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;
145 /* reset wait_ms to default sleep time */
146 if (wait_ms == gc_th->no_gc_sleep_time)
147 wait_ms = gc_th->min_sleep_time;
151 wake_up_all(&gc_th->fggc_wq);
153 trace_f2fs_background_gc(sbi->sb, wait_ms,
154 prefree_segments(sbi), free_segments(sbi));
156 /* balancing f2fs's metadata periodically */
157 f2fs_balance_fs_bg(sbi, true);
159 if (sbi->gc_mode != GC_NORMAL) {
160 spin_lock(&sbi->gc_remaining_trials_lock);
161 if (sbi->gc_remaining_trials) {
162 sbi->gc_remaining_trials--;
163 if (!sbi->gc_remaining_trials)
164 sbi->gc_mode = GC_NORMAL;
166 spin_unlock(&sbi->gc_remaining_trials_lock);
168 sb_end_write(sbi->sb);
170 } while (!kthread_should_stop());
174 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
176 struct f2fs_gc_kthread *gc_th;
177 dev_t dev = sbi->sb->s_bdev->bd_dev;
179 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
183 gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
184 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
185 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
186 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
190 sbi->gc_thread = gc_th;
191 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
192 init_waitqueue_head(&sbi->gc_thread->fggc_wq);
193 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
194 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
195 if (IS_ERR(gc_th->f2fs_gc_task)) {
196 int err = PTR_ERR(gc_th->f2fs_gc_task);
199 sbi->gc_thread = NULL;
206 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
208 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
212 kthread_stop(gc_th->f2fs_gc_task);
213 wake_up_all(&gc_th->fggc_wq);
215 sbi->gc_thread = NULL;
218 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
222 if (gc_type == BG_GC) {
223 if (sbi->am.atgc_enabled)
231 switch (sbi->gc_mode) {
247 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
248 int type, struct victim_sel_policy *p)
250 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
252 if (p->alloc_mode == SSR) {
253 p->gc_mode = GC_GREEDY;
254 p->dirty_bitmap = dirty_i->dirty_segmap[type];
255 p->max_search = dirty_i->nr_dirty[type];
257 } else if (p->alloc_mode == AT_SSR) {
258 p->gc_mode = GC_GREEDY;
259 p->dirty_bitmap = dirty_i->dirty_segmap[type];
260 p->max_search = dirty_i->nr_dirty[type];
263 p->gc_mode = select_gc_type(sbi, gc_type);
264 p->ofs_unit = sbi->segs_per_sec;
265 if (__is_large_section(sbi)) {
266 p->dirty_bitmap = dirty_i->dirty_secmap;
267 p->max_search = count_bits(p->dirty_bitmap,
270 p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
271 p->max_search = dirty_i->nr_dirty[DIRTY];
276 * adjust candidates range, should select all dirty segments for
277 * foreground GC and urgent GC cases.
279 if (gc_type != FG_GC &&
280 (sbi->gc_mode != GC_URGENT_HIGH) &&
281 (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
282 p->max_search > sbi->max_victim_search)
283 p->max_search = sbi->max_victim_search;
285 /* let's select beginning hot/small space first in no_heap mode*/
286 if (f2fs_need_rand_seg(sbi))
287 p->offset = get_random_u32_below(MAIN_SECS(sbi) * sbi->segs_per_sec);
288 else if (test_opt(sbi, NOHEAP) &&
289 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
292 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
295 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
296 struct victim_sel_policy *p)
298 /* SSR allocates in a segment unit */
299 if (p->alloc_mode == SSR)
300 return sbi->blocks_per_seg;
301 else if (p->alloc_mode == AT_SSR)
305 if (p->gc_mode == GC_GREEDY)
306 return 2 * sbi->blocks_per_seg * p->ofs_unit;
307 else if (p->gc_mode == GC_CB)
309 else if (p->gc_mode == GC_AT)
311 else /* No other gc_mode */
315 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
317 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
321 * If the gc_type is FG_GC, we can select victim segments
322 * selected by background GC before.
323 * Those segments guarantee they have small valid blocks.
325 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
326 if (sec_usage_check(sbi, secno))
328 clear_bit(secno, dirty_i->victim_secmap);
329 return GET_SEG_FROM_SEC(sbi, secno);
334 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
336 struct sit_info *sit_i = SIT_I(sbi);
337 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
338 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
339 unsigned long long mtime = 0;
340 unsigned int vblocks;
341 unsigned char age = 0;
344 unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
346 for (i = 0; i < usable_segs_per_sec; i++)
347 mtime += get_seg_entry(sbi, start + i)->mtime;
348 vblocks = get_valid_blocks(sbi, segno, true);
350 mtime = div_u64(mtime, usable_segs_per_sec);
351 vblocks = div_u64(vblocks, usable_segs_per_sec);
353 u = (vblocks * 100) >> sbi->log_blocks_per_seg;
355 /* Handle if the system time has changed by the user */
356 if (mtime < sit_i->min_mtime)
357 sit_i->min_mtime = mtime;
358 if (mtime > sit_i->max_mtime)
359 sit_i->max_mtime = mtime;
360 if (sit_i->max_mtime != sit_i->min_mtime)
361 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
362 sit_i->max_mtime - sit_i->min_mtime);
364 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
367 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
368 unsigned int segno, struct victim_sel_policy *p)
370 if (p->alloc_mode == SSR)
371 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
373 /* alloc_mode == LFS */
374 if (p->gc_mode == GC_GREEDY)
375 return get_valid_blocks(sbi, segno, true);
376 else if (p->gc_mode == GC_CB)
377 return get_cb_cost(sbi, segno);
383 static unsigned int count_bits(const unsigned long *addr,
384 unsigned int offset, unsigned int len)
386 unsigned int end = offset + len, sum = 0;
388 while (offset < end) {
389 if (test_bit(offset++, addr))
395 static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi,
396 unsigned long long mtime, unsigned int segno,
397 struct rb_node *parent, struct rb_node **p,
400 struct atgc_management *am = &sbi->am;
401 struct victim_entry *ve;
403 ve = f2fs_kmem_cache_alloc(victim_entry_slab,
404 GFP_NOFS, true, NULL);
409 rb_link_node(&ve->rb_node, parent, p);
410 rb_insert_color_cached(&ve->rb_node, &am->root, left_most);
412 list_add_tail(&ve->list, &am->victim_list);
419 static void insert_victim_entry(struct f2fs_sb_info *sbi,
420 unsigned long long mtime, unsigned int segno)
422 struct atgc_management *am = &sbi->am;
