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         switch (type) {
 175         case META_NAT:
 176                 break;
 177         case META_SIT:
 178                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 179                         return false;
 180                 break;
 181         case META_SSA:
 182                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 183                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 184                         return false;
 185                 break;
 186         case META_CP:
 187                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 188                         blkaddr < __start_cp_addr(sbi)))
 189                         return false;
 190                 break;
 191         case META_POR:
 192                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 193                         blkaddr < MAIN_BLKADDR(sbi)))
 194                         return false;
 195                 break;
 196         case DATA_GENERIC:
 197         case DATA_GENERIC_ENHANCE:
 198         case DATA_GENERIC_ENHANCE_READ:
 199         case DATA_GENERIC_ENHANCE_UPDATE:
 200                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 201                                 blkaddr < MAIN_BLKADDR(sbi))) {
 202                         f2fs_warn(sbi, "access invalid blkaddr:%u",
 203                                   blkaddr);
 204                         set_sbi_flag(sbi, SBI_NEED_FSCK);
 205                         dump_stack();
 206                         return false;
 207                 } else {
 208                         return __is_bitmap_valid(sbi, blkaddr, type);
 209                 }
 210                 break;
 211         case META_GENERIC:
 212                 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
 213                         blkaddr >= MAIN_BLKADDR(sbi)))
 214                         return false;
 215                 break;
 216         default:
 217                 BUG();
 218         }
 219
 220         return true;
 221 }
 222
 223 /*
 224  * Readahead CP/NAT/SIT/SSA/POR pages
 225  */
 226 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 227                                                         int type, bool sync)
 228 {
 229         struct page *page;
 230         block_t blkno = start;
 231         struct f2fs_io_info fio = {
 232                 .sbi = sbi,
 233                 .type = META,
 234                 .op = REQ_OP_READ,
 235                 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
 236                 .encrypted_page = NULL,
 237                 .in_list = 0,
 238                 .is_por = (type == META_POR) ? 1 : 0,
 239         };
 240         struct blk_plug plug;
 241         int err;
 242
 243         if (unlikely(type == META_POR))
 244                 fio.op_flags &= ~REQ_META;
 245
 246         blk_start_plug(&plug);
 247         for (; nrpages-- > 0; blkno++) {
 248
 249                 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
 250                         goto out;
 251
 252                 switch (type) {
 253                 case META_NAT:
 254                         if (unlikely(blkno >=
 255                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 256                                 blkno = 0;
 257                         /* get nat block addr */
 258                         fio.new_blkaddr = current_nat_addr(sbi,
 259                                         blkno * NAT_ENTRY_PER_BLOCK);
 260                         break;
 261                 case META_SIT:
 262                         if (unlikely(blkno >= TOTAL_SEGS(sbi)))
 263                                 goto out;
 264                         /* get sit block addr */
 265                         fio.new_blkaddr = current_sit_addr(sbi,
 266                                         blkno * SIT_ENTRY_PER_BLOCK);
 267                         break;
 268                 case META_SSA:
 269                 case META_CP:
 270                 case META_POR:
 271                         fio.new_blkaddr = blkno;
 272                         break;
 273                 default:
 274                         BUG();
 275                 }
 276
 277                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 278                                                 fio.new_blkaddr, false);
 279                 if (!page)
 280                         continue;
 281                 if (PageUptodate(page)) {
 282                         f2fs_put_page(page, 1);
 283                         continue;
 284                 }
 285
 286                 fio.page = page;
 287                 err = f2fs_submit_page_bio(&fio);
 288                 f2fs_put_page(page, err ? 1 : 0);
 289
 290                 if (!err)
 291                         f2fs_update_iostat(sbi, NULL, FS_META_READ_IO,
 292                                                         F2FS_BLKSIZE);
 293         }
 294 out:
 295         blk_finish_plug(&plug);
 296         return blkno - start;
 297 }
 298
 299 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
 300                                                         unsigned int ra_blocks)
 301 {
 302         struct page *page;
 303         bool readahead = false;
 304
 305         if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
 306                 return;
 307
 308         page = find_get_page(META_MAPPING(sbi), index);
 309         if (!page || !PageUptodate(page))
 310                 readahead = true;
 311         f2fs_put_page(page, 0);
 312
 313         if (readahead)
 314                 f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
 315 }
 316
 317 static int __f2fs_write_meta_page(struct page *page,
 318                                 struct writeback_control *wbc,
 319                                 enum iostat_type io_type)
 320 {
 321         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 322
 323         trace_f2fs_writepage(page, META);
 324
 325         if (unlikely(f2fs_cp_error(sbi))) {
 326                 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
 327                         ClearPageUptodate(page);
 328                         dec_page_count(sbi, F2FS_DIRTY_META);
 329                         unlock_page(page);
 330                         return 0;
 331                 }
 332                 goto redirty_out;
 333         }
 334         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 335                 goto redirty_out;
 336         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 337                 goto redirty_out;
 338
 339         f2fs_do_write_meta_page(sbi, page, io_type);
 340         dec_page_count(sbi, F2FS_DIRTY_META);
 341
 342         if (wbc->for_reclaim)
 343                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
 344
 345         unlock_page(page);
 346
 347         if (unlikely(f2fs_cp_error(sbi)))
 348                 f2fs_submit_merged_write(sbi, META);
 349
 350         return 0;
 351
 352 redirty_out:
 353         redirty_page_for_writepage(wbc, page);
 354         return AOP_WRITEPAGE_ACTIVATE;
 355 }
 356
 357 static int f2fs_write_meta_page(struct page *page,
 358                                 struct writeback_control *wbc)
 359 {
 360         return __f2fs_write_meta_page(page, wbc, FS_META_IO);
 361 }
 362
 363 static int f2fs_write_meta_pages(struct address_space *mapping,
 364                                 struct writeback_control *wbc)
 365 {
 366         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 367         long diff, written;
 368
