xref: /linux/fs/f2fs/gc.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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/init.h>
11 #include <linux/f2fs_fs.h>
12 #include <linux/kthread.h>
13 #include <linux/delay.h>
14 #include <linux/freezer.h>
15 #include <linux/sched/signal.h>
16 #include <linux/random.h>
17 #include <linux/sched/mm.h>
18 
19 #include "f2fs.h"
20 #include "node.h"
21 #include "segment.h"
22 #include "gc.h"
23 #include "iostat.h"
24 #include <trace/events/f2fs.h>
25 
26 static struct kmem_cache *victim_entry_slab;
27 
28 static unsigned int count_bits(const unsigned long *addr,
29 				unsigned int offset, unsigned int len);
30 
31 static int gc_thread_func(void *data)
32 {
33 	struct f2fs_sb_info *sbi = data;
34 	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
35 	wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
36 	wait_queue_head_t *fggc_wq = &sbi->gc_thread->fggc_wq;
37 	unsigned int wait_ms;
38 	struct f2fs_gc_control gc_control = {
39 		.victim_segno = NULL_SEGNO,
40 		.should_migrate_blocks = false,
41 		.err_gc_skipped = false };
42 
43 	wait_ms = gc_th->min_sleep_time;
44 
45 	set_freezable();
46 	do {
47 		bool sync_mode, foreground = false;
48 
49 		wait_event_freezable_timeout(*wq,
50 				kthread_should_stop() ||
51 				waitqueue_active(fggc_wq) ||
52 				gc_th->gc_wake,
53 				msecs_to_jiffies(wait_ms));
54 
55 		if (test_opt(sbi, GC_MERGE) && waitqueue_active(fggc_wq))
56 			foreground = true;
57 
58 		/* give it a try one time */
59 		if (gc_th->gc_wake)
60 			gc_th->gc_wake = false;
61 
62 		if (f2fs_readonly(sbi->sb)) {
63 			stat_other_skip_bggc_count(sbi);
64 			continue;
65 		}
66 		if (kthread_should_stop())
67 			break;
68 
69 		if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
70 			increase_sleep_time(gc_th, &wait_ms);
71 			stat_other_skip_bggc_count(sbi);
72 			continue;
73 		}
74 
75 		if (time_to_inject(sbi, FAULT_CHECKPOINT))
76 			f2fs_stop_checkpoint(sbi, false,
77 					STOP_CP_REASON_FAULT_INJECT);
78 
79 		if (!sb_start_write_trylock(sbi->sb)) {
80 			stat_other_skip_bggc_count(sbi);
81 			continue;
82 		}
83 
84 		gc_control.one_time = false;
85 
86 		/*
87 		 * [GC triggering condition]
88 		 * 0. GC is not conducted currently.
89 		 * 1. There are enough dirty segments.
90 		 * 2. IO subsystem is idle by checking the # of writeback pages.
91 		 * 3. IO subsystem is idle by checking the # of requests in
92 		 *    bdev's request list.
93 		 *
94 		 * Note) We have to avoid triggering GCs frequently.
95 		 * Because it is possible that some segments can be
96 		 * invalidated soon after by user update or deletion.
97 		 * So, I'd like to wait some time to collect dirty segments.
98 		 */
99 		if (sbi->gc_mode == GC_URGENT_HIGH ||
100 				sbi->gc_mode == GC_URGENT_MID) {
101 			wait_ms = gc_th->urgent_sleep_time;
102 			f2fs_down_write(&sbi->gc_lock);
103 			goto do_gc;
104 		}
105 
106 		if (foreground) {
107 			f2fs_down_write(&sbi->gc_lock);
108 			goto do_gc;
109 		} else if (!f2fs_down_write_trylock(&sbi->gc_lock)) {
110 			stat_other_skip_bggc_count(sbi);
111 			goto next;
112 		}
113 
114 		if (!is_idle(sbi, GC_TIME)) {
115 			increase_sleep_time(gc_th, &wait_ms);
116 			f2fs_up_write(&sbi->gc_lock);
117 			stat_io_skip_bggc_count(sbi);
118 			goto next;
119 		}
120 
121 		if (f2fs_sb_has_blkzoned(sbi)) {
122 			if (has_enough_free_blocks(sbi,
123 				gc_th->no_zoned_gc_percent)) {
124 				wait_ms = gc_th->no_gc_sleep_time;
125 				f2fs_up_write(&sbi->gc_lock);
126 				goto next;
127 			}
128 			if (wait_ms == gc_th->no_gc_sleep_time)
129 				wait_ms = gc_th->max_sleep_time;
130 		}
131 
132 		if (need_to_boost_gc(sbi)) {
133 			decrease_sleep_time(gc_th, &wait_ms);
134 			if (f2fs_sb_has_blkzoned(sbi))
135 				gc_control.one_time = true;
136 		} else {
137 			increase_sleep_time(gc_th, &wait_ms);
138 		}
139 do_gc:
140 		stat_inc_gc_call_count(sbi, foreground ?
141 					FOREGROUND : BACKGROUND);
142 
143 		sync_mode = (F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC) ||
144 				gc_control.one_time;
145 
146 		/* foreground GC was been triggered via f2fs_balance_fs() */
147 		if (foreground)
148 			sync_mode = false;
149 
150 		gc_control.init_gc_type = sync_mode ? FG_GC : BG_GC;
151 		gc_control.no_bg_gc = foreground;
152 		gc_control.nr_free_secs = foreground ? 1 : 0;
153 
154 		/* if return value is not zero, no victim was selected */
155 		if (f2fs_gc(sbi, &gc_control)) {
156 			/* don't bother wait_ms by foreground gc */
157 			if (!foreground)
158 				wait_ms = gc_th->no_gc_sleep_time;
159 		} else {
160 			/* reset wait_ms to default sleep time */
161 			if (wait_ms == gc_th->no_gc_sleep_time)
162 				wait_ms = gc_th->min_sleep_time;
163 		}
164 
165 		if (foreground)
166 			wake_up_all(&gc_th->fggc_wq);
167 
168 		trace_f2fs_background_gc(sbi->sb, wait_ms,
169 				prefree_segments(sbi), free_segments(sbi));
170 
171 		/* balancing f2fs's metadata periodically */
172 		f2fs_balance_fs_bg(sbi, true);
173 next:
174 		if (sbi->gc_mode != GC_NORMAL) {
175 			spin_lock(&sbi->gc_remaining_trials_lock);
176 			if (sbi->gc_remaining_trials) {
177 				sbi->gc_remaining_trials--;
178 				if (!sbi->gc_remaining_trials)
179 					sbi->gc_mode = GC_NORMAL;
180 			}
181 			spin_unlock(&sbi->gc_remaining_trials_lock);
182 		}
183 		sb_end_write(sbi->sb);
184 
185 	} while (!kthread_should_stop());
186 	return 0;
187 }
188 
189 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
190 {
191 	struct f2fs_gc_kthread *gc_th;
192 	dev_t dev = sbi->sb->s_bdev->bd_dev;
193 
194 	gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
195 	if (!gc_th)
196 		return -ENOMEM;
197 
198 	gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
199 	gc_th->valid_thresh_ratio = DEF_GC_THREAD_VALID_THRESH_RATIO;
200 
201 	if (f2fs_sb_has_blkzoned(sbi)) {
202 		gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME_ZONED;
203 		gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME_ZONED;
204 		gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME_ZONED;
205 		gc_th->no_zoned_gc_percent = LIMIT_NO_ZONED_GC;
206 		gc_th->boost_zoned_gc_percent = LIMIT_BOOST_ZONED_GC;
207 	} else {
208 		gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
209 		gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
210 		gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
211 		gc_th->no_zoned_gc_percent = 0;
212 		gc_th->boost_zoned_gc_percent = 0;
213 	}
214 
215 	gc_th->gc_wake = false;
216 
217 	sbi->gc_thread = gc_th;
218 	init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
219 	init_waitqueue_head(&sbi->gc_thread->fggc_wq);
220 	sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
221 			"f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
222 	if (IS_ERR(gc_th->f2fs_gc_task)) {
223 		int err = PTR_ERR(gc_th->f2fs_gc_task);
224 
225 		kfree(gc_th);
226 		sbi->gc_thread = NULL;
227 		return err;
228 	}
229 
230 	return 0;
231 }
232 
233 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
234 {
235 	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
236 
237 	if (!gc_th)
238 		return;
239 	kthread_stop(gc_th->f2fs_gc_task);
240 	wake_up_all(&gc_th->fggc_wq);
241 	kfree(gc_th);
242 	sbi->gc_thread = NULL;
243 }
244 
245 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
246 {
247 	int gc_mode;
248 
249 	if (gc_type == BG_GC) {
250 		if (sbi->am.atgc_enabled)
251 			gc_mode = GC_AT;
252 		else
253 			gc_mode = GC_CB;
254 	} else {
255 		gc_mode = GC_GREEDY;
256 	}
257 
258 	switch (sbi->gc_mode) {
259 	case GC_IDLE_CB:
260 	case GC_URGENT_LOW:
261 	case GC_URGENT_MID:
262 		gc_mode = GC_CB;
263 		break;
264 	case GC_IDLE_GREEDY:
265 	case GC_URGENT_HIGH:
266 		gc_mode = GC_GREEDY;
267 		break;
268 	case GC_IDLE_AT:
269 		gc_mode = GC_AT;
270 		break;
271 	}
272 
273 	return gc_mode;
274 }
275 
276 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
277 			int type, struct victim_sel_policy *p)
278 {
279 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
280 
281 	if (p->alloc_mode == SSR) {
282 		p->gc_mode = GC_GREEDY;
283 		p->dirty_bitmap = dirty_i->dirty_segmap[type];
284 		p->max_search = dirty_i->nr_dirty[type];
285 		p->ofs_unit = 1;
286 	} else if (p->alloc_mode == AT_SSR) {
287 		p->gc_mode = GC_GREEDY;
288 		p->dirty_bitmap = dirty_i->dirty_segmap[type];
289 		p->max_search = dirty_i->nr_dirty[type];
290 		p->ofs_unit = 1;
291 	} else {
292 		p->gc_mode = select_gc_type(sbi, gc_type);
293 		p->ofs_unit = SEGS_PER_SEC(sbi);
294 		if (__is_large_section(sbi)) {
295 			p->dirty_bitmap = dirty_i->dirty_secmap;
296 			p->max_search = count_bits(p->dirty_bitmap,
297 						0, MAIN_SECS(sbi));
298 		} else {
299 			p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
300 			p->max_search = dirty_i->nr_dirty[DIRTY];
301 		}
302 	}
303 
304 	/*
305 	 * adjust candidates range, should select all dirty segments for
306 	 * foreground GC and urgent GC cases.
