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, is_valid_data_blkaddr(sbi, nat_get_blkaddr(e)) &&
 308                         new_blkaddr == NEW_ADDR);
 309
 310         /* increment version no as node is removed */
 311         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 312                 unsigned char version = nat_get_version(e);
 313                 nat_set_version(e, inc_node_version(version));
 314
 315                 /* in order to reuse the nid */
 316                 if (nm_i->next_scan_nid > ni->nid)
 317                         nm_i->next_scan_nid = ni->nid;
 318         }
 319
 320         /* change address */
 321         nat_set_blkaddr(e, new_blkaddr);
 322         if (!is_valid_data_blkaddr(sbi, new_blkaddr))
 323                 set_nat_flag(e, IS_CHECKPOINTED, false);
 324         __set_nat_cache_dirty(nm_i, e);
 325
 326         /* update fsync_mark if its inode nat entry is still alive */
 327         if (ni->nid != ni->ino)
 328                 e = __lookup_nat_cache(nm_i, ni->ino);
 329         if (e) {
 330                 if (fsync_done && ni->nid == ni->ino)
 331                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 332                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 333         }
 334         up_write(&nm_i->nat_tree_lock);
 335 }
 336
 337 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 338 {
 339         struct f2fs_nm_info *nm_i = NM_I(sbi);
 340         int nr = nr_shrink;
 341
 342         if (!down_write_trylock(&nm_i->nat_tree_lock))
 343                 return 0;
 344
 345         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 346                 struct nat_entry *ne;
 347                 ne = list_first_entry(&nm_i->nat_entries,
 348                                         struct nat_entry, list);
 349                 __del_from_nat_cache(nm_i, ne);
 350                 nr_shrink--;
 351         }
 352         up_write(&nm_i->nat_tree_lock);
 353         return nr - nr_shrink;
 354 }
 355
 356 /*
 357  * This function always returns success
 358  */
 359 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 360 {
 361         struct f2fs_nm_info *nm_i = NM_I(sbi);
 362         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 363         struct f2fs_journal *journal = curseg->journal;
 364         nid_t start_nid = START_NID(nid);
 365         struct f2fs_nat_block *nat_blk;
 366         struct page *page = NULL;
 367         struct f2fs_nat_entry ne;
 368         struct nat_entry *e;
 369         int i;
 370
 371         ni->nid = nid;
 372
 373         /* Check nat cache */
 374         down_read(&nm_i->nat_tree_lock);
 375         e = __lookup_nat_cache(nm_i, nid);
 376         if (e) {
 377                 ni->ino = nat_get_ino(e);
 378                 ni->blk_addr = nat_get_blkaddr(e);
 379                 ni->version = nat_get_version(e);
 380                 up_read(&nm_i->nat_tree_lock);
 381                 return;
 382         }
 383
 384         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 385
 386         /* Check current segment summary */
 387         down_read(&curseg->journal_rwsem);
 388         i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 389         if (i >= 0) {
 390                 ne = nat_in_journal(journal, i);
 391                 node_info_from_raw_nat(ni, &ne);
 392         }
 393         up_read(&curseg->journal_rwsem);
 394         if (i >= 0)
 395                 goto cache;
 396
 397         /* Fill node_info from nat page */
 398         page = get_current_nat_page(sbi, start_nid);
 399         nat_blk = (struct f2fs_nat_block *)page_address(page);
 400         ne = nat_blk->entries[nid - start_nid];
 401         node_info_from_raw_nat(ni, &ne);
 402         f2fs_put_page(page, 1);
 403 cache:
 404         up_read(&nm_i->nat_tree_lock);
 405         /* cache nat entry */
 406         down_write(&nm_i->nat_tree_lock);
 407         cache_nat_entry(sbi, nid, &ne);
 408         up_write(&nm_i->nat_tree_lock);
 409 }
 410
 411 /*
 412  * readahead MAX_RA_NODE number of node pages.
 413  */
 414 static void ra_node_pages(struct page *parent, int start, int n)
 415 {
 416         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 417         struct blk_plug plug;
 418         int i, end;
 419         nid_t nid;
 420
 421         blk_start_plug(&plug);
 422
 423         /* Then, try readahead for siblings of the desired node */
 424         end = start + n;
 425         end = min(end, NIDS_PER_BLOCK);
 426         for (i = start; i < end; i++) {
 427                 nid = get_nid(parent, i, false);
 428                 ra_node_page(sbi, nid);
 429         }
 430
 431         blk_finish_plug(&plug);
 432 }
 433
 434 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 435 {
 436         const long direct_index = ADDRS_PER_INODE(dn->inode);
 437         const long direct_blks = ADDRS_PER_BLOCK;
 438         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 439         unsigned int skipped_unit = ADDRS_PER_BLOCK;
 440         int cur_level = dn->cur_level;
 441         int max_level = dn->max_level;
 442         pgoff_t base = 0;
 443
 444         if (!dn->max_level)
 445                 return pgofs + 1;
 446
 447         while (max_level-- > cur_level)
 448                 skipped_unit *= NIDS_PER_BLOCK;
 449
 450         switch (dn->max_level) {
 451         case 3:
 452                 base += 2 * indirect_blks;
 453         case 2:
 454                 base += 2 * direct_blks;
 455         case 1:
 456                 base += direct_index;
 457                 break;
 458         default:
 459                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 460         }
 461
 462         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 463 }
 464
 465 /*
 466  * The maximum depth is four.
 467  * Offset[0] will have raw inode offset.
 468  */
 469 static int get_node_path(struct inode *inode, long block,
 470                                 int offset[4], unsigned int noffset[4])
 471 {
 472         const long direct_index = ADDRS_PER_INODE(inode);
 473         const long direct_blks = ADDRS_PER_BLOCK;
 474         const long dptrs_per_blk = NIDS_PER_BLOCK;
 475         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 476         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 477         int n = 0;
 478         int level = 0;
 479
 480         noffset[0] = 0;
 481
 482         if (block < direct_index) {
 483                 offset[n] = block;
 484                 goto got;
 485         }
 486         block -= direct_index;
 487         if (block < direct_blks) {
 488                 offset[n++] = NODE_DIR1_BLOCK;
 489                 noffset[n] = 1;
 490                 offset[n] = block;
 491                 level = 1;
 492                 goto got;
 493         }
 494         block -= direct_blks;
 495         if (block < direct_blks) {
 496                 offset[n++] = NODE_DIR2_BLOCK;
 497                 noffset[n] = 2;
 498                 offset[n] = block;
 499                 level = 1;
 500                 goto got;
 501         }
 502         block -= direct_blks;
 503         if (block < indirect_blks) {
 504                 offset[n++] = NODE_IND1_BLOCK;
 505                 noffset[n] = 3;
 506                 offset[n++] = block / direct_blks;
 507                 noffset[n] = 4 + offset[n - 1];
 508                 offset[n] = block % direct_blks;
 509                 level = 2;
 510                 goto got;
 511         }
 512         block -= indirect_blks;
 513         if (block < indirect_blks) {
 514                 offset[n++] = NODE_IND2_BLOCK;
 515                 noffset[n] = 4 + dptrs_per_blk;
 516                 offset[n++] = block / direct_blks;
 517                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 518                 offset[n] = block % direct_blks;
 519                 level = 2;
 520                 goto got;
 521         }
 522         block -= indirect_blks;
 523         if (block < dindirect_blks) {
 524                 offset[n++] = NODE_DIND_BLOCK;
 525                 noffset[n] = 5 + (dptrs_per_blk * 2);
 526                 offset[n++] = block / indirect_blks;
 527                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 528                               offset[n - 1] * (dptrs_per_blk + 1);
 529                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 530                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 531                               offset[n - 2] * (dptrs_per_blk + 1) +
 532                               offset[n - 1];
 533                 offset[n] = block % direct_blks;
 534                 level = 3;
 535                 goto got;
 536         } else {
 537                 BUG();
 538         }
 539 got:
 540         return level;
 541 }
 542
 543 /*
 544  * Caller should call f2fs_put_dnode(dn).
 545  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 546  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 547  * In the case of RDONLY_NODE, we don't need to care about mutex.
