1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * fs/f2fs/checkpoint.c 4 * 5 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 6 * http://www.samsung.com/ 7 */ 8 #include <linux/fs.h> 9 #include <linux/bio.h> 10 #include <linux/mpage.h> 11 #include <linux/writeback.h> 12 #include <linux/blkdev.h> 13 #include <linux/f2fs_fs.h> 14 #include <linux/pagevec.h> 15 #include <linux/swap.h> 16 #include <linux/kthread.h> 17 18 #include "f2fs.h" 19 #include "node.h" 20 #include "segment.h" 21 #include <trace/events/f2fs.h> 22 23 #define DEFAULT_CHECKPOINT_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3)) 24 25 static struct kmem_cache *ino_entry_slab; 26 struct kmem_cache *f2fs_inode_entry_slab; 27 28 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io) 29 { 30 f2fs_build_fault_attr(sbi, 0, 0); 31 set_ckpt_flags(sbi, CP_ERROR_FLAG); 32 if (!end_io) 33 f2fs_flush_merged_writes(sbi); 34 } 35 36 /* 37 * We guarantee no failure on the returned page. 38 */ 39 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index) 40 { 41 struct address_space *mapping = META_MAPPING(sbi); 42 struct page *page; 43 repeat: 44 page = f2fs_grab_cache_page(mapping, index, false); 45 if (!page) { 46 cond_resched(); 47 goto repeat; 48 } 49 f2fs_wait_on_page_writeback(page, META, true, true); 50 if (!PageUptodate(page)) 51 SetPageUptodate(page); 52 return page; 53 } 54 55 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index, 56 bool is_meta) 57 { 58 struct address_space *mapping = META_MAPPING(sbi); 59 struct page *page; 60 struct f2fs_io_info fio = { 61 .sbi = sbi, 62 .type = META, 63 .op = REQ_OP_READ, 64 .op_flags = REQ_META | REQ_PRIO, 65 .old_blkaddr = index, 66 .new_blkaddr = index, 67 .encrypted_page = NULL, 68 .is_por = !is_meta, 69 }; 70 int err; 71 72 if (unlikely(!is_meta)) 73 fio.op_flags &= ~REQ_META; 74 repeat: 75 page = f2fs_grab_cache_page(mapping, index, false); 76 if (!page) { 77 cond_resched(); 78 goto repeat; 79 } 80 if (PageUptodate(page)) 81 goto out; 82 83 fio.page = page; 84 85 err = f2fs_submit_page_bio(&fio); 86 if (err) { 87 f2fs_put_page(page, 1); 88 return ERR_PTR(err); 89 } 90 91 f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE); 92 93 lock_page(page); 94 if (unlikely(page->mapping != mapping)) { 95 f2fs_put_page(page, 1); 96 goto repeat; 97 } 98 99 if (unlikely(!PageUptodate(page))) { 100 f2fs_put_page(page, 1); 101 return ERR_PTR(-EIO); 102 } 103 out: 104 return page; 105 } 106 107 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index) 108 { 109 return __get_meta_page(sbi, index, true); 110 } 111 112 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index) 113 { 114 struct page *page; 115 int count = 0; 116 117 retry: 118 page = __get_meta_page(sbi, index, true); 119 if (IS_ERR(page)) { 120 if (PTR_ERR(page) == -EIO && 121 ++count <= DEFAULT_RETRY_IO_COUNT) 122 goto retry; 123 f2fs_stop_checkpoint(sbi, false); 124 } 125 return page; 126 } 127 128 /* for POR only */ 129 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index) 130 { 131 return __get_meta_page(sbi, index, false); 132 } 133 134 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr, 135 int type) 136 { 137 struct seg_entry *se; 138 unsigned int segno, offset; 139 bool exist; 140 141 if (type != DATA_GENERIC_ENHANCE && type != DATA_GENERIC_ENHANCE_READ) 142 return true; 143 144 segno = GET_SEGNO(sbi, blkaddr); 145 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 146 se = get_seg_entry(sbi, segno); 147 148 exist = f2fs_test_bit(offset, se->cur_valid_map); 149 if (!exist && type == DATA_GENERIC_ENHANCE) { 150 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d", 151 blkaddr, exist); 152 set_sbi_flag(sbi, SBI_NEED_FSCK); 153 WARN_ON(1); 154 } 155 return exist; 156 } 157 158 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi, 159 block_t blkaddr, int type) 160 { 161 switch (type) { 162 case META_NAT: 163 break; 164 case META_SIT: 165 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi))) 166 return false; 167 break; 168 case META_SSA: 169 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) || 170 blkaddr < SM_I(sbi)->ssa_blkaddr)) 171 return false; 172 break; 173 case META_CP: 174 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr || 175 blkaddr < __start_cp_addr(sbi))) 176 return false; 177 break; 178 case META_POR: 179 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) || 180 blkaddr < MAIN_BLKADDR(sbi))) 181 return false; 182 break; 183 case DATA_GENERIC: 184 case DATA_GENERIC_ENHANCE: 185 case DATA_GENERIC_ENHANCE_READ: 186 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) || 187 blkaddr < MAIN_BLKADDR(sbi))) { 188 f2fs_warn(sbi, "access invalid blkaddr:%u", 189 blkaddr); 190 set_sbi_flag(sbi, SBI_NEED_FSCK); 191 WARN_ON(1); 192 return false; 193 } else { 194 return __is_bitmap_valid(sbi, blkaddr, type); 195 } 196 break; 197 case META_GENERIC: 198 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) || 199 blkaddr >= MAIN_BLKADDR(sbi))) 200 return false; 201 break; 202 default: 203 BUG(); 204 } 205 206 return true; 207 } 208 209 /* 210 * Readahead CP/NAT/SIT/SSA/POR pages 211 */ 212 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages, 213 int type, bool sync) 214 { 215 struct page *page; 216 block_t blkno = start; 217 struct f2fs_io_info fio = { 218 .sbi = sbi, 219 .type = META, 220 .op = REQ_OP_READ, 221 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD, 222 .encrypted_page = NULL, 223 .in_list = false, 224 .is_por = (type == META_POR), 225 }; 226 struct blk_plug plug; 227 int err; 228 229 if (unlikely(type == META_POR)) 230 fio.op_flags &= ~REQ_META; 231 232 blk_start_plug(&plug); 233 for (; nrpages-- > 0; blkno++) { 234 235 if (!f2fs_is_valid_blkaddr(sbi, blkno, type)) 236 goto out; 237 238 switch (type) { 239 case META_NAT: 240 if (unlikely(blkno >= 241 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid))) 242 blkno = 0; 243 /* get nat block addr */ 244 fio.new_blkaddr = current_nat_addr(sbi, 245 blkno * NAT_ENTRY_PER_BLOCK); 246 break; 247 case META_SIT: 248 if (unlikely(blkno >= TOTAL_SEGS(sbi))) 249 goto out; 250 /* get sit block addr */ 251 fio.new_blkaddr = current_sit_addr(sbi, 252 blkno * SIT_ENTRY_PER_BLOCK); 253 break; 254 case META_SSA: 255 case META_CP: 256 case META_POR: 257 fio.new_blkaddr = blkno; 258 break; 259 default: 260 BUG(); 261 } 262 263 page = f2fs_grab_cache_page(META_MAPPING(sbi), 264 fio.new_blkaddr, false); 265 if (!page) 266 continue; 267 if (PageUptodate(page)) { 268 f2fs_put_page(page, 1); 269 continue; 270 } 271 272 fio.page = page; 273 err = f2fs_submit_page_bio(&fio); 274 f2fs_put_page(page, err ? 