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 <trace/events/f2fs.h>
  22
  23 #define DEFAULT_CHECKPOINT_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
  24
  25 static struct kmem_cache *ino_entry_slab;
  26 struct kmem_cache *f2fs_inode_entry_slab;
  27
  28 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
  29 {
  30         f2fs_build_fault_attr(sbi, 0, 0);
  31         set_ckpt_flags(sbi, CP_ERROR_FLAG);
  32         if (!end_io)
  33                 f2fs_flush_merged_writes(sbi);
  34 }
  35
  36 /*
  37  * We guarantee no failure on the returned page.
  38  */
  39 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  40 {
  41         struct address_space *mapping = META_MAPPING(sbi);
  42         struct page *page;
  43 repeat:
  44         page = f2fs_grab_cache_page(mapping, index, false);
  45         if (!page) {
  46                 cond_resched();
  47                 goto repeat;
  48         }
  49         f2fs_wait_on_page_writeback(page, META, true, true);
  50         if (!PageUptodate(page))
  51                 SetPageUptodate(page);
  52         return page;
  53 }
  54
  55 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
  56                                                         bool is_meta)
  57 {
  58         struct address_space *mapping = META_MAPPING(sbi);
  59         struct page *page;
  60         struct f2fs_io_info fio = {
  61                 .sbi = sbi,
  62                 .type = META,
  63                 .op = REQ_OP_READ,
  64                 .op_flags = REQ_META | REQ_PRIO,
  65                 .old_blkaddr = index,
  66                 .new_blkaddr = index,
  67                 .encrypted_page = NULL,
  68                 .is_por = !is_meta,
  69         };
  70         int err;
  71
  72         if (unlikely(!is_meta))
  73                 fio.op_flags &= ~REQ_META;
  74 repeat:
  75         page = f2fs_grab_cache_page(mapping, index, false);
  76         if (!page) {
  77                 cond_resched();
  78                 goto repeat;
  79         }
  80         if (PageUptodate(page))
  81                 goto out;
  82
  83         fio.page = page;
  84
  85         err = f2fs_submit_page_bio(&fio);
  86         if (err) {
  87                 f2fs_put_page(page, 1);
  88                 return ERR_PTR(err);
  89         }
  90
  91         f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
  92
  93         lock_page(page);
  94         if (unlikely(page->mapping != mapping)) {
  95                 f2fs_put_page(page, 1);
  96                 goto repeat;
  97         }
  98
  99         if (unlikely(!PageUptodate(page))) {
 100                 f2fs_put_page(page, 1);
 101                 return ERR_PTR(-EIO);
 102         }
 103 out:
 104         return page;
 105 }
 106
 107 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 108 {
 109         return __get_meta_page(sbi, index, true);
 110 }
 111
 112 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
 113 {
 114         struct page *page;
 115         int count = 0;
 116
 117 retry:
 118         page = __get_meta_page(sbi, index, true);
 119         if (IS_ERR(page)) {
 120                 if (PTR_ERR(page) == -EIO &&
 121                                 ++count <= DEFAULT_RETRY_IO_COUNT)
 122                         goto retry;
 123                 f2fs_stop_checkpoint(sbi, false);
 124         }
 125         return page;
 126 }
 127
 128 /* for POR only */
 129 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 130 {
 131         return __get_meta_page(sbi, index, false);
 132 }
 133
 134 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
 135                                                         int type)
 136 {
 137         struct seg_entry *se;
 138         unsigned int segno, offset;
 139         bool exist;
 140
 141         if (type != DATA_GENERIC_ENHANCE && type != DATA_GENERIC_ENHANCE_READ)
 142                 return true;
 143
 144         segno = GET_SEGNO(sbi, blkaddr);
 145         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 146         se = get_seg_entry(sbi, segno);
 147
 148         exist = f2fs_test_bit(offset, se->cur_valid_map);
 149         if (!exist && type == DATA_GENERIC_ENHANCE) {
 150                 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
 151                          blkaddr, exist);
 152                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 153                 WARN_ON(1);
 154         }
 155         return exist;
 156 }
 157
 158 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
 159                                         block_t blkaddr, int type)
 160 {
 161         switch (type) {
 162         case META_NAT:
 163                 break;
 164         case META_SIT:
 165                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 166                         return false;
 167                 break;
 168         case META_SSA:
 169                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 170                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 171                         return false;
 172                 break;
 173         case META_CP:
 174                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 175                         blkaddr < __start_cp_addr(sbi)))
 176                         return false;
 177                 break;
 178         case META_POR:
 179                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 180                         blkaddr < MAIN_BLKADDR(sbi)))
 181                         return false;
 182                 break;
 183         case DATA_GENERIC:
 184         case DATA_GENERIC_ENHANCE:
 185         case DATA_GENERIC_ENHANCE_READ:
 186                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 187                                 blkaddr < MAIN_BLKADDR(sbi))) {
 188                         f2fs_warn(sbi, "access invalid blkaddr:%u",
 189                                   blkaddr);
 190                         set_sbi_flag(sbi, SBI_NEED_FSCK);
 191                         WARN_ON(1);
 192                         return false;
 193                 } else {
 194                         return __is_bitmap_valid(sbi, blkaddr, type);
 195                 }
 196                 break;
 197         case META_GENERIC:
 198                 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
 199                         blkaddr >= MAIN_BLKADDR(sbi)))
 200                         return false;
 201                 break;
 202         default:
 203                 BUG();
 204         }
 205
 206         return true;
 207 }
 208
 209 /*
 210  * Readahead CP/NAT/SIT/SSA/POR pages
 211  */
 212 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 213                                                         int type, bool sync)
 214 {
 215         struct page *page;
 216         block_t blkno = start;
 217         struct f2fs_io_info fio = {
 218                 .sbi = sbi,
 219                 .type = META,
 220                 .op = REQ_OP_READ,
 221                 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
 222                 .encrypted_page = NULL,
 223                 .in_list = false,
 224                 .is_por = (type == META_POR),
 225         };
 226         struct blk_plug plug;
 227         int err;
 228
 229         if (unlikely(type == META_POR))
 230                 fio.op_flags &= ~REQ_META;
 231
 232         blk_start_plug(&plug);
 233         for (; nrpages-- > 0; blkno++) {
 234
 235                 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
 236                         goto out;
 237
 238                 switch (type) {
 239                 case META_NAT:
 240                         if (unlikely(blkno >=
 241                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 242                                 blkno = 0;
 243                         /* get nat block addr */
 244                         fio.new_blkaddr = current_nat_addr(sbi,
 245                                         blkno * NAT_ENTRY_PER_BLOCK);
 246                         break;
 247                 case META_SIT:
 248                         if (unlikely(blkno >= TOTAL_SEGS(sbi)))