424 struct rb_node *parent = NULL;
425 bool left_most = true;
427 p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most);
428 attach_victim_entry(sbi, mtime, segno, parent, p, left_most);
431 static void add_victim_entry(struct f2fs_sb_info *sbi,
432 struct victim_sel_policy *p, unsigned int segno)
434 struct sit_info *sit_i = SIT_I(sbi);
435 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
436 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
437 unsigned long long mtime = 0;
440 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
441 if (p->gc_mode == GC_AT &&
442 get_valid_blocks(sbi, segno, true) == 0)
446 for (i = 0; i < sbi->segs_per_sec; i++)
447 mtime += get_seg_entry(sbi, start + i)->mtime;
448 mtime = div_u64(mtime, sbi->segs_per_sec);
450 /* Handle if the system time has changed by the user */
451 if (mtime < sit_i->min_mtime)
452 sit_i->min_mtime = mtime;
453 if (mtime > sit_i->max_mtime)
454 sit_i->max_mtime = mtime;
455 if (mtime < sit_i->dirty_min_mtime)
456 sit_i->dirty_min_mtime = mtime;
457 if (mtime > sit_i->dirty_max_mtime)
458 sit_i->dirty_max_mtime = mtime;
460 /* don't choose young section as candidate */
461 if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
464 insert_victim_entry(sbi, mtime, segno);
467 static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi,
468 struct victim_sel_policy *p)
470 struct atgc_management *am = &sbi->am;
471 struct rb_node *parent = NULL;
474 f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most);
479 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
480 struct victim_sel_policy *p)
482 struct sit_info *sit_i = SIT_I(sbi);
483 struct atgc_management *am = &sbi->am;
484 struct rb_root_cached *root = &am->root;
485 struct rb_node *node;
487 struct victim_entry *ve;
488 unsigned long long total_time;
489 unsigned long long age, u, accu;
490 unsigned long long max_mtime = sit_i->dirty_max_mtime;
491 unsigned long long min_mtime = sit_i->dirty_min_mtime;
492 unsigned int sec_blocks = CAP_BLKS_PER_SEC(sbi);
493 unsigned int vblocks;
494 unsigned int dirty_threshold = max(am->max_candidate_count,
495 am->candidate_ratio *
496 am->victim_count / 100);
497 unsigned int age_weight = am->age_weight;
499 unsigned int iter = 0;
501 if (max_mtime < min_mtime)
505 total_time = max_mtime - min_mtime;
507 accu = div64_u64(ULLONG_MAX, total_time);
508 accu = min_t(unsigned long long, div_u64(accu, 100),
509 DEFAULT_ACCURACY_CLASS);
511 node = rb_first_cached(root);
513 re = rb_entry_safe(node, struct rb_entry, rb_node);
517 ve = (struct victim_entry *)re;
519 if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
522 /* age = 10000 * x% * 60 */
523 age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
526 vblocks = get_valid_blocks(sbi, ve->segno, true);
527 f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
529 /* u = 10000 * x% * 40 */
530 u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
533 f2fs_bug_on(sbi, age + u >= UINT_MAX);
535 cost = UINT_MAX - (age + u);
538 if (cost < p->min_cost ||
539 (cost == p->min_cost && age > p->oldest_age)) {
542 p->min_segno = ve->segno;
545 if (iter < dirty_threshold) {
546 node = rb_next(node);
552 * select candidates around source section in range of
553 * [target - dirty_threshold, target + dirty_threshold]
555 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
556 struct victim_sel_policy *p)
558 struct sit_info *sit_i = SIT_I(sbi);
559 struct atgc_management *am = &sbi->am;
560 struct rb_node *node;
562 struct victim_entry *ve;
563 unsigned long long age;
564 unsigned long long max_mtime = sit_i->dirty_max_mtime;
565 unsigned long long min_mtime = sit_i->dirty_min_mtime;
566 unsigned int seg_blocks = sbi->blocks_per_seg;
567 unsigned int vblocks;
568 unsigned int dirty_threshold = max(am->max_candidate_count,
569 am->candidate_ratio *
570 am->victim_count / 100);
572 unsigned int iter = 0;
575 if (max_mtime < min_mtime)
579 node = lookup_central_victim(sbi, p);
581 re = rb_entry_safe(node, struct rb_entry, rb_node);
588 ve = (struct victim_entry *)re;
590 if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
593 age = max_mtime - ve->mtime;
595 vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
596 f2fs_bug_on(sbi, !vblocks);
599 if (vblocks == seg_blocks)
604 age = max_mtime - abs(p->age - age);
605 cost = UINT_MAX - vblocks;
607 if (cost < p->min_cost ||
608 (cost == p->min_cost && age > p->oldest_age)) {
611 p->min_segno = ve->segno;
614 if (iter < dirty_threshold) {
616 node = rb_prev(node);
618 node = rb_next(node);
628 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
629 struct victim_sel_policy *p)
631 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
632 &sbi->am.root, true));
634 if (p->gc_mode == GC_AT)
635 atgc_lookup_victim(sbi, p);
636 else if (p->alloc_mode == AT_SSR)
637 atssr_lookup_victim(sbi, p);
642 static void release_victim_entry(struct f2fs_sb_info *sbi)
644 struct atgc_management *am = &sbi->am;
645 struct victim_entry *ve, *tmp;
647 list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
649 kmem_cache_free(victim_entry_slab, ve);
653 am->root = RB_ROOT_CACHED;
655 f2fs_bug_on(sbi, am->victim_count);
656 f2fs_bug_on(sbi, !list_empty(&am->victim_list));
659 static bool f2fs_pin_section(struct f2fs_sb_info *sbi, unsigned int segno)
661 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
662 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
664 if (!dirty_i->enable_pin_section)
666 if (!test_and_set_bit(secno, dirty_i->pinned_secmap))
667 dirty_i->pinned_secmap_cnt++;
671 static bool f2fs_pinned_section_exists(struct dirty_seglist_info *dirty_i)
673 return dirty_i->pinned_secmap_cnt;
676 static bool f2fs_section_is_pinned(struct dirty_seglist_info *dirty_i,
679 return dirty_i->enable_pin_section &&
680 f2fs_pinned_section_exists(dirty_i) &&
681 test_bit(secno, dirty_i->pinned_secmap);
684 static void f2fs_unpin_all_sections(struct f2fs_sb_info *sbi, bool enable)
686 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
688 if (f2fs_pinned_section_exists(DIRTY_I(sbi))) {
689 memset(DIRTY_I(sbi)->pinned_secmap, 0, bitmap_size);
690 DIRTY_I(sbi)->pinned_secmap_cnt = 0;
692 DIRTY_I(sbi)->enable_pin_section = enable;
695 static int f2fs_gc_pinned_control(struct inode *inode, int gc_type,
698 if (!f2fs_is_pinned_file(inode))
700 if (gc_type != FG_GC)
702 if (!f2fs_pin_section(F2FS_I_SB(inode), segno))
703 f2fs_pin_file_control(inode, true);
708 * This function is called from two paths.
