fs/f2fs/node.c

   1 /*
   2  * fs/f2fs/node.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/mpage.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/swap.h>
  18
  19 #include "f2fs.h"
  20 #include "node.h"
  21 #include "segment.h"
  22 #include "trace.h"
  23 #include <trace/events/f2fs.h>
  24
  25 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
  26
  27 static struct kmem_cache *nat_entry_slab;
  28 static struct kmem_cache *free_nid_slab;
  29 static struct kmem_cache *nat_entry_set_slab;
  30
  31 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
  32 {
  33         struct f2fs_nm_info *nm_i = NM_I(sbi);
  34         struct sysinfo val;
  35         unsigned long avail_ram;
  36         unsigned long mem_size = 0;
  37         bool res = false;
  38
  39         si_meminfo(&val);
  40
  41         /* only uses low memory */
  42         avail_ram = val.totalram - val.totalhigh;
  43
  44         /*
  45          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  46          */
  47         if (type == FREE_NIDS) {
  48                 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
  49                                                         PAGE_CACHE_SHIFT;
  50                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  51         } else if (type == NAT_ENTRIES) {
  52                 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
  53                                                         PAGE_CACHE_SHIFT;
  54                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  55         } else if (type == DIRTY_DENTS) {
  56                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  57                         return false;
  58                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  59                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  60         } else if (type == INO_ENTRIES) {
  61                 int i;
  62
  63                 for (i = 0; i <= UPDATE_INO; i++)
  64                         mem_size += (sbi->im[i].ino_num *
  65                                 sizeof(struct ino_entry)) >> PAGE_CACHE_SHIFT;
  66                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  67         } else if (type == EXTENT_CACHE) {
  68                 mem_size = (sbi->total_ext_tree * sizeof(struct extent_tree) +
  69                                 atomic_read(&sbi->total_ext_node) *
  70                                 sizeof(struct extent_node)) >> PAGE_CACHE_SHIFT;
  71                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  72         } else {
  73                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  74                         return false;
  75         }
  76         return res;
  77 }
  78
  79 static void clear_node_page_dirty(struct page *page)
  80 {
  81         struct address_space *mapping = page->mapping;
  82         unsigned int long flags;
  83
  84         if (PageDirty(page)) {
  85                 spin_lock_irqsave(&mapping->tree_lock, flags);
  86                 radix_tree_tag_clear(&mapping->page_tree,
  87                                 page_index(page),
  88                                 PAGECACHE_TAG_DIRTY);
  89                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  90
  91                 clear_page_dirty_for_io(page);
  92                 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
  93         }
  94         ClearPageUptodate(page);
  95 }
  96
  97 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  98 {
  99         pgoff_t index = current_nat_addr(sbi, nid);
 100         return get_meta_page(sbi, index);
 101 }
 102
 103 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 104 {
 105         struct page *src_page;
 106         struct page *dst_page;
 107         pgoff_t src_off;
 108         pgoff_t dst_off;
 109         void *src_addr;
 110         void *dst_addr;
 111         struct f2fs_nm_info *nm_i = NM_I(sbi);
 112
 113         src_off = current_nat_addr(sbi, nid);
 114         dst_off = next_nat_addr(sbi, src_off);
 115
 116         /* get current nat block page with lock */
 117         src_page = get_meta_page(sbi, src_off);
 118         dst_page = grab_meta_page(sbi, dst_off);
 119         f2fs_bug_on(sbi, PageDirty(src_page));
 120
 121         src_addr = page_address(src_page);
 122         dst_addr = page_address(dst_page);
 123         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
 124         set_page_dirty(dst_page);
 125         f2fs_put_page(src_page, 1);
 126
 127         set_to_next_nat(nm_i, nid);
 128
 129         return dst_page;
 130 }
 131
 132 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 133 {
 134         return radix_tree_lookup(&nm_i->nat_root, n);
 135 }
 136
 137 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 138                 nid_t start, unsigned int nr, struct nat_entry **ep)
 139 {
 140         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 141 }
 142
 143 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 144 {
 145         list_del(&e->list);
 146         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 147         nm_i->nat_cnt--;
 148         kmem_cache_free(nat_entry_slab, e);
 149 }
 150
 151 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 152                                                 struct nat_entry *ne)
 153 {
 154         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 155         struct nat_entry_set *head;
 156
 157         if (get_nat_flag(ne, IS_DIRTY))
 158                 return;
 159
 160         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 161         if (!head) {
 162                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_ATOMIC);
 163
 164                 INIT_LIST_HEAD(&head->entry_list);
 165                 INIT_LIST_HEAD(&head->set_list);
 166                 head->set = set;
 167                 head->entry_cnt = 0;
 168                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 169         }
 170         list_move_tail(&ne->list, &head->entry_list);
 171         nm_i->dirty_nat_cnt++;
 172         head->entry_cnt++;
 173         set_nat_flag(ne, IS_DIRTY, true);
 174 }
 175
 176 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 177                                                 struct nat_entry *ne)
 178 {
 179         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 180         struct nat_entry_set *head;
 181
 182         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 183         if (head) {
 184                 list_move_tail(&ne->list, &nm_i->nat_entries);
 185                 set_nat_flag(ne, IS_DIRTY, false);
 186                 head->entry_cnt--;
 187                 nm_i->dirty_nat_cnt--;
 188         }
 189 }
 190
 191 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 192                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 193 {
 194         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 195                                                         start, nr);
 196 }
 197
 198 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 199 {
 200         struct f2fs_nm_info *nm_i = NM_I(sbi);
 201         struct nat_entry *e;
 202         bool need = false;
 203
 204         down_read(&nm_i->nat_tree_lock);
 205         e = __lookup_nat_cache(nm_i, nid);
 206         if (e) {
 207                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 208                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 209                         need = true;
 210         }
 211         up_read(&nm_i->nat_tree_lock);
 212         return need;
 213 }
 214
 215 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 216 {
 217         struct f2fs_nm_info *nm_i = NM_I(sbi);
 218         struct nat_entry *e;
 219         bool is_cp = true;
 220
 221         down_read(&nm_i->nat_tree_lock);
 222         e = __lookup_nat_cache(nm_i, nid);
 223         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 224                 is_cp = false;
 225         up_read(&nm_i->nat_tree_lock);
 226         return is_cp;
 227 }
 228
 229 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 230 {
 231         struct f2fs_nm_info *nm_i = NM_I(sbi);
 232         struct nat_entry *e;
 233         bool need_update = true;
 234
 235         down_read(&nm_i->nat_tree_lock);
 236         e = __lookup_nat_cache(nm_i, ino);
 237         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 238                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 239                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 240                 need_update = false;
 241         up_read(&nm_i->nat_tree_lock);
 242         return need_update;
 243 }
 244
 245 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 246 {
 247         struct nat_entry *new;
 248
 249         new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 250         f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
 251         memset(new, 0, sizeof(struct nat_entry));
 252         nat_set_nid(new, nid);
 253         nat_reset_flag(new);
 254         list_add_tail(&new->list, &nm_i->nat_entries);
 255         nm_i->nat_cnt++;
 256         return new;
 257 }
 258
 259 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 260                                                 struct f2fs_nat_entry *ne)
 261 {
 262         struct nat_entry *e;
 263
 264         down_write(&nm_i->nat_tree_lock);
 265         e = __lookup_nat_cache(nm_i, nid);
 266         if (!e) {
 267                 e = grab_nat_entry(nm_i, nid);
 268                 node_info_from_raw_nat(&e->ni, ne);
 269         }
 270         up_write(&nm_i->nat_tree_lock);
 271 }
 272
 273 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 274                         block_t new_blkaddr, bool fsync_done)
 275 {
 276         struct f2fs_nm_info *nm_i = NM_I(sbi);
 277         struct nat_entry *e;
 278
 279         down_write(&nm_i->nat_tree_lock);
 280         e = __lookup_nat_cache(nm_i, ni->nid);
 281         if (!e) {
 282                 e = grab_nat_entry(nm_i, ni->nid);
 283                 copy_node_info(&e->ni, ni);
 284                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 285         } else if (new_blkaddr == NEW_ADDR) {
 286                 /*
 287                  * when nid is reallocated,
 288                  * previous nat entry can be remained in nat cache.
