fs/f2fs/checkpoint.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/checkpoint.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/bio.h>
  10 #include <linux/mpage.h>
  11 #include <linux/writeback.h>
  12 #include <linux/blkdev.h>
  13 #include <linux/f2fs_fs.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/swap.h>
  16 #include <linux/kthread.h>
  17
  18 #include "f2fs.h"
  19 #include "node.h"
  20 #include "segment.h"
  21 #include "iostat.h"
  22 #include <trace/events/f2fs.h>
  23
  24 #define DEFAULT_CHECKPOINT_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
  25
  26 static struct kmem_cache *ino_entry_slab;
  27 struct kmem_cache *f2fs_inode_entry_slab;
  28
  29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
  30 {
  31         f2fs_build_fault_attr(sbi, 0, 0);
  32         set_ckpt_flags(sbi, CP_ERROR_FLAG);
  33         if (!end_io)
  34                 f2fs_flush_merged_writes(sbi);
  35 }
  36
  37 /*
  38  * We guarantee no failure on the returned page.
  39  */
  40 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  41 {
  42         struct address_space *mapping = META_MAPPING(sbi);
  43         struct page *page;
  44 repeat:
  45         page = f2fs_grab_cache_page(mapping, index, false);
  46         if (!page) {
  47                 cond_resched();
  48                 goto repeat;
  49         }
  50         f2fs_wait_on_page_writeback(page, META, true, true);
  51         if (!PageUptodate(page))
  52                 SetPageUptodate(page);
  53         return page;
  54 }
  55
  56 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
  57                                                         bool is_meta)
  58 {
  59         struct address_space *mapping = META_MAPPING(sbi);
  60         struct page *page;
  61         struct f2fs_io_info fio = {
  62                 .sbi = sbi,
  63                 .type = META,
  64                 .op = REQ_OP_READ,
  65                 .op_flags = REQ_META | REQ_PRIO,
  66                 .old_blkaddr = index,
  67                 .new_blkaddr = index,
  68                 .encrypted_page = NULL,
  69                 .is_por = !is_meta,
  70         };
  71         int err;
  72
  73         if (unlikely(!is_meta))
  74                 fio.op_flags &= ~REQ_META;
  75 repeat:
  76         page = f2fs_grab_cache_page(mapping, index, false);
  77         if (!page) {
  78                 cond_resched();
  79                 goto repeat;
  80         }
  81         if (PageUptodate(page))
  82                 goto out;
  83
  84         fio.page = page;
  85
  86         err = f2fs_submit_page_bio(&fio);
  87         if (err) {
  88                 f2fs_put_page(page, 1);
  89                 return ERR_PTR(err);
  90         }
  91
  92         f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
  93
  94         lock_page(page);
  95         if (unlikely(page->mapping != mapping)) {
  96                 f2fs_put_page(page, 1);
  97                 goto repeat;
  98         }
  99
 100         if (unlikely(!PageUptodate(page))) {
 101                 f2fs_handle_page_eio(sbi, page->index, META);
 102                 f2fs_put_page(page, 1);
 103                 return ERR_PTR(-EIO);
 104         }
 105 out:
 106         return page;
 107 }
 108
 109 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 110 {
 111         return __get_meta_page(sbi, index, true);
 112 }
 113
 114 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
 115 {
 116         struct page *page;
 117         int count = 0;
 118
 119 retry:
 120         page = __get_meta_page(sbi, index, true);
 121         if (IS_ERR(page)) {
 122                 if (PTR_ERR(page) == -EIO &&
 123                                 ++count <= DEFAULT_RETRY_IO_COUNT)
 124                         goto retry;
 125                 f2fs_stop_checkpoint(sbi, false);
 126         }
 127         return page;
 128 }
 129
 130 /* for POR only */
 131 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 132 {
 133         return __get_meta_page(sbi, index, false);
 134 }
 135
 136 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
 137                                                         int type)
 138 {
 139         struct seg_entry *se;
 140         unsigned int segno, offset;
 141         bool exist;
 142
 143         if (type != DATA_GENERIC_ENHANCE && type != DATA_GENERIC_ENHANCE_READ)
 144                 return true;
 145
 146         segno = GET_SEGNO(sbi, blkaddr);
 147         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 148         se = get_seg_entry(sbi, segno);
 149
 150         exist = f2fs_test_bit(offset, se->cur_valid_map);
 151         if (!exist && type == DATA_GENERIC_ENHANCE) {
 152                 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
 153                          blkaddr, exist);
 154                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 155                 dump_stack();
 156         }
 157         return exist;
 158 }
 159
 160 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
 161                                         block_t blkaddr, int type)
 162 {
 163         switch (type) {
 164         case META_NAT:
 165                 break;
 166         case META_SIT:
 167                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 168                         return false;
 169                 break;
 170         case META_SSA:
 171                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 172                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 173                         return false;
 174                 break;
 175         case META_CP:
 176                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 177                         blkaddr < __start_cp_addr(sbi)))
 178                         return false;
 179                 break;
 180         case META_POR:
 181                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 182                         blkaddr < MAIN_BLKADDR(sbi)))
 183                         return false;
 184                 break;
 185         case DATA_GENERIC:
 186         case DATA_GENERIC_ENHANCE:
 187         case DATA_GENERIC_ENHANCE_READ:
 188                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 189                                 blkaddr < MAIN_BLKADDR(sbi))) {
 190                         f2fs_warn(sbi, "access invalid blkaddr:%u",
 191                                   blkaddr);
 192                         set_sbi_flag(sbi, SBI_NEED_FSCK);
 193                         dump_stack();
 194                         return false;
 195                 } else {
 196                         return __is_bitmap_valid(sbi, blkaddr, type);
 197                 }
 198                 break;
 199         case META_GENERIC:
 200                 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
 201                         blkaddr >= MAIN_BLKADDR(sbi)))
 202                         return false;
 203                 break;
 204         default:
 205                 BUG();
 206         }
 207
 208         return true;
 209 }
 210
 211 /*
 212  * Readahead CP/NAT/SIT/SSA/POR pages
 213  */
 214 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 215                                                         int type, bool sync)
 216 {
 217         struct page *page;
 218         block_t blkno = start;
 219         struct f2fs_io_info fio = {
 220                 .sbi = sbi,
 221                 .type = META,
 222                 .op = REQ_OP_READ,
 223                 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
 224                 .encrypted_page = NULL,
 225                 .in_list = false,
 226                 .is_por = (type == META_POR),
 227         };
 228         struct blk_plug plug;
 229         int err;
 230
 231         if (unlikely(type == META_POR))
 232                 fio.op_flags &= ~REQ_META;
 233
 234         blk_start_plug(&plug);
 235         for (; nrpages-- > 0; blkno++) {
 236
 237                 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
 238                         goto out;
 239
 240                 switch (type) {
 241                 case META_NAT:
 242                         if (unlikely(blkno >=
 243                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 244                                 blkno = 0;
 245                         /* get nat block addr */
