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/backing-dev.h>
11 #include <linux/init.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/kthread.h>
14 #include <linux/delay.h>
15 #include <linux/freezer.h>
16 #include <linux/sched/signal.h>
23 #include <trace/events/f2fs.h>
25 static struct kmem_cache *victim_entry_slab;
27 static unsigned int count_bits(const unsigned long *addr,
28 unsigned int offset, unsigned int len);
30 static int gc_thread_func(void *data)
32 struct f2fs_sb_info *sbi = data;
33 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
34 wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
35 wait_queue_head_t *fggc_wq = &sbi->gc_thread->fggc_wq;
38 wait_ms = gc_th->min_sleep_time;
42 bool sync_mode, foreground = false;
44 wait_event_interruptible_timeout(*wq,
45 kthread_should_stop() || freezing(current) ||
46 waitqueue_active(fggc_wq) ||
48 msecs_to_jiffies(wait_ms));
50 if (test_opt(sbi, GC_MERGE) && waitqueue_active(fggc_wq))
53 /* give it a try one time */
57 if (try_to_freeze()) {
58 stat_other_skip_bggc_count(sbi);
61 if (kthread_should_stop())
64 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
65 increase_sleep_time(gc_th, &wait_ms);
66 stat_other_skip_bggc_count(sbi);
70 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
71 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
72 f2fs_stop_checkpoint(sbi, false);
75 if (!sb_start_write_trylock(sbi->sb)) {
76 stat_other_skip_bggc_count(sbi);
81 * [GC triggering condition]
82 * 0. GC is not conducted currently.
83 * 1. There are enough dirty segments.
84 * 2. IO subsystem is idle by checking the # of writeback pages.
85 * 3. IO subsystem is idle by checking the # of requests in
86 * bdev's request list.
88 * Note) We have to avoid triggering GCs frequently.
89 * Because it is possible that some segments can be
90 * invalidated soon after by user update or deletion.
91 * So, I'd like to wait some time to collect dirty segments.
93 if (sbi->gc_mode == GC_URGENT_HIGH) {
94 wait_ms = gc_th->urgent_sleep_time;
95 down_write(&sbi->gc_lock);
100 down_write(&sbi->gc_lock);
102 } else if (!down_write_trylock(&sbi->gc_lock)) {
103 stat_other_skip_bggc_count(sbi);
107 if (!is_idle(sbi, GC_TIME)) {
108 increase_sleep_time(gc_th, &wait_ms);
109 up_write(&sbi->gc_lock);
110 stat_io_skip_bggc_count(sbi);
114 if (has_enough_invalid_blocks(sbi))
115 decrease_sleep_time(gc_th, &wait_ms);
117 increase_sleep_time(gc_th, &wait_ms);
120 stat_inc_bggc_count(sbi->stat_info);
122 sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
124 /* foreground GC was been triggered via f2fs_balance_fs() */
128 /* if return value is not zero, no victim was selected */
129 if (f2fs_gc(sbi, sync_mode, !foreground, false, NULL_SEGNO))
130 wait_ms = gc_th->no_gc_sleep_time;
133 wake_up_all(&gc_th->fggc_wq);
135 trace_f2fs_background_gc(sbi->sb, wait_ms,
136 prefree_segments(sbi), free_segments(sbi));
138 /* balancing f2fs's metadata periodically */
139 f2fs_balance_fs_bg(sbi, true);
141 sb_end_write(sbi->sb);
143 } while (!kthread_should_stop());
147 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
149 struct f2fs_gc_kthread *gc_th;
150 dev_t dev = sbi->sb->s_bdev->bd_dev;
153 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
159 gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
160 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
161 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
162 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
166 sbi->gc_thread = gc_th;
167 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
168 init_waitqueue_head(&sbi->gc_thread->fggc_wq);
169 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
170 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
171 if (IS_ERR(gc_th->f2fs_gc_task)) {
172 err = PTR_ERR(gc_th->f2fs_gc_task);
174 sbi->gc_thread = NULL;
180 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
182 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
186 kthread_stop(gc_th->f2fs_gc_task);
187 wake_up_all(&gc_th->fggc_wq);
189 sbi->gc_thread = NULL;
192 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
196 if (gc_type == BG_GC) {
197 if (sbi->am.atgc_enabled)
205 switch (sbi->gc_mode) {
221 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
222 int type, struct victim_sel_policy *p)
224 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
226 if (p->alloc_mode == SSR) {
227 p->gc_mode = GC_GREEDY;
228 p->dirty_bitmap = dirty_i->dirty_segmap[type];
229 p->max_search = dirty_i->nr_dirty[type];
231 } else if (p->alloc_mode == AT_SSR) {
232 p->gc_mode = GC_GREEDY;
233 p->dirty_bitmap = dirty_i->dirty_segmap[type];
234 p->max_search = dirty_i->nr_dirty[type];
237 p->gc_mode = select_gc_type(sbi, gc_type);
238 p->ofs_unit = sbi->segs_per_sec;
239 if (__is_large_section(sbi)) {
240 p->dirty_bitmap = dirty_i->dirty_secmap;
241 p->max_search = count_bits(p->dirty_bitmap,
244 p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
245 p->max_search = dirty_i->nr_dirty[DIRTY];
250 * adjust candidates range, should select all dirty segments for
251 * foreground GC and urgent GC cases.
253 if (gc_type != FG_GC &&
254 (sbi->gc_mode != GC_URGENT_HIGH) &&
255 (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
256 p->max_search > sbi->max_victim_search)
257 p->max_search = sbi->max_victim_search;
259 /* let's select beginning hot/small space first in no_heap mode*/
260 if (test_opt(sbi, NOHEAP) &&
261 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
264 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
267 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
268 struct victim_sel_policy *p)
270 /* SSR allocates in a segment unit */
271 if (p->alloc_mode == SSR)
272 return sbi->blocks_per_seg;
273 else if (p->alloc_mode == AT_SSR)
277 if (p->gc_mode == GC_GREEDY)
278 return 2 * sbi->blocks_per_seg * p->ofs_unit;
279 else if (p->gc_mode == GC_CB)
281 else if (p->gc_mode == GC_AT)
283 else /* No other gc_mode */
287 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
289 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
293 * If the gc_type is FG_GC, we can select victim segments
294 * selected by background GC before.
