kernel/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->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->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 bool is_checkpointed_node(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 is_cp = true;
 203
 204         down_read(&nm_i->nat_tree_lock);
 205         e = __lookup_nat_cache(nm_i, nid);
 206         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 207                 is_cp = false;
 208         up_read(&nm_i->nat_tree_lock);
 209         return is_cp;
 210 }
 211
 212 bool has_fsynced_inode(struct f2fs_sb_info *sbi, nid_t ino)
 213 {
 214         struct f2fs_nm_info *nm_i = NM_I(sbi);
 215         struct nat_entry *e;
 216         bool fsynced = false;
 217
 218         down_read(&nm_i->nat_tree_lock);
 219         e = __lookup_nat_cache(nm_i, ino);
 220         if (e && get_nat_flag(e, HAS_FSYNCED_INODE))
 221                 fsynced = true;
 222         up_read(&nm_i->nat_tree_lock);
 223         return fsynced;
 224 }
 225
 226 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 227 {
 228         struct f2fs_nm_info *nm_i = NM_I(sbi);
 229         struct nat_entry *e;
 230         bool need_update = true;
 231
 232         down_read(&nm_i->nat_tree_lock);
 233         e = __lookup_nat_cache(nm_i, ino);
 234         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 235                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 236                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 237                 need_update = false;
 238         up_read(&nm_i->nat_tree_lock);
 239         return need_update;
 240 }
 241
 242 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 243 {
 244         struct nat_entry *new;
 245
 246         new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 247         f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
 248         memset(new, 0, sizeof(struct nat_entry));
 249         nat_set_nid(new, nid);
 250         nat_reset_flag(new);
 251         list_add_tail(&new->list, &nm_i->nat_entries);
 252         nm_i->nat_cnt++;
 253         return new;
 254 }
 255
 256 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 257                                                 struct f2fs_nat_entry *ne)
 258 {
 259         struct nat_entry *e;
 260
 261         down_write(&nm_i->nat_tree_lock);
 262         e = __lookup_nat_cache(nm_i, nid);
 263         if (!e) {
 264                 e = grab_nat_entry(nm_i, nid);
 265                 node_info_from_raw_nat(&e->ni, ne);
 266         }
 267         up_write(&nm_i->nat_tree_lock);
 268 }
 269
 270 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 271                         block_t new_blkaddr, bool fsync_done)
 272 {
 273         struct f2fs_nm_info *nm_i = NM_I(sbi);
 274         struct nat_entry *e;
 275
 276         down_write(&nm_i->nat_tree_lock);
 277         e = __lookup_nat_cache(nm_i, ni->nid);
 278         if (!e) {
 279                 e = grab_nat_entry(nm_i, ni->nid);
 280                 copy_node_info(&e->ni, ni);
 281                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 282         } else if (new_blkaddr == NEW_ADDR) {
 283                 /*
 284                  * when nid is reallocated,
 285                  * previous nat entry can be remained in nat cache.
 286                  * So, reinitialize it with new information.
 287                  */
 288                 copy_node_info(&e->ni, ni);
 289                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 290         }
 291
 292         /* sanity check */
 293         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 294         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 295                         new_blkaddr == NULL_ADDR);
 296         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 297                         new_blkaddr == NEW_ADDR);
 298         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 299                         nat_get_blkaddr(e) != NULL_ADDR &&
 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
 308         /* change address */
 309         nat_set_blkaddr(e, new_blkaddr);
 310         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 311                 set_nat_flag(e, IS_CHECKPOINTED, false);
 312         __set_nat_cache_dirty(nm_i, e);
 313
 314         /* update fsync_mark if its inode nat entry is still alive */
 315         e = __lookup_nat_cache(nm_i, ni->ino);
 316         if (e) {
 317                 if (fsync_done && ni->nid == ni->ino)
 318                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 319                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 320         }
 321         up_write(&nm_i->nat_tree_lock);
 322 }
 323
 324 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 325 {
 326         struct f2fs_nm_info *nm_i = NM_I(sbi);
 327
 328         if (available_free_memory(sbi, NAT_ENTRIES))
 329                 return 0;
 330
 331         down_write(&nm_i->nat_tree_lock);
 332         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 333                 struct nat_entry *ne;
 334                 ne = list_first_entry(&nm_i->nat_entries,
 335                                         struct nat_entry, list);
 336                 __del_from_nat_cache(nm_i, ne);
 337                 nr_shrink--;
 338         }
 339         up_write(&nm_i->nat_tree_lock);
 340         return nr_shrink;
 341 }
 342
 343 /*
 344  * This function always returns success
 345  */
 346 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 347 {
 348         struct f2fs_nm_info *nm_i = NM_I(sbi);
 349         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 350         struct f2fs_summary_block *sum = curseg->sum_blk;
 351         nid_t start_nid = START_NID(nid);
 352         struct f2fs_nat_block *nat_blk;
 353         struct page *page = NULL;
 354         struct f2fs_nat_entry ne;
 355         struct nat_entry *e;
 356         int i;
 357
 358         ni->nid = nid;
 359
 360         /* Check nat cache */
 361         down_read(&nm_i->nat_tree_lock);
 362         e = __lookup_nat_cache(nm_i, nid);
 363         if (e) {
 364                 ni->ino = nat_get_ino(e);
 365                 ni->blk_addr = nat_get_blkaddr(e);
 366                 ni->version = nat_get_version(e);
 367         }
 368         up_read(&nm_i->nat_tree_lock);
 369         if (e)
 370                 return;
 371
 372         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 373
 374         /* Check current segment summary */
 375         mutex_lock(&curseg->curseg_mutex);
 376         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 377         if (i >= 0) {
 378                 ne = nat_in_journal(sum, i);
 379                 node_info_from_raw_nat(ni, &ne);
 380         }
 381         mutex_unlock(&curseg->curseg_mutex);
 382         if (i >= 0)
 383                 goto cache;
 384
 385         /* Fill node_info from nat page */
 386         page = get_current_nat_page(sbi, start_nid);
 387         nat_blk = (struct f2fs_nat_block *)page_address(page);
 388         ne = nat_blk->entries[nid - start_nid];
 389         node_info_from_raw_nat(ni, &ne);
 390         f2fs_put_page(page, 1);
 391 cache:
 392         /* cache nat entry */
 393         cache_nat_entry(NM_I(sbi), nid, &ne);
 394 }
 395
 396 /*
 397  * The maximum depth is four.
 398  * Offset[0] will have raw inode offset.