 249                                 goto out;
 250                         /* get sit block addr */
 251                         fio.new_blkaddr = current_sit_addr(sbi,
 252                                         blkno * SIT_ENTRY_PER_BLOCK);
 253                         break;
 254                 case META_SSA:
 255                 case META_CP:
 256                 case META_POR:
 257                         fio.new_blkaddr = blkno;
 258                         break;
 259                 default:
 260                         BUG();
 261                 }
 262
 263                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 264                                                 fio.new_blkaddr, false);
 265                 if (!page)
 266                         continue;
 267                 if (PageUptodate(page)) {
 268                         f2fs_put_page(page, 1);
 269                         continue;
 270                 }
 271
 272                 fio.page = page;
 273                 err = f2fs_submit_page_bio(&fio);
 274                 f2fs_put_page(page, err ? 1 : 0);
 275
 276                 if (!err)
 277                         f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
 278         }
 279 out:
 280         blk_finish_plug(&plug);
 281         return blkno - start;
 282 }
 283
 284 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 285 {
 286         struct page *page;
 287         bool readahead = false;
 288
 289         page = find_get_page(META_MAPPING(sbi), index);
 290         if (!page || !PageUptodate(page))
 291                 readahead = true;
 292         f2fs_put_page(page, 0);
 293
 294         if (readahead)
 295                 f2fs_ra_meta_pages(sbi, index, BIO_MAX_VECS, META_POR, true);
 296 }
 297
 298 static int __f2fs_write_meta_page(struct page *page,
 299                                 struct writeback_control *wbc,
 300                                 enum iostat_type io_type)
 301 {
 302         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 303
 304         trace_f2fs_writepage(page, META);
 305
 306         if (unlikely(f2fs_cp_error(sbi)))
 307                 goto redirty_out;
 308         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 309                 goto redirty_out;
 310         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 311                 goto redirty_out;
 312
 313         f2fs_do_write_meta_page(sbi, page, io_type);
 314         dec_page_count(sbi, F2FS_DIRTY_META);
 315
 316         if (wbc->for_reclaim)
 317                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
 318
 319         unlock_page(page);
 320
 321         if (unlikely(f2fs_cp_error(sbi)))
 322                 f2fs_submit_merged_write(sbi, META);
 323
 324         return 0;
 325
 326 redirty_out:
 327         redirty_page_for_writepage(wbc, page);
 328         return AOP_WRITEPAGE_ACTIVATE;
 329 }
 330
 331 static int f2fs_write_meta_page(struct page *page,
 332                                 struct writeback_control *wbc)
 333 {
 334         return __f2fs_write_meta_page(page, wbc, FS_META_IO);
 335 }
 336
 337 static int f2fs_write_meta_pages(struct address_space *mapping,
 338                                 struct writeback_control *wbc)
 339 {
 340         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 341         long diff, written;
 342
 343         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 344                 goto skip_write;
 345
 346         /* collect a number of dirty meta pages and write together */
 347         if (wbc->sync_mode != WB_SYNC_ALL &&
 348                         get_pages(sbi, F2FS_DIRTY_META) <
 349                                         nr_pages_to_skip(sbi, META))
 350                 goto skip_write;
 351
 352         /* if locked failed, cp will flush dirty pages instead */
 353         if (!down_write_trylock(&sbi->cp_global_sem))
 354                 goto skip_write;
 355
 356         trace_f2fs_writepages(mapping->host, wbc, META);
 357         diff = nr_pages_to_write(sbi, META, wbc);
 358         written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
 359         up_write(&sbi->cp_global_sem);
 360         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 361         return 0;
 362
 363 skip_write:
 364         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 365         trace_f2fs_writepages(mapping->host, wbc, META);
 366         return 0;
 367 }
 368
 369 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 370                                 long nr_to_write, enum iostat_type io_type)
 371 {
 372         struct address_space *mapping = META_MAPPING(sbi);
 373         pgoff_t index = 0, prev = ULONG_MAX;
 374         struct pagevec pvec;
 375         long nwritten = 0;
 376         int nr_pages;
 377         struct writeback_control wbc = {
 378                 .for_reclaim = 0,
 379         };
 380         struct blk_plug plug;
 381
 382         pagevec_init(&pvec);
 383
 384         blk_start_plug(&plug);
 385
 386         while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 387                                 PAGECACHE_TAG_DIRTY))) {
 388                 int i;
 389
 390                 for (i = 0; i < nr_pages; i++) {
 391                         struct page *page = pvec.pages[i];
 392
 393                         if (prev == ULONG_MAX)
 394                                 prev = page->index - 1;
 395                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 396                                 pagevec_release(&pvec);
 397                                 goto stop;
 398                         }
 399
 400                         lock_page(page);
 401
 402                         if (unlikely(page->mapping != mapping)) {
 403 continue_unlock:
 404                                 unlock_page(page);
 405                                 continue;
 406                         }
 407                         if (!PageDirty(page)) {
 408                                 /* someone wrote it for us */
 409                                 goto continue_unlock;
 410                         }
 411
 412                         f2fs_wait_on_page_writeback(page, META, true, true);
 413
 414                         if (!clear_page_dirty_for_io(page))
 415                                 goto continue_unlock;
 416
 417                         if (__f2fs_write_meta_page(page, &wbc, io_type)) {
 418                                 unlock_page(page);
 419                                 break;
 420                         }
 421                         nwritten++;
 422                         prev = page->index;
 423                         if (unlikely(nwritten >= nr_to_write))
 424                                 break;
 425                 }
 426                 pagevec_release(&pvec);
 427                 cond_resched();
 428         }
 429 stop:
 430         if (nwritten)
 431                 f2fs_submit_merged_write(sbi, type);
 432
 433         blk_finish_plug(&plug);
 434
 435         return nwritten;
 436 }
 437
 438 static int f2fs_set_meta_page_dirty(struct page *page)
 439 {
 440         trace_f2fs_set_page_dirty(page, META);
 441
 442         if (!PageUptodate(page))
 443                 SetPageUptodate(page);
 444         if (!PageDirty(page)) {
 445                 __set_page_dirty_nobuffers(page);
 446                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 447                 set_page_private_reference(page);
 448                 return 1;
 449         }
 450         return 0;
 451 }
 452
 453 const struct address_space_operations f2fs_meta_aops = {
 454         .writepage      = f2fs_write_meta_page,
 455         .writepages     = f2fs_write_meta_pages,
 456         .set_page_dirty = f2fs_set_meta_page_dirty,
 457         .invalidatepage = f2fs_invalidate_page,
 458         .releasepage    = f2fs_release_page,
 459 #ifdef CONFIG_MIGRATION
 460         .migratepage    = f2fs_migrate_page,
 461 #endif
 462 };
 463
 464 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 465                                                 unsigned int devidx, int type)
