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
  17 #include "f2fs.h"
  18 #include "node.h"
  19 #include "segment.h"
  20 #include "trace.h"
  21 #include <trace/events/f2fs.h>
  22
  23 static struct kmem_cache *ino_entry_slab;
  24 struct kmem_cache *f2fs_inode_entry_slab;
  25
  26 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
  27 {
  28         f2fs_build_fault_attr(sbi, 0, 0);
  29         set_ckpt_flags(sbi, CP_ERROR_FLAG);
  30         if (!end_io)
  31                 f2fs_flush_merged_writes(sbi);
  32 }
  33
  34 /*
  35  * We guarantee no failure on the returned page.
  36  */
  37 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  38 {
  39         struct address_space *mapping = META_MAPPING(sbi);
  40         struct page *page = NULL;
  41 repeat:
  42         page = f2fs_grab_cache_page(mapping, index, false);
  43         if (!page) {
  44                 cond_resched();
  45                 goto repeat;
  46         }
  47         f2fs_wait_on_page_writeback(page, META, true, true);
  48         if (!PageUptodate(page))
  49                 SetPageUptodate(page);
  50         return page;
  51 }
  52
  53 /*
  54  * We guarantee no failure on the returned page.
  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         lock_page(page);
  93         if (unlikely(page->mapping != mapping)) {
  94                 f2fs_put_page(page, 1);
  95                 goto repeat;
  96         }
  97
  98         if (unlikely(!PageUptodate(page))) {
  99                 f2fs_put_page(page, 1);
 100                 return ERR_PTR(-EIO);
 101         }
 102 out:
 103         return page;
 104 }
 105
 106 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 107 {
 108         return __get_meta_page(sbi, index, true);
 109 }
 110
 111 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
 112 {
 113         struct page *page;
 114         int count = 0;
 115
 116 retry:
 117         page = __get_meta_page(sbi, index, true);
 118         if (IS_ERR(page)) {
 119                 if (PTR_ERR(page) == -EIO &&
 120                                 ++count <= DEFAULT_RETRY_IO_COUNT)
 121                         goto retry;
 122                 f2fs_stop_checkpoint(sbi, false);
 123         }
 124         return page;
 125 }
 126
 127 /* for POR only */
 128 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 129 {
 130         return __get_meta_page(sbi, index, false);
 131 }
 132
 133 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
 134                                                         int type)
 135 {
 136         struct seg_entry *se;
 137         unsigned int segno, offset;
 138         bool exist;
 139
 140         if (type == DATA_GENERIC)
 141                 return true;
 142
 143         segno = GET_SEGNO(sbi, blkaddr);
 144         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 145         se = get_seg_entry(sbi, segno);
 146
 147         exist = f2fs_test_bit(offset, se->cur_valid_map);
 148         if (exist && type == DATA_GENERIC_ENHANCE_UPDATE) {
 149                 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
 150                          blkaddr, exist);
 151                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 152                 return exist;
 153         }
 154
 155         if (!exist && type == DATA_GENERIC_ENHANCE) {
 156                 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
 157                          blkaddr, exist);
 158                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 159                 dump_stack();
 160         }
 161         return exist;
 162 }
 163
 164 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
 165                                         block_t blkaddr, int type)
 166 {
 167         switch (type) {
 168         case META_NAT:
 169                 break;
 170         case META_SIT:
 171                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 172                         return false;
 173                 break;
 174         case META_SSA:
 175                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 176                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 177                         return false;
 178                 break;
 179         case META_CP:
 180                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 181                         blkaddr < __start_cp_addr(sbi)))
 182                         return false;
 183                 break;
 184         case META_POR:
 185                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 186                         blkaddr < MAIN_BLKADDR(sbi)))
 187                         return false;
 188                 break;
 189         case DATA_GENERIC:
 190         case DATA_GENERIC_ENHANCE:
 191         case DATA_GENERIC_ENHANCE_READ:
 192         case DATA_GENERIC_ENHANCE_UPDATE:
 193                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 194                                 blkaddr < MAIN_BLKADDR(sbi))) {
 195                         f2fs_warn(sbi, "access invalid blkaddr:%u",
 196                                   blkaddr);
 197                         set_sbi_flag(sbi, SBI_NEED_FSCK);
 198                         dump_stack();
 199                         return false;
 200                 } else {
 201                         return __is_bitmap_valid(sbi, blkaddr, type);
 202                 }
 203                 break;
 204         case META_GENERIC:
 205                 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
 206                         blkaddr >= MAIN_BLKADDR(sbi)))
 207                         return false;
 208                 break;
 209         default:
 210                 BUG();
 211         }
 212
 213         return true;
 214 }
 215
 216 /*
 217  * Readahead CP/NAT/SIT/SSA pages
 218  */
 219 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 220                                                         int type, bool sync)
 221 {
 222         struct page *page;
 223         block_t blkno = start;
 224         struct f2fs_io_info fio = {
 225                 .sbi = sbi,
 226                 .type = META,
 227                 .op = REQ_OP_READ,
 228                 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
 229                 .encrypted_page = NULL,
 230                 .in_list = false,
 231                 .is_por = (type == META_POR),
 232         };
