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/events/f2fs.h>
  23
  24 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
  25
  26 static struct kmem_cache *nat_entry_slab;
  27 static struct kmem_cache *free_nid_slab;
  28 static struct kmem_cache *nat_entry_set_slab;
  29
  30 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
  31 {
  32         struct f2fs_nm_info *nm_i = NM_I(sbi);
  33         struct sysinfo val;
  34         unsigned long mem_size = 0;
  35         bool res = false;
  36
  37         si_meminfo(&val);
  38         /* give 25%, 25%, 50% memory for each components respectively */
  39         if (type == FREE_NIDS) {
  40                 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >> 12;
  41                 res = mem_size < ((val.totalram * nm_i->ram_thresh / 100) >> 2);
  42         } else if (type == NAT_ENTRIES) {
  43                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >> 12;
  44                 res = mem_size < ((val.totalram * nm_i->ram_thresh / 100) >> 2);
  45         } else if (type == DIRTY_DENTS) {
  46                 if (sbi->sb->s_bdi->dirty_exceeded)
  47                         return false;
  48                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  49                 res = mem_size < ((val.totalram * nm_i->ram_thresh / 100) >> 1);
  50         }
  51         return res;
  52 }
  53
  54 static void clear_node_page_dirty(struct page *page)
  55 {
  56         struct address_space *mapping = page->mapping;
  57         unsigned int long flags;
  58
  59         if (PageDirty(page)) {
  60                 spin_lock_irqsave(&mapping->tree_lock, flags);
  61                 radix_tree_tag_clear(&mapping->page_tree,
  62                                 page_index(page),
  63                                 PAGECACHE_TAG_DIRTY);
  64                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  65
  66                 clear_page_dirty_for_io(page);
  67                 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
  68         }
  69         ClearPageUptodate(page);
  70 }
  71
  72 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  73 {
  74         pgoff_t index = current_nat_addr(sbi, nid);
  75         return get_meta_page(sbi, index);
  76 }
  77
  78 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  79 {
  80         struct page *src_page;
  81         struct page *dst_page;
  82         pgoff_t src_off;
  83         pgoff_t dst_off;
  84         void *src_addr;
  85         void *dst_addr;
  86         struct f2fs_nm_info *nm_i = NM_I(sbi);
  87
  88         src_off = current_nat_addr(sbi, nid);
  89         dst_off = next_nat_addr(sbi, src_off);
  90
  91         /* get current nat block page with lock */
  92         src_page = get_meta_page(sbi, src_off);
  93         dst_page = grab_meta_page(sbi, dst_off);
  94         f2fs_bug_on(sbi, PageDirty(src_page));
  95
  96         src_addr = page_address(src_page);
  97         dst_addr = page_address(dst_page);
  98         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
  99         set_page_dirty(dst_page);
 100         f2fs_put_page(src_page, 1);
 101
 102         set_to_next_nat(nm_i, nid);
 103
 104         return dst_page;
 105 }
 106
 107 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 108 {
 109         return radix_tree_lookup(&nm_i->nat_root, n);
 110 }
 111
 112 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 113                 nid_t start, unsigned int nr, struct nat_entry **ep)
 114 {
 115         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 116 }
 117
 118 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 119 {
 120         list_del(&e->list);
 121         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 122         nm_i->nat_cnt--;
 123         kmem_cache_free(nat_entry_slab, e);
 124 }
 125
 126 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 127                                                 struct nat_entry *ne)
 128 {
 129         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 130         struct nat_entry_set *head;
 131
 132         if (get_nat_flag(ne, IS_DIRTY))
 133                 return;
 134 retry:
 135         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 136         if (!head) {
 137                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_ATOMIC);
 138
 139                 INIT_LIST_HEAD(&head->entry_list);
 140                 INIT_LIST_HEAD(&head->set_list);
 141                 head->set = set;
 142                 head->entry_cnt = 0;
 143
 144                 if (radix_tree_insert(&nm_i->nat_set_root, set, head)) {
 145                         cond_resched();
 146                         goto retry;
 147                 }
 148         }
 149         list_move_tail(&ne->list, &head->entry_list);
 150         nm_i->dirty_nat_cnt++;
 151         head->entry_cnt++;
 152         set_nat_flag(ne, IS_DIRTY, true);
 153 }
 154
 155 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 156                                                 struct nat_entry *ne)
 157 {
 158         nid_t set = ne->ni.nid / NAT_ENTRY_PER_BLOCK;
 159         struct nat_entry_set *head;
 160
 161         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 162         if (head) {
 163                 list_move_tail(&ne->list, &nm_i->nat_entries);
 164                 set_nat_flag(ne, IS_DIRTY, false);
 165                 head->entry_cnt--;
 166                 nm_i->dirty_nat_cnt--;
 167         }
 168 }
 169
 170 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 171                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 172 {
 173         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 174                                                         start, nr);
 175 }
 176
 177 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 178 {
 179         struct f2fs_nm_info *nm_i = NM_I(sbi);
 180         struct nat_entry *e;
 181         bool is_cp = true;
 182
 183         read_lock(&nm_i->nat_tree_lock);
 184         e = __lookup_nat_cache(nm_i, nid);
 185         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 186                 is_cp = false;
 187         read_unlock(&nm_i->nat_tree_lock);
 188         return is_cp;
 189 }
 190
 191 bool has_fsynced_inode(struct f2fs_sb_info *sbi, nid_t ino)
 192 {
 193         struct f2fs_nm_info *nm_i = NM_I(sbi);
 194         struct nat_entry *e;
 195         bool fsynced = false;
 196
 197         read_lock(&nm_i->nat_tree_lock);
 198         e = __lookup_nat_cache(nm_i, ino);
 199         if (e && get_nat_flag(e, HAS_FSYNCED_INODE))
 200                 fsynced = true;
 201         read_unlock(&nm_i->nat_tree_lock);
 202         return fsynced;
 203 }
 204
 205 bool need_inode_block_update(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 need_update = true;
 210
 211         read_lock(&nm_i->nat_tree_lock);
 212         e = __lookup_nat_cache(nm_i, ino);
 213         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 214                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 215                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 216                 need_update = false;
 217         read_unlock(&nm_i->nat_tree_lock);
 218         return need_update;
 219 }
 220
 221 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 222 {
 223         struct nat_entry *new;
 224
 225         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 226         if (!new)
 227                 return NULL;
 228         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 229                 kmem_cache_free(nat_entry_slab, new);
 230                 return NULL;
 231         }
 232         memset(new, 0, sizeof(struct nat_entry));
 233         nat_set_nid(new, nid);
 234         nat_reset_flag(new);
 235         list_add_tail(&new->list, &nm_i->nat_entries);
 236         nm_i->nat_cnt++;
 237         return new;
 238 }
 239
 240 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 241                                                 struct f2fs_nat_entry *ne)
 242 {
 243         struct nat_entry *e;
 244 retry:
 245         write_lock(&nm_i->nat_tree_lock);
 246         e = __lookup_nat_cache(nm_i, nid);
 247         if (!e) {
 248                 e = grab_nat_entry(nm_i, nid);
 249                 if (!e) {
 250                         write_unlock(&nm_i->nat_tree_lock);
 251                         goto retry;
 252                 }
 253                 node_info_from_raw_nat(&e->ni, ne);
 254         }
 255         write_unlock(&nm_i->nat_tree_lock);
 256 }
 257
 258 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 259                         block_t new_blkaddr, bool fsync_done)
 260 {
 261         struct f2fs_nm_info *nm_i = NM_I(sbi);
 262         struct nat_entry *e;
 263 retry:
 264         write_lock(&nm_i->nat_tree_lock);
 265         e = __lookup_nat_cache(nm_i, ni->nid);
 266         if (!e) {
 267                 e = grab_nat_entry(nm_i, ni->nid);
 268                 if (!e) {
 269                         write_unlock(&nm_i->nat_tree_lock);
 270                         goto retry;
 271                 }
 272                 e->ni = *ni;
 273                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 274         } else if (new_blkaddr == NEW_ADDR) {
 275                 /*
 276                  * when nid is reallocated,
 277                  * previous nat entry can be remained in nat cache.