 548  */
 549 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 550 {
 551         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 552         struct page *npage[4];
 553         struct page *parent = NULL;
 554         int offset[4];
 555         unsigned int noffset[4];
 556         nid_t nids[4];
 557         int level, i = 0;
 558         int err = 0;
 559
 560         level = get_node_path(dn->inode, index, offset, noffset);
 561
 562         nids[0] = dn->inode->i_ino;
 563         npage[0] = dn->inode_page;
 564
 565         if (!npage[0]) {
 566                 npage[0] = get_node_page(sbi, nids[0]);
 567                 if (IS_ERR(npage[0]))
 568                         return PTR_ERR(npage[0]);
 569         }
 570
 571         /* if inline_data is set, should not report any block indices */
 572         if (f2fs_has_inline_data(dn->inode) && index) {
 573                 err = -ENOENT;
 574                 f2fs_put_page(npage[0], 1);
 575                 goto release_out;
 576         }
 577
 578         parent = npage[0];
 579         if (level != 0)
 580                 nids[1] = get_nid(parent, offset[0], true);
 581         dn->inode_page = npage[0];
 582         dn->inode_page_locked = true;
 583
 584         /* get indirect or direct nodes */
 585         for (i = 1; i <= level; i++) {
 586                 bool done = false;
 587
 588                 if (!nids[i] && mode == ALLOC_NODE) {
 589                         /* alloc new node */
 590                         if (!alloc_nid(sbi, &(nids[i]))) {
 591                                 err = -ENOSPC;
 592                                 goto release_pages;
 593                         }
 594
 595                         dn->nid = nids[i];
 596                         npage[i] = new_node_page(dn, noffset[i], NULL);
 597                         if (IS_ERR(npage[i])) {
 598                                 alloc_nid_failed(sbi, nids[i]);
 599                                 err = PTR_ERR(npage[i]);
 600                                 goto release_pages;
 601                         }
 602
 603                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 604                         alloc_nid_done(sbi, nids[i]);
 605                         done = true;
 606                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 607                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 608                         if (IS_ERR(npage[i])) {
 609                                 err = PTR_ERR(npage[i]);
 610                                 goto release_pages;
 611                         }
 612                         done = true;
 613                 }
 614                 if (i == 1) {
 615                         dn->inode_page_locked = false;
 616                         unlock_page(parent);
 617                 } else {
 618                         f2fs_put_page(parent, 1);
 619                 }
 620
 621                 if (!done) {
 622                         npage[i] = get_node_page(sbi, nids[i]);
 623                         if (IS_ERR(npage[i])) {
 624                                 err = PTR_ERR(npage[i]);
 625                                 f2fs_put_page(npage[0], 0);
 626                                 goto release_out;
 627                         }
 628                 }
 629                 if (i < level) {
 630                         parent = npage[i];
 631                         nids[i + 1] = get_nid(parent, offset[i], false);
 632                 }
 633         }
 634         dn->nid = nids[level];
 635         dn->ofs_in_node = offset[level];
 636         dn->node_page = npage[level];
 637         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 638         return 0;
 639
 640 release_pages:
 641         f2fs_put_page(parent, 1);
 642         if (i > 1)
 643                 f2fs_put_page(npage[0], 0);
 644 release_out:
 645         dn->inode_page = NULL;
 646         dn->node_page = NULL;
 647         if (err == -ENOENT) {
 648                 dn->cur_level = i;
 649                 dn->max_level = level;
 650                 dn->ofs_in_node = offset[level];
 651         }
 652         return err;
 653 }
 654
 655 static void truncate_node(struct dnode_of_data *dn)
 656 {
 657         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 658         struct node_info ni;
 659         pgoff_t index;
 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         index = dn->node_page->index;
 683         f2fs_put_page(dn->node_page, 1);
 684
 685         invalidate_mapping_pages(NODE_MAPPING(sbi),
 686                         index, index);
 687
 688         dn->node_page = NULL;
 689         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 690 }
 691
 692 static int truncate_dnode(struct dnode_of_data *dn)
 693 {
 694         struct page *page;
 695
 696         if (dn->nid == 0)
 697                 return 1;
 698
 699         /* get direct node */
 700         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 701         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 702                 return 1;
 703         else if (IS_ERR(page))
 704                 return PTR_ERR(page);
 705
 706         /* Make dnode_of_data for parameter */
 707         dn->node_page = page;
 708         dn->ofs_in_node = 0;
 709         truncate_data_blocks(dn);
 710         truncate_node(dn);
 711         return 1;
 712 }
 713
 714 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 715                                                 int ofs, int depth)
 716 {
 717         struct dnode_of_data rdn = *dn;
 718         struct page *page;
 719         struct f2fs_node *rn;
 720         nid_t child_nid;
 721         unsigned int child_nofs;
 722         int freed = 0;
 723         int i, ret;
 724
 725         if (dn->nid == 0)
 726                 return NIDS_PER_BLOCK + 1;
 727
 728         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 729
 730         page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 731         if (IS_ERR(page)) {
 732                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 733                 return PTR_ERR(page);
 734         }
 735
 736         ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 737
 738         rn = F2FS_NODE(page);
 739         if (depth < 3) {
 740                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 741                         child_nid = le32_to_cpu(rn->in.nid[i]);
 742                         if (child_nid == 0)
 743                                 continue;
 744                         rdn.nid = child_nid;
 745                         ret = truncate_dnode(&rdn);
 746                         if (ret < 0)
 747                                 goto out_err;
 748                         if (set_nid(page, i, 0, false))
 749                                 dn->node_changed = true;
 750                 }
 751         } else {
 752                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 753                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 754                         child_nid = le32_to_cpu(rn->in.nid[i]);
 755                         if (child_nid == 0) {
 756                                 child_nofs += NIDS_PER_BLOCK + 1;
 757                                 continue;
 758                         }
 759                         rdn.nid = child_nid;
 760                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 761                         if (ret == (NIDS_PER_BLOCK + 1)) {
 762                                 if (set_nid(page, i, 0, false))
 763                                         dn->node_changed = true;
 764                                 child_nofs += ret;
 765                         } else if (ret < 0 && ret != -ENOENT) {
 766                                 goto out_err;
 767                         }
 768                 }
 769                 freed = child_nofs;
 770         }
 771
 772         if (!ofs) {
 773                 /* remove current indirect node */
 774                 dn->node_page = page;
 775                 truncate_node(dn);
 776                 freed++;
 777         } else {
 778                 f2fs_put_page(page, 1);
 779         }
 780         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 781         return freed;
 782
 783 out_err:
 784         f2fs_put_page(page, 1);
 785         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 786         return ret;
 787 }
 788
 789 static int truncate_partial_nodes(struct dnode_of_data *dn,
 790                         struct f2fs_inode *ri, int *offset, int depth)
 791 {
 792         struct page *pages[2];
 793         nid_t nid[3];
 794         nid_t child_nid;
 795         int err = 0;
 796         int i;
 797         int idx = depth - 2;
 798
 799         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 800         if (!nid[0])
 801                 return 0;
 802
 803         /* get indirect nodes in the path */
 804         for (i = 0; i < idx + 1; i++) {
 805                 /* reference count'll be increased */
 806                 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
 807                 if (IS_ERR(pages[i])) {
 808                         err = PTR_ERR(pages[i]);
 809                         idx = i - 1;
 810                         goto fail;
 811                 }
 812                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 813         }
 814
 815         ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
 816
 817         /* free direct nodes linked to a partial indirect node */
 818         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 819                 child_nid = get_nid(pages[idx], i, false);
 820                 if (!child_nid)
 821                         continue;
 822                 dn->nid = child_nid;
 823                 err = truncate_dnode(dn);
 824                 if (err < 0)
 825                         goto fail;
 826                 if (set_nid(pages[idx], i, 0, false))
 827                         dn->node_changed = true;
 828         }
 829
 830         if (offset[idx + 1] == 0) {
 831                 dn->node_page = pages[idx];
 832                 dn->nid = nid[idx];
 833                 truncate_node(dn);
 834         } else {
 835                 f2fs_put_page(pages[idx], 1);
 836         }
 837         offset[idx]++;
 838         offset[idx + 1] = 0;
 839         idx--;
 840 fail:
 841         for (i = idx; i >= 0; i--)
 842                 f2fs_put_page(pages[i], 1);
 843
 844         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 845
 846         return err;
 847 }
 848
 849 /*
 850  * All the block addresses of data and nodes should be nullified.
