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 = (atomic_read(&sbi->total_ext_tree) *
  69                                 sizeof(struct extent_tree) +
  70                                 atomic_read(&sbi->total_ext_node) *
  71                                 sizeof(struct extent_node)) >> PAGE_CACHE_SHIFT;
  72                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  73         } else {
  74                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
  75                         return true;
  76         }
  77         return res;
  78 }
  79
  80 static void clear_node_page_dirty(struct page *page)
  81 {
  82         struct address_space *mapping = page->mapping;
  83         unsigned int long flags;
  84
  85         if (PageDirty(page)) {
  86                 spin_lock_irqsave(&mapping->tree_lock, flags);
  87                 radix_tree_tag_clear(&mapping->page_tree,
  88                                 page_index(page),
  89                                 PAGECACHE_TAG_DIRTY);
  90                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  91
  92                 clear_page_dirty_for_io(page);
  93                 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
  94         }
  95         ClearPageUptodate(page);
  96 }
  97
  98 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  99 {
 100         pgoff_t index = current_nat_addr(sbi, nid);
 101         return get_meta_page(sbi, index);
 102 }
 103
 104 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 105 {
 106         struct page *src_page;
 107         struct page *dst_page;
 108         pgoff_t src_off;
 109         pgoff_t dst_off;
 110         void *src_addr;
 111         void *dst_addr;
 112         struct f2fs_nm_info *nm_i = NM_I(sbi);
 113
 114         src_off = current_nat_addr(sbi, nid);
 115         dst_off = next_nat_addr(sbi, src_off);
 116
 117         /* get current nat block page with lock */
 118         src_page = get_meta_page(sbi, src_off);
 119         dst_page = grab_meta_page(sbi, dst_off);
 120         f2fs_bug_on(sbi, PageDirty(src_page));
 121
 122         src_addr = page_address(src_page);
 123         dst_addr = page_address(dst_page);
 124         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
 125         set_page_dirty(dst_page);
 126         f2fs_put_page(src_page, 1);
 127
 128         set_to_next_nat(nm_i, nid);
 129
 130         return dst_page;
 131 }
 132
 133 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 134 {
 135         return radix_tree_lookup(&nm_i->nat_root, n);
 136 }
 137
 138 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 139                 nid_t start, unsigned int nr, struct nat_entry **ep)
 140 {
 141         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 142 }
 143
 144 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 145 {
 146         list_del(&e->list);
 147         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 148         nm_i->nat_cnt--;
 149         kmem_cache_free(nat_entry_slab, e);
 150 }
 151
 152 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 153                                                 struct nat_entry *ne)
 154 {
 155         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 156         struct nat_entry_set *head;
 157
 158         if (get_nat_flag(ne, IS_DIRTY))
 159                 return;
 160
 161         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 162         if (!head) {
 163                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 164
 165                 INIT_LIST_HEAD(&head->entry_list);
 166                 INIT_LIST_HEAD(&head->set_list);
 167                 head->set = set;
 168                 head->entry_cnt = 0;
 169                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 170         }
 171         list_move_tail(&ne->list, &head->entry_list);
 172         nm_i->dirty_nat_cnt++;
 173         head->entry_cnt++;
 174         set_nat_flag(ne, IS_DIRTY, true);
 175 }
 176
 177 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 178                                                 struct nat_entry *ne)
 179 {
 180         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 181         struct nat_entry_set *head;
 182
 183         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 184         if (head) {
 185                 list_move_tail(&ne->list, &nm_i->nat_entries);
 186                 set_nat_flag(ne, IS_DIRTY, false);
 187                 head->entry_cnt--;
 188                 nm_i->dirty_nat_cnt--;
 189         }
 190 }
 191
 192 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 193                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 194 {
 195         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 196                                                         start, nr);
 197 }
 198
 199 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 200 {
 201         struct f2fs_nm_info *nm_i = NM_I(sbi);
 202         struct nat_entry *e;
 203         bool need = false;
 204
 205         down_read(&nm_i->nat_tree_lock);
 206         e = __lookup_nat_cache(nm_i, nid);
 207         if (e) {
 208                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 209                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 210                         need = true;
 211         }
 212         up_read(&nm_i->nat_tree_lock);
 213         return need;
 214 }
 215
 216 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 217 {
 218         struct f2fs_nm_info *nm_i = NM_I(sbi);
 219         struct nat_entry *e;
 220         bool is_cp = true;
 221
 222         down_read(&nm_i->nat_tree_lock);
 223         e = __lookup_nat_cache(nm_i, nid);
 224         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 225                 is_cp = false;
 226         up_read(&nm_i->nat_tree_lock);
 227         return is_cp;
 228 }
 229
 230 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 231 {
 232         struct f2fs_nm_info *nm_i = NM_I(sbi);
 233         struct nat_entry *e;
 234         bool need_update = true;
 235
 236         down_read(&nm_i->nat_tree_lock);
 237         e = __lookup_nat_cache(nm_i, ino);
 238         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 239                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 240                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 241                 need_update = false;
 242         up_read(&nm_i->nat_tree_lock);
 243         return need_update;
 244 }
 245
 246 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 247 {
 248         struct nat_entry *new;
 249
 250         new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
 251         f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
 252         memset(new, 0, sizeof(struct nat_entry));
 253         nat_set_nid(new, nid);
 254         nat_reset_flag(new);
 255         list_add_tail(&new->list, &nm_i->nat_entries);
 256         nm_i->nat_cnt++;
 257         return new;
 258 }
 259
 260 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 261                                                 struct f2fs_nat_entry *ne)
 262 {
 263         struct nat_entry *e;
 264
 265         e = __lookup_nat_cache(nm_i, nid);
 266         if (!e) {
 267                 e = grab_nat_entry(nm_i, nid);
 268                 node_info_from_raw_nat(&e->ni, ne);
 269         }
 270 }
 271
 272 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 273                         block_t new_blkaddr, bool fsync_done)
 274 {
 275         struct f2fs_nm_info *nm_i = NM_I(sbi);
 276         struct nat_entry *e;
 277
 278         down_write(&nm_i->nat_tree_lock);
 279         e = __lookup_nat_cache(nm_i, ni->nid);
 280         if (!e) {
 281                 e = grab_nat_entry(nm_i, ni->nid);
 282                 copy_node_info(&e->ni, ni);
 283                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 284         } else if (new_blkaddr == NEW_ADDR) {
 285                 /*
 286                  * when nid is reallocated,
 287                  * previous nat entry can be remained in nat cache.
 288                  * So, reinitialize it with new information.
