xref: /kernel/linux/linux-5.10/fs/f2fs/gc.c (revision 8c2ecf20)
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * fs/f2fs/gc.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 *             http://www.samsung.com/
7 */
8#include <linux/fs.h>
9#include <linux/module.h>
10#include <linux/mount.h>
11#include <linux/backing-dev.h>
12#include <linux/init.h>
13#include <linux/f2fs_fs.h>
14#include <linux/kthread.h>
15#include <linux/delay.h>
16#include <linux/freezer.h>
17#include <linux/sched/signal.h>
18
19#include "f2fs.h"
20#include "node.h"
21#include "segment.h"
22#include "gc.h"
23#include <trace/events/f2fs.h>
24
25static struct kmem_cache *victim_entry_slab;
26
27static unsigned int count_bits(const unsigned long *addr,
28				unsigned int offset, unsigned int len);
29
30static int gc_thread_func(void *data)
31{
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;
36	unsigned int wait_ms;
37
38	wait_ms = gc_th->min_sleep_time;
39
40	set_freezable();
41	do {
42		bool sync_mode, foreground = false;
43
44		wait_event_interruptible_timeout(*wq,
45				kthread_should_stop() || freezing(current) ||
46				waitqueue_active(fggc_wq) ||
47				gc_th->gc_wake,
48				msecs_to_jiffies(wait_ms));
49
50		if (test_opt(sbi, GC_MERGE) && waitqueue_active(fggc_wq))
51			foreground = true;
52
53		/* give it a try one time */
54		if (gc_th->gc_wake)
55			gc_th->gc_wake = 0;
56
57		if (try_to_freeze()) {
58			stat_other_skip_bggc_count(sbi);
59			continue;
60		}
61		if (kthread_should_stop())
62			break;
63
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);
67			continue;
68		}
69
70		if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
71			f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
72			f2fs_stop_checkpoint(sbi, false);
73		}
74
75		if (!sb_start_write_trylock(sbi->sb)) {
76			stat_other_skip_bggc_count(sbi);
77			continue;
78		}
79
80		/*
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.
87		 *
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.
92		 */
93		if (sbi->gc_mode == GC_URGENT_HIGH) {
94			wait_ms = gc_th->urgent_sleep_time;
95			down_write(&sbi->gc_lock);
96			goto do_gc;
97		}
98
99		if (foreground) {
100			down_write(&sbi->gc_lock);
101			goto do_gc;
102		} else if (!down_write_trylock(&sbi->gc_lock)) {
103			stat_other_skip_bggc_count(sbi);
104			goto next;
105		}
106
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);
111			goto next;
112		}
113
114		if (has_enough_invalid_blocks(sbi))
115			decrease_sleep_time(gc_th, &wait_ms);
116		else
117			increase_sleep_time(gc_th, &wait_ms);
118do_gc:
119		if (!foreground)
120			stat_inc_bggc_count(sbi->stat_info);
121
122		sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
123
124		/* foreground GC was been triggered via f2fs_balance_fs() */
125		if (foreground)
126			sync_mode = false;
127
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;
131
132		if (foreground)
133			wake_up_all(&gc_th->fggc_wq);
134
135		trace_f2fs_background_gc(sbi->sb, wait_ms,
136				prefree_segments(sbi), free_segments(sbi));
137
138		/* balancing f2fs's metadata periodically */
139		f2fs_balance_fs_bg(sbi, true);
140next:
141		sb_end_write(sbi->sb);
142
143	} while (!kthread_should_stop());
144	return 0;
145}
146
147int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
148{
149	struct f2fs_gc_kthread *gc_th;
150	dev_t dev = sbi->sb->s_bdev->bd_dev;
151	int err = 0;
152
153	gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
154	if (!gc_th) {
155		err = -ENOMEM;
156		goto out;
157	}
158
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;
163
164	gc_th->gc_wake= 0;
165
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);
173		kfree(gc_th);
174		sbi->gc_thread = NULL;
175	}
176out:
177	return err;
178}
179
180void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
181{
182	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
183	if (!gc_th)
184		return;
185	kthread_stop(gc_th->f2fs_gc_task);
186	wake_up_all(&gc_th->fggc_wq);
187	kfree(gc_th);
188	sbi->gc_thread = NULL;
189}
190
191static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
192{
193	int gc_mode;
194
195	if (gc_type == BG_GC) {
196		if (sbi->am.atgc_enabled)
197			gc_mode = GC_AT;
198		else
199			gc_mode = GC_CB;
200	} else {
201		gc_mode = GC_GREEDY;
202	}
203
204	switch (sbi->gc_mode) {
205	case GC_IDLE_CB:
206		gc_mode = GC_CB;
207		break;
208	case GC_IDLE_GREEDY:
209	case GC_URGENT_HIGH:
210		gc_mode = GC_GREEDY;
211		break;
212	case GC_IDLE_AT:
213		gc_mode = GC_AT;
214		break;
215	}
216
217	return gc_mode;
218}
219
220static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
221			int type, struct victim_sel_policy *p)
222{
223	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
224
225	if (p->alloc_mode == SSR) {
226		p->gc_mode = GC_GREEDY;
227		p->dirty_bitmap = dirty_i->dirty_segmap[type];
228		p->max_search = dirty_i->nr_dirty[type];
229		p->ofs_unit = 1;
230	} else if (p->alloc_mode == AT_SSR) {
231		p->gc_mode = GC_GREEDY;
232		p->dirty_bitmap = dirty_i->dirty_segmap[type];
233		p->max_search = dirty_i->nr_dirty[type];
234		p->ofs_unit = 1;
235	} else {
236		p->gc_mode = select_gc_type(sbi, gc_type);
237		p->ofs_unit = sbi->segs_per_sec;
238		if (__is_large_section(sbi)) {
239			p->dirty_bitmap = dirty_i->dirty_secmap;
240			p->max_search = count_bits(p->dirty_bitmap,
241						0, MAIN_SECS(sbi));
242		} else {
243			p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
244			p->max_search = dirty_i->nr_dirty[DIRTY];
245		}
246	}
247
248	/*
249	 * adjust candidates range, should select all dirty segments for
250	 * foreground GC and urgent GC cases.
251	 */
252	if (gc_type != FG_GC &&
253			(sbi->gc_mode != GC_URGENT_HIGH) &&
254			(p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
255			p->max_search > sbi->max_victim_search)
256		p->max_search = sbi->max_victim_search;
257
258	/* let's select beginning hot/small space first in no_heap mode*/
259	if (test_opt(sbi, NOHEAP) &&
260		(type == CURSEG_HOT_DATA || IS_NODESEG(type)))
261		p->offset = 0;
262	else
263		p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
264}
265
266static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
267				struct victim_sel_policy *p)
268{
269	/* SSR allocates in a segment unit */
270	if (p->alloc_mode == SSR)
271		return sbi->blocks_per_seg;
272	else if (p->alloc_mode == AT_SSR)
273		return UINT_MAX;
274
275	/* LFS */
276	if (p->gc_mode == GC_GREEDY)
277		return 2 * sbi->blocks_per_seg * p->ofs_unit;
278	else if (p->gc_mode == GC_CB)
279		return UINT_MAX;
280	else if (p->gc_mode == GC_AT)
281		return UINT_MAX;
282	else /* No other gc_mode */
283		return 0;
284}
285
286static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
287{
288	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
289	unsigned int secno;
290
291	/*
292	 * If the gc_type is FG_GC, we can select victim segments
293	 * selected by background GC before.
294	 * Those segments guarantee they have small valid blocks.
