xref: /openbmc/linux/fs/f2fs/gc.c (revision f608c38c59c6020bfde14af88630b8d7817003f9)
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 unsigned int count_bits(const unsigned long *addr,
25 				unsigned int offset, unsigned int len);
26 
27 static int gc_thread_func(void *data)
28 {
29 	struct f2fs_sb_info *sbi = data;
30 	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
31 	wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
32 	unsigned int wait_ms;
33 
34 	wait_ms = gc_th->min_sleep_time;
35 
36 	set_freezable();
37 	do {
38 		bool sync_mode;
39 
40 		wait_event_interruptible_timeout(*wq,
41 				kthread_should_stop() || freezing(current) ||
42 				gc_th->gc_wake,
43 				msecs_to_jiffies(wait_ms));
44 
45 		/* give it a try one time */
46 		if (gc_th->gc_wake)
47 			gc_th->gc_wake = 0;
48 
49 		if (try_to_freeze()) {
50 			stat_other_skip_bggc_count(sbi);
51 			continue;
52 		}
53 		if (kthread_should_stop())
54 			break;
55 
56 		if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
57 			increase_sleep_time(gc_th, &wait_ms);
58 			stat_other_skip_bggc_count(sbi);
59 			continue;
60 		}
61 
62 		if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
63 			f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
64 			f2fs_stop_checkpoint(sbi, false);
65 		}
66 
67 		if (!sb_start_write_trylock(sbi->sb)) {
68 			stat_other_skip_bggc_count(sbi);
69 			continue;
70 		}
71 
72 		/*
73 		 * [GC triggering condition]
74 		 * 0. GC is not conducted currently.
75 		 * 1. There are enough dirty segments.
76 		 * 2. IO subsystem is idle by checking the # of writeback pages.
77 		 * 3. IO subsystem is idle by checking the # of requests in
78 		 *    bdev's request list.
79 		 *
80 		 * Note) We have to avoid triggering GCs frequently.
81 		 * Because it is possible that some segments can be
82 		 * invalidated soon after by user update or deletion.
83 		 * So, I'd like to wait some time to collect dirty segments.
84 		 */
85 		if (sbi->gc_mode == GC_URGENT) {
86 			wait_ms = gc_th->urgent_sleep_time;
87 			down_write(&sbi->gc_lock);
88 			goto do_gc;
89 		}
90 
91 		if (!down_write_trylock(&sbi->gc_lock)) {
92 			stat_other_skip_bggc_count(sbi);
93 			goto next;
94 		}
95 
96 		if (!is_idle(sbi, GC_TIME)) {
97 			increase_sleep_time(gc_th, &wait_ms);
98 			up_write(&sbi->gc_lock);
99 			stat_io_skip_bggc_count(sbi);
100 			goto next;
101 		}
102 
103 		if (has_enough_invalid_blocks(sbi))
104 			decrease_sleep_time(gc_th, &wait_ms);
105 		else
106 			increase_sleep_time(gc_th, &wait_ms);
107 do_gc:
108 		stat_inc_bggc_count(sbi->stat_info);
109 
110 		sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
111 
112 		/* if return value is not zero, no victim was selected */
113 		if (f2fs_gc(sbi, sync_mode, true, NULL_SEGNO))
114 			wait_ms = gc_th->no_gc_sleep_time;
115 
116 		trace_f2fs_background_gc(sbi->sb, wait_ms,
117 				prefree_segments(sbi), free_segments(sbi));
118 
119 		/* balancing f2fs's metadata periodically */
120 		f2fs_balance_fs_bg(sbi, true);
121 next:
122 		sb_end_write(sbi->sb);
123 
124 	} while (!kthread_should_stop());
125 	return 0;
126 }
127 
128 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
129 {
130 	struct f2fs_gc_kthread *gc_th;
131 	dev_t dev = sbi->sb->s_bdev->bd_dev;
132 	int err = 0;
133 
134 	gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
135 	if (!gc_th) {