 278                  * So, reinitialize it with new information.
 279                  */
 280                 e->ni = *ni;
 281                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 282         }
 283
 284         /* sanity check */
 285         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 286         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 287                         new_blkaddr == NULL_ADDR);
 288         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 289                         new_blkaddr == NEW_ADDR);
 290         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 291                         nat_get_blkaddr(e) != NULL_ADDR &&
 292                         new_blkaddr == NEW_ADDR);
 293
 294         /* increment version no as node is removed */
 295         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 296                 unsigned char version = nat_get_version(e);
 297                 nat_set_version(e, inc_node_version(version));
 298         }
 299
 300         /* change address */
 301         nat_set_blkaddr(e, new_blkaddr);
 302         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 303                 set_nat_flag(e, IS_CHECKPOINTED, false);
 304         __set_nat_cache_dirty(nm_i, e);
 305
 306         /* update fsync_mark if its inode nat entry is still alive */
 307         e = __lookup_nat_cache(nm_i, ni->ino);
 308         if (e) {
 309                 if (fsync_done && ni->nid == ni->ino)
 310                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 311                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 312         }
 313         write_unlock(&nm_i->nat_tree_lock);
 314 }
 315
 316 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 317 {
 318         struct f2fs_nm_info *nm_i = NM_I(sbi);
 319
 320         if (available_free_memory(sbi, NAT_ENTRIES))
 321                 return 0;
 322
 323         write_lock(&nm_i->nat_tree_lock);
 324         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 325                 struct nat_entry *ne;
 326                 ne = list_first_entry(&nm_i->nat_entries,
 327                                         struct nat_entry, list);
 328                 __del_from_nat_cache(nm_i, ne);
 329                 nr_shrink--;
 330         }
 331         write_unlock(&nm_i->nat_tree_lock);
 332         return nr_shrink;
 333 }
 334
 335 /*
 336  * This function always returns success
 337  */
 338 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 339 {
 340         struct f2fs_nm_info *nm_i = NM_I(sbi);
 341         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 342         struct f2fs_summary_block *sum = curseg->sum_blk;
 343         nid_t start_nid = START_NID(nid);
 344         struct f2fs_nat_block *nat_blk;
 345         struct page *page = NULL;
 346         struct f2fs_nat_entry ne;
 347         struct nat_entry *e;
 348         int i;
 349
 350         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 351         ni->nid = nid;
 352
 353         /* Check nat cache */
 354         read_lock(&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         read_unlock(&nm_i->nat_tree_lock);
 362         if (e)
 363                 return;
 364
 365         /* Check current segment summary */
 366         mutex_lock(&curseg->curseg_mutex);
 367         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 368         if (i >= 0) {
 369                 ne = nat_in_journal(sum, i);
 370                 node_info_from_raw_nat(ni, &ne);
 371         }
 372         mutex_unlock(&curseg->curseg_mutex);
 373         if (i >= 0)
 374                 goto cache;
 375
 376         /* Fill node_info from nat page */
 377         page = get_current_nat_page(sbi, start_nid);
 378         nat_blk = (struct f2fs_nat_block *)page_address(page);
 379         ne = nat_blk->entries[nid - start_nid];
 380         node_info_from_raw_nat(ni, &ne);
 381         f2fs_put_page(page, 1);
 382 cache:
 383         /* cache nat entry */
 384         cache_nat_entry(NM_I(sbi), nid, &ne);
 385 }
 386
 387 /*
 388  * The maximum depth is four.
 389  * Offset[0] will have raw inode offset.
 390  */
 391 static int get_node_path(struct f2fs_inode_info *fi, long block,
 392                                 int offset[4], unsigned int noffset[4])
 393 {
 394         const long direct_index = ADDRS_PER_INODE(fi);
 395         const long direct_blks = ADDRS_PER_BLOCK;
 396         const long dptrs_per_blk = NIDS_PER_BLOCK;
 397         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 398         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 399         int n = 0;
 400         int level = 0;
 401
 402         noffset[0] = 0;
 403
 404         if (block < direct_index) {
 405                 offset[n] = block;
 406                 goto got;
 407         }
 408         block -= direct_index;
 409         if (block < direct_blks) {
 410                 offset[n++] = NODE_DIR1_BLOCK;
 411                 noffset[n] = 1;
 412                 offset[n] = block;
 413                 level = 1;
 414                 goto got;
 415         }
 416         block -= direct_blks;
 417         if (block < direct_blks) {
 418                 offset[n++] = NODE_DIR2_BLOCK;
 419                 noffset[n] = 2;
 420                 offset[n] = block;
 421                 level = 1;
 422                 goto got;
 423         }
 424         block -= direct_blks;
 425         if (block < indirect_blks) {
 426                 offset[n++] = NODE_IND1_BLOCK;
 427                 noffset[n] = 3;
 428                 offset[n++] = block / direct_blks;
 429                 noffset[n] = 4 + offset[n - 1];
 430                 offset[n] = block % direct_blks;
 431                 level = 2;
 432                 goto got;
 433         }
 434         block -= indirect_blks;
 435         if (block < indirect_blks) {
 436                 offset[n++] = NODE_IND2_BLOCK;
 437                 noffset[n] = 4 + dptrs_per_blk;
 438                 offset[n++] = block / direct_blks;
 439                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 440                 offset[n] = block % direct_blks;
 441                 level = 2;
 442                 goto got;
 443         }
 444         block -= indirect_blks;
 445         if (block < dindirect_blks) {
 446                 offset[n++] = NODE_DIND_BLOCK;
 447                 noffset[n] = 5 + (dptrs_per_blk * 2);
 448                 offset[n++] = block / indirect_blks;
 449                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 450                               offset[n - 1] * (dptrs_per_blk + 1);
 451                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 452                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 453                               offset[n - 2] * (dptrs_per_blk + 1) +
 454                               offset[n - 1];
 455                 offset[n] = block % direct_blks;
 456                 level = 3;
 457                 goto got;
 458         } else {
 459                 BUG();
 460         }
 461 got:
 462         return level;
 463 }
 464
 465 /*
 466  * Caller should call f2fs_put_dnode(dn).
 467  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 468  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 469  * In the case of RDONLY_NODE, we don't need to care about mutex.
