1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
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>
24 #include <trace/events/f2fs.h>
26 static struct kmem_cache *victim_entry_slab;
28 static unsigned int count_bits(const unsigned long *addr,
29 unsigned int offset, unsigned int len);
31 static int gc_thread_func(void *data)
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;
38 struct f2fs_gc_control gc_control = {
39 .victim_segno = NULL_SEGNO,
40 .should_migrate_blocks = false,
41 .err_gc_skipped = false };
43 wait_ms = gc_th->min_sleep_time;
47 bool sync_mode, foreground = false;
49 wait_event_interruptible_timeout(*wq,
50 kthread_should_stop() || freezing(current) ||
51 waitqueue_active(fggc_wq) ||
53 msecs_to_jiffies(wait_ms));
55 if (test_opt(sbi, GC_MERGE) && waitqueue_active(fggc_wq))
58 /* give it a try one time */
62 if (try_to_freeze()) {
63 stat_other_skip_bggc_count(sbi);
66 if (kthread_should_stop())
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);
75 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
76 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
77 f2fs_stop_checkpoint(sbi, false,
78 STOP_CP_REASON_FAULT_INJECT);
81 if (!sb_start_write_trylock(sbi->sb)) {
82 stat_other_skip_bggc_count(sbi);
87 * [GC triggering condition]
88 * 0. GC is not conducted currently.
89 * 1. There are enough dirty segments.
90 * 2. IO subsystem is idle by checking the # of writeback pages.
91 * 3. IO subsystem is idle by checking the # of requests in
92 * bdev's request list.
94 * Note) We have to avoid triggering GCs frequently.
95 * Because it is possible that some segments can be
96 * invalidated soon after by user update or deletion.
97 * So, I'd like to wait some time to collect dirty segments.
99 if (sbi->gc_mode == GC_URGENT_HIGH ||
100 sbi->gc_mode == GC_URGENT_MID) {
101 wait_ms = gc_th->urgent_sleep_time;
102 f2fs_down_write(&sbi->gc_lock);
107 f2fs_down_write(&sbi->gc_lock);
109 } else if (!f2fs_down_write_trylock(&sbi->gc_lock)) {
110 stat_other_skip_bggc_count(sbi);
114 if (!is_idle(sbi, GC_TIME)) {
115 increase_sleep_time(gc_th, &wait_ms);
116 f2fs_up_write(&sbi->gc_lock);
117 stat_io_skip_bggc_count(sbi);
121 if (has_enough_invalid_blocks(sbi))
122 decrease_sleep_time(gc_th, &wait_ms);
124 increase_sleep_time(gc_th, &wait_ms);
127 stat_inc_bggc_count(sbi->stat_info);
129 sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
131 /* foreground GC was been triggered via f2fs_balance_fs() */
135 gc_control.init_gc_type = sync_mode ? FG_GC : BG_GC;
136 gc_control.no_bg_gc = foreground;
137 gc_control.nr_free_secs = foreground ? 1 : 0;
139 /* if return value is not zero, no victim was selected */
140 if (f2fs_gc(sbi, &gc_control)) {
141 /* don't bother wait_ms by foreground gc */
143 wait_ms = gc_th->no_gc_sleep_time;
147 wake_up_all(&gc_th->fggc_wq);
149 trace_f2fs_background_gc(sbi->sb, wait_ms,
150 prefree_segments(sbi), free_segments(sbi));
152 /* balancing f2fs's metadata periodically */
153 f2fs_balance_fs_bg(sbi, true);
155 if (sbi->gc_mode == GC_URGENT_HIGH) {
156 spin_lock(&sbi->gc_urgent_high_lock);
157 if (sbi->gc_urgent_high_remaining) {
158 sbi->gc_urgent_high_remaining--;
159 if (!sbi->gc_urgent_high_remaining)
160 sbi->gc_mode = GC_NORMAL;
162 spin_unlock(&sbi->gc_urgent_high_lock);
164 sb_end_write(sbi->sb);
166 } while (!kthread_should_stop());
170 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
172 struct f2fs_gc_kthread *gc_th;
173 dev_t dev = sbi->sb->s_bdev->bd_dev;
176 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
182 gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
183 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
184 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
185 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
189 sbi->gc_thread = gc_th;
190 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
191 init_waitqueue_head(&sbi->gc_thread->fggc_wq);
192 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
193 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
194 if (IS_ERR(gc_th->f2fs_gc_task)) {
195 err = PTR_ERR(gc_th->f2fs_gc_task);
197 sbi->gc_thread = NULL;
203 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
205 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
209 kthread_stop(gc_th->f2fs_gc_task);
210 wake_up_all(&gc_th->fggc_wq);
212 sbi->gc_thread = NULL;
215 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
219 if (gc_type == BG_GC) {
220 if (sbi->am.atgc_enabled)
228 switch (sbi->gc_mode) {
244 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
245 int type, struct victim_sel_policy *p)
247 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
249 if (p->alloc_mode == SSR) {
250 p->gc_mode = GC_GREEDY;
251 p->dirty_bitmap = dirty_i->dirty_segmap[type];
252 p->max_search = dirty_i->nr_dirty[type];
254 } else if (p->alloc_mode == AT_SSR) {
255 p->gc_mode = GC_GREEDY;
256 p->dirty_bitmap = dirty_i->dirty_segmap[type];
257 p->max_search = dirty_i->nr_dirty[type];
260 p->gc_mode = select_gc_type(sbi, gc_type);
261 p->ofs_unit = sbi->segs_per_sec;
262 if (__is_large_section(sbi)) {
263 p->dirty_bitmap = dirty_i->dirty_secmap;
264 p->max_search = count_bits(p->dirty_bitmap,
267 p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
268 p->max_search = dirty_i->nr_dirty[DIRTY];
273 * adjust candidates range, should select all dirty segments for
274 * foreground GC and urgent GC cases.
