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         /* give 25%, 25%, 50%, 50% memory for each components respectively */
  45         if (type == FREE_NIDS) {
  46                 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
  47                                                         PAGE_CACHE_SHIFT;
  48                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  49         } else if (type == NAT_ENTRIES) {
  50                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
  51                                                         PAGE_CACHE_SHIFT;
  52                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  53         } else if (type == DIRTY_DENTS) {
  54                 if (sbi->sb->s_bdi->dirty_exceeded)
  55                         return false;
  56                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  57                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  58         } else if (type == INO_ENTRIES) {
  59                 int i;
  60
  61                 for (i = 0; i <= UPDATE_INO; i++)
  62                         mem_size += (sbi->im[i].ino_num *
  63                                 sizeof(struct ino_entry)) >> PAGE_CACHE_SHIFT;
  64                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  65         } else {
  66                 if (sbi->sb->s_bdi->dirty_exceeded)
  67                         return false;
  68         }
  69         return res;
  70 }
  71
  72 static void clear_node_page_dirty(struct page *page)
  73 {
  74         struct address_space *mapping = page->mapping;
  75         unsigned int long flags;
  76
  77         if (PageDirty(page)) {
  78                 spin_lock_irqsave(&mapping->tree_lock, flags);
  79                 radix_tree_tag_clear(&mapping->page_tree,
  80                                 page_index(page),
  81                                 PAGECACHE_TAG_DIRTY);
  82                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  83
  84                 clear_page_dirty_for_io(page);
  85                 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
  86         }
  87         ClearPageUptodate(page);
  88 }
  89
  90 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  91 {
  92         pgoff_t index = current_nat_addr(sbi, nid);
  93         return get_meta_page(sbi, index);
  94 }
  95
  96 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  97 {
  98         struct page *src_page;
  99         struct page *dst_page;
 100         pgoff_t src_off;
 101         pgoff_t dst_off;
 102         void *src_addr;
 103         void *dst_addr;
 104         struct f2fs_nm_info *nm_i = NM_I(sbi);
 105
 106         src_off = current_nat_addr(sbi, nid);
 107         dst_off = next_nat_addr(sbi, src_off);
 108
 109         /* get current nat block page with lock */
 110         src_page = get_meta_page(sbi, src_off);
 111         dst_page = grab_meta_page(sbi, dst_off);
 112         f2fs_bug_on(sbi, PageDirty(src_page));
 113
 114         src_addr = page_address(src_page);
 115         dst_addr = page_address(dst_page);
 116         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
 117         set_page_dirty(dst_page);
 118         f2fs_put_page(src_page, 1);
 119
 120         set_to_next_nat(nm_i, nid);
 121
 122         return dst_page;
 123 }
 124
 125 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 126 {
 127         return radix_tree_lookup(&nm_i->nat_root, n);
 128 }
 129
 130 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 131                 nid_t start, unsigned int nr, struct nat_entry **ep)
 132 {
 133         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 134 }
 135
 136 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 137 {
 138         list_del(&e->list);
 139         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 140         nm_i->nat_cnt--;
 141         kmem_cache_free(nat_entry_slab, e);
 142 }
 143
 144 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 145                                                 struct nat_entry *ne)
 146 {
 147         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 148         struct nat_entry_set *head;
 149
 150         if (get_nat_flag(ne, IS_DIRTY))
 151                 return;
 152
 153         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 154         if (!head) {
 155                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_ATOMIC);
 156
 157                 INIT_LIST_HEAD(&head->entry_list);
 158                 INIT_LIST_HEAD(&head->set_list);
 159                 head->set = set;
 160                 head->entry_cnt = 0;
 161                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 162         }
 163         list_move_tail(&ne->list, &head->entry_list);
 164         nm_i->dirty_nat_cnt++;
 165         head->entry_cnt++;
 166         set_nat_flag(ne, IS_DIRTY, true);
 167 }
 168
 169 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 170                                                 struct nat_entry *ne)
 171 {
 172         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 173         struct nat_entry_set *head;
 174
 175         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 176         if (head) {
 177                 list_move_tail(&ne->list, &nm_i->nat_entries);
 178                 set_nat_flag(ne, IS_DIRTY, false);
 179                 head->entry_cnt--;
 180                 nm_i->dirty_nat_cnt--;
 181         }
 182 }
 183
 184 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 185                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 186 {
 187         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 188                                                         start, nr);
 189 }
 190
 191 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 192 {
 193         struct f2fs_nm_info *nm_i = NM_I(sbi);
 194         struct nat_entry *e;
 195         bool is_cp = true;
 196
 197         down_read(&nm_i->nat_tree_lock);
 198         e = __lookup_nat_cache(nm_i, nid);
 199         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 200                 is_cp = false;
 201         up_read(&nm_i->nat_tree_lock);
 202         return is_cp;
 203 }
 204
 205 bool has_fsynced_inode(struct f2fs_sb_info *sbi, nid_t ino)
 206 {
 207         struct f2fs_nm_info *nm_i = NM_I(sbi);
 208         struct nat_entry *e;
 209         bool fsynced = false;
 210
 211         down_read(&nm_i->nat_tree_lock);
 212         e = __lookup_nat_cache(nm_i, ino);
 213         if (e && get_nat_flag(e, HAS_FSYNCED_INODE))
 214                 fsynced = true;
 215         up_read(&nm_i->nat_tree_lock);
 216         return fsynced;
 217 }
 218
 219 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 220 {
 221         struct f2fs_nm_info *nm_i = NM_I(sbi);
 222         struct nat_entry *e;
 223         bool need_update = true;
 224
 225         down_read(&nm_i->nat_tree_lock);
 226         e = __lookup_nat_cache(nm_i, ino);
 227         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 228                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 229                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 230                 need_update = false;
 231         up_read(&nm_i->nat_tree_lock);
 232         return need_update;
 233 }
 234
 235 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 236 {
 237         struct nat_entry *new;
 238
 239         new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 240         f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
 241         memset(new, 0, sizeof(struct nat_entry));
 242         nat_set_nid(new, nid);
 243         nat_reset_flag(new);
 244         list_add_tail(&new->list, &nm_i->nat_entries);
 245         nm_i->nat_cnt++;
 246         return new;
 247 }
 248
 249 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 250                                                 struct f2fs_nat_entry *ne)
 251 {
 252         struct nat_entry *e;
 253
 254         down_write(&nm_i->nat_tree_lock);
 255         e = __lookup_nat_cache(nm_i, nid);
 256         if (!e) {
 257                 e = grab_nat_entry(nm_i, nid);
 258                 node_info_from_raw_nat(&e->ni, ne);
 259         }
 260         up_write(&nm_i->nat_tree_lock);
 261 }
 262
 263 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 264                         block_t new_blkaddr, bool fsync_done)
 265 {
 266         struct f2fs_nm_info *nm_i = NM_I(sbi);
 267         struct nat_entry *e;
 268
 269         down_write(&nm_i->nat_tree_lock);
 270         e = __lookup_nat_cache(nm_i, ni->nid);
 271         if (!e) {
 272                 e = grab_nat_entry(nm_i, ni->nid);
 273                 copy_node_info(&e->ni, ni);
 274                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 275         } else if (new_blkaddr == NEW_ADDR) {
 276                 /*
 277                  * when nid is reallocated,
 278                  * previous nat entry can be remained in nat cache.
 279                  * So, reinitialize it with new information.
