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/folio_batch.h> 15 #include <linux/swap.h> 16 #include <linux/kthread.h> 17 #include <linux/delayacct.h> 18 #include <linux/ioprio.h> 19 #include <linux/math64.h> 20 21 #include "f2fs.h" 22 #include "node.h" 23 #include "segment.h" 24 #include "iostat.h" 25 #include <trace/events/f2fs.h> 26 27 static inline void get_lock_elapsed_time(struct f2fs_time_stat *ts) 28 { 29 ts->total_time = ktime_get(); 30 #ifdef CONFIG_64BIT 31 ts->running_time = current->se.sum_exec_runtime; 32 #endif 33 #if defined(CONFIG_SCHED_INFO) && defined(CONFIG_SCHEDSTATS) 34 ts->runnable_time = current->sched_info.run_delay; 35 #endif 36 #ifdef CONFIG_TASK_DELAY_ACCT 37 if (current->delays) 38 ts->io_sleep_time = current->delays->blkio_delay; 39 #endif 40 } 41 42 static inline void trace_lock_elapsed_time_start(struct f2fs_rwsem *sem, 43 struct f2fs_lock_context *lc) 44 { 45 lc->lock_trace = trace_f2fs_lock_elapsed_time_enabled(); 46 if (!lc->lock_trace) 47 return; 48 49 get_lock_elapsed_time(&lc->ts); 50 } 51 52 static inline void trace_lock_elapsed_time_end(struct f2fs_rwsem *sem, 53 struct f2fs_lock_context *lc, bool is_write) 54 { 55 struct f2fs_time_stat tts; 56 unsigned long long total_time; 57 unsigned long long running_time = 0; 58 unsigned long long runnable_time = 0; 59 unsigned long long io_sleep_time = 0; 60 unsigned long long other_time = 0; 61 unsigned npm = NSEC_PER_MSEC; 62 63 if (!lc->lock_trace) 64 return; 65 66 if (time_to_inject(sem->sbi, FAULT_LOCK_TIMEOUT)) 67 f2fs_schedule_timeout_killable(DEFAULT_FAULT_TIMEOUT, true); 68 69 get_lock_elapsed_time(&tts); 70 71 total_time = div_u64(tts.total_time - lc->ts.total_time, npm); 72 if (total_time <= sem->sbi->max_lock_elapsed_time) 73 return; 74 75 #ifdef CONFIG_64BIT 76 running_time = div_u64(tts.running_time - lc->ts.running_time, npm); 77 #endif 78 #if defined(CONFIG_SCHED_INFO) && defined(CONFIG_SCHEDSTATS) 79 runnable_time = div_u64(tts.runnable_time - lc->ts.runnable_time, npm); 80 #endif 81 #ifdef CONFIG_TASK_DELAY_ACCT 82 io_sleep_time = div_u64(tts.io_sleep_time - lc->ts.io_sleep_time, npm); 83 #endif 84 if (total_time > running_time + io_sleep_time + runnable_time) 85 other_time = total_time - running_time - 86 io_sleep_time - runnable_time; 87 88 trace_f2fs_lock_elapsed_time(sem->sbi, sem->name, is_write, current, 89 get_current_ioprio(), total_time, running_time, 90 runnable_time, io_sleep_time, other_time); 91 } 92 93 static bool need_uplift_priority(struct f2fs_rwsem *sem, bool is_write) 94 { 95 if (!(sem->sbi->adjust_lock_priority & BIT(sem->name - 1))) 96 return false; 97 98 switch (sem->name) { 99 /* 100 * writer is checkpoint which has high priority, let's just uplift 101 * priority for reader 102 */ 103 case LOCK_NAME_CP_RWSEM: 104 case LOCK_NAME_NODE_CHANGE: 105 case LOCK_NAME_NODE_WRITE: 106 return !is_write; 107 case LOCK_NAME_GC_LOCK: 108 case LOCK_NAME_CP_GLOBAL: 109 case LOCK_NAME_IO_RWSEM: 110 case LOCK_NAME_NAT_TREE_LOCK: 111 return true; 112 default: 113 f2fs_bug_on(sem->sbi, 1); 114 } 115 return false; 116 } 117 118 static void uplift_priority(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc, 119 bool is_write) 120 { 121 lc->need_restore = false; 122 if (!sem->sbi->adjust_lock_priority) 123 return; 124 if (rt_task(current)) 125 return; 126 if (!need_uplift_priority(sem, is_write)) 127 return; 128 lc->orig_nice = task_nice(current); 129 lc->new_nice = PRIO_TO_NICE(sem->sbi->lock_duration_priority); 130 if (lc->orig_nice <= lc->new_nice) 131 return; 132 set_user_nice(current, lc->new_nice); 133 lc->need_restore = true; 134 135 trace_f2fs_priority_uplift(sem->sbi, sem->name, is_write, current, 136 NICE_TO_PRIO(lc->orig_nice), NICE_TO_PRIO(lc->new_nice)); 137 } 138 139 static void restore_priority(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc, 140 bool is_write) 141 { 142 if (!lc->need_restore) 143 return; 144 /* someone has updated the priority */ 145 if (task_nice(current) != lc->new_nice) 146 return; 147 set_user_nice(current, lc->orig_nice); 148 149 trace_f2fs_priority_restore(sem->sbi, sem->name, is_write, current, 150 NICE_TO_PRIO(lc->orig_nice), NICE_TO_PRIO(lc->new_nice)); 151 } 152 153 void f2fs_down_read_trace(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc) 154 { 155 uplift_priority(sem, lc, false); 156 f2fs_down_read(sem); 157 trace_lock_elapsed_time_start(sem, lc); 158 } 159 160 int f2fs_down_read_trylock_trace(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc) 161 { 162 uplift_priority(sem, lc, false); 163 if (!f2fs_down_read_trylock(sem)) { 164 restore_priority(sem, lc, false); 165 return 0; 166 } 167 trace_lock_elapsed_time_start(sem, lc); 168 return 1; 169 } 170 171 void f2fs_up_read_trace(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc) 172 { 173 f2fs_up_read(sem); 174 restore_priority(sem, lc, false); 175 trace_lock_elapsed_time_end(sem, lc, false); 176 } 177 178 void f2fs_down_write_trace(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc) 179 { 180 uplift_priority(sem, lc, true); 181 f2fs_down_write(sem); 182 trace_lock_elapsed_time_start(sem, lc); 183 } 184 185 int f2fs_down_write_trylock_trace(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc) 186 { 187 uplift_priority(sem, lc, true); 188 if (!f2fs_down_write_trylock(sem)) { 189 restore_priority(sem, lc, true); 190 return 0; 191 } 192 trace_lock_elapsed_time_start(sem, lc); 193 return 1; 194 } 195 196 void f2fs_up_write_trace(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc) 197 { 198 f2fs_up_write(sem); 199 restore_priority(sem, lc, true); 200 trace_lock_elapsed_time_end(sem, lc, true); 201 } 202 203 void f2fs_lock_op(struct f2fs_sb_info *sbi, struct f2fs_lock_context *lc) 204 { 205 f2fs_down_read_trace(&sbi->cp_rwsem, lc); 206 } 207 208 int f2fs_trylock_op(struct f2fs_sb_info *sbi, struct f2fs_lock_context *lc) 209 { 210 if (time_to_inject(sbi, FAULT_LOCK_OP)) 211 return 0; 212 213 return f2fs_down_read_trylock_trace(&sbi->cp_rwsem, lc); 214 } 215 216 void f2fs_unlock_op(struct f2fs_sb_info *sbi, struct f2fs_lock_context *lc) 217 { 218 f2fs_up_read_trace(&sbi->cp_rwsem, lc); 219 } 220 221 static inline void f2fs_lock_all(struct f2fs_sb_info *sbi) 222 { 223 f2fs_down_write(&sbi->cp_rwsem); 224 } 225 226 static inline void f2fs_unlock_all(struct f2fs_sb_info *sbi) 227 { 228 f2fs_up_write(&sbi->cp_rwsem); 229 } 230 231 #define DEFAULT_CHECKPOINT_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_RT, 3)) 232 233 static struct kmem_cache *ino_entry_slab; 234 struct kmem_cache *f2fs_inode_entry_slab; 235 236 /* 237 * We guarantee no failure on the returned page. 238 */ 239 struct folio *f2fs_grab_meta_folio(struct f2fs_sb_info *sbi, pgoff_t index) 240 { 241 struct address_space *mapping = META_MAPPING(sbi); 242 struct folio *folio; 243 repeat: 244 folio = f2fs_grab_cache_folio(mapping, index, false); 245 if (IS_ERR(folio)) { 246 cond_resched(); 247 goto repeat; 248 } 249 f2fs_folio_wait_writeback(folio, META, true, true); 250 if (!folio_test_uptodate(folio)) 251 folio_mark_uptodate(folio); 252 return folio; 253 } 254 255 static struct folio *__get_meta_folio(struct f2fs_sb_info *sbi, pgoff_t index, 256 bool is_meta) 257 { 258 struct address_space *mapping = META_MAPPING(sbi); 259 struct folio *folio; 260 struct f2fs_io_info fio = { 261 .sbi = sbi, 262 .type = META, 263 .op = REQ_OP_READ, 264 .op_flags = REQ_META | REQ_PRIO, 265 .old_blkaddr = index, 266 .new_blkaddr = index, 267 .encrypted_page = NULL, 268 .is_por = !is_meta ? 