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_interruptible_timeout(*wq, 50 kthread_should_stop() || freezing(current) || 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 (try_to_freeze() || 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 /* 85 * [GC triggering condition] 86 * 0. GC is not conducted currently. 87 * 1. There are enough dirty segments. 88 * 2. IO subsystem is idle by checking the # of writeback pages. 89 * 3. IO subsystem is idle by checking the # of requests in 90 * bdev's request list. 91 * 92 * Note) We have to avoid triggering GCs frequently. 93 * Because it is possible that some segments can be 94 * invalidated soon after by user update or deletion. 95 * So, I'd like to wait some time to collect dirty segments. 96 */ 97 if (sbi->gc_mode == GC_URGENT_HIGH || 98 sbi->gc_mode == GC_URGENT_MID) { 99 wait_ms = gc_th->urgent_sleep_time; 100 f2fs_down_write(&sbi->gc_lock); 101 goto do_gc; 102 } 103 104 if (foreground) { 105 f2fs_down_write(&sbi->gc_lock); 106 goto do_gc; 107 } else if (!f2fs_down_write_trylock(&sbi->gc_lock)) { 108 stat_other_skip_bggc_count(sbi); 109 goto next; 110 } 111 112 if (!is_idle(sbi, GC_TIME)) { 113 increase_sleep_time(gc_th, &wait_ms); 114 f2fs_up_write(&sbi->gc_lock); 115 stat_io_skip_bggc_count(sbi); 116 goto next; 117 } 118 119 if (has_enough_invalid_blocks(sbi)) 120 decrease_sleep_time(gc_th, &wait_ms); 121 else 122 increase_sleep_time(gc_th, &wait_ms); 123 do_gc: 124 stat_inc_gc_call_count(sbi, foreground ? 125 FOREGROUND : BACKGROUND); 126 127 sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC; 128 129 /* foreground GC was been triggered via f2fs_balance_fs() */ 130 if (foreground) 131 sync_mode = false; 132 133 gc_control.init_gc_type = sync_mode ? FG_GC : BG_GC; 134 gc_control.no_bg_gc = foreground; 135 gc_control.nr_free_secs = foreground ? 1 : 0; 136 137 /* if return value is not zero, no victim was selected */ 138 if (f2fs_gc(sbi, &gc_control)) { 139 /* don't bother wait_ms by foreground gc */ 140 if (!foreground) 141 wait_ms = gc_th->no_gc_sleep_time; 142 } else { 143 /* reset wait_ms to default sleep time */ 144 if (wait_ms == gc_th->no_gc_sleep_time) 145 wait_ms = gc_th->min_sleep_time; 146 } 147 148 if (foreground) 149 wake_up_all(&gc_th->fggc_wq); 150 151 trace_f2fs_background_gc(sbi->sb, wait_ms, 152 prefree_segments(sbi), free_segments(sbi)); 153 154 /* balancing f2fs's metadata periodically */ 155 f2fs_balance_fs_bg(sbi, true); 156 next: 157 if (sbi->gc_mode != GC_NORMAL) { 158 spin_lock(&sbi->gc_remaining_trials_lock); 159 if (sbi->gc_remaining_trials) { 160 sbi->gc_remaining_trials--; 161 if (!sbi->gc_remaining_trials) 162 sbi->gc_mode = GC_NORMAL; 163 } 164 spin_unlock(&sbi->gc_remaining_trials_lock); 165 } 166 sb_end_write(sbi->sb); 167 168 } while (!kthread_should_stop()); 169 return 0; 170 } 171 172 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi) 173 { 174 struct f2fs_gc_kthread *gc_th; 175 dev_t dev = sbi->sb->s_bdev->bd_dev; 176 177 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL); 178 if (!gc_th) 179 return -ENOMEM; 180 181 gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME; 182 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; 183 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; 184 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; 185 186 gc_th->gc_wake = false; 187 188 sbi->gc_thread = gc_th; 189 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); 190 init_waitqueue_head(&sbi->gc_thread->fggc_wq); 191 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, 192 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); 193 if (IS_ERR(gc_th->f2fs_gc_task)) { 194 int err = PTR_ERR(gc_th->f2fs_gc_task); 195 196 kfree(gc_th); 197 sbi->gc_thread = NULL; 198 return err; 199 } 200 201 return 0; 202 } 203 204 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi) 205 { 206 struct f2fs_gc_kthread *gc_th = sbi->gc_thread; 207 208 if (!gc_th) 209 return; 210 kthread_stop(gc_th->f2fs_gc_task); 211 wake_up_all(&gc_th->fggc_wq); 212 kfree(gc_th); 213 sbi->gc_thread = NULL; 214 } 215 216 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type) 217 { 218 int gc_mode; 219 220 if (gc_type == BG_GC) { 221 if (sbi->am.atgc_enabled) 222 gc_mode = GC_AT; 223 else 224 gc_mode = GC_CB; 225 } else { 226 gc_mode = GC_GREEDY; 227 } 228 229 switch (sbi->gc_mode) { 230 case GC_IDLE_CB: 231 case GC_URGENT_LOW: 232 case GC_URGENT_MID: 233 gc_mode = GC_CB; 234 break; 235 case GC_IDLE_GREEDY: 236 case GC_URGENT_HIGH: 237 gc_mode = GC_GREEDY; 238 break; 239 case GC_IDLE_AT: 240 gc_mode = GC_AT; 241 break; 242 } 243 244 return gc_mode; 245 } 246 247 static void select_policy(struct f2fs_sb_info *sbi, int gc_type, 248 int type, struct victim_sel_policy *p) 249 { 250 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 251 252 if (p->alloc_mode == SSR) { 253 p->gc_mode = GC_GREEDY; 254 p->dirty_bitmap = dirty_i->dirty_segmap[type]; 255 p->max_search = dirty_i->nr_dirty[type]; 256 p->ofs_unit = 1; 257 } else if (p->alloc_mode == AT_SSR) { 258 p->gc_mode = GC_GREEDY; 259 p->dirty_bitmap = dirty_i->dirty_segmap[type]; 260 p->max_search = dirty_i->nr_dirty[type]; 261 p->ofs_unit = 1; 262 } else { 263 p->gc_mode = select_gc_type(sbi, gc_type); 264 p->ofs_unit = SEGS_PER_SEC(sbi); 265 if (__is_large_section(sbi)) { 266 p->dirty_bitmap = dirty_i->dirty_secmap; 267 p->max_search = count_bits(p->dirty_bitmap, 268 0, MAIN_SECS(sbi)); 269 } else { 270 p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY]; 271 p->max_search = dirty_i->nr_dirty[DIRTY]; 272 } 273 } 274 275 /* 276 * adjust candidates range, should select all dirty segments for 277 * foreground GC and urgent GC cases. 278 */ 279 if (gc_type != FG_GC && 280 (sbi->gc_mode != GC_URGENT_HIGH) && 281 (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) && 282 p->max_search > sbi->max_victim_search) 283 p->max_search = sbi->max_victim_search; 284 285 /* let's select beginning hot/small space first. */ 286 if (f2fs_need_rand_seg(sbi)) 287 p->offset = get_random_u32_below(MAIN_SECS(sbi) * 288 SEGS_PER_SEC(sbi)); 289 else if (type == CURSEG_HOT_DATA || IS_NODESEG(type)) 290 p->offset = 0; 291 else 292 p->offset = SIT_I(sbi)->last_victim[p->gc_mode]; 293 } 294 295 static unsigned int get_max_cost(struct f2fs_sb_info *sbi, 296 struct victim_sel_policy *p) 297 { 298 /* SSR allocates in a segment unit */ 299 if (p->alloc_mode == SSR) 300 return BLKS_PER_SEG(sbi); 301 else if (p->alloc_mode == AT_SSR) 302 return UINT_MAX; 303 304 /* LFS */ 305 if (p->gc_mode == GC_GREEDY) 306 return 2 * BLKS_PER_SEG(sbi) * p->ofs_unit; 307 else if (p->gc_mode == GC_CB) 308 return UINT_MAX; 309 else if (p->gc_mode == GC_AT) 310 return UINT_MAX; 311 else /* No other gc_mode */ 312 return 0; 313 } 314 315 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) 316 { 317 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 318 unsigned int secno; 319 320 /* 321 * If the gc_type is FG_GC, we can select victim segments 322 * selected by background GC before. 323 * Those segments guarantee they have small valid blocks. 