 851  */
 852 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 853 {
 854         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 855         int err = 0, cont = 1;
 856         int level, offset[4], noffset[4];
 857         unsigned int nofs = 0;
 858         struct f2fs_inode *ri;
 859         struct dnode_of_data dn;
 860         struct page *page;
 861
 862         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 863
 864         level = get_node_path(inode, from, offset, noffset);
 865
 866         page = get_node_page(sbi, inode->i_ino);
 867         if (IS_ERR(page)) {
 868                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 869                 return PTR_ERR(page);
 870         }
 871
 872         set_new_dnode(&dn, inode, page, NULL, 0);
 873         unlock_page(page);
 874
 875         ri = F2FS_INODE(page);
 876         switch (level) {
 877         case 0:
 878         case 1:
 879                 nofs = noffset[1];
 880                 break;
 881         case 2:
 882                 nofs = noffset[1];
 883                 if (!offset[level - 1])
 884                         goto skip_partial;
 885                 err = truncate_partial_nodes(&dn, ri, offset, level);
 886                 if (err < 0 && err != -ENOENT)
 887                         goto fail;
 888                 nofs += 1 + NIDS_PER_BLOCK;
 889                 break;
 890         case 3:
 891                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 892                 if (!offset[level - 1])
 893                         goto skip_partial;
 894                 err = truncate_partial_nodes(&dn, ri, offset, level);
 895                 if (err < 0 && err != -ENOENT)
 896                         goto fail;
 897                 break;
 898         default:
 899                 BUG();
 900         }
 901
 902 skip_partial:
 903         while (cont) {
 904                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 905                 switch (offset[0]) {
 906                 case NODE_DIR1_BLOCK:
 907                 case NODE_DIR2_BLOCK:
 908                         err = truncate_dnode(&dn);
 909                         break;
 910
 911                 case NODE_IND1_BLOCK:
 912                 case NODE_IND2_BLOCK:
 913                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 914                         break;
 915
 916                 case NODE_DIND_BLOCK:
 917                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 918                         cont = 0;
 919                         break;
 920
 921                 default:
 922                         BUG();
 923                 }
 924                 if (err < 0 && err != -ENOENT)
 925                         goto fail;
 926                 if (offset[1] == 0 &&
 927                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 928                         lock_page(page);
 929                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
 930                         f2fs_wait_on_page_writeback(page, NODE, true);
 931                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 932                         set_page_dirty(page);
 933                         unlock_page(page);
 934                 }
 935                 offset[1] = 0;
 936                 offset[0]++;
 937                 nofs += err;
 938         }
 939 fail:
 940         f2fs_put_page(page, 0);
 941         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 942         return err > 0 ? 0 : err;
 943 }
 944
 945 int truncate_xattr_node(struct inode *inode, struct page *page)
 946 {
 947         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 948         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 949         struct dnode_of_data dn;
 950         struct page *npage;
 951
 952         if (!nid)
 953                 return 0;
 954
 955         npage = get_node_page(sbi, nid);
 956         if (IS_ERR(npage))
 957                 return PTR_ERR(npage);
 958
 959         f2fs_i_xnid_write(inode, 0);
 960
 961         /* need to do checkpoint during fsync */
 962         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 963
 964         set_new_dnode(&dn, inode, page, npage, nid);
 965
 966         if (page)
 967                 dn.inode_page_locked = true;
 968         truncate_node(&dn);
 969         return 0;
 970 }
 971
 972 /*
 973  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 974  * f2fs_unlock_op().
 975  */
 976 int remove_inode_page(struct inode *inode)
 977 {
 978         struct dnode_of_data dn;
 979         int err;
 980
 981         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 982         err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
 983         if (err)
 984                 return err;
 985
 986         err = truncate_xattr_node(inode, dn.inode_page);
 987         if (err) {
 988                 f2fs_put_dnode(&dn);
 989                 return err;
 990         }
 991
 992         /* remove potential inline_data blocks */
 993         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 994                                 S_ISLNK(inode->i_mode))
 995                 truncate_data_blocks_range(&dn, 1);
 996
 997         /* 0 is possible, after f2fs_new_inode() has failed */
 998         f2fs_bug_on(F2FS_I_SB(inode),
 999                         inode->i_blocks != 0 && inode->i_blocks != 1);
1000
1001         /* will put inode & node pages */
1002         truncate_node(&dn);
1003         return 0;
1004 }
1005
1006 struct page *new_inode_page(struct inode *inode)
1007 {
1008         struct dnode_of_data dn;
1009
1010         /* allocate inode page for new inode */
1011         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1012
1013         /* caller should f2fs_put_page(page, 1); */
1014         return new_node_page(&dn, 0, NULL);
1015 }
1016
1017 struct page *new_node_page(struct dnode_of_data *dn,
1018                                 unsigned int ofs, struct page *ipage)
1019 {
1020         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1021         struct node_info old_ni, new_ni;
1022         struct page *page;
1023         int err;
1024
1025         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1026                 return ERR_PTR(-EPERM);
1027
1028         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1029         if (!page)
1030                 return ERR_PTR(-ENOMEM);
1031
1032         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1033                 err = -ENOSPC;
1034                 goto fail;
1035         }
1036
1037         get_node_info(sbi, dn->nid, &old_ni);
1038
1039         /* Reinitialize old_ni with new node page */
1040         f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1041         new_ni = old_ni;
1042         new_ni.ino = dn->inode->i_ino;
1043         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1044
1045         f2fs_wait_on_page_writeback(page, NODE, true);
1046         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1047         set_cold_node(dn->inode, page);
1048         if (!PageUptodate(page))
1049                 SetPageUptodate(page);
1050         if (set_page_dirty(page))
1051                 dn->node_changed = true;
1052
1053         if (f2fs_has_xattr_block(ofs))
1054                 f2fs_i_xnid_write(dn->inode, dn->nid);
1055
1056         if (ofs == 0)
1057                 inc_valid_inode_count(sbi);
1058         return page;
1059
1060 fail:
1061         clear_node_page_dirty(page);
1062         f2fs_put_page(page, 1);
1063         return ERR_PTR(err);
1064 }
1065
1066 /*
1067  * Caller should do after getting the following values.
