xref: /openbmc/linux/fs/f2fs/gc.c (revision 8ffdff6a)
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/backing-dev.h>
11 #include <linux/init.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/kthread.h>
14 #include <linux/delay.h>
15 #include <linux/freezer.h>
16 #include <linux/sched/signal.h>
17 
18 #include "f2fs.h"
19 #include "node.h"
20 #include "segment.h"
21 #include "gc.h"
22 #include <trace/events/f2fs.h>
23 
24 static struct kmem_cache *victim_entry_slab;
25 
26 static unsigned int count_bits(const unsigned long *addr,
27 				unsigned int offset, unsigned int len);
28 
29 static int gc_thread_func(void *data)
30 {
31 	struct f2fs_sb_info *sbi = data;
32 	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
33 	wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
34 	unsigned int wait_ms;
35 
36 	wait_ms = gc_th->min_sleep_time;
37 
38 	set_freezable();
39 	do {
40 		bool sync_mode;
41 
42 		wait_event_interruptible_timeout(*wq,
43 				kthread_should_stop() || freezing(current) ||
44 				gc_th->gc_wake,
45 				msecs_to_jiffies(wait_ms));
46 
47 		/* give it a try one time */
48 		if (gc_th->gc_wake)
49 			gc_th->gc_wake = 0;
50 
51 		if (try_to_freeze()) {
52 			stat_other_skip_bggc_count(sbi);
53 			continue;
54 		}
55 		if (kthread_should_stop())
56 			break;
57 
58 		if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
59 			increase_sleep_time(gc_th, &wait_ms);
60 			stat_other_skip_bggc_count(sbi);
61 			continue;
62 		}
63 
64 		if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
65 			f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
66 			f2fs_stop_checkpoint(sbi, false);
67 		}
68 
69 		if (!sb_start_write_trylock(sbi->sb)) {
70 			stat_other_skip_bggc_count(sbi);
71 			continue;
72 		}
73 
74 		/*
75 		 * [GC triggering condition]
76 		 * 0. GC is not conducted currently.
77 		 * 1. There are enough dirty segments.
78 		 * 2. IO subsystem is idle by checking the # of writeback pages.
79 		 * 3. IO subsystem is idle by checking the # of requests in
80 		 *    bdev's request list.
81 		 *
82 		 * Note) We have to avoid triggering GCs frequently.
83 		 * Because it is possible that some segments can be
84 		 * invalidated soon after by user update or deletion.
85 		 * So, I'd like to wait some time to collect dirty segments.
86 		 */
87 		if (sbi->gc_mode == GC_URGENT_HIGH) {
88 			wait_ms = gc_th->urgent_sleep_time;
89 			down_write(&sbi->gc_lock);
90 			goto do_gc;
91 		}
92 
93 		if (!down_write_trylock(&sbi->gc_lock)) {
94 			stat_other_skip_bggc_count(sbi);
95 			goto next;
96 		}
97 
98 		if (!is_idle(sbi, GC_TIME)) {
99 			increase_sleep_time(gc_th, &wait_ms);
100 			up_write(&sbi->gc_lock);
101 			stat_io_skip_bggc_count(sbi);
102 			goto next;
103 		}
104 
105 		if (has_enough_invalid_blocks(sbi))
106 			decrease_sleep_time(gc_th, &wait_ms);
107 		else
108 			increase_sleep_time(gc_th, &wait_ms);
109 do_gc:
110 		stat_inc_bggc_count(sbi->stat_info);
111 
112 		sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
113 
114 		/* if return value is not zero, no victim was selected */
115 		if (f2fs_gc(sbi, sync_mode, true, NULL_SEGNO))
116 			wait_ms = gc_th->no_gc_sleep_time;
117 
118 		trace_f2fs_background_gc(sbi->sb, wait_ms,
119 				prefree_segments(sbi), free_segments(sbi));
120 
121 		/* balancing f2fs's metadata periodically */
122 		f2fs_balance_fs_bg(sbi, true);
123 next:
124 		sb_end_write(sbi->sb);
125 
126 	} while (!kthread_should_stop());
127 	return 0;
128 }
129 
130 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
131 {
132 	struct f2fs_gc_kthread *gc_th;
133 	dev_t dev = sbi->sb->s_bdev->bd_dev;
134 	int err = 0;
135 
136 	gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
137 	if (!gc_th) {
138 		err = -ENOMEM;
139 		goto out;
140 	}
141 
142 	gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
143 	gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
144 	gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
145 	gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
146 
147 	gc_th->gc_wake= 0;
148 
149 	sbi->gc_thread = gc_th;
150 	init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
151 	sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
152 			"f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
153 	if (IS_ERR(gc_th->f2fs_gc_task)) {
154 		err = PTR_ERR(gc_th->f2fs_gc_task);
155 		kfree(gc_th);
156 		sbi->gc_thread = NULL;
157 	}
158 out:
159 	return err;
160 }
161 
162 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
163 {
164 	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
165 	if (!gc_th)
166 		return;
167 	kthread_stop(gc_th->f2fs_gc_task);
168 	kfree(gc_th);
169 	sbi->gc_thread = NULL;
170 }
171 
172 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
173 {
174 	int gc_mode;
175 
176 	if (gc_type == BG_GC) {
177 		if (sbi->am.atgc_enabled)
178 			gc_mode = GC_AT;
179 		else
180 			gc_mode = GC_CB;
181 	} else {
182 		gc_mode = GC_GREEDY;
183 	}
184 
185 	switch (sbi->gc_mode) {
186 	case GC_IDLE_CB:
187 		gc_mode = GC_CB;
188 		break;
189 	case GC_IDLE_GREEDY:
190 	case GC_URGENT_HIGH:
191 		gc_mode = GC_GREEDY;
192 		break;
193 	case GC_IDLE_AT:
194 		gc_mode = GC_AT;
195 		break;
196 	}
197 
198 	return gc_mode;
199 }
200 
201 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
202 			int type, struct victim_sel_policy *p)
203 {
204 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
205 
206 	if (p->alloc_mode == SSR) {
207 		p->gc_mode = GC_GREEDY;
208 		p->dirty_bitmap = dirty_i->dirty_segmap[type];
209 		p->max_search = dirty_i->nr_dirty[type];
210 		p->ofs_unit = 1;
211 	} else if (p->alloc_mode == AT_SSR) {
212 		p->gc_mode = GC_GREEDY;
213 		p->dirty_bitmap = dirty_i->dirty_segmap[type];
214 		p->max_search = dirty_i->nr_dirty[type];
215 		p->ofs_unit = 1;
216 	} else {
217 		p->gc_mode = select_gc_type(sbi, gc_type);
218 		p->ofs_unit = sbi->segs_per_sec;
219 		if (__is_large_section(sbi)) {
220 			p->dirty_bitmap = dirty_i->dirty_secmap;
221 			p->max_search = count_bits(p->dirty_bitmap,
222 						0, MAIN_SECS(sbi));
223 		} else {
224 			p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
225 			p->max_search = dirty_i->nr_dirty[DIRTY];
226 		}
227 	}
228 
229 	/*
230 	 * adjust candidates range, should select all dirty segments for
231 	 * foreground GC and urgent GC cases.
232 	 */
233 	if (gc_type != FG_GC &&
234 			(sbi->gc_mode != GC_URGENT_HIGH) &&
235 			(p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
236 			p->max_search > sbi->max_victim_search)
237 		p->max_search = sbi->max_victim_search;
238 
239 	/* let's select beginning hot/small space first in no_heap mode*/
240 	if (test_opt(sbi, NOHEAP) &&
241 		(type == CURSEG_HOT_DATA || IS_NODESEG(type)))
242 		p->offset = 0;
243 	else
244 		p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
245 }
246 
247 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
248 				struct victim_sel_policy *p)
249 {
250 	/* SSR allocates in a segment unit */
251 	if (p->alloc_mode == SSR)
252 		return sbi->blocks_per_seg;
253 	else if (p->alloc_mode == AT_SSR)
254 		return UINT_MAX;
255 
256 	/* LFS */
257 	if (p->gc_mode == GC_GREEDY)
258 		return 2 * sbi->blocks_per_seg * p->ofs_unit;
259 	else if (p->gc_mode == GC_CB)
260 		return UINT_MAX;
261 	else if (p->gc_mode == GC_AT)
262 		return UINT_MAX;
263 	else /* No other gc_mode */
264 		return 0;
265 }
266 
267 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
268 {
269 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
270 	unsigned int secno;
271 
272 	/*
273 	 * If the gc_type is FG_GC, we can select victim segments
274 	 * selected by background GC before.