276 if (gc_type != FG_GC &&
277 (sbi->gc_mode != GC_URGENT_HIGH) &&
278 (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
279 p->max_search > sbi->max_victim_search)
280 p->max_search = sbi->max_victim_search;
282 /* let's select beginning hot/small space first in no_heap mode*/
283 if (f2fs_need_rand_seg(sbi))
284 p->offset = prandom_u32_max(MAIN_SECS(sbi) * sbi->segs_per_sec);
285 else if (test_opt(sbi, NOHEAP) &&
286 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
289 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
292 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
293 struct victim_sel_policy *p)
295 /* SSR allocates in a segment unit */
296 if (p->alloc_mode == SSR)
297 return sbi->blocks_per_seg;
298 else if (p->alloc_mode == AT_SSR)
302 if (p->gc_mode == GC_GREEDY)
303 return 2 * sbi->blocks_per_seg * p->ofs_unit;
304 else if (p->gc_mode == GC_CB)
306 else if (p->gc_mode == GC_AT)
308 else /* No other gc_mode */
312 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
314 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
318 * If the gc_type is FG_GC, we can select victim segments
319 * selected by background GC before.
320 * Those segments guarantee they have small valid blocks.
322 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
323 if (sec_usage_check(sbi, secno))
325 clear_bit(secno, dirty_i->victim_secmap);
326 return GET_SEG_FROM_SEC(sbi, secno);
331 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
333 struct sit_info *sit_i = SIT_I(sbi);
334 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
335 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
336 unsigned long long mtime = 0;
337 unsigned int vblocks;
338 unsigned char age = 0;
341 unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
343 for (i = 0; i < usable_segs_per_sec; i++)
344 mtime += get_seg_entry(sbi, start + i)->mtime;
345 vblocks = get_valid_blocks(sbi, segno, true);
347 mtime = div_u64(mtime, usable_segs_per_sec);
348 vblocks = div_u64(vblocks, usable_segs_per_sec);
350 u = (vblocks * 100) >> sbi->log_blocks_per_seg;
352 /* Handle if the system time has changed by the user */
353 if (mtime < sit_i->min_mtime)
354 sit_i->min_mtime = mtime;
355 if (mtime > sit_i->max_mtime)
356 sit_i->max_mtime = mtime;
357 if (sit_i->max_mtime != sit_i->min_mtime)
358 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
359 sit_i->max_mtime - sit_i->min_mtime);
361 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
364 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
365 unsigned int segno, struct victim_sel_policy *p)
367 if (p->alloc_mode == SSR)
368 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
370 /* alloc_mode == LFS */
371 if (p->gc_mode == GC_GREEDY)
372 return get_valid_blocks(sbi, segno, true);
373 else if (p->gc_mode == GC_CB)
374 return get_cb_cost(sbi, segno);
380 static unsigned int count_bits(const unsigned long *addr,
381 unsigned int offset, unsigned int len)
383 unsigned int end = offset + len, sum = 0;
385 while (offset < end) {
386 if (test_bit(offset++, addr))
392 static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi,
393 unsigned long long mtime, unsigned int segno,
394 struct rb_node *parent, struct rb_node **p,
397 struct atgc_management *am = &sbi->am;
398 struct victim_entry *ve;
400 ve = f2fs_kmem_cache_alloc(victim_entry_slab,
401 GFP_NOFS, true, NULL);
406 rb_link_node(&ve->rb_node, parent, p);
407 rb_insert_color_cached(&ve->rb_node, &am->root, left_most);
409 list_add_tail(&ve->list, &am->victim_list);
416 static void insert_victim_entry(struct f2fs_sb_info *sbi,
417 unsigned long long mtime, unsigned int segno)
419 struct atgc_management *am = &sbi->am;
421 struct rb_node *parent = NULL;
422 bool left_most = true;
424 p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most);
425 attach_victim_entry(sbi, mtime, segno, parent, p, left_most);
428 static void add_victim_entry(struct f2fs_sb_info *sbi,
429 struct victim_sel_policy *p, unsigned int segno)
431 struct sit_info *sit_i = SIT_I(sbi);
432 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
433 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
434 unsigned long long mtime = 0;
437 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
438 if (p->gc_mode == GC_AT &&
439 get_valid_blocks(sbi, segno, true) == 0)
443 for (i = 0; i < sbi->segs_per_sec; i++)
444 mtime += get_seg_entry(sbi, start + i)->mtime;
445 mtime = div_u64(mtime, sbi->segs_per_sec);
447 /* Handle if the system time has changed by the user */
448 if (mtime < sit_i->min_mtime)
449 sit_i->min_mtime = mtime;
450 if (mtime > sit_i->max_mtime)
451 sit_i->max_mtime = mtime;
452 if (mtime < sit_i->dirty_min_mtime)
453 sit_i->dirty_min_mtime = mtime;
454 if (mtime > sit_i->dirty_max_mtime)
455 sit_i->dirty_max_mtime = mtime;
457 /* don't choose young section as candidate */
458 if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
461 insert_victim_entry(sbi, mtime, segno);
464 static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi,
465 struct victim_sel_policy *p)
467 struct atgc_management *am = &sbi->am;
468 struct rb_node *parent = NULL;
471 f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most);
476 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
477 struct victim_sel_policy *p)
479 struct sit_info *sit_i = SIT_I(sbi);
480 struct atgc_management *am = &sbi->am;
481 struct rb_root_cached *root = &am->root;
482 struct rb_node *node;
484 struct victim_entry *ve;
485 unsigned long long total_time;
486 unsigned long long age, u, accu;
487 unsigned long long max_mtime = sit_i->dirty_max_mtime;
488 unsigned long long min_mtime = sit_i->dirty_min_mtime;
489 unsigned int sec_blocks = CAP_BLKS_PER_SEC(sbi);
490 unsigned int vblocks;
491 unsigned int dirty_threshold = max(am->max_candidate_count,
492 am->candidate_ratio *
493 am->victim_count / 100);
494 unsigned int age_weight = am->age_weight;
496 unsigned int iter = 0;
498 if (max_mtime < min_mtime)
502 total_time = max_mtime - min_mtime;
504 accu = div64_u64(ULLONG_MAX, total_time);
505 accu = min_t(unsigned long long, div_u64(accu, 100),
506 DEFAULT_ACCURACY_CLASS);
508 node = rb_first_cached(root);
510 re = rb_entry_safe(node, struct rb_entry, rb_node);
514 ve = (struct victim_entry *)re;
516 if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
519 /* age = 10000 * x% * 60 */
520 age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
523 vblocks = get_valid_blocks(sbi, ve->segno, true);
524 f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