1068  * 0: f2fs_put_page(page, 0)
1069  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1070  */
1071 static int read_node_page(struct page *page, int op_flags)
1072 {
1073         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1074         struct node_info ni;
1075         struct f2fs_io_info fio = {
1076                 .sbi = sbi,
1077                 .type = NODE,
1078                 .op = REQ_OP_READ,
1079                 .op_flags = op_flags,
1080                 .page = page,
1081                 .encrypted_page = NULL,
1082         };
1083
1084         if (PageUptodate(page))
1085                 return LOCKED_PAGE;
1086
1087         get_node_info(sbi, page->index, &ni);
1088
1089         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1090                 ClearPageUptodate(page);
1091                 return -ENOENT;
1092         }
1093
1094         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1095         return f2fs_submit_page_bio(&fio);
1096 }
1097
1098 /*
1099  * Readahead a node page
1100  */
1101 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1102 {
1103         struct page *apage;
1104         int err;
1105
1106         if (!nid)
1107                 return;
1108         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1109
1110         rcu_read_lock();
1111         apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1112         rcu_read_unlock();
1113         if (apage)
1114                 return;
1115
1116         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1117         if (!apage)
1118                 return;
1119
1120         err = read_node_page(apage, REQ_RAHEAD);
1121         f2fs_put_page(apage, err ? 1 : 0);
1122 }
1123
1124 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1125                                         struct page *parent, int start)
1126 {
1127         struct page *page;
1128         int err;
1129
1130         if (!nid)
1131                 return ERR_PTR(-ENOENT);
1132         f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1133 repeat:
1134         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1135         if (!page)
1136                 return ERR_PTR(-ENOMEM);
1137
1138         err = read_node_page(page, READ_SYNC);
1139         if (err < 0) {
1140                 f2fs_put_page(page, 1);
1141                 return ERR_PTR(err);
1142         } else if (err == LOCKED_PAGE) {
1143                 goto page_hit;
1144         }
1145
1146         if (parent)
1147                 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1148
1149         lock_page(page);
1150
1151         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1152                 f2fs_put_page(page, 1);
1153                 goto repeat;
1154         }
1155
1156         if (unlikely(!PageUptodate(page)))
1157                 goto out_err;
1158 page_hit:
1159         if(unlikely(nid != nid_of_node(page))) {
1160                 f2fs_bug_on(sbi, 1);
1161                 ClearPageUptodate(page);
1162 out_err:
1163                 f2fs_put_page(page, 1);
1164                 return ERR_PTR(-EIO);
1165         }
1166         return page;
1167 }
1168
1169 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1170 {
1171         return __get_node_page(sbi, nid, NULL, 0);
1172 }
1173
1174 struct page *get_node_page_ra(struct page *parent, int start)
1175 {
1176         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1177         nid_t nid = get_nid(parent, start, false);
1178
1179         return __get_node_page(sbi, nid, parent, start);
1180 }
1181
1182 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1183 {
1184         struct inode *inode;
1185         struct page *page;
1186         int ret;
1187
1188         /* should flush inline_data before evict_inode */
1189         inode = ilookup(sbi->sb, ino);
1190         if (!inode)
1191                 return;
1192
1193         page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1194         if (!page)
1195                 goto iput_out;
1196
1197         if (!PageUptodate(page))
1198                 goto page_out;
1199
1200         if (!PageDirty(page))
1201                 goto page_out;
1202
1203         if (!clear_page_dirty_for_io(page))
1204                 goto page_out;
1205
1206         ret = f2fs_write_inline_data(inode, page);
1207         inode_dec_dirty_pages(inode);
1208         if (ret)
1209                 set_page_dirty(page);
1210 page_out:
1211         f2fs_put_page(page, 1);
1212 iput_out:
1213         iput(inode);
1214 }
1215
1216 void move_node_page(struct page *node_page, int gc_type)
1217 {
1218         if (gc_type == FG_GC) {
1219                 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1220                 struct writeback_control wbc = {
1221                         .sync_mode = WB_SYNC_ALL,
1222                         .nr_to_write = 1,
1223                         .for_reclaim = 0,
1224                 };
1225
1226                 set_page_dirty(node_page);
1227                 f2fs_wait_on_page_writeback(node_page, NODE, true);
1228
1229                 f2fs_bug_on(sbi, PageWriteback(node_page));
1230                 if (!clear_page_dirty_for_io(node_page))
1231                         goto out_page;
1232
1233                 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1234                         unlock_page(node_page);
1235                 goto release_page;
1236         } else {
1237                 /* set page dirty and write it */
1238                 if (!PageWriteback(node_page))
1239                         set_page_dirty(node_page);
1240         }
1241 out_page:
1242         unlock_page(node_page);
1243 release_page:
1244         f2fs_put_page(node_page, 0);
1245 }
1246
1247 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1248 {
1249         pgoff_t index, end;
1250         struct pagevec pvec;
1251         struct page *last_page = NULL;
1252
1253         pagevec_init(&pvec, 0);
1254         index = 0;
1255         end = ULONG_MAX;
1256
1257         while (index <= end) {
1258                 int i, nr_pages;
1259                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1260                                 PAGECACHE_TAG_DIRTY,
1261                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1262                 if (nr_pages == 0)
1263                         break;
1264
1265                 for (i = 0; i < nr_pages; i++) {
1266                         struct page *page = pvec.pages[i];
1267
1268                         if (unlikely(f2fs_cp_error(sbi))) {
1269                                 f2fs_put_page(last_page, 0);
1270                                 pagevec_release(&pvec);
1271                                 return ERR_PTR(-EIO);
1272                         }
1273
1274                         if (!IS_DNODE(page) || !is_cold_node(page))
1275                                 continue;
1276                         if (ino_of_node(page) != ino)
1277                                 continue;
1278
1279                         lock_page(page);
1280
1281                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1282 continue_unlock:
1283                                 unlock_page(page);
1284                                 continue;
1285                         }
1286                         if (ino_of_node(page) != ino)
1287                                 goto continue_unlock;
1288
1289                         if (!PageDirty(page)) {
1290                                 /* someone wrote it for us */
1291                                 goto continue_unlock;
1292                         }
1293
1294                         if (last_page)
1295                                 f2fs_put_page(last_page, 0);
1296
1297                         get_page(page);
1298                         last_page = page;
1299                         unlock_page(page);
1300                 }
1301                 pagevec_release(&pvec);
1302                 cond_resched();
1303         }
1304         return last_page;
1305 }
1306
1307 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1308                         struct writeback_control *wbc, bool atomic)
1309 {
1310         pgoff_t index, end;
1311         struct pagevec pvec;
1312         int ret = 0;
1313         struct page *last_page = NULL;
1314         bool marked = false;
1315         nid_t ino = inode->i_ino;
1316         int nwritten = 0;
1317
1318         if (atomic) {
1319                 last_page = last_fsync_dnode(sbi, ino);
1320                 if (IS_ERR_OR_NULL(last_page))
1321                         return PTR_ERR_OR_ZERO(last_page);
1322         }
1323 retry:
1324         pagevec_init(&pvec, 0);
1325         index = 0;
1326         end = ULONG_MAX;
1327
1328         while (index <= end) {
1329                 int i, nr_pages;
1330                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1331                                 PAGECACHE_TAG_DIRTY,
1332                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1333                 if (nr_pages == 0)
1334                         break;
1335
1336                 for (i = 0; i < nr_pages; i++) {
1337                         struct page *page = pvec.pages[i];
1338
1339                         if (unlikely(f2fs_cp_error(sbi))) {
1340                                 f2fs_put_page(last_page, 0);
1341                                 pagevec_release(&pvec);
1342                                 return -EIO;
1343                         }
1344
1345                         if (!IS_DNODE(page) || !is_cold_node(page))
1346                                 continue;
1347                         if (ino_of_node(page) != ino)
1348                                 continue;
1349
1350                         lock_page(page);
1351
1352                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1353 continue_unlock:
1354                                 unlock_page(page);
1355                                 continue;
1356                         }
1357                         if (ino_of_node(page) != ino)
1358                                 goto continue_unlock;
1359
1360                         if (!PageDirty(page) && page != last_page) {