 233         struct blk_plug plug;
 234
 235         if (unlikely(type == META_POR))
 236                 fio.op_flags &= ~REQ_META;
 237
 238         blk_start_plug(&plug);
 239         for (; nrpages-- > 0; blkno++) {
 240
 241                 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
 242                         goto out;
 243
 244                 switch (type) {
 245                 case META_NAT:
 246                         if (unlikely(blkno >=
 247                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 248                                 blkno = 0;
 249                         /* get nat block addr */
 250                         fio.new_blkaddr = current_nat_addr(sbi,
 251                                         blkno * NAT_ENTRY_PER_BLOCK);
 252                         break;
 253                 case META_SIT:
 254                         if (unlikely(blkno >= TOTAL_SEGS(sbi)))
 255                                 goto out;
 256                         /* get sit block addr */
 257                         fio.new_blkaddr = current_sit_addr(sbi,
 258                                         blkno * SIT_ENTRY_PER_BLOCK);
 259                         break;
 260                 case META_SSA:
 261                 case META_CP:
 262                 case META_POR:
 263                         fio.new_blkaddr = blkno;
 264                         break;
 265                 default:
 266                         BUG();
 267                 }
 268
 269                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 270                                                 fio.new_blkaddr, false);
 271                 if (!page)
 272                         continue;
 273                 if (PageUptodate(page)) {
 274                         f2fs_put_page(page, 1);
 275                         continue;
 276                 }
 277
 278                 fio.page = page;
 279                 f2fs_submit_page_bio(&fio);
 280                 f2fs_put_page(page, 0);
 281         }
 282 out:
 283         blk_finish_plug(&plug);
 284         return blkno - start;
 285 }
 286
 287 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 288 {
 289         struct page *page;
 290         bool readahead = false;
 291
 292         page = find_get_page(META_MAPPING(sbi), index);
 293         if (!page || !PageUptodate(page))
 294                 readahead = true;
 295         f2fs_put_page(page, 0);
 296
 297         if (readahead)
 298                 f2fs_ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
 299 }
 300
 301 static int __f2fs_write_meta_page(struct page *page,
 302                                 struct writeback_control *wbc,
 303                                 enum iostat_type io_type)
 304 {
 305         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 306
 307         trace_f2fs_writepage(page, META);
 308
 309         if (unlikely(f2fs_cp_error(sbi))) {
 310                 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
 311                         ClearPageUptodate(page);
 312                         dec_page_count(sbi, F2FS_DIRTY_META);
 313                         unlock_page(page);
 314                         return 0;
 315                 }
 316                 goto redirty_out;
 317         }
 318         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 319                 goto redirty_out;
 320         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 321                 goto redirty_out;
 322
 323         f2fs_do_write_meta_page(sbi, page, io_type);
 324         dec_page_count(sbi, F2FS_DIRTY_META);
 325
 326         if (wbc->for_reclaim)
 327                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
 328
 329         unlock_page(page);
 330
 331         if (unlikely(f2fs_cp_error(sbi)))
 332                 f2fs_submit_merged_write(sbi, META);
 333
 334         return 0;
 335
 336 redirty_out:
 337         redirty_page_for_writepage(wbc, page);
 338         return AOP_WRITEPAGE_ACTIVATE;
 339 }
 340
 341 static int f2fs_write_meta_page(struct page *page,
 342                                 struct writeback_control *wbc)
 343 {
 344         return __f2fs_write_meta_page(page, wbc, FS_META_IO);
 345 }
 346
 347 static int f2fs_write_meta_pages(struct address_space *mapping,
 348                                 struct writeback_control *wbc)
 349 {
 350         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 351         long diff, written;
 352
 353         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 354                 goto skip_write;
 355
 356         /* collect a number of dirty meta pages and write together */
 357         if (wbc->sync_mode != WB_SYNC_ALL &&
 358                         get_pages(sbi, F2FS_DIRTY_META) <
 359                                         nr_pages_to_skip(sbi, META))
 360                 goto skip_write;
 361
 362         /* if locked failed, cp will flush dirty pages instead */
 363         if (!mutex_trylock(&sbi->cp_mutex))
 364                 goto skip_write;
 365
 366         trace_f2fs_writepages(mapping->host, wbc, META);
 367         diff = nr_pages_to_write(sbi, META, wbc);
 368         written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
 369         mutex_unlock(&sbi->cp_mutex);
 370         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 371         return 0;
 372
 373 skip_write:
 374         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 375         trace_f2fs_writepages(mapping->host, wbc, META);
 376         return 0;
 377 }
 378
 379 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 380                                 long nr_to_write, enum iostat_type io_type)
 381 {
 382         struct address_space *mapping = META_MAPPING(sbi);
 383         pgoff_t index = 0, prev = ULONG_MAX;
 384         struct pagevec pvec;
 385         long nwritten = 0;
 386         int nr_pages;
 387         struct writeback_control wbc = {
 388                 .for_reclaim = 0,
 389         };
 390         struct blk_plug plug;
 391
 392         pagevec_init(&pvec);
 393
 394         blk_start_plug(&plug);
 395
 396         while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 397                                 PAGECACHE_TAG_DIRTY))) {
 398                 int i;
 399
 400                 for (i = 0; i < nr_pages; i++) {
 401                         struct page *page = pvec.pages[i];
 402
 403                         if (prev == ULONG_MAX)
