fs/f2fs/checkpoint.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/checkpoint.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/bio.h>
  10 #include <linux/mpage.h>
  11 #include <linux/writeback.h>
  12 #include <linux/blkdev.h>
  13 #include <linux/f2fs_fs.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/swap.h>
  16 #include <linux/kthread.h>
  17
  18 #include "f2fs.h"
  19 #include "node.h"
  20 #include "segment.h"
  21 #include "iostat.h"
  22 #include <trace/events/f2fs.h>
  23
  24 #define DEFAULT_CHECKPOINT_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
  25
  26 static struct kmem_cache *ino_entry_slab;
  27 struct kmem_cache *f2fs_inode_entry_slab;
  28
  29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
  30 {
  31         f2fs_build_fault_attr(sbi, 0, 0);
  32         set_ckpt_flags(sbi, CP_ERROR_FLAG);
  33         if (!end_io)
  34                 f2fs_flush_merged_writes(sbi);
  35 }
  36
  37 /*
  38  * We guarantee no failure on the returned page.
  39  */
  40 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  41 {
  42         struct address_space *mapping = META_MAPPING(sbi);
  43         struct page *page;
  44 repeat:
  45         page = f2fs_grab_cache_page(mapping, index, false);
  46         if (!page) {
  47                 cond_resched();
  48                 goto repeat;
  49         }
  50         f2fs_wait_on_page_writeback(page, META, true, true);
  51         if (!PageUptodate(page))
  52                 SetPageUptodate(page);
  53         return page;
  54 }
  55
  56 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
  57                                                         bool is_meta)
  58 {
  59         struct address_space *mapping = META_MAPPING(sbi);
  60         struct page *page;
  61         struct f2fs_io_info fio = {
  62                 .sbi = sbi,
  63                 .type = META,
  64                 .op = REQ_OP_READ,
  65                 .op_flags = REQ_META | REQ_PRIO,
  66                 .old_blkaddr = index,
  67                 .new_blkaddr = index,
  68                 .encrypted_page = NULL,
  69                 .is_por = !is_meta,
  70         };
  71         int err;
  72
  73         if (unlikely(!is_meta))
  74                 fio.op_flags &= ~REQ_META;
  75 repeat:
  76         page = f2fs_grab_cache_page(mapping, index, false);
  77         if (!page) {
  78                 cond_resched();
  79                 goto repeat;
  80         }
  81         if (PageUptodate(page))
  82                 goto out;
  83
  84         fio.page = page;
  85
  86         err = f2fs_submit_page_bio(&fio);
  87         if (err) {
  88                 f2fs_put_page(page, 1);
  89                 return ERR_PTR(err);
  90         }
  91
  92         f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
  93
  94         lock_page(page);
  95         if (unlikely(page->mapping != mapping)) {
  96                 f2fs_put_page(page, 1);
  97                 goto repeat;
  98         }
  99
 100         if (unlikely(!PageUptodate(page))) {
 101                 f2fs_handle_page_eio(sbi, page->index, META);
 102                 f2fs_put_page(page, 1);
 103                 return ERR_PTR(-EIO);
 104         }
 105 out:
 106         return page;
 107 }
 108
 109 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 110 {
 111         return __get_meta_page(sbi, index, true);
 112 }
 113
 114 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
 115 {
 116         struct page *page;
 117         int count = 0;
 118
 119 retry:
 120         page = __get_meta_page(sbi, index, true);
 121         if (IS_ERR(page)) {
 122                 if (PTR_ERR(page) == -EIO &&
 123                                 ++count <= DEFAULT_RETRY_IO_COUNT)
 124                         goto retry;
 125                 f2fs_stop_checkpoint(sbi, false);
 126         }
 127         return page;
 128 }
 129
 130 /* for POR only */
 131 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 132 {
 133         return __get_meta_page(sbi, index, false);
 134 }
 135
 136 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
 137                                                         int type)
 138 {
 139         struct seg_entry *se;
 140         unsigned int segno, offset;
 141         bool exist;
 142
 143         if (type != DATA_GENERIC_ENHANCE && type != DATA_GENERIC_ENHANCE_READ)
 144                 return true;
 145
 146         segno = GET_SEGNO(sbi, blkaddr);
 147         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 148         se = get_seg_entry(sbi, segno);
 149
 150         exist = f2fs_test_bit(offset, se->cur_valid_map);
 151         if (!exist && type == DATA_GENERIC_ENHANCE) {
 152                 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
 153                          blkaddr, exist);
 154                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 155                 dump_stack();
 156         }
 157         return exist;
 158 }
 159
 160 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
 161                                         block_t blkaddr, int type)
 162 {
 163         switch (type) {
 164         case META_NAT:
 165                 break;
 166         case META_SIT:
 167                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 168                         return false;
 169                 break;
 170         case META_SSA:
 171                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 172                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 173                         return false;
 174                 break;
 175         case META_CP:
 176                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 177                         blkaddr < __start_cp_addr(sbi)))
 178                         return false;
 179                 break;
 180         case META_POR:
 181                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 182                         blkaddr < MAIN_BLKADDR(sbi)))
 183                         return false;
 184                 break;
 185         case DATA_GENERIC:
 186         case DATA_GENERIC_ENHANCE:
 187         case DATA_GENERIC_ENHANCE_READ:
 188                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 189                                 blkaddr < MAIN_BLKADDR(sbi))) {
 190                         f2fs_warn(sbi, "access invalid blkaddr:%u",
 191                                   blkaddr);
 192                         set_sbi_flag(sbi, SBI_NEED_FSCK);
 193                         dump_stack();
 194                         return false;
 195                 } else {
 196                         return __is_bitmap_valid(sbi, blkaddr, type);
 197                 }
 198                 break;
 199         case META_GENERIC:
 200                 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
 201                         blkaddr >= MAIN_BLKADDR(sbi)))
 202                         return false;
 203                 break;
 204         default:
 205                 BUG();
 206         }
 207
 208         return true;
 209 }
 210
 211 /*
 212  * Readahead CP/NAT/SIT/SSA/POR pages
 213  */
 214 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 215                                                         int type, bool sync)
 216 {
 217         struct page *page;
 218         block_t blkno = start;
 219         struct f2fs_io_info fio = {
 220                 .sbi = sbi,
 221                 .type = META,
 222                 .op = REQ_OP_READ,
 223                 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
 224                 .encrypted_page = NULL,
 225                 .in_list = false,
 226                 .is_por = (type == META_POR),
 227         };
 228         struct blk_plug plug;
 229         int err;
 230
 231         if (unlikely(type == META_POR))
 232                 fio.op_flags &= ~REQ_META;
 233
 234         blk_start_plug(&plug);
 235         for (; nrpages-- > 0; blkno++) {
 236
 237                 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
 238                         goto out;
 239
 240                 switch (type) {
 241                 case META_NAT:
 242                         if (unlikely(blkno >=
 243                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 244                                 blkno = 0;
 245                         /* get nat block addr */