 289                  */
 290                 copy_node_info(&e->ni, ni);
 291                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 292         }
 293
 294         /* sanity check */
 295         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 296         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 297                         new_blkaddr == NULL_ADDR);
 298         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 299                         new_blkaddr == NEW_ADDR);
 300         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 301                         nat_get_blkaddr(e) != NULL_ADDR &&
 302                         new_blkaddr == NEW_ADDR);
 303
 304         /* increment version no as node is removed */
 305         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 306                 unsigned char version = nat_get_version(e);
 307                 nat_set_version(e, inc_node_version(version));
 308
 309                 /* in order to reuse the nid */
 310                 if (nm_i->next_scan_nid > ni->nid)
 311                         nm_i->next_scan_nid = ni->nid;
 312         }
 313
 314         /* change address */
 315         nat_set_blkaddr(e, new_blkaddr);
 316         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 317                 set_nat_flag(e, IS_CHECKPOINTED, false);
 318         __set_nat_cache_dirty(nm_i, e);
 319
 320         /* update fsync_mark if its inode nat entry is still alive */
 321         if (ni->nid != ni->ino)
 322                 e = __lookup_nat_cache(nm_i, ni->ino);
 323         if (e) {
 324                 if (fsync_done && ni->nid == ni->ino)
 325                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 326                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 327         }
 328         up_write(&nm_i->nat_tree_lock);
 329 }
 330
 331 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 332 {
 333         struct f2fs_nm_info *nm_i = NM_I(sbi);
 334         int nr = nr_shrink;
 335
 336         if (!down_write_trylock(&nm_i->nat_tree_lock))
 337                 return 0;
 338
 339         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 340                 struct nat_entry *ne;
 341                 ne = list_first_entry(&nm_i->nat_entries,
 342                                         struct nat_entry, list);
 343                 __del_from_nat_cache(nm_i, ne);
 344                 nr_shrink--;
 345         }
 346         up_write(&nm_i->nat_tree_lock);
 347         return nr - nr_shrink;
 348 }
 349
 350 /*
 351  * This function always returns success
 352  */
 353 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 354 {
 355         struct f2fs_nm_info *nm_i = NM_I(sbi);
 356         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 357         struct f2fs_summary_block *sum = curseg->sum_blk;
 358         nid_t start_nid = START_NID(nid);
 359         struct f2fs_nat_block *nat_blk;
 360         struct page *page = NULL;
 361         struct f2fs_nat_entry ne;
 362         struct nat_entry *e;
 363         int i;
 364
 365         ni->nid = nid;
 366
 367         /* Check nat cache */
 368         down_read(&nm_i->nat_tree_lock);
 369         e = __lookup_nat_cache(nm_i, nid);
 370         if (e) {
 371                 ni->ino = nat_get_ino(e);
 372                 ni->blk_addr = nat_get_blkaddr(e);
 373                 ni->version = nat_get_version(e);
 374         }
 375         up_read(&nm_i->nat_tree_lock);
 376         if (e)
 377                 return;
 378
 379         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 380
 381         down_write(&nm_i->nat_tree_lock);
 382
 383         /* Check current segment summary */
 384         mutex_lock(&curseg->curseg_mutex);
 385         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 386         if (i >= 0) {
 387                 ne = nat_in_journal(sum, i);
 388                 node_info_from_raw_nat(ni, &ne);
 389         }
 390         mutex_unlock(&curseg->curseg_mutex);
 391         if (i >= 0)
 392                 goto cache;
 393
 394         /* Fill node_info from nat page */
 395         page = get_current_nat_page(sbi, start_nid);
 396         nat_blk = (struct f2fs_nat_block *)page_address(page);
 397         ne = nat_blk->entries[nid - start_nid];
 398         node_info_from_raw_nat(ni, &ne);
 399         f2fs_put_page(page, 1);
 400 cache:
 401         /* cache nat entry */
 402         cache_nat_entry(NM_I(sbi), nid, &ne);
 403         up_write(&nm_i->nat_tree_lock);
 404 }
 405
 406 /*
 407  * The maximum depth is four.
 408  * Offset[0] will have raw inode offset.
 409  */
 410 static int get_node_path(struct f2fs_inode_info *fi, long block,
 411                                 int offset[4], unsigned int noffset[4])
 412 {
 413         const long direct_index = ADDRS_PER_INODE(fi);
 414         const long direct_blks = ADDRS_PER_BLOCK;
 415         const long dptrs_per_blk = NIDS_PER_BLOCK;
 416         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 417         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 418         int n = 0;
 419         int level = 0;
 420
 421         noffset[0] = 0;
 422
 423         if (block < direct_index) {
 424                 offset[n] = block;
 425                 goto got;
 426         }
 427         block -= direct_index;
 428         if (block < direct_blks) {
 429                 offset[n++] = NODE_DIR1_BLOCK;
 430                 noffset[n] = 1;
 431                 offset[n] = block;
 432                 level = 1;
 433                 goto got;
 434         }
 435         block -= direct_blks;
 436         if (block < direct_blks) {
 437                 offset[n++] = NODE_DIR2_BLOCK;
 438                 noffset[n] = 2;
 439                 offset[n] = block;
 440                 level = 1;
 441                 goto got;
 442         }
 443         block -= direct_blks;
 444         if (block < indirect_blks) {
 445                 offset[n++] = NODE_IND1_BLOCK;
 446                 noffset[n] = 3;
 447                 offset[n++] = block / direct_blks;
 448                 noffset[n] = 4 + offset[n - 1];
 449                 offset[n] = block % direct_blks;
 450                 level = 2;
 451                 goto got;
 452         }
 453         block -= indirect_blks;
 454         if (block < indirect_blks) {
 455                 offset[n++] = NODE_IND2_BLOCK;
 456                 noffset[n] = 4 + dptrs_per_blk;
 457                 offset[n++] = block / direct_blks;
 458                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 459                 offset[n] = block % direct_blks;
 460                 level = 2;
 461                 goto got;
 462         }
 463         block -= indirect_blks;
 464         if (block < dindirect_blks) {
 465                 offset[n++] = NODE_DIND_BLOCK;
 466                 noffset[n] = 5 + (dptrs_per_blk * 2);
 467                 offset[n++] = block / indirect_blks;
 468                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 469                               offset[n - 1] * (dptrs_per_blk + 1);
 470                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 471                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 472                               offset[n - 2] * (dptrs_per_blk + 1) +
 473                               offset[n - 1];
 474                 offset[n] = block % direct_blks;
 475                 level = 3;
 476                 goto got;
 477         } else {
 478                 BUG();
 479         }
 480 got:
 481         return level;
 482 }
 483
 484 /*
 485  * Caller should call f2fs_put_dnode(dn).
 486  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 487  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 488  * In the case of RDONLY_NODE, we don't need to care about mutex.
 489  */
 490 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 491 {
 492         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 493         struct page *npage[4];
 494         struct page *parent = NULL;
 495         int offset[4];
 496         unsigned int noffset[4];
 497         nid_t nids[4];
 498         int level, i;
 499         int err = 0;
 500
 501         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 502
 503         nids[0] = dn->inode->i_ino;
 504         npage[0] = dn->inode_page;
 505
 506         if (!npage[0]) {
 507                 npage[0] = get_node_page(sbi, nids[0]);
 508                 if (IS_ERR(npage[0]))
 509                         return PTR_ERR(npage[0]);
 510         }
 511
 512         /* if inline_data is set, should not report any block indices */
 513         if (f2fs_has_inline_data(dn->inode) && index) {
 514                 err = -ENOENT;
 515                 f2fs_put_page(npage[0], 1);
 516                 goto release_out;
 517         }
 518
 519         parent = npage[0];
 520         if (level != 0)
 521                 nids[1] = get_nid(parent, offset[0], true);
 522         dn->inode_page = npage[0];
 523         dn->inode_page_locked = true;
 524
 525         /* get indirect or direct nodes */
 526         for (i = 1; i <= level; i++) {
 527                 bool done = false;
 528
 529                 if (!nids[i] && mode == ALLOC_NODE) {
 530                         /* alloc new node */
 531                         if (!alloc_nid(sbi, &(nids[i]))) {
 532                                 err = -ENOSPC;
 533                                 goto release_pages;
 534                         }
 535
 536                         dn->nid = nids[i];
 537                         npage[i] = new_node_page(dn, noffset[i], NULL);
 538                         if (IS_ERR(npage[i])) {
 539                                 alloc_nid_failed(sbi, nids[i]);
 540                                 err = PTR_ERR(npage[i]);
 541                                 goto release_pages;
 542                         }
 543
 544                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 545                         alloc_nid_done(sbi, nids[i]);
 546                         done = true;
 547                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 548                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 549                         if (IS_ERR(npage[i])) {
 550                                 err = PTR_ERR(npage[i]);
 551                                 goto release_pages;
 552                         }
 553                         done = true;
 554                 }
 555                 if (i == 1) {
 556                         dn->inode_page_locked = false;
 557                         unlock_page(parent);
 558                 } else {
 559                         f2fs_put_page(parent, 1);
 560                 }
 561
 562                 if (!done) {