295	 */
296	for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
297		if (sec_usage_check(sbi, secno))
298			continue;
299		clear_bit(secno, dirty_i->victim_secmap);
300		return GET_SEG_FROM_SEC(sbi, secno);
301	}
302	return NULL_SEGNO;
303}
304
305static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
306{
307	struct sit_info *sit_i = SIT_I(sbi);
308	unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
309	unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
310	unsigned long long mtime = 0;
311	unsigned int vblocks;
312	unsigned char age = 0;
313	unsigned char u;
314	unsigned int i;
315	unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
316
317	for (i = 0; i < usable_segs_per_sec; i++)
318		mtime += get_seg_entry(sbi, start + i)->mtime;
319	vblocks = get_valid_blocks(sbi, segno, true);
320
321	mtime = div_u64(mtime, usable_segs_per_sec);
322	vblocks = div_u64(vblocks, usable_segs_per_sec);
323
324	u = (vblocks * 100) >> sbi->log_blocks_per_seg;
325
326	/* Handle if the system time has changed by the user */
327	if (mtime < sit_i->min_mtime)
328		sit_i->min_mtime = mtime;
329	if (mtime > sit_i->max_mtime)
330		sit_i->max_mtime = mtime;
331	if (sit_i->max_mtime != sit_i->min_mtime)
332		age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
333				sit_i->max_mtime - sit_i->min_mtime);
334
335	return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
336}
337
338static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
339			unsigned int segno, struct victim_sel_policy *p)
340{
341	if (p->alloc_mode == SSR)
342		return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
343
344	/* alloc_mode == LFS */
345	if (p->gc_mode == GC_GREEDY)
346		return get_valid_blocks(sbi, segno, true);
347	else if (p->gc_mode == GC_CB)
348		return get_cb_cost(sbi, segno);
349
350	f2fs_bug_on(sbi, 1);
351	return 0;
352}
353
354static unsigned int count_bits(const unsigned long *addr,
355				unsigned int offset, unsigned int len)
356{
357	unsigned int end = offset + len, sum = 0;
358
359	while (offset < end) {
360		if (test_bit(offset++, addr))
361			++sum;
362	}
363	return sum;
364}
365
366static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi,
367				unsigned long long mtime, unsigned int segno,
368				struct rb_node *parent, struct rb_node **p,
369				bool left_most)
370{
371	struct atgc_management *am = &sbi->am;
372	struct victim_entry *ve;
373
374	ve =  f2fs_kmem_cache_alloc(victim_entry_slab, GFP_NOFS);
375
376	ve->mtime = mtime;
377	ve->segno = segno;
378
379	rb_link_node(&ve->rb_node, parent, p);
380	rb_insert_color_cached(&ve->rb_node, &am->root, left_most);
381
382	list_add_tail(&ve->list, &am->victim_list);
383
384	am->victim_count++;
385
386	return ve;
387}
388
389static void insert_victim_entry(struct f2fs_sb_info *sbi,
390				unsigned long long mtime, unsigned int segno)
391{
392	struct atgc_management *am = &sbi->am;
393	struct rb_node **p;
394	struct rb_node *parent = NULL;
395	bool left_most = true;
396
397	p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most);
398	attach_victim_entry(sbi, mtime, segno, parent, p, left_most);
399}
400
401static void add_victim_entry(struct f2fs_sb_info *sbi,
402				struct victim_sel_policy *p, unsigned int segno)
403{
404	struct sit_info *sit_i = SIT_I(sbi);
405	unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
406	unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
407	unsigned long long mtime = 0;
408	unsigned int i;
409
410	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
411		if (p->gc_mode == GC_AT &&
412			get_valid_blocks(sbi, segno, true) == 0)
413			return;
414	}
415
416	for (i = 0; i < sbi->segs_per_sec; i++)
417		mtime += get_seg_entry(sbi, start + i)->mtime;
418	mtime = div_u64(mtime, sbi->segs_per_sec);
419
420	/* Handle if the system time has changed by the user */
421	if (mtime < sit_i->min_mtime)
422		sit_i->min_mtime = mtime;
423	if (mtime > sit_i->max_mtime)
424		sit_i->max_mtime = mtime;
425	if (mtime < sit_i->dirty_min_mtime)
426		sit_i->dirty_min_mtime = mtime;
427	if (mtime > sit_i->dirty_max_mtime)
428		sit_i->dirty_max_mtime = mtime;
429
430	/* don't choose young section as candidate */
431	if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
432		return;
433
434	insert_victim_entry(sbi, mtime, segno);
435}
436
437static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi,
438						struct victim_sel_policy *p)
439{
440	struct atgc_management *am = &sbi->am;
441	struct rb_node *parent = NULL;
442	bool left_most;
443
444	f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most);
445
446	return parent;
447}
448
449static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
450						struct victim_sel_policy *p)
451{
452	struct sit_info *sit_i = SIT_I(sbi);
453	struct atgc_management *am = &sbi->am;
454	struct rb_root_cached *root = &am->root;
455	struct rb_node *node;
456	struct rb_entry *re;
457	struct victim_entry *ve;
458	unsigned long long total_time;
459	unsigned long long age, u, accu;
460	unsigned long long max_mtime = sit_i->dirty_max_mtime;
461	unsigned long long min_mtime = sit_i->dirty_min_mtime;
462	unsigned int sec_blocks = BLKS_PER_SEC(sbi);
463	unsigned int vblocks;
464	unsigned int dirty_threshold = max(am->max_candidate_count,
465					am->candidate_ratio *
466					am->victim_count / 100);
467	unsigned int age_weight = am->age_weight;
468	unsigned int cost;
469	unsigned int iter = 0;
470
471	if (max_mtime < min_mtime)
472		return;
473
474	max_mtime += 1;
475	total_time = max_mtime - min_mtime;
476
477	accu = div64_u64(ULLONG_MAX, total_time);
478	accu = min_t(unsigned long long, div_u64(accu, 100),
479					DEFAULT_ACCURACY_CLASS);
480
481	node = rb_first_cached(root);
482next:
483	re = rb_entry_safe(node, struct rb_entry, rb_node);
484	if (!re)
485		return;
486
487	ve = (struct victim_entry *)re;
488
489	if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
490		goto skip;
491
492	/* age = 10000 * x% * 60 */
493	age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
494								age_weight;
495
496	vblocks = get_valid_blocks(sbi, ve->segno, true);
497	f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
498
499	/* u = 10000 * x% * 40 */
500	u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
501							(100 - age_weight);
502
503	f2fs_bug_on(sbi, age + u >= UINT_MAX);
504
505	cost = UINT_MAX - (age + u);
506	iter++;
507
508	if (cost < p->min_cost ||
509			(cost == p->min_cost && age > p->oldest_age)) {
510		p->min_cost = cost;
511		p->oldest_age = age;
512		p->min_segno = ve->segno;
513	}
514skip:
515	if (iter < dirty_threshold) {
516		node = rb_next(node);
517		goto next;
518	}
519}
520
521/*
522 * select candidates around source section in range of
523 * [target - dirty_threshold, target + dirty_threshold]
524 */
525static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
526						struct victim_sel_policy *p)
527{
528	struct sit_info *sit_i = SIT_I(sbi);
529	struct atgc_management *am = &sbi->am;
530	struct rb_node *node;
531	struct rb_entry *re;
532	struct victim_entry *ve;
533	unsigned long long age;
534	unsigned long long max_mtime = sit_i->dirty_max_mtime;
535	unsigned long long min_mtime = sit_i->dirty_min_mtime;
536	unsigned int seg_blocks = sbi->blocks_per_seg;
537	unsigned int vblocks;
538	unsigned int dirty_threshold = max(am->max_candidate_count,
539					am->candidate_ratio *
540					am->victim_count / 100);
541	unsigned int cost;
542	unsigned int iter = 0;
543	int stage = 0;
544
545	if (max_mtime < min_mtime)
546		return;
547	max_mtime += 1;
548next_stage:
549	node = lookup_central_victim(sbi, p);
550next_node:
551	re = rb_entry_safe(node, struct rb_entry, rb_node);
552	if (!re) {
553		if (stage == 0)
554			goto skip_stage;
555		return;
556	}
557
558	ve = (struct victim_entry *)re;
559
560	if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
561		goto skip_node;
562
563	age = max_mtime - ve->mtime;
564
565	vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
566	f2fs_bug_on(sbi, !vblocks);
567
568	/* rare case */
569	if (vblocks == seg_blocks)
570		goto skip_node;
571
572	iter++;
573
574	age = max_mtime - abs(p->age - age);
575	cost = UINT_MAX - vblocks;
576
577	if (cost < p->min_cost ||
578			(cost == p->min_cost && age > p->oldest_age)) {
579		p->min_cost = cost;
580		p->oldest_age = age;
581		p->min_segno = ve->segno;
582	}
583skip_node:
584	if (iter < dirty_threshold) {
585		if (stage == 0)
586			node = rb_prev(node);
587		else if (stage == 1)
588			node = rb_next(node);
589		goto next_node;
590	}
591skip_stage:
592	if (stage < 1) {
593		stage++;
594		iter = 0;
595		goto next_stage;
596	}
597}
598static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
599						struct victim_sel_policy *p)
600{
601	f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
602						&sbi->am.root, true));
603
604	if (p->gc_mode == GC_AT)
605		atgc_lookup_victim(sbi, p);
606	else if (p->alloc_mode == AT_SSR)
607		atssr_lookup_victim(sbi, p);
608	else
609		f2fs_bug_on(sbi, 1);
610}
611
612static void release_victim_entry(struct f2fs_sb_info *sbi)
613{
614	struct atgc_management *am = &sbi->am;
615	struct victim_entry *ve, *tmp;
616
617	list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
618		list_del(&ve->list);
619		kmem_cache_free(victim_entry_slab, ve);
620		am->victim_count--;
621	}
622
623	am->root = RB_ROOT_CACHED;
624
625	f2fs_bug_on(sbi, am->victim_count);
626	f2fs_bug_on(sbi, !list_empty(&am->victim_list));
627}
628
629/*
630 * This function is called from two paths.
