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_NOFS);
 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_NOFS);
 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         e = __lookup_nat_cache(nm_i, nid);
 265         if (!e) {
 266                 e = grab_nat_entry(nm_i, nid);
 267                 node_info_from_raw_nat(&e->ni, ne);
 268         }
 269 }
 270
 271 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 272                         block_t new_blkaddr, bool fsync_done)
 273 {
 274         struct f2fs_nm_info *nm_i = NM_I(sbi);
 275         struct nat_entry *e;
 276
 277         down_write(&nm_i->nat_tree_lock);
 278         e = __lookup_nat_cache(nm_i, ni->nid);
 279         if (!e) {
 280                 e = grab_nat_entry(nm_i, ni->nid);
 281                 copy_node_info(&e->ni, ni);
 282                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 283         } else if (new_blkaddr == NEW_ADDR) {
 284                 /*
 285                  * when nid is reallocated,
 286                  * previous nat entry can be remained in nat cache.
 287                  * So, reinitialize it with new information.
 288                  */
 289                 copy_node_info(&e->ni, ni);
 290                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 291         }
 292
 293         /* sanity check */
 294         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 295         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 296                         new_blkaddr == NULL_ADDR);
 297         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 298                         new_blkaddr == NEW_ADDR);
 299         f2fs_bug_on(sbi, is_valid_data_blkaddr(sbi, nat_get_blkaddr(e)) &&
 300                         new_blkaddr == NEW_ADDR);
 301
 302         /* increment version no as node is removed */
 303         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 304                 unsigned char version = nat_get_version(e);
 305                 nat_set_version(e, inc_node_version(version));
 306
 307                 /* in order to reuse the nid */
 308                 if (nm_i->next_scan_nid > ni->nid)
 309                         nm_i->next_scan_nid = ni->nid;
 310         }
 311
 312         /* change address */
 313         nat_set_blkaddr(e, new_blkaddr);
 314         if (!is_valid_data_blkaddr(sbi, new_blkaddr))
 315                 set_nat_flag(e, IS_CHECKPOINTED, false);
 316         __set_nat_cache_dirty(nm_i, e);
 317
 318         /* update fsync_mark if its inode nat entry is still alive */
 319         if (ni->nid != ni->ino)
 320                 e = __lookup_nat_cache(nm_i, ni->ino);
 321         if (e) {
 322                 if (fsync_done && ni->nid == ni->ino)
 323                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 324                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 325         }
 326         up_write(&nm_i->nat_tree_lock);
 327 }
 328
 329 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 330 {
 331         struct f2fs_nm_info *nm_i = NM_I(sbi);
 332         int nr = nr_shrink;
 333
 334         if (!down_write_trylock(&nm_i->nat_tree_lock))
 335                 return 0;
 336
 337         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 338                 struct nat_entry *ne;
 339                 ne = list_first_entry(&nm_i->nat_entries,
 340                                         struct nat_entry, list);
 341                 __del_from_nat_cache(nm_i, ne);
 342                 nr_shrink--;
 343         }
 344         up_write(&nm_i->nat_tree_lock);
 345         return nr - nr_shrink;
 346 }
 347
 348 /*
 349  * This function always returns success
 350  */
 351 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 352 {
 353         struct f2fs_nm_info *nm_i = NM_I(sbi);
 354         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 355         struct f2fs_summary_block *sum = curseg->sum_blk;
 356         nid_t start_nid = START_NID(nid);
 357         struct f2fs_nat_block *nat_blk;
 358         struct page *page = NULL;
 359         struct f2fs_nat_entry ne;
 360         struct nat_entry *e;
 361         int i;
 362
 363         ni->nid = nid;
 364
 365         /* Check nat cache */
 366         down_read(&nm_i->nat_tree_lock);
 367         e = __lookup_nat_cache(nm_i, nid);
 368         if (e) {
 369                 ni->ino = nat_get_ino(e);
 370                 ni->blk_addr = nat_get_blkaddr(e);
 371                 ni->version = nat_get_version(e);
 372         }
 373         up_read(&nm_i->nat_tree_lock);
 374         if (e)
 375                 return;
 376
 377         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 378
 379         down_write(&nm_i->nat_tree_lock);
 380
 381         /* Check current segment summary */
 382         mutex_lock(&curseg->curseg_mutex);
 383         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 384         if (i >= 0) {
 385                 ne = nat_in_journal(sum, i);
 386                 node_info_from_raw_nat(ni, &ne);
 387         }
 388         mutex_unlock(&curseg->curseg_mutex);
 389         if (i >= 0)
 390                 goto cache;
 391
 392         /* Fill node_info from nat page */
 393         page = get_current_nat_page(sbi, start_nid);
 394         nat_blk = (struct f2fs_nat_block *)page_address(page);
 395         ne = nat_blk->entries[nid - start_nid];
 396         node_info_from_raw_nat(ni, &ne);
 397         f2fs_put_page(page, 1);
 398 cache:
 399         /* cache nat entry */
 400         cache_nat_entry(NM_I(sbi), nid, &ne);
 401         up_write(&nm_i->nat_tree_lock);
 402 }
 403
 404 /*
 405  * The maximum depth is four.
 406  * Offset[0] will have raw inode offset.
 407  */
 408 static int get_node_path(struct f2fs_inode_info *fi, long block,
 409                                 int offset[4], unsigned int noffset[4])
 410 {
 411         const long direct_index = ADDRS_PER_INODE(fi);
 412         const long direct_blks = ADDRS_PER_BLOCK;
 413         const long dptrs_per_blk = NIDS_PER_BLOCK;
 414         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 415         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 416         int n = 0;
 417         int level = 0;
 418
 419         noffset[0] = 0;
 420
 421         if (block < direct_index) {
 422                 offset[n] = block;
 423                 goto got;
 424         }
 425         block -= direct_index;
 426         if (block < direct_blks) {
 427                 offset[n++] = NODE_DIR1_BLOCK;
 428                 noffset[n] = 1;
 429                 offset[n] = block;
 430                 level = 1;
 431                 goto got;
 432         }
 433         block -= direct_blks;
 434         if (block < direct_blks) {
 435                 offset[n++] = NODE_DIR2_BLOCK;
 436                 noffset[n] = 2;
 437                 offset[n] = block;
 438                 level = 1;
 439                 goto got;
 440         }
 441         block -= direct_blks;
 442         if (block < indirect_blks) {
 443                 offset[n++] = NODE_IND1_BLOCK;
 444                 noffset[n] = 3;
 445                 offset[n++] = block / direct_blks;
 446                 noffset[n] = 4 + offset[n - 1];
 447                 offset[n] = block % direct_blks;
 448                 level = 2;
 449                 goto got;
 450         }
 451         block -= indirect_blks;
 452         if (block < indirect_blks) {
 453                 offset[n++] = NODE_IND2_BLOCK;
 454                 noffset[n] = 4 + dptrs_per_blk;
 455                 offset[n++] = block / direct_blks;
 456                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 457                 offset[n] = block % direct_blks;
 458                 level = 2;
 459                 goto got;
 460         }
 461         block -= indirect_blks;
 462         if (block < dindirect_blks) {
 463                 offset[n++] = NODE_DIND_BLOCK;
 464                 noffset[n] = 5 + (dptrs_per_blk * 2);
 465                 offset[n++] = block / indirect_blks;
 466                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 467                               offset[n - 1] * (dptrs_per_blk + 1);
 468                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 469                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 470                               offset[n - 2] * (dptrs_per_blk + 1) +
 471                               offset[n - 1];
 472                 offset[n] = block % direct_blks;
 473                 level = 3;
 474                 goto got;
 475         } else {
 476                 BUG();
 477         }
 478 got:
 479         return level;
 480 }
 481
 482 /*
 483  * Caller should call f2fs_put_dnode(dn).
