xref: /linux/fs/f2fs/gc.c (revision d4821902e43453b85b31329441a9f6ac071228a8)
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 
39 	wait_ms = gc_th->min_sleep_time;
40 
41 	set_freezable();
42 	do {
43 		bool sync_mode, foreground = false;
44 
45 		wait_event_interruptible_timeout(*wq,
46 				kthread_should_stop() || freezing(current) ||
47 				waitqueue_active(fggc_wq) ||
48 				gc_th->gc_wake,
49 				msecs_to_jiffies(wait_ms));
50 
51 		if (test_opt(sbi, GC_MERGE) && waitqueue_active(fggc_wq))
52 			foreground = true;
53 
54 		/* give it a try one time */
55 		if (gc_th->gc_wake)
56 			gc_th->gc_wake = 0;
57 
58 		if (try_to_freeze()) {
59 			stat_other_skip_bggc_count(sbi);
60 			continue;
61 		}
62 		if (kthread_should_stop())
63 			break;
64 
65 		if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
66 			increase_sleep_time(gc_th, &wait_ms);
67 			stat_other_skip_bggc_count(sbi);
68 			continue;
69 		}
70 
71 		if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
72 			f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
73 			f2fs_stop_checkpoint(sbi, false);
74 		}
75 
76 		if (!sb_start_write_trylock(sbi->sb)) {
77 			stat_other_skip_bggc_count(sbi);
78 			continue;
79 		}
80 
81 		/*
82 		 * [GC triggering condition]
83 		 * 0. GC is not conducted currently.
84 		 * 1. There are enough dirty segments.
85 		 * 2. IO subsystem is idle by checking the # of writeback pages.
86 		 * 3. IO subsystem is idle by checking the # of requests in
87 		 *    bdev's request list.
88 		 *
89 		 * Note) We have to avoid triggering GCs frequently.
90 		 * Because it is possible that some segments can be
91 		 * invalidated soon after by user update or deletion.
92 		 * So, I'd like to wait some time to collect dirty segments.
93 		 */
94 		if (sbi->gc_mode == GC_URGENT_HIGH) {
95 			spin_lock(&sbi->gc_urgent_high_lock);
96 			if (sbi->gc_urgent_high_limited) {
97 				if (!sbi->gc_urgent_high_remaining) {
98 					sbi->gc_urgent_high_limited = false;
99 					spin_unlock(&sbi->gc_urgent_high_lock);
100 					sbi->gc_mode = GC_NORMAL;
101 					continue;
102 				}
103 				sbi->gc_urgent_high_remaining--;
104 			}
105 			spin_unlock(&sbi->gc_urgent_high_lock);
106 		}
107 
108 		if (sbi->gc_mode == GC_URGENT_HIGH ||
109 				sbi->gc_mode == GC_URGENT_MID) {
110 			wait_ms = gc_th->urgent_sleep_time;
111 			f2fs_down_write(&sbi->gc_lock);
112 			goto do_gc;
113 		}
114 
115 		if (foreground) {
116 			f2fs_down_write(&sbi->gc_lock);
117 			goto do_gc;
118 		} else if (!f2fs_down_write_trylock(&sbi->gc_lock)) {
119 			stat_other_skip_bggc_count(sbi);
120 			goto next;
121 		}
122 
123 		if (!is_idle(sbi, GC_TIME)) {
124 			increase_sleep_time(gc_th, &wait_ms);
125 			f2fs_up_write(&sbi->gc_lock);
126 			stat_io_skip_bggc_count(sbi);
127 			goto next;
128 		}
129 
130 		if (has_enough_invalid_blocks(sbi))
131 			decrease_sleep_time(gc_th, &wait_ms);
132 		else
133 			increase_sleep_time(gc_th, &wait_ms);
134 do_gc:
135 		if (!foreground)
136 			stat_inc_bggc_count(sbi->stat_info);
137 
138 		sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
139 
140 		/* foreground GC was been triggered via f2fs_balance_fs() */
141 		if (foreground)
142 			sync_mode = false;
143 
144 		/* if return value is not zero, no victim was selected */
145 		if (f2fs_gc(sbi, sync_mode, !foreground, false, NULL_SEGNO))
146 			wait_ms = gc_th->no_gc_sleep_time;
147 
148 		if (foreground)
149 			wake_up_all(&gc_th->fggc_wq);
150 
151 		trace_f2fs_background_gc(sbi->sb, wait_ms,
152 				prefree_segments(sbi), free_segments(sbi));
153 
154 		/* balancing f2fs's metadata periodically */
155 		f2fs_balance_fs_bg(sbi, true);
156 next:
157 		sb_end_write(sbi->sb);
158 
159 	} while (!kthread_should_stop());
160 	return 0;
161 }
162 
163 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
164 {
165 	struct f2fs_gc_kthread *gc_th;
166 	dev_t dev = sbi->sb->s_bdev->bd_dev;
167 	int err = 0;
168 
169 	gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
170 	if (!gc_th) {
171 		err = -ENOMEM;
172 		goto out;
173 	}
174 
175 	gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
176 	gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
177 	gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
178 	gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
179 
180 	gc_th->gc_wake = 0;
181 
182 	sbi->gc_thread = gc_th;
183 	init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
184 	init_waitqueue_head(&sbi->gc_thread->fggc_wq);
185 	sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
186 			"f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
187 	if (IS_ERR(gc_th->f2fs_gc_task)) {
188 		err = PTR_ERR(gc_th->f2fs_gc_task);
189 		kfree(gc_th);
190 		sbi->gc_thread = NULL;
191 	}
192 out:
193 	return err;
194 }
195 
196 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
197 {
198 	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
199 
200 	if (!gc_th)
201 		return;
202 	kthread_stop(gc_th->f2fs_gc_task);
203 	wake_up_all(&gc_th->fggc_wq);
204 	kfree(gc_th);
205 	sbi->gc_thread = NULL;
206 }
207 
208 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
209 {
210 	int gc_mode;
211 
212 	if (gc_type == BG_GC) {
213 		if (sbi->am.atgc_enabled)
214 			gc_mode = GC_AT;
215 		else
216 			gc_mode = GC_CB;
217 	} else {
218 		gc_mode = GC_GREEDY;
219 	}
220 
221 	switch (sbi->gc_mode) {
222 	case GC_IDLE_CB:
223 		gc_mode = GC_CB;
224 		break;
225 	case GC_IDLE_GREEDY:
226 	case GC_URGENT_HIGH:
227 		gc_mode = GC_GREEDY;
228 		break;
229 	case GC_IDLE_AT:
230 		gc_mode = GC_AT;
231 		break;
232 	}
233 
234 	return gc_mode;
235 }
236 
237 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
238 			int type, struct victim_sel_policy *p)
239 {
240 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
241 
242 	if (p->alloc_mode == SSR) {
243 		p->gc_mode = GC_GREEDY;
244 		p->dirty_bitmap = dirty_i->dirty_segmap[type];
245 		p->max_search = dirty_i->nr_dirty[type];
246 		p->ofs_unit = 1;
247 	} else if (p->alloc_mode == AT_SSR) {
248 		p->gc_mode = GC_GREEDY;
249 		p->dirty_bitmap = dirty_i->dirty_segmap[type];
250 		p->max_search = dirty_i->nr_dirty[type];
251 		p->ofs_unit = 1;
252 	} else {
253 		p->gc_mode = select_gc_type(sbi, gc_type);
254 		p->ofs_unit = sbi->segs_per_sec;
255 		if (__is_large_section(sbi)) {
256 			p->dirty_bitmap = dirty_i->dirty_secmap;
257 			p->max_search = count_bits(p->dirty_bitmap,
258 						0, MAIN_SECS(sbi));
259 		} else {
260 			p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
261 			p->max_search = dirty_i->nr_dirty[DIRTY];
262 		}
263 	}
264 
265 	/*
266 	 * adjust candidates range, should select all dirty segments for
267 	 * foreground GC and urgent GC cases.