1 : 0); 275 276 if (!err) 277 f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE); 278 } 279 out: 280 blk_finish_plug(&plug); 281 return blkno - start; 282 } 283 284 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index) 285 { 286 struct page *page; 287 bool readahead = false; 288 289 page = find_get_page(META_MAPPING(sbi), index); 290 if (!page || !PageUptodate(page)) 291 readahead = true; 292 f2fs_put_page(page, 0); 293 294 if (readahead) 295 f2fs_ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true); 296 } 297 298 static int __f2fs_write_meta_page(struct page *page, 299 struct writeback_control *wbc, 300 enum iostat_type io_type) 301 { 302 struct f2fs_sb_info *sbi = F2FS_P_SB(page); 303 304 trace_f2fs_writepage(page, META); 305 306 if (unlikely(f2fs_cp_error(sbi))) 307 goto redirty_out; 308 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 309 goto redirty_out; 310 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0)) 311 goto redirty_out; 312 313 f2fs_do_write_meta_page(sbi, page, io_type); 314 dec_page_count(sbi, F2FS_DIRTY_META); 315 316 if (wbc->for_reclaim) 317 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META); 318 319 unlock_page(page); 320 321 if (unlikely(f2fs_cp_error(sbi))) 322 f2fs_submit_merged_write(sbi, META); 323 324 return 0; 325 326 redirty_out: 327 redirty_page_for_writepage(wbc, page); 328 return AOP_WRITEPAGE_ACTIVATE; 329 } 330 331 static int f2fs_write_meta_page(struct page *page, 332 struct writeback_control *wbc) 333 { 334 return __f2fs_write_meta_page(page, wbc, FS_META_IO); 335 } 336 337 static int f2fs_write_meta_pages(struct address_space *mapping, 338 struct writeback_control *wbc) 339 { 340 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 341 long diff, written; 342 343 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 344 goto skip_write; 345 346 /* collect a number of dirty meta pages and write together */ 347 if (wbc->sync_mode != WB_SYNC_ALL && 348 get_pages(sbi, F2FS_DIRTY_META) < 349 nr_pages_to_skip(sbi, META)) 350 goto skip_write; 351 352 /* if locked failed, cp will flush dirty pages instead */ 353 if (!down_write_trylock(&sbi->cp_global_sem)) 354 goto skip_write; 355 356 trace_f2fs_writepages(mapping->host, wbc, META); 357 diff = nr_pages_to_write(sbi, META, wbc); 358 written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO); 359 up_write(&sbi->cp_global_sem); 360 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff); 361 return 0; 362 363 skip_write: 364 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META); 365 trace_f2fs_writepages(mapping->host, wbc, META); 366 return 0; 367 } 368 369 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type, 370 long nr_to_write, enum iostat_type io_type) 371 { 372 struct address_space *mapping = META_MAPPING(sbi); 373 pgoff_t index = 0, prev = ULONG_MAX; 374 struct pagevec pvec; 375 long nwritten = 0; 376 int nr_pages; 377 struct writeback_control wbc = { 378 .for_reclaim = 0, 379 }; 380 struct blk_plug plug; 381 382 pagevec_init(&pvec); 383 384 blk_start_plug(&plug); 385 386 while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, 387 PAGECACHE_TAG_DIRTY))) { 388 int i; 389 390 for (i = 0; i < nr_pages; i++) { 391 struct page *page = pvec.pages[i]; 392 393 if (prev == ULONG_MAX) 394 prev = page->index - 1; 395 if (nr_to_write != LONG_MAX && page->index != prev + 1) { 396 pagevec_release(&pvec); 397 goto stop; 398 } 399 400 lock_page(page); 401 402 if (unlikely(page->mapping != mapping)) { 403 continue_unlock: 404 unlock_page(page); 405 continue; 406 } 407 if (!PageDirty(page)) { 408 /* someone wrote it for us */ 409 goto continue_unlock; 410 } 411 412 f2fs_wait_on_page_writeback(page, META, true, true); 413 414 if (!clear_page_dirty_for_io(page)) 415 goto continue_unlock; 416 417 if (__f2fs_write_meta_page(page, &wbc, io_type)) { 418 unlock_page(page); 419 break; 420 } 421 nwritten++; 422 prev = page->index; 423 if (unlikely(nwritten >= nr_to_write)) 424 break; 425 } 426 pagevec_release(&pvec); 427 cond_resched(); 428 } 429 stop: 430 if (nwritten) 431 f2fs_submit_merged_write(sbi, type); 432 433 blk_finish_plug(&plug); 434 435 return nwritten; 436 } 437 438 static int f2fs_set_meta_page_dirty(struct page *page) 439 { 440 trace_f2fs_set_page_dirty(page, META); 441 442 if (!PageUptodate(page)) 443 SetPageUptodate(page); 444 if (!PageDirty(page)) { 445 __set_page_dirty_nobuffers(page); 446 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META); 447 f2fs_set_page_private(page, 0); 448 return 1; 449 } 450 return 0; 451 } 452 453 const struct address_space_operations f2fs_meta_aops = { 454 .writepage = f2fs_write_meta_page, 455 .writepages = f2fs_write_meta_pages, 456 .set_page_dirty = f2fs_set_meta_page_dirty, 457 .invalidatepage = f2fs_invalidate_page, 458 .releasepage = f2fs_release_page, 459 #ifdef CONFIG_MIGRATION 460 .migratepage = f2fs_migrate_page, 461 #endif 462 }; 463 464 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, 465 unsigned int devidx, int type) 466 { 467 struct inode_management *im = &sbi->im[type]; 468 struct ino_entry *e, *tmp; 469 470 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS); 471 472 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL); 473 474 spin_lock(&im->ino_lock); 475 e = radix_tree_lookup(&im->ino_root, ino); 476 if (!e) { 477 e = tmp; 478 if (unlikely(radix_tree_insert(&im->ino_root, ino, e))) 479 f2fs_bug_on(sbi, 1); 480 481 memset(e, 0, sizeof(struct ino_entry)); 482 e->ino = ino; 483 484 list_add_tail(&e->list, &im->ino_list); 485 if (type != ORPHAN_INO) 486 im->ino_num++; 487 } 488 489 if (type == FLUSH_INO) 490 f2fs_set_bit(devidx, (char *)&e->dirty_device); 491 492 spin_unlock(&im->ino_lock); 493 radix_tree_preload_end(); 494 495 if (e != tmp) 496 kmem_cache_free(ino_entry_slab, tmp); 497 } 498 499 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) 500 { 501 struct inode_management *im = &sbi->im[type]; 502 struct ino_entry *e; 503 504 spin_lock(&im->ino_lock); 505 e = radix_tree_lookup(&im->ino_root, ino); 506 if (e) { 507 list_del(&e->list); 508 radix_tree_delete(&im->ino_root, ino); 509 im->ino_num--; 510 spin_unlock(&im->ino_lock); 511 kmem_cache_free(ino_entry_slab, e); 512 return; 513 } 514 spin_unlock(&im->ino_lock); 515 } 516 517 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) 518 { 519 /* add new dirty ino entry into list */ 520 __add_ino_entry(sbi, ino, 0, type); 521 } 522 523 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) 524 { 525 /* remove dirty ino entry from list */ 526 __remove_ino_entry(sbi, ino, type); 527 } 528 529 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */ 530 