 484  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 485  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 486  * In the case of RDONLY_NODE, we don't need to care about mutex.
 487  */
 488 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 489 {
 490         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 491         struct page *npage[4];
 492         struct page *parent = NULL;
 493         int offset[4];
 494         unsigned int noffset[4];
 495         nid_t nids[4];
 496         int level, i;
 497         int err = 0;
 498
 499         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 500
 501         nids[0] = dn->inode->i_ino;
 502         npage[0] = dn->inode_page;
 503
 504         if (!npage[0]) {
 505                 npage[0] = get_node_page(sbi, nids[0]);
 506                 if (IS_ERR(npage[0]))
 507                         return PTR_ERR(npage[0]);
 508         }
 509
 510         /* if inline_data is set, should not report any block indices */
 511         if (f2fs_has_inline_data(dn->inode) && index) {
 512                 err = -ENOENT;
 513                 f2fs_put_page(npage[0], 1);
 514                 goto release_out;
 515         }
 516
 517         parent = npage[0];
 518         if (level != 0)
 519                 nids[1] = get_nid(parent, offset[0], true);
 520         dn->inode_page = npage[0];
 521         dn->inode_page_locked = true;
 522
 523         /* get indirect or direct nodes */
 524         for (i = 1; i <= level; i++) {
 525                 bool done = false;
 526
 527                 if (!nids[i] && mode == ALLOC_NODE) {
 528                         /* alloc new node */
 529                         if (!alloc_nid(sbi, &(nids[i]))) {
 530                                 err = -ENOSPC;
 531                                 goto release_pages;
 532                         }
 533
 534                         dn->nid = nids[i];
 535                         npage[i] = new_node_page(dn, noffset[i], NULL);
 536                         if (IS_ERR(npage[i])) {
 537                                 alloc_nid_failed(sbi, nids[i]);
 538                                 err = PTR_ERR(npage[i]);
 539                                 goto release_pages;
 540                         }
 541
 542                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 543                         alloc_nid_done(sbi, nids[i]);
 544                         done = true;
 545                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 546                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 547                         if (IS_ERR(npage[i])) {
 548                                 err = PTR_ERR(npage[i]);
 549                                 goto release_pages;
 550                         }
 551                         done = true;
 552                 }
 553                 if (i == 1) {
 554                         dn->inode_page_locked = false;
 555                         unlock_page(parent);
 556                 } else {
 557                         f2fs_put_page(parent, 1);
 558                 }
 559
 560                 if (!done) {
 561                         npage[i] = get_node_page(sbi, nids[i]);
 562                         if (IS_ERR(npage[i])) {
 563                                 err = PTR_ERR(npage[i]);
 564                                 f2fs_put_page(npage[0], 0);
 565                                 goto release_out;
 566                         }
 567                 }
 568                 if (i < level) {
 569                         parent = npage[i];
 570                         nids[i + 1] = get_nid(parent, offset[i], false);
 571                 }
 572         }
 573         dn->nid = nids[level];
 574         dn->ofs_in_node = offset[level];
 575         dn->node_page = npage[level];
 576         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 577         return 0;
 578
 579 release_pages:
 580         f2fs_put_page(parent, 1);
 581         if (i > 1)
 582                 f2fs_put_page(npage[0], 0);
 583 release_out:
 584         dn->inode_page = NULL;
 585         dn->node_page = NULL;
 586         return err;
 587 }
 588
 589 static void truncate_node(struct dnode_of_data *dn)
 590 {
 591         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 592         struct node_info ni;
 593         pgoff_t index;
 594
 595         get_node_info(sbi, dn->nid, &ni);
 596         if (dn->inode->i_blocks == 0) {
 597                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 598                 goto invalidate;
 599         }
 600         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 601
 602         /* Deallocate node address */
 603         invalidate_blocks(sbi, ni.blk_addr);
 604         dec_valid_node_count(sbi, dn->inode);
 605         set_node_addr(sbi, &ni, NULL_ADDR, false);
 606
 607         if (dn->nid == dn->inode->i_ino) {
 608                 remove_orphan_inode(sbi, dn->nid);
 609                 dec_valid_inode_count(sbi);
 610         } else {
 611                 sync_inode_page(dn);
 612         }
 613 invalidate:
 614         clear_node_page_dirty(dn->node_page);
 615         set_sbi_flag(sbi, SBI_IS_DIRTY);
 616
 617         index = dn->node_page->index;
 618         f2fs_put_page(dn->node_page, 1);
 619
 620         invalidate_mapping_pages(NODE_MAPPING(sbi),
 621                         index, index);
 622
 623         dn->node_page = NULL;
 624         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 625 }
 626
 627 static int truncate_dnode(struct dnode_of_data *dn)
 628 {
 629         struct page *page;
 630
 631         if (dn->nid == 0)
 632                 return 1;
 633
 634         /* get direct node */
 635         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 636         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 637                 return 1;
 638         else if (IS_ERR(page))
 639                 return PTR_ERR(page);
 640
 641         /* Make dnode_of_data for parameter */
 642         dn->node_page = page;
 643         dn->ofs_in_node = 0;
 644         truncate_data_blocks(dn);
 645         truncate_node(dn);
 646         return 1;
 647 }
 648
 649 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 650                                                 int ofs, int depth)
 651 {
 652         struct dnode_of_data rdn = *dn;
 653         struct page *page;
 654         struct f2fs_node *rn;
 655         nid_t child_nid;
 656         unsigned int child_nofs;
 657         int freed = 0;
 658         int i, ret;
 659
 660         if (dn->nid == 0)
 661                 return NIDS_PER_BLOCK + 1;
 662
 663         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 664
 665         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 666         if (IS_ERR(page)) {
 667                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 668                 return PTR_ERR(page);
 669         }
 670
 671         rn = F2FS_NODE(page);
 672         if (depth < 3) {
 673                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 674                         child_nid = le32_to_cpu(rn->in.nid[i]);
 675                         if (child_nid == 0)
 676                                 continue;
 677                         rdn.nid = child_nid;
 678                         ret = truncate_dnode(&rdn);
 679                         if (ret < 0)
 680                                 goto out_err;
 681                         set_nid(page, i, 0, false);
 682                 }
 683         } else {
 684                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 685                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 686                         child_nid = le32_to_cpu(rn->in.nid[i]);
 687                         if (child_nid == 0) {
 688                                 child_nofs += NIDS_PER_BLOCK + 1;
 689                                 continue;
 690                         }
 691                         rdn.nid = child_nid;
 692                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 693                         if (ret == (NIDS_PER_BLOCK + 1)) {
 694                                 set_nid(page, i, 0, false);
 695                                 child_nofs += ret;
 696                         } else if (ret < 0 && ret != -ENOENT) {
 697                                 goto out_err;
 698                         }
 699                 }
 700                 freed = child_nofs;
 701         }
 702
 703         if (!ofs) {
 704                 /* remove current indirect node */
 705                 dn->node_page = page;
 706                 truncate_node(dn);
 707                 freed++;
 708         } else {
 709                 f2fs_put_page(page, 1);
 710         }
 711         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 712         return freed;
 713
 714 out_err:
 715         f2fs_put_page(page, 1);
 716         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 717         return ret;
 718 }
 719
 720 static int truncate_partial_nodes(struct dnode_of_data *dn,
 721                         struct f2fs_inode *ri, int *offset, int depth)
 722 {
 723         struct page *pages[2];
 724         nid_t nid[3];
 725         nid_t child_nid;
 726         int err = 0;
 727         int i;
 728         int idx = depth - 2;
 729
 730         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 731         if (!nid[0])
 732                 return 0;
 733
 734         /* get indirect nodes in the path */
 735         for (i = 0; i < idx + 1; i++) {
 736                 /* reference count'll be increased */
 737                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 738                 if (IS_ERR(pages[i])) {
 739                         err = PTR_ERR(pages[i]);
 740                         idx = i - 1;
 741                         goto fail;
 742                 }
 743                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 744         }
 745
 746         /* free direct nodes linked to a partial indirect node */
 747         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 748                 child_nid = get_nid(pages[idx], i, false);
 749                 if (!child_nid)
 750                         continue;
 751                 dn->nid = child_nid;
 752                 err = truncate_dnode(dn);
 753                 if (err < 0)
 754                         goto fail;
 755                 set_nid(pages[idx], i, 0, false);
 756         }
 757
 758         if (offset[idx + 1] == 0) {
 759                 dn->node_page = pages[idx];
 760                 dn->nid = nid[idx];
 761                 truncate_node(dn);
 762         } else {
 763                 f2fs_put_page(pages[idx], 1);
 764         }
 765         offset[idx]++;
 766         offset[idx + 1] = 0;
 767         idx--;
 768 fail:
 769         for (i = idx; i >= 0; i--)
 770                 f2fs_put_page(pages[i], 1);
 771
 772         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 773
 774         return err;
 775 }
 776
 777 /*
 778  * All the block addresses of data and nodes should be nullified.