1 : 0, 269 }; 270 int err; 271 272 if (unlikely(!is_meta)) 273 fio.op_flags &= ~REQ_META; 274 repeat: 275 folio = f2fs_grab_cache_folio(mapping, index, false); 276 if (IS_ERR(folio)) { 277 cond_resched(); 278 goto repeat; 279 } 280 if (folio_test_uptodate(folio)) 281 goto out; 282 283 fio.folio = folio; 284 285 err = f2fs_submit_page_bio(&fio); 286 if (err) { 287 f2fs_folio_put(folio, true); 288 return ERR_PTR(err); 289 } 290 291 f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, F2FS_BLKSIZE); 292 293 folio_lock(folio); 294 if (unlikely(!is_meta_folio(folio))) { 295 f2fs_folio_put(folio, true); 296 goto repeat; 297 } 298 299 if (unlikely(!folio_test_uptodate(folio))) { 300 f2fs_handle_page_eio(sbi, folio, META); 301 f2fs_folio_put(folio, true); 302 return ERR_PTR(-EIO); 303 } 304 out: 305 return folio; 306 } 307 308 struct folio *f2fs_get_meta_folio(struct f2fs_sb_info *sbi, pgoff_t index) 309 { 310 return __get_meta_folio(sbi, index, true); 311 } 312 313 struct folio *f2fs_get_meta_folio_retry(struct f2fs_sb_info *sbi, pgoff_t index) 314 { 315 struct folio *folio; 316 int count = 0; 317 318 retry: 319 folio = __get_meta_folio(sbi, index, true); 320 if (IS_ERR(folio)) { 321 if (PTR_ERR(folio) == -EIO && 322 ++count <= DEFAULT_RETRY_IO_COUNT) 323 goto retry; 324 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_META_PAGE); 325 } 326 return folio; 327 } 328 329 /* for POR only */ 330 struct folio *f2fs_get_tmp_folio(struct f2fs_sb_info *sbi, pgoff_t index) 331 { 332 return __get_meta_folio(sbi, index, false); 333 } 334 335 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr, 336 int type) 337 { 338 struct seg_entry *se; 339 unsigned int segno, offset; 340 bool exist; 341 342 if (type == DATA_GENERIC) 343 return true; 344 345 segno = GET_SEGNO(sbi, blkaddr); 346 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 347 se = get_seg_entry(sbi, segno); 348 349 exist = f2fs_test_bit(offset, se->cur_valid_map); 350 351 /* skip data, if we already have an error in checkpoint. */ 352 if (unlikely(f2fs_cp_error(sbi))) 353 return exist; 354 355 if ((exist && type == DATA_GENERIC_ENHANCE_UPDATE) || 356 (!exist && type == DATA_GENERIC_ENHANCE)) 357 goto out_err; 358 if (!exist && type != DATA_GENERIC_ENHANCE_UPDATE) 359 goto out_handle; 360 return exist; 361 362 out_err: 363 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d", 364 blkaddr, exist); 365 set_sbi_flag(sbi, SBI_NEED_FSCK); 366 dump_stack(); 367 out_handle: 368 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); 369 return exist; 370 } 371 372 static bool __f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi, 373 block_t blkaddr, int type) 374 { 375 switch (type) { 376 case META_NAT: 377 break; 378 case META_SIT: 379 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi))) 380 goto check_only; 381 break; 382 case META_SSA: 383 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) || 384 blkaddr < SM_I(sbi)->ssa_blkaddr)) 385 goto check_only; 386 break; 387 case META_CP: 388 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr || 389 blkaddr < __start_cp_addr(sbi))) 390 goto check_only; 391 break; 392 case META_POR: 393 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) || 394 blkaddr < MAIN_BLKADDR(sbi))) 395 goto check_only; 396 break; 397 case DATA_GENERIC: 398 case DATA_GENERIC_ENHANCE: 399 case DATA_GENERIC_ENHANCE_READ: 400 case DATA_GENERIC_ENHANCE_UPDATE: 401 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) || 402 blkaddr < MAIN_BLKADDR(sbi))) { 403 404 /* Skip to emit an error message. */ 405 if (unlikely(f2fs_cp_error(sbi))) 406 return false; 407 408 f2fs_warn(sbi, "access invalid blkaddr:%u", 409 blkaddr); 410 set_sbi_flag(sbi, SBI_NEED_FSCK); 411 dump_stack(); 412 goto err; 413 } else { 414 return __is_bitmap_valid(sbi, blkaddr, type); 415 } 416 break; 417 case META_GENERIC: 418 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) || 419 blkaddr >= MAIN_BLKADDR(sbi))) 420 goto err; 421 break; 422 default: 423 BUG(); 424 } 425 426 return true; 427 err: 428 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); 429 check_only: 430 return false; 431 } 432 433 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi, 434 block_t blkaddr, int type) 435 { 436 if (time_to_inject(sbi, FAULT_BLKADDR_VALIDITY)) 437 return false; 438 return __f2fs_is_valid_blkaddr(sbi, blkaddr, type); 439 } 440 441 bool f2fs_is_valid_blkaddr_raw(struct f2fs_sb_info *sbi, 442 block_t blkaddr, int type) 443 { 444 return __f2fs_is_valid_blkaddr(sbi, blkaddr, type); 445 } 446 447 /* 448 * Readahead CP/NAT/SIT/SSA/POR pages 449 */ 450 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages, 451 int type, bool sync) 452 { 453 block_t blkno = start; 454 struct f2fs_io_info fio = { 455 .sbi = sbi, 456 .type = META, 457 .op = REQ_OP_READ, 458 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD, 459 .encrypted_page = NULL, 460 .in_list = 0, 461 .is_por = (type == META_POR) ? 1 : 0, 462 }; 463 struct blk_plug plug; 464 int err; 465 466 if (unlikely(type == META_POR)) 467 fio.op_flags &= ~REQ_META; 468 469 blk_start_plug(&plug); 470 for (; nrpages-- > 0; blkno++) { 471 struct folio *folio; 472 473 if (!f2fs_is_valid_blkaddr(sbi, blkno, type)) 474 goto out; 475 476 switch (type) { 477 case META_NAT: 478 if (unlikely(blkno >= 479 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid))) 480 blkno = 0; 481 /* get nat block addr */ 482 fio.new_blkaddr = current_nat_addr(sbi, 483 blkno * NAT_ENTRY_PER_BLOCK); 484 break; 485 case META_SIT: 486 if (unlikely(blkno >= TOTAL_SEGS(sbi))) 487 goto out; 488 /* get sit block addr */ 489 fio.new_blkaddr = current_sit_addr(sbi, 490 blkno * SIT_ENTRY_PER_BLOCK); 491 break; 492 case META_SSA: 493 case META_CP: 494 case META_POR: 495 fio.new_blkaddr = blkno; 496 break; 497 default: 498 BUG(); 499 } 500 501 folio = f2fs_grab_cache_folio(META_MAPPING(sbi), 502 fio.new_blkaddr, false); 503 if (IS_ERR(folio)) 504 continue; 505 if (folio_test_uptodate(folio)) { 506 f2fs_folio_put(folio, true); 507 continue; 508 } 509 510 fio.folio = folio; 511 err = f2fs_submit_page_bio(&fio); 512 f2fs_folio_put(folio, err ? true : false); 513 514 if (!err) 515 f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, 516 F2FS_BLKSIZE); 517 } 518 out: 519 blk_finish_plug(&plug); 520 return blkno - start; 521 } 522 523 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index, 524 unsigned int ra_blocks) 525 { 526 struct folio *folio; 527 bool readahead = false; 528 529 if (ra_blocks == RECOVERY_MIN_RA_BLOCKS) 530 return; 531 532 folio = filemap_get_folio(META_MAPPING(sbi), index); 533 if (IS_ERR(folio) || !folio_test_uptodate(folio)) 534 readahead = true; 535 f2fs_folio_put(folio, false); 536 537 if (readahead) 538 f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true); 539 } 540 541 static bool __f2fs_write_meta_folio(struct folio *folio, 542 struct writeback_control *wbc, 543 enum iostat_type io_type) 544 { 545 struct f2fs_sb_info *sbi = F2FS_F_SB(folio); 546 547 trace_f2fs_writepage(folio, META); 548 549 if (unlikely(f2fs_cp_error(sbi))) { 550 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) { 551 folio_clear_uptodate(folio); 552 dec_page_count(sbi, F2FS_DIRTY_META); 553 folio_unlock(folio); 554 return true; 555 } 556 goto redirty_out; 557 } 558 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 559 goto redirty_out; 560 561 f2fs_do_write_meta_page(sbi, folio, io_type); 562 dec_page_count(sbi, F2FS_DIRTY_META); 563 564 folio_unlock(folio); 565 566 if (unlikely(f2fs_cp_error(sbi))) 567 f2fs_submit_merged_write(sbi, META); 568 569 return true; 570 571 redirty_out: 572 folio_redirty_for_writepage(wbc, folio); 573 return false; 574 } 575 576 static int f2fs_write_meta_pages(struct address_space *mapping, 577 struct writeback_control *wbc) 578 { 579 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 580 struct f2fs_lock_context lc; 581 long diff, written; 582 583 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 584 goto skip_write; 585 586 /* collect a number of dirty meta pages and write together */ 587 if (wbc->sync_mode != WB_SYNC_ALL && 588 get_pages(sbi, F2FS_DIRTY_META) < 589 nr_pages_to_skip(sbi, META)) 590 goto skip_write; 591 592 /* if locked failed, cp will flush dirty pages instead */ 593 if (!f2fs_down_write_trylock_trace(&sbi->cp_global_sem, &lc)) 594 goto skip_write; 595 596 trace_f2fs_writepages(mapping->host, wbc, META); 597 diff = nr_pages_to_write(sbi, META, wbc); 598 written = f2fs_sync_meta_pages(sbi, wbc->nr_to_write, FS_META_IO); 599 f2fs_up_write_trace(&sbi->cp_global_sem, &lc); 600 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff); 601 return 0; 602 603 skip_write: 604 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META); 605 trace_f2fs_writepages(mapping->host, wbc, META); 606 return 0; 607 } 608 609 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, long nr_to_write, 610 enum iostat_type io_type) 611 { 612 struct address_space *mapping = META_MAPPING(sbi); 613 pgoff_t index = 0, prev = ULONG_MAX; 614 struct folio_batch fbatch; 615 long nwritten = 0; 616 int nr_folios; 617 struct writeback_control wbc = {}; 618 struct blk_plug plug; 619 620 folio_batch_init(&fbatch); 621 622 blk_start_plug(&plug); 623 624 while ((nr_folios = filemap_get_folios_tag(mapping, &index, 625 (pgoff_t)-1, 626 PAGECACHE_TAG_DIRTY, &fbatch))) { 627 int i; 628 629 for (i = 0; i < nr_folios; i++) { 630 struct folio *folio = fbatch.folios[i]; 631 632 if (nr_to_write != LONG_MAX && i != 0 && 633 folio->index != prev + 634 folio_nr_pages(fbatch.folios[i-1])) { 635 folio_batch_release(&fbatch); 636 goto stop; 637 } 638 639 folio_lock(folio); 640 641 if (unlikely(!is_meta_folio(folio))) { 642 continue_unlock: 643 folio_unlock(folio); 644 continue; 645 } 646 if (!folio_test_dirty(folio)) { 647 /* someone wrote it for us */ 648 goto continue_unlock; 649 } 650 651 f2fs_folio_wait_writeback(folio, META, true, true); 652 653 if (!folio_clear_dirty_for_io(folio)) 654 goto continue_unlock; 655 656 if (!