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_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_SHIFT;
  54                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  55                 if (excess_cached_nats(sbi))
  56                         res = false;
  57         } else if (type == DIRTY_DENTS) {
  58                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  59                         return false;
  60                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  61                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  62         } else if (type == INO_ENTRIES) {
  63                 int i;
  64
  65                 for (i = 0; i <= UPDATE_INO; i++)
  66                         mem_size += (sbi->im[i].ino_num *
  67                                 sizeof(struct ino_entry)) >> PAGE_SHIFT;
  68                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  69         } else if (type == EXTENT_CACHE) {
  70                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  71                                 sizeof(struct extent_tree) +
  72                                 atomic_read(&sbi->total_ext_node) *
  73                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  74                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  75         } else {
  76                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
  77                         return true;
  78         }
  79         return res;
  80 }
  81
  82 static void clear_node_page_dirty(struct page *page)
  83 {
  84         struct address_space *mapping = page->mapping;
  85         unsigned int long flags;
  86
  87         if (PageDirty(page)) {
  88                 spin_lock_irqsave(&mapping->tree_lock, flags);
  89                 radix_tree_tag_clear(&mapping->page_tree,
  90                                 page_index(page),
  91                                 PAGECACHE_TAG_DIRTY);
  92                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  93
  94                 clear_page_dirty_for_io(page);
  95                 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
  96         }
  97         ClearPageUptodate(page);
  98 }
  99
 100 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 101 {
 102         pgoff_t index = current_nat_addr(sbi, nid);
 103         return get_meta_page(sbi, index);
 104 }
 105
 106 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 107 {
 108         struct page *src_page;
 109         struct page *dst_page;
 110         pgoff_t src_off;
 111         pgoff_t dst_off;
 112         void *src_addr;
 113         void *dst_addr;
 114         struct f2fs_nm_info *nm_i = NM_I(sbi);
 115
 116         src_off = current_nat_addr(sbi, nid);
 117         dst_off = next_nat_addr(sbi, src_off);
 118
 119         /* get current nat block page with lock */
 120         src_page = get_meta_page(sbi, src_off);
 121         dst_page = grab_meta_page(sbi, dst_off);
 122         f2fs_bug_on(sbi, PageDirty(src_page));
 123
 124         src_addr = page_address(src_page);
 125         dst_addr = page_address(dst_page);
 126         memcpy(dst_addr, src_addr, PAGE_SIZE);
 127         set_page_dirty(dst_page);
 128         f2fs_put_page(src_page, 1);
 129
 130         set_to_next_nat(nm_i, nid);
 131
 132         return dst_page;
 133 }
 134
 135 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 136 {
 137         return radix_tree_lookup(&nm_i->nat_root, n);
 138 }
 139
 140 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 141                 nid_t start, unsigned int nr, struct nat_entry **ep)
 142 {
 143         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 144 }
 145
 146 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 147 {
 148         list_del(&e->list);
 149         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 150         nm_i->nat_cnt--;
 151         kmem_cache_free(nat_entry_slab, e);
 152 }
 153
 154 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 155                                                 struct nat_entry *ne)
 156 {
 157         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 158         struct nat_entry_set *head;
 159
 160         if (get_nat_flag(ne, IS_DIRTY))
 161                 return;
 162
 163         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 164         if (!head) {
 165                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 166
 167                 INIT_LIST_HEAD(&head->entry_list);
 168                 INIT_LIST_HEAD(&head->set_list);
 169                 head->set = set;
 170                 head->entry_cnt = 0;
 171                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 172         }
 173         list_move_tail(&ne->list, &head->entry_list);
 174         nm_i->dirty_nat_cnt++;
 175         head->entry_cnt++;
 176         set_nat_flag(ne, IS_DIRTY, true);
 177 }
 178
 179 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 180                                                 struct nat_entry *ne)
 181 {
 182         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 183         struct nat_entry_set *head;
 184
 185         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 186         if (head) {
 187                 list_move_tail(&ne->list, &nm_i->nat_entries);
 188                 set_nat_flag(ne, IS_DIRTY, false);
 189                 head->entry_cnt--;
 190                 nm_i->dirty_nat_cnt--;
 191         }
 192 }
 193
 194 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 195                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 196 {
 197         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 198                                                         start, nr);
 199 }
 200
 201 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 202 {
 203         struct f2fs_nm_info *nm_i = NM_I(sbi);
 204         struct nat_entry *e;
 205         bool need = false;
 206
 207         down_read(&nm_i->nat_tree_lock);
 208         e = __lookup_nat_cache(nm_i, nid);
 209         if (e) {
 210                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 211                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 212                         need = true;
 213         }
 214         up_read(&nm_i->nat_tree_lock);
 215         return need;
 216 }
 217
 218 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 219 {
 220         struct f2fs_nm_info *nm_i = NM_I(sbi);
 221         struct nat_entry *e;
 222         bool is_cp = true;
 223
 224         down_read(&nm_i->nat_tree_lock);
 225         e = __lookup_nat_cache(nm_i, nid);
 226         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 227                 is_cp = false;
 228         up_read(&nm_i->nat_tree_lock);
 229         return is_cp;
 230 }
 231
 232 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 233 {
 234         struct f2fs_nm_info *nm_i = NM_I(sbi);
 235         struct nat_entry *e;
 236         bool need_update = true;
 237
 238         down_read(&nm_i->nat_tree_lock);
 239         e = __lookup_nat_cache(nm_i, ino);
 240         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 241                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 242                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 243                 need_update = false;
 244         up_read(&nm_i->nat_tree_lock);
 245         return need_update;
 246 }
 247
 248 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 249 {
 250         struct nat_entry *new;
 251
 252         new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
 253         f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
 254         memset(new, 0, sizeof(struct nat_entry));
 255         nat_set_nid(new, nid);
 256         nat_reset_flag(new);
 257         list_add_tail(&new->list, &nm_i->nat_entries);
 258         nm_i->nat_cnt++;
 259         return new;
 260 }
 261
 262 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 263                                                 struct f2fs_nat_entry *ne)
 264 {
 265         struct f2fs_nm_info *nm_i = NM_I(sbi);
 266         struct nat_entry *e;
 267
 268         e = __lookup_nat_cache(nm_i, nid);
 269         if (!e) {
 270                 e = grab_nat_entry(nm_i, nid);
 271                 node_info_from_raw_nat(&e->ni, ne);
 272         } else {
 273                 f2fs_bug_on(sbi, nat_get_ino(e) != ne->ino ||
 274                                 nat_get_blkaddr(e) != ne->block_addr ||
 275                                 nat_get_version(e) != ne->version);
 276         }
 277 }
 278
 279 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 280                         block_t new_blkaddr, bool fsync_done)
 281 {
 282         struct f2fs_nm_info *nm_i = NM_I(sbi);
 283         struct nat_entry *e;
 284
 285         down_write(&nm_i->nat_tree_lock);
 286         e = __lookup_nat_cache(nm_i, ni->nid);
 287         if (!e) {
 288                 e = grab_nat_entry(nm_i, ni->nid);
 289                 copy_node_info(&e->ni, ni);
 290                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 291         } else if (new_blkaddr == NEW_ADDR) {
 292                 /*
 293                  * when nid is reallocated,
 294                  * previous nat entry can be remained in nat cache.
 295                  * So, reinitialize it with new information.
