1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * fs/f2fs/checkpoint.c 4 * 5 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 6 * http://www.samsung.com/ 7 */ 8 #include <linux/fs.h> 9 #include <linux/bio.h> 10 #include <linux/mpage.h> 11 #include <linux/writeback.h> 12 #include <linux/blkdev.h> 13 #include <linux/f2fs_fs.h> 14 #include <linux/pagevec.h> 15 #include <linux/swap.h> 16 #include <linux/kthread.h> 17 18 #include "f2fs.h" 19 #include "node.h" 20 #include "segment.h" 21 #include "iostat.h" 22 #include <trace/events/f2fs.h> 23 24 #define DEFAULT_CHECKPOINT_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 25 26 static struct kmem_cache *ino_entry_slab; 27 struct kmem_cache *f2fs_inode_entry_slab; 28 29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io, 30 unsigned char reason) 31 { 32 f2fs_build_fault_attr(sbi, 0, 0); 33 set_ckpt_flags(sbi, CP_ERROR_FLAG); 34 if (!end_io) { 35 f2fs_flush_merged_writes(sbi); 36 37 f2fs_handle_stop(sbi, reason); 38 } 39 } 40 41 /* 42 * We guarantee no failure on the returned page. 43 */ 44 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index) 45 { 46 struct address_space *mapping = META_MAPPING(sbi); 47 struct page *page; 48 repeat: 49 page = f2fs_grab_cache_page(mapping, index, false); 50 if (!page) { 51 cond_resched(); 52 goto repeat; 53 } 54 f2fs_wait_on_page_writeback(page, META, true, true); 55 if (!PageUptodate(page)) 56 SetPageUptodate(page); 57 return page; 58 } 59 60 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index, 61 bool is_meta) 62 { 63 struct address_space *mapping = META_MAPPING(sbi); 64 struct page *page; 65 struct f2fs_io_info fio = { 66 .sbi = sbi, 67 .type = META, 68 .op = REQ_OP_READ, 69 .op_flags = REQ_META | REQ_PRIO, 70 .old_blkaddr = index, 71 .new_blkaddr = index, 72 .encrypted_page = NULL, 73 .is_por = !is_meta, 74 }; 75 int err; 76 77 if (unlikely(!is_meta)) 78 fio.op_flags &= ~REQ_META; 79 repeat: 80 page = f2fs_grab_cache_page(mapping, index, false); 81 if (!page) { 82 cond_resched(); 83 goto repeat; 84 } 85 if (PageUptodate(page)) 86 goto out; 87 88 fio.page = page; 89 90 err = f2fs_submit_page_bio(&fio); 91 if (err) { 92 f2fs_put_page(page, 1); 93 return ERR_PTR(err); 94 } 95 96 f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, F2FS_BLKSIZE); 97 98 lock_page(page); 99 if (unlikely(page->mapping != mapping)) { 100 f2fs_put_page(page, 1); 101 goto repeat; 102 } 103 104 if (unlikely(!PageUptodate(page))) { 105 f2fs_handle_page_eio(sbi, page->index, META); 106 f2fs_put_page(page, 1); 107 return ERR_PTR(-EIO); 108 } 109 out: 110 return page; 111 } 112 113 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index) 114 { 115 return __get_meta_page(sbi, index, true); 116 } 117 118 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index) 119 { 120 struct page *page; 121 int count = 0; 122 123 retry: 124 page = __get_meta_page(sbi, index, true); 125 if (IS_ERR(page)) { 126 if (PTR_ERR(page) == -EIO && 127 ++count <= DEFAULT_RETRY_IO_COUNT) 128 goto retry; 129 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_META_PAGE); 130 } 131 return page; 132 } 133 134 /* for POR only */ 135 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index) 136 { 137 return __get_meta_page(sbi, index, false); 138 } 139 140 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr, 141 int type) 142 { 143 struct seg_entry *se; 144 unsigned int segno, offset; 145 bool exist; 146 147 if (type == DATA_GENERIC) 148 return true; 149 150 segno = GET_SEGNO(sbi, blkaddr); 151 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 152 se = get_seg_entry(sbi, segno); 153 154 exist = f2fs_test_bit(offset, se->cur_valid_map); 155 if (exist && type == DATA_GENERIC_ENHANCE_UPDATE) { 156 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d", 157 blkaddr, exist); 158 set_sbi_flag(sbi, SBI_NEED_FSCK); 159 return exist; 160 } 161 162 if (!exist && type == DATA_GENERIC_ENHANCE) { 163 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d", 164 blkaddr, exist); 165 set_sbi_flag(sbi, SBI_NEED_FSCK); 166 dump_stack(); 167 } 168 return exist; 169 } 170 171 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi, 172 block_t blkaddr, int type) 173 { 174 if (time_to_inject(sbi, FAULT_BLKADDR)) { 175 f2fs_show_injection_info(sbi, FAULT_BLKADDR); 176 return false; 177 } 178 179 switch (type) { 180 case META_NAT: 181 break; 182 case META_SIT: 183 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi))) 184 return false; 185 break; 186 case META_SSA: 187 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) || 188 blkaddr < SM_I(sbi)->ssa_blkaddr)) 189 return false; 190 break; 191 case META_CP: 192 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr || 193 blkaddr < __start_cp_addr(sbi))) 194 return false; 195 break; 196 case META_POR: 197 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) || 198 blkaddr < MAIN_BLKADDR(sbi))) 199 return false; 200 break; 201 case DATA_GENERIC: 202 case DATA_GENERIC_ENHANCE: 203 case DATA_GENERIC_ENHANCE_READ: 204 case DATA_GENERIC_ENHANCE_UPDATE: 205 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) || 206 blkaddr < MAIN_BLKADDR(sbi))) { 207 f2fs_warn(sbi, "access invalid blkaddr:%u", 208 blkaddr); 209 set_sbi_flag(sbi, SBI_NEED_FSCK); 210 dump_stack(); 211 return false; 212 } else { 213 return __is_bitmap_valid(sbi, blkaddr, type); 214 } 215 break; 216 case META_GENERIC: 217 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) || 218 blkaddr >= MAIN_BLKADDR(sbi))) 219 return false; 220 break; 221 default: 222 BUG(); 223 } 224 225 return true; 226 } 227 228 /* 229 * Readahead CP/NAT/SIT/SSA/POR pages 230 */ 231 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages, 232 int type, bool sync) 233 { 234 struct page *page; 235 block_t blkno = start; 236 struct f2fs_io_info fio = { 237 .sbi = sbi, 238 .type = META, 239 .op = REQ_OP_READ, 240 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD, 241 .encrypted_page = NULL, 242 .in_list = false, 243 .is_por = (type == META_POR), 244 }; 245 struct blk_plug plug; 246 int err; 247 248 if (unlikely(type == META_POR)) 249 fio.op_flags &= ~REQ_META; 250 251 blk_start_plug(&plug); 252 for (; nrpages-- > 0; blkno++) { 253 254 if (!f2fs_is_valid_blkaddr(sbi, blkno, type)) 255 goto out; 256 257 switch (type) { 258 case META_NAT: 259 if (unlikely(blkno >= 260 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid))) 261 blkno = 0; 262 /* get nat block addr */ 263 fio.new_blkaddr = current_nat_addr(sbi, 264 blkno * NAT_ENTRY_PER_BLOCK); 265 break; 266 case META_SIT: 267 if (unlikely(blkno >= TOTAL_SEGS(sbi))) 268 goto out; 269 /* get sit block addr */ 270 fio.new_blkaddr = current_sit_addr(sbi, 271 blkno * SIT_ENTRY_PER_BLOCK); 272 break; 273 case META_SSA: 274 case META_CP: 275 case META_POR: 276 fio.new_blkaddr = blkno; 277 break; 278 default: 279 BUG(); 280 } 281 282 page = f2fs_grab_cache_page(META_MAPPING(sbi), 283 fio.new_blkaddr, false); 284 if (!page) 285 continue; 286 if (PageUptodate(page)) { 287 f2fs_put_page(page, 1); 288 continue; 289 } 290 291 fio.page = page; 292 err = f2fs_submit_page_bio(&fio); 293 f2fs_put_page(page, err ? 