631 * One is garbage collection and the other is SSR segment selection.
632 * When it is called during GC, it just gets a victim segment
633 * and it does not remove it from dirty seglist.
634 * When it is called from SSR segment selection, it finds a segment
635 * which has minimum valid blocks and removes it from dirty seglist.
636 */
637static int get_victim_by_default(struct f2fs_sb_info *sbi,
638			unsigned int *result, int gc_type, int type,
639			char alloc_mode, unsigned long long age)
640{
641	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
642	struct sit_info *sm = SIT_I(sbi);
643	struct victim_sel_policy p;
644	unsigned int secno, last_victim;
645	unsigned int last_segment;
646	unsigned int nsearched;
647	bool is_atgc;
648	int ret = 0;
649
650	mutex_lock(&dirty_i->seglist_lock);
651	last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
652
653	p.alloc_mode = alloc_mode;
654	p.age = age;
655	p.age_threshold = sbi->am.age_threshold;
656
657retry:
658	select_policy(sbi, gc_type, type, &p);
659	p.min_segno = NULL_SEGNO;
660	p.oldest_age = 0;
661	p.min_cost = get_max_cost(sbi, &p);
662
663	is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
664	nsearched = 0;
665
666	if (is_atgc)
667		SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
668
669	if (*result != NULL_SEGNO) {
670		if (!get_valid_blocks(sbi, *result, false)) {
671			ret = -ENODATA;
672			goto out;
673		}
674
675		if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
676			ret = -EBUSY;
677		else
678			p.min_segno = *result;
679		goto out;
680	}
681
682	ret = -ENODATA;
683	if (p.max_search == 0)
684		goto out;
685
686	if (__is_large_section(sbi) && p.alloc_mode == LFS) {
687		if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
688			p.min_segno = sbi->next_victim_seg[BG_GC];
689			*result = p.min_segno;
690			sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
691			goto got_result;
692		}
693		if (gc_type == FG_GC &&
694				sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
695			p.min_segno = sbi->next_victim_seg[FG_GC];
696			*result = p.min_segno;
697			sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
698			goto got_result;
699		}
700	}
701
702	last_victim = sm->last_victim[p.gc_mode];
703	if (p.alloc_mode == LFS && gc_type == FG_GC) {
704		p.min_segno = check_bg_victims(sbi);
705		if (p.min_segno != NULL_SEGNO)
706			goto got_it;
707	}
708
709	while (1) {
710		unsigned long cost, *dirty_bitmap;
711		unsigned int unit_no, segno;
712
713		dirty_bitmap = p.dirty_bitmap;
714		unit_no = find_next_bit(dirty_bitmap,
715				last_segment / p.ofs_unit,
716				p.offset / p.ofs_unit);
717		segno = unit_no * p.ofs_unit;
718		if (segno >= last_segment) {
719			if (sm->last_victim[p.gc_mode]) {
720				last_segment =
721					sm->last_victim[p.gc_mode];
722				sm->last_victim[p.gc_mode] = 0;
723				p.offset = 0;
724				continue;
725			}
726			break;
727		}
728
729		p.offset = segno + p.ofs_unit;
730		nsearched++;
731
732#ifdef CONFIG_F2FS_CHECK_FS
733		/*
734		 * skip selecting the invalid segno (that is failed due to block
735		 * validity check failure during GC) to avoid endless GC loop in
736		 * such cases.
737		 */
738		if (test_bit(segno, sm->invalid_segmap))
739			goto next;
740#endif
741
742		secno = GET_SEC_FROM_SEG(sbi, segno);
743
744		if (sec_usage_check(sbi, secno))
745			goto next;
746
747		/* Don't touch checkpointed data */
748		if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
749			if (p.alloc_mode == LFS) {
750				/*
751				 * LFS is set to find source section during GC.
752				 * The victim should have no checkpointed data.
753				 */
754				if (get_ckpt_valid_blocks(sbi, segno, true))
755					goto next;
756			} else {
757				/*
758				 * SSR | AT_SSR are set to find target segment
759				 * for writes which can be full by checkpointed
760				 * and newly written blocks.
761				 */
762				if (!f2fs_segment_has_free_slot(sbi, segno))
763					goto next;
764			}
765		}
766
767		if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
768			goto next;
769
770		if (is_atgc) {
771			add_victim_entry(sbi, &p, segno);
772			goto next;
773		}
774
775		cost = get_gc_cost(sbi, segno, &p);
776
777		if (p.min_cost > cost) {
778			p.min_segno = segno;
779			p.min_cost = cost;
780		}
781next:
782		if (nsearched >= p.max_search) {
783			if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
784				sm->last_victim[p.gc_mode] =
785					last_victim + p.ofs_unit;
786			else
787				sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
788			sm->last_victim[p.gc_mode] %=
789				(MAIN_SECS(sbi) * sbi->segs_per_sec);
790			break;
791		}
792	}
793
794	/* get victim for GC_AT/AT_SSR */
795	if (is_atgc) {
796		lookup_victim_by_age(sbi, &p);
797		release_victim_entry(sbi);
798	}
799
800	if (is_atgc && p.min_segno == NULL_SEGNO &&
801			sm->elapsed_time < p.age_threshold) {
802		p.age_threshold = 0;
803		goto retry;
804	}
805
806	if (p.min_segno != NULL_SEGNO) {
807got_it:
808		*result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
809got_result:
810		if (p.alloc_mode == LFS) {
811			secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
812			if (gc_type == FG_GC)
813				sbi->cur_victim_sec = secno;
814			else
815				set_bit(secno, dirty_i->victim_secmap);
816		}
817		ret = 0;
818
819	}
820out:
821	if (p.min_segno != NULL_SEGNO)
822		trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
823				sbi->cur_victim_sec,
824				prefree_segments(sbi), free_segments(sbi));
825	mutex_unlock(&dirty_i->seglist_lock);
826
827	return ret;
828}
829
830static const struct victim_selection default_v_ops = {
831	.get_victim = get_victim_by_default,
832};
833
834static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
835{
836	struct inode_entry *ie;
837
838	ie = radix_tree_lookup(&gc_list->iroot, ino);
839	if (ie)
840		return ie->inode;
841	return NULL;
842}
843
844static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
845{
846	struct inode_entry *new_ie;
847
848	if (inode == find_gc_inode(gc_list, inode->i_ino)) {
849		iput(inode);
850		return;
851	}
852	new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS);
853	new_ie->inode = inode;
854
855	f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
856	list_add_tail(&new_ie->list, &gc_list->ilist);
857}
858
859static void put_gc_inode(struct gc_inode_list *gc_list)
860{
861	struct inode_entry *ie, *next_ie;
862	list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
863		radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
864		iput(ie->inode);
865		list_del(&ie->list);
866		kmem_cache_free(f2fs_inode_entry_slab, ie);
867	}
868}
869
870static int check_valid_map(struct f2fs_sb_info *sbi,
871				unsigned int segno, int offset)
872{
873	struct sit_info *sit_i = SIT_I(sbi);
874	struct seg_entry *sentry;
875	int ret;
876
877	down_read(&sit_i->sentry_lock);
878	sentry = get_seg_entry(sbi, segno);
879	ret = f2fs_test_bit(offset, sentry->cur_valid_map);
880	up_read(&sit_i->sentry_lock);
881	return ret;
882}
883
884/*
885 * This function compares node address got in summary with that in NAT.