 296                  */
 297                 copy_node_info(&e->ni, ni);
 298                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 299         }
 300
 301         /* sanity check */
 302         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 303         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 304                         new_blkaddr == NULL_ADDR);
 305         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 306                         new_blkaddr == NEW_ADDR);
 307         f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
 308                         nat_get_blkaddr(e) != NULL_ADDR &&
 309                         new_blkaddr == NEW_ADDR);
 310
 311         /* increment version no as node is removed */
 312         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 313                 unsigned char version = nat_get_version(e);
 314                 nat_set_version(e, inc_node_version(version));
 315
 316                 /* in order to reuse the nid */
 317                 if (nm_i->next_scan_nid > ni->nid)
 318                         nm_i->next_scan_nid = ni->nid;
 319         }
 320
 321         /* change address */
 322         nat_set_blkaddr(e, new_blkaddr);
 323         if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
 324                 set_nat_flag(e, IS_CHECKPOINTED, false);
 325         __set_nat_cache_dirty(nm_i, e);
 326
 327         /* update fsync_mark if its inode nat entry is still alive */
 328         if (ni->nid != ni->ino)
 329                 e = __lookup_nat_cache(nm_i, ni->ino);
 330         if (e) {
 331                 if (fsync_done && ni->nid == ni->ino)
 332                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 333                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 334         }
 335         up_write(&nm_i->nat_tree_lock);
 336 }
 337
 338 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 339 {
 340         struct f2fs_nm_info *nm_i = NM_I(sbi);
 341         int nr = nr_shrink;
 342
 343         if (!down_write_trylock(&nm_i->nat_tree_lock))
 344                 return 0;
 345
 346         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 347                 struct nat_entry *ne;
 348                 ne = list_first_entry(&nm_i->nat_entries,
 349                                         struct nat_entry, list);
 350                 __del_from_nat_cache(nm_i, ne);
 351                 nr_shrink--;
 352         }
 353         up_write(&nm_i->nat_tree_lock);
 354         return nr - nr_shrink;
 355 }
 356
 357 /*
 358  * This function always returns success
 359  */
 360 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 361 {
 362         struct f2fs_nm_info *nm_i = NM_I(sbi);
 363         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 364         struct f2fs_journal *journal = curseg->journal;
 365         nid_t start_nid = START_NID(nid);
 366         struct f2fs_nat_block *nat_blk;
 367         struct page *page = NULL;
 368         struct f2fs_nat_entry ne;
 369         struct nat_entry *e;
 370         int i;
 371
 372         ni->nid = nid;
 373
 374         /* Check nat cache */
 375         down_read(&nm_i->nat_tree_lock);
 376         e = __lookup_nat_cache(nm_i, nid);
 377         if (e) {
 378                 ni->ino = nat_get_ino(e);
 379                 ni->blk_addr = nat_get_blkaddr(e);
 380                 ni->version = nat_get_version(e);
 381                 up_read(&nm_i->nat_tree_lock);
 382                 return;
 383         }
 384
 385         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 386
 387         /* Check current segment summary */
 388         down_read(&curseg->journal_rwsem);
 389         i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 390         if (i >= 0) {
 391                 ne = nat_in_journal(journal, i);
 392                 node_info_from_raw_nat(ni, &ne);
 393         }
 394         up_read(&curseg->journal_rwsem);
 395         if (i >= 0)
 396                 goto cache;
 397
 398         /* Fill node_info from nat page */
 399         page = get_current_nat_page(sbi, start_nid);
 400         nat_blk = (struct f2fs_nat_block *)page_address(page);
 401         ne = nat_blk->entries[nid - start_nid];
 402         node_info_from_raw_nat(ni, &ne);
 403         f2fs_put_page(page, 1);
 404 cache:
 405         up_read(&nm_i->nat_tree_lock);
 406         /* cache nat entry */
 407         down_write(&nm_i->nat_tree_lock);
 408         cache_nat_entry(sbi, nid, &ne);
 409         up_write(&nm_i->nat_tree_lock);
 410 }
 411
 412 /*
 413  * readahead MAX_RA_NODE number of node pages.
 414  */
 415 static void ra_node_pages(struct page *parent, int start, int n)
 416 {
 417         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 418         struct blk_plug plug;
 419         int i, end;
 420         nid_t nid;
 421
 422         blk_start_plug(&plug);
 423
 424         /* Then, try readahead for siblings of the desired node */
 425         end = start + n;
 426         end = min(end, NIDS_PER_BLOCK);
 427         for (i = start; i < end; i++) {
 428                 nid = get_nid(parent, i, false);
 429                 ra_node_page(sbi, nid);
 430         }
 431
 432         blk_finish_plug(&plug);
 433 }
 434
 435 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 436 {
 437         const long direct_index = ADDRS_PER_INODE(dn->inode);
 438         const long direct_blks = ADDRS_PER_BLOCK;
 439         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 440         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 441         int cur_level = dn->cur_level;
 442         int max_level = dn->max_level;
 443         pgoff_t base = 0;
 444
 445         if (!dn->max_level)
 446                 return pgofs + 1;
 447
 448         while (max_level-- > cur_level)
 449                 skipped_unit *= NIDS_PER_BLOCK;
 450
 451         switch (dn->max_level) {
 452         case 3:
 453                 base += 2 * indirect_blks;
 454         case 2:
 455                 base += 2 * direct_blks;
 456         case 1:
 457                 base += direct_index;
 458                 break;
 459         default:
 460                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 461         }
 462
 463         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 464 }
 465
 466 /*
 467  * The maximum depth is four.
 468  * Offset[0] will have raw inode offset.
 469  */
 470 static int get_node_path(struct inode *inode, long block,
 471                                 int offset[4], unsigned int noffset[4])
 472 {
 473         const long direct_index = ADDRS_PER_INODE(inode);
 474         const long direct_blks = ADDRS_PER_BLOCK;
 475         const long dptrs_per_blk = NIDS_PER_BLOCK;
 476         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 477         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 478         int n = 0;
 479         int level = 0;
 480
 481         noffset[0] = 0;
 482
 483         if (block < direct_index) {
 484                 offset[n] = block;
 485                 goto got;
 486         }
 487         block -= direct_index;
 488         if (block < direct_blks) {
 489                 offset[n++] = NODE_DIR1_BLOCK;
 490                 noffset[n] = 1;
 491                 offset[n] = block;
 492                 level = 1;
 493                 goto got;
 494         }
 495         block -= direct_blks;
 496         if (block < direct_blks) {
 497                 offset[n++] = NODE_DIR2_BLOCK;
 498                 noffset[n] = 2;
 499                 offset[n] = block;
 500                 level = 1;
 501                 goto got;
 502         }
 503         block -= direct_blks;
 504         if (block < indirect_blks) {
 505                 offset[n++] = NODE_IND1_BLOCK;
 506                 noffset[n] = 3;
 507                 offset[n++] = block / direct_blks;
 508                 noffset[n] = 4 + offset[n - 1];
 509                 offset[n] = block % direct_blks;
 510                 level = 2;
 511                 goto got;
 512         }
 513         block -= indirect_blks;
 514         if (block < indirect_blks) {
 515                 offset[n++] = NODE_IND2_BLOCK;
 516                 noffset[n] = 4 + dptrs_per_blk;
 517                 offset[n++] = block / direct_blks;
 518                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 519                 offset[n] = block % direct_blks;
 520                 level = 2;
 521                 goto got;
 522         }
 523         block -= indirect_blks;
 524         if (block < dindirect_blks) {
 525                 offset[n++] = NODE_DIND_BLOCK;
 526                 noffset[n] = 5 + (dptrs_per_blk * 2);
 527                 offset[n++] = block / indirect_blks;
 528                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 529                               offset[n - 1] * (dptrs_per_blk + 1);
 530                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 531                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 532                               offset[n - 2] * (dptrs_per_blk + 1) +
 533                               offset[n - 1];
 534                 offset[n] = block % direct_blks;
 535                 level = 3;
 536                 goto got;
 537         } else {
 538                 BUG();
 539         }
 540 got:
 541         return level;
 542 }
 543
 544 /*
 545  * Caller should call f2fs_put_dnode(dn).
 546  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 547  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 548  * In the case of RDONLY_NODE, we don't need to care about mutex.
