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; 145 146 /* foreground GC was been triggered via f2fs_balance_fs() */ 147 if (foreground) 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; 200 201 if (f2fs_sb_has_blkzoned(sbi)) { 202 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME_ZONED; 203 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME_ZONED; 204 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME_ZONED; 205 gc_th->no_zoned_gc_percent = LIMIT_NO_ZONED_GC; 206 gc_th->boost_zoned_gc_percent = LIMIT_BOOST_ZONED_GC; 207 } else { 208 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; 209 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; 210 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; 211 gc_th->no_zoned_gc_percent = 0; 212 gc_th->boost_zoned_gc_percent = 0; 213 } 214 215 gc_th->gc_wake = false; 216 217 sbi->gc_thread = gc_th; 218 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); 219 init_waitqueue_head(&sbi->gc_thread->fggc_wq); 220 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, 221 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); 222 if (IS_ERR(gc_th->f2fs_gc_task)) { 223 int err = PTR_ERR(gc_th->f2fs_gc_task); 224 225 kfree(gc_th); 226 sbi->gc_thread = NULL; 227 return err; 228 } 229 230 return 0; 231 } 232 233 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi) 234 { 235 struct f2fs_gc_kthread *gc_th = sbi->gc_thread; 236 237 if (!gc_th) 238 return; 239 kthread_stop(gc_th->f2fs_gc_task); 240 wake_up_all(&gc_th->fggc_wq); 241 kfree(gc_th); 242 sbi->gc_thread = NULL; 243 } 244 245 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type) 246 { 247 int gc_mode; 248 249 if (gc_type == BG_GC) { 250 if (sbi->am.atgc_enabled) 251 gc_mode = GC_AT; 252 else 253 gc_mode = GC_CB; 254 } else { 255 gc_mode = GC_GREEDY; 256 } 257 258 switch (sbi->gc_mode) { 259 case GC_IDLE_CB: 260 case GC_URGENT_LOW: 261 case GC_URGENT_MID: 262 gc_mode = GC_CB; 263 break; 264 case GC_IDLE_GREEDY: 265 case GC_URGENT_HIGH: 266 gc_mode = GC_GREEDY; 267 break; 268 case GC_IDLE_AT: 269 gc_mode = GC_AT; 270 break; 271 } 272 273 return gc_mode; 274 } 275 276 static void select_policy(struct f2fs_sb_info *sbi, int gc_type, 277 int type, struct victim_sel_policy *p) 278 { 279 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 280 281 if (p->alloc_mode == SSR) { 282 p->gc_mode = GC_GREEDY; 283 p->dirty_bitmap = dirty_i->dirty_segmap[type]; 284 p->max_search = dirty_i->nr_dirty[type]; 285 p->ofs_unit = 1; 286 } else if (p->alloc_mode == AT_SSR) { 287 p->gc_mode = GC_GREEDY; 288 p->dirty_bitmap = dirty_i->dirty_segmap[type]; 289 p->max_search = dirty_i->nr_dirty[type]; 290 p->ofs_unit = 1; 291 } else { 292 p->gc_mode = select_gc_type(sbi, gc_type); 293 p->ofs_unit = SEGS_PER_SEC(sbi); 294 if (__is_large_section(sbi)) { 295 p->dirty_bitmap = dirty_i->dirty_secmap; 296 p->max_search = count_bits(p->dirty_bitmap, 297 0, MAIN_SECS(sbi)); 298 } else { 299 p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY]; 300 p->max_search = dirty_i->nr_dirty[DIRTY]; 301 } 302 } 303 304 /* 305 * adjust candidates range, should select all dirty segments for 306 * foreground GC and urgent GC cases. 307 */ 308 if (gc_type != FG_GC && 309 (sbi->gc_mode != GC_URGENT_HIGH) && 310 (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) && 311 p->max_search > sbi->max_victim_search) 312 p->max_search = sbi->max_victim_search; 313 314 /* let's select beginning hot/small space first. */ 315 if (f2fs_need_rand_seg(sbi)) 316 p->offset = get_random_u32_below(MAIN_SECS(sbi) * 317 SEGS_PER_SEC(sbi)); 318 else if (type == CURSEG_HOT_DATA || IS_NODESEG(type)) 319 p->offset = 0; 320 else 321 p->offset = SIT_I(sbi)->last_victim[p->gc_mode]; 322 } 323 324 static unsigned int get_max_cost(struct f2fs_sb_info *sbi, 325 struct victim_sel_policy *p) 326 { 327 /* SSR allocates in a segment unit */ 328 if (p->alloc_mode == SSR) 329 return BLKS_PER_SEG(sbi); 330 else if (p->alloc_mode == AT_SSR) 331 return UINT_MAX; 332 333 /* LFS */ 334 if (p->gc_mode == GC_GREEDY) 335 return SEGS_TO_BLKS(sbi, 2 * p->ofs_unit); 336 else if (p->gc_mode == GC_CB) 337 return UINT_MAX; 338 else if (p->gc_mode == GC_AT) 339 return UINT_MAX; 340 else /* No other gc_mode */ 341 return 0; 342 } 343 344 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) 345 { 346 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 347 unsigned int secno; 348 349 /* 350 * If the gc_type is FG_GC, we can select victim segments 351 * selected by background GC before. 352 * Those segments guarantee they have small valid blocks. 353 */ 354 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { 355 if (sec_usage_check(sbi, secno)) 356 continue; 357 clear_bit(secno, dirty_i->victim_secmap); 358 return GET_SEG_FROM_SEC(sbi, secno); 359 } 360 return NULL_SEGNO; 361 } 362 363 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) 364 { 365 struct sit_info *sit_i = SIT_I(sbi); 366 unsigned long long mtime = 0; 367 unsigned int vblocks; 368 unsigned char age = 0; 369 unsigned char u; 370 unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi); 371 372 mtime = f2fs_get_section_mtime(sbi, segno); 373 f2fs_bug_on(sbi, mtime == INVALID_MTIME); 374 vblocks = get_valid_blocks(sbi, segno, true); 375 vblocks = div_u64(vblocks, usable_segs_per_sec); 376 377 u = BLKS_TO_SEGS(sbi, vblocks * 100); 378 379 /* Handle if the system time has changed by the user */ 380 if (mtime < sit_i->min_mtime) 381 sit_i->min_mtime = mtime; 382 if (mtime > sit_i->max_mtime) 383 sit_i->max_mtime = mtime; 384 if (sit_i->max_mtime != sit_i->min_mtime) 385 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), 386 sit_i->max_mtime - sit_i->min_mtime); 387 388 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); 389 } 390 391 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, 392 unsigned int segno, struct victim_sel_policy *p) 393 { 394 if (p->alloc_mode == SSR) 395 return get_seg_entry(sbi, segno)->ckpt_valid_blocks; 396 397 if (p->one_time_gc && (get_valid_blocks(sbi, segno, true) >= 398 CAP_BLKS_PER_SEC(sbi) * sbi->gc_thread->valid_thresh_ratio / 399 100)) 400 return UINT_MAX; 401 402 /* alloc_mode == LFS */ 403 if (p->gc_mode == GC_GREEDY) 404 return get_valid_blocks(sbi, segno, true); 405 else if (p->gc_mode == GC_CB) 406 return get_cb_cost(sbi, segno); 407 408 f2fs_bug_on(sbi, 1); 409 return 0; 410 } 411 412 static unsigned int count_bits(const unsigned long *addr, 413 unsigned int offset, unsigned int len) 414 { 415 unsigned int end = offset + len, sum = 0; 416 417 while (offset < end) { 418 if (test_bit(offset++, addr)) 419 ++sum; 420 } 421 return sum; 422 } 423 424 static bool f2fs_check_victim_tree(struct f2fs_sb_info *sbi, 425 struct rb_root_cached *root) 426 { 427 #ifdef CONFIG_F2FS_CHECK_FS 428 struct rb_node *cur = rb_first_cached(root), *next; 429 struct victim_entry *cur_ve, *next_ve; 430 431 while (cur) { 432 next = rb_next(cur); 433 if (!next) 434 return true; 435 436 cur_ve = rb_entry(cur, struct victim_entry, rb_node); 437 next_ve = rb_entry(next, struct victim_entry, rb_node); 438 439 if (cur_ve->mtime > next_ve->mtime) { 440 f2fs_info(sbi, "broken victim_rbtree, " 441 "cur_mtime(%llu) next_mtime(%llu)", 442 cur_ve->mtime, next_ve->mtime); 443 return false; 444 } 445 cur = next; 446 } 447 #endif 448 return true; 449 } 450 451 static struct victim_entry *__lookup_victim_entry(struct f2fs_sb_info *sbi, 452 unsigned long long mtime) 453 { 454 struct atgc_management *am = &sbi->am; 455 struct rb_node *node = am->root.rb_root.rb_node; 456 struct victim_entry *ve = NULL; 457 458 while (node) { 459 ve = rb_entry(node, struct victim_entry, rb_node); 460 461 if (mtime < ve->mtime) 462 node = node->rb_left; 463 else 464 node = node->rb_right; 465 } 466 return ve; 467 } 468 469 static struct victim_entry *__create_victim_entry(struct f2fs_sb_info *sbi, 470 unsigned long long mtime, unsigned int segno) 471 { 472 struct atgc_management *am = &sbi->am; 473 struct victim_entry *ve; 474 475 ve = f2fs_kmem_cache_alloc(victim_entry_slab, GFP_NOFS, true, NULL); 476 477 ve->mtime = mtime; 478 ve->segno = segno; 479 480 list_add_tail(&ve->list, &am->victim_list); 481 am->victim_count++; 482 483 return ve; 484 } 485 486 static void __insert_victim_entry(struct f2fs_sb_info *sbi, 487 unsigned long long mtime, unsigned int segno) 488 { 489 struct atgc_management *am = &sbi->am; 490 struct rb_root_cached *root = &am->root; 491 struct rb_node **p = &root->rb_root.rb_node; 492 struct rb_node *parent = NULL; 493 struct victim_entry *ve; 494 bool left_most = true; 495 496 /* look up rb tree to find parent node */ 497 while (*p) { 498 parent = *p; 499 ve = rb_entry(parent, struct victim_entry, rb_node); 500 501 if (mtime < ve->mtime) { 502 p = &(*p)->rb_left; 503 } else { 504 p = &(*p)->rb_right; 505 left_most = false; 506 } 507 } 508 509 ve = __create_victim_entry(sbi, mtime, segno); 510 511 rb_link_node(&ve->rb_node, parent, p); 512 rb_insert_color_cached(&ve->rb_node, root, left_most); 513 } 514 515 static void add_victim_entry(struct f2fs_sb_info *sbi, 516 struct victim_sel_policy *p, unsigned int segno) 517 { 518 struct sit_info *sit_i = SIT_I(sbi); 519 unsigned long long mtime = 0; 520 521 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 522 if (p->gc_mode == GC_AT && 523 get_valid_blocks(sbi, segno, true) == 0) 524 return; 525 } 526 527 mtime = f2fs_get_section_mtime(sbi, segno); 528 f2fs_bug_on(sbi, mtime == INVALID_MTIME); 529 530 /* Handle if the system time has changed by the user */ 531 if (mtime < sit_i->min_mtime) 532 sit_i->min_mtime = mtime; 533 if (mtime > sit_i->max_mtime) 534 sit_i->max_mtime = mtime; 535 if (mtime < sit_i->dirty_min_mtime) 536 sit_i->dirty_min_mtime = mtime; 537 if (mtime > sit_i->dirty_max_mtime) 538 sit_i->dirty_max_mtime = mtime; 539 540 /* don't choose young section as candidate */ 541 if (sit_i->dirty_max_mtime - mtime < p->age_threshold) 542 return; 543 544 __insert_victim_entry(sbi, mtime, segno); 545 } 546 547 static void atgc_lookup_victim(struct f2fs_sb_info *sbi, 548 struct victim_sel_policy *p) 549 { 550 struct sit_info *sit_i = SIT_I(sbi); 551 struct atgc_management *am = &sbi->am; 552 struct rb_root_cached *root = &am->root; 553 struct rb_node *node; 554 struct victim_entry *ve; 555 unsigned long long total_time; 556 unsigned long long age, u, accu; 557 unsigned long long max_mtime = sit_i->dirty_max_mtime; 558 unsigned long long min_mtime = sit_i->dirty_min_mtime; 559 unsigned int sec_blocks = CAP_BLKS_PER_SEC(sbi); 560 unsigned int vblocks; 561 unsigned int dirty_threshold = max(am->max_candidate_count, 562 am->candidate_ratio * 563 am->victim_count / 100); 564 unsigned int age_weight = am->age_weight; 565 unsigned int cost; 566 unsigned int iter = 0; 567 568 if (max_mtime < min_mtime) 569 return; 570 571 max_mtime += 1; 572 total_time = max_mtime - min_mtime; 573 574 accu = div64_u64(ULLONG_MAX, total_time); 575 accu = min_t(unsigned long long, div_u64(accu, 100), 576 DEFAULT_ACCURACY_CLASS); 577 578 node = rb_first_cached(root); 579 next: 580 ve = rb_entry_safe(node, struct victim_entry, rb_node); 581 if (!ve) 582 return; 583 584 if (ve->mtime >= max_mtime || ve->mtime < min_mtime) 585 goto skip; 586 587 /* age = 10000 * x% * 60 */ 588 age = div64_u64(accu * (max_mtime - ve->mtime), total_time) * 589 age_weight; 590 591 vblocks = get_valid_blocks(sbi, ve->segno, true); 592 f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks); 593 594 /* u = 10000 * x% * 40 */ 595 u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) * 596 (100 - age_weight); 597 598 f2fs_bug_on(sbi, age + u >= UINT_MAX); 599 600 cost = UINT_MAX - (age + u); 601 iter++; 602 603 if (cost < p->min_cost || 604 (cost == p->min_cost && age > p->oldest_age)) { 605 p->min_cost = cost; 606 p->oldest_age = age; 607 p->min_segno = ve->segno; 608 } 609 skip: 610 if (iter < dirty_threshold) { 611 node = rb_next(node); 612 goto next; 613 } 614 } 615 616 /* 617 * select candidates around source section in range of 618 * [target - dirty_threshold, target + dirty_threshold] 619 */ 620 static void atssr_lookup_victim(struct f2fs_sb_info *sbi, 621 struct victim_sel_policy *p) 622 { 623 struct sit_info *sit_i = SIT_I(sbi); 624 struct atgc_management *am = &sbi->am; 625 struct victim_entry *ve; 626 unsigned long long age; 627 unsigned long long max_mtime = sit_i->dirty_max_mtime; 628 unsigned long long min_mtime = sit_i->dirty_min_mtime; 629 unsigned int vblocks; 630 unsigned int dirty_threshold = max(am->max_candidate_count, 631 am->candidate_ratio * 632 am->victim_count / 100); 633 unsigned int cost, iter; 634 int stage = 0; 635 636 if (max_mtime < min_mtime) 637 return; 638 max_mtime += 1; 639 next_stage: 640 iter = 0; 641 ve = __lookup_victim_entry(sbi, p->age); 642 next_node: 643 if (!ve) { 644 if (stage++ == 0) 645 goto next_stage; 646 return; 647 } 648 649 if (ve->mtime >= max_mtime || ve->mtime < min_mtime) 650 goto skip_node; 651 652 age = max_mtime - ve->mtime; 653 654 vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks; 655 f2fs_bug_on(sbi, !vblocks); 656 657 /* rare case */ 658 if (vblocks == BLKS_PER_SEG(sbi)) 659 goto skip_node; 660 661 iter++; 662 663 age = max_mtime - abs(p->age - age); 664 cost = UINT_MAX - vblocks; 665 666 if (cost < p->min_cost || 667 (cost == p->min_cost && age > p->oldest_age)) { 668 p->min_cost = cost; 669 p->oldest_age = age; 670 p->min_segno = ve->segno; 671 } 672 skip_node: 673 if (iter < dirty_threshold) { 674 ve = rb_entry(stage == 0 ? rb_prev(&ve->rb_node) : 675 rb_next(&ve->rb_node), 676 struct victim_entry, rb_node); 677 goto next_node; 678 } 679 680 if (stage++ == 0) 681 goto next_stage; 682 } 683 684 static void lookup_victim_by_age(struct f2fs_sb_info *sbi, 685 struct victim_sel_policy *p) 686 { 687 f2fs_bug_on(sbi, !f2fs_check_victim_tree(sbi, &sbi->am.root)); 688 689 if (p->gc_mode == GC_AT) 690 atgc_lookup_victim(sbi, p); 691 else if (p->alloc_mode == AT_SSR) 692 atssr_lookup_victim(sbi, p); 693 else 694 f2fs_bug_on(sbi, 1); 695 } 696 697 static void release_victim_entry(struct f2fs_sb_info *sbi) 698 { 699 struct atgc_management *am = &sbi->am; 700 struct victim_entry *ve, *tmp; 701 702 list_for_each_entry_safe(ve, tmp, &am->victim_list, list) { 703 list_del(&ve->list); 704 kmem_cache_free(victim_entry_slab, ve); 705 am->victim_count--; 706 } 707 708 am->root = RB_ROOT_CACHED; 709 710 f2fs_bug_on(sbi, am->victim_count); 711 f2fs_bug_on(sbi, !list_empty(&am->victim_list)); 712 } 713 714 static bool f2fs_pin_section(struct f2fs_sb_info *sbi, unsigned int segno) 715 { 716 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 717 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 718 719 if (!dirty_i->enable_pin_section) 720 return false; 721 if (!test_and_set_bit(secno, dirty_i->pinned_secmap)) 722 dirty_i->pinned_secmap_cnt++; 723 return true; 724 } 725 726 static bool f2fs_pinned_section_exists(struct dirty_seglist_info *dirty_i) 727 { 728 return dirty_i->pinned_secmap_cnt; 729 } 730 731 static bool f2fs_section_is_pinned(struct dirty_seglist_info *dirty_i, 732 unsigned int secno) 733 { 734 return dirty_i->enable_pin_section && 735 f2fs_pinned_section_exists(dirty_i) && 736 test_bit(secno, dirty_i->pinned_secmap); 737 } 738 739 static void f2fs_unpin_all_sections(struct f2fs_sb_info *sbi, bool enable) 740 { 741 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 742 743 if (f2fs_pinned_section_exists(DIRTY_I(sbi))) { 744 memset(DIRTY_I(sbi)->pinned_secmap, 0, bitmap_size); 745 DIRTY_I(sbi)->pinned_secmap_cnt = 0; 746 } 747 DIRTY_I(sbi)->enable_pin_section = enable; 748 } 749 750 static int f2fs_gc_pinned_control(struct inode *inode, int gc_type, 751 unsigned int segno) 752 { 753 if (!f2fs_is_pinned_file(inode)) 754 return 0; 755 if (gc_type != FG_GC) 756 return -EBUSY; 757 if (!f2fs_pin_section(F2FS_I_SB(inode), segno)) 758 f2fs_pin_file_control(inode, true); 759 return -EAGAIN; 760 } 761 762 /* 763 * This function is called from two paths. 