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/backing-dev.h> 11 #include <linux/init.h> 12 #include <linux/f2fs_fs.h> 13 #include <linux/kthread.h> 14 #include <linux/delay.h> 15 #include <linux/freezer.h> 16 #include <linux/sched/signal.h> 17 18 #include "f2fs.h" 19 #include "node.h" 20 #include "segment.h" 21 #include "gc.h" 22 #include <trace/events/f2fs.h> 23 24 static struct kmem_cache *victim_entry_slab; 25 26 static unsigned int count_bits(const unsigned long *addr, 27 unsigned int offset, unsigned int len); 28 29 static int gc_thread_func(void *data) 30 { 31 struct f2fs_sb_info *sbi = data; 32 struct f2fs_gc_kthread *gc_th = sbi->gc_thread; 33 wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; 34 unsigned int wait_ms; 35 36 wait_ms = gc_th->min_sleep_time; 37 38 set_freezable(); 39 do { 40 bool sync_mode; 41 42 wait_event_interruptible_timeout(*wq, 43 kthread_should_stop() || freezing(current) || 44 gc_th->gc_wake, 45 msecs_to_jiffies(wait_ms)); 46 47 /* give it a try one time */ 48 if (gc_th->gc_wake) 49 gc_th->gc_wake = 0; 50 51 if (try_to_freeze()) { 52 stat_other_skip_bggc_count(sbi); 53 continue; 54 } 55 if (kthread_should_stop()) 56 break; 57 58 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { 59 increase_sleep_time(gc_th, &wait_ms); 60 stat_other_skip_bggc_count(sbi); 61 continue; 62 } 63 64 if (time_to_inject(sbi, FAULT_CHECKPOINT)) { 65 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT); 66 f2fs_stop_checkpoint(sbi, false); 67 } 68 69 if (!sb_start_write_trylock(sbi->sb)) { 70 stat_other_skip_bggc_count(sbi); 71 continue; 72 } 73 74 /* 75 * [GC triggering condition] 76 * 0. GC is not conducted currently. 77 * 1. There are enough dirty segments. 78 * 2. IO subsystem is idle by checking the # of writeback pages. 79 * 3. IO subsystem is idle by checking the # of requests in 80 * bdev's request list. 81 * 82 * Note) We have to avoid triggering GCs frequently. 83 * Because it is possible that some segments can be 84 * invalidated soon after by user update or deletion. 85 * So, I'd like to wait some time to collect dirty segments. 86 */ 87 if (sbi->gc_mode == GC_URGENT_HIGH) { 88 wait_ms = gc_th->urgent_sleep_time; 89 down_write(&sbi->gc_lock); 90 goto do_gc; 91 } 92 93 if (!down_write_trylock(&sbi->gc_lock)) { 94 stat_other_skip_bggc_count(sbi); 95 goto next; 96 } 97 98 if (!is_idle(sbi, GC_TIME)) { 99 increase_sleep_time(gc_th, &wait_ms); 100 up_write(&sbi->gc_lock); 101 stat_io_skip_bggc_count(sbi); 102 goto next; 103 } 104 105 if (has_enough_invalid_blocks(sbi)) 106 decrease_sleep_time(gc_th, &wait_ms); 107 else 108 increase_sleep_time(gc_th, &wait_ms); 109 do_gc: 110 stat_inc_bggc_count(sbi->stat_info); 111 112 sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC; 113 114 /* if return value is not zero, no victim was selected */ 115 if (f2fs_gc(sbi, sync_mode, true, NULL_SEGNO)) 116 wait_ms = gc_th->no_gc_sleep_time; 117 118 trace_f2fs_background_gc(sbi->sb, wait_ms, 119 prefree_segments(sbi), free_segments(sbi)); 120 121 /* balancing f2fs's metadata periodically */ 122 f2fs_balance_fs_bg(sbi, true); 123 next: 124 sb_end_write(sbi->sb); 125 126 } while (!kthread_should_stop()); 127 return 0; 128 } 129 130 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi) 131 { 132 struct f2fs_gc_kthread *gc_th; 133 dev_t dev = sbi->sb->s_bdev->bd_dev; 134 int err = 0; 135 136 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL); 137 if (!gc_th) { 138 err = -ENOMEM; 139 goto out; 140 } 141 142 gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME; 143 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; 144 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; 145 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; 146 147 gc_th->gc_wake= 0; 148 149 sbi->gc_thread = gc_th; 150 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); 151 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, 152 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); 153 if (IS_ERR(gc_th->f2fs_gc_task)) { 154 err = PTR_ERR(gc_th->f2fs_gc_task); 155 kfree(gc_th); 156 sbi->gc_thread = NULL; 157 } 158 out: 159 return err; 160 } 161 162 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi) 163 { 164 struct f2fs_gc_kthread *gc_th = sbi->gc_thread; 165 if (!gc_th) 166 return; 167 kthread_stop(gc_th->f2fs_gc_task); 168 kfree(gc_th); 169 sbi->gc_thread = NULL; 170 } 171 172 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type) 173 { 174 int gc_mode; 175 176 if (gc_type == BG_GC) { 177 if (sbi->am.atgc_enabled) 178 gc_mode = GC_AT; 179 else 180 gc_mode = GC_CB; 181 } else { 182 gc_mode = GC_GREEDY; 183 } 184 185 switch (sbi->gc_mode) { 186 case GC_IDLE_CB: 187 gc_mode = GC_CB; 188 break; 189 case GC_IDLE_GREEDY: 190 case GC_URGENT_HIGH: 191 gc_mode = GC_GREEDY; 192 break; 193 case GC_IDLE_AT: 194 gc_mode = GC_AT; 195 break; 196 } 197 198 return gc_mode; 199 } 200 201 static void select_policy(struct f2fs_sb_info *sbi, int gc_type, 202 int type, struct victim_sel_policy *p) 203 { 204 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 205 206 if (p->alloc_mode == SSR) { 207 p->gc_mode = GC_GREEDY; 208 p->dirty_bitmap = dirty_i->dirty_segmap[type]; 209 p->max_search = dirty_i->nr_dirty[type]; 210 p->ofs_unit = 1; 211 } else if (p->alloc_mode == AT_SSR) { 212 p->gc_mode = GC_GREEDY; 213 p->dirty_bitmap = dirty_i->dirty_segmap[type]; 214 p->max_search = dirty_i->nr_dirty[type]; 215 p->ofs_unit = 1; 216 } else { 217 p->gc_mode = select_gc_type(sbi, gc_type); 218 p->ofs_unit = sbi->segs_per_sec; 219 if (__is_large_section(sbi)) { 220 p->dirty_bitmap = dirty_i->dirty_secmap; 221 p->max_search = count_bits(p->dirty_bitmap, 222 0, MAIN_SECS(sbi)); 223 } else { 224 p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY]; 225 p->max_search = dirty_i->nr_dirty[DIRTY]; 226 } 227 } 228 229 /* 230 * adjust candidates range, should select all dirty segments for 231 * foreground GC and urgent GC cases. 232 */ 233 if (gc_type != FG_GC && 234 (sbi->gc_mode != GC_URGENT_HIGH) && 235 (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) && 236 p->max_search > sbi->max_victim_search) 237 p->max_search = sbi->max_victim_search; 238 239 /* let's select beginning hot/small space first in no_heap mode*/ 240 if (test_opt(sbi, NOHEAP) && 241 (type == CURSEG_HOT_DATA || IS_NODESEG(type))) 242 p->offset = 0; 243 else 244 p->offset = SIT_I(sbi)->last_victim[p->gc_mode]; 245 } 246 247 static unsigned int get_max_cost(struct f2fs_sb_info *sbi, 248 struct victim_sel_policy *p) 249 { 250 /* SSR allocates in a segment unit */ 251 if (p->alloc_mode == SSR) 252 return sbi->blocks_per_seg; 253 else if (p->alloc_mode == AT_SSR) 254 return UINT_MAX; 255 256 /* LFS */ 257 if (p->gc_mode == GC_GREEDY) 258 return 2 * sbi->blocks_per_seg * p->ofs_unit; 259 else if (p->gc_mode == GC_CB) 260 return UINT_MAX; 261 else if (p->gc_mode == GC_AT) 262 return UINT_MAX; 263 else /* No other gc_mode */ 264 return 0; 265 } 266 267 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) 268 { 269 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 270 unsigned int secno; 271 272 /* 273 * If the gc_type is FG_GC, we can select victim segments 274 * selected by background GC before. 