275 	 * Those segments guarantee they have small valid blocks.
276 	 */
277 	for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
278 		if (sec_usage_check(sbi, secno))
279 			continue;
280 		clear_bit(secno, dirty_i->victim_secmap);
281 		return GET_SEG_FROM_SEC(sbi, secno);
282 	}
283 	return NULL_SEGNO;
284 }
285 
286 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
287 {
288 	struct sit_info *sit_i = SIT_I(sbi);
289 	unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
290 	unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
291 	unsigned long long mtime = 0;
292 	unsigned int vblocks;
293 	unsigned char age = 0;
294 	unsigned char u;
295 	unsigned int i;
296 	unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
297 
298 	for (i = 0; i < usable_segs_per_sec; i++)
299 		mtime += get_seg_entry(sbi, start + i)->mtime;
300 	vblocks = get_valid_blocks(sbi, segno, true);
301 
302 	mtime = div_u64(mtime, usable_segs_per_sec);
303 	vblocks = div_u64(vblocks, usable_segs_per_sec);
304 
305 	u = (vblocks * 100) >> sbi->log_blocks_per_seg;
306 
307 	/* Handle if the system time has changed by the user */
308 	if (mtime < sit_i->min_mtime)
309 		sit_i->min_mtime = mtime;
310 	if (mtime > sit_i->max_mtime)
311 		sit_i->max_mtime = mtime;
312 	if (sit_i->max_mtime != sit_i->min_mtime)
313 		age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
314 				sit_i->max_mtime - sit_i->min_mtime);
315 
316 	return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
317 }
318 
319 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
320 			unsigned int segno, struct victim_sel_policy *p)
321 {
322 	if (p->alloc_mode == SSR)
323 		return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
324 
325 	/* alloc_mode == LFS */
326 	if (p->gc_mode == GC_GREEDY)
327 		return get_valid_blocks(sbi, segno, true);
328 	else if (p->gc_mode == GC_CB)
329 		return get_cb_cost(sbi, segno);
330 
331 	f2fs_bug_on(sbi, 1);
332 	return 0;
333 }
334 
335 static unsigned int count_bits(const unsigned long *addr,
336 				unsigned int offset, unsigned int len)
337 {
338 	unsigned int end = offset + len, sum = 0;
339 
340 	while (offset < end) {
341 		if (test_bit(offset++, addr))
342 			++sum;
343 	}
344 	return sum;
345 }
346 
347 static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi,
348 				unsigned long long mtime, unsigned int segno,
349 				struct rb_node *parent, struct rb_node **p,
350 				bool left_most)
351 {
352 	struct atgc_management *am = &sbi->am;
353 	struct victim_entry *ve;
354 
355 	ve =  f2fs_kmem_cache_alloc(victim_entry_slab, GFP_NOFS);
356 
357 	ve->mtime = mtime;
358 	ve->segno = segno;
359 
360 	rb_link_node(&ve->rb_node, parent, p);
361 	rb_insert_color_cached(&ve->rb_node, &am->root, left_most);
362 
363 	list_add_tail(&ve->list, &am->victim_list);
364 
365 	am->victim_count++;
366 
367 	return ve;
368 }
369 
370 static void insert_victim_entry(struct f2fs_sb_info *sbi,
371 				unsigned long long mtime, unsigned int segno)
372 {
373 	struct atgc_management *am = &sbi->am;
374 	struct rb_node **p;
375 	struct rb_node *parent = NULL;
376 	bool left_most = true;
377 
378 	p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most);
379 	attach_victim_entry(sbi, mtime, segno, parent, p, left_most);
380 }
381 
382 static void add_victim_entry(struct f2fs_sb_info *sbi,
383 				struct victim_sel_policy *p, unsigned int segno)
384 {
385 	struct sit_info *sit_i = SIT_I(sbi);
386 	unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
387 	unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
388 	unsigned long long mtime = 0;
389 	unsigned int i;
390 
391 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
392 		if (p->gc_mode == GC_AT &&
393 			get_valid_blocks(sbi, segno, true) == 0)
394 			return;
395 
396 		if (p->alloc_mode == AT_SSR &&
397 			get_seg_entry(sbi, segno)->ckpt_valid_blocks == 0)
398 			return;
399 	}
400 
401 	for (i = 0; i < sbi->segs_per_sec; i++)
402 		mtime += get_seg_entry(sbi, start + i)->mtime;
403 	mtime = div_u64(mtime, sbi->segs_per_sec);
404 
405 	/* Handle if the system time has changed by the user */
406 	if (mtime < sit_i->min_mtime)
407 		sit_i->min_mtime = mtime;
408 	if (mtime > sit_i->max_mtime)
409 		sit_i->max_mtime = mtime;
410 	if (mtime < sit_i->dirty_min_mtime)
411 		sit_i->dirty_min_mtime = mtime;
412 	if (mtime > sit_i->dirty_max_mtime)
413 		sit_i->dirty_max_mtime = mtime;
414 
415 	/* don't choose young section as candidate */
416 	if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
417 		return;
418 
419 	insert_victim_entry(sbi, mtime, segno);
420 }
421 
422 static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi,
423 						struct victim_sel_policy *p)
424 {
425 	struct atgc_management *am = &sbi->am;
426 	struct rb_node *parent = NULL;
427 	bool left_most;
428 
429 	f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most);
430 
431 	return parent;
432 }
433 
434 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
435 						struct victim_sel_policy *p)
436 {
437 	struct sit_info *sit_i = SIT_I(sbi);
438 	struct atgc_management *am = &sbi->am;
439 	struct rb_root_cached *root = &am->root;
440 	struct rb_node *node;
441 	struct rb_entry *re;
442 	struct victim_entry *ve;
443 	unsigned long long total_time;
444 	unsigned long long age, u, accu;
445 	unsigned long long max_mtime = sit_i->dirty_max_mtime;
446 	unsigned long long min_mtime = sit_i->dirty_min_mtime;
447 	unsigned int sec_blocks = BLKS_PER_SEC(sbi);
448 	unsigned int vblocks;
449 	unsigned int dirty_threshold = max(am->max_candidate_count,
450 					am->candidate_ratio *
451 					am->victim_count / 100);
452 	unsigned int age_weight = am->age_weight;
453 	unsigned int cost;
454 	unsigned int iter = 0;
455 
456 	if (max_mtime < min_mtime)
457 		return;
458 
459 	max_mtime += 1;
460 	total_time = max_mtime - min_mtime;
461 
462 	accu = div64_u64(ULLONG_MAX, total_time);
463 	accu = min_t(unsigned long long, div_u64(accu, 100),
464 					DEFAULT_ACCURACY_CLASS);
465 
466 	node = rb_first_cached(root);
467 next:
468 	re = rb_entry_safe(node, struct rb_entry, rb_node);
469 	if (!re)
470 		return;
471 
472 	ve = (struct victim_entry *)re;
473 
474 	if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
475 		goto skip;
476 
477 	/* age = 10000 * x% * 60 */
478 	age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
479 								age_weight;
480 
481 	vblocks = get_valid_blocks(sbi, ve->segno, true);
482 	f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
483 
484 	/* u = 10000 * x% * 40 */
485 	u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
486 							(100 - age_weight);