 549  */
 550 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 551 {
 552         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 553         struct page *npage[4];
 554         struct page *parent = NULL;
 555         int offset[4];
 556         unsigned int noffset[4];
 557         nid_t nids[4];
 558         int level, i = 0;
 559         int err = 0;
 560
 561         level = get_node_path(dn->inode, index, offset, noffset);
 562
 563         nids[0] = dn->inode->i_ino;
 564         npage[0] = dn->inode_page;
 565
 566         if (!npage[0]) {
 567                 npage[0] = get_node_page(sbi, nids[0]);
 568                 if (IS_ERR(npage[0]))
 569                         return PTR_ERR(npage[0]);
 570         }
 571
 572         /* if inline_data is set, should not report any block indices */
 573         if (f2fs_has_inline_data(dn->inode) && index) {
 574                 err = -ENOENT;
 575                 f2fs_put_page(npage[0], 1);
 576                 goto release_out;
 577         }
 578
 579         parent = npage[0];
 580         if (level != 0)
 581                 nids[1] = get_nid(parent, offset[0], true);
 582         dn->inode_page = npage[0];
 583         dn->inode_page_locked = true;
 584
 585         /* get indirect or direct nodes */
 586         for (i = 1; i <= level; i++) {
 587                 bool done = false;
 588
 589                 if (!nids[i] && mode == ALLOC_NODE) {
 590                         /* alloc new node */
 591                         if (!alloc_nid(sbi, &(nids[i]))) {
 592                                 err = -ENOSPC;
 593                                 goto release_pages;
 594                         }
 595
 596                         dn->nid = nids[i];
 597                         npage[i] = new_node_page(dn, noffset[i], NULL);
 598                         if (IS_ERR(npage[i])) {
 599                                 alloc_nid_failed(sbi, nids[i]);
 600                                 err = PTR_ERR(npage[i]);
 601                                 goto release_pages;
 602                         }
 603
 604                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 605                         alloc_nid_done(sbi, nids[i]);
 606                         done = true;
 607                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 608                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 609                         if (IS_ERR(npage[i])) {
 610                                 err = PTR_ERR(npage[i]);
 611                                 goto release_pages;
 612                         }
 613                         done = true;
 614                 }
 615                 if (i == 1) {
 616                         dn->inode_page_locked = false;
 617                         unlock_page(parent);
 618                 } else {
 619                         f2fs_put_page(parent, 1);
 620                 }
 621
 622                 if (!done) {
 623                         npage[i] = get_node_page(sbi, nids[i]);
 624                         if (IS_ERR(npage[i])) {
 625                                 err = PTR_ERR(npage[i]);
 626                                 f2fs_put_page(npage[0], 0);
 627                                 goto release_out;
 628                         }
 629                 }
 630                 if (i < level) {
 631                         parent = npage[i];
 632                         nids[i + 1] = get_nid(parent, offset[i], false);
 633                 }
 634         }
 635         dn->nid = nids[level];
 636         dn->ofs_in_node = offset[level];
 637         dn->node_page = npage[level];
 638         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 639         return 0;
 640
 641 release_pages:
 642         f2fs_put_page(parent, 1);
 643         if (i > 1)
 644                 f2fs_put_page(npage[0], 0);
 645 release_out:
 646         dn->inode_page = NULL;
 647         dn->node_page = NULL;
 648         if (err == -ENOENT) {
 649                 dn->cur_level = i;
 650                 dn->max_level = level;
 651                 dn->ofs_in_node = offset[level];
 652         }
 653         return err;
 654 }
 655
 656 static void truncate_node(struct dnode_of_data *dn)
 657 {
 658         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 659         struct node_info ni;
 660
 661         get_node_info(sbi, dn->nid, &ni);
 662         if (dn->inode->i_blocks == 0) {
 663                 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
 664                 goto invalidate;
 665         }
 666         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
 667
 668         /* Deallocate node address */
 669         invalidate_blocks(sbi, ni.blk_addr);
 670         dec_valid_node_count(sbi, dn->inode);
 671         set_node_addr(sbi, &ni, NULL_ADDR, false);
 672
 673         if (dn->nid == dn->inode->i_ino) {
 674                 remove_orphan_inode(sbi, dn->nid);
 675                 dec_valid_inode_count(sbi);
 676                 f2fs_inode_synced(dn->inode);
 677         }
 678 invalidate:
 679         clear_node_page_dirty(dn->node_page);
 680         set_sbi_flag(sbi, SBI_IS_DIRTY);
 681
 682         f2fs_put_page(dn->node_page, 1);
 683
 684         invalidate_mapping_pages(NODE_MAPPING(sbi),
 685                         dn->node_page->index, dn->node_page->index);
 686
 687         dn->node_page = NULL;
 688         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 689 }
 690
 691 static int truncate_dnode(struct dnode_of_data *dn)
 692 {
 693         struct page *page;
 694
 695         if (dn->nid == 0)
 696                 return 1;
 697
 698         /* get direct node */
 699         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 700         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 701                 return 1;
 702         else if (IS_ERR(page))
 703                 return PTR_ERR(page);
 704
 705         /* Make dnode_of_data for parameter */
 706         dn->node_page = page;
 707         dn->ofs_in_node = 0;
 708         truncate_data_blocks(dn);
 709         truncate_node(dn);
 710         return 1;
 711 }
 712
 713 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 714                                                 int ofs, int depth)
 715 {
 716         struct dnode_of_data rdn = *dn;
 717         struct page *page;
 718         struct f2fs_node *rn;
 719         nid_t child_nid;
 720         unsigned int child_nofs;
 721         int freed = 0;
 722         int i, ret;
 723
 724         if (dn->nid == 0)
 725                 return NIDS_PER_BLOCK + 1;
 726
 727         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 728
 729         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 730         if (IS_ERR(page)) {
 731                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 732                 return PTR_ERR(page);
 733         }
 734
 735         ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 736
 737         rn = F2FS_NODE(page);
 738         if (depth < 3) {
 739                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 740                         child_nid = le32_to_cpu(rn->in.nid[i]);
 741                         if (child_nid == 0)
 742                                 continue;
 743                         rdn.nid = child_nid;
 744                         ret = truncate_dnode(&rdn);
 745                         if (ret < 0)
 746                                 goto out_err;
 747                         if (set_nid(page, i, 0, false))
 748                                 dn->node_changed = true;
 749                 }
 750         } else {
 751                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 752                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 753                         child_nid = le32_to_cpu(rn->in.nid[i]);
 754                         if (child_nid == 0) {
 755                                 child_nofs += NIDS_PER_BLOCK + 1;
 756                                 continue;
 757                         }
 758                         rdn.nid = child_nid;
 759                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 760                         if (ret == (NIDS_PER_BLOCK + 1)) {
 761                                 if (set_nid(page, i, 0, false))
 762                                         dn->node_changed = true;
 763                                 child_nofs += ret;
 764                         } else if (ret < 0 && ret != -ENOENT) {
 765                                 goto out_err;
 766                         }
 767                 }
 768                 freed = child_nofs;
 769         }
 770
 771         if (!ofs) {
 772                 /* remove current indirect node */
 773                 dn->node_page = page;
 774                 truncate_node(dn);
 775                 freed++;
 776         } else {
 777                 f2fs_put_page(page, 1);
 778         }
 779         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 780         return freed;
 781
 782 out_err:
 783         f2fs_put_page(page, 1);
 784         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 785         return ret;
 786 }
 787
 788 static int truncate_partial_nodes(struct dnode_of_data *dn,
 789                         struct f2fs_inode *ri, int *offset, int depth)
 790 {
 791         struct page *pages[2];
 792         nid_t nid[3];
 793         nid_t child_nid;
 794         int err = 0;
 795         int i;
 796         int idx = depth - 2;
 797
 798         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 799         if (!nid[0])
 800                 return 0;
 801
 802         /* get indirect nodes in the path */
 803         for (i = 0; i < idx + 1; i++) {
 804                 /* reference count'll be increased */
 805                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 806                 if (IS_ERR(pages[i])) {
 807                         err = PTR_ERR(pages[i]);
 808                         idx = i - 1;
 809                         goto fail;
 810                 }
 811                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 812         }
 813
 814         ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 815
 816         /* free direct nodes linked to a partial indirect node */
 817         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 818                 child_nid = get_nid(pages[idx], i, false);
 819                 if (!child_nid)
 820                         continue;
 821                 dn->nid = child_nid;
 822                 err = truncate_dnode(dn);
 823                 if (err < 0)
 824                         goto fail;
 825                 if (set_nid(pages[idx], i, 0, false))
 826                         dn->node_changed = true;
 827         }
 828
 829         if (offset[idx + 1] == 0) {
 830                 dn->node_page = pages[idx];
 831                 dn->nid = nid[idx];
 832                 truncate_node(dn);
 833         } else {
 834                 f2fs_put_page(pages[idx], 1);
 835         }
 836         offset[idx]++;
 837         offset[idx + 1] = 0;
 838         idx--;
 839 fail:
 840         for (i = idx; i >= 0; i--)
 841                 f2fs_put_page(pages[i], 1);
 842
 843         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 844
 845         return err;
 846 }
 847
 848 /*
 849  * All the block addresses of data and nodes should be nullified.
 850  */
 851 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 852 {
 853         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 854         int err = 0, cont = 1;
 855         int level, offset[4], noffset[4];
 856         unsigned int nofs = 0;
 857         struct f2fs_inode *ri;
 858         struct dnode_of_data dn;
 859         struct page *page;
 860
 861         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 862
 863         level = get_node_path(inode, from, offset, noffset);
 864
 865         page = get_node_page(sbi, inode->i_ino);
 866         if (IS_ERR(page)) {
 867                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 868                 return PTR_ERR(page);
 869         }
 870
 871         set_new_dnode(&dn, inode, page, NULL, 0);
 872         unlock_page(page);
 873
 874         ri = F2FS_INODE(page);
 875         switch (level) {
 876         case 0:
 877         case 1:
 878                 nofs = noffset[1];
 879                 break;
 880         case 2:
 881                 nofs = noffset[1];
 882                 if (!offset[level - 1])
 883                         goto skip_partial;
 884                 err = truncate_partial_nodes(&dn, ri, offset, level);
 885                 if (err < 0 && err != -ENOENT)
 886                         goto fail;
 887                 nofs += 1 + NIDS_PER_BLOCK;
 888                 break;
 889         case 3:
 890                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 891                 if (!offset[level - 1])
 892                         goto skip_partial;
 893                 err = truncate_partial_nodes(&dn, ri, offset, level);
 894                 if (err < 0 && err != -ENOENT)
 895                         goto fail;
 896                 break;
 897         default:
 898                 BUG();
 899         }
 900
 901 skip_partial:
 902         while (cont) {
 903                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 904                 switch (offset[0]) {
 905                 case NODE_DIR1_BLOCK:
 906                 case NODE_DIR2_BLOCK:
 907                         err = truncate_dnode(&dn);
 908                         break;
 909
 910                 case NODE_IND1_BLOCK:
 911                 case NODE_IND2_BLOCK:
 912                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 913                         break;
 914
 915                 case NODE_DIND_BLOCK:
 916                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 917                         cont = 0;
 918                         break;
 919
 920                 default:
 921                         BUG();
 922                 }
 923                 if (err < 0 && err != -ENOENT)
 924                         goto fail;
 925                 if (offset[1] == 0 &&
 926                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 927                         lock_page(page);
 928                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
 929                         f2fs_wait_on_page_writeback(page, NODE, true);
 930                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 931                         set_page_dirty(page);
 932                         unlock_page(page);
 933                 }
 934                 offset[1] = 0;
 935                 offset[0]++;
 936                 nofs += err;
 937         }
 938 fail:
 939         f2fs_put_page(page, 0);
 940         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 941         return err > 0 ? 0 : err;
 942 }
 943
 944 int truncate_xattr_node(struct inode *inode, struct page *page)
 945 {
 946         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 947         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 948         struct dnode_of_data dn;
 949         struct page *npage;
 950
 951         if (!nid)
 952                 return 0;
 953
 954         npage = get_node_page(sbi, nid);
 955         if (IS_ERR(npage))
 956                 return PTR_ERR(npage);
 957
 958         f2fs_i_xnid_write(inode, 0);
 959
 960         /* need to do checkpoint during fsync */
 961         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 962
 963         set_new_dnode(&dn, inode, page, npage, nid);
 964
 965         if (page)
 966                 dn.inode_page_locked = true;
 967         truncate_node(&dn);
 968         return 0;
 969 }
 970
 971 /*
 972  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 973  * f2fs_unlock_op().