 246                         fio.new_blkaddr = current_nat_addr(sbi,
 247                                         blkno * NAT_ENTRY_PER_BLOCK);
 248                         break;
 249                 case META_SIT:
 250                         if (unlikely(blkno >= TOTAL_SEGS(sbi)))
 251                                 goto out;
 252                         /* get sit block addr */
 253                         fio.new_blkaddr = current_sit_addr(sbi,
 254                                         blkno * SIT_ENTRY_PER_BLOCK);
 255                         break;
 256                 case META_SSA:
 257                 case META_CP:
 258                 case META_POR:
 259                         fio.new_blkaddr = blkno;
 260                         break;
 261                 default:
 262                         BUG();
 263                 }
 264
 265                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 266                                                 fio.new_blkaddr, false);
 267                 if (!page)
 268                         continue;
 269                 if (PageUptodate(page)) {
 270                         f2fs_put_page(page, 1);
 271                         continue;
 272                 }
 273
 274                 fio.page = page;
 275                 err = f2fs_submit_page_bio(&fio);
 276                 f2fs_put_page(page, err ? 1 : 0);
 277
 278                 if (!err)
 279                         f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
 280         }
 281 out:
 282         blk_finish_plug(&plug);
 283         return blkno - start;
 284 }
 285
 286 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
 287                                                         unsigned int ra_blocks)
 288 {
 289         struct page *page;
 290         bool readahead = false;
 291
 292         if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
 293                 return;
 294
 295         page = find_get_page(META_MAPPING(sbi), index);
 296         if (!page || !PageUptodate(page))
 297                 readahead = true;
 298         f2fs_put_page(page, 0);
 299
 300         if (readahead)
 301                 f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
 302 }
 303
 304 static int __f2fs_write_meta_page(struct page *page,
 305                                 struct writeback_control *wbc,
 306                                 enum iostat_type io_type)
 307 {
 308         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 309
 310         trace_f2fs_writepage(page, META);
 311
 312         if (unlikely(f2fs_cp_error(sbi)))
 313                 goto redirty_out;
 314         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 315                 goto redirty_out;
 316         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 317                 goto redirty_out;
 318
 319         f2fs_do_write_meta_page(sbi, page, io_type);
 320         dec_page_count(sbi, F2FS_DIRTY_META);
 321
 322         if (wbc->for_reclaim)
 323                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
 324
 325         unlock_page(page);
 326
 327         if (unlikely(f2fs_cp_error(sbi)))
 328                 f2fs_submit_merged_write(sbi, META);
 329
 330         return 0;
 331
 332 redirty_out:
 333         redirty_page_for_writepage(wbc, page);
 334         return AOP_WRITEPAGE_ACTIVATE;
 335 }
 336
 337 static int f2fs_write_meta_page(struct page *page,
 338                                 struct writeback_control *wbc)
 339 {
 340         return __f2fs_write_meta_page(page, wbc, FS_META_IO);
 341 }
 342
 343 static int f2fs_write_meta_pages(struct address_space *mapping,
 344                                 struct writeback_control *wbc)
 345 {
 346         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 347         long diff, written;
 348
 349         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 350                 goto skip_write;
 351
 352         /* collect a number of dirty meta pages and write together */
 353         if (wbc->sync_mode != WB_SYNC_ALL &&
 354                         get_pages(sbi, F2FS_DIRTY_META) <
 355                                         nr_pages_to_skip(sbi, META))
 356                 goto skip_write;
 357
 358         /* if locked failed, cp will flush dirty pages instead */
 359         if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
 360                 goto skip_write;
 361
 362         trace_f2fs_writepages(mapping->host, wbc, META);
 363         diff = nr_pages_to_write(sbi, META, wbc);
 364         written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
 365         f2fs_up_write(&sbi->cp_global_sem);
 366         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 367         return 0;
 368
 369 skip_write:
 370         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 371         trace_f2fs_writepages(mapping->host, wbc, META);
 372         return 0;
 373 }
 374
 375 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 376                                 long nr_to_write, enum iostat_type io_type)
 377 {
 378         struct address_space *mapping = META_MAPPING(sbi);
 379         pgoff_t index = 0, prev = ULONG_MAX;
 380         struct pagevec pvec;
 381         long nwritten = 0;
 382         int nr_pages;
 383         struct writeback_control wbc = {
 384                 .for_reclaim = 0,
 385         };
 386         struct blk_plug plug;
 387
 388         pagevec_init(&pvec);
 389
 390         blk_start_plug(&plug);
 391
 392         while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 393                                 PAGECACHE_TAG_DIRTY))) {
 394                 int i;
 395
 396                 for (i = 0; i < nr_pages; i++) {
 397                         struct page *page = pvec.pages[i];
 398
 399                         if (prev == ULONG_MAX)
 400                                 prev = page->index - 1;
 401                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 402                                 pagevec_release(&pvec);
 403                                 goto stop;
 404                         }
 405
 406                         lock_page(page);
 407
 408                         if (unlikely(page->mapping != mapping)) {
 409 continue_unlock:
 410                                 unlock_page(page);
 411                                 continue;
 412                         }
 413                         if (!PageDirty(page)) {
 414                                 /* someone wrote it for us */
 415                                 goto continue_unlock;
 416                         }
 417
 418                         f2fs_wait_on_page_writeback(page, META, true, true);
 419
 420                         if (!clear_page_dirty_for_io(page))
 421                                 goto continue_unlock;
 422
 423                         if (__f2fs_write_meta_page(page, &wbc, io_type)) {
 424                                 unlock_page(page);
 425                                 break;
 426                         }
 427                         nwritten++;
 428                         prev = page->index;
 429                         if (unlikely(nwritten >= nr_to_write))
 430                                 break;
 431                 }
 432                 pagevec_release(&pvec);
 433                 cond_resched();
 434         }
 435 stop:
 436         if (nwritten)
 437                 f2fs_submit_merged_write(sbi, type);
 438
 439         blk_finish_plug(&plug);
 440
 441         return nwritten;
 442 }
 443
 444 static bool f2fs_dirty_meta_folio(struct address_space *mapping,
 445                 struct folio *folio)
 446 {
 447         trace_f2fs_set_page_dirty(&folio->page, META);
 448
 449         if (!folio_test_uptodate(folio))
 450                 folio_mark_uptodate(folio);
 451         if (!folio_test_dirty(folio)) {
 452                 filemap_dirty_folio(mapping, folio);
 453                 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
 454                 set_page_private_reference(&folio->page);
 455                 return true;
 456         }
 457         return false;
 458 }
 459
 460 const struct address_space_operations f2fs_meta_aops = {
 461         .writepage      = f2fs_write_meta_page,
 462         .writepages     = f2fs_write_meta_pages,
 463         .dirty_folio    = f2fs_dirty_meta_folio,
 464         .invalidate_folio = f2fs_invalidate_folio,
