2014         return nr - nr_shrink;
2015 }
2016
2017 void recover_inline_xattr(struct inode *inode, struct page *page)
2018 {
2019         void *src_addr, *dst_addr;
2020         size_t inline_size;
2021         struct page *ipage;
2022         struct f2fs_inode *ri;
2023
2024         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2025         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2026
2027         ri = F2FS_INODE(page);
2028         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2029                 clear_inode_flag(inode, FI_INLINE_XATTR);
2030                 goto update_inode;
2031         }
2032
2033         dst_addr = inline_xattr_addr(ipage);
2034         src_addr = inline_xattr_addr(page);
2035         inline_size = inline_xattr_size(inode);
2036
2037         f2fs_wait_on_page_writeback(ipage, NODE, true);
2038         memcpy(dst_addr, src_addr, inline_size);
2039 update_inode:
2040         update_inode(inode, ipage);
2041         f2fs_put_page(ipage, 1);
2042 }
2043
2044 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2045 {
2046         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2047         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2048         nid_t new_xnid = nid_of_node(page);
2049         struct node_info ni;
2050
2051         /* 1: invalidate the previous xattr nid */
2052         if (!prev_xnid)
2053                 goto recover_xnid;
2054
2055         /* Deallocate node address */
2056         get_node_info(sbi, prev_xnid, &ni);
2057         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2058         invalidate_blocks(sbi, ni.blk_addr);
2059         dec_valid_node_count(sbi, inode);
2060         set_node_addr(sbi, &ni, NULL_ADDR, false);
2061
2062 recover_xnid:
2063         /* 2: allocate new xattr nid */
2064         if (unlikely(!inc_valid_node_count(sbi, inode)))
2065                 f2fs_bug_on(sbi, 1);
2066
2067         remove_free_nid(sbi, new_xnid);
2068         get_node_info(sbi, new_xnid, &ni);
2069         ni.ino = inode->i_ino;
2070         set_node_addr(sbi, &ni, NEW_ADDR, false);
2071         f2fs_i_xnid_write(inode, new_xnid);
2072
2073         /* 3: update xattr blkaddr */
2074         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
2075         set_node_addr(sbi, &ni, blkaddr, false);
2076 }
2077
2078 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2079 {
2080         struct f2fs_inode *src, *dst;
2081         nid_t ino = ino_of_node(page);
2082         struct node_info old_ni, new_ni;
2083         struct page *ipage;
2084
2085         get_node_info(sbi, ino, &old_ni);
2086
2087         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2088                 return -EINVAL;
2089 retry:
2090         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2091         if (!ipage) {
2092                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2093                 goto retry;
2094         }
2095
2096         /* Should not use this inode from free nid list */
2097         remove_free_nid(sbi, ino);
2098
2099         if (!PageUptodate(ipage))
2100                 SetPageUptodate(ipage);
2101         fill_node_footer(ipage, ino, ino, 0, true);
2102
2103         src = F2FS_INODE(page);
2104         dst = F2FS_INODE(ipage);
2105
2106         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2107         dst->i_size = 0;
2108         dst->i_blocks = cpu_to_le64(1);
2109         dst->i_links = cpu_to_le32(1);
2110         dst->i_xattr_nid = 0;
2111         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2112
2113         new_ni = old_ni;
2114         new_ni.ino = ino;
2115
2116         if (unlikely(!inc_valid_node_count(sbi, NULL)))
2117                 WARN_ON(1);
2118         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2119         inc_valid_inode_count(sbi);
2120         set_page_dirty(ipage);
2121         f2fs_put_page(ipage, 1);
2122         return 0;
2123 }
2124
2125 int restore_node_summary(struct f2fs_sb_info *sbi,
2126                         unsigned int segno, struct f2fs_summary_block *sum)
2127 {
2128         struct f2fs_node *rn;
2129         struct f2fs_summary *sum_entry;
2130         block_t addr;
2131         int i, idx, last_offset, nrpages;
2132
2133         /* scan the node segment */
2134         last_offset = sbi->blocks_per_seg;
2135         addr = START_BLOCK(sbi, segno);
2136         sum_entry = &sum->entries[0];
2137
2138         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2139                 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2140
2141                 /* readahead node pages */
2142                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2143
2144                 for (idx = addr; idx < addr + nrpages; idx++) {
2145                         struct page *page = get_tmp_page(sbi, idx);
2146
2147                         rn = F2FS_NODE(page);
2148                         sum_entry->nid = rn->footer.nid;
2149                         sum_entry->version = 0;
2150                         sum_entry->ofs_in_node = 0;
2151                         sum_entry++;
2152                         f2fs_put_page(page, 1);
2153                 }
2154
2155                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2156                                                         addr + nrpages);