764 * One is garbage collection and the other is SSR segment selection. 765 * When it is called during GC, it just gets a victim segment 766 * and it does not remove it from dirty seglist. 767 * When it is called from SSR segment selection, it finds a segment 768 * which has minimum valid blocks and removes it from dirty seglist. 769 */ 770 int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result, 771 int gc_type, int type, char alloc_mode, 772 unsigned long long age, bool one_time) 773 { 774 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 775 struct sit_info *sm = SIT_I(sbi); 776 struct victim_sel_policy p; 777 unsigned int secno, last_victim; 778 unsigned int last_segment; 779 unsigned int nsearched; 780 bool is_atgc; 781 int ret = 0; 782 783 mutex_lock(&dirty_i->seglist_lock); 784 last_segment = MAIN_SECS(sbi) * SEGS_PER_SEC(sbi); 785 786 p.alloc_mode = alloc_mode; 787 p.age = age; 788 p.age_threshold = sbi->am.age_threshold; 789 p.one_time_gc = one_time; 790 791 retry: 792 select_policy(sbi, gc_type, type, &p); 793 p.min_segno = NULL_SEGNO; 794 p.oldest_age = 0; 795 p.min_cost = get_max_cost(sbi, &p); 796 797 is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR); 798 nsearched = 0; 799 800 if (is_atgc) 801 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX; 802 803 if (*result != NULL_SEGNO) { 804 if (!get_valid_blocks(sbi, *result, false)) { 805 ret = -ENODATA; 806 goto out; 807 } 808 809 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result))) { 810 ret = -EBUSY; 811 goto out; 812 } 813 if (gc_type == FG_GC) 814 clear_bit(GET_SEC_FROM_SEG(sbi, *result), dirty_i->victim_secmap); 815 p.min_segno = *result; 816 goto got_result; 817 } 818 819 ret = -ENODATA; 820 if (p.max_search == 0) 821 goto out; 822 823 if (__is_large_section(sbi) && p.alloc_mode == LFS) { 824 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) { 825 p.min_segno = sbi->next_victim_seg[BG_GC]; 826 *result = p.min_segno; 827 sbi->next_victim_seg[BG_GC] = NULL_SEGNO; 828 goto got_result; 829 } 830 if (gc_type == FG_GC && 831 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) { 832 p.min_segno = sbi->next_victim_seg[FG_GC]; 833 *result = p.min_segno; 834 sbi->next_victim_seg[FG_GC] = NULL_SEGNO; 835 goto got_result; 836 } 837 } 838 839 last_victim = sm->last_victim[p.gc_mode]; 840 if (p.alloc_mode == LFS && gc_type == FG_GC) { 841 p.min_segno = check_bg_victims(sbi); 842 if (p.min_segno != NULL_SEGNO) 843 goto got_it; 844 } 845 846 while (1) { 847 unsigned long cost, *dirty_bitmap; 848 unsigned int unit_no, segno; 849 850 dirty_bitmap = p.dirty_bitmap; 851 unit_no = find_next_bit(dirty_bitmap, 852 last_segment / p.ofs_unit, 853 p.offset / p.ofs_unit); 854 segno = unit_no * p.ofs_unit; 855 if (segno >= last_segment) { 856 if (sm->last_victim[p.gc_mode]) { 857 last_segment = 858 sm->last_victim[p.gc_mode]; 859 sm->last_victim[p.gc_mode] = 0; 860 p.offset = 0; 861 continue; 862 } 863 break; 864 } 865 866 p.offset = segno + p.ofs_unit; 867 nsearched++; 868 869 #ifdef CONFIG_F2FS_CHECK_FS 870 /* 871 * skip selecting the invalid segno (that is failed due to block 872 * validity check failure during GC) to avoid endless GC loop in 873 * such cases. 874 */ 875 if (test_bit(segno, sm->invalid_segmap)) 876 goto next; 877 #endif 878 879 secno = GET_SEC_FROM_SEG(sbi, segno); 880 881 if (sec_usage_check(sbi, secno)) 882 goto next; 883 884 /* Don't touch checkpointed data */ 885 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 886 if (p.alloc_mode == LFS) { 887 /* 888 * LFS is set to find source section during GC. 889 * The victim should have no checkpointed data. 890 */ 891 if (get_ckpt_valid_blocks(sbi, segno, true)) 892 goto next; 893 } else { 894 /* 895 * SSR | AT_SSR are set to find target segment 896 * for writes which can be full by checkpointed 897 * and newly written blocks. 898 */ 899 if (!f2fs_segment_has_free_slot(sbi, segno)) 900 goto next; 901 } 902 } 903 904 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) 905 goto next; 906 907 if (gc_type == FG_GC && f2fs_section_is_pinned(dirty_i, secno)) 908 goto next; 909 910 if (is_atgc) { 911 add_victim_entry(sbi, &p, segno); 912 goto next; 913 } 914 915 cost = get_gc_cost(sbi, segno, &p); 916 917 if (p.min_cost > cost) { 918 p.min_segno = segno; 919 p.min_cost = cost; 920 } 921 next: 922 if (nsearched >= p.max_search) { 923 if (!sm->last_victim[p.gc_mode] && segno <= last_victim) 924 sm->last_victim[p.gc_mode] = 925 last_victim + p.ofs_unit; 926 else 927 sm->last_victim[p.gc_mode] = segno + p.ofs_unit; 928 sm->last_victim[p.gc_mode] %= 929 (MAIN_SECS(sbi) * SEGS_PER_SEC(sbi)); 930 break; 931 } 932 } 933 934 /* get victim for GC_AT/AT_SSR */ 935 if (is_atgc) { 936 lookup_victim_by_age(sbi, &p); 937 release_victim_entry(sbi); 938 } 939 940 if (is_atgc && p.min_segno == NULL_SEGNO && 941 sm->elapsed_time < p.age_threshold) { 942 p.age_threshold = 0; 943 goto retry; 944 } 945 946 if (p.min_segno != NULL_SEGNO) { 947 got_it: 948 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; 949 got_result: 950 if (p.alloc_mode == LFS) { 951 secno = GET_SEC_FROM_SEG(sbi, p.min_segno); 952 if (gc_type == FG_GC) 953 sbi->cur_victim_sec = secno; 954 else 955 set_bit(secno, dirty_i->victim_secmap); 956 } 957 ret = 0; 958 959 } 960 out: 961 if (p.min_segno != NULL_SEGNO) 962 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, 963 sbi->cur_victim_sec, 964 prefree_segments(sbi), free_segments(sbi)); 965 mutex_unlock(&dirty_i->seglist_lock); 966 967 return ret; 968 } 969 970 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) 971 { 972 struct inode_entry *ie; 973 974 ie = radix_tree_lookup(&gc_list->iroot, ino); 975 if (ie) 976 return ie->inode; 977 return NULL; 978 } 979 980 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) 981 { 982 struct inode_entry *new_ie; 983 984 if (inode == find_gc_inode(gc_list, inode->i_ino)) { 985 iput(inode); 986 return; 987 } 988 new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, 989 GFP_NOFS, true, NULL); 990 new_ie->inode = inode; 991 992 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); 993 list_add_tail(&new_ie->list, &gc_list->ilist); 994 } 995 996 static void put_gc_inode(struct gc_inode_list *gc_list) 997 { 998 struct inode_entry *ie, *next_ie; 999 1000 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { 1001 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); 1002 iput(ie->inode); 1003 list_del(&ie->list); 1004 kmem_cache_free(f2fs_inode_entry_slab, ie); 1005 } 1006 } 1007 1008 static int check_valid_map(struct f2fs_sb_info *sbi, 1009 unsigned int segno, int offset) 1010 { 1011 struct sit_info *sit_i = SIT_I(sbi); 1012 struct seg_entry *sentry; 1013 int ret; 1014 