 289                  * So, reinitialize it with new information.
 290                  */
 291                 copy_node_info(&e->ni, ni);
 292                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 293         }
 294
 295         /* sanity check */
 296         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 297         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 298                         new_blkaddr == NULL_ADDR);
 299         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 300                         new_blkaddr == NEW_ADDR);
 301         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 302                         nat_get_blkaddr(e) != NULL_ADDR &&
 303                         new_blkaddr == NEW_ADDR);
 304
 305         /* increment version no as node is removed */
 306         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 307                 unsigned char version = nat_get_version(e);
 308                 nat_set_version(e, inc_node_version(version));
 309         }
 310
 311         /* change address */
 312         nat_set_blkaddr(e, new_blkaddr);
 313         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 314                 set_nat_flag(e, IS_CHECKPOINTED, false);
 315         __set_nat_cache_dirty(nm_i, e);
 316
 317         /* update fsync_mark if its inode nat entry is still alive */
 318         if (ni->nid != ni->ino)
 319                 e = __lookup_nat_cache(nm_i, ni->ino);
 320         if (e) {
 321                 if (fsync_done && ni->nid == ni->ino)
 322                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 323                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 324         }
 325         up_write(&nm_i->nat_tree_lock);
 326 }
 327
 328 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 329 {
 330         struct f2fs_nm_info *nm_i = NM_I(sbi);
 331
 332         if (available_free_memory(sbi, NAT_ENTRIES))
 333                 return 0;
 334
 335         down_write(&nm_i->nat_tree_lock);
 336         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 337                 struct nat_entry *ne;
 338                 ne = list_first_entry(&nm_i->nat_entries,
 339                                         struct nat_entry, list);
 340                 __del_from_nat_cache(nm_i, ne);
 341                 nr_shrink--;
 342         }
 343         up_write(&nm_i->nat_tree_lock);
 344         return nr_shrink;
 345 }
 346
 347 /*
 348  * This function always returns success
 349  */
 350 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 351 {
 352         struct f2fs_nm_info *nm_i = NM_I(sbi);
 353         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 354         struct f2fs_summary_block *sum = curseg->sum_blk;
 355         nid_t start_nid = START_NID(nid);
 356         struct f2fs_nat_block *nat_blk;
 357         struct page *page = NULL;
 358         struct f2fs_nat_entry ne;
 359         struct nat_entry *e;
 360         int i;
 361
 362         ni->nid = nid;
 363
 364         /* Check nat cache */
 365         down_read(&nm_i->nat_tree_lock);
 366         e = __lookup_nat_cache(nm_i, nid);
 367         if (e) {
 368                 ni->ino = nat_get_ino(e);
 369                 ni->blk_addr = nat_get_blkaddr(e);
 370                 ni->version = nat_get_version(e);
 371         }
 372         up_read(&nm_i->nat_tree_lock);
 373         if (e)
 374                 return;
 375
 376         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 377
 378         /* Check current segment summary */
 379         mutex_lock(&curseg->curseg_mutex);
 380         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 381         if (i >= 0) {
 382                 ne = nat_in_journal(sum, i);
 383                 node_info_from_raw_nat(ni, &ne);
 384         }
 385         mutex_unlock(&curseg->curseg_mutex);
 386         if (i >= 0)
 387                 goto cache;
 388
 389         /* Fill node_info from nat page */
 390         page = get_current_nat_page(sbi, start_nid);
 391         nat_blk = (struct f2fs_nat_block *)page_address(page);
 392         ne = nat_blk->entries[nid - start_nid];
 393         node_info_from_raw_nat(ni, &ne);
 394         f2fs_put_page(page, 1);
 395 cache:
 396         /* cache nat entry */
 397         cache_nat_entry(NM_I(sbi), nid, &ne);
 398 }
 399
 400 /*
 401  * The maximum depth is four.
 402  * Offset[0] will have raw inode offset.
 403  */
 404 static int get_node_path(struct f2fs_inode_info *fi, long block,
 405                                 int offset[4], unsigned int noffset[4])
 406 {
 407         const long direct_index = ADDRS_PER_INODE(fi);
 408         const long direct_blks = ADDRS_PER_BLOCK;
 409         const long dptrs_per_blk = NIDS_PER_BLOCK;
 410         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 411         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 412         int n = 0;
 413         int level = 0;
 414
 415         noffset[0] = 0;
 416
 417         if (block < direct_index) {
 418                 offset[n] = block;
 419                 goto got;
 420         }
 421         block -= direct_index;
 422         if (block < direct_blks) {
 423                 offset[n++] = NODE_DIR1_BLOCK;
 424                 noffset[n] = 1;
 425                 offset[n] = block;
 426                 level = 1;
 427                 goto got;
 428         }
 429         block -= direct_blks;
 430         if (block < direct_blks) {
 431                 offset[n++] = NODE_DIR2_BLOCK;
 432                 noffset[n] = 2;
 433                 offset[n] = block;
 434                 level = 1;
 435                 goto got;
 436         }
 437         block -= direct_blks;
 438         if (block < indirect_blks) {
 439                 offset[n++] = NODE_IND1_BLOCK;
 440                 noffset[n] = 3;
 441                 offset[n++] = block / direct_blks;
 442                 noffset[n] = 4 + offset[n - 1];
 443                 offset[n] = block % direct_blks;
 444                 level = 2;
 445                 goto got;
 446         }
 447         block -= indirect_blks;
 448         if (block < indirect_blks) {
 449                 offset[n++] = NODE_IND2_BLOCK;
 450                 noffset[n] = 4 + dptrs_per_blk;
 451                 offset[n++] = block / direct_blks;
 452                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 453                 offset[n] = block % direct_blks;
 454                 level = 2;
 455                 goto got;
 456         }
 457         block -= indirect_blks;
 458         if (block < dindirect_blks) {
 459                 offset[n++] = NODE_DIND_BLOCK;
 460                 noffset[n] = 5 + (dptrs_per_blk * 2);
 461                 offset[n++] = block / indirect_blks;
 462                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 463                               offset[n - 1] * (dptrs_per_blk + 1);
 464                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 465                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 466                               offset[n - 2] * (dptrs_per_blk + 1) +
 467                               offset[n - 1];
 468                 offset[n] = block % direct_blks;
 469                 level = 3;
 470                 goto got;
 471         } else {
 472                 BUG();
 473         }
 474 got:
 475         return level;
 476 }
 477
 478 /*
 479  * Caller should call f2fs_put_dnode(dn).
 480  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 481  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 482  * In the case of RDONLY_NODE, we don't need to care about mutex.