487 
488 	f2fs_bug_on(sbi, age + u >= UINT_MAX);
489 
490 	cost = UINT_MAX - (age + u);
491 	iter++;
492 
493 	if (cost < p->min_cost ||
494 			(cost == p->min_cost && age > p->oldest_age)) {
495 		p->min_cost = cost;
496 		p->oldest_age = age;
497 		p->min_segno = ve->segno;
498 	}
499 skip:
500 	if (iter < dirty_threshold) {
501 		node = rb_next(node);
502 		goto next;
503 	}
504 }
505 
506 /*
507  * select candidates around source section in range of
508  * [target - dirty_threshold, target + dirty_threshold]
509  */
510 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
511 						struct victim_sel_policy *p)
512 {
513 	struct sit_info *sit_i = SIT_I(sbi);
514 	struct atgc_management *am = &sbi->am;
515 	struct rb_node *node;
516 	struct rb_entry *re;
517 	struct victim_entry *ve;
518 	unsigned long long age;
519 	unsigned long long max_mtime = sit_i->dirty_max_mtime;
520 	unsigned long long min_mtime = sit_i->dirty_min_mtime;
521 	unsigned int seg_blocks = sbi->blocks_per_seg;
522 	unsigned int vblocks;
523 	unsigned int dirty_threshold = max(am->max_candidate_count,
524 					am->candidate_ratio *
525 					am->victim_count / 100);
526 	unsigned int cost;
527 	unsigned int iter = 0;
528 	int stage = 0;
529 
530 	if (max_mtime < min_mtime)
531 		return;
532 	max_mtime += 1;
533 next_stage:
534 	node = lookup_central_victim(sbi, p);
535 next_node:
536 	re = rb_entry_safe(node, struct rb_entry, rb_node);
537 	if (!re) {
538 		if (stage == 0)
539 			goto skip_stage;
540 		return;
541 	}
542 
543 	ve = (struct victim_entry *)re;
544 
545 	if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
546 		goto skip_node;
547 
548 	age = max_mtime - ve->mtime;
549 
550 	vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
551 	f2fs_bug_on(sbi, !vblocks);
552 
553 	/* rare case */
554 	if (vblocks == seg_blocks)
555 		goto skip_node;
556 
557 	iter++;
558 
559 	age = max_mtime - abs(p->age - age);
560 	cost = UINT_MAX - vblocks;
561 
562 	if (cost < p->min_cost ||
563 			(cost == p->min_cost && age > p->oldest_age)) {
564 		p->min_cost = cost;
565 		p->oldest_age = age;
566 		p->min_segno = ve->segno;
567 	}
568 skip_node:
569 	if (iter < dirty_threshold) {
570 		if (stage == 0)
571 			node = rb_prev(node);
572 		else if (stage == 1)
573 			node = rb_next(node);
574 		goto next_node;
575 	}
576 skip_stage:
577 	if (stage < 1) {
578 		stage++;
579 		iter = 0;
580 		goto next_stage;
581 	}
582 }
583 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
584 						struct victim_sel_policy *p)
585 {
586 	f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
587 						&sbi->am.root, true));
588 
589 	if (p->gc_mode == GC_AT)
590 		atgc_lookup_victim(sbi, p);
591 	else if (p->alloc_mode == AT_SSR)
592 		atssr_lookup_victim(sbi, p);
593 	else
594 		f2fs_bug_on(sbi, 1);
595 }
596 
597 static void release_victim_entry(struct f2fs_sb_info *sbi)
598 {
599 	struct atgc_management *am = &sbi->am;
600 	struct victim_entry *ve, *tmp;
601 
602 	list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
603 		list_del(&ve->list);
604 		kmem_cache_free(victim_entry_slab, ve);
605 		am->victim_count--;
606 	}
607 
608 	am->root = RB_ROOT_CACHED;
609 
610 	f2fs_bug_on(sbi, am->victim_count);
611 	f2fs_bug_on(sbi, !list_empty(&am->victim_list));
612 }
613 
614 /*
615  * This function is called from two paths.
616  * One is garbage collection and the other is SSR segment selection.
617  * When it is called during GC, it just gets a victim segment
618  * and it does not remove it from dirty seglist.
619  * When it is called from SSR segment selection, it finds a segment
620  * which has minimum valid blocks and removes it from dirty seglist.
621  */
622 static int get_victim_by_default(struct f2fs_sb_info *sbi,
623 			unsigned int *result, int gc_type, int type,
624 			char alloc_mode, unsigned long long age)
625 {
626 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
627 	struct sit_info *sm = SIT_I(sbi);
628 	struct victim_sel_policy p;
629 	unsigned int secno, last_victim;
630 	unsigned int last_segment;
631 	unsigned int nsearched;
632 	bool is_atgc;
633 	int ret = 0;
634 
635 	mutex_lock(&dirty_i->seglist_lock);
636 	last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
637 
638 	p.alloc_mode = alloc_mode;
639 	p.age = age;
640 	p.age_threshold = sbi->am.age_threshold;
641 
642 retry:
643 	select_policy(sbi, gc_type, type, &p);
644 	p.min_segno = NULL_SEGNO;
645 	p.oldest_age = 0;
646 	p.min_cost = get_max_cost(sbi, &p);
647 
648 	is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
649 	nsearched = 0;
650 
651 	if (is_atgc)
652 		SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
653 
654 	if (*result != NULL_SEGNO) {
655 		if (!get_valid_blocks(sbi, *result, false)) {
656 			ret = -ENODATA;
657 			goto out;
658 		}
659 
660 		if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
661 			ret = -EBUSY;
662 		else
663 			p.min_segno = *result;
664 		goto out;
665 	}
666 
667 	ret = -ENODATA;
668 	if (p.max_search == 0)
669 		goto out;
670 
671 	if (__is_large_section(sbi) && p.alloc_mode == LFS) {
672 		if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
673 			p.min_segno = sbi->next_victim_seg[BG_GC];
674 			*result = p.min_segno;
675 			sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
676 			goto got_result;
677 		}
678 		if (gc_type == FG_GC &&
679 				sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
680 			p.min_segno = sbi->next_victim_seg[FG_GC];
681 			*result = p.min_segno;
682 			sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
683 			goto got_result;
684 		}
685 	}
686 
687 	last_victim = sm->last_victim[p.gc_mode];
688 	if (p.alloc_mode == LFS && gc_type == FG_GC) {
689 		p.min_segno = check_bg_victims(sbi);
690 		if (p.min_segno != NULL_SEGNO)
691 			goto got_it;
692 	}
693 
694 	while (1) {
695 		unsigned long cost, *dirty_bitmap;
696 		unsigned int unit_no, segno;
697 
698 		dirty_bitmap = p.dirty_bitmap;
699 		unit_no = find_next_bit(dirty_bitmap,
700 				last_segment / p.ofs_unit,
701 				p.offset / p.ofs_unit);
702 		segno = unit_no * p.ofs_unit;
703 		if (segno >= last_segment) {
704 			if (sm->last_victim[p.gc_mode]) {
705 				last_segment =
706 					sm->last_victim[p.gc_mode];
707 				sm->last_victim[p.gc_mode] = 0;
708 				p.offset = 0;
709 				continue;
710 			}
711 			break;
712 		}
713 
714 		p.offset = segno + p.ofs_unit;
715 		nsearched++;
716 
717 #ifdef CONFIG_F2FS_CHECK_FS
718 		/*
719 		 * skip selecting the invalid segno (that is failed due to block
720 		 * validity check failure during GC) to avoid endless GC loop in
721 		 * such cases.