__f2fs_write_meta_folio(folio, &wbc, 657 io_type)) { 658 folio_unlock(folio); 659 break; 660 } 661 nwritten += folio_nr_pages(folio); 662 prev = folio->index; 663 if (unlikely(nwritten >= nr_to_write)) 664 break; 665 } 666 folio_batch_release(&fbatch); 667 cond_resched(); 668 } 669 stop: 670 if (nwritten) 671 f2fs_submit_merged_write(sbi, META); 672 673 blk_finish_plug(&plug); 674 675 return nwritten; 676 } 677 678 static bool f2fs_dirty_meta_folio(struct address_space *mapping, 679 struct folio *folio) 680 { 681 trace_f2fs_set_page_dirty(folio, META); 682 683 if (!folio_test_uptodate(folio)) 684 folio_mark_uptodate(folio); 685 if (filemap_dirty_folio(mapping, folio)) { 686 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META); 687 folio_set_f2fs_reference(folio); 688 return true; 689 } 690 return false; 691 } 692 693 const struct address_space_operations f2fs_meta_aops = { 694 .writepages = f2fs_write_meta_pages, 695 .dirty_folio = f2fs_dirty_meta_folio, 696 .invalidate_folio = f2fs_invalidate_folio, 697 .release_folio = f2fs_release_folio, 698 .migrate_folio = filemap_migrate_folio, 699 }; 700 701 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, 702 unsigned int devidx, int type) 703 { 704 struct inode_management *im = &sbi->im[type]; 705 struct ino_entry *e = NULL, *new = NULL; 706 int ret; 707 708 if (type == FLUSH_INO) { 709 rcu_read_lock(); 710 e = radix_tree_lookup(&im->ino_root, ino); 711 rcu_read_unlock(); 712 } 713 714 retry: 715 if (!e) 716 new = f2fs_kmem_cache_alloc(ino_entry_slab, 717 GFP_NOFS, true, NULL); 718 719 ret = radix_tree_preload(GFP_NOFS | __GFP_NOFAIL); 720 f2fs_bug_on(sbi, ret); 721 722 spin_lock(&im->ino_lock); 723 e = radix_tree_lookup(&im->ino_root, ino); 724 if (!e) { 725 if (!new) { 726 spin_unlock(&im->ino_lock); 727 radix_tree_preload_end(); 728 goto retry; 729 } 730 e = new; 731 if (unlikely(radix_tree_insert(&im->ino_root, ino, e))) 732 f2fs_bug_on(sbi, 1); 733 734 memset(e, 0, sizeof(struct ino_entry)); 735 e->ino = ino; 736 737 list_add_tail(&e->list, &im->ino_list); 738 if (type != ORPHAN_INO) 739 im->ino_num++; 740 } 741 742 if (type == FLUSH_INO) 743 f2fs_set_bit(devidx, (char *)&e->dirty_device); 744 745 spin_unlock(&im->ino_lock); 746 radix_tree_preload_end(); 747 748 if (new && e != new) 749 kmem_cache_free(ino_entry_slab, new); 750 } 751 752 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) 753 { 754 struct inode_management *im = &sbi->im[type]; 755 struct ino_entry *e; 756 757 spin_lock(&im->ino_lock); 758 e = radix_tree_lookup(&im->ino_root, ino); 759 if (e) { 760 list_del(&e->list); 761 radix_tree_delete(&im->ino_root, ino); 762 im->ino_num--; 763 spin_unlock(&im->ino_lock); 764 kmem_cache_free(ino_entry_slab, e); 765 return; 766 } 767 spin_unlock(&im->ino_lock); 768 } 769 770 static void __set_ino_bitmap(struct f2fs_sb_info *sbi, nid_t ino, int type) 771 { 772 struct inode_management *im = &sbi->im[type]; 773 unsigned long index = INO_SLOT_INDEX(ino); 774 unsigned int ofs = INO_BIT_OFFSET(ino); 775 void **slot, *entry; 776 unsigned long bitmap = 0; 777 int ret; 778 779 ret = radix_tree_preload(GFP_NOFS | __GFP_NOFAIL); 780 f2fs_bug_on(sbi, ret); 781 782 spin_lock(&im->ino_lock); 783 slot = radix_tree_lookup_slot(&im->ino_root, index); 784 if (slot) { 785 entry = radix_tree_deref_slot_protected(slot, &im->ino_lock); 786 bitmap = xa_to_value(entry); 787 if (!(bitmap & (1UL << ofs))) { 788 bitmap |= (1UL << ofs); 789 entry = xa_mk_value(bitmap); 790 radix_tree_replace_slot(&im->ino_root, slot, entry); 791 } 792 } else { 793 bitmap |= (1UL << ofs); 794 entry = xa_mk_value(bitmap); 795 if (unlikely(radix_tree_insert(&im->ino_root, index, entry))) 796 f2fs_bug_on(sbi, 1); 797 } 798 spin_unlock(&im->ino_lock); 799 radix_tree_preload_end(); 800 } 801 802 static void __clear_ino_bitmap(struct f2fs_sb_info *sbi, nid_t ino, int type) 803 { 804 struct inode_management *im = &sbi->im[type]; 805 unsigned long index = INO_SLOT_INDEX(ino); 806 unsigned int ofs = INO_BIT_OFFSET(ino); 807 void **slot, *entry; 808 unsigned long bitmap; 809 810 spin_lock(&im->ino_lock); 811 slot = radix_tree_lookup_slot(&im->ino_root, index); 812 if (slot) { 813 entry = radix_tree_deref_slot_protected(slot, &im->ino_lock); 814 bitmap = xa_to_value(entry); 815 if (bitmap & (1UL << ofs)) 816 bitmap &= ~(1UL << ofs); 817 818 if (bitmap) { 819 entry = xa_mk_value(bitmap); 820 radix_tree_replace_slot(&im->ino_root, slot, entry); 821 } else { 822 radix_tree_delete(&im->ino_root, index); 823 } 824 } 825 spin_unlock(&im->ino_lock); 826 } 827 828 static void f2fs_wait_for_inode_record(struct f2fs_sb_info *sbi, int mode) 829 { 830 if (mode != APPEND_INO && mode != UPDATE_INO) 831 return; 832 833 /* Let's wait for some pending updates for APPEND_INO and UPDATE_INO. */ 834 flush_workqueue(sbi->evict_wq); 835 } 836 837 static void __f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, 838 unsigned int devidx, int type) 839 { 840 if (type <= FLUSH_INO) 841 __add_ino_entry(sbi, ino, devidx, type); 842 else 843 __set_ino_bitmap(sbi, ino, type); 844 } 845 846 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) 847 { 848 __f2fs_add_ino_entry(sbi, ino, 0, type); 849 } 850 851 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) 852 { 853 if (type <= FLUSH_INO) 854 __remove_ino_entry(sbi, ino, type); 855 else 856 __clear_ino_bitmap(sbi, ino, type); 857 } 858 859 /* mode should be APPEND_INO, UPDATE_INO, TRANS_DIR_INO and XATTR_DIR_INO */ 860 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode) 861 { 862 struct inode_management *im = &sbi->im[mode]; 863 unsigned long index = INO_SLOT_INDEX(ino); 864 unsigned int ofs = INO_BIT_OFFSET(ino); 865 void *entry; 866 unsigned long bitmap; 867 868 f2fs_bug_on(sbi, mode <= FLUSH_INO); 869 870 spin_lock(&im->ino_lock); 871 entry = radix_tree_lookup(&im->ino_root, index); 872 if (!entry) { 873 spin_unlock(&im->ino_lock); 874 return false; 875 } 876 bitmap = xa_to_value(entry); 877 spin_unlock(&im->ino_lock); 878 879 return bitmap & (1UL << ofs); 880 } 881 882 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all) 883 { 884 struct ino_entry *e, *tmp; 885 int i; 886 887 for (i = all ? ORPHAN_INO : FLUSH_INO; i <= FLUSH_INO; i++) { 888 struct inode_management *im = &sbi->im[i]; 889 890 f2fs_wait_for_inode_record(sbi, i); 891 892 spin_lock(&im->ino_lock); 893 list_for_each_entry_safe(e, tmp, &im->ino_list, list) { 894 list_del(&e->list); 895 radix_tree_delete(&im->ino_root, e->ino); 896 kmem_cache_free(ino_entry_slab, e); 897 im->ino_num--; 898 } 899 spin_unlock(&im->ino_lock); 900 } 901 902 for (i = APPEND_INO; i < MAX_INO_ENTRY; i++) { 903 struct inode_management *im = &sbi->im[i]; 904 905 spin_lock(&im->ino_lock); 906 xa_destroy(&im->ino_root); 907 spin_unlock(&im->ino_lock); 908 } 909 } 910 911 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino, 912 unsigned int devidx, int type) 913 { 914 __f2fs_add_ino_entry(sbi, ino, devidx, type); 915 } 916 917 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino, 918 unsigned int devidx, int type) 919 { 920 struct inode_management *im = &sbi->im[type]; 921 struct ino_entry *e; 922 bool is_dirty = false; 923 924 spin_lock(&im->ino_lock); 925 e = radix_tree_lookup(&im->ino_root, ino); 926 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device)) 927 is_dirty = true; 928 spin_unlock(&im->ino_lock); 929 return is_dirty; 930 } 931 932 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi) 933 { 934 struct