 369         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 370                 goto skip_write;
 371
 372         /* collect a number of dirty meta pages and write together */
 373         if (wbc->sync_mode != WB_SYNC_ALL &&
 374                         get_pages(sbi, F2FS_DIRTY_META) <
 375                                         nr_pages_to_skip(sbi, META))
 376                 goto skip_write;
 377
 378         /* if locked failed, cp will flush dirty pages instead */
 379         if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
 380                 goto skip_write;
 381
 382         trace_f2fs_writepages(mapping->host, wbc, META);
 383         diff = nr_pages_to_write(sbi, META, wbc);
 384         written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
 385         f2fs_up_write(&sbi->cp_global_sem);
 386         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 387         return 0;
 388
 389 skip_write:
 390         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 391         trace_f2fs_writepages(mapping->host, wbc, META);
 392         return 0;
 393 }
 394
 395 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 396                                 long nr_to_write, enum iostat_type io_type)
 397 {
 398         struct address_space *mapping = META_MAPPING(sbi);
 399         pgoff_t index = 0, prev = ULONG_MAX;
 400         struct pagevec pvec;
 401         long nwritten = 0;
 402         int nr_pages;
 403         struct writeback_control wbc = {
 404                 .for_reclaim = 0,
 405         };
 406         struct blk_plug plug;
 407
 408         pagevec_init(&pvec);
 409
 410         blk_start_plug(&plug);
 411
 412         while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 413                                 PAGECACHE_TAG_DIRTY))) {
 414                 int i;
 415
 416                 for (i = 0; i < nr_pages; i++) {
 417                         struct page *page = pvec.pages[i];
 418
 419                         if (prev == ULONG_MAX)
 420                                 prev = page->index - 1;
 421                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 422                                 pagevec_release(&pvec);
 423                                 goto stop;
 424                         }
 425
 426                         lock_page(page);
 427
 428                         if (unlikely(page->mapping != mapping)) {
 429 continue_unlock:
 430                                 unlock_page(page);
 431                                 continue;
 432                         }
 433                         if (!PageDirty(page)) {
 434                                 /* someone wrote it for us */
 435                                 goto continue_unlock;
 436                         }
 437
 438                         f2fs_wait_on_page_writeback(page, META, true, true);
 439
 440                         if (!clear_page_dirty_for_io(page))
 441                                 goto continue_unlock;
 442
 443                         if (__f2fs_write_meta_page(page, &wbc, io_type)) {
 444                                 unlock_page(page);
 445                                 break;
 446                         }
 447                         nwritten++;
 448                         prev = page->index;
 449                         if (unlikely(nwritten >= nr_to_write))
 450                                 break;
 451                 }
 452                 pagevec_release(&pvec);
 453                 cond_resched();
 454         }
 455 stop:
 456         if (nwritten)
 457                 f2fs_submit_merged_write(sbi, type);
 458
 459         blk_finish_plug(&plug);
 460
 461         return nwritten;
 462 }
 463
 464 static bool f2fs_dirty_meta_folio(struct address_space *mapping,
 465                 struct folio *folio)
 466 {
 467         trace_f2fs_set_page_dirty(&folio->page, META);
 468
 469         if (!folio_test_uptodate(folio))
 470                 folio_mark_uptodate(folio);
 471         if (filemap_dirty_folio(mapping, folio)) {
 472                 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
 473                 set_page_private_reference(&folio->page);
 474                 return true;
 475         }
 476         return false;
 477 }
 478
 479 const struct address_space_operations f2fs_meta_aops = {
 480         .writepage      = f2fs_write_meta_page,
 481         .writepages     = f2fs_write_meta_pages,
 482         .dirty_folio    = f2fs_dirty_meta_folio,
 483         .invalidate_folio = f2fs_invalidate_folio,
 484         .release_folio  = f2fs_release_folio,
 485         .migrate_folio  = filemap_migrate_folio,
 486 };
 487
 488 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 489                                                 unsigned int devidx, int type)
 490 {
 491         struct inode_management *im = &sbi->im[type];
 492         struct ino_entry *e = NULL, *new = NULL;
 493
 494         if (type == FLUSH_INO) {
 495                 rcu_read_lock();
 496                 e = radix_tree_lookup(&im->ino_root, ino);
 497                 rcu_read_unlock();
 498         }
 499
 500 retry:
 501         if (!e)
 502                 new = f2fs_kmem_cache_alloc(ino_entry_slab,
 503                                                 GFP_NOFS, true, NULL);
 504
 505         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 506
 507         spin_lock(&im->ino_lock);
 508         e = radix_tree_lookup(&im->ino_root, ino);
 509         if (!e) {
 510                 if (!new) {
 511                         spin_unlock(&im->ino_lock);
 512                         goto retry;
 513                 }
 514                 e = new;
 515                 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
 516                         f2fs_bug_on(sbi, 1);
 517
 518                 memset(e, 0, sizeof(struct ino_entry));
 519                 e->ino = ino;
 520
 521                 list_add_tail(&e->list, &im->ino_list);
 522                 if (type != ORPHAN_INO)
 523                         im->ino_num++;
 524         }
 525
 526         if (type == FLUSH_INO)
 527                 f2fs_set_bit(devidx, (char *)&e->dirty_device);
 528
 529         spin_unlock(&im->ino_lock);
 530         radix_tree_preload_end();
 531
 532         if (new && e != new)
 533                 kmem_cache_free(ino_entry_slab, new);
 534 }
 535
 536 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 537 {
 538         struct inode_management *im = &sbi->im[type];
 539         struct ino_entry *e;
 540
 541         spin_lock(&im->ino_lock);
 542         e = radix_tree_lookup(&im->ino_root, ino);
 543         if (e) {
 544                 list_del(&e->list);
 545                 radix_tree_delete(&im->ino_root, ino);
 546                 im->ino_num--;
 547                 spin_unlock(&im->ino_lock);
 548                 kmem_cache_free(ino_entry_slab, e);
 549                 return;
 550         }
 551         spin_unlock(&im->ino_lock);
 552 }
 553
 554 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 555 {
 556         /* add new dirty ino entry into list */
 557         __add_ino_entry(sbi, ino, 0, type);
 558 }
 559
 560 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 561 {
 562         /* remove dirty ino entry from list */
 563         __remove_ino_entry(sbi, ino, type);
 564 }
 565
 566 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
 567 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 568 {
 569         struct inode_management *im = &sbi->im[mode];
 570         struct ino_entry *e;
 571
 572         spin_lock(&im->ino_lock);
 573         e = radix_tree_lookup(&im->ino_root, ino);
 574         spin_unlock(&im->ino_lock);