 470  */
 471 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 472 {
 473         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 474         struct page *npage[4];
 475         struct page *parent;
 476         int offset[4];
 477         unsigned int noffset[4];
 478         nid_t nids[4];
 479         int level, i;
 480         int err = 0;
 481
 482         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 483
 484         nids[0] = dn->inode->i_ino;
 485         npage[0] = dn->inode_page;
 486
 487         if (!npage[0]) {
 488                 npage[0] = get_node_page(sbi, nids[0]);
 489                 if (IS_ERR(npage[0]))
 490                         return PTR_ERR(npage[0]);
 491         }
 492         parent = npage[0];
 493         if (level != 0)
 494                 nids[1] = get_nid(parent, offset[0], true);
 495         dn->inode_page = npage[0];
 496         dn->inode_page_locked = true;
 497
 498         /* get indirect or direct nodes */
 499         for (i = 1; i <= level; i++) {
 500                 bool done = false;
 501
 502                 if (!nids[i] && mode == ALLOC_NODE) {
 503                         /* alloc new node */
 504                         if (!alloc_nid(sbi, &(nids[i]))) {
 505                                 err = -ENOSPC;
 506                                 goto release_pages;
 507                         }
 508
 509                         dn->nid = nids[i];
 510                         npage[i] = new_node_page(dn, noffset[i], NULL);
 511                         if (IS_ERR(npage[i])) {
 512                                 alloc_nid_failed(sbi, nids[i]);
 513                                 err = PTR_ERR(npage[i]);
 514                                 goto release_pages;
 515                         }
 516
 517                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 518                         alloc_nid_done(sbi, nids[i]);
 519                         done = true;
 520                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 521                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 522                         if (IS_ERR(npage[i])) {
 523                                 err = PTR_ERR(npage[i]);
 524                                 goto release_pages;
 525                         }
 526                         done = true;
 527                 }
 528                 if (i == 1) {
 529                         dn->inode_page_locked = false;
 530                         unlock_page(parent);
 531                 } else {
 532                         f2fs_put_page(parent, 1);
 533                 }
 534
 535                 if (!done) {
 536                         npage[i] = get_node_page(sbi, nids[i]);
 537                         if (IS_ERR(npage[i])) {
 538                                 err = PTR_ERR(npage[i]);
 539                                 f2fs_put_page(npage[0], 0);
 540                                 goto release_out;
 541                         }
 542                 }
 543                 if (i < level) {
 544                         parent = npage[i];
 545                         nids[i + 1] = get_nid(parent, offset[i], false);
 546                 }
 547         }
 548         dn->nid = nids[level];
 549         dn->ofs_in_node = offset[level];
 550         dn->node_page = npage[level];
 551         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 552         return 0;
 553
 554 release_pages:
 555         f2fs_put_page(parent, 1);
 556         if (i > 1)
 557                 f2fs_put_page(npage[0], 0);
 558 release_out:
 559         dn->inode_page = NULL;
 560         dn->node_page = NULL;
 561         return err;
 562 }
 563
 564 static void truncate_node(struct dnode_of_data *dn)
 565 {
 566         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 567         struct node_info ni;
 568
 569         get_node_info(sbi, dn->nid, &ni);
 570         if (dn->inode->i_blocks == 0) {
 571                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 572                 goto invalidate;
 573         }
 574         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 575
 576         /* Deallocate node address */
 577         invalidate_blocks(sbi, ni.blk_addr);
 578         dec_valid_node_count(sbi, dn->inode);
 579         set_node_addr(sbi, &ni, NULL_ADDR, false);
 580
 581         if (dn->nid == dn->inode->i_ino) {
 582                 remove_orphan_inode(sbi, dn->nid);
 583                 dec_valid_inode_count(sbi);
 584         } else {
 585                 sync_inode_page(dn);
 586         }
 587 invalidate:
 588         clear_node_page_dirty(dn->node_page);
 589         F2FS_SET_SB_DIRT(sbi);
 590
 591         f2fs_put_page(dn->node_page, 1);
 592
 593         invalidate_mapping_pages(NODE_MAPPING(sbi),
 594                         dn->node_page->index, dn->node_page->index);
 595
 596         dn->node_page = NULL;
 597         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 598 }
 599
 600 static int truncate_dnode(struct dnode_of_data *dn)
 601 {
 602         struct page *page;
 603
 604         if (dn->nid == 0)
 605                 return 1;
 606
 607         /* get direct node */
 608         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 609         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 610                 return 1;
 611         else if (IS_ERR(page))
 612                 return PTR_ERR(page);
 613
 614         /* Make dnode_of_data for parameter */
 615         dn->node_page = page;
 616         dn->ofs_in_node = 0;
 617         truncate_data_blocks(dn);
 618         truncate_node(dn);
 619         return 1;
 620 }
 621
 622 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 623                                                 int ofs, int depth)
 624 {
 625         struct dnode_of_data rdn = *dn;
 626         struct page *page;
 627         struct f2fs_node *rn;
 628         nid_t child_nid;
 629         unsigned int child_nofs;
 630         int freed = 0;
 631         int i, ret;
 632
 633         if (dn->nid == 0)
 634                 return NIDS_PER_BLOCK + 1;
 635
 636         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 637
 638         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 639         if (IS_ERR(page)) {
 640                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 641                 return PTR_ERR(page);
 642         }
 643
 644         rn = F2FS_NODE(page);
 645         if (depth < 3) {
 646                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 647                         child_nid = le32_to_cpu(rn->in.nid[i]);
 648                         if (child_nid == 0)
 649                                 continue;
 650                         rdn.nid = child_nid;
 651                         ret = truncate_dnode(&rdn);
 652                         if (ret < 0)
 653                                 goto out_err;
 654                         set_nid(page, i, 0, false);
 655                 }
 656         } else {
 657                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 658                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 659                         child_nid = le32_to_cpu(rn->in.nid[i]);
 660                         if (child_nid == 0) {
 661                                 child_nofs += NIDS_PER_BLOCK + 1;
 662                                 continue;
 663                         }
 664                         rdn.nid = child_nid;
 665                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 666                         if (ret == (NIDS_PER_BLOCK + 1)) {
 667                                 set_nid(page, i, 0, false);
 668                                 child_nofs += ret;
 669                         } else if (ret < 0 && ret != -ENOENT) {
 670                                 goto out_err;
 671                         }
 672                 }
 673                 freed = child_nofs;
 674         }
 675
 676         if (!ofs) {
 677                 /* remove current indirect node */
 678                 dn->node_page = page;
 679                 truncate_node(dn);
 680                 freed++;
 681         } else {
 682                 f2fs_put_page(page, 1);
 683         }
 684         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 685         return freed;
 686
 687 out_err:
 688         f2fs_put_page(page, 1);
 689         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 690         return ret;
 691 }
 692
 693 static int truncate_partial_nodes(struct dnode_of_data *dn,
 694                         struct f2fs_inode *ri, int *offset, int depth)
 695 {
 696         struct page *pages[2];
 697         nid_t nid[3];
 698         nid_t child_nid;
 699         int err = 0;
 700         int i;
 701         int idx = depth - 2;
 702
 703         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 704         if (!nid[0])
 705                 return 0;
 706
 707         /* get indirect nodes in the path */
 708         for (i = 0; i < idx + 1; i++) {
 709                 /* reference count'll be increased */
 710                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 711                 if (IS_ERR(pages[i])) {
 712                         err = PTR_ERR(pages[i]);
 713                         idx = i - 1;
 714                         goto fail;
 715                 }
 716                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 717         }
 718
 719         /* free direct nodes linked to a partial indirect node */
 720         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 721                 child_nid = get_nid(pages[idx], i, false);
 722                 if (!child_nid)
 723                         continue;
 724                 dn->nid = child_nid;
 725                 err = truncate_dnode(dn);
 726                 if (err < 0)
 727                         goto fail;
 728                 set_nid(pages[idx], i, 0, false);
 729         }
 730
 731         if (offset[idx + 1] == 0) {
 732                 dn->node_page = pages[idx];
 733                 dn->nid = nid[idx];
 734                 truncate_node(dn);
 735         } else {
 736                 f2fs_put_page(pages[idx], 1);
 737         }
 738         offset[idx]++;
 739         offset[idx + 1] = 0;
 740         idx--;
 741 fail:
 742         for (i = idx; i >= 0; i--)
 743                 f2fs_put_page(pages[i], 1);
 744
 745         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 746
 747         return err;
 748 }
 749
 750 /*
 751  * All the block addresses of data and nodes should be nullified.
