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;
39 wait_ms = gc_th->min_sleep_time;
43 bool sync_mode, foreground = false;
45 wait_event_interruptible_timeout(*wq,
46 kthread_should_stop() || freezing(current) ||
47 waitqueue_active(fggc_wq) ||
49 msecs_to_jiffies(wait_ms));
51 if (test_opt(sbi, GC_MERGE) && waitqueue_active(fggc_wq))
54 /* give it a try one time */
58 if (try_to_freeze()) {
59 stat_other_skip_bggc_count(sbi);
62 if (kthread_should_stop())
65 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
66 increase_sleep_time(gc_th, &wait_ms);
67 stat_other_skip_bggc_count(sbi);
71 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
72 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
73 f2fs_stop_checkpoint(sbi, false);
76 if (!sb_start_write_trylock(sbi->sb)) {
77 stat_other_skip_bggc_count(sbi);
82 * [GC triggering condition]
83 * 0. GC is not conducted currently.
84 * 1. There are enough dirty segments.
85 * 2. IO subsystem is idle by checking the # of writeback pages.
86 * 3. IO subsystem is idle by checking the # of requests in
87 * bdev's request list.
89 * Note) We have to avoid triggering GCs frequently.
90 * Because it is possible that some segments can be
91 * invalidated soon after by user update or deletion.
92 * So, I'd like to wait some time to collect dirty segments.
94 if (sbi->gc_mode == GC_URGENT_HIGH) {
95 spin_lock(&sbi->gc_urgent_high_lock);
96 if (sbi->gc_urgent_high_limited) {
97 if (!sbi->gc_urgent_high_remaining) {
98 sbi->gc_urgent_high_limited = false;
99 spin_unlock(&sbi->gc_urgent_high_lock);
100 sbi->gc_mode = GC_NORMAL;
103 sbi->gc_urgent_high_remaining--;
105 spin_unlock(&sbi->gc_urgent_high_lock);
108 if (sbi->gc_mode == GC_URGENT_HIGH ||
109 sbi->gc_mode == GC_URGENT_MID) {
110 wait_ms = gc_th->urgent_sleep_time;
111 f2fs_down_write(&sbi->gc_lock);
116 f2fs_down_write(&sbi->gc_lock);
118 } else if (!f2fs_down_write_trylock(&sbi->gc_lock)) {
119 stat_other_skip_bggc_count(sbi);
123 if (!is_idle(sbi, GC_TIME)) {
124 increase_sleep_time(gc_th, &wait_ms);
125 f2fs_up_write(&sbi->gc_lock);
126 stat_io_skip_bggc_count(sbi);
130 if (has_enough_invalid_blocks(sbi))
131 decrease_sleep_time(gc_th, &wait_ms);
133 increase_sleep_time(gc_th, &wait_ms);
136 stat_inc_bggc_count(sbi->stat_info);
138 sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
140 /* foreground GC was been triggered via f2fs_balance_fs() */
144 /* if return value is not zero, no victim was selected */
145 if (f2fs_gc(sbi, sync_mode, !foreground, false, NULL_SEGNO))
146 wait_ms = gc_th->no_gc_sleep_time;
149 wake_up_all(&gc_th->fggc_wq);
151 trace_f2fs_background_gc(sbi->sb, wait_ms,
152 prefree_segments(sbi), free_segments(sbi));
154 /* balancing f2fs's metadata periodically */
155 f2fs_balance_fs_bg(sbi, true);
157 sb_end_write(sbi->sb);
159 } while (!kthread_should_stop());
163 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
165 struct f2fs_gc_kthread *gc_th;
166 dev_t dev = sbi->sb->s_bdev->bd_dev;
169 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
175 gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
176 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
177 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
178 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
182 sbi->gc_thread = gc_th;
183 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
184 init_waitqueue_head(&sbi->gc_thread->fggc_wq);
185 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
186 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
187 if (IS_ERR(gc_th->f2fs_gc_task)) {
188 err = PTR_ERR(gc_th->f2fs_gc_task);
190 sbi->gc_thread = NULL;
196 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
198 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
202 kthread_stop(gc_th->f2fs_gc_task);
203 wake_up_all(&gc_th->fggc_wq);
205 sbi->gc_thread = NULL;
208 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
212 if (gc_type == BG_GC) {
213 if (sbi->am.atgc_enabled)
221 switch (sbi->gc_mode) {
237 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
238 int type, struct victim_sel_policy *p)
240 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
242 if (p->alloc_mode == SSR) {
243 p->gc_mode = GC_GREEDY;
244 p->dirty_bitmap = dirty_i->dirty_segmap[type];
245 p->max_search = dirty_i->nr_dirty[type];
247 } else if (p->alloc_mode == AT_SSR) {
248 p->gc_mode = GC_GREEDY;
249 p->dirty_bitmap = dirty_i->dirty_segmap[type];
250 p->max_search = dirty_i->nr_dirty[type];
253 p->gc_mode = select_gc_type(sbi, gc_type);
254 p->ofs_unit = sbi->segs_per_sec;
255 if (__is_large_section(sbi)) {
256 p->dirty_bitmap = dirty_i->dirty_secmap;
257 p->max_search = count_bits(p->dirty_bitmap,
260 p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
261 p->max_search = dirty_i->nr_dirty[DIRTY];
266 * adjust candidates range, should select all dirty segments for
267 * foreground GC and urgent GC cases.