307 	 */
308 	if (gc_type != FG_GC &&
309 			(sbi->gc_mode != GC_URGENT_HIGH) &&
310 			(p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
311 			p->max_search > sbi->max_victim_search)
312 		p->max_search = sbi->max_victim_search;
313 
314 	/* let's select beginning hot/small space first. */
315 	if (f2fs_need_rand_seg(sbi))
316 		p->offset = get_random_u32_below(MAIN_SECS(sbi) *
317 						SEGS_PER_SEC(sbi));
318 	else if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
319 		p->offset = 0;
320 	else
321 		p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
322 }
323 
324 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
325 				struct victim_sel_policy *p)
326 {
327 	/* SSR allocates in a segment unit */
328 	if (p->alloc_mode == SSR)
329 		return BLKS_PER_SEG(sbi);
330 	else if (p->alloc_mode == AT_SSR)
331 		return UINT_MAX;
332 
333 	/* LFS */
334 	if (p->gc_mode == GC_GREEDY)
335 		return SEGS_TO_BLKS(sbi, 2 * p->ofs_unit);
336 	else if (p->gc_mode == GC_CB)
337 		return UINT_MAX;
338 	else if (p->gc_mode == GC_AT)
339 		return UINT_MAX;
340 	else /* No other gc_mode */
341 		return 0;
342 }
343 
344 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
345 {
346 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
347 	unsigned int secno;
348 
349 	/*
350 	 * If the gc_type is FG_GC, we can select victim segments
351 	 * selected by background GC before.
352 	 * Those segments guarantee they have small valid blocks.
353 	 */
354 	for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
355 		if (sec_usage_check(sbi, secno))
356 			continue;
357 		clear_bit(secno, dirty_i->victim_secmap);
358 		return GET_SEG_FROM_SEC(sbi, secno);
359 	}
360 	return NULL_SEGNO;
361 }
362 
363 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
364 {
365 	struct sit_info *sit_i = SIT_I(sbi);
366 	unsigned long long mtime = 0;
367 	unsigned int vblocks;
368 	unsigned char age = 0;
369 	unsigned char u;
370 	unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi);
371 
372 	mtime = f2fs_get_section_mtime(sbi, segno);
373 	f2fs_bug_on(sbi, mtime == INVALID_MTIME);
374 	vblocks = get_valid_blocks(sbi, segno, true);
375 	vblocks = div_u64(vblocks, usable_segs_per_sec);
376 
377 	u = BLKS_TO_SEGS(sbi, vblocks * 100);
378 
379 	/* Handle if the system time has changed by the user */
380 	if (mtime < sit_i->min_mtime)
381 		sit_i->min_mtime = mtime;
382 	if (mtime > sit_i->max_mtime)
383 		sit_i->max_mtime = mtime;
384 	if (sit_i->max_mtime != sit_i->min_mtime)
385 		age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
386 				sit_i->max_mtime - sit_i->min_mtime);
387 
388 	return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
389 }
390 
391 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
392 			unsigned int segno, struct victim_sel_policy *p)
393 {
394 	if (p->alloc_mode == SSR)
395 		return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
396 
397 	if (p->one_time_gc && (get_valid_blocks(sbi, segno, true) >=
398 		CAP_BLKS_PER_SEC(sbi) * sbi->gc_thread->valid_thresh_ratio /
399 		100))
400 		return UINT_MAX;
401 
402 	/* alloc_mode == LFS */
403 	if (p->gc_mode == GC_GREEDY)
404 		return get_valid_blocks(sbi, segno, true);
405 	else if (p->gc_mode == GC_CB)
406 		return get_cb_cost(sbi, segno);
407 
408 	f2fs_bug_on(sbi, 1);
409 	return 0;
410 }
411 
412 static unsigned int count_bits(const unsigned long *addr,
413 				unsigned int offset, unsigned int len)
414 {
415 	unsigned int end = offset + len, sum = 0;
416 
417 	while (offset < end) {
418 		if (test_bit(offset++, addr))
419 			++sum;
420 	}
421 	return sum;
422 }
423 
424 static bool f2fs_check_victim_tree(struct f2fs_sb_info *sbi,
425 				struct rb_root_cached *root)
426 {
427 #ifdef CONFIG_F2FS_CHECK_FS
428 	struct rb_node *cur = rb_first_cached(root), *next;
429 	struct victim_entry *cur_ve, *next_ve;
430 
431 	while (cur) {
432 		next = rb_next(cur);
433 		if (!next)
434 			return true;
435 
436 		cur_ve = rb_entry(cur, struct victim_entry, rb_node);
437 		next_ve = rb_entry(next, struct victim_entry, rb_node);
438 
439 		if (cur_ve->mtime > next_ve->mtime) {
440 			f2fs_info(sbi, "broken victim_rbtree, "
441 				"cur_mtime(%llu) next_mtime(%llu)",
442 				cur_ve->mtime, next_ve->mtime);
443 			return false;
444 		}
445 		cur = next;
446 	}
447 #endif
448 	return true;
449 }
450 
451 static struct victim_entry *__lookup_victim_entry(struct f2fs_sb_info *sbi,
452 					unsigned long long mtime)
453 {
454 	struct atgc_management *am = &sbi->am;
455 	struct rb_node *node = am->root.rb_root.rb_node;
456 	struct victim_entry *ve = NULL;
457 
458 	while (node) {
459 		ve = rb_entry(node, struct victim_entry, rb_node);
460 
461 		if (mtime < ve->mtime)
462 			node = node->rb_left;
463 		else
464 			node = node->rb_right;
465 	}
466 	return ve;
467 }
468 
469 static struct victim_entry *__create_victim_entry(struct f2fs_sb_info *sbi,
470 		unsigned long long mtime, unsigned int segno)
471 {
472 	struct atgc_management *am = &sbi->am;
473 	struct victim_entry *ve;
474 
475 	ve =  f2fs_kmem_cache_alloc(victim_entry_slab, GFP_NOFS, true, NULL);
476 
477 	ve->mtime = mtime;
478 	ve->segno = segno;
479 
480 	list_add_tail(&ve->list, &am->victim_list);
481 	am->victim_count++;
482 
483 	return ve;
484 }
485 
486 static void __insert_victim_entry(struct f2fs_sb_info *sbi,
487 				unsigned long long mtime, unsigned int segno)
488 {
489 	struct atgc_management *am = &sbi->am;
490 	struct rb_root_cached *root = &am->root;
491 	struct rb_node **p = &root->rb_root.rb_node;
492 	struct rb_node *parent = NULL;
493 	struct victim_entry *ve;
494 	bool left_most = true;
495 
496 	/* look up rb tree to find parent node */
497 	while (*p) {
498 		parent = *p;
499 		ve = rb_entry(parent, struct victim_entry, rb_node);
500 
501 		if (mtime < ve->mtime) {
502 			p = &(*p)->rb_left;
503 		} else {
504 			p = &(*p)->rb_right;
505 			left_most = false;
506 		}
507 	}
508 
509 	ve = __create_victim_entry(sbi, mtime, segno);
510 
511 	rb_link_node(&ve->rb_node, parent, p);
512 	rb_insert_color_cached(&ve->rb_node, root, left_most);
513 }
514 
515 static void add_victim_entry(struct f2fs_sb_info *sbi,
516 				struct victim_sel_policy *p, unsigned int segno)
517 {
518 	struct sit_info *sit_i = SIT_I(sbi);
519 	unsigned long long mtime = 0;
520 
521 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
522 		if (p->gc_mode == GC_AT &&
523 			get_valid_blocks(sbi, segno, true) == 0)
524 			return;
525 	}
526 
527 	mtime = f2fs_get_section_mtime(sbi, segno);
528 	f2fs_bug_on(sbi, mtime == INVALID_MTIME);
529 
530 	/* Handle if the system time has changed by the user */
531 	if (mtime < sit_i->min_mtime)
532 		sit_i->min_mtime = mtime;
533 	if (mtime > sit_i->max_mtime)
534 		sit_i->max_mtime = mtime;
535 	if (mtime < sit_i->dirty_min_mtime)
536 		sit_i->dirty_min_mtime = mtime;
537 	if (mtime > sit_i->dirty_max_mtime)
538 		sit_i->dirty_max_mtime = mtime;
539 
540 	/* don't choose young section as candidate */
541 	if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
542 		return;
543 
544 	__insert_victim_entry(sbi, mtime, segno);
545 }
546 
547 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
548 						struct victim_sel_policy *p)
549 {
550 	struct sit_info *sit_i = SIT_I(sbi);
551 	struct atgc_management *am = &sbi->am;
552 	struct rb_root_cached *root = &am->root;
553 	struct rb_node *node;
554 	struct victim_entry *ve;
555 	unsigned long long total_time;
556 	unsigned long long age, u, accu;
557 	unsigned long long max_mtime = sit_i->dirty_max_mtime;
558 	unsigned long long min_mtime = sit_i->dirty_min_mtime;
559 	unsigned int sec_blocks = CAP_BLKS_PER_SEC(sbi);
560 	unsigned int vblocks;
561 	unsigned int dirty_threshold = max(am->max_candidate_count,
562 					am->candidate_ratio *
563 					am->victim_count / 100);
564 	unsigned int age_weight = am->age_weight;
565 	unsigned int cost;
566 	unsigned int iter = 0;
567 
568 	if (max_mtime < min_mtime)
569 		return;
570 
571 	max_mtime += 1;
572 	total_time = max_mtime - min_mtime;
573 
574 	accu = div64_u64(ULLONG_MAX, total_time);
575 	accu = min_t(unsigned long long, div_u64(accu, 100),
576 					DEFAULT_ACCURACY_CLASS);
577 
578 	node = rb_first_cached(root);
579 next:
580 	ve = rb_entry_safe(node, struct victim_entry, rb_node);
581 	if (!ve)
582 		return;
583 
584 	if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
585 		goto skip;
586 
587 	/* age = 10000 * x% * 60 */
588 	age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
589 								age_weight;