 466 {
 467         struct inode_management *im = &sbi->im[type];
 468         struct ino_entry *e = NULL, *new = NULL;
 469
 470         if (type == FLUSH_INO) {
 471                 rcu_read_lock();
 472                 e = radix_tree_lookup(&im->ino_root, ino);
 473                 rcu_read_unlock();
 474         }
 475
 476 retry:
 477         if (!e)
 478                 new = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 479
 480         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 481
 482         spin_lock(&im->ino_lock);
 483         e = radix_tree_lookup(&im->ino_root, ino);
 484         if (!e) {
 485                 if (!new) {
 486                         spin_unlock(&im->ino_lock);
 487                         goto retry;
 488                 }
 489                 e = new;
 490                 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
 491                         f2fs_bug_on(sbi, 1);
 492
 493                 memset(e, 0, sizeof(struct ino_entry));
 494                 e->ino = ino;
 495
 496                 list_add_tail(&e->list, &im->ino_list);
 497                 if (type != ORPHAN_INO)
 498                         im->ino_num++;
 499         }
 500
 501         if (type == FLUSH_INO)
 502                 f2fs_set_bit(devidx, (char *)&e->dirty_device);
 503
 504         spin_unlock(&im->ino_lock);
 505         radix_tree_preload_end();
 506
 507         if (new && e != new)
 508                 kmem_cache_free(ino_entry_slab, new);
 509 }
 510
 511 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 512 {
 513         struct inode_management *im = &sbi->im[type];
 514         struct ino_entry *e;
 515
 516         spin_lock(&im->ino_lock);
 517         e = radix_tree_lookup(&im->ino_root, ino);
 518         if (e) {
 519                 list_del(&e->list);
 520                 radix_tree_delete(&im->ino_root, ino);
 521                 im->ino_num--;
 522                 spin_unlock(&im->ino_lock);
 523                 kmem_cache_free(ino_entry_slab, e);
 524                 return;
 525         }
 526         spin_unlock(&im->ino_lock);
 527 }
 528
 529 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 530 {
 531         /* add new dirty ino entry into list */
 532         __add_ino_entry(sbi, ino, 0, type);
 533 }
 534
 535 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 536 {
 537         /* remove dirty ino entry from list */
 538         __remove_ino_entry(sbi, ino, type);
 539 }
 540
 541 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
 542 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 543 {
 544         struct inode_management *im = &sbi->im[mode];
 545         struct ino_entry *e;
 546
 547         spin_lock(&im->ino_lock);
 548         e = radix_tree_lookup(&im->ino_root, ino);
 549         spin_unlock(&im->ino_lock);
 550         return e ? true : false;
 551 }
 552
 553 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 554 {
 555         struct ino_entry *e, *tmp;
 556         int i;
 557
 558         for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
 559                 struct inode_management *im = &sbi->im[i];
 560
 561                 spin_lock(&im->ino_lock);
 562                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 563                         list_del(&e->list);
 564                         radix_tree_delete(&im->ino_root, e->ino);
 565                         kmem_cache_free(ino_entry_slab, e);
 566                         im->ino_num--;
 567                 }
 568                 spin_unlock(&im->ino_lock);
 569         }
 570 }
 571
 572 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 573                                         unsigned int devidx, int type)
 574 {
 575         __add_ino_entry(sbi, ino, devidx, type);
 576 }
 577
 578 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 579                                         unsigned int devidx, int type)
 580 {
 581         struct inode_management *im = &sbi->im[type];
 582         struct ino_entry *e;
 583         bool is_dirty = false;
 584
 585         spin_lock(&im->ino_lock);
 586         e = radix_tree_lookup(&im->ino_root, ino);
 587         if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
 588                 is_dirty = true;
 589         spin_unlock(&im->ino_lock);
 590         return is_dirty;
 591 }
 592
 593 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
 594 {
 595         struct inode_management *im = &sbi->im[ORPHAN_INO];
 596         int err = 0;
 597
 598         spin_lock(&im->ino_lock);
 599
 600         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 601                 spin_unlock(&im->ino_lock);
 602                 f2fs_show_injection_info(sbi, FAULT_ORPHAN);
 603                 return -ENOSPC;
 604         }
 605
 606         if (unlikely(im->ino_num >= sbi->max_orphans))
 607                 err = -ENOSPC;
 608         else
 609                 im->ino_num++;
 610         spin_unlock(&im->ino_lock);
 611
 612         return err;
 613 }
 614
 615 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
 616 {
 617         struct inode_management *im = &sbi->im[ORPHAN_INO];
 618
 619         spin_lock(&im->ino_lock);
 620         f2fs_bug_on(sbi, im->ino_num == 0);
 621         im->ino_num--;
 622         spin_unlock(&im->ino_lock);
 623 }
 624
 625 void f2fs_add_orphan_inode(struct inode *inode)
 626 {
 627         /* add new orphan ino entry into list */
 628         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
 629         f2fs_update_inode_page(inode);
 630 }
 631
 632 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 633 {
 634         /* remove orphan entry from orphan list */
 635         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 636 }
 637
 638 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 639 {
 640         struct inode *inode;
 641         struct node_info ni;
 642         int err;
 643
 644         inode = f2fs_iget_retry(sbi->sb, ino);
 645         if (IS_ERR(inode)) {
 646                 /*
 647                  * there should be a bug that we can't find the entry
 648                  * to orphan inode.
 649                  */
 650                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 651                 return PTR_ERR(inode);
 652         }
 653
 654         err = dquot_initialize(inode);
 655         if (err) {
 656                 iput(inode);
 657                 goto err_out;
 658         }
 659
 660         clear_nlink(inode);
 661
 662         /* truncate all the data during iput */
 663         iput(inode);
 664
 665         err = f2fs_get_node_info(sbi, ino, &ni);
 666         if (err)
 667                 goto err_out;
 668
 669         /* ENOMEM was fully retried in f2fs_evict_inode. */
 670         if (ni.blk_addr != NULL_ADDR) {
 671                 err = -EIO;
 672                 goto err_out;
 673         }
 674         return 0;
 675
 676 err_out:
 677         set_sbi_flag(sbi, SBI_NEED_FSCK);
 678         f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
 679                   __func__, ino);
 680         return err;
 681 }
 682
 683 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
 684 {
 685         block_t start_blk, orphan_blocks, i, j;
 686         unsigned int s_flags = sbi->sb->s_flags;
 687         int err = 0;
 688 #ifdef CONFIG_QUOTA
 689         int quota_enabled;
 690 #endif
 691
 692         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 693                 return 0;
 694
 695         if (bdev_read_only(sbi->sb->s_bdev)) {
 696                 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
 697                 return 0;
 698         }
 699
 700         if (s_flags & SB_RDONLY) {
 701                 f2fs_info(sbi, "orphan cleanup on readonly fs");
 702                 sbi->sb->s_flags &= ~SB_RDONLY;
 703         }
 704
 705 #ifdef CONFIG_QUOTA
 706         /* Needed for iput() to work correctly and not trash data */
 707         sbi->sb->s_flags |= SB_ACTIVE;
 708
 709         /*
 710          * Turn on quotas which were not enabled for read-only mounts if
 711          * filesystem has quota feature, so that they are updated correctly.