886 * On validity, copy that node with cold status, otherwise (invalid node)
887 * ignore that.
888 */
889static int gc_node_segment(struct f2fs_sb_info *sbi,
890		struct f2fs_summary *sum, unsigned int segno, int gc_type)
891{
892	struct f2fs_summary *entry;
893	block_t start_addr;
894	int off;
895	int phase = 0;
896	bool fggc = (gc_type == FG_GC);
897	int submitted = 0;
898	unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
899
900	start_addr = START_BLOCK(sbi, segno);
901
902next_step:
903	entry = sum;
904
905	if (fggc && phase == 2)
906		atomic_inc(&sbi->wb_sync_req[NODE]);
907
908	for (off = 0; off < usable_blks_in_seg; off++, entry++) {
909		nid_t nid = le32_to_cpu(entry->nid);
910		struct page *node_page;
911		struct node_info ni;
912		int err;
913
914		/* stop BG_GC if there is not enough free sections. */
915		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
916			return submitted;
917
918		if (check_valid_map(sbi, segno, off) == 0)
919			continue;
920
921		if (phase == 0) {
922			f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
923							META_NAT, true);
924			continue;
925		}
926
927		if (phase == 1) {
928			f2fs_ra_node_page(sbi, nid);
929			continue;
930		}
931
932		/* phase == 2 */
933		node_page = f2fs_get_node_page(sbi, nid);
934		if (IS_ERR(node_page))
935			continue;
936
937		/* block may become invalid during f2fs_get_node_page */
938		if (check_valid_map(sbi, segno, off) == 0) {
939			f2fs_put_page(node_page, 1);
940			continue;
941		}
942
943		if (f2fs_get_node_info(sbi, nid, &ni)) {
944			f2fs_put_page(node_page, 1);
945			continue;
946		}
947
948		if (ni.blk_addr != start_addr + off) {
949			f2fs_put_page(node_page, 1);
950			continue;
951		}
952
953		err = f2fs_move_node_page(node_page, gc_type);
954		if (!err && gc_type == FG_GC)
955			submitted++;
956		stat_inc_node_blk_count(sbi, 1, gc_type);
957	}
958
959	if (++phase < 3)
960		goto next_step;
961
962	if (fggc)
963		atomic_dec(&sbi->wb_sync_req[NODE]);
964	return submitted;
965}
966
967/*
968 * Calculate start block index indicating the given node offset.
969 * Be careful, caller should give this node offset only indicating direct node
970 * blocks. If any node offsets, which point the other types of node blocks such
971 * as indirect or double indirect node blocks, are given, it must be a caller's
972 * bug.
973 */
974block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
975{
976	unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
977	unsigned int bidx;
978
979	if (node_ofs == 0)
980		return 0;
981
982	if (node_ofs <= 2) {
983		bidx = node_ofs - 1;
984	} else if (node_ofs <= indirect_blks) {
985		int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
986		bidx = node_ofs - 2 - dec;
987	} else {
988		int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
989		bidx = node_ofs - 5 - dec;
990	}
991	return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
992}
993
994static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
995		struct node_info *dni, block_t blkaddr, unsigned int *nofs)
996{
997	struct page *node_page;
998	nid_t nid;
999	unsigned int ofs_in_node, max_addrs, base;
1000	block_t source_blkaddr;
1001
1002	nid = le32_to_cpu(sum->nid);
1003	ofs_in_node = le16_to_cpu(sum->ofs_in_node);
1004
1005	node_page = f2fs_get_node_page(sbi, nid);
1006	if (IS_ERR(node_page))
1007		return false;
1008
1009	if (f2fs_get_node_info(sbi, nid, dni)) {
1010		f2fs_put_page(node_page, 1);
1011		return false;
1012	}
1013
1014	if (sum->version != dni->version) {
1015		f2fs_warn(sbi, "%s: valid data with mismatched node version.",
1016			  __func__);
1017		set_sbi_flag(sbi, SBI_NEED_FSCK);
1018	}
1019
1020	if (f2fs_check_nid_range(sbi, dni->ino)) {
1021		f2fs_put_page(node_page, 1);
1022		return false;
1023	}
1024
1025	if (IS_INODE(node_page)) {
1026		base = offset_in_addr(F2FS_INODE(node_page));
1027		max_addrs = DEF_ADDRS_PER_INODE;
1028	} else {
1029		base = 0;
1030		max_addrs = DEF_ADDRS_PER_BLOCK;
1031	}
1032
1033	if (base + ofs_in_node >= max_addrs) {
1034		f2fs_err(sbi, "Inconsistent blkaddr offset: base:%u, ofs_in_node:%u, max:%u, ino:%u, nid:%u",
1035			base, ofs_in_node, max_addrs, dni->ino, dni->nid);
1036		f2fs_put_page(node_page, 1);
1037		return false;
1038	}
1039
1040	*nofs = ofs_of_node(node_page);
1041	source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1042	f2fs_put_page(node_page, 1);
1043
1044	if (source_blkaddr != blkaddr) {
1045#ifdef CONFIG_F2FS_CHECK_FS
1046		unsigned int segno = GET_SEGNO(sbi, blkaddr);
1047		unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1048
1049		if (unlikely(check_valid_map(sbi, segno, offset))) {
1050			if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1051				f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u\n",
1052						blkaddr, source_blkaddr, segno);
1053				set_sbi_flag(sbi, SBI_NEED_FSCK);
1054			}
1055		}
1056#endif
1057		return false;
1058	}
1059	return true;
1060}
1061
1062static int ra_data_block(struct inode *inode, pgoff_t index)
1063{
1064	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1065	struct address_space *mapping = inode->i_mapping;
1066	struct dnode_of_data dn;
1067	struct page *page;
1068	struct extent_info ei = {0, 0, 0};
1069	struct f2fs_io_info fio = {
1070		.sbi = sbi,
1071		.ino = inode->i_ino,
1072		.type = DATA,
1073		.temp = COLD,
1074		.op = REQ_OP_READ,
1075		.op_flags = 0,
1076		.encrypted_page = NULL,
1077		.in_list = false,
1078		.retry = false,
1079	};
1080	int err;
1081
1082	page = f2fs_grab_cache_page(mapping, index, true);
1083	if (!page)
1084		return -ENOMEM;
1085
1086	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1087		dn.data_blkaddr = ei.blk + index - ei.fofs;
1088		if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1089						DATA_GENERIC_ENHANCE_READ))) {
1090			err = -EFSCORRUPTED;
1091			goto put_page;
1092		}
1093		goto got_it;
1094	}
1095
1096	set_new_dnode(&dn, inode, NULL, NULL, 0);
1097	err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1098	if (err)
1099		goto put_page;
1100	f2fs_put_dnode(&dn);
1101
1102	if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1103		err = -ENOENT;
1104		goto put_page;
1105	}
1106	if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1107						DATA_GENERIC_ENHANCE))) {
1108		err = -EFSCORRUPTED;
1109		goto put_page;
1110	}
1111got_it:
1112	/* read page */
1113	fio.page = page;
1114	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1115
1116	/*
1117	 * don't cache encrypted data into meta inode until previous dirty
1118	 * data were writebacked to avoid racing between GC and flush.