1015 down_read(&sit_i->sentry_lock); 1016 sentry = get_seg_entry(sbi, segno); 1017 ret = f2fs_test_bit(offset, sentry->cur_valid_map); 1018 up_read(&sit_i->sentry_lock); 1019 return ret; 1020 } 1021 1022 /* 1023 * This function compares node address got in summary with that in NAT. 1024 * On validity, copy that node with cold status, otherwise (invalid node) 1025 * ignore that. 1026 */ 1027 static int gc_node_segment(struct f2fs_sb_info *sbi, 1028 struct f2fs_summary *sum, unsigned int segno, int gc_type) 1029 { 1030 struct f2fs_summary *entry; 1031 block_t start_addr; 1032 int off; 1033 int phase = 0; 1034 bool fggc = (gc_type == FG_GC); 1035 int submitted = 0; 1036 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 1037 1038 start_addr = START_BLOCK(sbi, segno); 1039 1040 next_step: 1041 entry = sum; 1042 1043 if (fggc && phase == 2) 1044 atomic_inc(&sbi->wb_sync_req[NODE]); 1045 1046 for (off = 0; off < usable_blks_in_seg; off++, entry++) { 1047 nid_t nid = le32_to_cpu(entry->nid); 1048 struct page *node_page; 1049 struct node_info ni; 1050 int err; 1051 1052 /* stop BG_GC if there is not enough free sections. */ 1053 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) 1054 return submitted; 1055 1056 if (check_valid_map(sbi, segno, off) == 0) 1057 continue; 1058 1059 if (phase == 0) { 1060 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, 1061 META_NAT, true); 1062 continue; 1063 } 1064 1065 if (phase == 1) { 1066 f2fs_ra_node_page(sbi, nid); 1067 continue; 1068 } 1069 1070 /* phase == 2 */ 1071 node_page = f2fs_get_node_page(sbi, nid); 1072 if (IS_ERR(node_page)) 1073 continue; 1074 1075 /* block may become invalid during f2fs_get_node_page */ 1076 if (check_valid_map(sbi, segno, off) == 0) { 1077 f2fs_put_page(node_page, 1); 1078 continue; 1079 } 1080 1081 if (f2fs_get_node_info(sbi, nid, &ni, false)) { 1082 f2fs_put_page(node_page, 1); 1083 continue; 1084 } 1085 1086 if (ni.blk_addr != start_addr + off) { 1087 f2fs_put_page(node_page, 1); 1088 continue; 1089 } 1090 1091 err = f2fs_move_node_page(node_page, gc_type); 1092 if (!err && gc_type == FG_GC) 1093 submitted++; 1094 stat_inc_node_blk_count(sbi, 1, gc_type); 1095 } 1096 1097 if (++phase < 3) 1098 goto next_step; 1099 1100 if (fggc) 1101 atomic_dec(&sbi->wb_sync_req[NODE]); 1102 return submitted; 1103 } 1104 1105 /* 1106 * Calculate start block index indicating the given node offset. 1107 * Be careful, caller should give this node offset only indicating direct node 1108 * blocks. If any node offsets, which point the other types of node blocks such 1109 * as indirect or double indirect node blocks, are given, it must be a caller's 1110 * bug. 1111 */ 1112 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode) 1113 { 1114 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; 1115 unsigned int bidx; 1116 1117 if (node_ofs == 0) 1118 return 0; 1119 1120 if (node_ofs <= 2) { 1121 bidx = node_ofs - 1; 1122 } else if (node_ofs <= indirect_blks) { 1123 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); 1124 1125 bidx = node_ofs - 2 - dec; 1126 } else { 1127 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); 1128 1129 bidx = node_ofs - 5 - dec; 1130 } 1131 return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode); 1132 } 1133 1134 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 1135 struct node_info *dni, block_t blkaddr, unsigned int *nofs) 1136 { 1137 struct page *node_page; 1138 nid_t nid; 1139 unsigned int ofs_in_node, max_addrs, base; 1140 block_t source_blkaddr; 1141 1142 nid = le32_to_cpu(sum->nid); 1143 ofs_in_node = le16_to_cpu(sum->ofs_in_node); 1144 1145 node_page = f2fs_get_node_page(sbi, nid); 1146 if (IS_ERR(node_page)) 1147 return false; 1148 1149 if (f2fs_get_node_info(sbi, nid, dni, false)) { 1150 f2fs_put_page(node_page, 1); 1151 return false; 1152 } 1153 1154 if (sum->version != dni->version) { 1155 f2fs_warn(sbi, "%s: valid data with mismatched node version.", 1156 __func__); 1157 set_sbi_flag(sbi, SBI_NEED_FSCK); 1158 } 1159 1160 if (f2fs_check_nid_range(sbi, dni->ino)) { 1161 f2fs_put_page(node_page, 1); 1162 return false; 1163 } 1164 1165 if (IS_INODE(node_page)) { 1166 base = offset_in_addr(F2FS_INODE(node_page)); 1167 max_addrs = DEF_ADDRS_PER_INODE; 1168 } else { 1169 base = 0; 1170 max_addrs = DEF_ADDRS_PER_BLOCK; 1171 } 1172 1173 if (base + ofs_in_node >= max_addrs) { 1174 f2fs_err(sbi, "Inconsistent blkaddr offset: base:%u, ofs_in_node:%u, max:%u, ino:%u, nid:%u", 1175 base, ofs_in_node, max_addrs, dni->ino, dni->nid); 1176 f2fs_put_page(node_page, 1); 1177 return false; 1178 } 1179 1180 *nofs = ofs_of_node(node_page); 1181 source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node); 1182 f2fs_put_page(node_page, 1); 1183 1184 if (source_blkaddr != blkaddr) { 1185 #ifdef CONFIG_F2FS_CHECK_FS 1186 unsigned int segno = GET_SEGNO(sbi, blkaddr); 1187 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 1188 1189 if (unlikely(check_valid_map(sbi, segno, offset))) { 1190 if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) { 1191 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u", 1192 blkaddr, source_blkaddr, segno); 1193 set_sbi_flag(sbi, SBI_NEED_FSCK); 1194 } 1195 } 1196 #endif 1197 return false; 1198 } 1199 return true; 1200 } 1201 1202 static int ra_data_block(struct inode *inode, pgoff_t index) 1203 { 1204 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1205 struct address_space *mapping = f2fs_is_cow_file(inode) ? 1206 F2FS_I(inode)->atomic_inode->i_mapping : inode->i_mapping; 1207 struct dnode_of_data dn; 1208 struct page *page; 1209 struct f2fs_io_info fio = { 1210 .sbi = sbi, 1211 .ino = inode->i_ino, 1212 .type = DATA, 1213 .temp = COLD, 1214 .op = REQ_OP_READ, 1215 .op_flags = 0, 1216 .encrypted_page = NULL, 1217 .in_list = 0, 1218 }; 1219 int err; 1220 1221 page = f2fs_grab_cache_page(mapping, index, true); 1222 if (!page) 1223 return -ENOMEM; 1224 1225 if (f2fs_lookup_read_extent_cache_block(inode, index, 1226 &dn.data_blkaddr)) { 1227 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, 1228 DATA_GENERIC_ENHANCE_READ))) { 1229 err = -EFSCORRUPTED; 1230 goto put_page; 1231 } 1232 goto got_it; 1233 } 1234 1235 set_new_dnode(&dn, inode, NULL, NULL, 0); 1236 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 1237 if (err) 1238 goto put_page; 1239 f2fs_put_dnode(&dn); 1240 1241 if (!__is_valid_data_blkaddr(dn.data_blkaddr)) { 1242 err = -ENOENT; 1243 goto put_page; 1244 } 1245 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, 1246 DATA_GENERIC_ENHANCE))) { 1247 err = -EFSCORRUPTED; 1248 goto put_page; 1249 } 1250 got_it: 1251 /* read page */ 1252 fio.page = page; 1253 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; 1254 1255 /* 1256 * don't cache encrypted data into meta inode until previous dirty 1257 * data were writebacked to avoid racing between GC and flush. 1258 */ 1259 f2fs_wait_on_page_writeback(page, DATA, true, true); 1260 1261 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); 1262 1263 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi), 1264 dn.data_blkaddr, 1265 FGP_LOCK | FGP_CREAT, GFP_NOFS); 1266 if (!fio.encrypted_page) { 1267 err = -ENOMEM; 1268 goto put_page; 1269 } 1270 1271 err = f2fs_submit_page_bio(&fio); 1272 if (err) 1273 goto put_encrypted_page; 1274 f2fs_put_page(fio.encrypted_page, 0); 1275 f2fs_put_page(page, 1); 1276 1277 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE); 1278 f2fs_update_iostat(sbi, NULL, FS_GDATA_READ_IO, F2FS_BLKSIZE); 1279 1280 return 0; 1281 put_encrypted_page: 1282 f2fs_put_page(fio.encrypted_page, 1); 1283 put_page: 1284 f2fs_put_page(page, 1); 1285 return err; 1286 } 1287 1288 /* 1289 * Move data block via META_MAPPING while keeping locked data page. 1290 * This can be used to move blocks, aka LBAs, directly on disk. 1291 */ 1292 static int move_data_block(struct inode *inode, block_t bidx, 1293 int gc_type, unsigned int segno, int off) 1294 { 1295 struct address_space *mapping = f2fs_is_cow_file(inode) ? 