 404                                 prev = page->index - 1;
 405                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 406                                 pagevec_release(&pvec);
 407                                 goto stop;
 408                         }
 409
 410                         lock_page(page);
 411
 412                         if (unlikely(page->mapping != mapping)) {
 413 continue_unlock:
 414                                 unlock_page(page);
 415                                 continue;
 416                         }
 417                         if (!PageDirty(page)) {
 418                                 /* someone wrote it for us */
 419                                 goto continue_unlock;
 420                         }
 421
 422                         f2fs_wait_on_page_writeback(page, META, true, true);
 423
 424                         if (!clear_page_dirty_for_io(page))
 425                                 goto continue_unlock;
 426
 427                         if (__f2fs_write_meta_page(page, &wbc, io_type)) {
 428                                 unlock_page(page);
 429                                 break;
 430                         }
 431                         nwritten++;
 432                         prev = page->index;
 433                         if (unlikely(nwritten >= nr_to_write))
 434                                 break;
 435                 }
 436                 pagevec_release(&pvec);
 437                 cond_resched();
 438         }
 439 stop:
 440         if (nwritten)
 441                 f2fs_submit_merged_write(sbi, type);
 442
 443         blk_finish_plug(&plug);
 444
 445         return nwritten;
 446 }
 447
 448 static int f2fs_set_meta_page_dirty(struct page *page)
 449 {
 450         trace_f2fs_set_page_dirty(page, META);
 451
 452         if (!PageUptodate(page))
 453                 SetPageUptodate(page);
 454         if (!PageDirty(page)) {
 455                 __set_page_dirty_nobuffers(page);
 456                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 457                 f2fs_set_page_private(page, 0);
 458                 f2fs_trace_pid(page);
 459                 return 1;
 460         }
 461         return 0;
 462 }
 463
 464 const struct address_space_operations f2fs_meta_aops = {
 465         .writepage      = f2fs_write_meta_page,
 466         .writepages     = f2fs_write_meta_pages,
 467         .set_page_dirty = f2fs_set_meta_page_dirty,
 468         .invalidatepage = f2fs_invalidate_page,
 469         .releasepage    = f2fs_release_page,
 470 #ifdef CONFIG_MIGRATION
 471         .migratepage    = f2fs_migrate_page,
 472 #endif
 473 };
 474
 475 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 476                                                 unsigned int devidx, int type)
 477 {
 478         struct inode_management *im = &sbi->im[type];
 479         struct ino_entry *e, *tmp;
 480
 481         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 482
 483         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 484
 485         spin_lock(&im->ino_lock);
 486         e = radix_tree_lookup(&im->ino_root, ino);
 487         if (!e) {
 488                 e = tmp;
 489                 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
 490                         f2fs_bug_on(sbi, 1);
 491
 492                 memset(e, 0, sizeof(struct ino_entry));
 493                 e->ino = ino;
 494
 495                 list_add_tail(&e->list, &im->ino_list);
 496                 if (type != ORPHAN_INO)
 497                         im->ino_num++;
 498         }
 499
 500         if (type == FLUSH_INO)
 501                 f2fs_set_bit(devidx, (char *)&e->dirty_device);
 502
 503         spin_unlock(&im->ino_lock);
 504         radix_tree_preload_end();
 505
 506         if (e != tmp)
 507                 kmem_cache_free(ino_entry_slab, tmp);
 508 }
 509
 510 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 511 {
 512         struct inode_management *im = &sbi->im[type];
 513         struct ino_entry *e;
 514
 515         spin_lock(&im->ino_lock);
 516         e = radix_tree_lookup(&im->ino_root, ino);
 517         if (e) {
 518                 list_del(&e->list);
 519                 radix_tree_delete(&im->ino_root, ino);
 520                 im->ino_num--;
 521                 spin_unlock(&im->ino_lock);
 522                 kmem_cache_free(ino_entry_slab, e);
 523                 return;
 524         }
 525         spin_unlock(&im->ino_lock);
 526 }
 527
 528 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 529 {
 530         /* add new dirty ino entry into list */
 531         __add_ino_entry(sbi, ino, 0, type);
 532 }
 533
 534 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 535 {
 536         /* remove dirty ino entry from list */
 537         __remove_ino_entry(sbi, ino, type);
 538 }
 539
 540 /* mode should be APPEND_INO or UPDATE_INO */
 541 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 542 {
 543         struct inode_management *im = &sbi->im[mode];
 544         struct ino_entry *e;
 545
 546         spin_lock(&im->ino_lock);
 547         e = radix_tree_lookup(&im->ino_root, ino);
 548         spin_unlock(&im->ino_lock);
 549         return e ? true : false;
 550 }
 551
 552 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 553 {
 554         struct ino_entry *e, *tmp;
 555         int i;
 556
 557         for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
 558                 struct inode_management *im = &sbi->im[i];
 559
 560                 spin_lock(&im->ino_lock);
 561                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 562                         list_del(&e->list);
 563                         radix_tree_delete(&im->ino_root, e->ino);
 564                         kmem_cache_free(ino_entry_slab, e);
 565                         im->ino_num--;
 566                 }
 567                 spin_unlock(&im->ino_lock);
 568         }
 569 }
 570
 571 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 572                                         unsigned int devidx, int type)
 573 {
 574         __add_ino_entry(sbi, ino, devidx, type);
 575 }
 576
 577 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 578                                         unsigned int devidx, int type)
 579 {
 580         struct inode_management *im = &sbi->im[type];
 581         struct ino_entry *e;
 582         bool is_dirty = false;