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode) 531 { 532 struct inode_management *im = &sbi->im[mode]; 533 struct ino_entry *e; 534 535 spin_lock(&im->ino_lock); 536 e = radix_tree_lookup(&im->ino_root, ino); 537 spin_unlock(&im->ino_lock); 538 return e ? true : false; 539 } 540 541 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all) 542 { 543 struct ino_entry *e, *tmp; 544 int i; 545 546 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) { 547 struct inode_management *im = &sbi->im[i]; 548 549 spin_lock(&im->ino_lock); 550 list_for_each_entry_safe(e, tmp, &im->ino_list, list) { 551 list_del(&e->list); 552 radix_tree_delete(&im->ino_root, e->ino); 553 kmem_cache_free(ino_entry_slab, e); 554 im->ino_num--; 555 } 556 spin_unlock(&im->ino_lock); 557 } 558 } 559 560 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino, 561 unsigned int devidx, int type) 562 { 563 __add_ino_entry(sbi, ino, devidx, type); 564 } 565 566 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino, 567 unsigned int devidx, int type) 568 { 569 struct inode_management *im = &sbi->im[type]; 570 struct ino_entry *e; 571 bool is_dirty = false; 572 573 spin_lock(&im->ino_lock); 574 e = radix_tree_lookup(&im->ino_root, ino); 575 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device)) 576 is_dirty = true; 577 spin_unlock(&im->ino_lock); 578 return is_dirty; 579 } 580 581 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi) 582 { 583 struct inode_management *im = &sbi->im[ORPHAN_INO]; 584 int err = 0; 585 586 spin_lock(&im->ino_lock); 587 588 if (time_to_inject(sbi, FAULT_ORPHAN)) { 589 spin_unlock(&im->ino_lock); 590 f2fs_show_injection_info(sbi, FAULT_ORPHAN); 591 return -ENOSPC; 592 } 593 594 if (unlikely(im->ino_num >= sbi->max_orphans)) 595 err = -ENOSPC; 596 else 597 im->ino_num++; 598 spin_unlock(&im->ino_lock); 599 600 return err; 601 } 602 603 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi) 604 { 605 struct inode_management *im = &sbi->im[ORPHAN_INO]; 606 607 spin_lock(&im->ino_lock); 608 f2fs_bug_on(sbi, im->ino_num == 0); 609 im->ino_num--; 610 spin_unlock(&im->ino_lock); 611 } 612 613 void f2fs_add_orphan_inode(struct inode *inode) 614 { 615 /* add new orphan ino entry into list */ 616 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO); 617 f2fs_update_inode_page(inode); 618 } 619 620 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) 621 { 622 /* remove orphan entry from orphan list */ 623 __remove_ino_entry(sbi, ino, ORPHAN_INO); 624 } 625 626 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) 627 { 628 struct inode *inode; 629 struct node_info ni; 630 int err; 631 632 inode = f2fs_iget_retry(sbi->sb, ino); 633 if (IS_ERR(inode)) { 634 /* 635 * there should be a bug that we can't find the entry 636 * to orphan inode. 637 */ 638 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT); 639 return PTR_ERR(inode); 640 } 641 642 err = dquot_initialize(inode); 643 if (err) { 644 iput(inode); 645 goto err_out; 646 } 647 648 clear_nlink(inode); 649 650 /* truncate all the data during iput */ 651 iput(inode); 652 653 err = f2fs_get_node_info(sbi, ino, &ni); 654 if (err) 655 goto err_out; 656 657 /* ENOMEM was fully retried in f2fs_evict_inode. */ 658 if (ni.blk_addr != NULL_ADDR) { 659 err = -EIO; 660 goto err_out; 661 } 662 return 0; 663 664 err_out: 665 set_sbi_flag(sbi, SBI_NEED_FSCK); 666 f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.", 667 __func__, ino); 668 return err; 669 } 670 671 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi) 672 { 673 block_t start_blk, orphan_blocks, i, j; 674 unsigned int s_flags = sbi->sb->s_flags; 675 int err = 0; 676 #ifdef CONFIG_QUOTA 677 int quota_enabled; 678 #endif 679 680 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG)) 681 return 0; 682 683 if (bdev_read_only(sbi->sb->s_bdev)) { 684 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup"); 685 return 0; 686 } 687 688 if (s_flags & SB_RDONLY) { 689 f2fs_info(sbi, "orphan cleanup on readonly fs"); 690 sbi->sb->s_flags &= ~SB_RDONLY; 691 } 692 693 #ifdef CONFIG_QUOTA 694 /* Needed for iput() to work correctly and not trash data */ 695 sbi->sb->s_flags |= SB_ACTIVE; 696 697 /* 698 * Turn on quotas which were not enabled for read-only mounts if 699 * filesystem has quota feature, so that they are updated correctly. 700 */ 701 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY); 702 #endif 703 704 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi); 705 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi); 706 707 f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true); 708 709 for (i = 0; i < orphan_blocks; i++) { 710 struct page *page; 711 struct f2fs_orphan_block *orphan_blk; 712 713 page = f2fs_get_meta_page(sbi, start_blk + i); 714 if (IS_ERR(page)) { 715 err = PTR_ERR(page); 716 goto out; 717 } 718 719 orphan_blk = (struct f2fs_orphan_block *)page_address(page); 720 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) { 721 nid_t ino = le32_to_cpu(orphan_blk->ino[j]); 722 err = recover_orphan_inode(sbi, ino); 723 if (err) { 724 f2fs_put_page(page, 1); 725 goto out; 726 } 727 } 728 f2fs_put_page(page, 1); 729 } 730 /* clear Orphan Flag */ 731 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG); 732 out: 733 set_sbi_flag(sbi, SBI_IS_RECOVERED); 734 735 #ifdef CONFIG_QUOTA 736 /* Turn quotas off */ 737 if (quota_enabled) 738 f2fs_quota_off_umount(sbi->sb); 739 #endif 740 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */ 741 742 return err; 743 } 744 745 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk) 746 { 747 struct list_head *head; 748 struct f2fs_orphan_block *orphan_blk = NULL; 749 unsigned int nentries = 0; 750 unsigned short index = 1; 751 unsigned short orphan_blocks; 752 struct page *page = NULL; 753 struct ino_entry *orphan = NULL; 754 struct inode_management *im = &sbi->im[ORPHAN_INO]; 755 756 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num); 757 758 /* 759 * we don't need to do spin_lock(&im->ino_lock) here, since all the 760 * orphan inode operations are covered under f2fs_lock_op(). 761 * And, spin_lock should be avoided due to page operations below. 