fs/f2fs/node.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/node.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/mpage.h>
  11 #include <linux/backing-dev.h>
  12 #include <linux/blkdev.h>
  13 #include <linux/pagevec.h>
  14 #include <linux/swap.h>
  15
  16 #include "f2fs.h"
  17 #include "node.h"
  18 #include "segment.h"
  19 #include "xattr.h"
  20 #include <trace/events/f2fs.h>
  21
  22 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
  23
  24 static struct kmem_cache *nat_entry_slab;
  25 static struct kmem_cache *free_nid_slab;
  26 static struct kmem_cache *nat_entry_set_slab;
  27 static struct kmem_cache *fsync_node_entry_slab;
  28
  29 /*
  30  * Check whether the given nid is within node id range.
  31  */
  32 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
  33 {
  34         if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
  35                 set_sbi_flag(sbi, SBI_NEED_FSCK);
  36                 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
  37                           __func__, nid);
  38                 return -EFSCORRUPTED;
  39         }
  40         return 0;
  41 }
  42
  43 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
  44 {
  45         struct f2fs_nm_info *nm_i = NM_I(sbi);
  46         struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
  47         struct sysinfo val;
  48         unsigned long avail_ram;
  49         unsigned long mem_size = 0;
  50         bool res = false;
  51
  52         if (!nm_i)
  53                 return true;
  54
  55         si_meminfo(&val);
  56
  57         /* only uses low memory */
  58         avail_ram = val.totalram - val.totalhigh;
  59
  60         /*
  61          * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
  62          */
  63         if (type == FREE_NIDS) {
  64                 mem_size = (nm_i->nid_cnt[FREE_NID] *
  65                                 sizeof(struct free_nid)) >> PAGE_SHIFT;
  66                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  67         } else if (type == NAT_ENTRIES) {
  68                 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
  69                                 sizeof(struct nat_entry)) >> PAGE_SHIFT;
  70                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
  71                 if (excess_cached_nats(sbi))
  72                         res = false;
  73         } else if (type == DIRTY_DENTS) {
  74                 if (sbi->sb->s_bdi->wb.dirty_exceeded)
  75                         return false;
  76                 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
  77                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  78         } else if (type == INO_ENTRIES) {
  79                 int i;
  80
  81                 for (i = 0; i < MAX_INO_ENTRY; i++)
  82                         mem_size += sbi->im[i].ino_num *
  83                                                 sizeof(struct ino_entry);
  84                 mem_size >>= PAGE_SHIFT;
  85                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  86         } else if (type == EXTENT_CACHE) {
  87                 mem_size = (atomic_read(&sbi->total_ext_tree) *
  88                                 sizeof(struct extent_tree) +
  89                                 atomic_read(&sbi->total_ext_node) *
  90                                 sizeof(struct extent_node)) >> PAGE_SHIFT;
  91                 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
  92         } else if (type == INMEM_PAGES) {
  93                 /* it allows 20% / total_ram for inmemory pages */
  94                 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
  95                 res = mem_size < (val.totalram / 5);
  96         } else if (type == DISCARD_CACHE) {
  97                 mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
  98                                 sizeof(struct discard_cmd)) >> PAGE_SHIFT;
  99                 res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
 100         } else if (type == COMPRESS_PAGE) {
 101 #ifdef CONFIG_F2FS_FS_COMPRESSION
 102                 unsigned long free_ram = val.freeram;
 103
 104                 /*
 105                  * free memory is lower than watermark or cached page count
 106                  * exceed threshold, deny caching compress page.
 107                  */
 108                 res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
 109                         (COMPRESS_MAPPING(sbi)->nrpages <
 110                          free_ram * sbi->compress_percent / 100);
 111 #else
 112                 res = false;
 113 #endif
 114         } else {
 115                 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
 116                         return true;
 117         }
 118         return res;
 119 }
 120
 121 static void clear_node_page_dirty(struct page *page)
 122 {
 123         if (PageDirty(page)) {
 124                 f2fs_clear_page_cache_dirty_tag(page);
 125                 clear_page_dirty_for_io(page);
 126                 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
 127         }
 128         ClearPageUptodate(page);
 129 }
 130
 131 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 132 {
 133         return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
 134 }
 135
 136 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
 137 {
 138         struct page *src_page;
 139         struct page *dst_page;
 140         pgoff_t dst_off;
 141         void *src_addr;
 142         void *dst_addr;
 143         struct f2fs_nm_info *nm_i = NM_I(sbi);
 144
 145         dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
 146
 147         /* get current nat block page with lock */
 148         src_page = get_current_nat_page(sbi, nid);
 149         if (IS_ERR(src_page))
 150                 return src_page;
 151         dst_page = f2fs_grab_meta_page(sbi, dst_off);
 152         f2fs_bug_on(sbi, PageDirty(src_page));
 153
 154         src_addr = page_address(src_page);
 155         dst_addr = page_address(dst_page);
 156         memcpy(dst_addr, src_addr, PAGE_SIZE);
 157         set_page_dirty(dst_page);
 158         f2fs_put_page(src_page, 1);
 159
 160         set_to_next_nat(nm_i, nid);
 161
 162         return dst_page;
 163 }
 164
 165 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
 166 {
 167         struct nat_entry *new;
 168
 169         if (no_fail)
 170                 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
 171         else
 172                 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
 173         if (new) {
 174                 nat_set_nid(new, nid);
 175                 nat_reset_flag(new);
 176         }
 177         return new;
 178 }
 179
 180 static void __free_nat_entry(struct nat_entry *e)
 181 {
 182         kmem_cache_free(nat_entry_slab, e);
 183 }
 184
 185 /* must be locked by nat_tree_lock */
 186 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
 187         struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
 188 {
 189         if (no_fail)
 190                 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
 191         else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
 192                 return NULL;
 193
 194         if (raw_ne)
 195                 node_info_from_raw_nat(&ne->ni, raw_ne);
 196
 197         spin_lock(&nm_i->nat_list_lock);
 198         list_add_tail(&ne->list, &nm_i->nat_entries);
 199         spin_unlock(&nm_i->nat_list_lock);
 200
 201         nm_i->nat_cnt[TOTAL_NAT]++;
 202         nm_i->nat_cnt[RECLAIMABLE_NAT]++;
 203         return ne;
 204 }
 205
 206 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 207 {
 208         struct nat_entry *ne;
 209
 210         ne = radix_tree_lookup(&nm_i->nat_root, n);
 211
 212         /* for recent accessed nat entry, move it to tail of lru list */
 213         if (ne && !get_nat_flag(ne, IS_DIRTY)) {
 214                 spin_lock(&nm_i->nat_list_lock);
 215                 if (!list_empty(&ne->list))
 216                         list_move_tail(&ne->list, &nm_i->nat_entries);
 217                 spin_unlock(&nm_i->nat_list_lock);
 218         }
 219
 220         return ne;
 221 }
 222
 223 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 224                 nid_t start, unsigned int nr, struct nat_entry **ep)
 225 {
 226         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 227 }
 228
 229 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 230 {
 231         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 232         nm_i->nat_cnt[TOTAL_NAT]--;
 233         nm_i->nat_cnt[RECLAIMABLE_NAT]--;
 234         __free_nat_entry(e);
 235 }
 236
 237 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
 238                                                         struct nat_entry *ne)
 239 {
 240         nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
 241         struct nat_entry_set *head;
 242
 243         head = radix_tree_lookup(&nm_i->nat_set_root, set);
 244         if (!head) {
 245                 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
 246
 247                 INIT_LIST_HEAD(&head->entry_list);
 248                 INIT_LIST_HEAD(&head->set_list);
 249                 head->set = set;
 250                 head->entry_cnt = 0;
 251                 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
 252         }
 253         return head;
 254 }
 255
 256 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 257                                                 struct nat_entry *ne)
 258 {
 259         struct nat_entry_set *head;
 260         bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
 261
 262         if (!new_ne)
 263                 head = __grab_nat_entry_set(nm_i, ne);
 264
 265         /*
 266          * update entry_cnt in below condition:
 267          * 1. update NEW_ADDR to valid block address;
 268          * 2. update old block address to new one;
 269          */
 270         if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
 271                                 !get_nat_flag(ne, IS_DIRTY)))
 272                 head->entry_cnt++;
 273
 274         set_nat_flag(ne, IS_PREALLOC, new_ne);
 275
 276         if (get_nat_flag(ne, IS_DIRTY))
 277                 goto refresh_list;
 278
 279         nm_i->nat_cnt[DIRTY_NAT]++;
 280         nm_i->nat_cnt[RECLAIMABLE_NAT]--;
 281         set_nat_flag(ne, IS_DIRTY, true);
 282 refresh_list:
 283         spin_lock(&nm_i->nat_list_lock);
 284         if (new_ne)
 285                 list_del_init(&ne->list);
 286         else
 287                 list_move_tail(&ne->list, &head->entry_list);
 288         spin_unlock(&nm_i->nat_list_lock);
 289 }
 290
 291 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
 292                 struct nat_entry_set *set, struct nat_entry *ne)
 293 {
 294         spin_lock(&nm_i->nat_list_lock);
 295         list_move_tail(&ne->list, &nm_i->nat_entries);
 296         spin_unlock(&nm_i->nat_list_lock);
 297
 298         set_nat_flag(ne, IS_DIRTY, false);
 299         set->entry_cnt--;
 300         nm_i->nat_cnt[DIRTY_NAT]--;
 301         nm_i->nat_cnt[RECLAIMABLE_NAT]++;
 302 }
 303
 304 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
 305                 nid_t start, unsigned int nr, struct nat_entry_set **ep)
 306 {
 307         return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
 308                                                         start, nr);
 309 }
 310
 311 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
 312 {
 313         return NODE_MAPPING(sbi) == page->mapping &&
 314                         IS_DNODE(page) && is_cold_node(page);
 315 }
 316
 317 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
 318 {
 319         spin_lock_init(&sbi->fsync_node_lock);
 320         INIT_LIST_HEAD(&sbi->fsync_node_list);
 321         sbi->fsync_seg_id = 0;
 322         sbi->fsync_node_num = 0;
 323 }
 324
 325 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
 326                                                         struct page *page)
 327 {
 328         struct fsync_node_entry *fn;
 329         unsigned long flags;
 330         unsigned int seq_id;
 331
 332         fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
 333
 334         get_page(page);
 335         fn->page = page;
 336         INIT_LIST_HEAD(&fn->list);
 337
 338         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 339         list_add_tail(&fn->list, &sbi->fsync_node_list);
 340         fn->seq_id = sbi->fsync_seg_id++;
 341         seq_id = fn->seq_id;
 342         sbi->fsync_node_num++;
 343         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 344
 345         return seq_id;
 346 }
 347
 348 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
 349 {
 350         struct fsync_node_entry *fn;
 351         unsigned long flags;
 352
 353         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 354         list_for_each_entry(fn, &sbi->fsync_node_list, list) {
 355                 if (fn->page == page) {
 356                         list_del(&fn->list);
 357                         sbi->fsync_node_num--;
 358                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 359                         kmem_cache_free(fsync_node_entry_slab, fn);
 360                         put_page(page);
 361                         return;
 362                 }
 363         }
 364         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 365         f2fs_bug_on(sbi, 1);
 366 }
 367
 368 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
 369 {
 370         unsigned long flags;
 371
 372         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
 373         sbi->fsync_seg_id = 0;
 374         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
 375 }
 376
 377 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
 378 {
 379         struct f2fs_nm_info *nm_i = NM_I(sbi);
 380         struct nat_entry *e;
 381         bool need = false;
 382
 383         down_read(&nm_i->nat_tree_lock);
 384         e = __lookup_nat_cache(nm_i, nid);
 385         if (e) {
 386                 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
 387                                 !get_nat_flag(e, HAS_FSYNCED_INODE))
 388                         need = true;
 389         }
 390         up_read(&nm_i->nat_tree_lock);
 391         return need;
 392 }
 393
 394 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 395 {
 396         struct f2fs_nm_info *nm_i = NM_I(sbi);
 397         struct nat_entry *e;
 398         bool is_cp = true;
 399
 400         down_read(&nm_i->nat_tree_lock);
 401         e = __lookup_nat_cache(nm_i, nid);
 402         if (e && !get_nat_flag(e, IS_CHECKPOINTED))
 403                 is_cp = false;
 404         up_read(&nm_i->nat_tree_lock);
 405         return is_cp;
 406 }
 407
 408 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
 409 {
 410         struct f2fs_nm_info *nm_i = NM_I(sbi);
 411         struct nat_entry *e;
 412         bool need_update = true;
 413
 414         down_read(&nm_i->nat_tree_lock);
 415         e = __lookup_nat_cache(nm_i, ino);
 416         if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
 417                         (get_nat_flag(e, IS_CHECKPOINTED) ||
 418                          get_nat_flag(e, HAS_FSYNCED_INODE)))
 419                 need_update = false;
 420         up_read(&nm_i->nat_tree_lock);
 421         return need_update;
 422 }
 423
 424 /* must be locked by nat_tree_lock */
 425 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
 426                                                 struct f2fs_nat_entry *ne)
 427 {
 428         struct f2fs_nm_info *nm_i = NM_I(sbi);
 429         struct nat_entry *new, *e;
 430
 431         new = __alloc_nat_entry(nid, false);
 432         if (!new)
 433                 return;
 434
 435         down_write(&nm_i->nat_tree_lock);
 436         e = __lookup_nat_cache(nm_i, nid);
 437         if (!e)
 438                 e = __init_nat_entry(nm_i, new, ne, false);
 439         else
 440                 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
 441                                 nat_get_blkaddr(e) !=
 442                                         le32_to_cpu(ne->block_addr) ||
 443                                 nat_get_version(e) != ne->version);
 444         up_write(&nm_i->nat_tree_lock);
 445         if (e != new)
 446                 __free_nat_entry(new);
 447 }
 448
 449 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 450                         block_t new_blkaddr, bool fsync_done)
 451 {
 452         struct f2fs_nm_info *nm_i = NM_I(sbi);
 453         struct nat_entry *e;
 454         struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
 455
 456         down_write(&nm_i->nat_tree_lock);
 457         e = __lookup_nat_cache(nm_i, ni->nid);
 458         if (!e) {
 459                 e = __init_nat_entry(nm_i, new, NULL, true);
 460                 copy_node_info(&e->ni, ni);
 461                 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
 462         } else if (new_blkaddr == NEW_ADDR) {
 463                 /*
 464                  * when nid is reallocated,
 465                  * previous nat entry can be remained in nat cache.