inode_management *im = &sbi->im[ORPHAN_INO]; 935 int err = 0; 936 937 spin_lock(&im->ino_lock); 938 939 if (time_to_inject(sbi, FAULT_ORPHAN)) { 940 spin_unlock(&im->ino_lock); 941 return -ENOSPC; 942 } 943 944 if (unlikely(im->ino_num >= sbi->max_orphans)) 945 err = -ENOSPC; 946 else 947 im->ino_num++; 948 spin_unlock(&im->ino_lock); 949 950 return err; 951 } 952 953 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi) 954 { 955 struct inode_management *im = &sbi->im[ORPHAN_INO]; 956 957 spin_lock(&im->ino_lock); 958 f2fs_bug_on(sbi, im->ino_num == 0); 959 im->ino_num--; 960 spin_unlock(&im->ino_lock); 961 } 962 963 void f2fs_add_orphan_inode(struct inode *inode) 964 { 965 /* add new orphan ino entry into list */ 966 f2fs_add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO); 967 f2fs_update_inode_page(inode); 968 } 969 970 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) 971 { 972 /* remove orphan entry from orphan list */ 973 f2fs_remove_ino_entry(sbi, ino, ORPHAN_INO); 974 } 975 976 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) 977 { 978 struct inode *inode; 979 struct node_info ni; 980 int err; 981 982 inode = f2fs_iget_retry(sbi->sb, ino); 983 if (IS_ERR(inode)) { 984 /* 985 * there should be a bug that we can't find the entry 986 * to orphan inode. 987 */ 988 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT); 989 return PTR_ERR(inode); 990 } 991 992 err = f2fs_dquot_initialize(inode); 993 if (err) { 994 iput(inode); 995 goto err_out; 996 } 997 998 clear_nlink(inode); 999 1000 /* truncate all the data during iput */ 1001 iput(inode); 1002 1003 err = f2fs_get_node_info(sbi, ino, &ni, false); 1004 if (err) 1005 goto err_out; 1006 1007 /* ENOMEM was fully retried in f2fs_evict_inode. */ 1008 if (ni.blk_addr != NULL_ADDR) { 1009 err = -EIO; 1010 goto err_out; 1011 } 1012 return 0; 1013 1014 err_out: 1015 set_sbi_flag(sbi, SBI_NEED_FSCK); 1016 f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.", 1017 __func__, ino); 1018 return err; 1019 } 1020 1021 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi) 1022 { 1023 block_t start_blk, orphan_blocks, i, j; 1024 int err = 0; 1025 1026 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG)) 1027 return 0; 1028 1029 if (f2fs_hw_is_readonly(sbi)) { 1030 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup"); 1031 return 0; 1032 } 1033 1034 if (is_sbi_flag_set(sbi, SBI_IS_WRITABLE)) 1035 f2fs_info(sbi, "orphan cleanup on readonly fs"); 1036 1037 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi); 1038 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi); 1039 1040 f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true); 1041 1042 for (i = 0; i < orphan_blocks; i++) { 1043 struct folio *folio; 1044 struct f2fs_orphan_block *orphan_blk; 1045 unsigned int entry_count; 1046 1047 folio = f2fs_get_meta_folio(sbi, start_blk + i); 1048 if (IS_ERR(folio)) { 1049 err = PTR_ERR(folio); 1050 goto out; 1051 } 1052 1053 orphan_blk = folio_address(folio); 1054 entry_count = le32_to_cpu(orphan_blk->entry_count); 1055 if (entry_count > F2FS_ORPHANS_PER_BLOCK) { 1056 f2fs_err(sbi, "invalid orphan inode entry count %u", 1057 entry_count); 1058 set_sbi_flag(sbi, SBI_NEED_FSCK); 1059 f2fs_handle_error(sbi, ERROR_INCONSISTENT_ORPHAN); 1060 err = -EFSCORRUPTED; 1061 f2fs_folio_put(folio, true); 1062 goto out; 1063 } 1064 1065 for (j = 0; j < entry_count; j++) { 1066 nid_t ino = le32_to_cpu(orphan_blk->ino[j]); 1067 1068 err = recover_orphan_inode(sbi, ino); 1069 if (err) { 1070 f2fs_folio_put(folio, true); 1071 goto out; 1072 } 1073 } 1074 f2fs_folio_put(folio, true); 1075 } 1076 /* clear Orphan Flag */ 1077 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG); 1078 out: 1079 set_sbi_flag(sbi, SBI_IS_RECOVERED); 1080 1081 return err; 1082 } 1083 1084 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk) 1085 { 1086 struct list_head *head; 1087 struct f2fs_orphan_block *orphan_blk = NULL; 1088 unsigned int nentries = 0; 1089 unsigned short index = 1; 1090 unsigned short orphan_blocks; 1091 struct folio *folio = NULL; 1092 struct ino_entry *orphan = NULL; 1093 struct inode_management *im = &sbi->im[ORPHAN_INO]; 1094 1095 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num); 1096 1097 /* 1098 * we don't need to do spin_lock(&im->ino_lock) here, since all the 1099 * orphan inode operations are covered under f2fs_lock_op(). 1100 * And, spin_lock should be avoided due to page operations below. 1101 */ 1102 head = &im->ino_list; 1103 1104 /* loop for each orphan inode entry and write them in journal block */ 1105 list_for_each_entry(orphan, head, list) { 1106 if (!folio) { 1107 folio = f2fs_grab_meta_folio(sbi, start_blk++); 1108 orphan_blk = folio_address(folio); 1109 memset(orphan_blk, 0, sizeof(*orphan_blk)); 1110 } 1111 1112 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino); 1113 1114 if (nentries == F2FS_ORPHANS_PER_BLOCK) { 1115 /* 1116 * an orphan block is full of 1020 entries, 1117 * then we need to flush current orphan blocks 1118 * and bring another one in memory 1119 */ 1120 orphan_blk->blk_addr = cpu_to_le16(index); 1121 orphan_blk->blk_count = cpu_to_le16(orphan_blocks); 1122 orphan_blk->entry_count = cpu_to_le32(nentries); 1123 folio_mark_dirty(folio); 1124 f2fs_folio_put(folio, true); 1125 index++; 1126 nentries = 0; 1127 folio = NULL; 1128 } 1129 } 1130 1131 if (folio) { 1132 orphan_blk->blk_addr = cpu_to_le16(index); 1133 orphan_blk->blk_count = cpu_to_le16(orphan_blocks); 1134 orphan_blk->entry_count = cpu_to_le32(nentries); 1135 folio_mark_dirty(folio); 1136 f2fs_folio_put(folio, true); 1137 } 1138 } 1139 1140 static __u32 f2fs_checkpoint_chksum(struct f2fs_checkpoint *ckpt) 1141 { 1142 unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset); 1143 __u32 chksum; 1144 1145 chksum = f2fs_crc32(ckpt, chksum_ofs); 1146 if (chksum_ofs < CP_CHKSUM_OFFSET) { 1147 chksum_ofs += sizeof(chksum); 1148 chksum = f2fs_chksum(chksum, (__u8 *)ckpt + chksum_ofs, 1149 F2FS_BLKSIZE - chksum_ofs); 1150 } 1151 return chksum; 1152 } 1153 1154 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr, 1155 struct f2fs_checkpoint **cp_block, struct folio **cp_folio, 1156 unsigned long long *version) 1157 { 1158 size_t crc_offset = 0; 1159 __u32 crc; 1160 1161 *cp_folio = f2fs_get_meta_folio(sbi, cp_addr); 1162 if (IS_ERR(*cp_folio)) 1163 return PTR_ERR(*cp_folio); 1164 1165 *cp_block = folio_address(*cp_folio); 1166 1167 crc_offset = le32_to_cpu((*cp_block)->checksum_offset); 1168 if (crc_offset < CP_MIN_CHKSUM_OFFSET || 1169 crc_offset > CP_CHKSUM_OFFSET) { 1170 f2fs_folio_put(*cp_folio, true); 1171 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset); 1172 return -EINVAL; 1173 } 1174 1175 crc = f2fs_checkpoint_chksum(*cp_block); 1176 if (crc != cur_cp_crc(*cp_block)) { 1177 f2fs_folio_put(*cp_folio, true); 1178 f2fs_warn(sbi, "invalid crc value"); 1179 return -EINVAL; 1180 } 1181 1182 *version = cur_cp_version(*cp_block); 1183 return 0; 1184 } 1185 1186 static struct folio *validate_checkpoint(struct f2fs_sb_info *sbi, 1187 block_t cp_addr, unsigned long long *version) 1188 { 1189 struct folio *cp_folio_1 = NULL, *cp_folio_2 = NULL; 1190 struct f2fs_checkpoint *cp_block = NULL; 1191 unsigned long long cur_version = 0, pre_version = 0; 1192 unsigned int cp_blocks; 1193 int err; 1194 1195 err = get_checkpoint_version(sbi, cp_addr, &cp_block, 1196 &cp_folio_1, version); 1197 if (err) 1198 return NULL; 1199 1200 cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count); 1201 1202 if (cp_blocks > BLKS_PER_SEG(sbi) || cp_blocks <= F2FS_CP_PACKS) { 1203 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u", 1204 le32_to_cpu(cp_block->cp_pack_total_block_count)); 1205 goto invalid_cp; 1206 } 1207 pre_version = *version; 1208 1209 cp_addr += cp_blocks - 1; 1210 err = get_checkpoint_version(sbi, cp_addr, &cp_block, 1211 &cp_folio_2, version); 1212 if (err) 1213 goto invalid_cp; 1214 cur_version = *version; 1215 1216 if (cur_version == pre_version) { 1217 *version = cur_version; 1218 f2fs_folio_put(cp_folio_2, true); 1219 return cp_folio_1; 1220 } 1221 f2fs_folio_put(cp_folio_2, true); 1222 invalid_cp: 1223 f2fs_folio_put(cp_folio_1, true); 1224 return NULL; 1225 } 1226 1227 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi) 1228 { 1229 struct f2fs_checkpoint *cp_block; 1230 