 779  */
 780 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 781 {
 782         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 783         int err = 0, cont = 1;
 784         int level, offset[4], noffset[4];
 785         unsigned int nofs = 0;
 786         struct f2fs_inode *ri;
 787         struct dnode_of_data dn;
 788         struct page *page;
 789
 790         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 791
 792         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 793 restart:
 794         page = get_node_page(sbi, inode->i_ino);
 795         if (IS_ERR(page)) {
 796                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 797                 return PTR_ERR(page);
 798         }
 799
 800         set_new_dnode(&dn, inode, page, NULL, 0);
 801         unlock_page(page);
 802
 803         ri = F2FS_INODE(page);
 804         switch (level) {
 805         case 0:
 806         case 1:
 807                 nofs = noffset[1];
 808                 break;
 809         case 2:
 810                 nofs = noffset[1];
 811                 if (!offset[level - 1])
 812                         goto skip_partial;
 813                 err = truncate_partial_nodes(&dn, ri, offset, level);
 814                 if (err < 0 && err != -ENOENT)
 815                         goto fail;
 816                 nofs += 1 + NIDS_PER_BLOCK;
 817                 break;
 818         case 3:
 819                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 820                 if (!offset[level - 1])
 821                         goto skip_partial;
 822                 err = truncate_partial_nodes(&dn, ri, offset, level);
 823                 if (err < 0 && err != -ENOENT)
 824                         goto fail;
 825                 break;
 826         default:
 827                 BUG();
 828         }
 829
 830 skip_partial:
 831         while (cont) {
 832                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 833                 switch (offset[0]) {
 834                 case NODE_DIR1_BLOCK:
 835                 case NODE_DIR2_BLOCK:
 836                         err = truncate_dnode(&dn);
 837                         break;
 838
 839                 case NODE_IND1_BLOCK:
 840                 case NODE_IND2_BLOCK:
 841                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 842                         break;
 843
 844                 case NODE_DIND_BLOCK:
 845                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 846                         cont = 0;
 847                         break;
 848
 849                 default:
 850                         BUG();
 851                 }
 852                 if (err < 0 && err != -ENOENT)
 853                         goto fail;
 854                 if (offset[1] == 0 &&
 855                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 856                         lock_page(page);
 857                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 858                                 f2fs_put_page(page, 1);
 859                                 goto restart;
 860                         }
 861                         f2fs_wait_on_page_writeback(page, NODE);
 862                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 863                         set_page_dirty(page);
 864                         unlock_page(page);
 865                 }
 866                 offset[1] = 0;
 867                 offset[0]++;
 868                 nofs += err;
 869         }
 870 fail:
 871         f2fs_put_page(page, 0);
 872         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 873         return err > 0 ? 0 : err;
 874 }
 875
 876 int truncate_xattr_node(struct inode *inode, struct page *page)
 877 {
 878         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 879         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 880         struct dnode_of_data dn;
 881         struct page *npage;
 882
 883         if (!nid)
 884                 return 0;
 885
 886         npage = get_node_page(sbi, nid);
 887         if (IS_ERR(npage))
 888                 return PTR_ERR(npage);
 889
 890         F2FS_I(inode)->i_xattr_nid = 0;
 891
 892         /* need to do checkpoint during fsync */
 893         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 894
 895         set_new_dnode(&dn, inode, page, npage, nid);
 896
 897         if (page)
 898                 dn.inode_page_locked = true;
 899         truncate_node(&dn);
 900         return 0;
 901 }
 902
 903 /*
 904  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 905  * f2fs_unlock_op().
 906  */
 907 int remove_inode_page(struct inode *inode)
 908 {
 909         struct dnode_of_data dn;
 910         int err;
 911
 912         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 913         err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
 914         if (err)
 915                 return err;
 916
 917         err = truncate_xattr_node(inode, dn.inode_page);
 918         if (err) {
 919                 f2fs_put_dnode(&dn);
 920                 return err;
 921         }
 922
 923         /* remove potential inline_data blocks */
 924         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 925                                 S_ISLNK(inode->i_mode))
 926                 truncate_data_blocks_range(&dn, 1);
 927
 928         /* 0 is possible, after f2fs_new_inode() has failed */
 929         f2fs_bug_on(F2FS_I_SB(inode),
 930                         inode->i_blocks != 0 && inode->i_blocks != 1);
 931
 932         /* will put inode & node pages */
 933         truncate_node(&dn);
 934         return 0;
 935 }
 936
 937 struct page *new_inode_page(struct inode *inode)
 938 {
 939         struct dnode_of_data dn;
 940
 941         /* allocate inode page for new inode */
 942         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 943
 944         /* caller should f2fs_put_page(page, 1); */
 945         return new_node_page(&dn, 0, NULL);
 946 }
 947
 948 struct page *new_node_page(struct dnode_of_data *dn,
 949                                 unsigned int ofs, struct page *ipage)
 950 {
 951         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 952         struct node_info old_ni, new_ni;
 953         struct page *page;
 954         int err;
 955
 956         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 957                 return ERR_PTR(-EPERM);
 958
 959         page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
 960         if (!page)
 961                 return ERR_PTR(-ENOMEM);
 962
 963         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 964                 err = -ENOSPC;
 965                 goto fail;
 966         }
 967
 968         get_node_info(sbi, dn->nid, &old_ni);
 969
 970         /* Reinitialize old_ni with new node page */
 971         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
 972         new_ni = old_ni;
 973         new_ni.ino = dn->inode->i_ino;
 974         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
 975
 976         f2fs_wait_on_page_writeback(page, NODE);
 977         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 978         set_cold_node(dn->inode, page);
 979         SetPageUptodate(page);
 980         set_page_dirty(page);
 981
 982         if (f2fs_has_xattr_block(ofs))
 983                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 984
 985         dn->node_page = page;
 986         if (ipage)
 987                 update_inode(dn->inode, ipage);
 988         else
 989                 sync_inode_page(dn);
 990         if (ofs == 0)
 991                 inc_valid_inode_count(sbi);
 992
 993         return page;
 994
 995 fail:
 996         clear_node_page_dirty(page);
 997         f2fs_put_page(page, 1);
 998         return ERR_PTR(err);
 999 }
1000
1001 /*
1002  * Caller should do after getting the following values.