275 * Those segments guarantee they have small valid blocks. 276 */ 277 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { 278 if (sec_usage_check(sbi, secno)) 279 continue; 280 clear_bit(secno, dirty_i->victim_secmap); 281 return GET_SEG_FROM_SEC(sbi, secno); 282 } 283 return NULL_SEGNO; 284 } 285 286 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) 287 { 288 struct sit_info *sit_i = SIT_I(sbi); 289 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 290 unsigned int start = GET_SEG_FROM_SEC(sbi, secno); 291 unsigned long long mtime = 0; 292 unsigned int vblocks; 293 unsigned char age = 0; 294 unsigned char u; 295 unsigned int i; 296 unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno); 297 298 for (i = 0; i < usable_segs_per_sec; i++) 299 mtime += get_seg_entry(sbi, start + i)->mtime; 300 vblocks = get_valid_blocks(sbi, segno, true); 301 302 mtime = div_u64(mtime, usable_segs_per_sec); 303 vblocks = div_u64(vblocks, usable_segs_per_sec); 304 305 u = (vblocks * 100) >> sbi->log_blocks_per_seg; 306 307 /* Handle if the system time has changed by the user */ 308 if (mtime < sit_i->min_mtime) 309 sit_i->min_mtime = mtime; 310 if (mtime > sit_i->max_mtime) 311 sit_i->max_mtime = mtime; 312 if (sit_i->max_mtime != sit_i->min_mtime) 313 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), 314 sit_i->max_mtime - sit_i->min_mtime); 315 316 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); 317 } 318 319 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, 320 unsigned int segno, struct victim_sel_policy *p) 321 { 322 if (p->alloc_mode == SSR) 323 return get_seg_entry(sbi, segno)->ckpt_valid_blocks; 324 325 /* alloc_mode == LFS */ 326 if (p->gc_mode == GC_GREEDY) 327 return get_valid_blocks(sbi, segno, true); 328 else if (p->gc_mode == GC_CB) 329 return get_cb_cost(sbi, segno); 330 331 f2fs_bug_on(sbi, 1); 332 return 0; 333 } 334 335 static unsigned int count_bits(const unsigned long *addr, 336 unsigned int offset, unsigned int len) 337 { 338 unsigned int end = offset + len, sum = 0; 339 340 while (offset < end) { 341 if (test_bit(offset++, addr)) 342 ++sum; 343 } 344 return sum; 345 } 346 347 static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi, 348 unsigned long long mtime, unsigned int segno, 349 struct rb_node *parent, struct rb_node **p, 350 bool left_most) 351 { 352 struct atgc_management *am = &sbi->am; 353 struct victim_entry *ve; 354 355 ve = f2fs_kmem_cache_alloc(victim_entry_slab, GFP_NOFS); 356 357 ve->mtime = mtime; 358 ve->segno = segno; 359 360 rb_link_node(&ve->rb_node, parent, p); 361 rb_insert_color_cached(&ve->rb_node, &am->root, left_most); 362 363 list_add_tail(&ve->list, &am->victim_list); 364 365 am->victim_count++; 366 367 return ve; 368 } 369 370 static void insert_victim_entry(struct f2fs_sb_info *sbi, 371 unsigned long long mtime, unsigned int segno) 372 { 373 struct atgc_management *am = &sbi->am; 374 struct rb_node **p; 375 struct rb_node *parent = NULL; 376 bool left_most = true; 377 378 p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most); 379 attach_victim_entry(sbi, mtime, segno, parent, p, left_most); 380 } 381 382 static void add_victim_entry(struct f2fs_sb_info *sbi, 383 struct victim_sel_policy *p, unsigned int segno) 384 { 385 struct sit_info *sit_i = SIT_I(sbi); 386 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); 387 unsigned int start = GET_SEG_FROM_SEC(sbi, secno); 388 unsigned long long mtime = 0; 389 unsigned int i; 390 391 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 392 if (p->gc_mode == GC_AT && 393 get_valid_blocks(sbi, segno, true) == 0) 394 return; 395 396 if (p->alloc_mode == AT_SSR && 397 get_seg_entry(sbi, segno)->ckpt_valid_blocks == 0) 398 return; 399 } 400 401 for (i = 0; i < sbi->segs_per_sec; i++) 402 mtime += get_seg_entry(sbi, start + i)->mtime; 403 mtime = div_u64(mtime, sbi->segs_per_sec); 404 405 /* Handle if the system time has changed by the user */ 406 if (mtime < sit_i->min_mtime) 407 sit_i->min_mtime = mtime; 408 if (mtime > sit_i->max_mtime) 409 sit_i->max_mtime = mtime; 410 if (mtime < sit_i->dirty_min_mtime) 411 sit_i->dirty_min_mtime = mtime; 412 if (mtime > sit_i->dirty_max_mtime) 413 sit_i->dirty_max_mtime = mtime; 414 415 /* don't choose young section as candidate */ 416 if (sit_i->dirty_max_mtime - mtime < p->age_threshold) 417 return; 418 419 insert_victim_entry(sbi, mtime, segno); 420 } 421 422 static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi, 423 struct victim_sel_policy *p) 424 { 425 struct atgc_management *am = &sbi->am; 426 struct rb_node *parent = NULL; 427 bool left_most; 428 429 f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most); 430 431 return parent; 432 } 433 434 static void atgc_lookup_victim(struct f2fs_sb_info *sbi, 435 struct victim_sel_policy *p) 436 { 437 struct sit_info *sit_i = SIT_I(sbi); 438 struct atgc_management *am = &sbi->am; 439 struct rb_root_cached *root = &am->root; 440 struct rb_node *node; 441 struct rb_entry *re; 442 struct victim_entry *ve; 443 unsigned long long total_time; 444 unsigned long long age, u, accu; 445 unsigned long long max_mtime = sit_i->dirty_max_mtime; 446 unsigned long long min_mtime = sit_i->dirty_min_mtime; 447 unsigned int sec_blocks = BLKS_PER_SEC(sbi); 448 unsigned int vblocks; 449 unsigned int dirty_threshold = max(am->max_candidate_count, 450 am->candidate_ratio * 451 am->victim_count / 100); 452 unsigned int age_weight = am->age_weight; 453 unsigned int cost; 454 unsigned int iter = 0; 455 456 if (max_mtime < min_mtime) 457 return; 458 459 max_mtime += 1; 460 total_time = max_mtime - min_mtime; 461 462 accu = div64_u64(ULLONG_MAX, total_time); 463 accu = min_t(unsigned long long, div_u64(accu, 100), 464 DEFAULT_ACCURACY_CLASS); 465 466 node = rb_first_cached(root); 467 next: 468 re = rb_entry_safe(node, struct rb_entry, rb_node); 469 if (!re) 470 return; 471 472 ve = (struct victim_entry *)re; 473 474 if (ve->mtime >= max_mtime || ve->mtime < min_mtime) 475 goto skip; 476 477 /* age = 10000 * x% * 60 */ 478 age = div64_u64(accu * (max_mtime - ve->mtime), total_time) * 479 age_weight; 480 481 vblocks = get_valid_blocks(sbi, ve->segno, true); 482 f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks); 483 484 /* u = 10000 * x% * 40 */ 485 u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) * 486 (100 - age_weight); 487 488 f2fs_bug_on(sbi, age + u >= UINT_MAX); 489 490 cost = UINT_MAX - (age + u); 491 iter++; 492 493 if (cost < p->min_cost || 494 (cost == p->min_cost && age > p->oldest_age)) { 495 p->min_cost = cost; 496 p->oldest_age = age; 497 p->min_segno = ve->segno; 498 } 499 skip: 500 if (iter < dirty_threshold) { 501 node = rb_next(node); 502 goto next; 503 } 504 } 505 506 /* 507 * select candidates around source section in range of 508 * [target - dirty_threshold, target + dirty_threshold] 509 */ 510 static void atssr_lookup_victim(struct f2fs_sb_info *sbi, 511 struct victim_sel_policy *p) 512 { 513 struct sit_info *sit_i = SIT_I(sbi); 514 struct