324 */ 325 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { 326 if (sec_usage_check(sbi, secno)) 327 continue; 328 clear_bit(secno, dirty_i->victim_secmap); 329 return GET_SEG_FROM_SEC(sbi, secno); 330 } 331 return NULL_SEGNO; 332 } 333 334 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) 335 { 336 struct sit_info *sit_i = SIT_I(sbi); 337 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 338 unsigned int start = GET_SEG_FROM_SEC(sbi, secno); 339 unsigned long long mtime = 0; 340 unsigned int vblocks; 341 unsigned char age = 0; 342 unsigned char u; 343 unsigned int i; 344 unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno); 345 346 for (i = 0; i < usable_segs_per_sec; i++) 347 mtime += get_seg_entry(sbi, start + i)->mtime; 348 vblocks = get_valid_blocks(sbi, segno, true); 349 350 mtime = div_u64(mtime, usable_segs_per_sec); 351 vblocks = div_u64(vblocks, usable_segs_per_sec); 352 353 u = (vblocks * 100) >> sbi->log_blocks_per_seg; 354 355 /* Handle if the system time has changed by the user */ 356 if (mtime < sit_i->min_mtime) 357 sit_i->min_mtime = mtime; 358 if (mtime > sit_i->max_mtime) 359 sit_i->max_mtime = mtime; 360 if (sit_i->max_mtime != sit_i->min_mtime) 361 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), 362 sit_i->max_mtime - sit_i->min_mtime); 363 364 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); 365 } 366 367 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, 368 unsigned int segno, struct victim_sel_policy *p) 369 { 370 if (p->alloc_mode == SSR) 371 return get_seg_entry(sbi, segno)->ckpt_valid_blocks; 372 373 /* alloc_mode == LFS */ 374 if (p->gc_mode == GC_GREEDY) 375 return get_valid_blocks(sbi, segno, true); 376 else if (p->gc_mode == GC_CB) 377 return get_cb_cost(sbi, segno); 378 379 f2fs_bug_on(sbi, 1); 380 return 0; 381 } 382 383 static unsigned int count_bits(const unsigned long *addr, 384 unsigned int offset, unsigned int len) 385 { 386 unsigned int end = offset + len, sum = 0; 387 388 while (offset < end) { 389 if (test_bit(offset++, addr)) 390 ++sum; 391 } 392 return sum; 393 } 394 395 static bool f2fs_check_victim_tree(struct f2fs_sb_info *sbi, 396 struct rb_root_cached *root) 397 { 398 #ifdef CONFIG_F2FS_CHECK_FS 399 struct rb_node *cur = rb_first_cached(root), *next; 400 struct victim_entry *cur_ve, *next_ve; 401 402 while (cur) { 403 next = rb_next(cur); 404 if (!next) 405 return true; 406 407 cur_ve = rb_entry(cur, struct victim_entry, rb_node); 408 next_ve = rb_entry(next, struct victim_entry, rb_node); 409 410 if (cur_ve->mtime > next_ve->mtime) { 411 f2fs_info(sbi, "broken victim_rbtree, " 412 "cur_mtime(%llu) next_mtime(%llu)", 413 cur_ve->mtime, next_ve->mtime); 414 return false; 415 } 416 cur = next; 417 } 418 #endif 419 return true; 420 } 421 422 static struct victim_entry *__lookup_victim_entry(struct f2fs_sb_info *sbi, 423 unsigned long long mtime) 424 { 425 struct atgc_management *am = &sbi->am; 426 struct rb_node *node = am->root.rb_root.rb_node; 427 struct victim_entry *ve = NULL; 428 429 while (node) { 430 ve = rb_entry(node, struct victim_entry, rb_node); 431 432 if (mtime < ve->mtime) 433 node = node->rb_left; 434 else 435 node = node->rb_right; 436 } 437 return ve; 438 } 439 440 static struct victim_entry *__create_victim_entry(struct f2fs_sb_info *sbi, 441 unsigned long long mtime, unsigned int segno) 442 { 443 struct atgc_management *am = &sbi->am; 444 struct victim_entry *ve; 445 446 ve = f2fs_kmem_cache_alloc(victim_entry_slab, GFP_NOFS, true, NULL); 447 448 ve->mtime = mtime; 449 ve->segno = segno; 450 451 list_add_tail(&ve->list, &am->victim_list); 452 am->victim_count++; 453 454 return ve; 455 } 456 457 static void __insert_victim_entry(struct f2fs_sb_info *sbi, 458 unsigned long long mtime, unsigned int segno) 459 { 460 struct atgc_management *am = &sbi->am; 461 struct rb_root_cached *root = &am->root; 462 struct rb_node **p = &root->rb_root.rb_node; 463 struct rb_node *parent = NULL; 464 struct victim_entry *ve; 465 bool left_most = true; 466 467 /* look up rb tree to find parent node */ 468 while (*p) { 469 parent = *p; 470 ve = rb_entry(parent, struct victim_entry, rb_node); 471 472 if (mtime < ve->mtime) { 473 p = &(*p)->rb_left; 474 } else { 475 p = &(*p)->rb_right; 476 left_most = false; 477 } 478 } 479 480 ve = __create_victim_entry(sbi, mtime, segno); 481 482 rb_link_node(&ve->rb_node, parent, p); 483 rb_insert_color_cached(&ve->rb_node, root, left_most); 484 } 485 486 static void add_victim_entry(struct f2fs_sb_info *sbi, 487 struct victim_sel_policy *p, unsigned int segno) 488 { 489 struct sit_info *sit_i = SIT_I(sbi); 490 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 491 unsigned int start = GET_SEG_FROM_SEC(sbi, secno); 492 unsigned long long mtime = 0; 493 unsigned int i; 494 495 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 496 if (p->gc_mode == GC_AT && 497 get_valid_blocks(sbi, segno, true) == 0) 498 return; 499 } 500 501 for (i = 0; i < SEGS_PER_SEC(sbi); i++) 502 mtime += get_seg_entry(sbi, start + i)->mtime; 503 mtime = div_u64(mtime, SEGS_PER_SEC(sbi)); 504 505 /* Handle if the system time has changed by the user */ 506 if (mtime < sit_i->min_mtime) 507 sit_i->min_mtime = mtime; 508 if (mtime > sit_i->max_mtime) 509 sit_i->max_mtime = mtime; 510 if (mtime < sit_i->dirty_min_mtime) 511 sit_i->dirty_min_mtime = mtime; 512 if (mtime > sit_i->dirty_max_mtime) 513 sit_i->dirty_max_mtime = mtime; 514 515 /* don't choose young section as candidate */ 516 if (sit_i->dirty_max_mtime - mtime < p->age_threshold) 517 return; 518 519 __insert_victim_entry(sbi, mtime, segno); 520 } 521 522 static void atgc_lookup_victim(struct f2fs_sb_info *sbi, 523 struct victim_sel_policy *p) 524 { 525 struct sit_info *sit_i = SIT_I(sbi); 526 struct atgc_management *am = &sbi->am; 527 struct rb_root_cached *root = &am->root; 528 struct rb_node *node; 529 struct victim_entry *ve; 530 unsigned long long total_time; 531 unsigned long long age, u, accu; 532 unsigned long long max_mtime = sit_i->dirty_max_mtime; 533 unsigned long long min_mtime = sit_i->dirty_min_mtime; 534 unsigned int sec_blocks = CAP_BLKS_PER_SEC(sbi); 535 unsigned int vblocks; 536 unsigned int dirty_threshold = max(am->max_candidate_count, 537 am->candidate_ratio * 538 am->victim_count / 100); 539 unsigned int age_weight = am->age_weight; 540 unsigned int cost; 541 unsigned int iter = 0; 542 543 if (max_mtime < min_mtime) 544 return; 545 546 max_mtime += 1; 547 total_time = max_mtime - min_mtime; 548 549 accu = div64_u64(ULLONG_MAX, total_time); 550 accu = min_t(unsigned long long, div_u64(accu, 100), 551 DEFAULT_ACCURACY_CLASS); 552 553 node = rb_first_cached(root); 554 next: 555 ve = rb_entry_safe(node, struct victim_entry, rb_node); 556 if (!ve) 557 return; 558 559 if (ve->mtime >= max_mtime || ve->mtime < min_mtime) 560 goto skip; 561 562 /* age = 10000 * x% * 60 */ 563 age = div64_u64(accu * (max_mtime - ve->mtime), total_time) * 564 age_weight; 565 566 vblocks = get_valid_blocks(sbi, ve->segno, true); 567 f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks); 568 569 /* u = 10000 * x% * 40 */ 570 u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) * 571 (100 - age_weight); 572 573 f2fs_bug_on(sbi, age + u >= UINT_MAX); 574 575 cost = UINT_MAX - (age + u); 576 iter++; 577 578 if (cost < p->min_cost || 579 (cost == p->min_cost && age > p->oldest_age)) { 580 p->min_cost = cost; 581 p->oldest_age = age; 582 p->min_segno = ve->segno; 583 } 584 skip: 585 if (iter < dirty_threshold) { 586 node = rb_next(node); 587 goto next; 588 } 589 } 590 591 /* 592 * select candidates around source section in range of 593 * [target - dirty_threshold, target + dirty_threshold] 594 */ 595 static void atssr_lookup_victim(struct f2fs_sb_info *sbi, 596 struct victim_sel_policy *p) 597 { 598 struct sit_info *sit_i = SIT_I(sbi); 599 struct atgc_management *am = &sbi->am; 600 struct victim_entry *ve; 601 unsigned long long age; 602 unsigned long long max_mtime = sit_i->dirty_max_mtime; 603 unsigned long long min_mtime = sit_i->dirty_min_mtime; 604 unsigned int vblocks; 605 unsigned int dirty_threshold = max(am->max_candidate_count, 606 am->candidate_ratio * 607 am->victim_count / 100); 608 unsigned int cost, iter; 609 int stage = 0; 610 611 if (max_mtime < min_mtime) 612 return; 613 max_mtime += 1; 614 next_stage: 615 iter = 0; 616 ve = __lookup_victim_entry(sbi, p->age); 617 next_node: 618 if (!ve) { 619 if (stage++ == 0) 620 goto next_stage; 621 return; 622 } 623 624 