 246                         fio.new_blkaddr = current_nat_addr(sbi,
 247                                         blkno * NAT_ENTRY_PER_BLOCK);
 248                         break;
 249                 case META_SIT:
 250                         if (unlikely(blkno >= TOTAL_SEGS(sbi)))
 251                                 goto out;
 252                         /* get sit block addr */
 253                         fio.new_blkaddr = current_sit_addr(sbi,
 254                                         blkno * SIT_ENTRY_PER_BLOCK);
 255                         break;
 256                 case META_SSA:
 257                 case META_CP:
 258                 case META_POR:
 259                         fio.new_blkaddr = blkno;
 260                         break;
 261                 default:
 262                         BUG();
 263                 }
 264
 265                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 266                                                 fio.new_blkaddr, false);
 267                 if (!page)
 268                         continue;
 269                 if (PageUptodate(page)) {
 270                         f2fs_put_page(page, 1);
 271                         continue;
 272                 }
 273
 274                 fio.page = page;
 275                 err = f2fs_submit_page_bio(&fio);
 276                 f2fs_put_page(page, err ? 1 : 0);
 277
 278                 if (!err)
 279                         f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
 280         }
 281 out:
 282         blk_finish_plug(&plug);
 283         return blkno - start;
 284 }
 285
 286 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
 287                                                         unsigned int ra_blocks)
 288 {
 289         struct page *page;
 290         bool readahead = false;
 291
 292         if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
 293                 return;
 294
 295         page = find_get_page(META_MAPPING(sbi), index);
 296         if (!page || !PageUptodate(page))
 297                 readahead = true;
 298         f2fs_put_page(page, 0);
 299
 300         if (readahead)
 301                 f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
 302 }
 303
 304 static int __f2fs_write_meta_page(struct page *page,
 305                                 struct writeback_control *wbc,
 306                                 enum iostat_type io_type)
 307 {
 308         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 309
 310         trace_f2fs_writepage(page, META);
 311
 312         if (unlikely(f2fs_cp_error(sbi)))
 313                 goto redirty_out;
 314         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 315                 goto redirty_out;
 316         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 317                 goto redirty_out;
 318
 319         f2fs_do_write_meta_page(sbi, page, io_type);
 320         dec_page_count(sbi, F2FS_DIRTY_META);
 321
 322         if (wbc->for_reclaim)
 323                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
 324
 325         unlock_page(page);
 326
 327         if (unlikely(f2fs_cp_error(sbi)))
 328                 f2fs_submit_merged_write(sbi, META);
 329
 330         return 0;
 331
 332 redirty_out:
 333         redirty_page_for_writepage(wbc, page);
 334         return AOP_WRITEPAGE_ACTIVATE;
 335 }
 336
 337 static int f2fs_write_meta_page(struct page *page,
 338                                 struct writeback_control *wbc)
 339 {
 340         return __f2fs_write_meta_page(page, wbc, FS_META_IO);
 341 }
 342
 343 static int f2fs_write_meta_pages(struct address_space *mapping,
 344                                 struct writeback_control *wbc)
 345 {
 346         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 347         long diff, written;
 348
 349         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 350                 goto skip_write;
 351
 352         /* collect a number of dirty meta pages and write together */
 353         if (wbc->sync_mode != WB_SYNC_ALL &&
 354                         get_pages(sbi, F2FS_DIRTY_META) <
 355                                         nr_pages_to_skip(sbi, META))
 356                 goto skip_write;
 357
 358         /* if locked failed, cp will flush dirty pages instead */
 359         if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
 360                 goto skip_write;
 361
 362         trace_f2fs_writepages(mapping->host, wbc, META);
 363         diff = nr_pages_to_write(sbi, META, wbc);
 364         written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
 365         f2fs_up_write(&sbi->cp_global_sem);
 366         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 367         return 0;
 368
 369 skip_write:
 370         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 371         trace_f2fs_writepages(mapping->host, wbc, META);
 372         return 0;
 373 }
 374
 375 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 376                                 long nr_to_write, enum iostat_type io_type)
 377 {
 378         struct address_space *mapping = META_MAPPING(sbi);
 379         pgoff_t index = 0, prev = ULONG_MAX;
 380         struct pagevec pvec;
 381         long nwritten = 0;
 382         int nr_pages;
 383         struct writeback_control wbc = {
 384                 .for_reclaim = 0,
 385         };
 386         struct blk_plug plug;
 387
 388         pagevec_init(&pvec);
 389
 390         blk_start_plug(&plug);
 391
 392         while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 393                                 PAGECACHE_TAG_DIRTY))) {
 394                 int i;
 395
 396                 for (i = 0; i < nr_pages; i++) {
 397                         struct page *page = pvec.pages[i];
 398
 399                         if (prev == ULONG_MAX)
 400                                 prev = page->index - 1;
 401                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 402                                 pagevec_release(&pvec);
 403                                 goto stop;
 404                         }
 405
 406                         lock_page(page);
 407
 408                         if (unlikely(page->mapping != mapping)) {
 409 continue_unlock:
 410                                 unlock_page(page);
 411                                 continue;
 412                         }
 413                         if (!PageDirty(page)) {
 414                                 /* someone wrote it for us */
 415                                 goto continue_unlock;
 416                         }
 417
 418                         f2fs_wait_on_page_writeback(page, META, true, true);
 419
 420                         if (!clear_page_dirty_for_io(page))
 421                                 goto continue_unlock;
 422
 423                         if (__f2fs_write_meta_page(page, &wbc, io_type)) {
 424                                 unlock_page(page);
 425                                 break;
 426                         }
 427                         nwritten++;
 428                         prev = page->index;
 429                         if (unlikely(nwritten >= nr_to_write))
 430                                 break;
 431                 }
 432                 pagevec_release(&pvec);
 433                 cond_resched();
 434         }
 435 stop:
 436         if (nwritten)
 437                 f2fs_submit_merged_write(sbi, type);
 438
 439         blk_finish_plug(&plug);
 440
 441         return nwritten;
 442 }
 443
 444 static bool f2fs_dirty_meta_folio(struct address_space *mapping,
 445                 struct folio *folio)
 446 {
 447         trace_f2fs_set_page_dirty(&folio->page, META);
 448
 449         if (!folio_test_uptodate(folio))
 450                 folio_mark_uptodate(folio);
 451         if (!folio_test_dirty(folio)) {
 452                 filemap_dirty_folio(mapping, folio);
 453                 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
 454                 set_page_private_reference(&folio->page);
 455                 return true;
 456         }
 457         return false;
 458 }
 459
 460 const struct address_space_operations f2fs_meta_aops = {
 461         .writepage      = f2fs_write_meta_page,
 462         .writepages     = f2fs_write_meta_pages,
 463         .dirty_folio    = f2fs_dirty_meta_folio,
 464         .invalidate_folio = f2fs_invalidate_folio,