1 : 0); 294 295 if (!err) 296 f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, 297 F2FS_BLKSIZE); 298 } 299 out: 300 blk_finish_plug(&plug); 301 return blkno - start; 302 } 303 304 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index, 305 unsigned int ra_blocks) 306 { 307 struct page *page; 308 bool readahead = false; 309 310 if (ra_blocks == RECOVERY_MIN_RA_BLOCKS) 311 return; 312 313 page = find_get_page(META_MAPPING(sbi), index); 314 if (!page || !PageUptodate(page)) 315 readahead = true; 316 f2fs_put_page(page, 0); 317 318 if (readahead) 319 f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true); 320 } 321 322 static int __f2fs_write_meta_page(struct page *page, 323 struct writeback_control *wbc, 324 enum iostat_type io_type) 325 { 326 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 327 328 trace_f2fs_writepage(page, META); 329 330 if (unlikely(f2fs_cp_error(sbi))) 331 goto redirty_out; 332 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 333 goto redirty_out; 334 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0)) 335 goto redirty_out; 336 337 f2fs_do_write_meta_page(sbi, page, io_type); 338 dec_page_count(sbi, F2FS_DIRTY_META); 339 340 if (wbc->for_reclaim) 341 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META); 342 343 unlock_page(page); 344 345 if (unlikely(f2fs_cp_error(sbi))) 346 f2fs_submit_merged_write(sbi, META); 347 348 return 0; 349 350 redirty_out: 351 redirty_page_for_writepage(wbc, page); 352 return AOP_WRITEPAGE_ACTIVATE; 353 } 354 355 static int f2fs_write_meta_page(struct page *page, 356 struct writeback_control *wbc) 357 { 358 return __f2fs_write_meta_page(page, wbc, FS_META_IO); 359 } 360 361 static int f2fs_write_meta_pages(struct address_space *mapping, 362 struct writeback_control *wbc) 363 { 364 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 365 long diff, written; 366 367 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 368 goto skip_write; 369 370 /* collect a number of dirty meta pages and write together */ 371 if (wbc->sync_mode != WB_SYNC_ALL && 372 get_pages(sbi, F2FS_DIRTY_META) < 373 nr_pages_to_skip(sbi, META)) 374 goto skip_write; 375 376 /* if locked failed, cp will flush dirty pages instead */ 377 if (!f2fs_down_write_trylock(&sbi->cp_global_sem)) 378 goto skip_write; 379 380 trace_f2fs_writepages(mapping->host, wbc, META); 381 diff = nr_pages_to_write(sbi, META, wbc); 382 written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO); 383 f2fs_up_write(&sbi->cp_global_sem); 384 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff); 385 return 0; 386 387 skip_write: 388 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META); 389 trace_f2fs_writepages(mapping->host, wbc, META); 390 return 0; 391 } 392 393 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type, 394 long nr_to_write, enum iostat_type io_type) 395 { 396 struct address_space *mapping = META_MAPPING(sbi); 397 pgoff_t index = 0, prev = ULONG_MAX; 398 struct folio_batch fbatch; 399 long nwritten = 0; 400 int nr_folios; 401 struct writeback_control wbc = { 402 .for_reclaim = 0, 403 }; 404 struct blk_plug plug; 405 406 folio_batch_init(&fbatch); 407 408 blk_start_plug(&plug); 409 410 while ((nr_folios = filemap_get_folios_tag(mapping, &index, 411 (pgoff_t)-1, 412 PAGECACHE_TAG_DIRTY, &fbatch))) { 413 int i; 414 415 for (i = 0; i < nr_folios; i++) { 416 struct folio *folio = fbatch.folios[i]; 417 418 if (nr_to_write != LONG_MAX && i != 0 && 419 folio->index != prev + 420 folio_nr_pages(fbatch.folios[i-1])) { 421 folio_batch_release(&fbatch); 422 goto stop; 423 } 424 425 folio_lock(folio); 426 427 if (unlikely(folio->mapping != mapping)) { 428 continue_unlock: 429 folio_unlock(folio); 430 continue; 431 } 432 if (!folio_test_dirty(folio)) { 433 /* someone wrote it for us */ 434 goto continue_unlock; 435 } 436 437 f2fs_wait_on_page_writeback(&folio->page, META, 438 true, true); 439 440 if (!folio_clear_dirty_for_io(folio)) 441 goto continue_unlock; 442 443 if (__f2fs_write_meta_page(&folio->page, &wbc, 444 io_type)) { 445 folio_unlock(folio); 446 break; 447 } 448 nwritten += folio_nr_pages(folio); 449 prev = folio->index; 450 if (unlikely(nwritten >= nr_to_write)) 451 break; 452 } 453 folio_batch_release(&fbatch); 454 cond_resched(); 455 } 456 stop: 457 if (nwritten) 458 f2fs_submit_merged_write(sbi, type); 459 460 blk_finish_plug(&plug); 461 462 return nwritten; 463 } 464 465 static bool f2fs_dirty_meta_folio(struct address_space *mapping, 466 struct folio *folio) 467 { 468 trace_f2fs_set_page_dirty(&folio->page, META); 469 470 if (!folio_test_uptodate(folio)) 471 folio_mark_uptodate(folio); 472 if (filemap_dirty_folio(mapping, folio)) { 473 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META); 474 set_page_private_reference(&folio->page); 475 return true; 476 } 477 return false; 478 } 479 480 const struct address_space_operations f2fs_meta_aops = { 481 .writepage = f2fs_write_meta_page, 482 .writepages = f2fs_write_meta_pages, 483 .dirty_folio = f2fs_dirty_meta_folio, 484 .invalidate_folio = f2fs_invalidate_folio, 485 .release_folio = f2fs_release_folio, 486 .migrate_folio = filemap_migrate_folio, 487 }; 488 489 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, 490 unsigned int devidx, int type) 491 { 492 struct inode_management *im = &sbi->im[type]; 493 struct ino_entry *e = NULL, *new = NULL; 494 495 if (type == FLUSH_INO) { 496 rcu_read_lock(); 497 e = radix_tree_lookup(&im->ino_root, ino); 498 rcu_read_unlock(); 499 } 500 501 retry: 502 if (!e) 503 new = f2fs_kmem_cache_alloc(ino_entry_slab, 504 GFP_NOFS, true, NULL); 505 506 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL); 507 508 spin_lock(&im->ino_lock); 509 e = radix_tree_lookup(&im->ino_root, ino); 510 if (!e) { 511 if (!new) { 512 spin_unlock(&im->ino_lock); 513 goto retry; 514 } 515 e = new; 516 if (unlikely(radix_tree_insert(&im->ino_root, ino, e))) 517 f2fs_bug_on(sbi, 1); 518 519 memset(e, 0, sizeof(struct ino_entry)); 520 e->ino = ino; 521 522 list_add_tail(&e->list, &im->ino_list); 523 if (type != ORPHAN_INO) 524 im->ino_num++; 525 } 526 527 if (type == FLUSH_INO) 528 f2fs_set_bit(devidx, (char *)&e->dirty_device); 529 530 spin_unlock(&im->ino_lock); 531 radix_tree_preload_end(); 532 533 if (new && e != new) 534 kmem_cache_free(ino_entry_slab, new); 535 } 536 537 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) 538 { 539 struct inode_management *im = &sbi->im[type]; 540 struct ino_entry *e; 541 542 spin_lock(&im->ino_lock); 543 e = radix_tree_lookup(&im->ino_root, ino); 544 if (e) { 545 list_del(&e->list); 546 radix_tree_delete(&im->ino_root, ino); 547 im->ino_num--; 548 spin_unlock(&im->ino_lock); 549 kmem_cache_free(ino_entry_slab, e); 550 return; 551 } 552 spin_unlock(&im->ino_lock); 