 575         return e ? true : false;
 576 }
 577
 578 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 579 {
 580         struct ino_entry *e, *tmp;
 581         int i;
 582
 583         for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
 584                 struct inode_management *im = &sbi->im[i];
 585
 586                 spin_lock(&im->ino_lock);
 587                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 588                         list_del(&e->list);
 589                         radix_tree_delete(&im->ino_root, e->ino);
 590                         kmem_cache_free(ino_entry_slab, e);
 591                         im->ino_num--;
 592                 }
 593                 spin_unlock(&im->ino_lock);
 594         }
 595 }
 596
 597 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 598                                         unsigned int devidx, int type)
 599 {
 600         __add_ino_entry(sbi, ino, devidx, type);
 601 }
 602
 603 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 604                                         unsigned int devidx, int type)
 605 {
 606         struct inode_management *im = &sbi->im[type];
 607         struct ino_entry *e;
 608         bool is_dirty = false;
 609
 610         spin_lock(&im->ino_lock);
 611         e = radix_tree_lookup(&im->ino_root, ino);
 612         if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
 613                 is_dirty = true;
 614         spin_unlock(&im->ino_lock);
 615         return is_dirty;
 616 }
 617
 618 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
 619 {
 620         struct inode_management *im = &sbi->im[ORPHAN_INO];
 621         int err = 0;
 622
 623         spin_lock(&im->ino_lock);
 624
 625         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 626                 spin_unlock(&im->ino_lock);
 627                 f2fs_show_injection_info(sbi, FAULT_ORPHAN);
 628                 return -ENOSPC;
 629         }
 630
 631         if (unlikely(im->ino_num >= sbi->max_orphans))
 632                 err = -ENOSPC;
 633         else
 634                 im->ino_num++;
 635         spin_unlock(&im->ino_lock);
 636
 637         return err;
 638 }
 639
 640 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
 641 {
 642         struct inode_management *im = &sbi->im[ORPHAN_INO];
 643
 644         spin_lock(&im->ino_lock);
 645         f2fs_bug_on(sbi, im->ino_num == 0);
 646         im->ino_num--;
 647         spin_unlock(&im->ino_lock);
 648 }
 649
 650 void f2fs_add_orphan_inode(struct inode *inode)
 651 {
 652         /* add new orphan ino entry into list */
 653         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
 654         f2fs_update_inode_page(inode);
 655 }
 656
 657 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 658 {
 659         /* remove orphan entry from orphan list */
 660         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 661 }
 662
 663 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 664 {
 665         struct inode *inode;
 666         struct node_info ni;
 667         int err;
 668
 669         inode = f2fs_iget_retry(sbi->sb, ino);
 670         if (IS_ERR(inode)) {
 671                 /*
 672                  * there should be a bug that we can't find the entry
 673                  * to orphan inode.
 674                  */
 675                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 676                 return PTR_ERR(inode);
 677         }
 678
 679         err = f2fs_dquot_initialize(inode);
 680         if (err) {
 681                 iput(inode);
 682                 goto err_out;
 683         }
 684
 685         clear_nlink(inode);
 686
 687         /* truncate all the data during iput */
 688         iput(inode);
 689
 690         err = f2fs_get_node_info(sbi, ino, &ni, false);
 691         if (err)
 692                 goto err_out;
 693
 694         /* ENOMEM was fully retried in f2fs_evict_inode. */
 695         if (ni.blk_addr != NULL_ADDR) {
 696                 err = -EIO;
 697                 goto err_out;
 698         }
 699         return 0;
 700
 701 err_out:
 702         set_sbi_flag(sbi, SBI_NEED_FSCK);
 703         f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
 704                   __func__, ino);
 705         return err;
 706 }
 707
 708 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
 709 {
 710         block_t start_blk, orphan_blocks, i, j;
 711         unsigned int s_flags = sbi->sb->s_flags;
 712         int err = 0;
 713 #ifdef CONFIG_QUOTA
 714         int quota_enabled;
 715 #endif
 716
 717         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 718                 return 0;
 719
 720         if (bdev_read_only(sbi->sb->s_bdev)) {
 721                 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
 722                 return 0;
 723         }
 724
 725         if (s_flags & SB_RDONLY) {
 726                 f2fs_info(sbi, "orphan cleanup on readonly fs");
 727                 sbi->sb->s_flags &= ~SB_RDONLY;
 728         }
 729
 730 #ifdef CONFIG_QUOTA
 731         /*
 732          * Turn on quotas which were not enabled for read-only mounts if
 733          * filesystem has quota feature, so that they are updated correctly.
 734          */
 735         quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
 736 #endif
 737
 738         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 739         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 740
 741         f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 742
 743         for (i = 0; i < orphan_blocks; i++) {
 744                 struct page *page;
 745                 struct f2fs_orphan_block *orphan_blk;
 746
 747                 page = f2fs_get_meta_page(sbi, start_blk + i);
 748                 if (IS_ERR(page)) {
 749                         err = PTR_ERR(page);
 750                         goto out;
 751                 }
 752
 753                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 754                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 755                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 756
 757                         err = recover_orphan_inode(sbi, ino);
 758                         if (err) {
 759                                 f2fs_put_page(page, 1);
 760                                 goto out;
 761                         }
 762                 }
 763                 f2fs_put_page(page, 1);
 764         }
 765         /* clear Orphan Flag */
 766         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 767 out:
 768         set_sbi_flag(sbi, SBI_IS_RECOVERED);
 769
 770 #ifdef CONFIG_QUOTA
 771         /* Turn quotas off */
 772         if (quota_enabled)
 773                 f2fs_quota_off_umount(sbi->sb);
 774 #endif
 775         sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
 776
 777         return err;
 778 }
 779
 780 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 781 {
 782         struct list_head *head;
 783         struct f2fs_orphan_block *orphan_blk = NULL;
 784         unsigned int nentries = 0;
 785         unsigned short index = 1;
 786         unsigned short orphan_blocks;
 787         struct page *page = NULL;
 788         struct ino_entry *orphan = NULL;
 789         struct inode_management *im = &sbi->im[ORPHAN_INO];
 790
 791         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 792
 793         /*
 794          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 795          * orphan inode operations are covered under f2fs_lock_op().
 796          * And, spin_lock should be avoided due to page operations below.