269 if (gc_type != FG_GC &&
270 (sbi->gc_mode != GC_URGENT_HIGH) &&
271 (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
272 p->max_search > sbi->max_victim_search)
273 p->max_search = sbi->max_victim_search;
275 /* let's select beginning hot/small space first in no_heap mode*/
276 if (f2fs_need_rand_seg(sbi))
277 p->offset = prandom_u32() % (MAIN_SECS(sbi) * sbi->segs_per_sec);
278 else if (test_opt(sbi, NOHEAP) &&
279 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
282 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
285 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
286 struct victim_sel_policy *p)
288 /* SSR allocates in a segment unit */
289 if (p->alloc_mode == SSR)
290 return sbi->blocks_per_seg;
291 else if (p->alloc_mode == AT_SSR)
295 if (p->gc_mode == GC_GREEDY)
296 return 2 * sbi->blocks_per_seg * p->ofs_unit;
297 else if (p->gc_mode == GC_CB)
299 else if (p->gc_mode == GC_AT)
301 else /* No other gc_mode */
305 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
307 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
311 * If the gc_type is FG_GC, we can select victim segments
312 * selected by background GC before.
313 * Those segments guarantee they have small valid blocks.
315 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
316 if (sec_usage_check(sbi, secno))
318 clear_bit(secno, dirty_i->victim_secmap);
319 return GET_SEG_FROM_SEC(sbi, secno);
324 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
326 struct sit_info *sit_i = SIT_I(sbi);
327 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
328 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
329 unsigned long long mtime = 0;
330 unsigned int vblocks;
331 unsigned char age = 0;
334 unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
336 for (i = 0; i < usable_segs_per_sec; i++)
337 mtime += get_seg_entry(sbi, start + i)->mtime;
338 vblocks = get_valid_blocks(sbi, segno, true);
340 mtime = div_u64(mtime, usable_segs_per_sec);
341 vblocks = div_u64(vblocks, usable_segs_per_sec);
343 u = (vblocks * 100) >> sbi->log_blocks_per_seg;
345 /* Handle if the system time has changed by the user */
346 if (mtime < sit_i->min_mtime)
347 sit_i->min_mtime = mtime;
348 if (mtime > sit_i->max_mtime)
349 sit_i->max_mtime = mtime;
350 if (sit_i->max_mtime != sit_i->min_mtime)
351 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
352 sit_i->max_mtime - sit_i->min_mtime);
354 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
357 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
358 unsigned int segno, struct victim_sel_policy *p)
360 if (p->alloc_mode == SSR)
361 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
363 /* alloc_mode == LFS */
364 if (p->gc_mode == GC_GREEDY)
365 return get_valid_blocks(sbi, segno, true);
366 else if (p->gc_mode == GC_CB)
367 return get_cb_cost(sbi, segno);
373 static unsigned int count_bits(const unsigned long *addr,
374 unsigned int offset, unsigned int len)
376 unsigned int end = offset + len, sum = 0;
378 while (offset < end) {
379 if (test_bit(offset++, addr))
385 static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi,
386 unsigned long long mtime, unsigned int segno,
387 struct rb_node *parent, struct rb_node **p,
390 struct atgc_management *am = &sbi->am;
391 struct victim_entry *ve;
393 ve = f2fs_kmem_cache_alloc(victim_entry_slab,
394 GFP_NOFS, true, NULL);
399 rb_link_node(&ve->rb_node, parent, p);
400 rb_insert_color_cached(&ve->rb_node, &am->root, left_most);
402 list_add_tail(&ve->list, &am->victim_list);
409 static void insert_victim_entry(struct f2fs_sb_info *sbi,
410 unsigned long long mtime, unsigned int segno)
412 struct atgc_management *am = &sbi->am;
414 struct rb_node *parent = NULL;
415 bool left_most = true;
417 p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most);
418 attach_victim_entry(sbi, mtime, segno, parent, p, left_most);
421 static void add_victim_entry(struct f2fs_sb_info *sbi,
422 struct victim_sel_policy *p, unsigned int segno)
424 struct sit_info *sit_i = SIT_I(sbi);
425 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
426 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
427 unsigned long long mtime = 0;
430 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
431 if (p->gc_mode == GC_AT &&
432 get_valid_blocks(sbi, segno, true) == 0)
436 for (i = 0; i < sbi->segs_per_sec; i++)
437 mtime += get_seg_entry(sbi, start + i)->mtime;
438 mtime = div_u64(mtime, sbi->segs_per_sec);
440 /* Handle if the system time has changed by the user */
441 if (mtime < sit_i->min_mtime)
442 sit_i->min_mtime = mtime;
443 if (mtime > sit_i->max_mtime)
444 sit_i->max_mtime = mtime;
445 if (mtime < sit_i->dirty_min_mtime)
446 sit_i->dirty_min_mtime = mtime;
447 if (mtime > sit_i->dirty_max_mtime)
448 sit_i->dirty_max_mtime = mtime;
450 /* don't choose young section as candidate */
451 if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
454 insert_victim_entry(sbi, mtime, segno);
457 static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi,
458 struct victim_sel_policy *p)
460 struct atgc_management *am = &sbi->am;
461 struct rb_node *parent = NULL;
464 f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most);
469 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
470 struct victim_sel_policy *p)
472 struct sit_info *sit_i = SIT_I(sbi);
473 struct atgc_management *am = &sbi->am;
474 struct rb_root_cached *root = &am->root;
475 struct rb_node *node;
477 struct victim_entry *ve;
478 unsigned long long total_time;
479 unsigned long long age, u, accu;
480 unsigned long long max_mtime = sit_i->dirty_max_mtime;
481 unsigned long long min_mtime = sit_i->dirty_min_mtime;
482 unsigned int sec_blocks = BLKS_PER_SEC(sbi);
483 unsigned int vblocks;
484 unsigned int dirty_threshold = max(am->max_candidate_count,
485 am->candidate_ratio *
486 am->victim_count / 100);
487 unsigned int age_weight = am->age_weight;
489 unsigned int iter = 0;
491 if (max_mtime < min_mtime)
495 total_time = max_mtime - min_mtime;
