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 #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 return true; 111 default: 112 f2fs_bug_on(sem->sbi, 1); 113 } 114 return false; 115 } 116 117 static void uplift_priority(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc, 118 bool is_write) 119 { 120 lc->need_restore = false; 121 if (!sem->sbi->adjust_lock_priority) 122 return; 123 if (rt_task(current)) 124 return; 125 if (!need_uplift_priority(sem, is_write)) 126 return; 127 lc->orig_nice = task_nice(current); 128 lc->new_nice = PRIO_TO_NICE(sem->sbi->lock_duration_priority); 129 if (lc->orig_nice <= lc->new_nice) 130 return; 131 set_user_nice(current, lc->new_nice); 132 lc->need_restore = true; 133 134 trace_f2fs_priority_uplift(sem->sbi, sem->name, is_write, current, 135 NICE_TO_PRIO(lc->orig_nice), NICE_TO_PRIO(lc->new_nice)); 136 } 137 138 static void restore_priority(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc, 139 bool is_write) 140 { 141 if (!lc->need_restore) 142 return; 143 /* someone has updated the priority */ 144 if (task_nice(current) != lc->new_nice) 145 return; 146 set_user_nice(current, lc->orig_nice); 147 148 trace_f2fs_priority_restore(sem->sbi, sem->name, is_write, current, 149 NICE_TO_PRIO(lc->orig_nice), NICE_TO_PRIO(lc->new_nice)); 150 } 151 152 void f2fs_down_read_trace(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc) 153 { 154 uplift_priority(sem, lc, false); 155 f2fs_down_read(sem); 156 trace_lock_elapsed_time_start(sem, lc); 157 } 158 159 int f2fs_down_read_trylock_trace(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc) 160 { 161 uplift_priority(sem, lc, false); 162 if (!f2fs_down_read_trylock(sem)) { 163 restore_priority(sem, lc, false); 164 return 0; 165 } 166 trace_lock_elapsed_time_start(sem, lc); 167 return 1; 168 } 169 170 void f2fs_up_read_trace(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc) 171 { 172 f2fs_up_read(sem); 173 restore_priority(sem, lc, false); 174 trace_lock_elapsed_time_end(sem, lc, false); 175 } 176 177 void f2fs_down_write_trace(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc) 178 { 179 uplift_priority(sem, lc, true); 180 f2fs_down_write(sem); 181 trace_lock_elapsed_time_start(sem, lc); 182 } 183 184 int f2fs_down_write_trylock_trace(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc) 185 { 186 uplift_priority(sem, lc, true); 187 if (!f2fs_down_write_trylock(sem)) { 188 restore_priority(sem, lc, true); 189 return 0; 190 } 191 trace_lock_elapsed_time_start(sem, lc); 192 return 1; 193 } 194 195 void f2fs_up_write_trace(struct f2fs_rwsem *sem, struct f2fs_lock_context *lc) 196 { 197 f2fs_up_write(sem); 198 restore_priority(sem, lc, true); 199 trace_lock_elapsed_time_end(sem, lc, true); 200 } 201 202 void f2fs_lock_op(struct f2fs_sb_info *sbi, struct f2fs_lock_context *lc) 203 { 204 f2fs_down_read_trace(&sbi->cp_rwsem, lc); 205 } 206 207 int f2fs_trylock_op(struct f2fs_sb_info *sbi, struct f2fs_lock_context *lc) 208 { 209 if (time_to_inject(sbi, FAULT_LOCK_OP)) 210 return 0; 211 212 return f2fs_down_read_trylock_trace(&sbi->cp_rwsem, lc); 213 } 214 215 void f2fs_unlock_op(struct f2fs_sb_info *sbi, struct f2fs_lock_context *lc) 216 { 217 f2fs_up_read_trace(&sbi->cp_rwsem, lc); 218 } 219 220 static inline void f2fs_lock_all(struct f2fs_sb_info *sbi) 221 { 222 f2fs_down_write(&sbi->cp_rwsem); 223 } 224 225 static inline void f2fs_unlock_all(struct f2fs_sb_info *sbi) 226 { 227 f2fs_up_write(&sbi->cp_rwsem); 228 } 229 230 #define DEFAULT_CHECKPOINT_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_RT, 3)) 231 232 static struct kmem_cache *ino_entry_slab; 233 struct kmem_cache *f2fs_inode_entry_slab; 234 235 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io, 236 unsigned char reason) 237 { 238 f2fs_build_fault_attr(sbi, 0, 0, FAULT_ALL); 239 if (!end_io) 240 f2fs_flush_merged_writes(sbi); 241 f2fs_handle_critical_error(sbi, reason); 242 } 243 244 /* 245 * We guarantee no failure on the returned page. 246 */ 247 struct folio *f2fs_grab_meta_folio(struct f2fs_sb_info *sbi, pgoff_t index) 248 { 249 struct address_space *mapping = META_MAPPING(sbi); 250 struct folio *folio; 251 repeat: 252 folio = f2fs_grab_cache_folio(mapping, index, false); 253 if (IS_ERR(folio)) { 254 cond_resched(); 255 goto repeat; 256 } 257 f2fs_folio_wait_writeback(folio, META, true, true); 258 if (!folio_test_uptodate(folio)) 259 folio_mark_uptodate(folio); 260 return folio; 261 } 262 263 static struct folio *__get_meta_folio(struct f2fs_sb_info *sbi, pgoff_t index, 264 bool is_meta) 265 { 266 struct address_space *mapping = META_MAPPING(sbi); 267 struct folio *folio; 268 struct f2fs_io_info fio = { 269 .sbi = sbi, 270 .type = META, 271 .op = REQ_OP_READ, 272 .op_flags = REQ_META | REQ_PRIO, 273 .old_blkaddr = index, 274 .new_blkaddr = index, 275 .encrypted_page = NULL, 276 .is_por = !is_meta ? 1 : 0, 277 }; 278 int err; 279 280 if (unlikely(!is_meta)) 281 fio.op_flags &= ~REQ_META; 282 repeat: 283 folio = f2fs_grab_cache_folio(mapping, index, false); 284 if (IS_ERR(folio)) { 285 cond_resched(); 286 goto repeat; 287 } 288 if (folio_test_uptodate(folio)) 289 goto out; 290 291 fio.folio = folio; 292 293 err = f2fs_submit_page_bio(&fio); 294 if (err) { 295 f2fs_folio_put(folio, true); 296 return ERR_PTR(err); 297 } 298 299 f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, F2FS_BLKSIZE); 300 301 folio_lock(folio); 302 if (unlikely(!is_meta_folio(folio))) { 303 f2fs_folio_put(folio, true); 304 goto repeat; 305 } 306 307 if (unlikely(!folio_test_uptodate(folio))) { 308 f2fs_handle_page_eio(sbi, folio, META); 309 f2fs_folio_put(folio, true); 310 return ERR_PTR(-EIO); 311 } 312 out: 313 return folio; 314 } 315 316 struct folio *f2fs_get_meta_folio(struct f2fs_sb_info *sbi, pgoff_t index) 317 { 318 return __get_meta_folio(sbi, index, true); 319 } 320 321 struct folio *f2fs_get_meta_folio_retry(struct f2fs_sb_info *sbi, pgoff_t index) 322 { 323 struct folio *folio; 324 int count = 0; 325 326 retry: 327 folio = __get_meta_folio(sbi, index, true); 328 if (IS_ERR(folio)) { 329 if (PTR_ERR(folio) == -EIO && 330 ++count <= DEFAULT_RETRY_IO_COUNT) 331 goto retry; 332 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_META_PAGE); 333 } 334 return folio; 335 } 336 337 /* for POR only */ 338 struct folio *f2fs_get_tmp_folio(struct f2fs_sb_info *sbi, pgoff_t index) 339 { 340 return __get_meta_folio(sbi, index, false); 341 } 342 343 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr, 344 int type) 345 { 346 struct seg_entry *se; 347 unsigned int segno, offset; 348 bool exist; 349 350 if (type == DATA_GENERIC) 351 return true; 352 353 segno = GET_SEGNO(sbi, blkaddr); 354 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 355 se = get_seg_entry(sbi, segno); 356 357 exist = f2fs_test_bit(offset, se->cur_valid_map); 358 359 /* skip data, if we already have an error in checkpoint. */ 360 if (unlikely(f2fs_cp_error(sbi))) 361 return exist; 362 363 if ((exist && type == DATA_GENERIC_ENHANCE_UPDATE) || 364 (!exist && type == DATA_GENERIC_ENHANCE)) 365 goto out_err; 366 if (!exist && type != DATA_GENERIC_ENHANCE_UPDATE) 367 goto out_handle; 368 return exist; 369 370 out_err: 371 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d", 372 blkaddr, exist); 373 set_sbi_flag(sbi, SBI_NEED_FSCK); 374 dump_stack(); 375 out_handle: 376 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); 377 return exist; 378 } 379 380 static bool __f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi, 381 block_t blkaddr, int type) 382 { 383 switch (type) { 384 case META_NAT: 385 