atgc_management *am = &sbi->am; 515 struct rb_node *node; 516 struct rb_entry *re; 517 struct victim_entry *ve; 518 unsigned long long age; 519 unsigned long long max_mtime = sit_i->dirty_max_mtime; 520 unsigned long long min_mtime = sit_i->dirty_min_mtime; 521 unsigned int seg_blocks = sbi->blocks_per_seg; 522 unsigned int vblocks; 523 unsigned int dirty_threshold = max(am->max_candidate_count, 524 am->candidate_ratio * 525 am->victim_count / 100); 526 unsigned int cost; 527 unsigned int iter = 0; 528 int stage = 0; 529 530 if (max_mtime < min_mtime) 531 return; 532 max_mtime += 1; 533 next_stage: 534 node = lookup_central_victim(sbi, p); 535 next_node: 536 re = rb_entry_safe(node, struct rb_entry, rb_node); 537 if (!re) { 538 if (stage == 0) 539 goto skip_stage; 540 return; 541 } 542 543 ve = (struct victim_entry *)re; 544 545 if (ve->mtime >= max_mtime || ve->mtime < min_mtime) 546 goto skip_node; 547 548 age = max_mtime - ve->mtime; 549 550 vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks; 551 f2fs_bug_on(sbi, !vblocks); 552 553 /* rare case */ 554 if (vblocks == seg_blocks) 555 goto skip_node; 556 557 iter++; 558 559 age = max_mtime - abs(p->age - age); 560 cost = UINT_MAX - vblocks; 561 562 if (cost < p->min_cost || 563 (cost == p->min_cost && age > p->oldest_age)) { 564 p->min_cost = cost; 565 p->oldest_age = age; 566 p->min_segno = ve->segno; 567 } 568 skip_node: 569 if (iter < dirty_threshold) { 570 if (stage == 0) 571 node = rb_prev(node); 572 else if (stage == 1) 573 node = rb_next(node); 574 goto next_node; 575 } 576 skip_stage: 577 if (stage < 1) { 578 stage++; 579 iter = 0; 580 goto next_stage; 581 } 582 } 583 static void lookup_victim_by_age(struct f2fs_sb_info *sbi, 584 struct victim_sel_policy *p) 585 { 586 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, 587 &sbi->am.root, true)); 588 589 if (p->gc_mode == GC_AT) 590 atgc_lookup_victim(sbi, p); 591 else if (p->alloc_mode == AT_SSR) 592 atssr_lookup_victim(sbi, p); 593 else 594 f2fs_bug_on(sbi, 1); 595 } 596 597 static void release_victim_entry(struct f2fs_sb_info *sbi) 598 { 599 struct atgc_management *am = &sbi->am; 600 struct victim_entry *ve, *tmp; 601 602 list_for_each_entry_safe(ve, tmp, &am->victim_list, list) { 603 list_del(&ve->list); 604 kmem_cache_free(victim_entry_slab, ve); 605 am->victim_count--; 606 } 607 608 am->root = RB_ROOT_CACHED; 609 610 f2fs_bug_on(sbi, am->victim_count); 611 f2fs_bug_on(sbi, !list_empty(&am->victim_list)); 612 } 613 614 /* 615 * This function is called from two paths. 616 * One is garbage collection and the other is SSR segment selection. 617 * When it is called during GC, it just gets a victim segment 618 * and it does not remove it from dirty seglist. 619 * When it is called from SSR segment selection, it finds a segment 620 * which has minimum valid blocks and removes it from dirty seglist. 621 */ 622 static int get_victim_by_default(struct f2fs_sb_info *sbi, 623 unsigned int *result, int gc_type, int type, 624 char alloc_mode, unsigned long long age) 625 { 626 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); 627 struct sit_info *sm = SIT_I(sbi); 628 struct victim_sel_policy p; 629 unsigned int secno, last_victim; 630 unsigned int last_segment; 631 unsigned int nsearched; 632 bool is_atgc; 633 int ret = 0; 634 635 mutex_lock(&dirty_i->seglist_lock); 636 last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec; 637 638 p.alloc_mode = alloc_mode; 639 p.age = age; 640 p.age_threshold = sbi->am.age_threshold; 641 642 retry: 643 select_policy(sbi, gc_type, type, &p); 644 p.min_segno = NULL_SEGNO; 645 p.oldest_age = 0; 646 p.min_cost = get_max_cost(sbi, &p); 647 648 is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR); 649 nsearched = 0; 650 651 if (is_atgc) 652 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX; 653 654 if (*result != NULL_SEGNO) { 655 if (!get_valid_blocks(sbi, *result, false)) { 656 ret = -ENODATA; 657 goto out; 658 } 659 660 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result))) 661 ret = -EBUSY; 662 else 663 p.min_segno = *result; 664 goto out; 665 } 666 667 ret = -ENODATA; 668 if (p.max_search == 0) 669 goto out; 670 671 if (__is_large_section(sbi) && p.alloc_mode == LFS) { 672 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) { 673 p.min_segno = sbi->next_victim_seg[BG_GC]; 674 *result = p.min_segno; 675 sbi->next_victim_seg[BG_GC] = NULL_SEGNO; 676 goto got_result; 677 } 678 if (gc_type == FG_GC && 679 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) { 680 p.min_segno = sbi->next_victim_seg[FG_GC]; 681 *result = p.min_segno; 682 sbi->next_victim_seg[FG_GC] = NULL_SEGNO; 683 goto got_result; 684 } 685 } 686 687 last_victim = sm->last_victim[p.gc_mode]; 688 if (p.alloc_mode == LFS && gc_type == FG_GC) { 689 p.min_segno = check_bg_victims(sbi); 690 if (p.min_segno != NULL_SEGNO) 691 goto got_it; 692 } 693 694 while (1) { 695 unsigned long cost, *dirty_bitmap; 696 unsigned int unit_no, segno; 697 698 dirty_bitmap = p.dirty_bitmap; 699 unit_no = find_next_bit(dirty_bitmap, 700 last_segment / p.ofs_unit, 701 p.offset / p.ofs_unit); 702 segno = unit_no * p.ofs_unit; 703 if (segno >= last_segment) { 704 if (sm->last_victim[p.gc_mode]) { 705 last_segment = 706 sm->last_victim[p.gc_mode]; 707 sm->last_victim[p.gc_mode] = 0; 708 p.offset = 0; 709 continue; 710 } 711 break; 712 } 713 714 p.offset = segno + p.ofs_unit; 715 nsearched++; 716 717 #ifdef CONFIG_F2FS_CHECK_FS 718 /* 719 * skip selecting the invalid segno (that is failed due to block 720 * validity check failure during GC) to avoid endless GC loop in 721 * such cases. 722 */ 723 if (test_bit(segno, sm->invalid_segmap)) 724 goto next; 725 #endif 726 727 secno = GET_SEC_FROM_SEG(sbi, segno); 728 729 if (sec_usage_check(sbi, secno)) 730 goto next; 731 /* Don't touch checkpointed data */ 732 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) && 733 get_ckpt_valid_blocks(sbi, segno) && 734 p.alloc_mode == LFS)) 735 goto next; 736 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) 737 goto next; 738 739 if (is_atgc) { 740 add_victim_entry(sbi, &p, segno); 741 goto next; 742 } 743 744 cost = get_gc_cost(sbi, segno, &p); 745 746 if (p.min_cost > cost) { 747 p.min_segno = segno; 748 p.min_cost = cost; 749 } 750 next: 751 if (nsearched >= p.max_search) { 752 if (!sm->last_victim[p.gc_mode] && segno <= last_victim) 753 sm->last_victim[p.gc_mode] = 754 last_victim + p.ofs_unit; 755 else 756 sm->last_victim[p.gc_mode] = segno + p.ofs_unit; 757 sm->last_victim[p.gc_mode] %= 758 (MAIN_SECS(sbi) * sbi->segs_per_sec); 759 break; 760 } 761 } 762 763 /* get victim for GC_AT/AT_SSR */ 764 if (is_atgc) { 765 lookup_victim_by_age(sbi, &p); 766 release_victim_entry(sbi); 767 } 768 769 if (is_atgc && p.min_segno == NULL_SEGNO && 770 sm->elapsed_time < p.age_threshold) { 771 p.age_threshold = 0; 772 goto retry; 773 } 774 775 if (p.min_segno != NULL_SEGNO) { 776 got_it: 777 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; 778 got_result: 779 if (p.alloc_mode == LFS) { 780 secno = GET_SEC_FROM_SEG(sbi, p.min_segno); 781 if (gc_type == FG_GC) 782 