if (ve->mtime >= max_mtime || ve->mtime < min_mtime) 625 goto skip_node; 626 627 age = max_mtime - ve->mtime; 628 629 vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks; 630 f2fs_bug_on(sbi, !vblocks); 631 632 /* rare case */ 633 if (vblocks == BLKS_PER_SEG(sbi)) 634 goto skip_node; 635 636 iter++; 637 638 age = max_mtime - abs(p->age - age); 639 cost = UINT_MAX - vblocks; 640 641 if (cost < p->min_cost || 642 (cost == p->min_cost && age > p->oldest_age)) { 643 p->min_cost = cost; 644 p->oldest_age = age; 645 p->min_segno = ve->segno; 646 } 647 skip_node: 648 if (iter < dirty_threshold) { 649 ve = rb_entry(stage == 0 ? rb_prev(&ve->rb_node) : 650 rb_next(&ve->rb_node), 651 struct victim_entry, rb_node); 652 goto next_node; 653 } 654 655 if (stage++ == 0) 656 goto next_stage; 657 } 658 659 static void lookup_victim_by_age(struct f2fs_sb_info *sbi, 660 struct victim_sel_policy *p) 661 { 662 f2fs_bug_on(sbi, !f2fs_check_victim_tree(sbi, &sbi->am.root)); 663 664 if (p->gc_mode == GC_AT) 665 atgc_lookup_victim(sbi, p); 666 else if (p->alloc_mode == AT_SSR) 667 atssr_lookup_victim(sbi, p); 668 else 669 f2fs_bug_on(sbi, 1); 670 } 671 672 static void release_victim_entry(struct f2fs_sb_info *sbi) 673 { 674 struct atgc_management *am = &sbi->am; 675 struct victim_entry *ve, *tmp; 676 677 list_for_each_entry_safe(ve, tmp, &am->victim_list, list) { 678 list_del(&ve->list); 679 kmem_cache_free(victim_entry_slab, ve); 680 am->victim_count--; 681 } 682 683 am->root = RB_ROOT_CACHED; 684 685 f2fs_bug_on(sbi, am->victim_count); 686 f2fs_bug_on(sbi, !list_empty(&am->victim_list)); 687 } 688 689 static bool f2fs_pin_section(struct f2fs_sb_info *sbi, unsigned int segno) 690 { 691 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 692 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 693 694 if (!dirty_i->enable_pin_section) 695 return false; 696 if (!test_and_set_bit(secno, dirty_i->pinned_secmap)) 697 dirty_i->pinned_secmap_cnt++; 698 return true; 699 } 700 701 static bool f2fs_pinned_section_exists(struct dirty_seglist_info *dirty_i) 702 { 703 return dirty_i->pinned_secmap_cnt; 704 } 705 706 static bool f2fs_section_is_pinned(struct dirty_seglist_info *dirty_i, 707 unsigned int secno) 708 { 709 return dirty_i->enable_pin_section && 710 f2fs_pinned_section_exists(dirty_i) && 711 test_bit(secno, dirty_i->pinned_secmap); 712 } 713 714 static void f2fs_unpin_all_sections(struct f2fs_sb_info *sbi, bool enable) 715 { 716 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi)); 717 718 if (f2fs_pinned_section_exists(DIRTY_I(sbi))) { 719 memset(DIRTY_I(sbi)->pinned_secmap, 0, bitmap_size); 720 DIRTY_I(sbi)->pinned_secmap_cnt = 0; 721 } 722 DIRTY_I(sbi)->enable_pin_section = enable; 723 } 724 725 static int f2fs_gc_pinned_control(struct inode *inode, int gc_type, 726 unsigned int segno) 727 { 728 if (!f2fs_is_pinned_file(inode)) 729 return 0; 730 if (gc_type != FG_GC) 731 return -EBUSY; 732 if (!f2fs_pin_section(F2FS_I_SB(inode), segno)) 733 f2fs_pin_file_control(inode, true); 734 return -EAGAIN; 735 } 736 737 /* 738 * This function is called from two paths. 739 * One is garbage collection and the other is SSR segment selection. 740 * When it is called during GC, it just gets a victim segment 741 * and it does not remove it from dirty seglist. 742 * When it is called from SSR segment selection, it finds a segment 743 * which has minimum valid blocks and removes it from dirty seglist. 744 */ 745 int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result, 746 int gc_type, int type, char alloc_mode, 747 unsigned long long age) 748 { 749 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 750 struct sit_info *sm = SIT_I(sbi); 751 struct victim_sel_policy p; 752 unsigned int secno, last_victim; 753 unsigned int last_segment; 754 unsigned int nsearched; 755 bool is_atgc; 756 int ret = 0; 757 758 mutex_lock(&dirty_i->seglist_lock); 759 last_segment = MAIN_SECS(sbi) * SEGS_PER_SEC(sbi); 760 761 p.alloc_mode = alloc_mode; 762 p.age = age; 763 p.age_threshold = sbi->am.age_threshold; 764 765 retry: 766 select_policy(sbi, gc_type, type, &p); 767 p.min_segno = NULL_SEGNO; 768 p.oldest_age = 0; 769 p.min_cost = get_max_cost(sbi, &p); 770 771 is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR); 772 nsearched = 0; 773 774 if (is_atgc) 775 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX; 776 777 if (*result != NULL_SEGNO) { 778 if (!get_valid_blocks(sbi, *result, false)) { 779 ret = -ENODATA; 780 goto out; 781 } 782 783 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result))) 784 ret = -EBUSY; 785 else 786 p.min_segno = *result; 787 goto out; 788 } 789 790 ret = -ENODATA; 791 if (p.max_search == 0) 792 goto out; 793 794 if (__is_large_section(sbi) && p.alloc_mode == LFS) { 795 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) { 796 p.min_segno = sbi->next_victim_seg[BG_GC]; 797 *result = p.min_segno; 798 sbi->next_victim_seg[BG_GC] = NULL_SEGNO; 799 goto got_result; 800 } 801 if (gc_type == FG_GC && 802 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) { 803 p.min_segno = sbi->next_victim_seg[FG_GC]; 804 *result = p.min_segno; 805 sbi->next_victim_seg[FG_GC] = NULL_SEGNO; 806 goto got_result; 807 } 808 } 809 810 last_victim = sm->last_victim[p.gc_mode]; 811 if (p.alloc_mode == LFS && gc_type == FG_GC) { 812 p.min_segno = check_bg_victims(sbi); 813 if (p.min_segno != NULL_SEGNO) 814 goto got_it; 815 } 816 817 while (1) { 818 unsigned long cost, *dirty_bitmap; 819 unsigned int unit_no, segno; 820 821 dirty_bitmap = p.dirty_bitmap; 822 unit_no = find_next_bit(dirty_bitmap, 823 last_segment / p.ofs_unit, 824 p.offset / p.ofs_unit); 825 segno = unit_no * p.ofs_unit; 826 if (segno >= last_segment) { 827 if (sm->last_victim[p.gc_mode]) { 828 last_segment = 829 sm->last_victim[p.gc_mode]; 830 sm->last_victim[p.gc_mode] = 0; 831 p.offset = 0; 832 continue; 833 } 834 break; 835 } 836 837 p.offset = segno + p.ofs_unit; 838 nsearched++; 839 840 #ifdef CONFIG_F2FS_CHECK_FS 841 /* 842 * skip selecting the invalid segno (that is failed due to block 843 * validity check failure during GC) to avoid endless GC loop in 844 * such cases. 845 */ 846 if (test_bit(segno, sm->invalid_segmap)) 847 goto next; 848 #endif 849 850 secno = GET_SEC_FROM_SEG(sbi, segno); 851 852 if (sec_usage_check(sbi, secno)) 853 goto next; 854 855 /* Don't touch checkpointed data */ 856 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 857 if (p.alloc_mode == LFS) { 858 /* 859 * LFS is set to find source section during GC. 860 * The victim should have no checkpointed data. 861 */ 862 if (get_ckpt_valid_blocks(sbi, segno, true)) 863 goto next; 864 } else { 865 /* 866 * SSR | AT_SSR are set to find target segment 867 * for writes which can be full by checkpointed 868 * and newly written blocks. 