136 		err = -ENOMEM;
137 		goto out;
138 	}
139 
140 	gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
141 	gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
142 	gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
143 	gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
144 
145 	gc_th->gc_wake= 0;
146 
147 	sbi->gc_thread = gc_th;
148 	init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
149 	sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
150 			"f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
151 	if (IS_ERR(gc_th->f2fs_gc_task)) {
152 		err = PTR_ERR(gc_th->f2fs_gc_task);
153 		kvfree(gc_th);
154 		sbi->gc_thread = NULL;
155 	}
156 out:
157 	return err;
158 }
159 
160 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
161 {
162 	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
163 	if (!gc_th)
164 		return;
165 	kthread_stop(gc_th->f2fs_gc_task);
166 	kvfree(gc_th);
167 	sbi->gc_thread = NULL;
168 }
169 
170 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
171 {
172 	int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
173 
174 	switch (sbi->gc_mode) {
175 	case GC_IDLE_CB:
176 		gc_mode = GC_CB;
177 		break;
178 	case GC_IDLE_GREEDY:
179 	case GC_URGENT:
180 		gc_mode = GC_GREEDY;
181 		break;
182 	}
183 	return gc_mode;
184 }
185 
186 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
187 			int type, struct victim_sel_policy *p)
188 {
189 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
190 
191 	if (p->alloc_mode == SSR) {
192 		p->gc_mode = GC_GREEDY;
193 		p->dirty_bitmap = dirty_i->dirty_segmap[type];
194 		p->max_search = dirty_i->nr_dirty[type];
195 		p->ofs_unit = 1;
196 	} else {
197 		p->gc_mode = select_gc_type(sbi, gc_type);
198 		p->ofs_unit = sbi->segs_per_sec;
199 		if (__is_large_section(sbi)) {
200 			p->dirty_bitmap = dirty_i->dirty_secmap;
201 			p->max_search = count_bits(p->dirty_bitmap,
202 						0, MAIN_SECS(sbi));
203 		} else {
204 			p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
205 			p->max_search = dirty_i->nr_dirty[DIRTY];
206 		}
207 	}
208 
209 	/*
210 	 * adjust candidates range, should select all dirty segments for
211 	 * foreground GC and urgent GC cases.
212 	 */
213 	if (gc_type != FG_GC &&
214 			(sbi->gc_mode != GC_URGENT) &&
215 			p->max_search > sbi->max_victim_search)
216 		p->max_search = sbi->max_victim_search;
217 
218 	/* let's select beginning hot/small space first in no_heap mode*/
219 	if (test_opt(sbi, NOHEAP) &&
220 		(type == CURSEG_HOT_DATA || IS_NODESEG(type)))
221 		p->offset = 0;
222 	else
223 		p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
224 }
225 
226 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
227 				struct victim_sel_policy *p)
228 {
229 	/* SSR allocates in a segment unit */
230 	if (p->alloc_mode == SSR)
231 		return sbi->blocks_per_seg;
232 	if (p->gc_mode == GC_GREEDY)
233 		return 2 * sbi->blocks_per_seg * p->ofs_unit;
234 	else if (p->gc_mode == GC_CB)
235 		return UINT_MAX;
236 	else /* No other gc_mode */
237 		return 0;
238 }
239 
240 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
241 {
242 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
243 	unsigned int secno;
244 
245 	/*
246 	 * If the gc_type is FG_GC, we can select victim segments
247 	 * selected by background GC before.
248 	 * Those segments guarantee they have small valid blocks.
