1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * fs/f2fs/gc.c 4 * 5 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 6 * http://www.samsung.com/ 7 */ 8 #include <linux/fs.h> 9 #include <linux/module.h> 10 #include <linux/init.h> 11 #include <linux/f2fs_fs.h> 12 #include <linux/kthread.h> 13 #include <linux/delay.h> 14 #include <linux/freezer.h> 15 #include <linux/sched/signal.h> 16 #include <linux/random.h> 17 #include <linux/sched/mm.h> 18 19 #include "f2fs.h" 20 #include "node.h" 21 #include "segment.h" 22 #include "gc.h" 23 #include "iostat.h" 24 #include <trace/events/f2fs.h> 25 26 static struct kmem_cache *victim_entry_slab; 27 28 static unsigned int count_bits(const unsigned long *addr, 29 unsigned int offset, unsigned int len); 30 31 static int gc_thread_func(void *data) 32 { 33 struct f2fs_sb_info *sbi = data; 34 struct f2fs_gc_kthread *gc_th = sbi->gc_thread; 35 wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; 36 wait_queue_head_t *fggc_wq = &sbi->gc_thread->fggc_wq; 37 unsigned int wait_ms; 38 struct f2fs_gc_control gc_control = { 39 .victim_segno = NULL_SEGNO, 40 .should_migrate_blocks = false, 41 .err_gc_skipped = false }; 42 43 wait_ms = gc_th->min_sleep_time; 44 45 set_freezable(); 46 do { 47 bool sync_mode, foreground = false; 48 49 wait_event_freezable_timeout(*wq, 50 kthread_should_stop() || 51 waitqueue_active(fggc_wq) || 52 gc_th->gc_wake, 53 msecs_to_jiffies(wait_ms)); 54 55 if (test_opt(sbi, GC_MERGE) && waitqueue_active(fggc_wq)) 56 foreground = true; 57 58 /* give it a try one time */ 59 if (gc_th->gc_wake) 60 gc_th->gc_wake = false; 61 62 if (f2fs_readonly(sbi->sb)) { 63 stat_other_skip_bggc_count(sbi); 64 continue; 65 } 66 if (kthread_should_stop()) 67 break; 68 69 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { 70 increase_sleep_time(gc_th, &wait_ms); 71 stat_other_skip_bggc_count(sbi); 72 continue; 73 } 74 75 if (time_to_inject(sbi, FAULT_CHECKPOINT)) 76 f2fs_stop_checkpoint(sbi, false, 77 STOP_CP_REASON_FAULT_INJECT); 78 79 if (!sb_start_write_trylock(sbi->sb)) { 80 stat_other_skip_bggc_count(sbi); 81 continue; 82 } 83 84 gc_control.one_time = false; 85 86 /* 87 * [GC triggering condition] 88 * 0. GC is not conducted currently. 89 * 1. There are enough dirty segments. 90 * 2. IO subsystem is idle by checking the # of writeback pages. 91 * 3. IO subsystem is idle by checking the # of requests in 92 * bdev's request list. 93 * 94 * Note) We have to avoid triggering GCs frequently. 95 * Because it is possible that some segments can be 96 * invalidated soon after by user update or deletion. 97 * So, I'd like to wait some time to collect dirty segments. 98 */ 99 if (sbi->gc_mode == GC_URGENT_HIGH || 100 sbi->gc_mode == GC_URGENT_MID) { 101 wait_ms = gc_th->urgent_sleep_time; 102 f2fs_down_write(&sbi->gc_lock); 103 goto do_gc; 104 } 105 106 if (foreground) { 107 f2fs_down_write(&sbi->gc_lock); 108 goto do_gc; 109 } else if (!f2fs_down_write_trylock(&sbi->gc_lock)) { 110 stat_other_skip_bggc_count(sbi); 111 goto next; 112 } 113 114 if (!is_idle(sbi, GC_TIME)) { 115 increase_sleep_time(gc_th, &wait_ms); 116 f2fs_up_write(&sbi->gc_lock); 117 stat_io_skip_bggc_count(sbi); 118 goto next; 119 } 120 121 if (f2fs_sb_has_blkzoned(sbi)) { 122 if (has_enough_free_blocks(sbi, 123 gc_th->no_zoned_gc_percent)) { 124 wait_ms = gc_th->no_gc_sleep_time; 125 f2fs_up_write(&sbi->gc_lock); 126 goto next; 127 } 128 if (wait_ms == gc_th->no_gc_sleep_time) 129 wait_ms = gc_th->max_sleep_time; 130 } 131 132 if (need_to_boost_gc(sbi)) { 133 decrease_sleep_time(gc_th, &wait_ms); 134 if (f2fs_sb_has_blkzoned(sbi)) 135 gc_control.one_time = true; 136 } else { 137 increase_sleep_time(gc_th, &wait_ms); 138 } 139 do_gc: 140 stat_inc_gc_call_count(sbi, foreground ? 141 FOREGROUND : BACKGROUND); 142 143 sync_mode = (F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC) || 144 (gc_control.one_time && gc_th->boost_gc_greedy); 145 146 /* foreground GC was been triggered via f2fs_balance_fs() */ 147 if (foreground && !f2fs_sb_has_blkzoned(sbi)) 148 sync_mode = false; 149 150 gc_control.init_gc_type = sync_mode ? FG_GC : BG_GC; 151 gc_control.no_bg_gc = foreground; 152 gc_control.nr_free_secs = foreground ? 1 : 0; 153 154 /* if return value is not zero, no victim was selected */ 155 if (f2fs_gc(sbi, &gc_control)) { 156 /* don't bother wait_ms by foreground gc */ 157 if (!foreground) 158 wait_ms = gc_th->no_gc_sleep_time; 159 } else { 160 /* reset wait_ms to default sleep time */ 161 if (wait_ms == gc_th->no_gc_sleep_time) 162 wait_ms = gc_th->min_sleep_time; 163 } 164 165 if (foreground) 166 wake_up_all(&gc_th->fggc_wq); 167 168 trace_f2fs_background_gc(sbi->sb, wait_ms, 169 prefree_segments(sbi), free_segments(sbi)); 170 171 /* balancing f2fs's metadata periodically */ 172 f2fs_balance_fs_bg(sbi, true); 173 next: 174 if (sbi->gc_mode != GC_NORMAL) { 175 spin_lock(&sbi->gc_remaining_trials_lock); 176 if (sbi->gc_remaining_trials) { 177 sbi->gc_remaining_trials--; 178 if (!sbi->gc_remaining_trials) 179 sbi->gc_mode = GC_NORMAL; 180 } 181 spin_unlock(&sbi->gc_remaining_trials_lock); 182 } 183 sb_end_write(sbi->sb); 184 185 } while (!kthread_should_stop()); 186 return 0; 187 } 188 189 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi) 190 { 191 struct f2fs_gc_kthread *gc_th; 192 dev_t dev = sbi->sb->s_bdev->bd_dev; 193 194 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL); 195 if (!gc_th) 196 return -ENOMEM; 197 198 gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME; 199 gc_th->valid_thresh_ratio = DEF_GC_THREAD_VALID_THRESH_RATIO; 200 gc_th->boost_gc_multiple = BOOST_GC_MULTIPLE; 201 gc_th->boost_gc_greedy = GC_GREEDY; 202 203 if (f2fs_sb_has_blkzoned(sbi)) { 204 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME_ZONED; 205 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME_ZONED; 206 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME_ZONED; 207 gc_th->no_zoned_gc_percent = LIMIT_NO_ZONED_GC; 208 gc_th->boost_zoned_gc_percent = LIMIT_BOOST_ZONED_GC; 209 } else { 210 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; 211 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; 212 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; 213 gc_th->no_zoned_gc_percent = 0; 214 gc_th->boost_zoned_gc_percent = 0; 215 } 216 217 gc_th->gc_wake = false; 218 219 sbi->gc_thread = gc_th; 220 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); 221 init_waitqueue_head(&sbi->gc_thread->fggc_wq); 222 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, 223 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); 224 if (IS_ERR(gc_th->f2fs_gc_task)) { 225 int err = PTR_ERR(gc_th->f2fs_gc_task); 226 227 kfree(gc_th); 228 sbi->gc_thread = NULL; 229 return err; 230 } 231 232 return 0; 233 } 234 235 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi) 236 { 237 struct f2fs_gc_kthread *gc_th = sbi->gc_thread; 238 239 if (!gc_th) 240 return; 241 kthread_stop(gc_th->f2fs_gc_task); 242 wake_up_all(&gc_th->fggc_wq); 243 kfree(gc_th); 244 sbi->gc_thread = NULL; 245 } 246 247 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type) 248 { 249 int gc_mode; 250 251 if (gc_type == BG_GC) { 252 if (sbi->am.atgc_enabled) 253 gc_mode = GC_AT; 254 else 255 gc_mode = GC_CB; 256 } else { 257 gc_mode = GC_GREEDY; 258 } 259 260 switch (sbi->gc_mode) { 261 case GC_IDLE_CB: 262 case GC_URGENT_LOW: 263 case GC_URGENT_MID: 264 gc_mode = GC_CB; 265 break; 266 case GC_IDLE_GREEDY: 267 case GC_URGENT_HIGH: 268 gc_mode = GC_GREEDY; 269 break; 270 case GC_IDLE_AT: 271 gc_mode = GC_AT; 272 break; 273 } 274 275 return gc_mode; 276 } 277 278 static void select_policy(struct f2fs_sb_info *sbi, int gc_type, 279 int type, struct victim_sel_policy *p) 280 { 281 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 282 283 if (p->alloc_mode == SSR || p->alloc_mode == AT_SSR) { 284 p->gc_mode = GC_GREEDY; 285 p->dirty_bitmap = dirty_i->dirty_segmap[type]; 286 p->max_search = dirty_i->nr_dirty[type]; 287 p->ofs_unit = 1; 288 } else { 289 p->gc_mode = select_gc_type(sbi, gc_type); 290 p->ofs_unit = SEGS_PER_SEC(sbi); 291 if (__is_large_section(sbi)) { 292 p->dirty_bitmap = dirty_i->dirty_secmap; 293 p->max_search = count_bits(p->dirty_bitmap, 294 0, MAIN_SECS(sbi)); 295 } else { 296 p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY]; 297 p->max_search = dirty_i->nr_dirty[DIRTY]; 298 } 299 } 300 301 /* 302 * adjust candidates range, should select all dirty segments for 303 * foreground GC and urgent GC cases. 