497 accu = div64_u64(ULLONG_MAX, total_time);
498 accu = min_t(unsigned long long, div_u64(accu, 100),
499 DEFAULT_ACCURACY_CLASS);
501 node = rb_first_cached(root);
503 re = rb_entry_safe(node, struct rb_entry, rb_node);
507 ve = (struct victim_entry *)re;
509 if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
512 /* age = 10000 * x% * 60 */
513 age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
516 vblocks = get_valid_blocks(sbi, ve->segno, true);
517 f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
519 /* u = 10000 * x% * 40 */
520 u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
523 f2fs_bug_on(sbi, age + u >= UINT_MAX);
525 cost = UINT_MAX - (age + u);
528 if (cost < p->min_cost ||
529 (cost == p->min_cost && age > p->oldest_age)) {
532 p->min_segno = ve->segno;
535 if (iter < dirty_threshold) {
536 node = rb_next(node);
542 * select candidates around source section in range of
543 * [target - dirty_threshold, target + dirty_threshold]
545 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
546 struct victim_sel_policy *p)
548 struct sit_info *sit_i = SIT_I(sbi);
549 struct atgc_management *am = &sbi->am;
550 struct rb_node *node;
552 struct victim_entry *ve;
553 unsigned long long age;
554 unsigned long long max_mtime = sit_i->dirty_max_mtime;
555 unsigned long long min_mtime = sit_i->dirty_min_mtime;
556 unsigned int seg_blocks = sbi->blocks_per_seg;
557 unsigned int vblocks;
558 unsigned int dirty_threshold = max(am->max_candidate_count,
559 am->candidate_ratio *
560 am->victim_count / 100);
562 unsigned int iter = 0;
565 if (max_mtime < min_mtime)
569 node = lookup_central_victim(sbi, p);
571 re = rb_entry_safe(node, struct rb_entry, rb_node);
578 ve = (struct victim_entry *)re;
580 if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
583 age = max_mtime - ve->mtime;
585 vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
586 f2fs_bug_on(sbi, !vblocks);
589 if (vblocks == seg_blocks)
594 age = max_mtime - abs(p->age - age);
595 cost = UINT_MAX - vblocks;
597 if (cost < p->min_cost ||
598 (cost == p->min_cost && age > p->oldest_age)) {
601 p->min_segno = ve->segno;
604 if (iter < dirty_threshold) {
606 node = rb_prev(node);
608 node = rb_next(node);
618 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
619 struct victim_sel_policy *p)
621 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
622 &sbi->am.root, true));
624 if (p->gc_mode == GC_AT)
625 atgc_lookup_victim(sbi, p);
626 else if (p->alloc_mode == AT_SSR)
627 atssr_lookup_victim(sbi, p);
632 static void release_victim_entry(struct f2fs_sb_info *sbi)
634 struct atgc_management *am = &sbi->am;
635 struct victim_entry *ve, *tmp;
637 list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
639 kmem_cache_free(victim_entry_slab, ve);
643 am->root = RB_ROOT_CACHED;
645 f2fs_bug_on(sbi, am->victim_count);
646 f2fs_bug_on(sbi, !list_empty(&am->victim_list));
650 * This function is called from two paths.
651 * One is garbage collection and the other is SSR segment selection.
652 * When it is called during GC, it just gets a victim segment
653 * and it does not remove it from dirty seglist.
654 * When it is called from SSR segment selection, it finds a segment
655 * which has minimum valid blocks and removes it from dirty seglist.
657 static int get_victim_by_default(struct f2fs_sb_info *sbi,
658 unsigned int *result, int gc_type, int type,
659 char alloc_mode, unsigned long long age)
661 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
662 struct sit_info *sm = SIT_I(sbi);
663 struct victim_sel_policy p;
664 unsigned int secno, last_victim;
665 unsigned int last_segment;
666 unsigned int nsearched;
670 mutex_lock(&dirty_i->seglist_lock);
671 last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
673 p.alloc_mode = alloc_mode;
675 p.age_threshold = sbi->am.age_threshold;
678 select_policy(sbi, gc_type, type, &p);
679 p.min_segno = NULL_SEGNO;
681 p.min_cost = get_max_cost(sbi, &p);
683 is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
687 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
689 if (*result != NULL_SEGNO) {
690 if (!get_valid_blocks(sbi, *result, false)) {
695 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
698 p.min_segno = *result;
703 if (p.max_search == 0)
706 if (__is_large_section(sbi) && p.alloc_mode == LFS) {
707 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
708 p.min_segno = sbi->next_victim_seg[BG_GC];
709 *result = p.min_segno;
710 sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
713 if (gc_type == FG_GC &&
714 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
715 p.min_segno = sbi->next_victim_seg[FG_GC];
716 *result = p.min_segno;
717 sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
722 last_victim = sm->last_victim[p.gc_mode];
723 if (p.alloc_mode == LFS && gc_type == FG_GC) {
724 p.min_segno = check_bg_victims(sbi);
725 if (p.min_segno != NULL_SEGNO)
730 unsigned long cost, *dirty_bitmap;
731 unsigned int unit_no, segno;
733 dirty_bitmap = p.dirty_bitmap;
734 unit_no = find_next_bit(dirty_bitmap,
735 last_segment / p.ofs_unit,
736 p.offset / p.ofs_unit);
737 segno = unit_no * p.ofs_unit;
738 if (segno >= last_segment) {
739 if (sm->last_victim[p.gc_mode]) {
741 sm->last_victim[p.gc_mode];
742 sm->last_victim[p.gc_mode] = 0;
749 p.offset = segno + p.ofs_unit;
752 #ifdef CONFIG_F2FS_CHECK_FS
754 * skip selecting the invalid segno (that is failed due to block
755 * validity check failure during GC) to avoid endless GC loop in
758 if (test_bit(segno, sm->invalid_segmap))
762 secno = GET_SEC_FROM_SEG(sbi, segno);
764 if (sec_usage_check(sbi, secno))
767 /* Don't touch checkpointed data */
768 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
769 if (p.alloc_mode == LFS) {
771 * LFS is set to find source section during GC.
772 * The victim should have no checkpointed data.
774 if (get_ckpt_valid_blocks(sbi, segno, true))
778 * SSR | AT_SSR are set to find target segment
779 * for writes which can be full by checkpointed
780 * and newly written blocks.