1003  * 0: f2fs_put_page(page, 0)
1004  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1005  */
1006 static int read_node_page(struct page *page, int rw)
1007 {
1008         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1009         struct node_info ni;
1010         struct f2fs_io_info fio = {
1011                 .sbi = sbi,
1012                 .type = NODE,
1013                 .rw = rw,
1014                 .page = page,
1015                 .encrypted_page = NULL,
1016         };
1017
1018         get_node_info(sbi, page->index, &ni);
1019
1020         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1021                 ClearPageUptodate(page);
1022                 return -ENOENT;
1023         }
1024
1025         if (PageUptodate(page))
1026                 return LOCKED_PAGE;
1027
1028         fio.blk_addr = ni.blk_addr;
1029         return f2fs_submit_page_bio(&fio);
1030 }
1031
1032 /*
1033  * Readahead a node page
1034  */
1035 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1036 {
1037         struct page *apage;
1038         int err;
1039
1040         apage = find_get_page(NODE_MAPPING(sbi), nid);
1041         if (apage && PageUptodate(apage)) {
1042                 f2fs_put_page(apage, 0);
1043                 return;
1044         }
1045         f2fs_put_page(apage, 0);
1046
1047         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1048         if (!apage)
1049                 return;
1050
1051         err = read_node_page(apage, READA);
1052         f2fs_put_page(apage, err ? 1 : 0);
1053 }
1054
1055 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1056 {
1057         struct page *page;
1058         int err;
1059 repeat:
1060         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1061         if (!page)
1062                 return ERR_PTR(-ENOMEM);
1063
1064         err = read_node_page(page, READ_SYNC);
1065         if (err < 0) {
1066                 f2fs_put_page(page, 1);
1067                 return ERR_PTR(err);
1068         } else if (err != LOCKED_PAGE) {
1069                 lock_page(page);
1070         }
1071
1072         if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
1073                 ClearPageUptodate(page);
1074                 f2fs_put_page(page, 1);
1075                 return ERR_PTR(-EIO);
1076         }
1077         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1078                 f2fs_put_page(page, 1);
1079                 goto repeat;
1080         }
1081         return page;
1082 }
1083
1084 /*
1085  * Return a locked page for the desired node page.
1086  * And, readahead MAX_RA_NODE number of node pages.
1087  */
1088 struct page *get_node_page_ra(struct page *parent, int start)
1089 {
1090         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1091         struct blk_plug plug;
1092         struct page *page;
1093         int err, i, end;
1094         nid_t nid;
1095
1096         /* First, try getting the desired direct node. */
1097         nid = get_nid(parent, start, false);
1098         if (!nid)
1099                 return ERR_PTR(-ENOENT);
1100 repeat:
1101         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1102         if (!page)
1103                 return ERR_PTR(-ENOMEM);
1104
1105         err = read_node_page(page, READ_SYNC);
1106         if (err < 0) {
1107                 f2fs_put_page(page, 1);
1108                 return ERR_PTR(err);
1109         } else if (err == LOCKED_PAGE) {
1110                 goto page_hit;
1111         }
1112
1113         blk_start_plug(&plug);
1114
1115         /* Then, try readahead for siblings of the desired node */
1116         end = start + MAX_RA_NODE;
1117         end = min(end, NIDS_PER_BLOCK);
1118         for (i = start + 1; i < end; i++) {
1119                 nid = get_nid(parent, i, false);
1120                 if (!nid)
1121                         continue;
1122                 ra_node_page(sbi, nid);
1123         }
1124
1125         blk_finish_plug(&plug);
1126
1127         lock_page(page);
1128         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1129                 f2fs_put_page(page, 1);
1130                 goto repeat;
1131         }
1132 page_hit:
1133         if (unlikely(!PageUptodate(page))) {
1134                 f2fs_put_page(page, 1);
1135                 return ERR_PTR(-EIO);
1136         }
1137         return page;
1138 }
1139
1140 void sync_inode_page(struct dnode_of_data *dn)
1141 {
1142         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1143                 update_inode(dn->inode, dn->node_page);
1144         } else if (dn->inode_page) {
1145                 if (!dn->inode_page_locked)
1146                         lock_page(dn->inode_page);
1147                 update_inode(dn->inode, dn->inode_page);
1148                 if (!dn->inode_page_locked)
1149                         unlock_page(dn->inode_page);
1150         } else {
1151                 update_inode_page(dn->inode);
1152         }
1153 }
1154
1155 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1156                                         struct writeback_control *wbc)
1157 {
1158         pgoff_t index, end;
1159         struct pagevec pvec;
1160         int step = ino ? 2 : 0;
1161         int nwritten = 0, wrote = 0;
1162
1163         pagevec_init(&pvec, 0);
1164
1165 next_step:
1166         index = 0;
1167         end = LONG_MAX;
1168
1169         while (index <= end) {
1170                 int i, nr_pages;
1171                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1172                                 PAGECACHE_TAG_DIRTY,
1173                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1174                 if (nr_pages == 0)
1175                         break;
1176
1177                 for (i = 0; i < nr_pages; i++) {
1178                         struct page *page = pvec.pages[i];
1179
1180                         /*
1181                          * flushing sequence with step:
1182                          * 0. indirect nodes
1183                          * 1. dentry dnodes
1184                          * 2. file dnodes
1185                          */
1186                         if (step == 0 && IS_DNODE(page))
1187                                 continue;
1188                         if (step == 1 && (!IS_DNODE(page) ||
1189                                                 is_cold_node(page)))
1190                                 continue;
1191                         if (step == 2 && (!IS_DNODE(page) ||
1192                                                 !is_cold_node(page)))
1193                                 continue;
1194
1195                         /*
1196                          * If an fsync mode,
1197                          * we should not skip writing node pages.