869 */ 870 if (!f2fs_segment_has_free_slot(sbi, segno)) 871 goto next; 872 } 873 } 874 875 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) 876 goto next; 877 878 if (gc_type == FG_GC && f2fs_section_is_pinned(dirty_i, secno)) 879 goto next; 880 881 if (is_atgc) { 882 add_victim_entry(sbi, &p, segno); 883 goto next; 884 } 885 886 cost = get_gc_cost(sbi, segno, &p); 887 888 if (p.min_cost > cost) { 889 p.min_segno = segno; 890 p.min_cost = cost; 891 } 892 next: 893 if (nsearched >= p.max_search) { 894 if (!sm->last_victim[p.gc_mode] && segno <= last_victim) 895 sm->last_victim[p.gc_mode] = 896 last_victim + p.ofs_unit; 897 else 898 sm->last_victim[p.gc_mode] = segno + p.ofs_unit; 899 sm->last_victim[p.gc_mode] %= 900 (MAIN_SECS(sbi) * SEGS_PER_SEC(sbi)); 901 break; 902 } 903 } 904 905 /* get victim for GC_AT/AT_SSR */ 906 if (is_atgc) { 907 lookup_victim_by_age(sbi, &p); 908 release_victim_entry(sbi); 909 } 910 911 if (is_atgc && p.min_segno == NULL_SEGNO && 912 sm->elapsed_time < p.age_threshold) { 913 p.age_threshold = 0; 914 goto retry; 915 } 916 917 if (p.min_segno != NULL_SEGNO) { 918 got_it: 919 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; 920 got_result: 921 if (p.alloc_mode == LFS) { 922 secno = GET_SEC_FROM_SEG(sbi, p.min_segno); 923 if (gc_type == FG_GC) 924 sbi->cur_victim_sec = secno; 925 else 926 set_bit(secno, dirty_i->victim_secmap); 927 } 928 ret = 0; 929 930 } 931 out: 932 if (p.min_segno != NULL_SEGNO) 933 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, 934 sbi->cur_victim_sec, 935 prefree_segments(sbi), free_segments(sbi)); 936 mutex_unlock(&dirty_i->seglist_lock); 937 938 return ret; 939 } 940 941 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) 942 { 943 struct inode_entry *ie; 944 945 ie = radix_tree_lookup(&gc_list->iroot, ino); 946 if (ie) 947 return ie->inode; 948 return NULL; 949 } 950 951 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) 952 { 953 struct inode_entry *new_ie; 954 955 if (inode == find_gc_inode(gc_list, inode->i_ino)) { 956 iput(inode); 957 return; 958 } 959 new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, 960 GFP_NOFS, true, NULL); 961 new_ie->inode = inode; 962 963 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); 964 list_add_tail(&new_ie->list, &gc_list->ilist); 965 } 966 967 static void put_gc_inode(struct gc_inode_list *gc_list) 968 { 969 struct inode_entry *ie, *next_ie; 970 971 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { 972 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); 973 iput(ie->inode); 974 list_del(&ie->list); 975 kmem_cache_free(f2fs_inode_entry_slab, ie); 976 } 977 } 978 979 static int check_valid_map(struct f2fs_sb_info *sbi, 980 unsigned int segno, int offset) 981 { 982 struct sit_info *sit_i = SIT_I(sbi); 983 struct seg_entry *sentry; 984 int ret; 985 986 down_read(&sit_i->sentry_lock); 987 sentry = get_seg_entry(sbi, segno); 988 ret = f2fs_test_bit(offset, sentry->cur_valid_map); 989 up_read(&sit_i->sentry_lock); 990 return ret; 991 } 992 993 /* 994 * This function compares node address got in summary with that in NAT. 995 * On validity, copy that node with cold status, otherwise (invalid node) 996 * ignore that. 997 */ 998 static int gc_node_segment(struct f2fs_sb_info *sbi, 999 struct f2fs_summary *sum, unsigned int segno, int gc_type) 1000 { 1001 struct f2fs_summary *entry; 1002 block_t start_addr; 1003 int off; 1004 int phase = 0; 1005 bool fggc = (gc_type == FG_GC); 1006 int submitted = 0; 1007 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 1008 1009 start_addr = START_BLOCK(sbi, segno); 1010 1011 next_step: 1012 entry = sum; 1013 1014 if (fggc && phase == 2) 1015 atomic_inc(&sbi->wb_sync_req[NODE]); 1016 1017 for (off = 0; off < usable_blks_in_seg; off++, entry++) { 1018 nid_t nid = le32_to_cpu(entry->nid); 1019 struct page *node_page; 1020 struct node_info ni; 1021 int err; 1022 1023 /* stop BG_GC if there is not enough free sections. */ 1024 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) 1025 return submitted; 1026 1027 if (check_valid_map(sbi, segno, off) == 0) 1028 continue; 1029 1030 if (phase == 0) { 1031 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, 1032 META_NAT, true); 1033 continue; 1034 } 1035 1036 if (phase == 1) { 1037 f2fs_ra_node_page(sbi, nid); 1038 continue; 1039 } 1040 1041 /* phase == 2 */ 1042 node_page = f2fs_get_node_page(sbi, nid); 1043 if (IS_ERR(node_page)) 1044 continue; 1045 1046 /* block may become invalid during f2fs_get_node_page */ 1047 if (check_valid_map(sbi, segno, off) == 0) { 1048 f2fs_put_page(node_page, 1); 1049 continue; 1050 } 1051 1052 if (f2fs_get_node_info(sbi, nid, &ni, false)) { 1053 f2fs_put_page(node_page, 1); 1054 continue; 1055 } 1056 1057 if (ni.blk_addr != start_addr + off) { 1058 f2fs_put_page(node_page, 1); 1059 continue; 1060 } 1061 1062 err = f2fs_move_node_page(node_page, gc_type); 1063 if (!err && gc_type == FG_GC) 1064 submitted++; 1065 stat_inc_node_blk_count(sbi, 1, gc_type); 1066 } 1067 1068 if (++phase < 3) 1069 goto next_step; 1070 1071 if (fggc) 1072 atomic_dec(&sbi->wb_sync_req[NODE]); 1073 return submitted; 1074 } 1075 1076 /* 1077 * Calculate start block index indicating the given node offset. 1078 * Be careful, caller should give this node offset only indicating direct node 1079 * blocks. If any node offsets, which point the other types of node blocks such 1080 * as indirect or double indirect node blocks, are given, it must be a caller's 1081 * bug. 1082 */ 1083 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode) 1084 { 1085 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; 1086 unsigned int bidx; 1087 1088 if (node_ofs == 0) 1089 return 0; 1090 1091 if (node_ofs <= 2) { 1092 bidx = node_ofs - 1; 1093 } else if (node_ofs <= indirect_blks) { 1094 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); 1095 1096 bidx = node_ofs - 2 - dec; 1097 } else { 1098 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); 1099 1100 bidx = node_ofs - 5 - dec; 1101 } 1102 return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode); 1103 } 1104 1105 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 1106 struct node_info *dni, block_t blkaddr, unsigned int *nofs) 1107 { 1108 struct page *node_page; 1109 nid_t nid; 1110 unsigned int ofs_in_node, max_addrs, base; 1111 block_t source_blkaddr; 1112 1113 nid = le32_to_cpu(sum->nid); 1114 ofs_in_node = le16_to_cpu(sum->ofs_in_node); 1115 1116 node_page = f2fs_get_node_page(sbi, nid); 1117 if (IS_ERR(node_page)) 1118 return false; 1119 1120 if (f2fs_get_node_info(sbi, nid, dni, false)) { 1121 f2fs_put_page(node_page, 1); 1122 return false; 1123 } 1124 1125 if (sum->version != dni->version) { 1126 f2fs_warn(sbi, "%s: valid data with mismatched node version.", 1127 __func__); 1128 set_sbi_flag(sbi, SBI_NEED_FSCK); 1129 } 1130 1131 if (f2fs_check_nid_range(sbi, dni->ino)) { 1132 f2fs_put_page(node_page, 1); 1133 return false; 1134 } 1135 1136 if (IS_INODE(node_page)) { 1137 base = offset_in_addr(F2FS_INODE(node_page)); 1138 max_addrs = DEF_ADDRS_PER_INODE; 1139 } else { 1140 base = 0; 1141 max_addrs = DEF_ADDRS_PER_BLOCK; 1142 } 1143 1144 if (base + ofs_in_node >= max_addrs) { 1145 f2fs_err(sbi, "Inconsistent blkaddr offset: base:%u, ofs_in_node:%u, max:%u, ino:%u, nid:%u", 1146 base, ofs_in_node, max_addrs, dni->ino, dni->nid); 1147 f2fs_put_page(node_page, 1); 1148 return false; 1149 } 1150 1151 *nofs = ofs_of_node(node_page); 1152 source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node); 1153 f2fs_put_page(node_page, 1); 1154 1155 if (source_blkaddr != blkaddr) { 1156 #ifdef CONFIG_F2FS_CHECK_FS 1157 unsigned int segno = GET_SEGNO(sbi, blkaddr); 1158 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 1159 1160 if (unlikely(check_valid_map(sbi, segno, offset))) { 1161 if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) { 1162 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u", 1163 blkaddr, source_blkaddr, segno); 1164 set_sbi_flag(sbi, SBI_NEED_FSCK); 1165 } 1166 } 1167 #endif 1168 return false; 1169 } 1170 return true; 1171 } 1172 1173 static int ra_data_block(struct inode *inode, pgoff_t index) 1174 { 1175 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1176 struct address_space *mapping = f2fs_is_cow_file(inode) ? 1177 F2FS_I(inode)->atomic_inode->i_mapping : inode->i_mapping; 1178 struct dnode_of_data dn; 1179 struct page *page; 1180 struct f2fs_io_info fio = { 1181 .sbi = sbi, 1182 .ino = inode->i_ino, 1183 .type = DATA, 1184 .temp = COLD, 1185 .op = REQ_OP_READ, 1186 .op_flags = 0, 1187 .encrypted_page = NULL, 1188 .in_list = 0, 1189 }; 1190 int err; 1191 1192 page = f2fs_grab_cache_page(mapping, index, true); 1193 if (!page) 1194 return -ENOMEM; 1195 1196 if (f2fs_lookup_read_extent_cache_block(inode, index, 1197 &dn.data_blkaddr)) { 1198 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, 1199 DATA_GENERIC_ENHANCE_READ))) { 1200 err = -EFSCORRUPTED; 1201 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); 1202 goto put_page; 1203 } 1204 goto got_it; 1205 } 1206 1207 set_new_dnode(&dn, inode, NULL, NULL, 0); 1208 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 1209 if (err) 1210 goto put_page; 1211 f2fs_put_dnode(&dn); 1212 1213 if (!