762 */ 763 head = &im->ino_list; 764 765 /* loop for each orphan inode entry and write them in Jornal block */ 766 list_for_each_entry(orphan, head, list) { 767 if (!page) { 768 page = f2fs_grab_meta_page(sbi, start_blk++); 769 orphan_blk = 770 (struct f2fs_orphan_block *)page_address(page); 771 memset(orphan_blk, 0, sizeof(*orphan_blk)); 772 } 773 774 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino); 775 776 if (nentries == F2FS_ORPHANS_PER_BLOCK) { 777 /* 778 * an orphan block is full of 1020 entries, 779 * then we need to flush current orphan blocks 780 * and bring another one in memory 781 */ 782 orphan_blk->blk_addr = cpu_to_le16(index); 783 orphan_blk->blk_count = cpu_to_le16(orphan_blocks); 784 orphan_blk->entry_count = cpu_to_le32(nentries); 785 set_page_dirty(page); 786 f2fs_put_page(page, 1); 787 index++; 788 nentries = 0; 789 page = NULL; 790 } 791 } 792 793 if (page) { 794 orphan_blk->blk_addr = cpu_to_le16(index); 795 orphan_blk->blk_count = cpu_to_le16(orphan_blocks); 796 orphan_blk->entry_count = cpu_to_le32(nentries); 797 set_page_dirty(page); 798 f2fs_put_page(page, 1); 799 } 800 } 801 802 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi, 803 struct f2fs_checkpoint *ckpt) 804 { 805 unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset); 806 __u32 chksum; 807 808 chksum = f2fs_crc32(sbi, ckpt, chksum_ofs); 809 if (chksum_ofs < CP_CHKSUM_OFFSET) { 810 chksum_ofs += sizeof(chksum); 811 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs, 812 F2FS_BLKSIZE - chksum_ofs); 813 } 814 return chksum; 815 } 816 817 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr, 818 struct f2fs_checkpoint **cp_block, struct page **cp_page, 819 unsigned long long *version) 820 { 821 size_t crc_offset = 0; 822 __u32 crc; 823 824 *cp_page = f2fs_get_meta_page(sbi, cp_addr); 825 if (IS_ERR(*cp_page)) 826 return PTR_ERR(*cp_page); 827 828 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page); 829 830 crc_offset = le32_to_cpu((*cp_block)->checksum_offset); 831 if (crc_offset < CP_MIN_CHKSUM_OFFSET || 832 crc_offset > CP_CHKSUM_OFFSET) { 833 f2fs_put_page(*cp_page, 1); 834 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset); 835 return -EINVAL; 836 } 837 838 crc = f2fs_checkpoint_chksum(sbi, *cp_block); 839 if (crc != cur_cp_crc(*cp_block)) { 840 f2fs_put_page(*cp_page, 1); 841 f2fs_warn(sbi, "invalid crc value"); 842 return -EINVAL; 843 } 844 845 *version = cur_cp_version(*cp_block); 846 return 0; 847 } 848 849 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi, 850 block_t cp_addr, unsigned long long *version) 851 { 852 struct page *cp_page_1 = NULL, *cp_page_2 = NULL; 853 struct f2fs_checkpoint *cp_block = NULL; 854 unsigned long long cur_version = 0, pre_version = 0; 855 int err; 856 857 err = get_checkpoint_version(sbi, cp_addr, &cp_block, 858 &cp_page_1, version); 859 if (err) 860 return NULL; 861 862 if (le32_to_cpu(cp_block->cp_pack_total_block_count) > 863 sbi->blocks_per_seg) { 864 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u", 865 le32_to_cpu(cp_block->cp_pack_total_block_count)); 866 goto invalid_cp; 867 } 868 pre_version = *version; 869 870 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1; 871 err = get_checkpoint_version(sbi, cp_addr, &cp_block, 872 &cp_page_2, version); 873 if (err) 874 goto invalid_cp; 875 cur_version = *version; 876 877 if (cur_version == pre_version) { 878 *version = cur_version; 879 f2fs_put_page(cp_page_2, 1); 880 return cp_page_1; 881 } 882 f2fs_put_page(cp_page_2, 1); 883 invalid_cp: 884 f2fs_put_page(cp_page_1, 1); 885 return NULL; 886 } 887 888 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi) 889 { 890 struct f2fs_checkpoint *cp_block; 891 struct f2fs_super_block *fsb = sbi->raw_super; 892 struct page *cp1, *cp2, *cur_page; 893 unsigned long blk_size = sbi->blocksize; 894 unsigned long long cp1_version = 0, cp2_version = 0; 895 unsigned long long cp_start_blk_no; 896 unsigned int cp_blks = 1 + __cp_payload(sbi); 897 block_t cp_blk_no; 898 int i; 899 int err; 900 901 sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks), 902 GFP_KERNEL); 903 if (!sbi->ckpt) 904 return -ENOMEM; 905 /* 906 * Finding out valid cp block involves read both 907 * sets( cp pack 1 and cp pack 2) 908 */ 909 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr); 910 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version); 911 912 /* The second checkpoint pack should start at the next segment */ 913 cp_start_blk_no += ((unsigned long long)1) << 914 le32_to_cpu(fsb->log_blocks_per_seg); 915 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version); 916 917 if (cp1 && cp2) { 918 if (ver_after(cp2_version, cp1_version)) 919 cur_page = cp2; 920 else 921 cur_page = cp1; 922 } else if (cp1) { 923 cur_page = cp1; 924 } else if (cp2) { 925 cur_page = cp2; 926 } else { 927 err = -EFSCORRUPTED; 928 goto fail_no_cp; 929 } 930 931 cp_block = (struct f2fs_checkpoint *)page_address(cur_page); 932 memcpy(sbi->ckpt, cp_block, blk_size); 933 934 if (cur_page == cp1) 935 sbi->cur_cp_pack = 1; 936 else 937 sbi->cur_cp_pack = 2; 938 939 /* Sanity checking of checkpoint */ 940 if (f2fs_sanity_check_ckpt(sbi)) { 941 err = -EFSCORRUPTED; 942 goto free_fail_no_cp; 943 } 944 945 if (cp_blks <= 1) 946 goto done; 947 948 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr); 949 if (cur_page == cp2) 950 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg); 951 952 for (i = 1; i < cp_blks; i++) { 953 void *sit_bitmap_ptr; 954 unsigned char *ckpt = (unsigned char *)sbi->ckpt; 955 956 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i); 957 if (IS_ERR(cur_page)) { 958 err = PTR_ERR(cur_page); 959 goto free_fail_no_cp; 960 } 961 sit_bitmap_ptr = page_address(cur_page); 962 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size); 963 f2fs_put_page(cur_page, 1); 964 } 965 done: 966 f2fs_put_page(cp1, 1); 967 f2fs_put_page(cp2, 1); 968 return 0; 969 970 free_fail_no_cp: 971 f2fs_put_page(cp1, 1); 972 f2fs_put_page(cp2, 1); 973 fail_no_cp: 974 kvfree(sbi->ckpt); 975 return err; 976 } 977 978 static void __add_dirty_inode(struct inode *inode, enum inode_type type) 979 { 980 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 981 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE; 982 983 if (is_inode_flag_set(inode, flag)) 984 return; 985 986 set_inode_flag(inode, flag); 987 if (!f2fs_is_volatile_file(inode)) 988 list_add_tail(&F2FS_I(inode)->dirty_list, 989 &sbi->inode_list[type]); 990 stat_inc_dirty_inode(sbi, type); 991 } 992 993 static void __remove_dirty_inode(struct inode *inode, enum inode_type type) 994 { 995 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE; 996 997 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag)) 998 return; 999 1000 list_del_init(&F2FS_I(inode)->dirty_list); 1001 clear_inode_flag(inode, flag); 1002 stat_dec_dirty_inode(F2FS_I_SB(inode), type); 1003 } 1004 1005 void f2fs_update_dirty_page(struct inode *inode, struct page *page) 1006 { 1007 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1008 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE; 1009 1010 