526 /* u = 10000 * x% * 40 */
527 u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
530 f2fs_bug_on(sbi, age + u >= UINT_MAX);
532 cost = UINT_MAX - (age + u);
535 if (cost < p->min_cost ||
536 (cost == p->min_cost && age > p->oldest_age)) {
539 p->min_segno = ve->segno;
542 if (iter < dirty_threshold) {
543 node = rb_next(node);
549 * select candidates around source section in range of
550 * [target - dirty_threshold, target + dirty_threshold]
552 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
553 struct victim_sel_policy *p)
555 struct sit_info *sit_i = SIT_I(sbi);
556 struct atgc_management *am = &sbi->am;
557 struct rb_node *node;
559 struct victim_entry *ve;
560 unsigned long long age;
561 unsigned long long max_mtime = sit_i->dirty_max_mtime;
562 unsigned long long min_mtime = sit_i->dirty_min_mtime;
563 unsigned int seg_blocks = sbi->blocks_per_seg;
564 unsigned int vblocks;
565 unsigned int dirty_threshold = max(am->max_candidate_count,
566 am->candidate_ratio *
567 am->victim_count / 100);
569 unsigned int iter = 0;
572 if (max_mtime < min_mtime)
576 node = lookup_central_victim(sbi, p);
578 re = rb_entry_safe(node, struct rb_entry, rb_node);
585 ve = (struct victim_entry *)re;
587 if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
590 age = max_mtime - ve->mtime;
592 vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
593 f2fs_bug_on(sbi, !vblocks);
596 if (vblocks == seg_blocks)
601 age = max_mtime - abs(p->age - age);
602 cost = UINT_MAX - vblocks;
604 if (cost < p->min_cost ||
605 (cost == p->min_cost && age > p->oldest_age)) {
608 p->min_segno = ve->segno;
611 if (iter < dirty_threshold) {
613 node = rb_prev(node);
615 node = rb_next(node);
625 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
626 struct victim_sel_policy *p)
628 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
629 &sbi->am.root, true));
631 if (p->gc_mode == GC_AT)
632 atgc_lookup_victim(sbi, p);
633 else if (p->alloc_mode == AT_SSR)
634 atssr_lookup_victim(sbi, p);
639 static void release_victim_entry(struct f2fs_sb_info *sbi)
641 struct atgc_management *am = &sbi->am;
642 struct victim_entry *ve, *tmp;
644 list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
646 kmem_cache_free(victim_entry_slab, ve);
650 am->root = RB_ROOT_CACHED;
652 f2fs_bug_on(sbi, am->victim_count);
653 f2fs_bug_on(sbi, !list_empty(&am->victim_list));
656 static bool f2fs_pin_section(struct f2fs_sb_info *sbi, unsigned int segno)
658 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
659 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
661 if (!dirty_i->enable_pin_section)
663 if (!test_and_set_bit(secno, dirty_i->pinned_secmap))
664 dirty_i->pinned_secmap_cnt++;
668 static bool f2fs_pinned_section_exists(struct dirty_seglist_info *dirty_i)
670 return dirty_i->pinned_secmap_cnt;
673 static bool f2fs_section_is_pinned(struct dirty_seglist_info *dirty_i,
676 return dirty_i->enable_pin_section &&
677 f2fs_pinned_section_exists(dirty_i) &&
678 test_bit(secno, dirty_i->pinned_secmap);
681 static void f2fs_unpin_all_sections(struct f2fs_sb_info *sbi, bool enable)
683 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
685 if (f2fs_pinned_section_exists(DIRTY_I(sbi))) {
686 memset(DIRTY_I(sbi)->pinned_secmap, 0, bitmap_size);
687 DIRTY_I(sbi)->pinned_secmap_cnt = 0;
689 DIRTY_I(sbi)->enable_pin_section = enable;
692 static int f2fs_gc_pinned_control(struct inode *inode, int gc_type,
695 if (!f2fs_is_pinned_file(inode))
697 if (gc_type != FG_GC)
699 if (!f2fs_pin_section(F2FS_I_SB(inode), segno))
700 f2fs_pin_file_control(inode, true);
705 * This function is called from two paths.
706 * One is garbage collection and the other is SSR segment selection.
707 * When it is called during GC, it just gets a victim segment
708 * and it does not remove it from dirty seglist.
709 * When it is called from SSR segment selection, it finds a segment
710 * which has minimum valid blocks and removes it from dirty seglist.
712 static int get_victim_by_default(struct f2fs_sb_info *sbi,
713 unsigned int *result, int gc_type, int type,
714 char alloc_mode, unsigned long long age)
716 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
717 struct sit_info *sm = SIT_I(sbi);
718 struct victim_sel_policy p;
719 unsigned int secno, last_victim;
720 unsigned int last_segment;
721 unsigned int nsearched;
725 mutex_lock(&dirty_i->seglist_lock);
726 last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
728 p.alloc_mode = alloc_mode;
730 p.age_threshold = sbi->am.age_threshold;
733 select_policy(sbi, gc_type, type, &p);
734 p.min_segno = NULL_SEGNO;
736 p.min_cost = get_max_cost(sbi, &p);
738 is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
742 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
744 if (*result != NULL_SEGNO) {
745 if (!get_valid_blocks(sbi, *result, false)) {
750 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
753 p.min_segno = *result;
758 if (p.max_search == 0)
761 if (__is_large_section(sbi) && p.alloc_mode == LFS) {
762 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
763 p.min_segno = sbi->next_victim_seg[BG_GC];
764 *result = p.min_segno;
765 sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
768 if (gc_type == FG_GC &&
769 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
770 p.min_segno = sbi->next_victim_seg[FG_GC];
771 *result = p.min_segno;
772 sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
777 last_victim = sm->last_victim[p.gc_mode];
778 if (p.alloc_mode == LFS && gc_type == FG_GC) {
779 p.min_segno = check_bg_victims(sbi);
780 if (p.min_segno != NULL_SEGNO)
785 unsigned long cost, *dirty_bitmap;
786 unsigned int unit_no, segno;
788 dirty_bitmap = p.dirty_bitmap;
789 unit_no = find_next_bit(dirty_bitmap,
790 last_segment / p.ofs_unit,
791 p.offset / p.ofs_unit);
792 segno = unit_no * p.ofs_unit;
793 if (segno >= last_segment) {
794 if (sm->last_victim[p.gc_mode]) {
796 sm->last_victim[p.gc_mode];
797 sm->last_victim[p.gc_mode] = 0;
804 p.offset = segno + p.ofs_unit;
807 #ifdef CONFIG_F2FS_CHECK_FS
809 * skip selecting the invalid segno (that is failed due to block
810 * validity check failure during GC) to avoid endless GC loop in
813 if (test_bit(segno, sm->invalid_segmap))
817 secno = GET_SEC_FROM_SEG(sbi, segno);
819 if (sec_usage_check(sbi, secno))
822 /* Don't touch checkpointed data */
823 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
824 if (p.alloc_mode == LFS) {
826 * LFS is set to find source section during GC.