sbi->cur_victim_sec = secno; 783 else 784 set_bit(secno, dirty_i->victim_secmap); 785 } 786 ret = 0; 787 788 } 789 out: 790 if (p.min_segno != NULL_SEGNO) 791 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, 792 sbi->cur_victim_sec, 793 prefree_segments(sbi), free_segments(sbi)); 794 mutex_unlock(&dirty_i->seglist_lock); 795 796 return ret; 797 } 798 799 static const struct victim_selection default_v_ops = { 800 .get_victim = get_victim_by_default, 801 }; 802 803 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) 804 { 805 struct inode_entry *ie; 806 807 ie = radix_tree_lookup(&gc_list->iroot, ino); 808 if (ie) 809 return ie->inode; 810 return NULL; 811 } 812 813 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) 814 { 815 struct inode_entry *new_ie; 816 817 if (inode == find_gc_inode(gc_list, inode->i_ino)) { 818 iput(inode); 819 return; 820 } 821 new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS); 822 new_ie->inode = inode; 823 824 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); 825 list_add_tail(&new_ie->list, &gc_list->ilist); 826 } 827 828 static void put_gc_inode(struct gc_inode_list *gc_list) 829 { 830 struct inode_entry *ie, *next_ie; 831 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { 832 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); 833 iput(ie->inode); 834 list_del(&ie->list); 835 kmem_cache_free(f2fs_inode_entry_slab, ie); 836 } 837 } 838 839 static int check_valid_map(struct f2fs_sb_info *sbi, 840 unsigned int segno, int offset) 841 { 842 struct sit_info *sit_i = SIT_I(sbi); 843 struct seg_entry *sentry; 844 int ret; 845 846 down_read(&sit_i->sentry_lock); 847 sentry = get_seg_entry(sbi, segno); 848 ret = f2fs_test_bit(offset, sentry->cur_valid_map); 849 up_read(&sit_i->sentry_lock); 850 return ret; 851 } 852 853 /* 854 * This function compares node address got in summary with that in NAT. 855 * On validity, copy that node with cold status, otherwise (invalid node) 856 * ignore that. 857 */ 858 static int gc_node_segment(struct f2fs_sb_info *sbi, 859 struct f2fs_summary *sum, unsigned int segno, int gc_type) 860 { 861 struct f2fs_summary *entry; 862 block_t start_addr; 863 int off; 864 int phase = 0; 865 bool fggc = (gc_type == FG_GC); 866 int submitted = 0; 867 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 868 869 start_addr = START_BLOCK(sbi, segno); 870 871 next_step: 872 entry = sum; 873 874 if (fggc && phase == 2) 875 atomic_inc(&sbi->wb_sync_req[NODE]); 876 877 for (off = 0; off < usable_blks_in_seg; off++, entry++) { 878 nid_t nid = le32_to_cpu(entry->nid); 879 struct page *node_page; 880 struct node_info ni; 881 int err; 882 883 /* stop BG_GC if there is not enough free sections. */ 884 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) 885 return submitted; 886 887 if (check_valid_map(sbi, segno, off) == 0) 888 continue; 889 890 if (phase == 0) { 891 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, 892 META_NAT, true); 893 continue; 894 } 895 896 if (phase == 1) { 897 f2fs_ra_node_page(sbi, nid); 898 continue; 899 } 900 901 /* phase == 2 */ 902 node_page = f2fs_get_node_page(sbi, nid); 903 if (IS_ERR(node_page)) 904 continue; 905 906 /* block may become invalid during f2fs_get_node_page */ 907 if (check_valid_map(sbi, segno, off) == 0) { 908 f2fs_put_page(node_page, 1); 909 continue; 910 } 911 912 if (f2fs_get_node_info(sbi, nid, &ni)) { 913 f2fs_put_page(node_page, 1); 914 continue; 915 } 916 917 if (ni.blk_addr != start_addr + off) { 918 f2fs_put_page(node_page, 1); 919 continue; 920 } 921 922 err = f2fs_move_node_page(node_page, gc_type); 923 if (!err && gc_type == FG_GC) 924 submitted++; 925 stat_inc_node_blk_count(sbi, 1, gc_type); 926 } 927 928 if (++phase < 3) 929 goto next_step; 930 931 if (fggc) 932 atomic_dec(&sbi->wb_sync_req[NODE]); 933 return submitted; 934 } 935 936 /* 937 * Calculate start block index indicating the given node offset. 938 * Be careful, caller should give this node offset only indicating direct node 939 * blocks. If any node offsets, which point the other types of node blocks such 940 * as indirect or double indirect node blocks, are given, it must be a caller's 941 * bug. 942 */ 943 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode) 944 { 945 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; 946 unsigned int bidx; 947 948 if (node_ofs == 0) 949 return 0; 950 951 if (node_ofs <= 2) { 952 bidx = node_ofs - 1; 953 } else if (node_ofs <= indirect_blks) { 954 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); 955 bidx = node_ofs - 2 - dec; 956 } else { 957 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); 958 bidx = node_ofs - 5 - dec; 959 } 960 return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode); 961 } 962 963 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 964 struct node_info *dni, block_t blkaddr, unsigned int *nofs) 965 { 966 struct page *node_page; 967 nid_t nid; 968 unsigned int ofs_in_node; 969 block_t source_blkaddr; 970 971 nid = le32_to_cpu(sum->nid); 972 ofs_in_node = le16_to_cpu(sum->ofs_in_node); 973 974 node_page = f2fs_get_node_page(sbi, nid); 975 if (IS_ERR(node_page)) 976 return false; 977 978 if (f2fs_get_node_info(sbi, nid, dni)) { 979 f2fs_put_page(node_page, 1); 980 return false; 981 } 982 983 if (sum->version != dni->version) { 984 f2fs_warn(sbi, "%s: valid data with mismatched node version.", 985 __func__); 986 set_sbi_flag(sbi, SBI_NEED_FSCK); 987 } 988 989 *nofs = ofs_of_node(node_page); 990 source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node); 991 f2fs_put_page(node_page, 1); 992 993 if (source_blkaddr != blkaddr) { 994 #ifdef CONFIG_F2FS_CHECK_FS 995 unsigned int segno = GET_SEGNO(sbi, blkaddr); 996 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 997 998 if (unlikely(check_valid_map(sbi, segno, offset))) { 999 if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) { 1000 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u\n", 1001 blkaddr, source_blkaddr, segno); 1002 f2fs_bug_on(sbi, 1); 1003 } 1004 } 1005 #endif 1006 return false; 1007 } 1008 return true; 1009 } 1010 1011 static int ra_data_block(struct inode *inode, pgoff_t index) 1012 { 1013 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1014 struct address_space *mapping = inode->i_mapping; 1015 struct dnode_of_data dn; 1016 struct page *page; 1017 struct extent_info ei = {0, 0, 0}; 1018 struct f2fs_io_info fio = { 1019 .sbi = sbi, 1020 .ino = inode->i_ino, 1021 .type = DATA, 1022 .temp = COLD, 1023 .op = REQ_OP_READ, 1024 .op_flags = 0, 1025 .encrypted_page = NULL, 1026 .in_list = false, 1027 .retry = false, 1028 }; 1029 int err; 1030 1031 page = f2fs_grab_cache_page(mapping, index, true); 1032 if (!page) 1033 return -ENOMEM; 1034 1035 if (f2fs_lookup_extent_cache(inode, index, &ei)) { 1036 dn.data_blkaddr = ei.blk + index - ei.fofs; 1037 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, 1038 DATA_GENERIC_ENHANCE_READ))) { 1039 err = -EFSCORRUPTED; 1040 goto put_page; 1041 } 1042 goto got_it; 1043 } 1044 1045 set_new_dnode(&dn, inode, NULL, NULL, 0); 1046 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 1047 if (err) 1048 goto put_page; 1049 f2fs_put_dnode(&dn); 1050 1051 if (!