268 	 */
269 	if (gc_type != FG_GC &&
270 			(sbi->gc_mode != GC_URGENT_HIGH) &&
271 			(p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
272 			p->max_search > sbi->max_victim_search)
273 		p->max_search = sbi->max_victim_search;
274 
275 	/* let's select beginning hot/small space first in no_heap mode*/
276 	if (f2fs_need_rand_seg(sbi))
277 		p->offset = prandom_u32() % (MAIN_SECS(sbi) * sbi->segs_per_sec);
278 	else if (test_opt(sbi, NOHEAP) &&
279 		(type == CURSEG_HOT_DATA || IS_NODESEG(type)))
280 		p->offset = 0;
281 	else
282 		p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
283 }
284 
285 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
286 				struct victim_sel_policy *p)
287 {
288 	/* SSR allocates in a segment unit */
289 	if (p->alloc_mode == SSR)
290 		return sbi->blocks_per_seg;
291 	else if (p->alloc_mode == AT_SSR)
292 		return UINT_MAX;
293 
294 	/* LFS */
295 	if (p->gc_mode == GC_GREEDY)
296 		return 2 * sbi->blocks_per_seg * p->ofs_unit;
297 	else if (p->gc_mode == GC_CB)
298 		return UINT_MAX;
299 	else if (p->gc_mode == GC_AT)
300 		return UINT_MAX;
301 	else /* No other gc_mode */
302 		return 0;
303 }
304 
305 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
306 {
307 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
308 	unsigned int secno;
309 
310 	/*
311 	 * If the gc_type is FG_GC, we can select victim segments
312 	 * selected by background GC before.
313 	 * Those segments guarantee they have small valid blocks.
314 	 */
315 	for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
316 		if (sec_usage_check(sbi, secno))
317 			continue;
318 		clear_bit(secno, dirty_i->victim_secmap);
319 		return GET_SEG_FROM_SEC(sbi, secno);
320 	}
321 	return NULL_SEGNO;
322 }
323 
324 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
325 {
326 	struct sit_info *sit_i = SIT_I(sbi);
327 	unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
328 	unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
329 	unsigned long long mtime = 0;
330 	unsigned int vblocks;
331 	unsigned char age = 0;
332 	unsigned char u;
333 	unsigned int i;
334 	unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
335 
336 	for (i = 0; i < usable_segs_per_sec; i++)
337 		mtime += get_seg_entry(sbi, start + i)->mtime;
338 	vblocks = get_valid_blocks(sbi, segno, true);
339 
340 	mtime = div_u64(mtime, usable_segs_per_sec);
341 	vblocks = div_u64(vblocks, usable_segs_per_sec);
342 
343 	u = (vblocks * 100) >> sbi->log_blocks_per_seg;
344 
345 	/* Handle if the system time has changed by the user */
346 	if (mtime < sit_i->min_mtime)
347 		sit_i->min_mtime = mtime;
348 	if (mtime > sit_i->max_mtime)
349 		sit_i->max_mtime = mtime;
350 	if (sit_i->max_mtime != sit_i->min_mtime)
351 		age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
352 				sit_i->max_mtime - sit_i->min_mtime);
353 
354 	return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
355 }
356 
357 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
358 			unsigned int segno, struct victim_sel_policy *p)
359 {
360 	if (p->alloc_mode == SSR)
361 		return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
362 
363 	/* alloc_mode == LFS */
364 	if (p->gc_mode == GC_GREEDY)
365 		return get_valid_blocks(sbi, segno, true);
366 	else if (p->gc_mode == GC_CB)
367 		return get_cb_cost(sbi, segno);
368 
369 	f2fs_bug_on(sbi, 1);
370 	return 0;
371 }
372 
373 static unsigned int count_bits(const unsigned long *addr,
374 				unsigned int offset, unsigned int len)
375 {
376 	unsigned int end = offset + len, sum = 0;
377 
378 	while (offset < end) {
379 		if (test_bit(offset++, addr))
380 			++sum;
381 	}
382 	return sum;
383 }
384 
385 static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi,
386 				unsigned long long mtime, unsigned int segno,
387 				struct rb_node *parent, struct rb_node **p,
388 				bool left_most)
389 {
390 	struct atgc_management *am = &sbi->am;
391 	struct victim_entry *ve;
392 
393 	ve =  f2fs_kmem_cache_alloc(victim_entry_slab,
394 				GFP_NOFS, true, NULL);
395 
396 	ve->mtime = mtime;
397 	ve->segno = segno;
398 
399 	rb_link_node(&ve->rb_node, parent, p);
400 	rb_insert_color_cached(&ve->rb_node, &am->root, left_most);
401 
402 	list_add_tail(&ve->list, &am->victim_list);
403 
404 	am->victim_count++;
405 
406 	return ve;
407 }
408 
409 static void insert_victim_entry(struct f2fs_sb_info *sbi,
410 				unsigned long long mtime, unsigned int segno)
411 {
412 	struct atgc_management *am = &sbi->am;
413 	struct rb_node **p;
414 	struct rb_node *parent = NULL;
415 	bool left_most = true;
416 
417 	p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most);
418 	attach_victim_entry(sbi, mtime, segno, parent, p, left_most);
419 }
420 
421 static void add_victim_entry(struct f2fs_sb_info *sbi,
422 				struct victim_sel_policy *p, unsigned int segno)
423 {
424 	struct sit_info *sit_i = SIT_I(sbi);
425 	unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
426 	unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
427 	unsigned long long mtime = 0;
428 	unsigned int i;
429 
430 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
431 		if (p->gc_mode == GC_AT &&
432 			get_valid_blocks(sbi, segno, true) == 0)
433 			return;
434 	}
435 
436 	for (i = 0; i < sbi->segs_per_sec; i++)
437 		mtime += get_seg_entry(sbi, start + i)->mtime;
438 	mtime = div_u64(mtime, sbi->segs_per_sec);
439 
440 	/* Handle if the system time has changed by the user */
441 	if (mtime < sit_i->min_mtime)
442 		sit_i->min_mtime = mtime;
443 	if (mtime > sit_i->max_mtime)
444 		sit_i->max_mtime = mtime;
445 	if (mtime < sit_i->dirty_min_mtime)
446 		sit_i->dirty_min_mtime = mtime;
447 	if (mtime > sit_i->dirty_max_mtime)
448 		sit_i->dirty_max_mtime = mtime;
449 
450 	/* don't choose young section as candidate */
451 	if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
452 		return;
453 
454 	insert_victim_entry(sbi, mtime, segno);
455 }
456 
457 static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi,
458 						struct victim_sel_policy *p)
459 {
460 	struct atgc_management *am = &sbi->am;
461 	struct rb_node *parent = NULL;
462 	bool left_most;
463 
464 	f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most);
465 
466 	return parent;
467 }
468 
469 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
470 						struct victim_sel_policy *p)
471 {
472 	struct sit_info *sit_i = SIT_I(sbi);
473 	struct atgc_management *am = &sbi->am;
474 	struct rb_root_cached *root = &am->root;
475 	struct rb_node *node;
476 	struct rb_entry *re;
477 	struct victim_entry *ve;
478 	unsigned long long total_time;
479 	unsigned long long age, u, accu;
480 	unsigned long long max_mtime = sit_i->dirty_max_mtime;
481 	unsigned long long min_mtime = sit_i->dirty_min_mtime;
482 	unsigned int sec_blocks = BLKS_PER_SEC(sbi);
483 	unsigned int vblocks;
484 	unsigned int dirty_threshold = max(am->max_candidate_count,
485 					am->candidate_ratio *
486 					am->victim_count / 100);
487 	unsigned int age_weight = am->age_weight;
488 	unsigned int cost;
489 	unsigned int iter = 0;
490 
491 	if (max_mtime < min_mtime)
492 		return;
493 
494 	max_mtime += 1;
495 	total_time = max_mtime - min_mtime;
496 
497 	accu = div64_u64(ULLONG_MAX, total_time);
498 	accu = min_t(unsigned long long, div_u64(accu, 100),
499 					DEFAULT_ACCURACY_CLASS);
500 
501 	node = rb_first_cached(root);
502 next:
503 	re = rb_entry_safe(node, struct rb_entry, rb_node);
504 	if (!re)
505 		return;
506 
507 	ve = (struct victim_entry *)re;
508 
509 	if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
510 		goto skip;
511 
512 	/* age = 10000 * x% * 60 */