1119	 */
1120	f2fs_wait_on_page_writeback(page, DATA, true, true);
1121
1122	f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1123
1124	fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1125					dn.data_blkaddr,
1126					FGP_LOCK | FGP_CREAT, GFP_NOFS);
1127	if (!fio.encrypted_page) {
1128		err = -ENOMEM;
1129		goto put_page;
1130	}
1131
1132	err = f2fs_submit_page_bio(&fio);
1133	if (err)
1134		goto put_encrypted_page;
1135	f2fs_put_page(fio.encrypted_page, 0);
1136	f2fs_put_page(page, 1);
1137
1138	f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1139	f2fs_update_iostat(sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1140
1141	return 0;
1142put_encrypted_page:
1143	f2fs_put_page(fio.encrypted_page, 1);
1144put_page:
1145	f2fs_put_page(page, 1);
1146	return err;
1147}
1148
1149/*
1150 * Move data block via META_MAPPING while keeping locked data page.
1151 * This can be used to move blocks, aka LBAs, directly on disk.
1152 */
1153static int move_data_block(struct inode *inode, block_t bidx,
1154				int gc_type, unsigned int segno, int off)
1155{
1156	struct f2fs_io_info fio = {
1157		.sbi = F2FS_I_SB(inode),
1158		.ino = inode->i_ino,
1159		.type = DATA,
1160		.temp = COLD,
1161		.op = REQ_OP_READ,
1162		.op_flags = 0,
1163		.encrypted_page = NULL,
1164		.in_list = false,
1165		.retry = false,
1166	};
1167	struct dnode_of_data dn;
1168	struct f2fs_summary sum;
1169	struct node_info ni;
1170	struct page *page, *mpage;
1171	block_t newaddr;
1172	int err = 0;
1173	bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1174	int type = fio.sbi->am.atgc_enabled ?
1175				CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1176
1177	/* do not read out */
1178	page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1179	if (!page)
1180		return -ENOMEM;
1181
1182	if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1183		err = -ENOENT;
1184		goto out;
1185	}
1186
1187	if (f2fs_is_atomic_file(inode)) {
1188		F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1189		F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1190		err = -EAGAIN;
1191		goto out;
1192	}
1193
1194	if (f2fs_is_pinned_file(inode)) {
1195		if (gc_type == FG_GC)
1196			f2fs_pin_file_control(inode, true);
1197		err = -EAGAIN;
1198		goto out;
1199	}
1200
1201	set_new_dnode(&dn, inode, NULL, NULL, 0);
1202	err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1203	if (err)
1204		goto out;
1205
1206	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1207		ClearPageUptodate(page);
1208		err = -ENOENT;
1209		goto put_out;
1210	}
1211
1212	/*
1213	 * don't cache encrypted data into meta inode until previous dirty
1214	 * data were writebacked to avoid racing between GC and flush.
1215	 */
1216	f2fs_wait_on_page_writeback(page, DATA, true, true);
1217
1218	f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1219
1220	err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1221	if (err)
1222		goto put_out;
1223
1224	set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1225
1226	/* read page */
1227	fio.page = page;
1228	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1229
1230	if (lfs_mode)
1231		down_write(&fio.sbi->io_order_lock);
1232
1233	mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1234					fio.old_blkaddr, false);
1235	if (!mpage) {
1236		err = -ENOMEM;
1237		goto up_out;
1238	}
1239
1240	fio.encrypted_page = mpage;
1241
1242	/* read source block in mpage */
1243	if (!PageUptodate(mpage)) {
1244		err = f2fs_submit_page_bio(&fio);
1245		if (err) {
1246			f2fs_put_page(mpage, 1);
1247			goto up_out;
1248		}
1249
1250		f2fs_update_iostat(fio.sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1251		f2fs_update_iostat(fio.sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1252
1253		lock_page(mpage);
1254		if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1255						!PageUptodate(mpage))) {
1256			err = -EIO;
1257			f2fs_put_page(mpage, 1);
1258			goto up_out;
1259		}
1260	}
1261
1262	f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1263				&sum, type, NULL, SEQ_NONE);
1264
1265	fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1266				newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1267	if (!fio.encrypted_page) {
1268		err = -ENOMEM;
1269		f2fs_put_page(mpage, 1);
1270		goto recover_block;
1271	}
1272
1273	/* write target block */
1274	f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1275	memcpy(page_address(fio.encrypted_page),
1276				page_address(mpage), PAGE_SIZE);
1277	f2fs_put_page(mpage, 1);
1278	invalidate_mapping_pages(META_MAPPING(fio.sbi),
1279				fio.old_blkaddr, fio.old_blkaddr);
1280
1281	set_page_dirty(fio.encrypted_page);
1282	if (clear_page_dirty_for_io(fio.encrypted_page))
1283		dec_page_count(fio.sbi, F2FS_DIRTY_META);
1284
1285	set_page_writeback(fio.encrypted_page);
1286	ClearPageError(page);
1287
1288	/* allocate block address */
1289	f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
1290
1291	fio.op = REQ_OP_WRITE;
1292	fio.op_flags = REQ_SYNC;
1293	fio.new_blkaddr = newaddr;
1294	f2fs_submit_page_write(&fio);
1295	if (fio.retry) {
1296		err = -EAGAIN;
1297		if (PageWriteback(fio.encrypted_page))
1298			end_page_writeback(fio.encrypted_page);
1299		goto put_page_out;
1300	}
1301
1302	f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
1303
1304	f2fs_update_data_blkaddr(&dn, newaddr);
1305	set_inode_flag(inode, FI_APPEND_WRITE);
1306	if (page->index == 0)
1307		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1308put_page_out:
1309	f2fs_put_page(fio.encrypted_page, 1);
1310recover_block:
1311	if (err)
1312		f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1313							true, true, true);
1314up_out:
1315	if (lfs_mode)
1316		up_write(&fio.sbi->io_order_lock);
1317put_out:
1318	f2fs_put_dnode(&dn);
1319out:
1320	f2fs_put_page(page, 1);
1321	return err;
1322}
1323
1324static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1325							unsigned int segno, int off)
1326{
1327	struct page *page;
1328	int err = 0;
1329
1330	page = f2fs_get_lock_data_page(inode, bidx, true);
1331	if (IS_ERR(page))
1332		return PTR_ERR(page);
1333
1334	if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1335		err = -ENOENT;
1336		goto out;
1337	}
1338
1339	if (f2fs_is_atomic_file(inode)) {
1340		F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1341		F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1342		err = -EAGAIN;
1343		goto out;
1344	}
1345	if (f2fs_is_pinned_file(inode)) {
1346		if (gc_type == FG_GC)
1347			f2fs_pin_file_control(inode, true);
1348		err = -EAGAIN;
1349		goto out;
1350	}
1351
1352	if (gc_type == BG_GC) {
1353		if (PageWriteback(page)) {
1354			err = -EAGAIN;
1355			goto out;
1356		}
1357		set_page_dirty(page);
1358		set_cold_data(page);
1359	} else {
1360		struct f2fs_io_info fio = {
1361			.sbi = F2FS_I_SB(inode),
1362			.ino = inode->i_ino,
1363			.type = DATA,
1364			.temp = COLD,
1365			.op = REQ_OP_WRITE,
1366			.op_flags = REQ_SYNC,
1367			.old_blkaddr = NULL_ADDR,
1368			.page = page,
1369			.encrypted_page = NULL,
1370			.need_lock = LOCK_REQ,
1371			.io_type = FS_GC_DATA_IO,
1372		};
1373		bool is_dirty = PageDirty(page);
1374
1375retry:
1376		f2fs_wait_on_page_writeback(page, DATA, true, true);
1377
1378		set_page_dirty(page);
1379		if (clear_page_dirty_for_io(page)) {
1380			inode_dec_dirty_pages(inode);
1381			f2fs_remove_dirty_inode(inode);
1382		}
1383
1384		set_cold_data(page);
1385
1386		err = f2fs_do_write_data_page(&fio);
1387		if (err) {
1388			clear_cold_data(page);
1389			if (err == -ENOMEM) {
1390				congestion_wait(BLK_RW_ASYNC,
1391						DEFAULT_IO_TIMEOUT);
1392				goto retry;
1393			}
1394			if (is_dirty)
1395				set_page_dirty(page);
1396		}
1397	}
1398out:
1399	f2fs_put_page(page, 1);
1400	return err;
1401}
1402
1403/*
1404 * This function tries to get parent node of victim data block, and identifies
1405 * data block validity. If the block is valid, copy that with cold status and
1406 * modify parent node.