782 if (!f2fs_segment_has_free_slot(sbi, segno))
787 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
791 add_victim_entry(sbi, &p, segno);
795 cost = get_gc_cost(sbi, segno, &p);
797 if (p.min_cost > cost) {
802 if (nsearched >= p.max_search) {
803 if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
804 sm->last_victim[p.gc_mode] =
805 last_victim + p.ofs_unit;
807 sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
808 sm->last_victim[p.gc_mode] %=
809 (MAIN_SECS(sbi) * sbi->segs_per_sec);
814 /* get victim for GC_AT/AT_SSR */
816 lookup_victim_by_age(sbi, &p);
817 release_victim_entry(sbi);
820 if (is_atgc && p.min_segno == NULL_SEGNO &&
821 sm->elapsed_time < p.age_threshold) {
826 if (p.min_segno != NULL_SEGNO) {
828 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
830 if (p.alloc_mode == LFS) {
831 secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
832 if (gc_type == FG_GC)
833 sbi->cur_victim_sec = secno;
835 set_bit(secno, dirty_i->victim_secmap);
841 if (p.min_segno != NULL_SEGNO)
842 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
844 prefree_segments(sbi), free_segments(sbi));
845 mutex_unlock(&dirty_i->seglist_lock);
850 static const struct victim_selection default_v_ops = {
851 .get_victim = get_victim_by_default,
854 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
856 struct inode_entry *ie;
858 ie = radix_tree_lookup(&gc_list->iroot, ino);
864 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
866 struct inode_entry *new_ie;
868 if (inode == find_gc_inode(gc_list, inode->i_ino)) {
872 new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab,
873 GFP_NOFS, true, NULL);
874 new_ie->inode = inode;
876 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
877 list_add_tail(&new_ie->list, &gc_list->ilist);
880 static void put_gc_inode(struct gc_inode_list *gc_list)
882 struct inode_entry *ie, *next_ie;
884 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
885 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
888 kmem_cache_free(f2fs_inode_entry_slab, ie);
892 static int check_valid_map(struct f2fs_sb_info *sbi,
893 unsigned int segno, int offset)
895 struct sit_info *sit_i = SIT_I(sbi);
896 struct seg_entry *sentry;
899 down_read(&sit_i->sentry_lock);
900 sentry = get_seg_entry(sbi, segno);
901 ret = f2fs_test_bit(offset, sentry->cur_valid_map);
902 up_read(&sit_i->sentry_lock);
907 * This function compares node address got in summary with that in NAT.
908 * On validity, copy that node with cold status, otherwise (invalid node)
911 static int gc_node_segment(struct f2fs_sb_info *sbi,
912 struct f2fs_summary *sum, unsigned int segno, int gc_type)
914 struct f2fs_summary *entry;
918 bool fggc = (gc_type == FG_GC);
920 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
922 start_addr = START_BLOCK(sbi, segno);
927 if (fggc && phase == 2)
928 atomic_inc(&sbi->wb_sync_req[NODE]);
930 for (off = 0; off < usable_blks_in_seg; off++, entry++) {
931 nid_t nid = le32_to_cpu(entry->nid);
932 struct page *node_page;
936 /* stop BG_GC if there is not enough free sections. */
937 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
940 if (check_valid_map(sbi, segno, off) == 0)
944 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
950 f2fs_ra_node_page(sbi, nid);
955 node_page = f2fs_get_node_page(sbi, nid);
956 if (IS_ERR(node_page))
959 /* block may become invalid during f2fs_get_node_page */
960 if (check_valid_map(sbi, segno, off) == 0) {
961 f2fs_put_page(node_page, 1);
965 if (f2fs_get_node_info(sbi, nid, &ni, false)) {
966 f2fs_put_page(node_page, 1);
970 if (ni.blk_addr != start_addr + off) {
971 f2fs_put_page(node_page, 1);
975 err = f2fs_move_node_page(node_page, gc_type);
976 if (!err && gc_type == FG_GC)
978 stat_inc_node_blk_count(sbi, 1, gc_type);
985 atomic_dec(&sbi->wb_sync_req[NODE]);
990 * Calculate start block index indicating the given node offset.
991 * Be careful, caller should give this node offset only indicating direct node
992 * blocks. If any node offsets, which point the other types of node blocks such
993 * as indirect or double indirect node blocks, are given, it must be a caller's
996 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
998 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
1004 if (node_ofs <= 2) {
1005 bidx = node_ofs - 1;
1006 } else if (node_ofs <= indirect_blks) {
1007 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
1009 bidx = node_ofs - 2 - dec;
1011 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
1013 bidx = node_ofs - 5 - dec;
1015 return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
1018 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1019 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
1021 struct page *node_page;
1023 unsigned int ofs_in_node;
1024 block_t source_blkaddr;
1026 nid = le32_to_cpu(sum->nid);
1027 ofs_in_node = le16_to_cpu(sum->ofs_in_node);
1029 node_page = f2fs_get_node_page(sbi, nid);
1030 if (IS_ERR(node_page))
1033 if (f2fs_get_node_info(sbi, nid, dni, false)) {
1034 f2fs_put_page(node_page, 1);
1038 if (sum->version != dni->version) {
1039 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
1041 set_sbi_flag(sbi, SBI_NEED_FSCK);
1044 if (f2fs_check_nid_range(sbi, dni->ino)) {
1045 f2fs_put_page(node_page, 1);
1049 *nofs = ofs_of_node(node_page);
1050 source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1051 f2fs_put_page(node_page, 1);
1053 if (source_blkaddr != blkaddr) {
1054 #ifdef CONFIG_F2FS_CHECK_FS
1055 unsigned int segno = GET_SEGNO(sbi, blkaddr);
1056 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1058 if (unlikely(check_valid_map(sbi, segno, offset))) {
1059 if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1060 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",
1061 blkaddr, source_blkaddr, segno);
1062 set_sbi_flag(sbi, SBI_NEED_FSCK);
1071 static int ra_data_block(struct inode *inode, pgoff_t index)
1073 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1074 struct address_space *mapping = inode->i_mapping;
1075 struct dnode_of_data dn;
1077 struct extent_info ei = {0, 0, 0};
1078 struct f2fs_io_info fio = {
1080 .ino = inode->i_ino,
1085 .encrypted_page = NULL,
1091 page = f2fs_grab_cache_page(mapping, index, true);
1095 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1096 dn.data_blkaddr = ei.blk + index - ei.fofs;
1097 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1098 DATA_GENERIC_ENHANCE_READ))) {
1099 err = -EFSCORRUPTED;
1105 set_new_dnode(&dn, inode, NULL, NULL, 0);
1106 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1109 f2fs_put_dnode(&dn);
1111 if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1115 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1116 DATA_GENERIC_ENHANCE))) {
1117 err = -EFSCORRUPTED;
1123 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1126 * don't cache encrypted data into meta inode until previous dirty
1127 * data were writebacked to avoid racing between GC and flush.