2157         }
2158         return 0;
2159 }
2160
2161 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2162 {
2163         struct f2fs_nm_info *nm_i = NM_I(sbi);
2164         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2165         struct f2fs_journal *journal = curseg->journal;
2166         int i;
2167
2168         down_write(&curseg->journal_rwsem);
2169         for (i = 0; i < nats_in_cursum(journal); i++) {
2170                 struct nat_entry *ne;
2171                 struct f2fs_nat_entry raw_ne;
2172                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2173
2174                 raw_ne = nat_in_journal(journal, i);
2175
2176                 ne = __lookup_nat_cache(nm_i, nid);
2177                 if (!ne) {
2178                         ne = grab_nat_entry(nm_i, nid, true);
2179                         node_info_from_raw_nat(&ne->ni, &raw_ne);
2180                 }
2181
2182                 /*
2183                  * if a free nat in journal has not been used after last
2184                  * checkpoint, we should remove it from available nids,
2185                  * since later we will add it again.
2186                  */
2187                 if (!get_nat_flag(ne, IS_DIRTY) &&
2188                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2189                         spin_lock(&nm_i->nid_list_lock);
2190                         nm_i->available_nids--;
2191                         spin_unlock(&nm_i->nid_list_lock);
2192                 }
2193
2194                 __set_nat_cache_dirty(nm_i, ne);
2195         }
2196         update_nats_in_cursum(journal, -i);
2197         up_write(&curseg->journal_rwsem);
2198 }
2199
2200 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2201                                                 struct list_head *head, int max)
2202 {
2203         struct nat_entry_set *cur;
2204
2205         if (nes->entry_cnt >= max)
2206                 goto add_out;
2207
2208         list_for_each_entry(cur, head, set_list) {
2209                 if (cur->entry_cnt >= nes->entry_cnt) {
2210                         list_add(&nes->set_list, cur->set_list.prev);
2211                         return;
2212                 }
2213         }
2214 add_out:
2215         list_add_tail(&nes->set_list, head);
2216 }
2217
2218 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2219                                         struct nat_entry_set *set)
2220 {
2221         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2222         struct f2fs_journal *journal = curseg->journal;
2223         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2224         bool to_journal = true;
2225         struct f2fs_nat_block *nat_blk;
2226         struct nat_entry *ne, *cur;
2227         struct page *page = NULL;
2228
2229         /*
2230          * there are two steps to flush nat entries:
2231          * #1, flush nat entries to journal in current hot data summary block.
2232          * #2, flush nat entries to nat page.
2233          */
2234         if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2235                 to_journal = false;
2236
2237         if (to_journal) {
2238                 down_write(&curseg->journal_rwsem);
2239         } else {
2240                 page = get_next_nat_page(sbi, start_nid);
2241                 nat_blk = page_address(page);
2242                 f2fs_bug_on(sbi, !nat_blk);
2243         }
2244
2245         /* flush dirty nats in nat entry set */
2246         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2247                 struct f2fs_nat_entry *raw_ne;
2248                 nid_t nid = nat_get_nid(ne);
2249                 int offset;
2250
2251                 if (nat_get_blkaddr(ne) == NEW_ADDR)
2252                         continue;
2253
2254                 if (to_journal) {
2255                         offset = lookup_journal_in_cursum(journal,
2256                                                         NAT_JOURNAL, nid, 1);
2257                         f2fs_bug_on(sbi, offset < 0);
2258                         raw_ne = &nat_in_journal(journal, offset);
2259                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2260                 } else {
2261                         raw_ne = &nat_blk->entries[nid - start_nid];
2262                 }
2263                 raw_nat_from_node_info(raw_ne, &ne->ni);
2264                 nat_reset_flag(ne);
2265                 __clear_nat_cache_dirty(NM_I(sbi), ne);
2266                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2267                         add_free_nid(sbi, nid, false);
2268                         spin_lock(&NM_I(sbi)->nid_list_lock);
2269                         NM_I(sbi)->available_nids++;
2270                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2271                 }
2272         }
2273
2274         if (to_journal)
2275                 up_write(&curseg->journal_rwsem);
2276         else
2277                 f2fs_put_page(page, 1);
2278
2279         f2fs_bug_on(sbi, set->entry_cnt);
2280
2281         radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2282         kmem_cache_free(nat_entry_set_slab, set);
2283 }
2284
2285 /*
2286  * This function is called during the checkpointing process.