1407 * If the parent node is not valid or the data block address is different,
1408 * the victim data block is ignored.
1409 */
1410static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1411		struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1412		bool force_migrate)
1413{
1414	struct super_block *sb = sbi->sb;
1415	struct f2fs_summary *entry;
1416	block_t start_addr;
1417	int off;
1418	int phase = 0;
1419	int submitted = 0;
1420	unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1421
1422	start_addr = START_BLOCK(sbi, segno);
1423
1424next_step:
1425	entry = sum;
1426
1427	for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1428		struct page *data_page;
1429		struct inode *inode;
1430		struct node_info dni; /* dnode info for the data */
1431		unsigned int ofs_in_node, nofs;
1432		block_t start_bidx;
1433		nid_t nid = le32_to_cpu(entry->nid);
1434
1435		/*
1436		 * stop BG_GC if there is not enough free sections.
1437		 * Or, stop GC if the segment becomes fully valid caused by
1438		 * race condition along with SSR block allocation.
1439		 */
1440		if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1441			(!force_migrate && get_valid_blocks(sbi, segno, true) ==
1442							BLKS_PER_SEC(sbi)))
1443			return submitted;
1444
1445		if (check_valid_map(sbi, segno, off) == 0)
1446			continue;
1447
1448		if (phase == 0) {
1449			f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1450							META_NAT, true);
1451			continue;
1452		}
1453
1454		if (phase == 1) {
1455			f2fs_ra_node_page(sbi, nid);
1456			continue;
1457		}
1458
1459		/* Get an inode by ino with checking validity */
1460		if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1461			continue;
1462
1463		if (phase == 2) {
1464			f2fs_ra_node_page(sbi, dni.ino);
1465			continue;
1466		}
1467
1468		ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1469
1470		if (phase == 3) {
1471			inode = f2fs_iget(sb, dni.ino);
1472			if (IS_ERR(inode) || is_bad_inode(inode) ||
1473					special_file(inode->i_mode)) {
1474				set_sbi_flag(sbi, SBI_NEED_FSCK);
1475				continue;
1476			}
1477
1478			if (!down_write_trylock(
1479				&F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1480				iput(inode);
1481				sbi->skipped_gc_rwsem++;
1482				continue;
1483			}
1484
1485			start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1486								ofs_in_node;
1487
1488			if (f2fs_post_read_required(inode)) {
1489				int err = ra_data_block(inode, start_bidx);
1490
1491				up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1492				if (err) {
1493					iput(inode);
1494					continue;
1495				}
1496				add_gc_inode(gc_list, inode);
1497				continue;
1498			}
1499
1500			data_page = f2fs_get_read_data_page(inode,
1501						start_bidx, REQ_RAHEAD, true);
1502			up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1503			if (IS_ERR(data_page)) {
1504				iput(inode);
1505				continue;
1506			}
1507
1508			f2fs_put_page(data_page, 0);
1509			add_gc_inode(gc_list, inode);
1510			continue;
1511		}
1512
1513		/* phase 4 */
1514		inode = find_gc_inode(gc_list, dni.ino);
1515		if (inode) {
1516			struct f2fs_inode_info *fi = F2FS_I(inode);
1517			bool locked = false;
1518			int err;
1519
1520			if (S_ISREG(inode->i_mode)) {
1521				if (!down_write_trylock(&fi->i_gc_rwsem[READ])) {
1522					sbi->skipped_gc_rwsem++;
1523					continue;
1524				}
1525				if (!down_write_trylock(
1526						&fi->i_gc_rwsem[WRITE])) {
1527					sbi->skipped_gc_rwsem++;
1528					up_write(&fi->i_gc_rwsem[READ]);
1529					continue;
1530				}
1531				locked = true;
1532
1533				/* wait for all inflight aio data */
1534				inode_dio_wait(inode);
1535			}
1536
1537			start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1538								+ ofs_in_node;
1539			if (f2fs_post_read_required(inode))
1540				err = move_data_block(inode, start_bidx,
1541							gc_type, segno, off);
1542			else
1543				err = move_data_page(inode, start_bidx, gc_type,
1544								segno, off);
1545
1546			if (!err && (gc_type == FG_GC ||
1547					f2fs_post_read_required(inode)))
1548				submitted++;
1549
1550			if (locked) {
1551				up_write(&fi->i_gc_rwsem[WRITE]);
1552				up_write(&fi->i_gc_rwsem[READ]);
1553			}
1554
1555			stat_inc_data_blk_count(sbi, 1, gc_type);
1556		}
1557	}
1558
1559	if (++phase < 5)
1560		goto next_step;
1561
1562	return submitted;
1563}
1564
1565static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1566			int gc_type)
1567{
1568	struct sit_info *sit_i = SIT_I(sbi);
1569	int ret;
1570
1571	down_write(&sit_i->sentry_lock);
1572	ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1573					      NO_CHECK_TYPE, LFS, 0);
1574	up_write(&sit_i->sentry_lock);
1575	return ret;
1576}
1577
1578static int do_garbage_collect(struct f2fs_sb_info *sbi,
1579				unsigned int start_segno,
1580				struct gc_inode_list *gc_list, int gc_type,
1581				bool force_migrate)
1582{
1583	struct page *sum_page;
1584	struct f2fs_summary_block *sum;
1585	struct blk_plug plug;
1586	unsigned int segno = start_segno;
1587	unsigned int end_segno = start_segno + sbi->segs_per_sec;
1588	int seg_freed = 0, migrated = 0;
1589	unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1590						SUM_TYPE_DATA : SUM_TYPE_NODE;
1591	int submitted = 0;
1592
1593	if (__is_large_section(sbi))
1594		end_segno = rounddown(end_segno, sbi->segs_per_sec);
1595
1596	/*
1597	 * zone-capacity can be less than zone-size in zoned devices,
1598	 * resulting in less than expected usable segments in the zone,
1599	 * calculate the end segno in the zone which can be garbage collected
1600	 */
1601	if (f2fs_sb_has_blkzoned(sbi))
1602		end_segno -= sbi->segs_per_sec -
1603					f2fs_usable_segs_in_sec(sbi, segno);
1604
1605	sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1606
1607	/* readahead multi ssa blocks those have contiguous address */
1608	if (__is_large_section(sbi))
1609		f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1610					end_segno - segno, META_SSA, true);
1611
1612	/* reference all summary page */
1613	while (segno < end_segno) {
1614		sum_page = f2fs_get_sum_page(sbi, segno++);
1615		if (IS_ERR(sum_page)) {
1616			int err = PTR_ERR(sum_page);
1617
1618			end_segno = segno - 1;
1619			for (segno = start_segno; segno < end_segno; segno++) {
1620				sum_page = find_get_page(META_MAPPING(sbi),
1621						GET_SUM_BLOCK(sbi, segno));
1622				f2fs_put_page(sum_page, 0);
1623				f2fs_put_page(sum_page, 0);
1624			}
1625			return err;
1626		}
1627		unlock_page(sum_page);
1628	}
1629
1630	blk_start_plug(&plug);
1631
1632	for (segno = start_segno; segno < end_segno; segno++) {
1633
1634		/* find segment summary of victim */
1635		sum_page = find_get_page(META_MAPPING(sbi),
1636					GET_SUM_BLOCK(sbi, segno));
1637		f2fs_put_page(sum_page, 0);
1638
1639		if (get_valid_blocks(sbi, segno, false) == 0)
1640			goto freed;
1641		if (gc_type == BG_GC && __is_large_section(sbi) &&
1642				migrated >= sbi->migration_granularity)
1643			goto skip;
1644		if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1645			goto skip;
1646
1647		sum = page_address(sum_page);
1648		if (type != GET_SUM_TYPE((&sum->footer))) {
1649			f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1650				 segno, type, GET_SUM_TYPE((&sum->footer)));
1651			set_sbi_flag(sbi, SBI_NEED_FSCK);
1652			f2fs_stop_checkpoint(sbi, false);