590 
591 	vblocks = get_valid_blocks(sbi, ve->segno, true);
592 	f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
593 
594 	/* u = 10000 * x% * 40 */
595 	u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
596 							(100 - age_weight);
597 
598 	f2fs_bug_on(sbi, age + u >= UINT_MAX);
599 
600 	cost = UINT_MAX - (age + u);
601 	iter++;
602 
603 	if (cost < p->min_cost ||
604 			(cost == p->min_cost && age > p->oldest_age)) {
605 		p->min_cost = cost;
606 		p->oldest_age = age;
607 		p->min_segno = ve->segno;
608 	}
609 skip:
610 	if (iter < dirty_threshold) {
611 		node = rb_next(node);
612 		goto next;
613 	}
614 }
615 
616 /*
617  * select candidates around source section in range of
618  * [target - dirty_threshold, target + dirty_threshold]
619  */
620 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
621 						struct victim_sel_policy *p)
622 {
623 	struct sit_info *sit_i = SIT_I(sbi);
624 	struct atgc_management *am = &sbi->am;
625 	struct victim_entry *ve;
626 	unsigned long long age;
627 	unsigned long long max_mtime = sit_i->dirty_max_mtime;
628 	unsigned long long min_mtime = sit_i->dirty_min_mtime;
629 	unsigned int vblocks;
630 	unsigned int dirty_threshold = max(am->max_candidate_count,
631 					am->candidate_ratio *
632 					am->victim_count / 100);
633 	unsigned int cost, iter;
634 	int stage = 0;
635 
636 	if (max_mtime < min_mtime)
637 		return;
638 	max_mtime += 1;
639 next_stage:
640 	iter = 0;
641 	ve = __lookup_victim_entry(sbi, p->age);
642 next_node:
643 	if (!ve) {
644 		if (stage++ == 0)
645 			goto next_stage;
646 		return;
647 	}
648 
649 	if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
650 		goto skip_node;
651 
652 	age = max_mtime - ve->mtime;
653 
654 	vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
655 	f2fs_bug_on(sbi, !vblocks);
656 
657 	/* rare case */
658 	if (vblocks == BLKS_PER_SEG(sbi))
659 		goto skip_node;
660 
661 	iter++;
662 
663 	age = max_mtime - abs(p->age - age);
664 	cost = UINT_MAX - vblocks;
665 
666 	if (cost < p->min_cost ||
667 			(cost == p->min_cost && age > p->oldest_age)) {
668 		p->min_cost = cost;
669 		p->oldest_age = age;
670 		p->min_segno = ve->segno;
671 	}
672 skip_node:
673 	if (iter < dirty_threshold) {
674 		ve = rb_entry(stage == 0 ? rb_prev(&ve->rb_node) :
675 					rb_next(&ve->rb_node),
676 					struct victim_entry, rb_node);
677 		goto next_node;
678 	}
679 
680 	if (stage++ == 0)
681 		goto next_stage;
682 }
683 
684 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
685 						struct victim_sel_policy *p)
686 {
687 	f2fs_bug_on(sbi, !f2fs_check_victim_tree(sbi, &sbi->am.root));
688 
689 	if (p->gc_mode == GC_AT)
690 		atgc_lookup_victim(sbi, p);
691 	else if (p->alloc_mode == AT_SSR)
692 		atssr_lookup_victim(sbi, p);
693 	else
694 		f2fs_bug_on(sbi, 1);
695 }
696 
697 static void release_victim_entry(struct f2fs_sb_info *sbi)
698 {
699 	struct atgc_management *am = &sbi->am;
700 	struct victim_entry *ve, *tmp;
701 
702 	list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
703 		list_del(&ve->list);
704 		kmem_cache_free(victim_entry_slab, ve);
705 		am->victim_count--;
706 	}
707 
708 	am->root = RB_ROOT_CACHED;
709 
710 	f2fs_bug_on(sbi, am->victim_count);
711 	f2fs_bug_on(sbi, !list_empty(&am->victim_list));
712 }
713 
714 static bool f2fs_pin_section(struct f2fs_sb_info *sbi, unsigned int segno)
715 {
716 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
717 	unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
718 
719 	if (!dirty_i->enable_pin_section)
720 		return false;
721 	if (!test_and_set_bit(secno, dirty_i->pinned_secmap))
722 		dirty_i->pinned_secmap_cnt++;
723 	return true;
724 }
725 
726 static bool f2fs_pinned_section_exists(struct dirty_seglist_info *dirty_i)
727 {
728 	return dirty_i->pinned_secmap_cnt;
729 }
730 
731 static bool f2fs_section_is_pinned(struct dirty_seglist_info *dirty_i,
732 						unsigned int secno)
733 {
734 	return dirty_i->enable_pin_section &&
735 		f2fs_pinned_section_exists(dirty_i) &&
736 		test_bit(secno, dirty_i->pinned_secmap);
737 }
738 
739 static void f2fs_unpin_all_sections(struct f2fs_sb_info *sbi, bool enable)
740 {
741 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
742 
743 	if (f2fs_pinned_section_exists(DIRTY_I(sbi))) {
744 		memset(DIRTY_I(sbi)->pinned_secmap, 0, bitmap_size);
745 		DIRTY_I(sbi)->pinned_secmap_cnt = 0;
746 	}
747 	DIRTY_I(sbi)->enable_pin_section = enable;
748 }
749 
750 static int f2fs_gc_pinned_control(struct inode *inode, int gc_type,
751 							unsigned int segno)
752 {
753 	if (!f2fs_is_pinned_file(inode))
754 		return 0;
755 	if (gc_type != FG_GC)
756 		return -EBUSY;
757 	if (!f2fs_pin_section(F2FS_I_SB(inode), segno))
758 		f2fs_pin_file_control(inode, true);
759 	return -EAGAIN;
760 }
761 
762 /*
763  * This function is called from two paths.
764  * One is garbage collection and the other is SSR segment selection.
765  * When it is called during GC, it just gets a victim segment
766  * and it does not remove it from dirty seglist.
767  * When it is called from SSR segment selection, it finds a segment
768  * which has minimum valid blocks and removes it from dirty seglist.
769  */
770 int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
771 			int gc_type, int type, char alloc_mode,
772 			unsigned long long age, bool one_time)
773 {
774 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
775 	struct sit_info *sm = SIT_I(sbi);
776 	struct victim_sel_policy p;
777 	unsigned int secno, last_victim;
778 	unsigned int last_segment;
779 	unsigned int nsearched;
780 	bool is_atgc;
781 	int ret = 0;
782 
783 	mutex_lock(&dirty_i->seglist_lock);
784 	last_segment = MAIN_SECS(sbi) * SEGS_PER_SEC(sbi);
785 
786 	p.alloc_mode = alloc_mode;
787 	p.age = age;
788 	p.age_threshold = sbi->am.age_threshold;
789 	p.one_time_gc = one_time;
790 
791 retry:
792 	select_policy(sbi, gc_type, type, &p);
793 	p.min_segno = NULL_SEGNO;
794 	p.oldest_age = 0;
795 	p.min_cost = get_max_cost(sbi, &p);
796 
797 	is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
798 	nsearched = 0;
799 
800 	if (is_atgc)
801 		SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
802 
803 	if (*result != NULL_SEGNO) {
804 		if (!get_valid_blocks(sbi, *result, false)) {
805 			ret = -ENODATA;
806 			goto out;
807 		}
808 
809 		if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
810 			ret = -EBUSY;
811 		else
812 			p.min_segno = *result;
813 		goto out;
814 	}
815 
816 	ret = -ENODATA;
817 	if (p.max_search == 0)
818 		goto out;
819 
820 	if (__is_large_section(sbi) && p.alloc_mode == LFS) {
821 		if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
822 			p.min_segno = sbi->next_victim_seg[BG_GC];
823 			*result = p.min_segno;
824 			sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
825 			goto got_result;
826 		}
827 		if (gc_type == FG_GC &&
828 				sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
829 			p.min_segno = sbi->next_victim_seg[FG_GC];
830 			*result = p.min_segno;
831 			sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
832 			goto got_result;
833 		}
834 	}
835 
836 	last_victim = sm->last_victim[p.gc_mode];
837 	if (p.alloc_mode == LFS && gc_type == FG_GC) {
838 		p.min_segno = check_bg_victims(sbi);
839 		if (p.min_segno != NULL_SEGNO)
840 			goto got_it;
841 	}
842 
843 	while (1) {
844 		unsigned long cost, *dirty_bitmap;
845 		unsigned int unit_no, segno;
846 
847 		dirty_bitmap = p.dirty_bitmap;
848 		unit_no = find_next_bit(dirty_bitmap,
849 				last_segment / p.ofs_unit,
850 				p.offset / p.ofs_unit);
851 		segno = unit_no * p.ofs_unit;
852 		if (segno >= last_segment) {
853 			if (sm->last_victim[p.gc_mode]) {
854 				last_segment =
855 					sm->last_victim[p.gc_mode];
856 				sm->last_victim[p.gc_mode] = 0;
857 				p.offset = 0;
858 				continue;
859 			}
860 			break;
861 		}
862 
863 		p.offset = segno + p.ofs_unit;
864 		nsearched++;
865 
866 #ifdef CONFIG_F2FS_CHECK_FS
867 		/*
868 		 * skip selecting the invalid segno (that is failed due to block
869 		 * validity check failure during GC) to avoid endless GC loop in
870 		 * such cases.
871 		 */
872 		if (test_bit(segno, sm->invalid_segmap))
873 			goto next;
874 #endif
875 
876 		secno = GET_SEC_FROM_SEG(sbi, segno);
877 
878 		if (sec_usage_check(sbi, secno))
879 			goto next;
880 
881 		/* Don't touch checkpointed data */
882 		if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
883 			if (p.alloc_mode == LFS) {
884 				/*
885 				 * LFS is set to find source section during GC.
