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                                                 unsigned char reason)
  31 {
  32         f2fs_build_fault_attr(sbi, 0, 0);
  33         set_ckpt_flags(sbi, CP_ERROR_FLAG);
  34         if (!end_io) {
  35                 f2fs_flush_merged_writes(sbi);
  36
  37                 f2fs_handle_stop(sbi, reason);
  38         }
  39 }
  40
  41 /*
  42  * We guarantee no failure on the returned page.
  43  */
  44 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  45 {
  46         struct address_space *mapping = META_MAPPING(sbi);
  47         struct page *page;
  48 repeat:
  49         page = f2fs_grab_cache_page(mapping, index, false);
  50         if (!page) {
  51                 cond_resched();
  52                 goto repeat;
  53         }
  54         f2fs_wait_on_page_writeback(page, META, true, true);
  55         if (!PageUptodate(page))
  56                 SetPageUptodate(page);
  57         return page;
  58 }
  59
  60 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
  61                                                         bool is_meta)
  62 {
  63         struct address_space *mapping = META_MAPPING(sbi);
  64         struct page *page;
  65         struct f2fs_io_info fio = {
  66                 .sbi = sbi,
  67                 .type = META,
  68                 .op = REQ_OP_READ,
  69                 .op_flags = REQ_META | REQ_PRIO,
  70                 .old_blkaddr = index,
  71                 .new_blkaddr = index,
  72                 .encrypted_page = NULL,
  73                 .is_por = !is_meta ? 1 : 0,
  74         };
  75         int err;
  76
  77         if (unlikely(!is_meta))
  78                 fio.op_flags &= ~REQ_META;
  79 repeat:
  80         page = f2fs_grab_cache_page(mapping, index, false);
  81         if (!page) {
  82                 cond_resched();
  83                 goto repeat;
  84         }
  85         if (PageUptodate(page))
  86                 goto out;
  87
  88         fio.page = page;
  89
  90         err = f2fs_submit_page_bio(&fio);
  91         if (err) {
  92                 f2fs_put_page(page, 1);
  93                 return ERR_PTR(err);
  94         }
  95
  96         f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, F2FS_BLKSIZE);
  97
  98         lock_page(page);
  99         if (unlikely(page->mapping != mapping)) {
 100                 f2fs_put_page(page, 1);
 101                 goto repeat;
 102         }
 103
 104         if (unlikely(!PageUptodate(page))) {
 105                 f2fs_handle_page_eio(sbi, page->index, META);
 106                 f2fs_put_page(page, 1);
 107                 return ERR_PTR(-EIO);
 108         }
 109 out:
 110         return page;
 111 }
 112
 113 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 114 {
 115         return __get_meta_page(sbi, index, true);
 116 }
 117
 118 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
 119 {
 120         struct page *page;
 121         int count = 0;
 122
 123 retry:
 124         page = __get_meta_page(sbi, index, true);
 125         if (IS_ERR(page)) {
 126                 if (PTR_ERR(page) == -EIO &&
 127                                 ++count <= DEFAULT_RETRY_IO_COUNT)
 128                         goto retry;
 129                 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_META_PAGE);
 130         }
 131         return page;
 132 }
 133
 134 /* for POR only */
 135 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 136 {
 137         return __get_meta_page(sbi, index, false);
 138 }
 139
 140 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
 141                                                         int type)
 142 {
 143         struct seg_entry *se;
 144         unsigned int segno, offset;
 145         bool exist;
 146
 147         if (type == DATA_GENERIC)
 148                 return true;
 149
 150         segno = GET_SEGNO(sbi, blkaddr);
 151         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 152         se = get_seg_entry(sbi, segno);
 153
 154         exist = f2fs_test_bit(offset, se->cur_valid_map);
 155         if (exist && type == DATA_GENERIC_ENHANCE_UPDATE) {
 156                 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
 157                          blkaddr, exist);
 158                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 159                 return exist;
 160         }
 161
 162         if (!exist && type == DATA_GENERIC_ENHANCE) {
 163                 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
 164                          blkaddr, exist);
 165                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 166                 dump_stack();
 167         }
 168         return exist;
 169 }
 170
 171 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
 172                                         block_t blkaddr, int type)
 173 {
 174         if (time_to_inject(sbi, FAULT_BLKADDR))
 175                 return false;
 176
 177         switch (type) {
 178         case META_NAT:
 179                 break;
 180         case META_SIT:
 181                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 182                         return false;
 183                 break;
 184         case META_SSA:
 185                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 186                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 187                         return false;
 188                 break;
 189         case META_CP:
 190                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 191                         blkaddr < __start_cp_addr(sbi)))
 192                         return false;
 193                 break;
 194         case META_POR:
 195                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 196                         blkaddr < MAIN_BLKADDR(sbi)))
 197                         return false;
 198                 break;
 199         case DATA_GENERIC:
 200         case DATA_GENERIC_ENHANCE:
 201         case DATA_GENERIC_ENHANCE_READ:
 202         case DATA_GENERIC_ENHANCE_UPDATE:
 203                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 204                                 blkaddr < MAIN_BLKADDR(sbi))) {
 205                         f2fs_warn(sbi, "access invalid blkaddr:%u",
 206                                   blkaddr);
 207                         set_sbi_flag(sbi, SBI_NEED_FSCK);
 208                         dump_stack();
 209                         return false;
 210                 } else {
 211                         return __is_bitmap_valid(sbi, blkaddr, type);
 212                 }
 213                 break;
 214         case META_GENERIC:
 215                 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
 216                         blkaddr >= MAIN_BLKADDR(sbi)))
 217                         return false;
 218                 break;
 219         default:
 220                 BUG();
 221         }
 222
 223         return true;
 224 }
 225
 226 /*
 227  * Readahead CP/NAT/SIT/SSA/POR pages
 228  */
 229 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 230                                                         int type, bool sync)
 231 {
 232         struct page *page;
 233         block_t blkno = start;
 234         struct f2fs_io_info fio = {
 235                 .sbi = sbi,
 236                 .type = META,
 237                 .op = REQ_OP_READ,
 238                 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
 239                 .encrypted_page = NULL,
 240                 .in_list = 0,
 241                 .is_por = (type == META_POR) ? 1 : 0,
 242         };
 243         struct blk_plug plug;
 244         int err;
 245
 246         if (unlikely(type == META_POR))
 247                 fio.op_flags &= ~REQ_META;
 248
 249         blk_start_plug(&plug);
 250         for (; nrpages-- > 0; blkno++) {
 251
 252                 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
 253                         goto out;
 254
 255                 switch (type) {
 256                 case META_NAT:
 257                         if (unlikely(blkno >=
 258                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 259                                 blkno = 0;
 260                         /* get nat block addr */
 261                         fio.new_blkaddr = current_nat_addr(sbi,
 262                                         blkno * NAT_ENTRY_PER_BLOCK);
