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