struct f2fs_super_block *fsb = sbi->raw_super; 1231 struct folio *cp1, *cp2, *cur_folio; 1232 unsigned long blk_size = sbi->blocksize; 1233 unsigned long long cp1_version = 0, cp2_version = 0; 1234 unsigned long long cp_start_blk_no; 1235 unsigned int cp_blks = 1 + __cp_payload(sbi); 1236 block_t cp_blk_no; 1237 int i; 1238 int err; 1239 1240 sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks), 1241 GFP_KERNEL); 1242 if (!sbi->ckpt) 1243 return -ENOMEM; 1244 /* 1245 * Finding out valid cp block involves read both 1246 * sets( cp pack 1 and cp pack 2) 1247 */ 1248 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr); 1249 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version); 1250 1251 /* The second checkpoint pack should start at the next segment */ 1252 cp_start_blk_no += ((unsigned long long)1) << 1253 le32_to_cpu(fsb->log_blocks_per_seg); 1254 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version); 1255 1256 if (cp1 && cp2) { 1257 if (ver_after(cp2_version, cp1_version)) 1258 cur_folio = cp2; 1259 else 1260 cur_folio = cp1; 1261 } else if (cp1) { 1262 cur_folio = cp1; 1263 } else if (cp2) { 1264 cur_folio = cp2; 1265 } else { 1266 err = -EFSCORRUPTED; 1267 goto fail_no_cp; 1268 } 1269 1270 cp_block = folio_address(cur_folio); 1271 memcpy(sbi->ckpt, cp_block, blk_size); 1272 1273 if (cur_folio == cp1) 1274 sbi->cur_cp_pack = 1; 1275 else 1276 sbi->cur_cp_pack = 2; 1277 1278 /* Sanity checking of checkpoint */ 1279 if (f2fs_sanity_check_ckpt(sbi)) { 1280 err = -EFSCORRUPTED; 1281 goto free_fail_no_cp; 1282 } 1283 1284 if (cp_blks <= 1) 1285 goto done; 1286 1287 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr); 1288 if (cur_folio == cp2) 1289 cp_blk_no += BIT(le32_to_cpu(fsb->log_blocks_per_seg)); 1290 1291 for (i = 1; i < cp_blks; i++) { 1292 void *sit_bitmap_ptr; 1293 unsigned char *ckpt = (unsigned char *)sbi->ckpt; 1294 1295 cur_folio = f2fs_get_meta_folio(sbi, cp_blk_no + i); 1296 if (IS_ERR(cur_folio)) { 1297 err = PTR_ERR(cur_folio); 1298 goto free_fail_no_cp; 1299 } 1300 sit_bitmap_ptr = folio_address(cur_folio); 1301 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size); 1302 f2fs_folio_put(cur_folio, true); 1303 } 1304 done: 1305 f2fs_folio_put(cp1, true); 1306 f2fs_folio_put(cp2, true); 1307 return 0; 1308 1309 free_fail_no_cp: 1310 f2fs_folio_put(cp1, true); 1311 f2fs_folio_put(cp2, true); 1312 fail_no_cp: 1313 kvfree(sbi->ckpt); 1314 return err; 1315 } 1316 1317 static void __add_dirty_inode(struct inode *inode, enum inode_type type) 1318 { 1319 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1320 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE; 1321 1322 if (is_inode_flag_set(inode, flag)) 1323 return; 1324 1325 set_inode_flag(inode, flag); 1326 list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]); 1327 stat_inc_dirty_inode(sbi, type); 1328 } 1329 1330 static void __remove_dirty_inode(struct inode *inode, enum inode_type type) 1331 { 1332 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE; 1333 1334 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag)) 1335 return; 1336 1337 list_del_init(&F2FS_I(inode)->dirty_list); 1338 clear_inode_flag(inode, flag); 1339 stat_dec_dirty_inode(F2FS_I_SB(inode), type); 1340 } 1341 1342 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio) 1343 { 1344 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1345 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE; 1346 1347 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) && 1348 !S_ISLNK(inode->i_mode)) 1349 return; 1350 1351 spin_lock(&sbi->inode_lock[type]); 1352 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH)) 1353 __add_dirty_inode(inode, type); 1354 inode_inc_dirty_pages(inode); 1355 spin_unlock(&sbi->inode_lock[type]); 1356 1357 folio_set_f2fs_reference(folio); 1358 } 1359 1360 void f2fs_remove_dirty_inode(struct inode *inode) 1361 { 1362 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1363 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE; 1364 1365 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) && 1366 !S_ISLNK(inode->i_mode)) 1367 return; 1368 1369 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH)) 1370 return; 1371 1372 spin_lock(&sbi->inode_lock[type]); 1373 __remove_dirty_inode(inode, type); 1374 spin_unlock(&sbi->inode_lock[type]); 1375 } 1376 1377 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type, 1378 bool from_cp) 1379 { 1380 struct list_head *head; 1381 struct inode *inode; 1382 struct f2fs_inode_info *fi; 1383 bool is_dir = (type == DIR_INODE); 1384 unsigned long ino = 0; 1385 1386 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir, 1387 get_pages(sbi, is_dir ? 1388 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA)); 1389 retry: 1390 if (unlikely(f2fs_cp_error(sbi))) { 1391 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir, 1392 get_pages(sbi, is_dir ? 1393 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA)); 1394 return -EIO; 1395 } 1396 1397 spin_lock(&sbi->inode_lock[type]); 1398 1399 head = &sbi->inode_list[type]; 1400 if (list_empty(head)) { 1401 spin_unlock(&sbi->inode_lock[type]); 1402 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir, 1403 get_pages(sbi, is_dir ? 1404 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA)); 1405 return 0; 1406 } 1407 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list); 1408 inode = igrab(&fi->vfs_inode); 1409 spin_unlock(&sbi->inode_lock[type]); 1410 if (inode) { 1411 unsigned long cur_ino = inode->i_ino; 1412 1413 if (from_cp) 1414 F2FS_I(inode)->cp_task = current; 1415 F2FS_I(inode)->wb_task = current; 1416 1417 filemap_fdatawrite(inode->i_mapping); 1418 1419 F2FS_I(inode)->wb_task = NULL; 1420 if (from_cp) 1421 F2FS_I(inode)->cp_task = NULL; 1422 1423 iput(inode); 1424 /* We need to give cpu to another writers. */ 1425 if (ino == cur_ino) 1426 cond_resched(); 1427 else 1428 ino = cur_ino; 1429 } else { 1430 /* 1431 * We should submit bio, since it exists several 1432 * writebacking dentry pages in the freeing inode. 1433 */ 1434 f2fs_submit_merged_write(sbi, DATA); 1435 cond_resched(); 1436 } 1437 goto retry; 1438 } 1439 1440 static int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi) 1441 { 1442 struct list_head *head = &sbi->inode_list[DIRTY_META]; 1443 struct inode *inode; 1444 struct f2fs_inode_info *fi; 1445 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA); 1446 1447 while (total--) { 1448 if (unlikely(f2fs_cp_error(sbi))) 1449 return -EIO; 1450 1451 spin_lock(&sbi->inode_lock[DIRTY_META]); 1452 if (list_empty(head)) { 1453 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1454 return 0; 1455 } 1456 fi = list_first_entry(head, struct f2fs_inode_info, 1457 gdirty_list); 1458 inode = igrab(&fi->vfs_inode); 1459 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1460 if (inode) { 1461 sync_inode_metadata(inode, 0); 1462 1463 /* it's on eviction */ 1464 if (is_inode_flag_set(inode, FI_DIRTY_INODE)) 1465 f2fs_update_inode_page(inode); 1466 iput(inode); 1467 } 1468 } 1469 return 0; 1470 } 1471 1472 static void __prepare_cp_block(struct f2fs_sb_info *sbi) 1473 { 1474 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1475 struct f2fs_nm_info *nm_i = NM_I(sbi); 1476 nid_t last_nid = nm_i->next_scan_nid; 1477 1478 next_free_nid(sbi, &last_nid); 1479 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi)); 1480 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi)); 1481 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi)); 1482 ckpt->next_free_nid = cpu_to_le32(last_nid); 1483 1484 /* update user_block_counts */ 1485 sbi->last_valid_block_count = sbi->total_valid_block_count; 1486 percpu_counter_set(&sbi->alloc_valid_block_count, 0); 1487 percpu_counter_set(&sbi->rf_node_block_count, 0); 1488 } 1489 1490 static bool __need_flush_quota(struct f2fs_sb_info *sbi) 1491 { 1492 bool ret = false; 1493 1494 if (!is_journalled_quota(sbi)) 1495 return false; 1496 1497 if (!f2fs_down_write_trylock(&sbi->quota_sem)) 1498 return true; 1499 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) { 1500 ret = false; 1501 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) { 1502 ret = false; 1503 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) { 1504 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH); 1505 ret = true; 1506 } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) { 1507 ret = true; 1508 } 1509 f2fs_up_write(&sbi->quota_sem); 1510 return ret; 1511 } 1512 1513 /* 1514 * Freeze all the FS-operations for checkpoint. 