 280                  */
 281                 copy_node_info(&e->ni, ni);
 282                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 283         }
 284
 285         /* sanity check */
 286         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 287         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 288                         new_blkaddr == NULL_ADDR);
 289         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 290                         new_blkaddr == NEW_ADDR);
 291         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 292                         nat_get_blkaddr(e) != NULL_ADDR &&
 293                         new_blkaddr == NEW_ADDR);
 294
 295         /* increment version no as node is removed */
 296         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 297                 unsigned char version = nat_get_version(e);
 298                 nat_set_version(e, inc_node_version(version));
 299         }
 300
 301         /* change address */
 302         nat_set_blkaddr(e, new_blkaddr);
 303         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 304                 set_nat_flag(e, IS_CHECKPOINTED, false);
 305         __set_nat_cache_dirty(nm_i, e);
 306
 307         /* update fsync_mark if its inode nat entry is still alive */
 308         e = __lookup_nat_cache(nm_i, ni->ino);
 309         if (e) {
 310                 if (fsync_done && ni->nid == ni->ino)
 311                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 312                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 313         }
 314         up_write(&nm_i->nat_tree_lock);
 315 }
 316
 317 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 318 {
 319         struct f2fs_nm_info *nm_i = NM_I(sbi);
 320
 321         if (available_free_memory(sbi, NAT_ENTRIES))
 322                 return 0;
 323
 324         down_write(&nm_i->nat_tree_lock);
 325         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 326                 struct nat_entry *ne;
 327                 ne = list_first_entry(&nm_i->nat_entries,
 328                                         struct nat_entry, list);
 329                 __del_from_nat_cache(nm_i, ne);
 330                 nr_shrink--;
 331         }
 332         up_write(&nm_i->nat_tree_lock);
 333         return nr_shrink;
 334 }
 335
 336 /*
 337  * This function always returns success
 338  */
 339 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 340 {
 341         struct f2fs_nm_info *nm_i = NM_I(sbi);
 342         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 343         struct f2fs_summary_block *sum = curseg->sum_blk;
 344         nid_t start_nid = START_NID(nid);
 345         struct f2fs_nat_block *nat_blk;
 346         struct page *page = NULL;
 347         struct f2fs_nat_entry ne;
 348         struct nat_entry *e;
 349         int i;
 350
 351         ni->nid = nid;
 352
 353         /* Check nat cache */
 354         down_read(&nm_i->nat_tree_lock);
 355         e = __lookup_nat_cache(nm_i, nid);
 356         if (e) {
 357                 ni->ino = nat_get_ino(e);
 358                 ni->blk_addr = nat_get_blkaddr(e);
 359                 ni->version = nat_get_version(e);
 360         }
 361         up_read(&nm_i->nat_tree_lock);
 362         if (e)
 363                 return;
 364
 365         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 366
 367         /* Check current segment summary */
 368         mutex_lock(&curseg->curseg_mutex);
 369         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 370         if (i >= 0) {
 371                 ne = nat_in_journal(sum, i);
 372                 node_info_from_raw_nat(ni, &ne);
 373         }
 374         mutex_unlock(&curseg->curseg_mutex);
 375         if (i >= 0)
 376                 goto cache;
 377
 378         /* Fill node_info from nat page */
 379         page = get_current_nat_page(sbi, start_nid);
 380         nat_blk = (struct f2fs_nat_block *)page_address(page);
 381         ne = nat_blk->entries[nid - start_nid];
 382         node_info_from_raw_nat(ni, &ne);
 383         f2fs_put_page(page, 1);
 384 cache:
 385         /* cache nat entry */
 386         cache_nat_entry(NM_I(sbi), nid, &ne);
 387 }
 388
 389 /*
 390  * The maximum depth is four.
 391  * Offset[0] will have raw inode offset.
 392  */
 393 static int get_node_path(struct f2fs_inode_info *fi, long block,
 394                                 int offset[4], unsigned int noffset[4])
 395 {
 396         const long direct_index = ADDRS_PER_INODE(fi);
 397         const long direct_blks = ADDRS_PER_BLOCK;
 398         const long dptrs_per_blk = NIDS_PER_BLOCK;
 399         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 400         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 401         int n = 0;
 402         int level = 0;
 403
 404         noffset[0] = 0;
 405
 406         if (block < direct_index) {
 407                 offset[n] = block;
 408                 goto got;
 409         }
 410         block -= direct_index;
 411         if (block < direct_blks) {
 412                 offset[n++] = NODE_DIR1_BLOCK;
 413                 noffset[n] = 1;
 414                 offset[n] = block;
 415                 level = 1;
 416                 goto got;
 417         }
 418         block -= direct_blks;
 419         if (block < direct_blks) {
 420                 offset[n++] = NODE_DIR2_BLOCK;
 421                 noffset[n] = 2;
 422                 offset[n] = block;
 423                 level = 1;
 424                 goto got;
 425         }
 426         block -= direct_blks;
 427         if (block < indirect_blks) {
 428                 offset[n++] = NODE_IND1_BLOCK;
 429                 noffset[n] = 3;
 430                 offset[n++] = block / direct_blks;
 431                 noffset[n] = 4 + offset[n - 1];
 432                 offset[n] = block % direct_blks;
 433                 level = 2;
 434                 goto got;
 435         }
 436         block -= indirect_blks;
 437         if (block < indirect_blks) {
 438                 offset[n++] = NODE_IND2_BLOCK;
 439                 noffset[n] = 4 + dptrs_per_blk;
 440                 offset[n++] = block / direct_blks;
 441                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 442                 offset[n] = block % direct_blks;
 443                 level = 2;
 444                 goto got;
 445         }
 446         block -= indirect_blks;
 447         if (block < dindirect_blks) {
 448                 offset[n++] = NODE_DIND_BLOCK;
 449                 noffset[n] = 5 + (dptrs_per_blk * 2);
 450                 offset[n++] = block / indirect_blks;
 451                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 452                               offset[n - 1] * (dptrs_per_blk + 1);
 453                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 454                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 455                               offset[n - 2] * (dptrs_per_blk + 1) +
 456                               offset[n - 1];
 457                 offset[n] = block % direct_blks;
 458                 level = 3;
 459                 goto got;
 460         } else {
 461                 BUG();
 462         }
 463 got:
 464         return level;
 465 }
 466
 467 /*
 468  * Caller should call f2fs_put_dnode(dn).
 469  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 470  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 471  * In the case of RDONLY_NODE, we don't need to care about mutex.