 465         .release_folio  = f2fs_release_folio,
 466 #ifdef CONFIG_MIGRATION
 467         .migratepage    = f2fs_migrate_page,
 468 #endif
 469 };
 470
 471 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 472                                                 unsigned int devidx, int type)
 473 {
 474         struct inode_management *im = &sbi->im[type];
 475         struct ino_entry *e = NULL, *new = NULL;
 476
 477         if (type == FLUSH_INO) {
 478                 rcu_read_lock();
 479                 e = radix_tree_lookup(&im->ino_root, ino);
 480                 rcu_read_unlock();
 481         }
 482
 483 retry:
 484         if (!e)
 485                 new = f2fs_kmem_cache_alloc(ino_entry_slab,
 486                                                 GFP_NOFS, true, NULL);
 487
 488         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 489
 490         spin_lock(&im->ino_lock);
 491         e = radix_tree_lookup(&im->ino_root, ino);
 492         if (!e) {
 493                 if (!new) {
 494                         spin_unlock(&im->ino_lock);
 495                         goto retry;
 496                 }
 497                 e = new;
 498                 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
 499                         f2fs_bug_on(sbi, 1);
 500
 501                 memset(e, 0, sizeof(struct ino_entry));
 502                 e->ino = ino;
 503
 504                 list_add_tail(&e->list, &im->ino_list);
 505                 if (type != ORPHAN_INO)
 506                         im->ino_num++;
 507         }
 508
 509         if (type == FLUSH_INO)
 510                 f2fs_set_bit(devidx, (char *)&e->dirty_device);
 511
 512         spin_unlock(&im->ino_lock);
 513         radix_tree_preload_end();
 514
 515         if (new && e != new)
 516                 kmem_cache_free(ino_entry_slab, new);
 517 }
 518
 519 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 520 {
 521         struct inode_management *im = &sbi->im[type];
 522         struct ino_entry *e;
 523
 524         spin_lock(&im->ino_lock);
 525         e = radix_tree_lookup(&im->ino_root, ino);
 526         if (e) {
 527                 list_del(&e->list);
 528                 radix_tree_delete(&im->ino_root, ino);
 529                 im->ino_num--;
 530                 spin_unlock(&im->ino_lock);
 531                 kmem_cache_free(ino_entry_slab, e);
 532                 return;
 533         }
 534         spin_unlock(&im->ino_lock);
 535 }
 536
 537 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 538 {
 539         /* add new dirty ino entry into list */
 540         __add_ino_entry(sbi, ino, 0, type);
 541 }
 542
 543 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 544 {
 545         /* remove dirty ino entry from list */
 546         __remove_ino_entry(sbi, ino, type);
 547 }
 548
 549 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
 550 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 551 {
 552         struct inode_management *im = &sbi->im[mode];
 553         struct ino_entry *e;
 554
 555         spin_lock(&im->ino_lock);
 556         e = radix_tree_lookup(&im->ino_root, ino);
 557         spin_unlock(&im->ino_lock);
 558         return e ? true : false;
 559 }
 560
 561 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 562 {
 563         struct ino_entry *e, *tmp;
 564         int i;
 565
 566         for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
 567                 struct inode_management *im = &sbi->im[i];
 568
 569                 spin_lock(&im->ino_lock);
 570                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 571                         list_del(&e->list);
 572                         radix_tree_delete(&im->ino_root, e->ino);
 573                         kmem_cache_free(ino_entry_slab, e);
 574                         im->ino_num--;
 575                 }
 576                 spin_unlock(&im->ino_lock);
 577         }
 578 }
 579
 580 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 581                                         unsigned int devidx, int type)
 582 {
 583         __add_ino_entry(sbi, ino, devidx, type);
 584 }
 585
 586 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 587                                         unsigned int devidx, int type)
 588 {
 589         struct inode_management *im = &sbi->im[type];
 590         struct ino_entry *e;
 591         bool is_dirty = false;
 592
 593         spin_lock(&im->ino_lock);
 594         e = radix_tree_lookup(&im->ino_root, ino);
 595         if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
 596                 is_dirty = true;
 597         spin_unlock(&im->ino_lock);
 598         return is_dirty;
 599 }
 600
 601 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
 602 {
 603         struct inode_management *im = &sbi->im[ORPHAN_INO];
 604         int err = 0;
 605
 606         spin_lock(&im->ino_lock);
 607
 608         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 609                 spin_unlock(&im->ino_lock);
 610                 f2fs_show_injection_info(sbi, FAULT_ORPHAN);
 611                 return -ENOSPC;
 612         }
 613
 614         if (unlikely(im->ino_num >= sbi->max_orphans))
 615                 err = -ENOSPC;
 616         else
 617                 im->ino_num++;
 618         spin_unlock(&im->ino_lock);
 619
 620         return err;
 621 }
 622
 623 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
 624 {
 625         struct inode_management *im = &sbi->im[ORPHAN_INO];
 626
 627         spin_lock(&im->ino_lock);
 628         f2fs_bug_on(sbi, im->ino_num == 0);
 629         im->ino_num--;
 630         spin_unlock(&im->ino_lock);
 631 }
 632
 633 void f2fs_add_orphan_inode(struct inode *inode)
 634 {
 635         /* add new orphan ino entry into list */
 636         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
 637         f2fs_update_inode_page(inode);
 638 }
 639
 640 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 641 {
 642         /* remove orphan entry from orphan list */
 643         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 644 }
 645
 646 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 647 {
 648         struct inode *inode;
 649         struct node_info ni;
 650         int err;
 651
 652         inode = f2fs_iget_retry(sbi->sb, ino);
 653         if (IS_ERR(inode)) {
 654                 /*
 655                  * there should be a bug that we can't find the entry
 656                  * to orphan inode.
 657                  */
 658                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 659                 return PTR_ERR(inode);
 660         }
 661
 662         err = f2fs_dquot_initialize(inode);
 663         if (err) {
 664                 iput(inode);
 665                 goto err_out;
 666         }
 667
 668         clear_nlink(inode);
 669
 670         /* truncate all the data during iput */
 671         iput(inode);
 672
 673         err = f2fs_get_node_info(sbi, ino, &ni, false);
 674         if (err)
 675                 goto err_out;
 676
 677         /* ENOMEM was fully retried in f2fs_evict_inode. */
 678         if (ni.blk_addr != NULL_ADDR) {
 679                 err = -EIO;
 680                 goto err_out;
 681         }
 682         return 0;
 683
 684 err_out:
 685         set_sbi_flag(sbi, SBI_NEED_FSCK);
 686         f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
 687                   __func__, ino);
 688         return err;
 689 }
 690
 691 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
 692 {
 693         block_t start_blk, orphan_blocks, i, j;
 694         unsigned int s_flags = sbi->sb->s_flags;
 695         int err = 0;
 696 #ifdef CONFIG_QUOTA
 697         int quota_enabled;
 698 #endif
 699
 700         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 701                 return 0;
 702
 703         if (bdev_read_only(sbi->sb->s_bdev)) {
 704                 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
 705                 return 0;
 706         }
 707
 708         if (s_flags & SB_RDONLY) {
 709                 f2fs_info(sbi, "orphan cleanup on readonly fs");
 710                 sbi->sb->s_flags &= ~SB_RDONLY;
 711         }
 712
 713 #ifdef CONFIG_QUOTA
 714         /*
 715          * Turn on quotas which were not enabled for read-only mounts if
 716          * filesystem has quota feature, so that they are updated correctly.