513 	age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
514 								age_weight;
515 
516 	vblocks = get_valid_blocks(sbi, ve->segno, true);
517 	f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
518 
519 	/* u = 10000 * x% * 40 */
520 	u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
521 							(100 - age_weight);
522 
523 	f2fs_bug_on(sbi, age + u >= UINT_MAX);
524 
525 	cost = UINT_MAX - (age + u);
526 	iter++;
527 
528 	if (cost < p->min_cost ||
529 			(cost == p->min_cost && age > p->oldest_age)) {
530 		p->min_cost = cost;
531 		p->oldest_age = age;
532 		p->min_segno = ve->segno;
533 	}
534 skip:
535 	if (iter < dirty_threshold) {
536 		node = rb_next(node);
537 		goto next;
538 	}
539 }
540 
541 /*
542  * select candidates around source section in range of
543  * [target - dirty_threshold, target + dirty_threshold]
544  */
545 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
546 						struct victim_sel_policy *p)
547 {
548 	struct sit_info *sit_i = SIT_I(sbi);
549 	struct atgc_management *am = &sbi->am;
550 	struct rb_node *node;
551 	struct rb_entry *re;
552 	struct victim_entry *ve;
553 	unsigned long long age;
554 	unsigned long long max_mtime = sit_i->dirty_max_mtime;
555 	unsigned long long min_mtime = sit_i->dirty_min_mtime;
556 	unsigned int seg_blocks = sbi->blocks_per_seg;
557 	unsigned int vblocks;
558 	unsigned int dirty_threshold = max(am->max_candidate_count,
559 					am->candidate_ratio *
560 					am->victim_count / 100);
561 	unsigned int cost;
562 	unsigned int iter = 0;
563 	int stage = 0;
564 
565 	if (max_mtime < min_mtime)
566 		return;
567 	max_mtime += 1;
568 next_stage:
569 	node = lookup_central_victim(sbi, p);
570 next_node:
571 	re = rb_entry_safe(node, struct rb_entry, rb_node);
572 	if (!re) {
573 		if (stage == 0)
574 			goto skip_stage;
575 		return;
576 	}
577 
578 	ve = (struct victim_entry *)re;
579 
580 	if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
581 		goto skip_node;
582 
583 	age = max_mtime - ve->mtime;
584 
585 	vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
586 	f2fs_bug_on(sbi, !vblocks);
587 
588 	/* rare case */
589 	if (vblocks == seg_blocks)
590 		goto skip_node;
591 
592 	iter++;
593 
594 	age = max_mtime - abs(p->age - age);
595 	cost = UINT_MAX - vblocks;
596 
597 	if (cost < p->min_cost ||
598 			(cost == p->min_cost && age > p->oldest_age)) {
599 		p->min_cost = cost;
600 		p->oldest_age = age;
601 		p->min_segno = ve->segno;
602 	}
603 skip_node:
604 	if (iter < dirty_threshold) {
605 		if (stage == 0)
606 			node = rb_prev(node);
607 		else if (stage == 1)
608 			node = rb_next(node);
609 		goto next_node;
610 	}
611 skip_stage:
612 	if (stage < 1) {
613 		stage++;
614 		iter = 0;
615 		goto next_stage;
616 	}
617 }
618 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
619 						struct victim_sel_policy *p)
620 {
621 	f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
622 						&sbi->am.root, true));
623 
624 	if (p->gc_mode == GC_AT)
625 		atgc_lookup_victim(sbi, p);
626 	else if (p->alloc_mode == AT_SSR)
627 		atssr_lookup_victim(sbi, p);
628 	else
629 		f2fs_bug_on(sbi, 1);
630 }
631 
632 static void release_victim_entry(struct f2fs_sb_info *sbi)
633 {
634 	struct atgc_management *am = &sbi->am;
635 	struct victim_entry *ve, *tmp;
636 
637 	list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
638 		list_del(&ve->list);
639 		kmem_cache_free(victim_entry_slab, ve);
640 		am->victim_count--;
641 	}
642 
643 	am->root = RB_ROOT_CACHED;
644 
645 	f2fs_bug_on(sbi, am->victim_count);
646 	f2fs_bug_on(sbi, !list_empty(&am->victim_list));
647 }
648 
649 /*
650  * This function is called from two paths.
651  * One is garbage collection and the other is SSR segment selection.
652  * When it is called during GC, it just gets a victim segment
653  * and it does not remove it from dirty seglist.
654  * When it is called from SSR segment selection, it finds a segment
655  * which has minimum valid blocks and removes it from dirty seglist.
656  */
657 static int get_victim_by_default(struct f2fs_sb_info *sbi,
658 			unsigned int *result, int gc_type, int type,
659 			char alloc_mode, unsigned long long age)
660 {
661 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
662 	struct sit_info *sm = SIT_I(sbi);
663 	struct victim_sel_policy p;
664 	unsigned int secno, last_victim;
665 	unsigned int last_segment;
666 	unsigned int nsearched;
667 	bool is_atgc;
668 	int ret = 0;
669 
670 	mutex_lock(&dirty_i->seglist_lock);
671 	last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
672 
673 	p.alloc_mode = alloc_mode;
674 	p.age = age;
675 	p.age_threshold = sbi->am.age_threshold;
676 
677 retry:
678 	select_policy(sbi, gc_type, type, &p);
679 	p.min_segno = NULL_SEGNO;
680 	p.oldest_age = 0;
681 	p.min_cost = get_max_cost(sbi, &p);
682 
683 	is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
684 	nsearched = 0;
685 
686 	if (is_atgc)
687 		SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
688 
689 	if (*result != NULL_SEGNO) {
690 		if (!get_valid_blocks(sbi, *result, false)) {
691 			ret = -ENODATA;
692 			goto out;
693 		}
694 
695 		if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
696 			ret = -EBUSY;
697 		else
698 			p.min_segno = *result;
699 		goto out;
700 	}
701 
702 	ret = -ENODATA;
703 	if (p.max_search == 0)
704 		goto out;
705 
706 	if (__is_large_section(sbi) && p.alloc_mode == LFS) {
707 		if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
708 			p.min_segno = sbi->next_victim_seg[BG_GC];
709 			*result = p.min_segno;
710 			sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
711 			goto got_result;
712 		}
713 		if (gc_type == FG_GC &&
714 				sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
715 			p.min_segno = sbi->next_victim_seg[FG_GC];
716 			*result = p.min_segno;
717 			sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
718 			goto got_result;
719 		}
720 	}
721 
722 	last_victim = sm->last_victim[p.gc_mode];
723 	if (p.alloc_mode == LFS && gc_type == FG_GC) {
724 		p.min_segno = check_bg_victims(sbi);
725 		if (p.min_segno != NULL_SEGNO)
726 			goto got_it;
727 	}
728 
729 	while (1) {
730 		unsigned long cost, *dirty_bitmap;
731 		unsigned int unit_no, segno;
732 
733 		dirty_bitmap = p.dirty_bitmap;
734 		unit_no = find_next_bit(dirty_bitmap,
735 				last_segment / p.ofs_unit,
736 				p.offset / p.ofs_unit);
737 		segno = unit_no * p.ofs_unit;
738 		if (segno >= last_segment) {
739 			if (sm->last_victim[p.gc_mode]) {
740 				last_segment =
741 					sm->last_victim[p.gc_mode];
742 				sm->last_victim[p.gc_mode] = 0;
743 				p.offset = 0;
744 				continue;
745 			}
746 			break;
747 		}
748 
749 		p.offset = segno + p.ofs_unit;
750 		nsearched++;
751 
752 #ifdef CONFIG_F2FS_CHECK_FS
753 		/*
754 		 * skip selecting the invalid segno (that is failed due to block
755 		 * validity check failure during GC) to avoid endless GC loop in
756 		 * such cases.
757 		 */
758 		if (test_bit(segno, sm->invalid_segmap))
759 			goto next;
760 #endif
761 
762 		secno = GET_SEC_FROM_SEG(sbi, segno);
763 
764 		if (sec_usage_check(sbi, secno))
765 			goto next;
766 
767 		/* Don't touch checkpointed data */
768 		if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
769 			if (p.alloc_mode == LFS) {
770 				/*
771 				 * LFS is set to find source section during GC.
772 				 * The victim should have no checkpointed data.
773 				 */
774 				if (get_ckpt_valid_blocks(sbi, segno, true))
775 					goto next;
776 			} else {
777 				/*
778 				 * SSR | AT_SSR are set to find target segment
779 				 * for writes which can be full by checkpointed
780 				 * and newly written blocks.