__is_valid_data_blkaddr(dn.data_blkaddr)) { 1214 err = -ENOENT; 1215 goto put_page; 1216 } 1217 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, 1218 DATA_GENERIC_ENHANCE))) { 1219 err = -EFSCORRUPTED; 1220 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR); 1221 goto put_page; 1222 } 1223 got_it: 1224 /* read page */ 1225 fio.page = page; 1226 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; 1227 1228 /* 1229 * don't cache encrypted data into meta inode until previous dirty 1230 * data were writebacked to avoid racing between GC and flush. 1231 */ 1232 f2fs_wait_on_page_writeback(page, DATA, true, true); 1233 1234 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); 1235 1236 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi), 1237 dn.data_blkaddr, 1238 FGP_LOCK | FGP_CREAT, GFP_NOFS); 1239 if (!fio.encrypted_page) { 1240 err = -ENOMEM; 1241 goto put_page; 1242 } 1243 1244 err = f2fs_submit_page_bio(&fio); 1245 if (err) 1246 goto put_encrypted_page; 1247 f2fs_put_page(fio.encrypted_page, 0); 1248 f2fs_put_page(page, 1); 1249 1250 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE); 1251 f2fs_update_iostat(sbi, NULL, FS_GDATA_READ_IO, F2FS_BLKSIZE); 1252 1253 return 0; 1254 put_encrypted_page: 1255 f2fs_put_page(fio.encrypted_page, 1); 1256 put_page: 1257 f2fs_put_page(page, 1); 1258 return err; 1259 } 1260 1261 /* 1262 * Move data block via META_MAPPING while keeping locked data page. 1263 * This can be used to move blocks, aka LBAs, directly on disk. 1264 */ 1265 static int move_data_block(struct inode *inode, block_t bidx, 1266 int gc_type, unsigned int segno, int off) 1267 { 1268 struct address_space *mapping = f2fs_is_cow_file(inode) ? 1269 F2FS_I(inode)->atomic_inode->i_mapping : inode->i_mapping; 1270 struct f2fs_io_info fio = { 1271 .sbi = F2FS_I_SB(inode), 1272 .ino = inode->i_ino, 1273 .type = DATA, 1274 .temp = COLD, 1275 .op = REQ_OP_READ, 1276 .op_flags = 0, 1277 .encrypted_page = NULL, 1278 .in_list = 0, 1279 }; 1280 struct dnode_of_data dn; 1281 struct f2fs_summary sum; 1282 struct node_info ni; 1283 struct page *page, *mpage; 1284 block_t newaddr; 1285 int err = 0; 1286 bool lfs_mode = f2fs_lfs_mode(fio.sbi); 1287 int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) && 1288 (fio.sbi->gc_mode != GC_URGENT_HIGH) ? 1289 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA; 1290 1291 /* do not read out */ 1292 page = f2fs_grab_cache_page(mapping, bidx, false); 1293 if (!page) 1294 return -ENOMEM; 1295 1296 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) { 1297 err = -ENOENT; 1298 goto out; 1299 } 1300 1301 err = f2fs_gc_pinned_control(inode, gc_type, segno); 1302 if (err) 1303 goto out; 1304 1305 set_new_dnode(&dn, inode, NULL, NULL, 0); 1306 err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE); 1307 if (err) 1308 goto out; 1309 1310 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 1311 ClearPageUptodate(page); 1312 err = -ENOENT; 1313 goto put_out; 1314 } 1315 1316 /* 1317 * don't cache encrypted data into meta inode until previous dirty 1318 * data were writebacked to avoid racing between GC and flush. 1319 */ 1320 f2fs_wait_on_page_writeback(page, DATA, true, true); 1321 1322 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); 1323 1324 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false); 1325 if (err) 1326 goto put_out; 1327 1328 /* read page */ 1329 fio.page = page; 1330 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; 1331 1332 if (lfs_mode) 1333 f2fs_down_write(&fio.sbi->io_order_lock); 1334 1335 mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi), 1336 fio.old_blkaddr, false); 1337 if (!mpage) { 1338 err = -ENOMEM; 1339 goto up_out; 1340 } 1341 1342 fio.encrypted_page = mpage; 1343 1344 /* read source block in mpage */ 1345 if (!PageUptodate(mpage)) { 1346 err = f2fs_submit_page_bio(&fio); 1347 if (err) { 1348 f2fs_put_page(mpage, 1); 1349 goto up_out; 1350 } 1351 1352 f2fs_update_iostat(fio.sbi, inode, FS_DATA_READ_IO, 1353 F2FS_BLKSIZE); 1354 f2fs_update_iostat(fio.sbi, NULL, FS_GDATA_READ_IO, 1355 F2FS_BLKSIZE); 1356 1357 lock_page(mpage); 1358 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) || 1359 !PageUptodate(mpage))) { 1360 err = -EIO; 1361 f2fs_put_page(mpage, 1); 1362 goto up_out; 1363 } 1364 } 1365 1366 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); 1367 1368 /* allocate block address */ 1369 f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr, 1370 &sum, type, NULL); 1371 1372 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi), 1373 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS); 1374 if (!fio.encrypted_page) { 1375 err = -ENOMEM; 1376 f2fs_put_page(mpage, 1); 1377 goto recover_block; 1378 } 1379 1380 /* write target block */ 1381 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true); 1382 memcpy(page_address(fio.encrypted_page), 1383 page_address(mpage), PAGE_SIZE); 1384 f2fs_put_page(mpage, 1); 1385 1386 f2fs_invalidate_internal_cache(fio.sbi, fio.old_blkaddr); 1387 1388 set_page_dirty(fio.encrypted_page); 1389 if (clear_page_dirty_for_io(fio.encrypted_page)) 1390 dec_page_count(fio.sbi, F2FS_DIRTY_META); 1391 1392 set_page_writeback(fio.encrypted_page); 1393 1394 fio.op = REQ_OP_WRITE; 1395 fio.op_flags = REQ_SYNC; 1396 fio.new_blkaddr = newaddr; 1397 f2fs_submit_page_write(&fio); 1398 1399 f2fs_update_iostat(fio.sbi, NULL, FS_GC_DATA_IO, F2FS_BLKSIZE); 1400 1401 f2fs_update_data_blkaddr(&dn, newaddr); 1402 set_inode_flag(inode, FI_APPEND_WRITE); 1403 1404 f2fs_put_page(fio.encrypted_page, 1); 1405 recover_block: 1406 if (err) 1407 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr, 1408 true, true, true); 1409 up_out: 1410 if (lfs_mode) 1411 f2fs_up_write(&fio.sbi->io_order_lock); 1412 put_out: 1413 f2fs_put_dnode(&dn); 1414 out: 1415 f2fs_put_page(page, 1); 1416 return err; 1417 } 1418 1419 static int move_data_page(struct inode *inode, block_t bidx, int gc_type, 1420 unsigned int segno, int off) 1421 { 1422 struct page *page; 1423 int err = 0; 1424 1425 page = f2fs_get_lock_data_page(inode, bidx, true); 1426 if (IS_ERR(page)) 1427 return PTR_ERR(page); 1428 1429 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) { 1430 err = -ENOENT; 1431 goto out; 1432 } 1433 1434 err = f2fs_gc_pinned_control(inode, gc_type, segno); 1435 if (err) 1436 goto out; 1437 1438 if (gc_type == BG_GC) { 1439 if (PageWriteback(page)) { 1440 err = -EAGAIN; 1441 goto out; 1442 } 1443 set_page_dirty(page); 1444 set_page_private_gcing(page); 1445 } else { 1446 struct f2fs_io_info fio = { 1447 .sbi = F2FS_I_SB(inode), 1448 .ino = inode->i_ino, 1449 .type = DATA, 1450 .temp = COLD, 1451 .op = REQ_OP_WRITE, 1452 .op_flags = REQ_SYNC, 1453 .old_blkaddr = NULL_ADDR, 1454 .page = page, 1455 .encrypted_page = NULL, 1456 .need_lock = LOCK_REQ, 1457 .io_type = FS_GC_DATA_IO, 1458 }; 1459 bool is_dirty = PageDirty(page); 1460 1461 retry: 1462 f2fs_wait_on_page_writeback(page, DATA, true, true); 1463 1464 set_page_dirty(page); 1465 if (clear_page_dirty_for_io(page)) { 1466 inode_dec_dirty_pages(inode); 1467 f2fs_remove_dirty_inode(inode); 1468 } 1469 1470 set_page_private_gcing(page); 1471 1472 err = f2fs_do_write_data_page(&fio); 1473 if (err) { 1474 clear_page_private_gcing(page); 1475 if (err == -ENOMEM) { 1476 memalloc_retry_wait(GFP_NOFS); 1477 goto retry; 1478 } 1479 if (is_dirty) 1480 set_page_dirty(page); 1481 } 1482 } 1483 out: 1484 f2fs_put_page(page, 1); 1485 return err; 1486 } 1487 1488 /* 1489 * This function tries to get parent node of victim data block, and identifies 1490 * data block validity. If the block is valid, copy that with cold status and 1491 * modify parent node. 1492 * If the parent node is not valid or the data block address is different, 1493 * the victim data block is ignored. 