 563                         npage[i] = get_node_page(sbi, nids[i]);
 564                         if (IS_ERR(npage[i])) {
 565                                 err = PTR_ERR(npage[i]);
 566                                 f2fs_put_page(npage[0], 0);
 567                                 goto release_out;
 568                         }
 569                 }
 570                 if (i < level) {
 571                         parent = npage[i];
 572                         nids[i + 1] = get_nid(parent, offset[i], false);
 573                 }
 574         }
 575         dn->nid = nids[level];
 576         dn->ofs_in_node = offset[level];
 577         dn->node_page = npage[level];
 578         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 579         return 0;
 580
 581 release_pages:
 582         f2fs_put_page(parent, 1);
 583         if (i > 1)
 584                 f2fs_put_page(npage[0], 0);
 585 release_out:
 586         dn->inode_page = NULL;
 587         dn->node_page = NULL;
 588         return err;
 589 }
 590
 591 static void truncate_node(struct dnode_of_data *dn)
 592 {
 593         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 594         struct node_info ni;
 595
 596         get_node_info(sbi, dn->nid, &ni);
 597         if (dn->inode->i_blocks == 0) {
 598                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 599                 goto invalidate;
 600         }
 601         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 602
 603         /* Deallocate node address */
 604         invalidate_blocks(sbi, ni.blk_addr);
 605         dec_valid_node_count(sbi, dn->inode);
 606         set_node_addr(sbi, &ni, NULL_ADDR, false);
 607
 608         if (dn->nid == dn->inode->i_ino) {
 609                 remove_orphan_inode(sbi, dn->nid);
 610                 dec_valid_inode_count(sbi);
 611         } else {
 612                 sync_inode_page(dn);
 613         }
 614 invalidate:
 615         clear_node_page_dirty(dn->node_page);
 616         set_sbi_flag(sbi, SBI_IS_DIRTY);
 617
 618         f2fs_put_page(dn->node_page, 1);
 619
 620         invalidate_mapping_pages(NODE_MAPPING(sbi),
 621                         dn->node_page->index, dn->node_page->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                         if (set_nid(page, i, 0, false))
 682                                 dn->node_changed = true;
 683                 }
 684         } else {
 685                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 686                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 687                         child_nid = le32_to_cpu(rn->in.nid[i]);
 688                         if (child_nid == 0) {
 689                                 child_nofs += NIDS_PER_BLOCK + 1;
 690                                 continue;
 691                         }
 692                         rdn.nid = child_nid;
 693                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 694                         if (ret == (NIDS_PER_BLOCK + 1)) {
 695                                 if (set_nid(page, i, 0, false))
 696                                         dn->node_changed = true;
 697                                 child_nofs += ret;
 698                         } else if (ret < 0 && ret != -ENOENT) {
 699                                 goto out_err;
 700                         }
 701                 }
 702                 freed = child_nofs;
 703         }
 704
 705         if (!ofs) {
 706                 /* remove current indirect node */
 707                 dn->node_page = page;
 708                 truncate_node(dn);
 709                 freed++;
 710         } else {
 711                 f2fs_put_page(page, 1);
 712         }
 713         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 714         return freed;
 715
 716 out_err:
 717         f2fs_put_page(page, 1);
 718         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 719         return ret;
 720 }
 721
 722 static int truncate_partial_nodes(struct dnode_of_data *dn,
 723                         struct f2fs_inode *ri, int *offset, int depth)
 724 {
 725         struct page *pages[2];
 726         nid_t nid[3];
 727         nid_t child_nid;
 728         int err = 0;
 729         int i;
 730         int idx = depth - 2;
 731
 732         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 733         if (!nid[0])
 734                 return 0;
 735
 736         /* get indirect nodes in the path */
 737         for (i = 0; i < idx + 1; i++) {
 738                 /* reference count'll be increased */
 739                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 740                 if (IS_ERR(pages[i])) {
 741                         err = PTR_ERR(pages[i]);
 742                         idx = i - 1;
 743                         goto fail;
 744                 }
 745                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 746         }
 747
 748         /* free direct nodes linked to a partial indirect node */
 749         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 750                 child_nid = get_nid(pages[idx], i, false);
 751                 if (!child_nid)
 752                         continue;
 753                 dn->nid = child_nid;
 754                 err = truncate_dnode(dn);
 755                 if (err < 0)
 756                         goto fail;
 757                 if (set_nid(pages[idx], i, 0, false))
 758                         dn->node_changed = true;
 759         }
 760
 761         if (offset[idx + 1] == 0) {
 762                 dn->node_page = pages[idx];
 763                 dn->nid = nid[idx];
 764                 truncate_node(dn);
 765         } else {
 766                 f2fs_put_page(pages[idx], 1);
 767         }
 768         offset[idx]++;
 769         offset[idx + 1] = 0;
 770         idx--;
 771 fail:
 772         for (i = idx; i >= 0; i--)
 773                 f2fs_put_page(pages[i], 1);
 774
 775         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 776
 777         return err;
 778 }
 779
 780 /*
 781  * All the block addresses of data and nodes should be nullified.
 782  */
 783 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 784 {
 785         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 786         int err = 0, cont = 1;
 787         int level, offset[4], noffset[4];
 788         unsigned int nofs = 0;
 789         struct f2fs_inode *ri;
 790         struct dnode_of_data dn;
 791         struct page *page;
 792
 793         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 794
 795         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 796 restart:
 797         page = get_node_page(sbi, inode->i_ino);
 798         if (IS_ERR(page)) {
 799                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 800                 return PTR_ERR(page);
 801         }
 802
 803         set_new_dnode(&dn, inode, page, NULL, 0);
 804         unlock_page(page);
 805
 806         ri = F2FS_INODE(page);
 807         switch (level) {
 808         case 0:
 809         case 1:
 810                 nofs = noffset[1];
 811                 break;
 812         case 2:
 813                 nofs = noffset[1];
 814                 if (!offset[level - 1])
 815                         goto skip_partial;
 816                 err = truncate_partial_nodes(&dn, ri, offset, level);
 817                 if (err < 0 && err != -ENOENT)
 818                         goto fail;
 819                 nofs += 1 + NIDS_PER_BLOCK;
 820                 break;
 821         case 3:
 822                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 823                 if (!offset[level - 1])
 824                         goto skip_partial;
 825                 err = truncate_partial_nodes(&dn, ri, offset, level);
 826                 if (err < 0 && err != -ENOENT)
 827                         goto fail;
 828                 break;
 829         default:
 830                 BUG();
 831         }
 832
 833 skip_partial:
 834         while (cont) {
 835                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 836                 switch (offset[0]) {
 837                 case NODE_DIR1_BLOCK:
 838                 case NODE_DIR2_BLOCK:
 839                         err = truncate_dnode(&dn);
 840                         break;
 841
 842                 case NODE_IND1_BLOCK:
 843                 case NODE_IND2_BLOCK:
 844                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 845                         break;
 846
 847                 case NODE_DIND_BLOCK:
 848                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 849                         cont = 0;
 850                         break;
 851
 852                 default:
 853                         BUG();
 854                 }
 855                 if (err < 0 && err != -ENOENT)
 856                         goto fail;
 857                 if (offset[1] == 0 &&
 858                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 859                         lock_page(page);
 860                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
 861                                 f2fs_put_page(page, 1);
 862                                 goto restart;
 863                         }
 864                         f2fs_wait_on_page_writeback(page, NODE);
 865                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 866                         set_page_dirty(page);
 867                         unlock_page(page);
 868                 }
 869                 offset[1] = 0;
 870                 offset[0]++;
 871                 nofs += err;
 872         }
 873 fail:
 874         f2fs_put_page(page, 0);
 875         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 876         return err > 0 ? 0 : err;
 877 }
 878
 879 int truncate_xattr_node(struct inode *inode, struct page *page)
 880 {
 881         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 882         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 883         struct dnode_of_data dn;
 884         struct page *npage;
 885
 886         if (!nid)
 887                 return 0;
 888
 889         npage = get_node_page(sbi, nid);
 890         if (IS_ERR(npage))
 891                 return PTR_ERR(npage);
 892
 893         F2FS_I(inode)->i_xattr_nid = 0;
 894
 895         /* need to do checkpoint during fsync */
 896         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 897
 898         set_new_dnode(&dn, inode, page, npage, nid);
 899
 900         if (page)
 901                 dn.inode_page_locked = true;
 902         truncate_node(&dn);
 903         return 0;
 904 }
 905
 906 /*
 907  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 908  * f2fs_unlock_op().