 263                         break;
 264                 case META_SIT:
 265                         if (unlikely(blkno >= TOTAL_SEGS(sbi)))
 266                                 goto out;
 267                         /* get sit block addr */
 268                         fio.new_blkaddr = current_sit_addr(sbi,
 269                                         blkno * SIT_ENTRY_PER_BLOCK);
 270                         break;
 271                 case META_SSA:
 272                 case META_CP:
 273                 case META_POR:
 274                         fio.new_blkaddr = blkno;
 275                         break;
 276                 default:
 277                         BUG();
 278                 }
 279
 280                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 281                                                 fio.new_blkaddr, false);
 282                 if (!page)
 283                         continue;
 284                 if (PageUptodate(page)) {
 285                         f2fs_put_page(page, 1);
 286                         continue;
 287                 }
 288
 289                 fio.page = page;
 290                 err = f2fs_submit_page_bio(&fio);
 291                 f2fs_put_page(page, err ? 1 : 0);
 292
 293                 if (!err)
 294                         f2fs_update_iostat(sbi, NULL, FS_META_READ_IO,
 295                                                         F2FS_BLKSIZE);
 296         }
 297 out:
 298         blk_finish_plug(&plug);
 299         return blkno - start;
 300 }
 301
 302 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
 303                                                         unsigned int ra_blocks)
 304 {
 305         struct page *page;
 306         bool readahead = false;
 307
 308         if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
 309                 return;
 310
 311         page = find_get_page(META_MAPPING(sbi), index);
 312         if (!page || !PageUptodate(page))
 313                 readahead = true;
 314         f2fs_put_page(page, 0);
 315
 316         if (readahead)
 317                 f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
 318 }
 319
 320 static int __f2fs_write_meta_page(struct page *page,
 321                                 struct writeback_control *wbc,
 322                                 enum iostat_type io_type)
 323 {
 324         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 325
 326         trace_f2fs_writepage(page, META);
 327
 328         if (unlikely(f2fs_cp_error(sbi)))
 329                 goto redirty_out;
 330         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 331                 goto redirty_out;
 332         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 333                 goto redirty_out;
 334
 335         f2fs_do_write_meta_page(sbi, page, io_type);
 336         dec_page_count(sbi, F2FS_DIRTY_META);
 337
 338         if (wbc->for_reclaim)
 339                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
 340
 341         unlock_page(page);
 342
 343         if (unlikely(f2fs_cp_error(sbi)))
 344                 f2fs_submit_merged_write(sbi, META);
 345
 346         return 0;
 347
 348 redirty_out:
 349         redirty_page_for_writepage(wbc, page);
 350         return AOP_WRITEPAGE_ACTIVATE;
 351 }
 352
 353 static int f2fs_write_meta_page(struct page *page,
 354                                 struct writeback_control *wbc)
 355 {
 356         return __f2fs_write_meta_page(page, wbc, FS_META_IO);
 357 }
 358
 359 static int f2fs_write_meta_pages(struct address_space *mapping,
 360                                 struct writeback_control *wbc)
 361 {
 362         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 363         long diff, written;
 364
 365         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 366                 goto skip_write;
 367
 368         /* collect a number of dirty meta pages and write together */
 369         if (wbc->sync_mode != WB_SYNC_ALL &&
 370                         get_pages(sbi, F2FS_DIRTY_META) <
 371                                         nr_pages_to_skip(sbi, META))
 372                 goto skip_write;
 373
 374         /* if locked failed, cp will flush dirty pages instead */
 375         if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
 376                 goto skip_write;
 377
 378         trace_f2fs_writepages(mapping->host, wbc, META);
 379         diff = nr_pages_to_write(sbi, META, wbc);
 380         written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
 381         f2fs_up_write(&sbi->cp_global_sem);
 382         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 383         return 0;
 384
 385 skip_write:
 386         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 387         trace_f2fs_writepages(mapping->host, wbc, META);
 388         return 0;
 389 }
 390
 391 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 392                                 long nr_to_write, enum iostat_type io_type)
 393 {
 394         struct address_space *mapping = META_MAPPING(sbi);
 395         pgoff_t index = 0, prev = ULONG_MAX;
 396         struct folio_batch fbatch;
 397         long nwritten = 0;
 398         int nr_folios;
 399         struct writeback_control wbc = {
 400                 .for_reclaim = 0,
 401         };
 402         struct blk_plug plug;
 403
 404         folio_batch_init(&fbatch);
 405
 406         blk_start_plug(&plug);
 407
 408         while ((nr_folios = filemap_get_folios_tag(mapping, &index,
 409                                         (pgoff_t)-1,
 410                                         PAGECACHE_TAG_DIRTY, &fbatch))) {
 411                 int i;
 412
 413                 for (i = 0; i < nr_folios; i++) {
 414                         struct folio *folio = fbatch.folios[i];
 415
 416                         if (nr_to_write != LONG_MAX && i != 0 &&
 417                                         folio->index != prev +
 418                                         folio_nr_pages(fbatch.folios[i-1])) {
 419                                 folio_batch_release(&fbatch);
 420                                 goto stop;
 421                         }
 422
 423                         folio_lock(folio);
 424
 425                         if (unlikely(folio->mapping != mapping)) {
 426 continue_unlock:
 427                                 folio_unlock(folio);
 428                                 continue;
 429                         }
 430                         if (!folio_test_dirty(folio)) {
 431                                 /* someone wrote it for us */
 432                                 goto continue_unlock;
 433                         }
 434
 435                         f2fs_wait_on_page_writeback(&folio->page, META,
 436                                         true, true);
 437
 438                         if (!folio_clear_dirty_for_io(folio))
 439                                 goto continue_unlock;
 440
 441                         if (__f2fs_write_meta_page(&folio->page, &wbc,
 442                                                 io_type)) {
 443                                 folio_unlock(folio);
 444                                 break;
 445                         }
 446                         nwritten += folio_nr_pages(folio);
 447                         prev = folio->index;
 448                         if (unlikely(nwritten >= nr_to_write))
 449                                 break;
 450                 }
 451                 folio_batch_release(&fbatch);
 452                 cond_resched();
 453         }
 454 stop:
 455         if (nwritten)
 456                 f2fs_submit_merged_write(sbi, type);
 457
 458         blk_finish_plug(&plug);
 459
 460         return nwritten;
 461 }
 462
 463 static bool f2fs_dirty_meta_folio(struct address_space *mapping,
 464                 struct folio *folio)
 465 {
 466         trace_f2fs_set_page_dirty(&folio->page, META);
 467
 468         if (!folio_test_uptodate(folio))
 469                 folio_mark_uptodate(folio);
 470         if (filemap_dirty_folio(mapping, folio)) {
 471                 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
 472                 set_page_private_reference(&folio->page);
 473                 return true;
 474         }
 475         return false;
 476 }
 477
 478 const struct address_space_operations f2fs_meta_aops = {
 479         .writepage      = f2fs_write_meta_page,