1198                          */
1199                         if (ino && ino_of_node(page) == ino)
1200                                 lock_page(page);
1201                         else if (!trylock_page(page))
1202                                 continue;
1203
1204                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1205 continue_unlock:
1206                                 unlock_page(page);
1207                                 continue;
1208                         }
1209                         if (ino && ino_of_node(page) != ino)
1210                                 goto continue_unlock;
1211
1212                         if (!PageDirty(page)) {
1213                                 /* someone wrote it for us */
1214                                 goto continue_unlock;
1215                         }
1216
1217                         if (!clear_page_dirty_for_io(page))
1218                                 goto continue_unlock;
1219
1220                         /* called by fsync() */
1221                         if (ino && IS_DNODE(page)) {
1222                                 set_fsync_mark(page, 1);
1223                                 if (IS_INODE(page))
1224                                         set_dentry_mark(page,
1225                                                 need_dentry_mark(sbi, ino));
1226                                 nwritten++;
1227                         } else {
1228                                 set_fsync_mark(page, 0);
1229                                 set_dentry_mark(page, 0);
1230                         }
1231
1232                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1233                                 unlock_page(page);
1234                         else
1235                                 wrote++;
1236
1237                         if (--wbc->nr_to_write == 0)
1238                                 break;
1239                 }
1240                 pagevec_release(&pvec);
1241                 cond_resched();
1242
1243                 if (wbc->nr_to_write == 0) {
1244                         step = 2;
1245                         break;
1246                 }
1247         }
1248
1249         if (step < 2) {
1250                 step++;
1251                 goto next_step;
1252         }
1253
1254         if (wrote)
1255                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1256         return nwritten;
1257 }
1258
1259 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1260 {
1261         pgoff_t index = 0, end = LONG_MAX;
1262         struct pagevec pvec;
1263         int ret2 = 0, ret = 0;
1264
1265         pagevec_init(&pvec, 0);
1266
1267         while (index <= end) {
1268                 int i, nr_pages;
1269                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1270                                 PAGECACHE_TAG_WRITEBACK,
1271                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1272                 if (nr_pages == 0)
1273                         break;
1274
1275                 for (i = 0; i < nr_pages; i++) {
1276                         struct page *page = pvec.pages[i];
1277
1278                         /* until radix tree lookup accepts end_index */
1279                         if (unlikely(page->index > end))
1280                                 continue;
1281
1282                         if (ino && ino_of_node(page) == ino) {
1283                                 f2fs_wait_on_page_writeback(page, NODE);
1284                                 if (TestClearPageError(page))
1285                                         ret = -EIO;
1286                         }
1287                 }
1288                 pagevec_release(&pvec);
1289                 cond_resched();
1290         }
1291
1292         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1293                 ret2 = -ENOSPC;
1294         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1295                 ret2 = -EIO;
1296         if (!ret)
1297                 ret = ret2;
1298         return ret;
1299 }
1300
1301 static int f2fs_write_node_page(struct page *page,
1302                                 struct writeback_control *wbc)
1303 {
1304         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1305         nid_t nid;
1306         struct node_info ni;
1307         struct f2fs_io_info fio = {
1308                 .sbi = sbi,
1309                 .type = NODE,
1310                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1311                 .page = page,
1312                 .encrypted_page = NULL,
1313         };
1314
1315         trace_f2fs_writepage(page, NODE);
1316
1317         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1318                 goto redirty_out;
1319         if (unlikely(f2fs_cp_error(sbi)))
1320                 goto redirty_out;
1321
1322         f2fs_wait_on_page_writeback(page, NODE);
1323
1324         /* get old block addr of this node page */
1325         nid = nid_of_node(page);
1326         f2fs_bug_on(sbi, page->index != nid);
1327
1328         if (wbc->for_reclaim) {
1329                 if (!down_read_trylock(&sbi->node_write))
1330                         goto redirty_out;
1331         } else {
1332                 down_read(&sbi->node_write);
1333         }
1334
1335         get_node_info(sbi, nid, &ni);
1336
1337         /* This page is already truncated */
1338         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1339                 ClearPageUptodate(page);
1340                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1341                 up_read(&sbi->node_write);
1342                 unlock_page(page);
1343                 return 0;
1344         }
1345
1346         if (__is_valid_data_blkaddr(ni.blk_addr) &&
1347                 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC)) {
1348                 up_read(&sbi->node_write);
1349                 goto redirty_out;
1350         }
1351
1352         set_page_writeback(page);
1353         fio.blk_addr = ni.blk_addr;
1354         write_node_page(nid, &fio);
1355         set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1356         dec_page_count(sbi, F2FS_DIRTY_NODES);
1357         up_read(&sbi->node_write);
1358         unlock_page(page);
1359
1360         if (wbc->for_reclaim)
1361                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1362
1363         return 0;
1364
1365 redirty_out:
1366         redirty_page_for_writepage(wbc, page);
1367         return AOP_WRITEPAGE_ACTIVATE;
1368 }
1369
1370 static int f2fs_write_node_pages(struct address_space *mapping,
1371                             struct writeback_control *wbc)
1372 {
1373         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1374         long diff;
1375
1376         trace_f2fs_writepages(mapping->host, wbc, NODE);
1377
1378         /* balancing f2fs's metadata in background */
1379         f2fs_balance_fs_bg(sbi);
1380
1381         /* collect a number of dirty node pages and write together */
1382         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1383                 goto skip_write;
1384
1385         diff = nr_pages_to_write(sbi, NODE, wbc);
1386         wbc->sync_mode = WB_SYNC_NONE;
1387         sync_node_pages(sbi, 0, wbc);
1388         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1389         return 0;
1390
1391 skip_write:
1392         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1393         return 0;
1394 }
1395
1396 static int f2fs_set_node_page_dirty(struct page *page)
1397 {
1398         trace_f2fs_set_page_dirty(page, NODE);
1399
1400         SetPageUptodate(page);
1401         if (!PageDirty(page)) {
1402                 __set_page_dirty_nobuffers(page);
1403                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1404                 SetPagePrivate(page);
1405                 f2fs_trace_pid(page);
1406                 return 1;
1407         }
1408         return 0;
1409 }
1410
1411 /*
1412  * Structure of the f2fs node operations
1413  */
1414 const struct address_space_operations f2fs_node_aops = {
1415         .writepage      = f2fs_write_node_page,
1416         .writepages     = f2fs_write_node_pages,
1417         .set_page_dirty = f2fs_set_node_page_dirty,
1418         .invalidatepage = f2fs_invalidate_page,
1419         .releasepage    = f2fs_release_page,
1420 };
1421
1422 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1423                                                 nid_t n)
1424 {
1425         return radix_tree_lookup(&nm_i->free_nid_root, n);
1426 }
1427
1428 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1429                                                 struct free_nid *i)
1430 {
1431         list_del(&i->list);