 472  */
 473 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 474 {
 475         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 476         struct page *npage[4];
 477         struct page *parent;
 478         int offset[4];
 479         unsigned int noffset[4];
 480         nid_t nids[4];
 481         int level, i;
 482         int err = 0;
 483
 484         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 485
 486         nids[0] = dn->inode->i_ino;
 487         npage[0] = dn->inode_page;
 488
 489         if (!npage[0]) {
 490                 npage[0] = get_node_page(sbi, nids[0]);
 491                 if (IS_ERR(npage[0]))
 492                         return PTR_ERR(npage[0]);
 493         }
 494         parent = npage[0];
 495         if (level != 0)
 496                 nids[1] = get_nid(parent, offset[0], true);
 497         dn->inode_page = npage[0];
 498         dn->inode_page_locked = true;
 499
 500         /* get indirect or direct nodes */
 501         for (i = 1; i <= level; i++) {
 502                 bool done = false;
 503
 504                 if (!nids[i] && mode == ALLOC_NODE) {
 505                         /* alloc new node */
 506                         if (!alloc_nid(sbi, &(nids[i]))) {
 507                                 err = -ENOSPC;
 508                                 goto release_pages;
 509                         }
 510
 511                         dn->nid = nids[i];
 512                         npage[i] = new_node_page(dn, noffset[i], NULL);
 513                         if (IS_ERR(npage[i])) {
 514                                 alloc_nid_failed(sbi, nids[i]);
 515                                 err = PTR_ERR(npage[i]);
 516                                 goto release_pages;
 517                         }
 518
 519                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 520                         alloc_nid_done(sbi, nids[i]);
 521                         done = true;
 522                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 523                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 524                         if (IS_ERR(npage[i])) {
 525                                 err = PTR_ERR(npage[i]);
 526                                 goto release_pages;
 527                         }
 528                         done = true;
 529                 }
 530                 if (i == 1) {
 531                         dn->inode_page_locked = false;
 532                         unlock_page(parent);
 533                 } else {
 534                         f2fs_put_page(parent, 1);
 535                 }
 536
 537                 if (!done) {
 538                         npage[i] = get_node_page(sbi, nids[i]);
 539                         if (IS_ERR(npage[i])) {
 540                                 err = PTR_ERR(npage[i]);
 541                                 f2fs_put_page(npage[0], 0);
 542                                 goto release_out;
 543                         }
 544                 }
 545                 if (i < level) {
 546                         parent = npage[i];
 547                         nids[i + 1] = get_nid(parent, offset[i], false);
 548                 }
 549         }
 550         dn->nid = nids[level];
 551         dn->ofs_in_node = offset[level];
 552         dn->node_page = npage[level];
 553         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 554         return 0;
 555
 556 release_pages:
 557         f2fs_put_page(parent, 1);
 558         if (i > 1)
 559                 f2fs_put_page(npage[0], 0);
 560 release_out:
 561         dn->inode_page = NULL;
 562         dn->node_page = NULL;
 563         return err;
 564 }
 565
 566 static void truncate_node(struct dnode_of_data *dn)
 567 {
 568         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 569         struct node_info ni;
 570
 571         get_node_info(sbi, dn->nid, &ni);
 572         if (dn->inode->i_blocks == 0) {
 573                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 574                 goto invalidate;
 575         }
 576         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 577
 578         /* Deallocate node address */
 579         invalidate_blocks(sbi, ni.blk_addr);
 580         dec_valid_node_count(sbi, dn->inode);
 581         set_node_addr(sbi, &ni, NULL_ADDR, false);
 582
 583         if (dn->nid == dn->inode->i_ino) {
 584                 remove_orphan_inode(sbi, dn->nid);
 585                 dec_valid_inode_count(sbi);
 586         } else {
 587                 sync_inode_page(dn);
 588         }
 589 invalidate:
 590         clear_node_page_dirty(dn->node_page);
 591         set_sbi_flag(sbi, SBI_IS_DIRTY);
 592
 593         f2fs_put_page(dn->node_page, 1);
 594
 595         invalidate_mapping_pages(NODE_MAPPING(sbi),
 596                         dn->node_page->index, dn->node_page->index);
 597
 598         dn->node_page = NULL;
 599         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 600 }
 601
 602 static int truncate_dnode(struct dnode_of_data *dn)
 603 {
 604         struct page *page;
 605
 606         if (dn->nid == 0)
 607                 return 1;
 608
 609         /* get direct node */
 610         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 611         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 612                 return 1;
 613         else if (IS_ERR(page))
 614                 return PTR_ERR(page);
 615
 616         /* Make dnode_of_data for parameter */
 617         dn->node_page = page;
 618         dn->ofs_in_node = 0;
 619         truncate_data_blocks(dn);
 620         truncate_node(dn);
 621         return 1;
 622 }
 623
 624 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 625                                                 int ofs, int depth)
 626 {
 627         struct dnode_of_data rdn = *dn;
 628         struct page *page;
 629         struct f2fs_node *rn;
 630         nid_t child_nid;
 631         unsigned int child_nofs;
 632         int freed = 0;
 633         int i, ret;
 634
 635         if (dn->nid == 0)
 636                 return NIDS_PER_BLOCK + 1;
 637
 638         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 639
 640         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 641         if (IS_ERR(page)) {
 642                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 643                 return PTR_ERR(page);
 644         }
 645
 646         rn = F2FS_NODE(page);
 647         if (depth < 3) {
 648                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 649                         child_nid = le32_to_cpu(rn->in.nid[i]);
 650                         if (child_nid == 0)
 651                                 continue;
 652                         rdn.nid = child_nid;
 653                         ret = truncate_dnode(&rdn);
 654                         if (ret < 0)
 655                                 goto out_err;
 656                         set_nid(page, i, 0, false);
 657                 }
 658         } else {
 659                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 660                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 661                         child_nid = le32_to_cpu(rn->in.nid[i]);
 662                         if (child_nid == 0) {
 663                                 child_nofs += NIDS_PER_BLOCK + 1;
 664                                 continue;
 665                         }
 666                         rdn.nid = child_nid;
 667                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 668                         if (ret == (NIDS_PER_BLOCK + 1)) {
 669                                 set_nid(page, i, 0, false);
 670                                 child_nofs += ret;
 671                         } else if (ret < 0 && ret != -ENOENT) {
 672                                 goto out_err;
 673                         }
 674                 }
 675                 freed = child_nofs;
 676         }
 677
 678         if (!ofs) {
 679                 /* remove current indirect node */
 680                 dn->node_page = page;
 681                 truncate_node(dn);
 682                 freed++;
 683         } else {
 684                 f2fs_put_page(page, 1);
 685         }
 686         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 687         return freed;
 688
 689 out_err:
 690         f2fs_put_page(page, 1);
 691         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 692         return ret;
 693 }
 694
 695 static int truncate_partial_nodes(struct dnode_of_data *dn,
 696                         struct f2fs_inode *ri, int *offset, int depth)
 697 {
 698         struct page *pages[2];
 699         nid_t nid[3];
 700         nid_t child_nid;
 701         int err = 0;
 702         int i;
 703         int idx = depth - 2;
 704
 705         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 706         if (!nid[0])
 707                 return 0;
 708
 709         /* get indirect nodes in the path */
 710         for (i = 0; i < idx + 1; i++) {
 711                 /* reference count'll be increased */
 712                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 713                 if (IS_ERR(pages[i])) {
 714                         err = PTR_ERR(pages[i]);
 715                         idx = i - 1;
 716                         goto fail;
 717                 }
 718                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 719         }
 720
 721         /* free direct nodes linked to a partial indirect node */
 722         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 723                 child_nid = get_nid(pages[idx], i, false);
 724                 if (!child_nid)
 725                         continue;
 726                 dn->nid = child_nid;
 727                 err = truncate_dnode(dn);
 728                 if (err < 0)
 729                         goto fail;
 730                 set_nid(pages[idx], i, 0, false);
 731         }
 732
 733         if (offset[idx + 1] == 0) {
 734                 dn->node_page = pages[idx];
 735                 dn->nid = nid[idx];
 736                 truncate_node(dn);
 737         } else {
 738                 f2fs_put_page(pages[idx], 1);
 739         }
 740         offset[idx]++;
 741         offset[idx + 1] = 0;
 742         idx--;
 743 fail:
 744         for (i = idx; i >= 0; i--)
 745                 f2fs_put_page(pages[i], 1);
 746
 747         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 748
 749         return err;
 750 }
 751
 752 /*
 753  * All the block addresses of data and nodes should be nullified.