553 } 554 555 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) 556 { 557 /* add new dirty ino entry into list */ 558 __add_ino_entry(sbi, ino, 0, type); 559 } 560 561 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) 562 { 563 /* remove dirty ino entry from list */ 564 __remove_ino_entry(sbi, ino, type); 565 } 566 567 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */ 568 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode) 569 { 570 struct inode_management *im = &sbi->im[mode]; 571 struct ino_entry *e; 572 573 spin_lock(&im->ino_lock); 574 e = radix_tree_lookup(&im->ino_root, ino); 575 spin_unlock(&im->ino_lock); 576 return e ? true : false; 577 } 578 579 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all) 580 { 581 struct ino_entry *e, *tmp; 582 int i; 583 584 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) { 585 struct inode_management *im = &sbi->im[i]; 586 587 spin_lock(&im->ino_lock); 588 list_for_each_entry_safe(e, tmp, &im->ino_list, list) { 589 list_del(&e->list); 590 radix_tree_delete(&im->ino_root, e->ino); 591 kmem_cache_free(ino_entry_slab, e); 592 im->ino_num--; 593 } 594 spin_unlock(&im->ino_lock); 595 } 596 } 597 598 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino, 599 unsigned int devidx, int type) 600 { 601 __add_ino_entry(sbi, ino, devidx, type); 602 } 603 604 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino, 605 unsigned int devidx, int type) 606 { 607 struct inode_management *im = &sbi->im[type]; 608 struct ino_entry *e; 609 bool is_dirty = false; 610 611 spin_lock(&im->ino_lock); 612 e = radix_tree_lookup(&im->ino_root, ino); 613 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device)) 614 is_dirty = true; 615 spin_unlock(&im->ino_lock); 616 return is_dirty; 617 } 618 619 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi) 620 { 621 struct inode_management *im = &sbi->im[ORPHAN_INO]; 622 int err = 0; 623 624 spin_lock(&im->ino_lock); 625 626 if (time_to_inject(sbi, FAULT_ORPHAN)) { 627 spin_unlock(&im->ino_lock); 628 f2fs_show_injection_info(sbi, FAULT_ORPHAN); 629 return -ENOSPC; 630 } 631 632 if (unlikely(im->ino_num >= sbi->max_orphans)) 633 err = -ENOSPC; 634 else 635 im->ino_num++; 636 spin_unlock(&im->ino_lock); 637 638 return err; 639 } 640 641 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi) 642 { 643 struct inode_management *im = &sbi->im[ORPHAN_INO]; 644 645 spin_lock(&im->ino_lock); 646 f2fs_bug_on(sbi, im->ino_num == 0); 647 im->ino_num--; 648 spin_unlock(&im->ino_lock); 649 } 650 651 void f2fs_add_orphan_inode(struct inode *inode) 652 { 653 /* add new orphan ino entry into list */ 654 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO); 655 f2fs_update_inode_page(inode); 656 } 657 658 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) 659 { 660 /* remove orphan entry from orphan list */ 661 __remove_ino_entry(sbi, ino, ORPHAN_INO); 662 } 663 664 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) 665 { 666 struct inode *inode; 667 struct node_info ni; 668 int err; 669 670 inode = f2fs_iget_retry(sbi->sb, ino); 671 if (IS_ERR(inode)) { 672 /* 673 * there should be a bug that we can't find the entry 674 * to orphan inode. 675 */ 676 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT); 677 return PTR_ERR(inode); 678 } 679 680 err = f2fs_dquot_initialize(inode); 681 if (err) { 682 iput(inode); 683 goto err_out; 684 } 685 686 clear_nlink(inode); 687 688 /* truncate all the data during iput */ 689 iput(inode); 690 691 err = f2fs_get_node_info(sbi, ino, &ni, false); 692 if (err) 693 goto err_out; 694 695 /* ENOMEM was fully retried in f2fs_evict_inode. */ 696 if (ni.blk_addr != NULL_ADDR) { 697 err = -EIO; 698 goto err_out; 699 } 700 return 0; 701 702 err_out: 703 set_sbi_flag(sbi, SBI_NEED_FSCK); 704 f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.", 705 __func__, ino); 706 return err; 707 } 708 709 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi) 710 { 711 block_t start_blk, orphan_blocks, i, j; 712 unsigned int s_flags = sbi->sb->s_flags; 713 int err = 0; 714 #ifdef CONFIG_QUOTA 715 int quota_enabled; 716 #endif 717 718 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG)) 719 return 0; 720 721 if (bdev_read_only(sbi->sb->s_bdev)) { 722 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup"); 723 return 0; 724 } 725 726 if (s_flags & SB_RDONLY) { 727 f2fs_info(sbi, "orphan cleanup on readonly fs"); 728 sbi->sb->s_flags &= ~SB_RDONLY; 729 } 730 731 #ifdef CONFIG_QUOTA 732 /* 733 * Turn on quotas which were not enabled for read-only mounts if 734 * filesystem has quota feature, so that they are updated correctly. 735 */ 736 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY); 737 #endif 738 739 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi); 740 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi); 741 742 f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true); 743 744 for (i = 0; i < orphan_blocks; i++) { 745 struct page *page; 746 struct f2fs_orphan_block *orphan_blk; 747 748 page = f2fs_get_meta_page(sbi, start_blk + i); 749 if (IS_ERR(page)) { 750 err = PTR_ERR(page); 751 goto out; 752 } 753 754 orphan_blk = (struct f2fs_orphan_block *)page_address(page); 755 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) { 756 nid_t ino = le32_to_cpu(orphan_blk->ino[j]); 757 758 err = recover_orphan_inode(sbi, ino); 759 if (err) { 760 f2fs_put_page(page, 1); 761 goto out; 762 } 763 } 764 f2fs_put_page(page, 1); 765 } 766 /* clear Orphan Flag */ 767 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG); 768 out: 769 set_sbi_flag(sbi, SBI_IS_RECOVERED); 770 771 #ifdef CONFIG_QUOTA 772 /* Turn quotas off */ 773 if (quota_enabled) 774 f2fs_quota_off_umount(sbi->sb); 775 #endif 776 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */ 777 778 return err; 779 } 780 781 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk) 782 { 783 struct list_head *head; 784 struct f2fs_orphan_block *orphan_blk = NULL; 785 unsigned int nentries = 0; 786 unsigned short index = 1; 787 unsigned short orphan_blocks; 788 struct page *page = NULL; 789 struct ino_entry *orphan = NULL; 790 struct inode_management *im = &sbi->im[ORPHAN_INO]; 791 792 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num); 793 794 /* 795 * we don't need to do spin_lock(&im->ino_lock) here, since all the 796 * orphan inode operations are covered under f2fs_lock_op(). 797 * And, spin_lock should be avoided due to page operations below. 