2287  */
2288 void flush_nat_entries(struct f2fs_sb_info *sbi)
2289 {
2290         struct f2fs_nm_info *nm_i = NM_I(sbi);
2291         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2292         struct f2fs_journal *journal = curseg->journal;
2293         struct nat_entry_set *setvec[SETVEC_SIZE];
2294         struct nat_entry_set *set, *tmp;
2295         unsigned int found;
2296         nid_t set_idx = 0;
2297         LIST_HEAD(sets);
2298
2299         if (!nm_i->dirty_nat_cnt)
2300                 return;
2301
2302         down_write(&nm_i->nat_tree_lock);
2303
2304         /*
2305          * if there are no enough space in journal to store dirty nat
2306          * entries, remove all entries from journal and merge them
2307          * into nat entry set.
2308          */
2309         if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2310                 remove_nats_in_journal(sbi);
2311
2312         while ((found = __gang_lookup_nat_set(nm_i,
2313                                         set_idx, SETVEC_SIZE, setvec))) {
2314                 unsigned idx;
2315                 set_idx = setvec[found - 1]->set + 1;
2316                 for (idx = 0; idx < found; idx++)
2317                         __adjust_nat_entry_set(setvec[idx], &sets,
2318                                                 MAX_NAT_JENTRIES(journal));
2319         }
2320
2321         /* flush dirty nats in nat entry set */
2322         list_for_each_entry_safe(set, tmp, &sets, set_list)
2323                 __flush_nat_entry_set(sbi, set);
2324
2325         up_write(&nm_i->nat_tree_lock);
2326
2327         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2328 }
2329
2330 static int init_node_manager(struct f2fs_sb_info *sbi)
2331 {
2332         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2333         struct f2fs_nm_info *nm_i = NM_I(sbi);
2334         unsigned char *version_bitmap;
2335         unsigned int nat_segs, nat_blocks;
2336
2337         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2338
2339         /* segment_count_nat includes pair segment so divide to 2. */
2340         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2341         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2342
2343         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2344
2345         /* not used nids: 0, node, meta, (and root counted as valid node) */
2346         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2347                                                         F2FS_RESERVED_NODE_NUM;
2348         nm_i->nid_cnt[FREE_NID_LIST] = 0;
2349         nm_i->nid_cnt[ALLOC_NID_LIST] = 0;
2350         nm_i->nat_cnt = 0;
2351         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2352         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2353         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2354
2355         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2356         INIT_LIST_HEAD(&nm_i->nid_list[FREE_NID_LIST]);
2357         INIT_LIST_HEAD(&nm_i->nid_list[ALLOC_NID_LIST]);
2358         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2359         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2360         INIT_LIST_HEAD(&nm_i->nat_entries);
2361
2362         mutex_init(&nm_i->build_lock);
2363         spin_lock_init(&nm_i->nid_list_lock);
2364         init_rwsem(&nm_i->nat_tree_lock);
2365
2366         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2367         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2368         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2369         if (!version_bitmap)
2370                 return -EFAULT;
2371
2372         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2373                                         GFP_KERNEL);
2374         if (!nm_i->nat_bitmap)
2375                 return -ENOMEM;
2376
2377 #ifdef CONFIG_F2FS_CHECK_FS
2378         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2379                                         GFP_KERNEL);
2380         if (!nm_i->nat_bitmap_mir)
2381                 return -ENOMEM;
2382 #endif
2383