 752  */
 753 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 754 {
 755         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 756         int err = 0, cont = 1;
 757         int level, offset[4], noffset[4];
 758         unsigned int nofs = 0;
 759         struct f2fs_inode *ri;
 760         struct dnode_of_data dn;
 761         struct page *page;
 762
 763         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 764
 765         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 766 restart:
 767         page = get_node_page(sbi, inode->i_ino);
 768         if (IS_ERR(page)) {
 769                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 770                 return PTR_ERR(page);
 771         }
 772
 773         set_new_dnode(&dn, inode, page, NULL, 0);
 774         unlock_page(page);
 775
 776         ri = F2FS_INODE(page);
 777         switch (level) {
 778         case 0:
 779         case 1:
 780                 nofs = noffset[1];
 781                 break;
 782         case 2:
 783                 nofs = noffset[1];
 784                 if (!offset[level - 1])
 785                         goto skip_partial;
 786                 err = truncate_partial_nodes(&dn, ri, offset, level);
 787                 if (err < 0 && err != -ENOENT)
 788                         goto fail;
 789                 nofs += 1 + NIDS_PER_BLOCK;
 790                 break;
 791         case 3:
 792                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 793                 if (!offset[level - 1])
 794                         goto skip_partial;
 795                 err = truncate_partial_nodes(&dn, ri, offset, level);
 796                 if (err < 0 && err != -ENOENT)
 797                         goto fail;
 798                 break;
 799         default:
 800                 BUG();
 801         }
 802
 803 skip_partial:
 804         while (cont) {
 805                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 806                 switch (offset[0]) {
 807                 case NODE_DIR1_BLOCK:
 808                 case NODE_DIR2_BLOCK:
 809                         err = truncate_dnode(&dn);
 810                         break;
 811
 812                 case NODE_IND1_BLOCK:
 813                 case NODE_IND2_BLOCK:
 814                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 815                         break;
 816
 817                 case NODE_DIND_BLOCK:
 818                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 819                         cont = 0;
 820                         break;
 821
 822                 default:
 823                         BUG();
 824                 }
 825                 if (err < 0 && err != -ENOENT)
 826                         goto fail;
 827                 if (offset[1] == 0 &&
 828                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 829                         lock_page(page);
 830                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 831                                 f2fs_put_page(page, 1);
 832                                 goto restart;
 833                         }
 834                         f2fs_wait_on_page_writeback(page, NODE);
 835                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 836                         set_page_dirty(page);
 837                         unlock_page(page);
 838                 }
 839                 offset[1] = 0;
 840                 offset[0]++;
 841                 nofs += err;
 842         }
 843 fail:
 844         f2fs_put_page(page, 0);
 845         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 846         return err > 0 ? 0 : err;
 847 }
 848
 849 int truncate_xattr_node(struct inode *inode, struct page *page)
 850 {
 851         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 852         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 853         struct dnode_of_data dn;
 854         struct page *npage;
 855
 856         if (!nid)
 857                 return 0;
 858
 859         npage = get_node_page(sbi, nid);
 860         if (IS_ERR(npage))
 861                 return PTR_ERR(npage);
 862
 863         F2FS_I(inode)->i_xattr_nid = 0;
 864
 865         /* need to do checkpoint during fsync */
 866         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 867
 868         set_new_dnode(&dn, inode, page, npage, nid);
 869
 870         if (page)
 871                 dn.inode_page_locked = true;
 872         truncate_node(&dn);
 873         return 0;
 874 }
 875
 876 /*
 877  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 878  * f2fs_unlock_op().
 879  */
 880 void remove_inode_page(struct inode *inode)
 881 {
 882         struct dnode_of_data dn;
 883
 884         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 885         if (get_dnode_of_data(&dn, 0, LOOKUP_NODE))
 886                 return;
 887
 888         if (truncate_xattr_node(inode, dn.inode_page)) {
 889                 f2fs_put_dnode(&dn);
 890                 return;
 891         }
 892
 893         /* remove potential inline_data blocks */
 894         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 895                                 S_ISLNK(inode->i_mode))
 896                 truncate_data_blocks_range(&dn, 1);
 897
 898         /* 0 is possible, after f2fs_new_inode() has failed */
 899         f2fs_bug_on(F2FS_I_SB(inode),
 900                         inode->i_blocks != 0 && inode->i_blocks != 1);
 901
 902         /* will put inode & node pages */
 903         truncate_node(&dn);
 904 }
 905
 906 struct page *new_inode_page(struct inode *inode)
 907 {
 908         struct dnode_of_data dn;
 909
 910         /* allocate inode page for new inode */
 911         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 912
 913         /* caller should f2fs_put_page(page, 1); */
 914         return new_node_page(&dn, 0, NULL);
 915 }
 916
 917 struct page *new_node_page(struct dnode_of_data *dn,
 918                                 unsigned int ofs, struct page *ipage)
 919 {
 920         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 921         struct node_info old_ni, new_ni;
 922         struct page *page;
 923         int err;
 924
 925         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 926                 return ERR_PTR(-EPERM);
 927
 928         page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
 929         if (!page)
 930                 return ERR_PTR(-ENOMEM);
 931
 932         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 933                 err = -ENOSPC;
 934                 goto fail;
 935         }
 936
 937         get_node_info(sbi, dn->nid, &old_ni);
 938
 939         /* Reinitialize old_ni with new node page */
 940         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
 941         new_ni = old_ni;
 942         new_ni.ino = dn->inode->i_ino;
 943         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
 944
 945         f2fs_wait_on_page_writeback(page, NODE);
 946         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 947         set_cold_node(dn->inode, page);
 948         SetPageUptodate(page);
 949         set_page_dirty(page);
 950
 951         if (f2fs_has_xattr_block(ofs))
 952                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 953
 954         dn->node_page = page;
 955         if (ipage)
 956                 update_inode(dn->inode, ipage);
 957         else
 958                 sync_inode_page(dn);
 959         if (ofs == 0)
 960                 inc_valid_inode_count(sbi);
 961
 962         return page;
 963
 964 fail:
 965         clear_node_page_dirty(page);
 966         f2fs_put_page(page, 1);
 967         return ERR_PTR(err);
 968 }
 969
 970 /*
 971  * Caller should do after getting the following values.
 972  * 0: f2fs_put_page(page, 0)
 973  * LOCKED_PAGE: f2fs_put_page(page, 1)
 974  * error: nothing
 975  */
 976 static int read_node_page(struct page *page, int rw)
 977 {
 978         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 979         struct node_info ni;
 980
 981         get_node_info(sbi, page->index, &ni);
 982
 983         if (unlikely(ni.blk_addr == NULL_ADDR)) {
 984                 f2fs_put_page(page, 1);
 985                 return -ENOENT;
 986         }
 987
 988         if (PageUptodate(page))
 989                 return LOCKED_PAGE;
 990
 991         return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
 992 }
 993
 994 /*
 995  * Readahead a node page
 996  */
 997 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 998 {
 999         struct page *apage;
1000         int err;
1001
1002         apage = find_get_page(NODE_MAPPING(sbi), nid);
1003         if (apage && PageUptodate(apage)) {
1004                 f2fs_put_page(apage, 0);
1005                 return;
1006         }
1007         f2fs_put_page(apage, 0);
1008
1009         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1010         if (!apage)
1011                 return;
1012
1013         err = read_node_page(apage, READA);
1014         if (err == 0)
1015                 f2fs_put_page(apage, 0);
1016         else if (err == LOCKED_PAGE)
1017                 f2fs_put_page(apage, 1);
1018 }
1019
1020 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1021 {
1022         struct page *page;
1023         int err;
1024 repeat:
1025         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1026         if (!page)
1027                 return ERR_PTR(-ENOMEM);
1028
1029         err = read_node_page(page, READ_SYNC);
1030         if (err < 0)
1031                 return ERR_PTR(err);
1032         else if (err == LOCKED_PAGE)
1033                 goto got_it;
1034
1035         lock_page(page);
1036         if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
1037                 f2fs_put_page(page, 1);
1038                 return ERR_PTR(-EIO);
1039         }
1040         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1041                 f2fs_put_page(page, 1);
1042                 goto repeat;
1043         }
1044 got_it:
1045         return page;
1046 }
1047
1048 /*
1049  * Return a locked page for the desired node page.
1050  * And, readahead MAX_RA_NODE number of node pages.