781 				 */
782 				if (!f2fs_segment_has_free_slot(sbi, segno))
783 					goto next;
784 			}
785 		}
786 
787 		if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
788 			goto next;
789 
790 		if (is_atgc) {
791 			add_victim_entry(sbi, &p, segno);
792 			goto next;
793 		}
794 
795 		cost = get_gc_cost(sbi, segno, &p);
796 
797 		if (p.min_cost > cost) {
798 			p.min_segno = segno;
799 			p.min_cost = cost;
800 		}
801 next:
802 		if (nsearched >= p.max_search) {
803 			if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
804 				sm->last_victim[p.gc_mode] =
805 					last_victim + p.ofs_unit;
806 			else
807 				sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
808 			sm->last_victim[p.gc_mode] %=
809 				(MAIN_SECS(sbi) * sbi->segs_per_sec);
810 			break;
811 		}
812 	}
813 
814 	/* get victim for GC_AT/AT_SSR */
815 	if (is_atgc) {
816 		lookup_victim_by_age(sbi, &p);
817 		release_victim_entry(sbi);
818 	}
819 
820 	if (is_atgc && p.min_segno == NULL_SEGNO &&
821 			sm->elapsed_time < p.age_threshold) {
822 		p.age_threshold = 0;
823 		goto retry;
824 	}
825 
826 	if (p.min_segno != NULL_SEGNO) {
827 got_it:
828 		*result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
829 got_result:
830 		if (p.alloc_mode == LFS) {
831 			secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
832 			if (gc_type == FG_GC)
833 				sbi->cur_victim_sec = secno;
834 			else
835 				set_bit(secno, dirty_i->victim_secmap);
836 		}
837 		ret = 0;
838 
839 	}
840 out:
841 	if (p.min_segno != NULL_SEGNO)
842 		trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
843 				sbi->cur_victim_sec,
844 				prefree_segments(sbi), free_segments(sbi));
845 	mutex_unlock(&dirty_i->seglist_lock);
846 
847 	return ret;
848 }
849 
850 static const struct victim_selection default_v_ops = {
851 	.get_victim = get_victim_by_default,
852 };
853 
854 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
855 {
856 	struct inode_entry *ie;
857 
858 	ie = radix_tree_lookup(&gc_list->iroot, ino);
859 	if (ie)
860 		return ie->inode;
861 	return NULL;
862 }
863 
864 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
865 {
866 	struct inode_entry *new_ie;
867 
868 	if (inode == find_gc_inode(gc_list, inode->i_ino)) {
869 		iput(inode);
870 		return;
871 	}
872 	new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab,
873 					GFP_NOFS, true, NULL);
874 	new_ie->inode = inode;
875 
876 	f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
877 	list_add_tail(&new_ie->list, &gc_list->ilist);
878 }
879 
880 static void put_gc_inode(struct gc_inode_list *gc_list)
881 {
882 	struct inode_entry *ie, *next_ie;
883 
884 	list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
885 		radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
886 		iput(ie->inode);
887 		list_del(&ie->list);
888 		kmem_cache_free(f2fs_inode_entry_slab, ie);
889 	}
890 }
891 
892 static int check_valid_map(struct f2fs_sb_info *sbi,
893 				unsigned int segno, int offset)
894 {
895 	struct sit_info *sit_i = SIT_I(sbi);
896 	struct seg_entry *sentry;
897 	int ret;
898 
899 	down_read(&sit_i->sentry_lock);
900 	sentry = get_seg_entry(sbi, segno);
901 	ret = f2fs_test_bit(offset, sentry->cur_valid_map);
902 	up_read(&sit_i->sentry_lock);
903 	return ret;
904 }
905 
906 /*
907  * This function compares node address got in summary with that in NAT.
908  * On validity, copy that node with cold status, otherwise (invalid node)
909  * ignore that.
910  */
911 static int gc_node_segment(struct f2fs_sb_info *sbi,
912 		struct f2fs_summary *sum, unsigned int segno, int gc_type)
913 {
914 	struct f2fs_summary *entry;
915 	block_t start_addr;
916 	int off;
917 	int phase = 0;
918 	bool fggc = (gc_type == FG_GC);
919 	int submitted = 0;
920 	unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
921 
922 	start_addr = START_BLOCK(sbi, segno);
923 
924 next_step:
925 	entry = sum;
926 
927 	if (fggc && phase == 2)
928 		atomic_inc(&sbi->wb_sync_req[NODE]);
929 
930 	for (off = 0; off < usable_blks_in_seg; off++, entry++) {
931 		nid_t nid = le32_to_cpu(entry->nid);
932 		struct page *node_page;
933 		struct node_info ni;
934 		int err;
935 
936 		/* stop BG_GC if there is not enough free sections. */
937 		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
938 			return submitted;
939 
940 		if (check_valid_map(sbi, segno, off) == 0)
941 			continue;
942 
943 		if (phase == 0) {
944 			f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
945 							META_NAT, true);
946 			continue;
947 		}
948 
949 		if (phase == 1) {
950 			f2fs_ra_node_page(sbi, nid);
951 			continue;
952 		}
953 
954 		/* phase == 2 */
955 		node_page = f2fs_get_node_page(sbi, nid);
956 		if (IS_ERR(node_page))
957 			continue;
958 
959 		/* block may become invalid during f2fs_get_node_page */
960 		if (check_valid_map(sbi, segno, off) == 0) {
961 			f2fs_put_page(node_page, 1);
962 			continue;
963 		}
964 
965 		if (f2fs_get_node_info(sbi, nid, &ni, false)) {
966 			f2fs_put_page(node_page, 1);
967 			continue;
968 		}
969 
970 		if (ni.blk_addr != start_addr + off) {
971 			f2fs_put_page(node_page, 1);
972 			continue;
973 		}
974 
975 		err = f2fs_move_node_page(node_page, gc_type);
976 		if (!err && gc_type == FG_GC)
977 			submitted++;
978 		stat_inc_node_blk_count(sbi, 1, gc_type);
979 	}
980 
981 	if (++phase < 3)
982 		goto next_step;
983 
984 	if (fggc)
985 		atomic_dec(&sbi->wb_sync_req[NODE]);
986 	return submitted;
987 }
988 
989 /*
990  * Calculate start block index indicating the given node offset.
991  * Be careful, caller should give this node offset only indicating direct node
992  * blocks. If any node offsets, which point the other types of node blocks such
993  * as indirect or double indirect node blocks, are given, it must be a caller's
994  * bug.
995  */
996 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
997 {
998 	unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
999 	unsigned int bidx;
1000 
1001 	if (node_ofs == 0)
1002 		return 0;
1003 
1004 	if (node_ofs <= 2) {
1005 		bidx = node_ofs - 1;
1006 	} else if (node_ofs <= indirect_blks) {
1007 		int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
1008 
1009 		bidx = node_ofs - 2 - dec;
1010 	} else {
1011 		int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
1012 
1013 		bidx = node_ofs - 5 - dec;
1014 	}
1015 	return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
1016 }
1017 
1018 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1019 		struct node_info *dni, block_t blkaddr, unsigned int *nofs)
1020 {
1021 	struct page *node_page;
1022 	nid_t nid;
1023 	unsigned int ofs_in_node;
1024 	block_t source_blkaddr;
1025 
1026 	nid = le32_to_cpu(sum->nid);
1027 	ofs_in_node = le16_to_cpu(sum->ofs_in_node);
1028 
1029 	node_page = f2fs_get_node_page(sbi, nid);
1030 	if (IS_ERR(node_page))
1031 		return false;
1032 
1033 	if (f2fs_get_node_info(sbi, nid, dni, false)) {
1034 		f2fs_put_page(node_page, 1);
1035 		return false;
1036 	}
1037 
1038 	if (sum->version != dni->version) {
1039 		f2fs_warn(sbi, "%s: valid data with mismatched node version.",
1040 			  __func__);
1041 		set_sbi_flag(sbi, SBI_NEED_FSCK);
1042 	}
1043 
1044 	if (f2fs_check_nid_range(sbi, dni->ino)) {
1045 		f2fs_put_page(node_page, 1);
1046 		return false;
1047 	}
1048 
1049 	*nofs = ofs_of_node(node_page);
1050 	source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1051 	f2fs_put_page(node_page, 1);
1052 
1053 	if (source_blkaddr != blkaddr) {
1054 #ifdef CONFIG_F2FS_CHECK_FS
1055 		unsigned int segno = GET_SEGNO(sbi, blkaddr);
1056 		unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1057 
1058 		if (unlikely(check_valid_map(sbi, segno, offset))) {
1059 			if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1060 				f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",
1061 					 blkaddr, source_blkaddr, segno);
1062 				set_sbi_flag(sbi, SBI_NEED_FSCK);
1063 			}
1064 		}
1065 #endif
1066 		return false;
1067 	}
1068 	return true;
1069 }
1070 
1071 static int ra_data_block(struct inode *inode, pgoff_t index)
1072 {
1073 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1074 	struct address_space *mapping = inode->i_mapping;
1075 	struct dnode_of_data dn;
1076 	struct page *page;
1077 	struct extent_info ei = {0, 0, 0};
1078 	struct f2fs_io_info fio = {
1079 		.sbi = sbi,
1080 		.ino = inode->i_ino,
1081 		.type = DATA,
1082 		.temp = COLD,
1083 		.op = REQ_OP_READ,
1084 		.op_flags = 0,
1085 		.encrypted_page = NULL,
1086 		.in_list = false,
1087 		.retry = false,
1088 	};
1089 	int err;
1090 
1091 	page = f2fs_grab_cache_page(mapping, index, true);
1092 	if (!page)
1093 		return -ENOMEM;
1094 
1095 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1096 		dn.data_blkaddr = ei.blk + index - ei.fofs;
1097 		if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1098 						DATA_GENERIC_ENHANCE_READ))) {
1099 			err = -EFSCORRUPTED;
1100 			goto put_page;
1101 		}
1102 		goto got_it;
1103 	}
1104 
1105 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1106 	err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1107 	if (err)
1108 		goto put_page;
1109 	f2fs_put_dnode(&dn);
1110 
1111 	if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1112 		err = -ENOENT;
1113 		goto put_page;
1114 	}
1115 	if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1116 						DATA_GENERIC_ENHANCE))) {
1117 		err = -EFSCORRUPTED;
1118 		goto put_page;
1119 	}
1120 got_it:
1121 	/* read page */
1122 	fio.page = page;
1123 	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1124 
1125 	/*
1126 	 * don't cache encrypted data into meta inode until previous dirty
1127 	 * data were writebacked to avoid racing between GC and flush.