1296 F2FS_I(inode)->atomic_inode->i_mapping : inode->i_mapping; 1297 struct f2fs_io_info fio = { 1298 .sbi = F2FS_I_SB(inode), 1299 .ino = inode->i_ino, 1300 .type = DATA, 1301 .temp = COLD, 1302 .op = REQ_OP_READ, 1303 .op_flags = 0, 1304 .encrypted_page = NULL, 1305 .in_list = 0, 1306 }; 1307 struct dnode_of_data dn; 1308 struct f2fs_summary sum; 1309 struct node_info ni; 1310 struct page *page, *mpage; 1311 block_t newaddr; 1312 int err = 0; 1313 bool lfs_mode = f2fs_lfs_mode(fio.sbi); 1314 int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) && 1315 (fio.sbi->gc_mode != GC_URGENT_HIGH) ? 1316 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA; 1317 1318 /* do not read out */ 1319 page = f2fs_grab_cache_page(mapping, bidx, false); 1320 if (!page) 1321 return -ENOMEM; 1322 1323 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) { 1324 err = -ENOENT; 1325 goto out; 1326 } 1327 1328 err = f2fs_gc_pinned_control(inode, gc_type, segno); 1329 if (err) 1330 goto out; 1331 1332 set_new_dnode(&dn, inode, NULL, NULL, 0); 1333 err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE); 1334 if (err) 1335 goto out; 1336 1337 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 1338 ClearPageUptodate(page); 1339 err = -ENOENT; 1340 goto put_out; 1341 } 1342 1343 /* 1344 * don't cache encrypted data into meta inode until previous dirty 1345 * data were writebacked to avoid racing between GC and flush. 1346 */ 1347 f2fs_wait_on_page_writeback(page, DATA, true, true); 1348 1349 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); 1350 1351 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false); 1352 if (err) 1353 goto put_out; 1354 1355 /* read page */ 1356 fio.page = page; 1357 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; 1358 1359 if (lfs_mode) 1360 f2fs_down_write(&fio.sbi->io_order_lock); 1361 1362 mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi), 1363 fio.old_blkaddr, false); 1364 if (!mpage) { 1365 err = -ENOMEM; 1366 goto up_out; 1367 } 1368 1369 fio.encrypted_page = mpage; 1370 1371 /* read source block in mpage */ 1372 if (!PageUptodate(mpage)) { 1373 err = f2fs_submit_page_bio(&fio); 1374 if (err) { 1375 f2fs_put_page(mpage, 1); 1376 goto up_out; 1377 } 1378 1379 f2fs_update_iostat(fio.sbi, inode, FS_DATA_READ_IO, 1380 F2FS_BLKSIZE); 1381 f2fs_update_iostat(fio.sbi, NULL, FS_GDATA_READ_IO, 1382 F2FS_BLKSIZE); 1383 1384 lock_page(mpage); 1385 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) || 1386 !PageUptodate(mpage))) { 1387 err = -EIO; 1388 f2fs_put_page(mpage, 1); 1389 goto up_out; 1390 } 1391 } 1392 1393 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); 1394 1395 /* allocate block address */ 1396 err = f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr, 1397 &sum, type, NULL); 1398 if (err) { 1399 f2fs_put_page(mpage, 1); 1400 /* filesystem should shutdown, no need to recovery block */ 1401 goto up_out; 1402 } 1403 1404 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi), 1405 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS); 1406 if (!fio.encrypted_page) { 1407 err = -ENOMEM; 1408 f2fs_put_page(mpage, 1); 1409 goto recover_block; 1410 } 1411 1412 /* write target block */ 1413 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true); 1414 memcpy(page_address(fio.encrypted_page), 1415 page_address(mpage), PAGE_SIZE); 1416 f2fs_put_page(mpage, 1); 1417 1418 f2fs_invalidate_internal_cache(fio.sbi, fio.old_blkaddr, 1); 1419 1420 set_page_dirty(fio.encrypted_page); 1421 if (clear_page_dirty_for_io(fio.encrypted_page)) 1422 dec_page_count(fio.sbi, F2FS_DIRTY_META); 1423 1424 set_page_writeback(fio.encrypted_page); 1425 1426 fio.op = REQ_OP_WRITE; 1427 fio.op_flags = REQ_SYNC; 1428 fio.new_blkaddr = newaddr; 1429 f2fs_submit_page_write(&fio); 1430 1431 f2fs_update_iostat(fio.sbi, NULL, FS_GC_DATA_IO, F2FS_BLKSIZE); 1432 1433 f2fs_update_data_blkaddr(&dn, newaddr); 1434 set_inode_flag(inode, FI_APPEND_WRITE); 1435 1436 f2fs_put_page(fio.encrypted_page, 1); 1437 recover_block: 1438 if (err) 1439 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr, 1440 true, true, true); 1441 up_out: 1442 if (lfs_mode) 1443 f2fs_up_write(&fio.sbi->io_order_lock); 1444 put_out: 1445 f2fs_put_dnode(&dn); 1446 out: 1447 f2fs_put_page(page, 1); 1448 return err; 1449 } 1450 1451 static int move_data_page(struct inode *inode, block_t bidx, int gc_type, 1452 unsigned int segno, int off) 1453 { 1454 struct folio *folio; 1455 int err = 0; 1456 1457 folio = f2fs_get_lock_data_folio(inode, bidx, true); 1458 if (IS_ERR(folio)) 1459 return PTR_ERR(folio); 1460 1461 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) { 1462 err = -ENOENT; 1463 goto out; 1464 } 1465 1466 err = f2fs_gc_pinned_control(inode, gc_type, segno); 1467 if (err) 1468 goto out; 1469 1470 if (gc_type == BG_GC) { 1471 if (folio_test_writeback(folio)) { 1472 err = -EAGAIN; 1473 goto out; 1474 } 1475 folio_mark_dirty(folio); 1476 set_page_private_gcing(&folio->page); 1477 } else { 1478 struct f2fs_io_info fio = { 1479 .sbi = F2FS_I_SB(inode), 1480 .ino = inode->i_ino, 1481 .type = DATA, 1482 .temp = COLD, 1483 .op = REQ_OP_WRITE, 1484 .op_flags = REQ_SYNC, 1485 .old_blkaddr = NULL_ADDR, 1486 .page = &folio->page, 1487 .encrypted_page = NULL, 1488 .need_lock = LOCK_REQ, 1489 .io_type = FS_GC_DATA_IO, 1490 }; 1491 bool is_dirty = folio_test_dirty(folio); 1492 1493 retry: 1494 f2fs_folio_wait_writeback(folio, DATA, true, true); 1495 1496 folio_mark_dirty(folio); 1497 if (folio_clear_dirty_for_io(folio)) { 1498 inode_dec_dirty_pages(inode); 1499 f2fs_remove_dirty_inode(inode); 1500 } 1501 1502 set_page_private_gcing(&folio->page); 1503 1504 err = f2fs_do_write_data_page(&fio); 1505 if (err) { 1506 clear_page_private_gcing(&folio->page); 1507 if (err == -ENOMEM) { 1508 memalloc_retry_wait(GFP_NOFS); 1509 goto retry; 1510 } 1511 if (is_dirty) 1512 folio_mark_dirty(folio); 1513 } 1514 } 1515 out: 1516 f2fs_folio_put(folio, true); 1517 return err; 1518 } 1519 1520 /* 1521 * This function tries to get parent node of victim data block, and identifies 1522 * data block validity. If the block is valid, copy that with cold status and 1523 * modify parent node. 1524 * If the parent node is not valid or the data block address is different, 1525 * the victim data block is ignored. 1526 */ 1527 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 1528 struct gc_inode_list *gc_list, unsigned int segno, int gc_type, 1529 bool force_migrate) 1530 { 1531 struct super_block *sb = sbi->sb; 1532 struct f2fs_summary *entry; 1533 block_t start_addr; 1534 int off; 1535 int phase = 0; 1536 int submitted = 0; 1537 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 1538 1539 start_addr = START_BLOCK(sbi, segno); 1540 1541 next_step: 1542 entry = sum; 1543 1544 for (off = 0; off < usable_blks_in_seg; off++, entry++) { 1545 struct inode *inode; 1546 struct node_info dni; /* dnode info for the data */ 1547 unsigned int ofs_in_node, nofs; 1548 block_t start_bidx; 1549 nid_t nid = le32_to_cpu(entry->nid); 1550 1551 /* 1552 * stop BG_GC if there is not enough free sections. 