 466                  * So, reinitialize it with new information.
 467                  */
 468                 copy_node_info(&e->ni, ni);
 469                 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
 470         }
 471         /* let's free early to reduce memory consumption */
 472         if (e != new)
 473                 __free_nat_entry(new);
 474
 475         /* sanity check */
 476         f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
 477         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
 478                         new_blkaddr == NULL_ADDR);
 479         f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
 480                         new_blkaddr == NEW_ADDR);
 481         f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
 482                         new_blkaddr == NEW_ADDR);
 483
 484         /* increment version no as node is removed */
 485         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 486                 unsigned char version = nat_get_version(e);
 487
 488                 nat_set_version(e, inc_node_version(version));
 489         }
 490
 491         /* change address */
 492         nat_set_blkaddr(e, new_blkaddr);
 493         if (!__is_valid_data_blkaddr(new_blkaddr))
 494                 set_nat_flag(e, IS_CHECKPOINTED, false);
 495         __set_nat_cache_dirty(nm_i, e);
 496
 497         /* update fsync_mark if its inode nat entry is still alive */
 498         if (ni->nid != ni->ino)
 499                 e = __lookup_nat_cache(nm_i, ni->ino);
 500         if (e) {
 501                 if (fsync_done && ni->nid == ni->ino)
 502                         set_nat_flag(e, HAS_FSYNCED_INODE, true);
 503                 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
 504         }
 505         up_write(&nm_i->nat_tree_lock);
 506 }
 507
 508 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 509 {
 510         struct f2fs_nm_info *nm_i = NM_I(sbi);
 511         int nr = nr_shrink;
 512
 513         if (!down_write_trylock(&nm_i->nat_tree_lock))
 514                 return 0;
 515
 516         spin_lock(&nm_i->nat_list_lock);
 517         while (nr_shrink) {
 518                 struct nat_entry *ne;
 519
 520                 if (list_empty(&nm_i->nat_entries))
 521                         break;
 522
 523                 ne = list_first_entry(&nm_i->nat_entries,
 524                                         struct nat_entry, list);
 525                 list_del(&ne->list);
 526                 spin_unlock(&nm_i->nat_list_lock);
 527
 528                 __del_from_nat_cache(nm_i, ne);
 529                 nr_shrink--;
 530
 531                 spin_lock(&nm_i->nat_list_lock);
 532         }
 533         spin_unlock(&nm_i->nat_list_lock);
 534
 535         up_write(&nm_i->nat_tree_lock);
 536         return nr - nr_shrink;
 537 }
 538
 539 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
 540                                                 struct node_info *ni)
 541 {
 542         struct f2fs_nm_info *nm_i = NM_I(sbi);
 543         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 544         struct f2fs_journal *journal = curseg->journal;
 545         nid_t start_nid = START_NID(nid);
 546         struct f2fs_nat_block *nat_blk;
 547         struct page *page = NULL;
 548         struct f2fs_nat_entry ne;
 549         struct nat_entry *e;
 550         pgoff_t index;
 551         block_t blkaddr;
 552         int i;
 553
 554         ni->nid = nid;
 555 retry:
 556         /* Check nat cache */
 557         down_read(&nm_i->nat_tree_lock);
 558         e = __lookup_nat_cache(nm_i, nid);
 559         if (e) {
 560                 ni->ino = nat_get_ino(e);
 561                 ni->blk_addr = nat_get_blkaddr(e);
 562                 ni->version = nat_get_version(e);
 563                 up_read(&nm_i->nat_tree_lock);
 564                 return 0;
 565         }
 566
 567         /*
 568          * Check current segment summary by trying to grab journal_rwsem first.
 569          * This sem is on the critical path on the checkpoint requiring the above
 570          * nat_tree_lock. Therefore, we should retry, if we failed to grab here
 571          * while not bothering checkpoint.
 572          */
 573         if (!rwsem_is_locked(&sbi->cp_global_sem)) {
 574                 down_read(&curseg->journal_rwsem);
 575         } else if (!down_read_trylock(&curseg->journal_rwsem)) {
 576                 up_read(&nm_i->nat_tree_lock);
 577                 goto retry;
 578         }
 579
 580         i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
 581         if (i >= 0) {
 582                 ne = nat_in_journal(journal, i);
 583                 node_info_from_raw_nat(ni, &ne);
 584         }
 585         up_read(&curseg->journal_rwsem);
 586         if (i >= 0) {
 587                 up_read(&nm_i->nat_tree_lock);
 588                 goto cache;
 589         }
 590
 591         /* Fill node_info from nat page */
 592         index = current_nat_addr(sbi, nid);
 593         up_read(&nm_i->nat_tree_lock);
 594
 595         page = f2fs_get_meta_page(sbi, index);
 596         if (IS_ERR(page))
 597                 return PTR_ERR(page);
 598
 599         nat_blk = (struct f2fs_nat_block *)page_address(page);
 600         ne = nat_blk->entries[nid - start_nid];
 601         node_info_from_raw_nat(ni, &ne);
 602         f2fs_put_page(page, 1);
 603 cache:
 604         blkaddr = le32_to_cpu(ne.block_addr);
 605         if (__is_valid_data_blkaddr(blkaddr) &&
 606                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
 607                 return -EFAULT;
 608
 609         /* cache nat entry */
 610         cache_nat_entry(sbi, nid, &ne);
 611         return 0;
 612 }
 613
 614 /*
 615  * readahead MAX_RA_NODE number of node pages.
 616  */
 617 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
 618 {
 619         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
 620         struct blk_plug plug;
 621         int i, end;
 622         nid_t nid;
 623
 624         blk_start_plug(&plug);
 625
 626         /* Then, try readahead for siblings of the desired node */
 627         end = start + n;
 628         end = min(end, NIDS_PER_BLOCK);
 629         for (i = start; i < end; i++) {
 630                 nid = get_nid(parent, i, false);
 631                 f2fs_ra_node_page(sbi, nid);
 632         }
 633
 634         blk_finish_plug(&plug);
 635 }
 636
 637 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
 638 {
 639         const long direct_index = ADDRS_PER_INODE(dn->inode);
 640         const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
 641         const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
 642         unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
 643         int cur_level = dn->cur_level;
 644         int max_level = dn->max_level;
 645         pgoff_t base = 0;
 646
 647         if (!dn->max_level)
 648                 return pgofs + 1;
 649
 650         while (max_level-- > cur_level)
 651                 skipped_unit *= NIDS_PER_BLOCK;
 652
 653         switch (dn->max_level) {
 654         case 3:
 655                 base += 2 * indirect_blks;
 656                 fallthrough;
 657         case 2:
 658                 base += 2 * direct_blks;
 659                 fallthrough;
 660         case 1:
 661                 base += direct_index;
 662                 break;
 663         default:
 664                 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
 665         }
 666
 667         return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
 668 }
 669
 670 /*
 671  * The maximum depth is four.
 672  * Offset[0] will have raw inode offset.
 673  */
 674 static int get_node_path(struct inode *inode, long block,
 675                                 int offset[4], unsigned int noffset[4])
 676 {
 677         const long direct_index = ADDRS_PER_INODE(inode);
 678         const long direct_blks = ADDRS_PER_BLOCK(inode);
 679         const long dptrs_per_blk = NIDS_PER_BLOCK;
 680         const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
 681         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 682         int n = 0;
 683         int level = 0;
 684
 685         noffset[0] = 0;
 686
 687         if (block < direct_index) {
 688                 offset[n] = block;
 689                 goto got;
 690         }
 691         block -= direct_index;
 692         if (block < direct_blks) {
 693                 offset[n++] = NODE_DIR1_BLOCK;
 694                 noffset[n] = 1;
 695                 offset[n] = block;
 696                 level = 1;
 697                 goto got;
 698         }
 699         block -= direct_blks;
 700         if (block < direct_blks) {
 701                 offset[n++] = NODE_DIR2_BLOCK;
 702                 noffset[n] = 2;
 703                 offset[n] = block;
 704                 level = 1;
 705                 goto got;
 706         }
 707         block -= direct_blks;
 708         if (block < indirect_blks) {
 709                 offset[n++] = NODE_IND1_BLOCK;
 710                 noffset[n] = 3;
 711                 offset[n++] = block / direct_blks;
 712                 noffset[n] = 4 + offset[n - 1];
 713                 offset[n] = block % direct_blks;
 714                 level = 2;
 715                 goto got;
 716         }
 717         block -= indirect_blks;
 718         if (block < indirect_blks) {
 719                 offset[n++] = NODE_IND2_BLOCK;
 720                 noffset[n] = 4 + dptrs_per_blk;
 721                 offset[n++] = block / direct_blks;
 722                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 723                 offset[n] = block % direct_blks;
 724                 level = 2;
 725                 goto got;
 726         }
 727         block -= indirect_blks;
 728         if (block < dindirect_blks) {
 729                 offset[n++] = NODE_DIND_BLOCK;
 730                 noffset[n] = 5 + (dptrs_per_blk * 2);
 731                 offset[n++] = block / indirect_blks;
 732                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 733                               offset[n - 1] * (dptrs_per_blk + 1);
 734                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 735                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 736                               offset[n - 2] * (dptrs_per_blk + 1) +
 737                               offset[n - 1];
 738                 offset[n] = block % direct_blks;
 739                 level = 3;
 740                 goto got;
 741         } else {
 742                 return -E2BIG;
 743         }
 744 got:
 745         return level;
 746 }
 747
 748 /*
 749  * Caller should call f2fs_put_dnode(dn).