1051  */
1052 struct page *get_node_page_ra(struct page *parent, int start)
1053 {
1054         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1055         struct blk_plug plug;
1056         struct page *page;
1057         int err, i, end;
1058         nid_t nid;
1059
1060         /* First, try getting the desired direct node. */
1061         nid = get_nid(parent, start, false);
1062         if (!nid)
1063                 return ERR_PTR(-ENOENT);
1064 repeat:
1065         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1066         if (!page)
1067                 return ERR_PTR(-ENOMEM);
1068
1069         err = read_node_page(page, READ_SYNC);
1070         if (err < 0)
1071                 return ERR_PTR(err);
1072         else if (err == LOCKED_PAGE)
1073                 goto page_hit;
1074
1075         blk_start_plug(&plug);
1076
1077         /* Then, try readahead for siblings of the desired node */
1078         end = start + MAX_RA_NODE;
1079         end = min(end, NIDS_PER_BLOCK);
1080         for (i = start + 1; i < end; i++) {
1081                 nid = get_nid(parent, i, false);
1082                 if (!nid)
1083                         continue;
1084                 ra_node_page(sbi, nid);
1085         }
1086
1087         blk_finish_plug(&plug);
1088
1089         lock_page(page);
1090         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1091                 f2fs_put_page(page, 1);
1092                 goto repeat;
1093         }
1094 page_hit:
1095         if (unlikely(!PageUptodate(page))) {
1096                 f2fs_put_page(page, 1);
1097                 return ERR_PTR(-EIO);
1098         }
1099         return page;
1100 }
1101
1102 void sync_inode_page(struct dnode_of_data *dn)
1103 {
1104         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1105                 update_inode(dn->inode, dn->node_page);
1106         } else if (dn->inode_page) {
1107                 if (!dn->inode_page_locked)
1108                         lock_page(dn->inode_page);
1109                 update_inode(dn->inode, dn->inode_page);
1110                 if (!dn->inode_page_locked)
1111                         unlock_page(dn->inode_page);
1112         } else {
1113                 update_inode_page(dn->inode);
1114         }
1115 }
1116
1117 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1118                                         struct writeback_control *wbc)
1119 {
1120         pgoff_t index, end;
1121         struct pagevec pvec;
1122         int step = ino ? 2 : 0;
1123         int nwritten = 0, wrote = 0;
1124
1125         pagevec_init(&pvec, 0);
1126
1127 next_step:
1128         index = 0;
1129         end = LONG_MAX;
1130
1131         while (index <= end) {
1132                 int i, nr_pages;
1133                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1134                                 PAGECACHE_TAG_DIRTY,
1135                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1136                 if (nr_pages == 0)
1137                         break;
1138
1139                 for (i = 0; i < nr_pages; i++) {
1140                         struct page *page = pvec.pages[i];
1141
1142                         /*
1143                          * flushing sequence with step:
1144                          * 0. indirect nodes
1145                          * 1. dentry dnodes
1146                          * 2. file dnodes
1147                          */
1148                         if (step == 0 && IS_DNODE(page))
1149                                 continue;
1150                         if (step == 1 && (!IS_DNODE(page) ||
1151                                                 is_cold_node(page)))
1152                                 continue;
1153                         if (step == 2 && (!IS_DNODE(page) ||
1154                                                 !is_cold_node(page)))
1155                                 continue;
1156
1157                         /*
1158                          * If an fsync mode,
1159                          * we should not skip writing node pages.
1160                          */
1161                         if (ino && ino_of_node(page) == ino)
1162                                 lock_page(page);
1163                         else if (!trylock_page(page))
1164                                 continue;
1165
1166                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1167 continue_unlock:
1168                                 unlock_page(page);
1169                                 continue;
1170                         }
1171                         if (ino && ino_of_node(page) != ino)
1172                                 goto continue_unlock;
1173
1174                         if (!PageDirty(page)) {
1175                                 /* someone wrote it for us */
1176                                 goto continue_unlock;
1177                         }
1178
1179                         if (!clear_page_dirty_for_io(page))
1180                                 goto continue_unlock;
1181
1182                         /* called by fsync() */
1183                         if (ino && IS_DNODE(page)) {
1184                                 set_fsync_mark(page, 1);
1185                                 if (IS_INODE(page)) {
1186                                         if (!is_checkpointed_node(sbi, ino) &&
1187                                                 !has_fsynced_inode(sbi, ino))
1188                                                 set_dentry_mark(page, 1);
1189                                         else
1190                                                 set_dentry_mark(page, 0);
1191                                 }
1192                                 nwritten++;
1193                         } else {
1194                                 set_fsync_mark(page, 0);
1195                                 set_dentry_mark(page, 0);
1196                         }
1197
1198                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1199                                 unlock_page(page);
1200                         else
1201                                 wrote++;
1202
1203                         if (--wbc->nr_to_write == 0)
1204                                 break;
1205                 }
1206                 pagevec_release(&pvec);
1207                 cond_resched();
1208
1209                 if (wbc->nr_to_write == 0) {
1210                         step = 2;
1211                         break;
1212                 }
1213         }
1214
1215         if (step < 2) {
1216                 step++;
1217                 goto next_step;
1218         }
1219
1220         if (wrote)
1221                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1222         return nwritten;
1223 }
1224
1225 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1226 {
1227         pgoff_t index = 0, end = LONG_MAX;
1228         struct pagevec pvec;
1229         int ret2 = 0, ret = 0;
1230
1231         pagevec_init(&pvec, 0);
1232
1233         while (index <= end) {
1234                 int i, nr_pages;
1235                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1236                                 PAGECACHE_TAG_WRITEBACK,
1237                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1238                 if (nr_pages == 0)
1239                         break;
1240
1241                 for (i = 0; i < nr_pages; i++) {
1242                         struct page *page = pvec.pages[i];
1243
1244                         /* until radix tree lookup accepts end_index */
1245                         if (unlikely(page->index > end))
1246                                 continue;
1247
1248                         if (ino && ino_of_node(page) == ino) {
1249                                 f2fs_wait_on_page_writeback(page, NODE);
1250                                 if (TestClearPageError(page))
1251                                         ret = -EIO;
1252                         }
1253                 }
1254                 pagevec_release(&pvec);
1255                 cond_resched();
1256         }
1257
1258         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1259                 ret2 = -ENOSPC;
1260         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1261                 ret2 = -EIO;
1262         if (!ret)
1263                 ret = ret2;
1264         return ret;
1265 }
1266
1267 static int f2fs_write_node_page(struct page *page,
1268                                 struct writeback_control *wbc)
1269 {
1270         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1271         nid_t nid;
1272         block_t new_addr;
1273         struct node_info ni;
1274         struct f2fs_io_info fio = {
1275                 .type = NODE,
1276                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1277         };
1278
1279         trace_f2fs_writepage(page, NODE);
1280
1281         if (unlikely(sbi->por_doing))
1282                 goto redirty_out;
1283         if (unlikely(f2fs_cp_error(sbi)))
1284                 goto redirty_out;
1285
1286         f2fs_wait_on_page_writeback(page, NODE);
1287
1288         /* get old block addr of this node page */
1289         nid = nid_of_node(page);
1290         f2fs_bug_on(sbi, page->index != nid);
1291
1292         get_node_info(sbi, nid, &ni);
1293
1294         /* This page is already truncated */
1295         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1296                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1297                 unlock_page(page);
1298                 return 0;
1299         }
1300
1301         if (wbc->for_reclaim)
1302                 goto redirty_out;
1303
1304         down_read(&sbi->node_write);
1305         set_page_writeback(page);
1306         write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1307         set_node_addr(sbi, &ni, new_addr, is_fsync_dnode(page));