 583
 584         spin_lock(&im->ino_lock);
 585         e = radix_tree_lookup(&im->ino_root, ino);
 586         if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
 587                 is_dirty = true;
 588         spin_unlock(&im->ino_lock);
 589         return is_dirty;
 590 }
 591
 592 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
 593 {
 594         struct inode_management *im = &sbi->im[ORPHAN_INO];
 595         int err = 0;
 596
 597         spin_lock(&im->ino_lock);
 598
 599         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 600                 spin_unlock(&im->ino_lock);
 601                 f2fs_show_injection_info(sbi, FAULT_ORPHAN);
 602                 return -ENOSPC;
 603         }
 604
 605         if (unlikely(im->ino_num >= sbi->max_orphans))
 606                 err = -ENOSPC;
 607         else
 608                 im->ino_num++;
 609         spin_unlock(&im->ino_lock);
 610
 611         return err;
 612 }
 613
 614 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
 615 {
 616         struct inode_management *im = &sbi->im[ORPHAN_INO];
 617
 618         spin_lock(&im->ino_lock);
 619         f2fs_bug_on(sbi, im->ino_num == 0);
 620         im->ino_num--;
 621         spin_unlock(&im->ino_lock);
 622 }
 623
 624 void f2fs_add_orphan_inode(struct inode *inode)
 625 {
 626         /* add new orphan ino entry into list */
 627         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
 628         f2fs_update_inode_page(inode);
 629 }
 630
 631 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 632 {
 633         /* remove orphan entry from orphan list */
 634         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 635 }
 636
 637 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 638 {
 639         struct inode *inode;
 640         struct node_info ni;
 641         int err;
 642
 643         inode = f2fs_iget_retry(sbi->sb, ino);
 644         if (IS_ERR(inode)) {
 645                 /*
 646                  * there should be a bug that we can't find the entry
 647                  * to orphan inode.
 648                  */
 649                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 650                 return PTR_ERR(inode);
 651         }
 652
 653         err = dquot_initialize(inode);
 654         if (err) {
 655                 iput(inode);
 656                 goto err_out;
 657         }
 658
 659         clear_nlink(inode);
 660
 661         /* truncate all the data during iput */
 662         iput(inode);
 663
 664         err = f2fs_get_node_info(sbi, ino, &ni);
 665         if (err)
 666                 goto err_out;
 667
 668         /* ENOMEM was fully retried in f2fs_evict_inode. */
 669         if (ni.blk_addr != NULL_ADDR) {
 670                 err = -EIO;
 671                 goto err_out;
 672         }
 673         return 0;
 674
 675 err_out:
 676         set_sbi_flag(sbi, SBI_NEED_FSCK);
 677         f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
 678                   __func__, ino);
 679         return err;
 680 }
 681
 682 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
 683 {
 684         block_t start_blk, orphan_blocks, i, j;
 685         unsigned int s_flags = sbi->sb->s_flags;
 686         int err = 0;
 687 #ifdef CONFIG_QUOTA
 688         int quota_enabled;
 689 #endif
 690
 691         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 692                 return 0;
 693
 694         if (bdev_read_only(sbi->sb->s_bdev)) {
 695                 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
 696                 return 0;
 697         }
 698
 699         if (s_flags & SB_RDONLY) {
 700                 f2fs_info(sbi, "orphan cleanup on readonly fs");
 701                 sbi->sb->s_flags &= ~SB_RDONLY;
 702         }
 703
 704 #ifdef CONFIG_QUOTA
 705         /* Needed for iput() to work correctly and not trash data */
 706         sbi->sb->s_flags |= SB_ACTIVE;
 707
 708         /*
 709          * Turn on quotas which were not enabled for read-only mounts if
 710          * filesystem has quota feature, so that they are updated correctly.
 711          */
 712         quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
 713 #endif
 714
 715         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 716         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 717
 718         f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 719
 720         for (i = 0; i < orphan_blocks; i++) {
 721                 struct page *page;
 722                 struct f2fs_orphan_block *orphan_blk;
 723
 724                 page = f2fs_get_meta_page(sbi, start_blk + i);
 725                 if (IS_ERR(page)) {
 726                         err = PTR_ERR(page);
 727                         goto out;
 728                 }
 729
 730                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 731                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 732                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 733                         err = recover_orphan_inode(sbi, ino);
 734                         if (err) {
 735                                 f2fs_put_page(page, 1);
 736                                 goto out;
 737                         }
 738                 }
 739                 f2fs_put_page(page, 1);
 740         }
 741         /* clear Orphan Flag */
 742         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 743 out:
 744         set_sbi_flag(sbi, SBI_IS_RECOVERED);
 745
 746 #ifdef CONFIG_QUOTA
 747         /* Turn quotas off */
 748         if (quota_enabled)
 749                 f2fs_quota_off_umount(sbi->sb);
 750 #endif
 751         sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
 752
 753         return err;
 754 }
 755
 756 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 757 {
 758         struct list_head *head;
 759         struct f2fs_orphan_block *orphan_blk = NULL;
 760         unsigned int nentries = 0;
 761         unsigned short index = 1;
 762         unsigned short orphan_blocks;
 763         struct page *page = NULL;
 764         struct ino_entry *orphan = NULL;
 765         struct inode_management *im = &sbi->im[ORPHAN_INO];
 766
 767         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 768
 769         /*
 770          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 771          * orphan inode operations are covered under f2fs_lock_op().