 483  */
 484 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 485 {
 486         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 487         struct page *npage[4];
 488         struct page *parent = NULL;
 489         int offset[4];
 490         unsigned int noffset[4];
 491         nid_t nids[4];
 492         int level, i;
 493         int err = 0;
 494
 495         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 496
 497         nids[0] = dn->inode->i_ino;
 498         npage[0] = dn->inode_page;
 499
 500         if (!npage[0]) {
 501                 npage[0] = get_node_page(sbi, nids[0]);
 502                 if (IS_ERR(npage[0]))
 503                         return PTR_ERR(npage[0]);
 504         }
 505
 506         /* if inline_data is set, should not report any block indices */
 507         if (f2fs_has_inline_data(dn->inode) && index) {
 508                 err = -ENOENT;
 509                 f2fs_put_page(npage[0], 1);
 510                 goto release_out;
 511         }
 512
 513         parent = npage[0];
 514         if (level != 0)
 515                 nids[1] = get_nid(parent, offset[0], true);
 516         dn->inode_page = npage[0];
 517         dn->inode_page_locked = true;
 518
 519         /* get indirect or direct nodes */
 520         for (i = 1; i <= level; i++) {
 521                 bool done = false;
 522
 523                 if (!nids[i] && mode == ALLOC_NODE) {
 524                         /* alloc new node */
 525                         if (!alloc_nid(sbi, &(nids[i]))) {
 526                                 err = -ENOSPC;
 527                                 goto release_pages;
 528                         }
 529
 530                         dn->nid = nids[i];
 531                         npage[i] = new_node_page(dn, noffset[i], NULL);
 532                         if (IS_ERR(npage[i])) {
 533                                 alloc_nid_failed(sbi, nids[i]);
 534                                 err = PTR_ERR(npage[i]);
 535                                 goto release_pages;
 536                         }
 537
 538                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 539                         alloc_nid_done(sbi, nids[i]);
 540                         done = true;
 541                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 542                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 543                         if (IS_ERR(npage[i])) {
 544                                 err = PTR_ERR(npage[i]);
 545                                 goto release_pages;
 546                         }
 547                         done = true;
 548                 }
 549                 if (i == 1) {
 550                         dn->inode_page_locked = false;
 551                         unlock_page(parent);
 552                 } else {
 553                         f2fs_put_page(parent, 1);
 554                 }
 555
 556                 if (!done) {
 557                         npage[i] = get_node_page(sbi, nids[i]);
 558                         if (IS_ERR(npage[i])) {
 559                                 err = PTR_ERR(npage[i]);
 560                                 f2fs_put_page(npage[0], 0);
 561                                 goto release_out;
 562                         }
 563                 }
 564                 if (i < level) {
 565                         parent = npage[i];
 566                         nids[i + 1] = get_nid(parent, offset[i], false);
 567                 }
 568         }
 569         dn->nid = nids[level];
 570         dn->ofs_in_node = offset[level];
 571         dn->node_page = npage[level];
 572         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 573         return 0;
 574
 575 release_pages:
 576         f2fs_put_page(parent, 1);
 577         if (i > 1)
 578                 f2fs_put_page(npage[0], 0);
 579 release_out:
 580         dn->inode_page = NULL;
 581         dn->node_page = NULL;
 582         return err;
 583 }
 584
 585 static void truncate_node(struct dnode_of_data *dn)
 586 {
 587         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 588         struct node_info ni;
 589
 590         get_node_info(sbi, dn->nid, &ni);
 591         if (dn->inode->i_blocks == 0) {
 592                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 593                 goto invalidate;
 594         }
 595         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 596
 597         /* Deallocate node address */
 598         invalidate_blocks(sbi, ni.blk_addr);
 599         dec_valid_node_count(sbi, dn->inode);
 600         set_node_addr(sbi, &ni, NULL_ADDR, false);
 601
 602         if (dn->nid == dn->inode->i_ino) {
 603                 remove_orphan_inode(sbi, dn->nid);
 604                 dec_valid_inode_count(sbi);
 605         } else {
 606                 sync_inode_page(dn);
 607         }
 608 invalidate:
 609         clear_node_page_dirty(dn->node_page);
 610         set_sbi_flag(sbi, SBI_IS_DIRTY);
 611
 612         f2fs_put_page(dn->node_page, 1);
 613
 614         invalidate_mapping_pages(NODE_MAPPING(sbi),
 615                         dn->node_page->index, dn->node_page->index);
 616
 617         dn->node_page = NULL;
 618         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 619 }
 620
 621 static int truncate_dnode(struct dnode_of_data *dn)
 622 {
 623         struct page *page;
 624
 625         if (dn->nid == 0)
 626                 return 1;
 627
 628         /* get direct node */
 629         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 630         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 631                 return 1;
 632         else if (IS_ERR(page))
 633                 return PTR_ERR(page);
 634
 635         /* Make dnode_of_data for parameter */
 636         dn->node_page = page;
 637         dn->ofs_in_node = 0;
 638         truncate_data_blocks(dn);
 639         truncate_node(dn);
 640         return 1;
 641 }
 642
 643 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 644                                                 int ofs, int depth)
 645 {
 646         struct dnode_of_data rdn = *dn;
 647         struct page *page;
 648         struct f2fs_node *rn;
 649         nid_t child_nid;
 650         unsigned int child_nofs;
 651         int freed = 0;
 652         int i, ret;
 653
 654         if (dn->nid == 0)
 655                 return NIDS_PER_BLOCK + 1;
 656
 657         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 658
 659         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 660         if (IS_ERR(page)) {
 661                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 662                 return PTR_ERR(page);
 663         }
 664
 665         rn = F2FS_NODE(page);
 666         if (depth < 3) {
 667                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 668                         child_nid = le32_to_cpu(rn->in.nid[i]);
 669                         if (child_nid == 0)
 670                                 continue;
 671                         rdn.nid = child_nid;
 672                         ret = truncate_dnode(&rdn);
 673                         if (ret < 0)
 674                                 goto out_err;
 675                         set_nid(page, i, 0, false);
 676                 }
 677         } else {
 678                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 679                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 680                         child_nid = le32_to_cpu(rn->in.nid[i]);
 681                         if (child_nid == 0) {
 682                                 child_nofs += NIDS_PER_BLOCK + 1;
 683                                 continue;
 684                         }
 685                         rdn.nid = child_nid;
 686                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 687                         if (ret == (NIDS_PER_BLOCK + 1)) {
 688                                 set_nid(page, i, 0, false);
 689                                 child_nofs += ret;
 690                         } else if (ret < 0 && ret != -ENOENT) {
 691                                 goto out_err;
 692                         }
 693                 }
 694                 freed = child_nofs;
 695         }
 696
 697         if (!ofs) {
 698                 /* remove current indirect node */
 699                 dn->node_page = page;
 700                 truncate_node(dn);
 701                 freed++;
 702         } else {
 703                 f2fs_put_page(page, 1);
 704         }
 705         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 706         return freed;
 707
 708 out_err:
 709         f2fs_put_page(page, 1);
 710         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 711         return ret;
 712 }
 713
 714 static int truncate_partial_nodes(struct dnode_of_data *dn,
 715                         struct f2fs_inode *ri, int *offset, int depth)
 716 {
 717         struct page *pages[2];
 718         nid_t nid[3];
 719         nid_t child_nid;
 720         int err = 0;
 721         int i;
 722         int idx = depth - 2;
 723
 724         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 725         if (!nid[0])
 726                 return 0;
 727
 728         /* get indirect nodes in the path */
 729         for (i = 0; i < idx + 1; i++) {
 730                 /* reference count'll be increased */
 731                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 732                 if (IS_ERR(pages[i])) {
 733                         err = PTR_ERR(pages[i]);
 734                         idx = i - 1;
 735                         goto fail;
 736                 }
 737                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 738         }
 739
 740         /* free direct nodes linked to a partial indirect node */
 741         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 742                 child_nid = get_nid(pages[idx], i, false);
 743                 if (!child_nid)
 744                         continue;
 745                 dn->nid = child_nid;
 746                 err = truncate_dnode(dn);
 747                 if (err < 0)
 748                         goto fail;
 749                 set_nid(pages[idx], i, 0, false);
 750         }
 751
 752         if (offset[idx + 1] == 0) {
 753                 dn->node_page = pages[idx];
 754                 dn->nid = nid[idx];
 755                 truncate_node(dn);
 756         } else {
 757                 f2fs_put_page(pages[idx], 1);
 758         }
 759         offset[idx]++;
 760         offset[idx + 1] = 0;
 761         idx--;
 762 fail:
 763         for (i = idx; i >= 0; i--)
 764                 f2fs_put_page(pages[i], 1);
 765
 766         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 767
 768         return err;
 769 }
 770
 771 /*
 772  * All the block addresses of data and nodes should be nullified.