1308         dec_page_count(sbi, F2FS_DIRTY_NODES);
1309         up_read(&sbi->node_write);
1310         unlock_page(page);
1311         return 0;
1312
1313 redirty_out:
1314         redirty_page_for_writepage(wbc, page);
1315         return AOP_WRITEPAGE_ACTIVATE;
1316 }
1317
1318 static int f2fs_write_node_pages(struct address_space *mapping,
1319                             struct writeback_control *wbc)
1320 {
1321         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1322         long diff;
1323
1324         trace_f2fs_writepages(mapping->host, wbc, NODE);
1325
1326         /* balancing f2fs's metadata in background */
1327         f2fs_balance_fs_bg(sbi);
1328
1329         /* collect a number of dirty node pages and write together */
1330         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1331                 goto skip_write;
1332
1333         diff = nr_pages_to_write(sbi, NODE, wbc);
1334         wbc->sync_mode = WB_SYNC_NONE;
1335         sync_node_pages(sbi, 0, wbc);
1336         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1337         return 0;
1338
1339 skip_write:
1340         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1341         return 0;
1342 }
1343
1344 static int f2fs_set_node_page_dirty(struct page *page)
1345 {
1346         trace_f2fs_set_page_dirty(page, NODE);
1347
1348         SetPageUptodate(page);
1349         if (!PageDirty(page)) {
1350                 __set_page_dirty_nobuffers(page);
1351                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1352                 SetPagePrivate(page);
1353                 return 1;
1354         }
1355         return 0;
1356 }
1357
1358 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1359                                       unsigned int length)
1360 {
1361         struct inode *inode = page->mapping->host;
1362         if (PageDirty(page))
1363                 dec_page_count(F2FS_I_SB(inode), F2FS_DIRTY_NODES);
1364         ClearPagePrivate(page);
1365 }
1366
1367 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1368 {
1369         ClearPagePrivate(page);
1370         return 1;
1371 }
1372
1373 /*
1374  * Structure of the f2fs node operations
1375  */
1376 const struct address_space_operations f2fs_node_aops = {
1377         .writepage      = f2fs_write_node_page,
1378         .writepages     = f2fs_write_node_pages,
1379         .set_page_dirty = f2fs_set_node_page_dirty,
1380         .invalidatepage = f2fs_invalidate_node_page,
1381         .releasepage    = f2fs_release_node_page,
1382 };
1383
1384 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1385                                                 nid_t n)
1386 {
1387         return radix_tree_lookup(&nm_i->free_nid_root, n);
1388 }
1389
1390 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1391                                                 struct free_nid *i)
1392 {
1393         list_del(&i->list);
1394         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1395 }
1396
1397 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1398 {
1399         struct f2fs_nm_info *nm_i = NM_I(sbi);
1400         struct free_nid *i;
1401         struct nat_entry *ne;
1402         bool allocated = false;
1403
1404         if (!available_free_memory(sbi, FREE_NIDS))
1405                 return -1;
1406
1407         /* 0 nid should not be used */
1408         if (unlikely(nid == 0))
1409                 return 0;
1410
1411         if (build) {
1412                 /* do not add allocated nids */
1413                 read_lock(&nm_i->nat_tree_lock);
1414                 ne = __lookup_nat_cache(nm_i, nid);
1415                 if (ne &&
1416                         (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1417                                 nat_get_blkaddr(ne) != NULL_ADDR))
1418                         allocated = true;
1419                 read_unlock(&nm_i->nat_tree_lock);
1420                 if (allocated)
1421                         return 0;
1422         }
1423
1424         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1425         i->nid = nid;
1426         i->state = NID_NEW;
1427
1428         spin_lock(&nm_i->free_nid_list_lock);
1429         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1430                 spin_unlock(&nm_i->free_nid_list_lock);
1431                 kmem_cache_free(free_nid_slab, i);
1432                 return 0;
1433         }
1434         list_add_tail(&i->list, &nm_i->free_nid_list);
1435         nm_i->fcnt++;
1436         spin_unlock(&nm_i->free_nid_list_lock);
1437         return 1;
1438 }
1439
1440 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1441 {
1442         struct free_nid *i;
1443         bool need_free = false;
1444
1445         spin_lock(&nm_i->free_nid_list_lock);
1446         i = __lookup_free_nid_list(nm_i, nid);
1447         if (i && i->state == NID_NEW) {
1448                 __del_from_free_nid_list(nm_i, i);
1449                 nm_i->fcnt--;
1450                 need_free = true;
1451         }
1452         spin_unlock(&nm_i->free_nid_list_lock);
1453
1454         if (need_free)
1455                 kmem_cache_free(free_nid_slab, i);
1456 }
1457
1458 static void scan_nat_page(struct f2fs_sb_info *sbi,
1459                         struct page *nat_page, nid_t start_nid)
1460 {
1461         struct f2fs_nm_info *nm_i = NM_I(sbi);
1462         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1463         block_t blk_addr;
1464         int i;
1465
1466         i = start_nid % NAT_ENTRY_PER_BLOCK;
1467
1468         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1469
1470                 if (unlikely(start_nid >= nm_i->max_nid))
1471                         break;
1472
1473                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1474                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1475                 if (blk_addr == NULL_ADDR) {
1476                         if (add_free_nid(sbi, start_nid, true) < 0)
1477                                 break;
1478                 }
1479         }
1480 }
1481
1482 static void build_free_nids(struct f2fs_sb_info *sbi)
1483 {
1484         struct f2fs_nm_info *nm_i = NM_I(sbi);
1485         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1486         struct f2fs_summary_block *sum = curseg->sum_blk;
1487         int i = 0;
1488         nid_t nid = nm_i->next_scan_nid;
1489
1490         /* Enough entries */
1491         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1492                 return;
1493
1494         /* readahead nat pages to be scanned */
1495         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1496
1497         while (1) {
1498                 struct page *page = get_current_nat_page(sbi, nid);
1499
1500                 scan_nat_page(sbi, page, nid);
1501                 f2fs_put_page(page, 1);
1502
1503                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1504                 if (unlikely(nid >= nm_i->max_nid))
1505                         nid = 0;
1506
1507                 if (i++ == FREE_NID_PAGES)
1508                         break;
1509         }
1510
1511         /* go to the next free nat pages to find free nids abundantly */
1512         nm_i->next_scan_nid = nid;
1513
1514         /* find free nids from current sum_pages */
1515         mutex_lock(&curseg->curseg_mutex);
1516         for (i = 0; i < nats_in_cursum(sum); i++) {
1517                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1518                 nid = le32_to_cpu(nid_in_journal(sum, i));
1519                 if (addr == NULL_ADDR)
1520                         add_free_nid(sbi, nid, true);
1521                 else
1522                         remove_free_nid(nm_i, nid);
1523         }
1524         mutex_unlock(&curseg->curseg_mutex);
1525 }
1526
1527 /*
1528  * If this function returns success, caller can obtain a new nid
1529  * from second parameter of this function.
1530  * The returned nid could be used ino as well as nid when inode is created.
1531  */
1532 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1533 {
1534         struct f2fs_nm_info *nm_i = NM_I(sbi);
1535         struct free_nid *i = NULL;
1536 retry:
1537         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1538                 return false;
1539
1540         spin_lock(&nm_i->free_nid_list_lock);
1541
1542         /* We should not use stale free nids created by build_free_nids */
1543         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1544                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1545                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1546                         if (i->state == NID_NEW)
1547                                 break;
1548
1549                 f2fs_bug_on(sbi, i->state != NID_NEW);
1550                 *nid = i->nid;
1551                 i->state = NID_ALLOC;
1552                 nm_i->fcnt--;
1553                 spin_unlock(&nm_i->free_nid_list_lock);
1554                 return true;
1555         }
1556         spin_unlock(&nm_i->free_nid_list_lock);
1557
1558         /* Let's scan nat pages and its caches to get free nids */
1559         mutex_lock(&nm_i->build_lock);
1560         build_free_nids(sbi);
1561         mutex_unlock(&nm_i->build_lock);
1562         goto retry;
1563 }
1564
1565 /*
1566  * alloc_nid() should be called prior to this function.
1567  */
1568 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1569 {
1570         struct f2fs_nm_info *nm_i = NM_I(sbi);
1571         struct free_nid *i;
1572
1573         spin_lock(&nm_i->free_nid_list_lock);
1574         i = __lookup_free_nid_list(nm_i, nid);
1575         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1576         __del_from_free_nid_list(nm_i, i);
1577         spin_unlock(&nm_i->free_nid_list_lock);
1578
1579         kmem_cache_free(free_nid_slab, i);
1580 }
1581
1582 /*
1583  * alloc_nid() should be called prior to this function.