 465         .release_folio  = f2fs_release_folio,
 466         .migrate_folio  = filemap_migrate_folio,
 467 };
 468
 469 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 470                                                 unsigned int devidx, int type)
 471 {
 472         struct inode_management *im = &sbi->im[type];
 473         struct ino_entry *e = NULL, *new = NULL;
 474
 475         if (type == FLUSH_INO) {
 476                 rcu_read_lock();
 477                 e = radix_tree_lookup(&im->ino_root, ino);
 478                 rcu_read_unlock();
 479         }
 480
 481 retry:
 482         if (!e)
 483                 new = f2fs_kmem_cache_alloc(ino_entry_slab,
 484                                                 GFP_NOFS, true, NULL);
 485
 486         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 487
 488         spin_lock(&im->ino_lock);
 489         e = radix_tree_lookup(&im->ino_root, ino);
 490         if (!e) {
 491                 if (!new) {
 492                         spin_unlock(&im->ino_lock);
 493                         goto retry;
 494                 }
 495                 e = new;
 496                 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
 497                         f2fs_bug_on(sbi, 1);
 498
 499                 memset(e, 0, sizeof(struct ino_entry));
 500                 e->ino = ino;
 501
 502                 list_add_tail(&e->list, &im->ino_list);
 503                 if (type != ORPHAN_INO)
 504                         im->ino_num++;
 505         }
 506
 507         if (type == FLUSH_INO)
 508                 f2fs_set_bit(devidx, (char *)&e->dirty_device);
 509
 510         spin_unlock(&im->ino_lock);
 511         radix_tree_preload_end();
 512
 513         if (new && e != new)
 514                 kmem_cache_free(ino_entry_slab, new);
 515 }
 516
 517 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 518 {
 519         struct inode_management *im = &sbi->im[type];
 520         struct ino_entry *e;
 521
 522         spin_lock(&im->ino_lock);
 523         e = radix_tree_lookup(&im->ino_root, ino);
 524         if (e) {
 525                 list_del(&e->list);
 526                 radix_tree_delete(&im->ino_root, ino);
 527                 im->ino_num--;
 528                 spin_unlock(&im->ino_lock);
 529                 kmem_cache_free(ino_entry_slab, e);
 530                 return;
 531         }
 532         spin_unlock(&im->ino_lock);
 533 }
 534
 535 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 536 {
 537         /* add new dirty ino entry into list */
 538         __add_ino_entry(sbi, ino, 0, type);
 539 }
 540
 541 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 542 {
 543         /* remove dirty ino entry from list */
 544         __remove_ino_entry(sbi, ino, type);
 545 }
 546
 547 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
 548 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 549 {
 550         struct inode_management *im = &sbi->im[mode];
 551         struct ino_entry *e;
 552
 553         spin_lock(&im->ino_lock);
 554         e = radix_tree_lookup(&im->ino_root, ino);
 555         spin_unlock(&im->ino_lock);
 556         return e ? true : false;
 557 }
 558
 559 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 560 {
 561         struct ino_entry *e, *tmp;
 562         int i;
 563
 564         for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
 565                 struct inode_management *im = &sbi->im[i];
 566
 567                 spin_lock(&im->ino_lock);
 568                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 569                         list_del(&e->list);
 570                         radix_tree_delete(&im->ino_root, e->ino);
 571                         kmem_cache_free(ino_entry_slab, e);
 572                         im->ino_num--;
 573                 }
 574                 spin_unlock(&im->ino_lock);
 575         }
 576 }
 577
 578 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 579                                         unsigned int devidx, int type)
 580 {
 581         __add_ino_entry(sbi, ino, devidx, type);
 582 }
 583
 584 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 585                                         unsigned int devidx, int type)
 586 {
 587         struct inode_management *im = &sbi->im[type];
 588         struct ino_entry *e;
 589         bool is_dirty = false;
 590
 591         spin_lock(&im->ino_lock);
 592         e = radix_tree_lookup(&im->ino_root, ino);
 593         if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
 594                 is_dirty = true;
 595         spin_unlock(&im->ino_lock);
 596         return is_dirty;
 597 }
 598
 599 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
 600 {
 601         struct inode_management *im = &sbi->im[ORPHAN_INO];
 602         int err = 0;
 603
 604         spin_lock(&im->ino_lock);
 605
 606         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 607                 spin_unlock(&im->ino_lock);
 608                 f2fs_show_injection_info(sbi, FAULT_ORPHAN);
 609                 return -ENOSPC;
 610         }
 611
 612         if (unlikely(im->ino_num >= sbi->max_orphans))
 613                 err = -ENOSPC;
 614         else
 615                 im->ino_num++;
 616         spin_unlock(&im->ino_lock);
 617
 618         return err;
 619 }
 620
 621 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
 622 {
 623         struct inode_management *im = &sbi->im[ORPHAN_INO];
 624
 625         spin_lock(&im->ino_lock);
 626         f2fs_bug_on(sbi, im->ino_num == 0);
 627         im->ino_num--;
 628         spin_unlock(&im->ino_lock);
 629 }
 630
 631 void f2fs_add_orphan_inode(struct inode *inode)
 632 {
 633         /* add new orphan ino entry into list */
 634         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
 635         f2fs_update_inode_page(inode);
 636 }
 637
 638 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 639 {
 640         /* remove orphan entry from orphan list */
 641         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 642 }
 643
 644 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 645 {
 646         struct inode *inode;
 647         struct node_info ni;
 648         int err;
 649
 650         inode = f2fs_iget_retry(sbi->sb, ino);
 651         if (IS_ERR(inode)) {
 652                 /*
 653                  * there should be a bug that we can't find the entry
 654                  * to orphan inode.
 655                  */
 656                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 657                 return PTR_ERR(inode);
 658         }
 659
 660         err = f2fs_dquot_initialize(inode);
 661         if (err) {
 662                 iput(inode);
 663                 goto err_out;
 664         }
 665
 666         clear_nlink(inode);
 667
 668         /* truncate all the data during iput */
 669         iput(inode);
 670
 671         err = f2fs_get_node_info(sbi, ino, &ni, false);
 672         if (err)
 673                 goto err_out;
 674
 675         /* ENOMEM was fully retried in f2fs_evict_inode. */
 676         if (ni.blk_addr != NULL_ADDR) {
 677                 err = -EIO;
 678                 goto err_out;
 679         }
 680         return 0;
 681
 682 err_out:
 683         set_sbi_flag(sbi, SBI_NEED_FSCK);
 684         f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
 685                   __func__, ino);
 686         return err;
 687 }
 688
 689 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
 690 {
 691         block_t start_blk, orphan_blocks, i, j;
 692         unsigned int s_flags = sbi->sb->s_flags;
 693         int err = 0;
 694 #ifdef CONFIG_QUOTA
 695         int quota_enabled;
 696 #endif
 697
 698         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 699                 return 0;
 700
 701         if (bdev_read_only(sbi->sb->s_bdev)) {
 702                 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
 703                 return 0;
 704         }
 705
 706         if (s_flags & SB_RDONLY) {
 707                 f2fs_info(sbi, "orphan cleanup on readonly fs");
 708                 sbi->sb->s_flags &= ~SB_RDONLY;
 709         }
 710
 711 #ifdef CONFIG_QUOTA
 712         /*
 713          * Turn on quotas which were not enabled for read-only mounts if
 714          * filesystem has quota feature, so that they are updated correctly.