 399  */
 400 static int get_node_path(struct f2fs_inode_info *fi, long block,
 401                                 int offset[4], unsigned int noffset[4])
 402 {
 403         const long direct_index = ADDRS_PER_INODE(fi);
 404         const long direct_blks = ADDRS_PER_BLOCK;
 405         const long dptrs_per_blk = NIDS_PER_BLOCK;
 406         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 407         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 408         int n = 0;
 409         int level = 0;
 410
 411         noffset[0] = 0;
 412
 413         if (block < direct_index) {
 414                 offset[n] = block;
 415                 goto got;
 416         }
 417         block -= direct_index;
 418         if (block < direct_blks) {
 419                 offset[n++] = NODE_DIR1_BLOCK;
 420                 noffset[n] = 1;
 421                 offset[n] = block;
 422                 level = 1;
 423                 goto got;
 424         }
 425         block -= direct_blks;
 426         if (block < direct_blks) {
 427                 offset[n++] = NODE_DIR2_BLOCK;
 428                 noffset[n] = 2;
 429                 offset[n] = block;
 430                 level = 1;
 431                 goto got;
 432         }
 433         block -= direct_blks;
 434         if (block < indirect_blks) {
 435                 offset[n++] = NODE_IND1_BLOCK;
 436                 noffset[n] = 3;
 437                 offset[n++] = block / direct_blks;
 438                 noffset[n] = 4 + offset[n - 1];
 439                 offset[n] = block % direct_blks;
 440                 level = 2;
 441                 goto got;
 442         }
 443         block -= indirect_blks;
 444         if (block < indirect_blks) {
 445                 offset[n++] = NODE_IND2_BLOCK;
 446                 noffset[n] = 4 + dptrs_per_blk;
 447                 offset[n++] = block / direct_blks;
 448                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 449                 offset[n] = block % direct_blks;
 450                 level = 2;
 451                 goto got;
 452         }
 453         block -= indirect_blks;
 454         if (block < dindirect_blks) {
 455                 offset[n++] = NODE_DIND_BLOCK;
 456                 noffset[n] = 5 + (dptrs_per_blk * 2);
 457                 offset[n++] = block / indirect_blks;
 458                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 459                               offset[n - 1] * (dptrs_per_blk + 1);
 460                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 461                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 462                               offset[n - 2] * (dptrs_per_blk + 1) +
 463                               offset[n - 1];
 464                 offset[n] = block % direct_blks;
 465                 level = 3;
 466                 goto got;
 467         } else {
 468                 BUG();
 469         }
 470 got:
 471         return level;
 472 }
 473
 474 /*
 475  * Caller should call f2fs_put_dnode(dn).
 476  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 477  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 478  * In the case of RDONLY_NODE, we don't need to care about mutex.
 479  */
 480 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 481 {
 482         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 483         struct page *npage[4];
 484         struct page *parent = NULL;
 485         int offset[4];
 486         unsigned int noffset[4];
 487         nid_t nids[4];
 488         int level, i;
 489         int err = 0;
 490
 491         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 492
 493         nids[0] = dn->inode->i_ino;
 494         npage[0] = dn->inode_page;
 495
 496         if (!npage[0]) {
 497                 npage[0] = get_node_page(sbi, nids[0]);
 498                 if (IS_ERR(npage[0]))
 499                         return PTR_ERR(npage[0]);
 500         }
 501
 502         /* if inline_data is set, should not report any block indices */
 503         if (f2fs_has_inline_data(dn->inode) && index) {
 504                 err = -ENOENT;
 505                 f2fs_put_page(npage[0], 1);
 506                 goto release_out;
 507         }
 508
 509         parent = npage[0];
 510         if (level != 0)
 511                 nids[1] = get_nid(parent, offset[0], true);
 512         dn->inode_page = npage[0];
 513         dn->inode_page_locked = true;
 514
 515         /* get indirect or direct nodes */
 516         for (i = 1; i <= level; i++) {
 517                 bool done = false;
 518
 519                 if (!nids[i] && mode == ALLOC_NODE) {
 520                         /* alloc new node */
 521                         if (!alloc_nid(sbi, &(nids[i]))) {
 522                                 err = -ENOSPC;
 523                                 goto release_pages;
 524                         }
 525
 526                         dn->nid = nids[i];
 527                         npage[i] = new_node_page(dn, noffset[i], NULL);
 528                         if (IS_ERR(npage[i])) {
 529                                 alloc_nid_failed(sbi, nids[i]);
 530                                 err = PTR_ERR(npage[i]);
 531                                 goto release_pages;
 532                         }
 533
 534                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 535                         alloc_nid_done(sbi, nids[i]);
 536                         done = true;
 537                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 538                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 539                         if (IS_ERR(npage[i])) {
 540                                 err = PTR_ERR(npage[i]);
 541                                 goto release_pages;
 542                         }
 543                         done = true;
 544                 }
 545                 if (i == 1) {
 546                         dn->inode_page_locked = false;
 547                         unlock_page(parent);
 548                 } else {
 549                         f2fs_put_page(parent, 1);
 550                 }
 551
 552                 if (!done) {
 553                         npage[i] = get_node_page(sbi, nids[i]);
 554                         if (IS_ERR(npage[i])) {
 555                                 err = PTR_ERR(npage[i]);
 556                                 f2fs_put_page(npage[0], 0);
 557                                 goto release_out;
 558                         }
 559                 }
 560                 if (i < level) {
 561                         parent = npage[i];
 562                         nids[i + 1] = get_nid(parent, offset[i], false);
 563                 }
 564         }
 565         dn->nid = nids[level];
 566         dn->ofs_in_node = offset[level];
 567         dn->node_page = npage[level];
 568         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 569         return 0;
 570
 571 release_pages:
 572         f2fs_put_page(parent, 1);
 573         if (i > 1)
 574                 f2fs_put_page(npage[0], 0);
 575 release_out:
 576         dn->inode_page = NULL;
 577         dn->node_page = NULL;
 578         return err;
 579 }
 580
 581 static void truncate_node(struct dnode_of_data *dn)
 582 {
 583         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 584         struct node_info ni;
 585
 586         get_node_info(sbi, dn->nid, &ni);
 587         if (dn->inode->i_blocks == 0) {
 588                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 589                 goto invalidate;
 590         }
 591         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 592
 593         /* Deallocate node address */
 594         invalidate_blocks(sbi, ni.blk_addr);
 595         dec_valid_node_count(sbi, dn->inode);
 596         set_node_addr(sbi, &ni, NULL_ADDR, false);
 597
 598         if (dn->nid == dn->inode->i_ino) {
 599                 remove_orphan_inode(sbi, dn->nid);
 600                 dec_valid_inode_count(sbi);
 601         } else {
 602                 sync_inode_page(dn);
 603         }
 604 invalidate:
 605         clear_node_page_dirty(dn->node_page);
 606         set_sbi_flag(sbi, SBI_IS_DIRTY);
 607
 608         f2fs_put_page(dn->node_page, 1);
 609
 610         invalidate_mapping_pages(NODE_MAPPING(sbi),
 611                         dn->node_page->index, dn->node_page->index);
 612
 613         dn->node_page = NULL;
 614         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 615 }
 616
 617 static int truncate_dnode(struct dnode_of_data *dn)
 618 {
 619         struct page *page;
 620
 621         if (dn->nid == 0)
 622                 return 1;
 623
 624         /* get direct node */
 625         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 626         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 627                 return 1;
 628         else if (IS_ERR(page))
 629                 return PTR_ERR(page);
 630
 631         /* Make dnode_of_data for parameter */
 632         dn->node_page = page;
 633         dn->ofs_in_node = 0;
 634         truncate_data_blocks(dn);
 635         truncate_node(dn);
 636         return 1;
 637 }
 638
 639 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 640                                                 int ofs, int depth)
 641 {
 642         struct dnode_of_data rdn = *dn;
 643         struct page *page;