827 * The victim should have no checkpointed data.
829 if (get_ckpt_valid_blocks(sbi, segno, true))
833 * SSR | AT_SSR are set to find target segment
834 * for writes which can be full by checkpointed
835 * and newly written blocks.
837 if (!f2fs_segment_has_free_slot(sbi, segno))
842 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
845 if (gc_type == FG_GC && f2fs_section_is_pinned(dirty_i, secno))
849 add_victim_entry(sbi, &p, segno);
853 cost = get_gc_cost(sbi, segno, &p);
855 if (p.min_cost > cost) {
860 if (nsearched >= p.max_search) {
861 if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
862 sm->last_victim[p.gc_mode] =
863 last_victim + p.ofs_unit;
865 sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
866 sm->last_victim[p.gc_mode] %=
867 (MAIN_SECS(sbi) * sbi->segs_per_sec);
872 /* get victim for GC_AT/AT_SSR */
874 lookup_victim_by_age(sbi, &p);
875 release_victim_entry(sbi);
878 if (is_atgc && p.min_segno == NULL_SEGNO &&
879 sm->elapsed_time < p.age_threshold) {
884 if (p.min_segno != NULL_SEGNO) {
886 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
888 if (p.alloc_mode == LFS) {
889 secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
890 if (gc_type == FG_GC)
891 sbi->cur_victim_sec = secno;
893 set_bit(secno, dirty_i->victim_secmap);
899 if (p.min_segno != NULL_SEGNO)
900 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
902 prefree_segments(sbi), free_segments(sbi));
903 mutex_unlock(&dirty_i->seglist_lock);
908 static const struct victim_selection default_v_ops = {
909 .get_victim = get_victim_by_default,
912 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
914 struct inode_entry *ie;
916 ie = radix_tree_lookup(&gc_list->iroot, ino);
922 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
924 struct inode_entry *new_ie;
926 if (inode == find_gc_inode(gc_list, inode->i_ino)) {
930 new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab,
931 GFP_NOFS, true, NULL);
932 new_ie->inode = inode;
934 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
935 list_add_tail(&new_ie->list, &gc_list->ilist);
938 static void put_gc_inode(struct gc_inode_list *gc_list)
940 struct inode_entry *ie, *next_ie;
942 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
943 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
946 kmem_cache_free(f2fs_inode_entry_slab, ie);
950 static int check_valid_map(struct f2fs_sb_info *sbi,
951 unsigned int segno, int offset)
953 struct sit_info *sit_i = SIT_I(sbi);
954 struct seg_entry *sentry;
957 down_read(&sit_i->sentry_lock);
958 sentry = get_seg_entry(sbi, segno);
959 ret = f2fs_test_bit(offset, sentry->cur_valid_map);
960 up_read(&sit_i->sentry_lock);
965 * This function compares node address got in summary with that in NAT.
966 * On validity, copy that node with cold status, otherwise (invalid node)
969 static int gc_node_segment(struct f2fs_sb_info *sbi,
970 struct f2fs_summary *sum, unsigned int segno, int gc_type)
972 struct f2fs_summary *entry;
976 bool fggc = (gc_type == FG_GC);
978 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
980 start_addr = START_BLOCK(sbi, segno);
985 if (fggc && phase == 2)
986 atomic_inc(&sbi->wb_sync_req[NODE]);
988 for (off = 0; off < usable_blks_in_seg; off++, entry++) {
989 nid_t nid = le32_to_cpu(entry->nid);
990 struct page *node_page;
994 /* stop BG_GC if there is not enough free sections. */
995 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
998 if (check_valid_map(sbi, segno, off) == 0)
1002 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1008 f2fs_ra_node_page(sbi, nid);
1013 node_page = f2fs_get_node_page(sbi, nid);
1014 if (IS_ERR(node_page))
1017 /* block may become invalid during f2fs_get_node_page */
1018 if (check_valid_map(sbi, segno, off) == 0) {
1019 f2fs_put_page(node_page, 1);
1023 if (f2fs_get_node_info(sbi, nid, &ni, false)) {
1024 f2fs_put_page(node_page, 1);
1028 if (ni.blk_addr != start_addr + off) {
1029 f2fs_put_page(node_page, 1);
1033 err = f2fs_move_node_page(node_page, gc_type);
1034 if (!err && gc_type == FG_GC)
1036 stat_inc_node_blk_count(sbi, 1, gc_type);
1043 atomic_dec(&sbi->wb_sync_req[NODE]);
1048 * Calculate start block index indicating the given node offset.