249 	 */
250 	for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
251 		if (sec_usage_check(sbi, secno))
252 			continue;
253 		clear_bit(secno, dirty_i->victim_secmap);
254 		return GET_SEG_FROM_SEC(sbi, secno);
255 	}
256 	return NULL_SEGNO;
257 }
258 
259 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
260 {
261 	struct sit_info *sit_i = SIT_I(sbi);
262 	unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
263 	unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
264 	unsigned long long mtime = 0;
265 	unsigned int vblocks;
266 	unsigned char age = 0;
267 	unsigned char u;
268 	unsigned int i;
269 
270 	for (i = 0; i < sbi->segs_per_sec; i++)
271 		mtime += get_seg_entry(sbi, start + i)->mtime;
272 	vblocks = get_valid_blocks(sbi, segno, true);
273 
274 	mtime = div_u64(mtime, sbi->segs_per_sec);
275 	vblocks = div_u64(vblocks, sbi->segs_per_sec);
276 
277 	u = (vblocks * 100) >> sbi->log_blocks_per_seg;
278 
279 	/* Handle if the system time has changed by the user */
280 	if (mtime < sit_i->min_mtime)
281 		sit_i->min_mtime = mtime;
282 	if (mtime > sit_i->max_mtime)
283 		sit_i->max_mtime = mtime;
284 	if (sit_i->max_mtime != sit_i->min_mtime)
285 		age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
286 				sit_i->max_mtime - sit_i->min_mtime);
287 
288 	return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
289 }
290 
291 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
292 			unsigned int segno, struct victim_sel_policy *p)
293 {
294 	if (p->alloc_mode == SSR)
295 		return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
296 
297 	/* alloc_mode == LFS */
298 	if (p->gc_mode == GC_GREEDY)
299 		return get_valid_blocks(sbi, segno, true);
300 	else
301 		return get_cb_cost(sbi, segno);
302 }
303 
304 static unsigned int count_bits(const unsigned long *addr,
305 				unsigned int offset, unsigned int len)
306 {
307 	unsigned int end = offset + len, sum = 0;
308 
309 	while (offset < end) {
310 		if (test_bit(offset++, addr))
311 			++sum;
312 	}
313 	return sum;
314 }
315 
316 /*
317  * This function is called from two paths.
318  * One is garbage collection and the other is SSR segment selection.
319  * When it is called during GC, it just gets a victim segment
320  * and it does not remove it from dirty seglist.
321  * When it is called from SSR segment selection, it finds a segment
322  * which has minimum valid blocks and removes it from dirty seglist.
323  */
324 static int get_victim_by_default(struct f2fs_sb_info *sbi,
325 		unsigned int *result, int gc_type, int type, char alloc_mode)
326 {
327 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
328 	struct sit_info *sm = SIT_I(sbi);
329 	struct victim_sel_policy p;
330 	unsigned int secno, last_victim;
331 	unsigned int last_segment;
332 	unsigned int nsearched = 0;
333 
334 	mutex_lock(&dirty_i->seglist_lock);
335 	last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
336 
337 	p.alloc_mode = alloc_mode;
338 	select_policy(sbi, gc_type, type, &p);
339 
340 	p.min_segno = NULL_SEGNO;
341 	p.min_cost = get_max_cost(sbi, &p);
342 
343 	if (*result != NULL_SEGNO) {
344 		if (get_valid_blocks(sbi, *result, false) &&
345 			!sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
346 			p.min_segno = *result;
347 		goto out;
348 	}
349 
350 	if (p.max_search == 0)
351 		goto out;
352 
353 	if (__is_large_section(sbi) && p.alloc_mode == LFS) {
354 		if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
355 			p.min_segno = sbi->next_victim_seg[BG_GC];
356 			*result = p.min_segno;
357 			sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
358 			goto got_result;
359 		}
360 		if (gc_type == FG_GC &&
361 				sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
362 			p.min_segno = sbi->next_victim_seg[FG_GC];
363 			*result = p.min_segno;
364 			sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
365 			goto got_result;
366 		}
367 	}
368 
369 	last_victim = sm->last_victim[p.gc_mode];
370 	if (p.alloc_mode == LFS && gc_type == FG_GC) {
371 		p.min_segno = check_bg_victims(sbi);
372 		if (p.min_segno != NULL_SEGNO)
373 			goto got_it;
374 	}
375 
376 	while (1) {
377 		unsigned long cost, *dirty_bitmap;
378 		unsigned int unit_no, segno;
379 
380 		dirty_bitmap = p.dirty_bitmap;
381 		unit_no = find_next_bit(dirty_bitmap,
382 				last_segment / p.ofs_unit,
383 				p.offset / p.ofs_unit);
384 		segno = unit_no * p.ofs_unit;
385 		if (segno >= last_segment) {
386 			if (sm->last_victim[p.gc_mode]) {
387 				last_segment =
388 					sm->last_victim[p.gc_mode];
389 				sm->last_victim[p.gc_mode] = 0;
390 				p.offset = 0;
391 				continue;
392 			}
393 			break;
394 		}
395 
396 		p.offset = segno + p.ofs_unit;
397 		nsearched++;
398 
399 #ifdef CONFIG_F2FS_CHECK_FS
400 		/*
401 		 * skip selecting the invalid segno (that is failed due to block
402 		 * validity check failure during GC) to avoid endless GC loop in
403 		 * such cases.