709 * One is garbage collection and the other is SSR segment selection.
710 * When it is called during GC, it just gets a victim segment
711 * and it does not remove it from dirty seglist.
712 * When it is called from SSR segment selection, it finds a segment
713 * which has minimum valid blocks and removes it from dirty seglist.
715 static int get_victim_by_default(struct f2fs_sb_info *sbi,
716 unsigned int *result, int gc_type, int type,
717 char alloc_mode, unsigned long long age)
719 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
720 struct sit_info *sm = SIT_I(sbi);
721 struct victim_sel_policy p;
722 unsigned int secno, last_victim;
723 unsigned int last_segment;
724 unsigned int nsearched;
728 mutex_lock(&dirty_i->seglist_lock);
729 last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
731 p.alloc_mode = alloc_mode;
733 p.age_threshold = sbi->am.age_threshold;
736 select_policy(sbi, gc_type, type, &p);
737 p.min_segno = NULL_SEGNO;
739 p.min_cost = get_max_cost(sbi, &p);
741 is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
745 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
747 if (*result != NULL_SEGNO) {
748 if (!get_valid_blocks(sbi, *result, false)) {
753 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
756 p.min_segno = *result;
761 if (p.max_search == 0)
764 if (__is_large_section(sbi) && p.alloc_mode == LFS) {
765 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
766 p.min_segno = sbi->next_victim_seg[BG_GC];
767 *result = p.min_segno;
768 sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
771 if (gc_type == FG_GC &&
772 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
773 p.min_segno = sbi->next_victim_seg[FG_GC];
774 *result = p.min_segno;
775 sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
780 last_victim = sm->last_victim[p.gc_mode];
781 if (p.alloc_mode == LFS && gc_type == FG_GC) {
782 p.min_segno = check_bg_victims(sbi);
783 if (p.min_segno != NULL_SEGNO)
788 unsigned long cost, *dirty_bitmap;
789 unsigned int unit_no, segno;
791 dirty_bitmap = p.dirty_bitmap;
792 unit_no = find_next_bit(dirty_bitmap,
793 last_segment / p.ofs_unit,
794 p.offset / p.ofs_unit);
795 segno = unit_no * p.ofs_unit;
796 if (segno >= last_segment) {
797 if (sm->last_victim[p.gc_mode]) {
799 sm->last_victim[p.gc_mode];
800 sm->last_victim[p.gc_mode] = 0;
807 p.offset = segno + p.ofs_unit;
810 #ifdef CONFIG_F2FS_CHECK_FS
812 * skip selecting the invalid segno (that is failed due to block
813 * validity check failure during GC) to avoid endless GC loop in
816 if (test_bit(segno, sm->invalid_segmap))
820 secno = GET_SEC_FROM_SEG(sbi, segno);
822 if (sec_usage_check(sbi, secno))
825 /* Don't touch checkpointed data */
826 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
827 if (p.alloc_mode == LFS) {
829 * LFS is set to find source section during GC.
830 * The victim should have no checkpointed data.
832 if (get_ckpt_valid_blocks(sbi, segno, true))
836 * SSR | AT_SSR are set to find target segment
837 * for writes which can be full by checkpointed
838 * and newly written blocks.
840 if (!f2fs_segment_has_free_slot(sbi, segno))
845 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
848 if (gc_type == FG_GC && f2fs_section_is_pinned(dirty_i, secno))
852 add_victim_entry(sbi, &p, segno);
856 cost = get_gc_cost(sbi, segno, &p);
858 if (p.min_cost > cost) {
863 if (nsearched >= p.max_search) {
864 if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
865 sm->last_victim[p.gc_mode] =
866 last_victim + p.ofs_unit;
868 sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
869 sm->last_victim[p.gc_mode] %=
870 (MAIN_SECS(sbi) * sbi->segs_per_sec);
875 /* get victim for GC_AT/AT_SSR */
877 lookup_victim_by_age(sbi, &p);
878 release_victim_entry(sbi);
881 if (is_atgc && p.min_segno == NULL_SEGNO &&
882 sm->elapsed_time < p.age_threshold) {
887 if (p.min_segno != NULL_SEGNO) {
889 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
891 if (p.alloc_mode == LFS) {
892 secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
893 if (gc_type == FG_GC)
894 sbi->cur_victim_sec = secno;
896 set_bit(secno, dirty_i->victim_secmap);
902 if (p.min_segno != NULL_SEGNO)
903 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
905 prefree_segments(sbi), free_segments(sbi));
906 mutex_unlock(&dirty_i->seglist_lock);
911 static const struct victim_selection default_v_ops = {
912 .get_victim = get_victim_by_default,
915 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
917 struct inode_entry *ie;
919 ie = radix_tree_lookup(&gc_list->iroot, ino);
925 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
927 struct inode_entry *new_ie;
929 if (inode == find_gc_inode(gc_list, inode->i_ino)) {
933 new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab,
934 GFP_NOFS, true, NULL);
935 new_ie->inode = inode;
937 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
938 list_add_tail(&new_ie->list, &gc_list->ilist);
941 static void put_gc_inode(struct gc_inode_list *gc_list)
943 struct inode_entry *ie, *next_ie;
945 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
946 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
949 kmem_cache_free(f2fs_inode_entry_slab, ie);
953 static int check_valid_map(struct f2fs_sb_info *sbi,
954 unsigned int segno, int offset)
956 struct sit_info *sit_i = SIT_I(sbi);
957 struct seg_entry *sentry;
960 down_read(&sit_i->sentry_lock);
961 sentry = get_seg_entry(sbi, segno);
962 ret = f2fs_test_bit(offset, sentry->cur_valid_map);
963 up_read(&sit_i->sentry_lock);
968 * This function compares node address got in summary with that in NAT.