 712          */
 713         quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
 714 #endif
 715
 716         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 717         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 718
 719         f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 720
 721         for (i = 0; i < orphan_blocks; i++) {
 722                 struct page *page;
 723                 struct f2fs_orphan_block *orphan_blk;
 724
 725                 page = f2fs_get_meta_page(sbi, start_blk + i);
 726                 if (IS_ERR(page)) {
 727                         err = PTR_ERR(page);
 728                         goto out;
 729                 }
 730
 731                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 732                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 733                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 734
 735                         err = recover_orphan_inode(sbi, ino);
 736                         if (err) {
 737                                 f2fs_put_page(page, 1);
 738                                 goto out;
 739                         }
 740                 }
 741                 f2fs_put_page(page, 1);
 742         }
 743         /* clear Orphan Flag */
 744         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 745 out:
 746         set_sbi_flag(sbi, SBI_IS_RECOVERED);
 747
 748 #ifdef CONFIG_QUOTA
 749         /* Turn quotas off */
 750         if (quota_enabled)
 751                 f2fs_quota_off_umount(sbi->sb);
 752 #endif
 753         sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
 754
 755         return err;
 756 }
 757
 758 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 759 {
 760         struct list_head *head;
 761         struct f2fs_orphan_block *orphan_blk = NULL;
 762         unsigned int nentries = 0;
 763         unsigned short index = 1;
 764         unsigned short orphan_blocks;
 765         struct page *page = NULL;
 766         struct ino_entry *orphan = NULL;
 767         struct inode_management *im = &sbi->im[ORPHAN_INO];
 768
 769         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 770
 771         /*
 772          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 773          * orphan inode operations are covered under f2fs_lock_op().
 774          * And, spin_lock should be avoided due to page operations below.
 775          */
 776         head = &im->ino_list;
 777
 778         /* loop for each orphan inode entry and write them in Jornal block */
 779         list_for_each_entry(orphan, head, list) {
 780                 if (!page) {
 781                         page = f2fs_grab_meta_page(sbi, start_blk++);
 782                         orphan_blk =
 783                                 (struct f2fs_orphan_block *)page_address(page);
 784                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 785                 }
 786
 787                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 788
 789                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 790                         /*
 791                          * an orphan block is full of 1020 entries,
 792                          * then we need to flush current orphan blocks
 793                          * and bring another one in memory
 794                          */
 795                         orphan_blk->blk_addr = cpu_to_le16(index);
 796                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 797                         orphan_blk->entry_count = cpu_to_le32(nentries);
 798                         set_page_dirty(page);
 799                         f2fs_put_page(page, 1);
 800                         index++;
 801                         nentries = 0;
 802                         page = NULL;
 803                 }
 804         }
 805
 806         if (page) {
 807                 orphan_blk->blk_addr = cpu_to_le16(index);
 808                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 809                 orphan_blk->entry_count = cpu_to_le32(nentries);
 810                 set_page_dirty(page);
 811                 f2fs_put_page(page, 1);
 812         }
 813 }
 814
 815 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
 816                                                 struct f2fs_checkpoint *ckpt)
 817 {
 818         unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
 819         __u32 chksum;
 820
 821         chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
 822         if (chksum_ofs < CP_CHKSUM_OFFSET) {
 823                 chksum_ofs += sizeof(chksum);
 824                 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
 825                                                 F2FS_BLKSIZE - chksum_ofs);
 826         }
 827         return chksum;
 828 }
 829
 830 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 831                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 832                 unsigned long long *version)
 833 {
 834         size_t crc_offset = 0;
 835         __u32 crc;
 836
 837         *cp_page = f2fs_get_meta_page(sbi, cp_addr);
 838         if (IS_ERR(*cp_page))
 839                 return PTR_ERR(*cp_page);
 840
 841         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 842
 843         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 844         if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
 845                         crc_offset > CP_CHKSUM_OFFSET) {
 846                 f2fs_put_page(*cp_page, 1);
 847                 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
 848                 return -EINVAL;
 849         }
 850
 851         crc = f2fs_checkpoint_chksum(sbi, *cp_block);
 852         if (crc != cur_cp_crc(*cp_block)) {
 853                 f2fs_put_page(*cp_page, 1);
 854                 f2fs_warn(sbi, "invalid crc value");
 855                 return -EINVAL;
 856         }
 857
 858         *version = cur_cp_version(*cp_block);
 859         return 0;
 860 }
 861
 862 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 863                                 block_t cp_addr, unsigned long long *version)
 864 {
 865         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 866         struct f2fs_checkpoint *cp_block = NULL;
 867         unsigned long long cur_version = 0, pre_version = 0;
 868         int err;
 869
 870         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 871                                         &cp_page_1, version);
 872         if (err)
 873                 return NULL;
 874
 875         if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
 876                                         sbi->blocks_per_seg) {
 877                 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
 878                           le32_to_cpu(cp_block->cp_pack_total_block_count));
 879                 goto invalid_cp;
 880         }
 881         pre_version = *version;
 882
 883         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 884         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 885                                         &cp_page_2, version);
 886         if (err)
 887                 goto invalid_cp;
 888         cur_version = *version;
 889
 890         if (cur_version == pre_version) {
 891                 *version = cur_version;
 892                 f2fs_put_page(cp_page_2, 1);
 893                 return cp_page_1;
 894         }
 895         f2fs_put_page(cp_page_2, 1);
 896 invalid_cp:
 897         f2fs_put_page(cp_page_1, 1);
 898         return NULL;
 899 }
 900
 901 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
 902 {
 903         struct f2fs_checkpoint *cp_block;
 904         struct f2fs_super_block *fsb = sbi->raw_super;
 905         struct page *cp1, *cp2, *cur_page;
 906         unsigned long blk_size = sbi->blocksize;
 907         unsigned long long cp1_version = 0, cp2_version = 0;
 908         unsigned long long cp_start_blk_no;
 909         unsigned int cp_blks = 1 + __cp_payload(sbi);
 910         block_t cp_blk_no;
 911         int i;
 912         int err;
 913
 914         sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
 915                                   GFP_KERNEL);
 916         if (!sbi->ckpt)
 917                 return -ENOMEM;
 918         /*
 919          * Finding out valid cp block involves read both
 920          * sets( cp pack 1 and cp pack 2)
 921          */
 922         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 923         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 924
 925         /* The second checkpoint pack should start at the next segment */
 926         cp_start_blk_no += ((unsigned long long)1) <<
 927                                 le32_to_cpu(fsb->log_blocks_per_seg);
 928         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 929
 930         if (cp1 && cp2) {
 931                 if (ver_after(cp2_version, cp1_version))
 932                         cur_page = cp2;
 933                 else
 934                         cur_page = cp1;
 935         } else if (cp1) {
 936                 cur_page = cp1;
 937         } else if (cp2) {
 938                 cur_page = cp2;
 939         } else {
 940                 err = -EFSCORRUPTED;