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) && 1011 !S_ISLNK(inode->i_mode)) 1012 return; 1013 1014 spin_lock(&sbi->inode_lock[type]); 1015 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH)) 1016 __add_dirty_inode(inode, type); 1017 inode_inc_dirty_pages(inode); 1018 spin_unlock(&sbi->inode_lock[type]); 1019 1020 f2fs_set_page_private(page, 0); 1021 } 1022 1023 void f2fs_remove_dirty_inode(struct inode *inode) 1024 { 1025 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1026 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE; 1027 1028 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) && 1029 !S_ISLNK(inode->i_mode)) 1030 return; 1031 1032 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH)) 1033 return; 1034 1035 spin_lock(&sbi->inode_lock[type]); 1036 __remove_dirty_inode(inode, type); 1037 spin_unlock(&sbi->inode_lock[type]); 1038 } 1039 1040 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type) 1041 { 1042 struct list_head *head; 1043 struct inode *inode; 1044 struct f2fs_inode_info *fi; 1045 bool is_dir = (type == DIR_INODE); 1046 unsigned long ino = 0; 1047 1048 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir, 1049 get_pages(sbi, is_dir ? 1050 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA)); 1051 retry: 1052 if (unlikely(f2fs_cp_error(sbi))) { 1053 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir, 1054 get_pages(sbi, is_dir ? 1055 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA)); 1056 return -EIO; 1057 } 1058 1059 spin_lock(&sbi->inode_lock[type]); 1060 1061 head = &sbi->inode_list[type]; 1062 if (list_empty(head)) { 1063 spin_unlock(&sbi->inode_lock[type]); 1064 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir, 1065 get_pages(sbi, is_dir ? 1066 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA)); 1067 return 0; 1068 } 1069 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list); 1070 inode = igrab(&fi->vfs_inode); 1071 spin_unlock(&sbi->inode_lock[type]); 1072 if (inode) { 1073 unsigned long cur_ino = inode->i_ino; 1074 1075 F2FS_I(inode)->cp_task = current; 1076 1077 filemap_fdatawrite(inode->i_mapping); 1078 1079 F2FS_I(inode)->cp_task = NULL; 1080 1081 iput(inode); 1082 /* We need to give cpu to another writers. */ 1083 if (ino == cur_ino) 1084 cond_resched(); 1085 else 1086 ino = cur_ino; 1087 } else { 1088 /* 1089 * We should submit bio, since it exists several 1090 * wribacking dentry pages in the freeing inode. 1091 */ 1092 f2fs_submit_merged_write(sbi, DATA); 1093 cond_resched(); 1094 } 1095 goto retry; 1096 } 1097 1098 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi) 1099 { 1100 struct list_head *head = &sbi->inode_list[DIRTY_META]; 1101 struct inode *inode; 1102 struct f2fs_inode_info *fi; 1103 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA); 1104 1105 while (total--) { 1106 if (unlikely(f2fs_cp_error(sbi))) 1107 return -EIO; 1108 1109 spin_lock(&sbi->inode_lock[DIRTY_META]); 1110 if (list_empty(head)) { 1111 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1112 return 0; 1113 } 1114 fi = list_first_entry(head, struct f2fs_inode_info, 1115 gdirty_list); 1116 inode = igrab(&fi->vfs_inode); 1117 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1118 if (inode) { 1119 sync_inode_metadata(inode, 0); 1120 1121 /* it's on eviction */ 1122 if (is_inode_flag_set(inode, FI_DIRTY_INODE)) 1123 f2fs_update_inode_page(inode); 1124 iput(inode); 1125 } 1126 } 1127 return 0; 1128 } 1129 1130 static void __prepare_cp_block(struct f2fs_sb_info *sbi) 1131 { 1132 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1133 struct f2fs_nm_info *nm_i = NM_I(sbi); 1134 nid_t last_nid = nm_i->next_scan_nid; 1135 1136 next_free_nid(sbi, &last_nid); 1137 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi)); 1138 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi)); 1139 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi)); 1140 ckpt->next_free_nid = cpu_to_le32(last_nid); 1141 } 1142 1143 static bool __need_flush_quota(struct f2fs_sb_info *sbi) 1144 { 1145 bool ret = false; 1146 1147 if (!is_journalled_quota(sbi)) 1148 return false; 1149 1150 down_write(&sbi->quota_sem); 1151 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) { 1152 ret = false; 1153 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) { 1154 ret = false; 1155 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) { 1156 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH); 1157 ret = true; 1158 } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) { 1159 ret = true; 1160 } 1161 up_write(&sbi->quota_sem); 1162 return ret; 1163 } 1164 1165 /* 1166 * Freeze all the FS-operations for checkpoint. 1167 */ 1168 static int block_operations(struct f2fs_sb_info *sbi) 1169 { 1170 struct writeback_control wbc = { 1171 .sync_mode = WB_SYNC_ALL, 1172 .nr_to_write = LONG_MAX, 1173 .for_reclaim = 0, 1174 }; 1175 int err = 0, cnt = 0; 1176 1177 /* 1178 * Let's flush inline_data in dirty node pages. 1179 */ 1180 f2fs_flush_inline_data(sbi); 1181 1182 retry_flush_quotas: 1183 f2fs_lock_all(sbi); 1184 if (__need_flush_quota(sbi)) { 1185 int locked; 1186 1187 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) { 1188 set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH); 1189 set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH); 1190 goto retry_flush_dents; 1191 } 1192 f2fs_unlock_all(sbi); 1193 1194 /* only failed during mount/umount/freeze/quotactl */ 1195 locked = down_read_trylock(&sbi->sb->s_umount); 1196 f2fs_quota_sync(sbi->sb, -1); 1197 if (locked) 1198 up_read(&sbi->sb->s_umount); 1199 cond_resched(); 1200 goto retry_flush_quotas; 1201 } 1202 1203 retry_flush_dents: 1204 /* write all the dirty dentry pages */ 1205 if (get_pages(sbi, F2FS_DIRTY_DENTS)) { 1206 f2fs_unlock_all(sbi); 1207 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE); 1208 if (err) 1209 return err; 1210 cond_resched(); 1211 goto retry_flush_quotas; 1212 } 1213 1214 /* 1215 * POR: we should ensure that there are no dirty node pages 1216 * until finishing nat/sit flush. inode->i_blocks can be updated. 1217 */ 1218 down_write(&sbi->node_change); 1219 1220 if (get_pages(sbi, F2FS_DIRTY_IMETA)) { 1221 up_write(&sbi->node_change); 1222 f2fs_unlock_all(sbi); 1223 err = f2fs_sync_inode_meta(sbi); 1224 if (err) 1225 return err; 1226 cond_resched(); 1227 goto retry_flush_quotas; 1228 } 1229 1230 retry_flush_nodes: 1231 down_write(&sbi->node_write); 1232 1233 if (get_pages(sbi, F2FS_DIRTY_NODES)) { 1234 up_write(&sbi->node_write); 1235 atomic_inc(&sbi->wb_sync_req[NODE]); 1236 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO); 1237 atomic_dec(&sbi->wb_sync_req[NODE]); 1238 if (err) { 1239 up_write(&sbi->node_change); 1240 f2fs_unlock_all(sbi); 1241 return err; 1242 } 1243 cond_resched(); 1244 goto retry_flush_nodes; 1245 } 1246 1247 /* 1248 * sbi->node_change is used only for AIO write_begin path which produces 1249 * dirty node blocks and some checkpoint values by block allocation. 