1432         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1433 }
1434
1435 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1436 {
1437         struct f2fs_nm_info *nm_i = NM_I(sbi);
1438         struct free_nid *i, *e;
1439         struct nat_entry *ne;
1440         int err = -EINVAL;
1441
1442         if (!available_free_memory(sbi, FREE_NIDS))
1443                 return -1;
1444
1445         /* 0 nid should not be used */
1446         if (unlikely(nid == 0))
1447                 return 0;
1448
1449         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1450         i->nid = nid;
1451         i->state = NID_NEW;
1452
1453         if (radix_tree_preload(GFP_NOFS))
1454                 goto err;
1455
1456         spin_lock(&nm_i->free_nid_list_lock);
1457
1458         if (build) {
1459                 /*
1460                  *   Thread A             Thread B
1461                  *  - f2fs_create
1462                  *   - f2fs_new_inode
1463                  *    - alloc_nid
1464                  *     - __insert_nid_to_list(ALLOC_NID_LIST)
1465                  *                     - f2fs_balance_fs_bg
1466                  *                      - build_free_nids
1467                  *                       - __build_free_nids
1468                  *                        - scan_nat_page
1469                  *                         - add_free_nid
1470                  *                          - __lookup_nat_cache
1471                  *  - f2fs_add_link
1472                  *   - init_inode_metadata
1473                  *    - new_inode_page
1474                  *     - new_node_page
1475                  *      - set_node_addr
1476                  *  - alloc_nid_done
1477                  *   - __remove_nid_from_list(ALLOC_NID_LIST)
1478                  *                         - __insert_nid_to_list(FREE_NID_LIST)
1479                  */
1480                 ne = __lookup_nat_cache(nm_i, nid);
1481                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1482                                 nat_get_blkaddr(ne) != NULL_ADDR))
1483                         goto err_out;
1484
1485                 e = __lookup_free_nid_list(nm_i, nid);
1486                 if (e)
1487                         goto err_out;
1488         }
1489         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i))
1490                 goto err_out;
1491         err = 0;
1492         list_add_tail(&i->list, &nm_i->free_nid_list);
1493         nm_i->fcnt++;
1494 err_out:
1495         spin_unlock(&nm_i->free_nid_list_lock);
1496         radix_tree_preload_end();
1497 err:
1498         if (err)
1499                 kmem_cache_free(free_nid_slab, i);
1500         return !err;
1501 }
1502
1503 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1504 {
1505         struct free_nid *i;
1506         bool need_free = false;
1507
1508         spin_lock(&nm_i->free_nid_list_lock);
1509         i = __lookup_free_nid_list(nm_i, nid);
1510         if (i && i->state == NID_NEW) {
1511                 __del_from_free_nid_list(nm_i, i);
1512                 nm_i->fcnt--;
1513                 need_free = true;
1514         }
1515         spin_unlock(&nm_i->free_nid_list_lock);
1516
1517         if (need_free)
1518                 kmem_cache_free(free_nid_slab, i);
1519 }
1520
1521 static void scan_nat_page(struct f2fs_sb_info *sbi,
1522                         struct page *nat_page, nid_t start_nid)
1523 {
1524         struct f2fs_nm_info *nm_i = NM_I(sbi);
1525         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1526         block_t blk_addr;
1527         int i;
1528
1529         i = start_nid % NAT_ENTRY_PER_BLOCK;
1530
1531         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1532
1533                 if (unlikely(start_nid >= nm_i->max_nid))
1534                         break;
1535
1536                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1537                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1538                 if (blk_addr == NULL_ADDR) {
1539                         if (add_free_nid(sbi, start_nid, true) < 0)
1540                                 break;
1541                 }
1542         }
1543 }
1544
1545 static void build_free_nids(struct f2fs_sb_info *sbi)
1546 {
1547         struct f2fs_nm_info *nm_i = NM_I(sbi);
1548         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1549         struct f2fs_summary_block *sum = curseg->sum_blk;
1550         int i = 0;
1551         nid_t nid = nm_i->next_scan_nid;
1552
1553         /* Enough entries */
1554         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1555                 return;
1556
1557         /* readahead nat pages to be scanned */
1558         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1559                                                         META_NAT, true);
1560
1561         down_read(&nm_i->nat_tree_lock);
1562
1563         while (1) {
1564                 struct page *page = get_current_nat_page(sbi, nid);
1565
1566                 scan_nat_page(sbi, page, nid);
1567                 f2fs_put_page(page, 1);
1568
1569                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1570                 if (unlikely(nid >= nm_i->max_nid))
1571                         nid = 0;
1572
1573                 if (++i >= FREE_NID_PAGES)
1574                         break;
1575         }
1576
1577         /* go to the next free nat pages to find free nids abundantly */
1578         nm_i->next_scan_nid = nid;
1579
1580         /* find free nids from current sum_pages */
1581         mutex_lock(&curseg->curseg_mutex);
1582         for (i = 0; i < nats_in_cursum(sum); i++) {
1583                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1584                 nid = le32_to_cpu(nid_in_journal(sum, i));
1585                 if (addr == NULL_ADDR)
1586                         add_free_nid(sbi, nid, true);
1587                 else
1588                         remove_free_nid(nm_i, nid);
1589         }
1590         mutex_unlock(&curseg->curseg_mutex);
1591         up_read(&nm_i->nat_tree_lock);
1592
1593         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1594                                         nm_i->ra_nid_pages, META_NAT, false);
1595 }
1596
1597 /*
1598  * If this function returns success, caller can obtain a new nid
1599  * from second parameter of this function.
1600  * The returned nid could be used ino as well as nid when inode is created.
1601  */
1602 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1603 {
1604         struct f2fs_nm_info *nm_i = NM_I(sbi);
1605         struct free_nid *i = NULL;
1606 retry:
1607         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1608                 return false;
1609
1610         spin_lock(&nm_i->free_nid_list_lock);
1611
1612         /* We should not use stale free nids created by build_free_nids */
1613         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1614                 struct node_info ni;
1615
1616                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1617                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1618                         if (i->state == NID_NEW)
1619                                 break;
1620
1621                 f2fs_bug_on(sbi, i->state != NID_NEW);
1622                 *nid = i->nid;
1623                 i->state = NID_ALLOC;
1624                 nm_i->fcnt--;
1625                 spin_unlock(&nm_i->free_nid_list_lock);
1626
1627                 /* check nid is allocated already */
1628                 get_node_info(sbi, *nid, &ni);
1629                 if (ni.blk_addr != NULL_ADDR) {
1630                         alloc_nid_done(sbi, *nid);
1631                         goto retry;
1632                 }
1633                 return true;
1634         }
1635         spin_unlock(&nm_i->free_nid_list_lock);
1636
1637         /* Let's scan nat pages and its caches to get free nids */
1638         mutex_lock(&nm_i->build_lock);
1639         build_free_nids(sbi);
1640         mutex_unlock(&nm_i->build_lock);
1641         goto retry;
1642 }
1643
1644 /*
1645  * alloc_nid() should be called prior to this function.
1646  */
1647 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1648 {
1649         struct f2fs_nm_info *nm_i = NM_I(sbi);
1650         struct free_nid *i;
1651
1652         spin_lock(&nm_i->free_nid_list_lock);
1653         i = __lookup_free_nid_list(nm_i, nid);
1654         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1655         __del_from_free_nid_list(nm_i, i);
1656         spin_unlock(&nm_i->free_nid_list_lock);
1657
1658         kmem_cache_free(free_nid_slab, i);
1659 }
1660
1661 /*
1662  * alloc_nid() should be called prior to this function.