1129 f2fs_wait_on_page_writeback(page, DATA, true, true);
1131 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1133 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1135 FGP_LOCK | FGP_CREAT, GFP_NOFS);
1136 if (!fio.encrypted_page) {
1141 err = f2fs_submit_page_bio(&fio);
1143 goto put_encrypted_page;
1144 f2fs_put_page(fio.encrypted_page, 0);
1145 f2fs_put_page(page, 1);
1147 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1148 f2fs_update_iostat(sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1152 f2fs_put_page(fio.encrypted_page, 1);
1154 f2fs_put_page(page, 1);
1159 * Move data block via META_MAPPING while keeping locked data page.
1160 * This can be used to move blocks, aka LBAs, directly on disk.
1162 static int move_data_block(struct inode *inode, block_t bidx,
1163 int gc_type, unsigned int segno, int off)
1165 struct f2fs_io_info fio = {
1166 .sbi = F2FS_I_SB(inode),
1167 .ino = inode->i_ino,
1172 .encrypted_page = NULL,
1176 struct dnode_of_data dn;
1177 struct f2fs_summary sum;
1178 struct node_info ni;
1179 struct page *page, *mpage;
1182 bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1183 int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&
1184 (fio.sbi->gc_mode != GC_URGENT_HIGH) ?
1185 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1187 /* do not read out */
1188 page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1192 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1197 if (f2fs_is_atomic_file(inode)) {
1198 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1199 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1204 if (f2fs_is_pinned_file(inode)) {
1205 f2fs_pin_file_control(inode, true);
1210 set_new_dnode(&dn, inode, NULL, NULL, 0);
1211 err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1215 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1216 ClearPageUptodate(page);
1222 * don't cache encrypted data into meta inode until previous dirty
1223 * data were writebacked to avoid racing between GC and flush.
1225 f2fs_wait_on_page_writeback(page, DATA, true, true);
1227 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1229 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);
1235 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1238 f2fs_down_write(&fio.sbi->io_order_lock);
1240 mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1241 fio.old_blkaddr, false);
1247 fio.encrypted_page = mpage;
1249 /* read source block in mpage */
1250 if (!PageUptodate(mpage)) {
1251 err = f2fs_submit_page_bio(&fio);
1253 f2fs_put_page(mpage, 1);
1257 f2fs_update_iostat(fio.sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1258 f2fs_update_iostat(fio.sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1261 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1262 !PageUptodate(mpage))) {
1264 f2fs_put_page(mpage, 1);
1269 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1271 /* allocate block address */
1272 f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1275 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1276 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1277 if (!fio.encrypted_page) {
1279 f2fs_put_page(mpage, 1);
1283 /* write target block */
1284 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1285 memcpy(page_address(fio.encrypted_page),
1286 page_address(mpage), PAGE_SIZE);
1287 f2fs_put_page(mpage, 1);
1288 invalidate_mapping_pages(META_MAPPING(fio.sbi),
1289 fio.old_blkaddr, fio.old_blkaddr);
1290 f2fs_invalidate_compress_page(fio.sbi, fio.old_blkaddr);
1292 set_page_dirty(fio.encrypted_page);
1293 if (clear_page_dirty_for_io(fio.encrypted_page))
1294 dec_page_count(fio.sbi, F2FS_DIRTY_META);
1296 set_page_writeback(fio.encrypted_page);
1297 ClearPageError(page);
1299 fio.op = REQ_OP_WRITE;
1300 fio.op_flags = REQ_SYNC;
1301 fio.new_blkaddr = newaddr;
1302 f2fs_submit_page_write(&fio);
1305 if (PageWriteback(fio.encrypted_page))
1306 end_page_writeback(fio.encrypted_page);
1310 f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
1312 f2fs_update_data_blkaddr(&dn, newaddr);
1313 set_inode_flag(inode, FI_APPEND_WRITE);
1314 if (page->index == 0)
1315 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1317 f2fs_put_page(fio.encrypted_page, 1);
1320 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1324 f2fs_up_write(&fio.sbi->io_order_lock);
1326 f2fs_put_dnode(&dn);
1328 f2fs_put_page(page, 1);
1332 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1333 unsigned int segno, int off)
1338 page = f2fs_get_lock_data_page(inode, bidx, true);
1340 return PTR_ERR(page);
1342 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1347 if (f2fs_is_atomic_file(inode)) {
1348 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1349 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1353 if (f2fs_is_pinned_file(inode)) {
1354 if (gc_type == FG_GC)
1355 f2fs_pin_file_control(inode, true);
1360 if (gc_type == BG_GC) {
1361 if (PageWriteback(page)) {
1365 set_page_dirty(page);
1366 set_page_private_gcing(page);
1368 struct f2fs_io_info fio = {
1369 .sbi = F2FS_I_SB(inode),
1370 .ino = inode->i_ino,
1374 .op_flags = REQ_SYNC,
1375 .old_blkaddr = NULL_ADDR,
1377 .encrypted_page = NULL,
1378 .need_lock = LOCK_REQ,
1379 .io_type = FS_GC_DATA_IO,
1381 bool is_dirty = PageDirty(page);
1384 f2fs_wait_on_page_writeback(page, DATA, true, true);
1386 set_page_dirty(page);
1387 if (clear_page_dirty_for_io(page)) {
1388 inode_dec_dirty_pages(inode);
1389 f2fs_remove_dirty_inode(inode);
1392 set_page_private_gcing(page);
1394 err = f2fs_do_write_data_page(&fio);
1396 clear_page_private_gcing(page);
1397 if (err == -ENOMEM) {
1398 memalloc_retry_wait(GFP_NOFS);
1402 set_page_dirty(page);
1406 f2fs_put_page(page, 1);
1411 * This function tries to get parent node of victim data block, and identifies
1412 * data block validity. If the block is valid, copy that with cold status and
1413 * modify parent node.