404 		 */
405 		if (test_bit(segno, sm->invalid_segmap))
406 			goto next;
407 #endif
408 
409 		secno = GET_SEC_FROM_SEG(sbi, segno);
410 
411 		if (sec_usage_check(sbi, secno))
412 			goto next;
413 		/* Don't touch checkpointed data */
414 		if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
415 					get_ckpt_valid_blocks(sbi, segno) &&
416 					p.alloc_mode != SSR))
417 			goto next;
418 		if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
419 			goto next;
420 
421 		cost = get_gc_cost(sbi, segno, &p);
422 
423 		if (p.min_cost > cost) {
424 			p.min_segno = segno;
425 			p.min_cost = cost;
426 		}
427 next:
428 		if (nsearched >= p.max_search) {
429 			if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
430 				sm->last_victim[p.gc_mode] =
431 					last_victim + p.ofs_unit;
432 			else
433 				sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
434 			sm->last_victim[p.gc_mode] %=
435 				(MAIN_SECS(sbi) * sbi->segs_per_sec);
436 			break;
437 		}
438 	}
439 	if (p.min_segno != NULL_SEGNO) {
440 got_it:
441 		*result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
442 got_result:
443 		if (p.alloc_mode == LFS) {
444 			secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
445 			if (gc_type == FG_GC)
446 				sbi->cur_victim_sec = secno;
447 			else
448 				set_bit(secno, dirty_i->victim_secmap);
449 		}
450 
451 	}
452 out:
453 	if (p.min_segno != NULL_SEGNO)
454 		trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
455 				sbi->cur_victim_sec,
456 				prefree_segments(sbi), free_segments(sbi));
457 	mutex_unlock(&dirty_i->seglist_lock);
458 
459 	return (p.min_segno == NULL_SEGNO) ? 0 : 1;
460 }
461 
462 static const struct victim_selection default_v_ops = {
463 	.get_victim = get_victim_by_default,
464 };
465 
466 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
467 {
468 	struct inode_entry *ie;
469 
470 	ie = radix_tree_lookup(&gc_list->iroot, ino);
471 	if (ie)
472 		return ie->inode;
473 	return NULL;
474 }
475 
476 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
477 {
478 	struct inode_entry *new_ie;
479 
480 	if (inode == find_gc_inode(gc_list, inode->i_ino)) {
481 		iput(inode);
482 		return;
483 	}
484 	new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS);
485 	new_ie->inode = inode;
486 
487 	f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
488 	list_add_tail(&new_ie->list, &gc_list->ilist);
489 }
490 
491 static void put_gc_inode(struct gc_inode_list *gc_list)
492 {
493 	struct inode_entry *ie, *next_ie;
494 	list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
495 		radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
496 		iput(ie->inode);
497 		list_del(&ie->list);
498 		kmem_cache_free(f2fs_inode_entry_slab, ie);
499 	}
500 }
501 
502 static int check_valid_map(struct f2fs_sb_info *sbi,
503 				unsigned int segno, int offset)
504 {
505 	struct sit_info *sit_i = SIT_I(sbi);
506 	struct seg_entry *sentry;
507 	int ret;
508 
509 	down_read(&sit_i->sentry_lock);
510 	sentry = get_seg_entry(sbi, segno);
511 	ret = f2fs_test_bit(offset, sentry->cur_valid_map);
512 	up_read(&sit_i->sentry_lock);
513 	return ret;
514 }
515 
516 /*
517  * This function compares node address got in summary with that in NAT.
518  * On validity, copy that node with cold status, otherwise (invalid node)
519  * ignore that.