1663  */
1664 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1665 {
1666         struct f2fs_nm_info *nm_i = NM_I(sbi);
1667         struct free_nid *i;
1668         bool need_free = false;
1669
1670         if (!nid)
1671                 return;
1672
1673         spin_lock(&nm_i->free_nid_list_lock);
1674         i = __lookup_free_nid_list(nm_i, nid);
1675         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1676         if (!available_free_memory(sbi, FREE_NIDS)) {
1677                 __del_from_free_nid_list(nm_i, i);
1678                 need_free = true;
1679         } else {
1680                 i->state = NID_NEW;
1681                 nm_i->fcnt++;
1682         }
1683         spin_unlock(&nm_i->free_nid_list_lock);
1684
1685         if (need_free)
1686                 kmem_cache_free(free_nid_slab, i);
1687 }
1688
1689 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1690 {
1691         struct f2fs_nm_info *nm_i = NM_I(sbi);
1692         struct free_nid *i, *next;
1693         int nr = nr_shrink;
1694
1695         if (!mutex_trylock(&nm_i->build_lock))
1696                 return 0;
1697
1698         spin_lock(&nm_i->free_nid_list_lock);
1699         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1700                 if (nr_shrink <= 0 || nm_i->fcnt <= NAT_ENTRY_PER_BLOCK)
1701                         break;
1702                 if (i->state == NID_ALLOC)
1703                         continue;
1704                 __del_from_free_nid_list(nm_i, i);
1705                 kmem_cache_free(free_nid_slab, i);
1706                 nm_i->fcnt--;
1707                 nr_shrink--;
1708         }
1709         spin_unlock(&nm_i->free_nid_list_lock);
1710         mutex_unlock(&nm_i->build_lock);
1711
1712         return nr - nr_shrink;
1713 }
1714
1715 void recover_inline_xattr(struct inode *inode, struct page *page)
1716 {
1717         void *src_addr, *dst_addr;
1718         size_t inline_size;
1719         struct page *ipage;
1720         struct f2fs_inode *ri;
1721
1722         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1723         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1724
1725         ri = F2FS_INODE(page);
1726         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1727                 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1728                 goto update_inode;
1729         }
1730
1731         dst_addr = inline_xattr_addr(ipage);
1732         src_addr = inline_xattr_addr(page);
1733         inline_size = inline_xattr_size(inode);
1734
1735         f2fs_wait_on_page_writeback(ipage, NODE);
1736         memcpy(dst_addr, src_addr, inline_size);
1737 update_inode:
1738         update_inode(inode, ipage);
1739         f2fs_put_page(ipage, 1);
1740 }
1741
1742 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1743 {
1744         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1745         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1746         nid_t new_xnid = nid_of_node(page);
1747         struct node_info ni;
1748
1749         /* 1: invalidate the previous xattr nid */
1750         if (!prev_xnid)
1751                 goto recover_xnid;
1752
1753         /* Deallocate node address */
1754         get_node_info(sbi, prev_xnid, &ni);
1755         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1756         invalidate_blocks(sbi, ni.blk_addr);
1757         dec_valid_node_count(sbi, inode);
1758         set_node_addr(sbi, &ni, NULL_ADDR, false);
1759
1760 recover_xnid:
1761         /* 2: allocate new xattr nid */
1762         if (unlikely(!inc_valid_node_count(sbi, inode)))
1763                 f2fs_bug_on(sbi, 1);
1764
1765         remove_free_nid(NM_I(sbi), new_xnid);
1766         get_node_info(sbi, new_xnid, &ni);
1767         ni.ino = inode->i_ino;
1768         set_node_addr(sbi, &ni, NEW_ADDR, false);
1769         F2FS_I(inode)->i_xattr_nid = new_xnid;
1770
1771         /* 3: update xattr blkaddr */
1772         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1773         set_node_addr(sbi, &ni, blkaddr, false);
1774
1775         update_inode_page(inode);
1776 }
1777
1778 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1779 {
1780         struct f2fs_inode *src, *dst;
1781         nid_t ino = ino_of_node(page);
1782         struct node_info old_ni, new_ni;
1783         struct page *ipage;
1784
1785         get_node_info(sbi, ino, &old_ni);
1786
1787         if (unlikely(old_ni.blk_addr != NULL_ADDR))
1788                 return -EINVAL;
1789
1790         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1791         if (!ipage)
1792                 return -ENOMEM;
1793
1794         /* Should not use this inode from free nid list */
1795         remove_free_nid(NM_I(sbi), ino);
1796
1797         SetPageUptodate(ipage);
1798         fill_node_footer(ipage, ino, ino, 0, true);
1799
1800         src = F2FS_INODE(page);
1801         dst = F2FS_INODE(ipage);
1802
1803         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1804         dst->i_size = 0;
1805         dst->i_blocks = cpu_to_le64(1);
1806         dst->i_links = cpu_to_le32(1);
1807         dst->i_xattr_nid = 0;
1808         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1809
1810         new_ni = old_ni;
1811         new_ni.ino = ino;
1812
1813         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1814                 WARN_ON(1);
1815         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1816         inc_valid_inode_count(sbi);
1817         set_page_dirty(ipage);
1818         f2fs_put_page(ipage, 1);
1819         return 0;
1820 }
1821
1822 int restore_node_summary(struct f2fs_sb_info *sbi,
1823                         unsigned int segno, struct f2fs_summary_block *sum)
1824 {
1825         struct f2fs_node *rn;
1826         struct f2fs_summary *sum_entry;
1827         block_t addr;
1828         int bio_blocks = MAX_BIO_BLOCKS(sbi);
1829         int i, idx, last_offset, nrpages;
1830
1831         /* scan the node segment */
1832         last_offset = sbi->blocks_per_seg;
1833         addr = START_BLOCK(sbi, segno);
1834         sum_entry = &sum->entries[0];
1835
1836         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1837                 nrpages = min(last_offset - i, bio_blocks);
1838
1839                 /* readahead node pages */
1840                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
1841
1842                 for (idx = addr; idx < addr + nrpages; idx++) {
1843                         struct page *page = get_tmp_page(sbi, idx);
1844
1845                         rn = F2FS_NODE(page);
1846                         sum_entry->nid = rn->footer.nid;
1847                         sum_entry->version = 0;
1848                         sum_entry->ofs_in_node = 0;
1849                         sum_entry++;
1850                         f2fs_put_page(page, 1);
1851                 }
1852
1853                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
1854                                                         addr + nrpages);
1855         }
1856         return 0;
1857 }
1858
1859 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1860 {
1861         struct f2fs_nm_info *nm_i = NM_I(sbi);
1862         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1863         struct f2fs_summary_block *sum = curseg->sum_blk;
1864         int i;
1865
1866         mutex_lock(&curseg->curseg_mutex);
1867         for (i = 0; i < nats_in_cursum(sum); i++) {
1868                 struct nat_entry *ne;
1869                 struct f2fs_nat_entry raw_ne;
1870                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1871
1872                 raw_ne = nat_in_journal(sum, i);
1873
1874                 ne = __lookup_nat_cache(nm_i, nid);
1875                 if (!ne) {
1876                         ne = grab_nat_entry(nm_i, nid);
1877                         node_info_from_raw_nat(&ne->ni, &raw_ne);
1878                 }
1879                 __set_nat_cache_dirty(nm_i, ne);
1880         }
1881         update_nats_in_cursum(sum, -i);
1882         mutex_unlock(&curseg->curseg_mutex);
1883 }
1884
1885 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1886                                                 struct list_head *head, int max)
1887 {
1888         struct nat_entry_set *cur;
1889
1890         if (nes->entry_cnt >= max)
1891                 goto add_out;
1892
1893         list_for_each_entry(cur, head, set_list) {
1894                 if (cur->entry_cnt >= nes->entry_cnt) {
1895                         list_add(&nes->set_list, cur->set_list.prev);
1896                         return;
1897                 }
1898         }
1899 add_out:
1900         list_add_tail(&nes->set_list, head);
1901 }
1902
1903 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1904                                         struct nat_entry_set *set)
1905 {
1906         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1907         struct f2fs_summary_block *sum = curseg->sum_blk;
1908         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1909         bool to_journal = true;
1910         struct f2fs_nat_block *nat_blk;
1911         struct nat_entry *ne, *cur;
1912         struct page *page = NULL;
1913
1914         /*
1915          * there are two steps to flush nat entries:
1916          * #1, flush nat entries to journal in current hot data summary block.
1917          * #2, flush nat entries to nat page.