 941                 goto fail_no_cp;
 942         }
 943
 944         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 945         memcpy(sbi->ckpt, cp_block, blk_size);
 946
 947         if (cur_page == cp1)
 948                 sbi->cur_cp_pack = 1;
 949         else
 950                 sbi->cur_cp_pack = 2;
 951
 952         /* Sanity checking of checkpoint */
 953         if (f2fs_sanity_check_ckpt(sbi)) {
 954                 err = -EFSCORRUPTED;
 955                 goto free_fail_no_cp;
 956         }
 957
 958         if (cp_blks <= 1)
 959                 goto done;
 960
 961         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 962         if (cur_page == cp2)
 963                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 964
 965         for (i = 1; i < cp_blks; i++) {
 966                 void *sit_bitmap_ptr;
 967                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 968
 969                 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
 970                 if (IS_ERR(cur_page)) {
 971                         err = PTR_ERR(cur_page);
 972                         goto free_fail_no_cp;
 973                 }
 974                 sit_bitmap_ptr = page_address(cur_page);
 975                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 976                 f2fs_put_page(cur_page, 1);
 977         }
 978 done:
 979         f2fs_put_page(cp1, 1);
 980         f2fs_put_page(cp2, 1);
 981         return 0;
 982
 983 free_fail_no_cp:
 984         f2fs_put_page(cp1, 1);
 985         f2fs_put_page(cp2, 1);
 986 fail_no_cp:
 987         kvfree(sbi->ckpt);
 988         return err;
 989 }
 990
 991 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 992 {
 993         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 994         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 995
 996         if (is_inode_flag_set(inode, flag))
 997                 return;
 998
 999         set_inode_flag(inode, flag);
1000         if (!f2fs_is_volatile_file(inode))
1001                 list_add_tail(&F2FS_I(inode)->dirty_list,
1002                                                 &sbi->inode_list[type]);
1003         stat_inc_dirty_inode(sbi, type);
1004 }
1005
1006 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1007 {
1008         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1009
1010         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1011                 return;
1012
1013         list_del_init(&F2FS_I(inode)->dirty_list);
1014         clear_inode_flag(inode, flag);
1015         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1016 }
1017
1018 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
1019 {
1020         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1021         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1022
1023         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1024                         !S_ISLNK(inode->i_mode))
1025                 return;
1026
1027         spin_lock(&sbi->inode_lock[type]);
1028         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1029                 __add_dirty_inode(inode, type);
1030         inode_inc_dirty_pages(inode);
1031         spin_unlock(&sbi->inode_lock[type]);
1032
1033         set_page_private_reference(page);
1034 }
1035
1036 void f2fs_remove_dirty_inode(struct inode *inode)
1037 {
1038         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1039         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1040
1041         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1042                         !S_ISLNK(inode->i_mode))
1043                 return;
1044
1045         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1046                 return;
1047
1048         spin_lock(&sbi->inode_lock[type]);
1049         __remove_dirty_inode(inode, type);
1050         spin_unlock(&sbi->inode_lock[type]);
1051 }
1052
1053 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1054 {
1055         struct list_head *head;
1056         struct inode *inode;
1057         struct f2fs_inode_info *fi;
1058         bool is_dir = (type == DIR_INODE);
1059         unsigned long ino = 0;
1060
1061         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1062                                 get_pages(sbi, is_dir ?
1063                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1064 retry:
1065         if (unlikely(f2fs_cp_error(sbi))) {
1066                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1067                                 get_pages(sbi, is_dir ?
1068                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1069                 return -EIO;
1070         }
1071
1072         spin_lock(&sbi->inode_lock[type]);
1073
1074         head = &sbi->inode_list[type];
1075         if (list_empty(head)) {
1076                 spin_unlock(&sbi->inode_lock[type]);
1077                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1078                                 get_pages(sbi, is_dir ?
1079                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1080                 return 0;
1081         }
1082         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1083         inode = igrab(&fi->vfs_inode);
1084         spin_unlock(&sbi->inode_lock[type]);
1085         if (inode) {
1086                 unsigned long cur_ino = inode->i_ino;
1087
1088                 F2FS_I(inode)->cp_task = current;
1089
1090                 filemap_fdatawrite(inode->i_mapping);
1091
1092                 F2FS_I(inode)->cp_task = NULL;
1093
1094                 iput(inode);
1095                 /* We need to give cpu to another writers. */
1096                 if (ino == cur_ino)
1097                         cond_resched();
1098                 else
1099                         ino = cur_ino;
1100         } else {
1101                 /*
1102                  * We should submit bio, since it exists several
1103                  * wribacking dentry pages in the freeing inode.
1104                  */
1105                 f2fs_submit_merged_write(sbi, DATA);
1106                 cond_resched();
1107         }
1108         goto retry;
1109 }
1110
1111 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1112 {
1113         struct list_head *head = &sbi->inode_list[DIRTY_META];
1114         struct inode *inode;
1115         struct f2fs_inode_info *fi;
1116         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1117
1118         while (total--) {
1119                 if (unlikely(f2fs_cp_error(sbi)))
1120                         return -EIO;
1121
1122                 spin_lock(&sbi->inode_lock[DIRTY_META]);
1123                 if (list_empty(head)) {
1124                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1125                         return 0;
1126                 }
1127                 fi = list_first_entry(head, struct f2fs_inode_info,
1128                                                         gdirty_list);
1129                 inode = igrab(&fi->vfs_inode);
1130                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1131                 if (inode) {
1132                         sync_inode_metadata(inode, 0);
1133
1134                         /* it's on eviction */
1135                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1136                                 f2fs_update_inode_page(inode);
1137                         iput(inode);
1138                 }
1139         }
1140         return 0;
1141 }
1142
1143 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1144 {
1145         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1146         struct f2fs_nm_info *nm_i = NM_I(sbi);
1147         nid_t last_nid = nm_i->next_scan_nid;
1148
1149         next_free_nid(sbi, &last_nid);
1150         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1151         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1152         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1153         ckpt->next_free_nid = cpu_to_le32(last_nid);
1154 }
1155
1156 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1157 {
1158         bool ret = false;
1159
1160         if (!is_journalled_quota(sbi))
1161                 return false;
1162
1163         down_write(&sbi->quota_sem);
1164         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1165                 ret = false;
1166         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1167                 ret = false;
1168         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1169                 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1170                 ret = true;
1171         } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1172                 ret = true;
1173         }
1174         up_write(&sbi->quota_sem);
1175         return ret;
1176 }
1177
1178 /*
1179  * Freeze all the FS-operations for checkpoint.
1180  */
1181 static int block_operations(struct f2fs_sb_info *sbi)
1182 {
1183         struct writeback_control wbc = {
1184                 .sync_mode = WB_SYNC_ALL,
1185                 .nr_to_write = LONG_MAX,
1186                 .for_reclaim = 0,
1187         };
1188         int err = 0, cnt = 0;
1189
1190         /*
1191          * Let's flush inline_data in dirty node pages.