 717          */
 718         quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
 719 #endif
 720
 721         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 722         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 723
 724         f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 725
 726         for (i = 0; i < orphan_blocks; i++) {
 727                 struct page *page;
 728                 struct f2fs_orphan_block *orphan_blk;
 729
 730                 page = f2fs_get_meta_page(sbi, start_blk + i);
 731                 if (IS_ERR(page)) {
 732                         err = PTR_ERR(page);
 733                         goto out;
 734                 }
 735
 736                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 737                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 738                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 739
 740                         err = recover_orphan_inode(sbi, ino);
 741                         if (err) {
 742                                 f2fs_put_page(page, 1);
 743                                 goto out;
 744                         }
 745                 }
 746                 f2fs_put_page(page, 1);
 747         }
 748         /* clear Orphan Flag */
 749         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 750 out:
 751         set_sbi_flag(sbi, SBI_IS_RECOVERED);
 752
 753 #ifdef CONFIG_QUOTA
 754         /* Turn quotas off */
 755         if (quota_enabled)
 756                 f2fs_quota_off_umount(sbi->sb);
 757 #endif
 758         sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
 759
 760         return err;
 761 }
 762
 763 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 764 {
 765         struct list_head *head;
 766         struct f2fs_orphan_block *orphan_blk = NULL;
 767         unsigned int nentries = 0;
 768         unsigned short index = 1;
 769         unsigned short orphan_blocks;
 770         struct page *page = NULL;
 771         struct ino_entry *orphan = NULL;
 772         struct inode_management *im = &sbi->im[ORPHAN_INO];
 773
 774         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 775
 776         /*
 777          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 778          * orphan inode operations are covered under f2fs_lock_op().
 779          * And, spin_lock should be avoided due to page operations below.
 780          */
 781         head = &im->ino_list;
 782
 783         /* loop for each orphan inode entry and write them in Jornal block */
 784         list_for_each_entry(orphan, head, list) {
 785                 if (!page) {
 786                         page = f2fs_grab_meta_page(sbi, start_blk++);
 787                         orphan_blk =
 788                                 (struct f2fs_orphan_block *)page_address(page);
 789                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 790                 }
 791
 792                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 793
 794                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 795                         /*
 796                          * an orphan block is full of 1020 entries,
 797                          * then we need to flush current orphan blocks
 798                          * and bring another one in memory
 799                          */
 800                         orphan_blk->blk_addr = cpu_to_le16(index);
 801                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 802                         orphan_blk->entry_count = cpu_to_le32(nentries);
 803                         set_page_dirty(page);
 804                         f2fs_put_page(page, 1);
 805                         index++;
 806                         nentries = 0;
 807                         page = NULL;
 808                 }
 809         }
 810
 811         if (page) {
 812                 orphan_blk->blk_addr = cpu_to_le16(index);
 813                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 814                 orphan_blk->entry_count = cpu_to_le32(nentries);
 815                 set_page_dirty(page);
 816                 f2fs_put_page(page, 1);
 817         }
 818 }
 819
 820 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
 821                                                 struct f2fs_checkpoint *ckpt)
 822 {
 823         unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
 824         __u32 chksum;
 825
 826         chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
 827         if (chksum_ofs < CP_CHKSUM_OFFSET) {
 828                 chksum_ofs += sizeof(chksum);
 829                 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
 830                                                 F2FS_BLKSIZE - chksum_ofs);
 831         }
 832         return chksum;
 833 }
 834
 835 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 836                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 837                 unsigned long long *version)
 838 {
 839         size_t crc_offset = 0;
 840         __u32 crc;
 841
 842         *cp_page = f2fs_get_meta_page(sbi, cp_addr);
 843         if (IS_ERR(*cp_page))
 844                 return PTR_ERR(*cp_page);
 845
 846         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 847
 848         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 849         if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
 850                         crc_offset > CP_CHKSUM_OFFSET) {
 851                 f2fs_put_page(*cp_page, 1);
 852                 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
 853                 return -EINVAL;
 854         }
 855
 856         crc = f2fs_checkpoint_chksum(sbi, *cp_block);
 857         if (crc != cur_cp_crc(*cp_block)) {
 858                 f2fs_put_page(*cp_page, 1);
 859                 f2fs_warn(sbi, "invalid crc value");
 860                 return -EINVAL;
 861         }
 862
 863         *version = cur_cp_version(*cp_block);
 864         return 0;
 865 }
 866
 867 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 868                                 block_t cp_addr, unsigned long long *version)
 869 {
 870         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 871         struct f2fs_checkpoint *cp_block = NULL;
 872         unsigned long long cur_version = 0, pre_version = 0;
 873         unsigned int cp_blocks;
 874         int err;
 875
 876         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 877                                         &cp_page_1, version);
 878         if (err)
 879                 return NULL;
 880
 881         cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
 882
 883         if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
 884                 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
 885                           le32_to_cpu(cp_block->cp_pack_total_block_count));
 886                 goto invalid_cp;
 887         }
 888         pre_version = *version;
 889
 890         cp_addr += cp_blocks - 1;
 891         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 892                                         &cp_page_2, version);
 893         if (err)
 894                 goto invalid_cp;
 895         cur_version = *version;
 896
 897         if (cur_version == pre_version) {
 898                 *version = cur_version;
 899                 f2fs_put_page(cp_page_2, 1);
 900                 return cp_page_1;
 901         }
 902         f2fs_put_page(cp_page_2, 1);
 903 invalid_cp:
 904         f2fs_put_page(cp_page_1, 1);
 905         return NULL;
 906 }
 907
 908 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
 909 {
 910         struct f2fs_checkpoint *cp_block;
 911         struct f2fs_super_block *fsb = sbi->raw_super;
 912         struct page *cp1, *cp2, *cur_page;
 913         unsigned long blk_size = sbi->blocksize;
 914         unsigned long long cp1_version = 0, cp2_version = 0;
 915         unsigned long long cp_start_blk_no;
 916         unsigned int cp_blks = 1 + __cp_payload(sbi);
 917         block_t cp_blk_no;
 918         int i;
 919         int err;
 920
 921         sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
 922                                   GFP_KERNEL);
 923         if (!sbi->ckpt)
 924                 return -ENOMEM;
 925         /*
 926          * Finding out valid cp block involves read both
 927          * sets( cp pack 1 and cp pack 2)
 928          */
 929         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 930         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 931
 932         /* The second checkpoint pack should start at the next segment */
 933         cp_start_blk_no += ((unsigned long long)1) <<
 934                                 le32_to_cpu(fsb->log_blocks_per_seg);
 935         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 936
 937         if (cp1 && cp2) {
 938                 if (ver_after(cp2_version, cp1_version))
 939                         cur_page = cp2;
 940                 else
 941                         cur_page = cp1;
 942         } else if (cp1) {
 943                 cur_page = cp1;
 944         } else if (cp2) {
 945                 cur_page = cp2;
 946         } else {
 947                 err = -EFSCORRUPTED;