1918          */
1919         if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1920                 to_journal = false;
1921
1922         if (to_journal) {
1923                 mutex_lock(&curseg->curseg_mutex);
1924         } else {
1925                 page = get_next_nat_page(sbi, start_nid);
1926                 nat_blk = page_address(page);
1927                 f2fs_bug_on(sbi, !nat_blk);
1928         }
1929
1930         /* flush dirty nats in nat entry set */
1931         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1932                 struct f2fs_nat_entry *raw_ne;
1933                 nid_t nid = nat_get_nid(ne);
1934                 int offset;
1935
1936                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1937                         continue;
1938
1939                 if (to_journal) {
1940                         offset = lookup_journal_in_cursum(sum,
1941                                                         NAT_JOURNAL, nid, 1);
1942                         f2fs_bug_on(sbi, offset < 0);
1943                         raw_ne = &nat_in_journal(sum, offset);
1944                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1945                 } else {
1946                         raw_ne = &nat_blk->entries[nid - start_nid];
1947                 }
1948                 raw_nat_from_node_info(raw_ne, &ne->ni);
1949                 nat_reset_flag(ne);
1950                 __clear_nat_cache_dirty(NM_I(sbi), ne);
1951                 if (nat_get_blkaddr(ne) == NULL_ADDR)
1952                         add_free_nid(sbi, nid, false);
1953         }
1954
1955         if (to_journal)
1956                 mutex_unlock(&curseg->curseg_mutex);
1957         else
1958                 f2fs_put_page(page, 1);
1959
1960         f2fs_bug_on(sbi, set->entry_cnt);
1961
1962         radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1963         kmem_cache_free(nat_entry_set_slab, set);
1964 }
1965
1966 /*
1967  * This function is called during the checkpointing process.
1968  */
1969 void flush_nat_entries(struct f2fs_sb_info *sbi)
1970 {
1971         struct f2fs_nm_info *nm_i = NM_I(sbi);
1972         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1973         struct f2fs_summary_block *sum = curseg->sum_blk;
1974         struct nat_entry_set *setvec[SETVEC_SIZE];
1975         struct nat_entry_set *set, *tmp;
1976         unsigned int found;
1977         nid_t set_idx = 0;
1978         LIST_HEAD(sets);
1979
1980         if (!nm_i->dirty_nat_cnt)
1981                 return;
1982
1983         down_write(&nm_i->nat_tree_lock);
1984
1985         /*
1986          * if there are no enough space in journal to store dirty nat
1987          * entries, remove all entries from journal and merge them
1988          * into nat entry set.
1989          */
1990         if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1991                 remove_nats_in_journal(sbi);
1992
1993         while ((found = __gang_lookup_nat_set(nm_i,
1994                                         set_idx, SETVEC_SIZE, setvec))) {
1995                 unsigned idx;
1996                 set_idx = setvec[found - 1]->set + 1;
1997                 for (idx = 0; idx < found; idx++)
1998                         __adjust_nat_entry_set(setvec[idx], &sets,
1999                                                         MAX_NAT_JENTRIES(sum));
2000         }
2001
2002         /* flush dirty nats in nat entry set */
2003         list_for_each_entry_safe(set, tmp, &sets, set_list)
2004                 __flush_nat_entry_set(sbi, set);
2005
2006         up_write(&nm_i->nat_tree_lock);
2007
2008         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2009 }
2010
2011 static int init_node_manager(struct f2fs_sb_info *sbi)
2012 {
2013         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2014         struct f2fs_nm_info *nm_i = NM_I(sbi);
2015         unsigned char *version_bitmap;
2016         unsigned int nat_segs, nat_blocks;
2017
2018         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2019
2020         /* segment_count_nat includes pair segment so divide to 2. */
2021         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2022         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2023
2024         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2025
2026         /* not used nids: 0, node, meta, (and root counted as valid node) */
2027         nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2028         nm_i->fcnt = 0;
2029         nm_i->nat_cnt = 0;
2030         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2031         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2032
2033         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2034         INIT_LIST_HEAD(&nm_i->free_nid_list);
2035         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2036         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2037         INIT_LIST_HEAD(&nm_i->nat_entries);
2038
2039         mutex_init(&nm_i->build_lock);
2040         spin_lock_init(&nm_i->free_nid_list_lock);
2041         init_rwsem(&nm_i->nat_tree_lock);
2042
2043         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2044         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2045         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2046         if (!version_bitmap)
2047                 return -EFAULT;
2048
2049         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2050                                         GFP_KERNEL);
2051         if (!nm_i->nat_bitmap)
2052                 return -ENOMEM;
2053         return 0;
2054 }
2055
2056 int build_node_manager(struct f2fs_sb_info *sbi)
2057 {
2058         int err;
2059
2060         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2061         if (!sbi->nm_info)
2062                 return -ENOMEM;
2063
2064         err = init_node_manager(sbi);
2065         if (err)
2066                 return err;
2067
2068         build_free_nids(sbi);
2069         return 0;
2070 }
2071
2072 void destroy_node_manager(struct f2fs_sb_info *sbi)
2073 {
2074         struct f2fs_nm_info *nm_i = NM_I(sbi);
2075         struct free_nid *i, *next_i;
2076         struct nat_entry *natvec[NATVEC_SIZE];
2077         struct nat_entry_set *setvec[SETVEC_SIZE];
2078         nid_t nid = 0;
2079         unsigned int found;
2080
2081         if (!nm_i)
2082                 return;
2083
2084         /* destroy free nid list */
2085         spin_lock(&nm_i->free_nid_list_lock);
2086         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2087                 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2088                 __del_from_free_nid_list(nm_i, i);
2089                 nm_i->fcnt--;
2090                 spin_unlock(&nm_i->free_nid_list_lock);
2091                 kmem_cache_free(free_nid_slab, i);
2092                 spin_lock(&nm_i->free_nid_list_lock);
2093         }
2094         f2fs_bug_on(sbi, nm_i->fcnt);
2095         spin_unlock(&nm_i->free_nid_list_lock);
2096
2097         /* destroy nat cache */
2098         down_write(&nm_i->nat_tree_lock);
2099         while ((found = __gang_lookup_nat_cache(nm_i,
2100                                         nid, NATVEC_SIZE, natvec))) {
2101                 unsigned idx;
2102
2103                 nid = nat_get_nid(natvec[found - 1]) + 1;
2104                 for (idx = 0; idx < found; idx++)
2105                         __del_from_nat_cache(nm_i, natvec[idx]);
2106         }
2107         f2fs_bug_on(sbi, nm_i->nat_cnt);
2108
2109         /* destroy nat set cache */
2110         nid = 0;
2111         while ((found = __gang_lookup_nat_set(nm_i,
2112                                         nid, SETVEC_SIZE, setvec))) {
2113                 unsigned idx;
2114
2115                 nid = setvec[found - 1]->set + 1;
2116                 for (idx = 0; idx < found; idx++) {
2117                         /* entry_cnt is not zero, when cp_error was occurred */
2118                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2119                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2120                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2121                 }
2122         }
2123         up_write(&nm_i->nat_tree_lock);
2124
2125         kfree(nm_i->nat_bitmap);
2126         sbi->nm_info = NULL;
2127         kfree(nm_i);
2128 }
2129
2130 int __init create_node_manager_caches(void)
2131 {
2132         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2133                         sizeof(struct nat_entry));
2134         if (!nat_entry_slab)
2135                 goto fail;
2136
2137         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2138                         sizeof(struct free_nid));
2139         if (!free_nid_slab)
2140                 goto destroy_nat_entry;
2141
2142         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2143                         sizeof(struct nat_entry_set));
2144         if (!nat_entry_set_slab)
2145                 goto destroy_free_nid;
2146         return 0;
2147
2148 destroy_free_nid:
2149         kmem_cache_destroy(free_nid_slab);
2150 destroy_nat_entry:
2151         kmem_cache_destroy(nat_entry_slab);
2152 fail:
2153         return -ENOMEM;
2154 }
2155
2156 void destroy_node_manager_caches(void)
2157 {
2158         kmem_cache_destroy(nat_entry_set_slab);
2159         kmem_cache_destroy(free_nid_slab);
2160         kmem_cache_destroy(nat_entry_slab);
2161 }