 480         .writepages     = f2fs_write_meta_pages,
 481         .dirty_folio    = f2fs_dirty_meta_folio,
 482         .invalidate_folio = f2fs_invalidate_folio,
 483         .release_folio  = f2fs_release_folio,
 484         .migrate_folio  = filemap_migrate_folio,
 485 };
 486
 487 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 488                                                 unsigned int devidx, int type)
 489 {
 490         struct inode_management *im = &sbi->im[type];
 491         struct ino_entry *e = NULL, *new = NULL;
 492
 493         if (type == FLUSH_INO) {
 494                 rcu_read_lock();
 495                 e = radix_tree_lookup(&im->ino_root, ino);
 496                 rcu_read_unlock();
 497         }
 498
 499 retry:
 500         if (!e)
 501                 new = f2fs_kmem_cache_alloc(ino_entry_slab,
 502                                                 GFP_NOFS, true, NULL);
 503
 504         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 505
 506         spin_lock(&im->ino_lock);
 507         e = radix_tree_lookup(&im->ino_root, ino);
 508         if (!e) {
 509                 if (!new) {
 510                         spin_unlock(&im->ino_lock);
 511                         goto retry;
 512                 }
 513                 e = new;
 514                 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
 515                         f2fs_bug_on(sbi, 1);
 516
 517                 memset(e, 0, sizeof(struct ino_entry));
 518                 e->ino = ino;
 519
 520                 list_add_tail(&e->list, &im->ino_list);
 521                 if (type != ORPHAN_INO)
 522                         im->ino_num++;
 523         }
 524
 525         if (type == FLUSH_INO)
 526                 f2fs_set_bit(devidx, (char *)&e->dirty_device);
 527
 528         spin_unlock(&im->ino_lock);
 529         radix_tree_preload_end();
 530
 531         if (new && e != new)
 532                 kmem_cache_free(ino_entry_slab, new);
 533 }
 534
 535 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 536 {
 537         struct inode_management *im = &sbi->im[type];
 538         struct ino_entry *e;
 539
 540         spin_lock(&im->ino_lock);
 541         e = radix_tree_lookup(&im->ino_root, ino);
 542         if (e) {
 543                 list_del(&e->list);
 544                 radix_tree_delete(&im->ino_root, ino);
 545                 im->ino_num--;
 546                 spin_unlock(&im->ino_lock);
 547                 kmem_cache_free(ino_entry_slab, e);
 548                 return;
 549         }
 550         spin_unlock(&im->ino_lock);
 551 }
 552
 553 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 554 {
 555         /* add new dirty ino entry into list */
 556         __add_ino_entry(sbi, ino, 0, type);
 557 }
 558
 559 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 560 {
 561         /* remove dirty ino entry from list */
 562         __remove_ino_entry(sbi, ino, type);
 563 }
 564
 565 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
 566 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 567 {
 568         struct inode_management *im = &sbi->im[mode];
 569         struct ino_entry *e;
 570
 571         spin_lock(&im->ino_lock);
 572         e = radix_tree_lookup(&im->ino_root, ino);
 573         spin_unlock(&im->ino_lock);
 574         return e ? true : false;
 575 }
 576
 577 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 578 {
 579         struct ino_entry *e, *tmp;
 580         int i;
 581
 582         for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
 583                 struct inode_management *im = &sbi->im[i];
 584
 585                 spin_lock(&im->ino_lock);
 586                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 587                         list_del(&e->list);
 588                         radix_tree_delete(&im->ino_root, e->ino);
 589                         kmem_cache_free(ino_entry_slab, e);
 590                         im->ino_num--;
 591                 }
 592                 spin_unlock(&im->ino_lock);
 593         }
 594 }
 595
 596 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 597                                         unsigned int devidx, int type)
 598 {
 599         __add_ino_entry(sbi, ino, devidx, type);
 600 }
 601
 602 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 603                                         unsigned int devidx, int type)
 604 {
 605         struct inode_management *im = &sbi->im[type];
 606         struct ino_entry *e;
 607         bool is_dirty = false;
 608
 609         spin_lock(&im->ino_lock);
 610         e = radix_tree_lookup(&im->ino_root, ino);
 611         if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
 612                 is_dirty = true;
 613         spin_unlock(&im->ino_lock);
 614         return is_dirty;
 615 }
 616
 617 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
 618 {
 619         struct inode_management *im = &sbi->im[ORPHAN_INO];
 620         int err = 0;
 621
 622         spin_lock(&im->ino_lock);
 623
 624         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 625                 spin_unlock(&im->ino_lock);
 626                 return -ENOSPC;
 627         }
 628
 629         if (unlikely(im->ino_num >= sbi->max_orphans))
 630                 err = -ENOSPC;
 631         else
 632                 im->ino_num++;
 633         spin_unlock(&im->ino_lock);
 634
 635         return err;
 636 }
 637
 638 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
 639 {
 640         struct inode_management *im = &sbi->im[ORPHAN_INO];
 641
 642         spin_lock(&im->ino_lock);
 643         f2fs_bug_on(sbi, im->ino_num == 0);
 644         im->ino_num--;
 645         spin_unlock(&im->ino_lock);
 646 }
 647
 648 void f2fs_add_orphan_inode(struct inode *inode)
 649 {
 650         /* add new orphan ino entry into list */
 651         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
 652         f2fs_update_inode_page(inode);
 653 }
 654
 655 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 656 {
 657         /* remove orphan entry from orphan list */
 658         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 659 }
 660
 661 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 662 {
 663         struct inode *inode;
 664         struct node_info ni;
 665         int err;
 666
 667         inode = f2fs_iget_retry(sbi->sb, ino);
 668         if (IS_ERR(inode)) {
 669                 /*
 670                  * there should be a bug that we can't find the entry
 671                  * to orphan inode.
 672                  */
 673                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 674                 return PTR_ERR(inode);
 675         }
 676
 677         err = f2fs_dquot_initialize(inode);
 678         if (err) {
 679                 iput(inode);
 680                 goto err_out;
 681         }
 682
 683         clear_nlink(inode);
 684
 685         /* truncate all the data during iput */
 686         iput(inode);
 687
 688         err = f2fs_get_node_info(sbi, ino, &ni, false);
 689         if (err)
 690                 goto err_out;
 691
 692         /* ENOMEM was fully retried in f2fs_evict_inode. */
 693         if (ni.blk_addr != NULL_ADDR) {
 694                 err = -EIO;
 695                 goto err_out;
 696         }
 697         return 0;
 698
 699 err_out:
 700         set_sbi_flag(sbi, SBI_NEED_FSCK);
 701         f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
 702                   __func__, ino);
 703         return err;
 704 }
 705
 706 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
 707 {
 708         block_t start_blk, orphan_blocks, i, j;
 709         unsigned int s_flags = sbi->sb->s_flags;
 710         int err = 0;
 711 #ifdef CONFIG_QUOTA
 712         int quota_enabled;
 713 #endif
 714
 715         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 716                 return 0;
 717
 718         if (bdev_read_only(sbi->sb->s_bdev)) {
 719                 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
 720                 return 0;
 721         }
 722
 723         if (s_flags & SB_RDONLY) {
 724                 f2fs_info(sbi, "orphan cleanup on readonly fs");
 725                 sbi->sb->s_flags &= ~SB_RDONLY;
 726         }
 727
 728 #ifdef CONFIG_QUOTA
 729         /*
 730          * Turn on quotas which were not enabled for read-only mounts if
 731          * filesystem has quota feature, so that they are updated correctly.