1128 	 */
1129 	f2fs_wait_on_page_writeback(page, DATA, true, true);
1130 
1131 	f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1132 
1133 	fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1134 					dn.data_blkaddr,
1135 					FGP_LOCK | FGP_CREAT, GFP_NOFS);
1136 	if (!fio.encrypted_page) {
1137 		err = -ENOMEM;
1138 		goto put_page;
1139 	}
1140 
1141 	err = f2fs_submit_page_bio(&fio);
1142 	if (err)
1143 		goto put_encrypted_page;
1144 	f2fs_put_page(fio.encrypted_page, 0);
1145 	f2fs_put_page(page, 1);
1146 
1147 	f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1148 	f2fs_update_iostat(sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1149 
1150 	return 0;
1151 put_encrypted_page:
1152 	f2fs_put_page(fio.encrypted_page, 1);
1153 put_page:
1154 	f2fs_put_page(page, 1);
1155 	return err;
1156 }
1157 
1158 /*
1159  * Move data block via META_MAPPING while keeping locked data page.
1160  * This can be used to move blocks, aka LBAs, directly on disk.
1161  */
1162 static int move_data_block(struct inode *inode, block_t bidx,
1163 				int gc_type, unsigned int segno, int off)
1164 {
1165 	struct f2fs_io_info fio = {
1166 		.sbi = F2FS_I_SB(inode),
1167 		.ino = inode->i_ino,
1168 		.type = DATA,
1169 		.temp = COLD,
1170 		.op = REQ_OP_READ,
1171 		.op_flags = 0,
1172 		.encrypted_page = NULL,
1173 		.in_list = false,
1174 		.retry = false,
1175 	};
1176 	struct dnode_of_data dn;
1177 	struct f2fs_summary sum;
1178 	struct node_info ni;
1179 	struct page *page, *mpage;
1180 	block_t newaddr;
1181 	int err = 0;
1182 	bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1183 	int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&
1184 				(fio.sbi->gc_mode != GC_URGENT_HIGH) ?
1185 				CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1186 
1187 	/* do not read out */
1188 	page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1189 	if (!page)
1190 		return -ENOMEM;
1191 
1192 	if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1193 		err = -ENOENT;
1194 		goto out;
1195 	}
1196 
1197 	if (f2fs_is_atomic_file(inode)) {
1198 		F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1199 		F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1200 		err = -EAGAIN;
1201 		goto out;
1202 	}
1203 
1204 	if (f2fs_is_pinned_file(inode)) {
1205 		f2fs_pin_file_control(inode, true);
1206 		err = -EAGAIN;
1207 		goto out;
1208 	}
1209 
1210 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1211 	err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1212 	if (err)
1213 		goto out;
1214 
1215 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1216 		ClearPageUptodate(page);
1217 		err = -ENOENT;
1218 		goto put_out;
1219 	}
1220 
1221 	/*
1222 	 * don't cache encrypted data into meta inode until previous dirty
1223 	 * data were writebacked to avoid racing between GC and flush.
1224 	 */
1225 	f2fs_wait_on_page_writeback(page, DATA, true, true);
1226 
1227 	f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1228 
1229 	err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);
1230 	if (err)
1231 		goto put_out;
1232 
1233 	/* read page */
1234 	fio.page = page;
1235 	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1236 
1237 	if (lfs_mode)
1238 		f2fs_down_write(&fio.sbi->io_order_lock);
1239 
1240 	mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1241 					fio.old_blkaddr, false);
1242 	if (!mpage) {
1243 		err = -ENOMEM;
1244 		goto up_out;
1245 	}
1246 
1247 	fio.encrypted_page = mpage;
1248 
1249 	/* read source block in mpage */
1250 	if (!PageUptodate(mpage)) {
1251 		err = f2fs_submit_page_bio(&fio);
1252 		if (err) {
1253 			f2fs_put_page(mpage, 1);
1254 			goto up_out;
1255 		}
1256 
1257 		f2fs_update_iostat(fio.sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1258 		f2fs_update_iostat(fio.sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1259 
1260 		lock_page(mpage);
1261 		if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1262 						!PageUptodate(mpage))) {
1263 			err = -EIO;
1264 			f2fs_put_page(mpage, 1);
1265 			goto up_out;
1266 		}
1267 	}
1268 
1269 	set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1270 
1271 	/* allocate block address */
1272 	f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1273 				&sum, type, NULL);
1274 
1275 	fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1276 				newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1277 	if (!fio.encrypted_page) {
1278 		err = -ENOMEM;
1279 		f2fs_put_page(mpage, 1);
1280 		goto recover_block;
1281 	}
1282 
1283 	/* write target block */
1284 	f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1285 	memcpy(page_address(fio.encrypted_page),
1286 				page_address(mpage), PAGE_SIZE);
1287 	f2fs_put_page(mpage, 1);
1288 	invalidate_mapping_pages(META_MAPPING(fio.sbi),
1289 				fio.old_blkaddr, fio.old_blkaddr);
1290 	f2fs_invalidate_compress_page(fio.sbi, fio.old_blkaddr);
1291 
1292 	set_page_dirty(fio.encrypted_page);
1293 	if (clear_page_dirty_for_io(fio.encrypted_page))
1294 		dec_page_count(fio.sbi, F2FS_DIRTY_META);
1295 
1296 	set_page_writeback(fio.encrypted_page);
1297 	ClearPageError(page);
1298 
1299 	fio.op = REQ_OP_WRITE;
1300 	fio.op_flags = REQ_SYNC;
1301 	fio.new_blkaddr = newaddr;
1302 	f2fs_submit_page_write(&fio);
1303 	if (fio.retry) {
1304 		err = -EAGAIN;
1305 		if (PageWriteback(fio.encrypted_page))
1306 			end_page_writeback(fio.encrypted_page);
1307 		goto put_page_out;
1308 	}
1309 
1310 	f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
1311 
1312 	f2fs_update_data_blkaddr(&dn, newaddr);
1313 	set_inode_flag(inode, FI_APPEND_WRITE);
1314 	if (page->index == 0)
1315 		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1316 put_page_out:
1317 	f2fs_put_page(fio.encrypted_page, 1);
1318 recover_block:
1319 	if (err)
1320 		f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1321 							true, true, true);
1322 up_out:
1323 	if (lfs_mode)
1324 		f2fs_up_write(&fio.sbi->io_order_lock);
1325 put_out:
1326 	f2fs_put_dnode(&dn);
1327 out:
1328 	f2fs_put_page(page, 1);
1329 	return err;
1330 }
1331 
1332 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1333 							unsigned int segno, int off)
1334 {
1335 	struct page *page;
1336 	int err = 0;
1337 
1338 	page = f2fs_get_lock_data_page(inode, bidx, true);
1339 	if (IS_ERR(page))
1340 		return PTR_ERR(page);
1341 
1342 	if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1343 		err = -ENOENT;
1344 		goto out;
1345 	}
1346 
1347 	if (f2fs_is_atomic_file(inode)) {
1348 		F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1349 		F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1350 		err = -EAGAIN;
1351 		goto out;
1352 	}
1353 	if (f2fs_is_pinned_file(inode)) {
1354 		if (gc_type == FG_GC)
1355 			f2fs_pin_file_control(inode, true);
1356 		err = -EAGAIN;
1357 		goto out;
1358 	}
1359 
1360 	if (gc_type == BG_GC) {
1361 		if (PageWriteback(page)) {
1362 			err = -EAGAIN;
1363 			goto out;
1364 		}
1365 		set_page_dirty(page);
1366 		set_page_private_gcing(page);
1367 	} else {
1368 		struct f2fs_io_info fio = {
1369 			.sbi = F2FS_I_SB(inode),
1370 			.ino = inode->i_ino,
1371 			.type = DATA,
1372 			.temp = COLD,
1373 			.op = REQ_OP_WRITE,
1374 			.op_flags = REQ_SYNC,
1375 			.old_blkaddr = NULL_ADDR,
1376 			.page = page,
1377 			.encrypted_page = NULL,
1378 			.need_lock = LOCK_REQ,
1379 			.io_type = FS_GC_DATA_IO,
1380 		};
1381 		bool is_dirty = PageDirty(page);
1382 
1383 retry:
1384 		f2fs_wait_on_page_writeback(page, DATA, true, true);
1385 
1386 		set_page_dirty(page);
1387 		if (clear_page_dirty_for_io(page)) {
1388 			inode_dec_dirty_pages(inode);
1389 			f2fs_remove_dirty_inode(inode);
1390 		}
1391 
1392 		set_page_private_gcing(page);
1393 
1394 		err = f2fs_do_write_data_page(&fio);
1395 		if (err) {
1396 			clear_page_private_gcing(page);
1397 			if (err == -ENOMEM) {
1398 				memalloc_retry_wait(GFP_NOFS);
1399 				goto retry;
1400 			}
1401 			if (is_dirty)
1402 				set_page_dirty(page);
1403 		}
1404 	}
1405 out:
1406 	f2fs_put_page(page, 1);
1407 	return err;
1408 }
1409 
1410 /*
1411  * This function tries to get parent node of victim data block, and identifies
1412  * data block validity. If the block is valid, copy that with cold status and
1413  * modify parent node.