 754  */
 755 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 756 {
 757         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 758         int err = 0, cont = 1;
 759         int level, offset[4], noffset[4];
 760         unsigned int nofs = 0;
 761         struct f2fs_inode *ri;
 762         struct dnode_of_data dn;
 763         struct page *page;
 764
 765         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 766
 767         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 768 restart:
 769         page = get_node_page(sbi, inode->i_ino);
 770         if (IS_ERR(page)) {
 771                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 772                 return PTR_ERR(page);
 773         }
 774
 775         set_new_dnode(&dn, inode, page, NULL, 0);
 776         unlock_page(page);
 777
 778         ri = F2FS_INODE(page);
 779         switch (level) {
 780         case 0:
 781         case 1:
 782                 nofs = noffset[1];
 783                 break;
 784         case 2:
 785                 nofs = noffset[1];
 786                 if (!offset[level - 1])
 787                         goto skip_partial;
 788                 err = truncate_partial_nodes(&dn, ri, offset, level);
 789                 if (err < 0 && err != -ENOENT)
 790                         goto fail;
 791                 nofs += 1 + NIDS_PER_BLOCK;
 792                 break;
 793         case 3:
 794                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 795                 if (!offset[level - 1])
 796                         goto skip_partial;
 797                 err = truncate_partial_nodes(&dn, ri, offset, level);
 798                 if (err < 0 && err != -ENOENT)
 799                         goto fail;
 800                 break;
 801         default:
 802                 BUG();
 803         }
 804
 805 skip_partial:
 806         while (cont) {
 807                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 808                 switch (offset[0]) {
 809                 case NODE_DIR1_BLOCK:
 810                 case NODE_DIR2_BLOCK:
 811                         err = truncate_dnode(&dn);
 812                         break;
 813
 814                 case NODE_IND1_BLOCK:
 815                 case NODE_IND2_BLOCK:
 816                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 817                         break;
 818
 819                 case NODE_DIND_BLOCK:
 820                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 821                         cont = 0;
 822                         break;
 823
 824                 default:
 825                         BUG();
 826                 }
 827                 if (err < 0 && err != -ENOENT)
 828                         goto fail;
 829                 if (offset[1] == 0 &&
 830                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 831                         lock_page(page);
 832                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 833                                 f2fs_put_page(page, 1);
 834                                 goto restart;
 835                         }
 836                         f2fs_wait_on_page_writeback(page, NODE);
 837                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 838                         set_page_dirty(page);
 839                         unlock_page(page);
 840                 }
 841                 offset[1] = 0;
 842                 offset[0]++;
 843                 nofs += err;
 844         }
 845 fail:
 846         f2fs_put_page(page, 0);
 847         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 848         return err > 0 ? 0 : err;
 849 }
 850
 851 int truncate_xattr_node(struct inode *inode, struct page *page)
 852 {
 853         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 854         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 855         struct dnode_of_data dn;
 856         struct page *npage;
 857
 858         if (!nid)
 859                 return 0;
 860
 861         npage = get_node_page(sbi, nid);
 862         if (IS_ERR(npage))
 863                 return PTR_ERR(npage);
 864
 865         F2FS_I(inode)->i_xattr_nid = 0;
 866
 867         /* need to do checkpoint during fsync */
 868         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 869
 870         set_new_dnode(&dn, inode, page, npage, nid);
 871
 872         if (page)
 873                 dn.inode_page_locked = true;
 874         truncate_node(&dn);
 875         return 0;
 876 }
 877
 878 /*
 879  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 880  * f2fs_unlock_op().
 881  */
 882 void remove_inode_page(struct inode *inode)
 883 {
 884         struct dnode_of_data dn;
 885
 886         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 887         if (get_dnode_of_data(&dn, 0, LOOKUP_NODE))
 888                 return;
 889
 890         if (truncate_xattr_node(inode, dn.inode_page)) {
 891                 f2fs_put_dnode(&dn);
 892                 return;
 893         }
 894
 895         /* remove potential inline_data blocks */
 896         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 897                                 S_ISLNK(inode->i_mode))
 898                 truncate_data_blocks_range(&dn, 1);
 899
 900         /* 0 is possible, after f2fs_new_inode() has failed */
 901         f2fs_bug_on(F2FS_I_SB(inode),
 902                         inode->i_blocks != 0 && inode->i_blocks != 1);
 903
 904         /* will put inode & node pages */
 905         truncate_node(&dn);
 906 }
 907
 908 struct page *new_inode_page(struct inode *inode)
 909 {
 910         struct dnode_of_data dn;
 911
 912         /* allocate inode page for new inode */
 913         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 914
 915         /* caller should f2fs_put_page(page, 1); */
 916         return new_node_page(&dn, 0, NULL);
 917 }
 918
 919 struct page *new_node_page(struct dnode_of_data *dn,
 920                                 unsigned int ofs, struct page *ipage)
 921 {
 922         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 923         struct node_info old_ni, new_ni;
 924         struct page *page;
 925         int err;
 926
 927         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 928                 return ERR_PTR(-EPERM);
 929
 930         page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
 931         if (!page)
 932                 return ERR_PTR(-ENOMEM);
 933
 934         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 935                 err = -ENOSPC;
 936                 goto fail;
 937         }
 938
 939         get_node_info(sbi, dn->nid, &old_ni);
 940
 941         /* Reinitialize old_ni with new node page */
 942         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
 943         new_ni = old_ni;
 944         new_ni.ino = dn->inode->i_ino;
 945         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
 946
 947         f2fs_wait_on_page_writeback(page, NODE);
 948         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 949         set_cold_node(dn->inode, page);
 950         SetPageUptodate(page);
 951         set_page_dirty(page);
 952
 953         if (f2fs_has_xattr_block(ofs))
 954                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 955
 956         dn->node_page = page;
 957         if (ipage)
 958                 update_inode(dn->inode, ipage);
 959         else
 960                 sync_inode_page(dn);
 961         if (ofs == 0)
 962                 inc_valid_inode_count(sbi);
 963
 964         return page;
 965
 966 fail:
 967         clear_node_page_dirty(page);
 968         f2fs_put_page(page, 1);
 969         return ERR_PTR(err);
 970 }
 971
 972 /*
 973  * Caller should do after getting the following values.
 974  * 0: f2fs_put_page(page, 0)
 975  * LOCKED_PAGE: f2fs_put_page(page, 1)
 976  * error: nothing
 977  */
 978 static int read_node_page(struct page *page, int rw)
 979 {
 980         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 981         struct node_info ni;
 982         struct f2fs_io_info fio = {
 983                 .type = NODE,
 984                 .rw = rw,
 985         };
 986
 987         get_node_info(sbi, page->index, &ni);
 988
 989         if (unlikely(ni.blk_addr == NULL_ADDR)) {
 990                 f2fs_put_page(page, 1);
 991                 return -ENOENT;
 992         }
 993
 994         if (PageUptodate(page))
 995                 return LOCKED_PAGE;
 996
 997         fio.blk_addr = ni.blk_addr;
 998         return f2fs_submit_page_bio(sbi, page, &fio);
 999 }
1000
1001 /*
1002  * Readahead a node page
1003  */
1004 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1005 {
1006         struct page *apage;
1007         int err;
1008
1009         apage = find_get_page(NODE_MAPPING(sbi), nid);
1010         if (apage && PageUptodate(apage)) {
1011                 f2fs_put_page(apage, 0);
1012                 return;
1013         }
1014         f2fs_put_page(apage, 0);
1015
1016         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1017         if (!apage)
1018                 return;
1019
1020         err = read_node_page(apage, READA);
1021         if (err == 0)
1022                 f2fs_put_page(apage, 0);
1023         else if (err == LOCKED_PAGE)
1024                 f2fs_put_page(apage, 1);
1025 }
1026
1027 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1028 {
1029         struct page *page;
1030         int err;
1031 repeat:
1032         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1033         if (!page)
1034                 return ERR_PTR(-ENOMEM);
1035
1036         err = read_node_page(page, READ_SYNC);
1037         if (err < 0)
1038                 return ERR_PTR(err);
1039         else if (err == LOCKED_PAGE)
1040                 goto got_it;
1041
1042         lock_page(page);
1043         if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
1044                 f2fs_put_page(page, 1);
1045                 return ERR_PTR(-EIO);
1046         }
1047         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1048                 f2fs_put_page(page, 1);
1049                 goto repeat;
1050         }
1051 got_it:
1052         return page;
1053 }
1054
1055 /*
1056  * Return a locked page for the desired node page.
1057  * And, readahead MAX_RA_NODE number of node pages.