722 		 */
723 		if (test_bit(segno, sm->invalid_segmap))
724 			goto next;
725 #endif
726 
727 		secno = GET_SEC_FROM_SEG(sbi, segno);
728 
729 		if (sec_usage_check(sbi, secno))
730 			goto next;
731 		/* Don't touch checkpointed data */
732 		if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
733 					get_ckpt_valid_blocks(sbi, segno) &&
734 					p.alloc_mode == LFS))
735 			goto next;
736 		if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
737 			goto next;
738 
739 		if (is_atgc) {
740 			add_victim_entry(sbi, &p, segno);
741 			goto next;
742 		}
743 
744 		cost = get_gc_cost(sbi, segno, &p);
745 
746 		if (p.min_cost > cost) {
747 			p.min_segno = segno;
748 			p.min_cost = cost;
749 		}
750 next:
751 		if (nsearched >= p.max_search) {
752 			if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
753 				sm->last_victim[p.gc_mode] =
754 					last_victim + p.ofs_unit;
755 			else
756 				sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
757 			sm->last_victim[p.gc_mode] %=
758 				(MAIN_SECS(sbi) * sbi->segs_per_sec);
759 			break;
760 		}
761 	}
762 
763 	/* get victim for GC_AT/AT_SSR */
764 	if (is_atgc) {
765 		lookup_victim_by_age(sbi, &p);
766 		release_victim_entry(sbi);
767 	}
768 
769 	if (is_atgc && p.min_segno == NULL_SEGNO &&
770 			sm->elapsed_time < p.age_threshold) {
771 		p.age_threshold = 0;
772 		goto retry;
773 	}
774 
775 	if (p.min_segno != NULL_SEGNO) {
776 got_it:
777 		*result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
778 got_result:
779 		if (p.alloc_mode == LFS) {
780 			secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
781 			if (gc_type == FG_GC)
782 				sbi->cur_victim_sec = secno;
783 			else
784 				set_bit(secno, dirty_i->victim_secmap);
785 		}
786 		ret = 0;
787 
788 	}
789 out:
790 	if (p.min_segno != NULL_SEGNO)
791 		trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
792 				sbi->cur_victim_sec,
793 				prefree_segments(sbi), free_segments(sbi));
794 	mutex_unlock(&dirty_i->seglist_lock);
795 
796 	return ret;
797 }
798 
799 static const struct victim_selection default_v_ops = {
800 	.get_victim = get_victim_by_default,
801 };
802 
803 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
804 {
805 	struct inode_entry *ie;
806 
807 	ie = radix_tree_lookup(&gc_list->iroot, ino);
808 	if (ie)
809 		return ie->inode;
810 	return NULL;
811 }
812 
813 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
814 {
815 	struct inode_entry *new_ie;
816 
817 	if (inode == find_gc_inode(gc_list, inode->i_ino)) {
818 		iput(inode);
819 		return;
820 	}
821 	new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS);
822 	new_ie->inode = inode;
823 
824 	f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
825 	list_add_tail(&new_ie->list, &gc_list->ilist);
826 }
827 
828 static void put_gc_inode(struct gc_inode_list *gc_list)
829 {
830 	struct inode_entry *ie, *next_ie;
831 	list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
832 		radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
833 		iput(ie->inode);
834 		list_del(&ie->list);
835 		kmem_cache_free(f2fs_inode_entry_slab, ie);
836 	}
837 }
838 
839 static int check_valid_map(struct f2fs_sb_info *sbi,
840 				unsigned int segno, int offset)
841 {
842 	struct sit_info *sit_i = SIT_I(sbi);
843 	struct seg_entry *sentry;
844 	int ret;
845 
846 	down_read(&sit_i->sentry_lock);
847 	sentry = get_seg_entry(sbi, segno);
848 	ret = f2fs_test_bit(offset, sentry->cur_valid_map);
849 	up_read(&sit_i->sentry_lock);
850 	return ret;
851 }
852 
853 /*
854  * This function compares node address got in summary with that in NAT.
855  * On validity, copy that node with cold status, otherwise (invalid node)
856  * ignore that.
857  */
858 static int gc_node_segment(struct f2fs_sb_info *sbi,
859 		struct f2fs_summary *sum, unsigned int segno, int gc_type)
860 {
861 	struct f2fs_summary *entry;
862 	block_t start_addr;
863 	int off;
864 	int phase = 0;
865 	bool fggc = (gc_type == FG_GC);
866 	int submitted = 0;
867 	unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
868 
869 	start_addr = START_BLOCK(sbi, segno);
870 
871 next_step:
872 	entry = sum;
873 
874 	if (fggc && phase == 2)
875 		atomic_inc(&sbi->wb_sync_req[NODE]);
876 
877 	for (off = 0; off < usable_blks_in_seg; off++, entry++) {
878 		nid_t nid = le32_to_cpu(entry->nid);
879 		struct page *node_page;
880 		struct node_info ni;
881 		int err;
882 
883 		/* stop BG_GC if there is not enough free sections. */
884 		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
885 			return submitted;
886 
887 		if (check_valid_map(sbi, segno, off) == 0)
888 			continue;
889 
890 		if (phase == 0) {
891 			f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
892 							META_NAT, true);
893 			continue;
894 		}
895 
896 		if (phase == 1) {
897 			f2fs_ra_node_page(sbi, nid);
898 			continue;
899 		}
900 
901 		/* phase == 2 */
902 		node_page = f2fs_get_node_page(sbi, nid);
903 		if (IS_ERR(node_page))
904 			continue;
905 
906 		/* block may become invalid during f2fs_get_node_page */
907 		if (check_valid_map(sbi, segno, off) == 0) {
908 			f2fs_put_page(node_page, 1);
909 			continue;
910 		}
911 
912 		if (f2fs_get_node_info(sbi, nid, &ni)) {
913 			f2fs_put_page(node_page, 1);
914 			continue;
915 		}
916 
917 		if (ni.blk_addr != start_addr + off) {
918 			f2fs_put_page(node_page, 1);
919 			continue;
920 		}
921 
922 		err = f2fs_move_node_page(node_page, gc_type);
923 		if (!err && gc_type == FG_GC)
924 			submitted++;
925 		stat_inc_node_blk_count(sbi, 1, gc_type);
926 	}
927 
928 	if (++phase < 3)
929 		goto next_step;
930 
931 	if (fggc)
932 		atomic_dec(&sbi->wb_sync_req[NODE]);
933 	return submitted;
934 }
935 
936 /*
937  * Calculate start block index indicating the given node offset.
938  * Be careful, caller should give this node offset only indicating direct node
939  * blocks. If any node offsets, which point the other types of node blocks such
940  * as indirect or double indirect node blocks, are given, it must be a caller's
941  * bug.
942  */
943 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
944 {
945 	unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
946 	unsigned int bidx;
947 
948 	if (node_ofs == 0)
949 		return 0;
950 
951 	if (node_ofs <= 2) {
952 		bidx = node_ofs - 1;
953 	} else if (node_ofs <= indirect_blks) {
954 		int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
955 		bidx = node_ofs - 2 - dec;
956 	} else {
957 		int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
958 		bidx = node_ofs - 5 - dec;
959 	}
960 	return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
961 }
962 
963 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
964 		struct node_info *dni, block_t blkaddr, unsigned int *nofs)
965 {
966 	struct page *node_page;
967 	nid_t nid;
968 	unsigned int ofs_in_node;
969 	block_t source_blkaddr;
970 
971 	nid = le32_to_cpu(sum->nid);
972 	ofs_in_node = le16_to_cpu(sum->ofs_in_node);
973 
974 	node_page = f2fs_get_node_page(sbi, nid);
975 	if (IS_ERR(node_page))
976 		return false;
977 
978 	if (f2fs_get_node_info(sbi, nid, dni)) {
979 		f2fs_put_page(node_page, 1);
980 		return false;
981 	}
982 
983 	if (sum->version != dni->version) {
984 		f2fs_warn(sbi, "%s: valid data with mismatched node version.",
985 			  __func__);
986 		set_sbi_flag(sbi, SBI_NEED_FSCK);
987 	}
988 
989 	*nofs = ofs_of_node(node_page);
990 	source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
991 	f2fs_put_page(node_page, 1);
992 
993 	if (source_blkaddr != blkaddr) {
994 #ifdef CONFIG_F2FS_CHECK_FS
995 		unsigned int segno = GET_SEGNO(sbi, blkaddr);
996 		unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
997 
998 		if (unlikely(check_valid_map(sbi, segno, offset))) {
999 			if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1000 				f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u\n",
1001 						blkaddr, source_blkaddr, segno);
1002 				f2fs_bug_on(sbi, 1);
1003 			}
1004 		}
1005 #endif
1006 		return false;
1007 	}
1008 	return true;
1009 }
1010 
1011 static int ra_data_block(struct inode *inode, pgoff_t index)
1012 {
1013 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1014 	struct address_space *mapping = inode->i_mapping;
1015 	struct dnode_of_data dn;
1016 	struct page *page;
1017 	struct extent_info ei = {0, 0, 0};
1018 	struct f2fs_io_info fio = {
1019 		.sbi = sbi,
1020 		.ino = inode->i_ino,
1021 		.type = DATA,
1022 		.temp = COLD,
1023 		.op = REQ_OP_READ,
1024 		.op_flags = 0,
1025 		.encrypted_page = NULL,
1026 		.in_list = false,
1027 		.retry = false,
1028 	};
1029 	int err;
1030 
1031 	page = f2fs_grab_cache_page(mapping, index, true);
1032 	if (!page)
1033 		return -ENOMEM;
1034 
1035 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1036 		dn.data_blkaddr = ei.blk + index - ei.fofs;
1037 		if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1038 						DATA_GENERIC_ENHANCE_READ))) {
1039 			err = -EFSCORRUPTED;
1040 			goto put_page;
1041 		}
1042 		goto got_it;
1043 	}
1044 
1045 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1046 	err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1047 	if (err)
1048 		goto put_page;
1049 	f2fs_put_dnode(&dn);
1050 
1051 	if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1052 		err = -ENOENT;
1053 		goto put_page;
1054 	}
1055 	if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1056 						DATA_GENERIC_ENHANCE))) {
1057 		err = -EFSCORRUPTED;
1058 		goto put_page;
1059 	}
1060 got_it:
1061 	/* read page */
1062 	fio.page = page;
1063 	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1064 
1065 	/*
1066 	 * don't cache encrypted data into meta inode until previous dirty
1067 	 * data were writebacked to avoid racing between GC and flush.