1250 */ 1251 __prepare_cp_block(sbi); 1252 up_write(&sbi->node_change); 1253 return err; 1254 } 1255 1256 static void unblock_operations(struct f2fs_sb_info *sbi) 1257 { 1258 up_write(&sbi->node_write); 1259 f2fs_unlock_all(sbi); 1260 } 1261 1262 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type) 1263 { 1264 DEFINE_WAIT(wait); 1265 1266 for (;;) { 1267 if (!get_pages(sbi, type)) 1268 break; 1269 1270 if (unlikely(f2fs_cp_error(sbi))) 1271 break; 1272 1273 if (type == F2FS_DIRTY_META) 1274 f2fs_sync_meta_pages(sbi, META, LONG_MAX, 1275 FS_CP_META_IO); 1276 else if (type == F2FS_WB_CP_DATA) 1277 f2fs_submit_merged_write(sbi, DATA); 1278 1279 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE); 1280 io_schedule_timeout(DEFAULT_IO_TIMEOUT); 1281 } 1282 finish_wait(&sbi->cp_wait, &wait); 1283 } 1284 1285 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1286 { 1287 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num; 1288 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1289 unsigned long flags; 1290 1291 spin_lock_irqsave(&sbi->cp_lock, flags); 1292 1293 if ((cpc->reason & CP_UMOUNT) && 1294 le32_to_cpu(ckpt->cp_pack_total_block_count) > 1295 sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks) 1296 disable_nat_bits(sbi, false); 1297 1298 if (cpc->reason & CP_TRIMMED) 1299 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG); 1300 else 1301 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG); 1302 1303 if (cpc->reason & CP_UMOUNT) 1304 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG); 1305 else 1306 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG); 1307 1308 if (cpc->reason & CP_FASTBOOT) 1309 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG); 1310 else 1311 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG); 1312 1313 if (orphan_num) 1314 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); 1315 else 1316 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); 1317 1318 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1319 __set_ckpt_flags(ckpt, CP_FSCK_FLAG); 1320 1321 if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS)) 1322 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG); 1323 else 1324 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG); 1325 1326 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED)) 1327 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG); 1328 else 1329 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG); 1330 1331 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK)) 1332 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG); 1333 else 1334 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG); 1335 1336 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) 1337 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG); 1338 else 1339 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG); 1340 1341 if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) 1342 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG); 1343 1344 /* set this flag to activate crc|cp_ver for recovery */ 1345 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG); 1346 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG); 1347 1348 spin_unlock_irqrestore(&sbi->cp_lock, flags); 1349 } 1350 1351 static void commit_checkpoint(struct f2fs_sb_info *sbi, 1352 void *src, block_t blk_addr) 1353 { 1354 struct writeback_control wbc = { 1355 .for_reclaim = 0, 1356 }; 1357 1358 /* 1359 * pagevec_lookup_tag and lock_page again will take 1360 * some extra time. Therefore, f2fs_update_meta_pages and 1361 * f2fs_sync_meta_pages are combined in this function. 1362 */ 1363 struct page *page = f2fs_grab_meta_page(sbi, blk_addr); 1364 int err; 1365 1366 f2fs_wait_on_page_writeback(page, META, true, true); 1367 1368 memcpy(page_address(page), src, PAGE_SIZE); 1369 1370 set_page_dirty(page); 1371 if (unlikely(!clear_page_dirty_for_io(page))) 1372 f2fs_bug_on(sbi, 1); 1373 1374 /* writeout cp pack 2 page */ 1375 err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO); 1376 if (unlikely(err && f2fs_cp_error(sbi))) { 1377 f2fs_put_page(page, 1); 1378 return; 1379 } 1380 1381 f2fs_bug_on(sbi, err); 1382 f2fs_put_page(page, 0); 1383 1384 /* submit checkpoint (with barrier if NOBARRIER is not set) */ 1385 f2fs_submit_merged_write(sbi, META_FLUSH); 1386 } 1387 1388 static inline u64 get_sectors_written(struct block_device *bdev) 1389 { 1390 return (u64)part_stat_read(bdev, sectors[STAT_WRITE]); 1391 } 1392 1393 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi) 1394 { 1395 if (f2fs_is_multi_device(sbi)) { 1396 u64 sectors = 0; 1397 int i; 1398 1399 for (i = 0; i < sbi->s_ndevs; i++) 1400 sectors += get_sectors_written(FDEV(i).bdev); 1401 1402 return sectors; 1403 } 1404 1405 return get_sectors_written(sbi->sb->s_bdev); 1406 } 1407 1408 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1409 { 1410 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1411 struct f2fs_nm_info *nm_i = NM_I(sbi); 1412 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags; 1413 block_t start_blk; 1414 unsigned int data_sum_blocks, orphan_blocks; 1415 __u32 crc32 = 0; 1416 int i; 1417 int cp_payload_blks = __cp_payload(sbi); 1418 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE); 1419 u64 kbytes_written; 1420 int err; 1421 1422 /* Flush all the NAT/SIT pages */ 1423 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO); 1424 1425 /* start to update checkpoint, cp ver is already updated previously */ 1426 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true)); 1427 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi)); 1428 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) { 1429 ckpt->cur_node_segno[i] = 1430 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE)); 1431 ckpt->cur_node_blkoff[i] = 1432 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE)); 1433 ckpt->alloc_type[i + CURSEG_HOT_NODE] = 1434 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE); 1435 } 1436 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) { 1437 ckpt->cur_data_segno[i] = 1438 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA)); 1439 ckpt->cur_data_blkoff[i] = 1440 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA)); 1441 ckpt->alloc_type[i + CURSEG_HOT_DATA] = 1442 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA); 