1414  * If the parent node is not valid or the data block address is different,
1415  * the victim data block is ignored.
1416  */
1417 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1418 		struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1419 		bool force_migrate)
1420 {
1421 	struct super_block *sb = sbi->sb;
1422 	struct f2fs_summary *entry;
1423 	block_t start_addr;
1424 	int off;
1425 	int phase = 0;
1426 	int submitted = 0;
1427 	unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1428 
1429 	start_addr = START_BLOCK(sbi, segno);
1430 
1431 next_step:
1432 	entry = sum;
1433 
1434 	for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1435 		struct page *data_page;
1436 		struct inode *inode;
1437 		struct node_info dni; /* dnode info for the data */
1438 		unsigned int ofs_in_node, nofs;
1439 		block_t start_bidx;
1440 		nid_t nid = le32_to_cpu(entry->nid);
1441 
1442 		/*
1443 		 * stop BG_GC if there is not enough free sections.
1444 		 * Or, stop GC if the segment becomes fully valid caused by
1445 		 * race condition along with SSR block allocation.
1446 		 */
1447 		if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1448 			(!force_migrate && get_valid_blocks(sbi, segno, true) ==
1449 							BLKS_PER_SEC(sbi)))
1450 			return submitted;
1451 
1452 		if (check_valid_map(sbi, segno, off) == 0)
1453 			continue;
1454 
1455 		if (phase == 0) {
1456 			f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1457 							META_NAT, true);
1458 			continue;
1459 		}
1460 
1461 		if (phase == 1) {
1462 			f2fs_ra_node_page(sbi, nid);
1463 			continue;
1464 		}
1465 
1466 		/* Get an inode by ino with checking validity */
1467 		if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1468 			continue;
1469 
1470 		if (phase == 2) {
1471 			f2fs_ra_node_page(sbi, dni.ino);
1472 			continue;
1473 		}
1474 
1475 		ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1476 
1477 		if (phase == 3) {
1478 			inode = f2fs_iget(sb, dni.ino);
1479 			if (IS_ERR(inode) || is_bad_inode(inode) ||
1480 					special_file(inode->i_mode))
1481 				continue;
1482 
1483 			if (!f2fs_down_write_trylock(
1484 				&F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1485 				iput(inode);
1486 				sbi->skipped_gc_rwsem++;
1487 				continue;
1488 			}
1489 
1490 			start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1491 								ofs_in_node;
1492 
1493 			if (f2fs_post_read_required(inode)) {
1494 				int err = ra_data_block(inode, start_bidx);
1495 
1496 				f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1497 				if (err) {
1498 					iput(inode);
1499 					continue;
1500 				}
1501 				add_gc_inode(gc_list, inode);
1502 				continue;
1503 			}
1504 
1505 			data_page = f2fs_get_read_data_page(inode,
1506 						start_bidx, REQ_RAHEAD, true);
1507 			f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1508 			if (IS_ERR(data_page)) {
1509 				iput(inode);
1510 				continue;
1511 			}
1512 
1513 			f2fs_put_page(data_page, 0);
1514 			add_gc_inode(gc_list, inode);
1515 			continue;
1516 		}
1517 
1518 		/* phase 4 */
1519 		inode = find_gc_inode(gc_list, dni.ino);
1520 		if (inode) {
1521 			struct f2fs_inode_info *fi = F2FS_I(inode);
1522 			bool locked = false;
1523 			int err;
1524 
1525 			if (S_ISREG(inode->i_mode)) {
1526 				if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[READ])) {
1527 					sbi->skipped_gc_rwsem++;
1528 					continue;
1529 				}
1530 				if (!f2fs_down_write_trylock(
1531 						&fi->i_gc_rwsem[WRITE])) {
1532 					sbi->skipped_gc_rwsem++;
1533 					f2fs_up_write(&fi->i_gc_rwsem[READ]);
1534 					continue;
1535 				}
1536 				locked = true;
1537 
1538 				/* wait for all inflight aio data */
1539 				inode_dio_wait(inode);
1540 			}
1541 
1542 			start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1543 								+ ofs_in_node;
1544 			if (f2fs_post_read_required(inode))
1545 				err = move_data_block(inode, start_bidx,
1546 							gc_type, segno, off);
1547 			else
1548 				err = move_data_page(inode, start_bidx, gc_type,
1549 								segno, off);
1550 
1551 			if (!err && (gc_type == FG_GC ||
1552 					f2fs_post_read_required(inode)))
1553 				submitted++;
1554 
1555 			if (locked) {
1556 				f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
1557 				f2fs_up_write(&fi->i_gc_rwsem[READ]);
1558 			}
1559 
1560 			stat_inc_data_blk_count(sbi, 1, gc_type);
1561 		}
1562 	}
1563 
1564 	if (++phase < 5)
1565 		goto next_step;
1566 
1567 	return submitted;
1568 }
1569 
1570 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1571 			int gc_type)
1572 {
1573 	struct sit_info *sit_i = SIT_I(sbi);
1574 	int ret;
1575 
1576 	down_write(&sit_i->sentry_lock);
1577 	ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1578 					      NO_CHECK_TYPE, LFS, 0);
1579 	up_write(&sit_i->sentry_lock);
1580 	return ret;
1581 }
1582 
1583 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1584 				unsigned int start_segno,
1585 				struct gc_inode_list *gc_list, int gc_type,
1586 				bool force_migrate)
1587 {
1588 	struct page *sum_page;
1589 	struct f2fs_summary_block *sum;
1590 	struct blk_plug plug;
1591 	unsigned int segno = start_segno;
1592 	unsigned int end_segno = start_segno + sbi->segs_per_sec;
1593 	int seg_freed = 0, migrated = 0;
1594 	unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1595 						SUM_TYPE_DATA : SUM_TYPE_NODE;
1596 	int submitted = 0;
1597 
1598 	if (__is_large_section(sbi))
1599 		end_segno = rounddown(end_segno, sbi->segs_per_sec);
1600 
1601 	/*
1602 	 * zone-capacity can be less than zone-size in zoned devices,
1603 	 * resulting in less than expected usable segments in the zone,
1604 	 * calculate the end segno in the zone which can be garbage collected
1605 	 */
1606 	if (f2fs_sb_has_blkzoned(sbi))
1607 		end_segno -= sbi->segs_per_sec -
1608 					f2fs_usable_segs_in_sec(sbi, segno);
1609 
1610 	sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1611 
1612 	/* readahead multi ssa blocks those have contiguous address */
1613 	if (__is_large_section(sbi))
1614 		f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1615 					end_segno - segno, META_SSA, true);
1616 
1617 	/* reference all summary page */
1618 	while (segno < end_segno) {
1619 		sum_page = f2fs_get_sum_page(sbi, segno++);
1620 		if (IS_ERR(sum_page)) {
1621 			int err = PTR_ERR(sum_page);
1622 
1623 			end_segno = segno - 1;
1624 			for (segno = start_segno; segno < end_segno; segno++) {
1625 				sum_page = find_get_page(META_MAPPING(sbi),
1626 						GET_SUM_BLOCK(sbi, segno));
1627 				f2fs_put_page(sum_page, 0);
1628 				f2fs_put_page(sum_page, 0);
1629 			}
1630 			return err;
1631 		}
1632 		unlock_page(sum_page);
1633 	}
1634 
1635 	blk_start_plug(&plug);
1636 
1637 	for (segno = start_segno; segno < end_segno; segno++) {
1638 
1639 		/* find segment summary of victim */
1640 		sum_page = find_get_page(META_MAPPING(sbi),
1641 					GET_SUM_BLOCK(sbi, segno));
1642 		f2fs_put_page(sum_page, 0);
1643 
1644 		if (get_valid_blocks(sbi, segno, false) == 0)
1645 			goto freed;
1646 		if (gc_type == BG_GC && __is_large_section(sbi) &&
1647 				migrated >= sbi->migration_granularity)
1648 			goto skip;
1649 		if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1650 			goto skip;
1651 
1652 		sum = page_address(sum_page);
1653 		if (type != GET_SUM_TYPE((&sum->footer))) {