1192          */
1193         f2fs_flush_inline_data(sbi);
1194
1195 retry_flush_quotas:
1196         f2fs_lock_all(sbi);
1197         if (__need_flush_quota(sbi)) {
1198                 int locked;
1199
1200                 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1201                         set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1202                         set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1203                         goto retry_flush_dents;
1204                 }
1205                 f2fs_unlock_all(sbi);
1206
1207                 /* only failed during mount/umount/freeze/quotactl */
1208                 locked = down_read_trylock(&sbi->sb->s_umount);
1209                 f2fs_quota_sync(sbi->sb, -1);
1210                 if (locked)
1211                         up_read(&sbi->sb->s_umount);
1212                 cond_resched();
1213                 goto retry_flush_quotas;
1214         }
1215
1216 retry_flush_dents:
1217         /* write all the dirty dentry pages */
1218         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1219                 f2fs_unlock_all(sbi);
1220                 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1221                 if (err)
1222                         return err;
1223                 cond_resched();
1224                 goto retry_flush_quotas;
1225         }
1226
1227         /*
1228          * POR: we should ensure that there are no dirty node pages
1229          * until finishing nat/sit flush. inode->i_blocks can be updated.
1230          */
1231         down_write(&sbi->node_change);
1232
1233         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1234                 up_write(&sbi->node_change);
1235                 f2fs_unlock_all(sbi);
1236                 err = f2fs_sync_inode_meta(sbi);
1237                 if (err)
1238                         return err;
1239                 cond_resched();
1240                 goto retry_flush_quotas;
1241         }
1242
1243 retry_flush_nodes:
1244         down_write(&sbi->node_write);
1245
1246         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1247                 up_write(&sbi->node_write);
1248                 atomic_inc(&sbi->wb_sync_req[NODE]);
1249                 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1250                 atomic_dec(&sbi->wb_sync_req[NODE]);
1251                 if (err) {
1252                         up_write(&sbi->node_change);
1253                         f2fs_unlock_all(sbi);
1254                         return err;
1255                 }
1256                 cond_resched();
1257                 goto retry_flush_nodes;
1258         }
1259
1260         /*
1261          * sbi->node_change is used only for AIO write_begin path which produces
1262          * dirty node blocks and some checkpoint values by block allocation.
1263          */
1264         __prepare_cp_block(sbi);
1265         up_write(&sbi->node_change);
1266         return err;
1267 }
1268
1269 static void unblock_operations(struct f2fs_sb_info *sbi)
1270 {
1271         up_write(&sbi->node_write);
1272         f2fs_unlock_all(sbi);
1273 }
1274
1275 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1276 {
1277         DEFINE_WAIT(wait);
1278
1279         for (;;) {
1280                 if (!get_pages(sbi, type))
1281                         break;
1282
1283                 if (unlikely(f2fs_cp_error(sbi)))
1284                         break;
1285
1286                 if (type == F2FS_DIRTY_META)
1287                         f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1288                                                         FS_CP_META_IO);
1289                 else if (type == F2FS_WB_CP_DATA)
1290                         f2fs_submit_merged_write(sbi, DATA);
1291
1292                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1293                 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1294         }
1295         finish_wait(&sbi->cp_wait, &wait);
1296 }
1297
1298 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1299 {
1300         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1301         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1302         unsigned long flags;
1303
1304         spin_lock_irqsave(&sbi->cp_lock, flags);
1305
1306         if ((cpc->reason & CP_UMOUNT) &&
1307                         le32_to_cpu(ckpt->cp_pack_total_block_count) >
1308                         sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1309                 disable_nat_bits(sbi, false);
1310
1311         if (cpc->reason & CP_TRIMMED)
1312                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1313         else
1314                 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1315
1316         if (cpc->reason & CP_UMOUNT)
1317                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1318         else
1319                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1320
1321         if (cpc->reason & CP_FASTBOOT)
1322                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1323         else
1324                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1325
1326         if (orphan_num)
1327                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1328         else
1329                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1330
1331         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1332                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1333
1334         if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1335                 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1336         else
1337                 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1338
1339         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1340                 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1341         else
1342                 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1343
1344         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1345                 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1346         else
1347                 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1348
1349         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1350                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1351         else
1352                 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1353
1354         if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1355                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1356
1357         /* set this flag to activate crc|cp_ver for recovery */
1358         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1359         __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1360
1361         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1362 }
1363
1364 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1365         void *src, block_t blk_addr)
1366 {
1367         struct writeback_control wbc = {
1368                 .for_reclaim = 0,
1369         };
1370
1371         /*
1372          * pagevec_lookup_tag and lock_page again will take
1373          * some extra time. Therefore, f2fs_update_meta_pages and
1374          * f2fs_sync_meta_pages are combined in this function.
1375          */
1376         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1377         int err;
1378
1379         f2fs_wait_on_page_writeback(page, META, true, true);
1380
1381         memcpy(page_address(page), src, PAGE_SIZE);
1382
1383         set_page_dirty(page);
1384         if (unlikely(!clear_page_dirty_for_io(page)))
1385                 f2fs_bug_on(sbi, 1);
1386
1387         /* writeout cp pack 2 page */
1388         err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1389         if (unlikely(err && f2fs_cp_error(sbi))) {
1390                 f2fs_put_page(page, 1);
1391                 return;
1392         }
1393
1394         f2fs_bug_on(sbi, err);
1395         f2fs_put_page(page, 0);
1396
1397         /* submit checkpoint (with barrier if NOBARRIER is not set) */
1398         f2fs_submit_merged_write(sbi, META_FLUSH);
1399 }
1400
1401 static inline u64 get_sectors_written(struct block_device *bdev)
1402 {
1403         return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1404 }
1405
1406 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1407 {
1408         if (f2fs_is_multi_device(sbi)) {
1409                 u64 sectors = 0;
1410                 int i;
1411
1412                 for (i = 0; i < sbi->s_ndevs; i++)
1413                         sectors += get_sectors_written(FDEV(i).bdev);
1414
1415                 return sectors;
1416         }
1417
1418         return get_sectors_written(sbi->sb->s_bdev);
1419 }
1420
1421 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1422 {
1423         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1424         struct f2fs_nm_info *nm_i = NM_I(sbi);
1425         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1426         block_t start_blk;
1427         unsigned int data_sum_blocks, orphan_blocks;
1428         __u32 crc32 = 0;
1429         int i;
1430         int cp_payload_blks = __cp_payload(sbi);