969 * On validity, copy that node with cold status, otherwise (invalid node)
972 static int gc_node_segment(struct f2fs_sb_info *sbi,
973 struct f2fs_summary *sum, unsigned int segno, int gc_type)
975 struct f2fs_summary *entry;
979 bool fggc = (gc_type == FG_GC);
981 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
983 start_addr = START_BLOCK(sbi, segno);
988 if (fggc && phase == 2)
989 atomic_inc(&sbi->wb_sync_req[NODE]);
991 for (off = 0; off < usable_blks_in_seg; off++, entry++) {
992 nid_t nid = le32_to_cpu(entry->nid);
993 struct page *node_page;
997 /* stop BG_GC if there is not enough free sections. */
998 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
1001 if (check_valid_map(sbi, segno, off) == 0)
1005 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1011 f2fs_ra_node_page(sbi, nid);
1016 node_page = f2fs_get_node_page(sbi, nid);
1017 if (IS_ERR(node_page))
1020 /* block may become invalid during f2fs_get_node_page */
1021 if (check_valid_map(sbi, segno, off) == 0) {
1022 f2fs_put_page(node_page, 1);
1026 if (f2fs_get_node_info(sbi, nid, &ni, false)) {
1027 f2fs_put_page(node_page, 1);
1031 if (ni.blk_addr != start_addr + off) {
1032 f2fs_put_page(node_page, 1);
1036 err = f2fs_move_node_page(node_page, gc_type);
1037 if (!err && gc_type == FG_GC)
1039 stat_inc_node_blk_count(sbi, 1, gc_type);
1046 atomic_dec(&sbi->wb_sync_req[NODE]);
1051 * Calculate start block index indicating the given node offset.
1052 * Be careful, caller should give this node offset only indicating direct node
1053 * blocks. If any node offsets, which point the other types of node blocks such
1054 * as indirect or double indirect node blocks, are given, it must be a caller's
1057 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
1059 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
1065 if (node_ofs <= 2) {
1066 bidx = node_ofs - 1;
1067 } else if (node_ofs <= indirect_blks) {
1068 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
1070 bidx = node_ofs - 2 - dec;
1072 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
1074 bidx = node_ofs - 5 - dec;
1076 return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
1079 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1080 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
1082 struct page *node_page;
1084 unsigned int ofs_in_node, max_addrs, base;
1085 block_t source_blkaddr;
1087 nid = le32_to_cpu(sum->nid);
1088 ofs_in_node = le16_to_cpu(sum->ofs_in_node);
1090 node_page = f2fs_get_node_page(sbi, nid);
1091 if (IS_ERR(node_page))
1094 if (f2fs_get_node_info(sbi, nid, dni, false)) {
1095 f2fs_put_page(node_page, 1);
1099 if (sum->version != dni->version) {
1100 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
1102 set_sbi_flag(sbi, SBI_NEED_FSCK);
1105 if (f2fs_check_nid_range(sbi, dni->ino)) {
1106 f2fs_put_page(node_page, 1);
1110 if (IS_INODE(node_page)) {
1111 base = offset_in_addr(F2FS_INODE(node_page));
1112 max_addrs = DEF_ADDRS_PER_INODE;
1115 max_addrs = DEF_ADDRS_PER_BLOCK;
1118 if (base + ofs_in_node >= max_addrs) {
1119 f2fs_err(sbi, "Inconsistent blkaddr offset: base:%u, ofs_in_node:%u, max:%u, ino:%u, nid:%u",
1120 base, ofs_in_node, max_addrs, dni->ino, dni->nid);
1121 f2fs_put_page(node_page, 1);
1125 *nofs = ofs_of_node(node_page);
1126 source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1127 f2fs_put_page(node_page, 1);
1129 if (source_blkaddr != blkaddr) {
1130 #ifdef CONFIG_F2FS_CHECK_FS
1131 unsigned int segno = GET_SEGNO(sbi, blkaddr);
1132 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1134 if (unlikely(check_valid_map(sbi, segno, offset))) {
1135 if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1136 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",
1137 blkaddr, source_blkaddr, segno);
1138 set_sbi_flag(sbi, SBI_NEED_FSCK);
1147 static int ra_data_block(struct inode *inode, pgoff_t index)
1149 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1150 struct address_space *mapping = inode->i_mapping;
1151 struct dnode_of_data dn;
1153 struct extent_info ei = {0, };
1154 struct f2fs_io_info fio = {
1156 .ino = inode->i_ino,
1161 .encrypted_page = NULL,
1167 page = f2fs_grab_cache_page(mapping, index, true);
1171 if (f2fs_lookup_read_extent_cache(inode, index, &ei)) {
1172 dn.data_blkaddr = ei.blk + index - ei.fofs;
1173 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1174 DATA_GENERIC_ENHANCE_READ))) {
1175 err = -EFSCORRUPTED;
1176 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1182 set_new_dnode(&dn, inode, NULL, NULL, 0);
1183 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1186 f2fs_put_dnode(&dn);
1188 if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1192 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1193 DATA_GENERIC_ENHANCE))) {
1194 err = -EFSCORRUPTED;
1195 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1201 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1204 * don't cache encrypted data into meta inode until previous dirty
1205 * data were writebacked to avoid racing between GC and flush.