520  */
521 static int gc_node_segment(struct f2fs_sb_info *sbi,
522 		struct f2fs_summary *sum, unsigned int segno, int gc_type)
523 {
524 	struct f2fs_summary *entry;
525 	block_t start_addr;
526 	int off;
527 	int phase = 0;
528 	bool fggc = (gc_type == FG_GC);
529 	int submitted = 0;
530 
531 	start_addr = START_BLOCK(sbi, segno);
532 
533 next_step:
534 	entry = sum;
535 
536 	if (fggc && phase == 2)
537 		atomic_inc(&sbi->wb_sync_req[NODE]);
538 
539 	for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
540 		nid_t nid = le32_to_cpu(entry->nid);
541 		struct page *node_page;
542 		struct node_info ni;
543 		int err;
544 
545 		/* stop BG_GC if there is not enough free sections. */
546 		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
547 			return submitted;
548 
549 		if (check_valid_map(sbi, segno, off) == 0)
550 			continue;
551 
552 		if (phase == 0) {
553 			f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
554 							META_NAT, true);
555 			continue;
556 		}
557 
558 		if (phase == 1) {
559 			f2fs_ra_node_page(sbi, nid);
560 			continue;
561 		}
562 
563 		/* phase == 2 */
564 		node_page = f2fs_get_node_page(sbi, nid);
565 		if (IS_ERR(node_page))
566 			continue;
567 
568 		/* block may become invalid during f2fs_get_node_page */
569 		if (check_valid_map(sbi, segno, off) == 0) {
570 			f2fs_put_page(node_page, 1);
571 			continue;
572 		}
573 
574 		if (f2fs_get_node_info(sbi, nid, &ni)) {
575 			f2fs_put_page(node_page, 1);
576 			continue;
577 		}
578 
579 		if (ni.blk_addr != start_addr + off) {
580 			f2fs_put_page(node_page, 1);
581 			continue;
582 		}
583 
584 		err = f2fs_move_node_page(node_page, gc_type);
585 		if (!err && gc_type == FG_GC)
586 			submitted++;
587 		stat_inc_node_blk_count(sbi, 1, gc_type);
588 	}
589 
590 	if (++phase < 3)
591 		goto next_step;
592 
593 	if (fggc)
594 		atomic_dec(&sbi->wb_sync_req[NODE]);
595 	return submitted;
596 }
597 
598 /*
599  * Calculate start block index indicating the given node offset.
600  * Be careful, caller should give this node offset only indicating direct node
601  * blocks. If any node offsets, which point the other types of node blocks such
602  * as indirect or double indirect node blocks, are given, it must be a caller's
603  * bug.
604  */
605 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
606 {
607 	unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
608 	unsigned int bidx;
609 
610 	if (node_ofs == 0)
611 		return 0;
612 
613 	if (node_ofs <= 2) {
614 		bidx = node_ofs - 1;
615 	} else if (node_ofs <= indirect_blks) {
616 		int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
617 		bidx = node_ofs - 2 - dec;
618 	} else {
619 		int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
620 		bidx = node_ofs - 5 - dec;
621 	}
622 	return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
623 }
624 
625 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
626 		struct node_info *dni, block_t blkaddr, unsigned int *nofs)
627 {
628 	struct page *node_page;
629 	nid_t nid;
630 	unsigned int ofs_in_node;
631 	block_t source_blkaddr;
632 
633 	nid = le32_to_cpu(sum->nid);
634 	ofs_in_node = le16_to_cpu(sum->ofs_in_node);
635 
636 	node_page = f2fs_get_node_page(sbi, nid);
637 	if (IS_ERR(node_page))
638 		return false;
639 
640 	if (f2fs_get_node_info(sbi, nid, dni)) {
641 		f2fs_put_page(node_page, 1);
642 		return false;
643 	}
644 
645 	if (sum->version != dni->version) {
646 		f2fs_warn(sbi, "%s: valid data with mismatched node version.",
647 			  __func__);
648 		set_sbi_flag(sbi, SBI_NEED_FSCK);
649 	}
650 
651 	*nofs = ofs_of_node(node_page);
652 	source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
653 	f2fs_put_page(node_page, 1);
654 
655 	if (source_blkaddr != blkaddr) {
656 #ifdef CONFIG_F2FS_CHECK_FS
657 		unsigned int segno = GET_SEGNO(sbi, blkaddr);
658 		unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
659 
660 		if (unlikely(check_valid_map(sbi, segno, offset))) {