__is_valid_data_blkaddr(dn.data_blkaddr)) { 1052 err = -ENOENT; 1053 goto put_page; 1054 } 1055 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, 1056 DATA_GENERIC_ENHANCE))) { 1057 err = -EFSCORRUPTED; 1058 goto put_page; 1059 } 1060 got_it: 1061 /* read page */ 1062 fio.page = page; 1063 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; 1064 1065 /* 1066 * don't cache encrypted data into meta inode until previous dirty 1067 * data were writebacked to avoid racing between GC and flush. 1068 */ 1069 f2fs_wait_on_page_writeback(page, DATA, true, true); 1070 1071 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); 1072 1073 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi), 1074 dn.data_blkaddr, 1075 FGP_LOCK | FGP_CREAT, GFP_NOFS); 1076 if (!fio.encrypted_page) { 1077 err = -ENOMEM; 1078 goto put_page; 1079 } 1080 1081 err = f2fs_submit_page_bio(&fio); 1082 if (err) 1083 goto put_encrypted_page; 1084 f2fs_put_page(fio.encrypted_page, 0); 1085 f2fs_put_page(page, 1); 1086 1087 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE); 1088 f2fs_update_iostat(sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE); 1089 1090 return 0; 1091 put_encrypted_page: 1092 f2fs_put_page(fio.encrypted_page, 1); 1093 put_page: 1094 f2fs_put_page(page, 1); 1095 return err; 1096 } 1097 1098 /* 1099 * Move data block via META_MAPPING while keeping locked data page. 1100 * This can be used to move blocks, aka LBAs, directly on disk. 1101 */ 1102 static int move_data_block(struct inode *inode, block_t bidx, 1103 int gc_type, unsigned int segno, int off) 1104 { 1105 struct f2fs_io_info fio = { 1106 .sbi = F2FS_I_SB(inode), 1107 .ino = inode->i_ino, 1108 .type = DATA, 1109 .temp = COLD, 1110 .op = REQ_OP_READ, 1111 .op_flags = 0, 1112 .encrypted_page = NULL, 1113 .in_list = false, 1114 .retry = false, 1115 }; 1116 struct dnode_of_data dn; 1117 struct f2fs_summary sum; 1118 struct node_info ni; 1119 struct page *page, *mpage; 1120 block_t newaddr; 1121 int err = 0; 1122 bool lfs_mode = f2fs_lfs_mode(fio.sbi); 1123 int type = fio.sbi->am.atgc_enabled ? 1124 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA; 1125 1126 /* do not read out */ 1127 page = f2fs_grab_cache_page(inode->i_mapping, bidx, false); 1128 if (!page) 1129 return -ENOMEM; 1130 1131 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) { 1132 err = -ENOENT; 1133 goto out; 1134 } 1135 1136 if (f2fs_is_atomic_file(inode)) { 1137 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++; 1138 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++; 1139 err = -EAGAIN; 1140 goto out; 1141 } 1142 1143 if (f2fs_is_pinned_file(inode)) { 1144 f2fs_pin_file_control(inode, true); 1145 err = -EAGAIN; 1146 goto out; 1147 } 1148 1149 set_new_dnode(&dn, inode, NULL, NULL, 0); 1150 err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE); 1151 if (err) 1152 goto out; 1153 1154 if (unlikely(dn.data_blkaddr == NULL_ADDR)) { 1155 ClearPageUptodate(page); 1156 err = -ENOENT; 1157 goto put_out; 1158 } 1159 1160 /* 1161 * don't cache encrypted data into meta inode until previous dirty 1162 * data were writebacked to avoid racing between GC and flush. 1163 */ 1164 f2fs_wait_on_page_writeback(page, DATA, true, true); 1165 1166 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); 1167 1168 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni); 1169 if (err) 1170 goto put_out; 1171 1172 /* read page */ 1173 fio.page = page; 1174 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; 1175 1176 if (lfs_mode) 1177 down_write(&fio.sbi->io_order_lock); 1178 1179 mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi), 1180 fio.old_blkaddr, false); 1181 if (!mpage) { 1182 err = -ENOMEM; 1183 goto up_out; 1184 } 1185 1186 fio.encrypted_page = mpage; 1187 1188 /* read source block in mpage */ 1189 if (!PageUptodate(mpage)) { 1190 err = f2fs_submit_page_bio(&fio); 1191 if (err) { 1192 f2fs_put_page(mpage, 1); 1193 goto up_out; 1194 } 1195 1196 f2fs_update_iostat(fio.sbi, FS_DATA_READ_IO, F2FS_BLKSIZE); 1197 f2fs_update_iostat(fio.sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE); 1198 1199 lock_page(mpage); 1200 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) || 1201 !PageUptodate(mpage))) { 1202 err = -EIO; 1203 f2fs_put_page(mpage, 1); 1204 goto up_out; 1205 } 1206 } 1207 1208 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); 1209 1210 /* allocate block address */ 1211 f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr, 1212 &sum, type, NULL); 1213 1214 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi), 1215 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS); 1216 if (!fio.encrypted_page) { 1217 err = -ENOMEM; 1218 f2fs_put_page(mpage, 1); 1219 goto recover_block; 1220 } 1221 1222 /* write target block */ 1223 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true); 1224 memcpy(page_address(fio.encrypted_page), 1225 page_address(mpage), PAGE_SIZE); 1226 f2fs_put_page(mpage, 1); 1227 invalidate_mapping_pages(META_MAPPING(fio.sbi), 1228 fio.old_blkaddr, fio.old_blkaddr); 1229 1230 set_page_dirty(fio.encrypted_page); 1231 if (clear_page_dirty_for_io(fio.encrypted_page)) 1232 dec_page_count(fio.sbi, F2FS_DIRTY_META); 1233 1234 set_page_writeback(fio.encrypted_page); 1235 ClearPageError(page); 1236 1237 fio.op = REQ_OP_WRITE; 1238 fio.op_flags = REQ_SYNC; 1239 fio.new_blkaddr = newaddr; 1240 f2fs_submit_page_write(&fio); 1241 if (fio.retry) { 1242 err = -EAGAIN; 1243 if (PageWriteback(fio.encrypted_page)) 1244 end_page_writeback(fio.encrypted_page); 1245 goto put_page_out; 1246 } 1247 1248 f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE); 1249 1250 f2fs_update_data_blkaddr(&dn, newaddr); 1251 set_inode_flag(inode, FI_APPEND_WRITE); 1252 if (page->index == 0) 1253 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); 1254 put_page_out: 1255 f2fs_put_page(fio.encrypted_page, 1); 1256 recover_block: 1257 if (err) 1258 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr, 1259 true, true, true); 1260 up_out: 1261 if (lfs_mode) 1262 up_write(&fio.sbi->io_order_lock); 1263 put_out: 1264 f2fs_put_dnode(&dn); 1265 out: 1266 f2fs_put_page(page, 1); 1267 return err; 1268 } 1269 1270 static int move_data_page(struct inode *inode, block_t bidx, int gc_type, 1271 unsigned int segno, int off) 1272 { 1273 struct page *page; 1274 int err = 0; 1275 1276 page = f2fs_get_lock_data_page(inode, bidx, true); 1277 if (IS_ERR(page)) 1278 return PTR_ERR(page); 1279 1280 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) { 1281 err = -ENOENT; 1282 goto out; 1283 } 1284 1285 if (f2fs_is_atomic_file(inode)) { 1286 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++; 1287 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++; 1288 err = -EAGAIN; 1289 goto out; 1290 } 1291 if (f2fs_is_pinned_file(inode)) { 1292 if (gc_type == FG_GC) 1293 f2fs_pin_file_control(inode, true); 1294 err = -EAGAIN; 1295 goto out; 1296 } 1297 1298 if (gc_type == BG_GC) { 1299 if (PageWriteback(page)) { 1300 err = -EAGAIN; 1301 