 773  */
 774 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 775 {
 776         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 777         int err = 0, cont = 1;
 778         int level, offset[4], noffset[4];
 779         unsigned int nofs = 0;
 780         struct f2fs_inode *ri;
 781         struct dnode_of_data dn;
 782         struct page *page;
 783
 784         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 785
 786         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 787 restart:
 788         page = get_node_page(sbi, inode->i_ino);
 789         if (IS_ERR(page)) {
 790                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 791                 return PTR_ERR(page);
 792         }
 793
 794         set_new_dnode(&dn, inode, page, NULL, 0);
 795         unlock_page(page);
 796
 797         ri = F2FS_INODE(page);
 798         switch (level) {
 799         case 0:
 800         case 1:
 801                 nofs = noffset[1];
 802                 break;
 803         case 2:
 804                 nofs = noffset[1];
 805                 if (!offset[level - 1])
 806                         goto skip_partial;
 807                 err = truncate_partial_nodes(&dn, ri, offset, level);
 808                 if (err < 0 && err != -ENOENT)
 809                         goto fail;
 810                 nofs += 1 + NIDS_PER_BLOCK;
 811                 break;
 812         case 3:
 813                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 814                 if (!offset[level - 1])
 815                         goto skip_partial;
 816                 err = truncate_partial_nodes(&dn, ri, offset, level);
 817                 if (err < 0 && err != -ENOENT)
 818                         goto fail;
 819                 break;
 820         default:
 821                 BUG();
 822         }
 823
 824 skip_partial:
 825         while (cont) {
 826                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 827                 switch (offset[0]) {
 828                 case NODE_DIR1_BLOCK:
 829                 case NODE_DIR2_BLOCK:
 830                         err = truncate_dnode(&dn);
 831                         break;
 832
 833                 case NODE_IND1_BLOCK:
 834                 case NODE_IND2_BLOCK:
 835                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 836                         break;
 837
 838                 case NODE_DIND_BLOCK:
 839                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 840                         cont = 0;
 841                         break;
 842
 843                 default:
 844                         BUG();
 845                 }
 846                 if (err < 0 && err != -ENOENT)
 847                         goto fail;
 848                 if (offset[1] == 0 &&
 849                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 850                         lock_page(page);
 851                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 852                                 f2fs_put_page(page, 1);
 853                                 goto restart;
 854                         }
 855                         f2fs_wait_on_page_writeback(page, NODE);
 856                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 857                         set_page_dirty(page);
 858                         unlock_page(page);
 859                 }
 860                 offset[1] = 0;
 861                 offset[0]++;
 862                 nofs += err;
 863         }
 864 fail:
 865         f2fs_put_page(page, 0);
 866         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 867         return err > 0 ? 0 : err;
 868 }
 869
 870 int truncate_xattr_node(struct inode *inode, struct page *page)
 871 {
 872         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 873         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 874         struct dnode_of_data dn;
 875         struct page *npage;
 876
 877         if (!nid)
 878                 return 0;
 879
 880         npage = get_node_page(sbi, nid);
 881         if (IS_ERR(npage))
 882                 return PTR_ERR(npage);
 883
 884         F2FS_I(inode)->i_xattr_nid = 0;
 885
 886         /* need to do checkpoint during fsync */
 887         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 888
 889         set_new_dnode(&dn, inode, page, npage, nid);
 890
 891         if (page)
 892                 dn.inode_page_locked = true;
 893         truncate_node(&dn);
 894         return 0;
 895 }
 896
 897 /*
 898  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 899  * f2fs_unlock_op().
 900  */
 901 void remove_inode_page(struct inode *inode)
 902 {
 903         struct dnode_of_data dn;
 904
 905         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 906         if (get_dnode_of_data(&dn, 0, LOOKUP_NODE))
 907                 return;
 908
 909         if (truncate_xattr_node(inode, dn.inode_page)) {
 910                 f2fs_put_dnode(&dn);
 911                 return;
 912         }
 913
 914         /* remove potential inline_data blocks */
 915         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 916                                 S_ISLNK(inode->i_mode))
 917                 truncate_data_blocks_range(&dn, 1);
 918
 919         /* 0 is possible, after f2fs_new_inode() has failed */
 920         f2fs_bug_on(F2FS_I_SB(inode),
 921                         inode->i_blocks != 0 && inode->i_blocks != 1);
 922
 923         /* will put inode & node pages */
 924         truncate_node(&dn);
 925 }
 926
 927 struct page *new_inode_page(struct inode *inode)
 928 {
 929         struct dnode_of_data dn;
 930
 931         /* allocate inode page for new inode */
 932         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 933
 934         /* caller should f2fs_put_page(page, 1); */
 935         return new_node_page(&dn, 0, NULL);
 936 }
 937
 938 struct page *new_node_page(struct dnode_of_data *dn,
 939                                 unsigned int ofs, struct page *ipage)
 940 {
 941         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 942         struct node_info old_ni, new_ni;
 943         struct page *page;
 944         int err;
 945
 946         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 947                 return ERR_PTR(-EPERM);
 948
 949         page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
 950         if (!page)
 951                 return ERR_PTR(-ENOMEM);
 952
 953         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 954                 err = -ENOSPC;
 955                 goto fail;
 956         }
 957
 958         get_node_info(sbi, dn->nid, &old_ni);
 959
 960         /* Reinitialize old_ni with new node page */
 961         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
 962         new_ni = old_ni;
 963         new_ni.ino = dn->inode->i_ino;
 964         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
 965
 966         f2fs_wait_on_page_writeback(page, NODE);
 967         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 968         set_cold_node(dn->inode, page);
 969         SetPageUptodate(page);
 970         set_page_dirty(page);
 971
 972         if (f2fs_has_xattr_block(ofs))
 973                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 974
 975         dn->node_page = page;
 976         if (ipage)
 977                 update_inode(dn->inode, ipage);
 978         else
 979                 sync_inode_page(dn);
 980         if (ofs == 0)
 981                 inc_valid_inode_count(sbi);
 982
 983         return page;
 984
 985 fail:
 986         clear_node_page_dirty(page);
 987         f2fs_put_page(page, 1);
 988         return ERR_PTR(err);
 989 }
 990
 991 /*
 992  * Caller should do after getting the following values.
 993  * 0: f2fs_put_page(page, 0)
 994  * LOCKED_PAGE: f2fs_put_page(page, 1)
 995  * error: nothing
 996  */
 997 static int read_node_page(struct page *page, int rw)
 998 {
 999         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1000         struct node_info ni;
1001         struct f2fs_io_info fio = {
1002                 .sbi = sbi,
1003                 .type = NODE,
1004                 .rw = rw,
1005                 .page = page,
1006                 .encrypted_page = NULL,
1007         };
1008
1009         get_node_info(sbi, page->index, &ni);
1010
1011         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1012                 ClearPageUptodate(page);
1013                 f2fs_put_page(page, 1);
1014                 return -ENOENT;
1015         }
1016
1017         if (PageUptodate(page))
1018                 return LOCKED_PAGE;
1019
1020         fio.blk_addr = ni.blk_addr;
1021         return f2fs_submit_page_bio(&fio);
1022 }
1023
1024 /*
1025  * Readahead a node page
1026  */
1027 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1028 {
1029         struct page *apage;
1030         int err;
1031
1032         apage = find_get_page(NODE_MAPPING(sbi), nid);
1033         if (apage && PageUptodate(apage)) {
1034                 f2fs_put_page(apage, 0);
1035                 return;
1036         }
1037         f2fs_put_page(apage, 0);
1038
1039         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1040         if (!apage)
1041                 return;
1042
1043         err = read_node_page(apage, READA);
1044         if (err == 0)
1045                 f2fs_put_page(apage, 0);
1046         else if (err == LOCKED_PAGE)
1047                 f2fs_put_page(apage, 1);
1048 }
1049
1050 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1051 {
1052         struct page *page;
1053         int err;
1054 repeat:
1055         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1056         if (!page)
1057                 return ERR_PTR(-ENOMEM);
1058
1059         err = read_node_page(page, READ_SYNC);
1060         if (err < 0)
1061                 return ERR_PTR(err);
1062         else if (err != LOCKED_PAGE)
1063                 lock_page(page);
1064
1065         if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
1066                 ClearPageUptodate(page);
1067                 f2fs_put_page(page, 1);
1068                 return ERR_PTR(-EIO);
1069         }
1070         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1071                 f2fs_put_page(page, 1);
1072                 goto repeat;
1073         }
1074         return page;
1075 }
1076
1077 /*
1078  * Return a locked page for the desired node page.
1079  * And, readahead MAX_RA_NODE number of node pages.