 732          */
 733         quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
 734 #endif
 735
 736         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 737         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 738
 739         f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 740
 741         for (i = 0; i < orphan_blocks; i++) {
 742                 struct page *page;
 743                 struct f2fs_orphan_block *orphan_blk;
 744
 745                 page = f2fs_get_meta_page(sbi, start_blk + i);
 746                 if (IS_ERR(page)) {
 747                         err = PTR_ERR(page);
 748                         goto out;
 749                 }
 750
 751                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 752                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 753                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 754
 755                         err = recover_orphan_inode(sbi, ino);
 756                         if (err) {
 757                                 f2fs_put_page(page, 1);
 758                                 goto out;
 759                         }
 760                 }
 761                 f2fs_put_page(page, 1);
 762         }
 763         /* clear Orphan Flag */
 764         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 765 out:
 766         set_sbi_flag(sbi, SBI_IS_RECOVERED);
 767
 768 #ifdef CONFIG_QUOTA
 769         /* Turn quotas off */
 770         if (quota_enabled)
 771                 f2fs_quota_off_umount(sbi->sb);
 772 #endif
 773         sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
 774
 775         return err;
 776 }
 777
 778 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 779 {
 780         struct list_head *head;
 781         struct f2fs_orphan_block *orphan_blk = NULL;
 782         unsigned int nentries = 0;
 783         unsigned short index = 1;
 784         unsigned short orphan_blocks;
 785         struct page *page = NULL;
 786         struct ino_entry *orphan = NULL;
 787         struct inode_management *im = &sbi->im[ORPHAN_INO];
 788
 789         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 790
 791         /*
 792          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 793          * orphan inode operations are covered under f2fs_lock_op().
 794          * And, spin_lock should be avoided due to page operations below.
 795          */
 796         head = &im->ino_list;
 797
 798         /* loop for each orphan inode entry and write them in journal block */
 799         list_for_each_entry(orphan, head, list) {
 800                 if (!page) {
 801                         page = f2fs_grab_meta_page(sbi, start_blk++);
 802                         orphan_blk =
 803                                 (struct f2fs_orphan_block *)page_address(page);
 804                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 805                 }
 806
 807                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 808
 809                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 810                         /*
 811                          * an orphan block is full of 1020 entries,
 812                          * then we need to flush current orphan blocks
 813                          * and bring another one in memory
 814                          */
 815                         orphan_blk->blk_addr = cpu_to_le16(index);
 816                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 817                         orphan_blk->entry_count = cpu_to_le32(nentries);
 818                         set_page_dirty(page);
 819                         f2fs_put_page(page, 1);
 820                         index++;
 821                         nentries = 0;
 822                         page = NULL;
 823                 }
 824         }
 825
 826         if (page) {
 827                 orphan_blk->blk_addr = cpu_to_le16(index);
 828                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 829                 orphan_blk->entry_count = cpu_to_le32(nentries);
 830                 set_page_dirty(page);
 831                 f2fs_put_page(page, 1);
 832         }
 833 }
 834
 835 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
 836                                                 struct f2fs_checkpoint *ckpt)
 837 {
 838         unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
 839         __u32 chksum;
 840
 841         chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
 842         if (chksum_ofs < CP_CHKSUM_OFFSET) {
 843                 chksum_ofs += sizeof(chksum);
 844                 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
 845                                                 F2FS_BLKSIZE - chksum_ofs);
 846         }
 847         return chksum;
 848 }
 849
 850 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 851                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 852                 unsigned long long *version)
 853 {
 854         size_t crc_offset = 0;
 855         __u32 crc;
 856
 857         *cp_page = f2fs_get_meta_page(sbi, cp_addr);
 858         if (IS_ERR(*cp_page))
 859                 return PTR_ERR(*cp_page);
 860
 861         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 862
 863         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 864         if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
 865                         crc_offset > CP_CHKSUM_OFFSET) {
 866                 f2fs_put_page(*cp_page, 1);
 867                 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
 868                 return -EINVAL;
 869         }
 870
 871         crc = f2fs_checkpoint_chksum(sbi, *cp_block);
 872         if (crc != cur_cp_crc(*cp_block)) {
 873                 f2fs_put_page(*cp_page, 1);
 874                 f2fs_warn(sbi, "invalid crc value");
 875                 return -EINVAL;
 876         }
 877
 878         *version = cur_cp_version(*cp_block);
 879         return 0;
 880 }
 881
 882 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 883                                 block_t cp_addr, unsigned long long *version)
 884 {
 885         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 886         struct f2fs_checkpoint *cp_block = NULL;
 887         unsigned long long cur_version = 0, pre_version = 0;
 888         unsigned int cp_blocks;
 889         int err;
 890
 891         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 892                                         &cp_page_1, version);
 893         if (err)
 894                 return NULL;
 895
 896         cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
 897
 898         if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
 899                 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
 900                           le32_to_cpu(cp_block->cp_pack_total_block_count));
 901                 goto invalid_cp;
 902         }
 903         pre_version = *version;
 904
 905         cp_addr += cp_blocks - 1;
 906         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 907                                         &cp_page_2, version);
 908         if (err)
 909                 goto invalid_cp;
 910         cur_version = *version;
 911
 912         if (cur_version == pre_version) {
 913                 *version = cur_version;
 914                 f2fs_put_page(cp_page_2, 1);
 915                 return cp_page_1;
 916         }
 917         f2fs_put_page(cp_page_2, 1);
 918 invalid_cp:
 919         f2fs_put_page(cp_page_1, 1);
 920         return NULL;
 921 }
 922
 923 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
 924 {
 925         struct f2fs_checkpoint *cp_block;
 926         struct f2fs_super_block *fsb = sbi->raw_super;
 927         struct page *cp1, *cp2, *cur_page;
 928         unsigned long blk_size = sbi->blocksize;
 929         unsigned long long cp1_version = 0, cp2_version = 0;
 930         unsigned long long cp_start_blk_no;
 931         unsigned int cp_blks = 1 + __cp_payload(sbi);
 932         block_t cp_blk_no;
 933         int i;
 934         int err;
 935
 936         sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
 937                                   GFP_KERNEL);
 938         if (!sbi->ckpt)
 939                 return -ENOMEM;
 940         /*
 941          * Finding out valid cp block involves read both
 942          * sets( cp pack 1 and cp pack 2)
 943          */
 944         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 945         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 946
 947         /* The second checkpoint pack should start at the next segment */
 948         cp_start_blk_no += ((unsigned long long)1) <<
 949                                 le32_to_cpu(fsb->log_blocks_per_seg);
 950         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 951
 952         if (cp1 && cp2) {
 953                 if (ver_after(cp2_version, cp1_version))
 954                         cur_page = cp2;
 955                 else
 956                         cur_page = cp1;
 957         } else if (cp1) {
 958                 cur_page = cp1;
 959         } else if (cp2) {
 960                 cur_page = cp2;
 961         } else {
 962                 err = -EFSCORRUPTED;
 963                 goto fail_no_cp;
 964         }
 965