 715          */
 716         quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
 717 #endif
 718
 719         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 720         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 721
 722         f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 723
 724         for (i = 0; i < orphan_blocks; i++) {
 725                 struct page *page;
 726                 struct f2fs_orphan_block *orphan_blk;
 727
 728                 page = f2fs_get_meta_page(sbi, start_blk + i);
 729                 if (IS_ERR(page)) {
 730                         err = PTR_ERR(page);
 731                         goto out;
 732                 }
 733
 734                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 735                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 736                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 737
 738                         err = recover_orphan_inode(sbi, ino);
 739                         if (err) {
 740                                 f2fs_put_page(page, 1);
 741                                 goto out;
 742                         }
 743                 }
 744                 f2fs_put_page(page, 1);
 745         }
 746         /* clear Orphan Flag */
 747         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 748 out:
 749         set_sbi_flag(sbi, SBI_IS_RECOVERED);
 750
 751 #ifdef CONFIG_QUOTA
 752         /* Turn quotas off */
 753         if (quota_enabled)
 754                 f2fs_quota_off_umount(sbi->sb);
 755 #endif
 756         sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
 757
 758         return err;
 759 }
 760
 761 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 762 {
 763         struct list_head *head;
 764         struct f2fs_orphan_block *orphan_blk = NULL;
 765         unsigned int nentries = 0;
 766         unsigned short index = 1;
 767         unsigned short orphan_blocks;
 768         struct page *page = NULL;
 769         struct ino_entry *orphan = NULL;
 770         struct inode_management *im = &sbi->im[ORPHAN_INO];
 771
 772         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 773
 774         /*
 775          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 776          * orphan inode operations are covered under f2fs_lock_op().
 777          * And, spin_lock should be avoided due to page operations below.
 778          */
 779         head = &im->ino_list;
 780
 781         /* loop for each orphan inode entry and write them in Jornal block */
 782         list_for_each_entry(orphan, head, list) {
 783                 if (!page) {
 784                         page = f2fs_grab_meta_page(sbi, start_blk++);
 785                         orphan_blk =
 786                                 (struct f2fs_orphan_block *)page_address(page);
 787                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 788                 }
 789
 790                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 791
 792                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 793                         /*
 794                          * an orphan block is full of 1020 entries,
 795                          * then we need to flush current orphan blocks
 796                          * and bring another one in memory
 797                          */
 798                         orphan_blk->blk_addr = cpu_to_le16(index);
 799                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 800                         orphan_blk->entry_count = cpu_to_le32(nentries);
 801                         set_page_dirty(page);
 802                         f2fs_put_page(page, 1);
 803                         index++;
 804                         nentries = 0;
 805                         page = NULL;
 806                 }
 807         }
 808
 809         if (page) {
 810                 orphan_blk->blk_addr = cpu_to_le16(index);
 811                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 812                 orphan_blk->entry_count = cpu_to_le32(nentries);
 813                 set_page_dirty(page);
 814                 f2fs_put_page(page, 1);
 815         }
 816 }
 817
 818 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
 819                                                 struct f2fs_checkpoint *ckpt)
 820 {
 821         unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
 822         __u32 chksum;
 823
 824         chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
 825         if (chksum_ofs < CP_CHKSUM_OFFSET) {
 826                 chksum_ofs += sizeof(chksum);
 827                 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
 828                                                 F2FS_BLKSIZE - chksum_ofs);
 829         }
 830         return chksum;
 831 }
 832
 833 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 834                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 835                 unsigned long long *version)
 836 {
 837         size_t crc_offset = 0;
 838         __u32 crc;
 839
 840         *cp_page = f2fs_get_meta_page(sbi, cp_addr);
 841         if (IS_ERR(*cp_page))
 842                 return PTR_ERR(*cp_page);
 843
 844         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 845
 846         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 847         if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
 848                         crc_offset > CP_CHKSUM_OFFSET) {
 849                 f2fs_put_page(*cp_page, 1);
 850                 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
 851                 return -EINVAL;
 852         }
 853
 854         crc = f2fs_checkpoint_chksum(sbi, *cp_block);
 855         if (crc != cur_cp_crc(*cp_block)) {
 856                 f2fs_put_page(*cp_page, 1);
 857                 f2fs_warn(sbi, "invalid crc value");
 858                 return -EINVAL;
 859         }
 860
 861         *version = cur_cp_version(*cp_block);
 862         return 0;
 863 }
 864
 865 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 866                                 block_t cp_addr, unsigned long long *version)
 867 {
 868         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 869         struct f2fs_checkpoint *cp_block = NULL;
 870         unsigned long long cur_version = 0, pre_version = 0;
 871         unsigned int cp_blocks;
 872         int err;
 873
 874         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 875                                         &cp_page_1, version);
 876         if (err)
 877                 return NULL;
 878
 879         cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
 880
 881         if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
 882                 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
 883                           le32_to_cpu(cp_block->cp_pack_total_block_count));
 884                 goto invalid_cp;
 885         }
 886         pre_version = *version;
 887
 888         cp_addr += cp_blocks - 1;
 889         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 890                                         &cp_page_2, version);
 891         if (err)
 892                 goto invalid_cp;
 893         cur_version = *version;
 894
 895         if (cur_version == pre_version) {
 896                 *version = cur_version;
 897                 f2fs_put_page(cp_page_2, 1);
 898                 return cp_page_1;
 899         }
 900         f2fs_put_page(cp_page_2, 1);
 901 invalid_cp:
 902         f2fs_put_page(cp_page_1, 1);
 903         return NULL;
 904 }
 905
 906 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
 907 {
 908         struct f2fs_checkpoint *cp_block;
 909         struct f2fs_super_block *fsb = sbi->raw_super;
 910         struct page *cp1, *cp2, *cur_page;
 911         unsigned long blk_size = sbi->blocksize;
 912         unsigned long long cp1_version = 0, cp2_version = 0;
 913         unsigned long long cp_start_blk_no;
 914         unsigned int cp_blks = 1 + __cp_payload(sbi);
 915         block_t cp_blk_no;
 916         int i;
 917         int err;
 918
 919         sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
 920                                   GFP_KERNEL);
 921         if (!sbi->ckpt)
 922                 return -ENOMEM;
 923         /*
 924          * Finding out valid cp block involves read both
 925          * sets( cp pack 1 and cp pack 2)
 926          */
 927         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 928         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 929
 930         /* The second checkpoint pack should start at the next segment */
 931         cp_start_blk_no += ((unsigned long long)1) <<
 932                                 le32_to_cpu(fsb->log_blocks_per_seg);
 933         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 934
 935         if (cp1 && cp2) {
 936                 if (ver_after(cp2_version, cp1_version))
 937                         cur_page = cp2;
 938                 else
 939                         cur_page = cp1;
 940         } else if (cp1) {
 941                 cur_page = cp1;
 942         } else if (cp2) {
 943                 cur_page = cp2;
 944         } else {
 945                 err = -EFSCORRUPTED;