1361                                 /* someone wrote it for us */
1362                                 goto continue_unlock;
1363                         }
1364
1365                         f2fs_wait_on_page_writeback(page, NODE, true);
1366                         BUG_ON(PageWriteback(page));
1367
1368                         if (!atomic || page == last_page) {
1369                                 set_fsync_mark(page, 1);
1370                                 if (IS_INODE(page)) {
1371                                         if (is_inode_flag_set(inode,
1372                                                                 FI_DIRTY_INODE))
1373                                                 update_inode(inode, page);
1374                                         set_dentry_mark(page,
1375                                                 need_dentry_mark(sbi, ino));
1376                                 }
1377                                 /*  may be written by other thread */
1378                                 if (!PageDirty(page))
1379                                         set_page_dirty(page);
1380                         }
1381
1382                         if (!clear_page_dirty_for_io(page))
1383                                 goto continue_unlock;
1384
1385                         ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1386                         if (ret) {
1387                                 unlock_page(page);
1388                                 f2fs_put_page(last_page, 0);
1389                                 break;
1390                         } else {
1391                                 nwritten++;
1392                         }
1393
1394                         if (page == last_page) {
1395                                 f2fs_put_page(page, 0);
1396                                 marked = true;
1397                                 break;
1398                         }
1399                 }
1400                 pagevec_release(&pvec);
1401                 cond_resched();
1402
1403                 if (ret || marked)
1404                         break;
1405         }
1406         if (!ret && atomic && !marked) {
1407                 f2fs_msg(sbi->sb, KERN_DEBUG,
1408                         "Retry to write fsync mark: ino=%u, idx=%lx",
1409                                         ino, last_page->index);
1410                 lock_page(last_page);
1411                 set_page_dirty(last_page);
1412                 unlock_page(last_page);
1413                 goto retry;
1414         }
1415
1416         if (nwritten)
1417                 f2fs_submit_merged_bio_cond(sbi, NULL, NULL, ino, NODE, WRITE);
1418         return ret ? -EIO: 0;
1419 }
1420
1421 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1422 {
1423         pgoff_t index, end;
1424         struct pagevec pvec;
1425         int step = 0;
1426         int nwritten = 0;
1427         int ret = 0;
1428
1429         pagevec_init(&pvec, 0);
1430
1431 next_step:
1432         index = 0;
1433         end = ULONG_MAX;
1434
1435         while (index <= end) {
1436                 int i, nr_pages;
1437                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1438                                 PAGECACHE_TAG_DIRTY,
1439                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1440                 if (nr_pages == 0)
1441                         break;
1442
1443                 for (i = 0; i < nr_pages; i++) {
1444                         struct page *page = pvec.pages[i];
1445
1446                         if (unlikely(f2fs_cp_error(sbi))) {
1447                                 pagevec_release(&pvec);
1448                                 ret = -EIO;
1449                                 goto out;
1450                         }
1451
1452                         /*
1453                          * flushing sequence with step:
1454                          * 0. indirect nodes
1455                          * 1. dentry dnodes
1456                          * 2. file dnodes
1457                          */
1458                         if (step == 0 && IS_DNODE(page))
1459                                 continue;
1460                         if (step == 1 && (!IS_DNODE(page) ||
1461                                                 is_cold_node(page)))
1462                                 continue;
1463                         if (step == 2 && (!IS_DNODE(page) ||
1464                                                 !is_cold_node(page)))
1465                                 continue;
1466 lock_node:
1467                         if (wbc->sync_mode == WB_SYNC_ALL)
1468                                 lock_page(page);
1469                         else if (!trylock_page(page))
1470                                 continue;
1471
1472                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1473 continue_unlock:
1474                                 unlock_page(page);
1475                                 continue;
1476                         }
1477
1478                         if (!PageDirty(page)) {
1479                                 /* someone wrote it for us */
1480                                 goto continue_unlock;
1481                         }
1482
1483                         /* flush inline_data */
1484                         if (is_inline_node(page)) {
1485                                 clear_inline_node(page);
1486                                 unlock_page(page);
1487                                 flush_inline_data(sbi, ino_of_node(page));
1488                                 goto lock_node;
1489                         }
1490
1491                         f2fs_wait_on_page_writeback(page, NODE, true);
1492
1493                         BUG_ON(PageWriteback(page));
1494                         if (!clear_page_dirty_for_io(page))
1495                                 goto continue_unlock;
1496
1497                         set_fsync_mark(page, 0);
1498                         set_dentry_mark(page, 0);
1499
1500                         if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1501                                 unlock_page(page);
1502                         else
1503                                 nwritten++;
1504
1505                         if (--wbc->nr_to_write == 0)
1506                                 break;
1507                 }
1508                 pagevec_release(&pvec);
1509                 cond_resched();
1510
1511                 if (wbc->nr_to_write == 0) {
1512                         step = 2;
1513                         break;
1514                 }
1515         }
1516
1517         if (step < 2) {
1518                 step++;
1519                 goto next_step;
1520         }
1521 out:
1522         if (nwritten)
1523                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1524         return ret;
1525 }
1526
1527 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1528 {
1529         pgoff_t index = 0, end = ULONG_MAX;
1530         struct pagevec pvec;
1531         int ret2, ret = 0;
1532
1533         pagevec_init(&pvec, 0);
1534
1535         while (index <= end) {
1536                 int i, nr_pages;
1537                 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1538                                 PAGECACHE_TAG_WRITEBACK,
1539                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1540                 if (nr_pages == 0)
1541                         break;
1542
1543                 for (i = 0; i < nr_pages; i++) {
1544                         struct page *page = pvec.pages[i];
1545
1546                         /* until radix tree lookup accepts end_index */
1547                         if (unlikely(page->index > end))
1548                                 continue;
1549
1550                         if (ino && ino_of_node(page) == ino) {
1551                                 f2fs_wait_on_page_writeback(page, NODE, true);
1552                                 if (TestClearPageError(page))
1553                                         ret = -EIO;
1554                         }
1555                 }
1556                 pagevec_release(&pvec);
1557                 cond_resched();
1558         }
1559
1560         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
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         if (__is_valid_data_blkaddr(ni.blk_addr) &&
1611                 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC)) {
1612                 up_read(&sbi->node_write);
1613                 goto redirty_out;
1614         }
1615
1616         set_page_writeback(page);
1617         fio.old_blkaddr = ni.blk_addr;
1618         write_node_page(nid, &fio);
1619         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1620         dec_page_count(sbi, F2FS_DIRTY_NODES);
1621         up_read(&sbi->node_write);
1622
1623         if (wbc->for_reclaim)
1624                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1625
1626         unlock_page(page);
1627
1628         if (unlikely(f2fs_cp_error(sbi)))
1629                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1630
1631         return 0;
1632
1633 redirty_out:
1634         redirty_page_for_writepage(wbc, page);
1635         return AOP_WRITEPAGE_ACTIVATE;
1636 }
1637
1638 static int f2fs_write_node_pages(struct address_space *mapping,
1639                             struct writeback_control *wbc)
1640 {
1641         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1642         struct blk_plug plug;
1643         long diff;
1644
1645         /* balancing f2fs's metadata in background */
1646         f2fs_balance_fs_bg(sbi);
1647
1648         /* collect a number of dirty node pages and write together */
1649         if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1650                 goto skip_write;
1651
1652         trace_f2fs_writepages(mapping->host, wbc, NODE);
1653
1654         diff = nr_pages_to_write(sbi, NODE, wbc);
1655         wbc->sync_mode = WB_SYNC_NONE;
1656         blk_start_plug(&plug);
1657         sync_node_pages(sbi, wbc);
1658         blk_finish_plug(&plug);
1659         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1660         return 0;