304 */ 305 if (gc_type != FG_GC && 306 (sbi->gc_mode != GC_URGENT_HIGH) && 307 (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) && 308 p->max_search > sbi->max_victim_search) 309 p->max_search = sbi->max_victim_search; 310 311 /* let's select beginning hot/small space first. */ 312 if (f2fs_need_rand_seg(sbi)) 313 p->offset = get_random_u32_below(MAIN_SECS(sbi) * 314 SEGS_PER_SEC(sbi)); 315 else if (type == CURSEG_HOT_DATA || IS_NODESEG(type)) 316 p->offset = 0; 317 else 318 p->offset = SIT_I(sbi)->last_victim[p->gc_mode]; 319 } 320 321 static unsigned int get_max_cost(struct f2fs_sb_info *sbi, 322 struct victim_sel_policy *p) 323 { 324 /* SSR allocates in a segment unit */ 325 if (p->alloc_mode == SSR) 326 return BLKS_PER_SEG(sbi); 327 else if (p->alloc_mode == AT_SSR) 328 return UINT_MAX; 329 330 /* LFS */ 331 if (p->gc_mode == GC_GREEDY) 332 return SEGS_TO_BLKS(sbi, 2 * p->ofs_unit); 333 else if (p->gc_mode == GC_CB) 334 return UINT_MAX; 335 else if (p->gc_mode == GC_AT) 336 return UINT_MAX; 337 else /* No other gc_mode */ 338 return 0; 339 } 340 341 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) 342 { 343 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 344 unsigned int secno; 345 346 /* 347 * If the gc_type is FG_GC, we can select victim segments 348 * selected by background GC before. 349 * Those segments guarantee they have small valid blocks. 350 */ 351 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { 352 if (sec_usage_check(sbi, secno)) 353 continue; 354 clear_bit(secno, dirty_i->victim_secmap); 355 return GET_SEG_FROM_SEC(sbi, secno); 356 } 357 return NULL_SEGNO; 358 } 359 360 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) 361 { 362 struct sit_info *sit_i = SIT_I(sbi); 363 unsigned long long mtime = 0; 364 unsigned int vblocks; 365 unsigned char age = 0; 366 unsigned char u; 367 unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi); 368 369 mtime = f2fs_get_section_mtime(sbi, segno); 370 f2fs_bug_on(sbi, mtime == INVALID_MTIME); 371 vblocks = get_valid_blocks(sbi, segno, true); 372 vblocks = div_u64(vblocks, usable_segs_per_sec); 373 374 u = BLKS_TO_SEGS(sbi, vblocks * 100); 375 376 /* Handle if the system time has changed by the user */ 377 if (mtime < sit_i->min_mtime) 378 sit_i->min_mtime = mtime; 379 if (mtime > sit_i->max_mtime) 380 sit_i->max_mtime = mtime; 381 if (sit_i->max_mtime != sit_i->min_mtime) 382 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), 383 sit_i->max_mtime - sit_i->min_mtime); 384 385 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); 386 } 387 388 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, 389 unsigned int segno, struct victim_sel_policy *p, 390 unsigned int valid_thresh_ratio) 391 { 392 if (p->alloc_mode == SSR) 393 return get_seg_entry(sbi, segno)->ckpt_valid_blocks; 394 395 if (p->one_time_gc && (valid_thresh_ratio < 100) && 396 (get_valid_blocks(sbi, segno, true) >= 397 CAP_BLKS_PER_SEC(sbi) * valid_thresh_ratio / 100)) 398 return UINT_MAX; 399 400 /* alloc_mode == LFS */ 401 if (p->gc_mode == GC_GREEDY) 402 return get_valid_blocks(sbi, segno, true); 403 else if (p->gc_mode == GC_CB) 404 return get_cb_cost(sbi, segno); 405 406 f2fs_bug_on(sbi, 1); 407 return 0; 408 } 409 410 static unsigned int count_bits(const unsigned long *addr, 411 unsigned int offset, unsigned int len) 412 { 413 unsigned int end = offset + len, sum = 0; 414 415 while (offset < end) { 416 if (test_bit(offset++, addr)) 417 ++sum; 418 } 419 return sum; 420 } 421 422 static bool f2fs_check_victim_tree(struct f2fs_sb_info *sbi, 423 struct rb_root_cached *root) 424 { 425 #ifdef CONFIG_F2FS_CHECK_FS 426 struct rb_node *cur = rb_first_cached(root), *next; 427 struct victim_entry *cur_ve, *next_ve; 428 429 while (cur) { 430 next = rb_next(cur); 431 if (!next) 432 return true; 433 434 cur_ve = rb_entry(cur, struct victim_entry, rb_node); 435 next_ve = rb_entry(next, struct victim_entry, rb_node); 436 437 if (cur_ve->mtime > next_ve->mtime) { 438 f2fs_info(sbi, "broken victim_rbtree, " 439 "cur_mtime(%llu) next_mtime(%llu)", 440 cur_ve->mtime, next_ve->mtime); 441 return false; 442 } 443 cur = next; 444 } 445 #endif 446 return true; 447 } 448 449 static struct victim_entry *__lookup_victim_entry(struct f2fs_sb_info *sbi, 450 unsigned long long mtime) 451 { 452 struct atgc_management *am = &sbi->am; 453 struct rb_node *node = am->root.rb_root.rb_node; 454 struct victim_entry *ve = NULL; 455 456 while (node) { 457 ve = rb_entry(node, struct victim_entry, rb_node); 458 459 if (mtime < ve->mtime) 460 node = node->rb_left; 461 else 462 node = node->rb_right; 463 } 464 return ve; 465 } 466 467 static struct victim_entry *__create_victim_entry(struct f2fs_sb_info *sbi, 468 unsigned long long mtime, unsigned int segno) 469 { 470 struct atgc_management *am = &sbi->am; 471 struct victim_entry *ve; 472 473 ve = f2fs_kmem_cache_alloc(victim_entry_slab, GFP_NOFS, true, NULL); 474 475 ve->mtime = mtime; 476 ve->segno = segno; 477 478 list_add_tail(&ve->list, &am->victim_list); 479 am->victim_count++; 480 481 return ve; 482 } 483 484 static void __insert_victim_entry(struct f2fs_sb_info *sbi, 485 unsigned long long mtime, unsigned int segno) 486 { 487 struct atgc_management *am = &sbi->am; 488 struct rb_root_cached *root = &am->root; 489 struct rb_node **p = &root->rb_root.rb_node; 490 struct rb_node *parent = NULL; 491 struct victim_entry *ve; 492 bool left_most = true; 493 494 /* look up rb tree to find parent node */ 495 while (*p) { 496 parent = *p; 497 ve = rb_entry(parent, struct victim_entry, rb_node); 498 499 if (mtime < ve->mtime) { 500 p = &(*p)->rb_left; 501 } else { 502 p = &(*p)->rb_right; 503 left_most = false; 504 } 505 } 506 507 ve = __create_victim_entry(sbi, mtime, segno); 508 509 rb_link_node(&ve->rb_node, parent, p); 510 rb_insert_color_cached(&ve->rb_node, root, left_most); 511 } 512 513 static void add_victim_entry(struct f2fs_sb_info *sbi, 514 struct victim_sel_policy *p, unsigned int segno) 515 { 516 struct sit_info *sit_i = SIT_I(sbi); 517 unsigned long long mtime = 0; 518 519 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 520 if (p->gc_mode == GC_AT && 521 get_valid_blocks(sbi, segno, true) == 0) 522 return; 523 } 524 525 mtime = f2fs_get_section_mtime(sbi, segno); 526 f2fs_bug_on(sbi, mtime == INVALID_MTIME); 527 528 /* Handle if the system time has changed by the user */ 529 if (mtime < sit_i->min_mtime) 530 sit_i->min_mtime = mtime; 531 if (mtime > sit_i->max_mtime) 532 sit_i->max_mtime = mtime; 533 if (mtime < sit_i->dirty_min_mtime) 534 sit_i->dirty_min_mtime = mtime; 535 if (mtime > sit_i->dirty_max_mtime) 536 sit_i->dirty_max_mtime = mtime; 537 538 /* don't choose young section as candidate */ 539 if (sit_i->dirty_max_mtime - mtime < p->age_threshold) 540 return; 541 542 __insert_victim_entry(sbi, mtime, segno); 543 } 544 545 static void atgc_lookup_victim(struct f2fs_sb_info *sbi, 546 struct victim_sel_policy *p) 547 { 548 struct sit_info *sit_i = SIT_I(sbi); 549 struct atgc_management *am = &sbi->am; 550 struct rb_root_cached *root = &am->root; 551 struct rb_node *node; 552 struct victim_entry *ve; 553 unsigned long long total_time; 554 unsigned long long age, u, accu; 555 unsigned long long max_mtime = sit_i->dirty_max_mtime; 556 unsigned long long min_mtime = sit_i->dirty_min_mtime; 557 unsigned int sec_blocks = CAP_BLKS_PER_SEC(sbi); 558 unsigned int vblocks; 559 unsigned int dirty_threshold = max(am->max_candidate_count, 560 am->candidate_ratio * 561 am->victim_count / 100); 562 unsigned int age_weight = am->age_weight; 563 unsigned int cost; 564 unsigned int iter = 0; 565 566 if (max_mtime < min_mtime) 567 return; 568 569 max_mtime += 1; 570 total_time = max_mtime - min_mtime; 571 572 accu = div64_u64(ULLONG_MAX, total_time); 573 accu = min_t(unsigned long long, div_u64(accu, 100), 574 DEFAULT_ACCURACY_CLASS); 575 576 node = rb_first_cached(root); 577 next: 578 ve = rb_entry_safe(node, struct victim_entry, rb_node); 579 if (!ve) 580 return; 581 582 if (ve->mtime >= max_mtime || ve->mtime < min_mtime) 583 goto skip; 584 585 /* age = 10000 * x% * 60 */ 586 age = div64_u64(accu * (max_mtime - ve->mtime), total_time) * 587 age_weight; 588 589 vblocks = get_valid_blocks(sbi, ve->segno, true); 590 f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks); 591 592 /* u = 10000 * x% * 40 */ 593 u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) * 594 (100 - age_weight); 595 596 f2fs_bug_on(sbi, age + u >= UINT_MAX); 597 598 cost = UINT_MAX - (age + u); 599 iter++; 600 601 if (cost < p->min_cost || 602 (cost == p->min_cost && age > p->oldest_age)) { 603 p->min_cost = cost; 604 p->oldest_age = age; 605 p->min_segno = ve->segno; 606 } 607 skip: 608 if (iter < dirty_threshold) { 609 node = rb_next(node); 610 goto next; 611 } 612 } 613 614 /* 615 * select candidates around source section in range of 616 * [target - dirty_threshold, target + dirty_threshold] 617 */ 618 static void atssr_lookup_victim(struct f2fs_sb_info *sbi, 619 struct victim_sel_policy *p) 620 { 621 struct sit_info *sit_i = SIT_I(sbi); 622 struct atgc_management *am = &sbi->am; 623 struct victim_entry *ve; 624 unsigned long long age; 625 unsigned long long max_mtime = sit_i->dirty_max_mtime; 626 unsigned long long min_mtime = sit_i->dirty_min_mtime; 627 unsigned int vblocks; 628 unsigned int dirty_threshold = max(am->max_candidate_count, 629 am->candidate_ratio * 630 am->victim_count / 100); 631 unsigned int cost, iter; 632 int stage = 0; 633 634 if (max_mtime < min_mtime) 635 return; 636 max_mtime += 1; 637 next_stage: 638 iter = 0; 639 ve = __lookup_victim_entry(sbi, p->age); 640 next_node: 641 if (!ve) { 642 if (stage++ == 0) 643 goto next_stage; 644 return; 645 } 646 647 if (ve->mtime >= max_mtime || ve->mtime < min_mtime) 648 goto skip_node; 649 650 age = max_mtime - ve->mtime; 651 652 vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks; 653 f2fs_bug_on(sbi, !vblocks); 654 655 /* rare case */ 656 if (vblocks == BLKS_PER_SEG(sbi)) 657 goto skip_node; 658 659 iter++; 660 661 age = max_mtime - abs(p->age - age); 662 cost = UINT_MAX - vblocks; 663 664 if (cost < p->min_cost || 665 (cost == p->min_cost && age > p->oldest_age)) { 666 p->min_cost = cost; 667 p->oldest_age = age; 668 p->min_segno = ve->segno; 669 } 670 skip_node: 671 if (iter < dirty_threshold) { 672 ve = rb_entry(stage == 0 ? rb_prev(&ve->rb_node) : 673 rb_next(&ve->rb_node), 674 struct victim_entry, rb_node); 675 goto next_node; 676 } 677 678 if (stage++ == 0) 679 goto next_stage; 680 } 681 682 static void lookup_victim_by_age(struct f2fs_sb_info *sbi, 683 struct victim_sel_policy *p) 684 { 685 f2fs_bug_on(sbi, !f2fs_check_victim_tree(sbi, &sbi->am.root)); 686 687 if (p->gc_mode == GC_AT) 688 atgc_lookup_victim(sbi, p); 689 else if (p->alloc_mode == AT_SSR) 690 atssr_lookup_victim(sbi, p); 691 else 692 f2fs_bug_on(sbi, 1); 693 } 694 695 static void release_victim_entry(struct f2fs_sb_info *sbi) 696 { 697 struct atgc_management *am = &sbi->am; 698 struct victim_entry *ve, *tmp; 699 700 list_for_each_entry_safe(ve, tmp, &am->victim_list, list) { 701 list_del(&ve->list); 702 kmem_cache_free(victim_entry_slab, ve); 703 am->victim_count--; 704 } 705 706 am->root = RB_ROOT_CACHED; 707 708 f2fs_bug_on(sbi, am->victim_count); 709 f2fs_bug_on(sbi, !list_empty(&am->victim_list)); 710 } 711 712 static bool f2fs_pin_section(struct f2fs_sb_info *sbi, unsigned int segno) 713 { 714 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 715 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 716 717 if (!dirty_i->enable_pin_section) 718 return false; 719 if (!test_and_set_bit(secno, dirty_i->pinned_secmap)) 720 dirty_i->pinned_secmap_cnt++; 721 return true; 722 } 723 724 static bool f2fs_pinned_section_exists(struct dirty_seglist_info *dirty_i) 725 { 726 return dirty_i->pinned_secmap_cnt; 727 } 728 729 static bool f2fs_section_is_pinned(struct dirty_seglist_info *dirty_i, 730 unsigned int secno) 731 { 732 return dirty_i->enable_pin_section && 733 f2fs_pinned_section_exists(dirty_i) && 734 test_bit(secno, dirty_i->pinned_secmap); 735 } 736 737 static void f2fs_unpin_all_sections(struct f2fs_sb_info *sbi, bool enable) 738 { 739 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 740 741 if (f2fs_pinned_section_exists(DIRTY_I(sbi))) { 742 memset(DIRTY_I(sbi)->pinned_secmap, 0, bitmap_size); 743 DIRTY_I(sbi)->pinned_secmap_cnt = 0; 744 } 745 DIRTY_I(sbi)->enable_pin_section = enable; 746 } 747 748 static int f2fs_gc_pinned_control(struct inode *inode, int gc_type, 749 unsigned int segno) 750 { 751 if (!f2fs_is_pinned_file(inode)) 752 return 0; 753 if (gc_type != FG_GC) 754 return -EBUSY; 755 if (!f2fs_pin_section(F2FS_I_SB(inode), segno)) 756 f2fs_pin_file_control(inode, true); 757 return -EAGAIN; 758 } 759 760 /* 761 * This function is called from two paths. 762 * One is garbage collection and the other is SSR segment selection. 763 * When it is called during GC, it just gets a victim segment 764 * and it does not remove it from dirty seglist. 765 * When it is called from SSR segment selection, it finds a segment 766 * which has minimum valid blocks and removes it from dirty seglist. 767 */ 768 int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result, 769 int gc_type, int type, char alloc_mode, 770 unsigned long long age, bool one_time) 771 { 772 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 773 struct sit_info *sm = SIT_I(sbi); 774 struct victim_sel_policy p; 775 unsigned int secno, last_victim; 776 unsigned int last_segment; 777 unsigned int nsearched; 778 unsigned int valid_thresh_ratio = 100; 779 bool is_atgc; 780 int ret = 0; 781 782 mutex_lock(&dirty_i->seglist_lock); 783 last_segment = MAIN_SECS(sbi) * SEGS_PER_SEC(sbi); 784 785 p.alloc_mode = alloc_mode; 786 p.age = age; 787 p.age_threshold = sbi->am.age_threshold; 788 if (one_time) { 789 p.one_time_gc = one_time; 790 if (has_enough_free_secs(sbi, 0, NR_PERSISTENT_LOG)) 791 valid_thresh_ratio = sbi->gc_thread->valid_thresh_ratio; 792 } 793 794 retry: 795 select_policy(sbi, gc_type, type, &p); 796 p.min_segno = NULL_SEGNO; 797 p.oldest_age = 0; 798 p.min_cost = get_max_cost(sbi, &p); 799 800 is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR); 801 nsearched = 0; 802 803 if (is_atgc) 804 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX; 805 806 if (*result != NULL_SEGNO) { 807 if (!get_valid_blocks(sbi, *result, false)) { 808 ret = -ENODATA; 809 goto out; 810 } 811 812 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result))) { 813 ret = -EBUSY; 814 goto out; 815 } 816 if (gc_type == FG_GC) 817 clear_bit(GET_SEC_FROM_SEG(sbi, *result), dirty_i->victim_secmap); 818 p.min_segno = *result; 819 goto got_result; 820 } 821 822 ret = -ENODATA; 823 if (p.max_search == 0) 824 goto out; 825 826 if (__is_large_section(sbi) && p.alloc_mode == LFS) { 827 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) { 828 p.min_segno = sbi->next_victim_seg[BG_GC]; 829 *result = p.min_segno; 830 sbi->next_victim_seg[BG_GC] = NULL_SEGNO; 831 goto got_result; 832 } 833 if (gc_type == FG_GC && 834 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) { 835 p.min_segno = sbi->next_victim_seg[FG_GC]; 836 *result = p.min_segno; 837 sbi->next_victim_seg[FG_GC] = NULL_SEGNO; 838 goto got_result; 839 } 840 } 841 842 last_victim = sm->last_victim[p.gc_mode]; 843 if (p.alloc_mode == LFS && gc_type == FG_GC) { 844 p.min_segno = check_bg_victims(sbi); 845 if (p.min_segno != NULL_SEGNO) 846 goto got_it; 847 } 848 849 while (1) { 850 unsigned long cost, *dirty_bitmap; 851 unsigned int unit_no, segno; 852 853 dirty_bitmap = p.dirty_bitmap; 854 unit_no = find_next_bit(dirty_bitmap, 855 last_segment / p.ofs_unit, 856 p.offset / p.ofs_unit); 857 segno = unit_no * p.ofs_unit; 858 if (segno >= last_segment) { 859 if (sm->last_victim[p.gc_mode]) { 860 last_segment = 861 sm->last_victim[p.gc_mode]; 862 sm->last_victim[p.gc_mode] = 0; 863 p.offset = 0; 864 continue; 865 } 866 break; 867 } 868 869 p.offset = segno + p.ofs_unit; 870 nsearched++; 871 872 #ifdef CONFIG_F2FS_CHECK_FS 873 /* 874 * skip selecting the invalid segno (that is failed due to block 875 * validity check failure during GC) to avoid endless GC loop in 876 * such cases. 