798 */ 799 head = &im->ino_list; 800 801 /* loop for each orphan inode entry and write them in Jornal block */ 802 list_for_each_entry(orphan, head, list) { 803 if (!page) { 804 page = f2fs_grab_meta_page(sbi, start_blk++); 805 orphan_blk = 806 (struct f2fs_orphan_block *)page_address(page); 807 memset(orphan_blk, 0, sizeof(*orphan_blk)); 808 } 809 810 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino); 811 812 if (nentries == F2FS_ORPHANS_PER_BLOCK) { 813 /* 814 * an orphan block is full of 1020 entries, 815 * then we need to flush current orphan blocks 816 * and bring another one in memory 817 */ 818 orphan_blk->blk_addr = cpu_to_le16(index); 819 orphan_blk->blk_count = cpu_to_le16(orphan_blocks); 820 orphan_blk->entry_count = cpu_to_le32(nentries); 821 set_page_dirty(page); 822 f2fs_put_page(page, 1); 823 index++; 824 nentries = 0; 825 page = NULL; 826 } 827 } 828 829 if (page) { 830 orphan_blk->blk_addr = cpu_to_le16(index); 831 orphan_blk->blk_count = cpu_to_le16(orphan_blocks); 832 orphan_blk->entry_count = cpu_to_le32(nentries); 833 set_page_dirty(page); 834 f2fs_put_page(page, 1); 835 } 836 } 837 838 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi, 839 struct f2fs_checkpoint *ckpt) 840 { 841 unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset); 842 __u32 chksum; 843 844 chksum = f2fs_crc32(sbi, ckpt, chksum_ofs); 845 if (chksum_ofs < CP_CHKSUM_OFFSET) { 846 chksum_ofs += sizeof(chksum); 847 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs, 848 F2FS_BLKSIZE - chksum_ofs); 849 } 850 return chksum; 851 } 852 853 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr, 854 struct f2fs_checkpoint **cp_block, struct page **cp_page, 855 unsigned long long *version) 856 { 857 size_t crc_offset = 0; 858 __u32 crc; 859 860 *cp_page = f2fs_get_meta_page(sbi, cp_addr); 861 if (IS_ERR(*cp_page)) 862 return PTR_ERR(*cp_page); 863 864 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page); 865 866 crc_offset = le32_to_cpu((*cp_block)->checksum_offset); 867 if (crc_offset < CP_MIN_CHKSUM_OFFSET || 868 crc_offset > CP_CHKSUM_OFFSET) { 869 f2fs_put_page(*cp_page, 1); 870 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset); 871 return -EINVAL; 872 } 873 874 crc = f2fs_checkpoint_chksum(sbi, *cp_block); 875 if (crc != cur_cp_crc(*cp_block)) { 876 f2fs_put_page(*cp_page, 1); 877 f2fs_warn(sbi, "invalid crc value"); 878 return -EINVAL; 879 } 880 881 *version = cur_cp_version(*cp_block); 882 return 0; 883 } 884 885 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi, 886 block_t cp_addr, unsigned long long *version) 887 { 888 struct page *cp_page_1 = NULL, *cp_page_2 = NULL; 889 struct f2fs_checkpoint *cp_block = NULL; 890 unsigned long long cur_version = 0, pre_version = 0; 891 unsigned int cp_blocks; 892 int err; 893 894 err = get_checkpoint_version(sbi, cp_addr, &cp_block, 895 &cp_page_1, version); 896 if (err) 897 return NULL; 898 899 cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count); 900 901 if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) { 902 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u", 903 le32_to_cpu(cp_block->cp_pack_total_block_count)); 904 goto invalid_cp; 905 } 906 pre_version = *version; 907 908 cp_addr += cp_blocks - 1; 909 err = get_checkpoint_version(sbi, cp_addr, &cp_block, 910 &cp_page_2, version); 911 if (err) 912 goto invalid_cp; 913 cur_version = *version; 914 915 if (cur_version == pre_version) { 916 *version = cur_version; 917 f2fs_put_page(cp_page_2, 1); 918 return cp_page_1; 919 } 920 f2fs_put_page(cp_page_2, 1); 921 invalid_cp: 922 f2fs_put_page(cp_page_1, 1); 923 return NULL; 924 } 925 926 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi) 927 { 928 struct f2fs_checkpoint *cp_block; 929 struct f2fs_super_block *fsb = sbi->raw_super; 930 struct page *cp1, *cp2, *cur_page; 931 unsigned long blk_size = sbi->blocksize; 932 unsigned long long cp1_version = 0, cp2_version = 0; 933 unsigned long long cp_start_blk_no; 934 unsigned int cp_blks = 1 + __cp_payload(sbi); 935 block_t cp_blk_no; 936 int i; 937 int err; 938 939 sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks), 940 GFP_KERNEL); 941 if (!sbi->ckpt) 942 return -ENOMEM; 943 /* 944 * Finding out valid cp block involves read both 945 * sets( cp pack 1 and cp pack 2) 946 */ 947 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr); 948 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version); 949 950 /* The second checkpoint pack should start at the next segment */ 951 cp_start_blk_no += ((unsigned long long)1) << 952 le32_to_cpu(fsb->log_blocks_per_seg); 953 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version); 954 955 if (cp1 && cp2) { 956 if (ver_after(cp2_version, cp1_version)) 957 cur_page = cp2; 958 else 959 cur_page = cp1; 960 } else if (cp1) { 961 cur_page = cp1; 962 } else if (cp2) { 963 cur_page = cp2; 964 } else { 965 err = -EFSCORRUPTED; 966 goto fail_no_cp; 967 } 968 969 cp_block = (struct f2fs_checkpoint *)page_address(cur_page); 970 memcpy(sbi->ckpt, cp_block, blk_size); 971 972 if (cur_page == cp1) 973 sbi->cur_cp_pack = 1; 974 else 975 sbi->cur_cp_pack = 2; 976 977 /* Sanity checking of checkpoint */ 978 if (f2fs_sanity_check_ckpt(sbi)) { 979 err = -EFSCORRUPTED; 980 goto free_fail_no_cp; 981 } 982 983 if (cp_blks <= 1) 984 goto done; 985 986 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr); 987 if (cur_page == cp2) 988 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg); 989 990 for (i = 1; i < cp_blks; i++) { 991 void *sit_bitmap_ptr; 992 unsigned char *ckpt = (unsigned char *)sbi->ckpt; 993 994 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i); 995 if (IS_ERR(cur_page)) { 996 err = PTR_ERR(cur_page); 997 goto free_fail_no_cp; 998 } 999 sit_bitmap_ptr = page_address(cur_page); 1000 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size); 1001 f2fs_put_page(cur_page, 1); 1002 } 1003 done: 1004 f2fs_put_page(cp1, 1); 1005 f2fs_put_page(cp2, 1); 1006 return 0; 1007 1008 free_fail_no_cp: 1009 f2fs_put_page(cp1, 1); 1010 f2fs_put_page(cp2, 1); 1011 fail_no_cp: 1012 kvfree(sbi->ckpt); 1013 return err; 1014 } 1015 1016 static void __add_dirty_inode(struct inode *inode, enum inode_type type) 1017 { 1018 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1019 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE; 1020 1021 if (is_inode_flag_set(inode, flag)) 1022 return; 1023 1024 set_inode_flag(inode, flag); 1025 list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]); 1026 stat_inc_dirty_inode(sbi, type); 1027 } 1028 1029 static void __remove_dirty_inode(struct inode *inode, enum inode_type type) 1030 { 1031 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE; 1032 1033 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag)) 1034 return; 1035 1036 list_del_init(&F2FS_I(inode)->dirty_list); 1037 clear_inode_flag(inode, flag); 1038 stat_dec_dirty_inode(F2FS_I_SB(inode), type); 1039 } 1040 1041 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio) 1042 { 1043 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1044 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE; 1045 1046 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) && 1047 !S_ISLNK(inode->i_mode)) 1048 return; 1049 1050 spin_lock(&sbi->inode_lock[type]); 1051 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH)) 1052 __add_dirty_inode(inode, type); 1053 inode_inc_dirty_pages(inode); 1054 spin_unlock(&sbi->inode_lock[type]); 1055 1056 set_page_private_reference(&folio->page); 1057 } 1058 1059 void f2fs_remove_dirty_inode(struct inode *inode) 1060 { 1061 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1062 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE; 1063 1064 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) && 1065 !S_ISLNK(inode->i_mode)) 1066 return; 1067 1068 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH)) 1069 return; 1070 1071 spin_lock(&sbi->inode_lock[type]); 1072 __remove_dirty_inode(inode, type); 1073 spin_unlock(&sbi->inode_lock[type]); 1074 } 1075 1076 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type, 1077 bool from_cp) 1078 { 1079 struct list_head *head; 1080 struct inode *inode; 1081 struct f2fs_inode_info *fi; 1082 bool is_dir = (type == DIR_INODE); 1083 unsigned long ino = 0; 1084 1085 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir, 1086 get_pages(sbi, is_dir ? 