1661
1662 skip_write:
1663         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1664         trace_f2fs_writepages(mapping->host, wbc, NODE);
1665         return 0;
1666 }
1667
1668 static int f2fs_set_node_page_dirty(struct page *page)
1669 {
1670         trace_f2fs_set_page_dirty(page, NODE);
1671
1672         if (!PageUptodate(page))
1673                 SetPageUptodate(page);
1674         if (!PageDirty(page)) {
1675                 f2fs_set_page_dirty_nobuffers(page);
1676                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1677                 SetPagePrivate(page);
1678                 f2fs_trace_pid(page);
1679                 return 1;
1680         }
1681         return 0;
1682 }
1683
1684 /*
1685  * Structure of the f2fs node operations
1686  */
1687 const struct address_space_operations f2fs_node_aops = {
1688         .writepage      = f2fs_write_node_page,
1689         .writepages     = f2fs_write_node_pages,
1690         .set_page_dirty = f2fs_set_node_page_dirty,
1691         .invalidatepage = f2fs_invalidate_page,
1692         .releasepage    = f2fs_release_page,
1693 #ifdef CONFIG_MIGRATION
1694         .migratepage    = f2fs_migrate_page,
1695 #endif
1696 };
1697
1698 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1699                                                 nid_t n)
1700 {
1701         return radix_tree_lookup(&nm_i->free_nid_root, n);
1702 }
1703
1704 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1705                                                 struct free_nid *i)
1706 {
1707         list_del(&i->list);
1708         radix_tree_delete(&nm_i->free_nid_root, i->nid);
1709 }
1710
1711 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1712 {
1713         struct f2fs_nm_info *nm_i = NM_I(sbi);
1714         struct free_nid *i, *e;
1715         struct nat_entry *ne;
1716         int err = -EINVAL;
1717
1718         if (!available_free_memory(sbi, FREE_NIDS))
1719                 return -1;
1720
1721         /* 0 nid should not be used */
1722         if (unlikely(nid == 0))
1723                 return 0;
1724
1725         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1726         i->nid = nid;
1727         i->state = NID_NEW;
1728
1729         if (radix_tree_preload(GFP_NOFS))
1730                 goto err;
1731
1732         spin_lock(&nm_i->free_nid_list_lock);
1733
1734         if (build) {
1735                 /*
1736                  *   Thread A             Thread B
1737                  *  - f2fs_create
1738                  *   - f2fs_new_inode
1739                  *    - alloc_nid
1740                  *     - __insert_nid_to_list(ALLOC_NID_LIST)
1741                  *                     - f2fs_balance_fs_bg
1742                  *                      - build_free_nids
1743                  *                       - __build_free_nids
1744                  *                        - scan_nat_page
1745                  *                         - add_free_nid
1746                  *                          - __lookup_nat_cache
1747                  *  - f2fs_add_link
1748                  *   - init_inode_metadata
1749                  *    - new_inode_page
1750                  *     - new_node_page
1751                  *      - set_node_addr
1752                  *  - alloc_nid_done
1753                  *   - __remove_nid_from_list(ALLOC_NID_LIST)
1754                  *                         - __insert_nid_to_list(FREE_NID_LIST)
1755                  */
1756                 ne = __lookup_nat_cache(nm_i, nid);
1757                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1758                                 nat_get_blkaddr(ne) != NULL_ADDR))
1759                         goto err_out;
1760
1761                 e = __lookup_free_nid_list(nm_i, nid);
1762                 if (e)
1763                         goto err_out;
1764         }
1765         if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i))
1766                 goto err_out;
1767         err = 0;
1768         list_add_tail(&i->list, &nm_i->free_nid_list);
1769         nm_i->fcnt++;
1770 err_out:
1771         spin_unlock(&nm_i->free_nid_list_lock);
1772         radix_tree_preload_end();
1773 err:
1774         if (err)
1775                 kmem_cache_free(free_nid_slab, i);
1776         return !err;
1777 }
1778
1779 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1780 {
1781         struct free_nid *i;
1782         bool need_free = false;
1783
1784         spin_lock(&nm_i->free_nid_list_lock);
1785         i = __lookup_free_nid_list(nm_i, nid);
1786         if (i && i->state == NID_NEW) {
1787                 __del_from_free_nid_list(nm_i, i);
1788                 nm_i->fcnt--;
1789                 need_free = true;
1790         }
1791         spin_unlock(&nm_i->free_nid_list_lock);
1792
1793         if (need_free)
1794                 kmem_cache_free(free_nid_slab, i);
1795 }
1796
1797 static void scan_nat_page(struct f2fs_sb_info *sbi,
1798                         struct page *nat_page, nid_t start_nid)
1799 {
1800         struct f2fs_nm_info *nm_i = NM_I(sbi);
1801         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1802         block_t blk_addr;
1803         int i;
1804
1805         i = start_nid % NAT_ENTRY_PER_BLOCK;
1806
1807         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1808
1809                 if (unlikely(start_nid >= nm_i->max_nid))
1810                         break;
1811
1812                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1813                 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1814                 if (blk_addr == NULL_ADDR) {
1815                         if (add_free_nid(sbi, start_nid, true) < 0)
1816                                 break;
1817                 }
1818         }
1819 }
1820
1821 void build_free_nids(struct f2fs_sb_info *sbi)
1822 {
1823         struct f2fs_nm_info *nm_i = NM_I(sbi);
1824         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1825         struct f2fs_journal *journal = curseg->journal;
1826         int i = 0;
1827         nid_t nid = nm_i->next_scan_nid;
1828
1829         /* Enough entries */
1830         if (nm_i->fcnt >= NAT_ENTRY_PER_BLOCK)
1831                 return;
1832
1833         /* readahead nat pages to be scanned */
1834         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1835                                                         META_NAT, true);
1836
1837         down_read(&nm_i->nat_tree_lock);
1838
1839         while (1) {
1840                 struct page *page = get_current_nat_page(sbi, nid);
1841
1842                 scan_nat_page(sbi, page, nid);
1843                 f2fs_put_page(page, 1);
1844
1845                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1846                 if (unlikely(nid >= nm_i->max_nid))
1847                         nid = 0;
1848
1849                 if (++i >= FREE_NID_PAGES)
1850                         break;
1851         }
1852
1853         /* go to the next free nat pages to find free nids abundantly */
1854         nm_i->next_scan_nid = nid;
1855
1856         /* find free nids from current sum_pages */
1857         down_read(&curseg->journal_rwsem);
1858         for (i = 0; i < nats_in_cursum(journal); i++) {
1859                 block_t addr;
1860
1861                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1862                 nid = le32_to_cpu(nid_in_journal(journal, i));
1863                 if (addr == NULL_ADDR)
1864                         add_free_nid(sbi, nid, true);
1865                 else
1866                         remove_free_nid(nm_i, nid);
1867         }
1868         up_read(&curseg->journal_rwsem);
1869         up_read(&nm_i->nat_tree_lock);
1870
1871         ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1872                                         nm_i->ra_nid_pages, META_NAT, false);
1873 }
1874
1875 /*
1876  * If this function returns success, caller can obtain a new nid
1877  * from second parameter of this function.
1878  * The returned nid could be used ino as well as nid when inode is created.
1879  */
1880 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1881 {
1882         struct f2fs_nm_info *nm_i = NM_I(sbi);
1883         struct free_nid *i = NULL;
1884 retry:
1885 #ifdef CONFIG_F2FS_FAULT_INJECTION
1886         if (time_to_inject(sbi, FAULT_ALLOC_NID))
1887                 return false;
1888 #endif
1889         if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1890                 return false;
1891
1892         spin_lock(&nm_i->free_nid_list_lock);
1893
1894         /* We should not use stale free nids created by build_free_nids */
1895         if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1896                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1897                 list_for_each_entry(i, &nm_i->free_nid_list, list)
1898                         if (i->state == NID_NEW)
1899                                 break;
1900
1901                 f2fs_bug_on(sbi, i->state != NID_NEW);
1902                 *nid = i->nid;
1903                 i->state = NID_ALLOC;
1904                 nm_i->fcnt--;
1905                 spin_unlock(&nm_i->free_nid_list_lock);
1906                 return true;
1907         }
1908         spin_unlock(&nm_i->free_nid_list_lock);
1909
1910         /* Let's scan nat pages and its caches to get free nids */
1911         mutex_lock(&nm_i->build_lock);
1912         build_free_nids(sbi);
1913         mutex_unlock(&nm_i->build_lock);
1914         goto retry;
1915 }
1916
1917 /*
1918  * alloc_nid() should be called prior to this function.
1919  */
1920 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1921 {
1922         struct f2fs_nm_info *nm_i = NM_I(sbi);
1923         struct free_nid *i;
1924
1925         spin_lock(&nm_i->free_nid_list_lock);
1926         i = __lookup_free_nid_list(nm_i, nid);
1927         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1928         __del_from_free_nid_list(nm_i, i);
1929         spin_unlock(&nm_i->free_nid_list_lock);
1930
1931         kmem_cache_free(free_nid_slab, i);
1932 }
1933
1934 /*
1935  * alloc_nid() should be called prior to this function.