 797          */
 798         head = &im->ino_list;
 799
 800         /* loop for each orphan inode entry and write them in Jornal block */
 801         list_for_each_entry(orphan, head, list) {
 802                 if (!page) {
 803                         page = f2fs_grab_meta_page(sbi, start_blk++);
 804                         orphan_blk =
 805                                 (struct f2fs_orphan_block *)page_address(page);
 806                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 807                 }
 808
 809                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 810
 811                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 812                         /*
 813                          * an orphan block is full of 1020 entries,
 814                          * then we need to flush current orphan blocks
 815                          * and bring another one in memory
 816                          */
 817                         orphan_blk->blk_addr = cpu_to_le16(index);
 818                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 819                         orphan_blk->entry_count = cpu_to_le32(nentries);
 820                         set_page_dirty(page);
 821                         f2fs_put_page(page, 1);
 822                         index++;
 823                         nentries = 0;
 824                         page = NULL;
 825                 }
 826         }
 827
 828         if (page) {
 829                 orphan_blk->blk_addr = cpu_to_le16(index);
 830                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 831                 orphan_blk->entry_count = cpu_to_le32(nentries);
 832                 set_page_dirty(page);
 833                 f2fs_put_page(page, 1);
 834         }
 835 }
 836
 837 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
 838                                                 struct f2fs_checkpoint *ckpt)
 839 {
 840         unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
 841         __u32 chksum;
 842
 843         chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
 844         if (chksum_ofs < CP_CHKSUM_OFFSET) {
 845                 chksum_ofs += sizeof(chksum);
 846                 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
 847                                                 F2FS_BLKSIZE - chksum_ofs);
 848         }
 849         return chksum;
 850 }
 851
 852 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 853                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 854                 unsigned long long *version)
 855 {
 856         size_t crc_offset = 0;
 857         __u32 crc;
 858
 859         *cp_page = f2fs_get_meta_page(sbi, cp_addr);
 860         if (IS_ERR(*cp_page))
 861                 return PTR_ERR(*cp_page);
 862
 863         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 864
 865         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 866         if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
 867                         crc_offset > CP_CHKSUM_OFFSET) {
 868                 f2fs_put_page(*cp_page, 1);
 869                 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
 870                 return -EINVAL;
 871         }
 872
 873         crc = f2fs_checkpoint_chksum(sbi, *cp_block);
 874         if (crc != cur_cp_crc(*cp_block)) {
 875                 f2fs_put_page(*cp_page, 1);
 876                 f2fs_warn(sbi, "invalid crc value");
 877                 return -EINVAL;
 878         }
 879
 880         *version = cur_cp_version(*cp_block);
 881         return 0;
 882 }
 883
 884 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 885                                 block_t cp_addr, unsigned long long *version)
 886 {
 887         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 888         struct f2fs_checkpoint *cp_block = NULL;
 889         unsigned long long cur_version = 0, pre_version = 0;
 890         unsigned int cp_blocks;
 891         int err;
 892
 893         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 894                                         &cp_page_1, version);
 895         if (err)
 896                 return NULL;
 897
 898         cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
 899
 900         if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
 901                 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
 902                           le32_to_cpu(cp_block->cp_pack_total_block_count));
 903                 goto invalid_cp;
 904         }
 905         pre_version = *version;
 906
 907         cp_addr += cp_blocks - 1;
 908         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 909                                         &cp_page_2, version);
 910         if (err)
 911                 goto invalid_cp;
 912         cur_version = *version;
 913
 914         if (cur_version == pre_version) {
 915                 *version = cur_version;
 916                 f2fs_put_page(cp_page_2, 1);
 917                 return cp_page_1;
 918         }
 919         f2fs_put_page(cp_page_2, 1);
 920 invalid_cp:
 921         f2fs_put_page(cp_page_1, 1);
 922         return NULL;
 923 }
 924
 925 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
 926 {
 927         struct f2fs_checkpoint *cp_block;
 928         struct f2fs_super_block *fsb = sbi->raw_super;
 929         struct page *cp1, *cp2, *cur_page;
 930         unsigned long blk_size = sbi->blocksize;
 931         unsigned long long cp1_version = 0, cp2_version = 0;
 932         unsigned long long cp_start_blk_no;
 933         unsigned int cp_blks = 1 + __cp_payload(sbi);
 934         block_t cp_blk_no;
 935         int i;
 936         int err;
 937
 938         sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
 939                                   GFP_KERNEL);
 940         if (!sbi->ckpt)
 941                 return -ENOMEM;
 942         /*
 943          * Finding out valid cp block involves read both
 944          * sets( cp pack 1 and cp pack 2)
 945          */
 946         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 947         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 948
 949         /* The second checkpoint pack should start at the next segment */
 950         cp_start_blk_no += ((unsigned long long)1) <<
 951                                 le32_to_cpu(fsb->log_blocks_per_seg);
 952         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 953
 954         if (cp1 && cp2) {
 955                 if (ver_after(cp2_version, cp1_version))
 956                         cur_page = cp2;
 957                 else
 958                         cur_page = cp1;
 959         } else if (cp1) {
 960                 cur_page = cp1;
 961         } else if (cp2) {
 962                 cur_page = cp2;
 963         } else {
 964                 err = -EFSCORRUPTED;
 965                 goto fail_no_cp;
 966         }
 967
 968         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 969         memcpy(sbi->ckpt, cp_block, blk_size);
 970
 971         if (cur_page == cp1)
 972                 sbi->cur_cp_pack = 1;
 973         else
 974                 sbi->cur_cp_pack = 2;
 975
 976         /* Sanity checking of checkpoint */
 977         if (f2fs_sanity_check_ckpt(sbi)) {
 978                 err = -EFSCORRUPTED;
 979                 goto free_fail_no_cp;
 980         }
 981
 982         if (cp_blks <= 1)
 983                 goto done;
 984
 985         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 986         if (cur_page == cp2)
 987                 cp_blk_no += BIT(le32_to_cpu(fsb->log_blocks_per_seg));
 988
 989         for (i = 1; i < cp_blks; i++) {
 990                 void *sit_bitmap_ptr;
 991                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 992
 993                 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
 994                 if (IS_ERR(cur_page)) {
 995                         err = PTR_ERR(cur_page);
 996                         goto free_fail_no_cp;
 997                 }
 998                 sit_bitmap_ptr = page_address(cur_page);
 999                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
1000                 f2fs_put_page(cur_page, 1);
1001         }
1002 done:
1003         f2fs_put_page(cp1, 1);
1004         f2fs_put_page(cp2, 1);
1005         return 0;
1006
1007 free_fail_no_cp:
1008         f2fs_put_page(cp1, 1);
1009         f2fs_put_page(cp2, 1);
1010 fail_no_cp:
1011         kvfree(sbi->ckpt);
1012         return err;
1013 }
1014
1015 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1016 {
1017         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1018         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1019
1020         if (is_inode_flag_set(inode, flag))
1021                 return;
1022
1023         set_inode_flag(inode, flag);
1024         list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1025         stat_inc_dirty_inode(sbi, type);
1026 }
1027
1028 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1029 {
1030         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1031
1032         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1033                 return;
1034
1035         list_del_init(&F2FS_I(inode)->dirty_list);
1036         clear_inode_flag(inode, flag);
1037         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1038 }
1039
1040 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1041 {
1042         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1043         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1044
1045         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1046                         !S_ISLNK(inode->i_mode))
1047                 return;
1048
1049         spin_lock(&sbi->inode_lock[type]);
1050         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1051                 __add_dirty_inode(inode, type);
1052         inode_inc_dirty_pages(inode);
1053         spin_unlock(&sbi->inode_lock[type]);
1054
1055         set_page_private_reference(&folio->page);
1056 }
1057
1058 void f2fs_remove_dirty_inode(struct inode *inode)
1059 {
1060         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1061         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1062
1063         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1064                         !S_ISLNK(inode->i_mode))
1065                 return;
1066
1067         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1068                 return;
1069
1070         spin_lock(&sbi->inode_lock[type]);
1071         __remove_dirty_inode(inode, type);
1072         spin_unlock(&sbi->inode_lock[type]);
1073 }
1074
1075 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1076                                                 bool from_cp)
1077 {
1078         struct list_head *head;
1079         struct inode *inode;
1080         struct f2fs_inode_info *fi;
1081         bool is_dir = (type == DIR_INODE);
1082         unsigned long ino = 0;
1083
1084         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1085                                 get_pages(sbi, is_dir ?