1068 	 */
1069 	f2fs_wait_on_page_writeback(page, DATA, true, true);
1070 
1071 	f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1072 
1073 	fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1074 					dn.data_blkaddr,
1075 					FGP_LOCK | FGP_CREAT, GFP_NOFS);
1076 	if (!fio.encrypted_page) {
1077 		err = -ENOMEM;
1078 		goto put_page;
1079 	}
1080 
1081 	err = f2fs_submit_page_bio(&fio);
1082 	if (err)
1083 		goto put_encrypted_page;
1084 	f2fs_put_page(fio.encrypted_page, 0);
1085 	f2fs_put_page(page, 1);
1086 
1087 	f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1088 	f2fs_update_iostat(sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1089 
1090 	return 0;
1091 put_encrypted_page:
1092 	f2fs_put_page(fio.encrypted_page, 1);
1093 put_page:
1094 	f2fs_put_page(page, 1);
1095 	return err;
1096 }
1097 
1098 /*
1099  * Move data block via META_MAPPING while keeping locked data page.
1100  * This can be used to move blocks, aka LBAs, directly on disk.
1101  */
1102 static int move_data_block(struct inode *inode, block_t bidx,
1103 				int gc_type, unsigned int segno, int off)
1104 {
1105 	struct f2fs_io_info fio = {
1106 		.sbi = F2FS_I_SB(inode),
1107 		.ino = inode->i_ino,
1108 		.type = DATA,
1109 		.temp = COLD,
1110 		.op = REQ_OP_READ,
1111 		.op_flags = 0,
1112 		.encrypted_page = NULL,
1113 		.in_list = false,
1114 		.retry = false,
1115 	};
1116 	struct dnode_of_data dn;
1117 	struct f2fs_summary sum;
1118 	struct node_info ni;
1119 	struct page *page, *mpage;
1120 	block_t newaddr;
1121 	int err = 0;
1122 	bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1123 	int type = fio.sbi->am.atgc_enabled ?
1124 				CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1125 
1126 	/* do not read out */
1127 	page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1128 	if (!page)
1129 		return -ENOMEM;
1130 
1131 	if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1132 		err = -ENOENT;
1133 		goto out;
1134 	}
1135 
1136 	if (f2fs_is_atomic_file(inode)) {
1137 		F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1138 		F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1139 		err = -EAGAIN;
1140 		goto out;
1141 	}
1142 
1143 	if (f2fs_is_pinned_file(inode)) {
1144 		f2fs_pin_file_control(inode, true);
1145 		err = -EAGAIN;
1146 		goto out;
1147 	}
1148 
1149 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1150 	err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1151 	if (err)
1152 		goto out;
1153 
1154 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1155 		ClearPageUptodate(page);
1156 		err = -ENOENT;
1157 		goto put_out;
1158 	}
1159 
1160 	/*
1161 	 * don't cache encrypted data into meta inode until previous dirty
1162 	 * data were writebacked to avoid racing between GC and flush.
1163 	 */
1164 	f2fs_wait_on_page_writeback(page, DATA, true, true);
1165 
1166 	f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1167 
1168 	err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1169 	if (err)
1170 		goto put_out;
1171 
1172 	/* read page */
1173 	fio.page = page;
1174 	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1175 
1176 	if (lfs_mode)
1177 		down_write(&fio.sbi->io_order_lock);
1178 
1179 	mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1180 					fio.old_blkaddr, false);
1181 	if (!mpage) {
1182 		err = -ENOMEM;
1183 		goto up_out;
1184 	}
1185 
1186 	fio.encrypted_page = mpage;
1187 
1188 	/* read source block in mpage */
1189 	if (!PageUptodate(mpage)) {
1190 		err = f2fs_submit_page_bio(&fio);
1191 		if (err) {
1192 			f2fs_put_page(mpage, 1);
1193 			goto up_out;
1194 		}
1195 
1196 		f2fs_update_iostat(fio.sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1197 		f2fs_update_iostat(fio.sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1198 
1199 		lock_page(mpage);
1200 		if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1201 						!PageUptodate(mpage))) {
1202 			err = -EIO;
1203 			f2fs_put_page(mpage, 1);
1204 			goto up_out;
1205 		}
1206 	}
1207 
1208 	set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1209 
1210 	/* allocate block address */
1211 	f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1212 				&sum, type, NULL);
1213 
1214 	fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1215 				newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1216 	if (!fio.encrypted_page) {
1217 		err = -ENOMEM;
1218 		f2fs_put_page(mpage, 1);
1219 		goto recover_block;
1220 	}
1221 
1222 	/* write target block */
1223 	f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1224 	memcpy(page_address(fio.encrypted_page),
1225 				page_address(mpage), PAGE_SIZE);
1226 	f2fs_put_page(mpage, 1);
1227 	invalidate_mapping_pages(META_MAPPING(fio.sbi),
1228 				fio.old_blkaddr, fio.old_blkaddr);
1229 
1230 	set_page_dirty(fio.encrypted_page);
1231 	if (clear_page_dirty_for_io(fio.encrypted_page))
1232 		dec_page_count(fio.sbi, F2FS_DIRTY_META);
1233 
1234 	set_page_writeback(fio.encrypted_page);
1235 	ClearPageError(page);
1236 
1237 	fio.op = REQ_OP_WRITE;
1238 	fio.op_flags = REQ_SYNC;
1239 	fio.new_blkaddr = newaddr;
1240 	f2fs_submit_page_write(&fio);
1241 	if (fio.retry) {
1242 		err = -EAGAIN;
1243 		if (PageWriteback(fio.encrypted_page))
1244 			end_page_writeback(fio.encrypted_page);
1245 		goto put_page_out;
1246 	}
1247 
1248 	f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
1249 
1250 	f2fs_update_data_blkaddr(&dn, newaddr);
1251 	set_inode_flag(inode, FI_APPEND_WRITE);
1252 	if (page->index == 0)
1253 		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1254 put_page_out:
1255 	f2fs_put_page(fio.encrypted_page, 1);
1256 recover_block:
1257 	if (err)
1258 		f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1259 							true, true, true);
1260 up_out:
1261 	if (lfs_mode)
1262 		up_write(&fio.sbi->io_order_lock);
1263 put_out:
1264 	f2fs_put_dnode(&dn);
1265 out:
1266 	f2fs_put_page(page, 1);
1267 	return err;
1268 }
1269 
1270 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1271 							unsigned int segno, int off)
1272 {
1273 	struct page *page;
1274 	int err = 0;
1275 
1276 	page = f2fs_get_lock_data_page(inode, bidx, true);
1277 	if (IS_ERR(page))
1278 		return PTR_ERR(page);
1279 
1280 	if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1281 		err = -ENOENT;
1282 		goto out;
1283 	}
1284 
1285 	if (f2fs_is_atomic_file(inode)) {
1286 		F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1287 		F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1288 		err = -EAGAIN;
1289 		goto out;
1290 	}
1291 	if (f2fs_is_pinned_file(inode)) {
1292 		if (gc_type == FG_GC)
1293 			f2fs_pin_file_control(inode, true);
1294 		err = -EAGAIN;
1295 		goto out;
1296 	}
1297 
1298 	if (gc_type == BG_GC) {
1299 		if (PageWriteback(page)) {
1300 			err = -EAGAIN;
1301 			goto out;
1302 		}
1303 		set_page_dirty(page);
1304 		set_cold_data(page);
1305 	} else {
1306 		struct f2fs_io_info fio = {
1307 			.sbi = F2FS_I_SB(inode),
1308 			.ino = inode->i_ino,
1309 			.type = DATA,
1310 			.temp = COLD,
1311 			.op = REQ_OP_WRITE,
1312 			.op_flags = REQ_SYNC,
1313 			.old_blkaddr = NULL_ADDR,
1314 			.page = page,
1315 			.encrypted_page = NULL,
1316 			.need_lock = LOCK_REQ,
1317 			.io_type = FS_GC_DATA_IO,
1318 		};
1319 		bool is_dirty = PageDirty(page);
1320 
1321 retry:
1322 		f2fs_wait_on_page_writeback(page, DATA, true, true);
1323 
1324 		set_page_dirty(page);
1325 		if (clear_page_dirty_for_io(page)) {
1326 			inode_dec_dirty_pages(inode);
1327 			f2fs_remove_dirty_inode(inode);
1328 		}
1329 
1330 		set_cold_data(page);
1331 
1332 		err = f2fs_do_write_data_page(&fio);
1333 		if (err) {
1334 			clear_cold_data(page);
1335 			if (err == -ENOMEM) {
1336 				congestion_wait(BLK_RW_ASYNC,
1337 						DEFAULT_IO_TIMEOUT);
1338 				goto retry;
1339 			}
1340 			if (is_dirty)
1341 				set_page_dirty(page);
1342 		}
1343 	}
1344 out:
1345 	f2fs_put_page(page, 1);
1346 	return err;
1347 }
1348 
1349 /*
1350  * This function tries to get parent node of victim data block, and identifies
1351  * data block validity. If the block is valid, copy that with cold status and
1352  * modify parent node.