1080  */
1081 struct page *get_node_page_ra(struct page *parent, int start)
1082 {
1083         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1084         struct blk_plug plug;
1085         struct page *page;
1086         int err, i, end;
1087         nid_t nid;
1088
1089         /* First, try getting the desired direct node. */
1090         nid = get_nid(parent, start, false);
1091         if (!nid)
1092                 return ERR_PTR(-ENOENT);
1093 repeat:
1094         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1095         if (!page)
1096                 return ERR_PTR(-ENOMEM);
1097
1098         err = read_node_page(page, READ_SYNC);
1099         if (err < 0)
1100                 return ERR_PTR(err);
1101         else if (err == LOCKED_PAGE)
1102                 goto page_hit;
1103
1104         blk_start_plug(&plug);
1105
1106         /* Then, try readahead for siblings of the desired node */
1107         end = start + MAX_RA_NODE;
1108         end = min(end, NIDS_PER_BLOCK);
1109         for (i = start + 1; i < end; i++) {
1110                 nid = get_nid(parent, i, false);
1111                 if (!nid)
1112                         continue;
1113                 ra_node_page(sbi, nid);
1114         }
1115
1116         blk_finish_plug(&plug);
1117
1118         lock_page(page);
1119         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1120                 f2fs_put_page(page, 1);
1121                 goto repeat;
1122         }
1123 page_hit:
1124         if (unlikely(!PageUptodate(page))) {
1125                 f2fs_put_page(page, 1);
1126                 return ERR_PTR(-EIO);
1127         }
1128         return page;
1129 }
1130
1131 void sync_inode_page(struct dnode_of_data *dn)
1132 {
1133         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1134                 update_inode(dn->inode, dn->node_page);
1135         } else if (dn->inode_page) {
1136                 if (!dn->inode_page_locked)
1137                         lock_page(dn->inode_page);
1138                 update_inode(dn->inode, dn->inode_page);
1139                 if (!dn->inode_page_locked)
1140                         unlock_page(dn->inode_page);
1141         } else {
1142                 update_inode_page(dn->inode);
1143         }
1144 }
1145
1146 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1147                                         struct writeback_control *wbc)
1148 {
1149         pgoff_t index, end;
1150         struct pagevec pvec;
1151         int step = ino ? 2 : 0;
1152         int nwritten = 0, wrote = 0;
1153
1154         pagevec_init(&pvec, 0);
1155
1156 next_step:
1157         index = 0;
1158         end = LONG_MAX;
1159
1160         while (index <= end) {
1161                 int i, nr_pages;
1162                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1163                                 PAGECACHE_TAG_DIRTY,
1164                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1165                 if (nr_pages == 0)
1166                         break;
1167
1168                 for (i = 0; i < nr_pages; i++) {
1169                         struct page *page = pvec.pages[i];
1170
1171                         /*
1172                          * flushing sequence with step:
1173                          * 0. indirect nodes
1174                          * 1. dentry dnodes
1175                          * 2. file dnodes
1176                          */
1177                         if (step == 0 && IS_DNODE(page))
1178                                 continue;
1179                         if (step == 1 && (!IS_DNODE(page) ||
1180                                                 is_cold_node(page)))
1181                                 continue;
1182                         if (step == 2 && (!IS_DNODE(page) ||
1183                                                 !is_cold_node(page)))
1184                                 continue;
1185
1186                         /*
1187                          * If an fsync mode,
1188                          * we should not skip writing node pages.
1189                          */
1190                         if (ino && ino_of_node(page) == ino)
1191                                 lock_page(page);
1192                         else if (!trylock_page(page))
1193                                 continue;
1194
1195                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1196 continue_unlock:
1197                                 unlock_page(page);
1198                                 continue;
1199                         }
1200                         if (ino && ino_of_node(page) != ino)
1201                                 goto continue_unlock;
1202
1203                         if (!PageDirty(page)) {
1204                                 /* someone wrote it for us */
1205                                 goto continue_unlock;
1206                         }
1207
1208                         if (!clear_page_dirty_for_io(page))
1209                                 goto continue_unlock;
1210
1211                         /* called by fsync() */
1212                         if (ino && IS_DNODE(page)) {
1213                                 set_fsync_mark(page, 1);
1214                                 if (IS_INODE(page))
1215                                         set_dentry_mark(page,
1216                                                 need_dentry_mark(sbi, ino));
1217                                 nwritten++;
1218                         } else {
1219                                 set_fsync_mark(page, 0);
1220                                 set_dentry_mark(page, 0);
1221                         }
1222
1223                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1224                                 unlock_page(page);
1225                         else
1226                                 wrote++;
1227
1228                         if (--wbc->nr_to_write == 0)
1229                                 break;
1230                 }
1231                 pagevec_release(&pvec);
1232                 cond_resched();
1233
1234                 if (wbc->nr_to_write == 0) {
1235                         step = 2;
1236                         break;
1237                 }
1238         }
1239
1240         if (step < 2) {
1241                 step++;
1242                 goto next_step;
1243         }
1244
1245         if (wrote)
1246                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1247         return nwritten;
1248 }
1249
1250 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1251 {
1252         pgoff_t index = 0, end = LONG_MAX;
1253         struct pagevec pvec;
1254         int ret2 = 0, ret = 0;
1255
1256         pagevec_init(&pvec, 0);
1257
1258         while (index <= end) {
1259                 int i, nr_pages;
1260                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1261                                 PAGECACHE_TAG_WRITEBACK,
1262                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1263                 if (nr_pages == 0)
1264                         break;
1265
1266                 for (i = 0; i < nr_pages; i++) {
1267                         struct page *page = pvec.pages[i];
1268
1269                         /* until radix tree lookup accepts end_index */
1270                         if (unlikely(page->index > end))
1271                                 continue;
1272
1273                         if (ino && ino_of_node(page) == ino) {
1274                                 f2fs_wait_on_page_writeback(page, NODE);
1275                                 if (TestClearPageError(page))
1276                                         ret = -EIO;
1277                         }
1278                 }
1279                 pagevec_release(&pvec);
1280                 cond_resched();
1281         }
1282
1283         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1284                 ret2 = -ENOSPC;
1285         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1286                 ret2 = -EIO;
1287         if (!ret)
1288                 ret = ret2;
1289         return ret;
1290 }
1291
1292 static int f2fs_write_node_page(struct page *page,
1293                                 struct writeback_control *wbc)
1294 {
1295         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1296         nid_t nid;
1297         struct node_info ni;
1298         struct f2fs_io_info fio = {
1299                 .sbi = sbi,
1300                 .type = NODE,
1301                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1302                 .page = page,
1303                 .encrypted_page = NULL,
1304         };
1305
1306         trace_f2fs_writepage(page, NODE);
1307
1308         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1309                 goto redirty_out;
1310         if (unlikely(f2fs_cp_error(sbi)))
1311                 goto redirty_out;
1312
1313         f2fs_wait_on_page_writeback(page, NODE);
1314
1315         /* get old block addr of this node page */
1316         nid = nid_of_node(page);
1317         f2fs_bug_on(sbi, page->index != nid);
1318
1319         get_node_info(sbi, nid, &ni);
1320
1321         /* This page is already truncated */
1322         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1323                 ClearPageUptodate(page);
1324                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1325                 unlock_page(page);
1326                 return 0;
1327         }
1328
1329         if (wbc->for_reclaim) {
1330                 if (!down_read_trylock(&sbi->node_write))
1331                         goto redirty_out;
1332         } else {
1333                 down_read(&sbi->node_write);
1334         }
1335
1336         set_page_writeback(page);
1337         fio.blk_addr = ni.blk_addr;
1338         write_node_page(nid, &fio);