1584  */
1585 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1586 {
1587         struct f2fs_nm_info *nm_i = NM_I(sbi);
1588         struct free_nid *i;
1589         bool need_free = false;
1590
1591         if (!nid)
1592                 return;
1593
1594         spin_lock(&nm_i->free_nid_list_lock);
1595         i = __lookup_free_nid_list(nm_i, nid);
1596         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1597         if (!available_free_memory(sbi, FREE_NIDS)) {
1598                 __del_from_free_nid_list(nm_i, i);
1599                 need_free = true;
1600         } else {
1601                 i->state = NID_NEW;
1602                 nm_i->fcnt++;
1603         }
1604         spin_unlock(&nm_i->free_nid_list_lock);
1605
1606         if (need_free)
1607                 kmem_cache_free(free_nid_slab, i);
1608 }
1609
1610 void recover_inline_xattr(struct inode *inode, struct page *page)
1611 {
1612         void *src_addr, *dst_addr;
1613         size_t inline_size;
1614         struct page *ipage;
1615         struct f2fs_inode *ri;
1616
1617         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1618         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1619
1620         ri = F2FS_INODE(page);
1621         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1622                 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1623                 goto update_inode;
1624         }
1625
1626         dst_addr = inline_xattr_addr(ipage);
1627         src_addr = inline_xattr_addr(page);
1628         inline_size = inline_xattr_size(inode);
1629
1630         f2fs_wait_on_page_writeback(ipage, NODE);
1631         memcpy(dst_addr, src_addr, inline_size);
1632 update_inode:
1633         update_inode(inode, ipage);
1634         f2fs_put_page(ipage, 1);
1635 }
1636
1637 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1638 {
1639         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1640         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1641         nid_t new_xnid = nid_of_node(page);
1642         struct node_info ni;
1643
1644         /* 1: invalidate the previous xattr nid */
1645         if (!prev_xnid)
1646                 goto recover_xnid;
1647
1648         /* Deallocate node address */
1649         get_node_info(sbi, prev_xnid, &ni);
1650         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1651         invalidate_blocks(sbi, ni.blk_addr);
1652         dec_valid_node_count(sbi, inode);
1653         set_node_addr(sbi, &ni, NULL_ADDR, false);
1654
1655 recover_xnid:
1656         /* 2: allocate new xattr nid */
1657         if (unlikely(!inc_valid_node_count(sbi, inode)))
1658                 f2fs_bug_on(sbi, 1);
1659
1660         remove_free_nid(NM_I(sbi), new_xnid);
1661         get_node_info(sbi, new_xnid, &ni);
1662         ni.ino = inode->i_ino;
1663         set_node_addr(sbi, &ni, NEW_ADDR, false);
1664         F2FS_I(inode)->i_xattr_nid = new_xnid;
1665
1666         /* 3: update xattr blkaddr */
1667         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1668         set_node_addr(sbi, &ni, blkaddr, false);
1669
1670         update_inode_page(inode);
1671 }
1672
1673 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1674 {
1675         struct f2fs_inode *src, *dst;
1676         nid_t ino = ino_of_node(page);
1677         struct node_info old_ni, new_ni;
1678         struct page *ipage;
1679
1680         get_node_info(sbi, ino, &old_ni);
1681
1682         if (unlikely(old_ni.blk_addr != NULL_ADDR))
1683                 return -EINVAL;
1684
1685         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1686         if (!ipage)
1687                 return -ENOMEM;
1688
1689         /* Should not use this inode from free nid list */
1690         remove_free_nid(NM_I(sbi), ino);
1691
1692         SetPageUptodate(ipage);
1693         fill_node_footer(ipage, ino, ino, 0, true);
1694
1695         src = F2FS_INODE(page);
1696         dst = F2FS_INODE(ipage);
1697
1698         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1699         dst->i_size = 0;
1700         dst->i_blocks = cpu_to_le64(1);
1701         dst->i_links = cpu_to_le32(1);
1702         dst->i_xattr_nid = 0;
1703         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1704
1705         new_ni = old_ni;
1706         new_ni.ino = ino;
1707
1708         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1709                 WARN_ON(1);
1710         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1711         inc_valid_inode_count(sbi);
1712         set_page_dirty(ipage);
1713         f2fs_put_page(ipage, 1);
1714         return 0;
1715 }
1716
1717 /*
1718  * ra_sum_pages() merge contiguous pages into one bio and submit.
1719  * these pre-read pages are allocated in bd_inode's mapping tree.
1720  */
1721 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct page **pages,
1722                                 int start, int nrpages)
1723 {
1724         struct inode *inode = sbi->sb->s_bdev->bd_inode;
1725         struct address_space *mapping = inode->i_mapping;
1726         int i, page_idx = start;
1727         struct f2fs_io_info fio = {
1728                 .type = META,
1729                 .rw = READ_SYNC | REQ_META | REQ_PRIO
1730         };
1731
1732         for (i = 0; page_idx < start + nrpages; page_idx++, i++) {
1733                 /* alloc page in bd_inode for reading node summary info */
1734                 pages[i] = grab_cache_page(mapping, page_idx);
1735                 if (!pages[i])
1736                         break;
1737                 f2fs_submit_page_mbio(sbi, pages[i], page_idx, &fio);
1738         }
1739
1740         f2fs_submit_merged_bio(sbi, META, READ);
1741         return i;
1742 }
1743
1744 int restore_node_summary(struct f2fs_sb_info *sbi,
1745                         unsigned int segno, struct f2fs_summary_block *sum)
1746 {
1747         struct f2fs_node *rn;
1748         struct f2fs_summary *sum_entry;
1749         struct inode *inode = sbi->sb->s_bdev->bd_inode;
1750         block_t addr;
1751         int bio_blocks = MAX_BIO_BLOCKS(sbi);
1752         struct page *pages[bio_blocks];
1753         int i, idx, last_offset, nrpages, err = 0;
1754
1755         /* scan the node segment */
1756         last_offset = sbi->blocks_per_seg;
1757         addr = START_BLOCK(sbi, segno);
1758         sum_entry = &sum->entries[0];
1759
1760         for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) {
1761                 nrpages = min(last_offset - i, bio_blocks);
1762
1763                 /* readahead node pages */
1764                 nrpages = ra_sum_pages(sbi, pages, addr, nrpages);
1765                 if (!nrpages)
1766                         return -ENOMEM;
1767
1768                 for (idx = 0; idx < nrpages; idx++) {
1769                         if (err)
1770                                 goto skip;
1771
1772                         lock_page(pages[idx]);
1773                         if (unlikely(!PageUptodate(pages[idx]))) {
1774                                 err = -EIO;
1775                         } else {
1776                                 rn = F2FS_NODE(pages[idx]);
1777                                 sum_entry->nid = rn->footer.nid;
1778                                 sum_entry->version = 0;
1779                                 sum_entry->ofs_in_node = 0;
1780                                 sum_entry++;
1781                         }
1782                         unlock_page(pages[idx]);
1783 skip:
1784                         page_cache_release(pages[idx]);
1785                 }
1786
1787                 invalidate_mapping_pages(inode->i_mapping, addr,
1788                                                         addr + nrpages);
1789         }
1790         return err;
1791 }
1792
1793 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1794 {
1795         struct f2fs_nm_info *nm_i = NM_I(sbi);
1796         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1797         struct f2fs_summary_block *sum = curseg->sum_blk;
1798         int i;
1799
1800         mutex_lock(&curseg->curseg_mutex);
1801         for (i = 0; i < nats_in_cursum(sum); i++) {
1802                 struct nat_entry *ne;
1803                 struct f2fs_nat_entry raw_ne;
1804                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1805
1806                 raw_ne = nat_in_journal(sum, i);
1807 retry:
1808                 write_lock(&nm_i->nat_tree_lock);
1809                 ne = __lookup_nat_cache(nm_i, nid);
1810                 if (ne)
1811                         goto found;
1812
1813                 ne = grab_nat_entry(nm_i, nid);
1814                 if (!ne) {
1815                         write_unlock(&nm_i->nat_tree_lock);
1816                         goto retry;
1817                 }
1818                 node_info_from_raw_nat(&ne->ni, &raw_ne);
1819 found:
1820                 __set_nat_cache_dirty(nm_i, ne);
1821                 write_unlock(&nm_i->nat_tree_lock);
1822         }
1823         update_nats_in_cursum(sum, -i);
1824         mutex_unlock(&curseg->curseg_mutex);
1825 }
1826
1827 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1828                                                 struct list_head *head, int max)
1829 {
1830         struct nat_entry_set *cur;
1831
1832         if (nes->entry_cnt >= max)
1833                 goto add_out;
1834
1835         list_for_each_entry(cur, head, set_list) {
1836                 if (cur->entry_cnt >= nes->entry_cnt) {
1837                         list_add(&nes->set_list, cur->set_list.prev);
1838                         return;
1839                 }
1840         }
1841 add_out:
1842         list_add_tail(&nes->set_list, head);
1843 }
1844
1845 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1846                                         struct nat_entry_set *set)
1847 {
1848         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1849         struct f2fs_summary_block *sum = curseg->sum_blk;
1850         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1851         bool to_journal = true;
1852         struct f2fs_nat_block *nat_blk;
1853         struct nat_entry *ne, *cur;
1854         struct page *page = NULL;
1855
1856         /*
1857          * there are two steps to flush nat entries:
1858          * #1, flush nat entries to journal in current hot data summary block.