1058  */
1059 struct page *get_node_page_ra(struct page *parent, int start)
1060 {
1061         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1062         struct blk_plug plug;
1063         struct page *page;
1064         int err, i, end;
1065         nid_t nid;
1066
1067         /* First, try getting the desired direct node. */
1068         nid = get_nid(parent, start, false);
1069         if (!nid)
1070                 return ERR_PTR(-ENOENT);
1071 repeat:
1072         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1073         if (!page)
1074                 return ERR_PTR(-ENOMEM);
1075
1076         err = read_node_page(page, READ_SYNC);
1077         if (err < 0)
1078                 return ERR_PTR(err);
1079         else if (err == LOCKED_PAGE)
1080                 goto page_hit;
1081
1082         blk_start_plug(&plug);
1083
1084         /* Then, try readahead for siblings of the desired node */
1085         end = start + MAX_RA_NODE;
1086         end = min(end, NIDS_PER_BLOCK);
1087         for (i = start + 1; i < end; i++) {
1088                 nid = get_nid(parent, i, false);
1089                 if (!nid)
1090                         continue;
1091                 ra_node_page(sbi, nid);
1092         }
1093
1094         blk_finish_plug(&plug);
1095
1096         lock_page(page);
1097         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1098                 f2fs_put_page(page, 1);
1099                 goto repeat;
1100         }
1101 page_hit:
1102         if (unlikely(!PageUptodate(page))) {
1103                 f2fs_put_page(page, 1);
1104                 return ERR_PTR(-EIO);
1105         }
1106         return page;
1107 }
1108
1109 void sync_inode_page(struct dnode_of_data *dn)
1110 {
1111         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1112                 update_inode(dn->inode, dn->node_page);
1113         } else if (dn->inode_page) {
1114                 if (!dn->inode_page_locked)
1115                         lock_page(dn->inode_page);
1116                 update_inode(dn->inode, dn->inode_page);
1117                 if (!dn->inode_page_locked)
1118                         unlock_page(dn->inode_page);
1119         } else {
1120                 update_inode_page(dn->inode);
1121         }
1122 }
1123
1124 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1125                                         struct writeback_control *wbc)
1126 {
1127         pgoff_t index, end;
1128         struct pagevec pvec;
1129         int step = ino ? 2 : 0;
1130         int nwritten = 0, wrote = 0;
1131
1132         pagevec_init(&pvec, 0);
1133
1134 next_step:
1135         index = 0;
1136         end = LONG_MAX;
1137
1138         while (index <= end) {
1139                 int i, nr_pages;
1140                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1141                                 PAGECACHE_TAG_DIRTY,
1142                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1143                 if (nr_pages == 0)
1144                         break;
1145
1146                 for (i = 0; i < nr_pages; i++) {
1147                         struct page *page = pvec.pages[i];
1148
1149                         /*
1150                          * flushing sequence with step:
1151                          * 0. indirect nodes
1152                          * 1. dentry dnodes
1153                          * 2. file dnodes
1154                          */
1155                         if (step == 0 && IS_DNODE(page))
1156                                 continue;
1157                         if (step == 1 && (!IS_DNODE(page) ||
1158                                                 is_cold_node(page)))
1159                                 continue;
1160                         if (step == 2 && (!IS_DNODE(page) ||
1161                                                 !is_cold_node(page)))
1162                                 continue;
1163
1164                         /*
1165                          * If an fsync mode,
1166                          * we should not skip writing node pages.
1167                          */
1168                         if (ino && ino_of_node(page) == ino)
1169                                 lock_page(page);
1170                         else if (!trylock_page(page))
1171                                 continue;
1172
1173                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1174 continue_unlock:
1175                                 unlock_page(page);
1176                                 continue;
1177                         }
1178                         if (ino && ino_of_node(page) != ino)
1179                                 goto continue_unlock;
1180
1181                         if (!PageDirty(page)) {
1182                                 /* someone wrote it for us */
1183                                 goto continue_unlock;
1184                         }
1185
1186                         if (!clear_page_dirty_for_io(page))
1187                                 goto continue_unlock;
1188
1189                         /* called by fsync() */
1190                         if (ino && IS_DNODE(page)) {
1191                                 set_fsync_mark(page, 1);
1192                                 if (IS_INODE(page)) {
1193                                         if (!is_checkpointed_node(sbi, ino) &&
1194                                                 !has_fsynced_inode(sbi, ino))
1195                                                 set_dentry_mark(page, 1);
1196                                         else
1197                                                 set_dentry_mark(page, 0);
1198                                 }
1199                                 nwritten++;
1200                         } else {
1201                                 set_fsync_mark(page, 0);
1202                                 set_dentry_mark(page, 0);
1203                         }
1204
1205                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1206                                 unlock_page(page);
1207                         else
1208                                 wrote++;
1209
1210                         if (--wbc->nr_to_write == 0)
1211                                 break;
1212                 }
1213                 pagevec_release(&pvec);
1214                 cond_resched();
1215
1216                 if (wbc->nr_to_write == 0) {
1217                         step = 2;
1218                         break;
1219                 }
1220         }
1221
1222         if (step < 2) {
1223                 step++;
1224                 goto next_step;
1225         }
1226
1227         if (wrote)
1228                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1229         return nwritten;
1230 }
1231
1232 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1233 {
1234         pgoff_t index = 0, end = LONG_MAX;
1235         struct pagevec pvec;
1236         int ret2 = 0, ret = 0;
1237
1238         pagevec_init(&pvec, 0);
1239
1240         while (index <= end) {
1241                 int i, nr_pages;
1242                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1243                                 PAGECACHE_TAG_WRITEBACK,
1244                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1245                 if (nr_pages == 0)
1246                         break;
1247
1248                 for (i = 0; i < nr_pages; i++) {
1249                         struct page *page = pvec.pages[i];
1250
1251                         /* until radix tree lookup accepts end_index */
1252                         if (unlikely(page->index > end))
1253                                 continue;
1254
1255                         if (ino && ino_of_node(page) == ino) {
1256                                 f2fs_wait_on_page_writeback(page, NODE);
1257                                 if (TestClearPageError(page))
1258                                         ret = -EIO;
1259                         }
1260                 }
1261                 pagevec_release(&pvec);
1262                 cond_resched();
1263         }
1264
1265         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1266                 ret2 = -ENOSPC;
1267         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1268                 ret2 = -EIO;
1269         if (!ret)
1270                 ret = ret2;
1271         return ret;
1272 }
1273
1274 static int f2fs_write_node_page(struct page *page,
1275                                 struct writeback_control *wbc)
1276 {
1277         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1278         nid_t nid;
1279         struct node_info ni;
1280         struct f2fs_io_info fio = {
1281                 .type = NODE,
1282                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1283         };
1284
1285         trace_f2fs_writepage(page, NODE);
1286
1287         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1288                 goto redirty_out;
1289         if (unlikely(f2fs_cp_error(sbi)))
1290                 goto redirty_out;
1291
1292         f2fs_wait_on_page_writeback(page, NODE);
1293
1294         /* get old block addr of this node page */
1295         nid = nid_of_node(page);
1296         f2fs_bug_on(sbi, page->index != nid);
1297
1298         get_node_info(sbi, nid, &ni);
1299
1300         /* This page is already truncated */
1301         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1302                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1303                 unlock_page(page);
1304                 return 0;
1305         }
1306
1307         if (wbc->for_reclaim) {
1308                 if (!down_read_trylock(&sbi->node_write))
1309                         goto redirty_out;
1310         } else {
1311                 down_read(&sbi->node_write);
1312         }
1313
1314         set_page_writeback(page);
1315         fio.blk_addr = ni.blk_addr;
1316         write_node_page(sbi, page, nid, &fio);
1317         set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1318         dec_page_count(sbi, F2FS_DIRTY_NODES);
1319         up_read(&sbi->node_write);
1320         unlock_page(page);
1321
1322         if (wbc->for_reclaim)
1323                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1324