break; 386 case META_SIT: 387 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi))) 388 goto check_only; 389 break; 390 case META_SSA: 391 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) || 392 blkaddr < SM_I(sbi)->ssa_blkaddr)) 393 goto check_only; 394 break; 395 case META_CP: 396 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr || 397 blkaddr < __start_cp_addr(sbi))) 398 goto check_only; 399 break; 400 case META_POR: 401 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) || 402 blkaddr < MAIN_BLKADDR(sbi))) 403 goto check_only; 404 break; 405 case DATA_GENERIC: 406 case DATA_GENERIC_ENHANCE: 407 case DATA_GENERIC_ENHANCE_READ: 408 case DATA_GENERIC_ENHANCE_UPDATE: 409 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) || 410 blkaddr < MAIN_BLKADDR(sbi))) { 411 412 /* Skip to emit an error message. */ 413 if (unlikely(f2fs_cp_error(sbi))) 414 return false; 415 416 f2fs_warn(sbi, "access invalid blkaddr:%u", 417 blkaddr); 418 set_sbi_flag(sbi, SBI_NEED_FSCK); 419 dump_stack(); 420 goto err; 421 } else { 422 return __is_bitmap_valid(sbi, blkaddr, type); 423 } 424 break; 425 case META_GENERIC: 426 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) || 427 blkaddr >= MAIN_BLKADDR(sbi))) 428 goto err; 429 break; 430 default: 431 BUG(); 432 } 433 434 return true; 435 err: 436 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); 437 check_only: 438 return false; 439 } 440 441 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi, 442 block_t blkaddr, int type) 443 { 444 if (time_to_inject(sbi, FAULT_BLKADDR_VALIDITY)) 445 return false; 446 return __f2fs_is_valid_blkaddr(sbi, blkaddr, type); 447 } 448 449 bool f2fs_is_valid_blkaddr_raw(struct f2fs_sb_info *sbi, 450 block_t blkaddr, int type) 451 { 452 return __f2fs_is_valid_blkaddr(sbi, blkaddr, type); 453 } 454 455 /* 456 * Readahead CP/NAT/SIT/SSA/POR pages 457 */ 458 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages, 459 int type, bool sync) 460 { 461 block_t blkno = start; 462 struct f2fs_io_info fio = { 463 .sbi = sbi, 464 .type = META, 465 .op = REQ_OP_READ, 466 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD, 467 .encrypted_page = NULL, 468 .in_list = 0, 469 .is_por = (type == META_POR) ? 1 : 0, 470 }; 471 struct blk_plug plug; 472 int err; 473 474 if (unlikely(type == META_POR)) 475 fio.op_flags &= ~REQ_META; 476 477 blk_start_plug(&plug); 478 for (; nrpages-- > 0; blkno++) { 479 struct folio *folio; 480 481 if (!f2fs_is_valid_blkaddr(sbi, blkno, type)) 482 goto out; 483 484 switch (type) { 485 case META_NAT: 486 if (unlikely(blkno >= 487 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid))) 488 blkno = 0; 489 /* get nat block addr */ 490 fio.new_blkaddr = current_nat_addr(sbi, 491 blkno * NAT_ENTRY_PER_BLOCK); 492 break; 493 case META_SIT: 494 if (unlikely(blkno >= TOTAL_SEGS(sbi))) 495 goto out; 496 /* get sit block addr */ 497 fio.new_blkaddr = current_sit_addr(sbi, 498 blkno * SIT_ENTRY_PER_BLOCK); 499 break; 500 case META_SSA: 501 case META_CP: 502 case META_POR: 503 fio.new_blkaddr = blkno; 504 break; 505 default: 506 BUG(); 507 } 508 509 folio = f2fs_grab_cache_folio(META_MAPPING(sbi), 510 fio.new_blkaddr, false); 511 if (IS_ERR(folio)) 512 continue; 513 if (folio_test_uptodate(folio)) { 514 f2fs_folio_put(folio, true); 515 continue; 516 } 517 518 fio.folio = folio; 519 err = f2fs_submit_page_bio(&fio); 520 f2fs_folio_put(folio, err ? true : false); 521 522 if (!err) 523 f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, 524 F2FS_BLKSIZE); 525 } 526 out: 527 blk_finish_plug(&plug); 528 return blkno - start; 529 } 530 531 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index, 532 unsigned int ra_blocks) 533 { 534 struct folio *folio; 535 bool readahead = false; 536 537 if (ra_blocks == RECOVERY_MIN_RA_BLOCKS) 538 return; 539 540 folio = filemap_get_folio(META_MAPPING(sbi), index); 541 if (IS_ERR(folio) || !folio_test_uptodate(folio)) 542 readahead = true; 543 f2fs_folio_put(folio, false); 544 545 if (readahead) 546 f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true); 547 } 548 549 static bool __f2fs_write_meta_folio(struct folio *folio, 550 struct writeback_control *wbc, 551 enum iostat_type io_type) 552 { 553 struct f2fs_sb_info *sbi = F2FS_F_SB(folio); 554 555 trace_f2fs_writepage(folio, META); 556 557 if (unlikely(f2fs_cp_error(sbi))) { 558 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) { 559 folio_clear_uptodate(folio); 560 dec_page_count(sbi, F2FS_DIRTY_META); 561 folio_unlock(folio); 562 return true; 563 } 564 goto redirty_out; 565 } 566 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 567 goto redirty_out; 568 569 f2fs_do_write_meta_page(sbi, folio, io_type); 570 dec_page_count(sbi, F2FS_DIRTY_META); 571 572 folio_unlock(folio); 573 574 if (unlikely(f2fs_cp_error(sbi))) 575 f2fs_submit_merged_write(sbi, META); 576 577 return true; 578 579 redirty_out: 580 folio_redirty_for_writepage(wbc, folio); 581 return false; 582 } 583 584 static int f2fs_write_meta_pages(struct address_space *mapping, 585 struct writeback_control *wbc) 586 { 587 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); 588 struct f2fs_lock_context lc; 589 long diff, written; 590 591 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) 592 goto skip_write; 593 594 /* collect a number of dirty meta pages and write together */ 595 if (wbc->sync_mode != WB_SYNC_ALL && 596 get_pages(sbi, F2FS_DIRTY_META) < 597 nr_pages_to_skip(sbi, META)) 598 goto skip_write; 599 600 /* if locked failed, cp will flush dirty pages instead */ 601 if (!f2fs_down_write_trylock_trace(&sbi->cp_global_sem, &lc)) 602 goto skip_write; 603 604 trace_f2fs_writepages(mapping->host, wbc, META); 605 diff = nr_pages_to_write(sbi, META, wbc); 606 written = f2fs_sync_meta_pages(sbi, wbc->nr_to_write, FS_META_IO); 607 f2fs_up_write_trace(&sbi->cp_global_sem, &lc); 608 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff); 609 return 0; 610 611 skip_write: 612 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META); 613 trace_f2fs_writepages(mapping->host, wbc, META); 614 return 0; 615 } 616 617 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, long nr_to_write, 618 enum iostat_type io_type) 619 { 620 struct address_space *mapping = META_MAPPING(sbi); 621 pgoff_t index = 0, prev = ULONG_MAX; 622 struct folio_batch fbatch; 623 long nwritten = 0; 624 int nr_folios; 625 struct writeback_control wbc = {}; 626 struct blk_plug plug; 627 628 folio_batch_init(&fbatch); 629 630 blk_start_plug(&plug); 631 632 while ((nr_folios = filemap_get_folios_tag(mapping, &index, 633 (pgoff_t)-1, 634 PAGECACHE_TAG_DIRTY, &fbatch))) { 635 int i; 636 637 for (i = 0; i < nr_folios; i++) { 638 struct folio *folio = fbatch.folios[i]; 639 640 if (nr_to_write != LONG_MAX && i != 0 && 641 folio->index != prev + 642 folio_nr_pages(fbatch.folios[i-1])) { 643 folio_batch_release(&fbatch); 644 goto stop; 645 } 646 647 folio_lock(folio); 648 649 if (unlikely(!is_meta_folio(folio))) { 650 continue_unlock: 651 folio_unlock(folio); 652 continue; 653 } 654 if (!folio_test_dirty(folio)) { 655 /* someone wrote it for us */ 656 goto continue_unlock; 657 } 658 659 f2fs_folio_wait_writeback(folio, META, true, true); 660 661 if (!folio_clear_dirty_for_io(folio)) 662 goto continue_unlock; 663 664 if (!