 772          * And, spin_lock should be avoided due to page operations below.
 773          */
 774         head = &im->ino_list;
 775
 776         /* loop for each orphan inode entry and write them in Jornal block */
 777         list_for_each_entry(orphan, head, list) {
 778                 if (!page) {
 779                         page = f2fs_grab_meta_page(sbi, start_blk++);
 780                         orphan_blk =
 781                                 (struct f2fs_orphan_block *)page_address(page);
 782                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 783                 }
 784
 785                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 786
 787                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 788                         /*
 789                          * an orphan block is full of 1020 entries,
 790                          * then we need to flush current orphan blocks
 791                          * and bring another one in memory
 792                          */
 793                         orphan_blk->blk_addr = cpu_to_le16(index);
 794                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 795                         orphan_blk->entry_count = cpu_to_le32(nentries);
 796                         set_page_dirty(page);
 797                         f2fs_put_page(page, 1);
 798                         index++;
 799                         nentries = 0;
 800                         page = NULL;
 801                 }
 802         }
 803
 804         if (page) {
 805                 orphan_blk->blk_addr = cpu_to_le16(index);
 806                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 807                 orphan_blk->entry_count = cpu_to_le32(nentries);
 808                 set_page_dirty(page);
 809                 f2fs_put_page(page, 1);
 810         }
 811 }
 812
 813 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
 814                                                 struct f2fs_checkpoint *ckpt)
 815 {
 816         unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
 817         __u32 chksum;
 818
 819         chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
 820         if (chksum_ofs < CP_CHKSUM_OFFSET) {
 821                 chksum_ofs += sizeof(chksum);
 822                 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
 823                                                 F2FS_BLKSIZE - chksum_ofs);
 824         }
 825         return chksum;
 826 }
 827
 828 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 829                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 830                 unsigned long long *version)
 831 {
 832         size_t crc_offset = 0;
 833         __u32 crc;
 834
 835         *cp_page = f2fs_get_meta_page(sbi, cp_addr);
 836         if (IS_ERR(*cp_page))
 837                 return PTR_ERR(*cp_page);
 838
 839         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 840
 841         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 842         if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
 843                         crc_offset > CP_CHKSUM_OFFSET) {
 844                 f2fs_put_page(*cp_page, 1);
 845                 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
 846                 return -EINVAL;
 847         }
 848
 849         crc = f2fs_checkpoint_chksum(sbi, *cp_block);
 850         if (crc != cur_cp_crc(*cp_block)) {
 851                 f2fs_put_page(*cp_page, 1);
 852                 f2fs_warn(sbi, "invalid crc value");
 853                 return -EINVAL;
 854         }
 855
 856         *version = cur_cp_version(*cp_block);
 857         return 0;
 858 }
 859
 860 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 861                                 block_t cp_addr, unsigned long long *version)
 862 {
 863         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 864         struct f2fs_checkpoint *cp_block = NULL;
 865         unsigned long long cur_version = 0, pre_version = 0;
 866         unsigned int cp_blocks;
 867         int err;
 868
 869         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 870                                         &cp_page_1, version);
 871         if (err)
 872                 return NULL;
 873
 874         cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
 875
 876         if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
 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 += cp_blocks - 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 pack1 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         f2fs_set_page_private(page, 0);
1034         f2fs_trace_pid(page);
1035 }
1036
1037 void f2fs_remove_dirty_inode(struct inode *inode)
1038 {
1039         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1040         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1041
1042         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1043                         !S_ISLNK(inode->i_mode))
1044                 return;
1045
1046         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1047                 return;
1048
1049         spin_lock(&sbi->inode_lock[type]);
1050         __remove_dirty_inode(inode, type);
1051         spin_unlock(&sbi->inode_lock[type]);
1052 }
1053
1054 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1055 {
1056         struct list_head *head;
1057         struct inode *inode;
1058         struct f2fs_inode_info *fi;
1059         bool is_dir = (type == DIR_INODE);
1060         unsigned long ino = 0;
1061
1062         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1063                                 get_pages(sbi, is_dir ?
1064                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1065 retry:
1066         if (unlikely(f2fs_cp_error(sbi))) {
1067                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1068                                 get_pages(sbi, is_dir ?
1069                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1070                 return -EIO;
1071         }
1072
1073         spin_lock(&sbi->inode_lock[type]);
1074
1075         head = &sbi->inode_list[type];
1076         if (list_empty(head)) {
1077                 spin_unlock(&sbi->inode_lock[type]);
1078                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1079                                 get_pages(sbi, is_dir ?