1553 * Or, stop GC if the segment becomes fully valid caused by 1554 * race condition along with SSR block allocation. 1555 */ 1556 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) || 1557 (!force_migrate && get_valid_blocks(sbi, segno, true) == 1558 CAP_BLKS_PER_SEC(sbi))) 1559 return submitted; 1560 1561 if (check_valid_map(sbi, segno, off) == 0) 1562 continue; 1563 1564 if (phase == 0) { 1565 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, 1566 META_NAT, true); 1567 continue; 1568 } 1569 1570 if (phase == 1) { 1571 f2fs_ra_node_page(sbi, nid); 1572 continue; 1573 } 1574 1575 /* Get an inode by ino with checking validity */ 1576 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) 1577 continue; 1578 1579 if (phase == 2) { 1580 f2fs_ra_node_page(sbi, dni.ino); 1581 continue; 1582 } 1583 1584 ofs_in_node = le16_to_cpu(entry->ofs_in_node); 1585 1586 if (phase == 3) { 1587 struct folio *data_folio; 1588 int err; 1589 1590 inode = f2fs_iget(sb, dni.ino); 1591 if (IS_ERR(inode)) 1592 continue; 1593 1594 if (is_bad_inode(inode) || 1595 special_file(inode->i_mode)) { 1596 iput(inode); 1597 continue; 1598 } 1599 1600 if (f2fs_has_inline_data(inode)) { 1601 iput(inode); 1602 set_sbi_flag(sbi, SBI_NEED_FSCK); 1603 f2fs_err_ratelimited(sbi, 1604 "inode %lx has both inline_data flag and " 1605 "data block, nid=%u, ofs_in_node=%u", 1606 inode->i_ino, dni.nid, ofs_in_node); 1607 continue; 1608 } 1609 1610 err = f2fs_gc_pinned_control(inode, gc_type, segno); 1611 if (err == -EAGAIN) { 1612 iput(inode); 1613 return submitted; 1614 } 1615 1616 if (!f2fs_down_write_trylock( 1617 &F2FS_I(inode)->i_gc_rwsem[WRITE])) { 1618 iput(inode); 1619 sbi->skipped_gc_rwsem++; 1620 continue; 1621 } 1622 1623 start_bidx = f2fs_start_bidx_of_node(nofs, inode) + 1624 ofs_in_node; 1625 1626 if (f2fs_meta_inode_gc_required(inode)) { 1627 int err = ra_data_block(inode, start_bidx); 1628 1629 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1630 if (err) { 1631 iput(inode); 1632 continue; 1633 } 1634 add_gc_inode(gc_list, inode); 1635 continue; 1636 } 1637 1638 data_folio = f2fs_get_read_data_folio(inode, start_bidx, 1639 REQ_RAHEAD, true, NULL); 1640 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1641 if (IS_ERR(data_folio)) { 1642 iput(inode); 1643 continue; 1644 } 1645 1646 f2fs_folio_put(data_folio, false); 1647 add_gc_inode(gc_list, inode); 1648 continue; 1649 } 1650 1651 /* phase 4 */ 1652 inode = find_gc_inode(gc_list, dni.ino); 1653 if (inode) { 1654 struct f2fs_inode_info *fi = F2FS_I(inode); 1655 bool locked = false; 1656 int err; 1657 1658 if (S_ISREG(inode->i_mode)) { 1659 if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[WRITE])) { 1660 sbi->skipped_gc_rwsem++; 1661 continue; 1662 } 1663 if (!f2fs_down_write_trylock( 1664 &fi->i_gc_rwsem[READ])) { 1665 sbi->skipped_gc_rwsem++; 1666 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 1667 continue; 1668 } 1669 locked = true; 1670 1671 /* wait for all inflight aio data */ 1672 inode_dio_wait(inode); 1673 } 1674 1675 start_bidx = f2fs_start_bidx_of_node(nofs, inode) 1676 + ofs_in_node; 1677 if (f2fs_meta_inode_gc_required(inode)) 1678 err = move_data_block(inode, start_bidx, 1679 gc_type, segno, off); 1680 else 1681 err = move_data_page(inode, start_bidx, gc_type, 1682 segno, off); 1683 1684 if (!err && (gc_type == FG_GC || 1685 f2fs_meta_inode_gc_required(inode))) 1686 submitted++; 1687 1688 if (locked) { 1689 f2fs_up_write(&fi->i_gc_rwsem[READ]); 1690 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 1691 } 1692 1693 stat_inc_data_blk_count(sbi, 1, gc_type); 1694 } 1695 } 1696 1697 if (++phase < 5) 1698 goto next_step; 1699 1700 return submitted; 1701 } 1702 1703 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, 1704 int gc_type, bool one_time) 1705 { 1706 struct sit_info *sit_i = SIT_I(sbi); 1707 int ret; 1708 1709 down_write(&sit_i->sentry_lock); 1710 ret = f2fs_get_victim(sbi, victim, gc_type, NO_CHECK_TYPE, 1711 LFS, 0, one_time); 1712 up_write(&sit_i->sentry_lock); 1713 return ret; 1714 } 1715 1716 static int do_garbage_collect(struct f2fs_sb_info *sbi, 1717 unsigned int start_segno, 1718 struct gc_inode_list *gc_list, int gc_type, 1719 bool force_migrate, bool one_time) 1720 { 1721 struct page *sum_page; 1722 struct f2fs_summary_block *sum; 1723 struct blk_plug plug; 1724 unsigned int segno = start_segno; 1725 unsigned int end_segno = start_segno + SEGS_PER_SEC(sbi); 1726 unsigned int sec_end_segno; 1727 int seg_freed = 0, migrated = 0; 1728 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ? 1729 SUM_TYPE_DATA : SUM_TYPE_NODE; 1730 unsigned char data_type = (type == SUM_TYPE_DATA) ? DATA : NODE; 1731 int submitted = 0; 1732 1733 if (__is_large_section(sbi)) { 1734 sec_end_segno = rounddown(end_segno, SEGS_PER_SEC(sbi)); 1735 1736 /* 1737 * zone-capacity can be less than zone-size in zoned devices, 1738 * resulting in less than expected usable segments in the zone, 1739 * calculate the end segno in the zone which can be garbage 1740 * collected 1741 */ 1742 if (f2fs_sb_has_blkzoned(sbi)) 1743 sec_end_segno -= SEGS_PER_SEC(sbi) - 1744 f2fs_usable_segs_in_sec(sbi); 1745 1746 if (gc_type == BG_GC || one_time) { 1747 unsigned int window_granularity = 1748 sbi->migration_window_granularity; 1749 1750 if (f2fs_sb_has_blkzoned(sbi) && 1751 !has_enough_free_blocks(sbi, 1752 sbi->gc_thread->boost_zoned_gc_percent)) 1753 window_granularity *= 1754 BOOST_GC_MULTIPLE; 1755 1756 end_segno = start_segno + window_granularity; 1757 } 1758 1759 if (end_segno > sec_end_segno) 1760 end_segno = sec_end_segno; 1761 } 1762 1763 sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type); 1764 1765 /* readahead multi ssa blocks those have contiguous address */ 1766 if (__is_large_section(sbi)) 1767 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), 1768 end_segno - segno, META_SSA, true); 1769 1770 /* reference all summary page */ 1771 while (segno < end_segno) { 1772 sum_page = f2fs_get_sum_page(sbi, segno++); 1773 if (IS_ERR(sum_page)) { 1774 int err = PTR_ERR(sum_page); 1775 1776 end_segno = segno - 1; 1777 for (segno = start_segno; segno < end_segno; segno++) { 1778 sum_page = find_get_page(META_MAPPING(sbi), 1779 GET_SUM_BLOCK(sbi, segno)); 1780 f2fs_put_page(sum_page, 0); 1781 f2fs_put_page(sum_page, 0); 1782 } 1783 return err; 1784 } 1785 unlock_page(sum_page); 1786 } 1787 1788 blk_start_plug(&plug); 1789 1790 for (segno = start_segno; segno < end_segno; segno++) { 1791 1792 /* find segment summary of victim */ 1793 sum_page = find_get_page(META_MAPPING(sbi), 1794 GET_SUM_BLOCK(sbi, segno)); 1795 f2fs_put_page(sum_page, 0); 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 (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi))) 1803 goto skip; 1804 1805 sum = page_address(sum_page); 1806 if (type != GET_SUM_TYPE((&sum->footer))) { 1807 f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT", 1808 segno, type, GET_SUM_TYPE((&sum->footer))); 1809 f2fs_stop_checkpoint(sbi, false, 1810 STOP_CP_REASON_CORRUPTED_SUMMARY); 1811 goto skip; 1812 } 1813 1814 /* 1815 * this is to avoid deadlock: 1816 * - lock_page(sum_page) - f2fs_replace_block 1817 * - check_valid_map() - down_write(sentry_lock) 1818 * - down_read(sentry_lock) - change_curseg() 1819 * - lock_page(sum_page) 1820 */ 1821 if (type == SUM_TYPE_NODE) 1822 submitted += gc_node_segment(sbi, sum->entries, segno, 1823 gc_type); 1824 else 1825 submitted += gc_data_segment(sbi, sum->entries, gc_list, 1826 segno, gc_type, 1827 force_migrate); 1828 1829 stat_inc_gc_seg_count(sbi, data_type, gc_type); 1830 sbi->gc_reclaimed_segs[sbi->gc_mode]++; 1831 migrated++; 1832 1833 freed: 1834 if (gc_type == FG_GC && 1835 get_valid_blocks(sbi, segno, false) == 0) 1836 seg_freed++; 1837 1838 if (__is_large_section(sbi)) 1839 sbi->next_victim_seg[gc_type] = 1840 (segno + 1 < sec_end_segno) ? 