__f2fs_write_meta_folio(folio, &wbc, 665 io_type)) { 666 folio_unlock(folio); 667 break; 668 } 669 nwritten += folio_nr_pages(folio); 670 prev = folio->index; 671 if (unlikely(nwritten >= nr_to_write)) 672 break; 673 } 674 folio_batch_release(&fbatch); 675 cond_resched(); 676 } 677 stop: 678 if (nwritten) 679 f2fs_submit_merged_write(sbi, META); 680 681 blk_finish_plug(&plug); 682 683 return nwritten; 684 } 685 686 static bool f2fs_dirty_meta_folio(struct address_space *mapping, 687 struct folio *folio) 688 { 689 trace_f2fs_set_page_dirty(folio, META); 690 691 if (!folio_test_uptodate(folio)) 692 folio_mark_uptodate(folio); 693 if (filemap_dirty_folio(mapping, folio)) { 694 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META); 695 folio_set_f2fs_reference(folio); 696 return true; 697 } 698 return false; 699 } 700 701 const struct address_space_operations f2fs_meta_aops = { 702 .writepages = f2fs_write_meta_pages, 703 .dirty_folio = f2fs_dirty_meta_folio, 704 .invalidate_folio = f2fs_invalidate_folio, 705 .release_folio = f2fs_release_folio, 706 .migrate_folio = filemap_migrate_folio, 707 }; 708 709 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, 710 unsigned int devidx, int type) 711 { 712 struct inode_management *im = &sbi->im[type]; 713 struct ino_entry *e = NULL, *new = NULL; 714 int ret; 715 716 if (type == FLUSH_INO) { 717 rcu_read_lock(); 718 e = radix_tree_lookup(&im->ino_root, ino); 719 rcu_read_unlock(); 720 } 721 722 retry: 723 if (!e) 724 new = f2fs_kmem_cache_alloc(ino_entry_slab, 725 GFP_NOFS, true, NULL); 726 727 ret = radix_tree_preload(GFP_NOFS | __GFP_NOFAIL); 728 f2fs_bug_on(sbi, ret); 729 730 spin_lock(&im->ino_lock); 731 e = radix_tree_lookup(&im->ino_root, ino); 732 if (!e) { 733 if (!new) { 734 spin_unlock(&im->ino_lock); 735 radix_tree_preload_end(); 736 goto retry; 737 } 738 e = new; 739 if (unlikely(radix_tree_insert(&im->ino_root, ino, e))) 740 f2fs_bug_on(sbi, 1); 741 742 memset(e, 0, sizeof(struct ino_entry)); 743 e->ino = ino; 744 745 list_add_tail(&e->list, &im->ino_list); 746 if (type != ORPHAN_INO) 747 im->ino_num++; 748 } 749 750 if (type == FLUSH_INO) 751 f2fs_set_bit(devidx, (char *)&e->dirty_device); 752 753 spin_unlock(&im->ino_lock); 754 radix_tree_preload_end(); 755 756 if (new && e != new) 757 kmem_cache_free(ino_entry_slab, new); 758 } 759 760 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) 761 { 762 struct inode_management *im = &sbi->im[type]; 763 struct ino_entry *e; 764 765 spin_lock(&im->ino_lock); 766 e = radix_tree_lookup(&im->ino_root, ino); 767 if (e) { 768 list_del(&e->list); 769 radix_tree_delete(&im->ino_root, ino); 770 im->ino_num--; 771 spin_unlock(&im->ino_lock); 772 kmem_cache_free(ino_entry_slab, e); 773 return; 774 } 775 spin_unlock(&im->ino_lock); 776 } 777 778 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) 779 { 780 /* add new dirty ino entry into list */ 781 __add_ino_entry(sbi, ino, 0, type); 782 } 783 784 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type) 785 { 786 /* remove dirty ino entry from list */ 787 __remove_ino_entry(sbi, ino, type); 788 } 789 790 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */ 791 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode) 792 { 793 struct inode_management *im = &sbi->im[mode]; 794 struct ino_entry *e; 795 796 spin_lock(&im->ino_lock); 797 e = radix_tree_lookup(&im->ino_root, ino); 798 spin_unlock(&im->ino_lock); 799 return e ? true : false; 800 } 801 802 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all) 803 { 804 struct ino_entry *e, *tmp; 805 int i; 806 807 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) { 808 struct inode_management *im = &sbi->im[i]; 809 810 spin_lock(&im->ino_lock); 811 list_for_each_entry_safe(e, tmp, &im->ino_list, list) { 812 list_del(&e->list); 813 radix_tree_delete(&im->ino_root, e->ino); 814 kmem_cache_free(ino_entry_slab, e); 815 im->ino_num--; 816 } 817 spin_unlock(&im->ino_lock); 818 } 819 } 820 821 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino, 822 unsigned int devidx, int type) 823 { 824 __add_ino_entry(sbi, ino, devidx, type); 825 } 826 827 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino, 828 unsigned int devidx, int type) 829 { 830 struct inode_management *im = &sbi->im[type]; 831 struct ino_entry *e; 832 bool is_dirty = false; 833 834 spin_lock(&im->ino_lock); 835 e = radix_tree_lookup(&im->ino_root, ino); 836 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device)) 837 is_dirty = true; 838 spin_unlock(&im->ino_lock); 839 return is_dirty; 840 } 841 842 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi) 843 { 844 struct inode_management *im = &sbi->im[ORPHAN_INO]; 845 int err = 0; 846 847 spin_lock(&im->ino_lock); 848 849 if (time_to_inject(sbi, FAULT_ORPHAN)) { 850 spin_unlock(&im->ino_lock); 851 return -ENOSPC; 852 } 853 854 if (unlikely(im->ino_num >= sbi->max_orphans)) 855 err = -ENOSPC; 856 else 857 im->ino_num++; 858 spin_unlock(&im->ino_lock); 859 860 return err; 861 } 862 863 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi) 864 { 865 struct inode_management *im = &sbi->im[ORPHAN_INO]; 866 867 spin_lock(&im->ino_lock); 868 f2fs_bug_on(sbi, im->ino_num == 0); 869 im->ino_num--; 870 spin_unlock(&im->ino_lock); 871 } 872 873 void f2fs_add_orphan_inode(struct inode *inode) 874 { 875 /* add new orphan ino entry into list */ 876 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO); 877 f2fs_update_inode_page(inode); 878 } 879 880 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) 881 { 882 /* remove orphan entry from orphan list */ 883 __remove_ino_entry(sbi, ino, ORPHAN_INO); 884 } 885 886 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) 887 { 888 struct inode *inode; 889 struct node_info ni; 890 int err; 891 892 inode = f2fs_iget_retry(sbi->sb, ino); 893 if (IS_ERR(inode)) { 894 /* 895 * there should be a bug that we can't find the entry 896 * to orphan inode. 897 */ 898 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT); 899 return PTR_ERR(inode); 900 } 901 902 err = f2fs_dquot_initialize(inode); 903 if (err) { 904 iput(inode); 905 goto err_out; 906 } 907 908 clear_nlink(inode); 909 910 /* truncate all the data during iput */ 911 iput(inode); 912 913 err = f2fs_get_node_info(sbi, ino, &ni, false); 914 if (err) 915 goto err_out; 916 917 /* ENOMEM was fully retried in f2fs_evict_inode. */ 918 if (ni.blk_addr != NULL_ADDR) { 919 err = -EIO; 920 goto err_out; 921 } 922 return 0; 923 924 err_out: 925 set_sbi_flag(sbi, SBI_NEED_FSCK); 926 f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.", 927 __func__, ino); 928 return err; 929 } 930 931 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi) 932 { 933 block_t start_blk, orphan_blocks, i, j; 934 int err = 0; 935 936 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG)) 937 return 0; 938 939 if (f2fs_hw_is_readonly(sbi)) { 940 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup"); 941 return 0; 942 } 943 944 if (is_sbi_flag_set(sbi, SBI_IS_WRITABLE)) 945 f2fs_info(sbi, "orphan cleanup on readonly fs"); 946 947 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi); 948 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi); 949 950 f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true); 951 952 for (i = 0; i < orphan_blocks; i++) { 953 struct folio *folio; 954 struct f2fs_orphan_block *orphan_blk; 955 956 folio = f2fs_get_meta_folio(sbi, start_blk + i); 957 if (IS_ERR(folio)) { 958 err = PTR_ERR(folio); 959 goto out; 960 } 961 962 orphan_blk = folio_address(folio); 963 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) { 964 nid_t ino = le32_to_cpu(orphan_blk->ino[j]); 965 966 err = recover_orphan_inode(sbi, ino); 967 if (err) { 968 f2fs_folio_put(folio, true); 969 goto out; 970 } 971 } 972 f2fs_folio_put(folio, true); 973 } 974 /* clear Orphan Flag */ 975 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG); 976 out: 977 set_sbi_flag(sbi, SBI_IS_RECOVERED); 978 979 return err; 980 } 981 982 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk) 983 { 984 struct list_head *head; 985 struct f2fs_orphan_block *orphan_blk = NULL; 986 unsigned int nentries = 0; 987 unsigned short index = 1; 988 unsigned short orphan_blocks; 989 struct folio *folio = NULL; 990 struct ino_entry *orphan = NULL; 991 struct inode_management *im = &sbi->im[ORPHAN_INO]; 992 993 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num); 994 995 /* 996 * we don't need to do spin_lock(&im->ino_lock) here, since all the 997 * orphan inode operations are covered under f2fs_lock_op(). 