877 */ 878 if (test_bit(segno, sm->invalid_segmap)) 879 goto next; 880 #endif 881 882 secno = GET_SEC_FROM_SEG(sbi, segno); 883 884 if (sec_usage_check(sbi, secno)) 885 goto next; 886 887 /* Don't touch checkpointed data */ 888 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 889 if (p.alloc_mode == LFS) { 890 /* 891 * LFS is set to find source section during GC. 892 * The victim should have no checkpointed data. 893 */ 894 if (get_ckpt_valid_blocks(sbi, segno, true)) 895 goto next; 896 } else { 897 /* 898 * SSR | AT_SSR are set to find target segment 899 * for writes which can be full by checkpointed 900 * and newly written blocks. 901 */ 902 if (!f2fs_segment_has_free_slot(sbi, segno)) 903 goto next; 904 } 905 } 906 907 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) 908 goto next; 909 910 if (gc_type == FG_GC && f2fs_section_is_pinned(dirty_i, secno)) 911 goto next; 912 913 if (is_atgc) { 914 add_victim_entry(sbi, &p, segno); 915 goto next; 916 } 917 918 cost = get_gc_cost(sbi, segno, &p, valid_thresh_ratio); 919 920 if (p.min_cost > cost) { 921 p.min_segno = segno; 922 p.min_cost = cost; 923 } 924 next: 925 if (nsearched >= p.max_search) { 926 if (!sm->last_victim[p.gc_mode] && segno <= last_victim) 927 sm->last_victim[p.gc_mode] = 928 last_victim + p.ofs_unit; 929 else 930 sm->last_victim[p.gc_mode] = segno + p.ofs_unit; 931 sm->last_victim[p.gc_mode] %= 932 (MAIN_SECS(sbi) * SEGS_PER_SEC(sbi)); 933 break; 934 } 935 } 936 937 /* get victim for GC_AT/AT_SSR */ 938 if (is_atgc) { 939 lookup_victim_by_age(sbi, &p); 940 release_victim_entry(sbi); 941 } 942 943 if (is_atgc && p.min_segno == NULL_SEGNO && 944 sm->elapsed_time < p.age_threshold) { 945 p.age_threshold = 0; 946 goto retry; 947 } 948 949 if (p.min_segno != NULL_SEGNO) { 950 got_it: 951 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; 952 got_result: 953 if (p.alloc_mode == LFS) { 954 secno = GET_SEC_FROM_SEG(sbi, p.min_segno); 955 if (gc_type == FG_GC) 956 sbi->cur_victim_sec = secno; 957 else 958 set_bit(secno, dirty_i->victim_secmap); 959 } 960 ret = 0; 961 962 } 963 out: 964 if (p.min_segno != NULL_SEGNO) 965 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, 966 sbi->cur_victim_sec, 967 prefree_segments(sbi), free_segments(sbi)); 968 mutex_unlock(&dirty_i->seglist_lock); 969 970 return ret; 971 } 972 973 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) 974 { 975 struct inode_entry *ie; 976 977 ie = radix_tree_lookup(&gc_list->iroot, ino); 978 if (ie) 979 return ie->inode; 980 return NULL; 981 } 982 983 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) 984 { 985 struct inode_entry *new_ie; 986 987 if (inode == find_gc_inode(gc_list, inode->i_ino)) { 988 iput(inode); 989 return; 990 } 991 new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, 992 GFP_NOFS, true, NULL); 993 new_ie->inode = inode; 994 995 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); 996 list_add_tail(&new_ie->list, &gc_list->ilist); 997 } 998 999 static void put_gc_inode(struct gc_inode_list *gc_list) 1000 { 1001 struct inode_entry *ie, *next_ie; 1002 1003 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { 1004 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); 1005 iput(ie->inode); 1006 list_del(&ie->list); 1007 kmem_cache_free(f2fs_inode_entry_slab, ie); 1008 } 1009 } 1010 1011 static int check_valid_map(struct f2fs_sb_info *sbi, 1012 unsigned int segno, int offset) 1013 { 1014 struct sit_info *sit_i = SIT_I(sbi); 1015 struct seg_entry *sentry; 1016 int ret; 1017 1018 down_read(&sit_i->sentry_lock); 1019 sentry = get_seg_entry(sbi, segno); 1020 ret = f2fs_test_bit(offset, sentry->cur_valid_map); 1021 up_read(&sit_i->sentry_lock); 1022 return ret; 1023 } 1024 1025 /* 1026 * This function compares node address got in summary with that in NAT. 1027 * On validity, copy that node with cold status, otherwise (invalid node) 1028 * ignore that. 1029 */ 1030 static int gc_node_segment(struct f2fs_sb_info *sbi, 1031 struct f2fs_summary *sum, unsigned int segno, int gc_type) 1032 { 1033 struct f2fs_summary *entry; 1034 block_t start_addr; 1035 int off; 1036 int phase = 0; 1037 bool fggc = (gc_type == FG_GC); 1038 int submitted = 0; 1039 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 1040 1041 start_addr = START_BLOCK(sbi, segno); 1042 1043 next_step: 1044 entry = sum; 1045 1046 if (fggc && phase == 2) 1047 atomic_inc(&sbi->wb_sync_req[NODE]); 1048 1049 for (off = 0; off < usable_blks_in_seg; off++, entry++) { 1050 nid_t nid = le32_to_cpu(entry->nid); 1051 struct folio *node_folio; 1052 struct node_info ni; 1053 int err; 1054 1055 /* stop BG_GC if there is not enough free sections. */ 1056 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) 1057 return submitted; 1058 1059 if (check_valid_map(sbi, segno, off) == 0) 1060 continue; 1061 1062 if (phase == 0) { 1063 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, 1064 META_NAT, true); 1065 continue; 1066 } 1067 1068 if (phase == 1) { 1069 f2fs_ra_node_page(sbi, nid); 1070 continue; 1071 } 1072 1073 /* phase == 2 */ 1074 node_folio = f2fs_get_node_folio(sbi, nid); 1075 if (IS_ERR(node_folio)) 1076 continue; 1077 1078 /* block may become invalid during f2fs_get_node_folio */ 1079 if (check_valid_map(sbi, segno, off) == 0) { 1080 f2fs_folio_put(node_folio, true); 1081 continue; 1082 } 1083 1084 if (f2fs_get_node_info(sbi, nid, &ni, false)) { 1085 f2fs_folio_put(node_folio, true); 1086 continue; 1087 } 1088 1089 if (ni.blk_addr != start_addr + off) { 1090 f2fs_folio_put(node_folio, true); 1091 continue; 1092 } 1093 1094 err = f2fs_move_node_folio(node_folio, gc_type); 1095 if (!err && gc_type == FG_GC) 1096 submitted++; 1097 stat_inc_node_blk_count(sbi, 1, gc_type); 1098 } 1099 1100 if (++phase < 3) 1101 goto next_step; 1102 1103 if (fggc) 1104 atomic_dec(&sbi->wb_sync_req[NODE]); 1105 return submitted; 1106 } 1107 1108 /* 1109 * Calculate start block index indicating the given node offset. 1110 * Be careful, caller should give this node offset only indicating direct node 1111 * blocks. If any node offsets, which point the other types of node blocks such 1112 * as indirect or double indirect node blocks, are given, it must be a caller's 1113 * bug. 1114 */ 1115 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode) 1116 { 1117 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; 1118 unsigned int bidx; 1119 1120 if (node_ofs == 0) 1121 return 0; 1122 1123 if (node_ofs <= 2) { 1124 bidx = node_ofs - 1; 1125 } else if (node_ofs <= indirect_blks) { 1126 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); 1127 1128 bidx = node_ofs - 2 - dec; 1129 } else { 1130 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); 1131 1132 bidx = node_ofs - 5 - dec; 1133 } 1134 return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode); 1135 } 1136 1137 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 1138 struct node_info *dni, block_t blkaddr, unsigned int *nofs) 1139 { 1140 struct folio *node_folio; 1141 nid_t nid; 1142 unsigned int ofs_in_node, max_addrs, base; 1143 block_t source_blkaddr; 1144 1145 nid = le32_to_cpu(sum->nid); 1146 ofs_in_node = le16_to_cpu(sum->ofs_in_node); 1147 1148 node_folio = f2fs_get_node_folio(sbi, nid); 1149 if (IS_ERR(node_folio)) 1150 return false; 1151 1152 if (f2fs_get_node_info(sbi, nid, dni, false)) { 1153 f2fs_folio_put(node_folio, true); 1154 return false; 1155 } 1156 1157 if (sum->version != dni->version) { 1158 f2fs_warn(sbi, "%s: valid data with mismatched node version.", 1159 __func__); 1160 set_sbi_flag(sbi, SBI_NEED_FSCK); 1161 } 1162 1163 if (f2fs_check_nid_range(sbi, dni->ino)) { 1164 f2fs_folio_put(node_folio, true); 1165 return false; 1166 } 1167 1168 if (IS_INODE(node_folio)) { 1169 base = offset_in_addr(F2FS_INODE(node_folio)); 1170 max_addrs = DEF_ADDRS_PER_INODE; 1171 } else { 1172 base = 0; 1173 max_addrs = DEF_ADDRS_PER_BLOCK; 1174 } 1175 1176 if (base + ofs_in_node >= max_addrs) { 1177 f2fs_err(sbi, "Inconsistent blkaddr offset: base:%u, ofs_in_node:%u, max:%u, ino:%u, nid:%u", 1178 base, ofs_in_node, max_addrs, dni->ino, dni->nid); 1179 f2fs_folio_put(node_folio, true); 1180 return false; 1181 } 1182 1183 *nofs = ofs_of_node(node_folio); 1184 source_blkaddr = data_blkaddr(NULL, node_folio, ofs_in_node); 1185 f2fs_folio_put(node_folio, true); 1186 1187 if (source_blkaddr != blkaddr) { 1188 #ifdef CONFIG_F2FS_CHECK_FS 1189 unsigned int segno = GET_SEGNO(sbi, blkaddr); 1190 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 1191 1192 if (unlikely(check_valid_map(sbi, segno, offset))) { 1193 if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) { 1194 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u", 1195 blkaddr, source_blkaddr, segno); 1196 set_sbi_flag(sbi, SBI_NEED_FSCK); 1197 } 1198 } 1199 #endif 1200 return false; 1201 } 1202 return true; 1203 } 1204 1205 static int ra_data_block(struct inode *inode, pgoff_t index) 1206 { 1207 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1208 struct address_space *mapping = f2fs_is_cow_file(inode) ? 