 909  */
 910 int remove_inode_page(struct inode *inode)
 911 {
 912         struct dnode_of_data dn;
 913         int err;
 914
 915         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 916         err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
 917         if (err)
 918                 return err;
 919
 920         err = truncate_xattr_node(inode, dn.inode_page);
 921         if (err) {
 922                 f2fs_put_dnode(&dn);
 923                 return err;
 924         }
 925
 926         /* remove potential inline_data blocks */
 927         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 928                                 S_ISLNK(inode->i_mode))
 929                 truncate_data_blocks_range(&dn, 1);
 930
 931         /* 0 is possible, after f2fs_new_inode() has failed */
 932         f2fs_bug_on(F2FS_I_SB(inode),
 933                         inode->i_blocks != 0 && inode->i_blocks != 1);
 934
 935         /* will put inode & node pages */
 936         truncate_node(&dn);
 937         return 0;
 938 }
 939
 940 struct page *new_inode_page(struct inode *inode)
 941 {
 942         struct dnode_of_data dn;
 943
 944         /* allocate inode page for new inode */
 945         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 946
 947         /* caller should f2fs_put_page(page, 1); */
 948         return new_node_page(&dn, 0, NULL);
 949 }
 950
 951 struct page *new_node_page(struct dnode_of_data *dn,
 952                                 unsigned int ofs, struct page *ipage)
 953 {
 954         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 955         struct node_info old_ni, new_ni;
 956         struct page *page;
 957         int err;
 958
 959         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 960                 return ERR_PTR(-EPERM);
 961
 962         page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
 963         if (!page)
 964                 return ERR_PTR(-ENOMEM);
 965
 966         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 967                 err = -ENOSPC;
 968                 goto fail;
 969         }
 970
 971         get_node_info(sbi, dn->nid, &old_ni);
 972
 973         /* Reinitialize old_ni with new node page */
 974         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
 975         new_ni = old_ni;
 976         new_ni.ino = dn->inode->i_ino;
 977         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
 978
 979         f2fs_wait_on_page_writeback(page, NODE);
 980         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 981         set_cold_node(dn->inode, page);
 982         SetPageUptodate(page);
 983         if (set_page_dirty(page))
 984                 dn->node_changed = true;
 985
 986         if (f2fs_has_xattr_block(ofs))
 987                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 988
 989         dn->node_page = page;
 990         if (ipage)
 991                 update_inode(dn->inode, ipage);
 992         else
 993                 sync_inode_page(dn);
 994         if (ofs == 0)
 995                 inc_valid_inode_count(sbi);
 996
 997         return page;
 998
 999 fail:
1000         clear_node_page_dirty(page);
1001         f2fs_put_page(page, 1);
1002         return ERR_PTR(err);
1003 }
1004
1005 /*
1006  * Caller should do after getting the following values.
1007  * 0: f2fs_put_page(page, 0)
1008  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1009  */
1010 static int read_node_page(struct page *page, int rw)
1011 {
1012         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1013         struct node_info ni;
1014         struct f2fs_io_info fio = {
1015                 .sbi = sbi,
1016                 .type = NODE,
1017                 .rw = rw,
1018                 .page = page,
1019                 .encrypted_page = NULL,
1020         };
1021
1022         get_node_info(sbi, page->index, &ni);
1023
1024         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1025                 ClearPageUptodate(page);
1026                 return -ENOENT;
1027         }
1028
1029         if (PageUptodate(page))
1030                 return LOCKED_PAGE;
1031
1032         fio.blk_addr = ni.blk_addr;
1033         return f2fs_submit_page_bio(&fio);
1034 }
1035
1036 /*
1037  * Readahead a node page
1038  */
1039 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1040 {
1041         struct page *apage;
1042         int err;
1043
1044         if (!nid)
1045                 return;
1046         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1047
1048         apage = find_get_page(NODE_MAPPING(sbi), nid);
1049         if (apage && PageUptodate(apage)) {
1050                 f2fs_put_page(apage, 0);
1051                 return;
1052         }
1053         f2fs_put_page(apage, 0);
1054
1055         apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1056         if (!apage)
1057                 return;
1058
1059         err = read_node_page(apage, READA);
1060         f2fs_put_page(apage, err ? 1 : 0);
1061 }
1062
1063 /*
1064  * readahead MAX_RA_NODE number of node pages.
1065  */
1066 void ra_node_pages(struct page *parent, int start)
1067 {
1068         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1069         struct blk_plug plug;
1070         int i, end;
1071         nid_t nid;
1072
1073         blk_start_plug(&plug);
1074
1075         /* Then, try readahead for siblings of the desired node */
1076         end = start + MAX_RA_NODE;
1077         end = min(end, NIDS_PER_BLOCK);
1078         for (i = start; i < end; i++) {
1079                 nid = get_nid(parent, i, false);
1080                 ra_node_page(sbi, nid);
1081         }
1082
1083         blk_finish_plug(&plug);
1084 }
1085
1086 struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1087                                         struct page *parent, int start)
1088 {
1089         struct page *page;
1090         int err;
1091
1092         if (!nid)
1093                 return ERR_PTR(-ENOENT);
1094         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1095 repeat:
1096         page = grab_cache_page(NODE_MAPPING(sbi), nid);
1097         if (!page)
1098                 return ERR_PTR(-ENOMEM);
1099
1100         err = read_node_page(page, READ_SYNC);
1101         if (err < 0) {
1102                 f2fs_put_page(page, 1);
1103                 return ERR_PTR(err);
1104         } else if (err == LOCKED_PAGE) {
1105                 goto page_hit;
1106         }
1107
1108         if (parent)
1109                 ra_node_pages(parent, start + 1);
1110
1111         lock_page(page);
1112
1113         if (unlikely(!PageUptodate(page))) {
1114                 f2fs_put_page(page, 1);
1115                 return ERR_PTR(-EIO);
1116         }
1117         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1118                 f2fs_put_page(page, 1);
1119                 goto repeat;
1120         }
1121 page_hit:
1122         f2fs_bug_on(sbi, nid != nid_of_node(page));
1123         return page;
1124 }
1125
1126 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1127 {
1128         return __get_node_page(sbi, nid, NULL, 0);
1129 }
1130
1131 struct page *get_node_page_ra(struct page *parent, int start)
1132 {
1133         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1134         nid_t nid = get_nid(parent, start, false);
1135
1136         return __get_node_page(sbi, nid, parent, start);
1137 }
1138
1139 void sync_inode_page(struct dnode_of_data *dn)
1140 {
1141         int ret = 0;
1142
1143         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1144                 ret = update_inode(dn->inode, dn->node_page);
1145         } else if (dn->inode_page) {
1146                 if (!dn->inode_page_locked)
1147                         lock_page(dn->inode_page);
1148                 ret = update_inode(dn->inode, dn->inode_page);
1149                 if (!dn->inode_page_locked)
1150                         unlock_page(dn->inode_page);
1151         } else {
1152                 ret = update_inode_page(dn->inode);
1153         }
1154         dn->node_changed = ret ? true: false;
1155 }
1156
1157 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1158 {
1159         struct inode *inode;
1160         struct page *page;
1161
1162         /* should flush inline_data before evict_inode */
1163         inode = ilookup(sbi->sb, ino);
1164         if (!inode)
1165                 return;
1166
1167         page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1168         if (!page)
1169                 goto iput_out;
1170
1171         if (!PageUptodate(page))
1172                 goto page_out;
1173
1174         if (!PageDirty(page))
1175                 goto page_out;
1176
1177         if (!clear_page_dirty_for_io(page))
1178                 goto page_out;
1179
1180         if (!f2fs_write_inline_data(inode, page))
1181                 inode_dec_dirty_pages(inode);
1182         else
1183                 set_page_dirty(page);
1184 page_out:
1185         f2fs_put_page(page, 1);
1186 iput_out:
1187         iput(inode);
1188 }
1189
1190 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1191                                         struct writeback_control *wbc)
1192 {
1193         pgoff_t index, end;
1194         struct pagevec pvec;
1195         int step = ino ? 2 : 0;
1196         int nwritten = 0, wrote = 0;
1197
1198         pagevec_init(&pvec, 0);
1199
1200 next_step:
1201         index = 0;
1202         end = LONG_MAX;
1203
1204         while (index <= end) {
1205                 int i, nr_pages;
1206                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1207                                 PAGECACHE_TAG_DIRTY,
1208                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1209                 if (nr_pages == 0)
1210                         break;
1211
1212                 for (i = 0; i < nr_pages; i++) {
1213                         struct page *page = pvec.pages[i];
1214
1215                         if (unlikely(f2fs_cp_error(sbi))) {
1216                                 pagevec_release(&pvec);
1217                                 return -EIO;
1218                         }
1219
1220                         /*
1221                          * flushing sequence with step:
1222                          * 0. indirect nodes
1223                          * 1. dentry dnodes
1224                          * 2. file dnodes
1225                          */
1226                         if (step == 0 && IS_DNODE(page))
1227                                 continue;
1228                         if (step == 1 && (!IS_DNODE(page) ||
1229                                                 is_cold_node(page)))
1230                                 continue;
1231                         if (step == 2 && (!IS_DNODE(page) ||
1232                                                 !is_cold_node(page)))
1233                                 continue;
1234
1235                         /*
1236                          * If an fsync mode,
1237                          * we should not skip writing node pages.