 966         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 967         memcpy(sbi->ckpt, cp_block, blk_size);
 968
 969         if (cur_page == cp1)
 970                 sbi->cur_cp_pack = 1;
 971         else
 972                 sbi->cur_cp_pack = 2;
 973
 974         /* Sanity checking of checkpoint */
 975         if (f2fs_sanity_check_ckpt(sbi)) {
 976                 err = -EFSCORRUPTED;
 977                 goto free_fail_no_cp;
 978         }
 979
 980         if (cp_blks <= 1)
 981                 goto done;
 982
 983         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 984         if (cur_page == cp2)
 985                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 986
 987         for (i = 1; i < cp_blks; i++) {
 988                 void *sit_bitmap_ptr;
 989                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 990
 991                 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
 992                 if (IS_ERR(cur_page)) {
 993                         err = PTR_ERR(cur_page);
 994                         goto free_fail_no_cp;
 995                 }
 996                 sit_bitmap_ptr = page_address(cur_page);
 997                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 998                 f2fs_put_page(cur_page, 1);
 999         }
1000 done:
1001         f2fs_put_page(cp1, 1);
1002         f2fs_put_page(cp2, 1);
1003         return 0;
1004
1005 free_fail_no_cp:
1006         f2fs_put_page(cp1, 1);
1007         f2fs_put_page(cp2, 1);
1008 fail_no_cp:
1009         kvfree(sbi->ckpt);
1010         return err;
1011 }
1012
1013 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1014 {
1015         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1016         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1017
1018         if (is_inode_flag_set(inode, flag))
1019                 return;
1020
1021         set_inode_flag(inode, flag);
1022         list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1023         stat_inc_dirty_inode(sbi, type);
1024 }
1025
1026 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1027 {
1028         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1029
1030         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1031                 return;
1032
1033         list_del_init(&F2FS_I(inode)->dirty_list);
1034         clear_inode_flag(inode, flag);
1035         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1036 }
1037
1038 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1039 {
1040         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1041         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1042
1043         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1044                         !S_ISLNK(inode->i_mode))
1045                 return;
1046
1047         spin_lock(&sbi->inode_lock[type]);
1048         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1049                 __add_dirty_inode(inode, type);
1050         inode_inc_dirty_pages(inode);
1051         spin_unlock(&sbi->inode_lock[type]);
1052
1053         set_page_private_reference(&folio->page);
1054 }
1055
1056 void f2fs_remove_dirty_inode(struct inode *inode)
1057 {
1058         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1059         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1060
1061         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1062                         !S_ISLNK(inode->i_mode))
1063                 return;
1064
1065         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1066                 return;
1067
1068         spin_lock(&sbi->inode_lock[type]);
1069         __remove_dirty_inode(inode, type);
1070         spin_unlock(&sbi->inode_lock[type]);
1071 }
1072
1073 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1074                                                 bool from_cp)
1075 {
1076         struct list_head *head;
1077         struct inode *inode;
1078         struct f2fs_inode_info *fi;
1079         bool is_dir = (type == DIR_INODE);
1080         unsigned long ino = 0;
1081
1082         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1083                                 get_pages(sbi, is_dir ?
1084                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1085 retry:
1086         if (unlikely(f2fs_cp_error(sbi))) {
1087                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1088                                 get_pages(sbi, is_dir ?
1089                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1090                 return -EIO;
1091         }
1092
1093         spin_lock(&sbi->inode_lock[type]);
1094
1095         head = &sbi->inode_list[type];
1096         if (list_empty(head)) {
1097                 spin_unlock(&sbi->inode_lock[type]);
1098                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1099                                 get_pages(sbi, is_dir ?
1100                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1101                 return 0;
1102         }
1103         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1104         inode = igrab(&fi->vfs_inode);
1105         spin_unlock(&sbi->inode_lock[type]);
1106         if (inode) {
1107                 unsigned long cur_ino = inode->i_ino;
1108
1109                 if (from_cp)
1110                         F2FS_I(inode)->cp_task = current;
1111                 F2FS_I(inode)->wb_task = current;
1112
1113                 filemap_fdatawrite(inode->i_mapping);
1114
1115                 F2FS_I(inode)->wb_task = NULL;
1116                 if (from_cp)
1117                         F2FS_I(inode)->cp_task = NULL;
1118
1119                 iput(inode);
1120                 /* We need to give cpu to another writers. */
1121                 if (ino == cur_ino)
1122                         cond_resched();
1123                 else
1124                         ino = cur_ino;
1125         } else {
1126                 /*
1127                  * We should submit bio, since it exists several
1128                  * writebacking dentry pages in the freeing inode.
1129                  */
1130                 f2fs_submit_merged_write(sbi, DATA);
1131                 cond_resched();
1132         }
1133         goto retry;
1134 }
1135
1136 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1137 {
1138         struct list_head *head = &sbi->inode_list[DIRTY_META];
1139         struct inode *inode;
1140         struct f2fs_inode_info *fi;
1141         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1142
1143         while (total--) {
1144                 if (unlikely(f2fs_cp_error(sbi)))
1145                         return -EIO;
1146
1147                 spin_lock(&sbi->inode_lock[DIRTY_META]);
1148                 if (list_empty(head)) {
1149                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1150                         return 0;
1151                 }
1152                 fi = list_first_entry(head, struct f2fs_inode_info,
1153                                                         gdirty_list);
1154                 inode = igrab(&fi->vfs_inode);
1155                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1156                 if (inode) {
1157                         sync_inode_metadata(inode, 0);
1158
1159                         /* it's on eviction */
1160                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1161                                 f2fs_update_inode_page(inode);
1162                         iput(inode);
1163                 }
1164         }
1165         return 0;
1166 }
1167
1168 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1169 {
1170         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1171         struct f2fs_nm_info *nm_i = NM_I(sbi);
1172         nid_t last_nid = nm_i->next_scan_nid;
1173
1174         next_free_nid(sbi, &last_nid);
1175         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1176         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1177         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1178         ckpt->next_free_nid = cpu_to_le32(last_nid);
1179 }
1180
1181 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1182 {
1183         bool ret = false;
1184
1185         if (!is_journalled_quota(sbi))
1186                 return false;
1187
1188         if (!f2fs_down_write_trylock(&sbi->quota_sem))
1189                 return true;
1190         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1191                 ret = false;
1192         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1193                 ret = false;
1194         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1195                 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1196                 ret = true;
1197         } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1198                 ret = true;
1199         }
1200         f2fs_up_write(&sbi->quota_sem);
1201         return ret;
1202 }
1203
1204 /*
1205  * Freeze all the FS-operations for checkpoint.
1206  */
1207 static int block_operations(struct f2fs_sb_info *sbi)
1208 {
1209         struct writeback_control wbc = {
1210                 .sync_mode = WB_SYNC_ALL,
1211                 .nr_to_write = LONG_MAX,
1212                 .for_reclaim = 0,
1213         };
1214         int err = 0, cnt = 0;
1215
1216         /*
1217          * Let's flush inline_data in dirty node pages.