1654 			f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1655 				 segno, type, GET_SUM_TYPE((&sum->footer)));
1656 			set_sbi_flag(sbi, SBI_NEED_FSCK);
1657 			f2fs_stop_checkpoint(sbi, false);
1658 			goto skip;
1659 		}
1660 
1661 		/*
1662 		 * this is to avoid deadlock:
1663 		 * - lock_page(sum_page)         - f2fs_replace_block
1664 		 *  - check_valid_map()            - down_write(sentry_lock)
1665 		 *   - down_read(sentry_lock)     - change_curseg()
1666 		 *                                  - lock_page(sum_page)
1667 		 */
1668 		if (type == SUM_TYPE_NODE)
1669 			submitted += gc_node_segment(sbi, sum->entries, segno,
1670 								gc_type);
1671 		else
1672 			submitted += gc_data_segment(sbi, sum->entries, gc_list,
1673 							segno, gc_type,
1674 							force_migrate);
1675 
1676 		stat_inc_seg_count(sbi, type, gc_type);
1677 		sbi->gc_reclaimed_segs[sbi->gc_mode]++;
1678 		migrated++;
1679 
1680 freed:
1681 		if (gc_type == FG_GC &&
1682 				get_valid_blocks(sbi, segno, false) == 0)
1683 			seg_freed++;
1684 
1685 		if (__is_large_section(sbi) && segno + 1 < end_segno)
1686 			sbi->next_victim_seg[gc_type] = segno + 1;
1687 skip:
1688 		f2fs_put_page(sum_page, 0);
1689 	}
1690 
1691 	if (submitted)
1692 		f2fs_submit_merged_write(sbi,
1693 				(type == SUM_TYPE_NODE) ? NODE : DATA);
1694 
1695 	blk_finish_plug(&plug);
1696 
1697 	stat_inc_call_count(sbi->stat_info);
1698 
1699 	return seg_freed;
1700 }
1701 
1702 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1703 			bool background, bool force, unsigned int segno)
1704 {
1705 	int gc_type = sync ? FG_GC : BG_GC;
1706 	int sec_freed = 0, seg_freed = 0, total_freed = 0;
1707 	int ret = 0;
1708 	struct cp_control cpc;
1709 	unsigned int init_segno = segno;
1710 	struct gc_inode_list gc_list = {
1711 		.ilist = LIST_HEAD_INIT(gc_list.ilist),
1712 		.iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1713 	};
1714 	unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1715 	unsigned long long first_skipped;
1716 	unsigned int skipped_round = 0, round = 0;
1717 
1718 	trace_f2fs_gc_begin(sbi->sb, sync, background,
1719 				get_pages(sbi, F2FS_DIRTY_NODES),
1720 				get_pages(sbi, F2FS_DIRTY_DENTS),
1721 				get_pages(sbi, F2FS_DIRTY_IMETA),
1722 				free_sections(sbi),
1723 				free_segments(sbi),
1724 				reserved_segments(sbi),
1725 				prefree_segments(sbi));
1726 
1727 	cpc.reason = __get_cp_reason(sbi);
1728 	sbi->skipped_gc_rwsem = 0;
1729 	first_skipped = last_skipped;
1730 gc_more:
1731 	if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1732 		ret = -EINVAL;
1733 		goto stop;
1734 	}
1735 	if (unlikely(f2fs_cp_error(sbi))) {
1736 		ret = -EIO;
1737 		goto stop;
1738 	}
1739 
1740 	if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1741 		/*
1742 		 * For example, if there are many prefree_segments below given
1743 		 * threshold, we can make them free by checkpoint. Then, we
1744 		 * secure free segments which doesn't need fggc any more.
1745 		 */
1746 		if (prefree_segments(sbi) &&
1747 				!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1748 			ret = f2fs_write_checkpoint(sbi, &cpc);
1749 			if (ret)
1750 				goto stop;
1751 		}
1752 		if (has_not_enough_free_secs(sbi, 0, 0))
1753 			gc_type = FG_GC;
1754 	}
1755 
1756 	/* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1757 	if (gc_type == BG_GC && !background) {
1758 		ret = -EINVAL;
1759 		goto stop;
1760 	}
1761 	ret = __get_victim(sbi, &segno, gc_type);
1762 	if (ret)
1763 		goto stop;
1764 
1765 	seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type, force);
1766 	if (gc_type == FG_GC &&
1767 		seg_freed == f2fs_usable_segs_in_sec(sbi, segno))
1768 		sec_freed++;
1769 	total_freed += seg_freed;
1770 
1771 	if (gc_type == FG_GC) {
1772 		if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
1773 						sbi->skipped_gc_rwsem)
1774 			skipped_round++;
1775 		last_skipped = sbi->skipped_atomic_files[FG_GC];
1776 		round++;
1777 	}
1778 
1779 	if (gc_type == FG_GC)
1780 		sbi->cur_victim_sec = NULL_SEGNO;
1781 
1782 	if (sync)
1783 		goto stop;
1784 
1785 	if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1786 		if (skipped_round <= MAX_SKIP_GC_COUNT ||
1787 					skipped_round * 2 < round) {
1788 			segno = NULL_SEGNO;
1789 			goto gc_more;
1790 		}
1791 
1792 		if (first_skipped < last_skipped &&
1793 				(last_skipped - first_skipped) >
1794 						sbi->skipped_gc_rwsem) {
1795 			f2fs_drop_inmem_pages_all(sbi, true);
1796 			segno = NULL_SEGNO;
1797 			goto gc_more;
1798 		}
1799 		if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1800 			ret = f2fs_write_checkpoint(sbi, &cpc);
1801 	}
1802 stop:
1803 	SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1804 	SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1805 
1806 	trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1807 				get_pages(sbi, F2FS_DIRTY_NODES),
1808 				get_pages(sbi, F2FS_DIRTY_DENTS),
1809 				get_pages(sbi, F2FS_DIRTY_IMETA),
1810 				free_sections(sbi),
1811 				free_segments(sbi),
1812 				reserved_segments(sbi),
1813 				prefree_segments(sbi));
1814 
1815 	f2fs_up_write(&sbi->gc_lock);
1816 
1817 	put_gc_inode(&gc_list);
1818 
1819 	if (sync && !ret)
1820 		ret = sec_freed ? 0 : -EAGAIN;
1821 	return ret;
1822 }
1823 
1824 int __init f2fs_create_garbage_collection_cache(void)
1825 {
1826 	victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1827 					sizeof(struct victim_entry));
1828 	if (!victim_entry_slab)
1829 		return -ENOMEM;
1830 	return 0;
1831 }
1832 
1833 void f2fs_destroy_garbage_collection_cache(void)
1834 {
1835 	kmem_cache_destroy(victim_entry_slab);
1836 }
1837 
1838 static void init_atgc_management(struct f2fs_sb_info *sbi)
1839 {
1840 	struct atgc_management *am = &sbi->am;
1841 
1842 	if (test_opt(sbi, ATGC) &&
1843 		SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1844 		am->atgc_enabled = true;
1845 
1846 	am->root = RB_ROOT_CACHED;
1847 	INIT_LIST_HEAD(&am->victim_list);
1848 	am->victim_count = 0;
1849 
1850 	am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1851 	am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1852 	am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1853 	am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
1854 }
1855 
1856 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1857 {
1858 	DIRTY_I(sbi)->v_ops = &default_v_ops;
1859 
1860 	sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1861 
1862 	/* give warm/cold data area from slower device */
1863 	if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1864 		SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1865 				GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1866 
1867 	init_atgc_management(sbi);
1868 }
1869 
1870 static int free_segment_range(struct f2fs_sb_info *sbi,
1871 				unsigned int secs, bool gc_only)
1872 {
1873 	unsigned int segno, next_inuse, start, end;
1874 	struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1875 	int gc_mode, gc_type;
1876 	int err = 0;
1877 	int type;
1878 
1879 	/* Force block allocation for GC */
1880 	MAIN_SECS(sbi) -= secs;
1881 	start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1882 	end = MAIN_SEGS(sbi) - 1;
1883 
1884 	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1885 	for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1886 		if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1887 			SIT_I(sbi)->last_victim[gc_mode] = 0;