1238                          */
1239                         if (ino && ino_of_node(page) == ino)
1240                                 lock_page(page);
1241                         else if (!trylock_page(page))
1242                                 continue;
1243
1244                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1245 continue_unlock:
1246                                 unlock_page(page);
1247                                 continue;
1248                         }
1249                         if (ino && ino_of_node(page) != ino)
1250                                 goto continue_unlock;
1251
1252                         if (!PageDirty(page)) {
1253                                 /* someone wrote it for us */
1254                                 goto continue_unlock;
1255                         }
1256
1257                         /* flush inline_data */
1258                         if (!ino && is_inline_node(page)) {
1259                                 clear_inline_node(page);
1260                                 unlock_page(page);
1261                                 flush_inline_data(sbi, ino_of_node(page));
1262                                 continue;
1263                         }
1264
1265                         if (!clear_page_dirty_for_io(page))
1266                                 goto continue_unlock;
1267
1268                         /* called by fsync() */
1269                         if (ino && IS_DNODE(page)) {
1270                                 set_fsync_mark(page, 1);
1271                                 if (IS_INODE(page))
1272                                         set_dentry_mark(page,
1273                                                 need_dentry_mark(sbi, ino));
1274                                 nwritten++;
1275                         } else {
1276                                 set_fsync_mark(page, 0);
1277                                 set_dentry_mark(page, 0);
1278                         }
1279
1280                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1281                                 unlock_page(page);
1282                         else
1283                                 wrote++;
1284
1285                         if (--wbc->nr_to_write == 0)
1286                                 break;
1287                 }
1288                 pagevec_release(&pvec);
1289                 cond_resched();
1290
1291                 if (wbc->nr_to_write == 0) {
1292                         step = 2;
1293                         break;
1294                 }
1295         }
1296
1297         if (step < 2) {
1298                 step++;
1299                 goto next_step;
1300         }
1301
1302         if (wrote)
1303                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1304         return nwritten;
1305 }
1306
1307 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1308 {
1309         pgoff_t index = 0, end = LONG_MAX;
1310         struct pagevec pvec;
1311         int ret2 = 0, ret = 0;
1312
1313         pagevec_init(&pvec, 0);
1314
1315         while (index <= end) {
1316                 int i, nr_pages;
1317                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1318                                 PAGECACHE_TAG_WRITEBACK,
1319                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1320                 if (nr_pages == 0)
1321                         break;
1322
1323                 for (i = 0; i < nr_pages; i++) {
1324                         struct page *page = pvec.pages[i];
1325
1326                         /* until radix tree lookup accepts end_index */
1327                         if (unlikely(page->index > end))
1328                                 continue;
1329
1330                         if (ino && ino_of_node(page) == ino) {
1331                                 f2fs_wait_on_page_writeback(page, NODE);
1332                                 if (TestClearPageError(page))
1333                                         ret = -EIO;
1334                         }
1335                 }
1336                 pagevec_release(&pvec);
1337                 cond_resched();
1338         }
1339
1340         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1341                 ret2 = -ENOSPC;
1342         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1343                 ret2 = -EIO;
1344         if (!ret)
1345                 ret = ret2;
1346         return ret;
1347 }
1348
1349 static int f2fs_write_node_page(struct page *page,
1350                                 struct writeback_control *wbc)
1351 {
1352         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1353         nid_t nid;
1354         struct node_info ni;
1355         struct f2fs_io_info fio = {
1356                 .sbi = sbi,
1357                 .type = NODE,
1358                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1359                 .page = page,
1360                 .encrypted_page = NULL,
1361         };
1362
1363         trace_f2fs_writepage(page, NODE);
1364
1365         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1366                 goto redirty_out;
1367         if (unlikely(f2fs_cp_error(sbi)))
1368                 goto redirty_out;
1369
1370         f2fs_wait_on_page_writeback(page, NODE);
1371
1372         /* get old block addr of this node page */
1373         nid = nid_of_node(page);
1374         f2fs_bug_on(sbi, page->index != nid);
1375
1376         if (wbc->for_reclaim) {
1377                 if (!down_read_trylock(&sbi->node_write))
1378                         goto redirty_out;
1379         } else {
1380                 down_read(&sbi->node_write);
1381         }
1382
1383         get_node_info(sbi, nid, &ni);
1384
1385         /* This page is already truncated */
1386         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1387                 ClearPageUptodate(page);
1388                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1389                 up_read(&sbi->node_write);
1390                 unlock_page(page);
1391                 return 0;
1392         }
1393
1394         set_page_writeback(page);
1395         fio.blk_addr = ni.blk_addr;
1396         write_node_page(nid, &fio);
1397         set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1398         dec_page_count(sbi, F2FS_DIRTY_NODES);
1399         up_read(&sbi->node_write);
1400         unlock_page(page);
1401
1402         if (wbc->for_reclaim || unlikely(f2fs_cp_error(sbi)))
1403                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1404
1405         return 0;
1406
1407 redirty_out:
1408         redirty_page_for_writepage(wbc, page);
1409         return AOP_WRITEPAGE_ACTIVATE;
1410 }
1411
1412 static int f2fs_write_node_pages(struct address_space *mapping,
1413                             struct writeback_control *wbc)
1414 {
1415         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1416         long diff;
1417
1418         trace_f2fs_writepages(mapping->host, wbc, NODE);
1419
1420         /* balancing f2fs's metadata in background */
1421         f2fs_balance_fs_bg(sbi);
1422
1423         /* collect a number of dirty node pages and write together */
1424         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1425                 goto skip_write;
1426
1427         diff = nr_pages_to_write(sbi, NODE, wbc);
1428         wbc->sync_mode = WB_SYNC_NONE;
1429         sync_node_pages(sbi, 0, wbc);
1430         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1431         return 0;
1432
1433 skip_write:
1434         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1435         return 0;
1436 }
1437
1438 static int f2fs_set_node_page_dirty(struct page *page)
1439 {
1440         trace_f2fs_set_page_dirty(page, NODE);
1441
1442         SetPageUptodate(page);
1443         if (!PageDirty(page)) {
1444                 __set_page_dirty_nobuffers(page);
1445                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1446                 SetPagePrivate(page);
1447                 f2fs_trace_pid(page);
1448                 return 1;
1449         }
1450         return 0;
1451 }
1452
1453 /*
1454  * Structure of the f2fs node operations