 974  */
 975 int remove_inode_page(struct inode *inode)
 976 {
 977         struct dnode_of_data dn;
 978         int err;
 979
 980         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 981         err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
 982         if (err)
 983                 return err;
 984
 985         err = truncate_xattr_node(inode, dn.inode_page);
 986         if (err) {
 987                 f2fs_put_dnode(&dn);
 988                 return err;
 989         }
 990
 991         /* remove potential inline_data blocks */
 992         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 993                                 S_ISLNK(inode->i_mode))
 994                 truncate_data_blocks_range(&dn, 1);
 995
 996         /* 0 is possible, after f2fs_new_inode() has failed */
 997         f2fs_bug_on(F2FS_I_SB(inode),
 998                         inode->i_blocks != 0 && inode->i_blocks != 1);
 999
1000         /* will put inode & node pages */
1001         truncate_node(&dn);
1002         return 0;
1003 }
1004
1005 struct page *new_inode_page(struct inode *inode)
1006 {
1007         struct dnode_of_data dn;
1008
1009         /* allocate inode page for new inode */
1010         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1011
1012         /* caller should f2fs_put_page(page, 1); */
1013         return new_node_page(&dn, 0, NULL);
1014 }
1015
1016 struct page *new_node_page(struct dnode_of_data *dn,
1017                                 unsigned int ofs, struct page *ipage)
1018 {
1019         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1020         struct node_info old_ni, new_ni;
1021         struct page *page;
1022         int err;
1023
1024         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1025                 return ERR_PTR(-EPERM);
1026
1027         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1028         if (!page)
1029                 return ERR_PTR(-ENOMEM);
1030
1031         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1032                 err = -ENOSPC;
1033                 goto fail;
1034         }
1035
1036         get_node_info(sbi, dn->nid, &old_ni);
1037
1038         /* Reinitialize old_ni with new node page */
1039         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1040         new_ni = old_ni;
1041         new_ni.ino = dn->inode->i_ino;
1042         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1043
1044         f2fs_wait_on_page_writeback(page, NODE, true);
1045         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1046         set_cold_node(dn->inode, page);
1047         if (!PageUptodate(page))
1048                 SetPageUptodate(page);
1049         if (set_page_dirty(page))
1050                 dn->node_changed = true;
1051
1052         if (f2fs_has_xattr_block(ofs))
1053                 f2fs_i_xnid_write(dn->inode, dn->nid);
1054
1055         if (ofs == 0)
1056                 inc_valid_inode_count(sbi);
1057         return page;
1058
1059 fail:
1060         clear_node_page_dirty(page);
1061         f2fs_put_page(page, 1);
1062         return ERR_PTR(err);
1063 }
1064
1065 /*
1066  * Caller should do after getting the following values.
1067  * 0: f2fs_put_page(page, 0)
1068  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1069  */
1070 static int read_node_page(struct page *page, int op_flags)
1071 {
1072         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1073         struct node_info ni;
1074         struct f2fs_io_info fio = {
1075                 .sbi = sbi,
1076                 .type = NODE,
1077                 .op = REQ_OP_READ,
1078                 .op_flags = op_flags,
1079                 .page = page,
1080                 .encrypted_page = NULL,
1081         };
1082
1083         if (PageUptodate(page))
1084                 return LOCKED_PAGE;
1085
1086         get_node_info(sbi, page->index, &ni);
1087
1088         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1089                 ClearPageUptodate(page);
1090                 return -ENOENT;
1091         }
1092
1093         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1094         return f2fs_submit_page_bio(&fio);
1095 }
1096
1097 /*
1098  * Readahead a node page
1099  */
1100 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1101 {
1102         struct page *apage;
1103         int err;
1104
1105         if (!nid)
1106                 return;
1107         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1108
1109         rcu_read_lock();
1110         apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1111         rcu_read_unlock();
1112         if (apage)
1113                 return;
1114
1115         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1116         if (!apage)
1117                 return;
1118
1119         err = read_node_page(apage, REQ_RAHEAD);
1120         f2fs_put_page(apage, err ? 1 : 0);
1121 }
1122
1123 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1124                                         struct page *parent, int start)
1125 {
1126         struct page *page;
1127         int err;
1128
1129         if (!nid)
1130                 return ERR_PTR(-ENOENT);
1131         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1132 repeat:
1133         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1134         if (!page)
1135                 return ERR_PTR(-ENOMEM);
1136
1137         err = read_node_page(page, READ_SYNC);
1138         if (err < 0) {
1139                 f2fs_put_page(page, 1);
1140                 return ERR_PTR(err);
1141         } else if (err == LOCKED_PAGE) {
1142                 goto page_hit;
1143         }
1144
1145         if (parent)
1146                 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1147
1148         lock_page(page);
1149
1150         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1151                 f2fs_put_page(page, 1);
1152                 goto repeat;
1153         }
1154
1155         if (unlikely(!PageUptodate(page)))
1156                 goto out_err;
1157 page_hit:
1158         if(unlikely(nid != nid_of_node(page))) {
1159                 f2fs_bug_on(sbi, 1);
1160                 ClearPageUptodate(page);
1161 out_err:
1162                 f2fs_put_page(page, 1);
1163                 return ERR_PTR(-EIO);
1164         }
1165         return page;
1166 }
1167
1168 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1169 {
1170         return __get_node_page(sbi, nid, NULL, 0);
1171 }
1172
1173 struct page *get_node_page_ra(struct page *parent, int start)
1174 {
1175         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1176         nid_t nid = get_nid(parent, start, false);
1177
1178         return __get_node_page(sbi, nid, parent, start);
1179 }
1180
1181 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1182 {
1183         struct inode *inode;
1184         struct page *page;
1185         int ret;
1186
1187         /* should flush inline_data before evict_inode */
1188         inode = ilookup(sbi->sb, ino);
1189         if (!inode)
1190                 return;
1191
1192         page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1193         if (!page)
1194                 goto iput_out;
1195
1196         if (!PageUptodate(page))
1197                 goto page_out;
1198
1199         if (!PageDirty(page))
1200                 goto page_out;
1201
1202         if (!clear_page_dirty_for_io(page))
1203                 goto page_out;
1204
1205         ret = f2fs_write_inline_data(inode, page);
1206         inode_dec_dirty_pages(inode);
1207         if (ret)
1208                 set_page_dirty(page);
1209 page_out:
1210         f2fs_put_page(page, 1);
1211 iput_out:
1212         iput(inode);
1213 }
1214
1215 void move_node_page(struct page *node_page, int gc_type)
1216 {
1217         if (gc_type == FG_GC) {
1218                 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1219                 struct writeback_control wbc = {
1220                         .sync_mode = WB_SYNC_ALL,
1221                         .nr_to_write = 1,
1222                         .for_reclaim = 0,
1223                 };
1224
1225                 set_page_dirty(node_page);
1226                 f2fs_wait_on_page_writeback(node_page, NODE, true);
1227
1228                 f2fs_bug_on(sbi, PageWriteback(node_page));
1229                 if (!clear_page_dirty_for_io(node_page))
1230                         goto out_page;
1231
1232                 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1233                         unlock_page(node_page);
1234                 goto release_page;
1235         } else {
1236                 /* set page dirty and write it */
1237                 if (!PageWriteback(node_page))
1238                         set_page_dirty(node_page);
1239         }
1240 out_page:
1241         unlock_page(node_page);
1242 release_page:
1243         f2fs_put_page(node_page, 0);
1244 }
1245
1246 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1247 {
1248         pgoff_t index, end;
1249         struct pagevec pvec;
1250         struct page *last_page = NULL;
1251
1252         pagevec_init(&pvec, 0);
1253         index = 0;
1254         end = ULONG_MAX;
1255
1256         while (index <= end) {
1257                 int i, nr_pages;
1258                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1259                                 PAGECACHE_TAG_DIRTY,
1260                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1261                 if (nr_pages == 0)
1262                         break;
1263
1264                 for (i = 0; i < nr_pages; i++) {
1265                         struct page *page = pvec.pages[i];
1266
1267                         if (unlikely(f2fs_cp_error(sbi))) {
1268                                 f2fs_put_page(last_page, 0);
1269                                 pagevec_release(&pvec);
1270                                 return ERR_PTR(-EIO);
1271                         }
1272
1273                         if (!IS_DNODE(page) || !is_cold_node(page))
1274                                 continue;
1275                         if (ino_of_node(page) != ino)
1276                                 continue;
1277
1278                         lock_page(page);
1279
1280                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1281 continue_unlock:
1282                                 unlock_page(page);
1283                                 continue;
1284                         }
1285                         if (ino_of_node(page) != ino)
1286                                 goto continue_unlock;
1287
1288                         if (!PageDirty(page)) {