295 * Those segments guarantee they have small valid blocks.
297 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
298 if (sec_usage_check(sbi, secno))
300 clear_bit(secno, dirty_i->victim_secmap);
301 return GET_SEG_FROM_SEC(sbi, secno);
306 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
308 struct sit_info *sit_i = SIT_I(sbi);
309 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
310 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
311 unsigned long long mtime = 0;
312 unsigned int vblocks;
313 unsigned char age = 0;
316 unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
318 for (i = 0; i < usable_segs_per_sec; i++)
319 mtime += get_seg_entry(sbi, start + i)->mtime;
320 vblocks = get_valid_blocks(sbi, segno, true);
322 mtime = div_u64(mtime, usable_segs_per_sec);
323 vblocks = div_u64(vblocks, usable_segs_per_sec);
325 u = (vblocks * 100) >> sbi->log_blocks_per_seg;
327 /* Handle if the system time has changed by the user */
328 if (mtime < sit_i->min_mtime)
329 sit_i->min_mtime = mtime;
330 if (mtime > sit_i->max_mtime)
331 sit_i->max_mtime = mtime;
332 if (sit_i->max_mtime != sit_i->min_mtime)
333 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
334 sit_i->max_mtime - sit_i->min_mtime);
336 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
339 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
340 unsigned int segno, struct victim_sel_policy *p)
342 if (p->alloc_mode == SSR)
343 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
345 /* alloc_mode == LFS */
346 if (p->gc_mode == GC_GREEDY)
347 return get_valid_blocks(sbi, segno, true);
348 else if (p->gc_mode == GC_CB)
349 return get_cb_cost(sbi, segno);
355 static unsigned int count_bits(const unsigned long *addr,
356 unsigned int offset, unsigned int len)
358 unsigned int end = offset + len, sum = 0;
360 while (offset < end) {
361 if (test_bit(offset++, addr))
367 static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi,
368 unsigned long long mtime, unsigned int segno,
369 struct rb_node *parent, struct rb_node **p,
372 struct atgc_management *am = &sbi->am;
373 struct victim_entry *ve;
375 ve = f2fs_kmem_cache_alloc(victim_entry_slab,
376 GFP_NOFS, true, NULL);
381 rb_link_node(&ve->rb_node, parent, p);
382 rb_insert_color_cached(&ve->rb_node, &am->root, left_most);
384 list_add_tail(&ve->list, &am->victim_list);
391 static void insert_victim_entry(struct f2fs_sb_info *sbi,
392 unsigned long long mtime, unsigned int segno)
394 struct atgc_management *am = &sbi->am;
396 struct rb_node *parent = NULL;
397 bool left_most = true;
399 p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most);
400 attach_victim_entry(sbi, mtime, segno, parent, p, left_most);
403 static void add_victim_entry(struct f2fs_sb_info *sbi,
404 struct victim_sel_policy *p, unsigned int segno)
406 struct sit_info *sit_i = SIT_I(sbi);
407 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
408 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
409 unsigned long long mtime = 0;
412 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
413 if (p->gc_mode == GC_AT &&
414 get_valid_blocks(sbi, segno, true) == 0)
418 for (i = 0; i < sbi->segs_per_sec; i++)
419 mtime += get_seg_entry(sbi, start + i)->mtime;
420 mtime = div_u64(mtime, sbi->segs_per_sec);
422 /* Handle if the system time has changed by the user */
423 if (mtime < sit_i->min_mtime)
424 sit_i->min_mtime = mtime;
425 if (mtime > sit_i->max_mtime)
426 sit_i->max_mtime = mtime;
427 if (mtime < sit_i->dirty_min_mtime)
428 sit_i->dirty_min_mtime = mtime;
429 if (mtime > sit_i->dirty_max_mtime)
430 sit_i->dirty_max_mtime = mtime;
432 /* don't choose young section as candidate */
433 if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
436 insert_victim_entry(sbi, mtime, segno);
439 static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi,
440 struct victim_sel_policy *p)
442 struct atgc_management *am = &sbi->am;
443 struct rb_node *parent = NULL;
446 f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most);
451 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
452 struct victim_sel_policy *p)
454 struct sit_info *sit_i = SIT_I(sbi);
455 struct atgc_management *am = &sbi->am;
456 struct rb_root_cached *root = &am->root;
457 struct rb_node *node;
459 struct victim_entry *ve;
460 unsigned long long total_time;
461 unsigned long long age, u, accu;
462 unsigned long long max_mtime = sit_i->dirty_max_mtime;
463 unsigned long long min_mtime = sit_i->dirty_min_mtime;
464 unsigned int sec_blocks = BLKS_PER_SEC(sbi);
465 unsigned int vblocks;
466 unsigned int dirty_threshold = max(am->max_candidate_count,
467 am->candidate_ratio *
468 am->victim_count / 100);
469 unsigned int age_weight = am->age_weight;
471 unsigned int iter = 0;
473 if (max_mtime < min_mtime)
477 total_time = max_mtime - min_mtime;
479 accu = div64_u64(ULLONG_MAX, total_time);
480 accu = min_t(unsigned long long, div_u64(accu, 100),
481 DEFAULT_ACCURACY_CLASS);
483 node = rb_first_cached(root);
485 re = rb_entry_safe(node, struct rb_entry, rb_node);
489 ve = (struct victim_entry *)re;
491 if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
494 /* age = 10000 * x% * 60 */
495 age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
498 vblocks = get_valid_blocks(sbi, ve->segno, true);
499 f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
501 /* u = 10000 * x% * 40 */
502 u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