goto out; 1302 } 1303 set_page_dirty(page); 1304 set_cold_data(page); 1305 } else { 1306 struct f2fs_io_info fio = { 1307 .sbi = F2FS_I_SB(inode), 1308 .ino = inode->i_ino, 1309 .type = DATA, 1310 .temp = COLD, 1311 .op = REQ_OP_WRITE, 1312 .op_flags = REQ_SYNC, 1313 .old_blkaddr = NULL_ADDR, 1314 .page = page, 1315 .encrypted_page = NULL, 1316 .need_lock = LOCK_REQ, 1317 .io_type = FS_GC_DATA_IO, 1318 }; 1319 bool is_dirty = PageDirty(page); 1320 1321 retry: 1322 f2fs_wait_on_page_writeback(page, DATA, true, true); 1323 1324 set_page_dirty(page); 1325 if (clear_page_dirty_for_io(page)) { 1326 inode_dec_dirty_pages(inode); 1327 f2fs_remove_dirty_inode(inode); 1328 } 1329 1330 set_cold_data(page); 1331 1332 err = f2fs_do_write_data_page(&fio); 1333 if (err) { 1334 clear_cold_data(page); 1335 if (err == -ENOMEM) { 1336 congestion_wait(BLK_RW_ASYNC, 1337 DEFAULT_IO_TIMEOUT); 1338 goto retry; 1339 } 1340 if (is_dirty) 1341 set_page_dirty(page); 1342 } 1343 } 1344 out: 1345 f2fs_put_page(page, 1); 1346 return err; 1347 } 1348 1349 /* 1350 * This function tries to get parent node of victim data block, and identifies 1351 * data block validity. If the block is valid, copy that with cold status and 1352 * modify parent node. 1353 * If the parent node is not valid or the data block address is different, 1354 * the victim data block is ignored. 1355 */ 1356 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, 1357 struct gc_inode_list *gc_list, unsigned int segno, int gc_type) 1358 { 1359 struct super_block *sb = sbi->sb; 1360 struct f2fs_summary *entry; 1361 block_t start_addr; 1362 int off; 1363 int phase = 0; 1364 int submitted = 0; 1365 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno); 1366 1367 start_addr = START_BLOCK(sbi, segno); 1368 1369 next_step: 1370 entry = sum; 1371 1372 for (off = 0; off < usable_blks_in_seg; off++, entry++) { 1373 struct page *data_page; 1374 struct inode *inode; 1375 struct node_info dni; /* dnode info for the data */ 1376 unsigned int ofs_in_node, nofs; 1377 block_t start_bidx; 1378 nid_t nid = le32_to_cpu(entry->nid); 1379 1380 /* 1381 * stop BG_GC if there is not enough free sections. 1382 * Or, stop GC if the segment becomes fully valid caused by 1383 * race condition along with SSR block allocation. 1384 */ 1385 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) || 1386 get_valid_blocks(sbi, segno, true) == 1387 BLKS_PER_SEC(sbi)) 1388 return submitted; 1389 1390 if (check_valid_map(sbi, segno, off) == 0) 1391 continue; 1392 1393 if (phase == 0) { 1394 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, 1395 META_NAT, true); 1396 continue; 1397 } 1398 1399 if (phase == 1) { 1400 f2fs_ra_node_page(sbi, nid); 1401 continue; 1402 } 1403 1404 /* Get an inode by ino with checking validity */ 1405 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) 1406 continue; 1407 1408 if (phase == 2) { 1409 f2fs_ra_node_page(sbi, dni.ino); 1410 continue; 1411 } 1412 1413 ofs_in_node = le16_to_cpu(entry->ofs_in_node); 1414 1415 if (phase == 3) { 1416 inode = f2fs_iget(sb, dni.ino); 1417 if (IS_ERR(inode) || is_bad_inode(inode)) { 1418 set_sbi_flag(sbi, SBI_NEED_FSCK); 1419 continue; 1420 } 1421 1422 if (!down_write_trylock( 1423 &F2FS_I(inode)->i_gc_rwsem[WRITE])) { 1424 iput(inode); 1425 sbi->skipped_gc_rwsem++; 1426 continue; 1427 } 1428 1429 start_bidx = f2fs_start_bidx_of_node(nofs, inode) + 1430 ofs_in_node; 1431 1432 if (f2fs_post_read_required(inode)) { 1433 int err = ra_data_block(inode, start_bidx); 1434 1435 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1436 if (err) { 1437 iput(inode); 1438 continue; 1439 } 1440 add_gc_inode(gc_list, inode); 1441 continue; 1442 } 1443 1444 data_page = f2fs_get_read_data_page(inode, 1445 start_bidx, REQ_RAHEAD, true); 1446 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1447 if (IS_ERR(data_page)) { 1448 iput(inode); 1449 continue; 1450 } 1451 1452 f2fs_put_page(data_page, 0); 1453 add_gc_inode(gc_list, inode); 1454 continue; 1455 } 1456 1457 /* phase 4 */ 1458 inode = find_gc_inode(gc_list, dni.ino); 1459 if (inode) { 1460 struct f2fs_inode_info *fi = F2FS_I(inode); 1461 bool locked = false; 1462 int err; 1463 1464 if (S_ISREG(inode->i_mode)) { 1465 if (!down_write_trylock(&fi->i_gc_rwsem[READ])) 1466 continue; 1467 if (!down_write_trylock( 1468 &fi->i_gc_rwsem[WRITE])) { 1469 sbi->skipped_gc_rwsem++; 1470 up_write(&fi->i_gc_rwsem[READ]); 1471 continue; 1472 } 1473 locked = true; 1474 1475 /* wait for all inflight aio data */ 1476 inode_dio_wait(inode); 1477 } 1478 1479 start_bidx = f2fs_start_bidx_of_node(nofs, inode) 1480 + ofs_in_node; 1481 if (f2fs_post_read_required(inode)) 1482 err = move_data_block(inode, start_bidx, 1483 gc_type, segno, off); 1484 else 1485 err = move_data_page(inode, start_bidx, gc_type, 1486 segno, off); 1487 1488 if (!err && (gc_type == FG_GC || 1489 f2fs_post_read_required(inode))) 1490 submitted++; 1491 1492 if (locked) { 1493 up_write(&fi->i_gc_rwsem[WRITE]); 1494 up_write(&fi->i_gc_rwsem[READ]); 1495 } 1496 1497 stat_inc_data_blk_count(sbi, 1, gc_type); 1498 } 1499 } 1500 1501 if (++phase < 5) 1502 goto next_step; 1503 1504 return submitted; 1505 } 1506 1507 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, 1508 int gc_type) 1509 { 1510 struct sit_info *sit_i = SIT_I(sbi); 1511 int ret; 1512 1513 down_write(&sit_i->sentry_lock); 1514 ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, 1515 NO_CHECK_TYPE, LFS, 0); 1516 up_write(&sit_i->sentry_lock); 1517 return ret; 1518 } 1519 1520 static int do_garbage_collect(struct f2fs_sb_info *sbi, 1521 unsigned int start_segno, 1522 struct gc_inode_list *gc_list, int gc_type) 1523 { 1524 struct page *sum_page; 1525 struct f2fs_summary_block *sum; 1526 struct blk_plug plug; 1527 unsigned int segno = start_segno; 1528 unsigned int end_segno = start_segno + sbi->segs_per_sec; 1529 int seg_freed = 0, migrated = 0; 1530 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ? 1531 SUM_TYPE_DATA : SUM_TYPE_NODE; 1532 int submitted = 0; 1533 1534 if (__is_large_section(sbi)) 1535 end_segno = rounddown(end_segno, sbi->segs_per_sec); 1536 1537 /* 1538 * zone-capacity can be less than zone-size in zoned devices, 1539 * resulting in less than expected usable segments in the zone, 1540 * calculate the end segno in the zone which can be garbage collected 1541 */ 1542 if (f2fs_sb_has_blkzoned(sbi)) 1543 end_segno -= sbi->segs_per_sec - 1544 f2fs_usable_segs_in_sec(sbi, segno); 1545 1546 sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type); 1547 1548 /* readahead multi ssa blocks those have contiguous address */ 1549 if (__is_large_section(sbi)) 1550 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), 1551 end_segno - segno, META_SSA, true); 1552 1553 /* reference all summary page */ 1554 while (segno < end_segno) { 1555 sum_page = f2fs_get_sum_page(sbi, segno++); 1556 if (IS_ERR(sum_page)) { 1557 int err = PTR_ERR(sum_page); 1558 1559 end_segno = segno - 1; 1560 for (segno = start_segno; segno < end_segno; segno++) { 1561 sum_page = find_get_page(META_MAPPING(sbi), 1562 GET_SUM_BLOCK(sbi, segno)); 