1325         return 0;
1326
1327 redirty_out:
1328         redirty_page_for_writepage(wbc, page);
1329         return AOP_WRITEPAGE_ACTIVATE;
1330 }
1331
1332 static int f2fs_write_node_pages(struct address_space *mapping,
1333                             struct writeback_control *wbc)
1334 {
1335         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1336         long diff;
1337
1338         trace_f2fs_writepages(mapping->host, wbc, NODE);
1339
1340         /* balancing f2fs's metadata in background */
1341         f2fs_balance_fs_bg(sbi);
1342
1343         /* collect a number of dirty node pages and write together */
1344         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1345                 goto skip_write;
1346
1347         diff = nr_pages_to_write(sbi, NODE, wbc);
1348         wbc->sync_mode = WB_SYNC_NONE;
1349         sync_node_pages(sbi, 0, wbc);
1350         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1351         return 0;
1352
1353 skip_write:
1354         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1355         return 0;
1356 }
1357
1358 static int f2fs_set_node_page_dirty(struct page *page)
1359 {
1360         trace_f2fs_set_page_dirty(page, NODE);
1361
1362         SetPageUptodate(page);
1363         if (!PageDirty(page)) {
1364                 __set_page_dirty_nobuffers(page);
1365                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1366                 SetPagePrivate(page);
1367                 f2fs_trace_pid(page);
1368                 return 1;
1369         }
1370         return 0;
1371 }
1372
1373 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1374                                       unsigned int length)
1375 {
1376         struct inode *inode = page->mapping->host;
1377         if (PageDirty(page))
1378                 dec_page_count(F2FS_I_SB(inode), F2FS_DIRTY_NODES);
1379         ClearPagePrivate(page);
1380 }
1381
1382 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1383 {
1384         /* If this is dirty page, keep PagePrivate */
1385         if (PageDirty(page))
1386                 return 0;
1387
1388         ClearPagePrivate(page);
1389         return 1;
1390 }
1391
1392 /*
1393  * Structure of the f2fs node operations
1394  */
1395 const struct address_space_operations f2fs_node_aops = {
1396         .writepage      = f2fs_write_node_page,
1397         .writepages     = f2fs_write_node_pages,
1398         .set_page_dirty = f2fs_set_node_page_dirty,
1399         .invalidatepage = f2fs_invalidate_node_page,
1400         .releasepage    = f2fs_release_node_page,
1401 };
1402
1403 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1404                                                 nid_t n)
1405 {
1406         return radix_tree_lookup(&nm_i->free_nid_root, n);
1407 }
1408
1409 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1410                                                 struct free_nid *i)
1411 {
1412         list_del(&i->list);
1413         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1414 }
1415
1416 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1417 {
1418         struct f2fs_nm_info *nm_i = NM_I(sbi);
1419         struct free_nid *i;
1420         struct nat_entry *ne;
1421         bool allocated = false;
1422
1423         if (!available_free_memory(sbi, FREE_NIDS))
1424                 return -1;
1425
1426         /* 0 nid should not be used */
1427         if (unlikely(nid == 0))
1428                 return 0;
1429
1430         if (build) {
1431                 /* do not add allocated nids */
1432                 down_read(&nm_i->nat_tree_lock);
1433                 ne = __lookup_nat_cache(nm_i, nid);
1434                 if (ne &&
1435                         (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1436                                 nat_get_blkaddr(ne) != NULL_ADDR))
1437                         allocated = true;
1438                 up_read(&nm_i->nat_tree_lock);
1439                 if (allocated)
1440                         return 0;
1441         }
1442
1443         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1444         i->nid = nid;
1445         i->state = NID_NEW;
1446
1447         if (radix_tree_preload(GFP_NOFS)) {
1448                 kmem_cache_free(free_nid_slab, i);
1449                 return 0;
1450         }
1451
1452         spin_lock(&nm_i->free_nid_list_lock);
1453         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1454                 spin_unlock(&nm_i->free_nid_list_lock);
1455                 radix_tree_preload_end();
1456                 kmem_cache_free(free_nid_slab, i);
1457                 return 0;
1458         }
1459         list_add_tail(&i->list, &nm_i->free_nid_list);
1460         nm_i->fcnt++;
1461         spin_unlock(&nm_i->free_nid_list_lock);
1462         radix_tree_preload_end();
1463         return 1;
1464 }
1465
1466 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1467 {
1468         struct free_nid *i;
1469         bool need_free = false;
1470
1471         spin_lock(&nm_i->free_nid_list_lock);
1472         i = __lookup_free_nid_list(nm_i, nid);
1473         if (i && i->state == NID_NEW) {
1474                 __del_from_free_nid_list(nm_i, i);
1475                 nm_i->fcnt--;
1476                 need_free = true;
1477         }
1478         spin_unlock(&nm_i->free_nid_list_lock);
1479
1480         if (need_free)
1481                 kmem_cache_free(free_nid_slab, i);
1482 }
1483
1484 static void scan_nat_page(struct f2fs_sb_info *sbi,
1485                         struct page *nat_page, nid_t start_nid)
1486 {
1487         struct f2fs_nm_info *nm_i = NM_I(sbi);
1488         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1489         block_t blk_addr;
1490         int i;
1491
1492         i = start_nid % NAT_ENTRY_PER_BLOCK;
1493
1494         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1495
1496                 if (unlikely(start_nid >= nm_i->max_nid))
1497                         break;
1498
1499                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1500                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1501                 if (blk_addr == NULL_ADDR) {
1502                         if (add_free_nid(sbi, start_nid, true) < 0)
1503                                 break;
1504                 }
1505         }
1506 }
1507
1508 static void build_free_nids(struct f2fs_sb_info *sbi)
1509 {
1510         struct f2fs_nm_info *nm_i = NM_I(sbi);
1511         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1512         struct f2fs_summary_block *sum = curseg->sum_blk;
1513         int i = 0;
1514         nid_t nid = nm_i->next_scan_nid;
1515
1516         /* Enough entries */
1517         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1518                 return;
1519
1520         /* readahead nat pages to be scanned */
1521         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1522
1523         while (1) {
1524                 struct page *page = get_current_nat_page(sbi, nid);
1525
1526                 scan_nat_page(sbi, page, nid);
1527                 f2fs_put_page(page, 1);
1528
1529                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1530                 if (unlikely(nid >= nm_i->max_nid))
1531                         nid = 0;
1532
1533                 if (i++ == FREE_NID_PAGES)
1534                         break;
1535         }
1536
1537         /* go to the next free nat pages to find free nids abundantly */
1538         nm_i->next_scan_nid = nid;
1539
1540         /* find free nids from current sum_pages */
1541         mutex_lock(&curseg->curseg_mutex);
1542         for (i = 0; i < nats_in_cursum(sum); i++) {
1543                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1544                 nid = le32_to_cpu(nid_in_journal(sum, i));
1545                 if (addr == NULL_ADDR)
1546                         add_free_nid(sbi, nid, true);
1547                 else
1548                         remove_free_nid(nm_i, nid);
1549         }
1550         mutex_unlock(&curseg->curseg_mutex);
1551 }
1552
1553 /*
1554  * If this function returns success, caller can obtain a new nid
1555  * from second parameter of this function.
1556  * The returned nid could be used ino as well as nid when inode is created.
1557  */
1558 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1559 {
1560         struct f2fs_nm_info *nm_i = NM_I(sbi);
1561         struct free_nid *i = NULL;
1562 retry:
1563         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1564                 return false;
1565
1566         spin_lock(&nm_i->free_nid_list_lock);
1567
1568         /* We should not use stale free nids created by build_free_nids */
1569         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1570                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1571                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1572                         if (i->state == NID_NEW)
1573                                 break;
1574
1575                 f2fs_bug_on(sbi, i->state != NID_NEW);
1576                 *nid = i->nid;
1577                 i->state = NID_ALLOC;
1578                 nm_i->fcnt--;
1579                 spin_unlock(&nm_i->free_nid_list_lock);
1580                 return true;
1581         }
1582         spin_unlock(&nm_i->free_nid_list_lock);
1583
1584         /* Let's scan nat pages and its caches to get free nids */
1585         mutex_lock(&nm_i->build_lock);
1586         build_free_nids(sbi);
1587         mutex_unlock(&nm_i->build_lock);
1588         goto retry;
1589 }
1590
1591 /*
1592  * alloc_nid() should be called prior to this function.
1593  */
1594 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1595 {
1596         struct f2fs_nm_info *nm_i = NM_I(sbi);
1597         struct free_nid *i;
1598
1599         spin_lock(&nm_i->free_nid_list_lock);
1600         i = __lookup_free_nid_list(nm_i, nid);
1601         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1602         __del_from_free_nid_list(nm_i, i);
1603         spin_unlock(&nm_i->free_nid_list_lock);
1604
1605         kmem_cache_free(free_nid_slab, i);
1606 }
1607
1608 /*
1609  * alloc_nid() should be called prior to this function.