 948                 goto fail_no_cp;
 949         }
 950
 951         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 952         memcpy(sbi->ckpt, cp_block, blk_size);
 953
 954         if (cur_page == cp1)
 955                 sbi->cur_cp_pack = 1;
 956         else
 957                 sbi->cur_cp_pack = 2;
 958
 959         /* Sanity checking of checkpoint */
 960         if (f2fs_sanity_check_ckpt(sbi)) {
 961                 err = -EFSCORRUPTED;
 962                 goto free_fail_no_cp;
 963         }
 964
 965         if (cp_blks <= 1)
 966                 goto done;
 967
 968         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 969         if (cur_page == cp2)
 970                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 971
 972         for (i = 1; i < cp_blks; i++) {
 973                 void *sit_bitmap_ptr;
 974                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 975
 976                 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
 977                 if (IS_ERR(cur_page)) {
 978                         err = PTR_ERR(cur_page);
 979                         goto free_fail_no_cp;
 980                 }
 981                 sit_bitmap_ptr = page_address(cur_page);
 982                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 983                 f2fs_put_page(cur_page, 1);
 984         }
 985 done:
 986         f2fs_put_page(cp1, 1);
 987         f2fs_put_page(cp2, 1);
 988         return 0;
 989
 990 free_fail_no_cp:
 991         f2fs_put_page(cp1, 1);
 992         f2fs_put_page(cp2, 1);
 993 fail_no_cp:
 994         kvfree(sbi->ckpt);
 995         return err;
 996 }
 997
 998 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 999 {
1000         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1001         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1002
1003         if (is_inode_flag_set(inode, flag))
1004                 return;
1005
1006         set_inode_flag(inode, flag);
1007         list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1008         stat_inc_dirty_inode(sbi, type);
1009 }
1010
1011 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1012 {
1013         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1014
1015         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1016                 return;
1017
1018         list_del_init(&F2FS_I(inode)->dirty_list);
1019         clear_inode_flag(inode, flag);
1020         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1021 }
1022
1023 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1024 {
1025         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1026         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1027
1028         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1029                         !S_ISLNK(inode->i_mode))
1030                 return;
1031
1032         spin_lock(&sbi->inode_lock[type]);
1033         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1034                 __add_dirty_inode(inode, type);
1035         inode_inc_dirty_pages(inode);
1036         spin_unlock(&sbi->inode_lock[type]);
1037
1038         set_page_private_reference(&folio->page);
1039 }
1040
1041 void f2fs_remove_dirty_inode(struct inode *inode)
1042 {
1043         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1044         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1045
1046         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1047                         !S_ISLNK(inode->i_mode))
1048                 return;
1049
1050         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1051                 return;
1052
1053         spin_lock(&sbi->inode_lock[type]);
1054         __remove_dirty_inode(inode, type);
1055         spin_unlock(&sbi->inode_lock[type]);
1056 }
1057
1058 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1059 {
1060         struct list_head *head;
1061         struct inode *inode;
1062         struct f2fs_inode_info *fi;
1063         bool is_dir = (type == DIR_INODE);
1064         unsigned long ino = 0;
1065
1066         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1067                                 get_pages(sbi, is_dir ?
1068                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1069 retry:
1070         if (unlikely(f2fs_cp_error(sbi))) {
1071                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1072                                 get_pages(sbi, is_dir ?
1073                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1074                 return -EIO;
1075         }
1076
1077         spin_lock(&sbi->inode_lock[type]);
1078
1079         head = &sbi->inode_list[type];
1080         if (list_empty(head)) {
1081                 spin_unlock(&sbi->inode_lock[type]);
1082                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1083                                 get_pages(sbi, is_dir ?
1084                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1085                 return 0;
1086         }
1087         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1088         inode = igrab(&fi->vfs_inode);
1089         spin_unlock(&sbi->inode_lock[type]);
1090         if (inode) {
1091                 unsigned long cur_ino = inode->i_ino;
1092
1093                 F2FS_I(inode)->cp_task = current;
1094
1095                 filemap_fdatawrite(inode->i_mapping);
1096
1097                 F2FS_I(inode)->cp_task = NULL;
1098
1099                 iput(inode);
1100                 /* We need to give cpu to another writers. */
1101                 if (ino == cur_ino)
1102                         cond_resched();
1103                 else
1104                         ino = cur_ino;
1105         } else {
1106                 /*
1107                  * We should submit bio, since it exists several
1108                  * wribacking dentry pages in the freeing inode.
1109                  */
1110                 f2fs_submit_merged_write(sbi, DATA);
1111                 cond_resched();
1112         }
1113         goto retry;
1114 }
1115
1116 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1117 {
1118         struct list_head *head = &sbi->inode_list[DIRTY_META];
1119         struct inode *inode;
1120         struct f2fs_inode_info *fi;
1121         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1122
1123         while (total--) {
1124                 if (unlikely(f2fs_cp_error(sbi)))
1125                         return -EIO;
1126
1127                 spin_lock(&sbi->inode_lock[DIRTY_META]);
1128                 if (list_empty(head)) {
1129                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1130                         return 0;
1131                 }
1132                 fi = list_first_entry(head, struct f2fs_inode_info,
1133                                                         gdirty_list);
1134                 inode = igrab(&fi->vfs_inode);
1135                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1136                 if (inode) {
1137                         sync_inode_metadata(inode, 0);
1138
1139                         /* it's on eviction */
1140                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1141                                 f2fs_update_inode_page(inode);
1142                         iput(inode);
1143                 }
1144         }
1145         return 0;
1146 }
1147
1148 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1149 {
1150         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1151         struct f2fs_nm_info *nm_i = NM_I(sbi);
1152         nid_t last_nid = nm_i->next_scan_nid;
1153
1154         next_free_nid(sbi, &last_nid);
1155         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1156         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1157         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1158         ckpt->next_free_nid = cpu_to_le32(last_nid);
1159 }
1160
1161 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1162 {
1163         bool ret = false;
1164
1165         if (!is_journalled_quota(sbi))
1166                 return false;
1167
1168         if (!f2fs_down_write_trylock(&sbi->quota_sem))
1169                 return true;
1170         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1171                 ret = false;
1172         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1173                 ret = false;
1174         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1175                 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1176                 ret = true;
1177         } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1178                 ret = true;
1179         }
1180         f2fs_up_write(&sbi->quota_sem);
1181         return ret;
1182 }
1183
1184 /*
1185  * Freeze all the FS-operations for checkpoint.
1186  */
1187 static int block_operations(struct f2fs_sb_info *sbi)
1188 {
1189         struct writeback_control wbc = {
1190                 .sync_mode = WB_SYNC_ALL,
1191                 .nr_to_write = LONG_MAX,
1192                 .for_reclaim = 0,
1193         };
1194         int err = 0, cnt = 0;
1195
1196         /*
1197          * Let's flush inline_data in dirty node pages.