1086                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1087 retry:
1088         if (unlikely(f2fs_cp_error(sbi))) {
1089                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1090                                 get_pages(sbi, is_dir ?
1091                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1092                 return -EIO;
1093         }
1094
1095         spin_lock(&sbi->inode_lock[type]);
1096
1097         head = &sbi->inode_list[type];
1098         if (list_empty(head)) {
1099                 spin_unlock(&sbi->inode_lock[type]);
1100                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1101                                 get_pages(sbi, is_dir ?
1102                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1103                 return 0;
1104         }
1105         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1106         inode = igrab(&fi->vfs_inode);
1107         spin_unlock(&sbi->inode_lock[type]);
1108         if (inode) {
1109                 unsigned long cur_ino = inode->i_ino;
1110
1111                 if (from_cp)
1112                         F2FS_I(inode)->cp_task = current;
1113                 F2FS_I(inode)->wb_task = current;
1114
1115                 filemap_fdatawrite(inode->i_mapping);
1116
1117                 F2FS_I(inode)->wb_task = NULL;
1118                 if (from_cp)
1119                         F2FS_I(inode)->cp_task = NULL;
1120
1121                 iput(inode);
1122                 /* We need to give cpu to another writers. */
1123                 if (ino == cur_ino)
1124                         cond_resched();
1125                 else
1126                         ino = cur_ino;
1127         } else {
1128                 /*
1129                  * We should submit bio, since it exists several
1130                  * wribacking dentry pages in the freeing inode.
1131                  */
1132                 f2fs_submit_merged_write(sbi, DATA);
1133                 cond_resched();
1134         }
1135         goto retry;
1136 }
1137
1138 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1139 {
1140         struct list_head *head = &sbi->inode_list[DIRTY_META];
1141         struct inode *inode;
1142         struct f2fs_inode_info *fi;
1143         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1144
1145         while (total--) {
1146                 if (unlikely(f2fs_cp_error(sbi)))
1147                         return -EIO;
1148
1149                 spin_lock(&sbi->inode_lock[DIRTY_META]);
1150                 if (list_empty(head)) {
1151                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1152                         return 0;
1153                 }
1154                 fi = list_first_entry(head, struct f2fs_inode_info,
1155                                                         gdirty_list);
1156                 inode = igrab(&fi->vfs_inode);
1157                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1158                 if (inode) {
1159                         sync_inode_metadata(inode, 0);
1160
1161                         /* it's on eviction */
1162                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1163                                 f2fs_update_inode_page(inode);
1164                         iput(inode);
1165                 }
1166         }
1167         return 0;
1168 }
1169
1170 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1171 {
1172         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1173         struct f2fs_nm_info *nm_i = NM_I(sbi);
1174         nid_t last_nid = nm_i->next_scan_nid;
1175
1176         next_free_nid(sbi, &last_nid);
1177         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1178         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1179         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1180         ckpt->next_free_nid = cpu_to_le32(last_nid);
1181 }
1182
1183 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1184 {
1185         bool ret = false;
1186
1187         if (!is_journalled_quota(sbi))
1188                 return false;
1189
1190         if (!f2fs_down_write_trylock(&sbi->quota_sem))
1191                 return true;
1192         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1193                 ret = false;
1194         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1195                 ret = false;
1196         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1197                 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1198                 ret = true;
1199         } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1200                 ret = true;
1201         }
1202         f2fs_up_write(&sbi->quota_sem);
1203         return ret;
1204 }
1205
1206 /*
1207  * Freeze all the FS-operations for checkpoint.
1208  */
1209 static int block_operations(struct f2fs_sb_info *sbi)
1210 {
1211         struct writeback_control wbc = {
1212                 .sync_mode = WB_SYNC_ALL,
1213                 .nr_to_write = LONG_MAX,
1214                 .for_reclaim = 0,
1215         };
1216         int err = 0, cnt = 0;
1217
1218         /*
1219          * Let's flush inline_data in dirty node pages.
1220          */
1221         f2fs_flush_inline_data(sbi);
1222
1223 retry_flush_quotas:
1224         f2fs_lock_all(sbi);
1225         if (__need_flush_quota(sbi)) {
1226                 int locked;
1227
1228                 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1229                         set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1230                         set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1231                         goto retry_flush_dents;
1232                 }
1233                 f2fs_unlock_all(sbi);
1234
1235                 /* only failed during mount/umount/freeze/quotactl */
1236                 locked = down_read_trylock(&sbi->sb->s_umount);
1237                 f2fs_quota_sync(sbi->sb, -1);
1238                 if (locked)
1239                         up_read(&sbi->sb->s_umount);
1240                 cond_resched();
1241                 goto retry_flush_quotas;
1242         }
1243
1244 retry_flush_dents:
1245         /* write all the dirty dentry pages */
1246         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1247                 f2fs_unlock_all(sbi);
1248                 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1249                 if (err)
1250                         return err;
1251                 cond_resched();
1252                 goto retry_flush_quotas;
1253         }
1254
1255         /*
1256          * POR: we should ensure that there are no dirty node pages
1257          * until finishing nat/sit flush. inode->i_blocks can be updated.
1258          */
1259         f2fs_down_write(&sbi->node_change);
1260
1261         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1262                 f2fs_up_write(&sbi->node_change);
1263                 f2fs_unlock_all(sbi);
1264                 err = f2fs_sync_inode_meta(sbi);
1265                 if (err)
1266                         return err;
1267                 cond_resched();
1268                 goto retry_flush_quotas;
1269         }
1270
1271 retry_flush_nodes:
1272         f2fs_down_write(&sbi->node_write);
1273
1274         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1275                 f2fs_up_write(&sbi->node_write);
1276                 atomic_inc(&sbi->wb_sync_req[NODE]);
1277                 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1278                 atomic_dec(&sbi->wb_sync_req[NODE]);
1279                 if (err) {
1280                         f2fs_up_write(&sbi->node_change);
1281                         f2fs_unlock_all(sbi);
1282                         return err;
1283                 }
1284                 cond_resched();
1285                 goto retry_flush_nodes;
1286         }
1287
1288         /*
1289          * sbi->node_change is used only for AIO write_begin path which produces
1290          * dirty node blocks and some checkpoint values by block allocation.