1049 * Be careful, caller should give this node offset only indicating direct node
1050 * blocks. If any node offsets, which point the other types of node blocks such
1051 * as indirect or double indirect node blocks, are given, it must be a caller's
1054 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
1056 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
1062 if (node_ofs <= 2) {
1063 bidx = node_ofs - 1;
1064 } else if (node_ofs <= indirect_blks) {
1065 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
1067 bidx = node_ofs - 2 - dec;
1069 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
1071 bidx = node_ofs - 5 - dec;
1073 return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
1076 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1077 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
1079 struct page *node_page;
1081 unsigned int ofs_in_node, max_addrs;
1082 block_t source_blkaddr;
1084 nid = le32_to_cpu(sum->nid);
1085 ofs_in_node = le16_to_cpu(sum->ofs_in_node);
1087 node_page = f2fs_get_node_page(sbi, nid);
1088 if (IS_ERR(node_page))
1091 if (f2fs_get_node_info(sbi, nid, dni, false)) {
1092 f2fs_put_page(node_page, 1);
1096 if (sum->version != dni->version) {
1097 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
1099 set_sbi_flag(sbi, SBI_NEED_FSCK);
1102 if (f2fs_check_nid_range(sbi, dni->ino)) {
1103 f2fs_put_page(node_page, 1);
1107 max_addrs = IS_INODE(node_page) ? DEF_ADDRS_PER_INODE :
1108 DEF_ADDRS_PER_BLOCK;
1109 if (ofs_in_node >= max_addrs) {
1110 f2fs_err(sbi, "Inconsistent ofs_in_node:%u in summary, ino:%u, nid:%u, max:%u",
1111 ofs_in_node, dni->ino, dni->nid, max_addrs);
1115 *nofs = ofs_of_node(node_page);
1116 source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1117 f2fs_put_page(node_page, 1);
1119 if (source_blkaddr != blkaddr) {
1120 #ifdef CONFIG_F2FS_CHECK_FS
1121 unsigned int segno = GET_SEGNO(sbi, blkaddr);
1122 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1124 if (unlikely(check_valid_map(sbi, segno, offset))) {
1125 if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1126 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",
1127 blkaddr, source_blkaddr, segno);
1128 set_sbi_flag(sbi, SBI_NEED_FSCK);
1137 static int ra_data_block(struct inode *inode, pgoff_t index)
1139 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1140 struct address_space *mapping = inode->i_mapping;
1141 struct dnode_of_data dn;
1143 struct extent_info ei = {0, 0, 0};
1144 struct f2fs_io_info fio = {
1146 .ino = inode->i_ino,
1151 .encrypted_page = NULL,
1157 page = f2fs_grab_cache_page(mapping, index, true);
1161 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1162 dn.data_blkaddr = ei.blk + index - ei.fofs;
1163 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1164 DATA_GENERIC_ENHANCE_READ))) {
1165 err = -EFSCORRUPTED;
1166 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1172 set_new_dnode(&dn, inode, NULL, NULL, 0);
1173 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1176 f2fs_put_dnode(&dn);
1178 if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1182 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1183 DATA_GENERIC_ENHANCE))) {
1184 err = -EFSCORRUPTED;
1185 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1191 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1194 * don't cache encrypted data into meta inode until previous dirty
1195 * data were writebacked to avoid racing between GC and flush.
1197 f2fs_wait_on_page_writeback(page, DATA, true, true);
1199 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1201 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1203 FGP_LOCK | FGP_CREAT, GFP_NOFS);
1204 if (!fio.encrypted_page) {
1209 err = f2fs_submit_page_bio(&fio);
1211 goto put_encrypted_page;
1212 f2fs_put_page(fio.encrypted_page, 0);
1213 f2fs_put_page(page, 1);
1215 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
1216 f2fs_update_iostat(sbi, NULL, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1220 f2fs_put_page(fio.encrypted_page, 1);
1222 f2fs_put_page(page, 1);
1227 * Move data block via META_MAPPING while keeping locked data page.
1228 * This can be used to move blocks, aka LBAs, directly on disk.
1230 static int move_data_block(struct inode *inode, block_t bidx,
1231 int gc_type, unsigned int segno, int off)
1233 struct f2fs_io_info fio = {
1234 .sbi = F2FS_I_SB(inode),
1235 .ino = inode->i_ino,
1240 .encrypted_page = NULL,
1244 struct dnode_of_data dn;
1245 struct f2fs_summary sum;
1246 struct node_info ni;
1247 struct page *page, *mpage;
1250 bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1251 int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&
1252 (fio.sbi->gc_mode != GC_URGENT_HIGH) ?
1253 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1255 /* do not read out */
1256 page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1260 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1265 err = f2fs_gc_pinned_control(inode, gc_type, segno);
1269 set_new_dnode(&dn, inode, NULL, NULL, 0);
1270 err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1274 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1275 ClearPageUptodate(page);
1281 * don't cache encrypted data into meta inode until previous dirty
1282 * data were writebacked to avoid racing between GC and flush.
1284 f2fs_wait_on_page_writeback(page, DATA, true, true);
1286 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1288 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);
1294 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1297 f2fs_down_write(&fio.sbi->io_order_lock);
1299 mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1300 fio.old_blkaddr, false);
1306 fio.encrypted_page = mpage;
1308 /* read source block in mpage */
1309 if (!PageUptodate(mpage)) {
1310 err = f2fs_submit_page_bio(&fio);
1312 f2fs_put_page(mpage, 1);
1316 f2fs_update_iostat(fio.sbi, inode, FS_DATA_READ_IO,
1318 f2fs_update_iostat(fio.sbi, NULL, FS_GDATA_READ_IO,
1322 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1323 !PageUptodate(mpage))) {
1325 f2fs_put_page(mpage, 1);
1330 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1332 /* allocate block address */
1333 f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1336 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1337 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1338 if (!fio.encrypted_page) {
1340 f2fs_put_page(mpage, 1);
1344 /* write target block */
1345 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1346 memcpy(page_address(fio.encrypted_page),
1347 page_address(mpage), PAGE_SIZE);
1348 f2fs_put_page(mpage, 1);
1349 invalidate_mapping_pages(META_MAPPING(fio.sbi),
1350 fio.old_blkaddr, fio.old_blkaddr);
1351 f2fs_invalidate_compress_page(fio.sbi, fio.old_blkaddr);
1353 set_page_dirty(fio.encrypted_page);
1354 if (clear_page_dirty_for_io(fio.encrypted_page))
1355 dec_page_count(fio.sbi, F2FS_DIRTY_META);
1357 set_page_writeback(fio.encrypted_page);
1358 ClearPageError(page);
1360 fio.op = REQ_OP_WRITE;
1361 fio.op_flags = REQ_SYNC;
1362 fio.new_blkaddr = newaddr;
1363 f2fs_submit_page_write(&fio);
1366 if (PageWriteback(fio.encrypted_page))
1367 end_page_writeback(fio.encrypted_page);
1371 f2fs_update_iostat(fio.sbi, NULL, FS_GC_DATA_IO, F2FS_BLKSIZE);
1373 f2fs_update_data_blkaddr(&dn, newaddr);
1374 set_inode_flag(inode, FI_APPEND_WRITE);
1375 if (page->index == 0)
1376 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1378 f2fs_put_page(fio.encrypted_page, 1);
1381 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1385 f2fs_up_write(&fio.sbi->io_order_lock);
1387 f2fs_put_dnode(&dn);
1389 f2fs_put_page(page, 1);
1393 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1394 unsigned int segno, int off)
1399 page = f2fs_get_lock_data_page(inode, bidx, true);
1401 return PTR_ERR(page);
1403 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1408 err = f2fs_gc_pinned_control(inode, gc_type, segno);
1412 if (gc_type == BG_GC) {
1413 if (PageWriteback(page)) {
1417 set_page_dirty(page);
1418 set_page_private_gcing(page);
1420 struct f2fs_io_info fio = {
1421 .sbi = F2FS_I_SB(inode),
1422 .ino = inode->i_ino,
1426 .op_flags = REQ_SYNC,
1427 .old_blkaddr = NULL_ADDR,
1429 .encrypted_page = NULL,
1430 .need_lock = LOCK_REQ,
1431 .io_type = FS_GC_DATA_IO,
1433 bool is_dirty = PageDirty(page);
1436 f2fs_wait_on_page_writeback(page, DATA, true, true);
1438 set_page_dirty(page);
1439 if (clear_page_dirty_for_io(page)) {
1440 inode_dec_dirty_pages(inode);
1441 f2fs_remove_dirty_inode(inode);
1444 set_page_private_gcing(page);
1446 err = f2fs_do_write_data_page(&fio);
1448 clear_page_private_gcing(page);
1449 if (err == -ENOMEM) {
1450 memalloc_retry_wait(GFP_NOFS);
1454 set_page_dirty(page);
1458 f2fs_put_page(page, 1);
1463 * This function tries to get parent node of victim data block, and identifies
1464 * data block validity. If the block is valid, copy that with cold status and
1465 * modify parent node.