886 				 * The victim should have no checkpointed data.
887 				 */
888 				if (get_ckpt_valid_blocks(sbi, segno, true))
889 					goto next;
890 			} else {
891 				/*
892 				 * SSR | AT_SSR are set to find target segment
893 				 * for writes which can be full by checkpointed
894 				 * and newly written blocks.
895 				 */
896 				if (!f2fs_segment_has_free_slot(sbi, segno))
897 					goto next;
898 			}
899 		}
900 
901 		if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
902 			goto next;
903 
904 		if (gc_type == FG_GC && f2fs_section_is_pinned(dirty_i, secno))
905 			goto next;
906 
907 		if (is_atgc) {
908 			add_victim_entry(sbi, &p, segno);
909 			goto next;
910 		}
911 
912 		cost = get_gc_cost(sbi, segno, &p);
913 
914 		if (p.min_cost > cost) {
915 			p.min_segno = segno;
916 			p.min_cost = cost;
917 		}
918 next:
919 		if (nsearched >= p.max_search) {
920 			if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
921 				sm->last_victim[p.gc_mode] =
922 					last_victim + p.ofs_unit;
923 			else
924 				sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
925 			sm->last_victim[p.gc_mode] %=
926 				(MAIN_SECS(sbi) * SEGS_PER_SEC(sbi));
927 			break;
928 		}
929 	}
930 
931 	/* get victim for GC_AT/AT_SSR */
932 	if (is_atgc) {
933 		lookup_victim_by_age(sbi, &p);
934 		release_victim_entry(sbi);
935 	}
936 
937 	if (is_atgc && p.min_segno == NULL_SEGNO &&
938 			sm->elapsed_time < p.age_threshold) {
939 		p.age_threshold = 0;
940 		goto retry;
941 	}
942 
943 	if (p.min_segno != NULL_SEGNO) {
944 got_it:
945 		*result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
946 got_result:
947 		if (p.alloc_mode == LFS) {
948 			secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
949 			if (gc_type == FG_GC)
950 				sbi->cur_victim_sec = secno;
951 			else
952 				set_bit(secno, dirty_i->victim_secmap);
953 		}
954 		ret = 0;
955 
956 	}
957 out:
958 	if (p.min_segno != NULL_SEGNO)
959 		trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
960 				sbi->cur_victim_sec,
961 				prefree_segments(sbi), free_segments(sbi));
962 	mutex_unlock(&dirty_i->seglist_lock);
963 
964 	return ret;
965 }
966 
967 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
968 {
969 	struct inode_entry *ie;
970 
971 	ie = radix_tree_lookup(&gc_list->iroot, ino);
972 	if (ie)
973 		return ie->inode;
974 	return NULL;
975 }
976 
977 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
978 {
979 	struct inode_entry *new_ie;
980 
981 	if (inode == find_gc_inode(gc_list, inode->i_ino)) {
982 		iput(inode);
983 		return;
984 	}
985 	new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab,
986 					GFP_NOFS, true, NULL);
987 	new_ie->inode = inode;
988 
989 	f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
990 	list_add_tail(&new_ie->list, &gc_list->ilist);
991 }
992 
993 static void put_gc_inode(struct gc_inode_list *gc_list)
994 {
995 	struct inode_entry *ie, *next_ie;
996 
997 	list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
998 		radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
999 		iput(ie->inode);
1000 		list_del(&ie->list);
1001 		kmem_cache_free(f2fs_inode_entry_slab, ie);
1002 	}
1003 }
1004 
1005 static int check_valid_map(struct f2fs_sb_info *sbi,
1006 				unsigned int segno, int offset)
1007 {
1008 	struct sit_info *sit_i = SIT_I(sbi);
1009 	struct seg_entry *sentry;
1010 	int ret;
1011 
1012 	down_read(&sit_i->sentry_lock);
1013 	sentry = get_seg_entry(sbi, segno);
1014 	ret = f2fs_test_bit(offset, sentry->cur_valid_map);
1015 	up_read(&sit_i->sentry_lock);
1016 	return ret;
1017 }
1018 
1019 /*
1020  * This function compares node address got in summary with that in NAT.
1021  * On validity, copy that node with cold status, otherwise (invalid node)
1022  * ignore that.
1023  */
1024 static int gc_node_segment(struct f2fs_sb_info *sbi,
1025 		struct f2fs_summary *sum, unsigned int segno, int gc_type)
1026 {
1027 	struct f2fs_summary *entry;
1028 	block_t start_addr;
1029 	int off;
1030 	int phase = 0;
1031 	bool fggc = (gc_type == FG_GC);
1032 	int submitted = 0;
1033 	unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1034 
1035 	start_addr = START_BLOCK(sbi, segno);
1036 
1037 next_step:
1038 	entry = sum;
1039 
1040 	if (fggc && phase == 2)
1041 		atomic_inc(&sbi->wb_sync_req[NODE]);
1042 
1043 	for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1044 		nid_t nid = le32_to_cpu(entry->nid);
1045 		struct page *node_page;
1046 		struct node_info ni;
1047 		int err;
1048 
1049 		/* stop BG_GC if there is not enough free sections. */
1050 		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
1051 			return submitted;
1052 
1053 		if (check_valid_map(sbi, segno, off) == 0)
1054 			continue;
1055 
1056 		if (phase == 0) {
1057 			f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1058 							META_NAT, true);
1059 			continue;
1060 		}
1061 
1062 		if (phase == 1) {
1063 			f2fs_ra_node_page(sbi, nid);
1064 			continue;
1065 		}
1066 
1067 		/* phase == 2 */
1068 		node_page = f2fs_get_node_page(sbi, nid);
1069 		if (IS_ERR(node_page))
1070 			continue;
1071 
1072 		/* block may become invalid during f2fs_get_node_page */
1073 		if (check_valid_map(sbi, segno, off) == 0) {
1074 			f2fs_put_page(node_page, 1);
1075 			continue;
1076 		}
1077 
1078 		if (f2fs_get_node_info(sbi, nid, &ni, false)) {
1079 			f2fs_put_page(node_page, 1);
1080 			continue;
1081 		}
1082 
1083 		if (ni.blk_addr != start_addr + off) {
1084 			f2fs_put_page(node_page, 1);
1085 			continue;
1086 		}
1087 
1088 		err = f2fs_move_node_page(node_page, gc_type);
1089 		if (!err && gc_type == FG_GC)
1090 			submitted++;
1091 		stat_inc_node_blk_count(sbi, 1, gc_type);
1092 	}
1093 
1094 	if (++phase < 3)
1095 		goto next_step;
1096 
1097 	if (fggc)
1098 		atomic_dec(&sbi->wb_sync_req[NODE]);
1099 	return submitted;
1100 }
1101 
1102 /*
1103  * Calculate start block index indicating the given node offset.
1104  * Be careful, caller should give this node offset only indicating direct node
1105  * blocks. If any node offsets, which point the other types of node blocks such
1106  * as indirect or double indirect node blocks, are given, it must be a caller's
1107  * bug.
1108  */
1109 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
1110 {
1111 	unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
1112 	unsigned int bidx;
1113 
1114 	if (node_ofs == 0)
1115 		return 0;
1116 
1117 	if (node_ofs <= 2) {
1118 		bidx = node_ofs - 1;
1119 	} else if (node_ofs <= indirect_blks) {
1120 		int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
1121 
1122 		bidx = node_ofs - 2 - dec;
1123 	} else {
1124 		int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
1125 
1126 		bidx = node_ofs - 5 - dec;
1127 	}
1128 	return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
1129 }
1130 
1131 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1132 		struct node_info *dni, block_t blkaddr, unsigned int *nofs)
1133 {
1134 	struct page *node_page;
1135 	nid_t nid;
1136 	unsigned int ofs_in_node, max_addrs, base;
1137 	block_t source_blkaddr;
1138 
1139 	nid = le32_to_cpu(sum->nid);
1140 	ofs_in_node = le16_to_cpu(sum->ofs_in_node);
1141 
1142 	node_page = f2fs_get_node_page(sbi, nid);
1143 	if (IS_ERR(node_page))
1144 		return false;
1145 
1146 	if (f2fs_get_node_info(sbi, nid, dni, false)) {
1147 		f2fs_put_page(node_page, 1);
1148 		return false;
1149 	}
1150 
1151 	if (sum->version != dni->version) {
1152 		f2fs_warn(sbi, "%s: valid data with mismatched node version.",
1153 			  __func__);
1154 		set_sbi_flag(sbi, SBI_NEED_FSCK);
1155 	}
1156 
1157 	if (f2fs_check_nid_range(sbi, dni->ino)) {
1158 		f2fs_put_page(node_page, 1);
1159 		return false;
1160 	}
1161 
1162 	if (IS_INODE(node_page)) {
1163 		base = offset_in_addr(F2FS_INODE(node_page));
1164 		max_addrs = DEF_ADDRS_PER_INODE;
1165 	} else {
1166 		base = 0;
1167 		max_addrs = DEF_ADDRS_PER_BLOCK;
1168 	}
1169 
1170 	if (base + ofs_in_node >= max_addrs) {
1171 		f2fs_err(sbi, "Inconsistent blkaddr offset: base:%u, ofs_in_node:%u, max:%u, ino:%u, nid:%u",
1172 			base, ofs_in_node, max_addrs, dni->ino, dni->nid);
1173 		f2fs_put_page(node_page, 1);
1174 		return false;
1175 	}
1176 
1177 	*nofs = ofs_of_node(node_page);
1178 	source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1179 	f2fs_put_page(node_page, 1);
1180 
1181 	if (source_blkaddr != blkaddr) {
1182 #ifdef CONFIG_F2FS_CHECK_FS
1183 		unsigned int segno = GET_SEGNO(sbi, blkaddr);
1184 		unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1185 
1186 		if (unlikely(check_valid_map(sbi, segno, offset))) {
1187 			if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1188 				f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",
1189 					 blkaddr, source_blkaddr, segno);
1190 				set_sbi_flag(sbi, SBI_NEED_FSCK);
1191 			}
1192 		}
1193 #endif
1194 		return false;
1195 	}
1196 	return true;
1197 }
1198 
1199 static int ra_data_block(struct inode *inode, pgoff_t index)
1200 {
1201 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1202 	struct address_space *mapping = f2fs_is_cow_file(inode) ?