1087 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA)); 1088 retry: 1089 if (unlikely(f2fs_cp_error(sbi))) { 1090 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir, 1091 get_pages(sbi, is_dir ? 1092 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA)); 1093 return -EIO; 1094 } 1095 1096 spin_lock(&sbi->inode_lock[type]); 1097 1098 head = &sbi->inode_list[type]; 1099 if (list_empty(head)) { 1100 spin_unlock(&sbi->inode_lock[type]); 1101 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir, 1102 get_pages(sbi, is_dir ? 1103 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA)); 1104 return 0; 1105 } 1106 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list); 1107 inode = igrab(&fi->vfs_inode); 1108 spin_unlock(&sbi->inode_lock[type]); 1109 if (inode) { 1110 unsigned long cur_ino = inode->i_ino; 1111 1112 if (from_cp) 1113 F2FS_I(inode)->cp_task = current; 1114 F2FS_I(inode)->wb_task = current; 1115 1116 filemap_fdatawrite(inode->i_mapping); 1117 1118 F2FS_I(inode)->wb_task = NULL; 1119 if (from_cp) 1120 F2FS_I(inode)->cp_task = NULL; 1121 1122 iput(inode); 1123 /* We need to give cpu to another writers. */ 1124 if (ino == cur_ino) 1125 cond_resched(); 1126 else 1127 ino = cur_ino; 1128 } else { 1129 /* 1130 * We should submit bio, since it exists several 1131 * wribacking dentry pages in the freeing inode. 1132 */ 1133 f2fs_submit_merged_write(sbi, DATA); 1134 cond_resched(); 1135 } 1136 goto retry; 1137 } 1138 1139 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi) 1140 { 1141 struct list_head *head = &sbi->inode_list[DIRTY_META]; 1142 struct inode *inode; 1143 struct f2fs_inode_info *fi; 1144 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA); 1145 1146 while (total--) { 1147 if (unlikely(f2fs_cp_error(sbi))) 1148 return -EIO; 1149 1150 spin_lock(&sbi->inode_lock[DIRTY_META]); 1151 if (list_empty(head)) { 1152 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1153 return 0; 1154 } 1155 fi = list_first_entry(head, struct f2fs_inode_info, 1156 gdirty_list); 1157 inode = igrab(&fi->vfs_inode); 1158 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1159 if (inode) { 1160 sync_inode_metadata(inode, 0); 1161 1162 /* it's on eviction */ 1163 if (is_inode_flag_set(inode, FI_DIRTY_INODE)) 1164 f2fs_update_inode_page(inode); 1165 iput(inode); 1166 } 1167 } 1168 return 0; 1169 } 1170 1171 static void __prepare_cp_block(struct f2fs_sb_info *sbi) 1172 { 1173 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1174 struct f2fs_nm_info *nm_i = NM_I(sbi); 1175 nid_t last_nid = nm_i->next_scan_nid; 1176 1177 next_free_nid(sbi, &last_nid); 1178 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi)); 1179 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi)); 1180 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi)); 1181 ckpt->next_free_nid = cpu_to_le32(last_nid); 1182 } 1183 1184 static bool __need_flush_quota(struct f2fs_sb_info *sbi) 1185 { 1186 bool ret = false; 1187 1188 if (!is_journalled_quota(sbi)) 1189 return false; 1190 1191 if (!f2fs_down_write_trylock(&sbi->quota_sem)) 1192 return true; 1193 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) { 1194 ret = false; 1195 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) { 1196 ret = false; 1197 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) { 1198 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH); 1199 ret = true; 1200 } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) { 1201 ret = true; 1202 } 1203 f2fs_up_write(&sbi->quota_sem); 1204 return ret; 1205 } 1206 1207 /* 1208 * Freeze all the FS-operations for checkpoint. 1209 */ 1210 static int block_operations(struct f2fs_sb_info *sbi) 1211 { 1212 struct writeback_control wbc = { 1213 .sync_mode = WB_SYNC_ALL, 1214 .nr_to_write = LONG_MAX, 1215 .for_reclaim = 0, 1216 }; 1217 int err = 0, cnt = 0; 1218 1219 /* 1220 * Let's flush inline_data in dirty node pages. 1221 */ 1222 f2fs_flush_inline_data(sbi); 1223 1224 retry_flush_quotas: 1225 f2fs_lock_all(sbi); 1226 if (__need_flush_quota(sbi)) { 1227 int locked; 1228 1229 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) { 1230 set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH); 1231 set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH); 1232 goto retry_flush_dents; 1233 } 1234 f2fs_unlock_all(sbi); 1235 1236 /* only failed during mount/umount/freeze/quotactl */ 1237 locked = down_read_trylock(&sbi->sb->s_umount); 1238 f2fs_quota_sync(sbi->sb, -1); 1239 if (locked) 1240 up_read(&sbi->sb->s_umount); 1241 cond_resched(); 1242 goto retry_flush_quotas; 1243 } 1244 1245 retry_flush_dents: 1246 /* write all the dirty dentry pages */ 1247 if (get_pages(sbi, F2FS_DIRTY_DENTS)) { 1248 f2fs_unlock_all(sbi); 1249 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true); 1250 if (err) 1251 return err; 1252 cond_resched(); 1253 goto retry_flush_quotas; 1254 } 1255 1256 /* 1257 * POR: we should ensure that there are no dirty node pages 1258 * until finishing nat/sit flush. inode->i_blocks can be updated. 1259 */ 1260 f2fs_down_write(&sbi->node_change); 1261 1262 if (get_pages(sbi, F2FS_DIRTY_IMETA)) { 1263 f2fs_up_write(&sbi->node_change); 1264 f2fs_unlock_all(sbi); 1265 err = f2fs_sync_inode_meta(sbi); 1266 if (err) 1267 return err; 1268 cond_resched(); 1269 goto retry_flush_quotas; 1270 } 1271 1272 retry_flush_nodes: 1273 f2fs_down_write(&sbi->node_write); 1274 1275 if (get_pages(sbi, F2FS_DIRTY_NODES)) { 1276 f2fs_up_write(&sbi->node_write); 1277 atomic_inc(&sbi->wb_sync_req[NODE]); 1278 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO); 1279 atomic_dec(&sbi->wb_sync_req[NODE]); 1280 if (err) { 1281 f2fs_up_write(&sbi->node_change); 1282 f2fs_unlock_all(sbi); 1283 return err; 1284 } 1285 cond_resched(); 1286 goto retry_flush_nodes; 1287 } 1288 1289 /* 1290 * sbi->node_change is used only for AIO write_begin path which produces 1291 * dirty node blocks and some checkpoint values by block allocation. 