1455  */
1456 const struct address_space_operations f2fs_node_aops = {
1457         .writepage      = f2fs_write_node_page,
1458         .writepages     = f2fs_write_node_pages,
1459         .set_page_dirty = f2fs_set_node_page_dirty,
1460         .invalidatepage = f2fs_invalidate_page,
1461         .releasepage    = f2fs_release_page,
1462 };
1463
1464 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1465                                                 nid_t n)
1466 {
1467         return radix_tree_lookup(&nm_i->free_nid_root, n);
1468 }
1469
1470 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1471                                                 struct free_nid *i)
1472 {
1473         list_del(&i->list);
1474         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1475 }
1476
1477 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1478 {
1479         struct f2fs_nm_info *nm_i = NM_I(sbi);
1480         struct free_nid *i;
1481         struct nat_entry *ne;
1482         bool allocated = false;
1483
1484         if (!available_free_memory(sbi, FREE_NIDS))
1485                 return -1;
1486
1487         /* 0 nid should not be used */
1488         if (unlikely(nid == 0))
1489                 return 0;
1490
1491         if (build) {
1492                 /* do not add allocated nids */
1493                 ne = __lookup_nat_cache(nm_i, nid);
1494                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1495                                 nat_get_blkaddr(ne) != NULL_ADDR))
1496                         allocated = true;
1497                 if (allocated)
1498                         return 0;
1499         }
1500
1501         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1502         i->nid = nid;
1503         i->state = NID_NEW;
1504
1505         if (radix_tree_preload(GFP_NOFS)) {
1506                 kmem_cache_free(free_nid_slab, i);
1507                 return 0;
1508         }
1509
1510         spin_lock(&nm_i->free_nid_list_lock);
1511         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1512                 spin_unlock(&nm_i->free_nid_list_lock);
1513                 radix_tree_preload_end();
1514                 kmem_cache_free(free_nid_slab, i);
1515                 return 0;
1516         }
1517         list_add_tail(&i->list, &nm_i->free_nid_list);
1518         nm_i->fcnt++;
1519         spin_unlock(&nm_i->free_nid_list_lock);
1520         radix_tree_preload_end();
1521         return 1;
1522 }
1523
1524 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1525 {
1526         struct free_nid *i;
1527         bool need_free = false;
1528
1529         spin_lock(&nm_i->free_nid_list_lock);
1530         i = __lookup_free_nid_list(nm_i, nid);
1531         if (i && i->state == NID_NEW) {
1532                 __del_from_free_nid_list(nm_i, i);
1533                 nm_i->fcnt--;
1534                 need_free = true;
1535         }
1536         spin_unlock(&nm_i->free_nid_list_lock);
1537
1538         if (need_free)
1539                 kmem_cache_free(free_nid_slab, i);
1540 }
1541
1542 static void scan_nat_page(struct f2fs_sb_info *sbi,
1543                         struct page *nat_page, nid_t start_nid)
1544 {
1545         struct f2fs_nm_info *nm_i = NM_I(sbi);
1546         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1547         block_t blk_addr;
1548         int i;
1549
1550         i = start_nid % NAT_ENTRY_PER_BLOCK;
1551
1552         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1553
1554                 if (unlikely(start_nid >= nm_i->max_nid))
1555                         break;
1556
1557                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1558                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1559                 if (blk_addr == NULL_ADDR) {
1560                         if (add_free_nid(sbi, start_nid, true) < 0)
1561                                 break;
1562                 }
1563         }
1564 }
1565
1566 static void build_free_nids(struct f2fs_sb_info *sbi)
1567 {
1568         struct f2fs_nm_info *nm_i = NM_I(sbi);
1569         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1570         struct f2fs_summary_block *sum = curseg->sum_blk;
1571         int i = 0;
1572         nid_t nid = nm_i->next_scan_nid;
1573
1574         /* Enough entries */
1575         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1576                 return;
1577
1578         /* readahead nat pages to be scanned */
1579         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1580                                                         META_NAT, true);
1581
1582         down_read(&nm_i->nat_tree_lock);
1583
1584         while (1) {
1585                 struct page *page = get_current_nat_page(sbi, nid);
1586
1587                 scan_nat_page(sbi, page, nid);
1588                 f2fs_put_page(page, 1);
1589
1590                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1591                 if (unlikely(nid >= nm_i->max_nid))
1592                         nid = 0;
1593
1594                 if (++i >= FREE_NID_PAGES)
1595                         break;
1596         }
1597
1598         /* go to the next free nat pages to find free nids abundantly */
1599         nm_i->next_scan_nid = nid;
1600
1601         /* find free nids from current sum_pages */
1602         mutex_lock(&curseg->curseg_mutex);
1603         for (i = 0; i < nats_in_cursum(sum); i++) {
1604                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1605                 nid = le32_to_cpu(nid_in_journal(sum, i));
1606                 if (addr == NULL_ADDR)
1607                         add_free_nid(sbi, nid, true);
1608                 else
1609                         remove_free_nid(nm_i, nid);
1610         }
1611         mutex_unlock(&curseg->curseg_mutex);
1612         up_read(&nm_i->nat_tree_lock);
1613
1614         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1615                                         nm_i->ra_nid_pages, META_NAT, false);
1616 }
1617
1618 /*
1619  * If this function returns success, caller can obtain a new nid
1620  * from second parameter of this function.
1621  * The returned nid could be used ino as well as nid when inode is created.
1622  */
1623 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1624 {
1625         struct f2fs_nm_info *nm_i = NM_I(sbi);
1626         struct free_nid *i = NULL;
1627 retry:
1628         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1629                 return false;
1630
1631         spin_lock(&nm_i->free_nid_list_lock);
1632
1633         /* We should not use stale free nids created by build_free_nids */
1634         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1635                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1636                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1637                         if (i->state == NID_NEW)
1638                                 break;
1639
1640                 f2fs_bug_on(sbi, i->state != NID_NEW);
1641                 *nid = i->nid;
1642                 i->state = NID_ALLOC;
1643                 nm_i->fcnt--;
1644                 spin_unlock(&nm_i->free_nid_list_lock);
1645                 return true;
1646         }
1647         spin_unlock(&nm_i->free_nid_list_lock);
1648
1649         /* Let's scan nat pages and its caches to get free nids */
1650         mutex_lock(&nm_i->build_lock);
1651         build_free_nids(sbi);
1652         mutex_unlock(&nm_i->build_lock);
1653         goto retry;
1654 }
1655
1656 /*
1657  * alloc_nid() should be called prior to this function.
1658  */
1659 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1660 {
1661         struct f2fs_nm_info *nm_i = NM_I(sbi);
1662         struct free_nid *i;
1663
1664         spin_lock(&nm_i->free_nid_list_lock);
1665         i = __lookup_free_nid_list(nm_i, nid);
1666         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1667         __del_from_free_nid_list(nm_i, i);
1668         spin_unlock(&nm_i->free_nid_list_lock);
1669
1670         kmem_cache_free(free_nid_slab, i);
1671 }
1672
1673 /*
1674  * alloc_nid() should be called prior to this function.