1207 f2fs_wait_on_page_writeback(page, DATA, true, true);
1209 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1211 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1213 FGP_LOCK | FGP_CREAT, GFP_NOFS);
1214 if (!fio.encrypted_page) {
1219 err = f2fs_submit_page_bio(&fio);
1221 goto put_encrypted_page;
1222 f2fs_put_page(fio.encrypted_page, 0);
1223 f2fs_put_page(page, 1);
1225 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
1226 f2fs_update_iostat(sbi, NULL, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1230 f2fs_put_page(fio.encrypted_page, 1);
1232 f2fs_put_page(page, 1);
1237 * Move data block via META_MAPPING while keeping locked data page.
1238 * This can be used to move blocks, aka LBAs, directly on disk.
1240 static int move_data_block(struct inode *inode, block_t bidx,
1241 int gc_type, unsigned int segno, int off)
1243 struct f2fs_io_info fio = {
1244 .sbi = F2FS_I_SB(inode),
1245 .ino = inode->i_ino,
1250 .encrypted_page = NULL,
1254 struct dnode_of_data dn;
1255 struct f2fs_summary sum;
1256 struct node_info ni;
1257 struct page *page, *mpage;
1260 bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1261 int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&
1262 (fio.sbi->gc_mode != GC_URGENT_HIGH) ?
1263 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1265 /* do not read out */
1266 page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1270 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1275 err = f2fs_gc_pinned_control(inode, gc_type, segno);
1279 set_new_dnode(&dn, inode, NULL, NULL, 0);
1280 err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1284 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1285 ClearPageUptodate(page);
1291 * don't cache encrypted data into meta inode until previous dirty
1292 * data were writebacked to avoid racing between GC and flush.
1294 f2fs_wait_on_page_writeback(page, DATA, true, true);
1296 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1298 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);
1304 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1307 f2fs_down_write(&fio.sbi->io_order_lock);
1309 mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1310 fio.old_blkaddr, false);
1316 fio.encrypted_page = mpage;
1318 /* read source block in mpage */
1319 if (!PageUptodate(mpage)) {
1320 err = f2fs_submit_page_bio(&fio);
1322 f2fs_put_page(mpage, 1);
1326 f2fs_update_iostat(fio.sbi, inode, FS_DATA_READ_IO,
1328 f2fs_update_iostat(fio.sbi, NULL, FS_GDATA_READ_IO,
1332 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1333 !PageUptodate(mpage))) {
1335 f2fs_put_page(mpage, 1);
1340 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1342 /* allocate block address */
1343 f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1346 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1347 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1348 if (!fio.encrypted_page) {
1350 f2fs_put_page(mpage, 1);
1354 /* write target block */
1355 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1356 memcpy(page_address(fio.encrypted_page),
1357 page_address(mpage), PAGE_SIZE);
1358 f2fs_put_page(mpage, 1);
1359 invalidate_mapping_pages(META_MAPPING(fio.sbi),
1360 fio.old_blkaddr, fio.old_blkaddr);
1361 f2fs_invalidate_compress_page(fio.sbi, fio.old_blkaddr);
1363 set_page_dirty(fio.encrypted_page);
1364 if (clear_page_dirty_for_io(fio.encrypted_page))
1365 dec_page_count(fio.sbi, F2FS_DIRTY_META);
1367 set_page_writeback(fio.encrypted_page);
1368 ClearPageError(page);
1370 fio.op = REQ_OP_WRITE;
1371 fio.op_flags = REQ_SYNC;
1372 fio.new_blkaddr = newaddr;
1373 f2fs_submit_page_write(&fio);
1376 if (PageWriteback(fio.encrypted_page))
1377 end_page_writeback(fio.encrypted_page);
1381 f2fs_update_iostat(fio.sbi, NULL, FS_GC_DATA_IO, F2FS_BLKSIZE);
1383 f2fs_update_data_blkaddr(&dn, newaddr);
1384 set_inode_flag(inode, FI_APPEND_WRITE);
1385 if (page->index == 0)
1386 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1388 f2fs_put_page(fio.encrypted_page, 1);
1391 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1395 f2fs_up_write(&fio.sbi->io_order_lock);
1397 f2fs_put_dnode(&dn);
1399 f2fs_put_page(page, 1);
1403 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1404 unsigned int segno, int off)
1409 page = f2fs_get_lock_data_page(inode, bidx, true);
1411 return PTR_ERR(page);
1413 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1418 err = f2fs_gc_pinned_control(inode, gc_type, segno);
1422 if (gc_type == BG_GC) {
1423 if (PageWriteback(page)) {
1427 set_page_dirty(page);
1428 set_page_private_gcing(page);
1430 struct f2fs_io_info fio = {
1431 .sbi = F2FS_I_SB(inode),
1432 .ino = inode->i_ino,
1436 .op_flags = REQ_SYNC,
1437 .old_blkaddr = NULL_ADDR,
1439 .encrypted_page = NULL,
1440 .need_lock = LOCK_REQ,
1441 .io_type = FS_GC_DATA_IO,
1443 bool is_dirty = PageDirty(page);
1446 f2fs_wait_on_page_writeback(page, DATA, true, true);
1448 set_page_dirty(page);
1449 if (clear_page_dirty_for_io(page)) {
1450 inode_dec_dirty_pages(inode);
1451 f2fs_remove_dirty_inode(inode);
1454 set_page_private_gcing(page);
1456 err = f2fs_do_write_data_page(&fio);
1458 clear_page_private_gcing(page);
1459 if (err == -ENOMEM) {
1460 memalloc_retry_wait(GFP_NOFS);
1464 set_page_dirty(page);
1468 f2fs_put_page(page, 1);
1473 * This function tries to get parent node of victim data block, and identifies
1474 * data block validity. If the block is valid, copy that with cold status and
1475 * modify parent node.