1289                                 /* someone wrote it for us */
1290                                 goto continue_unlock;
1291                         }
1292
1293                         if (last_page)
1294                                 f2fs_put_page(last_page, 0);
1295
1296                         get_page(page);
1297                         last_page = page;
1298                         unlock_page(page);
1299                 }
1300                 pagevec_release(&pvec);
1301                 cond_resched();
1302         }
1303         return last_page;
1304 }
1305
1306 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1307                         struct writeback_control *wbc, bool atomic)
1308 {
1309         pgoff_t index, end;
1310         struct pagevec pvec;
1311         int ret = 0;
1312         struct page *last_page = NULL;
1313         bool marked = false;
1314         nid_t ino = inode->i_ino;
1315         int nwritten = 0;
1316
1317         if (atomic) {
1318                 last_page = last_fsync_dnode(sbi, ino);
1319                 if (IS_ERR_OR_NULL(last_page))
1320                         return PTR_ERR_OR_ZERO(last_page);
1321         }
1322 retry:
1323         pagevec_init(&pvec, 0);
1324         index = 0;
1325         end = ULONG_MAX;
1326
1327         while (index <= end) {
1328                 int i, nr_pages;
1329                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1330                                 PAGECACHE_TAG_DIRTY,
1331                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1332                 if (nr_pages == 0)
1333                         break;
1334
1335                 for (i = 0; i < nr_pages; i++) {
1336                         struct page *page = pvec.pages[i];
1337
1338                         if (unlikely(f2fs_cp_error(sbi))) {
1339                                 f2fs_put_page(last_page, 0);
1340                                 pagevec_release(&pvec);
1341                                 return -EIO;
1342                         }
1343
1344                         if (!IS_DNODE(page) || !is_cold_node(page))
1345                                 continue;
1346                         if (ino_of_node(page) != ino)
1347                                 continue;
1348
1349                         lock_page(page);
1350
1351                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1352 continue_unlock:
1353                                 unlock_page(page);
1354                                 continue;
1355                         }
1356                         if (ino_of_node(page) != ino)
1357                                 goto continue_unlock;
1358
1359                         if (!PageDirty(page) && page != last_page) {
1360                                 /* someone wrote it for us */
1361                                 goto continue_unlock;
1362                         }
1363
1364                         f2fs_wait_on_page_writeback(page, NODE, true);
1365                         BUG_ON(PageWriteback(page));
1366
1367                         if (!atomic || page == last_page) {
1368                                 set_fsync_mark(page, 1);
1369                                 if (IS_INODE(page)) {
1370                                         if (is_inode_flag_set(inode,
1371                                                                 FI_DIRTY_INODE))
1372                                                 update_inode(inode, page);
1373                                         set_dentry_mark(page,
1374                                                 need_dentry_mark(sbi, ino));
1375                                 }
1376                                 /*  may be written by other thread */
1377                                 if (!PageDirty(page))
1378                                         set_page_dirty(page);
1379                         }
1380
1381                         if (!clear_page_dirty_for_io(page))
1382                                 goto continue_unlock;
1383
1384                         ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1385                         if (ret) {
1386                                 unlock_page(page);
1387                                 f2fs_put_page(last_page, 0);
1388                                 break;
1389                         } else {
1390                                 nwritten++;
1391                         }
1392
1393                         if (page == last_page) {
1394                                 f2fs_put_page(page, 0);
1395                                 marked = true;
1396                                 break;
1397                         }
1398                 }
1399                 pagevec_release(&pvec);
1400                 cond_resched();
1401
1402                 if (ret || marked)
1403                         break;
1404         }
1405         if (!ret && atomic && !marked) {
1406                 f2fs_msg(sbi->sb, KERN_DEBUG,
1407                         "Retry to write fsync mark: ino=%u, idx=%lx",
1408                                         ino, last_page->index);
1409                 lock_page(last_page);
1410                 set_page_dirty(last_page);
1411                 unlock_page(last_page);
1412                 goto retry;
1413         }
1414
1415         if (nwritten)
1416                 f2fs_submit_merged_bio_cond(sbi, NULL, NULL, ino, NODE, WRITE);
1417         return ret ? -EIO: 0;
1418 }
1419
1420 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1421 {
1422         pgoff_t index, end;
1423         struct pagevec pvec;
1424         int step = 0;
1425         int nwritten = 0;
1426         int ret = 0;
1427
1428         pagevec_init(&pvec, 0);
1429
1430 next_step:
1431         index = 0;
1432         end = ULONG_MAX;
1433
1434         while (index <= end) {
1435                 int i, nr_pages;
1436                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1437                                 PAGECACHE_TAG_DIRTY,
1438                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1439                 if (nr_pages == 0)
1440                         break;
1441
1442                 for (i = 0; i < nr_pages; i++) {
1443                         struct page *page = pvec.pages[i];
1444
1445                         if (unlikely(f2fs_cp_error(sbi))) {
1446                                 pagevec_release(&pvec);
1447                                 ret = -EIO;
1448                                 goto out;
1449                         }
1450
1451                         /*
1452                          * flushing sequence with step:
1453                          * 0. indirect nodes
1454                          * 1. dentry dnodes
1455                          * 2. file dnodes
1456                          */
1457                         if (step == 0 && IS_DNODE(page))
1458                                 continue;
1459                         if (step == 1 && (!IS_DNODE(page) ||
1460                                                 is_cold_node(page)))
1461                                 continue;
1462                         if (step == 2 && (!IS_DNODE(page) ||
1463                                                 !is_cold_node(page)))
1464                                 continue;
1465 lock_node:
1466                         if (!trylock_page(page))
1467                                 continue;
1468
1469                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1470 continue_unlock:
1471                                 unlock_page(page);
1472                                 continue;
1473                         }
1474
1475                         if (!PageDirty(page)) {
1476                                 /* someone wrote it for us */
1477                                 goto continue_unlock;
1478                         }
1479
1480                         /* flush inline_data */
1481                         if (is_inline_node(page)) {
1482                                 clear_inline_node(page);
1483                                 unlock_page(page);
1484                                 flush_inline_data(sbi, ino_of_node(page));
1485                                 goto lock_node;
1486                         }
1487
1488                         f2fs_wait_on_page_writeback(page, NODE, true);
1489
1490                         BUG_ON(PageWriteback(page));
1491                         if (!clear_page_dirty_for_io(page))
1492                                 goto continue_unlock;
1493
1494                         set_fsync_mark(page, 0);
1495                         set_dentry_mark(page, 0);
1496
1497                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1498                                 unlock_page(page);
1499                         else
1500                                 nwritten++;
1501
1502                         if (--wbc->nr_to_write == 0)
1503                                 break;
1504                 }
1505                 pagevec_release(&pvec);
1506                 cond_resched();
1507
1508                 if (wbc->nr_to_write == 0) {
1509                         step = 2;
1510                         break;
1511                 }
1512         }
1513
1514         if (step < 2) {
1515                 step++;
1516                 goto next_step;
1517         }
1518 out:
1519         if (nwritten)
1520                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1521         return ret;
1522 }
1523
1524 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1525 {
1526         pgoff_t index = 0, end = ULONG_MAX;
1527         struct pagevec pvec;
1528         int ret2 = 0, ret = 0;
1529
1530         pagevec_init(&pvec, 0);
1531
1532         while (index <= end) {
1533                 int i, nr_pages;
1534                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1535                                 PAGECACHE_TAG_WRITEBACK,
1536                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1537                 if (nr_pages == 0)
1538                         break;
1539
1540                 for (i = 0; i < nr_pages; i++) {
1541                         struct page *page = pvec.pages[i];
1542
1543                         /* until radix tree lookup accepts end_index */
1544                         if (unlikely(page->index > end))
1545                                 continue;
1546
1547                         if (ino && ino_of_node(page) == ino) {
1548                                 f2fs_wait_on_page_writeback(page, NODE, true);
1549                                 if (TestClearPageError(page))
1550                                         ret = -EIO;
1551                         }
1552                 }