1203 				F2FS_I(inode)->atomic_inode->i_mapping : inode->i_mapping;
1204 	struct dnode_of_data dn;
1205 	struct page *page;
1206 	struct f2fs_io_info fio = {
1207 		.sbi = sbi,
1208 		.ino = inode->i_ino,
1209 		.type = DATA,
1210 		.temp = COLD,
1211 		.op = REQ_OP_READ,
1212 		.op_flags = 0,
1213 		.encrypted_page = NULL,
1214 		.in_list = 0,
1215 	};
1216 	int err;
1217 
1218 	page = f2fs_grab_cache_page(mapping, index, true);
1219 	if (!page)
1220 		return -ENOMEM;
1221 
1222 	if (f2fs_lookup_read_extent_cache_block(inode, index,
1223 						&dn.data_blkaddr)) {
1224 		if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1225 						DATA_GENERIC_ENHANCE_READ))) {
1226 			err = -EFSCORRUPTED;
1227 			goto put_page;
1228 		}
1229 		goto got_it;
1230 	}
1231 
1232 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1233 	err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1234 	if (err)
1235 		goto put_page;
1236 	f2fs_put_dnode(&dn);
1237 
1238 	if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1239 		err = -ENOENT;
1240 		goto put_page;
1241 	}
1242 	if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1243 						DATA_GENERIC_ENHANCE))) {
1244 		err = -EFSCORRUPTED;
1245 		goto put_page;
1246 	}
1247 got_it:
1248 	/* read page */
1249 	fio.page = page;
1250 	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1251 
1252 	/*
1253 	 * don't cache encrypted data into meta inode until previous dirty
1254 	 * data were writebacked to avoid racing between GC and flush.
1255 	 */
1256 	f2fs_wait_on_page_writeback(page, DATA, true, true);
1257 
1258 	f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1259 
1260 	fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1261 					dn.data_blkaddr,
1262 					FGP_LOCK | FGP_CREAT, GFP_NOFS);
1263 	if (!fio.encrypted_page) {
1264 		err = -ENOMEM;
1265 		goto put_page;
1266 	}
1267 
1268 	err = f2fs_submit_page_bio(&fio);
1269 	if (err)
1270 		goto put_encrypted_page;
1271 	f2fs_put_page(fio.encrypted_page, 0);
1272 	f2fs_put_page(page, 1);
1273 
1274 	f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
1275 	f2fs_update_iostat(sbi, NULL, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1276 
1277 	return 0;
1278 put_encrypted_page:
1279 	f2fs_put_page(fio.encrypted_page, 1);
1280 put_page:
1281 	f2fs_put_page(page, 1);
1282 	return err;
1283 }
1284 
1285 /*
1286  * Move data block via META_MAPPING while keeping locked data page.
1287  * This can be used to move blocks, aka LBAs, directly on disk.
1288  */
1289 static int move_data_block(struct inode *inode, block_t bidx,
1290 				int gc_type, unsigned int segno, int off)
1291 {
1292 	struct address_space *mapping = f2fs_is_cow_file(inode) ?
1293 				F2FS_I(inode)->atomic_inode->i_mapping : inode->i_mapping;
1294 	struct f2fs_io_info fio = {
1295 		.sbi = F2FS_I_SB(inode),
1296 		.ino = inode->i_ino,
1297 		.type = DATA,
1298 		.temp = COLD,
1299 		.op = REQ_OP_READ,
1300 		.op_flags = 0,
1301 		.encrypted_page = NULL,
1302 		.in_list = 0,
1303 	};
1304 	struct dnode_of_data dn;
1305 	struct f2fs_summary sum;
1306 	struct node_info ni;
1307 	struct page *page, *mpage;
1308 	block_t newaddr;
1309 	int err = 0;
1310 	bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1311 	int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&
1312 				(fio.sbi->gc_mode != GC_URGENT_HIGH) ?
1313 				CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1314 
1315 	/* do not read out */
1316 	page = f2fs_grab_cache_page(mapping, bidx, false);
1317 	if (!page)
1318 		return -ENOMEM;
1319 
1320 	if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1321 		err = -ENOENT;
1322 		goto out;
1323 	}
1324 
1325 	err = f2fs_gc_pinned_control(inode, gc_type, segno);
1326 	if (err)
1327 		goto out;
1328 
1329 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1330 	err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1331 	if (err)
1332 		goto out;
1333 
1334 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1335 		ClearPageUptodate(page);
1336 		err = -ENOENT;
1337 		goto put_out;
1338 	}
1339 
1340 	/*
1341 	 * don't cache encrypted data into meta inode until previous dirty
1342 	 * data were writebacked to avoid racing between GC and flush.
1343 	 */
1344 	f2fs_wait_on_page_writeback(page, DATA, true, true);
1345 
1346 	f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1347 
1348 	err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);
1349 	if (err)
1350 		goto put_out;
1351 
1352 	/* read page */
1353 	fio.page = page;
1354 	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1355 
1356 	if (lfs_mode)
1357 		f2fs_down_write(&fio.sbi->io_order_lock);
1358 
1359 	mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1360 					fio.old_blkaddr, false);
1361 	if (!mpage) {
1362 		err = -ENOMEM;
1363 		goto up_out;
1364 	}
1365 
1366 	fio.encrypted_page = mpage;
1367 
1368 	/* read source block in mpage */
1369 	if (!PageUptodate(mpage)) {
1370 		err = f2fs_submit_page_bio(&fio);
1371 		if (err) {
1372 			f2fs_put_page(mpage, 1);
1373 			goto up_out;
1374 		}
1375 
1376 		f2fs_update_iostat(fio.sbi, inode, FS_DATA_READ_IO,
1377 							F2FS_BLKSIZE);
1378 		f2fs_update_iostat(fio.sbi, NULL, FS_GDATA_READ_IO,
1379 							F2FS_BLKSIZE);
1380 
1381 		lock_page(mpage);
1382 		if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1383 						!PageUptodate(mpage))) {
1384 			err = -EIO;
1385 			f2fs_put_page(mpage, 1);
1386 			goto up_out;
1387 		}
1388 	}
1389 
1390 	set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1391 
1392 	/* allocate block address */
1393 	err = f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1394 				&sum, type, NULL);
1395 	if (err) {
1396 		f2fs_put_page(mpage, 1);
1397 		/* filesystem should shutdown, no need to recovery block */
1398 		goto up_out;
1399 	}
1400 
1401 	fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1402 				newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1403 	if (!fio.encrypted_page) {
1404 		err = -ENOMEM;
1405 		f2fs_put_page(mpage, 1);
1406 		goto recover_block;
1407 	}
1408 
1409 	/* write target block */
1410 	f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1411 	memcpy(page_address(fio.encrypted_page),
1412 				page_address(mpage), PAGE_SIZE);
1413 	f2fs_put_page(mpage, 1);
1414 
1415 	f2fs_invalidate_internal_cache(fio.sbi, fio.old_blkaddr);
1416 
1417 	set_page_dirty(fio.encrypted_page);
1418 	if (clear_page_dirty_for_io(fio.encrypted_page))
1419 		dec_page_count(fio.sbi, F2FS_DIRTY_META);
1420 
1421 	set_page_writeback(fio.encrypted_page);
1422 
1423 	fio.op = REQ_OP_WRITE;
1424 	fio.op_flags = REQ_SYNC;
1425 	fio.new_blkaddr = newaddr;
1426 	f2fs_submit_page_write(&fio);
1427 
1428 	f2fs_update_iostat(fio.sbi, NULL, FS_GC_DATA_IO, F2FS_BLKSIZE);
1429 
1430 	f2fs_update_data_blkaddr(&dn, newaddr);
1431 	set_inode_flag(inode, FI_APPEND_WRITE);
1432 
1433 	f2fs_put_page(fio.encrypted_page, 1);
1434 recover_block:
1435 	if (err)
1436 		f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1437 							true, true, true);
1438 up_out:
1439 	if (lfs_mode)
1440 		f2fs_up_write(&fio.sbi->io_order_lock);
1441 put_out:
1442 	f2fs_put_dnode(&dn);
1443 out:
1444 	f2fs_put_page(page, 1);
1445 	return err;
1446 }
1447 
1448 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1449 							unsigned int segno, int off)
1450 {
1451 	struct page *page;
1452 	int err = 0;
1453 
1454 	page = f2fs_get_lock_data_page(inode, bidx, true);
1455 	if (IS_ERR(page))
1456 		return PTR_ERR(page);
1457 
1458 	if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1459 		err = -ENOENT;
1460 		goto out;
1461 	}
1462 
1463 	err = f2fs_gc_pinned_control(inode, gc_type, segno);
1464 	if (err)
1465 		goto out;
1466 
1467 	if (gc_type == BG_GC) {
1468 		if (folio_test_writeback(page_folio(page))) {
1469 			err = -EAGAIN;
1470 			goto out;
1471 		}
1472 		set_page_dirty(page);
1473 		set_page_private_gcing(page);
1474 	} else {
1475 		struct f2fs_io_info fio = {
1476 			.sbi = F2FS_I_SB(inode),
1477 			.ino = inode->i_ino,
1478 			.type = DATA,
1479 			.temp = COLD,
1480 			.op = REQ_OP_WRITE,
1481 			.op_flags = REQ_SYNC,
1482 			.old_blkaddr = NULL_ADDR,
1483 			.page = page,
1484 			.encrypted_page = NULL,
1485 			.need_lock = LOCK_REQ,
1486 			.io_type = FS_GC_DATA_IO,
1487 		};
1488 		bool is_dirty = PageDirty(page);
1489 
1490 retry:
1491 		f2fs_wait_on_page_writeback(page, DATA, true, true);
1492 
1493 		set_page_dirty(page);
1494 		if (clear_page_dirty_for_io(page)) {
1495 			inode_dec_dirty_pages(inode);
1496 			f2fs_remove_dirty_inode(inode);
1497 		}
1498 
1499 		set_page_private_gcing(page);
1500 
1501 		err = f2fs_do_write_data_page(&fio);
1502 		if (err) {
1503 			clear_page_private_gcing(page);
1504 			if (err == -ENOMEM) {
1505 				memalloc_retry_wait(GFP_NOFS);
1506 				goto retry;
1507 			}
1508 			if (is_dirty)
1509 				set_page_dirty(page);
1510 		}
1511 	}
1512 out:
1513 	f2fs_put_page(page, 1);
1514 	return err;
1515 }
1516 
1517 /*
1518  * This function tries to get parent node of victim data block, and identifies
1519  * data block validity. If the block is valid, copy that with cold status and
1520  * modify parent node.