1476 * If the parent node is not valid or the data block address is different,
1477 * the victim data block is ignored.
1479 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1480 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1483 struct super_block *sb = sbi->sb;
1484 struct f2fs_summary *entry;
1489 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1491 start_addr = START_BLOCK(sbi, segno);
1496 for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1497 struct page *data_page;
1498 struct inode *inode;
1499 struct node_info dni; /* dnode info for the data */
1500 unsigned int ofs_in_node, nofs;
1502 nid_t nid = le32_to_cpu(entry->nid);
1505 * stop BG_GC if there is not enough free sections.
1506 * Or, stop GC if the segment becomes fully valid caused by
1507 * race condition along with SSR block allocation.
1509 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1510 (!force_migrate && get_valid_blocks(sbi, segno, true) ==
1511 CAP_BLKS_PER_SEC(sbi)))
1514 if (check_valid_map(sbi, segno, off) == 0)
1518 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1524 f2fs_ra_node_page(sbi, nid);
1528 /* Get an inode by ino with checking validity */
1529 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1533 f2fs_ra_node_page(sbi, dni.ino);
1537 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1542 inode = f2fs_iget(sb, dni.ino);
1543 if (IS_ERR(inode) || is_bad_inode(inode) ||
1544 special_file(inode->i_mode))
1547 err = f2fs_gc_pinned_control(inode, gc_type, segno);
1548 if (err == -EAGAIN) {
1553 if (!f2fs_down_write_trylock(
1554 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1556 sbi->skipped_gc_rwsem++;
1560 start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1563 if (f2fs_post_read_required(inode)) {
1564 int err = ra_data_block(inode, start_bidx);
1566 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1571 add_gc_inode(gc_list, inode);
1575 data_page = f2fs_get_read_data_page(inode, start_bidx,
1576 REQ_RAHEAD, true, NULL);
1577 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1578 if (IS_ERR(data_page)) {
1583 f2fs_put_page(data_page, 0);
1584 add_gc_inode(gc_list, inode);
1589 inode = find_gc_inode(gc_list, dni.ino);
1591 struct f2fs_inode_info *fi = F2FS_I(inode);
1592 bool locked = false;
1595 if (S_ISREG(inode->i_mode)) {
1596 if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[READ])) {
1597 sbi->skipped_gc_rwsem++;
1600 if (!f2fs_down_write_trylock(
1601 &fi->i_gc_rwsem[WRITE])) {
1602 sbi->skipped_gc_rwsem++;
1603 f2fs_up_write(&fi->i_gc_rwsem[READ]);
1608 /* wait for all inflight aio data */
1609 inode_dio_wait(inode);
1612 start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1614 if (f2fs_post_read_required(inode))
1615 err = move_data_block(inode, start_bidx,
1616 gc_type, segno, off);
1618 err = move_data_page(inode, start_bidx, gc_type,
1621 if (!err && (gc_type == FG_GC ||
1622 f2fs_post_read_required(inode)))
1626 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
1627 f2fs_up_write(&fi->i_gc_rwsem[READ]);
1630 stat_inc_data_blk_count(sbi, 1, gc_type);
1640 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1643 struct sit_info *sit_i = SIT_I(sbi);
1646 down_write(&sit_i->sentry_lock);
1647 ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1648 NO_CHECK_TYPE, LFS, 0);
1649 up_write(&sit_i->sentry_lock);
1653 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1654 unsigned int start_segno,
1655 struct gc_inode_list *gc_list, int gc_type,
1658 struct page *sum_page;
1659 struct f2fs_summary_block *sum;
1660 struct blk_plug plug;
1661 unsigned int segno = start_segno;
1662 unsigned int end_segno = start_segno + sbi->segs_per_sec;
1663 int seg_freed = 0, migrated = 0;
1664 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1665 SUM_TYPE_DATA : SUM_TYPE_NODE;
1668 if (__is_large_section(sbi))
1669 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1672 * zone-capacity can be less than zone-size in zoned devices,
1673 * resulting in less than expected usable segments in the zone,
1674 * calculate the end segno in the zone which can be garbage collected
1676 if (f2fs_sb_has_blkzoned(sbi))
1677 end_segno -= sbi->segs_per_sec -
1678 f2fs_usable_segs_in_sec(sbi, segno);
1680 sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1682 /* readahead multi ssa blocks those have contiguous address */
1683 if (__is_large_section(sbi))
1684 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1685 end_segno - segno, META_SSA, true);
1687 /* reference all summary page */
1688 while (segno < end_segno) {
1689 sum_page = f2fs_get_sum_page(sbi, segno++);
1690 if (IS_ERR(sum_page)) {
1691 int err = PTR_ERR(sum_page);
1693 end_segno = segno - 1;
1694 for (segno = start_segno; segno < end_segno; segno++) {
1695 sum_page = find_get_page(META_MAPPING(sbi),
1696 GET_SUM_BLOCK(sbi, segno));
1697 f2fs_put_page(sum_page, 0);
1698 f2fs_put_page(sum_page, 0);
1702 unlock_page(sum_page);
1705 blk_start_plug(&plug);
1707 for (segno = start_segno; segno < end_segno; segno++) {
1709 /* find segment summary of victim */
1710 sum_page = find_get_page(META_MAPPING(sbi),
1711 GET_SUM_BLOCK(sbi, segno));
1712 f2fs_put_page(sum_page, 0);
1714 if (get_valid_blocks(sbi, segno, false) == 0)
1716 if (gc_type == BG_GC && __is_large_section(sbi) &&
1717 migrated >= sbi->migration_granularity)
1719 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1722 sum = page_address(sum_page);
1723 if (type != GET_SUM_TYPE((&sum->footer))) {
1724 f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1725 segno, type, GET_SUM_TYPE((&sum->footer)));
1726 set_sbi_flag(sbi, SBI_NEED_FSCK);
1727 f2fs_stop_checkpoint(sbi, false,
1728 STOP_CP_REASON_CORRUPTED_SUMMARY);
1733 * this is to avoid deadlock:
1734 * - lock_page(sum_page) - f2fs_replace_block