1888 
1889 	for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1890 		if (sbi->next_victim_seg[gc_type] >= start)
1891 			sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1892 	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1893 
1894 	/* Move out cursegs from the target range */
1895 	for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1896 		f2fs_allocate_segment_for_resize(sbi, type, start, end);
1897 
1898 	/* do GC to move out valid blocks in the range */
1899 	for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1900 		struct gc_inode_list gc_list = {
1901 			.ilist = LIST_HEAD_INIT(gc_list.ilist),
1902 			.iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1903 		};
1904 
1905 		do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
1906 		put_gc_inode(&gc_list);
1907 
1908 		if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1909 			err = -EAGAIN;
1910 			goto out;
1911 		}
1912 		if (fatal_signal_pending(current)) {
1913 			err = -ERESTARTSYS;
1914 			goto out;
1915 		}
1916 	}
1917 	if (gc_only)
1918 		goto out;
1919 
1920 	err = f2fs_write_checkpoint(sbi, &cpc);
1921 	if (err)
1922 		goto out;
1923 
1924 	next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1925 	if (next_inuse <= end) {
1926 		f2fs_err(sbi, "segno %u should be free but still inuse!",
1927 			 next_inuse);
1928 		f2fs_bug_on(sbi, 1);
1929 	}
1930 out:
1931 	MAIN_SECS(sbi) += secs;
1932 	return err;
1933 }
1934 
1935 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1936 {
1937 	struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1938 	int section_count;
1939 	int segment_count;
1940 	int segment_count_main;
1941 	long long block_count;
1942 	int segs = secs * sbi->segs_per_sec;
1943 
1944 	f2fs_down_write(&sbi->sb_lock);
1945 
1946 	section_count = le32_to_cpu(raw_sb->section_count);
1947 	segment_count = le32_to_cpu(raw_sb->segment_count);
1948 	segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
1949 	block_count = le64_to_cpu(raw_sb->block_count);
1950 
1951 	raw_sb->section_count = cpu_to_le32(section_count + secs);
1952 	raw_sb->segment_count = cpu_to_le32(segment_count + segs);
1953 	raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
1954 	raw_sb->block_count = cpu_to_le64(block_count +
1955 					(long long)segs * sbi->blocks_per_seg);
1956 	if (f2fs_is_multi_device(sbi)) {
1957 		int last_dev = sbi->s_ndevs - 1;
1958 		int dev_segs =
1959 			le32_to_cpu(raw_sb->devs[last_dev].total_segments);
1960 
1961 		raw_sb->devs[last_dev].total_segments =
1962 						cpu_to_le32(dev_segs + segs);
1963 	}
1964 
1965 	f2fs_up_write(&sbi->sb_lock);
1966 }
1967 
1968 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
1969 {
1970 	int segs = secs * sbi->segs_per_sec;
1971 	long long blks = (long long)segs * sbi->blocks_per_seg;
1972 	long long user_block_count =
1973 				le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
1974 
1975 	SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
1976 	MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
1977 	MAIN_SECS(sbi) += secs;
1978 	FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
1979 	FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
1980 	F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
1981 
1982 	if (f2fs_is_multi_device(sbi)) {
1983 		int last_dev = sbi->s_ndevs - 1;
1984 
1985 		FDEV(last_dev).total_segments =
1986 				(int)FDEV(last_dev).total_segments + segs;
1987 		FDEV(last_dev).end_blk =
1988 				(long long)FDEV(last_dev).end_blk + blks;
1989 #ifdef CONFIG_BLK_DEV_ZONED
1990 		FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
1991 					(int)(blks >> sbi->log_blocks_per_blkz);
1992 #endif
1993 	}
1994 }
1995 
1996 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
1997 {
1998 	__u64 old_block_count, shrunk_blocks;
1999 	struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
2000 	unsigned int secs;
2001 	int err = 0;
2002 	__u32 rem;
2003 
2004 	old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
2005 	if (block_count > old_block_count)
2006 		return -EINVAL;
2007 
2008 	if (f2fs_is_multi_device(sbi)) {
2009 		int last_dev = sbi->s_ndevs - 1;
2010 		__u64 last_segs = FDEV(last_dev).total_segments;
2011 
2012 		if (block_count + last_segs * sbi->blocks_per_seg <=
2013 								old_block_count)
2014 			return -EINVAL;
2015 	}
2016 
2017 	/* new fs size should align to section size */
2018 	div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
2019 	if (rem)
2020 		return -EINVAL;
2021 
2022 	if (block_count == old_block_count)
2023 		return 0;
2024 
2025 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2026 		f2fs_err(sbi, "Should run fsck to repair first.");
2027 		return -EFSCORRUPTED;
2028 	}
2029 
2030 	if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2031 		f2fs_err(sbi, "Checkpoint should be enabled.");
2032 		return -EINVAL;
2033 	}
2034 
2035 	shrunk_blocks = old_block_count - block_count;
2036 	secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2037 
2038 	/* stop other GC */
2039 	if (!f2fs_down_write_trylock(&sbi->gc_lock))
2040 		return -EAGAIN;
2041 
2042 	/* stop CP to protect MAIN_SEC in free_segment_range */
2043 	f2fs_lock_op(sbi);
2044 
2045 	spin_lock(&sbi->stat_lock);
2046 	if (shrunk_blocks + valid_user_blocks(sbi) +
2047 		sbi->current_reserved_blocks + sbi->unusable_block_count +
2048 		F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2049 		err = -ENOSPC;
2050 	spin_unlock(&sbi->stat_lock);
2051 
2052 	if (err)
2053 		goto out_unlock;
2054 
2055 	err = free_segment_range(sbi, secs, true);
2056 
2057 out_unlock:
2058 	f2fs_unlock_op(sbi);
2059 	f2fs_up_write(&sbi->gc_lock);
2060 	if (err)
2061 		return err;
2062 
2063 	set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2064 
2065 	freeze_super(sbi->sb);
2066 	f2fs_down_write(&sbi->gc_lock);
2067 	f2fs_down_write(&sbi->cp_global_sem);
2068 
2069 	spin_lock(&sbi->stat_lock);
2070 	if (shrunk_blocks + valid_user_blocks(sbi) +
2071 		sbi->current_reserved_blocks + sbi->unusable_block_count +
2072 		F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2073 		err = -ENOSPC;
2074 	else
2075 		sbi->user_block_count -= shrunk_blocks;
2076 	spin_unlock(&sbi->stat_lock);
2077 	if (err)
2078 		goto out_err;
2079 
2080 	err = free_segment_range(sbi, secs, false);
2081 	if (err)
2082 		goto recover_out;
2083 
2084 	update_sb_metadata(sbi, -secs);
2085 
2086 	err = f2fs_commit_super(sbi, false);
2087 	if (err) {
2088 		update_sb_metadata(sbi, secs);
2089 		goto recover_out;
2090 	}
2091 
2092 	update_fs_metadata(sbi, -secs);
2093 	clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2094 	set_sbi_flag(sbi, SBI_IS_DIRTY);
2095 
2096 	err = f2fs_write_checkpoint(sbi, &cpc);
2097 	if (err) {
2098 		update_fs_metadata(sbi, secs);
2099 		update_sb_metadata(sbi, secs);
2100 		f2fs_commit_super(sbi, false);
2101 	}
2102 recover_out:
2103 	if (err) {
2104 		set_sbi_flag(sbi, SBI_NEED_FSCK);
2105 		f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2106 
2107 		spin_lock(&sbi->stat_lock);
2108 		sbi->user_block_count += shrunk_blocks;
2109 		spin_unlock(&sbi->stat_lock);
2110 	}
2111 out_err:
2112 	f2fs_up_write(&sbi->cp_global_sem);
2113 	f2fs_up_write(&sbi->gc_lock);
2114 	thaw_super(sbi->sb);
2115 	clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2116 	return err;
2117 }
2118