1521  * If the parent node is not valid or the data block address is different,
1522  * the victim data block is ignored.
1523  */
1524 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1525 		struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1526 		bool force_migrate)
1527 {
1528 	struct super_block *sb = sbi->sb;
1529 	struct f2fs_summary *entry;
1530 	block_t start_addr;
1531 	int off;
1532 	int phase = 0;
1533 	int submitted = 0;
1534 	unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1535 
1536 	start_addr = START_BLOCK(sbi, segno);
1537 
1538 next_step:
1539 	entry = sum;
1540 
1541 	for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1542 		struct page *data_page;
1543 		struct inode *inode;
1544 		struct node_info dni; /* dnode info for the data */
1545 		unsigned int ofs_in_node, nofs;
1546 		block_t start_bidx;
1547 		nid_t nid = le32_to_cpu(entry->nid);
1548 
1549 		/*
1550 		 * stop BG_GC if there is not enough free sections.
1551 		 * Or, stop GC if the segment becomes fully valid caused by
1552 		 * race condition along with SSR block allocation.
1553 		 */
1554 		if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1555 			(!force_migrate && get_valid_blocks(sbi, segno, true) ==
1556 							CAP_BLKS_PER_SEC(sbi)))
1557 			return submitted;
1558 
1559 		if (check_valid_map(sbi, segno, off) == 0)
1560 			continue;
1561 
1562 		if (phase == 0) {
1563 			f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1564 							META_NAT, true);
1565 			continue;
1566 		}
1567 
1568 		if (phase == 1) {
1569 			f2fs_ra_node_page(sbi, nid);
1570 			continue;
1571 		}
1572 
1573 		/* Get an inode by ino with checking validity */
1574 		if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1575 			continue;
1576 
1577 		if (phase == 2) {
1578 			f2fs_ra_node_page(sbi, dni.ino);
1579 			continue;
1580 		}
1581 
1582 		ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1583 
1584 		if (phase == 3) {
1585 			int err;
1586 
1587 			inode = f2fs_iget(sb, dni.ino);
1588 			if (IS_ERR(inode))
1589 				continue;
1590 
1591 			if (is_bad_inode(inode) ||
1592 					special_file(inode->i_mode)) {
1593 				iput(inode);
1594 				continue;
1595 			}
1596 
1597 			if (f2fs_has_inline_data(inode)) {
1598 				iput(inode);
1599 				set_sbi_flag(sbi, SBI_NEED_FSCK);
1600 				f2fs_err_ratelimited(sbi,
1601 					"inode %lx has both inline_data flag and "
1602 					"data block, nid=%u, ofs_in_node=%u",
1603 					inode->i_ino, dni.nid, ofs_in_node);
1604 				continue;
1605 			}
1606 
1607 			err = f2fs_gc_pinned_control(inode, gc_type, segno);
1608 			if (err == -EAGAIN) {
1609 				iput(inode);
1610 				return submitted;
1611 			}
1612 
1613 			if (!f2fs_down_write_trylock(
1614 				&F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1615 				iput(inode);
1616 				sbi->skipped_gc_rwsem++;
1617 				continue;
1618 			}
1619 
1620 			start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1621 								ofs_in_node;
1622 
1623 			if (f2fs_meta_inode_gc_required(inode)) {
1624 				int err = ra_data_block(inode, start_bidx);
1625 
1626 				f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1627 				if (err) {
1628 					iput(inode);
1629 					continue;
1630 				}
1631 				add_gc_inode(gc_list, inode);
1632 				continue;
1633 			}
1634 
1635 			data_page = f2fs_get_read_data_page(inode, start_bidx,
1636 							REQ_RAHEAD, true, NULL);
1637 			f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1638 			if (IS_ERR(data_page)) {
1639 				iput(inode);
1640 				continue;
1641 			}
1642 
1643 			f2fs_put_page(data_page, 0);
1644 			add_gc_inode(gc_list, inode);
1645 			continue;
1646 		}
1647 
1648 		/* phase 4 */
1649 		inode = find_gc_inode(gc_list, dni.ino);
1650 		if (inode) {
1651 			struct f2fs_inode_info *fi = F2FS_I(inode);
1652 			bool locked = false;
1653 			int err;
1654 
1655 			if (S_ISREG(inode->i_mode)) {
1656 				if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[WRITE])) {
1657 					sbi->skipped_gc_rwsem++;
1658 					continue;
1659 				}
1660 				if (!f2fs_down_write_trylock(
1661 						&fi->i_gc_rwsem[READ])) {
1662 					sbi->skipped_gc_rwsem++;
1663 					f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
1664 					continue;
1665 				}
1666 				locked = true;
1667 
1668 				/* wait for all inflight aio data */
1669 				inode_dio_wait(inode);
1670 			}
1671 
1672 			start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1673 								+ ofs_in_node;
1674 			if (f2fs_meta_inode_gc_required(inode))
1675 				err = move_data_block(inode, start_bidx,
1676 							gc_type, segno, off);
1677 			else
1678 				err = move_data_page(inode, start_bidx, gc_type,
1679 								segno, off);
1680 
1681 			if (!err && (gc_type == FG_GC ||
1682 					f2fs_meta_inode_gc_required(inode)))
1683 				submitted++;
1684 
1685 			if (locked) {
1686 				f2fs_up_write(&fi->i_gc_rwsem[READ]);
1687 				f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
1688 			}
1689 
1690 			stat_inc_data_blk_count(sbi, 1, gc_type);
1691 		}
1692 	}
1693 
1694 	if (++phase < 5)
1695 		goto next_step;
1696 
1697 	return submitted;
1698 }
1699 
1700 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1701 			int gc_type, bool one_time)
1702 {
1703 	struct sit_info *sit_i = SIT_I(sbi);
1704 	int ret;
1705 
1706 	down_write(&sit_i->sentry_lock);
1707 	ret = f2fs_get_victim(sbi, victim, gc_type, NO_CHECK_TYPE,
1708 			LFS, 0, one_time);
1709 	up_write(&sit_i->sentry_lock);
1710 	return ret;
1711 }
1712 
1713 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1714 				unsigned int start_segno,
1715 				struct gc_inode_list *gc_list, int gc_type,
1716 				bool force_migrate, bool one_time)
1717 {
1718 	struct page *sum_page;
1719 	struct f2fs_summary_block *sum;
1720 	struct blk_plug plug;
1721 	unsigned int segno = start_segno;
1722 	unsigned int end_segno = start_segno + SEGS_PER_SEC(sbi);
1723 	unsigned int sec_end_segno;
1724 	int seg_freed = 0, migrated = 0;
1725 	unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1726 						SUM_TYPE_DATA : SUM_TYPE_NODE;
1727 	unsigned char data_type = (type == SUM_TYPE_DATA) ? DATA : NODE;
1728 	int submitted = 0;
1729 
1730 	if (__is_large_section(sbi)) {
1731 		sec_end_segno = rounddown(end_segno, SEGS_PER_SEC(sbi));
1732 
1733 		/*
1734 		 * zone-capacity can be less than zone-size in zoned devices,
1735 		 * resulting in less than expected usable segments in the zone,
1736 		 * calculate the end segno in the zone which can be garbage
1737 		 * collected
1738 		 */
1739 		if (f2fs_sb_has_blkzoned(sbi))
1740 			sec_end_segno -= SEGS_PER_SEC(sbi) -
1741 					f2fs_usable_segs_in_sec(sbi);
1742 
1743 		if (gc_type == BG_GC || one_time) {
1744 			unsigned int window_granularity =
1745 				sbi->migration_window_granularity;
1746 
1747 			if (f2fs_sb_has_blkzoned(sbi) &&
1748 					!has_enough_free_blocks(sbi,
1749 					sbi->gc_thread->boost_zoned_gc_percent))
1750 				window_granularity *=
1751 					BOOST_GC_MULTIPLE;
1752 
1753 			end_segno = start_segno + window_granularity;
1754 		}
1755 
1756 		if (end_segno > sec_end_segno)
1757 			end_segno = sec_end_segno;
1758 	}
1759 
1760 	sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1761 
1762 	/* readahead multi ssa blocks those have contiguous address */
1763 	if (__is_large_section(sbi))
1764 		f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1765 					end_segno - segno, META_SSA, true);
1766 
1767 	/* reference all summary page */
1768 	while (segno < end_segno) {
1769 		sum_page = f2fs_get_sum_page(sbi, segno++);
1770 		if (IS_ERR(sum_page)) {
1771 			int err = PTR_ERR(sum_page);
1772 
1773 			end_segno = segno - 1;
1774 			for (segno = start_segno; segno < end_segno; segno++) {
1775 				sum_page = find_get_page(META_MAPPING(sbi),
1776 						GET_SUM_BLOCK(sbi, segno));
1777 				f2fs_put_page(sum_page, 0);
1778 				f2fs_put_page(sum_page, 0);
1779 			}
1780 			return err;
1781 		}
1782 		unlock_page(sum_page);
1783 	}
1784 
1785 	blk_start_plug(&plug);
1786 
1787 	for (segno = start_segno; segno < end_segno; segno++) {
1788 
1789 		/* find segment summary of victim */
1790 		sum_page = find_get_page(META_MAPPING(sbi),
1791 					GET_SUM_BLOCK(sbi, segno));
1792 		f2fs_put_page(sum_page, 0);
1793 
1794 		if (get_valid_blocks(sbi, segno, false) == 0)
1795 			goto freed;
1796 		if (gc_type == BG_GC && __is_large_section(sbi) &&
1797 				migrated >= sbi->migration_granularity)
1798 			goto skip;
1799 		if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1800 			goto skip;
1801 
1802 		sum = page_address(sum_page);
1803 		if (type != GET_SUM_TYPE((&sum->footer))) {
1804 			f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1805 				 segno, type, GET_SUM_TYPE((&sum->footer)));
1806 			f2fs_stop_checkpoint(sbi, false,
1807 				STOP_CP_REASON_CORRUPTED_SUMMARY);
1808 			goto skip;
1809 		}
1810 
1811 		/*
1812 		 * this is to avoid deadlock:
1813 		 * - lock_page(sum_page)         - f2fs_replace_block
1814 		 *  - check_valid_map()            - down_write(sentry_lock)
1815 		 *   - down_read(sentry_lock)     - change_curseg()
1816 		 *                                  - lock_page(sum_page)
1817 		 */
1818 		if (type == SUM_TYPE_NODE)
1819 			submitted += gc_node_segment(sbi, sum->entries, segno,
1820 								gc_type);
1821 		else
1822 			submitted += gc_data_segment(sbi, sum->entries, gc_list,
1823 							segno, gc_type,
1824 							force_migrate);
1825 
1826 		stat_inc_gc_seg_count(sbi, data_type, gc_type);
1827 		sbi->gc_reclaimed_segs[sbi->gc_mode]++;
1828 		migrated++;
1829 
1830 freed:
1831 		if (gc_type == FG_GC &&
1832 				get_valid_blocks(sbi, segno, false) == 0)
1833 			seg_freed++;
1834 
1835 		if (__is_large_section(sbi))
1836 			sbi->next_victim_seg[gc_type] =
1837 				(segno + 1 < sec_end_segno) ?