505 f2fs_bug_on(sbi, age + u >= UINT_MAX);
507 cost = UINT_MAX - (age + u);
510 if (cost < p->min_cost ||
511 (cost == p->min_cost && age > p->oldest_age)) {
514 p->min_segno = ve->segno;
517 if (iter < dirty_threshold) {
518 node = rb_next(node);
524 * select candidates around source section in range of
525 * [target - dirty_threshold, target + dirty_threshold]
527 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
528 struct victim_sel_policy *p)
530 struct sit_info *sit_i = SIT_I(sbi);
531 struct atgc_management *am = &sbi->am;
532 struct rb_node *node;
534 struct victim_entry *ve;
535 unsigned long long age;
536 unsigned long long max_mtime = sit_i->dirty_max_mtime;
537 unsigned long long min_mtime = sit_i->dirty_min_mtime;
538 unsigned int seg_blocks = sbi->blocks_per_seg;
539 unsigned int vblocks;
540 unsigned int dirty_threshold = max(am->max_candidate_count,
541 am->candidate_ratio *
542 am->victim_count / 100);
544 unsigned int iter = 0;
547 if (max_mtime < min_mtime)
551 node = lookup_central_victim(sbi, p);
553 re = rb_entry_safe(node, struct rb_entry, rb_node);
560 ve = (struct victim_entry *)re;
562 if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
565 age = max_mtime - ve->mtime;
567 vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
568 f2fs_bug_on(sbi, !vblocks);
571 if (vblocks == seg_blocks)
576 age = max_mtime - abs(p->age - age);
577 cost = UINT_MAX - vblocks;
579 if (cost < p->min_cost ||
580 (cost == p->min_cost && age > p->oldest_age)) {
583 p->min_segno = ve->segno;
586 if (iter < dirty_threshold) {
588 node = rb_prev(node);
590 node = rb_next(node);
600 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
601 struct victim_sel_policy *p)
603 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
604 &sbi->am.root, true));
606 if (p->gc_mode == GC_AT)
607 atgc_lookup_victim(sbi, p);
608 else if (p->alloc_mode == AT_SSR)
609 atssr_lookup_victim(sbi, p);
614 static void release_victim_entry(struct f2fs_sb_info *sbi)
616 struct atgc_management *am = &sbi->am;
617 struct victim_entry *ve, *tmp;
619 list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
621 kmem_cache_free(victim_entry_slab, ve);
625 am->root = RB_ROOT_CACHED;
627 f2fs_bug_on(sbi, am->victim_count);
628 f2fs_bug_on(sbi, !list_empty(&am->victim_list));
632 * This function is called from two paths.
633 * One is garbage collection and the other is SSR segment selection.
634 * When it is called during GC, it just gets a victim segment
635 * and it does not remove it from dirty seglist.
636 * When it is called from SSR segment selection, it finds a segment
637 * which has minimum valid blocks and removes it from dirty seglist.
639 static int get_victim_by_default(struct f2fs_sb_info *sbi,
640 unsigned int *result, int gc_type, int type,
641 char alloc_mode, unsigned long long age)
643 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
644 struct sit_info *sm = SIT_I(sbi);
645 struct victim_sel_policy p;
646 unsigned int secno, last_victim;
647 unsigned int last_segment;
648 unsigned int nsearched;
652 mutex_lock(&dirty_i->seglist_lock);
653 last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
655 p.alloc_mode = alloc_mode;
657 p.age_threshold = sbi->am.age_threshold;
660 select_policy(sbi, gc_type, type, &p);
661 p.min_segno = NULL_SEGNO;
663 p.min_cost = get_max_cost(sbi, &p);
665 is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
669 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
671 if (*result != NULL_SEGNO) {
672 if (!get_valid_blocks(sbi, *result, false)) {
677 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
680 p.min_segno = *result;
685 if (p.max_search == 0)
688 if (__is_large_section(sbi) && p.alloc_mode == LFS) {
689 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
690 p.min_segno = sbi->next_victim_seg[BG_GC];
691 *result = p.min_segno;
692 sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
695 if (gc_type == FG_GC &&
696 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
697 p.min_segno = sbi->next_victim_seg[FG_GC];
698 *result = p.min_segno;
699 sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
704 last_victim = sm->last_victim[p.gc_mode];
705 if (p.alloc_mode == LFS && gc_type == FG_GC) {
706 p.min_segno = check_bg_victims(sbi);
707 if (p.min_segno != NULL_SEGNO)
712 unsigned long cost, *dirty_bitmap;
713 unsigned int unit_no, segno;
715 dirty_bitmap = p.dirty_bitmap;
716 unit_no = find_next_bit(dirty_bitmap,
717 last_segment / p.ofs_unit,
718 p.offset / p.ofs_unit);
719 segno = unit_no * p.ofs_unit;
720 if (segno >= last_segment) {
721 if (sm->last_victim[p.gc_mode]) {
723 sm->last_victim[p.gc_mode];
724 sm->last_victim[p.gc_mode] = 0;
731 p.offset = segno + p.ofs_unit;
734 #ifdef CONFIG_F2FS_CHECK_FS
736 * skip selecting the invalid segno (that is failed due to block
737 * validity check failure during GC) to avoid endless GC loop in
740 if (test_bit(segno, sm->invalid_segmap))
744 secno = GET_SEC_FROM_SEG(sbi, segno);
746 if (sec_usage_check(sbi, secno))
749 /* Don't touch checkpointed data */
750 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
751 if (p.alloc_mode == LFS) {
753 * LFS is set to find source section during GC.
754 * The victim should have no checkpointed data.
756 if (get_ckpt_valid_blocks(sbi, segno, true))
760 * SSR | AT_SSR are set to find target segment
761 * for writes which can be full by checkpointed
762 * and newly written blocks.