1841 segno + 1 : NULL_SEGNO; 1842 skip: 1843 f2fs_put_page(sum_page, 0); 1844 } 1845 1846 if (submitted) 1847 f2fs_submit_merged_write(sbi, data_type); 1848 1849 blk_finish_plug(&plug); 1850 1851 if (migrated) 1852 stat_inc_gc_sec_count(sbi, data_type, gc_type); 1853 1854 return seg_freed; 1855 } 1856 1857 int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control) 1858 { 1859 int gc_type = gc_control->init_gc_type; 1860 unsigned int segno = gc_control->victim_segno; 1861 int sec_freed = 0, seg_freed = 0, total_freed = 0, total_sec_freed = 0; 1862 int ret = 0; 1863 struct cp_control cpc; 1864 struct gc_inode_list gc_list = { 1865 .ilist = LIST_HEAD_INIT(gc_list.ilist), 1866 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), 1867 }; 1868 unsigned int skipped_round = 0, round = 0; 1869 unsigned int upper_secs; 1870 1871 trace_f2fs_gc_begin(sbi->sb, gc_type, gc_control->no_bg_gc, 1872 gc_control->nr_free_secs, 1873 get_pages(sbi, F2FS_DIRTY_NODES), 1874 get_pages(sbi, F2FS_DIRTY_DENTS), 1875 get_pages(sbi, F2FS_DIRTY_IMETA), 1876 free_sections(sbi), 1877 free_segments(sbi), 1878 reserved_segments(sbi), 1879 prefree_segments(sbi)); 1880 1881 cpc.reason = __get_cp_reason(sbi); 1882 gc_more: 1883 sbi->skipped_gc_rwsem = 0; 1884 if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) { 1885 ret = -EINVAL; 1886 goto stop; 1887 } 1888 if (unlikely(f2fs_cp_error(sbi))) { 1889 ret = -EIO; 1890 goto stop; 1891 } 1892 1893 /* Let's run FG_GC, if we don't have enough space. */ 1894 if (has_not_enough_free_secs(sbi, 0, 0)) { 1895 gc_type = FG_GC; 1896 1897 /* 1898 * For example, if there are many prefree_segments below given 1899 * threshold, we can make them free by checkpoint. Then, we 1900 * secure free segments which doesn't need fggc any more. 1901 */ 1902 if (prefree_segments(sbi)) { 1903 stat_inc_cp_call_count(sbi, TOTAL_CALL); 1904 ret = f2fs_write_checkpoint(sbi, &cpc); 1905 if (ret) 1906 goto stop; 1907 /* Reset due to checkpoint */ 1908 sec_freed = 0; 1909 } 1910 } 1911 1912 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */ 1913 if (gc_type == BG_GC && gc_control->no_bg_gc) { 1914 ret = -EINVAL; 1915 goto stop; 1916 } 1917 retry: 1918 ret = __get_victim(sbi, &segno, gc_type, gc_control->one_time); 1919 if (ret) { 1920 /* allow to search victim from sections has pinned data */ 1921 if (ret == -ENODATA && gc_type == FG_GC && 1922 f2fs_pinned_section_exists(DIRTY_I(sbi))) { 1923 f2fs_unpin_all_sections(sbi, false); 1924 goto retry; 1925 } 1926 goto stop; 1927 } 1928 1929 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type, 1930 gc_control->should_migrate_blocks, 1931 gc_control->one_time); 1932 if (seg_freed < 0) 1933 goto stop; 1934 1935 total_freed += seg_freed; 1936 1937 if (seg_freed == f2fs_usable_segs_in_sec(sbi)) { 1938 sec_freed++; 1939 total_sec_freed++; 1940 } 1941 1942 if (gc_control->one_time) 1943 goto stop; 1944 1945 if (gc_type == FG_GC) { 1946 sbi->cur_victim_sec = NULL_SEGNO; 1947 1948 if (has_enough_free_secs(sbi, sec_freed, 0)) { 1949 if (!gc_control->no_bg_gc && 1950 total_sec_freed < gc_control->nr_free_secs) 1951 goto go_gc_more; 1952 goto stop; 1953 } 1954 if (sbi->skipped_gc_rwsem) 1955 skipped_round++; 1956 round++; 1957 if (skipped_round > MAX_SKIP_GC_COUNT && 1958 skipped_round * 2 >= round) { 1959 stat_inc_cp_call_count(sbi, TOTAL_CALL); 1960 ret = f2fs_write_checkpoint(sbi, &cpc); 1961 goto stop; 1962 } 1963 } else if (has_enough_free_secs(sbi, 0, 0)) { 1964 goto stop; 1965 } 1966 1967 __get_secs_required(sbi, NULL, &upper_secs, NULL); 1968 1969 /* 1970 * Write checkpoint to reclaim prefree segments. 1971 * We need more three extra sections for writer's data/node/dentry. 1972 */ 1973 if (free_sections(sbi) <= upper_secs + NR_GC_CHECKPOINT_SECS && 1974 prefree_segments(sbi)) { 1975 stat_inc_cp_call_count(sbi, TOTAL_CALL); 1976 ret = f2fs_write_checkpoint(sbi, &cpc); 1977 if (ret) 1978 goto stop; 1979 /* Reset due to checkpoint */ 1980 sec_freed = 0; 1981 } 1982 go_gc_more: 1983 segno = NULL_SEGNO; 1984 goto gc_more; 1985 1986 stop: 1987 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0; 1988 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno; 1989 1990 if (gc_type == FG_GC) 1991 f2fs_unpin_all_sections(sbi, true); 1992 1993 trace_f2fs_gc_end(sbi->sb, ret, total_freed, total_sec_freed, 1994 get_pages(sbi, F2FS_DIRTY_NODES), 1995 get_pages(sbi, F2FS_DIRTY_DENTS), 1996 get_pages(sbi, F2FS_DIRTY_IMETA), 1997 free_sections(sbi), 1998 free_segments(sbi), 1999 reserved_segments(sbi), 2000 prefree_segments(sbi)); 2001 2002 f2fs_up_write(&sbi->gc_lock); 2003 2004 put_gc_inode(&gc_list); 2005 2006 if (gc_control->err_gc_skipped && !ret) 2007 ret = total_sec_freed ? 0 : -EAGAIN; 2008 return ret; 2009 } 2010 2011 int __init f2fs_create_garbage_collection_cache(void) 2012 { 2013 victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry", 2014 sizeof(struct victim_entry)); 2015 return victim_entry_slab ? 0 : -ENOMEM; 2016 } 2017 2018 void f2fs_destroy_garbage_collection_cache(void) 2019 { 2020 kmem_cache_destroy(victim_entry_slab); 2021 } 2022 2023 static void init_atgc_management(struct f2fs_sb_info *sbi) 2024 { 2025 struct atgc_management *am = &sbi->am; 2026 2027 if (test_opt(sbi, ATGC) && 2028 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD) 2029 am->atgc_enabled = true; 2030 2031 am->root = RB_ROOT_CACHED; 2032 INIT_LIST_HEAD(&am->victim_list); 2033 am->victim_count = 0; 2034 2035 am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO; 2036 am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT; 2037 am->age_weight = DEF_GC_THREAD_AGE_WEIGHT; 2038 am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD; 2039 } 2040 2041 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi) 2042 { 2043 sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES; 2044 2045 /* give warm/cold data area from slower device */ 2046 if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi)) 2047 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 2048 GET_SEGNO(sbi, FDEV(0).end_blk) + 1; 2049 2050 init_atgc_management(sbi); 2051 } 2052 2053 int f2fs_gc_range(struct f2fs_sb_info *sbi, 2054 unsigned int start_seg, unsigned int end_seg, 2055 bool dry_run, unsigned int dry_run_sections) 2056 { 2057 unsigned int segno; 2058 unsigned int gc_secs = dry_run_sections; 2059 2060 if (unlikely(f2fs_cp_error(sbi))) 2061 return -EIO; 2062 2063 for (segno = start_seg; segno <= end_seg; segno += SEGS_PER_SEC(sbi)) { 2064 struct gc_inode_list gc_list = { 2065 .ilist = LIST_HEAD_INIT(gc_list.ilist), 2066 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), 2067 }; 2068 2069 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true, false); 2070 put_gc_inode(&gc_list); 2071 2072 if (!dry_run && get_valid_blocks(sbi, segno, true)) 2073 return -EAGAIN; 2074 if (dry_run && dry_run_sections && 2075 !get_valid_blocks(sbi, segno, true) && --gc_secs == 0) 2076 break; 2077 2078 if (fatal_signal_pending(current)) 2079 return -ERESTARTSYS; 2080 } 2081 2082 return 0; 2083 } 2084 2085 static int free_segment_range(struct f2fs_sb_info *sbi, 2086 unsigned int secs, bool dry_run) 2087 { 2088 unsigned int next_inuse, start, end; 2089 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 }; 2090 int gc_mode, gc_type; 2091 