 750  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 751  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
 752  */
 753 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 754 {
 755         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 756         struct page *npage[4];
 757         struct page *parent = NULL;
 758         int offset[4];
 759         unsigned int noffset[4];
 760         nid_t nids[4];
 761         int level, i = 0;
 762         int err = 0;
 763
 764         level = get_node_path(dn->inode, index, offset, noffset);
 765         if (level < 0)
 766                 return level;
 767
 768         nids[0] = dn->inode->i_ino;
 769         npage[0] = dn->inode_page;
 770
 771         if (!npage[0]) {
 772                 npage[0] = f2fs_get_node_page(sbi, nids[0]);
 773                 if (IS_ERR(npage[0]))
 774                         return PTR_ERR(npage[0]);
 775         }
 776
 777         /* if inline_data is set, should not report any block indices */
 778         if (f2fs_has_inline_data(dn->inode) && index) {
 779                 err = -ENOENT;
 780                 f2fs_put_page(npage[0], 1);
 781                 goto release_out;
 782         }
 783
 784         parent = npage[0];
 785         if (level != 0)
 786                 nids[1] = get_nid(parent, offset[0], true);
 787         dn->inode_page = npage[0];
 788         dn->inode_page_locked = true;
 789
 790         /* get indirect or direct nodes */
 791         for (i = 1; i <= level; i++) {
 792                 bool done = false;
 793
 794                 if (!nids[i] && mode == ALLOC_NODE) {
 795                         /* alloc new node */
 796                         if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
 797                                 err = -ENOSPC;
 798                                 goto release_pages;
 799                         }
 800
 801                         dn->nid = nids[i];
 802                         npage[i] = f2fs_new_node_page(dn, noffset[i]);
 803                         if (IS_ERR(npage[i])) {
 804                                 f2fs_alloc_nid_failed(sbi, nids[i]);
 805                                 err = PTR_ERR(npage[i]);
 806                                 goto release_pages;
 807                         }
 808
 809                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 810                         f2fs_alloc_nid_done(sbi, nids[i]);
 811                         done = true;
 812                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 813                         npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
 814                         if (IS_ERR(npage[i])) {
 815                                 err = PTR_ERR(npage[i]);
 816                                 goto release_pages;
 817                         }
 818                         done = true;
 819                 }
 820                 if (i == 1) {
 821                         dn->inode_page_locked = false;
 822                         unlock_page(parent);
 823                 } else {
 824                         f2fs_put_page(parent, 1);
 825                 }
 826
 827                 if (!done) {
 828                         npage[i] = f2fs_get_node_page(sbi, nids[i]);
 829                         if (IS_ERR(npage[i])) {
 830                                 err = PTR_ERR(npage[i]);
 831                                 f2fs_put_page(npage[0], 0);
 832                                 goto release_out;
 833                         }
 834                 }
 835                 if (i < level) {
 836                         parent = npage[i];
 837                         nids[i + 1] = get_nid(parent, offset[i], false);
 838                 }
 839         }
 840         dn->nid = nids[level];
 841         dn->ofs_in_node = offset[level];
 842         dn->node_page = npage[level];
 843         dn->data_blkaddr = f2fs_data_blkaddr(dn);
 844         return 0;
 845
 846 release_pages:
 847         f2fs_put_page(parent, 1);
 848         if (i > 1)
 849                 f2fs_put_page(npage[0], 0);
 850 release_out:
 851         dn->inode_page = NULL;
 852         dn->node_page = NULL;
 853         if (err == -ENOENT) {
 854                 dn->cur_level = i;
 855                 dn->max_level = level;
 856                 dn->ofs_in_node = offset[level];
 857         }
 858         return err;
 859 }
 860
 861 static int truncate_node(struct dnode_of_data *dn)
 862 {
 863         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 864         struct node_info ni;
 865         int err;
 866         pgoff_t index;
 867
 868         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 869         if (err)
 870                 return err;
 871
 872         /* Deallocate node address */
 873         f2fs_invalidate_blocks(sbi, ni.blk_addr);
 874         dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
 875         set_node_addr(sbi, &ni, NULL_ADDR, false);
 876
 877         if (dn->nid == dn->inode->i_ino) {
 878                 f2fs_remove_orphan_inode(sbi, dn->nid);
 879                 dec_valid_inode_count(sbi);
 880                 f2fs_inode_synced(dn->inode);
 881         }
 882
 883         clear_node_page_dirty(dn->node_page);
 884         set_sbi_flag(sbi, SBI_IS_DIRTY);
 885
 886         index = dn->node_page->index;
 887         f2fs_put_page(dn->node_page, 1);
 888
 889         invalidate_mapping_pages(NODE_MAPPING(sbi),
 890                         index, index);
 891
 892         dn->node_page = NULL;
 893         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 894
 895         return 0;
 896 }
 897
 898 static int truncate_dnode(struct dnode_of_data *dn)
 899 {
 900         struct page *page;
 901         int err;
 902
 903         if (dn->nid == 0)
 904                 return 1;
 905
 906         /* get direct node */
 907         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 908         if (PTR_ERR(page) == -ENOENT)
 909                 return 1;
 910         else if (IS_ERR(page))
 911                 return PTR_ERR(page);
 912
 913         /* Make dnode_of_data for parameter */
 914         dn->node_page = page;
 915         dn->ofs_in_node = 0;
 916         f2fs_truncate_data_blocks(dn);
 917         err = truncate_node(dn);
 918         if (err)
 919                 return err;
 920
 921         return 1;
 922 }
 923
 924 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 925                                                 int ofs, int depth)
 926 {
 927         struct dnode_of_data rdn = *dn;
 928         struct page *page;
 929         struct f2fs_node *rn;
 930         nid_t child_nid;
 931         unsigned int child_nofs;
 932         int freed = 0;
 933         int i, ret;
 934
 935         if (dn->nid == 0)
 936                 return NIDS_PER_BLOCK + 1;
 937
 938         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 939
 940         page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
 941         if (IS_ERR(page)) {
 942                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 943                 return PTR_ERR(page);
 944         }
 945
 946         f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
 947
 948         rn = F2FS_NODE(page);
 949         if (depth < 3) {
 950                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 951                         child_nid = le32_to_cpu(rn->in.nid[i]);
 952                         if (child_nid == 0)
 953                                 continue;
 954                         rdn.nid = child_nid;
 955                         ret = truncate_dnode(&rdn);
 956                         if (ret < 0)
 957                                 goto out_err;
 958                         if (set_nid(page, i, 0, false))
 959                                 dn->node_changed = true;
 960                 }
 961         } else {
 962                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 963                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 964                         child_nid = le32_to_cpu(rn->in.nid[i]);
 965                         if (child_nid == 0) {
 966                                 child_nofs += NIDS_PER_BLOCK + 1;
 967                                 continue;
 968                         }
 969                         rdn.nid = child_nid;
 970                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 971                         if (ret == (NIDS_PER_BLOCK + 1)) {
 972                                 if (set_nid(page, i, 0, false))
 973                                         dn->node_changed = true;
 974                                 child_nofs += ret;
 975                         } else if (ret < 0 && ret != -ENOENT) {
 976                                 goto out_err;
 977                         }
 978                 }
 979                 freed = child_nofs;
 980         }
 981
 982         if (!ofs) {
 983                 /* remove current indirect node */
 984                 dn->node_page = page;
 985                 ret = truncate_node(dn);
 986                 if (ret)
 987                         goto out_err;
 988                 freed++;
 989         } else {
 990                 f2fs_put_page(page, 1);
 991         }
 992         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 993         return freed;
 994
 995 out_err:
 996         f2fs_put_page(page, 1);
 997         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 998         return ret;
 999 }
1000
1001 static int truncate_partial_nodes(struct dnode_of_data *dn,
1002                         struct f2fs_inode *ri, int *offset, int depth)
1003 {
1004         struct page *pages[2];
1005         nid_t nid[3];
1006         nid_t child_nid;
1007         int err = 0;
1008         int i;
1009         int idx = depth - 2;
1010
1011         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1012         if (!nid[0])
1013                 return 0;
1014
1015         /* get indirect nodes in the path */
1016         for (i = 0; i < idx + 1; i++) {
1017                 /* reference count'll be increased */
1018                 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1019                 if (IS_ERR(pages[i])) {
1020                         err = PTR_ERR(pages[i]);
1021                         idx = i - 1;
1022                         goto fail;
1023                 }
1024                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1025         }
1026
1027         f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1028
1029         /* free direct nodes linked to a partial indirect node */
1030         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1031                 child_nid = get_nid(pages[idx], i, false);
1032                 if (!child_nid)
1033                         continue;
1034                 dn->nid = child_nid;
1035                 err = truncate_dnode(dn);
1036                 if (err < 0)
1037                         goto fail;
1038                 if (set_nid(pages[idx], i, 0, false))
1039                         dn->node_changed = true;
1040         }
1041
1042         if (offset[idx + 1] == 0) {
1043                 dn->node_page = pages[idx];
1044                 dn->nid = nid[idx];
1045                 err = truncate_node(dn);
1046                 if (err)
1047                         goto fail;
1048         } else {
1049                 f2fs_put_page(pages[idx], 1);
1050         }
1051         offset[idx]++;
1052         offset[idx + 1] = 0;
1053         idx--;
1054 fail:
1055         for (i = idx; i >= 0; i--)
1056                 f2fs_put_page(pages[i], 1);
1057
1058         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1059
1060         return err;
1061 }
1062
1063 /*
1064  * All the block addresses of data and nodes should be nullified.
1065  */
1066 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1067 {
1068         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1069         int err = 0, cont = 1;
1070         int level, offset[4], noffset[4];
1071         unsigned int nofs = 0;
1072         struct f2fs_inode *ri;
1073         struct dnode_of_data dn;
1074         struct page *page;
1075
1076         trace_f2fs_truncate_inode_blocks_enter(inode, from);
1077
1078         level = get_node_path(inode, from, offset, noffset);
1079         if (level < 0) {
1080                 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1081                 return level;
1082         }
1083
1084         page = f2fs_get_node_page(sbi, inode->i_ino);
1085         if (IS_ERR(page)) {
1086                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1087                 return PTR_ERR(page);
1088         }
1089
1090         set_new_dnode(&dn, inode, page, NULL, 0);
1091         unlock_page(page);
1092
1093         ri = F2FS_INODE(page);
1094         switch (level) {
1095         case 0:
1096         case 1:
1097                 nofs = noffset[1];
1098                 break;
1099         case 2:
1100                 nofs = noffset[1];
1101                 if (!offset[level - 1])
1102                         goto skip_partial;
1103                 err = truncate_partial_nodes(&dn, ri, offset, level);
1104                 if (err < 0 && err != -ENOENT)
1105                         goto fail;
1106                 nofs += 1 + NIDS_PER_BLOCK;
1107                 break;
1108         case 3:
1109                 nofs = 5 + 2 * NIDS_PER_BLOCK;
1110                 if (!offset[level - 1])
1111                         goto skip_partial;
1112                 err = truncate_partial_nodes(&dn, ri, offset, level);
1113                 if (err < 0 && err != -ENOENT)
1114                         goto fail;
1115                 break;
1116         default:
1117                 BUG();
1118         }
1119
1120 skip_partial:
1121         while (cont) {
1122                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1123                 switch (offset[0]) {
1124                 case NODE_DIR1_BLOCK:
1125                 case NODE_DIR2_BLOCK:
1126                         err = truncate_dnode(&dn);
1127                         break;
1128
1129                 case NODE_IND1_BLOCK:
1130                 case NODE_IND2_BLOCK:
1131                         err = truncate_nodes(&dn, nofs, offset[1], 2);
1132                         break;
1133
1134                 case NODE_DIND_BLOCK:
1135                         err = truncate_nodes(&dn, nofs, offset[1], 3);
1136                         cont = 0;
1137                         break;
1138
1139                 default:
1140                         BUG();
1141                 }
1142                 if (err < 0 && err != -ENOENT)
1143                         goto fail;
1144                 if (offset[1] == 0 &&
1145                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1146                         lock_page(page);
1147                         BUG_ON(page->mapping != NODE_MAPPING(sbi));
1148                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1149                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1150                         set_page_dirty(page);
1151                         unlock_page(page);
1152                 }
1153                 offset[1] = 0;
1154                 offset[0]++;
1155                 nofs += err;
1156         }
1157 fail:
1158         f2fs_put_page(page, 0);
1159         trace_f2fs_truncate_inode_blocks_exit(inode, err);
1160         return err > 0 ? 0 : err;
1161 }
1162
1163 /* caller must lock inode page */
1164 int f2fs_truncate_xattr_node(struct inode *inode)
1165 {
1166         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1167         nid_t nid = F2FS_I(inode)->i_xattr_nid;
1168         struct dnode_of_data dn;
1169         struct page *npage;
1170         int err;
1171
1172         if (!nid)
1173                 return 0;
1174
1175         npage = f2fs_get_node_page(sbi, nid);
1176         if (IS_ERR(npage))
1177                 return PTR_ERR(npage);
1178
1179         set_new_dnode(&dn, inode, NULL, npage, nid);
1180         err = truncate_node(&dn);
1181         if (err) {
1182                 f2fs_put_page(npage, 1);
1183                 return err;
1184         }
1185
1186         f2fs_i_xnid_write(inode, 0);
1187
1188         return 0;
1189 }
1190
1191 /*
1192  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1193  * f2fs_unlock_op().