764 if (!f2fs_segment_has_free_slot(sbi, segno))
769 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
773 add_victim_entry(sbi, &p, segno);
777 cost = get_gc_cost(sbi, segno, &p);
779 if (p.min_cost > cost) {
784 if (nsearched >= p.max_search) {
785 if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
786 sm->last_victim[p.gc_mode] =
787 last_victim + p.ofs_unit;
789 sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
790 sm->last_victim[p.gc_mode] %=
791 (MAIN_SECS(sbi) * sbi->segs_per_sec);
796 /* get victim for GC_AT/AT_SSR */
798 lookup_victim_by_age(sbi, &p);
799 release_victim_entry(sbi);
802 if (is_atgc && p.min_segno == NULL_SEGNO &&
803 sm->elapsed_time < p.age_threshold) {
808 if (p.min_segno != NULL_SEGNO) {
810 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
812 if (p.alloc_mode == LFS) {
813 secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
814 if (gc_type == FG_GC)
815 sbi->cur_victim_sec = secno;
817 set_bit(secno, dirty_i->victim_secmap);
823 if (p.min_segno != NULL_SEGNO)
824 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
826 prefree_segments(sbi), free_segments(sbi));
827 mutex_unlock(&dirty_i->seglist_lock);
832 static const struct victim_selection default_v_ops = {
833 .get_victim = get_victim_by_default,
836 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
838 struct inode_entry *ie;
840 ie = radix_tree_lookup(&gc_list->iroot, ino);
846 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
848 struct inode_entry *new_ie;
850 if (inode == find_gc_inode(gc_list, inode->i_ino)) {
854 new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab,
855 GFP_NOFS, true, NULL);
856 new_ie->inode = inode;
858 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
859 list_add_tail(&new_ie->list, &gc_list->ilist);
862 static void put_gc_inode(struct gc_inode_list *gc_list)
864 struct inode_entry *ie, *next_ie;
866 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
867 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
870 kmem_cache_free(f2fs_inode_entry_slab, ie);
874 static int check_valid_map(struct f2fs_sb_info *sbi,
875 unsigned int segno, int offset)
877 struct sit_info *sit_i = SIT_I(sbi);
878 struct seg_entry *sentry;
881 down_read(&sit_i->sentry_lock);
882 sentry = get_seg_entry(sbi, segno);
883 ret = f2fs_test_bit(offset, sentry->cur_valid_map);
884 up_read(&sit_i->sentry_lock);
889 * This function compares node address got in summary with that in NAT.
890 * On validity, copy that node with cold status, otherwise (invalid node)
893 static int gc_node_segment(struct f2fs_sb_info *sbi,
894 struct f2fs_summary *sum, unsigned int segno, int gc_type)
896 struct f2fs_summary *entry;
900 bool fggc = (gc_type == FG_GC);
902 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
904 start_addr = START_BLOCK(sbi, segno);
909 if (fggc && phase == 2)
910 atomic_inc(&sbi->wb_sync_req[NODE]);
912 for (off = 0; off < usable_blks_in_seg; off++, entry++) {
913 nid_t nid = le32_to_cpu(entry->nid);
914 struct page *node_page;
918 /* stop BG_GC if there is not enough free sections. */
919 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
922 if (check_valid_map(sbi, segno, off) == 0)
926 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
932 f2fs_ra_node_page(sbi, nid);
937 node_page = f2fs_get_node_page(sbi, nid);
938 if (IS_ERR(node_page))
941 /* block may become invalid during f2fs_get_node_page */
942 if (check_valid_map(sbi, segno, off) == 0) {
943 f2fs_put_page(node_page, 1);
947 if (f2fs_get_node_info(sbi, nid, &ni)) {
948 f2fs_put_page(node_page, 1);
952 if (ni.blk_addr != start_addr + off) {
953 f2fs_put_page(node_page, 1);
957 err = f2fs_move_node_page(node_page, gc_type);
958 if (!err && gc_type == FG_GC)
960 stat_inc_node_blk_count(sbi, 1, gc_type);
967 atomic_dec(&sbi->wb_sync_req[NODE]);
972 * Calculate start block index indicating the given node offset.
973 * Be careful, caller should give this node offset only indicating direct node
974 * blocks. If any node offsets, which point the other types of node blocks such
975 * as indirect or double indirect node blocks, are given, it must be a caller's
978 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
980 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
988 } else if (node_ofs <= indirect_blks) {
989 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
991 bidx = node_ofs - 2 - dec;
993 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
995 bidx = node_ofs - 5 - dec;
997 return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
1000 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1001 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
1003 struct page *node_page;
1005 unsigned int ofs_in_node;
1006 block_t source_blkaddr;
1008 nid = le32_to_cpu(sum->nid);
1009 ofs_in_node = le16_to_cpu(sum->ofs_in_node);
1011 node_page = f2fs_get_node_page(sbi, nid);
1012 if (IS_ERR(node_page))
1015 if (f2fs_get_node_info(sbi, nid, dni)) {
1016 f2fs_put_page(node_page, 1);
1020 if (sum->version != dni->version) {
1021 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
1023 set_sbi_flag(sbi, SBI_NEED_FSCK);
1026 *nofs = ofs_of_node(node_page);
1027 source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1028 f2fs_put_page(node_page, 1);
1030 if (source_blkaddr != blkaddr) {
1031 #ifdef CONFIG_F2FS_CHECK_FS
1032 unsigned int segno = GET_SEGNO(sbi, blkaddr);
1033 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1035 if (unlikely(check_valid_map(sbi, segno, offset))) {
1036 if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1037 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",
1038 blkaddr, source_blkaddr, segno);
1039 f2fs_bug_on(sbi, 1);
1048 static int ra_data_block(struct inode *inode, pgoff_t index)
1050 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1051 struct address_space *mapping = inode->i_mapping;
1052 struct dnode_of_data dn;
1054 struct extent_info ei = {0, 0, 0};
1055 struct f2fs_io_info fio = {
1057 .ino = inode->i_ino,
1062 .encrypted_page = NULL,
1068 page = f2fs_grab_cache_page(mapping, index, true);
1072 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1073 dn.data_blkaddr = ei.blk + index - ei.fofs;
1074 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1075 DATA_GENERIC_ENHANCE_READ))) {
1076 err = -EFSCORRUPTED;
1082 set_new_dnode(&dn, inode, NULL, NULL, 0);
1083 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1086 f2fs_put_dnode(&dn);
1088 if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1092 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1093 DATA_GENERIC_ENHANCE))) {
1094 err = -EFSCORRUPTED;
1100 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1103 * don't cache encrypted data into meta inode until previous dirty
1104 * data were writebacked to avoid racing between GC and flush.