1080                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1081                 return 0;
1082         }
1083         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1084         inode = igrab(&fi->vfs_inode);
1085         spin_unlock(&sbi->inode_lock[type]);
1086         if (inode) {
1087                 unsigned long cur_ino = inode->i_ino;
1088
1089                 F2FS_I(inode)->cp_task = current;
1090
1091                 filemap_fdatawrite(inode->i_mapping);
1092
1093                 F2FS_I(inode)->cp_task = NULL;
1094
1095                 iput(inode);
1096                 /* We need to give cpu to another writers. */
1097                 if (ino == cur_ino)
1098                         cond_resched();
1099                 else
1100                         ino = cur_ino;
1101         } else {
1102                 /*
1103                  * We should submit bio, since it exists several
1104                  * wribacking dentry pages in the freeing inode.
1105                  */
1106                 f2fs_submit_merged_write(sbi, DATA);
1107                 cond_resched();
1108         }
1109         goto retry;
1110 }
1111
1112 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1113 {
1114         struct list_head *head = &sbi->inode_list[DIRTY_META];
1115         struct inode *inode;
1116         struct f2fs_inode_info *fi;
1117         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1118
1119         while (total--) {
1120                 if (unlikely(f2fs_cp_error(sbi)))
1121                         return -EIO;
1122
1123                 spin_lock(&sbi->inode_lock[DIRTY_META]);
1124                 if (list_empty(head)) {
1125                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1126                         return 0;
1127                 }
1128                 fi = list_first_entry(head, struct f2fs_inode_info,
1129                                                         gdirty_list);
1130                 inode = igrab(&fi->vfs_inode);
1131                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1132                 if (inode) {
1133                         sync_inode_metadata(inode, 0);
1134
1135                         /* it's on eviction */
1136                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1137                                 f2fs_update_inode_page(inode);
1138                         iput(inode);
1139                 }
1140         }
1141         return 0;
1142 }
1143
1144 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1145 {
1146         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1147         struct f2fs_nm_info *nm_i = NM_I(sbi);
1148         nid_t last_nid = nm_i->next_scan_nid;
1149
1150         next_free_nid(sbi, &last_nid);
1151         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1152         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1153         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1154         ckpt->next_free_nid = cpu_to_le32(last_nid);
1155 }
1156
1157 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1158 {
1159         bool ret = false;
1160
1161         if (!is_journalled_quota(sbi))
1162                 return false;
1163
1164         if (!down_write_trylock(&sbi->quota_sem))
1165                 return true;
1166         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1167                 ret = false;
1168         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1169                 ret = false;
1170         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1171                 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1172                 ret = true;
1173         } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1174                 ret = true;
1175         }
1176         up_write(&sbi->quota_sem);
1177         return ret;
1178 }
1179
1180 /*
1181  * Freeze all the FS-operations for checkpoint.
1182  */
1183 static int block_operations(struct f2fs_sb_info *sbi)
1184 {
1185         struct writeback_control wbc = {
1186                 .sync_mode = WB_SYNC_ALL,
1187                 .nr_to_write = LONG_MAX,
1188                 .for_reclaim = 0,
1189         };
1190         struct blk_plug plug;
1191         int err = 0, cnt = 0;
1192
1193         blk_start_plug(&plug);
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                         goto out;
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                         goto out;
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                         goto out;
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 out:
1267         blk_finish_plug(&plug);
1268         return err;
1269 }
1270
1271 static void unblock_operations(struct f2fs_sb_info *sbi)
1272 {
1273         up_write(&sbi->node_write);
1274         f2fs_unlock_all(sbi);
1275 }
1276
1277 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1278 {
1279         DEFINE_WAIT(wait);
1280
1281         for (;;) {
1282                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1283
1284                 if (!get_pages(sbi, type))
1285                         break;
1286
1287                 if (unlikely(f2fs_cp_error(sbi) &&
1288                         !is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
1289                         break;
1290
1291                 io_schedule_timeout(HZ/50);
1292         }
1293         finish_wait(&sbi->cp_wait, &wait);
1294 }
1295
1296 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1297 {
1298         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1299         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1300         unsigned long flags;
1301
1302         spin_lock_irqsave(&sbi->cp_lock, flags);
1303
1304         if ((cpc->reason & CP_UMOUNT) &&
1305                         le32_to_cpu(ckpt->cp_pack_total_block_count) >
1306                         sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1307                 disable_nat_bits(sbi, false);
1308
1309         if (cpc->reason & CP_TRIMMED)
1310                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1311         else
1312                 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1313
1314         if (cpc->reason & CP_UMOUNT)
1315                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1316         else
1317                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1318
1319         if (cpc->reason & CP_FASTBOOT)
1320                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1321         else
1322                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1323
1324         if (orphan_num)
1325                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1326         else
1327                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1328
1329         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1330                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1331
1332         if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1333                 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1334         else
1335                 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1336
1337         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1338                 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1339         else
1340                 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1341
1342         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1343                 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1344         else
1345                 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1346
1347         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1348                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1349         else
1350                 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1351
1352         if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1353                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1354
1355         /* set this flag to activate crc|cp_ver for recovery */
1356         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1357         __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1358
1359         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1360 }
1361
1362 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1363         void *src, block_t blk_addr)
1364 {
1365         struct writeback_control wbc = {
1366                 .for_reclaim = 0,
1367         };
1368
1369         /*
1370          * pagevec_lookup_tag and lock_page again will take
1371          * some extra time. Therefore, f2fs_update_meta_pages and
1372          * f2fs_sync_meta_pages are combined in this function.