1339         set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1340         dec_page_count(sbi, F2FS_DIRTY_NODES);
1341         up_read(&sbi->node_write);
1342         unlock_page(page);
1343
1344         if (wbc->for_reclaim)
1345                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1346
1347         return 0;
1348
1349 redirty_out:
1350         redirty_page_for_writepage(wbc, page);
1351         return AOP_WRITEPAGE_ACTIVATE;
1352 }
1353
1354 static int f2fs_write_node_pages(struct address_space *mapping,
1355                             struct writeback_control *wbc)
1356 {
1357         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1358         long diff;
1359
1360         trace_f2fs_writepages(mapping->host, wbc, NODE);
1361
1362         /* balancing f2fs's metadata in background */
1363         f2fs_balance_fs_bg(sbi);
1364
1365         /* collect a number of dirty node pages and write together */
1366         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1367                 goto skip_write;
1368
1369         diff = nr_pages_to_write(sbi, NODE, wbc);
1370         wbc->sync_mode = WB_SYNC_NONE;
1371         sync_node_pages(sbi, 0, wbc);
1372         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1373         return 0;
1374
1375 skip_write:
1376         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1377         return 0;
1378 }
1379
1380 static int f2fs_set_node_page_dirty(struct page *page)
1381 {
1382         trace_f2fs_set_page_dirty(page, NODE);
1383
1384         SetPageUptodate(page);
1385         if (!PageDirty(page)) {
1386                 __set_page_dirty_nobuffers(page);
1387                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1388                 SetPagePrivate(page);
1389                 f2fs_trace_pid(page);
1390                 return 1;
1391         }
1392         return 0;
1393 }
1394
1395 /*
1396  * Structure of the f2fs node operations
1397  */
1398 const struct address_space_operations f2fs_node_aops = {
1399         .writepage      = f2fs_write_node_page,
1400         .writepages     = f2fs_write_node_pages,
1401         .set_page_dirty = f2fs_set_node_page_dirty,
1402         .invalidatepage = f2fs_invalidate_page,
1403         .releasepage    = f2fs_release_page,
1404 };
1405
1406 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1407                                                 nid_t n)
1408 {
1409         return radix_tree_lookup(&nm_i->free_nid_root, n);
1410 }
1411
1412 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1413                                                 struct free_nid *i)
1414 {
1415         list_del(&i->list);
1416         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1417 }
1418
1419 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1420 {
1421         struct f2fs_nm_info *nm_i = NM_I(sbi);
1422         struct free_nid *i;
1423         struct nat_entry *ne;
1424         bool allocated = false;
1425
1426         if (!available_free_memory(sbi, FREE_NIDS))
1427                 return -1;
1428
1429         /* 0 nid should not be used */
1430         if (unlikely(nid == 0))
1431                 return 0;
1432
1433         if (build) {
1434                 /* do not add allocated nids */
1435                 down_read(&nm_i->nat_tree_lock);
1436                 ne = __lookup_nat_cache(nm_i, nid);
1437                 if (ne &&
1438                         (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1439                                 nat_get_blkaddr(ne) != NULL_ADDR))
1440                         allocated = true;
1441                 up_read(&nm_i->nat_tree_lock);
1442                 if (allocated)
1443                         return 0;
1444         }
1445
1446         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1447         i->nid = nid;
1448         i->state = NID_NEW;
1449
1450         if (radix_tree_preload(GFP_NOFS)) {
1451                 kmem_cache_free(free_nid_slab, i);
1452                 return 0;
1453         }
1454
1455         spin_lock(&nm_i->free_nid_list_lock);
1456         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1457                 spin_unlock(&nm_i->free_nid_list_lock);
1458                 radix_tree_preload_end();
1459                 kmem_cache_free(free_nid_slab, i);
1460                 return 0;
1461         }
1462         list_add_tail(&i->list, &nm_i->free_nid_list);
1463         nm_i->fcnt++;
1464         spin_unlock(&nm_i->free_nid_list_lock);
1465         radix_tree_preload_end();
1466         return 1;
1467 }
1468
1469 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1470 {
1471         struct free_nid *i;
1472         bool need_free = false;
1473
1474         spin_lock(&nm_i->free_nid_list_lock);
1475         i = __lookup_free_nid_list(nm_i, nid);
1476         if (i && i->state == NID_NEW) {
1477                 __del_from_free_nid_list(nm_i, i);
1478                 nm_i->fcnt--;
1479                 need_free = true;
1480         }
1481         spin_unlock(&nm_i->free_nid_list_lock);
1482
1483         if (need_free)
1484                 kmem_cache_free(free_nid_slab, i);
1485 }
1486
1487 static void scan_nat_page(struct f2fs_sb_info *sbi,
1488                         struct page *nat_page, nid_t start_nid)
1489 {
1490         struct f2fs_nm_info *nm_i = NM_I(sbi);
1491         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1492         block_t blk_addr;
1493         int i;
1494
1495         i = start_nid % NAT_ENTRY_PER_BLOCK;
1496
1497         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1498
1499                 if (unlikely(start_nid >= nm_i->max_nid))
1500                         break;
1501
1502                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1503                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1504                 if (blk_addr == NULL_ADDR) {
1505                         if (add_free_nid(sbi, start_nid, true) < 0)
1506                                 break;
1507                 }
1508         }
1509 }
1510
1511 static void build_free_nids(struct f2fs_sb_info *sbi)
1512 {
1513         struct f2fs_nm_info *nm_i = NM_I(sbi);
1514         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1515         struct f2fs_summary_block *sum = curseg->sum_blk;
1516         int i = 0;
1517         nid_t nid = nm_i->next_scan_nid;
1518
1519         /* Enough entries */
1520         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1521                 return;
1522
1523         /* readahead nat pages to be scanned */
1524         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1525
1526         while (1) {
1527                 struct page *page = get_current_nat_page(sbi, nid);
1528
1529                 scan_nat_page(sbi, page, nid);
1530                 f2fs_put_page(page, 1);
1531
1532                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1533                 if (unlikely(nid >= nm_i->max_nid))
1534                         nid = 0;
1535
1536                 if (i++ == FREE_NID_PAGES)
1537                         break;
1538         }
1539
1540         /* go to the next free nat pages to find free nids abundantly */
1541         nm_i->next_scan_nid = nid;
1542
1543         /* find free nids from current sum_pages */
1544         mutex_lock(&curseg->curseg_mutex);
1545         for (i = 0; i < nats_in_cursum(sum); i++) {
1546                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1547                 nid = le32_to_cpu(nid_in_journal(sum, i));
1548                 if (addr == NULL_ADDR)
1549                         add_free_nid(sbi, nid, true);
1550                 else
1551                         remove_free_nid(nm_i, nid);
1552         }
1553         mutex_unlock(&curseg->curseg_mutex);
1554 }
1555
1556 /*
1557  * If this function returns success, caller can obtain a new nid
1558  * from second parameter of this function.
1559  * The returned nid could be used ino as well as nid when inode is created.
1560  */
1561 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1562 {
1563         struct f2fs_nm_info *nm_i = NM_I(sbi);
1564         struct free_nid *i = NULL;
1565 retry:
1566         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1567                 return false;
1568
1569         spin_lock(&nm_i->free_nid_list_lock);
1570
1571         /* We should not use stale free nids created by build_free_nids */
1572         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1573                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1574                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1575                         if (i->state == NID_NEW)
1576                                 break;
1577
1578                 f2fs_bug_on(sbi, i->state != NID_NEW);
1579                 *nid = i->nid;
1580                 i->state = NID_ALLOC;
1581                 nm_i->fcnt--;
1582                 spin_unlock(&nm_i->free_nid_list_lock);
1583                 return true;
1584         }
1585         spin_unlock(&nm_i->free_nid_list_lock);
1586
1587         /* Let's scan nat pages and its caches to get free nids */
1588         mutex_lock(&nm_i->build_lock);
1589         build_free_nids(sbi);
1590         mutex_unlock(&nm_i->build_lock);
1591         goto retry;
1592 }
1593
1594 /*
1595  * alloc_nid() should be called prior to this function.
1596  */
1597 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1598 {
1599         struct f2fs_nm_info *nm_i = NM_I(sbi);
1600         struct free_nid *i;
1601
1602         spin_lock(&nm_i->free_nid_list_lock);
1603         i = __lookup_free_nid_list(nm_i, nid);
1604         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1605         __del_from_free_nid_list(nm_i, i);
1606         spin_unlock(&nm_i->free_nid_list_lock);
1607
1608         kmem_cache_free(free_nid_slab, i);
1609 }
1610
1611 /*
1612  * alloc_nid() should be called prior to this function.