1194  */
1195 int f2fs_remove_inode_page(struct inode *inode)
1196 {
1197         struct dnode_of_data dn;
1198         int err;
1199
1200         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1201         err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1202         if (err)
1203                 return err;
1204
1205         err = f2fs_truncate_xattr_node(inode);
1206         if (err) {
1207                 f2fs_put_dnode(&dn);
1208                 return err;
1209         }
1210
1211         /* remove potential inline_data blocks */
1212         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1213                                 S_ISLNK(inode->i_mode))
1214                 f2fs_truncate_data_blocks_range(&dn, 1);
1215
1216         /* 0 is possible, after f2fs_new_inode() has failed */
1217         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1218                 f2fs_put_dnode(&dn);
1219                 return -EIO;
1220         }
1221
1222         if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1223                 f2fs_warn(F2FS_I_SB(inode),
1224                         "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1225                         inode->i_ino, (unsigned long long)inode->i_blocks);
1226                 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1227         }
1228
1229         /* will put inode & node pages */
1230         err = truncate_node(&dn);
1231         if (err) {
1232                 f2fs_put_dnode(&dn);
1233                 return err;
1234         }
1235         return 0;
1236 }
1237
1238 struct page *f2fs_new_inode_page(struct inode *inode)
1239 {
1240         struct dnode_of_data dn;
1241
1242         /* allocate inode page for new inode */
1243         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1244
1245         /* caller should f2fs_put_page(page, 1); */
1246         return f2fs_new_node_page(&dn, 0);
1247 }
1248
1249 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1250 {
1251         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1252         struct node_info new_ni;
1253         struct page *page;
1254         int err;
1255
1256         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1257                 return ERR_PTR(-EPERM);
1258
1259         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1260         if (!page)
1261                 return ERR_PTR(-ENOMEM);
1262
1263         if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1264                 goto fail;
1265
1266 #ifdef CONFIG_F2FS_CHECK_FS
1267         err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1268         if (err) {
1269                 dec_valid_node_count(sbi, dn->inode, !ofs);
1270                 goto fail;
1271         }
1272         f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1273 #endif
1274         new_ni.nid = dn->nid;
1275         new_ni.ino = dn->inode->i_ino;
1276         new_ni.blk_addr = NULL_ADDR;
1277         new_ni.flag = 0;
1278         new_ni.version = 0;
1279         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1280
1281         f2fs_wait_on_page_writeback(page, NODE, true, true);
1282         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1283         set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1284         if (!PageUptodate(page))
1285                 SetPageUptodate(page);
1286         if (set_page_dirty(page))
1287                 dn->node_changed = true;
1288
1289         if (f2fs_has_xattr_block(ofs))
1290                 f2fs_i_xnid_write(dn->inode, dn->nid);
1291
1292         if (ofs == 0)
1293                 inc_valid_inode_count(sbi);
1294         return page;
1295
1296 fail:
1297         clear_node_page_dirty(page);
1298         f2fs_put_page(page, 1);
1299         return ERR_PTR(err);
1300 }
1301
1302 /*
1303  * Caller should do after getting the following values.
1304  * 0: f2fs_put_page(page, 0)
1305  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1306  */
1307 static int read_node_page(struct page *page, int op_flags)
1308 {
1309         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1310         struct node_info ni;
1311         struct f2fs_io_info fio = {
1312                 .sbi = sbi,
1313                 .type = NODE,
1314                 .op = REQ_OP_READ,
1315                 .op_flags = op_flags,
1316                 .page = page,
1317                 .encrypted_page = NULL,
1318         };
1319         int err;
1320
1321         if (PageUptodate(page)) {
1322                 if (!f2fs_inode_chksum_verify(sbi, page)) {
1323                         ClearPageUptodate(page);
1324                         return -EFSBADCRC;
1325                 }
1326                 return LOCKED_PAGE;
1327         }
1328
1329         err = f2fs_get_node_info(sbi, page->index, &ni);
1330         if (err)
1331                 return err;
1332
1333         /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1334         if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR) ||
1335                         is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1336                 ClearPageUptodate(page);
1337                 return -ENOENT;
1338         }
1339
1340         fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1341
1342         err = f2fs_submit_page_bio(&fio);
1343
1344         if (!err)
1345                 f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1346
1347         return err;
1348 }
1349
1350 /*
1351  * Readahead a node page
1352  */
1353 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1354 {
1355         struct page *apage;
1356         int err;
1357
1358         if (!nid)
1359                 return;
1360         if (f2fs_check_nid_range(sbi, nid))
1361                 return;
1362
1363         apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1364         if (apage)
1365                 return;
1366
1367         apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1368         if (!apage)
1369                 return;
1370
1371         err = read_node_page(apage, REQ_RAHEAD);
1372         f2fs_put_page(apage, err ? 1 : 0);
1373 }
1374
1375 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1376                                         struct page *parent, int start)
1377 {
1378         struct page *page;
1379         int err;
1380
1381         if (!nid)
1382                 return ERR_PTR(-ENOENT);
1383         if (f2fs_check_nid_range(sbi, nid))
1384                 return ERR_PTR(-EINVAL);
1385 repeat:
1386         page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1387         if (!page)
1388                 return ERR_PTR(-ENOMEM);
1389
1390         err = read_node_page(page, 0);
1391         if (err < 0) {
1392                 f2fs_put_page(page, 1);
1393                 return ERR_PTR(err);
1394         } else if (err == LOCKED_PAGE) {
1395                 err = 0;
1396                 goto page_hit;
1397         }
1398
1399         if (parent)
1400                 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1401
1402         lock_page(page);
1403
1404         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1405                 f2fs_put_page(page, 1);
1406                 goto repeat;
1407         }
1408
1409         if (unlikely(!PageUptodate(page))) {
1410                 err = -EIO;
1411                 goto out_err;
1412         }
1413
1414         if (!f2fs_inode_chksum_verify(sbi, page)) {
1415                 err = -EFSBADCRC;
1416                 goto out_err;
1417         }
1418 page_hit:
1419         if (unlikely(nid != nid_of_node(page))) {
1420                 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1421                           nid, nid_of_node(page), ino_of_node(page),
1422                           ofs_of_node(page), cpver_of_node(page),
1423                           next_blkaddr_of_node(page));
1424                 err = -EINVAL;
1425 out_err:
1426                 ClearPageUptodate(page);
1427                 f2fs_put_page(page, 1);
1428                 return ERR_PTR(err);
1429         }
1430         return page;
1431 }
1432
1433 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1434 {
1435         return __get_node_page(sbi, nid, NULL, 0);
1436 }
1437
1438 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1439 {
1440         struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1441         nid_t nid = get_nid(parent, start, false);
1442
1443         return __get_node_page(sbi, nid, parent, start);
1444 }
1445
1446 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1447 {
1448         struct inode *inode;
1449         struct page *page;
1450         int ret;
1451
1452         /* should flush inline_data before evict_inode */
1453         inode = ilookup(sbi->sb, ino);
1454         if (!inode)
1455                 return;
1456
1457         page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1458                                         FGP_LOCK|FGP_NOWAIT, 0);
1459         if (!page)
1460                 goto iput_out;
1461
1462         if (!PageUptodate(page))
1463                 goto page_out;
1464
1465         if (!PageDirty(page))
1466                 goto page_out;
1467
1468         if (!clear_page_dirty_for_io(page))
1469                 goto page_out;
1470
1471         ret = f2fs_write_inline_data(inode, page);
1472         inode_dec_dirty_pages(inode);
1473         f2fs_remove_dirty_inode(inode);
1474         if (ret)
1475                 set_page_dirty(page);
1476 page_out:
1477         f2fs_put_page(page, 1);
1478 iput_out:
1479         iput(inode);
1480 }
1481
1482 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1483 {
1484         pgoff_t index;
1485         struct pagevec pvec;
1486         struct page *last_page = NULL;
1487         int nr_pages;
1488
1489         pagevec_init(&pvec);
1490         index = 0;
1491
1492         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1493                                 PAGECACHE_TAG_DIRTY))) {
1494                 int i;
1495
1496                 for (i = 0; i < nr_pages; i++) {
1497                         struct page *page = pvec.pages[i];
1498
1499                         if (unlikely(f2fs_cp_error(sbi))) {
1500                                 f2fs_put_page(last_page, 0);
1501                                 pagevec_release(&pvec);
1502                                 return ERR_PTR(-EIO);
1503                         }
1504
1505                         if (!IS_DNODE(page) || !is_cold_node(page))
1506                                 continue;
1507                         if (ino_of_node(page) != ino)
1508                                 continue;
1509
1510                         lock_page(page);
1511
1512                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1513 continue_unlock:
1514                                 unlock_page(page);
1515                                 continue;
1516                         }
1517                         if (ino_of_node(page) != ino)
1518                                 goto continue_unlock;
1519
1520                         if (!PageDirty(page)) {
1521                                 /* someone wrote it for us */
1522                                 goto continue_unlock;
1523                         }
1524
1525                         if (last_page)
1526                                 f2fs_put_page(last_page, 0);
1527
1528                         get_page(page);
1529                         last_page = page;
1530                         unlock_page(page);
1531                 }
1532                 pagevec_release(&pvec);
1533                 cond_resched();
1534         }
1535         return last_page;
1536 }
1537
1538 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1539                                 struct writeback_control *wbc, bool do_balance,
1540                                 enum iostat_type io_type, unsigned int *seq_id)
1541 {
1542         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1543         nid_t nid;
1544         struct node_info ni;
1545         struct f2fs_io_info fio = {
1546                 .sbi = sbi,
1547                 .ino = ino_of_node(page),
1548                 .type = NODE,
1549                 .op = REQ_OP_WRITE,
1550                 .op_flags = wbc_to_write_flags(wbc),
1551                 .page = page,
1552                 .encrypted_page = NULL,
1553                 .submitted = false,
1554                 .io_type = io_type,
1555                 .io_wbc = wbc,
1556         };
1557         unsigned int seq;
1558
1559         trace_f2fs_writepage(page, NODE);
1560
1561         if (unlikely(f2fs_cp_error(sbi))) {
1562                 ClearPageUptodate(page);
1563                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1564                 unlock_page(page);
1565                 return 0;
1566         }
1567
1568         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1569                 goto redirty_out;
1570
1571         if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1572                         wbc->sync_mode == WB_SYNC_NONE &&
1573                         IS_DNODE(page) && is_cold_node(page))
1574                 goto redirty_out;
1575
1576         /* get old block addr of this node page */
1577         nid = nid_of_node(page);
1578         f2fs_bug_on(sbi, page->index != nid);
1579
1580         if (f2fs_get_node_info(sbi, nid, &ni))
1581                 goto redirty_out;
1582
1583         if (wbc->for_reclaim) {
1584                 if (!down_read_trylock(&sbi->node_write))
1585                         goto redirty_out;
1586         } else {
1587                 down_read(&sbi->node_write);
1588         }
1589
1590         /* This page is already truncated */
1591         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1592                 ClearPageUptodate(page);
1593                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1594                 up_read(&sbi->node_write);
1595                 unlock_page(page);
1596                 return 0;
1597         }
1598
1599         if (__is_valid_data_blkaddr(ni.blk_addr) &&
1600                 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1601                                         DATA_GENERIC_ENHANCE)) {
1602                 up_read(&sbi->node_write);
1603                 goto redirty_out;
1604         }
1605
1606         if (atomic && !test_opt(sbi, NOBARRIER))
1607                 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1608
1609         /* should add to global list before clearing PAGECACHE status */
1610         if (f2fs_in_warm_node_list(sbi, page)) {
1611                 seq = f2fs_add_fsync_node_entry(sbi, page);
1612                 if (seq_id)
1613                         *seq_id = seq;
1614         }
1615
1616         set_page_writeback(page);
1617         ClearPageError(page);
1618
1619         fio.old_blkaddr = ni.blk_addr;
1620         f2fs_do_write_node_page(nid, &fio);
1621         set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1622         dec_page_count(sbi, F2FS_DIRTY_NODES);
1623         up_read(&sbi->node_write);
1624
1625         if (wbc->for_reclaim) {
1626                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1627                 submitted = NULL;
1628         }
1629
1630         unlock_page(page);
1631
1632         if (unlikely(f2fs_cp_error(sbi))) {
1633                 f2fs_submit_merged_write(sbi, NODE);
1634                 submitted = NULL;
1635         }
1636         if (submitted)
1637                 *submitted = fio.submitted;
1638
1639         if (do_balance)
1640                 f2fs_balance_fs(sbi, false);
1641         return 0;
1642
1643 redirty_out:
1644         redirty_page_for_writepage(wbc, page);
1645         return AOP_WRITEPAGE_ACTIVATE;
1646 }
1647
1648 int f2fs_move_node_page(struct page *node_page, int gc_type)
1649 {
1650         int err = 0;
1651
1652         if (gc_type == FG_GC) {
1653                 struct writeback_control wbc = {
1654                         .sync_mode = WB_SYNC_ALL,
1655                         .nr_to_write = 1,
1656                         .for_reclaim = 0,
1657                 };
1658
1659                 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1660
1661                 set_page_dirty(node_page);
1662
1663                 if (!clear_page_dirty_for_io(node_page)) {
1664                         err = -EAGAIN;
1665                         goto out_page;
1666                 }
1667
1668                 if (__write_node_page(node_page, false, NULL,
1669                                         &wbc, false, FS_GC_NODE_IO, NULL)) {
1670                         err = -EAGAIN;
1671                         unlock_page(node_page);
1672                 }
1673                 goto release_page;