1515 */ 1516 static int block_operations(struct f2fs_sb_info *sbi) 1517 { 1518 struct writeback_control wbc = { 1519 .sync_mode = WB_SYNC_ALL, 1520 .nr_to_write = LONG_MAX, 1521 }; 1522 int err = 0, cnt = 0; 1523 1524 /* 1525 * Let's flush inline_data in dirty node pages. 1526 */ 1527 f2fs_flush_inline_data(sbi); 1528 1529 retry_flush_quotas: 1530 f2fs_lock_all(sbi); 1531 if (__need_flush_quota(sbi)) { 1532 bool need_lock = sbi->umount_lock_holder != current; 1533 1534 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) { 1535 set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH); 1536 set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH); 1537 goto retry_flush_dents; 1538 } 1539 f2fs_unlock_all(sbi); 1540 1541 /* don't grab s_umount lock during mount/umount/remount/freeze/quotactl */ 1542 if (!need_lock) { 1543 f2fs_do_quota_sync(sbi->sb, -1); 1544 } else if (down_read_trylock(&sbi->sb->s_umount)) { 1545 f2fs_do_quota_sync(sbi->sb, -1); 1546 up_read(&sbi->sb->s_umount); 1547 } 1548 cond_resched(); 1549 goto retry_flush_quotas; 1550 } 1551 1552 retry_flush_dents: 1553 /* write all the dirty dentry pages */ 1554 if (get_pages(sbi, F2FS_DIRTY_DENTS)) { 1555 f2fs_unlock_all(sbi); 1556 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true); 1557 if (err) 1558 return err; 1559 cond_resched(); 1560 goto retry_flush_quotas; 1561 } 1562 1563 /* 1564 * POR: we should ensure that there are no dirty node pages 1565 * until finishing nat/sit flush. inode->i_blocks can be updated. 1566 */ 1567 f2fs_down_write(&sbi->node_change); 1568 1569 if (get_pages(sbi, F2FS_DIRTY_IMETA)) { 1570 f2fs_up_write(&sbi->node_change); 1571 f2fs_unlock_all(sbi); 1572 err = f2fs_sync_inode_meta(sbi); 1573 if (err) 1574 return err; 1575 cond_resched(); 1576 goto retry_flush_quotas; 1577 } 1578 1579 retry_flush_nodes: 1580 f2fs_down_write(&sbi->node_write); 1581 1582 if (get_pages(sbi, F2FS_DIRTY_NODES)) { 1583 f2fs_up_write(&sbi->node_write); 1584 atomic_inc(&sbi->wb_sync_req[NODE]); 1585 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO); 1586 atomic_dec(&sbi->wb_sync_req[NODE]); 1587 if (err) { 1588 f2fs_up_write(&sbi->node_change); 1589 f2fs_unlock_all(sbi); 1590 return err; 1591 } 1592 cond_resched(); 1593 goto retry_flush_nodes; 1594 } 1595 1596 /* 1597 * sbi->node_change is used only for AIO write_begin path which produces 1598 * dirty node blocks and some checkpoint values by block allocation. 1599 */ 1600 __prepare_cp_block(sbi); 1601 f2fs_up_write(&sbi->node_change); 1602 return err; 1603 } 1604 1605 static void unblock_operations(struct f2fs_sb_info *sbi) 1606 { 1607 f2fs_up_write(&sbi->node_write); 1608 f2fs_unlock_all(sbi); 1609 } 1610 1611 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type) 1612 { 1613 DEFINE_WAIT(wait); 1614 1615 for (;;) { 1616 if (!get_pages(sbi, type)) 1617 break; 1618 1619 if (unlikely(f2fs_cp_error(sbi) && 1620 !is_sbi_flag_set(sbi, SBI_IS_CLOSE))) 1621 break; 1622 1623 if (type == F2FS_DIRTY_META) 1624 f2fs_sync_meta_pages(sbi, LONG_MAX, FS_CP_META_IO); 1625 else if (type == F2FS_WB_CP_DATA) 1626 f2fs_submit_merged_write(sbi, DATA); 1627 1628 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE); 1629 io_schedule_timeout(DEFAULT_SCHEDULE_TIMEOUT); 1630 } 1631 finish_wait(&sbi->cp_wait, &wait); 1632 } 1633 1634 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1635 { 1636 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num; 1637 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1638 unsigned long flags; 1639 1640 spin_lock_irqsave(&sbi->cp_lock, flags); 1641 1642 if ((cpc->reason & CP_UMOUNT) && 1643 le32_to_cpu(ckpt->cp_pack_total_block_count) > 1644 sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks) 1645 disable_nat_bits(sbi, false); 1646 1647 if (cpc->reason & CP_TRIMMED) 1648 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG); 1649 else 1650 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG); 1651 1652 if (cpc->reason & CP_UMOUNT) 1653 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG); 1654 else 1655 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG); 1656 1657 if (cpc->reason & CP_FASTBOOT) 1658 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG); 1659 else 1660 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG); 1661 1662 if (orphan_num) 1663 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); 1664 else 1665 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); 1666 1667 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1668 __set_ckpt_flags(ckpt, CP_FSCK_FLAG); 1669 1670 if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS)) 1671 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG); 1672 else 1673 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG); 1674 1675 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED)) 1676 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG); 1677 else 1678 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG); 1679 1680 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK)) 1681 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG); 1682 else 1683 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG); 1684 1685 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) 1686 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG); 1687 else 1688 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG); 1689 1690 if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) 1691 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG); 1692 1693 /* set this flag to activate crc|cp_ver for recovery */ 1694 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG); 1695 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG); 1696 1697 spin_unlock_irqrestore(&sbi->cp_lock, flags); 1698 } 1699 1700 static void commit_checkpoint(struct f2fs_sb_info *sbi, 1701 void *src, block_t blk_addr) 1702 { 1703 struct writeback_control wbc = {}; 1704 1705 /* 1706 * filemap_get_folios_tag and folio_lock again will take 1707 * some extra time. Therefore, f2fs_update_meta_pages and 1708 * f2fs_sync_meta_pages are combined in this function. 1709 */ 1710 struct folio *folio = f2fs_grab_meta_folio(sbi, blk_addr); 1711 1712 memcpy(folio_address(folio), src, PAGE_SIZE); 1713 1714 folio_mark_dirty(folio); 1715 if (unlikely(!folio_clear_dirty_for_io(folio))) 1716 f2fs_bug_on(sbi, 1); 1717 1718 /* writeout cp pack 2 page */ 1719 if (unlikely(!__f2fs_write_meta_folio(folio, &wbc, FS_CP_META_IO))) { 1720 if (f2fs_cp_error(sbi)) { 1721 f2fs_folio_put(folio, true); 1722 return; 1723 } 1724 f2fs_bug_on(sbi, true); 1725 } 1726 1727 f2fs_folio_put(folio, false); 1728 1729 /* submit checkpoint (with barrier if NOBARRIER is not set) */ 1730 f2fs_submit_merged_write(sbi, META_FLUSH); 1731 } 1732 1733 static inline u64 get_sectors_written(struct block_device *bdev) 1734 { 1735 return (u64)part_stat_read(bdev, sectors[STAT_WRITE]); 1736 } 1737 1738 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi) 1739 { 1740 if (f2fs_is_multi_device(sbi)) { 1741 u64 sectors = 0; 1742 int i; 1743 1744 for (i = 0; i < sbi->s_ndevs; i++) 1745 sectors += get_sectors_written(FDEV(i).bdev); 1746 1747 return sectors; 1748 } 1749 1750 return get_sectors_written(sbi->sb->s_bdev); 1751 } 1752 1753 static inline void stat_cp_time(struct cp_control *cpc, enum cp_time type) 1754 { 1755 cpc->stats.times[type] = ktime_get(); 1756 } 1757 1758 static inline void check_cp_time(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1759 { 1760 unsigned long long sb_diff, cur_diff; 1761 enum cp_time ct; 1762 1763 sb_diff = (u64)ktime_ms_delta(sbi->cp_stats.times[CP_TIME_END], 1764 sbi->cp_stats.times[CP_TIME_START]); 1765 cur_diff = (u64)ktime_ms_delta(cpc->stats.times[CP_TIME_END], 1766 cpc->stats.times[CP_TIME_START]); 1767 1768 if (cur_diff > sb_diff) { 1769 sbi->cp_stats = cpc->stats; 1770 if (cur_diff < CP_LONG_LATENCY_THRESHOLD) 1771 return; 1772 1773 f2fs_warn(sbi, "checkpoint was blocked for %llu ms", cur_diff); 1774 for (ct = CP_TIME_START; ct < CP_TIME_MAX - 1; ct++) 1775 f2fs_warn(sbi, "Step#%d: %llu ms", ct, 1776 (u64)ktime_ms_delta(cpc->stats.times[ct + 1], 1777 cpc->stats.times[ct])); 