1466 * If the parent node is not valid or the data block address is different,
1467 * the victim data block is ignored.
1469 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1470 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1473 struct super_block *sb = sbi->sb;
1474 struct f2fs_summary *entry;
1479 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1481 start_addr = START_BLOCK(sbi, segno);
1486 for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1487 struct page *data_page;
1488 struct inode *inode;
1489 struct node_info dni; /* dnode info for the data */
1490 unsigned int ofs_in_node, nofs;
1492 nid_t nid = le32_to_cpu(entry->nid);
1495 * stop BG_GC if there is not enough free sections.
1496 * Or, stop GC if the segment becomes fully valid caused by
1497 * race condition along with SSR block allocation.
1499 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1500 (!force_migrate && get_valid_blocks(sbi, segno, true) ==
1501 CAP_BLKS_PER_SEC(sbi)))
1504 if (check_valid_map(sbi, segno, off) == 0)
1508 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1514 f2fs_ra_node_page(sbi, nid);
1518 /* Get an inode by ino with checking validity */
1519 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1523 f2fs_ra_node_page(sbi, dni.ino);
1527 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1532 inode = f2fs_iget(sb, dni.ino);
1533 if (IS_ERR(inode) || is_bad_inode(inode) ||
1534 special_file(inode->i_mode))
1537 err = f2fs_gc_pinned_control(inode, gc_type, segno);
1538 if (err == -EAGAIN) {
1543 if (!f2fs_down_write_trylock(
1544 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1546 sbi->skipped_gc_rwsem++;
1550 start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1553 if (f2fs_post_read_required(inode)) {
1554 int err = ra_data_block(inode, start_bidx);
1556 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1561 add_gc_inode(gc_list, inode);
1565 data_page = f2fs_get_read_data_page(inode,
1566 start_bidx, REQ_RAHEAD, true);
1567 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1568 if (IS_ERR(data_page)) {
1573 f2fs_put_page(data_page, 0);
1574 add_gc_inode(gc_list, inode);
1579 inode = find_gc_inode(gc_list, dni.ino);
1581 struct f2fs_inode_info *fi = F2FS_I(inode);
1582 bool locked = false;
1585 if (S_ISREG(inode->i_mode)) {
1586 if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[READ])) {
1587 sbi->skipped_gc_rwsem++;
1590 if (!f2fs_down_write_trylock(
1591 &fi->i_gc_rwsem[WRITE])) {
1592 sbi->skipped_gc_rwsem++;
1593 f2fs_up_write(&fi->i_gc_rwsem[READ]);
1598 /* wait for all inflight aio data */
1599 inode_dio_wait(inode);
1602 start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1604 if (f2fs_post_read_required(inode))
1605 err = move_data_block(inode, start_bidx,
1606 gc_type, segno, off);
1608 err = move_data_page(inode, start_bidx, gc_type,
1611 if (!err && (gc_type == FG_GC ||
1612 f2fs_post_read_required(inode)))
1616 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
1617 f2fs_up_write(&fi->i_gc_rwsem[READ]);
1620 stat_inc_data_blk_count(sbi, 1, gc_type);
1630 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1633 struct sit_info *sit_i = SIT_I(sbi);
1636 down_write(&sit_i->sentry_lock);
1637 ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1638 NO_CHECK_TYPE, LFS, 0);
1639 up_write(&sit_i->sentry_lock);
1643 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1644 unsigned int start_segno,
1645 struct gc_inode_list *gc_list, int gc_type,
1648 struct page *sum_page;
1649 struct f2fs_summary_block *sum;
1650 struct blk_plug plug;
1651 unsigned int segno = start_segno;
1652 unsigned int end_segno = start_segno + sbi->segs_per_sec;
1653 int seg_freed = 0, migrated = 0;
1654 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1655 SUM_TYPE_DATA : SUM_TYPE_NODE;
1658 if (__is_large_section(sbi))
1659 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1662 * zone-capacity can be less than zone-size in zoned devices,
1663 * resulting in less than expected usable segments in the zone,
1664 * calculate the end segno in the zone which can be garbage collected
1666 if (f2fs_sb_has_blkzoned(sbi))
1667 end_segno -= sbi->segs_per_sec -
1668 f2fs_usable_segs_in_sec(sbi, segno);
1670 sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1672 /* readahead multi ssa blocks those have contiguous address */
1673 if (__is_large_section(sbi))
1674 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1675 end_segno - segno, META_SSA, true);
1677 /* reference all summary page */
1678 while (segno < end_segno) {
1679 sum_page = f2fs_get_sum_page(sbi, segno++);
1680 if (IS_ERR(sum_page)) {
1681 int err = PTR_ERR(sum_page);
1683 end_segno = segno - 1;
1684 for (segno = start_segno; segno < end_segno; segno++) {
1685 sum_page = find_get_page(META_MAPPING(sbi),
1686 GET_SUM_BLOCK(sbi, segno));
1687 f2fs_put_page(sum_page, 0);
1688 f2fs_put_page(sum_page, 0);
1692 unlock_page(sum_page);
1695 blk_start_plug(&plug);
1697 for (segno = start_segno; segno < end_segno; segno++) {
1699 /* find segment summary of victim */
1700 sum_page = find_get_page(META_MAPPING(sbi),
1701 GET_SUM_BLOCK(sbi, segno));
1702 f2fs_put_page(sum_page, 0);
1704 if (get_valid_blocks(sbi, segno, false) == 0)
1706 if (gc_type == BG_GC && __is_large_section(sbi) &&
1707 migrated >= sbi->migration_granularity)
1709 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1712 sum = page_address(sum_page);
1713 if (type != GET_SUM_TYPE((&sum->footer))) {
1714 f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1715 segno, type, GET_SUM_TYPE((&sum->footer)));
1716 set_sbi_flag(sbi, SBI_NEED_FSCK);
1717 f2fs_stop_checkpoint(sbi, false,
1718 STOP_CP_REASON_CORRUPTED_SUMMARY);
1723 * this is to avoid deadlock:
1724 * - lock_page(sum_page) - f2fs_replace_block