 946                 goto fail_no_cp;
 947         }
 948
 949         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 950         memcpy(sbi->ckpt, cp_block, blk_size);
 951
 952         if (cur_page == cp1)
 953                 sbi->cur_cp_pack = 1;
 954         else
 955                 sbi->cur_cp_pack = 2;
 956
 957         /* Sanity checking of checkpoint */
 958         if (f2fs_sanity_check_ckpt(sbi)) {
 959                 err = -EFSCORRUPTED;
 960                 goto free_fail_no_cp;
 961         }
 962
 963         if (cp_blks <= 1)
 964                 goto done;
 965
 966         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 967         if (cur_page == cp2)
 968                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 969
 970         for (i = 1; i < cp_blks; i++) {
 971                 void *sit_bitmap_ptr;
 972                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 973
 974                 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
 975                 if (IS_ERR(cur_page)) {
 976                         err = PTR_ERR(cur_page);
 977                         goto free_fail_no_cp;
 978                 }
 979                 sit_bitmap_ptr = page_address(cur_page);
 980                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 981                 f2fs_put_page(cur_page, 1);
 982         }
 983 done:
 984         f2fs_put_page(cp1, 1);
 985         f2fs_put_page(cp2, 1);
 986         return 0;
 987
 988 free_fail_no_cp:
 989         f2fs_put_page(cp1, 1);
 990         f2fs_put_page(cp2, 1);
 991 fail_no_cp:
 992         kvfree(sbi->ckpt);
 993         return err;
 994 }
 995
 996 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 997 {
 998         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 999         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1000
1001         if (is_inode_flag_set(inode, flag))
1002                 return;
1003
1004         set_inode_flag(inode, flag);
1005         list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1006         stat_inc_dirty_inode(sbi, type);
1007 }
1008
1009 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1010 {
1011         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1012
1013         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1014                 return;
1015
1016         list_del_init(&F2FS_I(inode)->dirty_list);
1017         clear_inode_flag(inode, flag);
1018         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1019 }
1020
1021 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1022 {
1023         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1024         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1025
1026         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1027                         !S_ISLNK(inode->i_mode))
1028                 return;
1029
1030         spin_lock(&sbi->inode_lock[type]);
1031         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1032                 __add_dirty_inode(inode, type);
1033         inode_inc_dirty_pages(inode);
1034         spin_unlock(&sbi->inode_lock[type]);
1035
1036         set_page_private_reference(&folio->page);
1037 }
1038
1039 void f2fs_remove_dirty_inode(struct inode *inode)
1040 {
1041         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1042         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1043
1044         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1045                         !S_ISLNK(inode->i_mode))
1046                 return;
1047
1048         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1049                 return;
1050
1051         spin_lock(&sbi->inode_lock[type]);
1052         __remove_dirty_inode(inode, type);
1053         spin_unlock(&sbi->inode_lock[type]);
1054 }
1055
1056 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1057 {
1058         struct list_head *head;
1059         struct inode *inode;
1060         struct f2fs_inode_info *fi;
1061         bool is_dir = (type == DIR_INODE);
1062         unsigned long ino = 0;
1063
1064         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1065                                 get_pages(sbi, is_dir ?
1066                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1067 retry:
1068         if (unlikely(f2fs_cp_error(sbi))) {
1069                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1070                                 get_pages(sbi, is_dir ?
1071                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1072                 return -EIO;
1073         }
1074
1075         spin_lock(&sbi->inode_lock[type]);
1076
1077         head = &sbi->inode_list[type];
1078         if (list_empty(head)) {
1079                 spin_unlock(&sbi->inode_lock[type]);
1080                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1081                                 get_pages(sbi, is_dir ?
1082                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1083                 return 0;
1084         }
1085         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1086         inode = igrab(&fi->vfs_inode);
1087         spin_unlock(&sbi->inode_lock[type]);
1088         if (inode) {
1089                 unsigned long cur_ino = inode->i_ino;
1090
1091                 F2FS_I(inode)->cp_task = current;
1092
1093                 filemap_fdatawrite(inode->i_mapping);
1094
1095                 F2FS_I(inode)->cp_task = NULL;
1096
1097                 iput(inode);
1098                 /* We need to give cpu to another writers. */
1099                 if (ino == cur_ino)
1100                         cond_resched();
1101                 else
1102                         ino = cur_ino;
1103         } else {
1104                 /*
1105                  * We should submit bio, since it exists several
1106                  * wribacking dentry pages in the freeing inode.
1107                  */
1108                 f2fs_submit_merged_write(sbi, DATA);
1109                 cond_resched();
1110         }
1111         goto retry;
1112 }
1113
1114 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1115 {
1116         struct list_head *head = &sbi->inode_list[DIRTY_META];
1117         struct inode *inode;
1118         struct f2fs_inode_info *fi;
1119         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1120
1121         while (total--) {
1122                 if (unlikely(f2fs_cp_error(sbi)))
1123                         return -EIO;
1124
1125                 spin_lock(&sbi->inode_lock[DIRTY_META]);
1126                 if (list_empty(head)) {
1127                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1128                         return 0;
1129                 }
1130                 fi = list_first_entry(head, struct f2fs_inode_info,
1131                                                         gdirty_list);
1132                 inode = igrab(&fi->vfs_inode);
1133                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1134                 if (inode) {
1135                         sync_inode_metadata(inode, 0);
1136
1137                         /* it's on eviction */
1138                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1139                                 f2fs_update_inode_page(inode);
1140                         iput(inode);
1141                 }
1142         }
1143         return 0;
1144 }
1145
1146 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1147 {
1148         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1149         struct f2fs_nm_info *nm_i = NM_I(sbi);
1150         nid_t last_nid = nm_i->next_scan_nid;
1151
1152         next_free_nid(sbi, &last_nid);
1153         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1154         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1155         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1156         ckpt->next_free_nid = cpu_to_le32(last_nid);
1157 }
1158
1159 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1160 {
1161         bool ret = false;
1162
1163         if (!is_journalled_quota(sbi))
1164                 return false;
1165
1166         if (!f2fs_down_write_trylock(&sbi->quota_sem))
1167                 return true;
1168         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1169                 ret = false;
1170         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1171                 ret = false;
1172         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1173                 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1174                 ret = true;
1175         } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1176                 ret = true;
1177         }
1178         f2fs_up_write(&sbi->quota_sem);
1179         return ret;
1180 }
1181
1182 /*
1183  * Freeze all the FS-operations for checkpoint.
1184  */
1185 static int block_operations(struct f2fs_sb_info *sbi)
1186 {
1187         struct writeback_control wbc = {
1188                 .sync_mode = WB_SYNC_ALL,
1189                 .nr_to_write = LONG_MAX,
1190                 .for_reclaim = 0,
1191         };
1192         int err = 0, cnt = 0;
1193
1194         /*
1195          * Let's flush inline_data in dirty node pages.