1610  */
1611 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1612 {
1613         struct f2fs_nm_info *nm_i = NM_I(sbi);
1614         struct free_nid *i;
1615         bool need_free = false;
1616
1617         if (!nid)
1618                 return;
1619
1620         spin_lock(&nm_i->free_nid_list_lock);
1621         i = __lookup_free_nid_list(nm_i, nid);
1622         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1623         if (!available_free_memory(sbi, FREE_NIDS)) {
1624                 __del_from_free_nid_list(nm_i, i);
1625                 need_free = true;
1626         } else {
1627                 i->state = NID_NEW;
1628                 nm_i->fcnt++;
1629         }
1630         spin_unlock(&nm_i->free_nid_list_lock);
1631
1632         if (need_free)
1633                 kmem_cache_free(free_nid_slab, i);
1634 }
1635
1636 void recover_inline_xattr(struct inode *inode, struct page *page)
1637 {
1638         void *src_addr, *dst_addr;
1639         size_t inline_size;
1640         struct page *ipage;
1641         struct f2fs_inode *ri;
1642
1643         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1644         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1645
1646         ri = F2FS_INODE(page);
1647         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1648                 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1649                 goto update_inode;
1650         }
1651
1652         dst_addr = inline_xattr_addr(ipage);
1653         src_addr = inline_xattr_addr(page);
1654         inline_size = inline_xattr_size(inode);
1655
1656         f2fs_wait_on_page_writeback(ipage, NODE);
1657         memcpy(dst_addr, src_addr, inline_size);
1658 update_inode:
1659         update_inode(inode, ipage);
1660         f2fs_put_page(ipage, 1);
1661 }
1662
1663 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1664 {
1665         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1666         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1667         nid_t new_xnid = nid_of_node(page);
1668         struct node_info ni;
1669
1670         /* 1: invalidate the previous xattr nid */
1671         if (!prev_xnid)
1672                 goto recover_xnid;
1673
1674         /* Deallocate node address */
1675         get_node_info(sbi, prev_xnid, &ni);
1676         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1677         invalidate_blocks(sbi, ni.blk_addr);
1678         dec_valid_node_count(sbi, inode);
1679         set_node_addr(sbi, &ni, NULL_ADDR, false);
1680
1681 recover_xnid:
1682         /* 2: allocate new xattr nid */
1683         if (unlikely(!inc_valid_node_count(sbi, inode)))
1684                 f2fs_bug_on(sbi, 1);
1685
1686         remove_free_nid(NM_I(sbi), new_xnid);
1687         get_node_info(sbi, new_xnid, &ni);
1688         ni.ino = inode->i_ino;
1689         set_node_addr(sbi, &ni, NEW_ADDR, false);
1690         F2FS_I(inode)->i_xattr_nid = new_xnid;
1691
1692         /* 3: update xattr blkaddr */
1693         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1694         set_node_addr(sbi, &ni, blkaddr, false);
1695
1696         update_inode_page(inode);
1697 }
1698
1699 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1700 {
1701         struct f2fs_inode *src, *dst;
1702         nid_t ino = ino_of_node(page);
1703         struct node_info old_ni, new_ni;
1704         struct page *ipage;
1705
1706         get_node_info(sbi, ino, &old_ni);
1707
1708         if (unlikely(old_ni.blk_addr != NULL_ADDR))
1709                 return -EINVAL;
1710
1711         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1712         if (!ipage)
1713                 return -ENOMEM;
1714
1715         /* Should not use this inode from free nid list */
1716         remove_free_nid(NM_I(sbi), ino);
1717
1718         SetPageUptodate(ipage);
1719         fill_node_footer(ipage, ino, ino, 0, true);
1720
1721         src = F2FS_INODE(page);
1722         dst = F2FS_INODE(ipage);
1723
1724         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1725         dst->i_size = 0;
1726         dst->i_blocks = cpu_to_le64(1);
1727         dst->i_links = cpu_to_le32(1);
1728         dst->i_xattr_nid = 0;
1729         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1730
1731         new_ni = old_ni;
1732         new_ni.ino = ino;
1733
1734         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1735                 WARN_ON(1);
1736         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1737         inc_valid_inode_count(sbi);
1738         set_page_dirty(ipage);
1739         f2fs_put_page(ipage, 1);
1740         return 0;
1741 }
1742
1743 int restore_node_summary(struct f2fs_sb_info *sbi,
1744                         unsigned int segno, struct f2fs_summary_block *sum)
1745 {
1746         struct f2fs_node *rn;
1747         struct f2fs_summary *sum_entry;
1748         block_t addr;
1749         int bio_blocks = MAX_BIO_BLOCKS(sbi);
1750         int i, idx, last_offset, nrpages;
1751
1752         /* scan the node segment */
1753         last_offset = sbi->blocks_per_seg;
1754         addr = START_BLOCK(sbi, segno);
1755         sum_entry = &sum->entries[0];
1756
1757         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1758                 nrpages = min(last_offset - i, bio_blocks);
1759
1760                 /* readahead node pages */
1761                 ra_meta_pages(sbi, addr, nrpages, META_POR);
1762
1763                 for (idx = addr; idx < addr + nrpages; idx++) {
1764                         struct page *page = get_meta_page(sbi, idx);
1765
1766                         rn = F2FS_NODE(page);
1767                         sum_entry->nid = rn->footer.nid;
1768                         sum_entry->version = 0;
1769                         sum_entry->ofs_in_node = 0;
1770                         sum_entry++;
1771                         f2fs_put_page(page, 1);
1772                 }
1773
1774                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
1775                                                         addr + nrpages);
1776         }
1777         return 0;
1778 }
1779
1780 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1781 {
1782         struct f2fs_nm_info *nm_i = NM_I(sbi);
1783         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1784         struct f2fs_summary_block *sum = curseg->sum_blk;
1785         int i;
1786
1787         mutex_lock(&curseg->curseg_mutex);
1788         for (i = 0; i < nats_in_cursum(sum); i++) {
1789                 struct nat_entry *ne;
1790                 struct f2fs_nat_entry raw_ne;
1791                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1792
1793                 raw_ne = nat_in_journal(sum, i);
1794
1795                 down_write(&nm_i->nat_tree_lock);
1796                 ne = __lookup_nat_cache(nm_i, nid);
1797                 if (!ne) {
1798                         ne = grab_nat_entry(nm_i, nid);
1799                         node_info_from_raw_nat(&ne->ni, &raw_ne);
1800                 }
1801                 __set_nat_cache_dirty(nm_i, ne);
1802                 up_write(&nm_i->nat_tree_lock);
1803         }
1804         update_nats_in_cursum(sum, -i);
1805         mutex_unlock(&curseg->curseg_mutex);
1806 }
1807
1808 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1809                                                 struct list_head *head, int max)
1810 {
1811         struct nat_entry_set *cur;
1812
1813         if (nes->entry_cnt >= max)
1814                 goto add_out;
1815
1816         list_for_each_entry(cur, head, set_list) {
1817                 if (cur->entry_cnt >= nes->entry_cnt) {
1818                         list_add(&nes->set_list, cur->set_list.prev);
1819                         return;
1820                 }
1821         }
1822 add_out:
1823         list_add_tail(&nes->set_list, head);
1824 }
1825
1826 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1827                                         struct nat_entry_set *set)
1828 {
1829         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1830         struct f2fs_summary_block *sum = curseg->sum_blk;
1831         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1832         bool to_journal = true;
1833         struct f2fs_nat_block *nat_blk;
1834         struct nat_entry *ne, *cur;
1835         struct page *page = NULL;
1836
1837         /*
1838          * there are two steps to flush nat entries:
1839          * #1, flush nat entries to journal in current hot data summary block.
1840          * #2, flush nat entries to nat page.