1653			goto skip;
1654		}
1655
1656		/*
1657		 * this is to avoid deadlock:
1658		 * - lock_page(sum_page)         - f2fs_replace_block
1659		 *  - check_valid_map()            - down_write(sentry_lock)
1660		 *   - down_read(sentry_lock)     - change_curseg()
1661		 *                                  - lock_page(sum_page)
1662		 */
1663		if (type == SUM_TYPE_NODE)
1664			submitted += gc_node_segment(sbi, sum->entries, segno,
1665								gc_type);
1666		else
1667			submitted += gc_data_segment(sbi, sum->entries, gc_list,
1668							segno, gc_type,
1669							force_migrate);
1670
1671		stat_inc_seg_count(sbi, type, gc_type);
1672		migrated++;
1673
1674freed:
1675		if (gc_type == FG_GC &&
1676				get_valid_blocks(sbi, segno, false) == 0)
1677			seg_freed++;
1678
1679		if (__is_large_section(sbi))
1680			sbi->next_victim_seg[gc_type] =
1681				(segno + 1 < end_segno) ? segno + 1 : NULL_SEGNO;
1682skip:
1683		f2fs_put_page(sum_page, 0);
1684	}
1685
1686	if (submitted)
1687		f2fs_submit_merged_write(sbi,
1688				(type == SUM_TYPE_NODE) ? NODE : DATA);
1689
1690	blk_finish_plug(&plug);
1691
1692	stat_inc_call_count(sbi->stat_info);
1693
1694	return seg_freed;
1695}
1696
1697int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1698			bool background, bool force, unsigned int segno)
1699{
1700	int gc_type = sync ? FG_GC : BG_GC;
1701	int sec_freed = 0, seg_freed = 0, total_freed = 0;
1702	int ret = 0;
1703	struct cp_control cpc;
1704	unsigned int init_segno = segno;
1705	struct gc_inode_list gc_list = {
1706		.ilist = LIST_HEAD_INIT(gc_list.ilist),
1707		.iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1708	};
1709	unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1710	unsigned long long first_skipped;
1711	unsigned int skipped_round = 0, round = 0;
1712
1713	trace_f2fs_gc_begin(sbi->sb, sync, background,
1714				get_pages(sbi, F2FS_DIRTY_NODES),
1715				get_pages(sbi, F2FS_DIRTY_DENTS),
1716				get_pages(sbi, F2FS_DIRTY_IMETA),
1717				free_sections(sbi),
1718				free_segments(sbi),
1719				reserved_segments(sbi),
1720				prefree_segments(sbi));
1721
1722	cpc.reason = __get_cp_reason(sbi);
1723	sbi->skipped_gc_rwsem = 0;
1724	first_skipped = last_skipped;
1725gc_more:
1726	if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1727		ret = -EINVAL;
1728		goto stop;
1729	}
1730	if (unlikely(f2fs_cp_error(sbi))) {
1731		ret = -EIO;
1732		goto stop;
1733	}
1734
1735	if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1736		/*
1737		 * For example, if there are many prefree_segments below given
1738		 * threshold, we can make them free by checkpoint. Then, we
1739		 * secure free segments which doesn't need fggc any more.
1740		 */
1741		if (prefree_segments(sbi) &&
1742				!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1743			ret = f2fs_write_checkpoint(sbi, &cpc);
1744			if (ret)
1745				goto stop;
1746		}
1747		if (has_not_enough_free_secs(sbi, 0, 0))
1748			gc_type = FG_GC;
1749	}
1750
1751	/* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1752	if (gc_type == BG_GC && !background) {
1753		ret = -EINVAL;
1754		goto stop;
1755	}
1756	ret = __get_victim(sbi, &segno, gc_type);
1757	if (ret)
1758		goto stop;
1759
1760	seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type, force);
1761	if (gc_type == FG_GC &&
1762		seg_freed == f2fs_usable_segs_in_sec(sbi, segno))
1763		sec_freed++;
1764	total_freed += seg_freed;
1765
1766	if (gc_type == FG_GC) {
1767		if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
1768						sbi->skipped_gc_rwsem)
1769			skipped_round++;
1770		last_skipped = sbi->skipped_atomic_files[FG_GC];
1771		round++;
1772	}
1773
1774	if (gc_type == FG_GC && seg_freed)
1775		sbi->cur_victim_sec = NULL_SEGNO;
1776
1777	if (sync)
1778		goto stop;
1779
1780	if (!has_not_enough_free_secs(sbi, sec_freed, 0))
1781		goto stop;
1782
1783	if (skipped_round <= MAX_SKIP_GC_COUNT || skipped_round * 2 < round) {
1784
1785		/* Write checkpoint to reclaim prefree segments */
1786		if (free_sections(sbi) < NR_CURSEG_PERSIST_TYPE &&
1787				prefree_segments(sbi) &&
1788				!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1789			ret = f2fs_write_checkpoint(sbi, &cpc);
1790			if (ret)
1791				goto stop;
1792		}
1793		segno = NULL_SEGNO;
1794		goto gc_more;
1795	}
1796	if (first_skipped < last_skipped &&
1797			(last_skipped - first_skipped) >
1798					sbi->skipped_gc_rwsem) {
1799		f2fs_drop_inmem_pages_all(sbi, true);
1800		segno = NULL_SEGNO;
1801		goto gc_more;
1802	}
1803	if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1804		ret = f2fs_write_checkpoint(sbi, &cpc);
1805stop:
1806	SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1807	SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1808
1809	trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1810				get_pages(sbi, F2FS_DIRTY_NODES),
1811				get_pages(sbi, F2FS_DIRTY_DENTS),
1812				get_pages(sbi, F2FS_DIRTY_IMETA),
1813				free_sections(sbi),
1814				free_segments(sbi),
1815				reserved_segments(sbi),
1816				prefree_segments(sbi));
1817
1818	up_write(&sbi->gc_lock);
1819
1820	put_gc_inode(&gc_list);
1821
1822	if (sync && !ret)
1823		ret = sec_freed ? 0 : -EAGAIN;
1824	return ret;
1825}
1826
1827int __init f2fs_create_garbage_collection_cache(void)
1828{
1829	victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1830					sizeof(struct victim_entry));
1831	if (!victim_entry_slab)
1832		return -ENOMEM;
1833	return 0;
1834}
1835
1836void f2fs_destroy_garbage_collection_cache(void)
1837{
1838	kmem_cache_destroy(victim_entry_slab);
1839}
1840
1841static void init_atgc_management(struct f2fs_sb_info *sbi)
1842{
1843	struct atgc_management *am = &sbi->am;
1844
1845	if (test_opt(sbi, ATGC) &&
1846		SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1847		am->atgc_enabled = true;
1848
1849	am->root = RB_ROOT_CACHED;
1850	INIT_LIST_HEAD(&am->victim_list);
1851	am->victim_count = 0;
1852
1853	am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1854	am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1855	am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1856	am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
1857}
1858
1859void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1860{
1861	DIRTY_I(sbi)->v_ops = &default_v_ops;
1862
1863	sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1864
1865	/* give warm/cold data area from slower device */
1866	if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1867		SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1868				GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1869
1870	init_atgc_management(sbi);
1871}
1872
1873static int free_segment_range(struct f2fs_sb_info *sbi,
1874				unsigned int secs, bool gc_only)
1875{
1876	unsigned int segno, next_inuse, start, end;
1877	struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1878	int gc_mode, gc_type;
1879	int err = 0;
1880	int type;
1881
1882	/* Force block allocation for GC */
1883	MAIN_SECS(sbi) -= secs;
1884	start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1885	end = MAIN_SEGS(sbi) - 1;
1886
1887	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1888	for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1889		if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1890			SIT_I(sbi)->last_victim[gc_mode] = 0;
1891