1563 f2fs_put_page(sum_page, 0); 1564 f2fs_put_page(sum_page, 0); 1565 } 1566 return err; 1567 } 1568 unlock_page(sum_page); 1569 } 1570 1571 blk_start_plug(&plug); 1572 1573 for (segno = start_segno; segno < end_segno; segno++) { 1574 1575 /* find segment summary of victim */ 1576 sum_page = find_get_page(META_MAPPING(sbi), 1577 GET_SUM_BLOCK(sbi, segno)); 1578 f2fs_put_page(sum_page, 0); 1579 1580 if (get_valid_blocks(sbi, segno, false) == 0) 1581 goto freed; 1582 if (gc_type == BG_GC && __is_large_section(sbi) && 1583 migrated >= sbi->migration_granularity) 1584 goto skip; 1585 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi))) 1586 goto skip; 1587 1588 sum = page_address(sum_page); 1589 if (type != GET_SUM_TYPE((&sum->footer))) { 1590 f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT", 1591 segno, type, GET_SUM_TYPE((&sum->footer))); 1592 set_sbi_flag(sbi, SBI_NEED_FSCK); 1593 f2fs_stop_checkpoint(sbi, false); 1594 goto skip; 1595 } 1596 1597 /* 1598 * this is to avoid deadlock: 1599 * - lock_page(sum_page) - f2fs_replace_block 1600 * - check_valid_map() - down_write(sentry_lock) 1601 * - down_read(sentry_lock) - change_curseg() 1602 * - lock_page(sum_page) 1603 */ 1604 if (type == SUM_TYPE_NODE) 1605 submitted += gc_node_segment(sbi, sum->entries, segno, 1606 gc_type); 1607 else 1608 submitted += gc_data_segment(sbi, sum->entries, gc_list, 1609 segno, gc_type); 1610 1611 stat_inc_seg_count(sbi, type, gc_type); 1612 migrated++; 1613 1614 freed: 1615 if (gc_type == FG_GC && 1616 get_valid_blocks(sbi, segno, false) == 0) 1617 seg_freed++; 1618 1619 if (__is_large_section(sbi) && segno + 1 < end_segno) 1620 sbi->next_victim_seg[gc_type] = segno + 1; 1621 skip: 1622 f2fs_put_page(sum_page, 0); 1623 } 1624 1625 if (submitted) 1626 f2fs_submit_merged_write(sbi, 1627 (type == SUM_TYPE_NODE) ? NODE : DATA); 1628 1629 blk_finish_plug(&plug); 1630 1631 stat_inc_call_count(sbi->stat_info); 1632 1633 return seg_freed; 1634 } 1635 1636 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, 1637 bool background, unsigned int segno) 1638 { 1639 int gc_type = sync ? FG_GC : BG_GC; 1640 int sec_freed = 0, seg_freed = 0, total_freed = 0; 1641 int ret = 0; 1642 struct cp_control cpc; 1643 unsigned int init_segno = segno; 1644 struct gc_inode_list gc_list = { 1645 .ilist = LIST_HEAD_INIT(gc_list.ilist), 1646 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), 1647 }; 1648 unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC]; 1649 unsigned long long first_skipped; 1650 unsigned int skipped_round = 0, round = 0; 1651 1652 trace_f2fs_gc_begin(sbi->sb, sync, background, 1653 get_pages(sbi, F2FS_DIRTY_NODES), 1654 get_pages(sbi, F2FS_DIRTY_DENTS), 1655 get_pages(sbi, F2FS_DIRTY_IMETA), 1656 free_sections(sbi), 1657 free_segments(sbi), 1658 reserved_segments(sbi), 1659 prefree_segments(sbi)); 1660 1661 cpc.reason = __get_cp_reason(sbi); 1662 sbi->skipped_gc_rwsem = 0; 1663 first_skipped = last_skipped; 1664 gc_more: 1665 if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) { 1666 ret = -EINVAL; 1667 goto stop; 1668 } 1669 if (unlikely(f2fs_cp_error(sbi))) { 1670 ret = -EIO; 1671 goto stop; 1672 } 1673 1674 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) { 1675 /* 1676 * For example, if there are many prefree_segments below given 1677 * threshold, we can make them free by checkpoint. Then, we 1678 * secure free segments which doesn't need fggc any more. 1679 */ 1680 if (prefree_segments(sbi) && 1681 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { 1682 ret = f2fs_write_checkpoint(sbi, &cpc); 1683 if (ret) 1684 goto stop; 1685 } 1686 if (has_not_enough_free_secs(sbi, 0, 0)) 1687 gc_type = FG_GC; 1688 } 1689 1690 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */ 1691 if (gc_type == BG_GC && !background) { 1692 ret = -EINVAL; 1693 goto stop; 1694 } 1695 ret = __get_victim(sbi, &segno, gc_type); 1696 if (ret) 1697 goto stop; 1698 1699 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type); 1700 if (gc_type == FG_GC && 1701 seg_freed == f2fs_usable_segs_in_sec(sbi, segno)) 1702 sec_freed++; 1703 total_freed += seg_freed; 1704 1705 if (gc_type == FG_GC) { 1706 if (sbi->skipped_atomic_files[FG_GC] > last_skipped || 1707 sbi->skipped_gc_rwsem) 1708 skipped_round++; 1709 last_skipped = sbi->skipped_atomic_files[FG_GC]; 1710 round++; 1711 } 1712 1713 if (gc_type == FG_GC && seg_freed) 1714 sbi->cur_victim_sec = NULL_SEGNO; 1715 1716 if (sync) 1717 goto stop; 1718 1719 if (has_not_enough_free_secs(sbi, sec_freed, 0)) { 1720 if (skipped_round <= MAX_SKIP_GC_COUNT || 1721 skipped_round * 2 < round) { 1722 segno = NULL_SEGNO; 1723 goto gc_more; 1724 } 1725 1726 if (first_skipped < last_skipped && 1727 (last_skipped - first_skipped) > 1728 sbi->skipped_gc_rwsem) { 1729 f2fs_drop_inmem_pages_all(sbi, true); 1730 segno = NULL_SEGNO; 1731 goto gc_more; 1732 } 1733 if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) 1734 ret = f2fs_write_checkpoint(sbi, &cpc); 1735 } 1736 stop: 1737 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0; 1738 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno; 1739 1740 trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed, 1741 get_pages(sbi, F2FS_DIRTY_NODES), 1742 get_pages(sbi, F2FS_DIRTY_DENTS), 1743 get_pages(sbi, F2FS_DIRTY_IMETA), 1744 free_sections(sbi), 1745 free_segments(sbi), 1746 reserved_segments(sbi), 1747 prefree_segments(sbi)); 1748 1749 up_write(&sbi->gc_lock); 1750 1751 put_gc_inode(&gc_list); 1752 1753 if (sync && !ret) 1754 ret = sec_freed ? 0 : -EAGAIN; 1755 return ret; 1756 } 1757 1758 int __init f2fs_create_garbage_collection_cache(void) 1759 { 1760 victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry", 1761 sizeof(struct victim_entry)); 1762 if (!victim_entry_slab) 1763 return -ENOMEM; 1764 return 0; 1765 } 1766 1767 void f2fs_destroy_garbage_collection_cache(void) 1768 { 1769 kmem_cache_destroy(victim_entry_slab); 1770 } 1771 1772 static void init_atgc_management(struct f2fs_sb_info *sbi) 1773 { 1774 struct atgc_management *am = &sbi->am; 1775 1776 if (test_opt(sbi, ATGC) && 1777 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD) 1778 am->atgc_enabled = true; 1779 1780 am->root = RB_ROOT_CACHED; 1781 INIT_LIST_HEAD(&am->victim_list); 1782 am->victim_count = 0; 1783 1784 am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO; 1785 am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT; 1786 am->age_weight = DEF_GC_THREAD_AGE_WEIGHT; 1787 } 1788 1789 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi) 1790 { 1791 DIRTY_I(sbi)->v_ops = &default_v_ops; 1792 1793 sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES; 1794 1795 /* give warm/cold data area from slower device */ 1796 if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi)) 1797 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 1798 GET_SEGNO(sbi, FDEV(0).end_blk) + 1; 1799 1800 init_atgc_management(sbi); 1801 } 1802 1803 static int free_segment_range(struct f2fs_sb_info *sbi, 1804 unsigned int secs, bool gc_only) 1805 { 1806 unsigned int segno, next_inuse, start, end; 1807 