1841          */
1842         if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1843                 to_journal = false;
1844
1845         if (to_journal) {
1846                 mutex_lock(&curseg->curseg_mutex);
1847         } else {
1848                 page = get_next_nat_page(sbi, start_nid);
1849                 nat_blk = page_address(page);
1850                 f2fs_bug_on(sbi, !nat_blk);
1851         }
1852
1853         /* flush dirty nats in nat entry set */
1854         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1855                 struct f2fs_nat_entry *raw_ne;
1856                 nid_t nid = nat_get_nid(ne);
1857                 int offset;
1858
1859                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1860                         continue;
1861
1862                 if (to_journal) {
1863                         offset = lookup_journal_in_cursum(sum,
1864                                                         NAT_JOURNAL, nid, 1);
1865                         f2fs_bug_on(sbi, offset < 0);
1866                         raw_ne = &nat_in_journal(sum, offset);
1867                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1868                 } else {
1869                         raw_ne = &nat_blk->entries[nid - start_nid];
1870                 }
1871                 raw_nat_from_node_info(raw_ne, &ne->ni);
1872
1873                 down_write(&NM_I(sbi)->nat_tree_lock);
1874                 nat_reset_flag(ne);
1875                 __clear_nat_cache_dirty(NM_I(sbi), ne);
1876                 up_write(&NM_I(sbi)->nat_tree_lock);
1877
1878                 if (nat_get_blkaddr(ne) == NULL_ADDR)
1879                         add_free_nid(sbi, nid, false);
1880         }
1881
1882         if (to_journal)
1883                 mutex_unlock(&curseg->curseg_mutex);
1884         else
1885                 f2fs_put_page(page, 1);
1886
1887         f2fs_bug_on(sbi, set->entry_cnt);
1888
1889         radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1890         kmem_cache_free(nat_entry_set_slab, set);
1891 }
1892
1893 /*
1894  * This function is called during the checkpointing process.
1895  */
1896 void flush_nat_entries(struct f2fs_sb_info *sbi)
1897 {
1898         struct f2fs_nm_info *nm_i = NM_I(sbi);
1899         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1900         struct f2fs_summary_block *sum = curseg->sum_blk;
1901         struct nat_entry_set *setvec[SETVEC_SIZE];
1902         struct nat_entry_set *set, *tmp;
1903         unsigned int found;
1904         nid_t set_idx = 0;
1905         LIST_HEAD(sets);
1906
1907         if (!nm_i->dirty_nat_cnt)
1908                 return;
1909         /*
1910          * if there are no enough space in journal to store dirty nat
1911          * entries, remove all entries from journal and merge them
1912          * into nat entry set.
1913          */
1914         if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1915                 remove_nats_in_journal(sbi);
1916
1917         while ((found = __gang_lookup_nat_set(nm_i,
1918                                         set_idx, SETVEC_SIZE, setvec))) {
1919                 unsigned idx;
1920                 set_idx = setvec[found - 1]->set + 1;
1921                 for (idx = 0; idx < found; idx++)
1922                         __adjust_nat_entry_set(setvec[idx], &sets,
1923                                                         MAX_NAT_JENTRIES(sum));
1924         }
1925
1926         /* flush dirty nats in nat entry set */
1927         list_for_each_entry_safe(set, tmp, &sets, set_list)
1928                 __flush_nat_entry_set(sbi, set);
1929
1930         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1931 }
1932
1933 static int init_node_manager(struct f2fs_sb_info *sbi)
1934 {
1935         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1936         struct f2fs_nm_info *nm_i = NM_I(sbi);
1937         unsigned char *version_bitmap;
1938         unsigned int nat_segs, nat_blocks;
1939
1940         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1941
1942         /* segment_count_nat includes pair segment so divide to 2. */
1943         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1944         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1945
1946         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1947
1948         /* not used nids: 0, node, meta, (and root counted as valid node) */
1949         nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
1950         nm_i->fcnt = 0;
1951         nm_i->nat_cnt = 0;
1952         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1953
1954         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1955         INIT_LIST_HEAD(&nm_i->free_nid_list);
1956         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
1957         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
1958         INIT_LIST_HEAD(&nm_i->nat_entries);
1959
1960         mutex_init(&nm_i->build_lock);
1961         spin_lock_init(&nm_i->free_nid_list_lock);
1962         init_rwsem(&nm_i->nat_tree_lock);
1963
1964         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1965         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1966         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1967         if (!version_bitmap)
1968                 return -EFAULT;
1969
1970         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1971                                         GFP_KERNEL);
1972         if (!nm_i->nat_bitmap)
1973                 return -ENOMEM;
1974         return 0;
1975 }
1976
1977 int build_node_manager(struct f2fs_sb_info *sbi)
1978 {
1979         int err;
1980
1981         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1982         if (!sbi->nm_info)
1983                 return -ENOMEM;
1984
1985         err = init_node_manager(sbi);
1986         if (err)
1987                 return err;
1988
1989         build_free_nids(sbi);
1990         return 0;
1991 }
1992
1993 void destroy_node_manager(struct f2fs_sb_info *sbi)
1994 {
1995         struct f2fs_nm_info *nm_i = NM_I(sbi);
1996         struct free_nid *i, *next_i;
1997         struct nat_entry *natvec[NATVEC_SIZE];
1998         struct nat_entry_set *setvec[SETVEC_SIZE];
1999         nid_t nid = 0;
2000         unsigned int found;
2001
2002         if (!nm_i)
2003                 return;
2004
2005         /* destroy free nid list */
2006         spin_lock(&nm_i->free_nid_list_lock);
2007         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2008                 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2009                 __del_from_free_nid_list(nm_i, i);
2010                 nm_i->fcnt--;
2011                 spin_unlock(&nm_i->free_nid_list_lock);
2012                 kmem_cache_free(free_nid_slab, i);
2013                 spin_lock(&nm_i->free_nid_list_lock);
2014         }
2015         f2fs_bug_on(sbi, nm_i->fcnt);
2016         spin_unlock(&nm_i->free_nid_list_lock);
2017
2018         /* destroy nat cache */
2019         down_write(&nm_i->nat_tree_lock);
2020         while ((found = __gang_lookup_nat_cache(nm_i,
2021                                         nid, NATVEC_SIZE, natvec))) {
2022                 unsigned idx;
2023
2024                 nid = nat_get_nid(natvec[found - 1]) + 1;
2025                 for (idx = 0; idx < found; idx++)
2026                         __del_from_nat_cache(nm_i, natvec[idx]);
2027         }
2028         f2fs_bug_on(sbi, nm_i->nat_cnt);
2029
2030         /* destroy nat set cache */
2031         nid = 0;
2032         while ((found = __gang_lookup_nat_set(nm_i,
2033                                         nid, SETVEC_SIZE, setvec))) {
2034                 unsigned idx;
2035
2036                 nid = setvec[found - 1]->set + 1;
2037                 for (idx = 0; idx < found; idx++) {
2038                         /* entry_cnt is not zero, when cp_error was occurred */
2039                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2040                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2041                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2042                 }
2043         }
2044         up_write(&nm_i->nat_tree_lock);
2045
2046         kfree(nm_i->nat_bitmap);
2047         sbi->nm_info = NULL;
2048         kfree(nm_i);
2049 }
2050
2051 int __init create_node_manager_caches(void)
2052 {
2053         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2054                         sizeof(struct nat_entry));
2055         if (!nat_entry_slab)
2056                 goto fail;
2057
2058         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2059                         sizeof(struct free_nid));
2060         if (!free_nid_slab)
2061                 goto destroy_nat_entry;
2062
2063         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2064                         sizeof(struct nat_entry_set));
2065         if (!nat_entry_set_slab)
2066                 goto destroy_free_nid;
2067         return 0;
2068
2069 destroy_free_nid:
2070         kmem_cache_destroy(free_nid_slab);
2071 destroy_nat_entry:
2072         kmem_cache_destroy(nat_entry_slab);
2073 fail:
2074         return -ENOMEM;
2075 }
2076
2077 void destroy_node_manager_caches(void)
2078 {
2079         kmem_cache_destroy(nat_entry_set_slab);
2080         kmem_cache_destroy(free_nid_slab);
2081         kmem_cache_destroy(nat_entry_slab);
2082 }