1859          * #2, flush nat entries to nat page.
1860          */
1861         if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1862                 to_journal = false;
1863
1864         if (to_journal) {
1865                 mutex_lock(&curseg->curseg_mutex);
1866         } else {
1867                 page = get_next_nat_page(sbi, start_nid);
1868                 nat_blk = page_address(page);
1869                 f2fs_bug_on(sbi, !nat_blk);
1870         }
1871
1872         /* flush dirty nats in nat entry set */
1873         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1874                 struct f2fs_nat_entry *raw_ne;
1875                 nid_t nid = nat_get_nid(ne);
1876                 int offset;
1877
1878                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1879                         continue;
1880
1881                 if (to_journal) {
1882                         offset = lookup_journal_in_cursum(sum,
1883                                                         NAT_JOURNAL, nid, 1);
1884                         f2fs_bug_on(sbi, offset < 0);
1885                         raw_ne = &nat_in_journal(sum, offset);
1886                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1887                 } else {
1888                         raw_ne = &nat_blk->entries[nid - start_nid];
1889                 }
1890                 raw_nat_from_node_info(raw_ne, &ne->ni);
1891
1892                 write_lock(&NM_I(sbi)->nat_tree_lock);
1893                 nat_reset_flag(ne);
1894                 __clear_nat_cache_dirty(NM_I(sbi), ne);
1895                 write_unlock(&NM_I(sbi)->nat_tree_lock);
1896
1897                 if (nat_get_blkaddr(ne) == NULL_ADDR)
1898                         add_free_nid(sbi, nid, false);
1899         }
1900
1901         if (to_journal)
1902                 mutex_unlock(&curseg->curseg_mutex);
1903         else
1904                 f2fs_put_page(page, 1);
1905
1906         if (!set->entry_cnt) {
1907                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1908                 kmem_cache_free(nat_entry_set_slab, set);
1909         }
1910 }
1911
1912 /*
1913  * This function is called during the checkpointing process.
1914  */
1915 void flush_nat_entries(struct f2fs_sb_info *sbi)
1916 {
1917         struct f2fs_nm_info *nm_i = NM_I(sbi);
1918         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1919         struct f2fs_summary_block *sum = curseg->sum_blk;
1920         struct nat_entry_set *setvec[NATVEC_SIZE];
1921         struct nat_entry_set *set, *tmp;
1922         unsigned int found;
1923         nid_t set_idx = 0;
1924         LIST_HEAD(sets);
1925
1926         /*
1927          * if there are no enough space in journal to store dirty nat
1928          * entries, remove all entries from journal and merge them
1929          * into nat entry set.
1930          */
1931         if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1932                 remove_nats_in_journal(sbi);
1933
1934         if (!nm_i->dirty_nat_cnt)
1935                 return;
1936
1937         while ((found = __gang_lookup_nat_set(nm_i,
1938                                         set_idx, NATVEC_SIZE, setvec))) {
1939                 unsigned idx;
1940                 set_idx = setvec[found - 1]->set + 1;
1941                 for (idx = 0; idx < found; idx++)
1942                         __adjust_nat_entry_set(setvec[idx], &sets,
1943                                                         MAX_NAT_JENTRIES(sum));
1944         }
1945
1946         /* flush dirty nats in nat entry set */
1947         list_for_each_entry_safe(set, tmp, &sets, set_list)
1948                 __flush_nat_entry_set(sbi, set);
1949
1950         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1951 }
1952
1953 static int init_node_manager(struct f2fs_sb_info *sbi)
1954 {
1955         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1956         struct f2fs_nm_info *nm_i = NM_I(sbi);
1957         unsigned char *version_bitmap;
1958         unsigned int nat_segs, nat_blocks;
1959
1960         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1961
1962         /* segment_count_nat includes pair segment so divide to 2. */
1963         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1964         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1965
1966         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1967
1968         /* not used nids: 0, node, meta, (and root counted as valid node) */
1969         nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
1970         nm_i->fcnt = 0;
1971         nm_i->nat_cnt = 0;
1972         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1973
1974         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1975         INIT_LIST_HEAD(&nm_i->free_nid_list);
1976         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1977         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_ATOMIC);
1978         INIT_LIST_HEAD(&nm_i->nat_entries);
1979
1980         mutex_init(&nm_i->build_lock);
1981         spin_lock_init(&nm_i->free_nid_list_lock);
1982         rwlock_init(&nm_i->nat_tree_lock);
1983
1984         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1985         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1986         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1987         if (!version_bitmap)
1988                 return -EFAULT;
1989
1990         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1991                                         GFP_KERNEL);
1992         if (!nm_i->nat_bitmap)
1993                 return -ENOMEM;
1994         return 0;
1995 }
1996
1997 int build_node_manager(struct f2fs_sb_info *sbi)
1998 {
1999         int err;
2000
2001         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2002         if (!sbi->nm_info)
2003                 return -ENOMEM;
2004
2005         err = init_node_manager(sbi);
2006         if (err)
2007                 return err;
2008
2009         build_free_nids(sbi);
2010         return 0;
2011 }
2012
2013 void destroy_node_manager(struct f2fs_sb_info *sbi)
2014 {
2015         struct f2fs_nm_info *nm_i = NM_I(sbi);
2016         struct free_nid *i, *next_i;
2017         struct nat_entry *natvec[NATVEC_SIZE];
2018         nid_t nid = 0;
2019         unsigned int found;
2020
2021         if (!nm_i)
2022                 return;
2023
2024         /* destroy free nid list */
2025         spin_lock(&nm_i->free_nid_list_lock);
2026         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2027                 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2028                 __del_from_free_nid_list(nm_i, i);
2029                 nm_i->fcnt--;
2030                 spin_unlock(&nm_i->free_nid_list_lock);
2031                 kmem_cache_free(free_nid_slab, i);
2032                 spin_lock(&nm_i->free_nid_list_lock);
2033         }
2034         f2fs_bug_on(sbi, nm_i->fcnt);
2035         spin_unlock(&nm_i->free_nid_list_lock);
2036
2037         /* destroy nat cache */
2038         write_lock(&nm_i->nat_tree_lock);
2039         while ((found = __gang_lookup_nat_cache(nm_i,
2040                                         nid, NATVEC_SIZE, natvec))) {
2041                 unsigned idx;
2042                 nid = nat_get_nid(natvec[found - 1]) + 1;
2043                 for (idx = 0; idx < found; idx++)
2044                         __del_from_nat_cache(nm_i, natvec[idx]);
2045         }
2046         f2fs_bug_on(sbi, nm_i->nat_cnt);
2047         write_unlock(&nm_i->nat_tree_lock);
2048
2049         kfree(nm_i->nat_bitmap);
2050         sbi->nm_info = NULL;
2051         kfree(nm_i);
2052 }
2053
2054 int __init create_node_manager_caches(void)
2055 {
2056         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2057                         sizeof(struct nat_entry));
2058         if (!nat_entry_slab)
2059                 goto fail;
2060
2061         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2062                         sizeof(struct free_nid));
2063         if (!free_nid_slab)
2064                 goto destory_nat_entry;
2065
2066         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2067                         sizeof(struct nat_entry_set));
2068         if (!nat_entry_set_slab)
2069                 goto destory_free_nid;
2070         return 0;
2071
2072 destory_free_nid:
2073         kmem_cache_destroy(free_nid_slab);
2074 destory_nat_entry:
2075         kmem_cache_destroy(nat_entry_slab);
2076 fail:
2077         return -ENOMEM;
2078 }
2079
2080 void destroy_node_manager_caches(void)
2081 {
2082         kmem_cache_destroy(nat_entry_set_slab);
2083         kmem_cache_destroy(free_nid_slab);
2084         kmem_cache_destroy(nat_entry_slab);
2085 }