1431         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1432         u64 kbytes_written;
1433         int err;
1434
1435         /* Flush all the NAT/SIT pages */
1436         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1437
1438         /* start to update checkpoint, cp ver is already updated previously */
1439         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1440         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1441         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1442                 ckpt->cur_node_segno[i] =
1443                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1444                 ckpt->cur_node_blkoff[i] =
1445                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1446                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1447                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1448         }
1449         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1450                 ckpt->cur_data_segno[i] =
1451                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1452                 ckpt->cur_data_blkoff[i] =
1453                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1454                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1455                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1456         }
1457
1458         /* 2 cp + n data seg summary + orphan inode blocks */
1459         data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1460         spin_lock_irqsave(&sbi->cp_lock, flags);
1461         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1462                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1463         else
1464                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1465         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1466
1467         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1468         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1469                         orphan_blocks);
1470
1471         if (__remain_node_summaries(cpc->reason))
1472                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1473                                 cp_payload_blks + data_sum_blocks +
1474                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1475         else
1476                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1477                                 cp_payload_blks + data_sum_blocks +
1478                                 orphan_blocks);
1479
1480         /* update ckpt flag for checkpoint */
1481         update_ckpt_flags(sbi, cpc);
1482
1483         /* update SIT/NAT bitmap */
1484         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1485         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1486
1487         crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1488         *((__le32 *)((unsigned char *)ckpt +
1489                                 le32_to_cpu(ckpt->checksum_offset)))
1490                                 = cpu_to_le32(crc32);
1491
1492         start_blk = __start_cp_next_addr(sbi);
1493
1494         /* write nat bits */
1495         if (enabled_nat_bits(sbi, cpc)) {
1496                 __u64 cp_ver = cur_cp_version(ckpt);
1497                 block_t blk;
1498
1499                 cp_ver |= ((__u64)crc32 << 32);
1500                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1501
1502                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1503                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1504                         f2fs_update_meta_page(sbi, nm_i->nat_bits +
1505                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1506         }
1507
1508         /* write out checkpoint buffer at block 0 */
1509         f2fs_update_meta_page(sbi, ckpt, start_blk++);
1510
1511         for (i = 1; i < 1 + cp_payload_blks; i++)
1512                 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1513                                                         start_blk++);
1514
1515         if (orphan_num) {
1516                 write_orphan_inodes(sbi, start_blk);
1517                 start_blk += orphan_blocks;
1518         }
1519
1520         f2fs_write_data_summaries(sbi, start_blk);
1521         start_blk += data_sum_blocks;
1522
1523         /* Record write statistics in the hot node summary */
1524         kbytes_written = sbi->kbytes_written;
1525         kbytes_written += (f2fs_get_sectors_written(sbi) -
1526                                 sbi->sectors_written_start) >> 1;
1527         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1528
1529         if (__remain_node_summaries(cpc->reason)) {
1530                 f2fs_write_node_summaries(sbi, start_blk);
1531                 start_blk += NR_CURSEG_NODE_TYPE;
1532         }
1533
1534         /* update user_block_counts */
1535         sbi->last_valid_block_count = sbi->total_valid_block_count;
1536         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1537
1538         /* Here, we have one bio having CP pack except cp pack 2 page */
1539         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1540         /* Wait for all dirty meta pages to be submitted for IO */
1541         f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1542
1543         /* wait for previous submitted meta pages writeback */
1544         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1545
1546         /* flush all device cache */
1547         err = f2fs_flush_device_cache(sbi);
1548         if (err)
1549                 return err;
1550
1551         /* barrier and flush checkpoint cp pack 2 page if it can */
1552         commit_checkpoint(sbi, ckpt, start_blk);
1553         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1554
1555         /*
1556          * invalidate intermediate page cache borrowed from meta inode which are
1557          * used for migration of encrypted, verity or compressed inode's blocks.
1558          */
1559         if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1560                 f2fs_sb_has_compression(sbi))
1561                 invalidate_mapping_pages(META_MAPPING(sbi),
1562                                 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1563
1564         f2fs_release_ino_entry(sbi, false);
1565
1566         f2fs_reset_fsync_node_info(sbi);
1567
1568         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1569         clear_sbi_flag(sbi, SBI_NEED_CP);
1570         clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1571
1572         spin_lock(&sbi->stat_lock);
1573         sbi->unusable_block_count = 0;
1574         spin_unlock(&sbi->stat_lock);
1575
1576         __set_cp_next_pack(sbi);
1577
1578         /*
1579          * redirty superblock if metadata like node page or inode cache is
1580          * updated during writing checkpoint.
1581          */
1582         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1583                         get_pages(sbi, F2FS_DIRTY_IMETA))
1584                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1585
1586         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1587
1588         return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1589 }
1590
1591 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1592 {
1593         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1594         unsigned long long ckpt_ver;
1595         int err = 0;
1596
1597         if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1598                 return -EROFS;
1599
1600         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1601                 if (cpc->reason != CP_PAUSE)
1602                         return 0;
1603                 f2fs_warn(sbi, "Start checkpoint disabled!");
1604         }
1605         if (cpc->reason != CP_RESIZE)
1606                 down_write(&sbi->cp_global_sem);
1607
1608         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1609                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1610                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1611                 goto out;
1612         if (unlikely(f2fs_cp_error(sbi))) {
1613                 err = -EIO;
1614                 goto out;
1615         }
1616
1617         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1618
1619         err = block_operations(sbi);
1620         if (err)
1621                 goto out;
1622
1623         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1624
1625         f2fs_flush_merged_writes(sbi);
1626
1627         /* this is the case of multiple fstrims without any changes */
1628         if (cpc->reason & CP_DISCARD) {
1629                 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1630                         unblock_operations(sbi);
1631                         goto out;
1632                 }
1633
1634                 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1635                                 SIT_I(sbi)->dirty_sentries == 0 &&
1636                                 prefree_segments(sbi) == 0) {
1637                         f2fs_flush_sit_entries(sbi, cpc);
1638                         f2fs_clear_prefree_segments(sbi, cpc);