1209 F2FS_I(inode)->atomic_inode->i_mapping : inode->i_mapping; 1210 struct dnode_of_data dn; 1211 struct folio *folio; 1212 struct f2fs_io_info fio = { 1213 .sbi = sbi, 1214 .ino = inode->i_ino, 1215 .type = DATA, 1216 .temp = COLD, 1217 .op = REQ_OP_READ, 1218 .op_flags = 0, 1219 .encrypted_page = NULL, 1220 .in_list = 0, 1221 }; 1222 int err; 1223 1224 folio = f2fs_grab_cache_folio(mapping, index, true); 1225 if (IS_ERR(folio)) 1226 return PTR_ERR(folio); 1227 1228 if (f2fs_lookup_read_extent_cache_block(inode, index, 1229 &dn.data_blkaddr)) { 1230 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, 1231 DATA_GENERIC_ENHANCE_READ))) { 1232 err = -EFSCORRUPTED; 1233 goto put_folio; 1234 } 1235 goto got_it; 1236 } 1237 1238 set_new_dnode(&dn, inode, NULL, NULL, 0); 1239 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 1240 if (err) 1241 goto put_folio; 1242 f2fs_put_dnode(&dn); 1243 1244 if (!__is_valid_data_blkaddr(dn.data_blkaddr)) { 1245 err = -ENOENT; 1246 goto put_folio; 1247 } 1248 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, 1249 DATA_GENERIC_ENHANCE))) { 1250 err = -EFSCORRUPTED; 1251 goto put_folio; 1252 } 1253 got_it: 1254 /* read folio */ 1255 fio.folio = folio; 1256 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; 1257 1258 /* 1259 * don't cache encrypted data into meta inode until previous dirty 1260 * data were writebacked to avoid racing between GC and flush. 1261 */ 1262 f2fs_folio_wait_writeback(folio, DATA, true, true); 1263 1264 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); 1265 1266 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi), 1267 dn.data_blkaddr, 1268 FGP_LOCK | FGP_CREAT, GFP_NOFS); 1269 if (!fio.encrypted_page) { 1270 err = -ENOMEM; 1271 goto put_folio; 1272 } 1273 1274 err = f2fs_submit_page_bio(&fio); 1275 if (err) 1276 goto put_encrypted_page; 1277 f2fs_put_page(fio.encrypted_page, 0); 1278 f2fs_folio_put(folio, true); 1279 1280 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE); 1281 f2fs_update_iostat(sbi, NULL, FS_GDATA_READ_IO, F2FS_BLKSIZE); 1282 1283 return 0; 1284 put_encrypted_page: 1285 f2fs_put_page(fio.encrypted_page, 1); 1286 put_folio: 1287 f2fs_folio_put(folio, true); 1288 return err; 1289 } 1290 1291 /* 1292 * Move data block via META_MAPPING while keeping locked data page. 1293 * This can be used to move blocks, aka LBAs, directly on disk. 1294 */ 1295 static int move_data_block(struct inode *inode, block_t bidx, 1296 int gc_type, unsigned int segno, int off) 1297 { 1298 struct address_space *mapping = f2fs_is_cow_file(inode) ? 1299 F2FS_I(inode)->atomic_inode->i_mapping : inode->i_mapping; 1300 struct f2fs_io_info fio = { 1301 .sbi = F2FS_I_SB(inode), 1302 .ino = inode->i_ino, 1303 .type = DATA, 1304 .temp = COLD, 1305 .op = REQ_OP_READ, 1306 .op_flags = 0, 1307 .encrypted_page = NULL, 1308 .in_list = 0, 1309 }; 1310 struct dnode_of_data dn; 1311 struct f2fs_summary sum; 1312 struct node_info ni; 1313 struct folio *folio, *mfolio; 1314 block_t newaddr; 1315 int err = 0; 1316 bool lfs_mode = f2fs_lfs_mode(fio.sbi); 1317 int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) && 1318 (fio.sbi->gc_mode != GC_URGENT_HIGH) ? 1319 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA; 1320 1321 /* do not read out */ 1322 folio = f2fs_grab_cache_folio(mapping, bidx, false); 1323 if (IS_ERR(folio)) 1324 return PTR_ERR(folio); 1325 1326 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) { 1327 err = -ENOENT; 1328 goto out; 1329 } 1330 1331 err = f2fs_gc_pinned_control(inode, gc_type, segno); 1332 if (err) 1333 goto out; 1334 1335 set_new_dnode(&dn, inode, NULL, NULL, 0); 1336 err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE); 1337 if (err) 1338 goto out; 1339 1340 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 1341 folio_clear_uptodate(folio); 1342 err = -ENOENT; 1343 goto put_out; 1344 } 1345 1346 /* 1347 * don't cache encrypted data into meta inode until previous dirty 1348 * data were writebacked to avoid racing between GC and flush. 1349 */ 1350 f2fs_folio_wait_writeback(folio, DATA, true, true); 1351 1352 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); 1353 1354 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false); 1355 if (err) 1356 goto put_out; 1357 1358 /* read page */ 1359 fio.folio = folio; 1360 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; 1361 1362 if (lfs_mode) 1363 f2fs_down_write(&fio.sbi->io_order_lock); 1364 1365 mfolio = f2fs_grab_cache_folio(META_MAPPING(fio.sbi), 1366 fio.old_blkaddr, false); 1367 if (IS_ERR(mfolio)) { 1368 err = PTR_ERR(mfolio); 1369 goto up_out; 1370 } 1371 1372 fio.encrypted_page = folio_file_page(mfolio, fio.old_blkaddr); 1373 1374 /* read source block in mfolio */ 1375 if (!folio_test_uptodate(mfolio)) { 1376 err = f2fs_submit_page_bio(&fio); 1377 if (err) { 1378 f2fs_folio_put(mfolio, true); 1379 goto up_out; 1380 } 1381 1382 f2fs_update_iostat(fio.sbi, inode, FS_DATA_READ_IO, 1383 F2FS_BLKSIZE); 1384 f2fs_update_iostat(fio.sbi, NULL, FS_GDATA_READ_IO, 1385 F2FS_BLKSIZE); 1386 1387 folio_lock(mfolio); 1388 if (unlikely(!is_meta_folio(mfolio) || 1389 !folio_test_uptodate(mfolio))) { 1390 err = -EIO; 1391 f2fs_folio_put(mfolio, true); 1392 goto up_out; 1393 } 1394 } 1395 1396 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); 1397 1398 /* allocate block address */ 1399 err = f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr, 1400 &sum, type, NULL); 1401 if (err) { 1402 f2fs_folio_put(mfolio, true); 1403 /* filesystem should shutdown, no need to recovery block */ 1404 goto up_out; 1405 } 1406 1407 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi), 1408 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS); 1409 if (!fio.encrypted_page) { 1410 err = -ENOMEM; 1411 f2fs_folio_put(mfolio, true); 1412 goto recover_block; 1413 } 1414 1415 /* write target block */ 1416 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true); 1417 memcpy(page_address(fio.encrypted_page), 1418 folio_address(mfolio), PAGE_SIZE); 1419 f2fs_folio_put(mfolio, true); 1420 1421 f2fs_invalidate_internal_cache(fio.sbi, fio.old_blkaddr, 1); 1422 1423 set_page_dirty(fio.encrypted_page); 1424 if (clear_page_dirty_for_io(fio.encrypted_page)) 1425 dec_page_count(fio.sbi, F2FS_DIRTY_META); 1426 1427 set_page_writeback(fio.encrypted_page); 1428 1429 fio.op = REQ_OP_WRITE; 1430 fio.op_flags = REQ_SYNC; 1431 fio.new_blkaddr = newaddr; 1432 f2fs_submit_page_write(&fio); 1433 1434 f2fs_update_iostat(fio.sbi, NULL, FS_GC_DATA_IO, F2FS_BLKSIZE); 1435 1436 f2fs_update_data_blkaddr(&dn, newaddr); 1437 set_inode_flag(inode, FI_APPEND_WRITE); 1438 1439 f2fs_put_page(fio.encrypted_page, 1); 1440 recover_block: 1441 if (err) 1442 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr, 1443 true, true, true); 1444 up_out: 1445 if (lfs_mode) 1446 f2fs_up_write(&fio.sbi->io_order_lock); 1447 put_out: 1448 f2fs_put_dnode(&dn); 1449 out: 1450 f2fs_folio_put(folio, true); 1451 return err; 1452 } 1453 1454 static int move_data_page(struct inode *inode, block_t bidx, int gc_type, 1455 unsigned int segno, int off) 1456 { 1457 struct folio *folio; 1458 int err = 0; 1459 1460 folio = f2fs_get_lock_data_folio(inode, bidx, true); 1461 if (IS_ERR(folio)) 1462 return PTR_ERR(folio); 1463 1464 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) { 1465 err = -ENOENT; 1466 goto out; 1467 } 1468 1469 err = f2fs_gc_pinned_control(inode, gc_type, segno); 1470 if (err) 1471 goto out; 1472 1473 if (gc_type == BG_GC) { 1474 if (folio_test_writeback(folio)) { 1475 err = -EAGAIN; 1476 goto out; 1477 } 1478 folio_mark_dirty(folio); 1479 folio_set_f2fs_gcing(folio); 1480 } else { 1481 struct f2fs_io_info fio = { 1482 .sbi = F2FS_I_SB(inode), 1483 .ino = inode->i_ino, 1484 .type = DATA, 1485 .temp = COLD, 1486 .op = REQ_OP_WRITE, 1487 .op_flags = REQ_SYNC, 1488 .old_blkaddr = NULL_ADDR, 1489 .folio = folio, 1490 .encrypted_page = NULL, 1491 .need_lock = LOCK_REQ, 1492 .io_type = FS_GC_DATA_IO, 1493 }; 1494 bool is_dirty = folio_test_dirty(folio); 1495 1496 retry: 1497 f2fs_folio_wait_writeback(folio, DATA, true, true); 1498 1499 folio_mark_dirty(folio); 1500 if (folio_clear_dirty_for_io(folio)) { 1501 inode_dec_dirty_pages(inode); 1502 f2fs_remove_dirty_inode(inode); 1503 } 1504 1505 folio_set_f2fs_gcing(folio); 1506 1507 err = f2fs_do_write_data_page(&fio); 1508 if (err) { 1509 folio_clear_f2fs_gcing(folio); 1510 if (err == -ENOMEM) { 1511 memalloc_retry_wait(GFP_NOFS); 1512 goto retry; 1513 } 1514 if (is_dirty) 1515 folio_mark_dirty(folio); 1516 } 1517 } 1518 out: 1519 f2fs_folio_put(folio, true); 1520 return err; 1521 } 1522 1523 /* 1524 * This function tries to get parent node of victim data block, and identifies 1525 * data block validity. If the block is valid, copy that with cold status and 1526 * modify parent node. 1527 * If the parent node is not valid or the data block address is different, 1528 * the victim data block is ignored. 