 644         struct f2fs_node *rn;
 645         nid_t child_nid;
 646         unsigned int child_nofs;
 647         int freed = 0;
 648         int i, ret;
 649
 650         if (dn->nid == 0)
 651                 return NIDS_PER_BLOCK + 1;
 652
 653         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 654
 655         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 656         if (IS_ERR(page)) {
 657                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 658                 return PTR_ERR(page);
 659         }
 660
 661         rn = F2FS_NODE(page);
 662         if (depth < 3) {
 663                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 664                         child_nid = le32_to_cpu(rn->in.nid[i]);
 665                         if (child_nid == 0)
 666                                 continue;
 667                         rdn.nid = child_nid;
 668                         ret = truncate_dnode(&rdn);
 669                         if (ret < 0)
 670                                 goto out_err;
 671                         set_nid(page, i, 0, false);
 672                 }
 673         } else {
 674                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 675                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 676                         child_nid = le32_to_cpu(rn->in.nid[i]);
 677                         if (child_nid == 0) {
 678                                 child_nofs += NIDS_PER_BLOCK + 1;
 679                                 continue;
 680                         }
 681                         rdn.nid = child_nid;
 682                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 683                         if (ret == (NIDS_PER_BLOCK + 1)) {
 684                                 set_nid(page, i, 0, false);
 685                                 child_nofs += ret;
 686                         } else if (ret < 0 && ret != -ENOENT) {
 687                                 goto out_err;
 688                         }
 689                 }
 690                 freed = child_nofs;
 691         }
 692
 693         if (!ofs) {
 694                 /* remove current indirect node */
 695                 dn->node_page = page;
 696                 truncate_node(dn);
 697                 freed++;
 698         } else {
 699                 f2fs_put_page(page, 1);
 700         }
 701         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 702         return freed;
 703
 704 out_err:
 705         f2fs_put_page(page, 1);
 706         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 707         return ret;
 708 }
 709
 710 static int truncate_partial_nodes(struct dnode_of_data *dn,
 711                         struct f2fs_inode *ri, int *offset, int depth)
 712 {
 713         struct page *pages[2];
 714         nid_t nid[3];
 715         nid_t child_nid;
 716         int err = 0;
 717         int i;
 718         int idx = depth - 2;
 719
 720         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 721         if (!nid[0])
 722                 return 0;
 723
 724         /* get indirect nodes in the path */
 725         for (i = 0; i < idx + 1; i++) {
 726                 /* reference count'll be increased */
 727                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 728                 if (IS_ERR(pages[i])) {
 729                         err = PTR_ERR(pages[i]);
 730                         idx = i - 1;
 731                         goto fail;
 732                 }
 733                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 734         }
 735
 736         /* free direct nodes linked to a partial indirect node */
 737         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 738                 child_nid = get_nid(pages[idx], i, false);
 739                 if (!child_nid)
 740                         continue;
 741                 dn->nid = child_nid;
 742                 err = truncate_dnode(dn);
 743                 if (err < 0)
 744                         goto fail;
 745                 set_nid(pages[idx], i, 0, false);
 746         }
 747
 748         if (offset[idx + 1] == 0) {
 749                 dn->node_page = pages[idx];
 750                 dn->nid = nid[idx];
 751                 truncate_node(dn);
 752         } else {
 753                 f2fs_put_page(pages[idx], 1);
 754         }
 755         offset[idx]++;
 756         offset[idx + 1] = 0;
 757         idx--;
 758 fail:
 759         for (i = idx; i >= 0; i--)
 760                 f2fs_put_page(pages[i], 1);
 761
 762         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 763
 764         return err;
 765 }
 766
 767 /*
 768  * All the block addresses of data and nodes should be nullified.
 769  */
 770 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 771 {
 772         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 773         int err = 0, cont = 1;
 774         int level, offset[4], noffset[4];
 775         unsigned int nofs = 0;
 776         struct f2fs_inode *ri;
 777         struct dnode_of_data dn;
 778         struct page *page;
 779
 780         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 781
 782         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 783 restart:
 784         page = get_node_page(sbi, inode->i_ino);
 785         if (IS_ERR(page)) {
 786                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 787                 return PTR_ERR(page);
 788         }
 789
 790         set_new_dnode(&dn, inode, page, NULL, 0);
 791         unlock_page(page);
 792
 793         ri = F2FS_INODE(page);
 794         switch (level) {
 795         case 0:
 796         case 1:
 797                 nofs = noffset[1];
 798                 break;
 799         case 2:
 800                 nofs = noffset[1];
 801                 if (!offset[level - 1])
 802                         goto skip_partial;
 803                 err = truncate_partial_nodes(&dn, ri, offset, level);
 804                 if (err < 0 && err != -ENOENT)
 805                         goto fail;
 806                 nofs += 1 + NIDS_PER_BLOCK;
 807                 break;
 808         case 3:
 809                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 810                 if (!offset[level - 1])
 811                         goto skip_partial;
 812                 err = truncate_partial_nodes(&dn, ri, offset, level);
 813                 if (err < 0 && err != -ENOENT)
 814                         goto fail;
 815                 break;
 816         default:
 817                 BUG();
 818         }
 819
 820 skip_partial:
 821         while (cont) {
 822                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 823                 switch (offset[0]) {
 824                 case NODE_DIR1_BLOCK:
 825                 case NODE_DIR2_BLOCK:
 826                         err = truncate_dnode(&dn);
 827                         break;
 828
 829                 case NODE_IND1_BLOCK:
 830                 case NODE_IND2_BLOCK:
 831                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 832                         break;
 833
 834                 case NODE_DIND_BLOCK:
 835                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 836                         cont = 0;
 837                         break;
 838
 839                 default:
 840                         BUG();
 841                 }
 842                 if (err < 0 && err != -ENOENT)
 843                         goto fail;
 844                 if (offset[1] == 0 &&
 845                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 846                         lock_page(page);
 847                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 848                                 f2fs_put_page(page, 1);
 849                                 goto restart;
 850                         }
 851                         f2fs_wait_on_page_writeback(page, NODE);
 852                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 853                         set_page_dirty(page);
 854                         unlock_page(page);
 855                 }
 856                 offset[1] = 0;
 857                 offset[0]++;
 858                 nofs += err;
 859         }
 860 fail:
 861         f2fs_put_page(page, 0);
 862         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 863         return err > 0 ? 0 : err;
 864 }
 865
 866 int truncate_xattr_node(struct inode *inode, struct page *page)
 867 {
 868         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 869         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 870         struct dnode_of_data dn;
 871         struct page *npage;
 872
 873         if (!nid)
 874                 return 0;
 875
 876         npage = get_node_page(sbi, nid);
 877         if (IS_ERR(npage))
 878                 return PTR_ERR(npage);
 879
 880         F2FS_I(inode)->i_xattr_nid = 0;
 881
 882         /* need to do checkpoint during fsync */
 883         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 884
 885         set_new_dnode(&dn, inode, page, npage, nid);
 886
 887         if (page)
 888                 dn.inode_page_locked = true;
 889         truncate_node(&dn);
 890         return 0;
 891 }
 892
 893 /*
 894  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 895  * f2fs_unlock_op().