1218          */
1219         f2fs_flush_inline_data(sbi);
1220
1221 retry_flush_quotas:
1222         f2fs_lock_all(sbi);
1223         if (__need_flush_quota(sbi)) {
1224                 int locked;
1225
1226                 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1227                         set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1228                         set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1229                         goto retry_flush_dents;
1230                 }
1231                 f2fs_unlock_all(sbi);
1232
1233                 /* only failed during mount/umount/freeze/quotactl */
1234                 locked = down_read_trylock(&sbi->sb->s_umount);
1235                 f2fs_quota_sync(sbi->sb, -1);
1236                 if (locked)
1237                         up_read(&sbi->sb->s_umount);
1238                 cond_resched();
1239                 goto retry_flush_quotas;
1240         }
1241
1242 retry_flush_dents:
1243         /* write all the dirty dentry pages */
1244         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1245                 f2fs_unlock_all(sbi);
1246                 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1247                 if (err)
1248                         return err;
1249                 cond_resched();
1250                 goto retry_flush_quotas;
1251         }
1252
1253         /*
1254          * POR: we should ensure that there are no dirty node pages
1255          * until finishing nat/sit flush. inode->i_blocks can be updated.
1256          */
1257         f2fs_down_write(&sbi->node_change);
1258
1259         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1260                 f2fs_up_write(&sbi->node_change);
1261                 f2fs_unlock_all(sbi);
1262                 err = f2fs_sync_inode_meta(sbi);
1263                 if (err)
1264                         return err;
1265                 cond_resched();
1266                 goto retry_flush_quotas;
1267         }
1268
1269 retry_flush_nodes:
1270         f2fs_down_write(&sbi->node_write);
1271
1272         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1273                 f2fs_up_write(&sbi->node_write);
1274                 atomic_inc(&sbi->wb_sync_req[NODE]);
1275                 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1276                 atomic_dec(&sbi->wb_sync_req[NODE]);
1277                 if (err) {
1278                         f2fs_up_write(&sbi->node_change);
1279                         f2fs_unlock_all(sbi);
1280                         return err;
1281                 }
1282                 cond_resched();
1283                 goto retry_flush_nodes;
1284         }
1285
1286         /*
1287          * sbi->node_change is used only for AIO write_begin path which produces
1288          * dirty node blocks and some checkpoint values by block allocation.
1289          */
1290         __prepare_cp_block(sbi);
1291         f2fs_up_write(&sbi->node_change);
1292         return err;
1293 }
1294
1295 static void unblock_operations(struct f2fs_sb_info *sbi)
1296 {
1297         f2fs_up_write(&sbi->node_write);
1298         f2fs_unlock_all(sbi);
1299 }
1300
1301 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1302 {
1303         DEFINE_WAIT(wait);
1304
1305         for (;;) {
1306                 if (!get_pages(sbi, type))
1307                         break;
1308
1309                 if (unlikely(f2fs_cp_error(sbi)))
1310                         break;
1311
1312                 if (type == F2FS_DIRTY_META)
1313                         f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1314                                                         FS_CP_META_IO);
1315                 else if (type == F2FS_WB_CP_DATA)
1316                         f2fs_submit_merged_write(sbi, DATA);
1317
1318                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1319                 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1320         }
1321         finish_wait(&sbi->cp_wait, &wait);
1322 }
1323
1324 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1325 {
1326         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1327         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1328         unsigned long flags;
1329
1330         if (cpc->reason & CP_UMOUNT) {
1331                 if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1332                         NM_I(sbi)->nat_bits_blocks > sbi->blocks_per_seg) {
1333                         clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1334                         f2fs_notice(sbi, "Disable nat_bits due to no space");
1335                 } else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1336                                                 f2fs_nat_bitmap_enabled(sbi)) {
1337                         f2fs_enable_nat_bits(sbi);
1338                         set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1339                         f2fs_notice(sbi, "Rebuild and enable nat_bits");
1340                 }
1341         }
1342
1343         spin_lock_irqsave(&sbi->cp_lock, flags);
1344
1345         if (cpc->reason & CP_TRIMMED)
1346                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1347         else
1348                 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1349
1350         if (cpc->reason & CP_UMOUNT)
1351                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1352         else
1353                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1354
1355         if (cpc->reason & CP_FASTBOOT)
1356                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1357         else
1358                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1359
1360         if (orphan_num)
1361                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1362         else
1363                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1364
1365         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1366                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1367
1368         if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1369                 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1370         else
1371                 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1372
1373         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1374                 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1375         else
1376                 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1377
1378         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1379                 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1380         else
1381                 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1382
1383         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1384                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1385         else
1386                 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1387
1388         if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1389                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1390
1391         /* set this flag to activate crc|cp_ver for recovery */
1392         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1393         __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1394
1395         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1396 }
1397
1398 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1399         void *src, block_t blk_addr)
1400 {
1401         struct writeback_control wbc = {
1402                 .for_reclaim = 0,
1403         };
1404
1405         /*
1406          * filemap_get_folios_tag and lock_page again will take
1407          * some extra time. Therefore, f2fs_update_meta_pages and
1408          * f2fs_sync_meta_pages are combined in this function.
1409          */
1410         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1411         int err;
1412
1413         f2fs_wait_on_page_writeback(page, META, true, true);
1414
1415         memcpy(page_address(page), src, PAGE_SIZE);
1416
1417         set_page_dirty(page);
1418         if (unlikely(!clear_page_dirty_for_io(page)))
1419                 f2fs_bug_on(sbi, 1);
1420
1421         /* writeout cp pack 2 page */
1422         err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1423         if (unlikely(err && f2fs_cp_error(sbi))) {
1424                 f2fs_put_page(page, 1);
1425                 return;
1426         }
1427
1428         f2fs_bug_on(sbi, err);
1429         f2fs_put_page(page, 0);
1430
1431         /* submit checkpoint (with barrier if NOBARRIER is not set) */
1432         f2fs_submit_merged_write(sbi, META_FLUSH);
1433 }
1434
1435 static inline u64 get_sectors_written(struct block_device *bdev)
1436 {
1437         return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1438 }
1439
1440 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1441 {
1442         if (f2fs_is_multi_device(sbi)) {
1443                 u64 sectors = 0;
1444                 int i;
1445
1446                 for (i = 0; i < sbi->s_ndevs; i++)
1447                         sectors += get_sectors_written(FDEV(i).bdev);
1448
1449                 return sectors;
1450         }
1451
1452         return get_sectors_written(sbi->sb->s_bdev);
1453 }
1454
1455 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1456 {
1457         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1458         struct f2fs_nm_info *nm_i = NM_I(sbi);
1459         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1460         block_t start_blk;