1529 */ 1530 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 1531 struct gc_inode_list *gc_list, unsigned int segno, int gc_type, 1532 bool force_migrate) 1533 { 1534 struct super_block *sb = sbi->sb; 1535 struct f2fs_summary *entry; 1536 block_t start_addr; 1537 int off; 1538 int phase = 0; 1539 int submitted = 0; 1540 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 1541 1542 start_addr = START_BLOCK(sbi, segno); 1543 1544 next_step: 1545 entry = sum; 1546 1547 for (off = 0; off < usable_blks_in_seg; off++, entry++) { 1548 struct inode *inode; 1549 struct node_info dni; /* dnode info for the data */ 1550 unsigned int ofs_in_node, nofs; 1551 block_t start_bidx; 1552 nid_t nid = le32_to_cpu(entry->nid); 1553 1554 /* 1555 * stop BG_GC if there is not enough free sections. 1556 * Or, stop GC if the segment becomes fully valid caused by 1557 * race condition along with SSR block allocation. 1558 */ 1559 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) || 1560 (!force_migrate && get_valid_blocks(sbi, segno, true) == 1561 CAP_BLKS_PER_SEC(sbi))) 1562 return submitted; 1563 1564 if (check_valid_map(sbi, segno, off) == 0) 1565 continue; 1566 1567 if (phase == 0) { 1568 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, 1569 META_NAT, true); 1570 continue; 1571 } 1572 1573 if (phase == 1) { 1574 f2fs_ra_node_page(sbi, nid); 1575 continue; 1576 } 1577 1578 /* Get an inode by ino with checking validity */ 1579 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) 1580 continue; 1581 1582 if (phase == 2) { 1583 f2fs_ra_node_page(sbi, dni.ino); 1584 continue; 1585 } 1586 1587 ofs_in_node = le16_to_cpu(entry->ofs_in_node); 1588 1589 if (phase == 3) { 1590 struct folio *data_folio; 1591 int err; 1592 1593 inode = f2fs_iget(sb, dni.ino); 1594 if (IS_ERR(inode)) 1595 continue; 1596 1597 if (is_bad_inode(inode) || 1598 special_file(inode->i_mode)) { 1599 iput(inode); 1600 continue; 1601 } 1602 1603 if (f2fs_has_inline_data(inode)) { 1604 iput(inode); 1605 set_sbi_flag(sbi, SBI_NEED_FSCK); 1606 f2fs_err_ratelimited(sbi, 1607 "inode %lx has both inline_data flag and " 1608 "data block, nid=%u, ofs_in_node=%u", 1609 inode->i_ino, dni.nid, ofs_in_node); 1610 continue; 1611 } 1612 1613 err = f2fs_gc_pinned_control(inode, gc_type, segno); 1614 if (err == -EAGAIN) { 1615 iput(inode); 1616 return submitted; 1617 } 1618 1619 if (!f2fs_down_write_trylock( 1620 &F2FS_I(inode)->i_gc_rwsem[WRITE])) { 1621 iput(inode); 1622 sbi->skipped_gc_rwsem++; 1623 continue; 1624 } 1625 1626 start_bidx = f2fs_start_bidx_of_node(nofs, inode) + 1627 ofs_in_node; 1628 1629 if (f2fs_meta_inode_gc_required(inode)) { 1630 int err = ra_data_block(inode, start_bidx); 1631 1632 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1633 if (err) { 1634 iput(inode); 1635 continue; 1636 } 1637 add_gc_inode(gc_list, inode); 1638 continue; 1639 } 1640 1641 data_folio = f2fs_get_read_data_folio(inode, start_bidx, 1642 REQ_RAHEAD, true, NULL); 1643 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1644 if (IS_ERR(data_folio)) { 1645 iput(inode); 1646 continue; 1647 } 1648 1649 f2fs_folio_put(data_folio, false); 1650 add_gc_inode(gc_list, inode); 1651 continue; 1652 } 1653 1654 /* phase 4 */ 1655 inode = find_gc_inode(gc_list, dni.ino); 1656 if (inode) { 1657 struct f2fs_inode_info *fi = F2FS_I(inode); 1658 bool locked = false; 1659 int err; 1660 1661 if (S_ISREG(inode->i_mode)) { 1662 if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[WRITE])) { 1663 sbi->skipped_gc_rwsem++; 1664 continue; 1665 } 1666 if (!f2fs_down_write_trylock( 1667 &fi->i_gc_rwsem[READ])) { 1668 sbi->skipped_gc_rwsem++; 1669 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 1670 continue; 1671 } 1672 locked = true; 1673 1674 /* wait for all inflight aio data */ 1675 inode_dio_wait(inode); 1676 } 1677 1678 start_bidx = f2fs_start_bidx_of_node(nofs, inode) 1679 + ofs_in_node; 1680 if (f2fs_meta_inode_gc_required(inode)) 1681 err = move_data_block(inode, start_bidx, 1682 gc_type, segno, off); 1683 else 1684 err = move_data_page(inode, start_bidx, gc_type, 1685 segno, off); 1686 1687 if (!err && (gc_type == FG_GC || 1688 f2fs_meta_inode_gc_required(inode))) 1689 submitted++; 1690 1691 if (locked) { 1692 f2fs_up_write(&fi->i_gc_rwsem[READ]); 1693 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 1694 } 1695 1696 stat_inc_data_blk_count(sbi, 1, gc_type); 1697 } 1698 } 1699 1700 if (++phase < 5) 1701 goto next_step; 1702 1703 return submitted; 1704 } 1705 1706 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, 1707 int gc_type, bool one_time) 1708 { 1709 struct sit_info *sit_i = SIT_I(sbi); 1710 int ret; 1711 1712 down_write(&sit_i->sentry_lock); 1713 ret = f2fs_get_victim(sbi, victim, gc_type, NO_CHECK_TYPE, 1714 LFS, 0, one_time); 1715 up_write(&sit_i->sentry_lock); 1716 return ret; 1717 } 1718 1719 static int do_garbage_collect(struct f2fs_sb_info *sbi, 1720 unsigned int start_segno, 1721 struct gc_inode_list *gc_list, int gc_type, 1722 bool force_migrate, bool one_time) 1723 { 1724 struct blk_plug plug; 1725 unsigned int segno = start_segno; 1726 unsigned int end_segno = start_segno + SEGS_PER_SEC(sbi); 1727 unsigned int sec_end_segno; 1728 int seg_freed = 0, migrated = 0; 1729 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ? 1730 SUM_TYPE_DATA : SUM_TYPE_NODE; 1731 unsigned char data_type = (type == SUM_TYPE_DATA) ? DATA : NODE; 1732 int submitted = 0; 1733 1734 if (__is_large_section(sbi)) { 1735 sec_end_segno = rounddown(end_segno, SEGS_PER_SEC(sbi)); 1736 1737 /* 1738 * zone-capacity can be less than zone-size in zoned devices, 1739 * resulting in less than expected usable segments in the zone, 1740 * calculate the end segno in the zone which can be garbage 1741 * collected 1742 */ 1743 if (f2fs_sb_has_blkzoned(sbi)) 1744 sec_end_segno -= SEGS_PER_SEC(sbi) - 1745 f2fs_usable_segs_in_sec(sbi); 1746 1747 if (gc_type == BG_GC || one_time) { 1748 unsigned int window_granularity = 1749 sbi->migration_window_granularity; 1750 1751 if (f2fs_sb_has_blkzoned(sbi) && 1752 !has_enough_free_blocks(sbi, 1753 sbi->gc_thread->boost_zoned_gc_percent)) 1754 window_granularity *= 1755 sbi->gc_thread->boost_gc_multiple; 1756 1757 end_segno = start_segno + window_granularity; 1758 } 1759 1760 if (end_segno > sec_end_segno) 1761 end_segno = sec_end_segno; 1762 } 1763 1764 sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type); 1765 1766 /* readahead multi ssa blocks those have contiguous address */ 1767 if (__is_large_section(sbi)) 1768 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), 1769 end_segno - segno, META_SSA, true); 1770 1771 /* reference all summary page */ 1772 while (segno < end_segno) { 1773 struct folio *sum_folio = f2fs_get_sum_folio(sbi, segno++); 1774 if (IS_ERR(sum_folio)) { 1775 int err = PTR_ERR(sum_folio); 1776 1777 end_segno = segno - 1; 1778 for (segno = start_segno; segno < end_segno; segno++) { 1779 sum_folio = filemap_get_folio(META_MAPPING(sbi), 1780 GET_SUM_BLOCK(sbi, segno)); 1781 folio_put_refs(sum_folio, 2); 1782 } 1783 return err; 1784 } 1785 folio_unlock(sum_folio); 1786 } 1787 1788 blk_start_plug(&plug); 