 896  */
 897 void remove_inode_page(struct inode *inode)
 898 {
 899         struct dnode_of_data dn;
 900
 901         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 902         if (get_dnode_of_data(&dn, 0, LOOKUP_NODE))
 903                 return;
 904
 905         if (truncate_xattr_node(inode, dn.inode_page)) {
 906                 f2fs_put_dnode(&dn);
 907                 return;
 908         }
 909
 910         /* remove potential inline_data blocks */
 911         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 912                                 S_ISLNK(inode->i_mode))
 913                 truncate_data_blocks_range(&dn, 1);
 914
 915         /* 0 is possible, after f2fs_new_inode() has failed */
 916         f2fs_bug_on(F2FS_I_SB(inode),
 917                         inode->i_blocks != 0 && inode->i_blocks != 1);
 918
 919         /* will put inode & node pages */
 920         truncate_node(&dn);
 921 }
 922
 923 struct page *new_inode_page(struct inode *inode)
 924 {
 925         struct dnode_of_data dn;
 926
 927         /* allocate inode page for new inode */
 928         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 929
 930         /* caller should f2fs_put_page(page, 1); */
 931         return new_node_page(&dn, 0, NULL);
 932 }
 933
 934 struct page *new_node_page(struct dnode_of_data *dn,
 935                                 unsigned int ofs, struct page *ipage)
 936 {
 937         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 938         struct node_info old_ni, new_ni;
 939         struct page *page;
 940         int err;
 941
 942         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 943                 return ERR_PTR(-EPERM);
 944
 945         page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
 946         if (!page)
 947                 return ERR_PTR(-ENOMEM);
 948
 949         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 950                 err = -ENOSPC;
 951                 goto fail;
 952         }
 953
 954         get_node_info(sbi, dn->nid, &old_ni);
 955
 956         /* Reinitialize old_ni with new node page */
 957         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
 958         new_ni = old_ni;
 959         new_ni.ino = dn->inode->i_ino;
 960         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
 961
 962         f2fs_wait_on_page_writeback(page, NODE);
 963         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 964         set_cold_node(dn->inode, page);
 965         SetPageUptodate(page);
 966         set_page_dirty(page);
 967
 968         if (f2fs_has_xattr_block(ofs))
 969                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 970
 971         dn->node_page = page;
 972         if (ipage)
 973                 update_inode(dn->inode, ipage);
 974         else
 975                 sync_inode_page(dn);
 976         if (ofs == 0)
 977                 inc_valid_inode_count(sbi);
 978
 979         return page;
 980
 981 fail:
 982         clear_node_page_dirty(page);
 983         f2fs_put_page(page, 1);
 984         return ERR_PTR(err);
 985 }
 986
 987 /*
 988  * Caller should do after getting the following values.
 989  * 0: f2fs_put_page(page, 0)
 990  * LOCKED_PAGE: f2fs_put_page(page, 1)
 991  * error: nothing
 992  */
 993 static int read_node_page(struct page *page, int rw)
 994 {
 995         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 996         struct node_info ni;
 997         struct f2fs_io_info fio = {
 998                 .type = NODE,
 999                 .rw = rw,
1000         };
1001
1002         get_node_info(sbi, page->index, &ni);
1003
1004         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1005                 ClearPageUptodate(page);
1006                 f2fs_put_page(page, 1);
1007                 return -ENOENT;
1008         }
1009
1010         if (PageUptodate(page))
1011                 return LOCKED_PAGE;
1012
1013         fio.blk_addr = ni.blk_addr;
1014         return f2fs_submit_page_bio(sbi, page, &fio);
1015 }
1016
1017 /*
1018  * Readahead a node page
1019  */
1020 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1021 {
1022         struct page *apage;
1023         int err;
1024
1025         apage = find_get_page(NODE_MAPPING(sbi), nid);
1026         if (apage && PageUptodate(apage)) {
1027                 f2fs_put_page(apage, 0);
1028                 return;
1029         }
1030         f2fs_put_page(apage, 0);
1031
1032         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1033         if (!apage)
1034                 return;
1035
1036         err = read_node_page(apage, READA);
1037         if (err == 0)
1038                 f2fs_put_page(apage, 0);
1039         else if (err == LOCKED_PAGE)
1040                 f2fs_put_page(apage, 1);
1041 }
1042
1043 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1044 {
1045         struct page *page;
1046         int err;
1047 repeat:
1048         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1049         if (!page)
1050                 return ERR_PTR(-ENOMEM);
1051
1052         err = read_node_page(page, READ_SYNC);
1053         if (err < 0)
1054                 return ERR_PTR(err);
1055         else if (err != LOCKED_PAGE)
1056                 lock_page(page);
1057
1058         if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
1059                 ClearPageUptodate(page);
1060                 f2fs_put_page(page, 1);
1061                 return ERR_PTR(-EIO);
1062         }
1063         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1064                 f2fs_put_page(page, 1);
1065                 goto repeat;
1066         }
1067         return page;
1068 }
1069
1070 /*
1071  * Return a locked page for the desired node page.
1072  * And, readahead MAX_RA_NODE number of node pages.
1073  */
1074 struct page *get_node_page_ra(struct page *parent, int start)
1075 {
1076         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1077         struct blk_plug plug;
1078         struct page *page;
1079         int err, i, end;
1080         nid_t nid;
1081
1082         /* First, try getting the desired direct node. */
1083         nid = get_nid(parent, start, false);
1084         if (!nid)
1085                 return ERR_PTR(-ENOENT);
1086 repeat:
1087         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1088         if (!page)
1089                 return ERR_PTR(-ENOMEM);
1090
1091         err = read_node_page(page, READ_SYNC);
1092         if (err < 0)
1093                 return ERR_PTR(err);
1094         else if (err == LOCKED_PAGE)
1095                 goto page_hit;
1096
1097         blk_start_plug(&plug);
1098
1099         /* Then, try readahead for siblings of the desired node */
1100         end = start + MAX_RA_NODE;
1101         end = min(end, NIDS_PER_BLOCK);
1102         for (i = start + 1; i < end; i++) {
1103                 nid = get_nid(parent, i, false);
1104                 if (!nid)
1105                         continue;
1106                 ra_node_page(sbi, nid);
1107         }
1108
1109         blk_finish_plug(&plug);
1110
1111         lock_page(page);
1112         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1113                 f2fs_put_page(page, 1);
1114                 goto repeat;
1115         }
1116 page_hit:
1117         if (unlikely(!PageUptodate(page))) {
1118                 f2fs_put_page(page, 1);
1119                 return ERR_PTR(-EIO);
1120         }
1121         return page;
1122 }
1123
1124 void sync_inode_page(struct dnode_of_data *dn)
1125 {
1126         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1127                 update_inode(dn->inode, dn->node_page);
1128         } else if (dn->inode_page) {
1129                 if (!dn->inode_page_locked)
1130                         lock_page(dn->inode_page);
1131                 update_inode(dn->inode, dn->inode_page);
1132                 if (!dn->inode_page_locked)
1133                         unlock_page(dn->inode_page);
1134         } else {
1135                 update_inode_page(dn->inode);
1136         }
1137 }
1138
1139 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1140                                         struct writeback_control *wbc)
1141 {
1142         pgoff_t index, end;
1143         struct pagevec pvec;
1144         int step = ino ? 2 : 0;
1145         int nwritten = 0, wrote = 0;
1146
1147         pagevec_init(&pvec, 0);
1148
1149 next_step:
1150         index = 0;
1151         end = LONG_MAX;
1152
1153         while (index <= end) {
1154                 int i, nr_pages;
1155                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1156                                 PAGECACHE_TAG_DIRTY,
1157                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1158                 if (nr_pages == 0)
1159                         break;
1160
1161                 for (i = 0; i < nr_pages; i++) {
1162                         struct page *page = pvec.pages[i];
1163
1164                         /*
1165                          * flushing sequence with step:
1166                          * 0. indirect nodes
1167                          * 1. dentry dnodes
1168                          * 2. file dnodes
1169                          */
1170                         if (step == 0 && IS_DNODE(page))
1171                                 continue;
1172                         if (step == 1 && (!IS_DNODE(page) ||
1173                                                 is_cold_node(page)))
1174                                 continue;
1175                         if (step == 2 && (!IS_DNODE(page) ||
1176                                                 !is_cold_node(page)))
1177                                 continue;
1178
1179                         /*
1180                          * If an fsync mode,
1181                          * we should not skip writing node pages.