1414 * If the parent node is not valid or the data block address is different,
1415 * the victim data block is ignored.
1417 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1418 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1421 struct super_block *sb = sbi->sb;
1422 struct f2fs_summary *entry;
1427 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1429 start_addr = START_BLOCK(sbi, segno);
1434 for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1435 struct page *data_page;
1436 struct inode *inode;
1437 struct node_info dni; /* dnode info for the data */
1438 unsigned int ofs_in_node, nofs;
1440 nid_t nid = le32_to_cpu(entry->nid);
1443 * stop BG_GC if there is not enough free sections.
1444 * Or, stop GC if the segment becomes fully valid caused by
1445 * race condition along with SSR block allocation.
1447 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1448 (!force_migrate && get_valid_blocks(sbi, segno, true) ==
1452 if (check_valid_map(sbi, segno, off) == 0)
1456 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1462 f2fs_ra_node_page(sbi, nid);
1466 /* Get an inode by ino with checking validity */
1467 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1471 f2fs_ra_node_page(sbi, dni.ino);
1475 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1478 inode = f2fs_iget(sb, dni.ino);
1479 if (IS_ERR(inode) || is_bad_inode(inode) ||
1480 special_file(inode->i_mode))
1483 if (!f2fs_down_write_trylock(
1484 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1486 sbi->skipped_gc_rwsem++;
1490 start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1493 if (f2fs_post_read_required(inode)) {
1494 int err = ra_data_block(inode, start_bidx);
1496 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1501 add_gc_inode(gc_list, inode);
1505 data_page = f2fs_get_read_data_page(inode,
1506 start_bidx, REQ_RAHEAD, true);
1507 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1508 if (IS_ERR(data_page)) {
1513 f2fs_put_page(data_page, 0);
1514 add_gc_inode(gc_list, inode);
1519 inode = find_gc_inode(gc_list, dni.ino);
1521 struct f2fs_inode_info *fi = F2FS_I(inode);
1522 bool locked = false;
1525 if (S_ISREG(inode->i_mode)) {
1526 if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[READ])) {
1527 sbi->skipped_gc_rwsem++;
1530 if (!f2fs_down_write_trylock(
1531 &fi->i_gc_rwsem[WRITE])) {
1532 sbi->skipped_gc_rwsem++;
1533 f2fs_up_write(&fi->i_gc_rwsem[READ]);
1538 /* wait for all inflight aio data */
1539 inode_dio_wait(inode);
1542 start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1544 if (f2fs_post_read_required(inode))
1545 err = move_data_block(inode, start_bidx,
1546 gc_type, segno, off);
1548 err = move_data_page(inode, start_bidx, gc_type,
1551 if (!err && (gc_type == FG_GC ||
1552 f2fs_post_read_required(inode)))
1556 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
1557 f2fs_up_write(&fi->i_gc_rwsem[READ]);
1560 stat_inc_data_blk_count(sbi, 1, gc_type);
1570 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1573 struct sit_info *sit_i = SIT_I(sbi);
1576 down_write(&sit_i->sentry_lock);
1577 ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1578 NO_CHECK_TYPE, LFS, 0);
1579 up_write(&sit_i->sentry_lock);
1583 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1584 unsigned int start_segno,
1585 struct gc_inode_list *gc_list, int gc_type,
1588 struct page *sum_page;
1589 struct f2fs_summary_block *sum;
1590 struct blk_plug plug;
1591 unsigned int segno = start_segno;
1592 unsigned int end_segno = start_segno + sbi->segs_per_sec;
1593 int seg_freed = 0, migrated = 0;
1594 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1595 SUM_TYPE_DATA : SUM_TYPE_NODE;
1598 if (__is_large_section(sbi))
1599 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1602 * zone-capacity can be less than zone-size in zoned devices,
1603 * resulting in less than expected usable segments in the zone,
1604 * calculate the end segno in the zone which can be garbage collected
1606 if (f2fs_sb_has_blkzoned(sbi))
1607 end_segno -= sbi->segs_per_sec -
1608 f2fs_usable_segs_in_sec(sbi, segno);
1610 sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1612 /* readahead multi ssa blocks those have contiguous address */
1613 if (__is_large_section(sbi))
1614 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1615 end_segno - segno, META_SSA, true);
1617 /* reference all summary page */
1618 while (segno < end_segno) {
1619 sum_page = f2fs_get_sum_page(sbi, segno++);
1620 if (IS_ERR(sum_page)) {
1621 int err = PTR_ERR(sum_page);
1623 end_segno = segno - 1;
1624 for (segno = start_segno; segno < end_segno; segno++) {
1625 sum_page = find_get_page(META_MAPPING(sbi),
1626 GET_SUM_BLOCK(sbi, segno));
1627 f2fs_put_page(sum_page, 0);
1628 f2fs_put_page(sum_page, 0);
1632 unlock_page(sum_page);
1635 blk_start_plug(&plug);
1637 for (segno = start_segno; segno < end_segno; segno++) {
1639 /* find segment summary of victim */
1640 sum_page = find_get_page(META_MAPPING(sbi),
1641 GET_SUM_BLOCK(sbi, segno));
1642 f2fs_put_page(sum_page, 0);
1644 if (get_valid_blocks(sbi, segno, false) == 0)
1646 if (gc_type == BG_GC && __is_large_section(sbi) &&
1647 migrated >= sbi->migration_granularity)
1649 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1652 sum = page_address(sum_page);
1653 if (type != GET_SUM_TYPE((&sum->footer))) {
1654 f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1655 segno, type, GET_SUM_TYPE((&sum->footer)));
1656 set_sbi_flag(sbi, SBI_NEED_FSCK);
1657 f2fs_stop_checkpoint(sbi, false);
1662 * this is to avoid deadlock:
1663 * - lock_page(sum_page) - f2fs_replace_block