1674         } else {
1675                 /* set page dirty and write it */
1676                 if (!PageWriteback(node_page))
1677                         set_page_dirty(node_page);
1678         }
1679 out_page:
1680         unlock_page(node_page);
1681 release_page:
1682         f2fs_put_page(node_page, 0);
1683         return err;
1684 }
1685
1686 static int f2fs_write_node_page(struct page *page,
1687                                 struct writeback_control *wbc)
1688 {
1689         return __write_node_page(page, false, NULL, wbc, false,
1690                                                 FS_NODE_IO, NULL);
1691 }
1692
1693 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1694                         struct writeback_control *wbc, bool atomic,
1695                         unsigned int *seq_id)
1696 {
1697         pgoff_t index;
1698         struct pagevec pvec;
1699         int ret = 0;
1700         struct page *last_page = NULL;
1701         bool marked = false;
1702         nid_t ino = inode->i_ino;
1703         int nr_pages;
1704         int nwritten = 0;
1705
1706         if (atomic) {
1707                 last_page = last_fsync_dnode(sbi, ino);
1708                 if (IS_ERR_OR_NULL(last_page))
1709                         return PTR_ERR_OR_ZERO(last_page);
1710         }
1711 retry:
1712         pagevec_init(&pvec);
1713         index = 0;
1714
1715         while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1716                                 PAGECACHE_TAG_DIRTY))) {
1717                 int i;
1718
1719                 for (i = 0; i < nr_pages; i++) {
1720                         struct page *page = pvec.pages[i];
1721                         bool submitted = false;
1722
1723                         if (unlikely(f2fs_cp_error(sbi))) {
1724                                 f2fs_put_page(last_page, 0);
1725                                 pagevec_release(&pvec);
1726                                 ret = -EIO;
1727                                 goto out;
1728                         }
1729
1730                         if (!IS_DNODE(page) || !is_cold_node(page))
1731                                 continue;
1732                         if (ino_of_node(page) != ino)
1733                                 continue;
1734
1735                         lock_page(page);
1736
1737                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1738 continue_unlock:
1739                                 unlock_page(page);
1740                                 continue;
1741                         }
1742                         if (ino_of_node(page) != ino)
1743                                 goto continue_unlock;
1744
1745                         if (!PageDirty(page) && page != last_page) {
1746                                 /* someone wrote it for us */
1747                                 goto continue_unlock;
1748                         }
1749
1750                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1751
1752                         set_fsync_mark(page, 0);
1753                         set_dentry_mark(page, 0);
1754
1755                         if (!atomic || page == last_page) {
1756                                 set_fsync_mark(page, 1);
1757                                 if (IS_INODE(page)) {
1758                                         if (is_inode_flag_set(inode,
1759                                                                 FI_DIRTY_INODE))
1760                                                 f2fs_update_inode(inode, page);
1761                                         set_dentry_mark(page,
1762                                                 f2fs_need_dentry_mark(sbi, ino));
1763                                 }
1764                                 /* may be written by other thread */
1765                                 if (!PageDirty(page))
1766                                         set_page_dirty(page);
1767                         }
1768
1769                         if (!clear_page_dirty_for_io(page))
1770                                 goto continue_unlock;
1771
1772                         ret = __write_node_page(page, atomic &&
1773                                                 page == last_page,
1774                                                 &submitted, wbc, true,
1775                                                 FS_NODE_IO, seq_id);
1776                         if (ret) {
1777                                 unlock_page(page);
1778                                 f2fs_put_page(last_page, 0);
1779                                 break;
1780                         } else if (submitted) {
1781                                 nwritten++;
1782                         }
1783
1784                         if (page == last_page) {
1785                                 f2fs_put_page(page, 0);
1786                                 marked = true;
1787                                 break;
1788                         }
1789                 }
1790                 pagevec_release(&pvec);
1791                 cond_resched();
1792
1793                 if (ret || marked)
1794                         break;
1795         }
1796         if (!ret && atomic && !marked) {
1797                 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1798                            ino, last_page->index);
1799                 lock_page(last_page);
1800                 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1801                 set_page_dirty(last_page);
1802                 unlock_page(last_page);
1803                 goto retry;
1804         }
1805 out:
1806         if (nwritten)
1807                 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1808         return ret ? -EIO : 0;
1809 }
1810
1811 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1812 {
1813         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1814         bool clean;
1815
1816         if (inode->i_ino != ino)
1817                 return 0;
1818
1819         if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1820                 return 0;
1821
1822         spin_lock(&sbi->inode_lock[DIRTY_META]);
1823         clean = list_empty(&F2FS_I(inode)->gdirty_list);
1824         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1825
1826         if (clean)
1827                 return 0;
1828
1829         inode = igrab(inode);
1830         if (!inode)
1831                 return 0;
1832         return 1;
1833 }
1834
1835 static bool flush_dirty_inode(struct page *page)
1836 {
1837         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1838         struct inode *inode;
1839         nid_t ino = ino_of_node(page);
1840
1841         inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1842         if (!inode)
1843                 return false;
1844
1845         f2fs_update_inode(inode, page);
1846         unlock_page(page);
1847
1848         iput(inode);
1849         return true;
1850 }
1851
1852 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1853 {
1854         pgoff_t index = 0;
1855         struct pagevec pvec;
1856         int nr_pages;
1857
1858         pagevec_init(&pvec);
1859
1860         while ((nr_pages = pagevec_lookup_tag(&pvec,
1861                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1862                 int i;
1863
1864                 for (i = 0; i < nr_pages; i++) {
1865                         struct page *page = pvec.pages[i];
1866
1867                         if (!IS_DNODE(page))
1868                                 continue;
1869
1870                         lock_page(page);
1871
1872                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1873 continue_unlock:
1874                                 unlock_page(page);
1875                                 continue;
1876                         }
1877
1878                         if (!PageDirty(page)) {
1879                                 /* someone wrote it for us */
1880                                 goto continue_unlock;
1881                         }
1882
1883                         /* flush inline_data, if it's async context. */
1884                         if (page_private_inline(page)) {
1885                                 clear_page_private_inline(page);
1886                                 unlock_page(page);
1887                                 flush_inline_data(sbi, ino_of_node(page));
1888                                 continue;
1889                         }
1890                         unlock_page(page);
1891                 }
1892                 pagevec_release(&pvec);
1893                 cond_resched();
1894         }
1895 }
1896
1897 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1898                                 struct writeback_control *wbc,
1899                                 bool do_balance, enum iostat_type io_type)
1900 {
1901         pgoff_t index;
1902         struct pagevec pvec;
1903         int step = 0;
1904         int nwritten = 0;
1905         int ret = 0;
1906         int nr_pages, done = 0;
1907
1908         pagevec_init(&pvec);
1909
1910 next_step:
1911         index = 0;
1912
1913         while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1914                         NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1915                 int i;
1916
1917                 for (i = 0; i < nr_pages; i++) {
1918                         struct page *page = pvec.pages[i];
1919                         bool submitted = false;
1920                         bool may_dirty = true;
1921
1922                         /* give a priority to WB_SYNC threads */
1923                         if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1924                                         wbc->sync_mode == WB_SYNC_NONE) {
1925                                 done = 1;
1926                                 break;
1927                         }
1928
1929                         /*
1930                          * flushing sequence with step:
1931                          * 0. indirect nodes
1932                          * 1. dentry dnodes
1933                          * 2. file dnodes
1934                          */
1935                         if (step == 0 && IS_DNODE(page))
1936                                 continue;
1937                         if (step == 1 && (!IS_DNODE(page) ||
1938                                                 is_cold_node(page)))
1939                                 continue;
1940                         if (step == 2 && (!IS_DNODE(page) ||
1941                                                 !is_cold_node(page)))
1942                                 continue;
1943 lock_node:
1944                         if (wbc->sync_mode == WB_SYNC_ALL)
1945                                 lock_page(page);
1946                         else if (!trylock_page(page))
1947                                 continue;
1948
1949                         if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1950 continue_unlock:
1951                                 unlock_page(page);
1952                                 continue;
1953                         }
1954
1955                         if (!PageDirty(page)) {
1956                                 /* someone wrote it for us */
1957                                 goto continue_unlock;
1958                         }
1959
1960                         /* flush inline_data/inode, if it's async context. */
1961                         if (!do_balance)
1962                                 goto write_node;
1963
1964                         /* flush inline_data */
1965                         if (page_private_inline(page)) {
1966                                 clear_page_private_inline(page);
1967                                 unlock_page(page);
1968                                 flush_inline_data(sbi, ino_of_node(page));
1969                                 goto lock_node;
1970                         }
1971
1972                         /* flush dirty inode */
1973                         if (IS_INODE(page) && may_dirty) {
1974                                 may_dirty = false;
1975                                 if (flush_dirty_inode(page))
1976                                         goto lock_node;
1977                         }
1978 write_node:
1979                         f2fs_wait_on_page_writeback(page, NODE, true, true);
1980
1981                         if (!clear_page_dirty_for_io(page))
1982                                 goto continue_unlock;
1983
1984                         set_fsync_mark(page, 0);
1985                         set_dentry_mark(page, 0);
1986
1987                         ret = __write_node_page(page, false, &submitted,
1988                                                 wbc, do_balance, io_type, NULL);
1989                         if (ret)
1990                                 unlock_page(page);
1991                         else if (submitted)
1992                                 nwritten++;
1993
1994                         if (--wbc->nr_to_write == 0)
1995                                 break;
1996                 }
1997                 pagevec_release(&pvec);
1998                 cond_resched();
1999
2000                 if (wbc->nr_to_write == 0) {
2001                         step = 2;
2002                         break;
2003                 }
2004         }
2005
2006         if (step < 2) {
2007                 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2008                                 wbc->sync_mode == WB_SYNC_NONE && step == 1)
2009                         goto out;
2010                 step++;
2011                 goto next_step;
2012         }
2013 out:
2014         if (nwritten)
2015                 f2fs_submit_merged_write(sbi, NODE);
2016
2017         if (unlikely(f2fs_cp_error(sbi)))
2018                 return -EIO;
2019         return ret;
2020 }
2021
2022 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2023                                                 unsigned int seq_id)
2024 {
2025         struct fsync_node_entry *fn;
2026         struct page *page;
2027         struct list_head *head = &sbi->fsync_node_list;
2028         unsigned long flags;
2029         unsigned int cur_seq_id = 0;
2030         int ret2, ret = 0;
2031
2032         while (seq_id && cur_seq_id < seq_id) {
2033                 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2034                 if (list_empty(head)) {
2035                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2036                         break;
2037                 }
2038                 fn = list_first_entry(head, struct fsync_node_entry, list);
2039                 if (fn->seq_id > seq_id) {
2040                         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2041                         break;
2042                 }
2043                 cur_seq_id = fn->seq_id;
2044                 page = fn->page;
2045                 get_page(page);
2046                 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2047
2048                 f2fs_wait_on_page_writeback(page, NODE, true, false);
2049                 if (TestClearPageError(page))
2050                         ret = -EIO;
2051
2052                 put_page(page);
2053
2054                 if (ret)
2055                         break;
2056         }
2057
2058         ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2059         if (!ret)
2060                 ret = ret2;
2061
2062         return ret;
2063 }
2064
2065 static int f2fs_write_node_pages(struct address_space *mapping,
2066                             struct writeback_control *wbc)
2067 {
2068         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2069         struct blk_plug plug;
2070         long diff;
2071
2072         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2073                 goto skip_write;
2074
2075         /* balancing f2fs's metadata in background */
2076         f2fs_balance_fs_bg(sbi, true);
2077
2078         /* collect a number of dirty node pages and write together */
2079         if (wbc->sync_mode != WB_SYNC_ALL &&
2080                         get_pages(sbi, F2FS_DIRTY_NODES) <
2081                                         nr_pages_to_skip(sbi, NODE))
2082                 goto skip_write;
2083
2084         if (wbc->sync_mode == WB_SYNC_ALL)
2085                 atomic_inc(&sbi->wb_sync_req[NODE]);
2086         else if (atomic_read(&sbi->wb_sync_req[NODE]))
2087                 goto skip_write;
2088
2089         trace_f2fs_writepages(mapping->host, wbc, NODE);
2090
2091         diff = nr_pages_to_write(sbi, NODE, wbc);
2092         blk_start_plug(&plug);
2093         f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2094         blk_finish_plug(&plug);