1789 1790 for (segno = start_segno; segno < end_segno; segno++) { 1791 struct f2fs_summary_block *sum; 1792 1793 /* find segment summary of victim */ 1794 struct folio *sum_folio = filemap_get_folio(META_MAPPING(sbi), 1795 GET_SUM_BLOCK(sbi, segno)); 1796 1797 if (get_valid_blocks(sbi, segno, false) == 0) 1798 goto freed; 1799 if (gc_type == BG_GC && __is_large_section(sbi) && 1800 migrated >= sbi->migration_granularity) 1801 goto skip; 1802 if (!folio_test_uptodate(sum_folio) || 1803 unlikely(f2fs_cp_error(sbi))) 1804 goto skip; 1805 1806 sum = folio_address(sum_folio); 1807 if (type != GET_SUM_TYPE((&sum->footer))) { 1808 f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT", 1809 segno, type, GET_SUM_TYPE((&sum->footer))); 1810 f2fs_stop_checkpoint(sbi, false, 1811 STOP_CP_REASON_CORRUPTED_SUMMARY); 1812 goto skip; 1813 } 1814 1815 /* 1816 * this is to avoid deadlock: 1817 * - lock_page(sum_page) - f2fs_replace_block 1818 * - check_valid_map() - down_write(sentry_lock) 1819 * - down_read(sentry_lock) - change_curseg() 1820 * - lock_page(sum_page) 1821 */ 1822 if (type == SUM_TYPE_NODE) 1823 submitted += gc_node_segment(sbi, sum->entries, segno, 1824 gc_type); 1825 else 1826 submitted += gc_data_segment(sbi, sum->entries, gc_list, 1827 segno, gc_type, 1828 force_migrate); 1829 1830 stat_inc_gc_seg_count(sbi, data_type, gc_type); 1831 sbi->gc_reclaimed_segs[sbi->gc_mode]++; 1832 migrated++; 1833 1834 freed: 1835 if (gc_type == FG_GC && 1836 get_valid_blocks(sbi, segno, false) == 0) 1837 seg_freed++; 1838 1839 if (__is_large_section(sbi)) 1840 sbi->next_victim_seg[gc_type] = 1841 (segno + 1 < sec_end_segno) ? 1842 segno + 1 : NULL_SEGNO; 1843 skip: 1844 folio_put_refs(sum_folio, 2); 1845 } 1846 1847 if (submitted) 1848 f2fs_submit_merged_write(sbi, data_type); 1849 1850 blk_finish_plug(&plug); 1851 1852 if (migrated) 1853 stat_inc_gc_sec_count(sbi, data_type, gc_type); 1854 1855 return seg_freed; 1856 } 1857 1858 int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control) 1859 { 1860 int gc_type = gc_control->init_gc_type; 1861 unsigned int segno = gc_control->victim_segno; 1862 int sec_freed = 0, seg_freed = 0, total_freed = 0, total_sec_freed = 0; 1863 int ret = 0; 1864 struct cp_control cpc; 1865 struct gc_inode_list gc_list = { 1866 .ilist = LIST_HEAD_INIT(gc_list.ilist), 1867 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), 1868 }; 1869 unsigned int skipped_round = 0, round = 0; 1870 unsigned int upper_secs; 1871 1872 trace_f2fs_gc_begin(sbi->sb, gc_type, gc_control->no_bg_gc, 1873 gc_control->nr_free_secs, 1874 get_pages(sbi, F2FS_DIRTY_NODES), 1875 get_pages(sbi, F2FS_DIRTY_DENTS), 1876 get_pages(sbi, F2FS_DIRTY_IMETA), 1877 free_sections(sbi), 1878 free_segments(sbi), 1879 reserved_segments(sbi), 1880 prefree_segments(sbi)); 1881 1882 cpc.reason = __get_cp_reason(sbi); 1883 gc_more: 1884 sbi->skipped_gc_rwsem = 0; 1885 if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) { 1886 ret = -EINVAL; 1887 goto stop; 1888 } 1889 if (unlikely(f2fs_cp_error(sbi))) { 1890 ret = -EIO; 1891 goto stop; 1892 } 1893 1894 /* Let's run FG_GC, if we don't have enough space. */ 1895 if (has_not_enough_free_secs(sbi, 0, 0)) { 1896 gc_type = FG_GC; 1897 gc_control->one_time = false; 1898 1899 /* 1900 * For example, if there are many prefree_segments below given 1901 * threshold, we can make them free by checkpoint. Then, we 1902 * secure free segments which doesn't need fggc any more. 1903 */ 1904 if (prefree_segments(sbi)) { 1905 stat_inc_cp_call_count(sbi, TOTAL_CALL); 1906 ret = f2fs_write_checkpoint(sbi, &cpc); 1907 if (ret) 1908 goto stop; 1909 /* Reset due to checkpoint */ 1910 sec_freed = 0; 1911 } 1912 } 1913 1914 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */ 1915 if (gc_type == BG_GC && gc_control->no_bg_gc) { 1916 ret = -EINVAL; 1917 goto stop; 1918 } 1919 retry: 1920 ret = __get_victim(sbi, &segno, gc_type, gc_control->one_time); 1921 if (ret) { 1922 /* allow to search victim from sections has pinned data */ 1923 if (ret == -ENODATA && gc_type == FG_GC && 1924 f2fs_pinned_section_exists(DIRTY_I(sbi))) { 1925 f2fs_unpin_all_sections(sbi, false); 1926 goto retry; 1927 } 1928 goto stop; 1929 } 1930 1931 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type, 1932 gc_control->should_migrate_blocks, 1933 gc_control->one_time); 1934 if (seg_freed < 0) 1935 goto stop; 1936 1937 total_freed += seg_freed; 1938 1939 if (seg_freed == f2fs_usable_segs_in_sec(sbi)) { 1940 sec_freed++; 1941 total_sec_freed++; 1942 } 1943 1944 if (gc_control->one_time) 1945 goto stop; 1946 1947 if (gc_type == FG_GC) { 1948 sbi->cur_victim_sec = NULL_SEGNO; 1949 1950 if (has_enough_free_secs(sbi, sec_freed, 0)) { 1951 if (!gc_control->no_bg_gc && 1952 total_sec_freed < gc_control->nr_free_secs) 1953 goto go_gc_more; 1954 goto stop; 1955 } 1956 if (sbi->skipped_gc_rwsem) 1957 skipped_round++; 1958 round++; 1959 if (skipped_round > MAX_SKIP_GC_COUNT && 1960 skipped_round * 2 >= round) { 1961 stat_inc_cp_call_count(sbi, TOTAL_CALL); 1962 ret = f2fs_write_checkpoint(sbi, &cpc); 1963 goto stop; 1964 } 1965 } else if (has_enough_free_secs(sbi, 0, 0)) { 1966 goto stop; 1967 } 1968 1969 __get_secs_required(sbi, NULL, &upper_secs, NULL); 1970 1971 /* 1972 * Write checkpoint to reclaim prefree segments. 1973 * We need more three extra sections for writer's data/node/dentry. 1974 */ 1975 if (free_sections(sbi) <= upper_secs + NR_GC_CHECKPOINT_SECS && 1976 prefree_segments(sbi)) { 1977 stat_inc_cp_call_count(sbi, TOTAL_CALL); 1978 ret = f2fs_write_checkpoint(sbi, &cpc); 1979 if (ret) 1980 goto stop; 1981 /* Reset due to checkpoint */ 1982 sec_freed = 0; 1983 } 1984 go_gc_more: 1985 segno = NULL_SEGNO; 1986 goto gc_more; 1987 1988 stop: 1989 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0; 1990 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno; 1991 1992 if (gc_type == FG_GC) 1993 f2fs_unpin_all_sections(sbi, true); 1994 1995 trace_f2fs_gc_end(sbi->sb, ret, total_freed, total_sec_freed, 1996 get_pages(sbi, F2FS_DIRTY_NODES), 1997 get_pages(sbi, F2FS_DIRTY_DENTS), 1998 get_pages(sbi, F2FS_DIRTY_IMETA), 1999 free_sections(sbi), 2000 free_segments(sbi), 2001 reserved_segments(sbi), 2002 prefree_segments(sbi)); 2003 2004 f2fs_up_write(&sbi->gc_lock); 2005 2006 put_gc_inode(&gc_list); 2007 2008 if (gc_control->err_gc_skipped && !ret) 2009 ret = total_sec_freed ? 0 : -EAGAIN; 2010 return ret; 2011 } 2012 2013 int __init f2fs_create_garbage_collection_cache(void) 2014 { 2015 victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry", 2016 sizeof(struct victim_entry)); 2017 return victim_entry_slab ? 0 : -ENOMEM; 2018 } 2019 2020 void f2fs_destroy_garbage_collection_cache(void) 2021 { 2022 kmem_cache_destroy(victim_entry_slab); 2023 } 2024 2025 static void init_atgc_management(struct f2fs_sb_info *sbi) 2026 { 2027 struct atgc_management *am = &sbi->am; 2028 2029 if (test_opt(sbi, ATGC) && 2030 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD) 2031 am->atgc_enabled = true; 2032 2033 am->root = RB_ROOT_CACHED; 2034 INIT_LIST_HEAD(&am->victim_list); 2035 am->victim_count = 0; 2036 2037 am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO; 2038 am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT; 2039 am->age_weight = DEF_GC_THREAD_AGE_WEIGHT; 2040 am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD; 2041 } 2042 2043 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi) 2044 { 2045 sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES; 2046 2047 /* give warm/cold data area from slower device */ 2048 if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi)) 2049 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 2050 GET_SEGNO(sbi, FDEV(0).end_blk) + 1; 2051 2052 init_atgc_management(sbi); 2053 } 2054 2055 int f2fs_gc_range(struct f2fs_sb_info *sbi, 2056 unsigned int start_seg, unsigned int end_seg, 2057 bool dry_run, unsigned int dry_run_sections) 2058 { 2059 unsigned int segno; 2060 unsigned int gc_secs = dry_run_sections; 2061 2062 if (unlikely(f2fs_cp_error(sbi))) 2063 return -EIO; 2064 2065 for (segno = start_seg; segno <= end_seg; segno += SEGS_PER_SEC(sbi)) { 2066 struct gc_inode_list gc_list = { 2067 .ilist = LIST_HEAD_INIT(gc_list.ilist), 2068 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), 