1639                         unblock_operations(sbi);
1640                         goto out;
1641                 }
1642         }
1643
1644         /*
1645          * update checkpoint pack index
1646          * Increase the version number so that
1647          * SIT entries and seg summaries are written at correct place
1648          */
1649         ckpt_ver = cur_cp_version(ckpt);
1650         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1651
1652         /* write cached NAT/SIT entries to NAT/SIT area */
1653         err = f2fs_flush_nat_entries(sbi, cpc);
1654         if (err) {
1655                 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1656                 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1657                 goto stop;
1658         }
1659
1660         f2fs_flush_sit_entries(sbi, cpc);
1661
1662         /* save inmem log status */
1663         f2fs_save_inmem_curseg(sbi);
1664
1665         err = do_checkpoint(sbi, cpc);
1666         if (err) {
1667                 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1668                 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1669                 f2fs_release_discard_addrs(sbi);
1670         } else {
1671                 f2fs_clear_prefree_segments(sbi, cpc);
1672         }
1673
1674         f2fs_restore_inmem_curseg(sbi);
1675 stop:
1676         unblock_operations(sbi);
1677         stat_inc_cp_count(sbi->stat_info);
1678
1679         if (cpc->reason & CP_RECOVERY)
1680                 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1681
1682         /* update CP_TIME to trigger checkpoint periodically */
1683         f2fs_update_time(sbi, CP_TIME);
1684         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1685 out:
1686         if (cpc->reason != CP_RESIZE)
1687                 up_write(&sbi->cp_global_sem);
1688         return err;
1689 }
1690
1691 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1692 {
1693         int i;
1694
1695         for (i = 0; i < MAX_INO_ENTRY; i++) {
1696                 struct inode_management *im = &sbi->im[i];
1697
1698                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1699                 spin_lock_init(&im->ino_lock);
1700                 INIT_LIST_HEAD(&im->ino_list);
1701                 im->ino_num = 0;
1702         }
1703
1704         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1705                         NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1706                                 F2FS_ORPHANS_PER_BLOCK;
1707 }
1708
1709 int __init f2fs_create_checkpoint_caches(void)
1710 {
1711         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1712                         sizeof(struct ino_entry));
1713         if (!ino_entry_slab)
1714                 return -ENOMEM;
1715         f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1716                         sizeof(struct inode_entry));
1717         if (!f2fs_inode_entry_slab) {
1718                 kmem_cache_destroy(ino_entry_slab);
1719                 return -ENOMEM;
1720         }
1721         return 0;
1722 }
1723
1724 void f2fs_destroy_checkpoint_caches(void)
1725 {
1726         kmem_cache_destroy(ino_entry_slab);
1727         kmem_cache_destroy(f2fs_inode_entry_slab);
1728 }
1729
1730 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1731 {
1732         struct cp_control cpc = { .reason = CP_SYNC, };
1733         int err;
1734
1735         down_write(&sbi->gc_lock);
1736         err = f2fs_write_checkpoint(sbi, &cpc);
1737         up_write(&sbi->gc_lock);
1738
1739         return err;
1740 }
1741
1742 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1743 {
1744         struct ckpt_req_control *cprc = &sbi->cprc_info;
1745         struct ckpt_req *req, *next;
1746         struct llist_node *dispatch_list;
1747         u64 sum_diff = 0, diff, count = 0;
1748         int ret;
1749
1750         dispatch_list = llist_del_all(&cprc->issue_list);
1751         if (!dispatch_list)
1752                 return;
1753         dispatch_list = llist_reverse_order(dispatch_list);
1754
1755         ret = __write_checkpoint_sync(sbi);
1756         atomic_inc(&cprc->issued_ckpt);
1757
1758         llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1759                 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1760                 req->ret = ret;
1761                 complete(&req->wait);
1762
1763                 sum_diff += diff;
1764                 count++;
1765         }
1766         atomic_sub(count, &cprc->queued_ckpt);
1767         atomic_add(count, &cprc->total_ckpt);
1768
1769         spin_lock(&cprc->stat_lock);
1770         cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1771         if (cprc->peak_time < cprc->cur_time)
1772                 cprc->peak_time = cprc->cur_time;
1773         spin_unlock(&cprc->stat_lock);
1774 }
1775
1776 static int issue_checkpoint_thread(void *data)
1777 {
1778         struct f2fs_sb_info *sbi = data;
1779         struct ckpt_req_control *cprc = &sbi->cprc_info;
1780         wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1781 repeat:
1782         if (kthread_should_stop())
1783                 return 0;
1784
1785         if (!llist_empty(&cprc->issue_list))
1786                 __checkpoint_and_complete_reqs(sbi);
1787
1788         wait_event_interruptible(*q,
1789                 kthread_should_stop() || !llist_empty(&cprc->issue_list));
1790         goto repeat;
1791 }
1792
1793 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1794                 struct ckpt_req *wait_req)
1795 {
1796         struct ckpt_req_control *cprc = &sbi->cprc_info;
1797
1798         if (!llist_empty(&cprc->issue_list)) {
1799                 __checkpoint_and_complete_reqs(sbi);
1800         } else {
1801                 /* already dispatched by issue_checkpoint_thread */
1802                 if (wait_req)
1803                         wait_for_completion(&wait_req->wait);
1804         }
1805 }
1806
1807 static void init_ckpt_req(struct ckpt_req *req)
1808 {
1809         memset(req, 0, sizeof(struct ckpt_req));
1810
1811         init_completion(&req->wait);
1812         req->queue_time = ktime_get();
1813 }
1814
1815 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1816 {
1817         struct ckpt_req_control *cprc = &sbi->cprc_info;
1818         struct ckpt_req req;
1819         struct cp_control cpc;
1820
1821         cpc.reason = __get_cp_reason(sbi);
1822         if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1823                 int ret;
1824
1825                 down_write(&sbi->gc_lock);
1826                 ret = f2fs_write_checkpoint(sbi, &cpc);
1827                 up_write(&sbi->gc_lock);
1828
1829                 return ret;
1830         }
1831
1832         if (!cprc->f2fs_issue_ckpt)
1833                 return __write_checkpoint_sync(sbi);
1834
1835         init_ckpt_req(&req);
1836
1837         llist_add(&req.llnode, &cprc->issue_list);
1838         atomic_inc(&cprc->queued_ckpt);
1839
1840         /*
1841          * update issue_list before we wake up issue_checkpoint thread,
1842          * this smp_mb() pairs with another barrier in ___wait_event(),
1843          * see more details in comments of waitqueue_active().
1844          */
1845         smp_mb();
1846
1847         if (waitqueue_active(&cprc->ckpt_wait_queue))
1848                 wake_up(&cprc->ckpt_wait_queue);
1849
1850         if (cprc->f2fs_issue_ckpt)
1851                 wait_for_completion(&req.wait);
1852         else
1853                 flush_remained_ckpt_reqs(sbi, &req);
1854
1855         return req.ret;
1856 }
1857
1858 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1859 {
1860         dev_t dev = sbi->sb->s_bdev->bd_dev;
1861         struct ckpt_req_control *cprc = &sbi->cprc_info;
1862
1863         if (cprc->f2fs_issue_ckpt)
1864                 return 0;
1865
1866         cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1867                         "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1868         if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1869                 cprc->f2fs_issue_ckpt = NULL;
1870                 return -ENOMEM;
1871         }
1872
1873         set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1874
1875         return 0;
1876 }
1877
1878 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1879 {
1880         struct ckpt_req_control *cprc = &sbi->cprc_info;
1881
1882         if (cprc->f2fs_issue_ckpt) {
1883                 struct task_struct *ckpt_task = cprc->f2fs_issue_ckpt;
1884
1885                 cprc->f2fs_issue_ckpt = NULL;
1886                 kthread_stop(ckpt_task);
1887
1888                 flush_remained_ckpt_reqs(sbi, NULL);
1889         }
1890 }
1891
1892 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1893 {
1894         struct ckpt_req_control *cprc = &sbi->cprc_info;
1895
1896         atomic_set(&cprc->issued_ckpt, 0);
1897         atomic_set(&cprc->total_ckpt, 0);
1898         atomic_set(&cprc->queued_ckpt, 0);
1899         cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1900         init_waitqueue_head(&cprc->ckpt_wait_queue);
1901         init_llist_head(&cprc->issue_list);
1902         spin_lock_init(&cprc->stat_lock);
1903 }