2384         return 0;
2385 }
2386
2387 int build_node_manager(struct f2fs_sb_info *sbi)
2388 {
2389         int err;
2390
2391         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2392         if (!sbi->nm_info)
2393                 return -ENOMEM;
2394
2395         err = init_node_manager(sbi);
2396         if (err)
2397                 return err;
2398
2399         build_free_nids(sbi, true);
2400         return 0;
2401 }
2402
2403 void destroy_node_manager(struct f2fs_sb_info *sbi)
2404 {
2405         struct f2fs_nm_info *nm_i = NM_I(sbi);
2406         struct free_nid *i, *next_i;
2407         struct nat_entry *natvec[NATVEC_SIZE];
2408         struct nat_entry_set *setvec[SETVEC_SIZE];
2409         nid_t nid = 0;
2410         unsigned int found;
2411
2412         if (!nm_i)
2413                 return;
2414
2415         /* destroy free nid list */
2416         spin_lock(&nm_i->nid_list_lock);
2417         list_for_each_entry_safe(i, next_i, &nm_i->nid_list[FREE_NID_LIST],
2418                                                                         list) {
2419                 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2420                 spin_unlock(&nm_i->nid_list_lock);
2421                 kmem_cache_free(free_nid_slab, i);
2422                 spin_lock(&nm_i->nid_list_lock);
2423         }
2424         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID_LIST]);
2425         f2fs_bug_on(sbi, nm_i->nid_cnt[ALLOC_NID_LIST]);
2426         f2fs_bug_on(sbi, !list_empty(&nm_i->nid_list[ALLOC_NID_LIST]));
2427         spin_unlock(&nm_i->nid_list_lock);
2428
2429         /* destroy nat cache */
2430         down_write(&nm_i->nat_tree_lock);
2431         while ((found = __gang_lookup_nat_cache(nm_i,
2432                                         nid, NATVEC_SIZE, natvec))) {
2433                 unsigned idx;
2434
2435                 nid = nat_get_nid(natvec[found - 1]) + 1;
2436                 for (idx = 0; idx < found; idx++)
2437                         __del_from_nat_cache(nm_i, natvec[idx]);
2438         }
2439         f2fs_bug_on(sbi, nm_i->nat_cnt);
2440
2441         /* destroy nat set cache */
2442         nid = 0;
2443         while ((found = __gang_lookup_nat_set(nm_i,
2444                                         nid, SETVEC_SIZE, setvec))) {
2445                 unsigned idx;
2446
2447                 nid = setvec[found - 1]->set + 1;
2448                 for (idx = 0; idx < found; idx++) {
2449                         /* entry_cnt is not zero, when cp_error was occurred */
2450                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2451                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2452                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2453                 }
2454         }
2455         up_write(&nm_i->nat_tree_lock);
2456
2457         kfree(nm_i->nat_bitmap);
2458 #ifdef CONFIG_F2FS_CHECK_FS
2459         kfree(nm_i->nat_bitmap_mir);
2460 #endif
2461         sbi->nm_info = NULL;
2462         kfree(nm_i);
2463 }
2464
2465 int __init create_node_manager_caches(void)
2466 {
2467         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2468                         sizeof(struct nat_entry));
2469         if (!nat_entry_slab)
2470                 goto fail;
2471
2472         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2473                         sizeof(struct free_nid));
2474         if (!free_nid_slab)
2475                 goto destroy_nat_entry;
2476
2477         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2478                         sizeof(struct nat_entry_set));
2479         if (!nat_entry_set_slab)
2480                 goto destroy_free_nid;
2481         return 0;
2482
2483 destroy_free_nid:
2484         kmem_cache_destroy(free_nid_slab);
2485 destroy_nat_entry:
2486         kmem_cache_destroy(nat_entry_slab);
2487 fail:
2488         return -ENOMEM;
2489 }
2490
2491 void destroy_node_manager_caches(void)
2492 {
2493         kmem_cache_destroy(nat_entry_set_slab);
2494         kmem_cache_destroy(free_nid_slab);
2495         kmem_cache_destroy(nat_entry_slab);
2496 }