1291          */
1292         __prepare_cp_block(sbi);
1293         f2fs_up_write(&sbi->node_change);
1294         return err;
1295 }
1296
1297 static void unblock_operations(struct f2fs_sb_info *sbi)
1298 {
1299         f2fs_up_write(&sbi->node_write);
1300         f2fs_unlock_all(sbi);
1301 }
1302
1303 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1304 {
1305         DEFINE_WAIT(wait);
1306
1307         for (;;) {
1308                 if (!get_pages(sbi, type))
1309                         break;
1310
1311                 if (unlikely(f2fs_cp_error(sbi) &&
1312                         !is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
1313                         break;
1314
1315                 if (type == F2FS_DIRTY_META)
1316                         f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1317                                                         FS_CP_META_IO);
1318                 else if (type == F2FS_WB_CP_DATA)
1319                         f2fs_submit_merged_write(sbi, DATA);
1320
1321                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1322                 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1323         }
1324         finish_wait(&sbi->cp_wait, &wait);
1325 }
1326
1327 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1328 {
1329         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1330         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1331         unsigned long flags;
1332
1333         if (cpc->reason & CP_UMOUNT) {
1334                 if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1335                         NM_I(sbi)->nat_bits_blocks > sbi->blocks_per_seg) {
1336                         clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1337                         f2fs_notice(sbi, "Disable nat_bits due to no space");
1338                 } else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1339                                                 f2fs_nat_bitmap_enabled(sbi)) {
1340                         f2fs_enable_nat_bits(sbi);
1341                         set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1342                         f2fs_notice(sbi, "Rebuild and enable nat_bits");
1343                 }
1344         }
1345
1346         spin_lock_irqsave(&sbi->cp_lock, flags);
1347
1348         if (cpc->reason & CP_TRIMMED)
1349                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1350         else
1351                 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1352
1353         if (cpc->reason & CP_UMOUNT)
1354                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1355         else
1356                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1357
1358         if (cpc->reason & CP_FASTBOOT)
1359                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1360         else
1361                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1362
1363         if (orphan_num)
1364                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1365         else
1366                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1367
1368         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1369                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1370
1371         if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1372                 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1373         else
1374                 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1375
1376         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1377                 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1378         else
1379                 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1380
1381         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1382                 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1383         else
1384                 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1385
1386         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1387                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1388         else
1389                 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1390
1391         if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1392                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1393
1394         /* set this flag to activate crc|cp_ver for recovery */
1395         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1396         __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1397
1398         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1399 }
1400
1401 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1402         void *src, block_t blk_addr)
1403 {
1404         struct writeback_control wbc = {
1405                 .for_reclaim = 0,
1406         };
1407
1408         /*
1409          * pagevec_lookup_tag and lock_page again will take
1410          * some extra time. Therefore, f2fs_update_meta_pages and
1411          * f2fs_sync_meta_pages are combined in this function.
1412          */
1413         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1414         int err;
1415
1416         f2fs_wait_on_page_writeback(page, META, true, true);
1417
1418         memcpy(page_address(page), src, PAGE_SIZE);
1419
1420         set_page_dirty(page);
1421         if (unlikely(!clear_page_dirty_for_io(page)))
1422                 f2fs_bug_on(sbi, 1);
1423
1424         /* writeout cp pack 2 page */
1425         err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1426         if (unlikely(err && f2fs_cp_error(sbi))) {
1427                 f2fs_put_page(page, 1);
1428                 return;
1429         }
1430
1431         f2fs_bug_on(sbi, err);
1432         f2fs_put_page(page, 0);
1433
1434         /* submit checkpoint (with barrier if NOBARRIER is not set) */
1435         f2fs_submit_merged_write(sbi, META_FLUSH);
1436 }
1437
1438 static inline u64 get_sectors_written(struct block_device *bdev)
1439 {
1440         return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1441 }
1442
1443 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1444 {
1445         if (f2fs_is_multi_device(sbi)) {
1446                 u64 sectors = 0;
1447                 int i;
1448
1449                 for (i = 0; i < sbi->s_ndevs; i++)
1450                         sectors += get_sectors_written(FDEV(i).bdev);
1451
1452                 return sectors;
1453         }
1454
1455         return get_sectors_written(sbi->sb->s_bdev);
1456 }
1457
1458 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1459 {
1460         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1461         struct f2fs_nm_info *nm_i = NM_I(sbi);
1462         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1463         block_t start_blk;
1464         unsigned int data_sum_blocks, orphan_blocks;
1465         __u32 crc32 = 0;
1466         int i;
1467         int cp_payload_blks = __cp_payload(sbi);
1468         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1469         u64 kbytes_written;
1470         int err;
1471
1472         /* Flush all the NAT/SIT pages */
1473         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1474
1475         /* start to update checkpoint, cp ver is already updated previously */
1476         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1477         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1478         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1479                 ckpt->cur_node_segno[i] =
1480                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1481                 ckpt->cur_node_blkoff[i] =
1482                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1483                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1484                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1485         }
1486         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1487                 ckpt->cur_data_segno[i] =
1488                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1489                 ckpt->cur_data_blkoff[i] =
1490                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1491                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1492                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1493         }
1494
1495         /* 2 cp + n data seg summary + orphan inode blocks */
1496         data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1497         spin_lock_irqsave(&sbi->cp_lock, flags);
1498         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1499                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1500         else