1106 f2fs_wait_on_page_writeback(page, DATA, true, true);
1108 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1110 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1112 FGP_LOCK | FGP_CREAT, GFP_NOFS);
1113 if (!fio.encrypted_page) {
1118 err = f2fs_submit_page_bio(&fio);
1120 goto put_encrypted_page;
1121 f2fs_put_page(fio.encrypted_page, 0);
1122 f2fs_put_page(page, 1);
1124 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1125 f2fs_update_iostat(sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1129 f2fs_put_page(fio.encrypted_page, 1);
1131 f2fs_put_page(page, 1);
1136 * Move data block via META_MAPPING while keeping locked data page.
1137 * This can be used to move blocks, aka LBAs, directly on disk.
1139 static int move_data_block(struct inode *inode, block_t bidx,
1140 int gc_type, unsigned int segno, int off)
1142 struct f2fs_io_info fio = {
1143 .sbi = F2FS_I_SB(inode),
1144 .ino = inode->i_ino,
1149 .encrypted_page = NULL,
1153 struct dnode_of_data dn;
1154 struct f2fs_summary sum;
1155 struct node_info ni;
1156 struct page *page, *mpage;
1159 bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1160 int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&
1161 (fio.sbi->gc_mode != GC_URGENT_HIGH) ?
1162 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1164 /* do not read out */
1165 page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1169 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1174 if (f2fs_is_atomic_file(inode)) {
1175 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1176 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1181 if (f2fs_is_pinned_file(inode)) {
1182 f2fs_pin_file_control(inode, true);
1187 set_new_dnode(&dn, inode, NULL, NULL, 0);
1188 err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1192 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1193 ClearPageUptodate(page);
1199 * don't cache encrypted data into meta inode until previous dirty
1200 * data were writebacked to avoid racing between GC and flush.
1202 f2fs_wait_on_page_writeback(page, DATA, true, true);
1204 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1206 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1212 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1215 down_write(&fio.sbi->io_order_lock);
1217 mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1218 fio.old_blkaddr, false);
1224 fio.encrypted_page = mpage;
1226 /* read source block in mpage */
1227 if (!PageUptodate(mpage)) {
1228 err = f2fs_submit_page_bio(&fio);
1230 f2fs_put_page(mpage, 1);
1234 f2fs_update_iostat(fio.sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1235 f2fs_update_iostat(fio.sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1238 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1239 !PageUptodate(mpage))) {
1241 f2fs_put_page(mpage, 1);
1246 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1248 /* allocate block address */
1249 f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1252 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1253 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1254 if (!fio.encrypted_page) {
1256 f2fs_put_page(mpage, 1);
1260 /* write target block */
1261 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1262 memcpy(page_address(fio.encrypted_page),
1263 page_address(mpage), PAGE_SIZE);
1264 f2fs_put_page(mpage, 1);
1265 invalidate_mapping_pages(META_MAPPING(fio.sbi),
1266 fio.old_blkaddr, fio.old_blkaddr);
1267 f2fs_invalidate_compress_page(fio.sbi, fio.old_blkaddr);
1269 set_page_dirty(fio.encrypted_page);
1270 if (clear_page_dirty_for_io(fio.encrypted_page))
1271 dec_page_count(fio.sbi, F2FS_DIRTY_META);
1273 set_page_writeback(fio.encrypted_page);
1274 ClearPageError(page);
1276 fio.op = REQ_OP_WRITE;
1277 fio.op_flags = REQ_SYNC;
1278 fio.new_blkaddr = newaddr;
1279 f2fs_submit_page_write(&fio);
1282 if (PageWriteback(fio.encrypted_page))
1283 end_page_writeback(fio.encrypted_page);
1287 f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
1289 f2fs_update_data_blkaddr(&dn, newaddr);
1290 set_inode_flag(inode, FI_APPEND_WRITE);
1291 if (page->index == 0)
1292 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1294 f2fs_put_page(fio.encrypted_page, 1);
1297 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1301 up_write(&fio.sbi->io_order_lock);
1303 f2fs_put_dnode(&dn);
1305 f2fs_put_page(page, 1);
1309 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1310 unsigned int segno, int off)
1315 page = f2fs_get_lock_data_page(inode, bidx, true);
1317 return PTR_ERR(page);
1319 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1324 if (f2fs_is_atomic_file(inode)) {
1325 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1326 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1330 if (f2fs_is_pinned_file(inode)) {
1331 if (gc_type == FG_GC)
1332 f2fs_pin_file_control(inode, true);
1337 if (gc_type == BG_GC) {
1338 if (PageWriteback(page)) {
1342 set_page_dirty(page);
1343 set_page_private_gcing(page);
1345 struct f2fs_io_info fio = {
1346 .sbi = F2FS_I_SB(inode),
1347 .ino = inode->i_ino,
1351 .op_flags = REQ_SYNC,
1352 .old_blkaddr = NULL_ADDR,
1354 .encrypted_page = NULL,
1355 .need_lock = LOCK_REQ,
1356 .io_type = FS_GC_DATA_IO,
1358 bool is_dirty = PageDirty(page);
1361 f2fs_wait_on_page_writeback(page, DATA, true, true);
1363 set_page_dirty(page);
1364 if (clear_page_dirty_for_io(page)) {
1365 inode_dec_dirty_pages(inode);
1366 f2fs_remove_dirty_inode(inode);
1369 set_page_private_gcing(page);
1371 err = f2fs_do_write_data_page(&fio);
1373 clear_page_private_gcing(page);
1374 if (err == -ENOMEM) {
1375 congestion_wait(BLK_RW_ASYNC,
1376 DEFAULT_IO_TIMEOUT);
1380 set_page_dirty(page);
1384 f2fs_put_page(page, 1);
1389 * This function tries to get parent node of victim data block, and identifies
1390 * data block validity. If the block is valid, copy that with cold status and
1391 * modify parent node.