1725 * - check_valid_map() - down_write(sentry_lock)
1726 * - down_read(sentry_lock) - change_curseg()
1727 * - lock_page(sum_page)
1729 if (type == SUM_TYPE_NODE)
1730 submitted += gc_node_segment(sbi, sum->entries, segno,
1733 submitted += gc_data_segment(sbi, sum->entries, gc_list,
1737 stat_inc_seg_count(sbi, type, gc_type);
1738 sbi->gc_reclaimed_segs[sbi->gc_mode]++;
1742 if (gc_type == FG_GC &&
1743 get_valid_blocks(sbi, segno, false) == 0)
1746 if (__is_large_section(sbi) && segno + 1 < end_segno)
1747 sbi->next_victim_seg[gc_type] = segno + 1;
1749 f2fs_put_page(sum_page, 0);
1753 f2fs_submit_merged_write(sbi,
1754 (type == SUM_TYPE_NODE) ? NODE : DATA);
1756 blk_finish_plug(&plug);
1758 stat_inc_call_count(sbi->stat_info);
1763 int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control)
1765 int gc_type = gc_control->init_gc_type;
1766 unsigned int segno = gc_control->victim_segno;
1767 int sec_freed = 0, seg_freed = 0, total_freed = 0;
1769 struct cp_control cpc;
1770 struct gc_inode_list gc_list = {
1771 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1772 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1774 unsigned int skipped_round = 0, round = 0;
1776 trace_f2fs_gc_begin(sbi->sb, gc_type, gc_control->no_bg_gc,
1777 gc_control->nr_free_secs,
1778 get_pages(sbi, F2FS_DIRTY_NODES),
1779 get_pages(sbi, F2FS_DIRTY_DENTS),
1780 get_pages(sbi, F2FS_DIRTY_IMETA),
1783 reserved_segments(sbi),
1784 prefree_segments(sbi));
1786 cpc.reason = __get_cp_reason(sbi);
1787 sbi->skipped_gc_rwsem = 0;
1789 if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1793 if (unlikely(f2fs_cp_error(sbi))) {
1798 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1800 * For example, if there are many prefree_segments below given
1801 * threshold, we can make them free by checkpoint. Then, we
1802 * secure free segments which doesn't need fggc any more.
1804 if (prefree_segments(sbi)) {
1805 ret = f2fs_write_checkpoint(sbi, &cpc);
1809 if (has_not_enough_free_secs(sbi, 0, 0))
1813 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1814 if (gc_type == BG_GC && gc_control->no_bg_gc) {
1819 ret = __get_victim(sbi, &segno, gc_type);
1821 /* allow to search victim from sections has pinned data */
1822 if (ret == -ENODATA && gc_type == FG_GC &&
1823 f2fs_pinned_section_exists(DIRTY_I(sbi))) {
1824 f2fs_unpin_all_sections(sbi, false);
1830 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type,
1831 gc_control->should_migrate_blocks);
1832 total_freed += seg_freed;
1834 if (seg_freed == f2fs_usable_segs_in_sec(sbi, segno))
1837 if (gc_type == FG_GC)
1838 sbi->cur_victim_sec = NULL_SEGNO;
1840 if (gc_control->init_gc_type == FG_GC ||
1841 !has_not_enough_free_secs(sbi,
1842 (gc_type == FG_GC) ? sec_freed : 0, 0)) {
1843 if (gc_type == FG_GC && sec_freed < gc_control->nr_free_secs)
1848 /* FG_GC stops GC by skip_count */
1849 if (gc_type == FG_GC) {
1850 if (sbi->skipped_gc_rwsem)
1853 if (skipped_round > MAX_SKIP_GC_COUNT &&
1854 skipped_round * 2 >= round) {
1855 ret = f2fs_write_checkpoint(sbi, &cpc);
1860 /* Write checkpoint to reclaim prefree segments */
1861 if (free_sections(sbi) < NR_CURSEG_PERSIST_TYPE &&
1862 prefree_segments(sbi)) {
1863 ret = f2fs_write_checkpoint(sbi, &cpc);
1872 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1873 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno;
1875 if (gc_type == FG_GC)
1876 f2fs_unpin_all_sections(sbi, true);
1878 trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1879 get_pages(sbi, F2FS_DIRTY_NODES),
1880 get_pages(sbi, F2FS_DIRTY_DENTS),
1881 get_pages(sbi, F2FS_DIRTY_IMETA),
1884 reserved_segments(sbi),
1885 prefree_segments(sbi));
1887 f2fs_up_write(&sbi->gc_lock);
1889 put_gc_inode(&gc_list);
1891 if (gc_control->err_gc_skipped && !ret)
1892 ret = sec_freed ? 0 : -EAGAIN;
1896 int __init f2fs_create_garbage_collection_cache(void)
1898 victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1899 sizeof(struct victim_entry));
1900 if (!victim_entry_slab)
1905 void f2fs_destroy_garbage_collection_cache(void)
1907 kmem_cache_destroy(victim_entry_slab);
1910 static void init_atgc_management(struct f2fs_sb_info *sbi)
1912 struct atgc_management *am = &sbi->am;
1914 if (test_opt(sbi, ATGC) &&
1915 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1916 am->atgc_enabled = true;
1918 am->root = RB_ROOT_CACHED;
1919 INIT_LIST_HEAD(&am->victim_list);
1920 am->victim_count = 0;
1922 am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1923 am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1924 am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1925 am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
1928 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1930 DIRTY_I(sbi)->v_ops = &default_v_ops;
1932 sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1934 /* give warm/cold data area from slower device */
1935 if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1936 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1937 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1939 init_atgc_management(sbi);
1942 static int free_segment_range(struct f2fs_sb_info *sbi,
1943 unsigned int secs, bool gc_only)
1945 unsigned int segno, next_inuse, start, end;
1946 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1947 int gc_mode, gc_type;
1951 /* Force block allocation for GC */
1952 MAIN_SECS(sbi) -= secs;
1953 start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1954 end = MAIN_SEGS(sbi) - 1;
1956 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1957 for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1958 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1959 SIT_I(sbi)->last_victim[gc_mode] = 0;