1675  */
1676 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1677 {
1678         struct f2fs_nm_info *nm_i = NM_I(sbi);
1679         struct free_nid *i;
1680         bool need_free = false;
1681
1682         if (!nid)
1683                 return;
1684
1685         spin_lock(&nm_i->free_nid_list_lock);
1686         i = __lookup_free_nid_list(nm_i, nid);
1687         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1688         if (!available_free_memory(sbi, FREE_NIDS)) {
1689                 __del_from_free_nid_list(nm_i, i);
1690                 need_free = true;
1691         } else {
1692                 i->state = NID_NEW;
1693                 nm_i->fcnt++;
1694         }
1695         spin_unlock(&nm_i->free_nid_list_lock);
1696
1697         if (need_free)
1698                 kmem_cache_free(free_nid_slab, i);
1699 }
1700
1701 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1702 {
1703         struct f2fs_nm_info *nm_i = NM_I(sbi);
1704         struct free_nid *i, *next;
1705         int nr = nr_shrink;
1706
1707         if (!mutex_trylock(&nm_i->build_lock))
1708                 return 0;
1709
1710         spin_lock(&nm_i->free_nid_list_lock);
1711         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1712                 if (nr_shrink <= 0 || nm_i->fcnt <= NAT_ENTRY_PER_BLOCK)
1713                         break;
1714                 if (i->state == NID_ALLOC)
1715                         continue;
1716                 __del_from_free_nid_list(nm_i, i);
1717                 kmem_cache_free(free_nid_slab, i);
1718                 nm_i->fcnt--;
1719                 nr_shrink--;
1720         }
1721         spin_unlock(&nm_i->free_nid_list_lock);
1722         mutex_unlock(&nm_i->build_lock);
1723
1724         return nr - nr_shrink;
1725 }
1726
1727 void recover_inline_xattr(struct inode *inode, struct page *page)
1728 {
1729         void *src_addr, *dst_addr;
1730         size_t inline_size;
1731         struct page *ipage;
1732         struct f2fs_inode *ri;
1733
1734         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1735         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1736
1737         ri = F2FS_INODE(page);
1738         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1739                 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1740                 goto update_inode;
1741         }
1742
1743         dst_addr = inline_xattr_addr(ipage);
1744         src_addr = inline_xattr_addr(page);
1745         inline_size = inline_xattr_size(inode);
1746
1747         f2fs_wait_on_page_writeback(ipage, NODE);
1748         memcpy(dst_addr, src_addr, inline_size);
1749 update_inode:
1750         update_inode(inode, ipage);
1751         f2fs_put_page(ipage, 1);
1752 }
1753
1754 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1755 {
1756         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1757         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1758         nid_t new_xnid = nid_of_node(page);
1759         struct node_info ni;
1760
1761         /* 1: invalidate the previous xattr nid */
1762         if (!prev_xnid)
1763                 goto recover_xnid;
1764
1765         /* Deallocate node address */
1766         get_node_info(sbi, prev_xnid, &ni);
1767         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1768         invalidate_blocks(sbi, ni.blk_addr);
1769         dec_valid_node_count(sbi, inode);
1770         set_node_addr(sbi, &ni, NULL_ADDR, false);
1771
1772 recover_xnid:
1773         /* 2: allocate new xattr nid */
1774         if (unlikely(!inc_valid_node_count(sbi, inode)))
1775                 f2fs_bug_on(sbi, 1);
1776
1777         remove_free_nid(NM_I(sbi), new_xnid);
1778         get_node_info(sbi, new_xnid, &ni);
1779         ni.ino = inode->i_ino;
1780         set_node_addr(sbi, &ni, NEW_ADDR, false);
1781         F2FS_I(inode)->i_xattr_nid = new_xnid;
1782
1783         /* 3: update xattr blkaddr */
1784         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1785         set_node_addr(sbi, &ni, blkaddr, false);
1786
1787         update_inode_page(inode);
1788 }
1789
1790 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1791 {
1792         struct f2fs_inode *src, *dst;
1793         nid_t ino = ino_of_node(page);
1794         struct node_info old_ni, new_ni;
1795         struct page *ipage;
1796
1797         get_node_info(sbi, ino, &old_ni);
1798
1799         if (unlikely(old_ni.blk_addr != NULL_ADDR))
1800                 return -EINVAL;
1801
1802         ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1803         if (!ipage)
1804                 return -ENOMEM;
1805
1806         /* Should not use this inode from free nid list */
1807         remove_free_nid(NM_I(sbi), ino);
1808
1809         SetPageUptodate(ipage);
1810         fill_node_footer(ipage, ino, ino, 0, true);
1811
1812         src = F2FS_INODE(page);
1813         dst = F2FS_INODE(ipage);
1814
1815         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1816         dst->i_size = 0;
1817         dst->i_blocks = cpu_to_le64(1);
1818         dst->i_links = cpu_to_le32(1);
1819         dst->i_xattr_nid = 0;
1820         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1821
1822         new_ni = old_ni;
1823         new_ni.ino = ino;
1824
1825         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1826                 WARN_ON(1);
1827         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1828         inc_valid_inode_count(sbi);
1829         set_page_dirty(ipage);
1830         f2fs_put_page(ipage, 1);
1831         return 0;
1832 }
1833
1834 int restore_node_summary(struct f2fs_sb_info *sbi,
1835                         unsigned int segno, struct f2fs_summary_block *sum)
1836 {
1837         struct f2fs_node *rn;
1838         struct f2fs_summary *sum_entry;
1839         block_t addr;
1840         int bio_blocks = MAX_BIO_BLOCKS(sbi);
1841         int i, idx, last_offset, nrpages;
1842
1843         /* scan the node segment */
1844         last_offset = sbi->blocks_per_seg;
1845         addr = START_BLOCK(sbi, segno);
1846         sum_entry = &sum->entries[0];
1847
1848         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1849                 nrpages = min(last_offset - i, bio_blocks);
1850
1851                 /* readahead node pages */
1852                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
1853
1854                 for (idx = addr; idx < addr + nrpages; idx++) {
1855                         struct page *page = get_tmp_page(sbi, idx);
1856
1857                         rn = F2FS_NODE(page);
1858                         sum_entry->nid = rn->footer.nid;
1859                         sum_entry->version = 0;
1860                         sum_entry->ofs_in_node = 0;
1861                         sum_entry++;
1862                         f2fs_put_page(page, 1);
1863                 }
1864
1865                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
1866                                                         addr + nrpages);
1867         }
1868         return 0;
1869 }
1870
1871 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1872 {
1873         struct f2fs_nm_info *nm_i = NM_I(sbi);
1874         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1875         struct f2fs_summary_block *sum = curseg->sum_blk;
1876         int i;
1877
1878         mutex_lock(&curseg->curseg_mutex);
1879         for (i = 0; i < nats_in_cursum(sum); i++) {
1880                 struct nat_entry *ne;
1881                 struct f2fs_nat_entry raw_ne;
1882                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1883
1884                 raw_ne = nat_in_journal(sum, i);
1885
1886                 ne = __lookup_nat_cache(nm_i, nid);
1887                 if (!ne) {
1888                         ne = grab_nat_entry(nm_i, nid);
1889                         node_info_from_raw_nat(&ne->ni, &raw_ne);
1890                 }
1891                 __set_nat_cache_dirty(nm_i, ne);
1892         }
1893         update_nats_in_cursum(sum, -i);
1894         mutex_unlock(&curseg->curseg_mutex);
1895 }
1896
1897 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1898                                                 struct list_head *head, int max)
1899 {
1900         struct nat_entry_set *cur;
1901
1902         if (nes->entry_cnt >= max)
1903                 goto add_out;
1904
1905         list_for_each_entry(cur, head, set_list) {
1906                 if (cur->entry_cnt >= nes->entry_cnt) {
1907                         list_add(&nes->set_list, cur->set_list.prev);
1908                         return;
1909                 }
1910         }
1911 add_out:
1912         list_add_tail(&nes->set_list, head);
1913 }
1914
1915 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1916                                         struct nat_entry_set *set)
1917 {
1918         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1919         struct f2fs_summary_block *sum = curseg->sum_blk;
1920         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1921         bool to_journal = true;
1922         struct f2fs_nat_block *nat_blk;
1923         struct nat_entry *ne, *cur;
1924         struct page *page = NULL;
1925
1926         /*
1927          * there are two steps to flush nat entries:
1928          * #1, flush nat entries to journal in current hot data summary block.
1929          * #2, flush nat entries to nat page.