998 * And, spin_lock should be avoided due to page operations below. 999 */ 1000 head = &im->ino_list; 1001 1002 /* loop for each orphan inode entry and write them in journal block */ 1003 list_for_each_entry(orphan, head, list) { 1004 if (!folio) { 1005 folio = f2fs_grab_meta_folio(sbi, start_blk++); 1006 orphan_blk = folio_address(folio); 1007 memset(orphan_blk, 0, sizeof(*orphan_blk)); 1008 } 1009 1010 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino); 1011 1012 if (nentries == F2FS_ORPHANS_PER_BLOCK) { 1013 /* 1014 * an orphan block is full of 1020 entries, 1015 * then we need to flush current orphan blocks 1016 * and bring another one in memory 1017 */ 1018 orphan_blk->blk_addr = cpu_to_le16(index); 1019 orphan_blk->blk_count = cpu_to_le16(orphan_blocks); 1020 orphan_blk->entry_count = cpu_to_le32(nentries); 1021 folio_mark_dirty(folio); 1022 f2fs_folio_put(folio, true); 1023 index++; 1024 nentries = 0; 1025 folio = NULL; 1026 } 1027 } 1028 1029 if (folio) { 1030 orphan_blk->blk_addr = cpu_to_le16(index); 1031 orphan_blk->blk_count = cpu_to_le16(orphan_blocks); 1032 orphan_blk->entry_count = cpu_to_le32(nentries); 1033 folio_mark_dirty(folio); 1034 f2fs_folio_put(folio, true); 1035 } 1036 } 1037 1038 static __u32 f2fs_checkpoint_chksum(struct f2fs_checkpoint *ckpt) 1039 { 1040 unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset); 1041 __u32 chksum; 1042 1043 chksum = f2fs_crc32(ckpt, chksum_ofs); 1044 if (chksum_ofs < CP_CHKSUM_OFFSET) { 1045 chksum_ofs += sizeof(chksum); 1046 chksum = f2fs_chksum(chksum, (__u8 *)ckpt + chksum_ofs, 1047 F2FS_BLKSIZE - chksum_ofs); 1048 } 1049 return chksum; 1050 } 1051 1052 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr, 1053 struct f2fs_checkpoint **cp_block, struct folio **cp_folio, 1054 unsigned long long *version) 1055 { 1056 size_t crc_offset = 0; 1057 __u32 crc; 1058 1059 *cp_folio = f2fs_get_meta_folio(sbi, cp_addr); 1060 if (IS_ERR(*cp_folio)) 1061 return PTR_ERR(*cp_folio); 1062 1063 *cp_block = folio_address(*cp_folio); 1064 1065 crc_offset = le32_to_cpu((*cp_block)->checksum_offset); 1066 if (crc_offset < CP_MIN_CHKSUM_OFFSET || 1067 crc_offset > CP_CHKSUM_OFFSET) { 1068 f2fs_folio_put(*cp_folio, true); 1069 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset); 1070 return -EINVAL; 1071 } 1072 1073 crc = f2fs_checkpoint_chksum(*cp_block); 1074 if (crc != cur_cp_crc(*cp_block)) { 1075 f2fs_folio_put(*cp_folio, true); 1076 f2fs_warn(sbi, "invalid crc value"); 1077 return -EINVAL; 1078 } 1079 1080 *version = cur_cp_version(*cp_block); 1081 return 0; 1082 } 1083 1084 static struct folio *validate_checkpoint(struct f2fs_sb_info *sbi, 1085 block_t cp_addr, unsigned long long *version) 1086 { 1087 struct folio *cp_folio_1 = NULL, *cp_folio_2 = NULL; 1088 struct f2fs_checkpoint *cp_block = NULL; 1089 unsigned long long cur_version = 0, pre_version = 0; 1090 unsigned int cp_blocks; 1091 int err; 1092 1093 err = get_checkpoint_version(sbi, cp_addr, &cp_block, 1094 &cp_folio_1, version); 1095 if (err) 1096 return NULL; 1097 1098 cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count); 1099 1100 if (cp_blocks > BLKS_PER_SEG(sbi) || cp_blocks <= F2FS_CP_PACKS) { 1101 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u", 1102 le32_to_cpu(cp_block->cp_pack_total_block_count)); 1103 goto invalid_cp; 1104 } 1105 pre_version = *version; 1106 1107 cp_addr += cp_blocks - 1; 1108 err = get_checkpoint_version(sbi, cp_addr, &cp_block, 1109 &cp_folio_2, version); 1110 if (err) 1111 goto invalid_cp; 1112 cur_version = *version; 1113 1114 if (cur_version == pre_version) { 1115 *version = cur_version; 1116 f2fs_folio_put(cp_folio_2, true); 1117 return cp_folio_1; 1118 } 1119 f2fs_folio_put(cp_folio_2, true); 1120 invalid_cp: 1121 f2fs_folio_put(cp_folio_1, true); 1122 return NULL; 1123 } 1124 1125 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi) 1126 { 1127 struct f2fs_checkpoint *cp_block; 1128 struct f2fs_super_block *fsb = sbi->raw_super; 1129 struct folio *cp1, *cp2, *cur_folio; 1130 unsigned long blk_size = sbi->blocksize; 1131 unsigned long long cp1_version = 0, cp2_version = 0; 1132 unsigned long long cp_start_blk_no; 1133 unsigned int cp_blks = 1 + __cp_payload(sbi); 1134 block_t cp_blk_no; 1135 int i; 1136 int err; 1137 1138 sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks), 1139 GFP_KERNEL); 1140 if (!sbi->ckpt) 1141 return -ENOMEM; 1142 /* 1143 * Finding out valid cp block involves read both 1144 * sets( cp pack 1 and cp pack 2) 1145 */ 1146 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr); 1147 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version); 1148 1149 /* The second checkpoint pack should start at the next segment */ 1150 cp_start_blk_no += ((unsigned long long)1) << 1151 le32_to_cpu(fsb->log_blocks_per_seg); 1152 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version); 1153 1154 if (cp1 && cp2) { 1155 if (ver_after(cp2_version, cp1_version)) 1156 cur_folio = cp2; 1157 else 1158 cur_folio = cp1; 1159 } else if (cp1) { 1160 cur_folio = cp1; 1161 } else if (cp2) { 1162 cur_folio = cp2; 1163 } else { 1164 err = -EFSCORRUPTED; 1165 goto fail_no_cp; 1166 } 1167 1168 cp_block = folio_address(cur_folio); 1169 memcpy(sbi->ckpt, cp_block, blk_size); 1170 1171 if (cur_folio == cp1) 1172 sbi->cur_cp_pack = 1; 1173 else 1174 sbi->cur_cp_pack = 2; 1175 1176 /* Sanity checking of checkpoint */ 1177 if (f2fs_sanity_check_ckpt(sbi)) { 1178 err = -EFSCORRUPTED; 1179 goto free_fail_no_cp; 1180 } 1181 1182 if (cp_blks <= 1) 1183 goto done; 1184 1185 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr); 1186 if (cur_folio == cp2) 1187 cp_blk_no += BIT(le32_to_cpu(fsb->log_blocks_per_seg)); 1188 1189 for (i = 1; i < cp_blks; i++) { 1190 void *sit_bitmap_ptr; 1191 unsigned char *ckpt = (unsigned char *)sbi->ckpt; 1192 1193 cur_folio = f2fs_get_meta_folio(sbi, cp_blk_no + i); 1194 if (IS_ERR(cur_folio)) { 1195 err = PTR_ERR(cur_folio); 1196 goto free_fail_no_cp; 1197 } 1198 sit_bitmap_ptr = folio_address(cur_folio); 1199 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size); 1200 f2fs_folio_put(cur_folio, true); 1201 } 1202 done: 1203 f2fs_folio_put(cp1, true); 1204 f2fs_folio_put(cp2, true); 1205 return 0; 1206 1207 free_fail_no_cp: 1208 f2fs_folio_put(cp1, true); 1209 f2fs_folio_put(cp2, true); 1210 fail_no_cp: 1211 kvfree(sbi->ckpt); 1212 return err; 1213 } 1214 1215 static void __add_dirty_inode(struct inode *inode, enum inode_type type) 1216 { 1217 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1218 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE; 1219 1220 if (is_inode_flag_set(inode, flag)) 1221 return; 1222 1223 set_inode_flag(inode, flag); 1224 list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]); 1225 stat_inc_dirty_inode(sbi, type); 1226 } 1227 1228 static void __remove_dirty_inode(struct inode *inode, enum inode_type type) 1229 { 1230 