1182                          */
1183                         if (ino && ino_of_node(page) == ino)
1184                                 lock_page(page);
1185                         else if (!trylock_page(page))
1186                                 continue;
1187
1188                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1189 continue_unlock:
1190                                 unlock_page(page);
1191                                 continue;
1192                         }
1193                         if (ino && ino_of_node(page) != ino)
1194                                 goto continue_unlock;
1195
1196                         if (!PageDirty(page)) {
1197                                 /* someone wrote it for us */
1198                                 goto continue_unlock;
1199                         }
1200
1201                         if (!clear_page_dirty_for_io(page))
1202                                 goto continue_unlock;
1203
1204                         /* called by fsync() */
1205                         if (ino && IS_DNODE(page)) {
1206                                 set_fsync_mark(page, 1);
1207                                 if (IS_INODE(page)) {
1208                                         if (!is_checkpointed_node(sbi, ino) &&
1209                                                 !has_fsynced_inode(sbi, ino))
1210                                                 set_dentry_mark(page, 1);
1211                                         else
1212                                                 set_dentry_mark(page, 0);
1213                                 }
1214                                 nwritten++;
1215                         } else {
1216                                 set_fsync_mark(page, 0);
1217                                 set_dentry_mark(page, 0);
1218                         }
1219
1220                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1221                                 unlock_page(page);
1222                         else
1223                                 wrote++;
1224
1225                         if (--wbc->nr_to_write == 0)
1226                                 break;
1227                 }
1228                 pagevec_release(&pvec);
1229                 cond_resched();
1230
1231                 if (wbc->nr_to_write == 0) {
1232                         step = 2;
1233                         break;
1234                 }
1235         }
1236
1237         if (step < 2) {
1238                 step++;
1239                 goto next_step;
1240         }
1241
1242         if (wrote)
1243                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1244         return nwritten;
1245 }
1246
1247 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1248 {
1249         pgoff_t index = 0, end = LONG_MAX;
1250         struct pagevec pvec;
1251         int ret2 = 0, ret = 0;
1252
1253         pagevec_init(&pvec, 0);
1254
1255         while (index <= end) {
1256                 int i, nr_pages;
1257                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1258                                 PAGECACHE_TAG_WRITEBACK,
1259                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1260                 if (nr_pages == 0)
1261                         break;
1262
1263                 for (i = 0; i < nr_pages; i++) {
1264                         struct page *page = pvec.pages[i];
1265
1266                         /* until radix tree lookup accepts end_index */
1267                         if (unlikely(page->index > end))
1268                                 continue;
1269
1270                         if (ino && ino_of_node(page) == ino) {
1271                                 f2fs_wait_on_page_writeback(page, NODE);
1272                                 if (TestClearPageError(page))
1273                                         ret = -EIO;
1274                         }
1275                 }
1276                 pagevec_release(&pvec);
1277                 cond_resched();
1278         }
1279
1280         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1281                 ret2 = -ENOSPC;
1282         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1283                 ret2 = -EIO;
1284         if (!ret)
1285                 ret = ret2;
1286         return ret;
1287 }
1288
1289 static int f2fs_write_node_page(struct page *page,
1290                                 struct writeback_control *wbc)
1291 {
1292         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1293         nid_t nid;
1294         struct node_info ni;
1295         struct f2fs_io_info fio = {
1296                 .type = NODE,
1297                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1298         };
1299
1300         trace_f2fs_writepage(page, NODE);
1301
1302         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1303                 goto redirty_out;
1304         if (unlikely(f2fs_cp_error(sbi)))
1305                 goto redirty_out;
1306
1307         f2fs_wait_on_page_writeback(page, NODE);
1308
1309         /* get old block addr of this node page */
1310         nid = nid_of_node(page);
1311         f2fs_bug_on(sbi, page->index != nid);
1312
1313         get_node_info(sbi, nid, &ni);
1314
1315         /* This page is already truncated */
1316         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1317                 ClearPageUptodate(page);
1318                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1319                 unlock_page(page);
1320                 return 0;
1321         }
1322
1323         if (wbc->for_reclaim) {
1324                 if (!down_read_trylock(&sbi->node_write))
1325                         goto redirty_out;
1326         } else {
1327                 down_read(&sbi->node_write);
1328         }
1329
1330         set_page_writeback(page);
1331         fio.blk_addr = ni.blk_addr;
1332         write_node_page(sbi, page, nid, &fio);
1333         set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1334         dec_page_count(sbi, F2FS_DIRTY_NODES);
1335         up_read(&sbi->node_write);
1336         unlock_page(page);
1337
1338         if (wbc->for_reclaim)
1339                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1340
1341         return 0;
1342
1343 redirty_out:
1344         redirty_page_for_writepage(wbc, page);
1345         return AOP_WRITEPAGE_ACTIVATE;
1346 }
1347
1348 static int f2fs_write_node_pages(struct address_space *mapping,
1349                             struct writeback_control *wbc)
1350 {
1351         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1352         long diff;
1353
1354         trace_f2fs_writepages(mapping->host, wbc, NODE);
1355
1356         /* balancing f2fs's metadata in background */
1357         f2fs_balance_fs_bg(sbi);
1358
1359         /* collect a number of dirty node pages and write together */
1360         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1361                 goto skip_write;
1362
1363         diff = nr_pages_to_write(sbi, NODE, wbc);
1364         wbc->sync_mode = WB_SYNC_NONE;
1365         sync_node_pages(sbi, 0, wbc);
1366         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1367         return 0;
1368
1369 skip_write:
1370         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1371         return 0;
1372 }
1373
1374 static int f2fs_set_node_page_dirty(struct page *page)
1375 {
1376         trace_f2fs_set_page_dirty(page, NODE);
1377
1378         SetPageUptodate(page);
1379         if (!PageDirty(page)) {
1380                 __set_page_dirty_nobuffers(page);
1381                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1382                 SetPagePrivate(page);
1383                 f2fs_trace_pid(page);
1384                 return 1;
1385         }
1386         return 0;
1387 }
1388
1389 /*
1390  * Structure of the f2fs node operations
1391  */
1392 const struct address_space_operations f2fs_node_aops = {