1494 */ 1495 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 1496 struct gc_inode_list *gc_list, unsigned int segno, int gc_type, 1497 bool force_migrate) 1498 { 1499 struct super_block *sb = sbi->sb; 1500 struct f2fs_summary *entry; 1501 block_t start_addr; 1502 int off; 1503 int phase = 0; 1504 int submitted = 0; 1505 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 1506 1507 start_addr = START_BLOCK(sbi, segno); 1508 1509 next_step: 1510 entry = sum; 1511 1512 for (off = 0; off < usable_blks_in_seg; off++, entry++) { 1513 struct page *data_page; 1514 struct inode *inode; 1515 struct node_info dni; /* dnode info for the data */ 1516 unsigned int ofs_in_node, nofs; 1517 block_t start_bidx; 1518 nid_t nid = le32_to_cpu(entry->nid); 1519 1520 /* 1521 * stop BG_GC if there is not enough free sections. 1522 * Or, stop GC if the segment becomes fully valid caused by 1523 * race condition along with SSR block allocation. 1524 */ 1525 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) || 1526 (!force_migrate && get_valid_blocks(sbi, segno, true) == 1527 CAP_BLKS_PER_SEC(sbi))) 1528 return submitted; 1529 1530 if (check_valid_map(sbi, segno, off) == 0) 1531 continue; 1532 1533 if (phase == 0) { 1534 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, 1535 META_NAT, true); 1536 continue; 1537 } 1538 1539 if (phase == 1) { 1540 f2fs_ra_node_page(sbi, nid); 1541 continue; 1542 } 1543 1544 /* Get an inode by ino with checking validity */ 1545 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) 1546 continue; 1547 1548 if (phase == 2) { 1549 f2fs_ra_node_page(sbi, dni.ino); 1550 continue; 1551 } 1552 1553 ofs_in_node = le16_to_cpu(entry->ofs_in_node); 1554 1555 if (phase == 3) { 1556 int err; 1557 1558 inode = f2fs_iget(sb, dni.ino); 1559 if (IS_ERR(inode)) 1560 continue; 1561 1562 if (is_bad_inode(inode) || 1563 special_file(inode->i_mode)) { 1564 iput(inode); 1565 continue; 1566 } 1567 1568 if (f2fs_has_inline_data(inode)) { 1569 iput(inode); 1570 set_sbi_flag(sbi, SBI_NEED_FSCK); 1571 f2fs_err_ratelimited(sbi, 1572 "inode %lx has both inline_data flag and " 1573 "data block, nid=%u, ofs_in_node=%u", 1574 inode->i_ino, dni.nid, ofs_in_node); 1575 continue; 1576 } 1577 1578 err = f2fs_gc_pinned_control(inode, gc_type, segno); 1579 if (err == -EAGAIN) { 1580 iput(inode); 1581 return submitted; 1582 } 1583 1584 if (!f2fs_down_write_trylock( 1585 &F2FS_I(inode)->i_gc_rwsem[WRITE])) { 1586 iput(inode); 1587 sbi->skipped_gc_rwsem++; 1588 continue; 1589 } 1590 1591 start_bidx = f2fs_start_bidx_of_node(nofs, inode) + 1592 ofs_in_node; 1593 1594 if (f2fs_meta_inode_gc_required(inode)) { 1595 int err = ra_data_block(inode, start_bidx); 1596 1597 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1598 if (err) { 1599 iput(inode); 1600 continue; 1601 } 1602 add_gc_inode(gc_list, inode); 1603 continue; 1604 } 1605 1606 data_page = f2fs_get_read_data_page(inode, start_bidx, 1607 REQ_RAHEAD, true, NULL); 1608 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1609 if (IS_ERR(data_page)) { 1610 iput(inode); 1611 continue; 1612 } 1613 1614 f2fs_put_page(data_page, 0); 1615 add_gc_inode(gc_list, inode); 1616 continue; 1617 } 1618 1619 /* phase 4 */ 1620 inode = find_gc_inode(gc_list, dni.ino); 1621 if (inode) { 1622 struct f2fs_inode_info *fi = F2FS_I(inode); 1623 bool locked = false; 1624 int err; 1625 1626 if (S_ISREG(inode->i_mode)) { 1627 if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[WRITE])) { 1628 sbi->skipped_gc_rwsem++; 1629 continue; 1630 } 1631 if (!f2fs_down_write_trylock( 1632 &fi->i_gc_rwsem[READ])) { 1633 sbi->skipped_gc_rwsem++; 1634 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 1635 continue; 1636 } 1637 locked = true; 1638 1639 /* wait for all inflight aio data */ 1640 inode_dio_wait(inode); 1641 } 1642 1643 start_bidx = f2fs_start_bidx_of_node(nofs, inode) 1644 + ofs_in_node; 1645 if (f2fs_meta_inode_gc_required(inode)) 1646 err = move_data_block(inode, start_bidx, 1647 gc_type, segno, off); 1648 else 1649 err = move_data_page(inode, start_bidx, gc_type, 1650 segno, off); 1651 1652 if (!err && (gc_type == FG_GC || 1653 f2fs_meta_inode_gc_required(inode))) 1654 submitted++; 1655 1656 if (locked) { 1657 f2fs_up_write(&fi->i_gc_rwsem[READ]); 1658 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 1659 } 1660 1661 stat_inc_data_blk_count(sbi, 1, gc_type); 1662 } 1663 } 1664 1665 if (++phase < 5) 1666 goto next_step; 1667 1668 return submitted; 1669 } 1670 1671 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, 1672 int gc_type) 1673 { 1674 struct sit_info *sit_i = SIT_I(sbi); 1675 int ret; 1676 1677 down_write(&sit_i->sentry_lock); 1678 ret = f2fs_get_victim(sbi, victim, gc_type, NO_CHECK_TYPE, LFS, 0); 1679 up_write(&sit_i->sentry_lock); 1680 return ret; 1681 } 1682 1683 static int do_garbage_collect(struct f2fs_sb_info *sbi, 1684 unsigned int start_segno, 1685 struct gc_inode_list *gc_list, int gc_type, 1686 bool force_migrate) 1687 { 1688 struct page *sum_page; 1689 struct f2fs_summary_block *sum; 1690 struct blk_plug plug; 1691 unsigned int segno = start_segno; 1692 unsigned int end_segno = start_segno + SEGS_PER_SEC(sbi); 1693 int seg_freed = 0, migrated = 0; 1694 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ? 1695 SUM_TYPE_DATA : SUM_TYPE_NODE; 1696 unsigned char data_type = (type == SUM_TYPE_DATA) ? DATA : NODE; 1697 int submitted = 0; 1698 1699 if (__is_large_section(sbi)) 1700 end_segno = rounddown(end_segno, SEGS_PER_SEC(sbi)); 1701 1702 /* 1703 * zone-capacity can be less than zone-size in zoned devices, 1704 * resulting in less than expected usable segments in the zone, 1705 * calculate the end segno in the zone which can be garbage collected 1706 */ 1707 if (f2fs_sb_has_blkzoned(sbi)) 1708 end_segno -= SEGS_PER_SEC(sbi) - 1709 f2fs_usable_segs_in_sec(sbi, segno); 1710 1711 sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type); 1712 1713 /* readahead multi ssa blocks those have contiguous address */ 1714 if (__is_large_section(sbi)) 1715 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), 1716 end_segno - segno, META_SSA, true); 1717 1718 /* reference all summary page */ 1719 while (segno < end_segno) { 1720 sum_page = f2fs_get_sum_page(sbi, segno++); 1721 if (IS_ERR(sum_page)) { 1722 int err = PTR_ERR(sum_page); 1723 1724 end_segno = segno - 1; 1725 for (segno = start_segno; segno < end_segno; segno++) { 1726 sum_page = find_get_page(META_MAPPING(sbi), 1727 GET_SUM_BLOCK(sbi, segno)); 1728 f2fs_put_page(sum_page, 0); 1729 f2fs_put_page(sum_page, 0); 1730 } 1731 return err; 1732 } 1733 unlock_page(sum_page); 1734 } 1735 1736 blk_start_plug(&plug); 1737 1738 for (segno = start_segno; segno < end_segno; segno++) { 1739 1740 /* find segment summary of victim */ 1741 sum_page = find_get_page(META_MAPPING(sbi), 1742 GET_SUM_BLOCK(sbi, segno)); 1743 f2fs_put_page(sum_page, 0); 1744 1745 if (get_valid_blocks(sbi, segno, false) == 0) 1746 goto freed; 1747 if (gc_type == BG_GC && __is_large_section(sbi) && 1748 migrated >= sbi->migration_granularity) 1749 goto skip; 1750 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi))) 1751 goto skip; 1752 1753 sum = page_address(sum_page); 1754 if (type != GET_SUM_TYPE((&sum->footer))) { 1755 f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT", 1756 segno, type, GET_SUM_TYPE((&sum->footer))); 