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE; 1231 1232 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag)) 1233 return; 1234 1235 list_del_init(&F2FS_I(inode)->dirty_list); 1236 clear_inode_flag(inode, flag); 1237 stat_dec_dirty_inode(F2FS_I_SB(inode), type); 1238 } 1239 1240 void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio) 1241 { 1242 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1243 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE; 1244 1245 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) && 1246 !S_ISLNK(inode->i_mode)) 1247 return; 1248 1249 spin_lock(&sbi->inode_lock[type]); 1250 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH)) 1251 __add_dirty_inode(inode, type); 1252 inode_inc_dirty_pages(inode); 1253 spin_unlock(&sbi->inode_lock[type]); 1254 1255 folio_set_f2fs_reference(folio); 1256 } 1257 1258 void f2fs_remove_dirty_inode(struct inode *inode) 1259 { 1260 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1261 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE; 1262 1263 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) && 1264 !S_ISLNK(inode->i_mode)) 1265 return; 1266 1267 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH)) 1268 return; 1269 1270 spin_lock(&sbi->inode_lock[type]); 1271 __remove_dirty_inode(inode, type); 1272 spin_unlock(&sbi->inode_lock[type]); 1273 } 1274 1275 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type, 1276 bool from_cp) 1277 { 1278 struct list_head *head; 1279 struct inode *inode; 1280 struct f2fs_inode_info *fi; 1281 bool is_dir = (type == DIR_INODE); 1282 unsigned long ino = 0; 1283 1284 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir, 1285 get_pages(sbi, is_dir ? 1286 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA)); 1287 retry: 1288 if (unlikely(f2fs_cp_error(sbi))) { 1289 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir, 1290 get_pages(sbi, is_dir ? 1291 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA)); 1292 return -EIO; 1293 } 1294 1295 spin_lock(&sbi->inode_lock[type]); 1296 1297 head = &sbi->inode_list[type]; 1298 if (list_empty(head)) { 1299 spin_unlock(&sbi->inode_lock[type]); 1300 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir, 1301 get_pages(sbi, is_dir ? 1302 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA)); 1303 return 0; 1304 } 1305 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list); 1306 inode = igrab(&fi->vfs_inode); 1307 spin_unlock(&sbi->inode_lock[type]); 1308 if (inode) { 1309 unsigned long cur_ino = inode->i_ino; 1310 1311 if (from_cp) 1312 F2FS_I(inode)->cp_task = current; 1313 F2FS_I(inode)->wb_task = current; 1314 1315 filemap_fdatawrite(inode->i_mapping); 1316 1317 F2FS_I(inode)->wb_task = NULL; 1318 if (from_cp) 1319 F2FS_I(inode)->cp_task = NULL; 1320 1321 iput(inode); 1322 /* We need to give cpu to another writers. */ 1323 if (ino == cur_ino) 1324 cond_resched(); 1325 else 1326 ino = cur_ino; 1327 } else { 1328 /* 1329 * We should submit bio, since it exists several 1330 * writebacking dentry pages in the freeing inode. 1331 */ 1332 f2fs_submit_merged_write(sbi, DATA); 1333 cond_resched(); 1334 } 1335 goto retry; 1336 } 1337 1338 static int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi) 1339 { 1340 struct list_head *head = &sbi->inode_list[DIRTY_META]; 1341 struct inode *inode; 1342 struct f2fs_inode_info *fi; 1343 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA); 1344 1345 while (total--) { 1346 if (unlikely(f2fs_cp_error(sbi))) 1347 return -EIO; 1348 1349 spin_lock(&sbi->inode_lock[DIRTY_META]); 1350 if (list_empty(head)) { 1351 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1352 return 0; 1353 } 1354 fi = list_first_entry(head, struct f2fs_inode_info, 1355 gdirty_list); 1356 inode = igrab(&fi->vfs_inode); 1357 spin_unlock(&sbi->inode_lock[DIRTY_META]); 1358 if (inode) { 1359 sync_inode_metadata(inode, 0); 1360 1361 /* it's on eviction */ 1362 if (is_inode_flag_set(inode, FI_DIRTY_INODE)) 1363 f2fs_update_inode_page(inode); 1364 iput(inode); 1365 } 1366 } 1367 return 0; 1368 } 1369 1370 static void __prepare_cp_block(struct f2fs_sb_info *sbi) 1371 { 1372 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1373 struct f2fs_nm_info *nm_i = NM_I(sbi); 1374 nid_t last_nid = nm_i->next_scan_nid; 1375 1376 next_free_nid(sbi, &last_nid); 1377 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi)); 1378 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi)); 1379 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi)); 1380 ckpt->next_free_nid = cpu_to_le32(last_nid); 1381 1382 /* update user_block_counts */ 1383 sbi->last_valid_block_count = sbi->total_valid_block_count; 1384 percpu_counter_set(&sbi->alloc_valid_block_count, 0); 1385 percpu_counter_set(&sbi->rf_node_block_count, 0); 1386 } 1387 1388 static bool __need_flush_quota(struct f2fs_sb_info *sbi) 1389 { 1390 bool ret = false; 1391 1392 if (!is_journalled_quota(sbi)) 1393 return false; 1394 1395 if (!f2fs_down_write_trylock(&sbi->quota_sem)) 1396 return true; 1397 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) { 1398 ret = false; 1399 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) { 1400 ret = false; 1401 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) { 1402 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH); 1403 ret = true; 1404 } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) { 1405 ret = true; 1406 } 1407 f2fs_up_write(&sbi->quota_sem); 1408 return ret; 1409 } 1410 1411 /* 1412 * Freeze all the FS-operations for checkpoint. 1413 */ 1414 static int block_operations(struct f2fs_sb_info *sbi) 1415 { 1416 struct writeback_control wbc = { 1417 .sync_mode = WB_SYNC_ALL, 1418 .nr_to_write = LONG_MAX, 1419 }; 1420 int err = 0, cnt = 0; 1421 1422 /* 1423 * Let's flush inline_data in dirty node pages. 1424 */ 1425 f2fs_flush_inline_data(sbi); 1426 1427 retry_flush_quotas: 1428 f2fs_lock_all(sbi); 1429 if (__need_flush_quota(sbi)) { 1430 bool need_lock = sbi->umount_lock_holder != current; 1431 1432 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) { 1433 set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH); 1434 set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH); 1435 goto retry_flush_dents; 1436 } 1437 f2fs_unlock_all(sbi); 1438 1439 /* don't grab s_umount lock during mount/umount/remount/freeze/quotactl */ 1440 if (!need_lock) { 1441 f2fs_do_quota_sync(sbi->sb, -1); 1442 } else if (down_read_trylock(&sbi->sb->s_umount)) { 1443 f2fs_do_quota_sync(sbi->sb, -1); 1444 up_read(&sbi->sb->s_umount); 1445 } 1446 cond_resched(); 1447 goto retry_flush_quotas; 1448 } 1449 1450 retry_flush_dents: 1451 /* write all the dirty dentry pages */ 1452 if (get_pages(sbi, F2FS_DIRTY_DENTS)) { 1453 f2fs_unlock_all(sbi); 1454 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true); 1455 if (err) 1456 return err; 1457 cond_resched(); 1458 goto retry_flush_quotas; 1459 } 1460 1461 /* 1462 * POR: we should ensure that there are no dirty node pages 1463 * until finishing nat/sit flush. inode->i_blocks can be updated. 1464 */ 1465 f2fs_down_write(&sbi->node_change); 1466 1467 if (get_pages(sbi, F2FS_DIRTY_IMETA)) { 1468 f2fs_up_write(&sbi->node_change); 1469 f2fs_unlock_all(sbi); 1470 err = f2fs_sync_inode_meta(sbi); 1471 if (err) 1472 return err; 1473 cond_resched(); 1474 goto retry_flush_quotas; 1475 } 1476 1477 retry_flush_nodes: 1478 f2fs_down_write(&sbi->node_write); 1479 1480 if (get_pages(sbi, F2FS_DIRTY_NODES)) { 1481 f2fs_up_write(&sbi->node_write); 1482 atomic_inc(&sbi->wb_sync_req[NODE]); 1483 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO); 1484 atomic_dec(&sbi->wb_sync_req[NODE]); 1485 if (err) { 1486 f2fs_up_write(&sbi->node_change); 1487 f2fs_unlock_all(sbi); 1488 return err; 1489 } 1490 cond_resched(); 1491 goto retry_flush_nodes; 1492 } 1493 1494 /* 1495 * sbi->node_change is used only for AIO write_begin path which produces 1496 * dirty node blocks and some checkpoint values by block allocation. 