1393         .writepage      = f2fs_write_node_page,
1394         .writepages     = f2fs_write_node_pages,
1395         .set_page_dirty = f2fs_set_node_page_dirty,
1396         .invalidatepage = f2fs_invalidate_page,
1397         .releasepage    = f2fs_release_page,
1398 };
1399
1400 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1401                                                 nid_t n)
1402 {
1403         return radix_tree_lookup(&nm_i->free_nid_root, n);
1404 }
1405
1406 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1407                                                 struct free_nid *i)
1408 {
1409         list_del(&i->list);
1410         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1411 }
1412
1413 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1414 {
1415         struct f2fs_nm_info *nm_i = NM_I(sbi);
1416         struct free_nid *i;
1417         struct nat_entry *ne;
1418         bool allocated = false;
1419
1420         if (!available_free_memory(sbi, FREE_NIDS))
1421                 return -1;
1422
1423         /* 0 nid should not be used */
1424         if (unlikely(nid == 0))
1425                 return 0;
1426
1427         if (build) {
1428                 /* do not add allocated nids */
1429                 down_read(&nm_i->nat_tree_lock);
1430                 ne = __lookup_nat_cache(nm_i, nid);
1431                 if (ne &&
1432                         (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1433                                 nat_get_blkaddr(ne) != NULL_ADDR))
1434                         allocated = true;
1435                 up_read(&nm_i->nat_tree_lock);
1436                 if (allocated)
1437                         return 0;
1438         }
1439
1440         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1441         i->nid = nid;
1442         i->state = NID_NEW;
1443
1444         if (radix_tree_preload(GFP_NOFS)) {
1445                 kmem_cache_free(free_nid_slab, i);
1446                 return 0;
1447         }
1448
1449         spin_lock(&nm_i->free_nid_list_lock);
1450         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1451                 spin_unlock(&nm_i->free_nid_list_lock);
1452                 radix_tree_preload_end();
1453                 kmem_cache_free(free_nid_slab, i);
1454                 return 0;
1455         }
1456         list_add_tail(&i->list, &nm_i->free_nid_list);
1457         nm_i->fcnt++;
1458         spin_unlock(&nm_i->free_nid_list_lock);
1459         radix_tree_preload_end();
1460         return 1;
1461 }
1462
1463 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1464 {
1465         struct free_nid *i;
1466         bool need_free = false;
1467
1468         spin_lock(&nm_i->free_nid_list_lock);
1469         i = __lookup_free_nid_list(nm_i, nid);
1470         if (i && i->state == NID_NEW) {
1471                 __del_from_free_nid_list(nm_i, i);
1472                 nm_i->fcnt--;
1473                 need_free = true;
1474         }
1475         spin_unlock(&nm_i->free_nid_list_lock);
1476
1477         if (need_free)
1478                 kmem_cache_free(free_nid_slab, i);
1479 }
1480
1481 static void scan_nat_page(struct f2fs_sb_info *sbi,
1482                         struct page *nat_page, nid_t start_nid)
1483 {
1484         struct f2fs_nm_info *nm_i = NM_I(sbi);
1485         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1486         block_t blk_addr;
1487         int i;
1488
1489         i = start_nid % NAT_ENTRY_PER_BLOCK;
1490
1491         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1492
1493                 if (unlikely(start_nid >= nm_i->max_nid))
1494                         break;
1495
1496                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1497                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1498                 if (blk_addr == NULL_ADDR) {
1499                         if (add_free_nid(sbi, start_nid, true) < 0)
1500                                 break;
1501                 }
1502         }
1503 }
1504
1505 static void build_free_nids(struct f2fs_sb_info *sbi)
1506 {
1507         struct f2fs_nm_info *nm_i = NM_I(sbi);
1508         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1509         struct f2fs_summary_block *sum = curseg->sum_blk;
1510         int i = 0;
1511         nid_t nid = nm_i->next_scan_nid;
1512
1513         /* Enough entries */
1514         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1515                 return;
1516
1517         /* readahead nat pages to be scanned */
1518         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1519
1520         while (1) {
1521                 struct page *page = get_current_nat_page(sbi, nid);
1522
1523                 scan_nat_page(sbi, page, nid);
1524                 f2fs_put_page(page, 1);
1525
1526                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1527                 if (unlikely(nid >= nm_i->max_nid))
1528                         nid = 0;
1529
1530                 if (i++ == FREE_NID_PAGES)
1531                         break;
1532         }
1533
1534         /* go to the next free nat pages to find free nids abundantly */
1535         nm_i->next_scan_nid = nid;
1536
1537         /* find free nids from current sum_pages */
1538         mutex_lock(&curseg->curseg_mutex);
1539         for (i = 0; i < nats_in_cursum(sum); i++) {
1540                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1541                 nid = le32_to_cpu(nid_in_journal(sum, i));
1542                 if (addr == NULL_ADDR)
1543                         add_free_nid(sbi, nid, true);
1544                 else
1545                         remove_free_nid(nm_i, nid);
1546         }
1547         mutex_unlock(&curseg->curseg_mutex);
1548 }
1549
1550 /*
1551  * If this function returns success, caller can obtain a new nid
1552  * from second parameter of this function.
1553  * The returned nid could be used ino as well as nid when inode is created.
1554  */
1555 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1556 {
1557         struct f2fs_nm_info *nm_i = NM_I(sbi);
1558         struct free_nid *i = NULL;
1559 retry:
1560         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1561                 return false;
1562
1563         spin_lock(&nm_i->free_nid_list_lock);
1564
1565         /* We should not use stale free nids created by build_free_nids */
1566         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1567                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1568                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1569                         if (i->state == NID_NEW)
1570                                 break;
1571
1572                 f2fs_bug_on(sbi, i->state != NID_NEW);
1573                 *nid = i->nid;
1574                 i->state = NID_ALLOC;
1575                 nm_i->fcnt--;
1576                 spin_unlock(&nm_i->free_nid_list_lock);
1577                 return true;
1578         }
1579         spin_unlock(&nm_i->free_nid_list_lock);
1580
1581         /* Let's scan nat pages and its caches to get free nids */
1582         mutex_lock(&nm_i->build_lock);
1583         build_free_nids(sbi);
1584         mutex_unlock(&nm_i->build_lock);
1585         goto retry;
1586 }
1587
1588 /*
1589  * alloc_nid() should be called prior to this function.
1590  */
1591 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1592 {
1593         struct f2fs_nm_info *nm_i = NM_I(sbi);
1594         struct free_nid *i;
1595
1596         spin_lock(&nm_i->free_nid_list_lock);
1597         i = __lookup_free_nid_list(nm_i, nid);
1598         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1599         __del_from_free_nid_list(nm_i, i);
1600         spin_unlock(&nm_i->free_nid_list_lock);
1601
1602         kmem_cache_free(free_nid_slab, i);
1603 }
1604
1605 /*
1606  * alloc_nid() should be called prior to this function.