1461         unsigned int data_sum_blocks, orphan_blocks;
1462         __u32 crc32 = 0;
1463         int i;
1464         int cp_payload_blks = __cp_payload(sbi);
1465         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1466         u64 kbytes_written;
1467         int err;
1468
1469         /* Flush all the NAT/SIT pages */
1470         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1471
1472         /* start to update checkpoint, cp ver is already updated previously */
1473         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1474         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1475         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1476                 struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_NODE);
1477
1478                 ckpt->cur_node_segno[i] = cpu_to_le32(curseg->segno);
1479                 ckpt->cur_node_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1480                 ckpt->alloc_type[i + CURSEG_HOT_NODE] = curseg->alloc_type;
1481         }
1482         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1483                 struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_DATA);
1484
1485                 ckpt->cur_data_segno[i] = cpu_to_le32(curseg->segno);
1486                 ckpt->cur_data_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1487                 ckpt->alloc_type[i + CURSEG_HOT_DATA] = curseg->alloc_type;
1488         }
1489
1490         /* 2 cp + n data seg summary + orphan inode blocks */
1491         data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1492         spin_lock_irqsave(&sbi->cp_lock, flags);
1493         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1494                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1495         else
1496                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1497         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1498
1499         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1500         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1501                         orphan_blocks);
1502
1503         if (__remain_node_summaries(cpc->reason))
1504                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1505                                 cp_payload_blks + data_sum_blocks +
1506                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1507         else
1508                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1509                                 cp_payload_blks + data_sum_blocks +
1510                                 orphan_blocks);
1511
1512         /* update ckpt flag for checkpoint */
1513         update_ckpt_flags(sbi, cpc);
1514
1515         /* update SIT/NAT bitmap */
1516         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1517         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1518
1519         crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1520         *((__le32 *)((unsigned char *)ckpt +
1521                                 le32_to_cpu(ckpt->checksum_offset)))
1522                                 = cpu_to_le32(crc32);
1523
1524         start_blk = __start_cp_next_addr(sbi);
1525
1526         /* write nat bits */
1527         if ((cpc->reason & CP_UMOUNT) &&
1528                         is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1529                 __u64 cp_ver = cur_cp_version(ckpt);
1530                 block_t blk;
1531
1532                 cp_ver |= ((__u64)crc32 << 32);
1533                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1534
1535                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1536                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1537                         f2fs_update_meta_page(sbi, nm_i->nat_bits +
1538                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1539         }
1540
1541         /* write out checkpoint buffer at block 0 */
1542         f2fs_update_meta_page(sbi, ckpt, start_blk++);
1543
1544         for (i = 1; i < 1 + cp_payload_blks; i++)
1545                 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1546                                                         start_blk++);
1547
1548         if (orphan_num) {
1549                 write_orphan_inodes(sbi, start_blk);
1550                 start_blk += orphan_blocks;
1551         }
1552
1553         f2fs_write_data_summaries(sbi, start_blk);
1554         start_blk += data_sum_blocks;
1555
1556         /* Record write statistics in the hot node summary */
1557         kbytes_written = sbi->kbytes_written;
1558         kbytes_written += (f2fs_get_sectors_written(sbi) -
1559                                 sbi->sectors_written_start) >> 1;
1560         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1561
1562         if (__remain_node_summaries(cpc->reason)) {
1563                 f2fs_write_node_summaries(sbi, start_blk);
1564                 start_blk += NR_CURSEG_NODE_TYPE;
1565         }
1566
1567         /* update user_block_counts */
1568         sbi->last_valid_block_count = sbi->total_valid_block_count;
1569         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1570         percpu_counter_set(&sbi->rf_node_block_count, 0);
1571
1572         /* Here, we have one bio having CP pack except cp pack 2 page */
1573         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1574         /* Wait for all dirty meta pages to be submitted for IO */
1575         f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1576
1577         /* wait for previous submitted meta pages writeback */
1578         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1579
1580         /* flush all device cache */
1581         err = f2fs_flush_device_cache(sbi);
1582         if (err)
1583                 return err;
1584
1585         /* barrier and flush checkpoint cp pack 2 page if it can */
1586         commit_checkpoint(sbi, ckpt, start_blk);
1587         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1588
1589         /*
1590          * invalidate intermediate page cache borrowed from meta inode which are
1591          * used for migration of encrypted, verity or compressed inode's blocks.
1592          */
1593         if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1594                 f2fs_sb_has_compression(sbi))
1595                 invalidate_mapping_pages(META_MAPPING(sbi),
1596                                 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1597
1598         f2fs_release_ino_entry(sbi, false);
1599
1600         f2fs_reset_fsync_node_info(sbi);
1601
1602         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1603         clear_sbi_flag(sbi, SBI_NEED_CP);
1604         clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1605
1606         spin_lock(&sbi->stat_lock);
1607         sbi->unusable_block_count = 0;
1608         spin_unlock(&sbi->stat_lock);
1609
1610         __set_cp_next_pack(sbi);
1611
1612         /*
1613          * redirty superblock if metadata like node page or inode cache is
1614          * updated during writing checkpoint.
1615          */
1616         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1617                         get_pages(sbi, F2FS_DIRTY_IMETA))
1618                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1619
1620         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1621
1622         return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1623 }
1624
1625 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1626 {
1627         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1628         unsigned long long ckpt_ver;
1629         int err = 0;
1630
1631         if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1632                 return -EROFS;
1633
1634         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1635                 if (cpc->reason != CP_PAUSE)
1636                         return 0;
1637                 f2fs_warn(sbi, "Start checkpoint disabled!");
1638         }
1639         if (cpc->reason != CP_RESIZE)
1640                 f2fs_down_write(&sbi->cp_global_sem);
1641
1642         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1643                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1644                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1645                 goto out;
1646         if (unlikely(f2fs_cp_error(sbi))) {
1647                 err = -EIO;
1648                 goto out;
1649         }
1650
1651         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1652
1653         err = block_operations(sbi);
1654         if (err)
1655                 goto out;
1656
1657         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1658
1659         f2fs_flush_merged_writes(sbi);
1660
1661         /* this is the case of multiple fstrims without any changes */
1662         if (cpc->reason & CP_DISCARD) {
1663                 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1664                         unblock_operations(sbi);
1665                         goto out;
1666                 }
1667
1668                 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1669                                 SIT_I(sbi)->dirty_sentries == 0 &&
1670                                 prefree_segments(sbi) == 0) {
1671                         f2fs_flush_sit_entries(sbi, cpc);
1672                         f2fs_clear_prefree_segments(sbi, cpc);
1673                         unblock_operations(sbi);
1674                         goto out;
1675                 }