1501                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1502         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1503
1504         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1505         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1506                         orphan_blocks);
1507
1508         if (__remain_node_summaries(cpc->reason))
1509                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1510                                 cp_payload_blks + data_sum_blocks +
1511                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1512         else
1513                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1514                                 cp_payload_blks + data_sum_blocks +
1515                                 orphan_blocks);
1516
1517         /* update ckpt flag for checkpoint */
1518         update_ckpt_flags(sbi, cpc);
1519
1520         /* update SIT/NAT bitmap */
1521         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1522         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1523
1524         crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1525         *((__le32 *)((unsigned char *)ckpt +
1526                                 le32_to_cpu(ckpt->checksum_offset)))
1527                                 = cpu_to_le32(crc32);
1528
1529         start_blk = __start_cp_next_addr(sbi);
1530
1531         /* write nat bits */
1532         if ((cpc->reason & CP_UMOUNT) &&
1533                         is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1534                 __u64 cp_ver = cur_cp_version(ckpt);
1535                 block_t blk;
1536
1537                 cp_ver |= ((__u64)crc32 << 32);
1538                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1539
1540                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1541                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1542                         f2fs_update_meta_page(sbi, nm_i->nat_bits +
1543                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1544         }
1545
1546         /* write out checkpoint buffer at block 0 */
1547         f2fs_update_meta_page(sbi, ckpt, start_blk++);
1548
1549         for (i = 1; i < 1 + cp_payload_blks; i++)
1550                 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1551                                                         start_blk++);
1552
1553         if (orphan_num) {
1554                 write_orphan_inodes(sbi, start_blk);
1555                 start_blk += orphan_blocks;
1556         }
1557
1558         f2fs_write_data_summaries(sbi, start_blk);
1559         start_blk += data_sum_blocks;
1560
1561         /* Record write statistics in the hot node summary */
1562         kbytes_written = sbi->kbytes_written;
1563         kbytes_written += (f2fs_get_sectors_written(sbi) -
1564                                 sbi->sectors_written_start) >> 1;
1565         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1566
1567         if (__remain_node_summaries(cpc->reason)) {
1568                 f2fs_write_node_summaries(sbi, start_blk);
1569                 start_blk += NR_CURSEG_NODE_TYPE;
1570         }
1571
1572         /* update user_block_counts */
1573         sbi->last_valid_block_count = sbi->total_valid_block_count;
1574         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1575         percpu_counter_set(&sbi->rf_node_block_count, 0);
1576
1577         /* Here, we have one bio having CP pack except cp pack 2 page */
1578         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1579         /* Wait for all dirty meta pages to be submitted for IO */
1580         f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1581
1582         /* wait for previous submitted meta pages writeback */
1583         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1584
1585         /* flush all device cache */
1586         err = f2fs_flush_device_cache(sbi);
1587         if (err)
1588                 return err;
1589
1590         /* barrier and flush checkpoint cp pack 2 page if it can */
1591         commit_checkpoint(sbi, ckpt, start_blk);
1592         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1593
1594         /*
1595          * invalidate intermediate page cache borrowed from meta inode which are
1596          * used for migration of encrypted, verity or compressed inode's blocks.
1597          */
1598         if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1599                 f2fs_sb_has_compression(sbi))
1600                 invalidate_mapping_pages(META_MAPPING(sbi),
1601                                 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1602
1603         f2fs_release_ino_entry(sbi, false);
1604
1605         f2fs_reset_fsync_node_info(sbi);
1606
1607         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1608         clear_sbi_flag(sbi, SBI_NEED_CP);
1609         clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1610
1611         spin_lock(&sbi->stat_lock);
1612         sbi->unusable_block_count = 0;
1613         spin_unlock(&sbi->stat_lock);
1614
1615         __set_cp_next_pack(sbi);
1616
1617         /*
1618          * redirty superblock if metadata like node page or inode cache is
1619          * updated during writing checkpoint.
1620          */
1621         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1622                         get_pages(sbi, F2FS_DIRTY_IMETA))
1623                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1624
1625         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1626
1627         return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1628 }
1629
1630 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1631 {
1632         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1633         unsigned long long ckpt_ver;
1634         int err = 0;
1635
1636         if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1637                 return -EROFS;
1638
1639         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1640                 if (cpc->reason != CP_PAUSE)
1641                         return 0;
1642                 f2fs_warn(sbi, "Start checkpoint disabled!");
1643         }
1644         if (cpc->reason != CP_RESIZE)
1645                 f2fs_down_write(&sbi->cp_global_sem);
1646
1647         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1648                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1649                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1650                 goto out;
1651         if (unlikely(f2fs_cp_error(sbi))) {
1652                 err = -EIO;
1653                 goto out;
1654         }
1655
1656         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1657
1658         err = block_operations(sbi);
1659         if (err)
1660                 goto out;
1661
1662         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1663
1664         f2fs_flush_merged_writes(sbi);
1665
1666         /* this is the case of multiple fstrims without any changes */
1667         if (cpc->reason & CP_DISCARD) {
1668                 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1669                         unblock_operations(sbi);
1670                         goto out;
1671                 }
1672
1673                 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1674                                 SIT_I(sbi)->dirty_sentries == 0 &&
1675                                 prefree_segments(sbi) == 0) {
1676                         f2fs_flush_sit_entries(sbi, cpc);
1677                         f2fs_clear_prefree_segments(sbi, cpc);
1678                         unblock_operations(sbi);
1679                         goto out;
1680                 }
1681         }
1682
1683         /*
1684          * update checkpoint pack index
1685          * Increase the version number so that
1686          * SIT entries and seg summaries are written at correct place
1687          */
1688         ckpt_ver = cur_cp_version(ckpt);
1689         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1690
1691         /* write cached NAT/SIT entries to NAT/SIT area */
1692         err = f2fs_flush_nat_entries(sbi, cpc);
1693         if (err) {
1694                 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1695                 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1696                 goto stop;
1697         }
1698
1699         f2fs_flush_sit_entries(sbi, cpc);
1700
1701         /* save inmem log status */
1702         f2fs_save_inmem_curseg(sbi);
1703
1704         err = do_checkpoint(sbi, cpc);