1353  * If the parent node is not valid or the data block address is different,
1354  * the victim data block is ignored.
1355  */
1356 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1357 		struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
1358 {
1359 	struct super_block *sb = sbi->sb;
1360 	struct f2fs_summary *entry;
1361 	block_t start_addr;
1362 	int off;
1363 	int phase = 0;
1364 	int submitted = 0;
1365 	unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1366 
1367 	start_addr = START_BLOCK(sbi, segno);
1368 
1369 next_step:
1370 	entry = sum;
1371 
1372 	for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1373 		struct page *data_page;
1374 		struct inode *inode;
1375 		struct node_info dni; /* dnode info for the data */
1376 		unsigned int ofs_in_node, nofs;
1377 		block_t start_bidx;
1378 		nid_t nid = le32_to_cpu(entry->nid);
1379 
1380 		/*
1381 		 * stop BG_GC if there is not enough free sections.
1382 		 * Or, stop GC if the segment becomes fully valid caused by
1383 		 * race condition along with SSR block allocation.
1384 		 */
1385 		if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1386 				get_valid_blocks(sbi, segno, true) ==
1387 							BLKS_PER_SEC(sbi))
1388 			return submitted;
1389 
1390 		if (check_valid_map(sbi, segno, off) == 0)
1391 			continue;
1392 
1393 		if (phase == 0) {
1394 			f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1395 							META_NAT, true);
1396 			continue;
1397 		}
1398 
1399 		if (phase == 1) {
1400 			f2fs_ra_node_page(sbi, nid);
1401 			continue;
1402 		}
1403 
1404 		/* Get an inode by ino with checking validity */
1405 		if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1406 			continue;
1407 
1408 		if (phase == 2) {
1409 			f2fs_ra_node_page(sbi, dni.ino);
1410 			continue;
1411 		}
1412 
1413 		ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1414 
1415 		if (phase == 3) {
1416 			inode = f2fs_iget(sb, dni.ino);
1417 			if (IS_ERR(inode) || is_bad_inode(inode)) {
1418 				set_sbi_flag(sbi, SBI_NEED_FSCK);
1419 				continue;
1420 			}
1421 
1422 			if (!down_write_trylock(
1423 				&F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1424 				iput(inode);
1425 				sbi->skipped_gc_rwsem++;
1426 				continue;
1427 			}
1428 
1429 			start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1430 								ofs_in_node;
1431 
1432 			if (f2fs_post_read_required(inode)) {
1433 				int err = ra_data_block(inode, start_bidx);
1434 
1435 				up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1436 				if (err) {
1437 					iput(inode);
1438 					continue;
1439 				}
1440 				add_gc_inode(gc_list, inode);
1441 				continue;
1442 			}
1443 
1444 			data_page = f2fs_get_read_data_page(inode,
1445 						start_bidx, REQ_RAHEAD, true);
1446 			up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1447 			if (IS_ERR(data_page)) {
1448 				iput(inode);
1449 				continue;
1450 			}
1451 
1452 			f2fs_put_page(data_page, 0);
1453 			add_gc_inode(gc_list, inode);
1454 			continue;
1455 		}
1456 
1457 		/* phase 4 */
1458 		inode = find_gc_inode(gc_list, dni.ino);
1459 		if (inode) {
1460 			struct f2fs_inode_info *fi = F2FS_I(inode);
1461 			bool locked = false;
1462 			int err;
1463 
1464 			if (S_ISREG(inode->i_mode)) {
1465 				if (!down_write_trylock(&fi->i_gc_rwsem[READ]))
1466 					continue;
1467 				if (!down_write_trylock(
1468 						&fi->i_gc_rwsem[WRITE])) {
1469 					sbi->skipped_gc_rwsem++;
1470 					up_write(&fi->i_gc_rwsem[READ]);
1471 					continue;
1472 				}
1473 				locked = true;
1474 
1475 				/* wait for all inflight aio data */
1476 				inode_dio_wait(inode);
1477 			}
1478 
1479 			start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1480 								+ ofs_in_node;
1481 			if (f2fs_post_read_required(inode))
1482 				err = move_data_block(inode, start_bidx,
1483 							gc_type, segno, off);
1484 			else
1485 				err = move_data_page(inode, start_bidx, gc_type,
1486 								segno, off);
1487 
1488 			if (!err && (gc_type == FG_GC ||
1489 					f2fs_post_read_required(inode)))
1490 				submitted++;
1491 
1492 			if (locked) {
1493 				up_write(&fi->i_gc_rwsem[WRITE]);
1494 				up_write(&fi->i_gc_rwsem[READ]);
1495 			}
1496 
1497 			stat_inc_data_blk_count(sbi, 1, gc_type);
1498 		}
1499 	}
1500 
1501 	if (++phase < 5)
1502 		goto next_step;
1503 
1504 	return submitted;
1505 }
1506 
1507 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1508 			int gc_type)
1509 {
1510 	struct sit_info *sit_i = SIT_I(sbi);
1511 	int ret;
1512 
1513 	down_write(&sit_i->sentry_lock);
1514 	ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1515 					      NO_CHECK_TYPE, LFS, 0);
1516 	up_write(&sit_i->sentry_lock);
1517 	return ret;
1518 }
1519 
1520 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1521 				unsigned int start_segno,
1522 				struct gc_inode_list *gc_list, int gc_type)
1523 {
1524 	struct page *sum_page;
1525 	struct f2fs_summary_block *sum;
1526 	struct blk_plug plug;
1527 	unsigned int segno = start_segno;
1528 	unsigned int end_segno = start_segno + sbi->segs_per_sec;
1529 	int seg_freed = 0, migrated = 0;
1530 	unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1531 						SUM_TYPE_DATA : SUM_TYPE_NODE;
1532 	int submitted = 0;
1533 
1534 	if (__is_large_section(sbi))
1535 		end_segno = rounddown(end_segno, sbi->segs_per_sec);
1536 
1537 	/*
1538 	 * zone-capacity can be less than zone-size in zoned devices,
1539 	 * resulting in less than expected usable segments in the zone,
1540 	 * calculate the end segno in the zone which can be garbage collected
1541 	 */
1542 	if (f2fs_sb_has_blkzoned(sbi))
1543 		end_segno -= sbi->segs_per_sec -
1544 					f2fs_usable_segs_in_sec(sbi, segno);
1545 
1546 	sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1547 
1548 	/* readahead multi ssa blocks those have contiguous address */
1549 	if (__is_large_section(sbi))
1550 		f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1551 					end_segno - segno, META_SSA, true);
1552 
1553 	/* reference all summary page */
1554 	while (segno < end_segno) {
1555 		sum_page = f2fs_get_sum_page(sbi, segno++);
1556 		if (IS_ERR(sum_page)) {
1557 			int err = PTR_ERR(sum_page);
1558 
1559 			end_segno = segno - 1;
1560 			for (segno = start_segno; segno < end_segno; segno++) {
1561 				sum_page = find_get_page(META_MAPPING(sbi),
1562 						GET_SUM_BLOCK(sbi, segno));
1563 				f2fs_put_page(sum_page, 0);
1564 				f2fs_put_page(sum_page, 0);
1565 			}
1566 			return err;
1567 		}
1568 		unlock_page(sum_page);
1569 	}
1570 
1571 	blk_start_plug(&plug);
1572 
1573 	for (segno = start_segno; segno < end_segno; segno++) {
1574 
1575 		/* find segment summary of victim */
1576 		sum_page = find_get_page(META_MAPPING(sbi),
1577 					GET_SUM_BLOCK(sbi, segno));
1578 		f2fs_put_page(sum_page, 0);
1579 
1580 		if (get_valid_blocks(sbi, segno, false) == 0)
1581 			goto freed;
1582 		if (gc_type == BG_GC && __is_large_section(sbi) &&
1583 				migrated >= sbi->migration_granularity)
1584 			goto skip;
1585 		if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1586 			goto skip;
1587 
1588 		sum = page_address(sum_page);