1196          */
1197         f2fs_flush_inline_data(sbi);
1198
1199 retry_flush_quotas:
1200         f2fs_lock_all(sbi);
1201         if (__need_flush_quota(sbi)) {
1202                 int locked;
1203
1204                 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1205                         set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1206                         set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1207                         goto retry_flush_dents;
1208                 }
1209                 f2fs_unlock_all(sbi);
1210
1211                 /* only failed during mount/umount/freeze/quotactl */
1212                 locked = down_read_trylock(&sbi->sb->s_umount);
1213                 f2fs_quota_sync(sbi->sb, -1);
1214                 if (locked)
1215                         up_read(&sbi->sb->s_umount);
1216                 cond_resched();
1217                 goto retry_flush_quotas;
1218         }
1219
1220 retry_flush_dents:
1221         /* write all the dirty dentry pages */
1222         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1223                 f2fs_unlock_all(sbi);
1224                 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1225                 if (err)
1226                         return err;
1227                 cond_resched();
1228                 goto retry_flush_quotas;
1229         }
1230
1231         /*
1232          * POR: we should ensure that there are no dirty node pages
1233          * until finishing nat/sit flush. inode->i_blocks can be updated.
1234          */
1235         f2fs_down_write(&sbi->node_change);
1236
1237         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1238                 f2fs_up_write(&sbi->node_change);
1239                 f2fs_unlock_all(sbi);
1240                 err = f2fs_sync_inode_meta(sbi);
1241                 if (err)
1242                         return err;
1243                 cond_resched();
1244                 goto retry_flush_quotas;
1245         }
1246
1247 retry_flush_nodes:
1248         f2fs_down_write(&sbi->node_write);
1249
1250         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1251                 f2fs_up_write(&sbi->node_write);
1252                 atomic_inc(&sbi->wb_sync_req[NODE]);
1253                 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1254                 atomic_dec(&sbi->wb_sync_req[NODE]);
1255                 if (err) {
1256                         f2fs_up_write(&sbi->node_change);
1257                         f2fs_unlock_all(sbi);
1258                         return err;
1259                 }
1260                 cond_resched();
1261                 goto retry_flush_nodes;
1262         }
1263
1264         /*
1265          * sbi->node_change is used only for AIO write_begin path which produces
1266          * dirty node blocks and some checkpoint values by block allocation.
1267          */
1268         __prepare_cp_block(sbi);
1269         f2fs_up_write(&sbi->node_change);
1270         return err;
1271 }
1272
1273 static void unblock_operations(struct f2fs_sb_info *sbi)
1274 {
1275         f2fs_up_write(&sbi->node_write);
1276         f2fs_unlock_all(sbi);
1277 }
1278
1279 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1280 {
1281         DEFINE_WAIT(wait);
1282
1283         for (;;) {
1284                 if (!get_pages(sbi, type))
1285                         break;
1286
1287                 if (unlikely(f2fs_cp_error(sbi)))
1288                         break;
1289
1290                 if (type == F2FS_DIRTY_META)
1291                         f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1292                                                         FS_CP_META_IO);
1293                 else if (type == F2FS_WB_CP_DATA)
1294                         f2fs_submit_merged_write(sbi, DATA);
1295
1296                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1297                 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1298         }
1299         finish_wait(&sbi->cp_wait, &wait);
1300 }
1301
1302 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1303 {
1304         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1305         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1306         unsigned long flags;
1307
1308         if (cpc->reason & CP_UMOUNT) {
1309                 if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1310                         NM_I(sbi)->nat_bits_blocks > sbi->blocks_per_seg) {
1311                         clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1312                         f2fs_notice(sbi, "Disable nat_bits due to no space");
1313                 } else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1314                                                 f2fs_nat_bitmap_enabled(sbi)) {
1315                         f2fs_enable_nat_bits(sbi);
1316                         set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1317                         f2fs_notice(sbi, "Rebuild and enable nat_bits");
1318                 }
1319         }
1320
1321         spin_lock_irqsave(&sbi->cp_lock, flags);
1322
1323         if (cpc->reason & CP_TRIMMED)
1324                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1325         else
1326                 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1327
1328         if (cpc->reason & CP_UMOUNT)
1329                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1330         else
1331                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1332
1333         if (cpc->reason & CP_FASTBOOT)
1334                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1335         else
1336                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1337
1338         if (orphan_num)
1339                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1340         else
1341                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1342
1343         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1344                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1345
1346         if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1347                 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1348         else
1349                 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1350
1351         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1352                 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1353         else
1354                 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1355
1356         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1357                 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1358         else
1359                 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1360
1361         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1362                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1363         else
1364                 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1365
1366         if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1367                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1368
1369         /* set this flag to activate crc|cp_ver for recovery */
1370         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1371         __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1372
1373         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1374 }
1375
1376 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1377         void *src, block_t blk_addr)
1378 {
1379         struct writeback_control wbc = {
1380                 .for_reclaim = 0,
1381         };
1382
1383         /*
1384          * pagevec_lookup_tag and lock_page again will take
1385          * some extra time. Therefore, f2fs_update_meta_pages and
1386          * f2fs_sync_meta_pages are combined in this function.
1387          */
1388         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1389         int err;
1390
1391         f2fs_wait_on_page_writeback(page, META, true, true);
1392
1393         memcpy(page_address(page), src, PAGE_SIZE);
1394
1395         set_page_dirty(page);
1396         if (unlikely(!clear_page_dirty_for_io(page)))
1397                 f2fs_bug_on(sbi, 1);
1398
1399         /* writeout cp pack 2 page */
1400         err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1401         if (unlikely(err && f2fs_cp_error(sbi))) {
1402                 f2fs_put_page(page, 1);
1403                 return;
1404         }
1405
1406         f2fs_bug_on(sbi, err);
1407         f2fs_put_page(page, 0);
1408
1409         /* submit checkpoint (with barrier if NOBARRIER is not set) */
1410         f2fs_submit_merged_write(sbi, META_FLUSH);
1411 }
1412
1413 static inline u64 get_sectors_written(struct block_device *bdev)
1414 {
1415         return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1416 }
1417
1418 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1419 {
1420         if (f2fs_is_multi_device(sbi)) {
1421                 u64 sectors = 0;
1422                 int i;
1423
1424                 for (i = 0; i < sbi->s_ndevs; i++)
1425                         sectors += get_sectors_written(FDEV(i).bdev);
1426
1427                 return sectors;
1428         }
1429
1430         return get_sectors_written(sbi->sb->s_bdev);
1431 }
1432
1433 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1434 {