1705         if (err) {
1706                 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1707                 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1708                 f2fs_release_discard_addrs(sbi);
1709         } else {
1710                 f2fs_clear_prefree_segments(sbi, cpc);
1711         }
1712
1713         f2fs_restore_inmem_curseg(sbi);
1714 stop:
1715         unblock_operations(sbi);
1716         stat_inc_cp_count(sbi->stat_info);
1717
1718         if (cpc->reason & CP_RECOVERY)
1719                 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1720
1721         /* update CP_TIME to trigger checkpoint periodically */
1722         f2fs_update_time(sbi, CP_TIME);
1723         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1724 out:
1725         if (cpc->reason != CP_RESIZE)
1726                 f2fs_up_write(&sbi->cp_global_sem);
1727         return err;
1728 }
1729
1730 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1731 {
1732         int i;
1733
1734         for (i = 0; i < MAX_INO_ENTRY; i++) {
1735                 struct inode_management *im = &sbi->im[i];
1736
1737                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1738                 spin_lock_init(&im->ino_lock);
1739                 INIT_LIST_HEAD(&im->ino_list);
1740                 im->ino_num = 0;
1741         }
1742
1743         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1744                         NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1745                                 F2FS_ORPHANS_PER_BLOCK;
1746 }
1747
1748 int __init f2fs_create_checkpoint_caches(void)
1749 {
1750         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1751                         sizeof(struct ino_entry));
1752         if (!ino_entry_slab)
1753                 return -ENOMEM;
1754         f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1755                         sizeof(struct inode_entry));
1756         if (!f2fs_inode_entry_slab) {
1757                 kmem_cache_destroy(ino_entry_slab);
1758                 return -ENOMEM;
1759         }
1760         return 0;
1761 }
1762
1763 void f2fs_destroy_checkpoint_caches(void)
1764 {
1765         kmem_cache_destroy(ino_entry_slab);
1766         kmem_cache_destroy(f2fs_inode_entry_slab);
1767 }
1768
1769 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1770 {
1771         struct cp_control cpc = { .reason = CP_SYNC, };
1772         int err;
1773
1774         f2fs_down_write(&sbi->gc_lock);
1775         err = f2fs_write_checkpoint(sbi, &cpc);
1776         f2fs_up_write(&sbi->gc_lock);
1777
1778         return err;
1779 }
1780
1781 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1782 {
1783         struct ckpt_req_control *cprc = &sbi->cprc_info;
1784         struct ckpt_req *req, *next;
1785         struct llist_node *dispatch_list;
1786         u64 sum_diff = 0, diff, count = 0;
1787         int ret;
1788
1789         dispatch_list = llist_del_all(&cprc->issue_list);
1790         if (!dispatch_list)
1791                 return;
1792         dispatch_list = llist_reverse_order(dispatch_list);
1793
1794         ret = __write_checkpoint_sync(sbi);
1795         atomic_inc(&cprc->issued_ckpt);
1796
1797         llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1798                 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1799                 req->ret = ret;
1800                 complete(&req->wait);
1801
1802                 sum_diff += diff;
1803                 count++;
1804         }
1805         atomic_sub(count, &cprc->queued_ckpt);
1806         atomic_add(count, &cprc->total_ckpt);
1807
1808         spin_lock(&cprc->stat_lock);
1809         cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1810         if (cprc->peak_time < cprc->cur_time)
1811                 cprc->peak_time = cprc->cur_time;
1812         spin_unlock(&cprc->stat_lock);
1813 }
1814
1815 static int issue_checkpoint_thread(void *data)
1816 {
1817         struct f2fs_sb_info *sbi = data;
1818         struct ckpt_req_control *cprc = &sbi->cprc_info;
1819         wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1820 repeat:
1821         if (kthread_should_stop())
1822                 return 0;
1823
1824         if (!llist_empty(&cprc->issue_list))
1825                 __checkpoint_and_complete_reqs(sbi);
1826
1827         wait_event_interruptible(*q,
1828                 kthread_should_stop() || !llist_empty(&cprc->issue_list));
1829         goto repeat;
1830 }
1831
1832 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1833                 struct ckpt_req *wait_req)
1834 {
1835         struct ckpt_req_control *cprc = &sbi->cprc_info;
1836
1837         if (!llist_empty(&cprc->issue_list)) {
1838                 __checkpoint_and_complete_reqs(sbi);
1839         } else {
1840                 /* already dispatched by issue_checkpoint_thread */
1841                 if (wait_req)
1842                         wait_for_completion(&wait_req->wait);
1843         }
1844 }
1845
1846 static void init_ckpt_req(struct ckpt_req *req)
1847 {
1848         memset(req, 0, sizeof(struct ckpt_req));
1849
1850         init_completion(&req->wait);
1851         req->queue_time = ktime_get();
1852 }
1853
1854 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1855 {
1856         struct ckpt_req_control *cprc = &sbi->cprc_info;
1857         struct ckpt_req req;
1858         struct cp_control cpc;
1859
1860         cpc.reason = __get_cp_reason(sbi);
1861         if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1862                 int ret;
1863
1864                 f2fs_down_write(&sbi->gc_lock);
1865                 ret = f2fs_write_checkpoint(sbi, &cpc);
1866                 f2fs_up_write(&sbi->gc_lock);
1867
1868                 return ret;
1869         }
1870
1871         if (!cprc->f2fs_issue_ckpt)
1872                 return __write_checkpoint_sync(sbi);
1873
1874         init_ckpt_req(&req);
1875
1876         llist_add(&req.llnode, &cprc->issue_list);
1877         atomic_inc(&cprc->queued_ckpt);
1878
1879         /*
1880          * update issue_list before we wake up issue_checkpoint thread,
1881          * this smp_mb() pairs with another barrier in ___wait_event(),
1882          * see more details in comments of waitqueue_active().
1883          */
1884         smp_mb();
1885
1886         if (waitqueue_active(&cprc->ckpt_wait_queue))
1887                 wake_up(&cprc->ckpt_wait_queue);
1888
1889         if (cprc->f2fs_issue_ckpt)
1890                 wait_for_completion(&req.wait);
1891         else
1892                 flush_remained_ckpt_reqs(sbi, &req);
1893
1894         return req.ret;
1895 }
1896
1897 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1898 {
1899         dev_t dev = sbi->sb->s_bdev->bd_dev;
1900         struct ckpt_req_control *cprc = &sbi->cprc_info;
1901
1902         if (cprc->f2fs_issue_ckpt)
1903                 return 0;
1904
1905         cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1906                         "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1907         if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1908                 cprc->f2fs_issue_ckpt = NULL;
1909                 return -ENOMEM;
1910         }
1911
1912         set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1913
1914         return 0;
1915 }
1916
1917 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1918 {
1919         struct ckpt_req_control *cprc = &sbi->cprc_info;
1920         struct task_struct *ckpt_task;
1921
1922         if (!cprc->f2fs_issue_ckpt)
1923                 return;
1924
1925         ckpt_task = cprc->f2fs_issue_ckpt;
1926         cprc->f2fs_issue_ckpt = NULL;
1927         kthread_stop(ckpt_task);
1928
1929         f2fs_flush_ckpt_thread(sbi);
1930 }
1931
1932 void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1933 {
1934         struct ckpt_req_control *cprc = &sbi->cprc_info;
1935
1936         flush_remained_ckpt_reqs(sbi, NULL);
1937
1938         /* Let's wait for the previous dispatched checkpoint. */
1939         while (atomic_read(&cprc->queued_ckpt))
1940                 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1941 }
1942
1943 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1944 {
1945         struct ckpt_req_control *cprc = &sbi->cprc_info;
1946
1947         atomic_set(&cprc->issued_ckpt, 0);
1948         atomic_set(&cprc->total_ckpt, 0);
1949         atomic_set(&cprc->queued_ckpt, 0);
1950         cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1951         init_waitqueue_head(&cprc->ckpt_wait_queue);
1952         init_llist_head(&cprc->issue_list);
1953         spin_lock_init(&cprc->stat_lock);
1954 }