1930          */
1931         if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1932                 to_journal = false;
1933
1934         if (to_journal) {
1935                 mutex_lock(&curseg->curseg_mutex);
1936         } else {
1937                 page = get_next_nat_page(sbi, start_nid);
1938                 nat_blk = page_address(page);
1939                 f2fs_bug_on(sbi, !nat_blk);
1940         }
1941
1942         /* flush dirty nats in nat entry set */
1943         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1944                 struct f2fs_nat_entry *raw_ne;
1945                 nid_t nid = nat_get_nid(ne);
1946                 int offset;
1947
1948                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1949                         continue;
1950
1951                 if (to_journal) {
1952                         offset = lookup_journal_in_cursum(sum,
1953                                                         NAT_JOURNAL, nid, 1);
1954                         f2fs_bug_on(sbi, offset < 0);
1955                         raw_ne = &nat_in_journal(sum, offset);
1956                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1957                 } else {
1958                         raw_ne = &nat_blk->entries[nid - start_nid];
1959                 }
1960                 raw_nat_from_node_info(raw_ne, &ne->ni);
1961                 nat_reset_flag(ne);
1962                 __clear_nat_cache_dirty(NM_I(sbi), ne);
1963                 if (nat_get_blkaddr(ne) == NULL_ADDR)
1964                         add_free_nid(sbi, nid, false);
1965         }
1966
1967         if (to_journal)
1968                 mutex_unlock(&curseg->curseg_mutex);
1969         else
1970                 f2fs_put_page(page, 1);
1971
1972         f2fs_bug_on(sbi, set->entry_cnt);
1973
1974         radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1975         kmem_cache_free(nat_entry_set_slab, set);
1976 }
1977
1978 /*
1979  * This function is called during the checkpointing process.
1980  */
1981 void flush_nat_entries(struct f2fs_sb_info *sbi)
1982 {
1983         struct f2fs_nm_info *nm_i = NM_I(sbi);
1984         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1985         struct f2fs_summary_block *sum = curseg->sum_blk;
1986         struct nat_entry_set *setvec[SETVEC_SIZE];
1987         struct nat_entry_set *set, *tmp;
1988         unsigned int found;
1989         nid_t set_idx = 0;
1990         LIST_HEAD(sets);
1991
1992         if (!nm_i->dirty_nat_cnt)
1993                 return;
1994
1995         down_write(&nm_i->nat_tree_lock);
1996
1997         /*
1998          * if there are no enough space in journal to store dirty nat
1999          * entries, remove all entries from journal and merge them
2000          * into nat entry set.
2001          */
2002         if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2003                 remove_nats_in_journal(sbi);
2004
2005         while ((found = __gang_lookup_nat_set(nm_i,
2006                                         set_idx, SETVEC_SIZE, setvec))) {
2007                 unsigned idx;
2008                 set_idx = setvec[found - 1]->set + 1;
2009                 for (idx = 0; idx < found; idx++)
2010                         __adjust_nat_entry_set(setvec[idx], &sets,
2011                                                         MAX_NAT_JENTRIES(sum));
2012         }
2013
2014         /* flush dirty nats in nat entry set */
2015         list_for_each_entry_safe(set, tmp, &sets, set_list)
2016                 __flush_nat_entry_set(sbi, set);
2017
2018         up_write(&nm_i->nat_tree_lock);
2019
2020         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2021 }
2022
2023 static int init_node_manager(struct f2fs_sb_info *sbi)
2024 {
2025         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2026         struct f2fs_nm_info *nm_i = NM_I(sbi);
2027         unsigned char *version_bitmap;
2028         unsigned int nat_segs, nat_blocks;
2029
2030         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2031
2032         /* segment_count_nat includes pair segment so divide to 2. */
2033         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2034         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2035
2036         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2037
2038         /* not used nids: 0, node, meta, (and root counted as valid node) */
2039         nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2040         nm_i->fcnt = 0;
2041         nm_i->nat_cnt = 0;
2042         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2043         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2044         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2045
2046         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2047         INIT_LIST_HEAD(&nm_i->free_nid_list);
2048         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2049         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2050         INIT_LIST_HEAD(&nm_i->nat_entries);
2051
2052         mutex_init(&nm_i->build_lock);
2053         spin_lock_init(&nm_i->free_nid_list_lock);
2054         init_rwsem(&nm_i->nat_tree_lock);
2055
2056         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2057         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2058         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2059         if (!version_bitmap)
2060                 return -EFAULT;
2061
2062         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2063                                         GFP_KERNEL);
2064         if (!nm_i->nat_bitmap)
2065                 return -ENOMEM;
2066         return 0;
2067 }
2068
2069 int build_node_manager(struct f2fs_sb_info *sbi)
2070 {
2071         int err;
2072
2073         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2074         if (!sbi->nm_info)
2075                 return -ENOMEM;
2076
2077         err = init_node_manager(sbi);
2078         if (err)
2079                 return err;
2080
2081         build_free_nids(sbi);
2082         return 0;
2083 }
2084
2085 void destroy_node_manager(struct f2fs_sb_info *sbi)
2086 {
2087         struct f2fs_nm_info *nm_i = NM_I(sbi);
2088         struct free_nid *i, *next_i;
2089         struct nat_entry *natvec[NATVEC_SIZE];
2090         struct nat_entry_set *setvec[SETVEC_SIZE];
2091         nid_t nid = 0;
2092         unsigned int found;
2093
2094         if (!nm_i)
2095                 return;
2096
2097         /* destroy free nid list */
2098         spin_lock(&nm_i->free_nid_list_lock);
2099         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2100                 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2101                 __del_from_free_nid_list(nm_i, i);
2102                 nm_i->fcnt--;
2103                 spin_unlock(&nm_i->free_nid_list_lock);
2104                 kmem_cache_free(free_nid_slab, i);
2105                 spin_lock(&nm_i->free_nid_list_lock);
2106         }
2107         f2fs_bug_on(sbi, nm_i->fcnt);
2108         spin_unlock(&nm_i->free_nid_list_lock);
2109
2110         /* destroy nat cache */
2111         down_write(&nm_i->nat_tree_lock);
2112         while ((found = __gang_lookup_nat_cache(nm_i,
2113                                         nid, NATVEC_SIZE, natvec))) {
2114                 unsigned idx;
2115
2116                 nid = nat_get_nid(natvec[found - 1]) + 1;
2117                 for (idx = 0; idx < found; idx++)
2118                         __del_from_nat_cache(nm_i, natvec[idx]);
2119         }
2120         f2fs_bug_on(sbi, nm_i->nat_cnt);
2121
2122         /* destroy nat set cache */
2123         nid = 0;
2124         while ((found = __gang_lookup_nat_set(nm_i,
2125                                         nid, SETVEC_SIZE, setvec))) {
2126                 unsigned idx;
2127
2128                 nid = setvec[found - 1]->set + 1;
2129                 for (idx = 0; idx < found; idx++) {
2130                         /* entry_cnt is not zero, when cp_error was occurred */
2131                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2132                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2133                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2134                 }
2135         }
2136         up_write(&nm_i->nat_tree_lock);
2137
2138         kfree(nm_i->nat_bitmap);
2139         sbi->nm_info = NULL;
2140         kfree(nm_i);
2141 }
2142
2143 int __init create_node_manager_caches(void)
2144 {
2145         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2146                         sizeof(struct nat_entry));
2147         if (!nat_entry_slab)
2148                 goto fail;
2149
2150         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2151                         sizeof(struct free_nid));
2152         if (!free_nid_slab)
2153                 goto destroy_nat_entry;
2154
2155         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2156                         sizeof(struct nat_entry_set));
2157         if (!nat_entry_set_slab)
2158                 goto destroy_free_nid;
2159         return 0;
2160
2161 destroy_free_nid:
2162         kmem_cache_destroy(free_nid_slab);
2163 destroy_nat_entry:
2164         kmem_cache_destroy(nat_entry_slab);
2165 fail:
2166         return -ENOMEM;
2167 }
2168
2169 void destroy_node_manager_caches(void)
2170 {
2171         kmem_cache_destroy(nat_entry_set_slab);
2172         kmem_cache_destroy(free_nid_slab);
2173         kmem_cache_destroy(nat_entry_slab);
2174 }