1589 		if (type != GET_SUM_TYPE((&sum->footer))) {
1590 			f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1591 				 segno, type, GET_SUM_TYPE((&sum->footer)));
1592 			set_sbi_flag(sbi, SBI_NEED_FSCK);
1593 			f2fs_stop_checkpoint(sbi, false);
1594 			goto skip;
1595 		}
1596 
1597 		/*
1598 		 * this is to avoid deadlock:
1599 		 * - lock_page(sum_page)         - f2fs_replace_block
1600 		 *  - check_valid_map()            - down_write(sentry_lock)
1601 		 *   - down_read(sentry_lock)     - change_curseg()
1602 		 *                                  - lock_page(sum_page)
1603 		 */
1604 		if (type == SUM_TYPE_NODE)
1605 			submitted += gc_node_segment(sbi, sum->entries, segno,
1606 								gc_type);
1607 		else
1608 			submitted += gc_data_segment(sbi, sum->entries, gc_list,
1609 							segno, gc_type);
1610 
1611 		stat_inc_seg_count(sbi, type, gc_type);
1612 		migrated++;
1613 
1614 freed:
1615 		if (gc_type == FG_GC &&
1616 				get_valid_blocks(sbi, segno, false) == 0)
1617 			seg_freed++;
1618 
1619 		if (__is_large_section(sbi) && segno + 1 < end_segno)
1620 			sbi->next_victim_seg[gc_type] = segno + 1;
1621 skip:
1622 		f2fs_put_page(sum_page, 0);
1623 	}
1624 
1625 	if (submitted)
1626 		f2fs_submit_merged_write(sbi,
1627 				(type == SUM_TYPE_NODE) ? NODE : DATA);
1628 
1629 	blk_finish_plug(&plug);
1630 
1631 	stat_inc_call_count(sbi->stat_info);
1632 
1633 	return seg_freed;
1634 }
1635 
1636 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1637 			bool background, unsigned int segno)
1638 {
1639 	int gc_type = sync ? FG_GC : BG_GC;
1640 	int sec_freed = 0, seg_freed = 0, total_freed = 0;
1641 	int ret = 0;
1642 	struct cp_control cpc;
1643 	unsigned int init_segno = segno;
1644 	struct gc_inode_list gc_list = {
1645 		.ilist = LIST_HEAD_INIT(gc_list.ilist),
1646 		.iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1647 	};
1648 	unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1649 	unsigned long long first_skipped;
1650 	unsigned int skipped_round = 0, round = 0;
1651 
1652 	trace_f2fs_gc_begin(sbi->sb, sync, background,
1653 				get_pages(sbi, F2FS_DIRTY_NODES),
1654 				get_pages(sbi, F2FS_DIRTY_DENTS),
1655 				get_pages(sbi, F2FS_DIRTY_IMETA),
1656 				free_sections(sbi),
1657 				free_segments(sbi),
1658 				reserved_segments(sbi),
1659 				prefree_segments(sbi));
1660 
1661 	cpc.reason = __get_cp_reason(sbi);
1662 	sbi->skipped_gc_rwsem = 0;
1663 	first_skipped = last_skipped;
1664 gc_more:
1665 	if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1666 		ret = -EINVAL;
1667 		goto stop;
1668 	}
1669 	if (unlikely(f2fs_cp_error(sbi))) {
1670 		ret = -EIO;
1671 		goto stop;
1672 	}
1673 
1674 	if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1675 		/*
1676 		 * For example, if there are many prefree_segments below given
1677 		 * threshold, we can make them free by checkpoint. Then, we
1678 		 * secure free segments which doesn't need fggc any more.
1679 		 */
1680 		if (prefree_segments(sbi) &&
1681 				!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1682 			ret = f2fs_write_checkpoint(sbi, &cpc);
1683 			if (ret)
1684 				goto stop;
1685 		}
1686 		if (has_not_enough_free_secs(sbi, 0, 0))
1687 			gc_type = FG_GC;
1688 	}
1689 
1690 	/* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1691 	if (gc_type == BG_GC && !background) {
1692 		ret = -EINVAL;
1693 		goto stop;
1694 	}
1695 	ret = __get_victim(sbi, &segno, gc_type);
1696 	if (ret)
1697 		goto stop;
1698 
1699 	seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type);
1700 	if (gc_type == FG_GC &&
1701 		seg_freed == f2fs_usable_segs_in_sec(sbi, segno))
1702 		sec_freed++;
1703 	total_freed += seg_freed;
1704 
1705 	if (gc_type == FG_GC) {
1706 		if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
1707 						sbi->skipped_gc_rwsem)
1708 			skipped_round++;
1709 		last_skipped = sbi->skipped_atomic_files[FG_GC];
1710 		round++;
1711 	}
1712 
1713 	if (gc_type == FG_GC && seg_freed)
1714 		sbi->cur_victim_sec = NULL_SEGNO;
1715 
1716 	if (sync)
1717 		goto stop;
1718 
1719 	if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1720 		if (skipped_round <= MAX_SKIP_GC_COUNT ||
1721 					skipped_round * 2 < round) {
1722 			segno = NULL_SEGNO;
1723 			goto gc_more;
1724 		}
1725 
1726 		if (first_skipped < last_skipped &&
1727 				(last_skipped - first_skipped) >
1728 						sbi->skipped_gc_rwsem) {
1729 			f2fs_drop_inmem_pages_all(sbi, true);
1730 			segno = NULL_SEGNO;
1731 			goto gc_more;
1732 		}
1733 		if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1734 			ret = f2fs_write_checkpoint(sbi, &cpc);
1735 	}
1736 stop:
1737 	SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1738 	SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1739 
1740 	trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1741 				get_pages(sbi, F2FS_DIRTY_NODES),
1742 				get_pages(sbi, F2FS_DIRTY_DENTS),
1743 				get_pages(sbi, F2FS_DIRTY_IMETA),
1744 				free_sections(sbi),
1745 				free_segments(sbi),
1746 				reserved_segments(sbi),
1747 				prefree_segments(sbi));
1748 
1749 	up_write(&sbi->gc_lock);
1750 
1751 	put_gc_inode(&gc_list);
1752 
1753 	if (sync && !ret)
1754 		ret = sec_freed ? 0 : -EAGAIN;
1755 	return ret;
1756 }
1757 
1758 int __init f2fs_create_garbage_collection_cache(void)
1759 {
1760 	victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1761 					sizeof(struct victim_entry));
1762 	if (!victim_entry_slab)
1763 		return -ENOMEM;
1764 	return 0;
1765 }
1766 
1767 void f2fs_destroy_garbage_collection_cache(void)
1768 {
1769 	kmem_cache_destroy(victim_entry_slab);
1770 }
1771 
1772 static void init_atgc_management(struct f2fs_sb_info *sbi)
1773 {
1774 	struct atgc_management *am = &sbi->am;
1775 
1776 	if (test_opt(sbi, ATGC) &&
1777 		SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1778 		am->atgc_enabled = true;
1779 
1780 	am->root = RB_ROOT_CACHED;
1781 	INIT_LIST_HEAD(&am->victim_list);
1782 	am->victim_count = 0;
1783 
1784 	am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1785 	am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1786 	am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1787 }
1788 
1789 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1790 {
1791 	DIRTY_I(sbi)->v_ops = &default_v_ops;
1792 
1793 	sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1794 
1795 	/* give warm/cold data area from slower device */
1796 	if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1797 		SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1798 				GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1799 
1800 	init_atgc_management(sbi);
1801 }
1802 
1803 static int free_segment_range(struct f2fs_sb_info *sbi,
1804 				unsigned int secs, bool gc_only)
1805 {
1806 	unsigned int segno, next_inuse, start, end;
1807 	struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1808 	int gc_mode, gc_type;
1809 	int err = 0;
1810 	int type;
1811 
1812 	/* Force block allocation for GC */
1813 	MAIN_SECS(sbi) -= secs;
1814 	start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1815 	end = MAIN_SEGS(sbi) - 1;
1816 