1373          */
1374         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1375         int err;
1376
1377         f2fs_wait_on_page_writeback(page, META, true, true);
1378
1379         memcpy(page_address(page), src, PAGE_SIZE);
1380
1381         set_page_dirty(page);
1382         if (unlikely(!clear_page_dirty_for_io(page)))
1383                 f2fs_bug_on(sbi, 1);
1384
1385         /* writeout cp pack 2 page */
1386         err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1387         if (unlikely(err && f2fs_cp_error(sbi))) {
1388                 f2fs_put_page(page, 1);
1389                 return;
1390         }
1391
1392         f2fs_bug_on(sbi, err);
1393         f2fs_put_page(page, 0);
1394
1395         /* submit checkpoint (with barrier if NOBARRIER is not set) */
1396         f2fs_submit_merged_write(sbi, META_FLUSH);
1397 }
1398
1399 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1400 {
1401         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1402         struct f2fs_nm_info *nm_i = NM_I(sbi);
1403         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1404         block_t start_blk;
1405         unsigned int data_sum_blocks, orphan_blocks;
1406         __u32 crc32 = 0;
1407         int i;
1408         int cp_payload_blks = __cp_payload(sbi);
1409         struct super_block *sb = sbi->sb;
1410         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1411         u64 kbytes_written;
1412         int err;
1413
1414         /* Flush all the NAT/SIT pages */
1415         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1416
1417         /*
1418          * modify checkpoint
1419          * version number is already updated
1420          */
1421         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1422         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1423         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1424                 ckpt->cur_node_segno[i] =
1425                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1426                 ckpt->cur_node_blkoff[i] =
1427                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1428                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1429                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1430         }
1431         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1432                 ckpt->cur_data_segno[i] =
1433                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1434                 ckpt->cur_data_blkoff[i] =
1435                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1436                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1437                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1438         }
1439
1440         /* 2 cp  + n data seg summary + orphan inode blocks */
1441         data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1442         spin_lock_irqsave(&sbi->cp_lock, flags);
1443         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1444                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1445         else
1446                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1447         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1448
1449         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1450         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1451                         orphan_blocks);
1452
1453         if (__remain_node_summaries(cpc->reason))
1454                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1455                                 cp_payload_blks + data_sum_blocks +
1456                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1457         else
1458                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1459                                 cp_payload_blks + data_sum_blocks +
1460                                 orphan_blocks);
1461
1462         /* update ckpt flag for checkpoint */
1463         update_ckpt_flags(sbi, cpc);
1464
1465         /* update SIT/NAT bitmap */
1466         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1467         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1468
1469         crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1470         *((__le32 *)((unsigned char *)ckpt +
1471                                 le32_to_cpu(ckpt->checksum_offset)))
1472                                 = cpu_to_le32(crc32);
1473
1474         start_blk = __start_cp_next_addr(sbi);
1475
1476         /* write nat bits */
1477         if (enabled_nat_bits(sbi, cpc)) {
1478                 __u64 cp_ver = cur_cp_version(ckpt);
1479                 block_t blk;
1480
1481                 cp_ver |= ((__u64)crc32 << 32);
1482                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1483
1484                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1485                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1486                         f2fs_update_meta_page(sbi, nm_i->nat_bits +
1487                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1488         }
1489
1490         /* write out checkpoint buffer at block 0 */
1491         f2fs_update_meta_page(sbi, ckpt, start_blk++);
1492
1493         for (i = 1; i < 1 + cp_payload_blks; i++)
1494                 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1495                                                         start_blk++);
1496
1497         if (orphan_num) {
1498                 write_orphan_inodes(sbi, start_blk);
1499                 start_blk += orphan_blocks;
1500         }
1501
1502         f2fs_write_data_summaries(sbi, start_blk);
1503         start_blk += data_sum_blocks;
1504
1505         /* Record write statistics in the hot node summary */
1506         kbytes_written = sbi->kbytes_written;
1507         if (sb->s_bdev->bd_part)
1508                 kbytes_written += BD_PART_WRITTEN(sbi);
1509
1510         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1511
1512         if (__remain_node_summaries(cpc->reason)) {
1513                 f2fs_write_node_summaries(sbi, start_blk);
1514                 start_blk += NR_CURSEG_NODE_TYPE;
1515         }
1516
1517         /* update user_block_counts */
1518         sbi->last_valid_block_count = sbi->total_valid_block_count;
1519         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1520
1521         /* Here, we have one bio having CP pack except cp pack 2 page */
1522         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1523         /* Wait for all dirty meta pages to be submitted for IO */
1524         f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1525
1526         /* wait for previous submitted meta pages writeback */
1527         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1528
1529         /* flush all device cache */
1530         err = f2fs_flush_device_cache(sbi);
1531         if (err)
1532                 return err;
1533
1534         /* barrier and flush checkpoint cp pack 2 page if it can */
1535         commit_checkpoint(sbi, ckpt, start_blk);
1536         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1537
1538         /*
1539          * invalidate intermediate page cache borrowed from meta inode
1540          * which are used for migration of encrypted inode's blocks.