1757 set_sbi_flag(sbi, SBI_NEED_FSCK); 1758 f2fs_stop_checkpoint(sbi, false, 1759 STOP_CP_REASON_CORRUPTED_SUMMARY); 1760 goto skip; 1761 } 1762 1763 /* 1764 * this is to avoid deadlock: 1765 * - lock_page(sum_page) - f2fs_replace_block 1766 * - check_valid_map() - down_write(sentry_lock) 1767 * - down_read(sentry_lock) - change_curseg() 1768 * - lock_page(sum_page) 1769 */ 1770 if (type == SUM_TYPE_NODE) 1771 submitted += gc_node_segment(sbi, sum->entries, segno, 1772 gc_type); 1773 else 1774 submitted += gc_data_segment(sbi, sum->entries, gc_list, 1775 segno, gc_type, 1776 force_migrate); 1777 1778 stat_inc_gc_seg_count(sbi, data_type, gc_type); 1779 sbi->gc_reclaimed_segs[sbi->gc_mode]++; 1780 migrated++; 1781 1782 freed: 1783 if (gc_type == FG_GC && 1784 get_valid_blocks(sbi, segno, false) == 0) 1785 seg_freed++; 1786 1787 if (__is_large_section(sbi)) 1788 sbi->next_victim_seg[gc_type] = 1789 (segno + 1 < end_segno) ? segno + 1 : NULL_SEGNO; 1790 skip: 1791 f2fs_put_page(sum_page, 0); 1792 } 1793 1794 if (submitted) 1795 f2fs_submit_merged_write(sbi, data_type); 1796 1797 blk_finish_plug(&plug); 1798 1799 if (migrated) 1800 stat_inc_gc_sec_count(sbi, data_type, gc_type); 1801 1802 return seg_freed; 1803 } 1804 1805 int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control) 1806 { 1807 int gc_type = gc_control->init_gc_type; 1808 unsigned int segno = gc_control->victim_segno; 1809 int sec_freed = 0, seg_freed = 0, total_freed = 0, total_sec_freed = 0; 1810 int ret = 0; 1811 struct cp_control cpc; 1812 struct gc_inode_list gc_list = { 1813 .ilist = LIST_HEAD_INIT(gc_list.ilist), 1814 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), 1815 }; 1816 unsigned int skipped_round = 0, round = 0; 1817 unsigned int upper_secs; 1818 1819 trace_f2fs_gc_begin(sbi->sb, gc_type, gc_control->no_bg_gc, 1820 gc_control->nr_free_secs, 1821 get_pages(sbi, F2FS_DIRTY_NODES), 1822 get_pages(sbi, F2FS_DIRTY_DENTS), 1823 get_pages(sbi, F2FS_DIRTY_IMETA), 1824 free_sections(sbi), 1825 free_segments(sbi), 1826 reserved_segments(sbi), 1827 prefree_segments(sbi)); 1828 1829 cpc.reason = __get_cp_reason(sbi); 1830 gc_more: 1831 sbi->skipped_gc_rwsem = 0; 1832 if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) { 1833 ret = -EINVAL; 1834 goto stop; 1835 } 1836 if (unlikely(f2fs_cp_error(sbi))) { 1837 ret = -EIO; 1838 goto stop; 1839 } 1840 1841 /* Let's run FG_GC, if we don't have enough space. */ 1842 if (has_not_enough_free_secs(sbi, 0, 0)) { 1843 gc_type = FG_GC; 1844 1845 /* 1846 * For example, if there are many prefree_segments below given 1847 * threshold, we can make them free by checkpoint. Then, we 1848 * secure free segments which doesn't need fggc any more. 1849 */ 1850 if (prefree_segments(sbi)) { 1851 stat_inc_cp_call_count(sbi, TOTAL_CALL); 1852 ret = f2fs_write_checkpoint(sbi, &cpc); 1853 if (ret) 1854 goto stop; 1855 /* Reset due to checkpoint */ 1856 sec_freed = 0; 1857 } 1858 } 1859 1860 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */ 1861 if (gc_type == BG_GC && gc_control->no_bg_gc) { 1862 ret = -EINVAL; 1863 goto stop; 1864 } 1865 retry: 1866 ret = __get_victim(sbi, &segno, gc_type); 1867 if (ret) { 1868 /* allow to search victim from sections has pinned data */ 1869 if (ret == -ENODATA && gc_type == FG_GC && 1870 f2fs_pinned_section_exists(DIRTY_I(sbi))) { 1871 f2fs_unpin_all_sections(sbi, false); 1872 goto retry; 1873 } 1874 goto stop; 1875 } 1876 1877 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type, 1878 gc_control->should_migrate_blocks); 1879 total_freed += seg_freed; 1880 1881 if (seg_freed == f2fs_usable_segs_in_sec(sbi, segno)) { 1882 sec_freed++; 1883 total_sec_freed++; 1884 } 1885 1886 if (gc_type == FG_GC) { 1887 sbi->cur_victim_sec = NULL_SEGNO; 1888 1889 if (has_enough_free_secs(sbi, sec_freed, 0)) { 1890 if (!gc_control->no_bg_gc && 1891 total_sec_freed < gc_control->nr_free_secs) 1892 goto go_gc_more; 1893 goto stop; 1894 } 1895 if (sbi->skipped_gc_rwsem) 1896 skipped_round++; 1897 round++; 1898 if (skipped_round > MAX_SKIP_GC_COUNT && 1899 skipped_round * 2 >= round) { 1900 stat_inc_cp_call_count(sbi, TOTAL_CALL); 1901 ret = f2fs_write_checkpoint(sbi, &cpc); 1902 goto stop; 1903 } 1904 } else if (has_enough_free_secs(sbi, 0, 0)) { 1905 goto stop; 1906 } 1907 1908 __get_secs_required(sbi, NULL, &upper_secs, NULL); 1909 1910 /* 1911 * Write checkpoint to reclaim prefree segments. 1912 * We need more three extra sections for writer's data/node/dentry. 1913 */ 1914 if (free_sections(sbi) <= upper_secs + NR_GC_CHECKPOINT_SECS && 1915 prefree_segments(sbi)) { 1916 stat_inc_cp_call_count(sbi, TOTAL_CALL); 1917 ret = f2fs_write_checkpoint(sbi, &cpc); 1918 if (ret) 1919 goto stop; 1920 /* Reset due to checkpoint */ 1921 sec_freed = 0; 1922 } 1923 go_gc_more: 1924 segno = NULL_SEGNO; 1925 goto gc_more; 1926 1927 stop: 1928 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0; 1929 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno; 1930 1931 if (gc_type == FG_GC) 1932 f2fs_unpin_all_sections(sbi, true); 1933 1934 trace_f2fs_gc_end(sbi->sb, ret, total_freed, total_sec_freed, 1935 get_pages(sbi, F2FS_DIRTY_NODES), 1936 get_pages(sbi, F2FS_DIRTY_DENTS), 1937 get_pages(sbi, F2FS_DIRTY_IMETA), 1938 free_sections(sbi), 1939 free_segments(sbi), 1940 reserved_segments(sbi), 1941 prefree_segments(sbi)); 1942 1943 f2fs_up_write(&sbi->gc_lock); 1944 1945 put_gc_inode(&gc_list); 1946 1947 if (gc_control->err_gc_skipped && !ret) 1948 ret = total_sec_freed ? 0 : -EAGAIN; 1949 return ret; 1950 } 1951 1952 int __init f2fs_create_garbage_collection_cache(void) 1953 { 1954 victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry", 1955 sizeof(struct victim_entry)); 1956 return victim_entry_slab ? 0 : -ENOMEM; 1957 } 1958 1959 void f2fs_destroy_garbage_collection_cache(void) 1960 { 1961 kmem_cache_destroy(victim_entry_slab); 1962 } 1963 1964 static void init_atgc_management(struct f2fs_sb_info *sbi) 1965 { 1966 struct atgc_management *am = &sbi->am; 1967 1968 if (test_opt(sbi, ATGC) && 1969 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD) 1970 am->atgc_enabled = true; 1971 1972 am->root = RB_ROOT_CACHED; 1973 INIT_LIST_HEAD(&am->victim_list); 1974 am->victim_count = 0; 1975 1976 am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO; 1977 am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT; 1978 am->age_weight = DEF_GC_THREAD_AGE_WEIGHT; 1979 am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD; 1980 } 1981 1982 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi) 1983 { 1984 sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES; 1985 1986 /* give warm/cold data area from slower device */ 1987 if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi)) 1988 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 1989 GET_SEGNO(sbi, FDEV(0).end_blk) + 1; 1990 1991 init_atgc_management(sbi); 1992 } 1993 1994 int f2fs_gc_range(struct f2fs_sb_info *sbi, 1995 unsigned int start_seg, unsigned int end_seg, 1996 bool dry_run, unsigned int dry_run_sections) 1997 { 1998 unsigned int segno; 1999 unsigned int gc_secs = dry_run_sections; 2000 2001 for (segno = start_seg; segno <= end_seg; segno += SEGS_PER_SEC(sbi)) { 2002 struct gc_inode_list gc_list = { 2003 .ilist = LIST_HEAD_INIT(gc_list.ilist), 2004 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), 2005 }; 2006 2007 do_garbage_collect(sbi, segno, &gc_list, FG_GC, 2008 dry_run_sections == 0); 2009 put_gc_inode(&gc_list); 2010 2011 if (!dry_run && get_valid_blocks(sbi, segno, true)) 2012 return -EAGAIN; 2013 if (dry_run && dry_run_sections && 2014 !get_valid_blocks(sbi, segno, true) && --gc_secs == 0) 2015 break; 2016 2017 if (fatal_signal_pending(current)) 2018 return -ERESTARTSYS; 2019 } 2020 2021 return 0; 2022 } 2023 2024 