1435         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1436         struct f2fs_nm_info *nm_i = NM_I(sbi);
1437         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1438         block_t start_blk;
1439         unsigned int data_sum_blocks, orphan_blocks;
1440         __u32 crc32 = 0;
1441         int i;
1442         int cp_payload_blks = __cp_payload(sbi);
1443         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1444         u64 kbytes_written;
1445         int err;
1446
1447         /* Flush all the NAT/SIT pages */
1448         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1449
1450         /* start to update checkpoint, cp ver is already updated previously */
1451         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1452         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1453         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1454                 ckpt->cur_node_segno[i] =
1455                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1456                 ckpt->cur_node_blkoff[i] =
1457                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1458                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1459                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1460         }
1461         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1462                 ckpt->cur_data_segno[i] =
1463                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1464                 ckpt->cur_data_blkoff[i] =
1465                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1466                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1467                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1468         }
1469
1470         /* 2 cp + n data seg summary + orphan inode blocks */
1471         data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1472         spin_lock_irqsave(&sbi->cp_lock, flags);
1473         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1474                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1475         else
1476                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1477         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1478
1479         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1480         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1481                         orphan_blocks);
1482
1483         if (__remain_node_summaries(cpc->reason))
1484                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1485                                 cp_payload_blks + data_sum_blocks +
1486                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1487         else
1488                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1489                                 cp_payload_blks + data_sum_blocks +
1490                                 orphan_blocks);
1491
1492         /* update ckpt flag for checkpoint */
1493         update_ckpt_flags(sbi, cpc);
1494
1495         /* update SIT/NAT bitmap */
1496         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1497         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1498
1499         crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1500         *((__le32 *)((unsigned char *)ckpt +
1501                                 le32_to_cpu(ckpt->checksum_offset)))
1502                                 = cpu_to_le32(crc32);
1503
1504         start_blk = __start_cp_next_addr(sbi);
1505
1506         /* write nat bits */
1507         if ((cpc->reason & CP_UMOUNT) &&
1508                         is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1509                 __u64 cp_ver = cur_cp_version(ckpt);
1510                 block_t blk;
1511
1512                 cp_ver |= ((__u64)crc32 << 32);
1513                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1514
1515                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1516                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1517                         f2fs_update_meta_page(sbi, nm_i->nat_bits +
1518                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1519         }
1520
1521         /* write out checkpoint buffer at block 0 */
1522         f2fs_update_meta_page(sbi, ckpt, start_blk++);
1523
1524         for (i = 1; i < 1 + cp_payload_blks; i++)
1525                 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1526                                                         start_blk++);
1527
1528         if (orphan_num) {
1529                 write_orphan_inodes(sbi, start_blk);
1530                 start_blk += orphan_blocks;
1531         }
1532
1533         f2fs_write_data_summaries(sbi, start_blk);
1534         start_blk += data_sum_blocks;
1535
1536         /* Record write statistics in the hot node summary */
1537         kbytes_written = sbi->kbytes_written;
1538         kbytes_written += (f2fs_get_sectors_written(sbi) -
1539                                 sbi->sectors_written_start) >> 1;
1540         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1541
1542         if (__remain_node_summaries(cpc->reason)) {
1543                 f2fs_write_node_summaries(sbi, start_blk);
1544                 start_blk += NR_CURSEG_NODE_TYPE;
1545         }
1546
1547         /* update user_block_counts */
1548         sbi->last_valid_block_count = sbi->total_valid_block_count;
1549         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1550         percpu_counter_set(&sbi->rf_node_block_count, 0);
1551
1552         /* Here, we have one bio having CP pack except cp pack 2 page */
1553         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1554         /* Wait for all dirty meta pages to be submitted for IO */
1555         f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1556
1557         /* wait for previous submitted meta pages writeback */
1558         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1559
1560         /* flush all device cache */
1561         err = f2fs_flush_device_cache(sbi);
1562         if (err)
1563                 return err;
1564
1565         /* barrier and flush checkpoint cp pack 2 page if it can */
1566         commit_checkpoint(sbi, ckpt, start_blk);
1567         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1568
1569         /*
1570          * invalidate intermediate page cache borrowed from meta inode which are
1571          * used for migration of encrypted, verity or compressed inode's blocks.
1572          */
1573         if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1574                 f2fs_sb_has_compression(sbi))
1575                 invalidate_mapping_pages(META_MAPPING(sbi),
1576                                 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1577
1578         f2fs_release_ino_entry(sbi, false);
1579
1580         f2fs_reset_fsync_node_info(sbi);
1581
1582         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1583         clear_sbi_flag(sbi, SBI_NEED_CP);
1584         clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1585
1586         spin_lock(&sbi->stat_lock);
1587         sbi->unusable_block_count = 0;
1588         spin_unlock(&sbi->stat_lock);
1589
1590         __set_cp_next_pack(sbi);
1591
1592         /*
1593          * redirty superblock if metadata like node page or inode cache is
1594          * updated during writing checkpoint.
1595          */
1596         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1597                         get_pages(sbi, F2FS_DIRTY_IMETA))
1598                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1599
1600         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1601
1602         return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1603 }
1604
1605 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1606 {
1607         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1608         unsigned long long ckpt_ver;
1609         int err = 0;
1610
1611         if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1612                 return -EROFS;
1613
1614         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1615                 if (cpc->reason != CP_PAUSE)
1616                         return 0;
1617                 f2fs_warn(sbi, "Start checkpoint disabled!");
1618         }
1619         if (cpc->reason != CP_RESIZE)
1620                 f2fs_down_write(&sbi->cp_global_sem);
1621
1622         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1623                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1624                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1625                 goto out;
1626         if (unlikely(f2fs_cp_error(sbi))) {
1627                 err = -EIO;
1628                 goto out;
1629         }
1630
1631         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1632
1633         err = block_operations(sbi);
1634         if (err)
1635                 goto out;
1636
1637         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1638
1639         f2fs_flush_merged_writes(sbi);
1640
1641         /* this is the case of multiple fstrims without any changes */
1642         if (cpc->reason & CP_DISCARD) {
1643                 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1644                         unblock_operations(sbi);
1645                         goto out;
1646                 }
1647
1648                 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1649                                 SIT_I(sbi)->dirty_sentries == 0 &&