1961 for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1962 if (sbi->next_victim_seg[gc_type] >= start)
1963 sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1964 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1966 /* Move out cursegs from the target range */
1967 for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1968 f2fs_allocate_segment_for_resize(sbi, type, start, end);
1970 /* do GC to move out valid blocks in the range */
1971 for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1972 struct gc_inode_list gc_list = {
1973 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1974 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1977 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
1978 put_gc_inode(&gc_list);
1980 if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1984 if (fatal_signal_pending(current)) {
1992 err = f2fs_write_checkpoint(sbi, &cpc);
1996 next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1997 if (next_inuse <= end) {
1998 f2fs_err(sbi, "segno %u should be free but still inuse!",
2000 f2fs_bug_on(sbi, 1);
2003 MAIN_SECS(sbi) += secs;
2007 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
2009 struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
2012 int segment_count_main;
2013 long long block_count;
2014 int segs = secs * sbi->segs_per_sec;
2016 f2fs_down_write(&sbi->sb_lock);
2018 section_count = le32_to_cpu(raw_sb->section_count);
2019 segment_count = le32_to_cpu(raw_sb->segment_count);
2020 segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
2021 block_count = le64_to_cpu(raw_sb->block_count);
2023 raw_sb->section_count = cpu_to_le32(section_count + secs);
2024 raw_sb->segment_count = cpu_to_le32(segment_count + segs);
2025 raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
2026 raw_sb->block_count = cpu_to_le64(block_count +
2027 (long long)segs * sbi->blocks_per_seg);
2028 if (f2fs_is_multi_device(sbi)) {
2029 int last_dev = sbi->s_ndevs - 1;
2031 le32_to_cpu(raw_sb->devs[last_dev].total_segments);
2033 raw_sb->devs[last_dev].total_segments =
2034 cpu_to_le32(dev_segs + segs);
2037 f2fs_up_write(&sbi->sb_lock);
2040 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
2042 int segs = secs * sbi->segs_per_sec;
2043 long long blks = (long long)segs * sbi->blocks_per_seg;
2044 long long user_block_count =
2045 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
2047 SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
2048 MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
2049 MAIN_SECS(sbi) += secs;
2050 FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
2051 FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
2052 F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
2054 if (f2fs_is_multi_device(sbi)) {
2055 int last_dev = sbi->s_ndevs - 1;
2057 FDEV(last_dev).total_segments =
2058 (int)FDEV(last_dev).total_segments + segs;
2059 FDEV(last_dev).end_blk =
2060 (long long)FDEV(last_dev).end_blk + blks;
2061 #ifdef CONFIG_BLK_DEV_ZONED
2062 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
2063 (int)(blks >> sbi->log_blocks_per_blkz);
2068 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
2070 __u64 old_block_count, shrunk_blocks;
2071 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
2076 old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
2077 if (block_count > old_block_count)
2080 if (f2fs_is_multi_device(sbi)) {
2081 int last_dev = sbi->s_ndevs - 1;
2082 __u64 last_segs = FDEV(last_dev).total_segments;
2084 if (block_count + last_segs * sbi->blocks_per_seg <=
2089 /* new fs size should align to section size */
2090 div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
2094 if (block_count == old_block_count)
2097 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2098 f2fs_err(sbi, "Should run fsck to repair first.");
2099 return -EFSCORRUPTED;
2102 if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2103 f2fs_err(sbi, "Checkpoint should be enabled.");
2107 shrunk_blocks = old_block_count - block_count;
2108 secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2111 if (!f2fs_down_write_trylock(&sbi->gc_lock))
2114 /* stop CP to protect MAIN_SEC in free_segment_range */
2117 spin_lock(&sbi->stat_lock);
2118 if (shrunk_blocks + valid_user_blocks(sbi) +
2119 sbi->current_reserved_blocks + sbi->unusable_block_count +
2120 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2122 spin_unlock(&sbi->stat_lock);
2127 err = free_segment_range(sbi, secs, true);
2130 f2fs_unlock_op(sbi);
2131 f2fs_up_write(&sbi->gc_lock);
2135 freeze_super(sbi->sb);
2136 f2fs_down_write(&sbi->gc_lock);
2137 f2fs_down_write(&sbi->cp_global_sem);
2139 spin_lock(&sbi->stat_lock);
2140 if (shrunk_blocks + valid_user_blocks(sbi) +
2141 sbi->current_reserved_blocks + sbi->unusable_block_count +
2142 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2145 sbi->user_block_count -= shrunk_blocks;
2146 spin_unlock(&sbi->stat_lock);
2150 set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2151 err = free_segment_range(sbi, secs, false);
2155 update_sb_metadata(sbi, -secs);
2157 err = f2fs_commit_super(sbi, false);
2159 update_sb_metadata(sbi, secs);
2163 update_fs_metadata(sbi, -secs);
2164 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2165 set_sbi_flag(sbi, SBI_IS_DIRTY);
2167 err = f2fs_write_checkpoint(sbi, &cpc);
2169 update_fs_metadata(sbi, secs);
2170 update_sb_metadata(sbi, secs);
2171 f2fs_commit_super(sbi, false);
2174 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2176 set_sbi_flag(sbi, SBI_NEED_FSCK);
2177 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2179 spin_lock(&sbi->stat_lock);
2180 sbi->user_block_count += shrunk_blocks;
2181 spin_unlock(&sbi->stat_lock);
2184 f2fs_up_write(&sbi->cp_global_sem);
2185 f2fs_up_write(&sbi->gc_lock);
2186 thaw_super(sbi->sb);