static int free_segment_range(struct f2fs_sb_info *sbi, 2025 unsigned int secs, bool dry_run) 2026 { 2027 unsigned int next_inuse, start, end; 2028 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 }; 2029 int gc_mode, gc_type; 2030 int err = 0; 2031 int type; 2032 2033 /* Force block allocation for GC */ 2034 MAIN_SECS(sbi) -= secs; 2035 start = MAIN_SECS(sbi) * SEGS_PER_SEC(sbi); 2036 end = MAIN_SEGS(sbi) - 1; 2037 2038 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 2039 for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++) 2040 if (SIT_I(sbi)->last_victim[gc_mode] >= start) 2041 SIT_I(sbi)->last_victim[gc_mode] = 0; 2042 2043 for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++) 2044 if (sbi->next_victim_seg[gc_type] >= start) 2045 sbi->next_victim_seg[gc_type] = NULL_SEGNO; 2046 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 2047 2048 /* Move out cursegs from the target range */ 2049 for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++) 2050 f2fs_allocate_segment_for_resize(sbi, type, start, end); 2051 2052 /* do GC to move out valid blocks in the range */ 2053 err = f2fs_gc_range(sbi, start, end, dry_run, 0); 2054 if (err || dry_run) 2055 goto out; 2056 2057 stat_inc_cp_call_count(sbi, TOTAL_CALL); 2058 err = f2fs_write_checkpoint(sbi, &cpc); 2059 if (err) 2060 goto out; 2061 2062 next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start); 2063 if (next_inuse <= end) { 2064 f2fs_err(sbi, "segno %u should be free but still inuse!", 2065 next_inuse); 2066 f2fs_bug_on(sbi, 1); 2067 } 2068 out: 2069 MAIN_SECS(sbi) += secs; 2070 return err; 2071 } 2072 2073 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs) 2074 { 2075 struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi); 2076 int section_count; 2077 int segment_count; 2078 int segment_count_main; 2079 long long block_count; 2080 int segs = secs * SEGS_PER_SEC(sbi); 2081 2082 f2fs_down_write(&sbi->sb_lock); 2083 2084 section_count = le32_to_cpu(raw_sb->section_count); 2085 segment_count = le32_to_cpu(raw_sb->segment_count); 2086 segment_count_main = le32_to_cpu(raw_sb->segment_count_main); 2087 block_count = le64_to_cpu(raw_sb->block_count); 2088 2089 raw_sb->section_count = cpu_to_le32(section_count + secs); 2090 raw_sb->segment_count = cpu_to_le32(segment_count + segs); 2091 raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs); 2092 raw_sb->block_count = cpu_to_le64(block_count + 2093 (long long)(segs << sbi->log_blocks_per_seg)); 2094 if (f2fs_is_multi_device(sbi)) { 2095 int last_dev = sbi->s_ndevs - 1; 2096 int dev_segs = 2097 le32_to_cpu(raw_sb->devs[last_dev].total_segments); 2098 2099 raw_sb->devs[last_dev].total_segments = 2100 cpu_to_le32(dev_segs + segs); 2101 } 2102 2103 f2fs_up_write(&sbi->sb_lock); 2104 } 2105 2106 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs) 2107 { 2108 int segs = secs * SEGS_PER_SEC(sbi); 2109 long long blks = (long long)segs << sbi->log_blocks_per_seg; 2110 long long user_block_count = 2111 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count); 2112 2113 SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs; 2114 MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs; 2115 MAIN_SECS(sbi) += secs; 2116 FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs; 2117 FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs; 2118 F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks); 2119 2120 if (f2fs_is_multi_device(sbi)) { 2121 int last_dev = sbi->s_ndevs - 1; 2122 2123 FDEV(last_dev).total_segments = 2124 (int)FDEV(last_dev).total_segments + segs; 2125 FDEV(last_dev).end_blk = 2126 (long long)FDEV(last_dev).end_blk + blks; 2127 #ifdef CONFIG_BLK_DEV_ZONED 2128 FDEV(last_dev).nr_blkz = FDEV(last_dev).nr_blkz + 2129 div_u64(blks, sbi->blocks_per_blkz); 2130 #endif 2131 } 2132 } 2133 2134 int f2fs_resize_fs(struct file *filp, __u64 block_count) 2135 { 2136 struct f2fs_sb_info *sbi = F2FS_I_SB(file_inode(filp)); 2137 __u64 old_block_count, shrunk_blocks; 2138 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 }; 2139 unsigned int secs; 2140 int err = 0; 2141 __u32 rem; 2142 2143 old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count); 2144 if (block_count > old_block_count) 2145 return -EINVAL; 2146 2147 if (f2fs_is_multi_device(sbi)) { 2148 int last_dev = sbi->s_ndevs - 1; 2149 __u64 last_segs = FDEV(last_dev).total_segments; 2150 2151 if (block_count + (last_segs << sbi->log_blocks_per_seg) <= 2152 old_block_count) 2153 return -EINVAL; 2154 } 2155 2156 /* new fs size should align to section size */ 2157 div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem); 2158 if (rem) 2159 return -EINVAL; 2160 2161 if (block_count == old_block_count) 2162 return 0; 2163 2164 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 2165 f2fs_err(sbi, "Should run fsck to repair first."); 2166 return -EFSCORRUPTED; 2167 } 2168 2169 if (test_opt(sbi, DISABLE_CHECKPOINT)) { 2170 f2fs_err(sbi, "Checkpoint should be enabled."); 2171 return -EINVAL; 2172 } 2173 2174 err = mnt_want_write_file(filp); 2175 if (err) 2176 return err; 2177 2178 shrunk_blocks = old_block_count - block_count; 2179 secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi)); 2180 2181 /* stop other GC */ 2182 if (!f2fs_down_write_trylock(&sbi->gc_lock)) { 2183 err = -EAGAIN; 2184 goto out_drop_write; 2185 } 2186 2187 /* stop CP to protect MAIN_SEC in free_segment_range */ 2188 f2fs_lock_op(sbi); 2189 2190 spin_lock(&sbi->stat_lock); 2191 if (shrunk_blocks + valid_user_blocks(sbi) + 2192 sbi->current_reserved_blocks + sbi->unusable_block_count + 2193 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count) 2194 err = -ENOSPC; 2195 spin_unlock(&sbi->stat_lock); 2196 2197 if (err) 2198 goto out_unlock; 2199 2200 err = free_segment_range(sbi, secs, true); 2201 2202 out_unlock: 2203 f2fs_unlock_op(sbi); 2204 f2fs_up_write(&sbi->gc_lock); 2205 out_drop_write: 2206 mnt_drop_write_file(filp); 2207 if (err) 2208 return err; 2209 2210 err = freeze_super(sbi->sb, FREEZE_HOLDER_USERSPACE); 2211 if (err) 2212 return err; 2213 2214 if (f2fs_readonly(sbi->sb)) { 2215 err = thaw_super(sbi->sb, FREEZE_HOLDER_USERSPACE); 2216 if (err) 2217 return err; 2218 return -EROFS; 2219 } 2220 2221 f2fs_down_write(&sbi->gc_lock); 2222 f2fs_down_write(&sbi->cp_global_sem); 2223 2224 spin_lock(&sbi->stat_lock); 2225 if (shrunk_blocks + valid_user_blocks(sbi) + 2226 sbi->current_reserved_blocks + sbi->unusable_block_count + 2227 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count) 2228 err = -ENOSPC; 2229 else 2230 sbi->user_block_count -= shrunk_blocks; 2231 spin_unlock(&sbi->stat_lock); 2232 if (err) 2233 goto out_err; 2234 2235 set_sbi_flag(sbi, SBI_IS_RESIZEFS); 2236 err = free_segment_range(sbi, secs, false); 2237 if (err) 2238 goto recover_out; 2239 2240 update_sb_metadata(sbi, -secs); 2241 2242 err = f2fs_commit_super(sbi, false); 2243 if (err) { 2244 update_sb_metadata(sbi, secs); 2245 goto recover_out; 2246 } 2247 2248 update_fs_metadata(sbi, -secs); 2249 clear_sbi_flag(sbi, SBI_IS_RESIZEFS); 2250 set_sbi_flag(sbi, SBI_IS_DIRTY); 2251 2252 stat_inc_cp_call_count(sbi, TOTAL_CALL); 2253 err = f2fs_write_checkpoint(sbi, &cpc); 2254 if (err) { 2255 update_fs_metadata(sbi, secs); 2256 update_sb_metadata(sbi, secs); 2257 f2fs_commit_super(sbi, false); 2258 } 2259 recover_out: 2260 clear_sbi_flag(sbi, SBI_IS_RESIZEFS); 2261 if (err) { 2262 set_sbi_flag(sbi, SBI_NEED_FSCK); 2263 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!"); 2264 2265 spin_lock(&sbi->stat_lock); 2266 sbi->user_block_count += shrunk_blocks; 2267 spin_unlock(&sbi->stat_lock); 2268 } 2269 out_err: 2270 f2fs_up_write(&sbi->cp_global_sem); 2271 f2fs_up_write(&sbi->gc_lock); 2272 thaw_super(sbi->sb, FREEZE_HOLDER_USERSPACE); 2273 return err; 2274 } 2275