2095         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2096
2097         if (wbc->sync_mode == WB_SYNC_ALL)
2098                 atomic_dec(&sbi->wb_sync_req[NODE]);
2099         return 0;
2100
2101 skip_write:
2102         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2103         trace_f2fs_writepages(mapping->host, wbc, NODE);
2104         return 0;
2105 }
2106
2107 static int f2fs_set_node_page_dirty(struct page *page)
2108 {
2109         trace_f2fs_set_page_dirty(page, NODE);
2110
2111         if (!PageUptodate(page))
2112                 SetPageUptodate(page);
2113 #ifdef CONFIG_F2FS_CHECK_FS
2114         if (IS_INODE(page))
2115                 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2116 #endif
2117         if (!PageDirty(page)) {
2118                 __set_page_dirty_nobuffers(page);
2119                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2120                 set_page_private_reference(page);
2121                 return 1;
2122         }
2123         return 0;
2124 }
2125
2126 /*
2127  * Structure of the f2fs node operations
2128  */
2129 const struct address_space_operations f2fs_node_aops = {
2130         .writepage      = f2fs_write_node_page,
2131         .writepages     = f2fs_write_node_pages,
2132         .set_page_dirty = f2fs_set_node_page_dirty,
2133         .invalidatepage = f2fs_invalidate_page,
2134         .releasepage    = f2fs_release_page,
2135 #ifdef CONFIG_MIGRATION
2136         .migratepage    = f2fs_migrate_page,
2137 #endif
2138 };
2139
2140 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2141                                                 nid_t n)
2142 {
2143         return radix_tree_lookup(&nm_i->free_nid_root, n);
2144 }
2145
2146 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2147                                 struct free_nid *i)
2148 {
2149         struct f2fs_nm_info *nm_i = NM_I(sbi);
2150         int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2151
2152         if (err)
2153                 return err;
2154
2155         nm_i->nid_cnt[FREE_NID]++;
2156         list_add_tail(&i->list, &nm_i->free_nid_list);
2157         return 0;
2158 }
2159
2160 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2161                         struct free_nid *i, enum nid_state state)
2162 {
2163         struct f2fs_nm_info *nm_i = NM_I(sbi);
2164
2165         f2fs_bug_on(sbi, state != i->state);
2166         nm_i->nid_cnt[state]--;
2167         if (state == FREE_NID)
2168                 list_del(&i->list);
2169         radix_tree_delete(&nm_i->free_nid_root, i->nid);
2170 }
2171
2172 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2173                         enum nid_state org_state, enum nid_state dst_state)
2174 {
2175         struct f2fs_nm_info *nm_i = NM_I(sbi);
2176
2177         f2fs_bug_on(sbi, org_state != i->state);
2178         i->state = dst_state;
2179         nm_i->nid_cnt[org_state]--;
2180         nm_i->nid_cnt[dst_state]++;
2181
2182         switch (dst_state) {
2183         case PREALLOC_NID:
2184                 list_del(&i->list);
2185                 break;
2186         case FREE_NID:
2187                 list_add_tail(&i->list, &nm_i->free_nid_list);
2188                 break;
2189         default:
2190                 BUG_ON(1);
2191         }
2192 }
2193
2194 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2195                                                         bool set, bool build)
2196 {
2197         struct f2fs_nm_info *nm_i = NM_I(sbi);
2198         unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2199         unsigned int nid_ofs = nid - START_NID(nid);
2200
2201         if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2202                 return;
2203
2204         if (set) {
2205                 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2206                         return;
2207                 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2208                 nm_i->free_nid_count[nat_ofs]++;
2209         } else {
2210                 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2211                         return;
2212                 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2213                 if (!build)
2214                         nm_i->free_nid_count[nat_ofs]--;
2215         }
2216 }
2217
2218 /* return if the nid is recognized as free */
2219 static bool add_free_nid(struct f2fs_sb_info *sbi,
2220                                 nid_t nid, bool build, bool update)
2221 {
2222         struct f2fs_nm_info *nm_i = NM_I(sbi);
2223         struct free_nid *i, *e;
2224         struct nat_entry *ne;
2225         int err = -EINVAL;
2226         bool ret = false;
2227
2228         /* 0 nid should not be used */
2229         if (unlikely(nid == 0))
2230                 return false;
2231
2232         if (unlikely(f2fs_check_nid_range(sbi, nid)))
2233                 return false;
2234
2235         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2236         i->nid = nid;
2237         i->state = FREE_NID;
2238
2239         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2240
2241         spin_lock(&nm_i->nid_list_lock);
2242
2243         if (build) {
2244                 /*
2245                  *   Thread A             Thread B
2246                  *  - f2fs_create
2247                  *   - f2fs_new_inode
2248                  *    - f2fs_alloc_nid
2249                  *     - __insert_nid_to_list(PREALLOC_NID)
2250                  *                     - f2fs_balance_fs_bg
2251                  *                      - f2fs_build_free_nids
2252                  *                       - __f2fs_build_free_nids
2253                  *                        - scan_nat_page
2254                  *                         - add_free_nid
2255                  *                          - __lookup_nat_cache
2256                  *  - f2fs_add_link
2257                  *   - f2fs_init_inode_metadata
2258                  *    - f2fs_new_inode_page
2259                  *     - f2fs_new_node_page
2260                  *      - set_node_addr
2261                  *  - f2fs_alloc_nid_done
2262                  *   - __remove_nid_from_list(PREALLOC_NID)
2263                  *                         - __insert_nid_to_list(FREE_NID)
2264                  */
2265                 ne = __lookup_nat_cache(nm_i, nid);
2266                 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2267                                 nat_get_blkaddr(ne) != NULL_ADDR))
2268                         goto err_out;
2269
2270                 e = __lookup_free_nid_list(nm_i, nid);
2271                 if (e) {
2272                         if (e->state == FREE_NID)
2273                                 ret = true;
2274                         goto err_out;
2275                 }
2276         }
2277         ret = true;
2278         err = __insert_free_nid(sbi, i);
2279 err_out:
2280         if (update) {
2281                 update_free_nid_bitmap(sbi, nid, ret, build);
2282                 if (!build)
2283                         nm_i->available_nids++;
2284         }
2285         spin_unlock(&nm_i->nid_list_lock);
2286         radix_tree_preload_end();
2287
2288         if (err)
2289                 kmem_cache_free(free_nid_slab, i);
2290         return ret;
2291 }
2292
2293 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2294 {
2295         struct f2fs_nm_info *nm_i = NM_I(sbi);
2296         struct free_nid *i;
2297         bool need_free = false;
2298
2299         spin_lock(&nm_i->nid_list_lock);
2300         i = __lookup_free_nid_list(nm_i, nid);
2301         if (i && i->state == FREE_NID) {
2302                 __remove_free_nid(sbi, i, FREE_NID);
2303                 need_free = true;
2304         }
2305         spin_unlock(&nm_i->nid_list_lock);
2306
2307         if (need_free)
2308                 kmem_cache_free(free_nid_slab, i);
2309 }
2310
2311 static int scan_nat_page(struct f2fs_sb_info *sbi,
2312                         struct page *nat_page, nid_t start_nid)
2313 {
2314         struct f2fs_nm_info *nm_i = NM_I(sbi);
2315         struct f2fs_nat_block *nat_blk = page_address(nat_page);
2316         block_t blk_addr;
2317         unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2318         int i;
2319
2320         __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2321
2322         i = start_nid % NAT_ENTRY_PER_BLOCK;
2323
2324         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2325                 if (unlikely(start_nid >= nm_i->max_nid))
2326                         break;
2327
2328                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2329
2330                 if (blk_addr == NEW_ADDR)
2331                         return -EINVAL;
2332
2333                 if (blk_addr == NULL_ADDR) {
2334                         add_free_nid(sbi, start_nid, true, true);
2335                 } else {
2336                         spin_lock(&NM_I(sbi)->nid_list_lock);
2337                         update_free_nid_bitmap(sbi, start_nid, false, true);
2338                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2339                 }
2340         }
2341
2342         return 0;
2343 }
2344
2345 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2346 {
2347         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2348         struct f2fs_journal *journal = curseg->journal;
2349         int i;
2350
2351         down_read(&curseg->journal_rwsem);
2352         for (i = 0; i < nats_in_cursum(journal); i++) {
2353                 block_t addr;
2354                 nid_t nid;
2355
2356                 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2357                 nid = le32_to_cpu(nid_in_journal(journal, i));
2358                 if (addr == NULL_ADDR)
2359                         add_free_nid(sbi, nid, true, false);
2360                 else
2361                         remove_free_nid(sbi, nid);
2362         }
2363         up_read(&curseg->journal_rwsem);
2364 }
2365
2366 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2367 {
2368         struct f2fs_nm_info *nm_i = NM_I(sbi);
2369         unsigned int i, idx;
2370         nid_t nid;
2371
2372         down_read(&nm_i->nat_tree_lock);
2373
2374         for (i = 0; i < nm_i->nat_blocks; i++) {
2375                 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2376                         continue;
2377                 if (!nm_i->free_nid_count[i])
2378                         continue;
2379                 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2380                         idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2381                                                 NAT_ENTRY_PER_BLOCK, idx);
2382                         if (idx >= NAT_ENTRY_PER_BLOCK)
2383                                 break;
2384
2385                         nid = i * NAT_ENTRY_PER_BLOCK + idx;
2386                         add_free_nid(sbi, nid, true, false);
2387
2388                         if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2389                                 goto out;
2390                 }
2391         }
2392 out:
2393         scan_curseg_cache(sbi);
2394
2395         up_read(&nm_i->nat_tree_lock);
2396 }
2397
2398 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2399                                                 bool sync, bool mount)
2400 {
2401         struct f2fs_nm_info *nm_i = NM_I(sbi);
2402         int i = 0, ret;
2403         nid_t nid = nm_i->next_scan_nid;
2404
2405         if (unlikely(nid >= nm_i->max_nid))
2406                 nid = 0;
2407
2408         if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2409                 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2410
2411         /* Enough entries */
2412         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2413                 return 0;
2414
2415         if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2416                 return 0;
2417
2418         if (!mount) {
2419                 /* try to find free nids in free_nid_bitmap */
2420                 scan_free_nid_bits(sbi);
2421
2422                 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2423                         return 0;
2424         }
2425
2426         /* readahead nat pages to be scanned */
2427         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2428                                                         META_NAT, true);
2429
2430         down_read(&nm_i->nat_tree_lock);
2431
2432         while (1) {
2433                 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2434                                                 nm_i->nat_block_bitmap)) {
2435                         struct page *page = get_current_nat_page(sbi, nid);
2436
2437                         if (IS_ERR(page)) {
2438                                 ret = PTR_ERR(page);
2439                         } else {
2440                                 ret = scan_nat_page(sbi, page, nid);
2441                                 f2fs_put_page(page, 1);
2442                         }
2443
2444                         if (ret) {
2445                                 up_read(&nm_i->nat_tree_lock);
2446                                 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2447                                 return ret;
2448                         }
2449                 }
2450
2451                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2452                 if (unlikely(nid >= nm_i->max_nid))
2453                         nid = 0;
2454
2455                 if (++i >= FREE_NID_PAGES)
2456                         break;
2457         }
2458
2459         /* go to the next free nat pages to find free nids abundantly */
2460         nm_i->next_scan_nid = nid;
2461
2462         /* find free nids from current sum_pages */
2463         scan_curseg_cache(sbi);
2464
2465         up_read(&nm_i->nat_tree_lock);
2466
2467         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2468                                         nm_i->ra_nid_pages, META_NAT, false);
2469
2470         return 0;
2471 }
2472
2473 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2474 {
2475         int ret;
2476
2477         mutex_lock(&NM_I(sbi)->build_lock);
2478         ret = __f2fs_build_free_nids(sbi, sync, mount);
2479         mutex_unlock(&NM_I(sbi)->build_lock);
2480
2481         return ret;
2482 }
2483
2484 /*
2485  * If this function returns success, caller can obtain a new nid
2486  * from second parameter of this function.
2487  * The returned nid could be used ino as well as nid when inode is created.
2488  */
2489 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2490 {
2491         struct f2fs_nm_info *nm_i = NM_I(sbi);
2492         struct free_nid *i = NULL;
2493 retry:
2494         if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2495                 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2496                 return false;
2497         }
2498
2499         spin_lock(&nm_i->nid_list_lock);
2500
2501         if (unlikely(nm_i->available_nids == 0)) {
2502                 spin_unlock(&nm_i->nid_list_lock);
2503                 return false;
2504         }
2505
2506         /* We should not use stale free nids created by f2fs_build_free_nids */
2507         if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2508                 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2509                 i = list_first_entry(&nm_i->free_nid_list,
2510                                         struct free_nid, list);
2511                 *nid = i->nid;
2512
2513                 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2514                 nm_i->available_nids--;
2515
2516                 update_free_nid_bitmap(sbi, *nid, false, false);
2517
2518                 spin_unlock(&nm_i->nid_list_lock);
2519                 return true;
2520         }
2521         spin_unlock(&nm_i->nid_list_lock);
2522
2523         /* Let's scan nat pages and its caches to get free nids */
2524         if (!f2fs_build_free_nids(sbi, true, false))
2525                 goto retry;
2526         return false;
2527 }
2528
2529 /*
2530  * f2fs_alloc_nid() should be called prior to this function.
2531  */
2532 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2533 {
2534         struct f2fs_nm_info *nm_i = NM_I(sbi);
2535         struct free_nid *i;
2536
2537         spin_lock(&nm_i->nid_list_lock);
2538         i = __lookup_free_nid_list(nm_i, nid);
2539         f2fs_bug_on(sbi, !i);
2540         __remove_free_nid(sbi, i, PREALLOC_NID);
2541         spin_unlock(&nm_i->nid_list_lock);
2542
2543         kmem_cache_free(free_nid_slab, i);
2544 }
2545
2546 /*
2547  * f2fs_alloc_nid() should be called prior to this function.