1553                 pagevec_release(&pvec);
1554                 cond_resched();
1555         }
1556
1557         if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1558                 ret2 = -ENOSPC;
1559         if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1560                 ret2 = -EIO;
1561         if (!ret)
1562                 ret = ret2;
1563         return ret;
1564 }
1565
1566 static int f2fs_write_node_page(struct page *page,
1567                                 struct writeback_control *wbc)
1568 {
1569         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1570         nid_t nid;
1571         struct node_info ni;
1572         struct f2fs_io_info fio = {
1573                 .sbi = sbi,
1574                 .type = NODE,
1575                 .op = REQ_OP_WRITE,
1576                 .op_flags = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0,
1577                 .page = page,
1578                 .encrypted_page = NULL,
1579         };
1580
1581         trace_f2fs_writepage(page, NODE);
1582
1583         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1584                 goto redirty_out;
1585         if (unlikely(f2fs_cp_error(sbi)))
1586                 goto redirty_out;
1587
1588         /* get old block addr of this node page */
1589         nid = nid_of_node(page);
1590         f2fs_bug_on(sbi, page->index != nid);
1591
1592         if (wbc->for_reclaim) {
1593                 if (!down_read_trylock(&sbi->node_write))
1594                         goto redirty_out;
1595         } else {
1596                 down_read(&sbi->node_write);
1597         }
1598
1599         get_node_info(sbi, nid, &ni);
1600
1601         /* This page is already truncated */
1602         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1603                 ClearPageUptodate(page);
1604                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1605                 up_read(&sbi->node_write);
1606                 unlock_page(page);
1607                 return 0;
1608         }
1609
1610         set_page_writeback(page);
1611         fio.old_blkaddr = ni.blk_addr;
1612         write_node_page(nid, &fio);
1613         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1614         dec_page_count(sbi, F2FS_DIRTY_NODES);
1615         up_read(&sbi->node_write);
1616
1617         if (wbc->for_reclaim)
1618                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1619
1620         unlock_page(page);
1621
1622         if (unlikely(f2fs_cp_error(sbi)))
1623                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1624
1625         return 0;
1626
1627 redirty_out:
1628         redirty_page_for_writepage(wbc, page);
1629         return AOP_WRITEPAGE_ACTIVATE;
1630 }
1631
1632 static int f2fs_write_node_pages(struct address_space *mapping,
1633                             struct writeback_control *wbc)
1634 {
1635         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1636         struct blk_plug plug;
1637         long diff;
1638
1639         /* balancing f2fs's metadata in background */
1640         f2fs_balance_fs_bg(sbi);
1641
1642         /* collect a number of dirty node pages and write together */
1643         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1644                 goto skip_write;
1645
1646         trace_f2fs_writepages(mapping->host, wbc, NODE);
1647
1648         diff = nr_pages_to_write(sbi, NODE, wbc);
1649         wbc->sync_mode = WB_SYNC_NONE;
1650         blk_start_plug(&plug);
1651         sync_node_pages(sbi, wbc);
1652         blk_finish_plug(&plug);
1653         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1654         return 0;
1655
1656 skip_write:
1657         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1658         trace_f2fs_writepages(mapping->host, wbc, NODE);
1659         return 0;
1660 }
1661
1662 static int f2fs_set_node_page_dirty(struct page *page)
1663 {
1664         trace_f2fs_set_page_dirty(page, NODE);
1665
1666         if (!PageUptodate(page))
1667                 SetPageUptodate(page);
1668         if (!PageDirty(page)) {
1669                 f2fs_set_page_dirty_nobuffers(page);
1670                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1671                 SetPagePrivate(page);
1672                 f2fs_trace_pid(page);
1673                 return 1;
1674         }
1675         return 0;
1676 }
1677
1678 /*
1679  * Structure of the f2fs node operations
1680  */
1681 const struct address_space_operations f2fs_node_aops = {
1682         .writepage      = f2fs_write_node_page,
1683         .writepages     = f2fs_write_node_pages,
1684         .set_page_dirty = f2fs_set_node_page_dirty,
1685         .invalidatepage = f2fs_invalidate_page,
1686         .releasepage    = f2fs_release_page,
1687 #ifdef CONFIG_MIGRATION
1688         .migratepage    = f2fs_migrate_page,
1689 #endif
1690 };
1691
1692 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1693                                                 nid_t n)
1694 {
1695         return radix_tree_lookup(&nm_i->free_nid_root, n);
1696 }
1697
1698 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1699                                                 struct free_nid *i)
1700 {
1701         list_del(&i->list);
1702         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1703 }
1704
1705 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1706 {
1707         struct f2fs_nm_info *nm_i = NM_I(sbi);
1708         struct free_nid *i;
1709         struct nat_entry *ne;
1710
1711         if (!available_free_memory(sbi, FREE_NIDS))
1712                 return -1;
1713
1714         /* 0 nid should not be used */
1715         if (unlikely(nid == 0))
1716                 return 0;
1717
1718         if (build) {
1719                 /* do not add allocated nids */
1720                 ne = __lookup_nat_cache(nm_i, nid);
1721                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1722                                 nat_get_blkaddr(ne) != NULL_ADDR))
1723                         return 0;
1724         }
1725
1726         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1727         i->nid = nid;
1728         i->state = NID_NEW;
1729
1730         if (radix_tree_preload(GFP_NOFS)) {
1731                 kmem_cache_free(free_nid_slab, i);
1732                 return 0;
1733         }
1734
1735         spin_lock(&nm_i->free_nid_list_lock);
1736         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1737                 spin_unlock(&nm_i->free_nid_list_lock);
1738                 radix_tree_preload_end();
1739                 kmem_cache_free(free_nid_slab, i);
1740                 return 0;
1741         }
1742         list_add_tail(&i->list, &nm_i->free_nid_list);
1743         nm_i->fcnt++;
1744         spin_unlock(&nm_i->free_nid_list_lock);
1745         radix_tree_preload_end();
1746         return 1;
1747 }
1748
1749 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1750 {
1751         struct free_nid *i;
1752         bool need_free = false;
1753
1754         spin_lock(&nm_i->free_nid_list_lock);
1755         i = __lookup_free_nid_list(nm_i, nid);
1756         if (i && i->state == NID_NEW) {
1757                 __del_from_free_nid_list(nm_i, i);
1758                 nm_i->fcnt--;
1759                 need_free = true;
1760         }
1761         spin_unlock(&nm_i->free_nid_list_lock);
1762
1763         if (need_free)
1764                 kmem_cache_free(free_nid_slab, i);
1765 }
1766
1767 static void scan_nat_page(struct f2fs_sb_info *sbi,
1768                         struct page *nat_page, nid_t start_nid)
1769 {
1770         struct f2fs_nm_info *nm_i = NM_I(sbi);
1771         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1772         block_t blk_addr;
1773         int i;
1774
1775         i = start_nid % NAT_ENTRY_PER_BLOCK;
1776
1777         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1778
1779                 if (unlikely(start_nid >= nm_i->max_nid))
1780                         break;
1781
1782                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1783                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1784                 if (blk_addr == NULL_ADDR) {
1785                         if (add_free_nid(sbi, start_nid, true) < 0)
1786                                 break;
1787                 }
1788         }
1789 }
1790
1791 void build_free_nids(struct f2fs_sb_info *sbi)
1792 {
1793         struct f2fs_nm_info *nm_i = NM_I(sbi);
1794         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1795         struct f2fs_journal *journal = curseg->journal;
1796         int i = 0;
1797         nid_t nid = nm_i->next_scan_nid;
1798
1799         /* Enough entries */
1800         if (nm_i->fcnt >= NAT_ENTRY_PER_BLOCK)
1801                 return;
1802
1803         /* readahead nat pages to be scanned */
1804         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1805                                                         META_NAT, true);
1806
1807         down_read(&nm_i->nat_tree_lock);
1808
1809         while (1) {
1810                 struct page *page = get_current_nat_page(sbi, nid);
1811
1812                 scan_nat_page(sbi, page, nid);
1813                 f2fs_put_page(page, 1);
1814
1815                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1816                 if (unlikely(nid >= nm_i->max_nid))
1817                         nid = 0;
1818
1819                 if (++i >= FREE_NID_PAGES)
1820                         break;
1821         }
1822
1823         /* go to the next free nat pages to find free nids abundantly */
1824         nm_i->next_scan_nid = nid;
1825
1826         /* find free nids from current sum_pages */
1827         down_read(&curseg->journal_rwsem);
1828         for (i = 0; i < nats_in_cursum(journal); i++) {
1829                 block_t addr;
1830
1831                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1832                 nid = le32_to_cpu(nid_in_journal(journal, i));
1833                 if (addr == NULL_ADDR)
1834                         add_free_nid(sbi, nid, true);
1835                 else
1836                         remove_free_nid(nm_i, nid);
1837         }
1838         up_read(&curseg->journal_rwsem);
1839         up_read(&nm_i->nat_tree_lock);
1840
1841         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1842                                         nm_i->ra_nid_pages, META_NAT, false);
1843 }
1844
1845 /*
1846  * If this function returns success, caller can obtain a new nid
1847  * from second parameter of this function.
1848  * The returned nid could be used ino as well as nid when inode is created.