1936  */
1937 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1938 {
1939         struct f2fs_nm_info *nm_i = NM_I(sbi);
1940         struct free_nid *i;
1941         bool need_free = false;
1942
1943         if (!nid)
1944                 return;
1945
1946         spin_lock(&nm_i->free_nid_list_lock);
1947         i = __lookup_free_nid_list(nm_i, nid);
1948         f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1949         if (!available_free_memory(sbi, FREE_NIDS)) {
1950                 __del_from_free_nid_list(nm_i, i);
1951                 need_free = true;
1952         } else {
1953                 i->state = NID_NEW;
1954                 nm_i->fcnt++;
1955         }
1956         spin_unlock(&nm_i->free_nid_list_lock);
1957
1958         if (need_free)
1959                 kmem_cache_free(free_nid_slab, i);
1960 }
1961
1962 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1963 {
1964         struct f2fs_nm_info *nm_i = NM_I(sbi);
1965         struct free_nid *i, *next;
1966         int nr = nr_shrink;
1967
1968         if (nm_i->fcnt <= MAX_FREE_NIDS)
1969                 return 0;
1970
1971         if (!mutex_trylock(&nm_i->build_lock))
1972                 return 0;
1973
1974         spin_lock(&nm_i->free_nid_list_lock);
1975         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1976                 if (nr_shrink <= 0 || nm_i->fcnt <= MAX_FREE_NIDS)
1977                         break;
1978                 if (i->state == NID_ALLOC)
1979                         continue;
1980                 __del_from_free_nid_list(nm_i, i);
1981                 kmem_cache_free(free_nid_slab, i);
1982                 nm_i->fcnt--;
1983                 nr_shrink--;
1984         }
1985         spin_unlock(&nm_i->free_nid_list_lock);
1986         mutex_unlock(&nm_i->build_lock);
1987
1988         return nr - nr_shrink;
1989 }
1990
1991 void recover_inline_xattr(struct inode *inode, struct page *page)
1992 {
1993         void *src_addr, *dst_addr;
1994         size_t inline_size;
1995         struct page *ipage;
1996         struct f2fs_inode *ri;
1997
1998         ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1999         f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2000
2001         ri = F2FS_INODE(page);
2002         if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2003                 clear_inode_flag(inode, FI_INLINE_XATTR);
2004                 goto update_inode;
2005         }
2006
2007         dst_addr = inline_xattr_addr(ipage);
2008         src_addr = inline_xattr_addr(page);
2009         inline_size = inline_xattr_size(inode);
2010
2011         f2fs_wait_on_page_writeback(ipage, NODE, true);
2012         memcpy(dst_addr, src_addr, inline_size);
2013 update_inode:
2014         update_inode(inode, ipage);
2015         f2fs_put_page(ipage, 1);
2016 }
2017
2018 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2019 {
2020         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2021         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2022         nid_t new_xnid = nid_of_node(page);
2023         struct node_info ni;
2024
2025         /* 1: invalidate the previous xattr nid */
2026         if (!prev_xnid)
2027                 goto recover_xnid;
2028
2029         /* Deallocate node address */
2030         get_node_info(sbi, prev_xnid, &ni);
2031         f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2032         invalidate_blocks(sbi, ni.blk_addr);
2033         dec_valid_node_count(sbi, inode);
2034         set_node_addr(sbi, &ni, NULL_ADDR, false);
2035
2036 recover_xnid:
2037         /* 2: allocate new xattr nid */
2038         if (unlikely(!inc_valid_node_count(sbi, inode)))
2039                 f2fs_bug_on(sbi, 1);
2040
2041         remove_free_nid(NM_I(sbi), new_xnid);
2042         get_node_info(sbi, new_xnid, &ni);
2043         ni.ino = inode->i_ino;
2044         set_node_addr(sbi, &ni, NEW_ADDR, false);
2045         f2fs_i_xnid_write(inode, new_xnid);
2046
2047         /* 3: update xattr blkaddr */
2048         refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
2049         set_node_addr(sbi, &ni, blkaddr, false);
2050 }
2051
2052 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2053 {
2054         struct f2fs_inode *src, *dst;
2055         nid_t ino = ino_of_node(page);
2056         struct node_info old_ni, new_ni;
2057         struct page *ipage;
2058
2059         get_node_info(sbi, ino, &old_ni);
2060
2061         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2062                 return -EINVAL;
2063 retry:
2064         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2065         if (!ipage) {
2066                 congestion_wait(BLK_RW_ASYNC, HZ/50);
2067                 goto retry;
2068         }
2069
2070         /* Should not use this inode from free nid list */
2071         remove_free_nid(NM_I(sbi), ino);
2072
2073         if (!PageUptodate(ipage))
2074                 SetPageUptodate(ipage);
2075         fill_node_footer(ipage, ino, ino, 0, true);
2076
2077         src = F2FS_INODE(page);
2078         dst = F2FS_INODE(ipage);
2079
2080         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2081         dst->i_size = 0;
2082         dst->i_blocks = cpu_to_le64(1);
2083         dst->i_links = cpu_to_le32(1);
2084         dst->i_xattr_nid = 0;
2085         dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2086
2087         new_ni = old_ni;
2088         new_ni.ino = ino;
2089
2090         if (unlikely(!inc_valid_node_count(sbi, NULL)))
2091                 WARN_ON(1);
2092         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2093         inc_valid_inode_count(sbi);
2094         set_page_dirty(ipage);
2095         f2fs_put_page(ipage, 1);
2096         return 0;
2097 }
2098
2099 int restore_node_summary(struct f2fs_sb_info *sbi,
2100                         unsigned int segno, struct f2fs_summary_block *sum)
2101 {
2102         struct f2fs_node *rn;
2103         struct f2fs_summary *sum_entry;
2104         block_t addr;
2105         int bio_blocks = MAX_BIO_BLOCKS(sbi);
2106         int i, idx, last_offset, nrpages;
2107
2108         /* scan the node segment */
2109         last_offset = sbi->blocks_per_seg;
2110         addr = START_BLOCK(sbi, segno);
2111         sum_entry = &sum->entries[0];
2112
2113         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2114                 nrpages = min(last_offset - i, bio_blocks);
2115
2116                 /* readahead node pages */
2117                 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2118
2119                 for (idx = addr; idx < addr + nrpages; idx++) {
2120                         struct page *page = get_tmp_page(sbi, idx);
2121
2122                         rn = F2FS_NODE(page);
2123                         sum_entry->nid = rn->footer.nid;
2124                         sum_entry->version = 0;
2125                         sum_entry->ofs_in_node = 0;
2126                         sum_entry++;
2127                         f2fs_put_page(page, 1);
2128                 }
2129
2130                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2131                                                         addr + nrpages);
2132         }
2133         return 0;
2134 }
2135
2136 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2137 {
2138         struct f2fs_nm_info *nm_i = NM_I(sbi);
2139         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2140         struct f2fs_journal *journal = curseg->journal;
2141         int i;
2142
2143         down_write(&curseg->journal_rwsem);
2144         for (i = 0; i < nats_in_cursum(journal); i++) {
2145                 struct nat_entry *ne;
2146                 struct f2fs_nat_entry raw_ne;
2147                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2148
2149                 raw_ne = nat_in_journal(journal, i);
2150
2151                 ne = __lookup_nat_cache(nm_i, nid);
2152                 if (!ne) {
2153                         ne = grab_nat_entry(nm_i, nid);
2154                         node_info_from_raw_nat(&ne->ni, &raw_ne);
2155                 }
2156                 __set_nat_cache_dirty(nm_i, ne);
2157         }
2158         update_nats_in_cursum(journal, -i);
2159         up_write(&curseg->journal_rwsem);
2160 }
2161
2162 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2163                                                 struct list_head *head, int max)
2164 {
2165         struct nat_entry_set *cur;
2166
2167         if (nes->entry_cnt >= max)
2168                 goto add_out;
2169
2170         list_for_each_entry(cur, head, set_list) {
2171                 if (cur->entry_cnt >= nes->entry_cnt) {
2172                         list_add(&nes->set_list, cur->set_list.prev);
2173                         return;
2174                 }
2175         }
2176 add_out:
2177         list_add_tail(&nes->set_list, head);
2178 }
2179
2180 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2181                                         struct nat_entry_set *set)
2182 {
2183         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2184         struct f2fs_journal *journal = curseg->journal;
2185         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2186         bool to_journal = true;
2187         struct f2fs_nat_block *nat_blk;
2188         struct nat_entry *ne, *cur;
2189         struct page *page = NULL;
2190
2191         /*
2192          * there are two steps to flush nat entries:
2193          * #1, flush nat entries to journal in current hot data summary block.
2194          * #2, flush nat entries to nat page.