2548  */
2549 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2550 {
2551         struct f2fs_nm_info *nm_i = NM_I(sbi);
2552         struct free_nid *i;
2553         bool need_free = false;
2554
2555         if (!nid)
2556                 return;
2557
2558         spin_lock(&nm_i->nid_list_lock);
2559         i = __lookup_free_nid_list(nm_i, nid);
2560         f2fs_bug_on(sbi, !i);
2561
2562         if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2563                 __remove_free_nid(sbi, i, PREALLOC_NID);
2564                 need_free = true;
2565         } else {
2566                 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2567         }
2568
2569         nm_i->available_nids++;
2570
2571         update_free_nid_bitmap(sbi, nid, true, false);
2572
2573         spin_unlock(&nm_i->nid_list_lock);
2574
2575         if (need_free)
2576                 kmem_cache_free(free_nid_slab, i);
2577 }
2578
2579 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2580 {
2581         struct f2fs_nm_info *nm_i = NM_I(sbi);
2582         int nr = nr_shrink;
2583
2584         if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2585                 return 0;
2586
2587         if (!mutex_trylock(&nm_i->build_lock))
2588                 return 0;
2589
2590         while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2591                 struct free_nid *i, *next;
2592                 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2593
2594                 spin_lock(&nm_i->nid_list_lock);
2595                 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2596                         if (!nr_shrink || !batch ||
2597                                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2598                                 break;
2599                         __remove_free_nid(sbi, i, FREE_NID);
2600                         kmem_cache_free(free_nid_slab, i);
2601                         nr_shrink--;
2602                         batch--;
2603                 }
2604                 spin_unlock(&nm_i->nid_list_lock);
2605         }
2606
2607         mutex_unlock(&nm_i->build_lock);
2608
2609         return nr - nr_shrink;
2610 }
2611
2612 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2613 {
2614         void *src_addr, *dst_addr;
2615         size_t inline_size;
2616         struct page *ipage;
2617         struct f2fs_inode *ri;
2618
2619         ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2620         if (IS_ERR(ipage))
2621                 return PTR_ERR(ipage);
2622
2623         ri = F2FS_INODE(page);
2624         if (ri->i_inline & F2FS_INLINE_XATTR) {
2625                 if (!f2fs_has_inline_xattr(inode)) {
2626                         set_inode_flag(inode, FI_INLINE_XATTR);
2627                         stat_inc_inline_xattr(inode);
2628                 }
2629         } else {
2630                 if (f2fs_has_inline_xattr(inode)) {
2631                         stat_dec_inline_xattr(inode);
2632                         clear_inode_flag(inode, FI_INLINE_XATTR);
2633                 }
2634                 goto update_inode;
2635         }
2636
2637         dst_addr = inline_xattr_addr(inode, ipage);
2638         src_addr = inline_xattr_addr(inode, page);
2639         inline_size = inline_xattr_size(inode);
2640
2641         f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2642         memcpy(dst_addr, src_addr, inline_size);
2643 update_inode:
2644         f2fs_update_inode(inode, ipage);
2645         f2fs_put_page(ipage, 1);
2646         return 0;
2647 }
2648
2649 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2650 {
2651         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2652         nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2653         nid_t new_xnid;
2654         struct dnode_of_data dn;
2655         struct node_info ni;
2656         struct page *xpage;
2657         int err;
2658
2659         if (!prev_xnid)
2660                 goto recover_xnid;
2661
2662         /* 1: invalidate the previous xattr nid */
2663         err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2664         if (err)
2665                 return err;
2666
2667         f2fs_invalidate_blocks(sbi, ni.blk_addr);
2668         dec_valid_node_count(sbi, inode, false);
2669         set_node_addr(sbi, &ni, NULL_ADDR, false);
2670
2671 recover_xnid:
2672         /* 2: update xattr nid in inode */
2673         if (!f2fs_alloc_nid(sbi, &new_xnid))
2674                 return -ENOSPC;
2675
2676         set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2677         xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2678         if (IS_ERR(xpage)) {
2679                 f2fs_alloc_nid_failed(sbi, new_xnid);
2680                 return PTR_ERR(xpage);
2681         }
2682
2683         f2fs_alloc_nid_done(sbi, new_xnid);
2684         f2fs_update_inode_page(inode);
2685
2686         /* 3: update and set xattr node page dirty */
2687         memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2688
2689         set_page_dirty(xpage);
2690         f2fs_put_page(xpage, 1);
2691
2692         return 0;
2693 }
2694
2695 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2696 {
2697         struct f2fs_inode *src, *dst;
2698         nid_t ino = ino_of_node(page);
2699         struct node_info old_ni, new_ni;
2700         struct page *ipage;
2701         int err;
2702
2703         err = f2fs_get_node_info(sbi, ino, &old_ni);
2704         if (err)
2705                 return err;
2706
2707         if (unlikely(old_ni.blk_addr != NULL_ADDR))
2708                 return -EINVAL;
2709 retry:
2710         ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2711         if (!ipage) {
2712                 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2713                 goto retry;
2714         }
2715
2716         /* Should not use this inode from free nid list */
2717         remove_free_nid(sbi, ino);
2718
2719         if (!PageUptodate(ipage))
2720                 SetPageUptodate(ipage);
2721         fill_node_footer(ipage, ino, ino, 0, true);
2722         set_cold_node(ipage, false);
2723
2724         src = F2FS_INODE(page);
2725         dst = F2FS_INODE(ipage);
2726
2727         memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2728         dst->i_size = 0;
2729         dst->i_blocks = cpu_to_le64(1);
2730         dst->i_links = cpu_to_le32(1);
2731         dst->i_xattr_nid = 0;
2732         dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2733         if (dst->i_inline & F2FS_EXTRA_ATTR) {
2734                 dst->i_extra_isize = src->i_extra_isize;
2735
2736                 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2737                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2738                                                         i_inline_xattr_size))
2739                         dst->i_inline_xattr_size = src->i_inline_xattr_size;
2740
2741                 if (f2fs_sb_has_project_quota(sbi) &&
2742                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2743                                                                 i_projid))
2744                         dst->i_projid = src->i_projid;
2745
2746                 if (f2fs_sb_has_inode_crtime(sbi) &&
2747                         F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2748                                                         i_crtime_nsec)) {
2749                         dst->i_crtime = src->i_crtime;
2750                         dst->i_crtime_nsec = src->i_crtime_nsec;
2751                 }
2752         }
2753
2754         new_ni = old_ni;
2755         new_ni.ino = ino;
2756
2757         if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2758                 WARN_ON(1);
2759         set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2760         inc_valid_inode_count(sbi);
2761         set_page_dirty(ipage);
2762         f2fs_put_page(ipage, 1);
2763         return 0;
2764 }
2765
2766 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2767                         unsigned int segno, struct f2fs_summary_block *sum)
2768 {
2769         struct f2fs_node *rn;
2770         struct f2fs_summary *sum_entry;
2771         block_t addr;
2772         int i, idx, last_offset, nrpages;
2773
2774         /* scan the node segment */
2775         last_offset = sbi->blocks_per_seg;
2776         addr = START_BLOCK(sbi, segno);
2777         sum_entry = &sum->entries[0];
2778
2779         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2780                 nrpages = bio_max_segs(last_offset - i);
2781
2782                 /* readahead node pages */
2783                 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2784
2785                 for (idx = addr; idx < addr + nrpages; idx++) {
2786                         struct page *page = f2fs_get_tmp_page(sbi, idx);
2787
2788                         if (IS_ERR(page))
2789                                 return PTR_ERR(page);
2790
2791                         rn = F2FS_NODE(page);
2792                         sum_entry->nid = rn->footer.nid;
2793                         sum_entry->version = 0;
2794                         sum_entry->ofs_in_node = 0;
2795                         sum_entry++;
2796                         f2fs_put_page(page, 1);
2797                 }
2798
2799                 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2800                                                         addr + nrpages);
2801         }
2802         return 0;
2803 }
2804
2805 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2806 {
2807         struct f2fs_nm_info *nm_i = NM_I(sbi);
2808         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2809         struct f2fs_journal *journal = curseg->journal;
2810         int i;
2811
2812         down_write(&curseg->journal_rwsem);
2813         for (i = 0; i < nats_in_cursum(journal); i++) {
2814                 struct nat_entry *ne;
2815                 struct f2fs_nat_entry raw_ne;
2816                 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2817
2818                 if (f2fs_check_nid_range(sbi, nid))
2819                         continue;
2820
2821                 raw_ne = nat_in_journal(journal, i);
2822
2823                 ne = __lookup_nat_cache(nm_i, nid);
2824                 if (!ne) {
2825                         ne = __alloc_nat_entry(nid, true);
2826                         __init_nat_entry(nm_i, ne, &raw_ne, true);
2827                 }
2828
2829                 /*
2830                  * if a free nat in journal has not been used after last
2831                  * checkpoint, we should remove it from available nids,
2832                  * since later we will add it again.
2833                  */
2834                 if (!get_nat_flag(ne, IS_DIRTY) &&
2835                                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2836                         spin_lock(&nm_i->nid_list_lock);
2837                         nm_i->available_nids--;
2838                         spin_unlock(&nm_i->nid_list_lock);
2839                 }
2840
2841                 __set_nat_cache_dirty(nm_i, ne);
2842         }
2843         update_nats_in_cursum(journal, -i);
2844         up_write(&curseg->journal_rwsem);
2845 }
2846
2847 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2848                                                 struct list_head *head, int max)
2849 {
2850         struct nat_entry_set *cur;
2851
2852         if (nes->entry_cnt >= max)
2853                 goto add_out;
2854
2855         list_for_each_entry(cur, head, set_list) {
2856                 if (cur->entry_cnt >= nes->entry_cnt) {
2857                         list_add(&nes->set_list, cur->set_list.prev);
2858                         return;
2859                 }
2860         }
2861 add_out:
2862         list_add_tail(&nes->set_list, head);
2863 }
2864
2865 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2866                                                 struct page *page)
2867 {
2868         struct f2fs_nm_info *nm_i = NM_I(sbi);
2869         unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2870         struct f2fs_nat_block *nat_blk = page_address(page);
2871         int valid = 0;
2872         int i = 0;
2873
2874         if (!enabled_nat_bits(sbi, NULL))
2875                 return;
2876
2877         if (nat_index == 0) {
2878                 valid = 1;
2879                 i = 1;
2880         }
2881         for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2882                 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2883                         valid++;
2884         }
2885         if (valid == 0) {
2886                 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2887                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2888                 return;
2889         }
2890
2891         __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2892         if (valid == NAT_ENTRY_PER_BLOCK)
2893                 __set_bit_le(nat_index, nm_i->full_nat_bits);
2894         else
2895                 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2896 }
2897
2898 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2899                 struct nat_entry_set *set, struct cp_control *cpc)
2900 {
2901         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2902         struct f2fs_journal *journal = curseg->journal;
2903         nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2904         bool to_journal = true;
2905         struct f2fs_nat_block *nat_blk;
2906         struct nat_entry *ne, *cur;
2907         struct page *page = NULL;
2908
2909         /*
2910          * there are two steps to flush nat entries:
2911          * #1, flush nat entries to journal in current hot data summary block.
2912          * #2, flush nat entries to nat page.
2913          */
2914         if (enabled_nat_bits(sbi, cpc) ||
2915                 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2916                 to_journal = false;
2917
2918         if (to_journal) {
2919                 down_write(&curseg->journal_rwsem);
2920         } else {
2921                 page = get_next_nat_page(sbi, start_nid);
2922                 if (IS_ERR(page))
2923                         return PTR_ERR(page);
2924
2925                 nat_blk = page_address(page);
2926                 f2fs_bug_on(sbi, !nat_blk);
2927         }
2928
2929         /* flush dirty nats in nat entry set */
2930         list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2931                 struct f2fs_nat_entry *raw_ne;
2932                 nid_t nid = nat_get_nid(ne);
2933                 int offset;
2934
2935                 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2936
2937                 if (to_journal) {
2938                         offset = f2fs_lookup_journal_in_cursum(journal,
2939                                                         NAT_JOURNAL, nid, 1);
2940                         f2fs_bug_on(sbi, offset < 0);
2941                         raw_ne = &nat_in_journal(journal, offset);
2942                         nid_in_journal(journal, offset) = cpu_to_le32(nid);
2943                 } else {
2944                         raw_ne = &nat_blk->entries[nid - start_nid];
2945                 }
2946                 raw_nat_from_node_info(raw_ne, &ne->ni);
2947                 nat_reset_flag(ne);
2948                 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2949                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2950                         add_free_nid(sbi, nid, false, true);
2951                 } else {
2952                         spin_lock(&NM_I(sbi)->nid_list_lock);
2953                         update_free_nid_bitmap(sbi, nid, false, false);
2954                         spin_unlock(&NM_I(sbi)->nid_list_lock);
2955                 }
2956         }
2957
2958         if (to_journal) {
2959                 up_write(&curseg->journal_rwsem);
2960         } else {
2961                 __update_nat_bits(sbi, start_nid, page);
2962                 f2fs_put_page(page, 1);
2963         }
2964
2965         /* Allow dirty nats by node block allocation in write_begin */
2966         if (!set->entry_cnt) {
2967                 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2968                 kmem_cache_free(nat_entry_set_slab, set);
2969         }
2970         return 0;
2971 }
2972
2973 /*
2974  * This function is called during the checkpointing process.