1778 } 1779 } 1780 1781 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1782 { 1783 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1784 struct f2fs_nm_info *nm_i = NM_I(sbi); 1785 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags; 1786 block_t start_blk; 1787 unsigned int data_sum_blocks, orphan_blocks; 1788 __u32 crc32 = 0; 1789 int i; 1790 int cp_payload_blks = __cp_payload(sbi); 1791 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE); 1792 u64 kbytes_written; 1793 int err; 1794 1795 /* Flush all the NAT/SIT pages */ 1796 f2fs_sync_meta_pages(sbi, LONG_MAX, FS_CP_META_IO); 1797 1798 stat_cp_time(cpc, CP_TIME_SYNC_META); 1799 1800 /* start to update checkpoint, cp ver is already updated previously */ 1801 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true)); 1802 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi)); 1803 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) { 1804 struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_NODE); 1805 1806 ckpt->cur_node_segno[i] = cpu_to_le32(curseg->segno); 1807 ckpt->cur_node_blkoff[i] = cpu_to_le16(curseg->next_blkoff); 1808 ckpt->alloc_type[i + CURSEG_HOT_NODE] = curseg->alloc_type; 1809 } 1810 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) { 1811 struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_DATA); 1812 1813 ckpt->cur_data_segno[i] = cpu_to_le32(curseg->segno); 1814 ckpt->cur_data_blkoff[i] = cpu_to_le16(curseg->next_blkoff); 1815 ckpt->alloc_type[i + CURSEG_HOT_DATA] = curseg->alloc_type; 1816 } 1817 1818 /* 2 cp + n data seg summary + orphan inode blocks */ 1819 data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false); 1820 spin_lock_irqsave(&sbi->cp_lock, flags); 1821 if (data_sum_blocks < NR_CURSEG_DATA_TYPE) 1822 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); 1823 else 1824 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); 1825 spin_unlock_irqrestore(&sbi->cp_lock, flags); 1826 1827 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num); 1828 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks + 1829 orphan_blocks); 1830 1831 if (__remain_node_summaries(cpc->reason)) 1832 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS + 1833 cp_payload_blks + data_sum_blocks + 1834 orphan_blocks + NR_CURSEG_NODE_TYPE); 1835 else 1836 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS + 1837 cp_payload_blks + data_sum_blocks + 1838 orphan_blocks); 1839 1840 /* update ckpt flag for checkpoint */ 1841 update_ckpt_flags(sbi, cpc); 1842 1843 /* update SIT/NAT bitmap */ 1844 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP)); 1845 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP)); 1846 1847 crc32 = f2fs_checkpoint_chksum(ckpt); 1848 *((__le32 *)((unsigned char *)ckpt + 1849 le32_to_cpu(ckpt->checksum_offset))) 1850 = cpu_to_le32(crc32); 1851 1852 start_blk = __start_cp_next_addr(sbi); 1853 1854 /* write nat bits */ 1855 if (enabled_nat_bits(sbi, cpc)) { 1856 __u64 cp_ver = cur_cp_version(ckpt); 1857 block_t blk; 1858 1859 cp_ver |= ((__u64)crc32 << 32); 1860 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver); 1861 1862 blk = start_blk + BLKS_PER_SEG(sbi) - nm_i->nat_bits_blocks; 1863 for (i = 0; i < nm_i->nat_bits_blocks; i++) 1864 f2fs_update_meta_page(sbi, nm_i->nat_bits + 1865 F2FS_BLK_TO_BYTES(i), blk + i); 1866 } 1867 1868 /* write out checkpoint buffer at block 0 */ 1869 f2fs_update_meta_page(sbi, ckpt, start_blk++); 1870 1871 for (i = 1; i < 1 + cp_payload_blks; i++) 1872 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE, 1873 start_blk++); 1874 1875 if (orphan_num) { 1876 write_orphan_inodes(sbi, start_blk); 1877 start_blk += orphan_blocks; 1878 } 1879 1880 f2fs_write_data_summaries(sbi, start_blk); 1881 start_blk += data_sum_blocks; 1882 1883 /* Record write statistics in the hot node summary */ 1884 kbytes_written = sbi->kbytes_written; 1885 kbytes_written += (f2fs_get_sectors_written(sbi) - 1886 sbi->sectors_written_start) >> 1; 1887 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written); 1888 1889 if (__remain_node_summaries(cpc->reason)) { 1890 f2fs_write_node_summaries(sbi, start_blk); 1891 start_blk += NR_CURSEG_NODE_TYPE; 1892 } 1893 1894 /* Here, we have one bio having CP pack except cp pack 2 page */ 1895 f2fs_sync_meta_pages(sbi, LONG_MAX, FS_CP_META_IO); 1896 stat_cp_time(cpc, CP_TIME_SYNC_CP_META); 1897 1898 /* Wait for all dirty meta pages to be submitted for IO */ 1899 f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META); 1900 stat_cp_time(cpc, CP_TIME_WAIT_DIRTY_META); 1901 1902 /* wait for previous submitted meta pages writeback */ 1903 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA); 1904 stat_cp_time(cpc, CP_TIME_WAIT_CP_DATA); 1905 1906 /* flush all device cache */ 1907 err = f2fs_flush_device_cache(sbi); 1908 if (err) 1909 return err; 1910 stat_cp_time(cpc, CP_TIME_FLUSH_DEVICE); 1911 1912 /* barrier and flush checkpoint cp pack 2 page if it can */ 1913 commit_checkpoint(sbi, ckpt, start_blk); 1914 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA); 1915 stat_cp_time(cpc, CP_TIME_WAIT_LAST_CP); 1916 1917 /* 1918 * invalidate intermediate page cache borrowed from meta inode which are 1919 * used for migration of encrypted, verity or compressed inode's blocks. 1920 */ 1921 if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) || 1922 f2fs_sb_has_compression(sbi)) 1923 f2fs_bug_on(sbi, 1924 invalidate_inode_pages2_range(META_MAPPING(sbi), 1925 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1)); 1926 1927 f2fs_release_ino_entry(sbi, false); 1928 1929 f2fs_reset_fsync_node_info(sbi); 1930 1931 clear_sbi_flag(sbi, SBI_IS_DIRTY); 1932 clear_sbi_flag(sbi, SBI_NEED_CP); 1933 clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH); 1934 1935 spin_lock(&sbi->stat_lock); 1936 sbi->unusable_block_count = 0; 1937 spin_unlock(&sbi->stat_lock); 1938 1939 __set_cp_next_pack(sbi); 1940 1941 /* 1942 * redirty superblock if metadata like node page or inode cache is 1943 * updated during writing checkpoint. 1944 */ 1945 if (get_pages(sbi, F2FS_DIRTY_NODES) || 1946 get_pages(sbi, F2FS_DIRTY_IMETA)) 1947 set_sbi_flag(sbi, SBI_IS_DIRTY); 1948 1949 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS)); 1950 1951 return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0; 1952 } 1953 1954 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1955 { 1956 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1957 struct f2fs_lock_context lc; 1958 unsigned long long ckpt_ver; 1959 int err = 0; 1960 1961 stat_cp_time(cpc, CP_TIME_START); 1962 1963 if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi)) 1964 return -EROFS; 1965 1966 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 1967 if (cpc->reason != CP_PAUSE) 1968 return 0; 1969 f2fs_warn(sbi, "Start checkpoint disabled!"); 1970 } 1971 if (cpc->reason != CP_RESIZE) 1972 f2fs_down_write_trace(&sbi->cp_global_sem, &lc); 1973 1974 stat_cp_time(cpc, CP_TIME_LOCK); 1975 1976 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) && 1977 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) || 1978 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks))) 1979 goto out; 1980 if (unlikely(f2fs_cp_error(sbi))) { 1981 err = -EIO; 1982 goto out; 1983 } 1984 1985 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, CP_PHASE_START_BLOCK_OPS); 1986 1987 err = block_operations(sbi); 1988 if (err) 1989 goto out; 1990 1991 stat_cp_time(cpc, CP_TIME_OP_LOCK); 1992 1993 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, CP_PHASE_FINISH_BLOCK_OPS); 1994 1995 f2fs_flush_merged_writes(sbi); 1996 1997 /* this is the case of multiple fstrims without any changes */ 1998 if (cpc->reason & CP_DISCARD) { 1999 if (!f2fs_exist_trim_candidates(sbi, cpc)) { 2000 unblock_operations(sbi); 2001 goto out; 2002 } 2003 2004 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 && 2005 SIT_I(sbi)->dirty_sentries == 0 && 2006 prefree_segments(sbi) == 0) { 2007 f2fs_flush_sit_entries(sbi, cpc); 2008 f2fs_clear_prefree_segments(sbi, cpc); 2009 unblock_operations(sbi); 2010 goto out; 2011 } 2012 } 2013 stat_cp_time(cpc, CP_TIME_MERGE_WRITE); 2014 2015 /* 2016 * update checkpoint pack index 2017 * Increase the version number so that 2018 * SIT entries and seg summaries are written at correct place 2019 */ 2020 ckpt_ver = cur_cp_version(ckpt); 