1892	for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1893		if (sbi->next_victim_seg[gc_type] >= start)
1894			sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1895	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1896
1897	/* Move out cursegs from the target range */
1898	for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1899		f2fs_allocate_segment_for_resize(sbi, type, start, end);
1900
1901	/* do GC to move out valid blocks in the range */
1902	for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1903		struct gc_inode_list gc_list = {
1904			.ilist = LIST_HEAD_INIT(gc_list.ilist),
1905			.iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1906		};
1907
1908		do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
1909		put_gc_inode(&gc_list);
1910
1911		if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1912			err = -EAGAIN;
1913			goto out;
1914		}
1915		if (fatal_signal_pending(current)) {
1916			err = -ERESTARTSYS;
1917			goto out;
1918		}
1919	}
1920	if (gc_only)
1921		goto out;
1922
1923	err = f2fs_write_checkpoint(sbi, &cpc);
1924	if (err)
1925		goto out;
1926
1927	next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1928	if (next_inuse <= end) {
1929		f2fs_err(sbi, "segno %u should be free but still inuse!",
1930			 next_inuse);
1931		f2fs_bug_on(sbi, 1);
1932	}
1933out:
1934	MAIN_SECS(sbi) += secs;
1935	return err;
1936}
1937
1938static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1939{
1940	struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1941	int section_count;
1942	int segment_count;
1943	int segment_count_main;
1944	long long block_count;
1945	int segs = secs * sbi->segs_per_sec;
1946
1947	down_write(&sbi->sb_lock);
1948
1949	section_count = le32_to_cpu(raw_sb->section_count);
1950	segment_count = le32_to_cpu(raw_sb->segment_count);
1951	segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
1952	block_count = le64_to_cpu(raw_sb->block_count);
1953
1954	raw_sb->section_count = cpu_to_le32(section_count + secs);
1955	raw_sb->segment_count = cpu_to_le32(segment_count + segs);
1956	raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
1957	raw_sb->block_count = cpu_to_le64(block_count +
1958					(long long)segs * sbi->blocks_per_seg);
1959	if (f2fs_is_multi_device(sbi)) {
1960		int last_dev = sbi->s_ndevs - 1;
1961		int dev_segs =
1962			le32_to_cpu(raw_sb->devs[last_dev].total_segments);
1963
1964		raw_sb->devs[last_dev].total_segments =
1965						cpu_to_le32(dev_segs + segs);
1966	}
1967
1968	up_write(&sbi->sb_lock);
1969}
1970
1971static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
1972{
1973	int segs = secs * sbi->segs_per_sec;
1974	long long blks = (long long)segs * sbi->blocks_per_seg;
1975	long long user_block_count =
1976				le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
1977
1978	SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
1979	MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
1980	MAIN_SECS(sbi) += secs;
1981	FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
1982	FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
1983	F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
1984
1985	if (f2fs_is_multi_device(sbi)) {
1986		int last_dev = sbi->s_ndevs - 1;
1987
1988		FDEV(last_dev).total_segments =
1989				(int)FDEV(last_dev).total_segments + segs;
1990		FDEV(last_dev).end_blk =
1991				(long long)FDEV(last_dev).end_blk + blks;
1992#ifdef CONFIG_BLK_DEV_ZONED
1993		FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
1994					(int)(blks >> sbi->log_blocks_per_blkz);
1995#endif
1996	}
1997}
1998
1999int f2fs_resize_fs(struct file *filp, __u64 block_count)
2000{
2001	struct f2fs_sb_info *sbi = F2FS_I_SB(file_inode(filp));
2002	__u64 old_block_count, shrunk_blocks;
2003	struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
2004	unsigned int secs;
2005	int err = 0;
2006	__u32 rem;
2007
2008	old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
2009	if (block_count > old_block_count)
2010		return -EINVAL;
2011
2012	if (f2fs_is_multi_device(sbi)) {
2013		int last_dev = sbi->s_ndevs - 1;
2014		__u64 last_segs = FDEV(last_dev).total_segments;
2015
2016		if (block_count + last_segs * sbi->blocks_per_seg <=
2017								old_block_count)
2018			return -EINVAL;
2019	}
2020
2021	/* new fs size should align to section size */
2022	div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
2023	if (rem)
2024		return -EINVAL;
2025
2026	if (block_count == old_block_count)
2027		return 0;
2028
2029	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2030		f2fs_err(sbi, "Should run fsck to repair first.");
2031		return -EFSCORRUPTED;
2032	}
2033
2034	if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2035		f2fs_err(sbi, "Checkpoint should be enabled.");
2036		return -EINVAL;
2037	}
2038
2039	err = mnt_want_write_file(filp);
2040	if (err)
2041		return err;
2042
2043	shrunk_blocks = old_block_count - block_count;
2044	secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2045
2046	/* stop other GC */
2047	if (!down_write_trylock(&sbi->gc_lock)) {
2048		err = -EAGAIN;
2049		goto out_drop_write;
2050	}
2051
2052	/* stop CP to protect MAIN_SEC in free_segment_range */
2053	f2fs_lock_op(sbi);
2054
2055	spin_lock(&sbi->stat_lock);
2056	if (shrunk_blocks + valid_user_blocks(sbi) +
2057		sbi->current_reserved_blocks + sbi->unusable_block_count +
2058		F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2059		err = -ENOSPC;
2060	spin_unlock(&sbi->stat_lock);
2061
2062	if (err)
2063		goto out_unlock;
2064
2065	err = free_segment_range(sbi, secs, true);
2066
2067out_unlock:
2068	f2fs_unlock_op(sbi);
2069	up_write(&sbi->gc_lock);
2070out_drop_write:
2071	mnt_drop_write_file(filp);
2072	if (err)
2073		return err;
2074
2075	freeze_super(sbi->sb);
2076
2077	if (f2fs_readonly(sbi->sb)) {
2078		thaw_super(sbi->sb);
2079		return -EROFS;
2080	}
2081
2082	down_write(&sbi->gc_lock);
2083	mutex_lock(&sbi->cp_mutex);
2084
2085	spin_lock(&sbi->stat_lock);
2086	if (shrunk_blocks + valid_user_blocks(sbi) +
2087		sbi->current_reserved_blocks + sbi->unusable_block_count +
2088		F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2089		err = -ENOSPC;
2090	else
2091		sbi->user_block_count -= shrunk_blocks;
2092	spin_unlock(&sbi->stat_lock);
2093	if (err)
2094		goto out_err;
2095
2096	set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2097	err = free_segment_range(sbi, secs, false);
2098	if (err)
2099		goto recover_out;
2100
2101	update_sb_metadata(sbi, -secs);
2102
2103	err = f2fs_commit_super(sbi, false);
2104	if (err) {
2105		update_sb_metadata(sbi, secs);
2106		goto recover_out;
2107	}
2108
2109	update_fs_metadata(sbi, -secs);
2110	clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2111	set_sbi_flag(sbi, SBI_IS_DIRTY);
2112
2113	err = f2fs_write_checkpoint(sbi, &cpc);
2114	if (err) {
2115		update_fs_metadata(sbi, secs);
2116		update_sb_metadata(sbi, secs);
2117		f2fs_commit_super(sbi, false);
2118	}
2119recover_out:
2120	clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2121	if (err) {
2122		set_sbi_flag(sbi, SBI_NEED_FSCK);
2123		f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2124
2125		spin_lock(&sbi->stat_lock);
2126		sbi->user_block_count += shrunk_blocks;
2127		spin_unlock(&sbi->stat_lock);
2128	}
2129out_err:
2130	mutex_unlock(&sbi->cp_mutex);
2131	up_write(&sbi->gc_lock);
2132	thaw_super(sbi->sb);
2133	return err;
2134}
2135