1392 * If the parent node is not valid or the data block address is different,
1393 * the victim data block is ignored.
1395 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1396 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1399 struct super_block *sb = sbi->sb;
1400 struct f2fs_summary *entry;
1405 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1407 start_addr = START_BLOCK(sbi, segno);
1412 for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1413 struct page *data_page;
1414 struct inode *inode;
1415 struct node_info dni; /* dnode info for the data */
1416 unsigned int ofs_in_node, nofs;
1418 nid_t nid = le32_to_cpu(entry->nid);
1421 * stop BG_GC if there is not enough free sections.
1422 * Or, stop GC if the segment becomes fully valid caused by
1423 * race condition along with SSR block allocation.
1425 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1426 (!force_migrate && get_valid_blocks(sbi, segno, true) ==
1430 if (check_valid_map(sbi, segno, off) == 0)
1434 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1440 f2fs_ra_node_page(sbi, nid);
1444 /* Get an inode by ino with checking validity */
1445 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1449 f2fs_ra_node_page(sbi, dni.ino);
1453 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1456 inode = f2fs_iget(sb, dni.ino);
1457 if (IS_ERR(inode) || is_bad_inode(inode))
1460 if (!down_write_trylock(
1461 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1463 sbi->skipped_gc_rwsem++;
1467 start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1470 if (f2fs_post_read_required(inode)) {
1471 int err = ra_data_block(inode, start_bidx);
1473 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1478 add_gc_inode(gc_list, inode);
1482 data_page = f2fs_get_read_data_page(inode,
1483 start_bidx, REQ_RAHEAD, true);
1484 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1485 if (IS_ERR(data_page)) {
1490 f2fs_put_page(data_page, 0);
1491 add_gc_inode(gc_list, inode);
1496 inode = find_gc_inode(gc_list, dni.ino);
1498 struct f2fs_inode_info *fi = F2FS_I(inode);
1499 bool locked = false;
1502 if (S_ISREG(inode->i_mode)) {
1503 if (!down_write_trylock(&fi->i_gc_rwsem[READ])) {
1504 sbi->skipped_gc_rwsem++;
1507 if (!down_write_trylock(
1508 &fi->i_gc_rwsem[WRITE])) {
1509 sbi->skipped_gc_rwsem++;
1510 up_write(&fi->i_gc_rwsem[READ]);
1515 /* wait for all inflight aio data */
1516 inode_dio_wait(inode);
1519 start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1521 if (f2fs_post_read_required(inode))
1522 err = move_data_block(inode, start_bidx,
1523 gc_type, segno, off);
1525 err = move_data_page(inode, start_bidx, gc_type,
1528 if (!err && (gc_type == FG_GC ||
1529 f2fs_post_read_required(inode)))
1533 up_write(&fi->i_gc_rwsem[WRITE]);
1534 up_write(&fi->i_gc_rwsem[READ]);
1537 stat_inc_data_blk_count(sbi, 1, gc_type);
1547 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1550 struct sit_info *sit_i = SIT_I(sbi);
1553 down_write(&sit_i->sentry_lock);
1554 ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1555 NO_CHECK_TYPE, LFS, 0);
1556 up_write(&sit_i->sentry_lock);
1560 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1561 unsigned int start_segno,
1562 struct gc_inode_list *gc_list, int gc_type,
1565 struct page *sum_page;
1566 struct f2fs_summary_block *sum;
1567 struct blk_plug plug;
1568 unsigned int segno = start_segno;
1569 unsigned int end_segno = start_segno + sbi->segs_per_sec;
1570 int seg_freed = 0, migrated = 0;
1571 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1572 SUM_TYPE_DATA : SUM_TYPE_NODE;
1575 if (__is_large_section(sbi))
1576 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1579 * zone-capacity can be less than zone-size in zoned devices,
1580 * resulting in less than expected usable segments in the zone,
1581 * calculate the end segno in the zone which can be garbage collected
1583 if (f2fs_sb_has_blkzoned(sbi))
1584 end_segno -= sbi->segs_per_sec -
1585 f2fs_usable_segs_in_sec(sbi, segno);
1587 sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1589 /* readahead multi ssa blocks those have contiguous address */
1590 if (__is_large_section(sbi))
1591 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1592 end_segno - segno, META_SSA, true);
1594 /* reference all summary page */
1595 while (segno < end_segno) {
1596 sum_page = f2fs_get_sum_page(sbi, segno++);
1597 if (IS_ERR(sum_page)) {
1598 int err = PTR_ERR(sum_page);
1600 end_segno = segno - 1;
1601 for (segno = start_segno; segno < end_segno; segno++) {
1602 sum_page = find_get_page(META_MAPPING(sbi),
1603 GET_SUM_BLOCK(sbi, segno));
1604 f2fs_put_page(sum_page, 0);
1605 f2fs_put_page(sum_page, 0);
1609 unlock_page(sum_page);
1612 blk_start_plug(&plug);
1614 for (segno = start_segno; segno < end_segno; segno++) {
1616 /* find segment summary of victim */
1617 sum_page = find_get_page(META_MAPPING(sbi),
1618 GET_SUM_BLOCK(sbi, segno));
1619 f2fs_put_page(sum_page, 0);
1621 if (get_valid_blocks(sbi, segno, false) == 0)
1623 if (gc_type == BG_GC && __is_large_section(sbi) &&
1624 migrated >= sbi->migration_granularity)
1626 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1629 sum = page_address(sum_page);
1630 if (type != GET_SUM_TYPE((&sum->footer))) {
1631 f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1632 segno, type, GET_SUM_TYPE((&sum->footer)));
1633 set_sbi_flag(sbi, SBI_NEED_FSCK);
1634 f2fs_stop_checkpoint(sbi, false);
1639 * this is to avoid deadlock:
1640 * - lock_page(sum_page) - f2fs_replace_block
1641 * - check_valid_map() - down_write(sentry_lock)