1443 } 1444 1445 /* 2 cp + n data seg summary + orphan inode blocks */ 1446 data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false); 1447 spin_lock_irqsave(&sbi->cp_lock, flags); 1448 if (data_sum_blocks < NR_CURSEG_DATA_TYPE) 1449 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); 1450 else 1451 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); 1452 spin_unlock_irqrestore(&sbi->cp_lock, flags); 1453 1454 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num); 1455 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks + 1456 orphan_blocks); 1457 1458 if (__remain_node_summaries(cpc->reason)) 1459 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+ 1460 cp_payload_blks + data_sum_blocks + 1461 orphan_blocks + NR_CURSEG_NODE_TYPE); 1462 else 1463 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS + 1464 cp_payload_blks + data_sum_blocks + 1465 orphan_blocks); 1466 1467 /* update ckpt flag for checkpoint */ 1468 update_ckpt_flags(sbi, cpc); 1469 1470 /* update SIT/NAT bitmap */ 1471 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP)); 1472 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP)); 1473 1474 crc32 = f2fs_checkpoint_chksum(sbi, ckpt); 1475 *((__le32 *)((unsigned char *)ckpt + 1476 le32_to_cpu(ckpt->checksum_offset))) 1477 = cpu_to_le32(crc32); 1478 1479 start_blk = __start_cp_next_addr(sbi); 1480 1481 /* write nat bits */ 1482 if (enabled_nat_bits(sbi, cpc)) { 1483 __u64 cp_ver = cur_cp_version(ckpt); 1484 block_t blk; 1485 1486 cp_ver |= ((__u64)crc32 << 32); 1487 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver); 1488 1489 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks; 1490 for (i = 0; i < nm_i->nat_bits_blocks; i++) 1491 f2fs_update_meta_page(sbi, nm_i->nat_bits + 1492 (i << F2FS_BLKSIZE_BITS), blk + i); 1493 } 1494 1495 /* write out checkpoint buffer at block 0 */ 1496 f2fs_update_meta_page(sbi, ckpt, start_blk++); 1497 1498 for (i = 1; i < 1 + cp_payload_blks; i++) 1499 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE, 1500 start_blk++); 1501 1502 if (orphan_num) { 1503 write_orphan_inodes(sbi, start_blk); 1504 start_blk += orphan_blocks; 1505 } 1506 1507 f2fs_write_data_summaries(sbi, start_blk); 1508 start_blk += data_sum_blocks; 1509 1510 /* Record write statistics in the hot node summary */ 1511 kbytes_written = sbi->kbytes_written; 1512 kbytes_written += (f2fs_get_sectors_written(sbi) - 1513 sbi->sectors_written_start) >> 1; 1514 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written); 1515 1516 if (__remain_node_summaries(cpc->reason)) { 1517 f2fs_write_node_summaries(sbi, start_blk); 1518 start_blk += NR_CURSEG_NODE_TYPE; 1519 } 1520 1521 /* update user_block_counts */ 1522 sbi->last_valid_block_count = sbi->total_valid_block_count; 1523 percpu_counter_set(&sbi->alloc_valid_block_count, 0); 1524 1525 /* Here, we have one bio having CP pack except cp pack 2 page */ 1526 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO); 1527 /* Wait for all dirty meta pages to be submitted for IO */ 1528 f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META); 1529 1530 /* wait for previous submitted meta pages writeback */ 1531 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA); 1532 1533 /* flush all device cache */ 1534 err = f2fs_flush_device_cache(sbi); 1535 if (err) 1536 return err; 1537 1538 /* barrier and flush checkpoint cp pack 2 page if it can */ 1539 commit_checkpoint(sbi, ckpt, start_blk); 1540 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA); 1541 1542 /* 1543 * invalidate intermediate page cache borrowed from meta inode which are 1544 * used for migration of encrypted, verity or compressed inode's blocks. 1545 */ 1546 if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) || 1547 f2fs_sb_has_compression(sbi)) 1548 invalidate_mapping_pages(META_MAPPING(sbi), 1549 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1); 1550 1551 f2fs_release_ino_entry(sbi, false); 1552 1553 f2fs_reset_fsync_node_info(sbi); 1554 1555 clear_sbi_flag(sbi, SBI_IS_DIRTY); 1556 clear_sbi_flag(sbi, SBI_NEED_CP); 1557 clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH); 1558 1559 spin_lock(&sbi->stat_lock); 1560 sbi->unusable_block_count = 0; 1561 spin_unlock(&sbi->stat_lock); 1562 1563 __set_cp_next_pack(sbi); 1564 1565 /* 1566 * redirty superblock if metadata like node page or inode cache is 1567 * updated during writing checkpoint. 1568 */ 1569 if (get_pages(sbi, F2FS_DIRTY_NODES) || 1570 get_pages(sbi, F2FS_DIRTY_IMETA)) 1571 set_sbi_flag(sbi, SBI_IS_DIRTY); 1572 1573 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS)); 1574 1575 return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0; 1576 } 1577 1578 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1579 { 1580 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1581 unsigned long long ckpt_ver; 1582 int err = 0; 1583 1584 if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi)) 1585 return -EROFS; 1586 1587 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 1588 if (cpc->reason != CP_PAUSE) 1589 return 0; 1590 f2fs_warn(sbi, "Start checkpoint disabled!"); 1591 } 1592 if (cpc->reason != CP_RESIZE) 1593 down_write(&sbi->cp_global_sem); 1594 1595 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) && 1596 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) || 1597 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks))) 1598 goto out; 1599 if (unlikely(f2fs_cp_error(sbi))) { 1600 err = -EIO; 1601 goto out; 1602 } 1603 1604 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops"); 1605 1606 err = block_operations(sbi); 1607 if (err) 1608 goto out; 1609 1610 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops"); 1611 1612 f2fs_flush_merged_writes(sbi); 1613 1614 /* this is the case of multiple fstrims without any changes */ 1615 if (cpc->reason & CP_DISCARD) { 1616 if (!f2fs_exist_trim_candidates(sbi, cpc)) { 1617 unblock_operations(sbi); 1618 goto out; 1619 } 1620 1621 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 && 1622 SIT_I(sbi)->dirty_sentries == 0 && 1623 prefree_segments(sbi) == 0) { 1624 f2fs_flush_sit_entries(sbi, cpc); 1625 f2fs_clear_prefree_segments(sbi, cpc); 1626 unblock_operations(sbi); 1627 goto out; 1628 } 1629 } 1630 1631 /* 1632 * update checkpoint pack index 1633 * Increase the version number so that 1634 * SIT entries and seg summaries are written at correct place 1635 */ 1636 ckpt_ver = cur_cp_version(ckpt); 1637 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver); 1638 1639 /* write cached NAT/SIT entries to NAT/SIT area */ 1640 err = f2fs_flush_nat_entries(sbi, cpc); 1641 if (err) 1642 