1292 */ 1293 __prepare_cp_block(sbi); 1294 f2fs_up_write(&sbi->node_change); 1295 return err; 1296 } 1297 1298 static void unblock_operations(struct f2fs_sb_info *sbi) 1299 { 1300 f2fs_up_write(&sbi->node_write); 1301 f2fs_unlock_all(sbi); 1302 } 1303 1304 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type) 1305 { 1306 DEFINE_WAIT(wait); 1307 1308 for (;;) { 1309 if (!get_pages(sbi, type)) 1310 break; 1311 1312 if (unlikely(f2fs_cp_error(sbi))) 1313 break; 1314 1315 if (type == F2FS_DIRTY_META) 1316 f2fs_sync_meta_pages(sbi, META, LONG_MAX, 1317 FS_CP_META_IO); 1318 else if (type == F2FS_WB_CP_DATA) 1319 f2fs_submit_merged_write(sbi, DATA); 1320 1321 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE); 1322 io_schedule_timeout(DEFAULT_IO_TIMEOUT); 1323 } 1324 finish_wait(&sbi->cp_wait, &wait); 1325 } 1326 1327 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1328 { 1329 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num; 1330 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1331 unsigned long flags; 1332 1333 if (cpc->reason & CP_UMOUNT) { 1334 if (le32_to_cpu(ckpt->cp_pack_total_block_count) + 1335 NM_I(sbi)->nat_bits_blocks > sbi->blocks_per_seg) { 1336 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG); 1337 f2fs_notice(sbi, "Disable nat_bits due to no space"); 1338 } else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) && 1339 f2fs_nat_bitmap_enabled(sbi)) { 1340 f2fs_enable_nat_bits(sbi); 1341 set_ckpt_flags(sbi, CP_NAT_BITS_FLAG); 1342 f2fs_notice(sbi, "Rebuild and enable nat_bits"); 1343 } 1344 } 1345 1346 spin_lock_irqsave(&sbi->cp_lock, flags); 1347 1348 if (cpc->reason & CP_TRIMMED) 1349 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG); 1350 else 1351 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG); 1352 1353 if (cpc->reason & CP_UMOUNT) 1354 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG); 1355 else 1356 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG); 1357 1358 if (cpc->reason & CP_FASTBOOT) 1359 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG); 1360 else 1361 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG); 1362 1363 if (orphan_num) 1364 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); 1365 else 1366 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); 1367 1368 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1369 __set_ckpt_flags(ckpt, CP_FSCK_FLAG); 1370 1371 if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS)) 1372 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG); 1373 else 1374 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG); 1375 1376 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED)) 1377 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG); 1378 else 1379 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG); 1380 1381 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK)) 1382 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG); 1383 else 1384 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG); 1385 1386 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) 1387 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG); 1388 else 1389 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG); 1390 1391 if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) 1392 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG); 1393 1394 /* set this flag to activate crc|cp_ver for recovery */ 1395 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG); 1396 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG); 1397 1398 spin_unlock_irqrestore(&sbi->cp_lock, flags); 1399 } 1400 1401 static void commit_checkpoint(struct f2fs_sb_info *sbi, 1402 void *src, block_t blk_addr) 1403 { 1404 struct writeback_control wbc = { 1405 .for_reclaim = 0, 1406 }; 1407 1408 /* 1409 * filemap_get_folios_tag and lock_page again will take 1410 * some extra time. Therefore, f2fs_update_meta_pages and 1411 * f2fs_sync_meta_pages are combined in this function. 1412 */ 1413 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 1414 int err; 1415 1416 f2fs_wait_on_page_writeback(page, META, true, true); 1417 1418 memcpy(page_address(page), src, PAGE_SIZE); 1419 1420 set_page_dirty(page); 1421 if (unlikely(!clear_page_dirty_for_io(page))) 1422 f2fs_bug_on(sbi, 1); 1423 1424 /* writeout cp pack 2 page */ 1425 err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO); 1426 if (unlikely(err && f2fs_cp_error(sbi))) { 1427 f2fs_put_page(page, 1); 1428 return; 1429 } 1430 1431 f2fs_bug_on(sbi, err); 1432 f2fs_put_page(page, 0); 1433 1434 /* submit checkpoint (with barrier if NOBARRIER is not set) */ 1435 f2fs_submit_merged_write(sbi, META_FLUSH); 1436 } 1437 1438 static inline u64 get_sectors_written(struct block_device *bdev) 1439 { 1440 return (u64)part_stat_read(bdev, sectors[STAT_WRITE]); 1441 } 1442 1443 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi) 1444 { 1445 if (f2fs_is_multi_device(sbi)) { 1446 u64 sectors = 0; 1447 int i; 1448 1449 for (i = 0; i < sbi->s_ndevs; i++) 1450 sectors += get_sectors_written(FDEV(i).bdev); 1451 1452 return sectors; 1453 } 1454 1455 return get_sectors_written(sbi->sb->s_bdev); 1456 } 1457 1458 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1459 { 1460 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1461 struct f2fs_nm_info *nm_i = NM_I(sbi); 1462 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags; 1463 block_t start_blk; 1464 unsigned int data_sum_blocks, orphan_blocks; 1465 __u32 crc32 = 0; 1466 int i; 1467 int cp_payload_blks = __cp_payload(sbi); 1468 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE); 1469 u64 kbytes_written; 1470 int err; 1471 1472 /* Flush all the NAT/SIT pages */ 1473 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO); 1474 1475 /* start to update checkpoint, cp ver is already updated previously */ 1476 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true)); 1477 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi)); 1478 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) { 1479 ckpt->cur_node_segno[i] = 1480 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE)); 1481 ckpt->cur_node_blkoff[i] = 1482 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE)); 1483 ckpt->alloc_type[i + CURSEG_HOT_NODE] = 1484 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE); 1485 } 1486 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) { 1487 ckpt->cur_data_segno[i] = 1488 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA)); 1489 ckpt->cur_data_blkoff[i] = 1490 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA)); 1491 ckpt->alloc_type[i + CURSEG_HOT_DATA] = 1492 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA); 1493 } 1494 1495 /* 2 cp + n data seg summary + orphan inode blocks */ 1496 data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false); 1497 spin_lock_irqsave(&sbi->cp_lock, flags); 1498 if (data_sum_blocks < NR_CURSEG_DATA_TYPE) 1499 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); 1500 else 1501 