1642 * - down_read(sentry_lock) - change_curseg()
1643 * - lock_page(sum_page)
1645 if (type == SUM_TYPE_NODE)
1646 submitted += gc_node_segment(sbi, sum->entries, segno,
1649 submitted += gc_data_segment(sbi, sum->entries, gc_list,
1653 stat_inc_seg_count(sbi, type, gc_type);
1654 sbi->gc_reclaimed_segs[sbi->gc_mode]++;
1658 if (gc_type == FG_GC &&
1659 get_valid_blocks(sbi, segno, false) == 0)
1662 if (__is_large_section(sbi) && segno + 1 < end_segno)
1663 sbi->next_victim_seg[gc_type] = segno + 1;
1665 f2fs_put_page(sum_page, 0);
1669 f2fs_submit_merged_write(sbi,
1670 (type == SUM_TYPE_NODE) ? NODE : DATA);
1672 blk_finish_plug(&plug);
1674 stat_inc_call_count(sbi->stat_info);
1679 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1680 bool background, bool force, unsigned int segno)
1682 int gc_type = sync ? FG_GC : BG_GC;
1683 int sec_freed = 0, seg_freed = 0, total_freed = 0;
1685 struct cp_control cpc;
1686 unsigned int init_segno = segno;
1687 struct gc_inode_list gc_list = {
1688 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1689 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1691 unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1692 unsigned long long first_skipped;
1693 unsigned int skipped_round = 0, round = 0;
1695 trace_f2fs_gc_begin(sbi->sb, sync, background,
1696 get_pages(sbi, F2FS_DIRTY_NODES),
1697 get_pages(sbi, F2FS_DIRTY_DENTS),
1698 get_pages(sbi, F2FS_DIRTY_IMETA),
1701 reserved_segments(sbi),
1702 prefree_segments(sbi));
1704 cpc.reason = __get_cp_reason(sbi);
1705 sbi->skipped_gc_rwsem = 0;
1706 first_skipped = last_skipped;
1708 if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1712 if (unlikely(f2fs_cp_error(sbi))) {
1717 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1719 * For example, if there are many prefree_segments below given
1720 * threshold, we can make them free by checkpoint. Then, we
1721 * secure free segments which doesn't need fggc any more.
1723 if (prefree_segments(sbi) &&
1724 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1725 ret = f2fs_write_checkpoint(sbi, &cpc);
1729 if (has_not_enough_free_secs(sbi, 0, 0))
1733 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1734 if (gc_type == BG_GC && !background) {
1738 ret = __get_victim(sbi, &segno, gc_type);
1742 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type, force);
1743 if (gc_type == FG_GC &&
1744 seg_freed == f2fs_usable_segs_in_sec(sbi, segno))
1746 total_freed += seg_freed;
1748 if (gc_type == FG_GC) {
1749 if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
1750 sbi->skipped_gc_rwsem)
1752 last_skipped = sbi->skipped_atomic_files[FG_GC];
1756 if (gc_type == FG_GC)
1757 sbi->cur_victim_sec = NULL_SEGNO;
1762 if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1763 if (skipped_round <= MAX_SKIP_GC_COUNT ||
1764 skipped_round * 2 < round) {
1769 if (first_skipped < last_skipped &&
1770 (last_skipped - first_skipped) >
1771 sbi->skipped_gc_rwsem) {
1772 f2fs_drop_inmem_pages_all(sbi, true);
1776 if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1777 ret = f2fs_write_checkpoint(sbi, &cpc);
1780 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1781 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1783 trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1784 get_pages(sbi, F2FS_DIRTY_NODES),
1785 get_pages(sbi, F2FS_DIRTY_DENTS),
1786 get_pages(sbi, F2FS_DIRTY_IMETA),
1789 reserved_segments(sbi),
1790 prefree_segments(sbi));
1792 up_write(&sbi->gc_lock);
1794 put_gc_inode(&gc_list);
1797 ret = sec_freed ? 0 : -EAGAIN;
1801 int __init f2fs_create_garbage_collection_cache(void)
1803 victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1804 sizeof(struct victim_entry));
1805 if (!victim_entry_slab)
1810 void f2fs_destroy_garbage_collection_cache(void)
1812 kmem_cache_destroy(victim_entry_slab);
1815 static void init_atgc_management(struct f2fs_sb_info *sbi)
1817 struct atgc_management *am = &sbi->am;
1819 if (test_opt(sbi, ATGC) &&
1820 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1821 am->atgc_enabled = true;
1823 am->root = RB_ROOT_CACHED;
1824 INIT_LIST_HEAD(&am->victim_list);
1825 am->victim_count = 0;
1827 am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1828 am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1829 am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1830 am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
1833 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1835 DIRTY_I(sbi)->v_ops = &default_v_ops;
1837 sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1839 /* give warm/cold data area from slower device */
1840 if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1841 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1842 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1844 init_atgc_management(sbi);
1847 static int free_segment_range(struct f2fs_sb_info *sbi,
1848 unsigned int secs, bool gc_only)
1850 unsigned int segno, next_inuse, start, end;
1851 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1852 int gc_mode, gc_type;
1856 /* Force block allocation for GC */
1857 MAIN_SECS(sbi) -= secs;
1858 start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1859 end = MAIN_SEGS(sbi) - 1;
1861 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1862 for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1863 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1864 SIT_I(sbi)->last_victim[gc_mode] = 0;
1866 for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1867 if (sbi->next_victim_seg[gc_type] >= start)