1607  */
1608 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1609 {
1610         struct f2fs_nm_info *nm_i = NM_I(sbi);
1611         struct free_nid *i;
1612         bool need_free = false;
1613
1614         if (!nid)
1615                 return;
1616
1617         spin_lock(&nm_i->free_nid_list_lock);
1618         i = __lookup_free_nid_list(nm_i, nid);
1619         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1620         if (!available_free_memory(sbi, FREE_NIDS)) {
1621                 __del_from_free_nid_list(nm_i, i);
1622                 need_free = true;
1623         } else {
1624                 i->state = NID_NEW;
1625                 nm_i->fcnt++;
1626         }
1627         spin_unlock(&nm_i->free_nid_list_lock);
1628
1629         if (need_free)
1630                 kmem_cache_free(free_nid_slab, i);
1631 }
1632
1633 void recover_inline_xattr(struct inode *inode, struct page *page)
1634 {
1635         void *src_addr, *dst_addr;
1636         size_t inline_size;
1637         struct page *ipage;
1638         struct f2fs_inode *ri;
1639
1640         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1641         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1642
1643         ri = F2FS_INODE(page);
1644         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1645                 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1646                 goto update_inode;
1647         }
1648
1649         dst_addr = inline_xattr_addr(ipage);
1650         src_addr = inline_xattr_addr(page);
1651         inline_size = inline_xattr_size(inode);
1652
1653         f2fs_wait_on_page_writeback(ipage, NODE);
1654         memcpy(dst_addr, src_addr, inline_size);
1655 update_inode:
1656         update_inode(inode, ipage);
1657         f2fs_put_page(ipage, 1);
1658 }
1659
1660 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1661 {
1662         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1663         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1664         nid_t new_xnid = nid_of_node(page);
1665         struct node_info ni;
1666
1667         /* 1: invalidate the previous xattr nid */
1668         if (!prev_xnid)
1669                 goto recover_xnid;
1670
1671         /* Deallocate node address */
1672         get_node_info(sbi, prev_xnid, &ni);
1673         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1674         invalidate_blocks(sbi, ni.blk_addr);
1675         dec_valid_node_count(sbi, inode);
1676         set_node_addr(sbi, &ni, NULL_ADDR, false);
1677
1678 recover_xnid:
1679         /* 2: allocate new xattr nid */
1680         if (unlikely(!inc_valid_node_count(sbi, inode)))
1681                 f2fs_bug_on(sbi, 1);
1682
1683         remove_free_nid(NM_I(sbi), new_xnid);
1684         get_node_info(sbi, new_xnid, &ni);
1685         ni.ino = inode->i_ino;
1686         set_node_addr(sbi, &ni, NEW_ADDR, false);
1687         F2FS_I(inode)->i_xattr_nid = new_xnid;
1688
1689         /* 3: update xattr blkaddr */
1690         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1691         set_node_addr(sbi, &ni, blkaddr, false);
1692
1693         update_inode_page(inode);
1694 }
1695
1696 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1697 {
1698         struct f2fs_inode *src, *dst;
1699         nid_t ino = ino_of_node(page);
1700         struct node_info old_ni, new_ni;
1701         struct page *ipage;
1702
1703         get_node_info(sbi, ino, &old_ni);
1704
1705         if (unlikely(old_ni.blk_addr != NULL_ADDR))
1706                 return -EINVAL;
1707
1708         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1709         if (!ipage)
1710                 return -ENOMEM;
1711
1712         /* Should not use this inode from free nid list */
1713         remove_free_nid(NM_I(sbi), ino);
1714
1715         SetPageUptodate(ipage);
1716         fill_node_footer(ipage, ino, ino, 0, true);
1717
1718         src = F2FS_INODE(page);
1719         dst = F2FS_INODE(ipage);
1720
1721         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1722         dst->i_size = 0;
1723         dst->i_blocks = cpu_to_le64(1);
1724         dst->i_links = cpu_to_le32(1);
1725         dst->i_xattr_nid = 0;
1726         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1727
1728         new_ni = old_ni;
1729         new_ni.ino = ino;
1730
1731         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1732                 WARN_ON(1);
1733         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1734         inc_valid_inode_count(sbi);
1735         set_page_dirty(ipage);
1736         f2fs_put_page(ipage, 1);
1737         return 0;
1738 }
1739
1740 int restore_node_summary(struct f2fs_sb_info *sbi,
1741                         unsigned int segno, struct f2fs_summary_block *sum)
1742 {
1743         struct f2fs_node *rn;
1744         struct f2fs_summary *sum_entry;
1745         block_t addr;
1746         int bio_blocks = MAX_BIO_BLOCKS(sbi);
1747         int i, idx, last_offset, nrpages;
1748
1749         /* scan the node segment */
1750         last_offset = sbi->blocks_per_seg;
1751         addr = START_BLOCK(sbi, segno);
1752         sum_entry = &sum->entries[0];
1753
1754         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1755                 nrpages = min(last_offset - i, bio_blocks);
1756
1757                 /* readahead node pages */
1758                 ra_meta_pages(sbi, addr, nrpages, META_POR);
1759
1760                 for (idx = addr; idx < addr + nrpages; idx++) {
1761                         struct page *page = get_meta_page(sbi, idx);
1762
1763                         rn = F2FS_NODE(page);
1764                         sum_entry->nid = rn->footer.nid;
1765                         sum_entry->version = 0;
1766                         sum_entry->ofs_in_node = 0;
1767                         sum_entry++;
1768                         f2fs_put_page(page, 1);
1769                 }
1770
1771                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
1772                                                         addr + nrpages);
1773         }
1774         return 0;
1775 }
1776
1777 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1778 {
1779         struct f2fs_nm_info *nm_i = NM_I(sbi);
1780         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1781         struct f2fs_summary_block *sum = curseg->sum_blk;
1782         int i;
1783
1784         mutex_lock(&curseg->curseg_mutex);
1785         for (i = 0; i < nats_in_cursum(sum); i++) {
1786                 struct nat_entry *ne;
1787                 struct f2fs_nat_entry raw_ne;
1788                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1789
1790                 raw_ne = nat_in_journal(sum, i);
1791
1792                 down_write(&nm_i->nat_tree_lock);
1793                 ne = __lookup_nat_cache(nm_i, nid);
1794                 if (!ne) {
1795                         ne = grab_nat_entry(nm_i, nid);
1796                         node_info_from_raw_nat(&ne->ni, &raw_ne);
1797                 }
1798                 __set_nat_cache_dirty(nm_i, ne);
1799                 up_write(&nm_i->nat_tree_lock);
1800         }
1801         update_nats_in_cursum(sum, -i);
1802         mutex_unlock(&curseg->curseg_mutex);
1803 }
1804
1805 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1806                                                 struct list_head *head, int max)
1807 {
1808         struct nat_entry_set *cur;
1809
1810         if (nes->entry_cnt >= max)
1811                 goto add_out;
1812
1813         list_for_each_entry(cur, head, set_list) {
1814                 if (cur->entry_cnt >= nes->entry_cnt) {
1815                         list_add(&nes->set_list, cur->set_list.prev);
1816                         return;
1817                 }
1818         }
1819 add_out:
1820         list_add_tail(&nes->set_list, head);
1821 }
1822
1823 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1824                                         struct nat_entry_set *set)
1825 {
1826         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1827         struct f2fs_summary_block *sum = curseg->sum_blk;
1828         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1829         bool to_journal = true;
1830         struct f2fs_nat_block *nat_blk;
1831         struct nat_entry *ne, *cur;
1832         struct page *page = NULL;
1833         struct f2fs_nm_info *nm_i = NM_I(sbi);
1834
1835         /*
1836          * there are two steps to flush nat entries:
1837          * #1, flush nat entries to journal in current hot data summary block.
1838          * #2, flush nat entries to nat page.
1839          */
1840         if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1841                 to_journal = false;
1842
1843         if (to_journal) {
1844                 mutex_lock(&curseg->curseg_mutex);
1845         } else {
1846                 page = get_next_nat_page(sbi, start_nid);
1847                 nat_blk = page_address(page);
1848                 f2fs_bug_on(sbi, !nat_blk);
1849         }
1850
1851         /* flush dirty nats in nat entry set */
1852         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1853                 struct f2fs_nat_entry *raw_ne;
1854                 nid_t nid = nat_get_nid(ne);
1855                 int offset;
1856
1857                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1858                         continue;
1859
1860                 if (to_journal) {
1861                         offset = lookup_journal_in_cursum(sum,
1862                                                         NAT_JOURNAL, nid, 1);
1863                         f2fs_bug_on(sbi, offset < 0);
1864                         raw_ne = &nat_in_journal(sum, offset);
1865                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1866                 } else {
1867                         raw_ne = &nat_blk->entries[nid - start_nid];
1868                 }
1869                 raw_nat_from_node_info(raw_ne, &ne->ni);
1870
1871                 down_write(&NM_I(sbi)->nat_tree_lock);
1872                 nat_reset_flag(ne);
1873                 __clear_nat_cache_dirty(NM_I(sbi), ne);
1874                 up_write(&NM_I(sbi)->nat_tree_lock);
1875
1876                 if (nat_get_blkaddr(ne) == NULL_ADDR)
1877                         add_free_nid(sbi, nid, false);
1878         }
1879
1880         if (to_journal)
1881                 mutex_unlock(&curseg->curseg_mutex);
1882         else
1883                 f2fs_put_page(page, 1);
1884
1885         f2fs_bug_on(sbi, set->entry_cnt);
1886
1887         down_write(&nm_i->nat_tree_lock);
1888         radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1889         up_write(&nm_i->nat_tree_lock);
1890         kmem_cache_free(nat_entry_set_slab, set);
1891 }
1892
1893 /*
1894  * This function is called during the checkpointing process.