1735 * - check_valid_map() - down_write(sentry_lock)
1736 * - down_read(sentry_lock) - change_curseg()
1737 * - lock_page(sum_page)
1739 if (type == SUM_TYPE_NODE)
1740 submitted += gc_node_segment(sbi, sum->entries, segno,
1743 submitted += gc_data_segment(sbi, sum->entries, gc_list,
1747 stat_inc_seg_count(sbi, type, gc_type);
1748 sbi->gc_reclaimed_segs[sbi->gc_mode]++;
1752 if (gc_type == FG_GC &&
1753 get_valid_blocks(sbi, segno, false) == 0)
1756 if (__is_large_section(sbi))
1757 sbi->next_victim_seg[gc_type] =
1758 (segno + 1 < end_segno) ? segno + 1 : NULL_SEGNO;
1760 f2fs_put_page(sum_page, 0);
1764 f2fs_submit_merged_write(sbi,
1765 (type == SUM_TYPE_NODE) ? NODE : DATA);
1767 blk_finish_plug(&plug);
1769 stat_inc_call_count(sbi->stat_info);
1774 int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control)
1776 int gc_type = gc_control->init_gc_type;
1777 unsigned int segno = gc_control->victim_segno;
1778 int sec_freed = 0, seg_freed = 0, total_freed = 0;
1780 struct cp_control cpc;
1781 struct gc_inode_list gc_list = {
1782 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1783 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1785 unsigned int skipped_round = 0, round = 0;
1787 trace_f2fs_gc_begin(sbi->sb, gc_type, gc_control->no_bg_gc,
1788 gc_control->nr_free_secs,
1789 get_pages(sbi, F2FS_DIRTY_NODES),
1790 get_pages(sbi, F2FS_DIRTY_DENTS),
1791 get_pages(sbi, F2FS_DIRTY_IMETA),
1794 reserved_segments(sbi),
1795 prefree_segments(sbi));
1797 cpc.reason = __get_cp_reason(sbi);
1798 sbi->skipped_gc_rwsem = 0;
1800 if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1804 if (unlikely(f2fs_cp_error(sbi))) {
1809 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1811 * For example, if there are many prefree_segments below given
1812 * threshold, we can make them free by checkpoint. Then, we
1813 * secure free segments which doesn't need fggc any more.
1815 if (prefree_segments(sbi)) {
1816 ret = f2fs_write_checkpoint(sbi, &cpc);
1820 if (has_not_enough_free_secs(sbi, 0, 0))
1824 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1825 if (gc_type == BG_GC && gc_control->no_bg_gc) {
1830 ret = __get_victim(sbi, &segno, gc_type);
1832 /* allow to search victim from sections has pinned data */
1833 if (ret == -ENODATA && gc_type == FG_GC &&
1834 f2fs_pinned_section_exists(DIRTY_I(sbi))) {
1835 f2fs_unpin_all_sections(sbi, false);
1841 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type,
1842 gc_control->should_migrate_blocks);
1843 total_freed += seg_freed;
1845 if (seg_freed == f2fs_usable_segs_in_sec(sbi, segno))
1848 if (gc_type == FG_GC)
1849 sbi->cur_victim_sec = NULL_SEGNO;
1851 if (gc_control->init_gc_type == FG_GC ||
1852 !has_not_enough_free_secs(sbi,
1853 (gc_type == FG_GC) ? sec_freed : 0, 0)) {
1854 if (gc_type == FG_GC && sec_freed < gc_control->nr_free_secs)
1859 /* FG_GC stops GC by skip_count */
1860 if (gc_type == FG_GC) {
1861 if (sbi->skipped_gc_rwsem)
1864 if (skipped_round > MAX_SKIP_GC_COUNT &&
1865 skipped_round * 2 >= round) {
1866 ret = f2fs_write_checkpoint(sbi, &cpc);
1871 /* Write checkpoint to reclaim prefree segments */
1872 if (free_sections(sbi) < NR_CURSEG_PERSIST_TYPE &&
1873 prefree_segments(sbi)) {
1874 ret = f2fs_write_checkpoint(sbi, &cpc);
1883 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1884 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno;
1886 if (gc_type == FG_GC)
1887 f2fs_unpin_all_sections(sbi, true);
1889 trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1890 get_pages(sbi, F2FS_DIRTY_NODES),
1891 get_pages(sbi, F2FS_DIRTY_DENTS),
1892 get_pages(sbi, F2FS_DIRTY_IMETA),
1895 reserved_segments(sbi),
1896 prefree_segments(sbi));
1898 f2fs_up_write(&sbi->gc_lock);
1900 put_gc_inode(&gc_list);
1902 if (gc_control->err_gc_skipped && !ret)
1903 ret = sec_freed ? 0 : -EAGAIN;
1907 int __init f2fs_create_garbage_collection_cache(void)
1909 victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1910 sizeof(struct victim_entry));
1911 return victim_entry_slab ? 0 : -ENOMEM;
1914 void f2fs_destroy_garbage_collection_cache(void)
1916 kmem_cache_destroy(victim_entry_slab);
1919 static void init_atgc_management(struct f2fs_sb_info *sbi)
1921 struct atgc_management *am = &sbi->am;
1923 if (test_opt(sbi, ATGC) &&
1924 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1925 am->atgc_enabled = true;
1927 am->root = RB_ROOT_CACHED;
1928 INIT_LIST_HEAD(&am->victim_list);
1929 am->victim_count = 0;
1931 am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1932 am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1933 am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1934 am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
1937 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1939 DIRTY_I(sbi)->v_ops = &default_v_ops;
1941 sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1943 /* give warm/cold data area from slower device */
1944 if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1945 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1946 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1948 init_atgc_management(sbi);
1951 static int free_segment_range(struct f2fs_sb_info *sbi,
1952 unsigned int secs, bool gc_only)
1954 unsigned int segno, next_inuse, start, end;
1955 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1956 int gc_mode, gc_type;
1960 /* Force block allocation for GC */
1961 MAIN_SECS(sbi) -= secs;
1962 start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1963 end = MAIN_SEGS(sbi) - 1;
1965 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1966 for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1967 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1968 SIT_I(sbi)->last_victim[gc_mode] = 0;