1676         }
1677
1678         /*
1679          * update checkpoint pack index
1680          * Increase the version number so that
1681          * SIT entries and seg summaries are written at correct place
1682          */
1683         ckpt_ver = cur_cp_version(ckpt);
1684         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1685
1686         /* write cached NAT/SIT entries to NAT/SIT area */
1687         err = f2fs_flush_nat_entries(sbi, cpc);
1688         if (err) {
1689                 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1690                 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1691                 goto stop;
1692         }
1693
1694         f2fs_flush_sit_entries(sbi, cpc);
1695
1696         /* save inmem log status */
1697         f2fs_save_inmem_curseg(sbi);
1698
1699         err = do_checkpoint(sbi, cpc);
1700         if (err) {
1701                 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1702                 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1703                 f2fs_release_discard_addrs(sbi);
1704         } else {
1705                 f2fs_clear_prefree_segments(sbi, cpc);
1706         }
1707
1708         f2fs_restore_inmem_curseg(sbi);
1709 stop:
1710         unblock_operations(sbi);
1711         stat_inc_cp_count(sbi->stat_info);
1712
1713         if (cpc->reason & CP_RECOVERY)
1714                 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1715
1716         /* update CP_TIME to trigger checkpoint periodically */
1717         f2fs_update_time(sbi, CP_TIME);
1718         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1719 out:
1720         if (cpc->reason != CP_RESIZE)
1721                 f2fs_up_write(&sbi->cp_global_sem);
1722         return err;
1723 }
1724
1725 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1726 {
1727         int i;
1728
1729         for (i = 0; i < MAX_INO_ENTRY; i++) {
1730                 struct inode_management *im = &sbi->im[i];
1731
1732                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1733                 spin_lock_init(&im->ino_lock);
1734                 INIT_LIST_HEAD(&im->ino_list);
1735                 im->ino_num = 0;
1736         }
1737
1738         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1739                         NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1740                                 F2FS_ORPHANS_PER_BLOCK;
1741 }
1742
1743 int __init f2fs_create_checkpoint_caches(void)
1744 {
1745         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1746                         sizeof(struct ino_entry));
1747         if (!ino_entry_slab)
1748                 return -ENOMEM;
1749         f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1750                         sizeof(struct inode_entry));
1751         if (!f2fs_inode_entry_slab) {
1752                 kmem_cache_destroy(ino_entry_slab);
1753                 return -ENOMEM;
1754         }
1755         return 0;
1756 }
1757
1758 void f2fs_destroy_checkpoint_caches(void)
1759 {
1760         kmem_cache_destroy(ino_entry_slab);
1761         kmem_cache_destroy(f2fs_inode_entry_slab);
1762 }
1763
1764 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1765 {
1766         struct cp_control cpc = { .reason = CP_SYNC, };
1767         int err;
1768
1769         f2fs_down_write(&sbi->gc_lock);
1770         err = f2fs_write_checkpoint(sbi, &cpc);
1771         f2fs_up_write(&sbi->gc_lock);
1772
1773         return err;
1774 }
1775
1776 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1777 {
1778         struct ckpt_req_control *cprc = &sbi->cprc_info;
1779         struct ckpt_req *req, *next;
1780         struct llist_node *dispatch_list;
1781         u64 sum_diff = 0, diff, count = 0;
1782         int ret;
1783
1784         dispatch_list = llist_del_all(&cprc->issue_list);
1785         if (!dispatch_list)
1786                 return;
1787         dispatch_list = llist_reverse_order(dispatch_list);
1788
1789         ret = __write_checkpoint_sync(sbi);
1790         atomic_inc(&cprc->issued_ckpt);
1791
1792         llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1793                 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1794                 req->ret = ret;
1795                 complete(&req->wait);
1796
1797                 sum_diff += diff;
1798                 count++;
1799         }
1800         atomic_sub(count, &cprc->queued_ckpt);
1801         atomic_add(count, &cprc->total_ckpt);
1802
1803         spin_lock(&cprc->stat_lock);
1804         cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1805         if (cprc->peak_time < cprc->cur_time)
1806                 cprc->peak_time = cprc->cur_time;
1807         spin_unlock(&cprc->stat_lock);
1808 }
1809
1810 static int issue_checkpoint_thread(void *data)
1811 {
1812         struct f2fs_sb_info *sbi = data;
1813         struct ckpt_req_control *cprc = &sbi->cprc_info;
1814         wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1815 repeat:
1816         if (kthread_should_stop())
1817                 return 0;
1818
1819         if (!llist_empty(&cprc->issue_list))
1820                 __checkpoint_and_complete_reqs(sbi);
1821
1822         wait_event_interruptible(*q,
1823                 kthread_should_stop() || !llist_empty(&cprc->issue_list));
1824         goto repeat;
1825 }
1826
1827 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1828                 struct ckpt_req *wait_req)
1829 {
1830         struct ckpt_req_control *cprc = &sbi->cprc_info;
1831
1832         if (!llist_empty(&cprc->issue_list)) {
1833                 __checkpoint_and_complete_reqs(sbi);
1834         } else {
1835                 /* already dispatched by issue_checkpoint_thread */
1836                 if (wait_req)
1837                         wait_for_completion(&wait_req->wait);
1838         }
1839 }
1840
1841 static void init_ckpt_req(struct ckpt_req *req)
1842 {
1843         memset(req, 0, sizeof(struct ckpt_req));
1844
1845         init_completion(&req->wait);
1846         req->queue_time = ktime_get();
1847 }
1848
1849 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1850 {
1851         struct ckpt_req_control *cprc = &sbi->cprc_info;
1852         struct ckpt_req req;
1853         struct cp_control cpc;
1854
1855         cpc.reason = __get_cp_reason(sbi);
1856         if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1857                 int ret;
1858
1859                 f2fs_down_write(&sbi->gc_lock);
1860                 ret = f2fs_write_checkpoint(sbi, &cpc);
1861                 f2fs_up_write(&sbi->gc_lock);
1862
1863                 return ret;
1864         }
1865
1866         if (!cprc->f2fs_issue_ckpt)
1867                 return __write_checkpoint_sync(sbi);
1868
1869         init_ckpt_req(&req);
1870
1871         llist_add(&req.llnode, &cprc->issue_list);
1872         atomic_inc(&cprc->queued_ckpt);
1873
1874         /*
1875          * update issue_list before we wake up issue_checkpoint thread,
1876          * this smp_mb() pairs with another barrier in ___wait_event(),
1877          * see more details in comments of waitqueue_active().
1878          */
1879         smp_mb();
1880
1881         if (waitqueue_active(&cprc->ckpt_wait_queue))
1882                 wake_up(&cprc->ckpt_wait_queue);
1883
1884         if (cprc->f2fs_issue_ckpt)
1885                 wait_for_completion(&req.wait);
1886         else
1887                 flush_remained_ckpt_reqs(sbi, &req);
1888
1889         return req.ret;
1890 }
1891
1892 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1893 {
1894         dev_t dev = sbi->sb->s_bdev->bd_dev;
1895         struct ckpt_req_control *cprc = &sbi->cprc_info;
1896
1897         if (cprc->f2fs_issue_ckpt)
1898                 return 0;
1899
1900         cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1901                         "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1902         if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1903                 int err = PTR_ERR(cprc->f2fs_issue_ckpt);
1904
1905                 cprc->f2fs_issue_ckpt = NULL;
1906                 return err;
1907         }
1908
1909         set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1910
1911         return 0;
1912 }
1913
1914 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1915 {
1916         struct ckpt_req_control *cprc = &sbi->cprc_info;
1917         struct task_struct *ckpt_task;
1918
1919         if (!cprc->f2fs_issue_ckpt)
1920                 return;
1921
1922         ckpt_task = cprc->f2fs_issue_ckpt;
1923         cprc->f2fs_issue_ckpt = NULL;
1924         kthread_stop(ckpt_task);
1925
1926         f2fs_flush_ckpt_thread(sbi);
1927 }
1928
1929 void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1930 {
1931         struct ckpt_req_control *cprc = &sbi->cprc_info;
1932
1933         flush_remained_ckpt_reqs(sbi, NULL);
1934
1935         /* Let's wait for the previous dispatched checkpoint. */
1936         while (atomic_read(&cprc->queued_ckpt))
1937                 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1938 }
1939
1940 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1941 {
1942         struct ckpt_req_control *cprc = &sbi->cprc_info;
1943
1944         atomic_set(&cprc->issued_ckpt, 0);
1945         atomic_set(&cprc->total_ckpt, 0);
1946         atomic_set(&cprc->queued_ckpt, 0);
1947         cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1948         init_waitqueue_head(&cprc->ckpt_wait_queue);
1949         init_llist_head(&cprc->issue_list);
1950         spin_lock_init(&cprc->stat_lock);
1951 }