1198          */
1199         f2fs_flush_inline_data(sbi);
1200
1201 retry_flush_quotas:
1202         f2fs_lock_all(sbi);
1203         if (__need_flush_quota(sbi)) {
1204                 int locked;
1205
1206                 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1207                         set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1208                         set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1209                         goto retry_flush_dents;
1210                 }
1211                 f2fs_unlock_all(sbi);
1212
1213                 /* only failed during mount/umount/freeze/quotactl */
1214                 locked = down_read_trylock(&sbi->sb->s_umount);
1215                 f2fs_quota_sync(sbi->sb, -1);
1216                 if (locked)
1217                         up_read(&sbi->sb->s_umount);
1218                 cond_resched();
1219                 goto retry_flush_quotas;
1220         }
1221
1222 retry_flush_dents:
1223         /* write all the dirty dentry pages */
1224         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1225                 f2fs_unlock_all(sbi);
1226                 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1227                 if (err)
1228                         return err;
1229                 cond_resched();
1230                 goto retry_flush_quotas;
1231         }
1232
1233         /*
1234          * POR: we should ensure that there are no dirty node pages
1235          * until finishing nat/sit flush. inode->i_blocks can be updated.
1236          */
1237         f2fs_down_write(&sbi->node_change);
1238
1239         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1240                 f2fs_up_write(&sbi->node_change);
1241                 f2fs_unlock_all(sbi);
1242                 err = f2fs_sync_inode_meta(sbi);
1243                 if (err)
1244                         return err;
1245                 cond_resched();
1246                 goto retry_flush_quotas;
1247         }
1248
1249 retry_flush_nodes:
1250         f2fs_down_write(&sbi->node_write);
1251
1252         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1253                 f2fs_up_write(&sbi->node_write);
1254                 atomic_inc(&sbi->wb_sync_req[NODE]);
1255                 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1256                 atomic_dec(&sbi->wb_sync_req[NODE]);
1257                 if (err) {
1258                         f2fs_up_write(&sbi->node_change);
1259                         f2fs_unlock_all(sbi);
1260                         return err;
1261                 }
1262                 cond_resched();
1263                 goto retry_flush_nodes;
1264         }
1265
1266         /*
1267          * sbi->node_change is used only for AIO write_begin path which produces
1268          * dirty node blocks and some checkpoint values by block allocation.
1269          */
1270         __prepare_cp_block(sbi);
1271         f2fs_up_write(&sbi->node_change);
1272         return err;
1273 }
1274
1275 static void unblock_operations(struct f2fs_sb_info *sbi)
1276 {
1277         f2fs_up_write(&sbi->node_write);
1278         f2fs_unlock_all(sbi);
1279 }
1280
1281 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1282 {
1283         DEFINE_WAIT(wait);
1284
1285         for (;;) {
1286                 if (!get_pages(sbi, type))
1287                         break;
1288
1289                 if (unlikely(f2fs_cp_error(sbi)))
1290                         break;
1291
1292                 if (type == F2FS_DIRTY_META)
1293                         f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1294                                                         FS_CP_META_IO);
1295                 else if (type == F2FS_WB_CP_DATA)
1296                         f2fs_submit_merged_write(sbi, DATA);
1297
1298                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1299                 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1300         }
1301         finish_wait(&sbi->cp_wait, &wait);
1302 }
1303
1304 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1305 {
1306         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1307         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1308         unsigned long flags;
1309
1310         if (cpc->reason & CP_UMOUNT) {
1311                 if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1312                         NM_I(sbi)->nat_bits_blocks > sbi->blocks_per_seg) {
1313                         clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1314                         f2fs_notice(sbi, "Disable nat_bits due to no space");
1315                 } else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1316                                                 f2fs_nat_bitmap_enabled(sbi)) {
1317                         f2fs_enable_nat_bits(sbi);
1318                         set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1319                         f2fs_notice(sbi, "Rebuild and enable nat_bits");
1320                 }
1321         }
1322
1323         spin_lock_irqsave(&sbi->cp_lock, flags);
1324
1325         if (cpc->reason & CP_TRIMMED)
1326                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1327         else
1328                 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1329
1330         if (cpc->reason & CP_UMOUNT)
1331                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1332         else
1333                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1334
1335         if (cpc->reason & CP_FASTBOOT)
1336                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1337         else
1338                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1339
1340         if (orphan_num)
1341                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1342         else
1343                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1344
1345         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1346                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1347
1348         if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1349                 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1350         else
1351                 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1352
1353         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1354                 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1355         else
1356                 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1357
1358         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1359                 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1360         else
1361                 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1362
1363         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1364                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1365         else
1366                 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1367
1368         if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1369                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1370
1371         /* set this flag to activate crc|cp_ver for recovery */
1372         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1373         __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1374
1375         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1376 }
1377
1378 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1379         void *src, block_t blk_addr)
1380 {
1381         struct writeback_control wbc = {
1382                 .for_reclaim = 0,
1383         };
1384
1385         /*
1386          * pagevec_lookup_tag and lock_page again will take
1387          * some extra time. Therefore, f2fs_update_meta_pages and
1388          * f2fs_sync_meta_pages are combined in this function.
1389          */
1390         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1391         int err;
1392
1393         f2fs_wait_on_page_writeback(page, META, true, true);
1394
1395         memcpy(page_address(page), src, PAGE_SIZE);
1396
1397         set_page_dirty(page);
1398         if (unlikely(!clear_page_dirty_for_io(page)))
1399                 f2fs_bug_on(sbi, 1);
1400
1401         /* writeout cp pack 2 page */
1402         err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1403         if (unlikely(err && f2fs_cp_error(sbi))) {
1404                 f2fs_put_page(page, 1);
1405                 return;
1406         }
1407
1408         f2fs_bug_on(sbi, err);
1409         f2fs_put_page(page, 0);
1410
1411         /* submit checkpoint (with barrier if NOBARRIER is not set) */
1412         f2fs_submit_merged_write(sbi, META_FLUSH);
1413 }
1414
1415 static inline u64 get_sectors_written(struct block_device *bdev)
1416 {
1417         return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1418 }
1419
1420 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1421 {
1422         if (f2fs_is_multi_device(sbi)) {
1423                 u64 sectors = 0;
1424                 int i;
1425
1426                 for (i = 0; i < sbi->s_ndevs; i++)
1427                         sectors += get_sectors_written(FDEV(i).bdev);
1428
1429                 return sectors;
1430         }
1431
1432         return get_sectors_written(sbi->sb->s_bdev);
1433 }
1434
1435 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1436 {