1497 */ 1498 __prepare_cp_block(sbi); 1499 f2fs_up_write(&sbi->node_change); 1500 return err; 1501 } 1502 1503 static void unblock_operations(struct f2fs_sb_info *sbi) 1504 { 1505 f2fs_up_write(&sbi->node_write); 1506 f2fs_unlock_all(sbi); 1507 } 1508 1509 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type) 1510 { 1511 DEFINE_WAIT(wait); 1512 1513 for (;;) { 1514 if (!get_pages(sbi, type)) 1515 break; 1516 1517 if (unlikely(f2fs_cp_error(sbi) && 1518 !is_sbi_flag_set(sbi, SBI_IS_CLOSE))) 1519 break; 1520 1521 if (type == F2FS_DIRTY_META) 1522 f2fs_sync_meta_pages(sbi, LONG_MAX, FS_CP_META_IO); 1523 else if (type == F2FS_WB_CP_DATA) 1524 f2fs_submit_merged_write(sbi, DATA); 1525 1526 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE); 1527 io_schedule_timeout(DEFAULT_SCHEDULE_TIMEOUT); 1528 } 1529 finish_wait(&sbi->cp_wait, &wait); 1530 } 1531 1532 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1533 { 1534 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num; 1535 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1536 unsigned long flags; 1537 1538 spin_lock_irqsave(&sbi->cp_lock, flags); 1539 1540 if ((cpc->reason & CP_UMOUNT) && 1541 le32_to_cpu(ckpt->cp_pack_total_block_count) > 1542 sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks) 1543 disable_nat_bits(sbi, false); 1544 1545 if (cpc->reason & CP_TRIMMED) 1546 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG); 1547 else 1548 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG); 1549 1550 if (cpc->reason & CP_UMOUNT) 1551 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG); 1552 else 1553 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG); 1554 1555 if (cpc->reason & CP_FASTBOOT) 1556 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG); 1557 else 1558 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG); 1559 1560 if (orphan_num) 1561 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); 1562 else 1563 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); 1564 1565 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) 1566 __set_ckpt_flags(ckpt, CP_FSCK_FLAG); 1567 1568 if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS)) 1569 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG); 1570 else 1571 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG); 1572 1573 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED)) 1574 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG); 1575 else 1576 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG); 1577 1578 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK)) 1579 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG); 1580 else 1581 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG); 1582 1583 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) 1584 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG); 1585 else 1586 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG); 1587 1588 if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) 1589 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG); 1590 1591 /* set this flag to activate crc|cp_ver for recovery */ 1592 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG); 1593 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG); 1594 1595 spin_unlock_irqrestore(&sbi->cp_lock, flags); 1596 } 1597 1598 static void commit_checkpoint(struct f2fs_sb_info *sbi, 1599 void *src, block_t blk_addr) 1600 { 1601 struct writeback_control wbc = {}; 1602 1603 /* 1604 * filemap_get_folios_tag and folio_lock again will take 1605 * some extra time. Therefore, f2fs_update_meta_pages and 1606 * f2fs_sync_meta_pages are combined in this function. 1607 */ 1608 struct folio *folio = f2fs_grab_meta_folio(sbi, blk_addr); 1609 1610 memcpy(folio_address(folio), src, PAGE_SIZE); 1611 1612 folio_mark_dirty(folio); 1613 if (unlikely(!folio_clear_dirty_for_io(folio))) 1614 f2fs_bug_on(sbi, 1); 1615 1616 /* writeout cp pack 2 page */ 1617 if (unlikely(!__f2fs_write_meta_folio(folio, &wbc, FS_CP_META_IO))) { 1618 if (f2fs_cp_error(sbi)) { 1619 f2fs_folio_put(folio, true); 1620 return; 1621 } 1622 f2fs_bug_on(sbi, true); 1623 } 1624 1625 f2fs_folio_put(folio, false); 1626 1627 /* submit checkpoint (with barrier if NOBARRIER is not set) */ 1628 f2fs_submit_merged_write(sbi, META_FLUSH); 1629 } 1630 1631 static inline u64 get_sectors_written(struct block_device *bdev) 1632 { 1633 return (u64)part_stat_read(bdev, sectors[STAT_WRITE]); 1634 } 1635 1636 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi) 1637 { 1638 if (f2fs_is_multi_device(sbi)) { 1639 u64 sectors = 0; 1640 int i; 1641 1642 for (i = 0; i < sbi->s_ndevs; i++) 1643 sectors += get_sectors_written(FDEV(i).bdev); 1644 1645 return sectors; 1646 } 1647 1648 return get_sectors_written(sbi->sb->s_bdev); 1649 } 1650 1651 static inline void stat_cp_time(struct cp_control *cpc, enum cp_time type) 1652 { 1653 cpc->stats.times[type] = ktime_get(); 1654 } 1655 1656 static inline void check_cp_time(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1657 { 1658 unsigned long long sb_diff, cur_diff; 1659 enum cp_time ct; 1660 1661 sb_diff = (u64)ktime_ms_delta(sbi->cp_stats.times[CP_TIME_END], 1662 sbi->cp_stats.times[CP_TIME_START]); 1663 cur_diff = (u64)ktime_ms_delta(cpc->stats.times[CP_TIME_END], 1664 cpc->stats.times[CP_TIME_START]); 1665 1666 if (cur_diff > sb_diff) { 1667 sbi->cp_stats = cpc->stats; 1668 if (cur_diff < CP_LONG_LATENCY_THRESHOLD) 1669 return; 1670 1671 f2fs_warn(sbi, "checkpoint was blocked for %llu ms", cur_diff); 1672 for (ct = CP_TIME_START; ct < CP_TIME_MAX - 1; ct++) 1673 f2fs_warn(sbi, "Step#%d: %llu ms", ct, 1674 (u64)ktime_ms_delta(cpc->stats.times[ct + 1], 1675 cpc->stats.times[ct])); 1676 } 1677 } 1678 1679 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1680 { 1681 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1682 struct f2fs_nm_info *nm_i = NM_I(sbi); 1683 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags; 1684 block_t start_blk; 1685 unsigned int data_sum_blocks, orphan_blocks; 1686 __u32 crc32 = 0; 1687 int i; 1688 int cp_payload_blks = __cp_payload(sbi); 1689 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE); 1690 u64 kbytes_written; 1691 int err; 1692 1693 /* Flush all the NAT/SIT pages */ 1694 f2fs_sync_meta_pages(sbi, LONG_MAX, FS_CP_META_IO); 1695 1696 stat_cp_time(cpc, CP_TIME_SYNC_META); 1697 1698 /* start to update checkpoint, cp ver is already updated previously */ 1699 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true)); 1700 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi)); 1701 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) { 1702 struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_NODE); 1703 1704 ckpt->cur_node_segno[i] = cpu_to_le32(curseg->segno); 1705 ckpt->cur_node_blkoff[i] = cpu_to_le16(curseg->next_blkoff); 1706 ckpt->alloc_type[i + CURSEG_HOT_NODE] = curseg->alloc_type; 1707 } 1708 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) { 1709 struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_DATA); 1710 1711 ckpt->cur_data_segno[i] = cpu_to_le32(curseg->segno); 1712 ckpt->cur_data_blkoff[i] = cpu_to_le16(curseg->next_blkoff); 1713 ckpt->alloc_type[i + CURSEG_HOT_DATA] = curseg->alloc_type; 1714 } 1715 1716 /* 2 cp + n data seg summary + orphan inode blocks */ 1717 data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false); 1718 spin_lock_irqsave(&sbi->cp_lock, flags); 1719 if (data_sum_blocks < NR_CURSEG_DATA_TYPE) 1720 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); 1721 else 1722 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); 1723 spin_unlock_irqrestore(&sbi->cp_lock, flags); 1724 1725 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num); 1726 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks + 1727 orphan_blocks); 1728 1729 if (__remain_node_summaries(cpc->reason)) 1730 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS + 1731 cp_payload_blks + data_sum_blocks + 1732 orphan_blocks + NR_CURSEG_NODE_TYPE); 1733 else 1734 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS + 1735 cp_payload_blks + data_sum_blocks + 1736 orphan_blocks); 1737 1738 /* update ckpt flag for checkpoint */ 1739 update_ckpt_flags(sbi, cpc); 1740 1741 /* update SIT/NAT bitmap */ 1742 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP)); 1743 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP)); 1744 1745 crc32 = f2fs_checkpoint_chksum(ckpt); 1746 *((__le32 *)((unsigned char *)ckpt + 1747 le32_to_cpu(ckpt->checksum_offset))) 1748 = cpu_to_le32(crc32); 1749 1750 start_blk = __start_cp_next_addr(sbi); 1751 1752 /* write nat bits */ 1753 if (enabled_nat_bits(sbi, cpc)) { 1754 __u64 cp_ver = cur_cp_version(ckpt); 1755 block_t blk; 1756 1757 cp_ver |= ((__u64)crc32 << 32); 1758 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver); 1759 1760 blk = start_blk + BLKS_PER_SEG(sbi) - nm_i->nat_bits_blocks; 1761 for (i = 0; i < nm_i->nat_bits_blocks; i++) 1762 f2fs_update_meta_page(sbi, nm_i->nat_bits + 1763 F2FS_BLK_TO_BYTES(i), blk + i); 1764 } 1765 1766 /* write out checkpoint buffer at block 0 */ 1767 f2fs_update_meta_page(sbi, ckpt, start_blk++); 1768 1769 for (i = 1; i < 1 + cp_payload_blks; i++) 1770 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE, 1771 start_blk++); 1772 1773 if (orphan_num) { 1774 write_orphan_inodes(sbi, start_blk); 1775 start_blk += orphan_blocks; 1776 } 1777 1778 f2fs_write_data_summaries(sbi, start_blk); 1779 start_blk += data_sum_blocks; 1780 1781 /* Record write statistics in the hot node summary */ 1782 kbytes_written = sbi->kbytes_written; 1783 kbytes_written += (f2fs_get_sectors_written(sbi) - 1784 sbi->sectors_written_start) >> 1; 1785 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written); 1786 1787 if (__remain_node_summaries(cpc->reason)) { 1788 f2fs_write_node_summaries(sbi, start_blk); 1789 start_blk += NR_CURSEG_NODE_TYPE; 1790 } 1791 1792 /* Here, we have one bio having CP pack except cp pack 2 page */ 1793 f2fs_sync_meta_pages(sbi, LONG_MAX, FS_CP_META_IO); 1794 stat_cp_time(cpc, CP_TIME_SYNC_CP_META); 1795 1796 /* Wait for all dirty meta pages to be submitted for IO */ 1797 f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META); 1798 stat_cp_time(cpc, CP_TIME_WAIT_DIRTY_META); 1799 1800 /* wait for previous submitted meta pages writeback */ 1801 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA); 1802 stat_cp_time(cpc, CP_TIME_WAIT_CP_DATA); 1803 1804 /* flush all device cache */ 1805 err = f2fs_flush_device_cache(sbi); 1806 if (err) 1807 return err; 1808 stat_cp_time(cpc, CP_TIME_FLUSH_DEVICE); 1809 1810 /* barrier and flush checkpoint cp pack 2 page if it can */ 1811 commit_checkpoint(sbi, ckpt, start_blk); 1812 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA); 1813 stat_cp_time(cpc, CP_TIME_WAIT_LAST_CP); 1814 1815 /* 1816 * invalidate intermediate page cache borrowed from meta inode which are 1817 * used for migration of encrypted, verity or compressed inode's blocks. 1818 */ 1819 if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) || 1820 f2fs_sb_has_compression(sbi)) 1821 f2fs_bug_on(sbi, 1822 invalidate_inode_pages2_range(META_MAPPING(sbi), 1823 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1)); 1824 1825 f2fs_release_ino_entry(sbi, false); 1826 1827 f2fs_reset_fsync_node_info(sbi); 1828 1829 clear_sbi_flag(sbi, SBI_IS_DIRTY); 1830 clear_sbi_flag(sbi, SBI_NEED_CP); 1831 clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH); 1832 1833 spin_lock(&sbi->stat_lock); 1834 sbi->unusable_block_count = 0; 1835 spin_unlock(&sbi->stat_lock); 1836 1837 __set_cp_next_pack(sbi); 1838 1839 /* 1840 * redirty superblock if metadata like node page or inode cache is 1841 * updated during writing checkpoint. 1842 */ 1843 if (get_pages(sbi, F2FS_DIRTY_NODES) || 1844 get_pages(sbi, F2FS_DIRTY_IMETA)) 1845 set_sbi_flag(sbi, SBI_IS_DIRTY); 1846 1847 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS)); 1848 1849 return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0; 1850 } 1851 1852 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc) 1853 { 1854 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 1855 struct f2fs_lock_context lc; 1856 unsigned long long ckpt_ver; 1857 int err = 0; 1858 1859 stat_cp_time(cpc, CP_TIME_START); 1860 1861 if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi)) 1862 return -EROFS; 1863 1864 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 1865 if (cpc->reason != CP_PAUSE) 1866 return 0; 1867 f2fs_warn(sbi, "Start checkpoint disabled!"); 1868 } 1869 if (cpc->reason != CP_RESIZE) 1870 f2fs_down_write_trace(&sbi->cp_global_sem, &lc); 1871 1872 stat_cp_time(cpc, CP_TIME_LOCK); 1873 1874 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) && 1875 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) || 1876 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks))) 1877 goto out; 1878 if (unlikely(f2fs_cp_error(sbi))) { 1879 err = -EIO; 1880 goto out; 1881 } 1882 1883 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, CP_PHASE_START_BLOCK_OPS); 1884 1885 err = block_operations(sbi); 1886 if (err) 1887 goto out; 1888 1889 stat_cp_time(cpc, CP_TIME_OP_LOCK); 1890 1891 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, CP_PHASE_FINISH_BLOCK_OPS); 1892 1893 f2fs_flush_merged_writes(sbi); 1894 1895 /* this is the case of multiple fstrims without any changes */ 1896 if (cpc->reason & CP_DISCARD) { 1897 if (!f2fs_exist_trim_candidates(sbi, cpc)) { 1898 unblock_operations(sbi); 1899 goto out; 1900 } 1901 1902 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 && 1903 SIT_I(sbi)->dirty_sentries == 0 && 1904 prefree_segments(sbi) == 0) { 1905 f2fs_flush_sit_entries(sbi, cpc); 1906 f2fs_clear_prefree_segments(sbi, cpc); 1907 unblock_operations(sbi); 1908 goto out; 1909 } 1910 } 1911 stat_cp_time(cpc, CP_TIME_MERGE_WRITE); 1912 1913 /* 1914 * update checkpoint pack index 1915 * Increase the version number so that 1916 * SIT entries and seg summaries are written at correct place 1917 */ 1918 ckpt_ver = cur_cp_version(ckpt); 1919 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver); 1920 1921 /* write cached NAT/SIT entries to NAT/SIT area */ 1922 err = f2fs_flush_nat_entries(sbi, cpc); 1923 if (err) { 1924 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err); 1925 f2fs_bug_on(sbi, !f2fs_cp_error(sbi)); 1926 goto stop; 1927 } 1928 stat_cp_time(cpc, CP_TIME_FLUSH_NAT); 1929 1930 f2fs_flush_sit_entries(sbi, cpc); 1931 1932 stat_cp_time(cpc, CP_TIME_FLUSH_SIT); 1933 1934 /* save inmem log status */ 1935 f2fs_save_inmem_curseg(sbi); 1936 1937 err = do_checkpoint(sbi, cpc); 1938 if (err) { 1939 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err); 1940 f2fs_bug_on(sbi, !f2fs_cp_error(sbi)); 1941 f2fs_release_discard_addrs(sbi); 1942 } else { 1943 f2fs_clear_prefree_segments(sbi, cpc); 1944 } 1945 1946 f2fs_restore_inmem_curseg(sbi); 1947 f2fs_reinit_atgc_curseg(sbi); 1948 stat_inc_cp_count(sbi); 1949 stop: 1950 unblock_operations(sbi); 1951 stat_cp_time(cpc, CP_TIME_END); 1952 check_cp_time(sbi, cpc); 1953 1954 if (cpc->reason & CP_RECOVERY) 1955 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver); 1956 1957 /* update CP_TIME to trigger checkpoint periodically */ 1958 f2fs_update_time(sbi, CP_TIME); 1959 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, CP_PHASE_FINISH_CHECKPOINT); 1960 out: 1961 if (cpc->reason != CP_RESIZE) 1962 f2fs_up_write_trace(&sbi->cp_global_sem, &lc); 1963 return err; 1964 } 1965 1966 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi) 1967 { 1968 int i; 1969 1970 for (i = 0; i < MAX_INO_ENTRY; i++) { 1971 struct inode_management *im = &sbi->im[i]; 1972 1973 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC); 1974 spin_lock_init(&im->ino_lock); 1975 INIT_LIST_HEAD(&im->ino_list); 1976 im->ino_num = 0; 1977 } 1978 1979 sbi->max_orphans = (BLKS_PER_SEG(sbi) - F2FS_CP_PACKS - 1980 NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) * 1981 F2FS_ORPHANS_PER_BLOCK; 1982 } 1983 1984 int __init f2fs_create_checkpoint_caches(void) 1985 { 1986 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry", 1987 sizeof(struct ino_entry)); 1988 if (!ino_entry_slab) 1989 return -ENOMEM; 1990 f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry", 1991 sizeof(struct inode_entry)); 1992 if (!f2fs_inode_entry_slab) { 1993 kmem_cache_destroy(ino_entry_slab); 1994 return -ENOMEM; 1995 } 1996 return 0; 1997 } 1998 1999 void f2fs_destroy_checkpoint_caches(void) 2000 { 2001 kmem_cache_destroy(ino_entry_slab); 2002 kmem_cache_destroy(f2fs_inode_entry_slab); 2003 } 2004 2005 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi) 2006 { 2007 struct cp_control cpc = { .reason = CP_SYNC, }; 2008 struct f2fs_lock_context lc; 2009 int err; 2010 2011 f2fs_down_write_trace(&sbi->gc_lock, &lc); 2012 err = f2fs_write_checkpoint(sbi, &cpc); 2013 f2fs_up_write_trace(&sbi->gc_lock, &lc); 2014 2015 return err; 2016 } 2017 2018 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi) 2019 { 2020 struct ckpt_req_control *cprc = &sbi->cprc_info; 2021 struct ckpt_req *req, *next; 2022 struct llist_node *dispatch_list; 2023 u64 sum_diff = 0, diff, count = 0; 2024 int ret; 2025 2026 dispatch_list = llist_del_all(&cprc->issue_list); 2027 if (!dispatch_list) 2028 return; 2029 dispatch_list = llist_reverse_order(dispatch_list); 2030 2031 ret = __write_checkpoint_sync(sbi); 2032 atomic_inc(&cprc->issued_ckpt); 2033 2034 llist_for_each_entry_safe(req, next, dispatch_list, llnode) { 2035 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time); 2036 req->ret = ret; 2037 req->delta_time = diff; 2038 complete(&req->wait); 2039 2040 sum_diff += diff; 2041 count++; 2042 } 2043 atomic_sub(count, &cprc->queued_ckpt); 2044 atomic_add(count, &cprc->total_ckpt); 2045 2046 spin_lock(&cprc->stat_lock); 2047 cprc->cur_time = (unsigned int)div64_u64(sum_diff, count); 2048 if (cprc->peak_time < cprc->cur_time) 2049 cprc->peak_time = cprc->cur_time; 2050 spin_unlock(&cprc->stat_lock); 2051 } 2052 2053 static int issue_checkpoint_thread(void *data) 2054 { 2055 struct f2fs_sb_info *sbi = data; 2056 struct ckpt_req_control *cprc = &sbi->cprc_info; 2057 wait_queue_head_t *q = &cprc->ckpt_wait_queue; 2058 repeat: 2059 if (kthread_should_stop()) 2060 return 0; 2061 2062 if (!llist_empty(&cprc->issue_list)) 2063 __checkpoint_and_complete_reqs(sbi); 2064 2065 wait_event_interruptible(*q, 2066 kthread_should_stop() || !llist_empty(&cprc->issue_list)); 2067 goto repeat; 2068 } 2069 2070 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi, 2071 struct ckpt_req *wait_req) 2072 { 2073 struct ckpt_req_control *cprc = &sbi->cprc_info; 2074 2075 if (!llist_empty(&cprc->issue_list)) { 2076 __checkpoint_and_complete_reqs(sbi); 2077 } else { 2078 /* already dispatched by issue_checkpoint_thread */ 2079 if (wait_req) 2080 wait_for_completion(&wait_req->wait); 2081 } 2082 } 2083 2084 static void init_ckpt_req(struct ckpt_req *req) 2085 { 2086 memset(req, 0, sizeof(struct ckpt_req)); 2087 2088 init_completion(&req->wait); 2089 req->queue_time = ktime_get(); 2090 } 2091 2092 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi) 2093 { 2094 struct ckpt_req_control *cprc = &sbi->cprc_info; 2095 struct ckpt_req req; 2096 struct cp_control cpc; 2097 2098 cpc.reason = __get_cp_reason(sbi); 2099 if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC || 2100 sbi->umount_lock_holder == current) { 2101 struct f2fs_lock_context lc; 2102 int ret; 2103 2104 f2fs_down_write_trace(&sbi->gc_lock, &lc); 2105 ret = f2fs_write_checkpoint(sbi, &cpc); 2106 f2fs_up_write_trace(&sbi->gc_lock, &lc); 2107 2108 return ret; 2109 } 2110 2111 if (!cprc->f2fs_issue_ckpt) 2112 return __write_checkpoint_sync(sbi); 2113 2114 init_ckpt_req(&req); 2115 2116 llist_add(&req.llnode, &cprc->issue_list); 2117 atomic_inc(&cprc->queued_ckpt); 2118 2119 /* 2120 * update issue_list before we wake up issue_checkpoint thread, 2121 * this smp_mb() pairs with another barrier in ___wait_event(), 2122 * see more details in comments of waitqueue_active(). 2123 */ 2124 smp_mb(); 2125 2126 if (waitqueue_active(&cprc->ckpt_wait_queue)) 2127 wake_up(&cprc->ckpt_wait_queue); 2128 2129 if (cprc->f2fs_issue_ckpt) 2130 wait_for_completion(&req.wait); 2131 else 2132 flush_remained_ckpt_reqs(sbi, &req); 2133 2134 if (unlikely(req.delta_time >= CP_LONG_LATENCY_THRESHOLD)) { 2135 f2fs_warn_ratelimited(sbi, 2136 "blocked on checkpoint for %u ms", cprc->peak_time); 2137 dump_stack(); 2138 } 2139 2140 return req.ret; 2141 } 2142 2143 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi) 2144 { 2145 dev_t dev = sbi->sb->s_bdev->bd_dev; 2146 struct ckpt_req_control *cprc = &sbi->cprc_info; 2147 2148 if (cprc->f2fs_issue_ckpt) 2149 return 0; 2150 2151 cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi, 2152 "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev)); 2153 if (IS_ERR(cprc->f2fs_issue_ckpt)) { 2154 int err = PTR_ERR(cprc->f2fs_issue_ckpt); 2155 2156 cprc->f2fs_issue_ckpt = NULL; 2157 return err; 2158 } 2159 2160 set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio); 2161 set_user_nice(cprc->f2fs_issue_ckpt, 2162 PRIO_TO_NICE(sbi->critical_task_priority)); 2163 2164 return 0; 2165 } 2166 2167 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi) 2168 { 2169 struct ckpt_req_control *cprc = &sbi->cprc_info; 2170 struct task_struct *ckpt_task; 2171 2172 if (!cprc->f2fs_issue_ckpt) 2173 return; 2174 2175 ckpt_task = cprc->f2fs_issue_ckpt; 2176 cprc->f2fs_issue_ckpt = NULL; 2177 kthread_stop(ckpt_task); 2178 2179 f2fs_flush_ckpt_thread(sbi); 2180 } 2181 2182 void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi) 2183 { 2184 struct ckpt_req_control *cprc = &sbi->cprc_info; 2185 2186 flush_remained_ckpt_reqs(sbi, NULL); 2187 2188 /* Let's wait for the previous dispatched checkpoint. */ 2189 while (atomic_read(&cprc->queued_ckpt)) 2190 io_schedule_timeout(DEFAULT_SCHEDULE_TIMEOUT); 2191 } 2192 2193 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi) 2194 { 2195 struct ckpt_req_control *cprc = &sbi->cprc_info; 2196 2197 atomic_set(&cprc->issued_ckpt, 0); 2198 atomic_set(&cprc->total_ckpt, 0); 2199 atomic_set(&cprc->queued_ckpt, 0); 2200 cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO; 2201 init_waitqueue_head(&cprc->ckpt_wait_queue); 2202 init_llist_head(&cprc->issue_list); 2203 spin_lock_init(&cprc->stat_lock); 2204 } 2205