2069 }; 2070 2071 if (is_cursec(sbi, GET_SEC_FROM_SEG(sbi, segno))) 2072 continue; 2073 2074 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true, false); 2075 put_gc_inode(&gc_list); 2076 2077 if (!dry_run && get_valid_blocks(sbi, segno, true)) 2078 return -EAGAIN; 2079 if (dry_run && dry_run_sections && 2080 !get_valid_blocks(sbi, segno, true) && --gc_secs == 0) 2081 break; 2082 2083 if (fatal_signal_pending(current)) 2084 return -ERESTARTSYS; 2085 } 2086 2087 return 0; 2088 } 2089 2090 static int free_segment_range(struct f2fs_sb_info *sbi, 2091 unsigned int secs, bool dry_run) 2092 { 2093 unsigned int next_inuse, start, end; 2094 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 }; 2095 int gc_mode, gc_type; 2096 int err = 0; 2097 int type; 2098 2099 /* Force block allocation for GC */ 2100 MAIN_SECS(sbi) -= secs; 2101 start = MAIN_SECS(sbi) * SEGS_PER_SEC(sbi); 2102 end = MAIN_SEGS(sbi) - 1; 2103 2104 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2105 for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++) 2106 if (SIT_I(sbi)->last_victim[gc_mode] >= start) 2107 SIT_I(sbi)->last_victim[gc_mode] = 0; 2108 2109 for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++) 2110 if (sbi->next_victim_seg[gc_type] >= start) 2111 sbi->next_victim_seg[gc_type] = NULL_SEGNO; 2112 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2113 2114 /* Move out cursegs from the target range */ 2115 for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++) { 2116 err = f2fs_allocate_segment_for_resize(sbi, type, start, end); 2117 if (err) 2118 goto out; 2119 } 2120 2121 /* do GC to move out valid blocks in the range */ 2122 err = f2fs_gc_range(sbi, start, end, dry_run, 0); 2123 if (err || dry_run) 2124 goto out; 2125 2126 stat_inc_cp_call_count(sbi, TOTAL_CALL); 2127 err = f2fs_write_checkpoint(sbi, &cpc); 2128 if (err) 2129 goto out; 2130 2131 next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start); 2132 if (next_inuse <= end) { 2133 f2fs_err(sbi, "segno %u should be free but still inuse!", 2134 next_inuse); 2135 f2fs_bug_on(sbi, 1); 2136 } 2137 out: 2138 MAIN_SECS(sbi) += secs; 2139 return err; 2140 } 2141 2142 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs) 2143 { 2144 struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi); 2145 int section_count; 2146 int segment_count; 2147 int segment_count_main; 2148 long long block_count; 2149 int segs = secs * SEGS_PER_SEC(sbi); 2150 2151 f2fs_down_write(&sbi->sb_lock); 2152 2153 section_count = le32_to_cpu(raw_sb->section_count); 2154 segment_count = le32_to_cpu(raw_sb->segment_count); 2155 segment_count_main = le32_to_cpu(raw_sb->segment_count_main); 2156 block_count = le64_to_cpu(raw_sb->block_count); 2157 2158 raw_sb->section_count = cpu_to_le32(section_count + secs); 2159 raw_sb->segment_count = cpu_to_le32(segment_count + segs); 2160 raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs); 2161 raw_sb->block_count = cpu_to_le64(block_count + 2162 (long long)SEGS_TO_BLKS(sbi, segs)); 2163 if (f2fs_is_multi_device(sbi)) { 2164 int last_dev = sbi->s_ndevs - 1; 2165 int dev_segs = 2166 le32_to_cpu(raw_sb->devs[last_dev].total_segments); 2167 2168 raw_sb->devs[last_dev].total_segments = 2169 cpu_to_le32(dev_segs + segs); 2170 } 2171 2172 f2fs_up_write(&sbi->sb_lock); 2173 } 2174 2175 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs) 2176 { 2177 int segs = secs * SEGS_PER_SEC(sbi); 2178 long long blks = SEGS_TO_BLKS(sbi, segs); 2179 long long user_block_count = 2180 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count); 2181 2182 SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs; 2183 MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs; 2184 MAIN_SECS(sbi) += secs; 2185 FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs; 2186 FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs; 2187 F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks); 2188 2189 if (f2fs_is_multi_device(sbi)) { 2190 int last_dev = sbi->s_ndevs - 1; 2191 2192 FDEV(last_dev).total_segments = 2193 (int)FDEV(last_dev).total_segments + segs; 2194 FDEV(last_dev).end_blk = 2195 (long long)FDEV(last_dev).end_blk + blks; 2196 #ifdef CONFIG_BLK_DEV_ZONED 2197 FDEV(last_dev).nr_blkz = FDEV(last_dev).nr_blkz + 2198 div_u64(blks, sbi->blocks_per_blkz); 2199 #endif 2200 } 2201 } 2202 2203 int f2fs_resize_fs(struct file *filp, __u64 block_count) 2204 { 2205 struct f2fs_sb_info *sbi = F2FS_I_SB(file_inode(filp)); 2206 __u64 old_block_count, shrunk_blocks; 2207 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 }; 2208 unsigned int secs; 2209 int err = 0; 2210 __u32 rem; 2211 2212 old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count); 2213 if (block_count > old_block_count) 2214 return -EINVAL; 2215 2216 if (f2fs_is_multi_device(sbi)) { 2217 int last_dev = sbi->s_ndevs - 1; 2218 __u64 last_segs = FDEV(last_dev).total_segments; 2219 2220 if (block_count + SEGS_TO_BLKS(sbi, last_segs) <= 2221 old_block_count) 2222 return -EINVAL; 2223 } 2224 2225 /* new fs size should align to section size */ 2226 div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem); 2227 if (rem) 2228 return -EINVAL; 2229 2230 if (block_count == old_block_count) 2231 return 0; 2232 2233 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 2234 f2fs_err(sbi, "Should run fsck to repair first."); 2235 return -EFSCORRUPTED; 2236 } 2237 2238 if (test_opt(sbi, DISABLE_CHECKPOINT)) { 2239 f2fs_err(sbi, "Checkpoint should be enabled."); 2240 return -EINVAL; 2241 } 2242 2243 err = mnt_want_write_file(filp); 2244 if (err) 2245 return err; 2246 2247 shrunk_blocks = old_block_count - block_count; 2248 secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi)); 2249 2250 /* stop other GC */ 2251 if (!f2fs_down_write_trylock(&sbi->gc_lock)) { 2252 err = -EAGAIN; 2253 goto out_drop_write; 2254 } 2255 2256 /* stop CP to protect MAIN_SEC in free_segment_range */ 2257 f2fs_lock_op(sbi); 2258 2259 spin_lock(&sbi->stat_lock); 2260 if (shrunk_blocks + valid_user_blocks(sbi) + 2261 sbi->current_reserved_blocks + sbi->unusable_block_count + 2262 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count) 2263 err = -ENOSPC; 2264 spin_unlock(&sbi->stat_lock); 2265 2266 if (err) 2267 goto out_unlock; 2268 2269 err = free_segment_range(sbi, secs, true); 2270 2271 out_unlock: 2272 f2fs_unlock_op(sbi); 2273 f2fs_up_write(&sbi->gc_lock); 2274 out_drop_write: 2275 mnt_drop_write_file(filp); 2276 if (err) 2277 return err; 2278 2279 err = freeze_super(sbi->sb, FREEZE_HOLDER_KERNEL, NULL); 2280 if (err) 2281 return err; 2282 2283 if (f2fs_readonly(sbi->sb)) { 2284 err = thaw_super(sbi->sb, FREEZE_HOLDER_KERNEL, NULL); 2285 if (err) 2286 return err; 2287 return -EROFS; 2288 } 2289 2290 f2fs_down_write(&sbi->gc_lock); 2291 f2fs_down_write(&sbi->cp_global_sem); 2292 2293 spin_lock(&sbi->stat_lock); 2294 if (shrunk_blocks + valid_user_blocks(sbi) + 2295 sbi->current_reserved_blocks + sbi->unusable_block_count + 2296 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count) 2297 err = -ENOSPC; 2298 else 2299 sbi->user_block_count -= shrunk_blocks; 2300 spin_unlock(&sbi->stat_lock); 2301 if (err) 2302 goto out_err; 2303 2304 set_sbi_flag(sbi, SBI_IS_RESIZEFS); 2305 err = free_segment_range(sbi, secs, false); 2306 if (err) 2307 goto recover_out; 2308 2309 update_sb_metadata(sbi, -secs); 2310 2311 err = f2fs_commit_super(sbi, false); 2312 if (err) { 2313 update_sb_metadata(sbi, secs); 2314 goto recover_out; 2315 } 2316 2317 update_fs_metadata(sbi, -secs); 2318 clear_sbi_flag(sbi, SBI_IS_RESIZEFS); 2319 set_sbi_flag(sbi, SBI_IS_DIRTY); 2320 2321 stat_inc_cp_call_count(sbi, TOTAL_CALL); 2322 err = f2fs_write_checkpoint(sbi, &cpc); 2323 if (err) { 2324 update_fs_metadata(sbi, secs); 2325 update_sb_metadata(sbi, secs); 2326 f2fs_commit_super(sbi, false); 2327 } 2328 recover_out: 2329 clear_sbi_flag(sbi, SBI_IS_RESIZEFS); 2330 if (err) { 2331 set_sbi_flag(sbi, SBI_NEED_FSCK); 2332 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!"); 2333 2334 spin_lock(&sbi->stat_lock); 2335 sbi->user_block_count += shrunk_blocks; 2336 spin_unlock(&sbi->stat_lock); 2337 } 2338 out_err: 2339 f2fs_up_write(&sbi->cp_global_sem); 2340 f2fs_up_write(&sbi->gc_lock); 2341 thaw_super(sbi->sb, FREEZE_HOLDER_KERNEL, NULL); 2342 return err; 2343 } 2344