1664 * - check_valid_map() - down_write(sentry_lock)
1665 * - down_read(sentry_lock) - change_curseg()
1666 * - lock_page(sum_page)
1668 if (type == SUM_TYPE_NODE)
1669 submitted += gc_node_segment(sbi, sum->entries, segno,
1672 submitted += gc_data_segment(sbi, sum->entries, gc_list,
1676 stat_inc_seg_count(sbi, type, gc_type);
1677 sbi->gc_reclaimed_segs[sbi->gc_mode]++;
1681 if (gc_type == FG_GC &&
1682 get_valid_blocks(sbi, segno, false) == 0)
1685 if (__is_large_section(sbi) && segno + 1 < end_segno)
1686 sbi->next_victim_seg[gc_type] = segno + 1;
1688 f2fs_put_page(sum_page, 0);
1692 f2fs_submit_merged_write(sbi,
1693 (type == SUM_TYPE_NODE) ? NODE : DATA);
1695 blk_finish_plug(&plug);
1697 stat_inc_call_count(sbi->stat_info);
1702 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1703 bool background, bool force, unsigned int segno)
1705 int gc_type = sync ? FG_GC : BG_GC;
1706 int sec_freed = 0, seg_freed = 0, total_freed = 0;
1708 struct cp_control cpc;
1709 unsigned int init_segno = segno;
1710 struct gc_inode_list gc_list = {
1711 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1712 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1714 unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1715 unsigned long long first_skipped;
1716 unsigned int skipped_round = 0, round = 0;
1718 trace_f2fs_gc_begin(sbi->sb, sync, background,
1719 get_pages(sbi, F2FS_DIRTY_NODES),
1720 get_pages(sbi, F2FS_DIRTY_DENTS),
1721 get_pages(sbi, F2FS_DIRTY_IMETA),
1724 reserved_segments(sbi),
1725 prefree_segments(sbi));
1727 cpc.reason = __get_cp_reason(sbi);
1728 sbi->skipped_gc_rwsem = 0;
1729 first_skipped = last_skipped;
1731 if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1735 if (unlikely(f2fs_cp_error(sbi))) {
1740 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1742 * For example, if there are many prefree_segments below given
1743 * threshold, we can make them free by checkpoint. Then, we
1744 * secure free segments which doesn't need fggc any more.
1746 if (prefree_segments(sbi) &&
1747 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1748 ret = f2fs_write_checkpoint(sbi, &cpc);
1752 if (has_not_enough_free_secs(sbi, 0, 0))
1756 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1757 if (gc_type == BG_GC && !background) {
1761 ret = __get_victim(sbi, &segno, gc_type);
1765 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type, force);
1766 if (gc_type == FG_GC &&
1767 seg_freed == f2fs_usable_segs_in_sec(sbi, segno))
1769 total_freed += seg_freed;
1771 if (gc_type == FG_GC) {
1772 if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
1773 sbi->skipped_gc_rwsem)
1775 last_skipped = sbi->skipped_atomic_files[FG_GC];
1779 if (gc_type == FG_GC)
1780 sbi->cur_victim_sec = NULL_SEGNO;
1785 if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1786 if (skipped_round <= MAX_SKIP_GC_COUNT ||
1787 skipped_round * 2 < round) {
1792 if (first_skipped < last_skipped &&
1793 (last_skipped - first_skipped) >
1794 sbi->skipped_gc_rwsem) {
1795 f2fs_drop_inmem_pages_all(sbi, true);
1799 if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1800 ret = f2fs_write_checkpoint(sbi, &cpc);
1803 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1804 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1806 trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1807 get_pages(sbi, F2FS_DIRTY_NODES),
1808 get_pages(sbi, F2FS_DIRTY_DENTS),
1809 get_pages(sbi, F2FS_DIRTY_IMETA),
1812 reserved_segments(sbi),
1813 prefree_segments(sbi));
1815 f2fs_up_write(&sbi->gc_lock);
1817 put_gc_inode(&gc_list);
1820 ret = sec_freed ? 0 : -EAGAIN;
1824 int __init f2fs_create_garbage_collection_cache(void)
1826 victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1827 sizeof(struct victim_entry));
1828 if (!victim_entry_slab)
1833 void f2fs_destroy_garbage_collection_cache(void)
1835 kmem_cache_destroy(victim_entry_slab);
1838 static void init_atgc_management(struct f2fs_sb_info *sbi)
1840 struct atgc_management *am = &sbi->am;
1842 if (test_opt(sbi, ATGC) &&
1843 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1844 am->atgc_enabled = true;
1846 am->root = RB_ROOT_CACHED;
1847 INIT_LIST_HEAD(&am->victim_list);
1848 am->victim_count = 0;
1850 am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1851 am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1852 am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1853 am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
1856 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1858 DIRTY_I(sbi)->v_ops = &default_v_ops;
1860 sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1862 /* give warm/cold data area from slower device */
1863 if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1864 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1865 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1867 init_atgc_management(sbi);
1870 static int free_segment_range(struct f2fs_sb_info *sbi,
1871 unsigned int secs, bool gc_only)
1873 unsigned int segno, next_inuse, start, end;
1874 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1875 int gc_mode, gc_type;
1879 /* Force block allocation for GC */
1880 MAIN_SECS(sbi) -= secs;
1881 start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1882 end = MAIN_SEGS(sbi) - 1;
1884 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1885 for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1886 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1887 SIT_I(sbi)->last_victim[gc_mode] = 0;
1889 for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1890 if (sbi->next_victim_seg[gc_type] >= start)