goto stop; 1643 1644 f2fs_flush_sit_entries(sbi, cpc); 1645 1646 /* save inmem log status */ 1647 f2fs_save_inmem_curseg(sbi); 1648 1649 err = do_checkpoint(sbi, cpc); 1650 if (err) 1651 f2fs_release_discard_addrs(sbi); 1652 else 1653 f2fs_clear_prefree_segments(sbi, cpc); 1654 1655 f2fs_restore_inmem_curseg(sbi); 1656 stop: 1657 unblock_operations(sbi); 1658 stat_inc_cp_count(sbi->stat_info); 1659 1660 if (cpc->reason & CP_RECOVERY) 1661 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver); 1662 1663 /* update CP_TIME to trigger checkpoint periodically */ 1664 f2fs_update_time(sbi, CP_TIME); 1665 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint"); 1666 out: 1667 if (cpc->reason != CP_RESIZE) 1668 up_write(&sbi->cp_global_sem); 1669 return err; 1670 } 1671 1672 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi) 1673 { 1674 int i; 1675 1676 for (i = 0; i < MAX_INO_ENTRY; i++) { 1677 struct inode_management *im = &sbi->im[i]; 1678 1679 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC); 1680 spin_lock_init(&im->ino_lock); 1681 INIT_LIST_HEAD(&im->ino_list); 1682 im->ino_num = 0; 1683 } 1684 1685 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS - 1686 NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) * 1687 F2FS_ORPHANS_PER_BLOCK; 1688 } 1689 1690 int __init f2fs_create_checkpoint_caches(void) 1691 { 1692 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry", 1693 sizeof(struct ino_entry)); 1694 if (!ino_entry_slab) 1695 return -ENOMEM; 1696 f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry", 1697 sizeof(struct inode_entry)); 1698 if (!f2fs_inode_entry_slab) { 1699 kmem_cache_destroy(ino_entry_slab); 1700 return -ENOMEM; 1701 } 1702 return 0; 1703 } 1704 1705 void f2fs_destroy_checkpoint_caches(void) 1706 { 1707 kmem_cache_destroy(ino_entry_slab); 1708 kmem_cache_destroy(f2fs_inode_entry_slab); 1709 } 1710 1711 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi) 1712 { 1713 struct cp_control cpc = { .reason = CP_SYNC, }; 1714 int err; 1715 1716 down_write(&sbi->gc_lock); 1717 err = f2fs_write_checkpoint(sbi, &cpc); 1718 up_write(&sbi->gc_lock); 1719 1720 return err; 1721 } 1722 1723 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi) 1724 { 1725 struct ckpt_req_control *cprc = &sbi->cprc_info; 1726 struct ckpt_req *req, *next; 1727 struct llist_node *dispatch_list; 1728 u64 sum_diff = 0, diff, count = 0; 1729 int ret; 1730 1731 dispatch_list = llist_del_all(&cprc->issue_list); 1732 if (!dispatch_list) 1733 return; 1734 dispatch_list = llist_reverse_order(dispatch_list); 1735 1736 ret = __write_checkpoint_sync(sbi); 1737 atomic_inc(&cprc->issued_ckpt); 1738 1739 llist_for_each_entry_safe(req, next, dispatch_list, llnode) { 1740 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time); 1741 req->ret = ret; 1742 complete(&req->wait); 1743 1744 sum_diff += diff; 1745 count++; 1746 } 1747 atomic_sub(count, &cprc->queued_ckpt); 1748 atomic_add(count, &cprc->total_ckpt); 1749 1750 spin_lock(&cprc->stat_lock); 1751 cprc->cur_time = (unsigned int)div64_u64(sum_diff, count); 1752 if (cprc->peak_time < cprc->cur_time) 1753 cprc->peak_time = cprc->cur_time; 1754 spin_unlock(&cprc->stat_lock); 1755 } 1756 1757 static int issue_checkpoint_thread(void *data) 1758 { 1759 struct f2fs_sb_info *sbi = data; 1760 struct ckpt_req_control *cprc = &sbi->cprc_info; 1761 wait_queue_head_t *q = &cprc->ckpt_wait_queue; 1762 repeat: 1763 if (kthread_should_stop()) 1764 return 0; 1765 1766 if (!llist_empty(&cprc->issue_list)) 1767 __checkpoint_and_complete_reqs(sbi); 1768 1769 wait_event_interruptible(*q, 1770 kthread_should_stop() || !llist_empty(&cprc->issue_list)); 1771 goto repeat; 1772 } 1773 1774 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi, 1775 struct ckpt_req *wait_req) 1776 { 1777 struct ckpt_req_control *cprc = &sbi->cprc_info; 1778 1779 if (!llist_empty(&cprc->issue_list)) { 1780 __checkpoint_and_complete_reqs(sbi); 1781 } else { 1782 /* already dispatched by issue_checkpoint_thread */ 1783 if (wait_req) 1784 wait_for_completion(&wait_req->wait); 1785 } 1786 } 1787 1788 static void init_ckpt_req(struct ckpt_req *req) 1789 { 1790 memset(req, 0, sizeof(struct ckpt_req)); 1791 1792 init_completion(&req->wait); 1793 req->queue_time = ktime_get(); 1794 } 1795 1796 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi) 1797 { 1798 struct ckpt_req_control *cprc = &sbi->cprc_info; 1799 struct ckpt_req req; 1800 struct cp_control cpc; 1801 1802 cpc.reason = __get_cp_reason(sbi); 1803 if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) { 1804 int ret; 1805 1806 down_write(&sbi->gc_lock); 1807 ret = f2fs_write_checkpoint(sbi, &cpc); 1808 up_write(&sbi->gc_lock); 1809 1810 return ret; 1811 } 1812 1813 if (!cprc->f2fs_issue_ckpt) 1814 return __write_checkpoint_sync(sbi); 1815 1816 init_ckpt_req(&req); 1817 1818 llist_add(&req.llnode, &cprc->issue_list); 1819 atomic_inc(&cprc->queued_ckpt); 1820 1821 /* update issue_list before we wake up issue_checkpoint thread */ 1822 smp_mb(); 1823 1824 if (waitqueue_active(&cprc->ckpt_wait_queue)) 1825 wake_up(&cprc->ckpt_wait_queue); 1826 1827 if (cprc->f2fs_issue_ckpt) 1828 wait_for_completion(&req.wait); 1829 else 1830 flush_remained_ckpt_reqs(sbi, &req); 1831 1832 return req.ret; 1833 } 1834 1835 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi) 1836 { 1837 dev_t dev = sbi->sb->s_bdev->bd_dev; 1838 struct ckpt_req_control *cprc = &sbi->cprc_info; 1839 1840 if (cprc->f2fs_issue_ckpt) 1841 return 0; 1842 1843 cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi, 1844 "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev)); 1845 if (IS_ERR(cprc->f2fs_issue_ckpt)) { 1846 cprc->f2fs_issue_ckpt = NULL; 1847 return -ENOMEM; 1848 } 1849 1850 set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio); 1851 1852 return 0; 1853 } 1854 1855 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi) 1856 { 1857 struct ckpt_req_control *cprc = &sbi->cprc_info; 1858 1859 if (cprc->f2fs_issue_ckpt) { 1860 struct task_struct *ckpt_task = cprc->f2fs_issue_ckpt; 1861 1862 cprc->f2fs_issue_ckpt = NULL; 1863 kthread_stop(ckpt_task); 1864 1865 flush_remained_ckpt_reqs(sbi, NULL); 1866 } 1867 } 1868 1869 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi) 1870 { 1871 struct ckpt_req_control *cprc = &sbi->cprc_info; 1872 1873 atomic_set(&cprc->issued_ckpt, 0); 1874 atomic_set(&cprc->total_ckpt, 0); 1875 atomic_set(&cprc->queued_ckpt, 0); 1876 cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO; 1877 init_waitqueue_head(&cprc->ckpt_wait_queue); 1878 init_llist_head(&cprc->issue_list); 1879 spin_lock_init(&cprc->stat_lock); 1880 } 1881