1895  */
1896 void flush_nat_entries(struct f2fs_sb_info *sbi)
1897 {
1898         struct f2fs_nm_info *nm_i = NM_I(sbi);
1899         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1900         struct f2fs_summary_block *sum = curseg->sum_blk;
1901         struct nat_entry_set *setvec[SETVEC_SIZE];
1902         struct nat_entry_set *set, *tmp;
1903         unsigned int found;
1904         nid_t set_idx = 0;
1905         LIST_HEAD(sets);
1906
1907         if (!nm_i->dirty_nat_cnt)
1908                 return;
1909         /*
1910          * if there are no enough space in journal to store dirty nat
1911          * entries, remove all entries from journal and merge them
1912          * into nat entry set.
1913          */
1914         if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1915                 remove_nats_in_journal(sbi);
1916
1917         down_write(&nm_i->nat_tree_lock);
1918         while ((found = __gang_lookup_nat_set(nm_i,
1919                                         set_idx, SETVEC_SIZE, setvec))) {
1920                 unsigned idx;
1921                 set_idx = setvec[found - 1]->set + 1;
1922                 for (idx = 0; idx < found; idx++)
1923                         __adjust_nat_entry_set(setvec[idx], &sets,
1924                                                         MAX_NAT_JENTRIES(sum));
1925         }
1926         up_write(&nm_i->nat_tree_lock);
1927
1928         /* flush dirty nats in nat entry set */
1929         list_for_each_entry_safe(set, tmp, &sets, set_list)
1930                 __flush_nat_entry_set(sbi, set);
1931
1932         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1933 }
1934
1935 static int init_node_manager(struct f2fs_sb_info *sbi)
1936 {
1937         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1938         struct f2fs_nm_info *nm_i = NM_I(sbi);
1939         unsigned char *version_bitmap;
1940         unsigned int nat_segs, nat_blocks;
1941
1942         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1943
1944         /* segment_count_nat includes pair segment so divide to 2. */
1945         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1946         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1947
1948         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1949
1950         /* not used nids: 0, node, meta, (and root counted as valid node) */
1951         nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
1952         nm_i->fcnt = 0;
1953         nm_i->nat_cnt = 0;
1954         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1955
1956         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1957         INIT_LIST_HEAD(&nm_i->free_nid_list);
1958         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
1959         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
1960         INIT_LIST_HEAD(&nm_i->nat_entries);
1961
1962         mutex_init(&nm_i->build_lock);
1963         spin_lock_init(&nm_i->free_nid_list_lock);
1964         init_rwsem(&nm_i->nat_tree_lock);
1965
1966         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1967         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1968         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1969         if (!version_bitmap)
1970                 return -EFAULT;
1971
1972         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1973                                         GFP_KERNEL);
1974         if (!nm_i->nat_bitmap)
1975                 return -ENOMEM;
1976         return 0;
1977 }
1978
1979 int build_node_manager(struct f2fs_sb_info *sbi)
1980 {
1981         int err;
1982
1983         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1984         if (!sbi->nm_info)
1985                 return -ENOMEM;
1986
1987         err = init_node_manager(sbi);
1988         if (err)
1989                 return err;
1990
1991         build_free_nids(sbi);
1992         return 0;
1993 }
1994
1995 void destroy_node_manager(struct f2fs_sb_info *sbi)
1996 {
1997         struct f2fs_nm_info *nm_i = NM_I(sbi);
1998         struct free_nid *i, *next_i;
1999         struct nat_entry *natvec[NATVEC_SIZE];
2000         struct nat_entry_set *setvec[SETVEC_SIZE];
2001         nid_t nid = 0;
2002         unsigned int found;
2003
2004         if (!nm_i)
2005                 return;
2006
2007         /* destroy free nid list */
2008         spin_lock(&nm_i->free_nid_list_lock);
2009         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2010                 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2011                 __del_from_free_nid_list(nm_i, i);
2012                 nm_i->fcnt--;
2013                 spin_unlock(&nm_i->free_nid_list_lock);
2014                 kmem_cache_free(free_nid_slab, i);
2015                 spin_lock(&nm_i->free_nid_list_lock);
2016         }
2017         f2fs_bug_on(sbi, nm_i->fcnt);
2018         spin_unlock(&nm_i->free_nid_list_lock);
2019
2020         /* destroy nat cache */
2021         down_write(&nm_i->nat_tree_lock);
2022         while ((found = __gang_lookup_nat_cache(nm_i,
2023                                         nid, NATVEC_SIZE, natvec))) {
2024                 unsigned idx;
2025
2026                 nid = nat_get_nid(natvec[found - 1]) + 1;
2027                 for (idx = 0; idx < found; idx++)
2028                         __del_from_nat_cache(nm_i, natvec[idx]);
2029         }
2030         f2fs_bug_on(sbi, nm_i->nat_cnt);
2031
2032         /* destroy nat set cache */
2033         nid = 0;
2034         while ((found = __gang_lookup_nat_set(nm_i,
2035                                         nid, SETVEC_SIZE, setvec))) {
2036                 unsigned idx;
2037
2038                 nid = setvec[found - 1]->set + 1;
2039                 for (idx = 0; idx < found; idx++) {
2040                         /* entry_cnt is not zero, when cp_error was occurred */
2041                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2042                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2043                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2044                 }
2045         }
2046         up_write(&nm_i->nat_tree_lock);
2047
2048         kfree(nm_i->nat_bitmap);
2049         sbi->nm_info = NULL;
2050         kfree(nm_i);
2051 }
2052
2053 int __init create_node_manager_caches(void)
2054 {
2055         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2056                         sizeof(struct nat_entry));
2057         if (!nat_entry_slab)
2058                 goto fail;
2059
2060         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2061                         sizeof(struct free_nid));
2062         if (!free_nid_slab)
2063                 goto destroy_nat_entry;
2064
2065         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2066                         sizeof(struct nat_entry_set));
2067         if (!nat_entry_set_slab)
2068                 goto destroy_free_nid;
2069         return 0;
2070
2071 destroy_free_nid:
2072         kmem_cache_destroy(free_nid_slab);
2073 destroy_nat_entry:
2074         kmem_cache_destroy(nat_entry_slab);
2075 fail:
2076         return -ENOMEM;
2077 }
2078
2079 void destroy_node_manager_caches(void)
2080 {
2081         kmem_cache_destroy(nat_entry_set_slab);
2082         kmem_cache_destroy(free_nid_slab);
2083         kmem_cache_destroy(nat_entry_slab);
2084 }