2975  */
2976 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2977 {
2978         struct f2fs_nm_info *nm_i = NM_I(sbi);
2979         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2980         struct f2fs_journal *journal = curseg->journal;
2981         struct nat_entry_set *setvec[SETVEC_SIZE];
2982         struct nat_entry_set *set, *tmp;
2983         unsigned int found;
2984         nid_t set_idx = 0;
2985         LIST_HEAD(sets);
2986         int err = 0;
2987
2988         /*
2989          * during unmount, let's flush nat_bits before checking
2990          * nat_cnt[DIRTY_NAT].
2991          */
2992         if (enabled_nat_bits(sbi, cpc)) {
2993                 down_write(&nm_i->nat_tree_lock);
2994                 remove_nats_in_journal(sbi);
2995                 up_write(&nm_i->nat_tree_lock);
2996         }
2997
2998         if (!nm_i->nat_cnt[DIRTY_NAT])
2999                 return 0;
3000
3001         down_write(&nm_i->nat_tree_lock);
3002
3003         /*
3004          * if there are no enough space in journal to store dirty nat
3005          * entries, remove all entries from journal and merge them
3006          * into nat entry set.
3007          */
3008         if (enabled_nat_bits(sbi, cpc) ||
3009                 !__has_cursum_space(journal,
3010                         nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3011                 remove_nats_in_journal(sbi);
3012
3013         while ((found = __gang_lookup_nat_set(nm_i,
3014                                         set_idx, SETVEC_SIZE, setvec))) {
3015                 unsigned idx;
3016
3017                 set_idx = setvec[found - 1]->set + 1;
3018                 for (idx = 0; idx < found; idx++)
3019                         __adjust_nat_entry_set(setvec[idx], &sets,
3020                                                 MAX_NAT_JENTRIES(journal));
3021         }
3022
3023         /* flush dirty nats in nat entry set */
3024         list_for_each_entry_safe(set, tmp, &sets, set_list) {
3025                 err = __flush_nat_entry_set(sbi, set, cpc);
3026                 if (err)
3027                         break;
3028         }
3029
3030         up_write(&nm_i->nat_tree_lock);
3031         /* Allow dirty nats by node block allocation in write_begin */
3032
3033         return err;
3034 }
3035
3036 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3037 {
3038         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3039         struct f2fs_nm_info *nm_i = NM_I(sbi);
3040         unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3041         unsigned int i;
3042         __u64 cp_ver = cur_cp_version(ckpt);
3043         block_t nat_bits_addr;
3044
3045         if (!enabled_nat_bits(sbi, NULL))
3046                 return 0;
3047
3048         nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3049         nm_i->nat_bits = f2fs_kvzalloc(sbi,
3050                         nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3051         if (!nm_i->nat_bits)
3052                 return -ENOMEM;
3053
3054         nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3055                                                 nm_i->nat_bits_blocks;
3056         for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3057                 struct page *page;
3058
3059                 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3060                 if (IS_ERR(page))
3061                         return PTR_ERR(page);
3062
3063                 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3064                                         page_address(page), F2FS_BLKSIZE);
3065                 f2fs_put_page(page, 1);
3066         }
3067
3068         cp_ver |= (cur_cp_crc(ckpt) << 32);
3069         if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3070                 disable_nat_bits(sbi, true);
3071                 return 0;
3072         }
3073
3074         nm_i->full_nat_bits = nm_i->nat_bits + 8;
3075         nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3076
3077         f2fs_notice(sbi, "Found nat_bits in checkpoint");
3078         return 0;
3079 }
3080
3081 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3082 {
3083         struct f2fs_nm_info *nm_i = NM_I(sbi);
3084         unsigned int i = 0;
3085         nid_t nid, last_nid;
3086
3087         if (!enabled_nat_bits(sbi, NULL))
3088                 return;
3089
3090         for (i = 0; i < nm_i->nat_blocks; i++) {
3091                 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3092                 if (i >= nm_i->nat_blocks)
3093                         break;
3094
3095                 __set_bit_le(i, nm_i->nat_block_bitmap);
3096
3097                 nid = i * NAT_ENTRY_PER_BLOCK;
3098                 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3099
3100                 spin_lock(&NM_I(sbi)->nid_list_lock);
3101                 for (; nid < last_nid; nid++)
3102                         update_free_nid_bitmap(sbi, nid, true, true);
3103                 spin_unlock(&NM_I(sbi)->nid_list_lock);
3104         }
3105
3106         for (i = 0; i < nm_i->nat_blocks; i++) {
3107                 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3108                 if (i >= nm_i->nat_blocks)
3109                         break;
3110
3111                 __set_bit_le(i, nm_i->nat_block_bitmap);
3112         }
3113 }
3114
3115 static int init_node_manager(struct f2fs_sb_info *sbi)
3116 {
3117         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3118         struct f2fs_nm_info *nm_i = NM_I(sbi);
3119         unsigned char *version_bitmap;
3120         unsigned int nat_segs;
3121         int err;
3122
3123         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3124
3125         /* segment_count_nat includes pair segment so divide to 2. */
3126         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3127         nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3128         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3129
3130         /* not used nids: 0, node, meta, (and root counted as valid node) */
3131         nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3132                                                 F2FS_RESERVED_NODE_NUM;
3133         nm_i->nid_cnt[FREE_NID] = 0;
3134         nm_i->nid_cnt[PREALLOC_NID] = 0;
3135         nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3136         nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3137         nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3138
3139         INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3140         INIT_LIST_HEAD(&nm_i->free_nid_list);
3141         INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3142         INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3143         INIT_LIST_HEAD(&nm_i->nat_entries);
3144         spin_lock_init(&nm_i->nat_list_lock);
3145
3146         mutex_init(&nm_i->build_lock);
3147         spin_lock_init(&nm_i->nid_list_lock);
3148         init_rwsem(&nm_i->nat_tree_lock);
3149
3150         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3151         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3152         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3153         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3154                                         GFP_KERNEL);
3155         if (!nm_i->nat_bitmap)
3156                 return -ENOMEM;
3157
3158         err = __get_nat_bitmaps(sbi);
3159         if (err)
3160                 return err;
3161
3162 #ifdef CONFIG_F2FS_CHECK_FS
3163         nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3164                                         GFP_KERNEL);
3165         if (!nm_i->nat_bitmap_mir)
3166                 return -ENOMEM;
3167 #endif
3168
3169         return 0;
3170 }
3171
3172 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3173 {
3174         struct f2fs_nm_info *nm_i = NM_I(sbi);
3175         int i;
3176
3177         nm_i->free_nid_bitmap =
3178                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3179                                               nm_i->nat_blocks),
3180                               GFP_KERNEL);
3181         if (!nm_i->free_nid_bitmap)
3182                 return -ENOMEM;
3183
3184         for (i = 0; i < nm_i->nat_blocks; i++) {
3185                 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3186                         f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3187                 if (!nm_i->free_nid_bitmap[i])
3188                         return -ENOMEM;
3189         }
3190
3191         nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3192                                                                 GFP_KERNEL);
3193         if (!nm_i->nat_block_bitmap)
3194                 return -ENOMEM;
3195
3196         nm_i->free_nid_count =
3197                 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3198                                               nm_i->nat_blocks),
3199                               GFP_KERNEL);
3200         if (!nm_i->free_nid_count)
3201                 return -ENOMEM;
3202         return 0;
3203 }
3204
3205 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3206 {
3207         int err;
3208
3209         sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3210                                                         GFP_KERNEL);
3211         if (!sbi->nm_info)
3212                 return -ENOMEM;
3213
3214         err = init_node_manager(sbi);
3215         if (err)
3216                 return err;
3217
3218         err = init_free_nid_cache(sbi);
3219         if (err)
3220                 return err;
3221
3222         /* load free nid status from nat_bits table */
3223         load_free_nid_bitmap(sbi);
3224
3225         return f2fs_build_free_nids(sbi, true, true);
3226 }
3227
3228 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3229 {
3230         struct f2fs_nm_info *nm_i = NM_I(sbi);
3231         struct free_nid *i, *next_i;
3232         struct nat_entry *natvec[NATVEC_SIZE];
3233         struct nat_entry_set *setvec[SETVEC_SIZE];
3234         nid_t nid = 0;
3235         unsigned int found;
3236
3237         if (!nm_i)
3238                 return;
3239
3240         /* destroy free nid list */
3241         spin_lock(&nm_i->nid_list_lock);
3242         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3243                 __remove_free_nid(sbi, i, FREE_NID);
3244                 spin_unlock(&nm_i->nid_list_lock);
3245                 kmem_cache_free(free_nid_slab, i);
3246                 spin_lock(&nm_i->nid_list_lock);
3247         }
3248         f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3249         f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3250         f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3251         spin_unlock(&nm_i->nid_list_lock);
3252
3253         /* destroy nat cache */
3254         down_write(&nm_i->nat_tree_lock);
3255         while ((found = __gang_lookup_nat_cache(nm_i,
3256                                         nid, NATVEC_SIZE, natvec))) {
3257                 unsigned idx;
3258
3259                 nid = nat_get_nid(natvec[found - 1]) + 1;
3260                 for (idx = 0; idx < found; idx++) {
3261                         spin_lock(&nm_i->nat_list_lock);
3262                         list_del(&natvec[idx]->list);
3263                         spin_unlock(&nm_i->nat_list_lock);
3264
3265                         __del_from_nat_cache(nm_i, natvec[idx]);
3266                 }
3267         }
3268         f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3269
3270         /* destroy nat set cache */
3271         nid = 0;
3272         while ((found = __gang_lookup_nat_set(nm_i,
3273                                         nid, SETVEC_SIZE, setvec))) {
3274                 unsigned idx;
3275
3276                 nid = setvec[found - 1]->set + 1;
3277                 for (idx = 0; idx < found; idx++) {
3278                         /* entry_cnt is not zero, when cp_error was occurred */
3279                         f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3280                         radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3281                         kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3282                 }
3283         }
3284         up_write(&nm_i->nat_tree_lock);
3285
3286         kvfree(nm_i->nat_block_bitmap);
3287         if (nm_i->free_nid_bitmap) {
3288                 int i;
3289
3290                 for (i = 0; i < nm_i->nat_blocks; i++)
3291                         kvfree(nm_i->free_nid_bitmap[i]);
3292                 kvfree(nm_i->free_nid_bitmap);
3293         }
3294         kvfree(nm_i->free_nid_count);
3295
3296         kvfree(nm_i->nat_bitmap);
3297         kvfree(nm_i->nat_bits);
3298 #ifdef CONFIG_F2FS_CHECK_FS
3299         kvfree(nm_i->nat_bitmap_mir);
3300 #endif
3301         sbi->nm_info = NULL;
3302         kfree(nm_i);
3303 }
3304
3305 int __init f2fs_create_node_manager_caches(void)
3306 {
3307         nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3308                         sizeof(struct nat_entry));
3309         if (!nat_entry_slab)
3310                 goto fail;
3311
3312         free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3313                         sizeof(struct free_nid));
3314         if (!free_nid_slab)
3315                 goto destroy_nat_entry;
3316
3317         nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3318                         sizeof(struct nat_entry_set));
3319         if (!nat_entry_set_slab)
3320                 goto destroy_free_nid;
3321
3322         fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3323                         sizeof(struct fsync_node_entry));
3324         if (!fsync_node_entry_slab)
3325                 goto destroy_nat_entry_set;
3326         return 0;
3327
3328 destroy_nat_entry_set:
3329         kmem_cache_destroy(nat_entry_set_slab);
3330 destroy_free_nid:
3331         kmem_cache_destroy(free_nid_slab);
3332 destroy_nat_entry:
3333         kmem_cache_destroy(nat_entry_slab);
3334 fail:
3335         return -ENOMEM;
3336 }
3337
3338 void f2fs_destroy_node_manager_caches(void)
3339 {
3340         kmem_cache_destroy(fsync_node_entry_slab);
3341         kmem_cache_destroy(nat_entry_set_slab);
3342         kmem_cache_destroy(free_nid_slab);
3343         kmem_cache_destroy(nat_entry_slab);
3344 }