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); 1502 spin_unlock_irqrestore(&sbi->cp_lock, flags); 1503 1504 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num); 1505 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks + 1506 orphan_blocks); 1507 1508 if (__remain_node_summaries(cpc->reason)) 1509 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS + 1510 cp_payload_blks + data_sum_blocks + 1511 orphan_blocks + NR_CURSEG_NODE_TYPE); 1512 else 1513 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS + 1514 cp_payload_blks + data_sum_blocks + 1515 orphan_blocks); 1516 1517 /* update ckpt flag for checkpoint */ 1518 update_ckpt_flags(sbi, cpc); 1519 1520 /* update SIT/NAT bitmap */ 1521 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP)); 1522 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP)); 1523 1524 crc32 = f2fs_checkpoint_chksum(sbi, ckpt); 1525 *((__le32 *)((unsigned char *)ckpt + 1526 le32_to_cpu(ckpt->checksum_offset))) 1527 = cpu_to_le32(crc32); 1528 1529 start_blk = __start_cp_next_addr(sbi); 1530 1531 /* write nat bits */ 1532 if ((cpc->reason & CP_UMOUNT) && 1533 is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) { 1534 __u64 cp_ver = cur_cp_version(ckpt); 1535 block_t blk; 1536 1537 cp_ver |= ((__u64)crc32 << 32); 1538 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver); 1539 1540 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks; 1541 for (i = 0; i < nm_i->nat_bits_blocks; i++) 1542 f2fs_update_meta_page(sbi, nm_i->nat_bits + 1543 (i << F2FS_BLKSIZE_BITS), blk + i); 1544 } 1545 1546 /* write out checkpoint buffer at block 0 */ 1547 f2fs_update_meta_page(sbi, ckpt, start_blk++); 1548 1549 for (i = 1; i < 1 + cp_payload_blks; i++) 1550 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE, 1551 start_blk++); 1552 1553 if (orphan_num) { 1554 write_orphan_inodes(sbi, start_blk); 1555 start_blk += orphan_blocks; 1556 } 1557 1558 f2fs_write_data_summaries(sbi, start_blk); 1559 start_blk += data_sum_blocks; 1560 1561 /* Record write statistics in the hot node summary */ 1562 kbytes_written = sbi->kbytes_written; 1563 kbytes_written += (f2fs_get_sectors_written(sbi) - 1564 sbi->sectors_written_start) >> 1; 1565 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written); 1566 1567 if (__remain_node_summaries(cpc->reason)) { 1568 f2fs_write_node_summaries(sbi, start_blk); 1569 start_blk += NR_CURSEG_NODE_TYPE; 1570 } 1571 1572 /* update user_block_counts */ 1573 sbi->last_valid_block_count = sbi->total_valid_block_count; 1574 percpu_counter_set(&sbi->alloc_valid_block_count, 0); 1575 percpu_counter_set(&sbi->rf_node_block_count, 0); 1576 1577 /* Here, we have one bio having CP pack except cp pack 2 page */ 1578 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO); 1579 /* Wait for all dirty meta pages to be submitted for IO */ 1580 f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META); 1581 1582 /* wait for previous submitted meta pages writeback */ 1583 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA); 1584 1585 /* flush all device cache */ 1586 err = f2fs_flush_device_cache(sbi); 1587 if (err) 1588 return err; 1589 1590 /* barrier and flush checkpoint cp pack 2 page if it can */ 1591 commit_checkpoint(sbi, ckpt, start_blk); 1592 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA); 1593 1594 /* 1595 * invalidate intermediate page cache borrowed from meta inode which are 1596 * used for migration of encrypted, verity or compressed inode's blocks. 1597 */ 1598 if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) || 1599 f2fs_sb_has_compression(sbi)) 1600 invalidate_mapping_pages(META_MAPPING(sbi), 1601 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1); 1602 1603 f2fs_release_ino_entry(sbi, false); 1604 1605 f2fs_reset_fsync_node_info(sbi); 1606 1607 clear_sbi_flag(sbi, SBI_IS_DIRTY); 1608 clear_sbi_flag(sbi, SBI_NEED_CP); 1609 clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH); 1610 1611 spin_lock(&sbi->stat_lock); 1612 sbi->unusable_block_count = 0; 1613 spin_unlock(&sbi->stat_lock); 1614 1615 __set_cp_next_pack(sbi); 1616 1617 /* 1618 * redirty superblock if metadata like node page or inode cache is 1619 * updated during writing checkpoint. 1620 */ 1621 if (get_pages(sbi, F2FS_DIRTY_NODES) || 1622 get_pages(sbi, F2FS_DIRTY_IMETA)) 1623 set_sbi_flag(sbi, SBI_IS_DIRTY); 1624 1625 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS)); 1626 1627 return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0; 1628 } 1629 1630 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1631 { 1632 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1633 unsigned long long ckpt_ver; 1634 int err = 0; 1635 1636 if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi)) 1637 return -EROFS; 1638 1639 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 1640 if (cpc->reason != CP_PAUSE) 1641 return 0; 1642 f2fs_warn(sbi, "Start checkpoint disabled!"); 1643 } 1644 if (cpc->reason != CP_RESIZE) 1645 f2fs_down_write(&sbi->cp_global_sem); 1646 1647 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) && 1648 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) || 1649 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks))) 1650 goto out; 1651 if (unlikely(f2fs_cp_error(sbi))) { 1652 err = -EIO; 1653 goto out; 1654 } 1655 1656 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops"); 1657 1658 err = block_operations(sbi); 1659 if (err) 1660 goto out; 1661 1662 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops"); 1663 1664 f2fs_flush_merged_writes(sbi); 1665 1666 /* this is the case of multiple fstrims without any changes */ 1667 if (cpc->reason & CP_DISCARD) { 1668 if (!f2fs_exist_trim_candidates(sbi, cpc)) { 1669 unblock_operations(sbi); 1670 goto out; 1671 } 1672 1673 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 && 1674 SIT_I(sbi)->dirty_sentries == 0 && 1675 prefree_segments(sbi) == 0) { 1676 f2fs_flush_sit_entries(sbi, cpc); 1677 f2fs_clear_prefree_segments(sbi, cpc); 1678 unblock_operations(sbi); 1679 goto out; 1680 } 1681 } 1682 1683 /* 1684 * update checkpoint pack index 1685 * Increase the version number so that 1686 * SIT entries and seg summaries are written at correct place 1687 */ 1688 ckpt_ver = cur_cp_version(ckpt); 1689 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver); 1690 1691 /* write cached NAT/SIT entries to NAT/SIT area */ 1692 err = f2fs_flush_nat_entries(sbi, cpc); 1693 if (err) { 1694 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err); 1695 f2fs_bug_on(sbi, !f2fs_cp_error(sbi)); 1696 goto stop; 1697 } 1698 1699 f2fs_flush_sit_entries(sbi, cpc); 1700 1701 /* save inmem log status */ 1702 f2fs_save_inmem_curseg(sbi); 1703 1704 err = do_checkpoint(sbi, cpc); 1705 if (err) { 1706 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err); 1707 f2fs_bug_on(sbi, !f2fs_cp_error(sbi)); 1708 f2fs_release_discard_addrs(sbi); 1709 } else { 1710 f2fs_clear_prefree_segments(sbi, cpc); 