1849  */
1850 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1851 {
1852         struct f2fs_nm_info *nm_i = NM_I(sbi);
1853         struct free_nid *i = NULL;
1854 retry:
1855 #ifdef CONFIG_F2FS_FAULT_INJECTION
1856         if (time_to_inject(sbi, FAULT_ALLOC_NID))
1857                 return false;
1858 #endif
1859         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1860                 return false;
1861
1862         spin_lock(&nm_i->free_nid_list_lock);
1863
1864         /* We should not use stale free nids created by build_free_nids */
1865         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1866                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1867                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1868                         if (i->state == NID_NEW)
1869                                 break;
1870
1871                 f2fs_bug_on(sbi, i->state != NID_NEW);
1872                 *nid = i->nid;
1873                 i->state = NID_ALLOC;
1874                 nm_i->fcnt--;
1875                 spin_unlock(&nm_i->free_nid_list_lock);
1876                 return true;
1877         }
1878         spin_unlock(&nm_i->free_nid_list_lock);
1879
1880         /* Let's scan nat pages and its caches to get free nids */
1881         mutex_lock(&nm_i->build_lock);
1882         build_free_nids(sbi);
1883         mutex_unlock(&nm_i->build_lock);
1884         goto retry;
1885 }
1886
1887 /*
1888  * alloc_nid() should be called prior to this function.
1889  */
1890 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1891 {
1892         struct f2fs_nm_info *nm_i = NM_I(sbi);
1893         struct free_nid *i;
1894
1895         spin_lock(&nm_i->free_nid_list_lock);
1896         i = __lookup_free_nid_list(nm_i, nid);
1897         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1898         __del_from_free_nid_list(nm_i, i);
1899         spin_unlock(&nm_i->free_nid_list_lock);
1900
1901         kmem_cache_free(free_nid_slab, i);
1902 }
1903
1904 /*
1905  * alloc_nid() should be called prior to this function.
1906  */
1907 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1908 {
1909         struct f2fs_nm_info *nm_i = NM_I(sbi);
1910         struct free_nid *i;
1911         bool need_free = false;
1912
1913         if (!nid)
1914                 return;
1915
1916         spin_lock(&nm_i->free_nid_list_lock);
1917         i = __lookup_free_nid_list(nm_i, nid);
1918         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1919         if (!available_free_memory(sbi, FREE_NIDS)) {
1920                 __del_from_free_nid_list(nm_i, i);
1921                 need_free = true;
1922         } else {
1923                 i->state = NID_NEW;
1924                 nm_i->fcnt++;
1925         }
1926         spin_unlock(&nm_i->free_nid_list_lock);
1927
1928         if (need_free)
1929                 kmem_cache_free(free_nid_slab, i);
1930 }
1931
1932 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1933 {
1934         struct f2fs_nm_info *nm_i = NM_I(sbi);
1935         struct free_nid *i, *next;
1936         int nr = nr_shrink;
1937
1938         if (nm_i->fcnt <= MAX_FREE_NIDS)
1939                 return 0;
1940
1941         if (!mutex_trylock(&nm_i->build_lock))
1942                 return 0;
1943
1944         spin_lock(&nm_i->free_nid_list_lock);
1945         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1946                 if (nr_shrink <= 0 || nm_i->fcnt <= MAX_FREE_NIDS)
1947                         break;
1948                 if (i->state == NID_ALLOC)
1949                         continue;
1950                 __del_from_free_nid_list(nm_i, i);
1951                 kmem_cache_free(free_nid_slab, i);
1952                 nm_i->fcnt--;
1953                 nr_shrink--;
1954         }
1955         spin_unlock(&nm_i->free_nid_list_lock);
1956         mutex_unlock(&nm_i->build_lock);
1957
1958         return nr - nr_shrink;
1959 }
1960
1961 void recover_inline_xattr(struct inode *inode, struct page *page)
1962 {
1963         void *src_addr, *dst_addr;
1964         size_t inline_size;
1965         struct page *ipage;
1966         struct f2fs_inode *ri;
1967
1968         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1969         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1970
1971         ri = F2FS_INODE(page);
1972         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1973                 clear_inode_flag(inode, FI_INLINE_XATTR);
1974                 goto update_inode;
1975         }
1976
1977         dst_addr = inline_xattr_addr(ipage);
1978         src_addr = inline_xattr_addr(page);
1979         inline_size = inline_xattr_size(inode);
1980
1981         f2fs_wait_on_page_writeback(ipage, NODE, true);
1982         memcpy(dst_addr, src_addr, inline_size);
1983 update_inode:
1984         update_inode(inode, ipage);
1985         f2fs_put_page(ipage, 1);
1986 }
1987
1988 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1989 {
1990         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1991         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1992         nid_t new_xnid = nid_of_node(page);
1993         struct node_info ni;
1994
1995         /* 1: invalidate the previous xattr nid */
1996         if (!prev_xnid)
1997                 goto recover_xnid;
1998
1999         /* Deallocate node address */
2000         get_node_info(sbi, prev_xnid, &ni);
2001         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2002         invalidate_blocks(sbi, ni.blk_addr);
2003         dec_valid_node_count(sbi, inode);
2004         set_node_addr(sbi, &ni, NULL_ADDR, false);
2005
2006 recover_xnid:
2007         /* 2: allocate new xattr nid */
2008         if (unlikely(!inc_valid_node_count(sbi, inode)))
2009                 f2fs_bug_on(sbi, 1);
2010
2011         remove_free_nid(NM_I(sbi), new_xnid);
2012         get_node_info(sbi, new_xnid, &ni);
2013         ni.ino = inode->i_ino;
2014         set_node_addr(sbi, &ni, NEW_ADDR, false);
2015         f2fs_i_xnid_write(inode, new_xnid);
2016
2017         /* 3: update xattr blkaddr */
2018         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
2019         set_node_addr(sbi, &ni, blkaddr, false);
2020 }
2021
2022 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2023 {
2024         struct f2fs_inode *src, *dst;
2025         nid_t ino = ino_of_node(page);
2026         struct node_info old_ni, new_ni;
2027         struct page *ipage;
2028
2029         get_node_info(sbi, ino, &old_ni);
2030
2031         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2032                 return -EINVAL;
2033 retry:
2034         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2035         if (!ipage) {
2036                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2037                 goto retry;
2038         }
2039
2040         /* Should not use this inode from free nid list */
2041         remove_free_nid(NM_I(sbi), ino);
2042
2043         if (!PageUptodate(ipage))
2044                 SetPageUptodate(ipage);
2045         fill_node_footer(ipage, ino, ino, 0, true);
2046
2047         src = F2FS_INODE(page);
2048         dst = F2FS_INODE(ipage);
2049
2050         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2051         dst->i_size = 0;
2052         dst->i_blocks = cpu_to_le64(1);
2053         dst->i_links = cpu_to_le32(1);
2054         dst->i_xattr_nid = 0;
2055         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2056
2057         new_ni = old_ni;
2058         new_ni.ino = ino;
2059
2060         if (unlikely(!inc_valid_node_count(sbi, NULL)))
2061                 WARN_ON(1);
2062         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2063         inc_valid_inode_count(sbi);
2064         set_page_dirty(ipage);
2065         f2fs_put_page(ipage, 1);
2066         return 0;
2067 }
2068
2069 int restore_node_summary(struct f2fs_sb_info *sbi,
2070                         unsigned int segno, struct f2fs_summary_block *sum)
2071 {
2072         struct f2fs_node *rn;
2073         struct f2fs_summary *sum_entry;
2074         block_t addr;
2075         int bio_blocks = MAX_BIO_BLOCKS(sbi);
2076         int i, idx, last_offset, nrpages;
2077
2078         /* scan the node segment */
2079         last_offset = sbi->blocks_per_seg;
2080         addr = START_BLOCK(sbi, segno);
2081         sum_entry = &sum->entries[0];
2082
2083         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2084                 nrpages = min(last_offset - i, bio_blocks);
2085
2086                 /* readahead node pages */
2087                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2088
2089                 for (idx = addr; idx < addr + nrpages; idx++) {
2090                         struct page *page = get_tmp_page(sbi, idx);
2091
2092                         rn = F2FS_NODE(page);
2093                         sum_entry->nid = rn->footer.nid;
2094                         sum_entry->version = 0;
2095                         sum_entry->ofs_in_node = 0;
2096                         sum_entry++;
2097                         f2fs_put_page(page, 1);
2098                 }
2099
2100                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2101                                                         addr + nrpages);
2102         }
2103         return 0;
2104 }
2105
2106 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2107 {
2108         struct f2fs_nm_info *nm_i = NM_I(sbi);
2109         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2110         struct f2fs_journal *journal = curseg->journal;
2111         int i;
2112
2113         down_write(&curseg->journal_rwsem);
2114         for (i = 0; i < nats_in_cursum(journal); i++) {
2115                 struct nat_entry *ne;
2116                 struct f2fs_nat_entry raw_ne;
2117                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2118
2119                 raw_ne = nat_in_journal(journal, i);
2120
2121                 ne = __lookup_nat_cache(nm_i, nid);
2122                 if (!ne) {
2123                         ne = grab_nat_entry(nm_i, nid);
2124                         node_info_from_raw_nat(&ne->ni, &raw_ne);
2125                 }
2126                 __set_nat_cache_dirty(nm_i, ne);
2127         }
2128         update_nats_in_cursum(journal, -i);
2129         up_write(&curseg->journal_rwsem);
2130 }
2131
2132 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2133                                                 struct list_head *head, int max)
2134 {
2135         struct nat_entry_set *cur;
2136
2137         if (nes->entry_cnt >= max)
2138                 goto add_out;
2139
2140         list_for_each_entry(cur, head, set_list) {
2141                 if (cur->entry_cnt >= nes->entry_cnt) {
2142                         list_add(&nes->set_list, cur->set_list.prev);
2143                         return;
2144                 }
2145         }
2146 add_out:
2147         list_add_tail(&nes->set_list, head);
2148 }
2149
2150 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2151                                         struct nat_entry_set *set)
2152 {
2153         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2154         struct f2fs_journal *journal = curseg->journal;
2155         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2156         bool to_journal = true;
2157         struct f2fs_nat_block *nat_blk;
2158         struct nat_entry *ne, *cur;
2159         struct page *page = NULL;
2160
2161         /*
2162          * there are two steps to flush nat entries:
2163          * #1, flush nat entries to journal in current hot data summary block.