1817 	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1818 	for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1819 		if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1820 			SIT_I(sbi)->last_victim[gc_mode] = 0;
1821 
1822 	for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1823 		if (sbi->next_victim_seg[gc_type] >= start)
1824 			sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1825 	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1826 
1827 	/* Move out cursegs from the target range */
1828 	for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1829 		f2fs_allocate_segment_for_resize(sbi, type, start, end);
1830 
1831 	/* do GC to move out valid blocks in the range */
1832 	for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1833 		struct gc_inode_list gc_list = {
1834 			.ilist = LIST_HEAD_INIT(gc_list.ilist),
1835 			.iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1836 		};
1837 
1838 		do_garbage_collect(sbi, segno, &gc_list, FG_GC);
1839 		put_gc_inode(&gc_list);
1840 
1841 		if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1842 			err = -EAGAIN;
1843 			goto out;
1844 		}
1845 		if (fatal_signal_pending(current)) {
1846 			err = -ERESTARTSYS;
1847 			goto out;
1848 		}
1849 	}
1850 	if (gc_only)
1851 		goto out;
1852 
1853 	err = f2fs_write_checkpoint(sbi, &cpc);
1854 	if (err)
1855 		goto out;
1856 
1857 	next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1858 	if (next_inuse <= end) {
1859 		f2fs_err(sbi, "segno %u should be free but still inuse!",
1860 			 next_inuse);
1861 		f2fs_bug_on(sbi, 1);
1862 	}
1863 out:
1864 	MAIN_SECS(sbi) += secs;
1865 	return err;
1866 }
1867 
1868 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1869 {
1870 	struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1871 	int section_count;
1872 	int segment_count;
1873 	int segment_count_main;
1874 	long long block_count;
1875 	int segs = secs * sbi->segs_per_sec;
1876 
1877 	down_write(&sbi->sb_lock);
1878 
1879 	section_count = le32_to_cpu(raw_sb->section_count);
1880 	segment_count = le32_to_cpu(raw_sb->segment_count);
1881 	segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
1882 	block_count = le64_to_cpu(raw_sb->block_count);
1883 
1884 	raw_sb->section_count = cpu_to_le32(section_count + secs);
1885 	raw_sb->segment_count = cpu_to_le32(segment_count + segs);
1886 	raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
1887 	raw_sb->block_count = cpu_to_le64(block_count +
1888 					(long long)segs * sbi->blocks_per_seg);
1889 	if (f2fs_is_multi_device(sbi)) {
1890 		int last_dev = sbi->s_ndevs - 1;
1891 		int dev_segs =
1892 			le32_to_cpu(raw_sb->devs[last_dev].total_segments);
1893 
1894 		raw_sb->devs[last_dev].total_segments =
1895 						cpu_to_le32(dev_segs + segs);
1896 	}
1897 
1898 	up_write(&sbi->sb_lock);
1899 }
1900 
1901 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
1902 {
1903 	int segs = secs * sbi->segs_per_sec;
1904 	long long blks = (long long)segs * sbi->blocks_per_seg;
1905 	long long user_block_count =
1906 				le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
1907 
1908 	SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
1909 	MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
1910 	MAIN_SECS(sbi) += secs;
1911 	FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
1912 	FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
1913 	F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
1914 
1915 	if (f2fs_is_multi_device(sbi)) {
1916 		int last_dev = sbi->s_ndevs - 1;
1917 
1918 		FDEV(last_dev).total_segments =
1919 				(int)FDEV(last_dev).total_segments + segs;
1920 		FDEV(last_dev).end_blk =
1921 				(long long)FDEV(last_dev).end_blk + blks;
1922 #ifdef CONFIG_BLK_DEV_ZONED
1923 		FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
1924 					(int)(blks >> sbi->log_blocks_per_blkz);
1925 #endif
1926 	}
1927 }
1928 
1929 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
1930 {
1931 	__u64 old_block_count, shrunk_blocks;
1932 	struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1933 	unsigned int secs;
1934 	int err = 0;
1935 	__u32 rem;
1936 
1937 	old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
1938 	if (block_count > old_block_count)
1939 		return -EINVAL;
1940 
1941 	if (f2fs_is_multi_device(sbi)) {
1942 		int last_dev = sbi->s_ndevs - 1;
1943 		__u64 last_segs = FDEV(last_dev).total_segments;
1944 
1945 		if (block_count + last_segs * sbi->blocks_per_seg <=
1946 								old_block_count)
1947 			return -EINVAL;
1948 	}
1949 
1950 	/* new fs size should align to section size */
1951 	div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
1952 	if (rem)
1953 		return -EINVAL;
1954 
1955 	if (block_count == old_block_count)
1956 		return 0;
1957 
1958 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1959 		f2fs_err(sbi, "Should run fsck to repair first.");
1960 		return -EFSCORRUPTED;
1961 	}
1962 
1963 	if (test_opt(sbi, DISABLE_CHECKPOINT)) {
1964 		f2fs_err(sbi, "Checkpoint should be enabled.");
1965 		return -EINVAL;
1966 	}
1967 
1968 	shrunk_blocks = old_block_count - block_count;
1969 	secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
1970 
1971 	/* stop other GC */
1972 	if (!down_write_trylock(&sbi->gc_lock))
1973 		return -EAGAIN;
1974 
1975 	/* stop CP to protect MAIN_SEC in free_segment_range */
1976 	f2fs_lock_op(sbi);
1977 	err = free_segment_range(sbi, secs, true);
1978 	f2fs_unlock_op(sbi);
1979 	up_write(&sbi->gc_lock);
1980 	if (err)
1981 		return err;
1982 
1983 	set_sbi_flag(sbi, SBI_IS_RESIZEFS);
1984 
1985 	freeze_super(sbi->sb);
1986 	down_write(&sbi->gc_lock);
1987 	down_write(&sbi->cp_global_sem);
1988 
1989 	spin_lock(&sbi->stat_lock);
1990 	if (shrunk_blocks + valid_user_blocks(sbi) +
1991 		sbi->current_reserved_blocks + sbi->unusable_block_count +
1992 		F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
1993 		err = -ENOSPC;
1994 	else
1995 		sbi->user_block_count -= shrunk_blocks;
1996 	spin_unlock(&sbi->stat_lock);
1997 	if (err)
1998 		goto out_err;
1999 
2000 	err = free_segment_range(sbi, secs, false);
2001 	if (err)
2002 		goto recover_out;
2003 
2004 	update_sb_metadata(sbi, -secs);
2005 
2006 	err = f2fs_commit_super(sbi, false);
2007 	if (err) {
2008 		update_sb_metadata(sbi, secs);
2009 		goto recover_out;
2010 	}
2011 
2012 	update_fs_metadata(sbi, -secs);
2013 	clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2014 	set_sbi_flag(sbi, SBI_IS_DIRTY);
2015 
2016 	err = f2fs_write_checkpoint(sbi, &cpc);
2017 	if (err) {
2018 		update_fs_metadata(sbi, secs);
2019 		update_sb_metadata(sbi, secs);
2020 		f2fs_commit_super(sbi, false);
2021 	}
2022 recover_out:
2023 	if (err) {
2024 		set_sbi_flag(sbi, SBI_NEED_FSCK);
2025 		f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2026 
2027 		spin_lock(&sbi->stat_lock);
2028 		sbi->user_block_count += shrunk_blocks;
2029 		spin_unlock(&sbi->stat_lock);
2030 	}
2031 out_err:
2032 	up_write(&sbi->cp_global_sem);
2033 	up_write(&sbi->gc_lock);
2034 	thaw_super(sbi->sb);
2035 	clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2036 	return err;
2037 }
2038