1970 for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1971 if (sbi->next_victim_seg[gc_type] >= start)
1972 sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1973 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1975 /* Move out cursegs from the target range */
1976 for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1977 f2fs_allocate_segment_for_resize(sbi, type, start, end);
1979 /* do GC to move out valid blocks in the range */
1980 for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1981 struct gc_inode_list gc_list = {
1982 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1983 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1986 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
1987 put_gc_inode(&gc_list);
1989 if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1993 if (fatal_signal_pending(current)) {
2001 err = f2fs_write_checkpoint(sbi, &cpc);
2005 next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
2006 if (next_inuse <= end) {
2007 f2fs_err(sbi, "segno %u should be free but still inuse!",
2009 f2fs_bug_on(sbi, 1);
2012 MAIN_SECS(sbi) += secs;
2016 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
2018 struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
2021 int segment_count_main;
2022 long long block_count;
2023 int segs = secs * sbi->segs_per_sec;
2025 f2fs_down_write(&sbi->sb_lock);
2027 section_count = le32_to_cpu(raw_sb->section_count);
2028 segment_count = le32_to_cpu(raw_sb->segment_count);
2029 segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
2030 block_count = le64_to_cpu(raw_sb->block_count);
2032 raw_sb->section_count = cpu_to_le32(section_count + secs);
2033 raw_sb->segment_count = cpu_to_le32(segment_count + segs);
2034 raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
2035 raw_sb->block_count = cpu_to_le64(block_count +
2036 (long long)segs * sbi->blocks_per_seg);
2037 if (f2fs_is_multi_device(sbi)) {
2038 int last_dev = sbi->s_ndevs - 1;
2040 le32_to_cpu(raw_sb->devs[last_dev].total_segments);
2042 raw_sb->devs[last_dev].total_segments =
2043 cpu_to_le32(dev_segs + segs);
2046 f2fs_up_write(&sbi->sb_lock);
2049 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
2051 int segs = secs * sbi->segs_per_sec;
2052 long long blks = (long long)segs * sbi->blocks_per_seg;
2053 long long user_block_count =
2054 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
2056 SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
2057 MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
2058 MAIN_SECS(sbi) += secs;
2059 FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
2060 FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
2061 F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
2063 if (f2fs_is_multi_device(sbi)) {
2064 int last_dev = sbi->s_ndevs - 1;
2066 FDEV(last_dev).total_segments =
2067 (int)FDEV(last_dev).total_segments + segs;
2068 FDEV(last_dev).end_blk =
2069 (long long)FDEV(last_dev).end_blk + blks;
2070 #ifdef CONFIG_BLK_DEV_ZONED
2071 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
2072 (int)(blks >> sbi->log_blocks_per_blkz);
2077 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
2079 __u64 old_block_count, shrunk_blocks;
2080 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
2085 old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
2086 if (block_count > old_block_count)
2089 if (f2fs_is_multi_device(sbi)) {
2090 int last_dev = sbi->s_ndevs - 1;
2091 __u64 last_segs = FDEV(last_dev).total_segments;
2093 if (block_count + last_segs * sbi->blocks_per_seg <=
2098 /* new fs size should align to section size */
2099 div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
2103 if (block_count == old_block_count)
2106 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2107 f2fs_err(sbi, "Should run fsck to repair first.");
2108 return -EFSCORRUPTED;
2111 if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2112 f2fs_err(sbi, "Checkpoint should be enabled.");
2116 shrunk_blocks = old_block_count - block_count;
2117 secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2120 if (!f2fs_down_write_trylock(&sbi->gc_lock))
2123 /* stop CP to protect MAIN_SEC in free_segment_range */
2126 spin_lock(&sbi->stat_lock);
2127 if (shrunk_blocks + valid_user_blocks(sbi) +
2128 sbi->current_reserved_blocks + sbi->unusable_block_count +
2129 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2131 spin_unlock(&sbi->stat_lock);
2136 err = free_segment_range(sbi, secs, true);
2139 f2fs_unlock_op(sbi);
2140 f2fs_up_write(&sbi->gc_lock);
2144 freeze_super(sbi->sb);
2145 f2fs_down_write(&sbi->gc_lock);
2146 f2fs_down_write(&sbi->cp_global_sem);
2148 spin_lock(&sbi->stat_lock);
2149 if (shrunk_blocks + valid_user_blocks(sbi) +
2150 sbi->current_reserved_blocks + sbi->unusable_block_count +
2151 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2154 sbi->user_block_count -= shrunk_blocks;
2155 spin_unlock(&sbi->stat_lock);
2159 set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2160 err = free_segment_range(sbi, secs, false);
2164 update_sb_metadata(sbi, -secs);
2166 err = f2fs_commit_super(sbi, false);
2168 update_sb_metadata(sbi, secs);
2172 update_fs_metadata(sbi, -secs);
2173 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2174 set_sbi_flag(sbi, SBI_IS_DIRTY);
2176 err = f2fs_write_checkpoint(sbi, &cpc);
2178 update_fs_metadata(sbi, secs);
2179 update_sb_metadata(sbi, secs);
2180 f2fs_commit_super(sbi, false);
2183 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2185 set_sbi_flag(sbi, SBI_NEED_FSCK);
2186 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2188 spin_lock(&sbi->stat_lock);
2189 sbi->user_block_count += shrunk_blocks;
2190 spin_unlock(&sbi->stat_lock);
2193 f2fs_up_write(&sbi->cp_global_sem);
2194 f2fs_up_write(&sbi->gc_lock);
2195 thaw_super(sbi->sb);