1541          */
1542         if (f2fs_sb_has_encrypt(sbi))
1543                 invalidate_mapping_pages(META_MAPPING(sbi),
1544                                 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1545
1546         f2fs_release_ino_entry(sbi, false);
1547
1548         f2fs_reset_fsync_node_info(sbi);
1549
1550         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1551         clear_sbi_flag(sbi, SBI_NEED_CP);
1552         clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1553
1554         spin_lock(&sbi->stat_lock);
1555         sbi->unusable_block_count = 0;
1556         spin_unlock(&sbi->stat_lock);
1557
1558         __set_cp_next_pack(sbi);
1559
1560         /*
1561          * redirty superblock if metadata like node page or inode cache is
1562          * updated during writing checkpoint.
1563          */
1564         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1565                         get_pages(sbi, F2FS_DIRTY_IMETA))
1566                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1567
1568         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1569
1570         return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1571 }
1572
1573 /*
1574  * We guarantee that this checkpoint procedure will not fail.
1575  */
1576 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1577 {
1578         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1579         unsigned long long ckpt_ver;
1580         int err = 0;
1581
1582         if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1583                 return -EROFS;
1584
1585         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1586                 if (cpc->reason != CP_PAUSE)
1587                         return 0;
1588                 f2fs_warn(sbi, "Start checkpoint disabled!");
1589         }
1590         mutex_lock(&sbi->cp_mutex);
1591
1592         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1593                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1594                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1595                 goto out;
1596         if (unlikely(f2fs_cp_error(sbi))) {
1597                 err = -EIO;
1598                 goto out;
1599         }
1600
1601         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1602
1603         err = block_operations(sbi);
1604         if (err)
1605                 goto out;
1606
1607         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1608
1609         f2fs_flush_merged_writes(sbi);
1610
1611         /* this is the case of multiple fstrims without any changes */
1612         if (cpc->reason & CP_DISCARD) {
1613                 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1614                         unblock_operations(sbi);
1615                         goto out;
1616                 }
1617
1618                 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1619                                 SIT_I(sbi)->dirty_sentries == 0 &&
1620                                 prefree_segments(sbi) == 0) {
1621                         f2fs_flush_sit_entries(sbi, cpc);
1622                         f2fs_clear_prefree_segments(sbi, cpc);
1623                         unblock_operations(sbi);
1624                         goto out;
1625                 }
1626         }
1627
1628         /*
1629          * update checkpoint pack index
1630          * Increase the version number so that
1631          * SIT entries and seg summaries are written at correct place
1632          */
1633         ckpt_ver = cur_cp_version(ckpt);
1634         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1635
1636         /* write cached NAT/SIT entries to NAT/SIT area */
1637         err = f2fs_flush_nat_entries(sbi, cpc);
1638         if (err)
1639                 goto stop;
1640
1641         f2fs_flush_sit_entries(sbi, cpc);
1642
1643         /* unlock all the fs_lock[] in do_checkpoint() */
1644         err = do_checkpoint(sbi, cpc);
1645         if (err)
1646                 f2fs_release_discard_addrs(sbi);
1647         else
1648                 f2fs_clear_prefree_segments(sbi, cpc);
1649 stop:
1650         unblock_operations(sbi);
1651         stat_inc_cp_count(sbi->stat_info);
1652
1653         if (cpc->reason & CP_RECOVERY)
1654                 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1655
1656         /* do checkpoint periodically */
1657         f2fs_update_time(sbi, CP_TIME);
1658         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1659 out:
1660         mutex_unlock(&sbi->cp_mutex);
1661         return err;
1662 }
1663
1664 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1665 {
1666         int i;
1667
1668         for (i = 0; i < MAX_INO_ENTRY; i++) {
1669                 struct inode_management *im = &sbi->im[i];
1670
1671                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1672                 spin_lock_init(&im->ino_lock);
1673                 INIT_LIST_HEAD(&im->ino_list);
1674                 im->ino_num = 0;
1675         }
1676
1677         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1678                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1679                                 F2FS_ORPHANS_PER_BLOCK;
1680 }
1681
1682 int __init f2fs_create_checkpoint_caches(void)
1683 {
1684         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1685                         sizeof(struct ino_entry));
1686         if (!ino_entry_slab)
1687                 return -ENOMEM;
1688         f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1689                         sizeof(struct inode_entry));
1690         if (!f2fs_inode_entry_slab) {
1691                 kmem_cache_destroy(ino_entry_slab);
1692                 return -ENOMEM;
1693         }
1694         return 0;
1695 }
1696
1697 void f2fs_destroy_checkpoint_caches(void)
1698 {
1699         kmem_cache_destroy(ino_entry_slab);
1700         kmem_cache_destroy(f2fs_inode_entry_slab);
1701 }