2021 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver); 2022 2023 /* write cached NAT/SIT entries to NAT/SIT area */ 2024 err = f2fs_flush_nat_entries(sbi, cpc); 2025 if (err) { 2026 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err); 2027 f2fs_bug_on(sbi, !f2fs_cp_error(sbi)); 2028 goto stop; 2029 } 2030 stat_cp_time(cpc, CP_TIME_FLUSH_NAT); 2031 2032 f2fs_flush_sit_entries(sbi, cpc); 2033 2034 stat_cp_time(cpc, CP_TIME_FLUSH_SIT); 2035 2036 /* save inmem log status */ 2037 f2fs_save_inmem_curseg(sbi); 2038 2039 err = do_checkpoint(sbi, cpc); 2040 if (err) { 2041 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err); 2042 f2fs_bug_on(sbi, !f2fs_cp_error(sbi)); 2043 f2fs_release_discard_addrs(sbi); 2044 } else { 2045 f2fs_clear_prefree_segments(sbi, cpc); 2046 } 2047 2048 f2fs_restore_inmem_curseg(sbi); 2049 f2fs_reinit_atgc_curseg(sbi); 2050 stat_inc_cp_count(sbi); 2051 stop: 2052 unblock_operations(sbi); 2053 stat_cp_time(cpc, CP_TIME_END); 2054 check_cp_time(sbi, cpc); 2055 2056 if (cpc->reason & CP_RECOVERY) 2057 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver); 2058 2059 /* update CP_TIME to trigger checkpoint periodically */ 2060 f2fs_update_time(sbi, CP_TIME); 2061 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, CP_PHASE_FINISH_CHECKPOINT); 2062 out: 2063 if (cpc->reason != CP_RESIZE) 2064 f2fs_up_write_trace(&sbi->cp_global_sem, &lc); 2065 return err; 2066 } 2067 2068 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi) 2069 { 2070 int i; 2071 2072 for (i = 0; i < MAX_INO_ENTRY; i++) { 2073 struct inode_management *im = &sbi->im[i]; 2074 2075 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC); 2076 spin_lock_init(&im->ino_lock); 2077 INIT_LIST_HEAD(&im->ino_list); 2078 im->ino_num = 0; 2079 } 2080 2081 sbi->max_orphans = (BLKS_PER_SEG(sbi) - F2FS_CP_PACKS - 2082 NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) * 2083 F2FS_ORPHANS_PER_BLOCK; 2084 } 2085 2086 int __init f2fs_create_checkpoint_caches(void) 2087 { 2088 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry", 2089 sizeof(struct ino_entry)); 2090 if (!ino_entry_slab) 2091 return -ENOMEM; 2092 f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry", 2093 sizeof(struct inode_entry)); 2094 if (!f2fs_inode_entry_slab) { 2095 kmem_cache_destroy(ino_entry_slab); 2096 return -ENOMEM; 2097 } 2098 return 0; 2099 } 2100 2101 void f2fs_destroy_checkpoint_caches(void) 2102 { 2103 kmem_cache_destroy(ino_entry_slab); 2104 kmem_cache_destroy(f2fs_inode_entry_slab); 2105 } 2106 2107 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi) 2108 { 2109 struct cp_control cpc = { .reason = CP_SYNC, }; 2110 struct f2fs_lock_context lc; 2111 int err; 2112 2113 f2fs_down_write_trace(&sbi->gc_lock, &lc); 2114 err = f2fs_write_checkpoint(sbi, &cpc); 2115 f2fs_up_write_trace(&sbi->gc_lock, &lc); 2116 2117 return err; 2118 } 2119 2120 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi) 2121 { 2122 struct ckpt_req_control *cprc = &sbi->cprc_info; 2123 struct ckpt_req *req, *next; 2124 struct llist_node *dispatch_list; 2125 u64 sum_diff = 0, diff, count = 0; 2126 int ret; 2127 2128 dispatch_list = llist_del_all(&cprc->issue_list); 2129 if (!dispatch_list) 2130 return; 2131 dispatch_list = llist_reverse_order(dispatch_list); 2132 2133 ret = __write_checkpoint_sync(sbi); 2134 atomic_inc(&cprc->issued_ckpt); 2135 2136 llist_for_each_entry_safe(req, next, dispatch_list, llnode) { 2137 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time); 2138 req->ret = ret; 2139 req->delta_time = diff; 2140 complete(&req->wait); 2141 2142 sum_diff += diff; 2143 count++; 2144 } 2145 atomic_sub(count, &cprc->queued_ckpt); 2146 atomic_add(count, &cprc->total_ckpt); 2147 2148 spin_lock(&cprc->stat_lock); 2149 cprc->cur_time = (unsigned int)div64_u64(sum_diff, count); 2150 if (cprc->peak_time < cprc->cur_time) 2151 cprc->peak_time = cprc->cur_time; 2152 spin_unlock(&cprc->stat_lock); 2153 } 2154 2155 static int issue_checkpoint_thread(void *data) 2156 { 2157 struct f2fs_sb_info *sbi = data; 2158 struct ckpt_req_control *cprc = &sbi->cprc_info; 2159 wait_queue_head_t *q = &cprc->ckpt_wait_queue; 2160 repeat: 2161 if (kthread_should_stop()) 2162 return 0; 2163 2164 if (!llist_empty(&cprc->issue_list)) 2165 __checkpoint_and_complete_reqs(sbi); 2166 2167 wait_event_interruptible(*q, 2168 kthread_should_stop() || !llist_empty(&cprc->issue_list)); 2169 goto repeat; 2170 } 2171 2172 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi, 2173 struct ckpt_req *wait_req) 2174 { 2175 struct ckpt_req_control *cprc = &sbi->cprc_info; 2176 2177 if (!llist_empty(&cprc->issue_list)) { 2178 __checkpoint_and_complete_reqs(sbi); 2179 } else { 2180 /* already dispatched by issue_checkpoint_thread */ 2181 if (wait_req) 2182 wait_for_completion(&wait_req->wait); 2183 } 2184 } 2185 2186 static void init_ckpt_req(struct ckpt_req *req) 2187 { 2188 memset(req, 0, sizeof(struct ckpt_req)); 2189 2190 init_completion(&req->wait); 2191 req->queue_time = ktime_get(); 2192 } 2193 2194 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi) 2195 { 2196 struct ckpt_req_control *cprc = &sbi->cprc_info; 2197 struct ckpt_req req; 2198 struct cp_control cpc; 2199 2200 cpc.reason = __get_cp_reason(sbi); 2201 if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC || 2202 sbi->umount_lock_holder == current) { 2203 struct f2fs_lock_context lc; 2204 int ret; 2205 2206 f2fs_down_write_trace(&sbi->gc_lock, &lc); 2207 ret = f2fs_write_checkpoint(sbi, &cpc); 2208 f2fs_up_write_trace(&sbi->gc_lock, &lc); 2209 2210 return ret; 2211 } 2212 2213 if (!cprc->f2fs_issue_ckpt) 2214 return __write_checkpoint_sync(sbi); 2215 2216 init_ckpt_req(&req); 2217 2218 llist_add(&req.llnode, &cprc->issue_list); 2219 atomic_inc(&cprc->queued_ckpt); 2220 2221 /* 2222 * update issue_list before we wake up issue_checkpoint thread, 2223 * this smp_mb() pairs with another barrier in ___wait_event(), 2224 * see more details in comments of waitqueue_active(). 2225 */ 2226 smp_mb(); 2227 2228 if (waitqueue_active(&cprc->ckpt_wait_queue)) 2229 wake_up(&cprc->ckpt_wait_queue); 2230 2231 if (cprc->f2fs_issue_ckpt) 2232 wait_for_completion(&req.wait); 2233 else 2234 flush_remained_ckpt_reqs(sbi, &req); 2235 2236 if (unlikely(req.delta_time >= CP_LONG_LATENCY_THRESHOLD)) { 2237 f2fs_warn_ratelimited(sbi, 2238 "blocked on checkpoint for %u ms", cprc->peak_time); 2239 dump_stack(); 2240 } 2241 2242 return req.ret; 2243 } 2244 2245 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi) 2246 { 2247 dev_t dev = sbi->sb->s_bdev->bd_dev; 2248 struct ckpt_req_control *cprc = &sbi->cprc_info; 2249 2250 if (cprc->f2fs_issue_ckpt) 2251 return 0; 2252 2253 cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi, 2254 "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev)); 2255 if (IS_ERR(cprc->f2fs_issue_ckpt)) { 2256 int err = PTR_ERR(cprc->f2fs_issue_ckpt); 2257 2258 cprc->f2fs_issue_ckpt = NULL; 2259 return err; 2260 } 2261 2262 set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio); 2263 set_user_nice(cprc->f2fs_issue_ckpt, 2264 PRIO_TO_NICE(sbi->critical_task_priority)); 2265 2266 return 0; 2267 } 2268 2269 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi) 2270 { 2271 struct ckpt_req_control *cprc = &sbi->cprc_info; 2272 struct task_struct *ckpt_task; 2273 2274 if (!cprc->f2fs_issue_ckpt) 2275 return; 2276 2277 ckpt_task = cprc->f2fs_issue_ckpt; 2278 cprc->f2fs_issue_ckpt = NULL; 2279 kthread_stop(ckpt_task); 2280 2281 f2fs_flush_ckpt_thread(sbi); 2282 } 2283 2284 void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi) 2285 { 2286 struct ckpt_req_control *cprc = &sbi->cprc_info; 2287 2288 flush_remained_ckpt_reqs(sbi, NULL); 2289 2290 /* Let's wait for the previous dispatched checkpoint. */ 2291 while (atomic_read(&cprc->queued_ckpt)) 2292 io_schedule_timeout(DEFAULT_SCHEDULE_TIMEOUT); 2293 } 2294 2295 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi) 2296 { 2297 struct ckpt_req_control *cprc = &sbi->cprc_info; 2298 2299 atomic_set(&cprc->issued_ckpt, 0); 2300 atomic_set(&cprc->total_ckpt, 0); 2301 atomic_set(&cprc->queued_ckpt, 0); 2302 cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO; 2303 init_waitqueue_head(&cprc->ckpt_wait_queue); 2304 init_llist_head(&cprc->issue_list); 2305 spin_lock_init(&cprc->stat_lock); 2306 } 2307