1650                                 prefree_segments(sbi) == 0) {
1651                         f2fs_flush_sit_entries(sbi, cpc);
1652                         f2fs_clear_prefree_segments(sbi, cpc);
1653                         unblock_operations(sbi);
1654                         goto out;
1655                 }
1656         }
1657
1658         /*
1659          * update checkpoint pack index
1660          * Increase the version number so that
1661          * SIT entries and seg summaries are written at correct place
1662          */
1663         ckpt_ver = cur_cp_version(ckpt);
1664         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1665
1666         /* write cached NAT/SIT entries to NAT/SIT area */
1667         err = f2fs_flush_nat_entries(sbi, cpc);
1668         if (err) {
1669                 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1670                 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1671                 goto stop;
1672         }
1673
1674         f2fs_flush_sit_entries(sbi, cpc);
1675
1676         /* save inmem log status */
1677         f2fs_save_inmem_curseg(sbi);
1678
1679         err = do_checkpoint(sbi, cpc);
1680         if (err) {
1681                 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1682                 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1683                 f2fs_release_discard_addrs(sbi);
1684         } else {
1685                 f2fs_clear_prefree_segments(sbi, cpc);
1686         }
1687
1688         f2fs_restore_inmem_curseg(sbi);
1689 stop:
1690         unblock_operations(sbi);
1691         stat_inc_cp_count(sbi->stat_info);
1692
1693         if (cpc->reason & CP_RECOVERY)
1694                 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1695
1696         /* update CP_TIME to trigger checkpoint periodically */
1697         f2fs_update_time(sbi, CP_TIME);
1698         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1699 out:
1700         if (cpc->reason != CP_RESIZE)
1701                 f2fs_up_write(&sbi->cp_global_sem);
1702         return err;
1703 }
1704
1705 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1706 {
1707         int i;
1708
1709         for (i = 0; i < MAX_INO_ENTRY; i++) {
1710                 struct inode_management *im = &sbi->im[i];
1711
1712                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1713                 spin_lock_init(&im->ino_lock);
1714                 INIT_LIST_HEAD(&im->ino_list);
1715                 im->ino_num = 0;
1716         }
1717
1718         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1719                         NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1720                                 F2FS_ORPHANS_PER_BLOCK;
1721 }
1722
1723 int __init f2fs_create_checkpoint_caches(void)
1724 {
1725         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1726                         sizeof(struct ino_entry));
1727         if (!ino_entry_slab)
1728                 return -ENOMEM;
1729         f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1730                         sizeof(struct inode_entry));
1731         if (!f2fs_inode_entry_slab) {
1732                 kmem_cache_destroy(ino_entry_slab);
1733                 return -ENOMEM;
1734         }
1735         return 0;
1736 }
1737
1738 void f2fs_destroy_checkpoint_caches(void)
1739 {
1740         kmem_cache_destroy(ino_entry_slab);
1741         kmem_cache_destroy(f2fs_inode_entry_slab);
1742 }
1743
1744 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1745 {
1746         struct cp_control cpc = { .reason = CP_SYNC, };
1747         int err;
1748
1749         f2fs_down_write(&sbi->gc_lock);
1750         err = f2fs_write_checkpoint(sbi, &cpc);
1751         f2fs_up_write(&sbi->gc_lock);
1752
1753         return err;
1754 }
1755
1756 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1757 {
1758         struct ckpt_req_control *cprc = &sbi->cprc_info;
1759         struct ckpt_req *req, *next;
1760         struct llist_node *dispatch_list;
1761         u64 sum_diff = 0, diff, count = 0;
1762         int ret;
1763
1764         dispatch_list = llist_del_all(&cprc->issue_list);
1765         if (!dispatch_list)
1766                 return;
1767         dispatch_list = llist_reverse_order(dispatch_list);
1768
1769         ret = __write_checkpoint_sync(sbi);
1770         atomic_inc(&cprc->issued_ckpt);
1771
1772         llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1773                 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1774                 req->ret = ret;
1775                 complete(&req->wait);
1776
1777                 sum_diff += diff;
1778                 count++;
1779         }
1780         atomic_sub(count, &cprc->queued_ckpt);
1781         atomic_add(count, &cprc->total_ckpt);
1782
1783         spin_lock(&cprc->stat_lock);
1784         cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1785         if (cprc->peak_time < cprc->cur_time)
1786                 cprc->peak_time = cprc->cur_time;
1787         spin_unlock(&cprc->stat_lock);
1788 }
1789
1790 static int issue_checkpoint_thread(void *data)
1791 {
1792         struct f2fs_sb_info *sbi = data;
1793         struct ckpt_req_control *cprc = &sbi->cprc_info;
1794         wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1795 repeat:
1796         if (kthread_should_stop())
1797                 return 0;
1798
1799         if (!llist_empty(&cprc->issue_list))
1800                 __checkpoint_and_complete_reqs(sbi);
1801
1802         wait_event_interruptible(*q,
1803                 kthread_should_stop() || !llist_empty(&cprc->issue_list));
1804         goto repeat;
1805 }
1806
1807 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1808                 struct ckpt_req *wait_req)
1809 {
1810         struct ckpt_req_control *cprc = &sbi->cprc_info;
1811
1812         if (!llist_empty(&cprc->issue_list)) {
1813                 __checkpoint_and_complete_reqs(sbi);
1814         } else {
1815                 /* already dispatched by issue_checkpoint_thread */
1816                 if (wait_req)
1817                         wait_for_completion(&wait_req->wait);
1818         }
1819 }
1820
1821 static void init_ckpt_req(struct ckpt_req *req)
1822 {
1823         memset(req, 0, sizeof(struct ckpt_req));
1824
1825         init_completion(&req->wait);
1826         req->queue_time = ktime_get();
1827 }
1828
1829 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1830 {
1831         struct ckpt_req_control *cprc = &sbi->cprc_info;
1832         struct ckpt_req req;
1833         struct cp_control cpc;
1834
1835         cpc.reason = __get_cp_reason(sbi);
1836         if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1837                 int ret;
1838
1839                 f2fs_down_write(&sbi->gc_lock);
1840                 ret = f2fs_write_checkpoint(sbi, &cpc);
1841                 f2fs_up_write(&sbi->gc_lock);
1842
1843                 return ret;
1844         }
1845
1846         if (!cprc->f2fs_issue_ckpt)
1847                 return __write_checkpoint_sync(sbi);
1848
1849         init_ckpt_req(&req);
1850
1851         llist_add(&req.llnode, &cprc->issue_list);
1852         atomic_inc(&cprc->queued_ckpt);
1853
1854         /*
1855          * update issue_list before we wake up issue_checkpoint thread,
1856          * this smp_mb() pairs with another barrier in ___wait_event(),
1857          * see more details in comments of waitqueue_active().
1858          */
1859         smp_mb();
1860
1861         if (waitqueue_active(&cprc->ckpt_wait_queue))
1862                 wake_up(&cprc->ckpt_wait_queue);
1863
1864         if (cprc->f2fs_issue_ckpt)
1865                 wait_for_completion(&req.wait);
1866         else
1867                 flush_remained_ckpt_reqs(sbi, &req);
1868
1869         return req.ret;
1870 }
1871
1872 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1873 {
1874         dev_t dev = sbi->sb->s_bdev->bd_dev;
1875         struct ckpt_req_control *cprc = &sbi->cprc_info;
1876
1877         if (cprc->f2fs_issue_ckpt)
1878                 return 0;
1879
1880         cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1881                         "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1882         if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1883                 cprc->f2fs_issue_ckpt = NULL;
1884                 return -ENOMEM;
1885         }
1886
1887         set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1888
1889         return 0;
1890 }
1891
1892 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1893 {
1894         struct ckpt_req_control *cprc = &sbi->cprc_info;
1895
1896         if (cprc->f2fs_issue_ckpt) {
1897                 struct task_struct *ckpt_task = cprc->f2fs_issue_ckpt;
1898
1899                 cprc->f2fs_issue_ckpt = NULL;
1900                 kthread_stop(ckpt_task);
1901
1902                 flush_remained_ckpt_reqs(sbi, NULL);
1903         }
1904 }
1905
1906 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1907 {
1908         struct ckpt_req_control *cprc = &sbi->cprc_info;
1909
1910         atomic_set(&cprc->issued_ckpt, 0);
1911         atomic_set(&cprc->total_ckpt, 0);
1912         atomic_set(&cprc->queued_ckpt, 0);
1913         cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1914         init_waitqueue_head(&cprc->ckpt_wait_queue);
1915         init_llist_head(&cprc->issue_list);
1916         spin_lock_init(&cprc->stat_lock);
1917 }