1891 sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1892 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1894 /* Move out cursegs from the target range */
1895 for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1896 f2fs_allocate_segment_for_resize(sbi, type, start, end);
1898 /* do GC to move out valid blocks in the range */
1899 for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1900 struct gc_inode_list gc_list = {
1901 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1902 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1905 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
1906 put_gc_inode(&gc_list);
1908 if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1912 if (fatal_signal_pending(current)) {
1920 err = f2fs_write_checkpoint(sbi, &cpc);
1924 next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1925 if (next_inuse <= end) {
1926 f2fs_err(sbi, "segno %u should be free but still inuse!",
1928 f2fs_bug_on(sbi, 1);
1931 MAIN_SECS(sbi) += secs;
1935 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1937 struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1940 int segment_count_main;
1941 long long block_count;
1942 int segs = secs * sbi->segs_per_sec;
1944 f2fs_down_write(&sbi->sb_lock);
1946 section_count = le32_to_cpu(raw_sb->section_count);
1947 segment_count = le32_to_cpu(raw_sb->segment_count);
1948 segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
1949 block_count = le64_to_cpu(raw_sb->block_count);
1951 raw_sb->section_count = cpu_to_le32(section_count + secs);
1952 raw_sb->segment_count = cpu_to_le32(segment_count + segs);
1953 raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
1954 raw_sb->block_count = cpu_to_le64(block_count +
1955 (long long)segs * sbi->blocks_per_seg);
1956 if (f2fs_is_multi_device(sbi)) {
1957 int last_dev = sbi->s_ndevs - 1;
1959 le32_to_cpu(raw_sb->devs[last_dev].total_segments);
1961 raw_sb->devs[last_dev].total_segments =
1962 cpu_to_le32(dev_segs + segs);
1965 f2fs_up_write(&sbi->sb_lock);
1968 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
1970 int segs = secs * sbi->segs_per_sec;
1971 long long blks = (long long)segs * sbi->blocks_per_seg;
1972 long long user_block_count =
1973 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
1975 SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
1976 MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
1977 MAIN_SECS(sbi) += secs;
1978 FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
1979 FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
1980 F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
1982 if (f2fs_is_multi_device(sbi)) {
1983 int last_dev = sbi->s_ndevs - 1;
1985 FDEV(last_dev).total_segments =
1986 (int)FDEV(last_dev).total_segments + segs;
1987 FDEV(last_dev).end_blk =
1988 (long long)FDEV(last_dev).end_blk + blks;
1989 #ifdef CONFIG_BLK_DEV_ZONED
1990 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
1991 (int)(blks >> sbi->log_blocks_per_blkz);
1996 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
1998 __u64 old_block_count, shrunk_blocks;
1999 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
2004 old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
2005 if (block_count > old_block_count)
2008 if (f2fs_is_multi_device(sbi)) {
2009 int last_dev = sbi->s_ndevs - 1;
2010 __u64 last_segs = FDEV(last_dev).total_segments;
2012 if (block_count + last_segs * sbi->blocks_per_seg <=
2017 /* new fs size should align to section size */
2018 div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
2022 if (block_count == old_block_count)
2025 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2026 f2fs_err(sbi, "Should run fsck to repair first.");
2027 return -EFSCORRUPTED;
2030 if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2031 f2fs_err(sbi, "Checkpoint should be enabled.");
2035 shrunk_blocks = old_block_count - block_count;
2036 secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2039 if (!f2fs_down_write_trylock(&sbi->gc_lock))
2042 /* stop CP to protect MAIN_SEC in free_segment_range */
2045 spin_lock(&sbi->stat_lock);
2046 if (shrunk_blocks + valid_user_blocks(sbi) +
2047 sbi->current_reserved_blocks + sbi->unusable_block_count +
2048 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2050 spin_unlock(&sbi->stat_lock);
2055 err = free_segment_range(sbi, secs, true);
2058 f2fs_unlock_op(sbi);
2059 f2fs_up_write(&sbi->gc_lock);
2063 set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2065 freeze_super(sbi->sb);
2066 f2fs_down_write(&sbi->gc_lock);
2067 f2fs_down_write(&sbi->cp_global_sem);
2069 spin_lock(&sbi->stat_lock);
2070 if (shrunk_blocks + valid_user_blocks(sbi) +
2071 sbi->current_reserved_blocks + sbi->unusable_block_count +
2072 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2075 sbi->user_block_count -= shrunk_blocks;
2076 spin_unlock(&sbi->stat_lock);
2080 err = free_segment_range(sbi, secs, false);
2084 update_sb_metadata(sbi, -secs);
2086 err = f2fs_commit_super(sbi, false);
2088 update_sb_metadata(sbi, secs);
2092 update_fs_metadata(sbi, -secs);
2093 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2094 set_sbi_flag(sbi, SBI_IS_DIRTY);
2096 err = f2fs_write_checkpoint(sbi, &cpc);
2098 update_fs_metadata(sbi, secs);
2099 update_sb_metadata(sbi, secs);
2100 f2fs_commit_super(sbi, false);
2104 set_sbi_flag(sbi, SBI_NEED_FSCK);
2105 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2107 spin_lock(&sbi->stat_lock);
2108 sbi->user_block_count += shrunk_blocks;
2109 spin_unlock(&sbi->stat_lock);
2112 f2fs_up_write(&sbi->cp_global_sem);
2113 f2fs_up_write(&sbi->gc_lock);
2114 thaw_super(sbi->sb);
2115 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);