int err = 0; 2092 int type; 2093 2094 /* Force block allocation for GC */ 2095 MAIN_SECS(sbi) -= secs; 2096 start = MAIN_SECS(sbi) * SEGS_PER_SEC(sbi); 2097 end = MAIN_SEGS(sbi) - 1; 2098 2099 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2100 for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++) 2101 if (SIT_I(sbi)->last_victim[gc_mode] >= start) 2102 SIT_I(sbi)->last_victim[gc_mode] = 0; 2103 2104 for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++) 2105 if (sbi->next_victim_seg[gc_type] >= start) 2106 sbi->next_victim_seg[gc_type] = NULL_SEGNO; 2107 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2108 2109 /* Move out cursegs from the target range */ 2110 for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++) { 2111 err = f2fs_allocate_segment_for_resize(sbi, type, start, end); 2112 if (err) 2113 goto out; 2114 } 2115 2116 /* do GC to move out valid blocks in the range */ 2117 err = f2fs_gc_range(sbi, start, end, dry_run, 0); 2118 if (err || dry_run) 2119 goto out; 2120 2121 stat_inc_cp_call_count(sbi, TOTAL_CALL); 2122 err = f2fs_write_checkpoint(sbi, &cpc); 2123 if (err) 2124 goto out; 2125 2126 next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start); 2127 if (next_inuse <= end) { 2128 f2fs_err(sbi, "segno %u should be free but still inuse!", 2129 next_inuse); 2130 f2fs_bug_on(sbi, 1); 2131 } 2132 out: 2133 MAIN_SECS(sbi) += secs; 2134 return err; 2135 } 2136 2137 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs) 2138 { 2139 struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi); 2140 int section_count; 2141 int segment_count; 2142 int segment_count_main; 2143 long long block_count; 2144 int segs = secs * SEGS_PER_SEC(sbi); 2145 2146 f2fs_down_write(&sbi->sb_lock); 2147 2148 section_count = le32_to_cpu(raw_sb->section_count); 2149 segment_count = le32_to_cpu(raw_sb->segment_count); 2150 segment_count_main = le32_to_cpu(raw_sb->segment_count_main); 2151 block_count = le64_to_cpu(raw_sb->block_count); 2152 2153 raw_sb->section_count = cpu_to_le32(section_count + secs); 2154 raw_sb->segment_count = cpu_to_le32(segment_count + segs); 2155 raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs); 2156 raw_sb->block_count = cpu_to_le64(block_count + 2157 (long long)SEGS_TO_BLKS(sbi, segs)); 2158 if (f2fs_is_multi_device(sbi)) { 2159 int last_dev = sbi->s_ndevs - 1; 2160 int dev_segs = 2161 le32_to_cpu(raw_sb->devs[last_dev].total_segments); 2162 2163 raw_sb->devs[last_dev].total_segments = 2164 cpu_to_le32(dev_segs + segs); 2165 } 2166 2167 f2fs_up_write(&sbi->sb_lock); 2168 } 2169 2170 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs) 2171 { 2172 int segs = secs * SEGS_PER_SEC(sbi); 2173 long long blks = SEGS_TO_BLKS(sbi, segs); 2174 long long user_block_count = 2175 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count); 2176 2177 SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs; 2178 MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs; 2179 MAIN_SECS(sbi) += secs; 2180 FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs; 2181 FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs; 2182 F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks); 2183 2184 if (f2fs_is_multi_device(sbi)) { 2185 int last_dev = sbi->s_ndevs - 1; 2186 2187 FDEV(last_dev).total_segments = 2188 (int)FDEV(last_dev).total_segments + segs; 2189 FDEV(last_dev).end_blk = 2190 (long long)FDEV(last_dev).end_blk + blks; 2191 #ifdef CONFIG_BLK_DEV_ZONED 2192 FDEV(last_dev).nr_blkz = FDEV(last_dev).nr_blkz + 2193 div_u64(blks, sbi->blocks_per_blkz); 2194 #endif 2195 } 2196 } 2197 2198 int f2fs_resize_fs(struct file *filp, __u64 block_count) 2199 { 2200 struct f2fs_sb_info *sbi = F2FS_I_SB(file_inode(filp)); 2201 __u64 old_block_count, shrunk_blocks; 2202 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 }; 2203 unsigned int secs; 2204 int err = 0; 2205 __u32 rem; 2206 2207 old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count); 2208 if (block_count > old_block_count) 2209 return -EINVAL; 2210 2211 if (f2fs_is_multi_device(sbi)) { 2212 int last_dev = sbi->s_ndevs - 1; 2213 __u64 last_segs = FDEV(last_dev).total_segments; 2214 2215 if (block_count + SEGS_TO_BLKS(sbi, last_segs) <= 2216 old_block_count) 2217 return -EINVAL; 2218 } 2219 2220 /* new fs size should align to section size */ 2221 div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem); 2222 if (rem) 2223 return -EINVAL; 2224 2225 if (block_count == old_block_count) 2226 return 0; 2227 2228 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 2229 f2fs_err(sbi, "Should run fsck to repair first."); 2230 return -EFSCORRUPTED; 2231 } 2232 2233 if (test_opt(sbi, DISABLE_CHECKPOINT)) { 2234 f2fs_err(sbi, "Checkpoint should be enabled."); 2235 return -EINVAL; 2236 } 2237 2238 err = mnt_want_write_file(filp); 2239 if (err) 2240 return err; 2241 2242 shrunk_blocks = old_block_count - block_count; 2243 secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi)); 2244 2245 /* stop other GC */ 2246 if (!f2fs_down_write_trylock(&sbi->gc_lock)) { 2247 err = -EAGAIN; 2248 goto out_drop_write; 2249 } 2250 2251 /* stop CP to protect MAIN_SEC in free_segment_range */ 2252 f2fs_lock_op(sbi); 2253 2254 spin_lock(&sbi->stat_lock); 2255 if (shrunk_blocks + valid_user_blocks(sbi) + 2256 sbi->current_reserved_blocks + sbi->unusable_block_count + 2257 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count) 2258 err = -ENOSPC; 2259 spin_unlock(&sbi->stat_lock); 2260 2261 if (err) 2262 goto out_unlock; 2263 2264 err = free_segment_range(sbi, secs, true); 2265 2266 out_unlock: 2267 f2fs_unlock_op(sbi); 2268 f2fs_up_write(&sbi->gc_lock); 2269 out_drop_write: 2270 mnt_drop_write_file(filp); 2271 if (err) 2272 return err; 2273 2274 err = freeze_super(sbi->sb, FREEZE_HOLDER_USERSPACE); 2275 if (err) 2276 return err; 2277 2278 if (f2fs_readonly(sbi->sb)) { 2279 err = thaw_super(sbi->sb, FREEZE_HOLDER_USERSPACE); 2280 if (err) 2281 return err; 2282 return -EROFS; 2283 } 2284 2285 f2fs_down_write(&sbi->gc_lock); 2286 f2fs_down_write(&sbi->cp_global_sem); 2287 2288 spin_lock(&sbi->stat_lock); 2289 if (shrunk_blocks + valid_user_blocks(sbi) + 2290 sbi->current_reserved_blocks + sbi->unusable_block_count + 2291 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count) 2292 err = -ENOSPC; 2293 else 2294 sbi->user_block_count -= shrunk_blocks; 2295 spin_unlock(&sbi->stat_lock); 2296 if (err) 2297 goto out_err; 2298 2299 set_sbi_flag(sbi, SBI_IS_RESIZEFS); 2300 err = free_segment_range(sbi, secs, false); 2301 if (err) 2302 goto recover_out; 2303 2304 update_sb_metadata(sbi, -secs); 2305 2306 err = f2fs_commit_super(sbi, false); 2307 if (err) { 2308 update_sb_metadata(sbi, secs); 2309 goto recover_out; 2310 } 2311 2312 update_fs_metadata(sbi, -secs); 2313 clear_sbi_flag(sbi, SBI_IS_RESIZEFS); 2314 set_sbi_flag(sbi, SBI_IS_DIRTY); 2315 2316 stat_inc_cp_call_count(sbi, TOTAL_CALL); 2317 err = f2fs_write_checkpoint(sbi, &cpc); 2318 if (err) { 2319 update_fs_metadata(sbi, secs); 2320 update_sb_metadata(sbi, secs); 2321 f2fs_commit_super(sbi, false); 2322 } 2323 recover_out: 2324 clear_sbi_flag(sbi, SBI_IS_RESIZEFS); 2325 if (err) { 2326 set_sbi_flag(sbi, SBI_NEED_FSCK); 2327 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!"); 2328 2329 spin_lock(&sbi->stat_lock); 2330 sbi->user_block_count += shrunk_blocks; 2331 spin_unlock(&sbi->stat_lock); 2332 } 2333 out_err: 2334 f2fs_up_write(&sbi->cp_global_sem); 2335 f2fs_up_write(&sbi->gc_lock); 2336 thaw_super(sbi->sb, FREEZE_HOLDER_USERSPACE); 2337 return err; 2338 } 2339