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 }; 1808 int gc_mode, gc_type; 1809 int err = 0; 1810 int type; 1811 1812 /* Force block allocation for GC */ 1813 MAIN_SECS(sbi) -= secs; 1814 start = MAIN_SECS(sbi) * sbi->segs_per_sec; 1815 end = MAIN_SEGS(sbi) - 1; 1816 1817 mutex_lock(&DIRTY_I(sbi)->seglist_lock); 1818 for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++) 1819 if (SIT_I(sbi)->last_victim[gc_mode] >= start) 1820 SIT_I(sbi)->last_victim[gc_mode] = 0; 1821 1822 for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++) 1823 if (sbi->next_victim_seg[gc_type] >= start) 1824 sbi->next_victim_seg[gc_type] = NULL_SEGNO; 1825 mutex_unlock(&DIRTY_I(sbi)->seglist_lock); 1826 1827 /* Move out cursegs from the target range */ 1828 for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++) 1829 f2fs_allocate_segment_for_resize(sbi, type, start, end); 1830 1831 /* do GC to move out valid blocks in the range */ 1832 for (segno = start; segno <= end; segno += sbi->segs_per_sec) { 1833 struct gc_inode_list gc_list = { 1834 .ilist = LIST_HEAD_INIT(gc_list.ilist), 1835 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), 1836 }; 1837 1838 do_garbage_collect(sbi, segno, &gc_list, FG_GC); 1839 put_gc_inode(&gc_list); 1840 1841 if (!gc_only && get_valid_blocks(sbi, segno, true)) { 1842 err = -EAGAIN; 1843 goto out; 1844 } 1845 if (fatal_signal_pending(current)) { 1846 err = -ERESTARTSYS; 1847 goto out; 1848 } 1849 } 1850 if (gc_only) 1851 goto out; 1852 1853 err = f2fs_write_checkpoint(sbi, &cpc); 1854 if (err) 1855 goto out; 1856 1857 next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start); 1858 if (next_inuse <= end) { 1859 f2fs_err(sbi, "segno %u should be free but still inuse!", 1860 next_inuse); 1861 f2fs_bug_on(sbi, 1); 1862 } 1863 out: 1864 MAIN_SECS(sbi) += secs; 1865 return err; 1866 } 1867 1868 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs) 1869 { 1870 struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi); 1871 int section_count; 1872 int segment_count; 1873 int segment_count_main; 1874 long long block_count; 1875 int segs = secs * sbi->segs_per_sec; 1876 1877 down_write(&sbi->sb_lock); 1878 1879 section_count = le32_to_cpu(raw_sb->section_count); 1880 segment_count = le32_to_cpu(raw_sb->segment_count); 1881 segment_count_main = le32_to_cpu(raw_sb->segment_count_main); 1882 block_count = le64_to_cpu(raw_sb->block_count); 1883 1884 raw_sb->section_count = cpu_to_le32(section_count + secs); 1885 raw_sb->segment_count = cpu_to_le32(segment_count + segs); 1886 raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs); 1887 raw_sb->block_count = cpu_to_le64(block_count + 1888 (long long)segs * sbi->blocks_per_seg); 1889 if (f2fs_is_multi_device(sbi)) { 1890 int last_dev = sbi->s_ndevs - 1; 1891 int dev_segs = 1892 le32_to_cpu(raw_sb->devs[last_dev].total_segments); 1893 1894 raw_sb->devs[last_dev].total_segments = 1895 cpu_to_le32(dev_segs + segs); 1896 } 1897 1898 up_write(&sbi->sb_lock); 1899 } 1900 1901 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs) 1902 { 1903 int segs = secs * sbi->segs_per_sec; 1904 long long blks = (long long)segs * sbi->blocks_per_seg; 1905 long long user_block_count = 1906 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count); 1907 1908 SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs; 1909 MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs; 1910 MAIN_SECS(sbi) += secs; 1911 FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs; 1912 FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs; 1913 F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks); 1914 1915 if (f2fs_is_multi_device(sbi)) { 1916 int last_dev = sbi->s_ndevs - 1; 1917 1918 FDEV(last_dev).total_segments = 1919 (int)FDEV(last_dev).total_segments + segs; 1920 FDEV(last_dev).end_blk = 1921 (long long)FDEV(last_dev).end_blk + blks; 1922 #ifdef CONFIG_BLK_DEV_ZONED 1923 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz + 1924 (int)(blks >> sbi->log_blocks_per_blkz); 1925 #endif 1926 } 1927 } 1928 1929 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count) 1930 { 1931 __u64 old_block_count, shrunk_blocks; 1932 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 }; 1933 unsigned int secs; 1934 int err = 0; 1935 __u32 rem; 1936 1937 old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count); 1938 if (block_count > old_block_count) 1939 return -EINVAL; 1940 1941 if (f2fs_is_multi_device(sbi)) { 1942 int last_dev = sbi->s_ndevs - 1; 1943 __u64 last_segs = FDEV(last_dev).total_segments; 1944 1945 if (block_count + last_segs * sbi->blocks_per_seg <= 1946 old_block_count) 1947 return -EINVAL; 1948 } 1949 1950 /* new fs size should align to section size */ 1951 div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem); 1952 if (rem) 1953 return -EINVAL; 1954 1955 if (block_count == old_block_count) 1956 return 0; 1957 1958 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) { 1959 f2fs_err(sbi, "Should run fsck to repair first."); 1960 return -EFSCORRUPTED; 1961 } 1962 1963 if (test_opt(sbi, DISABLE_CHECKPOINT)) { 1964 f2fs_err(sbi, "Checkpoint should be enabled."); 1965 return -EINVAL; 1966 } 1967 1968 shrunk_blocks = old_block_count - block_count; 1969 secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi)); 1970 1971 /* stop other GC */ 1972 if (!down_write_trylock(&sbi->gc_lock)) 1973 return -EAGAIN; 1974 1975 /* stop CP to protect MAIN_SEC in free_segment_range */ 1976 f2fs_lock_op(sbi); 1977 err = free_segment_range(sbi, secs, true); 1978 f2fs_unlock_op(sbi); 1979 up_write(&sbi->gc_lock); 1980 if (err) 1981 return err; 1982 1983 set_sbi_flag(sbi, SBI_IS_RESIZEFS); 1984 1985 freeze_super(sbi->sb); 1986 down_write(&sbi->gc_lock); 1987 down_write(&sbi->cp_global_sem); 1988 1989 spin_lock(&sbi->stat_lock); 1990 if (shrunk_blocks + valid_user_blocks(sbi) + 1991 sbi->current_reserved_blocks + sbi->unusable_block_count + 1992 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count) 1993 err = -ENOSPC; 1994 else 1995 sbi->user_block_count -= shrunk_blocks; 1996 spin_unlock(&sbi->stat_lock); 1997 if (err) 1998 goto out_err; 1999 2000 err = free_segment_range(sbi, secs, false); 2001 if (err) 2002 goto recover_out; 2003 2004 update_sb_metadata(sbi, -secs); 2005 2006 err = f2fs_commit_super(sbi, false); 2007 if (err) { 2008 update_sb_metadata(sbi, secs); 2009 goto recover_out; 2010 } 2011 2012 update_fs_metadata(sbi, -secs); 2013 clear_sbi_flag(sbi, SBI_IS_RESIZEFS); 2014 set_sbi_flag(sbi, SBI_IS_DIRTY); 2015 2016 err = f2fs_write_checkpoint(sbi, &cpc); 2017 if (err) { 2018 update_fs_metadata(sbi, secs); 2019 update_sb_metadata(sbi, secs); 2020 f2fs_commit_super(sbi, false); 2021 } 2022 recover_out: 2023 if (err) { 2024 set_sbi_flag(sbi, SBI_NEED_FSCK); 2025 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!"); 2026 2027 spin_lock(&sbi->stat_lock); 2028 sbi->user_block_count += shrunk_blocks; 2029 spin_unlock(&sbi->stat_lock); 2030 } 2031 out_err: 2032 up_write(&sbi->cp_global_sem); 2033 up_write(&sbi->gc_lock); 2034 thaw_super(sbi->sb); 2035 clear_sbi_flag(sbi, SBI_IS_RESIZEFS); 2036 return err; 2037 } 2038