1711 } 1712 1713 f2fs_restore_inmem_curseg(sbi); 1714 stop: 1715 unblock_operations(sbi); 1716 stat_inc_cp_count(sbi->stat_info); 1717 1718 if (cpc->reason & CP_RECOVERY) 1719 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver); 1720 1721 /* update CP_TIME to trigger checkpoint periodically */ 1722 f2fs_update_time(sbi, CP_TIME); 1723 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint"); 1724 out: 1725 if (cpc->reason != CP_RESIZE) 1726 f2fs_up_write(&sbi->cp_global_sem); 1727 return err; 1728 } 1729 1730 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi) 1731 { 1732 int i; 1733 1734 for (i = 0; i < MAX_INO_ENTRY; i++) { 1735 struct inode_management *im = &sbi->im[i]; 1736 1737 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC); 1738 spin_lock_init(&im->ino_lock); 1739 INIT_LIST_HEAD(&im->ino_list); 1740 im->ino_num = 0; 1741 } 1742 1743 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS - 1744 NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) * 1745 F2FS_ORPHANS_PER_BLOCK; 1746 } 1747 1748 int __init f2fs_create_checkpoint_caches(void) 1749 { 1750 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry", 1751 sizeof(struct ino_entry)); 1752 if (!ino_entry_slab) 1753 return -ENOMEM; 1754 f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry", 1755 sizeof(struct inode_entry)); 1756 if (!f2fs_inode_entry_slab) { 1757 kmem_cache_destroy(ino_entry_slab); 1758 return -ENOMEM; 1759 } 1760 return 0; 1761 } 1762 1763 void f2fs_destroy_checkpoint_caches(void) 1764 { 1765 kmem_cache_destroy(ino_entry_slab); 1766 kmem_cache_destroy(f2fs_inode_entry_slab); 1767 } 1768 1769 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi) 1770 { 1771 struct cp_control cpc = { .reason = CP_SYNC, }; 1772 int err; 1773 1774 f2fs_down_write(&sbi->gc_lock); 1775 err = f2fs_write_checkpoint(sbi, &cpc); 1776 f2fs_up_write(&sbi->gc_lock); 1777 1778 return err; 1779 } 1780 1781 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi) 1782 { 1783 struct ckpt_req_control *cprc = &sbi->cprc_info; 1784 struct ckpt_req *req, *next; 1785 struct llist_node *dispatch_list; 1786 u64 sum_diff = 0, diff, count = 0; 1787 int ret; 1788 1789 dispatch_list = llist_del_all(&cprc->issue_list); 1790 if (!dispatch_list) 1791 return; 1792 dispatch_list = llist_reverse_order(dispatch_list); 1793 1794 ret = __write_checkpoint_sync(sbi); 1795 atomic_inc(&cprc->issued_ckpt); 1796 1797 llist_for_each_entry_safe(req, next, dispatch_list, llnode) { 1798 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time); 1799 req->ret = ret; 1800 complete(&req->wait); 1801 1802 sum_diff += diff; 1803 count++; 1804 } 1805 atomic_sub(count, &cprc->queued_ckpt); 1806 atomic_add(count, &cprc->total_ckpt); 1807 1808 spin_lock(&cprc->stat_lock); 1809 cprc->cur_time = (unsigned int)div64_u64(sum_diff, count); 1810 if (cprc->peak_time < cprc->cur_time) 1811 cprc->peak_time = cprc->cur_time; 1812 spin_unlock(&cprc->stat_lock); 1813 } 1814 1815 static int issue_checkpoint_thread(void *data) 1816 { 1817 struct f2fs_sb_info *sbi = data; 1818 struct ckpt_req_control *cprc = &sbi->cprc_info; 1819 wait_queue_head_t *q = &cprc->ckpt_wait_queue; 1820 repeat: 1821 if (kthread_should_stop()) 1822 return 0; 1823 1824 if (!llist_empty(&cprc->issue_list)) 1825 __checkpoint_and_complete_reqs(sbi); 1826 1827 wait_event_interruptible(*q, 1828 kthread_should_stop() || !llist_empty(&cprc->issue_list)); 1829 goto repeat; 1830 } 1831 1832 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi, 1833 struct ckpt_req *wait_req) 1834 { 1835 struct ckpt_req_control *cprc = &sbi->cprc_info; 1836 1837 if (!llist_empty(&cprc->issue_list)) { 1838 __checkpoint_and_complete_reqs(sbi); 1839 } else { 1840 /* already dispatched by issue_checkpoint_thread */ 1841 if (wait_req) 1842 wait_for_completion(&wait_req->wait); 1843 } 1844 } 1845 1846 static void init_ckpt_req(struct ckpt_req *req) 1847 { 1848 memset(req, 0, sizeof(struct ckpt_req)); 1849 1850 init_completion(&req->wait); 1851 req->queue_time = ktime_get(); 1852 } 1853 1854 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi) 1855 { 1856 struct ckpt_req_control *cprc = &sbi->cprc_info; 1857 struct ckpt_req req; 1858 struct cp_control cpc; 1859 1860 cpc.reason = __get_cp_reason(sbi); 1861 if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) { 1862 int ret; 1863 1864 f2fs_down_write(&sbi->gc_lock); 1865 ret = f2fs_write_checkpoint(sbi, &cpc); 1866 f2fs_up_write(&sbi->gc_lock); 1867 1868 return ret; 1869 } 1870 1871 if (!cprc->f2fs_issue_ckpt) 1872 return __write_checkpoint_sync(sbi); 1873 1874 init_ckpt_req(&req); 1875 1876 llist_add(&req.llnode, &cprc->issue_list); 1877 atomic_inc(&cprc->queued_ckpt); 1878 1879 /* 1880 * update issue_list before we wake up issue_checkpoint thread, 1881 * this smp_mb() pairs with another barrier in ___wait_event(), 1882 * see more details in comments of waitqueue_active(). 1883 */ 1884 smp_mb(); 1885 1886 if (waitqueue_active(&cprc->ckpt_wait_queue)) 1887 wake_up(&cprc->ckpt_wait_queue); 1888 1889 if (cprc->f2fs_issue_ckpt) 1890 wait_for_completion(&req.wait); 1891 else 1892 flush_remained_ckpt_reqs(sbi, &req); 1893 1894 return req.ret; 1895 } 1896 1897 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi) 1898 { 1899 dev_t dev = sbi->sb->s_bdev->bd_dev; 1900 struct ckpt_req_control *cprc = &sbi->cprc_info; 1901 1902 if (cprc->f2fs_issue_ckpt) 1903 return 0; 1904 1905 cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi, 1906 "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev)); 1907 if (IS_ERR(cprc->f2fs_issue_ckpt)) { 1908 int err = PTR_ERR(cprc->f2fs_issue_ckpt); 1909 1910 cprc->f2fs_issue_ckpt = NULL; 1911 return err; 1912 } 1913 1914 set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio); 1915 1916 return 0; 1917 } 1918 1919 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi) 1920 { 1921 struct ckpt_req_control *cprc = &sbi->cprc_info; 1922 struct task_struct *ckpt_task; 1923 1924 if (!cprc->f2fs_issue_ckpt) 1925 return; 1926 1927 ckpt_task = cprc->f2fs_issue_ckpt; 1928 cprc->f2fs_issue_ckpt = NULL; 1929 kthread_stop(ckpt_task); 1930 1931 f2fs_flush_ckpt_thread(sbi); 1932 } 1933 1934 void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi) 1935 { 1936 struct ckpt_req_control *cprc = &sbi->cprc_info; 1937 1938 flush_remained_ckpt_reqs(sbi, NULL); 1939 1940 /* Let's wait for the previous dispatched checkpoint. */ 1941 while (atomic_read(&cprc->queued_ckpt)) 1942 io_schedule_timeout(DEFAULT_IO_TIMEOUT); 1943 } 1944 1945 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi) 1946 { 1947 struct ckpt_req_control *cprc = &sbi->cprc_info; 1948 1949 atomic_set(&cprc->issued_ckpt, 0); 1950 atomic_set(&cprc->total_ckpt, 0); 1951 atomic_set(&cprc->queued_ckpt, 0); 1952 cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO; 1953 init_waitqueue_head(&cprc->ckpt_wait_queue); 1954 init_llist_head(&cprc->issue_list); 1955 spin_lock_init(&cprc->stat_lock); 1956 } 1957