1437         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1438         struct f2fs_nm_info *nm_i = NM_I(sbi);
1439         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1440         block_t start_blk;
1441         unsigned int data_sum_blocks, orphan_blocks;
1442         __u32 crc32 = 0;
1443         int i;
1444         int cp_payload_blks = __cp_payload(sbi);
1445         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1446         u64 kbytes_written;
1447         int err;
1448
1449         /* Flush all the NAT/SIT pages */
1450         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1451
1452         /* start to update checkpoint, cp ver is already updated previously */
1453         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1454         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1455         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1456                 ckpt->cur_node_segno[i] =
1457                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1458                 ckpt->cur_node_blkoff[i] =
1459                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1460                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1461                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1462         }
1463         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1464                 ckpt->cur_data_segno[i] =
1465                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1466                 ckpt->cur_data_blkoff[i] =
1467                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1468                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1469                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1470         }
1471
1472         /* 2 cp + n data seg summary + orphan inode blocks */
1473         data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1474         spin_lock_irqsave(&sbi->cp_lock, flags);
1475         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1476                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1477         else
1478                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1479         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1480
1481         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1482         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1483                         orphan_blocks);
1484
1485         if (__remain_node_summaries(cpc->reason))
1486                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1487                                 cp_payload_blks + data_sum_blocks +
1488                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1489         else
1490                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1491                                 cp_payload_blks + data_sum_blocks +
1492                                 orphan_blocks);
1493
1494         /* update ckpt flag for checkpoint */
1495         update_ckpt_flags(sbi, cpc);
1496
1497         /* update SIT/NAT bitmap */
1498         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1499         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1500
1501         crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1502         *((__le32 *)((unsigned char *)ckpt +
1503                                 le32_to_cpu(ckpt->checksum_offset)))
1504                                 = cpu_to_le32(crc32);
1505
1506         start_blk = __start_cp_next_addr(sbi);
1507
1508         /* write nat bits */
1509         if ((cpc->reason & CP_UMOUNT) &&
1510                         is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1511                 __u64 cp_ver = cur_cp_version(ckpt);
1512                 block_t blk;
1513
1514                 cp_ver |= ((__u64)crc32 << 32);
1515                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1516
1517                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1518                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1519                         f2fs_update_meta_page(sbi, nm_i->nat_bits +
1520                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1521         }
1522
1523         /* write out checkpoint buffer at block 0 */
1524         f2fs_update_meta_page(sbi, ckpt, start_blk++);
1525
1526         for (i = 1; i < 1 + cp_payload_blks; i++)
1527                 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1528                                                         start_blk++);
1529
1530         if (orphan_num) {
1531                 write_orphan_inodes(sbi, start_blk);
1532                 start_blk += orphan_blocks;
1533         }
1534
1535         f2fs_write_data_summaries(sbi, start_blk);
1536         start_blk += data_sum_blocks;
1537
1538         /* Record write statistics in the hot node summary */
1539         kbytes_written = sbi->kbytes_written;
1540         kbytes_written += (f2fs_get_sectors_written(sbi) -
1541                                 sbi->sectors_written_start) >> 1;
1542         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1543
1544         if (__remain_node_summaries(cpc->reason)) {
1545                 f2fs_write_node_summaries(sbi, start_blk);
1546                 start_blk += NR_CURSEG_NODE_TYPE;
1547         }
1548
1549         /* update user_block_counts */
1550         sbi->last_valid_block_count = sbi->total_valid_block_count;
1551         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1552         percpu_counter_set(&sbi->rf_node_block_count, 0);
1553
1554         /* Here, we have one bio having CP pack except cp pack 2 page */
1555         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1556         /* Wait for all dirty meta pages to be submitted for IO */
1557         f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1558
1559         /* wait for previous submitted meta pages writeback */
1560         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1561
1562         /* flush all device cache */
1563         err = f2fs_flush_device_cache(sbi);
1564         if (err)
1565                 return err;
1566
1567         /* barrier and flush checkpoint cp pack 2 page if it can */
1568         commit_checkpoint(sbi, ckpt, start_blk);
1569         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1570
1571         /*
1572          * invalidate intermediate page cache borrowed from meta inode which are
1573          * used for migration of encrypted, verity or compressed inode's blocks.
1574          */
1575         if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1576                 f2fs_sb_has_compression(sbi))
1577                 invalidate_mapping_pages(META_MAPPING(sbi),
1578                                 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1579
1580         f2fs_release_ino_entry(sbi, false);
1581
1582         f2fs_reset_fsync_node_info(sbi);
1583
1584         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1585         clear_sbi_flag(sbi, SBI_NEED_CP);
1586         clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1587
1588         spin_lock(&sbi->stat_lock);
1589         sbi->unusable_block_count = 0;
1590         spin_unlock(&sbi->stat_lock);
1591
1592         __set_cp_next_pack(sbi);
1593
1594         /*
1595          * redirty superblock if metadata like node page or inode cache is
1596          * updated during writing checkpoint.
1597          */
1598         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1599                         get_pages(sbi, F2FS_DIRTY_IMETA))
1600                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1601
1602         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1603
1604         return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1605 }
1606
1607 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1608 {
1609         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1610         unsigned long long ckpt_ver;
1611         int err = 0;
1612
1613         if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1614                 return -EROFS;
1615
1616         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1617                 if (cpc->reason != CP_PAUSE)
1618                         return 0;
1619                 f2fs_warn(sbi, "Start checkpoint disabled!");
1620         }
1621         if (cpc->reason != CP_RESIZE)
1622                 f2fs_down_write(&sbi->cp_global_sem);
1623
1624         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1625                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1626                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1627                 goto out;
1628         if (unlikely(f2fs_cp_error(sbi))) {
1629                 err = -EIO;
1630                 goto out;
1631         }
1632
1633         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1634
1635         err = block_operations(sbi);
1636         if (err)
1637                 goto out;
1638
1639         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1640
1641         f2fs_flush_merged_writes(sbi);
1642
1643         /* this is the case of multiple fstrims without any changes */
1644         if (cpc->reason & CP_DISCARD) {
1645                 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1646                         unblock_operations(sbi);
1647                         goto out;
1648                 }
1649
1650                 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1651                                 SIT_I(sbi)->dirty_sentries == 0 &&