1838 					segno + 1 : NULL_SEGNO;
1839 skip:
1840 		f2fs_put_page(sum_page, 0);
1841 	}
1842 
1843 	if (submitted)
1844 		f2fs_submit_merged_write(sbi, data_type);
1845 
1846 	blk_finish_plug(&plug);
1847 
1848 	if (migrated)
1849 		stat_inc_gc_sec_count(sbi, data_type, gc_type);
1850 
1851 	return seg_freed;
1852 }
1853 
1854 int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control)
1855 {
1856 	int gc_type = gc_control->init_gc_type;
1857 	unsigned int segno = gc_control->victim_segno;
1858 	int sec_freed = 0, seg_freed = 0, total_freed = 0, total_sec_freed = 0;
1859 	int ret = 0;
1860 	struct cp_control cpc;
1861 	struct gc_inode_list gc_list = {
1862 		.ilist = LIST_HEAD_INIT(gc_list.ilist),
1863 		.iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1864 	};
1865 	unsigned int skipped_round = 0, round = 0;
1866 	unsigned int upper_secs;
1867 
1868 	trace_f2fs_gc_begin(sbi->sb, gc_type, gc_control->no_bg_gc,
1869 				gc_control->nr_free_secs,
1870 				get_pages(sbi, F2FS_DIRTY_NODES),
1871 				get_pages(sbi, F2FS_DIRTY_DENTS),
1872 				get_pages(sbi, F2FS_DIRTY_IMETA),
1873 				free_sections(sbi),
1874 				free_segments(sbi),
1875 				reserved_segments(sbi),
1876 				prefree_segments(sbi));
1877 
1878 	cpc.reason = __get_cp_reason(sbi);
1879 gc_more:
1880 	sbi->skipped_gc_rwsem = 0;
1881 	if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1882 		ret = -EINVAL;
1883 		goto stop;
1884 	}
1885 	if (unlikely(f2fs_cp_error(sbi))) {
1886 		ret = -EIO;
1887 		goto stop;
1888 	}
1889 
1890 	/* Let's run FG_GC, if we don't have enough space. */
1891 	if (has_not_enough_free_secs(sbi, 0, 0)) {
1892 		gc_type = FG_GC;
1893 
1894 		/*
1895 		 * For example, if there are many prefree_segments below given
1896 		 * threshold, we can make them free by checkpoint. Then, we
1897 		 * secure free segments which doesn't need fggc any more.
1898 		 */
1899 		if (prefree_segments(sbi)) {
1900 			stat_inc_cp_call_count(sbi, TOTAL_CALL);
1901 			ret = f2fs_write_checkpoint(sbi, &cpc);
1902 			if (ret)
1903 				goto stop;
1904 			/* Reset due to checkpoint */
1905 			sec_freed = 0;
1906 		}
1907 	}
1908 
1909 	/* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1910 	if (gc_type == BG_GC && gc_control->no_bg_gc) {
1911 		ret = -EINVAL;
1912 		goto stop;
1913 	}
1914 retry:
1915 	ret = __get_victim(sbi, &segno, gc_type, gc_control->one_time);
1916 	if (ret) {
1917 		/* allow to search victim from sections has pinned data */
1918 		if (ret == -ENODATA && gc_type == FG_GC &&
1919 				f2fs_pinned_section_exists(DIRTY_I(sbi))) {
1920 			f2fs_unpin_all_sections(sbi, false);
1921 			goto retry;
1922 		}
1923 		goto stop;
1924 	}
1925 
1926 	seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type,
1927 				gc_control->should_migrate_blocks,
1928 				gc_control->one_time);
1929 	if (seg_freed < 0)
1930 		goto stop;
1931 
1932 	total_freed += seg_freed;
1933 
1934 	if (seg_freed == f2fs_usable_segs_in_sec(sbi)) {
1935 		sec_freed++;
1936 		total_sec_freed++;
1937 	}
1938 
1939 	if (gc_control->one_time)
1940 		goto stop;
1941 
1942 	if (gc_type == FG_GC) {
1943 		sbi->cur_victim_sec = NULL_SEGNO;
1944 
1945 		if (has_enough_free_secs(sbi, sec_freed, 0)) {
1946 			if (!gc_control->no_bg_gc &&
1947 			    total_sec_freed < gc_control->nr_free_secs)
1948 				goto go_gc_more;
1949 			goto stop;
1950 		}
1951 		if (sbi->skipped_gc_rwsem)
1952 			skipped_round++;
1953 		round++;
1954 		if (skipped_round > MAX_SKIP_GC_COUNT &&
1955 				skipped_round * 2 >= round) {
1956 			stat_inc_cp_call_count(sbi, TOTAL_CALL);
1957 			ret = f2fs_write_checkpoint(sbi, &cpc);
1958 			goto stop;
1959 		}
1960 	} else if (has_enough_free_secs(sbi, 0, 0)) {
1961 		goto stop;
1962 	}
1963 
1964 	__get_secs_required(sbi, NULL, &upper_secs, NULL);
1965 
1966 	/*
1967 	 * Write checkpoint to reclaim prefree segments.
1968 	 * We need more three extra sections for writer's data/node/dentry.
1969 	 */
1970 	if (free_sections(sbi) <= upper_secs + NR_GC_CHECKPOINT_SECS &&
1971 				prefree_segments(sbi)) {
1972 		stat_inc_cp_call_count(sbi, TOTAL_CALL);
1973 		ret = f2fs_write_checkpoint(sbi, &cpc);
1974 		if (ret)
1975 			goto stop;
1976 		/* Reset due to checkpoint */
1977 		sec_freed = 0;
1978 	}
1979 go_gc_more:
1980 	segno = NULL_SEGNO;
1981 	goto gc_more;
1982 
1983 stop:
1984 	SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1985 	SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno;
1986 
1987 	if (gc_type == FG_GC)
1988 		f2fs_unpin_all_sections(sbi, true);
1989 
1990 	trace_f2fs_gc_end(sbi->sb, ret, total_freed, total_sec_freed,
1991 				get_pages(sbi, F2FS_DIRTY_NODES),
1992 				get_pages(sbi, F2FS_DIRTY_DENTS),
1993 				get_pages(sbi, F2FS_DIRTY_IMETA),
1994 				free_sections(sbi),
1995 				free_segments(sbi),
1996 				reserved_segments(sbi),
1997 				prefree_segments(sbi));
1998 
1999 	f2fs_up_write(&sbi->gc_lock);
2000 
2001 	put_gc_inode(&gc_list);
2002 
2003 	if (gc_control->err_gc_skipped && !ret)
2004 		ret = total_sec_freed ? 0 : -EAGAIN;
2005 	return ret;
2006 }
2007 
2008 int __init f2fs_create_garbage_collection_cache(void)
2009 {
2010 	victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
2011 					sizeof(struct victim_entry));
2012 	return victim_entry_slab ? 0 : -ENOMEM;
2013 }
2014 
2015 void f2fs_destroy_garbage_collection_cache(void)
2016 {
2017 	kmem_cache_destroy(victim_entry_slab);
2018 }
2019 
2020 static void init_atgc_management(struct f2fs_sb_info *sbi)
2021 {
2022 	struct atgc_management *am = &sbi->am;
2023 
2024 	if (test_opt(sbi, ATGC) &&
2025 		SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
2026 		am->atgc_enabled = true;
2027 
2028 	am->root = RB_ROOT_CACHED;
2029 	INIT_LIST_HEAD(&am->victim_list);
2030 	am->victim_count = 0;
2031 
2032 	am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
2033 	am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
2034 	am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
2035 	am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
2036 }
2037 
2038 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
2039 {
2040 	sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
2041 
2042 	/* give warm/cold data area from slower device */
2043 	if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
2044 		SIT_I(sbi)->last_victim[ALLOC_NEXT] =
2045 				GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
2046 
2047 	init_atgc_management(sbi);
2048 }
2049 
2050 int f2fs_gc_range(struct f2fs_sb_info *sbi,
2051 		unsigned int start_seg, unsigned int end_seg,
2052 		bool dry_run, unsigned int dry_run_sections)
2053 {
2054 	unsigned int segno;
2055 	unsigned int gc_secs = dry_run_sections;
2056 
2057 	if (unlikely(f2fs_cp_error(sbi)))
2058 		return -EIO;
2059 
2060 	for (segno = start_seg; segno <= end_seg; segno += SEGS_PER_SEC(sbi)) {
2061 		struct gc_inode_list gc_list = {
2062 			.ilist = LIST_HEAD_INIT(gc_list.ilist),
2063 			.iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
2064 		};
2065 
2066 		do_garbage_collect(sbi, segno, &gc_list, FG_GC, true, false);
2067 		put_gc_inode(&gc_list);
2068 
2069 		if (!dry_run && get_valid_blocks(sbi, segno, true))
2070 			return -EAGAIN;
2071 		if (dry_run && dry_run_sections &&
2072 		    !get_valid_blocks(sbi, segno, true) && --gc_secs == 0)
2073 			break;
2074 
2075 		if (fatal_signal_pending(current))
2076 			return -ERESTARTSYS;
2077 	}
2078 
2079 	return 0;
2080 }
2081 
2082 static int free_segment_range(struct f2fs_sb_info *sbi,
2083 				unsigned int secs, bool dry_run)
2084 {
2085 	unsigned int next_inuse, start, end;
2086 	struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
2087 	int gc_mode, gc_type;
2088 	int err = 0;
2089 	int type;
2090 
2091 	/* Force block allocation for GC */
2092 	MAIN_SECS(sbi) -= secs;