2195          */
2196         if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2197                 to_journal = false;
2198
2199         if (to_journal) {
2200                 down_write(&curseg->journal_rwsem);
2201         } else {
2202                 page = get_next_nat_page(sbi, start_nid);
2203                 nat_blk = page_address(page);
2204                 f2fs_bug_on(sbi, !nat_blk);
2205         }
2206
2207         /* flush dirty nats in nat entry set */
2208         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2209                 struct f2fs_nat_entry *raw_ne;
2210                 nid_t nid = nat_get_nid(ne);
2211                 int offset;
2212
2213                 if (nat_get_blkaddr(ne) == NEW_ADDR)
2214                         continue;
2215
2216                 if (to_journal) {
2217                         offset = lookup_journal_in_cursum(journal,
2218                                                         NAT_JOURNAL, nid, 1);
2219                         f2fs_bug_on(sbi, offset < 0);
2220                         raw_ne = &nat_in_journal(journal, offset);
2221                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2222                 } else {
2223                         raw_ne = &nat_blk->entries[nid - start_nid];
2224                 }
2225                 raw_nat_from_node_info(raw_ne, &ne->ni);
2226                 nat_reset_flag(ne);
2227                 __clear_nat_cache_dirty(NM_I(sbi), ne);
2228                 if (nat_get_blkaddr(ne) == NULL_ADDR)
2229                         add_free_nid(sbi, nid, false);
2230         }
2231
2232         if (to_journal)
2233                 up_write(&curseg->journal_rwsem);
2234         else
2235                 f2fs_put_page(page, 1);
2236
2237         f2fs_bug_on(sbi, set->entry_cnt);
2238
2239         radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2240         kmem_cache_free(nat_entry_set_slab, set);
2241 }
2242
2243 /*
2244  * This function is called during the checkpointing process.
2245  */
2246 void flush_nat_entries(struct f2fs_sb_info *sbi)
2247 {
2248         struct f2fs_nm_info *nm_i = NM_I(sbi);
2249         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2250         struct f2fs_journal *journal = curseg->journal;
2251         struct nat_entry_set *setvec[SETVEC_SIZE];
2252         struct nat_entry_set *set, *tmp;
2253         unsigned int found;
2254         nid_t set_idx = 0;
2255         LIST_HEAD(sets);
2256
2257         if (!nm_i->dirty_nat_cnt)
2258                 return;
2259
2260         down_write(&nm_i->nat_tree_lock);
2261
2262         /*
2263          * if there are no enough space in journal to store dirty nat
2264          * entries, remove all entries from journal and merge them
2265          * into nat entry set.
2266          */
2267         if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2268                 remove_nats_in_journal(sbi);
2269
2270         while ((found = __gang_lookup_nat_set(nm_i,
2271                                         set_idx, SETVEC_SIZE, setvec))) {
2272                 unsigned idx;
2273                 set_idx = setvec[found - 1]->set + 1;
2274                 for (idx = 0; idx < found; idx++)
2275                         __adjust_nat_entry_set(setvec[idx], &sets,
2276                                                 MAX_NAT_JENTRIES(journal));
2277         }
2278
2279         /* flush dirty nats in nat entry set */
2280         list_for_each_entry_safe(set, tmp, &sets, set_list)
2281                 __flush_nat_entry_set(sbi, set);
2282
2283         up_write(&nm_i->nat_tree_lock);
2284
2285         f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2286 }
2287
2288 static int init_node_manager(struct f2fs_sb_info *sbi)
2289 {
2290         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2291         struct f2fs_nm_info *nm_i = NM_I(sbi);
2292         unsigned char *version_bitmap;
2293         unsigned int nat_segs, nat_blocks;
2294
2295         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2296
2297         /* segment_count_nat includes pair segment so divide to 2. */
2298         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2299         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2300
2301         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2302
2303         /* not used nids: 0, node, meta, (and root counted as valid node) */
2304         nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2305         nm_i->fcnt = 0;
2306         nm_i->nat_cnt = 0;
2307         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2308         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2309         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2310
2311         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2312         INIT_LIST_HEAD(&nm_i->free_nid_list);
2313         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2314         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2315         INIT_LIST_HEAD(&nm_i->nat_entries);
2316
2317         mutex_init(&nm_i->build_lock);
2318         spin_lock_init(&nm_i->free_nid_list_lock);
2319         init_rwsem(&nm_i->nat_tree_lock);
2320
2321         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2322         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2323         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2324         if (!version_bitmap)
2325                 return -EFAULT;
2326
2327         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2328                                         GFP_KERNEL);
2329         if (!nm_i->nat_bitmap)
2330                 return -ENOMEM;
2331         return 0;
2332 }
2333
2334 int build_node_manager(struct f2fs_sb_info *sbi)
2335 {
2336         int err;
2337
2338         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2339         if (!sbi->nm_info)
2340                 return -ENOMEM;
2341
2342         err = init_node_manager(sbi);
2343         if (err)
2344                 return err;
2345
2346         build_free_nids(sbi);
2347         return 0;
2348 }
2349
2350 void destroy_node_manager(struct f2fs_sb_info *sbi)
2351 {
2352         struct f2fs_nm_info *nm_i = NM_I(sbi);
2353         struct free_nid *i, *next_i;
2354         struct nat_entry *natvec[NATVEC_SIZE];
2355         struct nat_entry_set *setvec[SETVEC_SIZE];
2356         nid_t nid = 0;
2357         unsigned int found;
2358
2359         if (!nm_i)
2360                 return;
2361
2362         /* destroy free nid list */
2363         spin_lock(&nm_i->free_nid_list_lock);
2364         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2365                 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2366                 __del_from_free_nid_list(nm_i, i);
2367                 nm_i->fcnt--;
2368                 spin_unlock(&nm_i->free_nid_list_lock);
2369                 kmem_cache_free(free_nid_slab, i);
2370                 spin_lock(&nm_i->free_nid_list_lock);
2371         }
2372         f2fs_bug_on(sbi, nm_i->fcnt);
2373         spin_unlock(&nm_i->free_nid_list_lock);
2374
2375         /* destroy nat cache */
2376         down_write(&nm_i->nat_tree_lock);
2377         while ((found = __gang_lookup_nat_cache(nm_i,
2378                                         nid, NATVEC_SIZE, natvec))) {
2379                 unsigned idx;
2380
2381                 nid = nat_get_nid(natvec[found - 1]) + 1;
2382                 for (idx = 0; idx < found; idx++)
2383                         __del_from_nat_cache(nm_i, natvec[idx]);
2384         }
2385         f2fs_bug_on(sbi, nm_i->nat_cnt);
2386
2387         /* destroy nat set cache */
2388         nid = 0;
2389         while ((found = __gang_lookup_nat_set(nm_i,
2390                                         nid, SETVEC_SIZE, setvec))) {
2391                 unsigned idx;
2392
2393                 nid = setvec[found - 1]->set + 1;
2394                 for (idx = 0; idx < found; idx++) {
2395                         /* entry_cnt is not zero, when cp_error was occurred */
2396                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2397                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2398                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2399                 }
2400         }
2401         up_write(&nm_i->nat_tree_lock);
2402
2403         kfree(nm_i->nat_bitmap);
2404         sbi->nm_info = NULL;
2405         kfree(nm_i);
2406 }
2407
2408 int __init create_node_manager_caches(void)
2409 {
2410         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2411                         sizeof(struct nat_entry));
2412         if (!nat_entry_slab)
2413                 goto fail;
2414
2415         free_nid_slab = f2fs_kmem_cache_create("free_nid",
2416                         sizeof(struct free_nid));
2417         if (!free_nid_slab)
2418                 goto destroy_nat_entry;
2419
2420         nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2421                         sizeof(struct nat_entry_set));
2422         if (!nat_entry_set_slab)
2423                 goto destroy_free_nid;
2424         return 0;
2425
2426 destroy_free_nid:
2427         kmem_cache_destroy(free_nid_slab);
2428 destroy_nat_entry:
2429         kmem_cache_destroy(nat_entry_slab);
2430 fail:
2431         return -ENOMEM;
2432 }
2433
2434 void destroy_node_manager_caches(void)
2435 {
2436         kmem_cache_destroy(nat_entry_set_slab);
2437         kmem_cache_destroy(free_nid_slab);
2438         kmem_cache_destroy(nat_entry_slab);
2439 }