1613  */
1614 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1615 {
1616         struct f2fs_nm_info *nm_i = NM_I(sbi);
1617         struct free_nid *i;
1618         bool need_free = false;
1619
1620         if (!nid)
1621                 return;
1622
1623         spin_lock(&nm_i->free_nid_list_lock);
1624         i = __lookup_free_nid_list(nm_i, nid);
1625         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1626         if (!available_free_memory(sbi, FREE_NIDS)) {
1627                 __del_from_free_nid_list(nm_i, i);
1628                 need_free = true;
1629         } else {
1630                 i->state = NID_NEW;
1631                 nm_i->fcnt++;
1632         }
1633         spin_unlock(&nm_i->free_nid_list_lock);
1634
1635         if (need_free)
1636                 kmem_cache_free(free_nid_slab, i);
1637 }
1638
1639 void recover_inline_xattr(struct inode *inode, struct page *page)
1640 {
1641         void *src_addr, *dst_addr;
1642         size_t inline_size;
1643         struct page *ipage;
1644         struct f2fs_inode *ri;
1645
1646         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1647         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1648
1649         ri = F2FS_INODE(page);
1650         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1651                 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1652                 goto update_inode;
1653         }
1654
1655         dst_addr = inline_xattr_addr(ipage);
1656         src_addr = inline_xattr_addr(page);
1657         inline_size = inline_xattr_size(inode);
1658
1659         f2fs_wait_on_page_writeback(ipage, NODE);
1660         memcpy(dst_addr, src_addr, inline_size);
1661 update_inode:
1662         update_inode(inode, ipage);
1663         f2fs_put_page(ipage, 1);
1664 }
1665
1666 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1667 {
1668         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1669         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1670         nid_t new_xnid = nid_of_node(page);
1671         struct node_info ni;
1672
1673         /* 1: invalidate the previous xattr nid */
1674         if (!prev_xnid)
1675                 goto recover_xnid;
1676
1677         /* Deallocate node address */
1678         get_node_info(sbi, prev_xnid, &ni);
1679         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1680         invalidate_blocks(sbi, ni.blk_addr);
1681         dec_valid_node_count(sbi, inode);
1682         set_node_addr(sbi, &ni, NULL_ADDR, false);
1683
1684 recover_xnid:
1685         /* 2: allocate new xattr nid */
1686         if (unlikely(!inc_valid_node_count(sbi, inode)))
1687                 f2fs_bug_on(sbi, 1);
1688
1689         remove_free_nid(NM_I(sbi), new_xnid);
1690         get_node_info(sbi, new_xnid, &ni);
1691         ni.ino = inode->i_ino;
1692         set_node_addr(sbi, &ni, NEW_ADDR, false);
1693         F2FS_I(inode)->i_xattr_nid = new_xnid;
1694
1695         /* 3: update xattr blkaddr */
1696         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1697         set_node_addr(sbi, &ni, blkaddr, false);
1698
1699         update_inode_page(inode);
1700 }
1701
1702 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1703 {
1704         struct f2fs_inode *src, *dst;
1705         nid_t ino = ino_of_node(page);
1706         struct node_info old_ni, new_ni;
1707         struct page *ipage;
1708
1709         get_node_info(sbi, ino, &old_ni);
1710
1711         if (unlikely(old_ni.blk_addr != NULL_ADDR))
1712                 return -EINVAL;
1713
1714         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1715         if (!ipage)
1716                 return -ENOMEM;
1717
1718         /* Should not use this inode from free nid list */
1719         remove_free_nid(NM_I(sbi), ino);
1720
1721         SetPageUptodate(ipage);
1722         fill_node_footer(ipage, ino, ino, 0, true);
1723
1724         src = F2FS_INODE(page);
1725         dst = F2FS_INODE(ipage);
1726
1727         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1728         dst->i_size = 0;
1729         dst->i_blocks = cpu_to_le64(1);
1730         dst->i_links = cpu_to_le32(1);
1731         dst->i_xattr_nid = 0;
1732         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1733
1734         new_ni = old_ni;
1735         new_ni.ino = ino;
1736
1737         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1738                 WARN_ON(1);
1739         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1740         inc_valid_inode_count(sbi);
1741         set_page_dirty(ipage);
1742         f2fs_put_page(ipage, 1);
1743         return 0;
1744 }
1745
1746 int restore_node_summary(struct f2fs_sb_info *sbi,
1747                         unsigned int segno, struct f2fs_summary_block *sum)
1748 {
1749         struct f2fs_node *rn;
1750         struct f2fs_summary *sum_entry;
1751         block_t addr;
1752         int bio_blocks = MAX_BIO_BLOCKS(sbi);
1753         int i, idx, last_offset, nrpages;
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; i < last_offset; i += nrpages, addr += nrpages) {
1761                 nrpages = min(last_offset - i, bio_blocks);
1762
1763                 /* readahead node pages */
1764                 ra_meta_pages(sbi, addr, nrpages, META_POR);
1765
1766                 for (idx = addr; idx < addr + nrpages; idx++) {
1767                         struct page *page = get_meta_page(sbi, idx);
1768
1769                         rn = F2FS_NODE(page);
1770                         sum_entry->nid = rn->footer.nid;
1771                         sum_entry->version = 0;
1772                         sum_entry->ofs_in_node = 0;
1773                         sum_entry++;
1774                         f2fs_put_page(page, 1);
1775                 }
1776
1777                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
1778                                                         addr + nrpages);
1779         }
1780         return 0;
1781 }
1782
1783 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1784 {
1785         struct f2fs_nm_info *nm_i = NM_I(sbi);
1786         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1787         struct f2fs_summary_block *sum = curseg->sum_blk;
1788         int i;
1789
1790         mutex_lock(&curseg->curseg_mutex);
1791         for (i = 0; i < nats_in_cursum(sum); i++) {
1792                 struct nat_entry *ne;
1793                 struct f2fs_nat_entry raw_ne;
1794                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1795
1796                 raw_ne = nat_in_journal(sum, i);
1797
1798                 down_write(&nm_i->nat_tree_lock);
1799                 ne = __lookup_nat_cache(nm_i, nid);
1800                 if (!ne) {
1801                         ne = grab_nat_entry(nm_i, nid);
1802                         node_info_from_raw_nat(&ne->ni, &raw_ne);
1803                 }
1804                 __set_nat_cache_dirty(nm_i, ne);
1805                 up_write(&nm_i->nat_tree_lock);
1806         }
1807         update_nats_in_cursum(sum, -i);
1808         mutex_unlock(&curseg->curseg_mutex);
1809 }
1810
1811 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1812                                                 struct list_head *head, int max)
1813 {
1814         struct nat_entry_set *cur;
1815
1816         if (nes->entry_cnt >= max)
1817                 goto add_out;
1818
1819         list_for_each_entry(cur, head, set_list) {
1820                 if (cur->entry_cnt >= nes->entry_cnt) {
1821                         list_add(&nes->set_list, cur->set_list.prev);
1822                         return;
1823                 }
1824         }
1825 add_out:
1826         list_add_tail(&nes->set_list, head);
1827 }
1828
1829 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1830                                         struct nat_entry_set *set)
1831 {
1832         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1833         struct f2fs_summary_block *sum = curseg->sum_blk;
1834         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1835         bool to_journal = true;
1836         struct f2fs_nat_block *nat_blk;
1837         struct nat_entry *ne, *cur;
1838         struct page *page = NULL;
1839         struct f2fs_nm_info *nm_i = NM_I(sbi);
1840
1841         /*
1842          * there are two steps to flush nat entries:
1843          * #1, flush nat entries to journal in current hot data summary block.
1844          * #2, flush nat entries to nat page.