2093 	start = MAIN_SECS(sbi) * SEGS_PER_SEC(sbi);
2094 	end = MAIN_SEGS(sbi) - 1;
2095 
2096 	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2097 	for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
2098 		if (SIT_I(sbi)->last_victim[gc_mode] >= start)
2099 			SIT_I(sbi)->last_victim[gc_mode] = 0;
2100 
2101 	for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
2102 		if (sbi->next_victim_seg[gc_type] >= start)
2103 			sbi->next_victim_seg[gc_type] = NULL_SEGNO;
2104 	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2105 
2106 	/* Move out cursegs from the target range */
2107 	for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++) {
2108 		err = f2fs_allocate_segment_for_resize(sbi, type, start, end);
2109 		if (err)
2110 			goto out;
2111 	}
2112 
2113 	/* do GC to move out valid blocks in the range */
2114 	err = f2fs_gc_range(sbi, start, end, dry_run, 0);
2115 	if (err || dry_run)
2116 		goto out;
2117 
2118 	stat_inc_cp_call_count(sbi, TOTAL_CALL);
2119 	err = f2fs_write_checkpoint(sbi, &cpc);
2120 	if (err)
2121 		goto out;
2122 
2123 	next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
2124 	if (next_inuse <= end) {
2125 		f2fs_err(sbi, "segno %u should be free but still inuse!",
2126 			 next_inuse);
2127 		f2fs_bug_on(sbi, 1);
2128 	}
2129 out:
2130 	MAIN_SECS(sbi) += secs;
2131 	return err;
2132 }
2133 
2134 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
2135 {
2136 	struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
2137 	int section_count;
2138 	int segment_count;
2139 	int segment_count_main;
2140 	long long block_count;
2141 	int segs = secs * SEGS_PER_SEC(sbi);
2142 
2143 	f2fs_down_write(&sbi->sb_lock);
2144 
2145 	section_count = le32_to_cpu(raw_sb->section_count);
2146 	segment_count = le32_to_cpu(raw_sb->segment_count);
2147 	segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
2148 	block_count = le64_to_cpu(raw_sb->block_count);
2149 
2150 	raw_sb->section_count = cpu_to_le32(section_count + secs);
2151 	raw_sb->segment_count = cpu_to_le32(segment_count + segs);
2152 	raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
2153 	raw_sb->block_count = cpu_to_le64(block_count +
2154 			(long long)SEGS_TO_BLKS(sbi, segs));
2155 	if (f2fs_is_multi_device(sbi)) {
2156 		int last_dev = sbi->s_ndevs - 1;
2157 		int dev_segs =
2158 			le32_to_cpu(raw_sb->devs[last_dev].total_segments);
2159 
2160 		raw_sb->devs[last_dev].total_segments =
2161 						cpu_to_le32(dev_segs + segs);
2162 	}
2163 
2164 	f2fs_up_write(&sbi->sb_lock);
2165 }
2166 
2167 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
2168 {
2169 	int segs = secs * SEGS_PER_SEC(sbi);
2170 	long long blks = SEGS_TO_BLKS(sbi, segs);
2171 	long long user_block_count =
2172 				le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
2173 
2174 	SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
2175 	MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
2176 	MAIN_SECS(sbi) += secs;
2177 	FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
2178 	FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
2179 	F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
2180 
2181 	if (f2fs_is_multi_device(sbi)) {
2182 		int last_dev = sbi->s_ndevs - 1;
2183 
2184 		FDEV(last_dev).total_segments =
2185 				(int)FDEV(last_dev).total_segments + segs;
2186 		FDEV(last_dev).end_blk =
2187 				(long long)FDEV(last_dev).end_blk + blks;
2188 #ifdef CONFIG_BLK_DEV_ZONED
2189 		FDEV(last_dev).nr_blkz = FDEV(last_dev).nr_blkz +
2190 					div_u64(blks, sbi->blocks_per_blkz);
2191 #endif
2192 	}
2193 }
2194 
2195 int f2fs_resize_fs(struct file *filp, __u64 block_count)
2196 {
2197 	struct f2fs_sb_info *sbi = F2FS_I_SB(file_inode(filp));
2198 	__u64 old_block_count, shrunk_blocks;
2199 	struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
2200 	unsigned int secs;
2201 	int err = 0;
2202 	__u32 rem;
2203 
2204 	old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
2205 	if (block_count > old_block_count)
2206 		return -EINVAL;
2207 
2208 	if (f2fs_is_multi_device(sbi)) {
2209 		int last_dev = sbi->s_ndevs - 1;
2210 		__u64 last_segs = FDEV(last_dev).total_segments;
2211 
2212 		if (block_count + SEGS_TO_BLKS(sbi, last_segs) <=
2213 								old_block_count)
2214 			return -EINVAL;
2215 	}
2216 
2217 	/* new fs size should align to section size */
2218 	div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
2219 	if (rem)
2220 		return -EINVAL;
2221 
2222 	if (block_count == old_block_count)
2223 		return 0;
2224 
2225 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2226 		f2fs_err(sbi, "Should run fsck to repair first.");
2227 		return -EFSCORRUPTED;
2228 	}
2229 
2230 	if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2231 		f2fs_err(sbi, "Checkpoint should be enabled.");
2232 		return -EINVAL;
2233 	}
2234 
2235 	err = mnt_want_write_file(filp);
2236 	if (err)
2237 		return err;
2238 
2239 	shrunk_blocks = old_block_count - block_count;
2240 	secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2241 
2242 	/* stop other GC */
2243 	if (!f2fs_down_write_trylock(&sbi->gc_lock)) {
2244 		err = -EAGAIN;
2245 		goto out_drop_write;
2246 	}
2247 
2248 	/* stop CP to protect MAIN_SEC in free_segment_range */
2249 	f2fs_lock_op(sbi);
2250 
2251 	spin_lock(&sbi->stat_lock);
2252 	if (shrunk_blocks + valid_user_blocks(sbi) +
2253 		sbi->current_reserved_blocks + sbi->unusable_block_count +
2254 		F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2255 		err = -ENOSPC;
2256 	spin_unlock(&sbi->stat_lock);
2257 
2258 	if (err)
2259 		goto out_unlock;
2260 
2261 	err = free_segment_range(sbi, secs, true);
2262 
2263 out_unlock:
2264 	f2fs_unlock_op(sbi);
2265 	f2fs_up_write(&sbi->gc_lock);
2266 out_drop_write:
2267 	mnt_drop_write_file(filp);
2268 	if (err)
2269 		return err;
2270 
2271 	err = freeze_super(sbi->sb, FREEZE_HOLDER_USERSPACE);
2272 	if (err)
2273 		return err;
2274 
2275 	if (f2fs_readonly(sbi->sb)) {
2276 		err = thaw_super(sbi->sb, FREEZE_HOLDER_USERSPACE);
2277 		if (err)
2278 			return err;
2279 		return -EROFS;
2280 	}
2281 
2282 	f2fs_down_write(&sbi->gc_lock);
2283 	f2fs_down_write(&sbi->cp_global_sem);
2284 
2285 	spin_lock(&sbi->stat_lock);
2286 	if (shrunk_blocks + valid_user_blocks(sbi) +
2287 		sbi->current_reserved_blocks + sbi->unusable_block_count +
2288 		F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2289 		err = -ENOSPC;
2290 	else
2291 		sbi->user_block_count -= shrunk_blocks;
2292 	spin_unlock(&sbi->stat_lock);
2293 	if (err)
2294 		goto out_err;
2295 
2296 	set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2297 	err = free_segment_range(sbi, secs, false);
2298 	if (err)
2299 		goto recover_out;
2300 
2301 	update_sb_metadata(sbi, -secs);
2302 
2303 	err = f2fs_commit_super(sbi, false);
2304 	if (err) {
2305 		update_sb_metadata(sbi, secs);
2306 		goto recover_out;
2307 	}
2308 
2309 	update_fs_metadata(sbi, -secs);
2310 	clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2311 	set_sbi_flag(sbi, SBI_IS_DIRTY);
2312 
2313 	stat_inc_cp_call_count(sbi, TOTAL_CALL);
2314 	err = f2fs_write_checkpoint(sbi, &cpc);
2315 	if (err) {
2316 		update_fs_metadata(sbi, secs);
2317 		update_sb_metadata(sbi, secs);
2318 		f2fs_commit_super(sbi, false);
2319 	}
2320 recover_out:
2321 	clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2322 	if (err) {
2323 		set_sbi_flag(sbi, SBI_NEED_FSCK);
2324 		f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2325 
2326 		spin_lock(&sbi->stat_lock);
2327 		sbi->user_block_count += shrunk_blocks;
2328 		spin_unlock(&sbi->stat_lock);
2329 	}
2330 out_err:
2331 	f2fs_up_write(&sbi->cp_global_sem);
2332 	f2fs_up_write(&sbi->gc_lock);
2333 	thaw_super(sbi->sb, FREEZE_HOLDER_USERSPACE);
2334 	return err;
2335 }
2336