1868 sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1869 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1871 /* Move out cursegs from the target range */
1872 for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1873 f2fs_allocate_segment_for_resize(sbi, type, start, end);
1875 /* do GC to move out valid blocks in the range */
1876 for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1877 struct gc_inode_list gc_list = {
1878 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1879 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1882 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
1883 put_gc_inode(&gc_list);
1885 if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1889 if (fatal_signal_pending(current)) {
1897 err = f2fs_write_checkpoint(sbi, &cpc);
1901 next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1902 if (next_inuse <= end) {
1903 f2fs_err(sbi, "segno %u should be free but still inuse!",
1905 f2fs_bug_on(sbi, 1);
1908 MAIN_SECS(sbi) += secs;
1912 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1914 struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1917 int segment_count_main;
1918 long long block_count;
1919 int segs = secs * sbi->segs_per_sec;
1921 down_write(&sbi->sb_lock);
1923 section_count = le32_to_cpu(raw_sb->section_count);
1924 segment_count = le32_to_cpu(raw_sb->segment_count);
1925 segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
1926 block_count = le64_to_cpu(raw_sb->block_count);
1928 raw_sb->section_count = cpu_to_le32(section_count + secs);
1929 raw_sb->segment_count = cpu_to_le32(segment_count + segs);
1930 raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
1931 raw_sb->block_count = cpu_to_le64(block_count +
1932 (long long)segs * sbi->blocks_per_seg);
1933 if (f2fs_is_multi_device(sbi)) {
1934 int last_dev = sbi->s_ndevs - 1;
1936 le32_to_cpu(raw_sb->devs[last_dev].total_segments);
1938 raw_sb->devs[last_dev].total_segments =
1939 cpu_to_le32(dev_segs + segs);
1942 up_write(&sbi->sb_lock);
1945 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
1947 int segs = secs * sbi->segs_per_sec;
1948 long long blks = (long long)segs * sbi->blocks_per_seg;
1949 long long user_block_count =
1950 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
1952 SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
1953 MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
1954 MAIN_SECS(sbi) += secs;
1955 FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
1956 FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
1957 F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
1959 if (f2fs_is_multi_device(sbi)) {
1960 int last_dev = sbi->s_ndevs - 1;
1962 FDEV(last_dev).total_segments =
1963 (int)FDEV(last_dev).total_segments + segs;
1964 FDEV(last_dev).end_blk =
1965 (long long)FDEV(last_dev).end_blk + blks;
1966 #ifdef CONFIG_BLK_DEV_ZONED
1967 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
1968 (int)(blks >> sbi->log_blocks_per_blkz);
1973 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
1975 __u64 old_block_count, shrunk_blocks;
1976 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1981 old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
1982 if (block_count > old_block_count)
1985 if (f2fs_is_multi_device(sbi)) {
1986 int last_dev = sbi->s_ndevs - 1;
1987 __u64 last_segs = FDEV(last_dev).total_segments;
1989 if (block_count + last_segs * sbi->blocks_per_seg <=
1994 /* new fs size should align to section size */
1995 div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
1999 if (block_count == old_block_count)
2002 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2003 f2fs_err(sbi, "Should run fsck to repair first.");
2004 return -EFSCORRUPTED;
2007 if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2008 f2fs_err(sbi, "Checkpoint should be enabled.");
2012 shrunk_blocks = old_block_count - block_count;
2013 secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2016 if (!down_write_trylock(&sbi->gc_lock))
2019 /* stop CP to protect MAIN_SEC in free_segment_range */
2022 spin_lock(&sbi->stat_lock);
2023 if (shrunk_blocks + valid_user_blocks(sbi) +
2024 sbi->current_reserved_blocks + sbi->unusable_block_count +
2025 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2027 spin_unlock(&sbi->stat_lock);
2032 err = free_segment_range(sbi, secs, true);
2035 f2fs_unlock_op(sbi);
2036 up_write(&sbi->gc_lock);
2040 set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2042 freeze_super(sbi->sb);
2043 down_write(&sbi->gc_lock);
2044 down_write(&sbi->cp_global_sem);
2046 spin_lock(&sbi->stat_lock);
2047 if (shrunk_blocks + valid_user_blocks(sbi) +
2048 sbi->current_reserved_blocks + sbi->unusable_block_count +
2049 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2052 sbi->user_block_count -= shrunk_blocks;
2053 spin_unlock(&sbi->stat_lock);
2057 err = free_segment_range(sbi, secs, false);
2061 update_sb_metadata(sbi, -secs);
2063 err = f2fs_commit_super(sbi, false);
2065 update_sb_metadata(sbi, secs);
2069 update_fs_metadata(sbi, -secs);
2070 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2071 set_sbi_flag(sbi, SBI_IS_DIRTY);
2073 err = f2fs_write_checkpoint(sbi, &cpc);
2075 update_fs_metadata(sbi, secs);
2076 update_sb_metadata(sbi, secs);
2077 f2fs_commit_super(sbi, false);
2081 set_sbi_flag(sbi, SBI_NEED_FSCK);
2082 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2084 spin_lock(&sbi->stat_lock);
2085 sbi->user_block_count += shrunk_blocks;
2086 spin_unlock(&sbi->stat_lock);
2089 up_write(&sbi->cp_global_sem);
2090 up_write(&sbi->gc_lock);
2091 thaw_super(sbi->sb);
2092 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);