2164          * #2, flush nat entries to nat page.
2165          */
2166         if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2167                 to_journal = false;
2168
2169         if (to_journal) {
2170                 down_write(&curseg->journal_rwsem);
2171         } else {
2172                 page = get_next_nat_page(sbi, start_nid);
2173                 nat_blk = page_address(page);
2174                 f2fs_bug_on(sbi, !nat_blk);
2175         }
2176
2177         /* flush dirty nats in nat entry set */
2178         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2179                 struct f2fs_nat_entry *raw_ne;
2180                 nid_t nid = nat_get_nid(ne);
2181                 int offset;
2182
2183                 if (nat_get_blkaddr(ne) == NEW_ADDR)
2184                         continue;
2185
2186                 if (to_journal) {
2187                         offset = lookup_journal_in_cursum(journal,
2188                                                         NAT_JOURNAL, nid, 1);
2189                         f2fs_bug_on(sbi, offset < 0);
2190                         raw_ne = &nat_in_journal(journal, offset);
2191                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2192                 } else {
2193                         raw_ne = &nat_blk->entries[nid - start_nid];
2194                 }
2195                 raw_nat_from_node_info(raw_ne, &ne->ni);
2196                 nat_reset_flag(ne);
2197                 __clear_nat_cache_dirty(NM_I(sbi), ne);
2198                 if (nat_get_blkaddr(ne) == NULL_ADDR)
2199                         add_free_nid(sbi, nid, false);
2200         }
2201
2202         if (to_journal)
2203                 up_write(&curseg->journal_rwsem);
2204         else
2205                 f2fs_put_page(page, 1);
2206
2207         f2fs_bug_on(sbi, set->entry_cnt);
2208
2209         radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2210         kmem_cache_free(nat_entry_set_slab, set);
2211 }
2212
2213 /*
2214  * This function is called during the checkpointing process.
2215  */
2216 void flush_nat_entries(struct f2fs_sb_info *sbi)
2217 {
2218         struct f2fs_nm_info *nm_i = NM_I(sbi);
2219         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2220         struct f2fs_journal *journal = curseg->journal;
2221         struct nat_entry_set *setvec[SETVEC_SIZE];
2222         struct nat_entry_set *set, *tmp;
2223         unsigned int found;
2224         nid_t set_idx = 0;
2225         LIST_HEAD(sets);
2226
2227         if (!nm_i->dirty_nat_cnt)
2228                 return;
2229
2230         down_write(&nm_i->nat_tree_lock);
2231
2232         /*
2233          * if there are no enough space in journal to store dirty nat
2234          * entries, remove all entries from journal and merge them
2235          * into nat entry set.
2236          */
2237         if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2238                 remove_nats_in_journal(sbi);
2239
2240         while ((found = __gang_lookup_nat_set(nm_i,
2241                                         set_idx, SETVEC_SIZE, setvec))) {
2242                 unsigned idx;
2243                 set_idx = setvec[found - 1]->set + 1;
2244                 for (idx = 0; idx < found; idx++)
2245                         __adjust_nat_entry_set(setvec[idx], &sets,
2246                                                 MAX_NAT_JENTRIES(journal));
2247         }
2248
2249         /* flush dirty nats in nat entry set */
2250         list_for_each_entry_safe(set, tmp, &sets, set_list)
2251                 __flush_nat_entry_set(sbi, set);
2252
2253         up_write(&nm_i->nat_tree_lock);
2254
2255         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2256 }
2257
2258 static int init_node_manager(struct f2fs_sb_info *sbi)
2259 {
2260         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2261         struct f2fs_nm_info *nm_i = NM_I(sbi);
2262         unsigned char *version_bitmap;
2263         unsigned int nat_segs, nat_blocks;
2264
2265         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2266
2267         /* segment_count_nat includes pair segment so divide to 2. */
2268         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2269         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2270
2271         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2272
2273         /* not used nids: 0, node, meta, (and root counted as valid node) */
2274         nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2275         nm_i->fcnt = 0;
2276         nm_i->nat_cnt = 0;
2277         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2278         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2279         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2280
2281         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2282         INIT_LIST_HEAD(&nm_i->free_nid_list);
2283         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2284         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2285         INIT_LIST_HEAD(&nm_i->nat_entries);
2286
2287         mutex_init(&nm_i->build_lock);
2288         spin_lock_init(&nm_i->free_nid_list_lock);
2289         init_rwsem(&nm_i->nat_tree_lock);
2290
2291         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2292         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2293         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2294         if (!version_bitmap)
2295                 return -EFAULT;
2296
2297         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2298                                         GFP_KERNEL);
2299         if (!nm_i->nat_bitmap)
2300                 return -ENOMEM;
2301         return 0;
2302 }
2303
2304 int build_node_manager(struct f2fs_sb_info *sbi)
2305 {
2306         int err;
2307
2308         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2309         if (!sbi->nm_info)
2310                 return -ENOMEM;
2311
2312         err = init_node_manager(sbi);
2313         if (err)
2314                 return err;
2315
2316         build_free_nids(sbi);
2317         return 0;
2318 }
2319
2320 void destroy_node_manager(struct f2fs_sb_info *sbi)
2321 {
2322         struct f2fs_nm_info *nm_i = NM_I(sbi);
2323         struct free_nid *i, *next_i;
2324         struct nat_entry *natvec[NATVEC_SIZE];
2325         struct nat_entry_set *setvec[SETVEC_SIZE];
2326         nid_t nid = 0;
2327         unsigned int found;
2328
2329         if (!nm_i)
2330                 return;
2331
2332         /* destroy free nid list */
2333         spin_lock(&nm_i->free_nid_list_lock);
2334         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2335                 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2336                 __del_from_free_nid_list(nm_i, i);
2337                 nm_i->fcnt--;
2338                 spin_unlock(&nm_i->free_nid_list_lock);
2339                 kmem_cache_free(free_nid_slab, i);
2340                 spin_lock(&nm_i->free_nid_list_lock);
2341         }
2342         f2fs_bug_on(sbi, nm_i->fcnt);
2343         spin_unlock(&nm_i->free_nid_list_lock);
2344
2345         /* destroy nat cache */
2346         down_write(&nm_i->nat_tree_lock);
2347         while ((found = __gang_lookup_nat_cache(nm_i,
2348                                         nid, NATVEC_SIZE, natvec))) {
2349                 unsigned idx;
2350
2351                 nid = nat_get_nid(natvec[found - 1]) + 1;
2352                 for (idx = 0; idx < found; idx++)
2353                         __del_from_nat_cache(nm_i, natvec[idx]);
2354         }
2355         f2fs_bug_on(sbi, nm_i->nat_cnt);
2356
2357         /* destroy nat set cache */
2358         nid = 0;
2359         while ((found = __gang_lookup_nat_set(nm_i,
2360                                         nid, SETVEC_SIZE, setvec))) {
2361                 unsigned idx;
2362
2363                 nid = setvec[found - 1]->set + 1;
2364                 for (idx = 0; idx < found; idx++) {
2365                         /* entry_cnt is not zero, when cp_error was occurred */
2366                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2367                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2368                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2369                 }
2370         }
2371         up_write(&nm_i->nat_tree_lock);
2372
2373         kfree(nm_i->nat_bitmap);
2374         sbi->nm_info = NULL;
2375         kfree(nm_i);
2376 }
2377
2378 int __init create_node_manager_caches(void)
2379 {
2380         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2381                         sizeof(struct nat_entry));
2382         if (!nat_entry_slab)
2383                 goto fail;
2384
2385         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2386                         sizeof(struct free_nid));
2387         if (!free_nid_slab)
2388                 goto destroy_nat_entry;
2389
2390         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2391                         sizeof(struct nat_entry_set));
2392         if (!nat_entry_set_slab)
2393                 goto destroy_free_nid;
2394         return 0;
2395
2396 destroy_free_nid:
2397         kmem_cache_destroy(free_nid_slab);
2398 destroy_nat_entry:
2399         kmem_cache_destroy(nat_entry_slab);
2400 fail:
2401         return -ENOMEM;
2402 }
2403
2404 void destroy_node_manager_caches(void)
2405 {
2406         kmem_cache_destroy(nat_entry_set_slab);
2407         kmem_cache_destroy(free_nid_slab);
2408         kmem_cache_destroy(nat_entry_slab);
2409 }