1845          */
1846         if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1847                 to_journal = false;
1848
1849         if (to_journal) {
1850                 mutex_lock(&curseg->curseg_mutex);
1851         } else {
1852                 page = get_next_nat_page(sbi, start_nid);
1853                 nat_blk = page_address(page);
1854                 f2fs_bug_on(sbi, !nat_blk);
1855         }
1856
1857         /* flush dirty nats in nat entry set */
1858         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1859                 struct f2fs_nat_entry *raw_ne;
1860                 nid_t nid = nat_get_nid(ne);
1861                 int offset;
1862
1863                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1864                         continue;
1865
1866                 if (to_journal) {
1867                         offset = lookup_journal_in_cursum(sum,
1868                                                         NAT_JOURNAL, nid, 1);
1869                         f2fs_bug_on(sbi, offset < 0);
1870                         raw_ne = &nat_in_journal(sum, offset);
1871                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1872                 } else {
1873                         raw_ne = &nat_blk->entries[nid - start_nid];
1874                 }
1875                 raw_nat_from_node_info(raw_ne, &ne->ni);
1876
1877                 down_write(&NM_I(sbi)->nat_tree_lock);
1878                 nat_reset_flag(ne);
1879                 __clear_nat_cache_dirty(NM_I(sbi), ne);
1880                 up_write(&NM_I(sbi)->nat_tree_lock);
1881
1882                 if (nat_get_blkaddr(ne) == NULL_ADDR)
1883                         add_free_nid(sbi, nid, false);
1884         }
1885
1886         if (to_journal)
1887                 mutex_unlock(&curseg->curseg_mutex);
1888         else
1889                 f2fs_put_page(page, 1);
1890
1891         f2fs_bug_on(sbi, set->entry_cnt);
1892
1893         down_write(&nm_i->nat_tree_lock);
1894         radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1895         up_write(&nm_i->nat_tree_lock);
1896         kmem_cache_free(nat_entry_set_slab, set);
1897 }
1898
1899 /*
1900  * This function is called during the checkpointing process.
1901  */
1902 void flush_nat_entries(struct f2fs_sb_info *sbi)
1903 {
1904         struct f2fs_nm_info *nm_i = NM_I(sbi);
1905         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1906         struct f2fs_summary_block *sum = curseg->sum_blk;
1907         struct nat_entry_set *setvec[SETVEC_SIZE];
1908         struct nat_entry_set *set, *tmp;
1909         unsigned int found;
1910         nid_t set_idx = 0;
1911         LIST_HEAD(sets);
1912
1913         if (!nm_i->dirty_nat_cnt)
1914                 return;
1915         /*
1916          * if there are no enough space in journal to store dirty nat
1917          * entries, remove all entries from journal and merge them
1918          * into nat entry set.
1919          */
1920         if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1921                 remove_nats_in_journal(sbi);
1922
1923         down_write(&nm_i->nat_tree_lock);
1924         while ((found = __gang_lookup_nat_set(nm_i,
1925                                         set_idx, SETVEC_SIZE, setvec))) {
1926                 unsigned idx;
1927                 set_idx = setvec[found - 1]->set + 1;
1928                 for (idx = 0; idx < found; idx++)
1929                         __adjust_nat_entry_set(setvec[idx], &sets,
1930                                                         MAX_NAT_JENTRIES(sum));
1931         }
1932         up_write(&nm_i->nat_tree_lock);
1933
1934         /* flush dirty nats in nat entry set */
1935         list_for_each_entry_safe(set, tmp, &sets, set_list)
1936                 __flush_nat_entry_set(sbi, set);
1937
1938         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1939 }
1940
1941 static int init_node_manager(struct f2fs_sb_info *sbi)
1942 {
1943         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1944         struct f2fs_nm_info *nm_i = NM_I(sbi);
1945         unsigned char *version_bitmap;
1946         unsigned int nat_segs, nat_blocks;
1947
1948         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1949
1950         /* segment_count_nat includes pair segment so divide to 2. */
1951         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1952         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1953
1954         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1955
1956         /* not used nids: 0, node, meta, (and root counted as valid node) */
1957         nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
1958         nm_i->fcnt = 0;
1959         nm_i->nat_cnt = 0;
1960         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1961
1962         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1963         INIT_LIST_HEAD(&nm_i->free_nid_list);
1964         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
1965         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
1966         INIT_LIST_HEAD(&nm_i->nat_entries);
1967
1968         mutex_init(&nm_i->build_lock);
1969         spin_lock_init(&nm_i->free_nid_list_lock);
1970         init_rwsem(&nm_i->nat_tree_lock);
1971
1972         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1973         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1974         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1975         if (!version_bitmap)
1976                 return -EFAULT;
1977
1978         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1979                                         GFP_KERNEL);
1980         if (!nm_i->nat_bitmap)
1981                 return -ENOMEM;
1982         return 0;
1983 }
1984
1985 int build_node_manager(struct f2fs_sb_info *sbi)
1986 {
1987         int err;
1988
1989         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1990         if (!sbi->nm_info)
1991                 return -ENOMEM;
1992
1993         err = init_node_manager(sbi);
1994         if (err)
1995                 return err;
1996
1997         build_free_nids(sbi);
1998         return 0;
1999 }
2000
2001 void destroy_node_manager(struct f2fs_sb_info *sbi)
2002 {
2003         struct f2fs_nm_info *nm_i = NM_I(sbi);
2004         struct free_nid *i, *next_i;
2005         struct nat_entry *natvec[NATVEC_SIZE];
2006         struct nat_entry_set *setvec[SETVEC_SIZE];
2007         nid_t nid = 0;
2008         unsigned int found;
2009
2010         if (!nm_i)
2011                 return;
2012
2013         /* destroy free nid list */
2014         spin_lock(&nm_i->free_nid_list_lock);
2015         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2016                 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2017                 __del_from_free_nid_list(nm_i, i);
2018                 nm_i->fcnt--;
2019                 spin_unlock(&nm_i->free_nid_list_lock);
2020                 kmem_cache_free(free_nid_slab, i);
2021                 spin_lock(&nm_i->free_nid_list_lock);
2022         }
2023         f2fs_bug_on(sbi, nm_i->fcnt);
2024         spin_unlock(&nm_i->free_nid_list_lock);
2025
2026         /* destroy nat cache */
2027         down_write(&nm_i->nat_tree_lock);
2028         while ((found = __gang_lookup_nat_cache(nm_i,
2029                                         nid, NATVEC_SIZE, natvec))) {
2030                 unsigned idx;
2031
2032                 nid = nat_get_nid(natvec[found - 1]) + 1;
2033                 for (idx = 0; idx < found; idx++)
2034                         __del_from_nat_cache(nm_i, natvec[idx]);
2035         }
2036         f2fs_bug_on(sbi, nm_i->nat_cnt);
2037
2038         /* destroy nat set cache */
2039         nid = 0;
2040         while ((found = __gang_lookup_nat_set(nm_i,
2041                                         nid, SETVEC_SIZE, setvec))) {
2042                 unsigned idx;
2043
2044                 nid = setvec[found - 1]->set + 1;
2045                 for (idx = 0; idx < found; idx++) {
2046                         /* entry_cnt is not zero, when cp_error was occurred */
2047                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2048                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2049                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2050                 }
2051         }
2052         up_write(&nm_i->nat_tree_lock);
2053
2054         kfree(nm_i->nat_bitmap);
2055         sbi->nm_info = NULL;
2056         kfree(nm_i);
2057 }
2058
2059 int __init create_node_manager_caches(void)
2060 {
2061         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2062                         sizeof(struct nat_entry));
2063         if (!nat_entry_slab)
2064                 goto fail;
2065
2066         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2067                         sizeof(struct free_nid));
2068         if (!free_nid_slab)
2069                 goto destroy_nat_entry;
2070
2071         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2072                         sizeof(struct nat_entry_set));
2073         if (!nat_entry_set_slab)
2074                 goto destroy_free_nid;
2075         return 0;
2076
2077 destroy_free_nid:
2078         kmem_cache_destroy(free_nid_slab);
2079 destroy_nat_entry:
2080         kmem_cache_destroy(nat_entry_slab);
2081 fail:
2082         return -ENOMEM;
2083 }
2084
2085 void destroy_node_manager_caches(void)
2086 {
2087         kmem_cache_destroy(nat_entry_set_slab);
2088         kmem_cache_destroy(free_nid_slab);
2089         kmem_cache_destroy(nat_entry_slab);
2090 }