1652                                 prefree_segments(sbi) == 0) {
1653                         f2fs_flush_sit_entries(sbi, cpc);
1654                         f2fs_clear_prefree_segments(sbi, cpc);
1655                         unblock_operations(sbi);
1656                         goto out;
1657                 }
1658         }
1659
1660         /*
1661          * update checkpoint pack index
1662          * Increase the version number so that
1663          * SIT entries and seg summaries are written at correct place
1664          */
1665         ckpt_ver = cur_cp_version(ckpt);
1666         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1667
1668         /* write cached NAT/SIT entries to NAT/SIT area */
1669         err = f2fs_flush_nat_entries(sbi, cpc);
1670         if (err) {
1671                 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1672                 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1673                 goto stop;
1674         }
1675
1676         f2fs_flush_sit_entries(sbi, cpc);
1677
1678         /* save inmem log status */
1679         f2fs_save_inmem_curseg(sbi);
1680
1681         err = do_checkpoint(sbi, cpc);
1682         if (err) {
1683                 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1684                 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1685                 f2fs_release_discard_addrs(sbi);
1686         } else {
1687                 f2fs_clear_prefree_segments(sbi, cpc);
1688         }
1689
1690         f2fs_restore_inmem_curseg(sbi);
1691 stop:
1692         unblock_operations(sbi);
1693         stat_inc_cp_count(sbi->stat_info);
1694
1695         if (cpc->reason & CP_RECOVERY)
1696                 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1697
1698         /* update CP_TIME to trigger checkpoint periodically */
1699         f2fs_update_time(sbi, CP_TIME);
1700         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1701 out:
1702         if (cpc->reason != CP_RESIZE)
1703                 f2fs_up_write(&sbi->cp_global_sem);
1704         return err;
1705 }
1706
1707 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1708 {
1709         int i;
1710
1711         for (i = 0; i < MAX_INO_ENTRY; i++) {
1712                 struct inode_management *im = &sbi->im[i];
1713
1714                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1715                 spin_lock_init(&im->ino_lock);
1716                 INIT_LIST_HEAD(&im->ino_list);
1717                 im->ino_num = 0;
1718         }
1719
1720         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1721                         NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1722                                 F2FS_ORPHANS_PER_BLOCK;
1723 }
1724
1725 int __init f2fs_create_checkpoint_caches(void)
1726 {
1727         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1728                         sizeof(struct ino_entry));
1729         if (!ino_entry_slab)
1730                 return -ENOMEM;
1731         f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1732                         sizeof(struct inode_entry));
1733         if (!f2fs_inode_entry_slab) {
1734                 kmem_cache_destroy(ino_entry_slab);
1735                 return -ENOMEM;
1736         }
1737         return 0;
1738 }
1739
1740 void f2fs_destroy_checkpoint_caches(void)
1741 {
1742         kmem_cache_destroy(ino_entry_slab);
1743         kmem_cache_destroy(f2fs_inode_entry_slab);
1744 }
1745
1746 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1747 {
1748         struct cp_control cpc = { .reason = CP_SYNC, };
1749         int err;
1750
1751         f2fs_down_write(&sbi->gc_lock);
1752         err = f2fs_write_checkpoint(sbi, &cpc);
1753         f2fs_up_write(&sbi->gc_lock);
1754
1755         return err;
1756 }
1757
1758 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1759 {
1760         struct ckpt_req_control *cprc = &sbi->cprc_info;
1761         struct ckpt_req *req, *next;
1762         struct llist_node *dispatch_list;
1763         u64 sum_diff = 0, diff, count = 0;
1764         int ret;
1765
1766         dispatch_list = llist_del_all(&cprc->issue_list);
1767         if (!dispatch_list)
1768                 return;
1769         dispatch_list = llist_reverse_order(dispatch_list);
1770
1771         ret = __write_checkpoint_sync(sbi);
1772         atomic_inc(&cprc->issued_ckpt);
1773
1774         llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1775                 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1776                 req->ret = ret;
1777                 complete(&req->wait);
1778
1779                 sum_diff += diff;
1780                 count++;
1781         }
1782         atomic_sub(count, &cprc->queued_ckpt);
1783         atomic_add(count, &cprc->total_ckpt);
1784
1785         spin_lock(&cprc->stat_lock);
1786         cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1787         if (cprc->peak_time < cprc->cur_time)
1788                 cprc->peak_time = cprc->cur_time;
1789         spin_unlock(&cprc->stat_lock);
1790 }
1791
1792 static int issue_checkpoint_thread(void *data)
1793 {
1794         struct f2fs_sb_info *sbi = data;
1795         struct ckpt_req_control *cprc = &sbi->cprc_info;
1796         wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1797 repeat:
1798         if (kthread_should_stop())
1799                 return 0;
1800
1801         if (!llist_empty(&cprc->issue_list))
1802                 __checkpoint_and_complete_reqs(sbi);
1803
1804         wait_event_interruptible(*q,
1805                 kthread_should_stop() || !llist_empty(&cprc->issue_list));
1806         goto repeat;
1807 }
1808
1809 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1810                 struct ckpt_req *wait_req)
1811 {
1812         struct ckpt_req_control *cprc = &sbi->cprc_info;
1813
1814         if (!llist_empty(&cprc->issue_list)) {
1815                 __checkpoint_and_complete_reqs(sbi);
1816         } else {
1817                 /* already dispatched by issue_checkpoint_thread */
1818                 if (wait_req)
1819                         wait_for_completion(&wait_req->wait);
1820         }
1821 }
1822
1823 static void init_ckpt_req(struct ckpt_req *req)
1824 {
1825         memset(req, 0, sizeof(struct ckpt_req));
1826
1827         init_completion(&req->wait);
1828         req->queue_time = ktime_get();
1829 }
1830
1831 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1832 {
1833         struct ckpt_req_control *cprc = &sbi->cprc_info;
1834         struct ckpt_req req;
1835         struct cp_control cpc;
1836
1837         cpc.reason = __get_cp_reason(sbi);
1838         if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1839                 int ret;
1840
1841                 f2fs_down_write(&sbi->gc_lock);
1842                 ret = f2fs_write_checkpoint(sbi, &cpc);
1843                 f2fs_up_write(&sbi->gc_lock);
1844
1845                 return ret;
1846         }
1847
1848         if (!cprc->f2fs_issue_ckpt)
1849                 return __write_checkpoint_sync(sbi);
1850
1851         init_ckpt_req(&req);
1852
1853         llist_add(&req.llnode, &cprc->issue_list);
1854         atomic_inc(&cprc->queued_ckpt);
1855
1856         /*
1857          * update issue_list before we wake up issue_checkpoint thread,
1858          * this smp_mb() pairs with another barrier in ___wait_event(),
1859          * see more details in comments of waitqueue_active().
1860          */
1861         smp_mb();
1862
1863         if (waitqueue_active(&cprc->ckpt_wait_queue))
1864                 wake_up(&cprc->ckpt_wait_queue);
1865
1866         if (cprc->f2fs_issue_ckpt)
1867                 wait_for_completion(&req.wait);
1868         else
1869                 flush_remained_ckpt_reqs(sbi, &req);
1870
1871         return req.ret;
1872 }
1873
1874 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1875 {
1876         dev_t dev = sbi->sb->s_bdev->bd_dev;
1877         struct ckpt_req_control *cprc = &sbi->cprc_info;
1878
1879         if (cprc->f2fs_issue_ckpt)
1880                 return 0;
1881
1882         cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1883                         "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1884         if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1885                 cprc->f2fs_issue_ckpt = NULL;
1886                 return -ENOMEM;
1887         }
1888
1889         set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1890
1891         return 0;
1892 }
1893
1894 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1895 {
1896         struct ckpt_req_control *cprc = &sbi->cprc_info;
1897
1898         if (cprc->f2fs_issue_ckpt) {
1899                 struct task_struct *ckpt_task = cprc->f2fs_issue_ckpt;
1900
1901                 cprc->f2fs_issue_ckpt = NULL;
1902                 kthread_stop(ckpt_task);
1903
1904                 flush_remained_ckpt_reqs(sbi, NULL);
1905         }
1906 }
1907
1908 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1909 {
1910         struct ckpt_req_control *cprc = &sbi->cprc_info;
1911
1912         atomic_set(&cprc->issued_ckpt, 0);
1913         atomic_set(&cprc->total_ckpt, 0);
1914         atomic_set(&cprc->queued_ckpt, 0);
1915         cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1916         init_waitqueue_head(&cprc->ckpt_wait_queue);
1917         init_llist_head(&cprc->issue_list);
1918         spin_lock_init(&cprc->stat_lock);
1919 }