661 			if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
662 				f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u\n",
663 						blkaddr, source_blkaddr, segno);
664 				f2fs_bug_on(sbi, 1);
665 			}
666 		}
667 #endif
668 		return false;
669 	}
670 	return true;
671 }
672 
673 static int ra_data_block(struct inode *inode, pgoff_t index)
674 {
675 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
676 	struct address_space *mapping = inode->i_mapping;
677 	struct dnode_of_data dn;
678 	struct page *page;
679 	struct extent_info ei = {0, 0, 0};
680 	struct f2fs_io_info fio = {
681 		.sbi = sbi,
682 		.ino = inode->i_ino,
683 		.type = DATA,
684 		.temp = COLD,
685 		.op = REQ_OP_READ,
686 		.op_flags = 0,
687 		.encrypted_page = NULL,
688 		.in_list = false,
689 		.retry = false,
690 	};
691 	int err;
692 
693 	page = f2fs_grab_cache_page(mapping, index, true);
694 	if (!page)
695 		return -ENOMEM;
696 
697 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
698 		dn.data_blkaddr = ei.blk + index - ei.fofs;
699 		if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
700 						DATA_GENERIC_ENHANCE_READ))) {
701 			err = -EFSCORRUPTED;
702 			goto put_page;
703 		}
704 		goto got_it;
705 	}
706 
707 	set_new_dnode(&dn, inode, NULL, NULL, 0);
708 	err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
709 	if (err)
710 		goto put_page;
711 	f2fs_put_dnode(&dn);
712 
713 	if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
714 		err = -ENOENT;
715 		goto put_page;
716 	}
717 	if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
718 						DATA_GENERIC_ENHANCE))) {
719 		err = -EFSCORRUPTED;
720 		goto put_page;
721 	}
722 got_it:
723 	/* read page */
724 	fio.page = page;
725 	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
726 
727 	/*
728 	 * don't cache encrypted data into meta inode until previous dirty
729 	 * data were writebacked to avoid racing between GC and flush.
730 	 */
731 	f2fs_wait_on_page_writeback(page, DATA, true, true);
732 
733 	f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
734 
735 	fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
736 					dn.data_blkaddr,
737 					FGP_LOCK | FGP_CREAT, GFP_NOFS);
738 	if (!fio.encrypted_page) {
739 		err = -ENOMEM;
740 		goto put_page;
741 	}
742 
743 	err = f2fs_submit_page_bio(&fio);
744 	if (err)
745 		goto put_encrypted_page;
746 	f2fs_put_page(fio.encrypted_page, 0);
747 	f2fs_put_page(page, 1);
748 
749 	f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
750 	f2fs_update_iostat(sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
751 
752 	return 0;
753 put_encrypted_page:
754 	f2fs_put_page(fio.encrypted_page, 1);
755 put_page:
756 	f2fs_put_page(page, 1);
757 	return err;
758 }
759 
760 /*
761  * Move data block via META_MAPPING while keeping locked data page.
762  * This can be used to move blocks, aka LBAs, directly on disk.
763  */
764 static int move_data_block(struct inode *inode, block_t bidx,
765 				int gc_type, unsigned int segno, int off)
766 {
767 	struct f2fs_io_info fio = {
768 		.sbi = F2FS_I_SB(inode),
769 		.ino = inode->i_ino,
770 		.type = DATA,
771 		.temp = COLD,
772 		.op = REQ_OP_READ,
773 		.op_flags = 0,
774 		.encrypted_page = NULL,
775 		.in_list = false,
776 		.retry = false,
777 	};
778 	struct dnode_of_data dn;
779 	struct f2fs_summary sum;
780 	struct node_info ni;
781 	struct page *page, *mpage;
782 	block_t newaddr;
783 	int err = 0;
784 	bool lfs_mode = f2fs_lfs_mode(fio.sbi);
785 
786 	/* do not read out */
787 	page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
788 	if (!page)
789 		return -ENOMEM;
790 
791 	if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
792 		err = -ENOENT;
793 		goto out;
794 	}
795 
796 	if (f2fs_is_atomic_file(inode)) {
797 		F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
798 		F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
799 		err = -EAGAIN;
800 		goto out;
801 	}
802 
803 	if (f2fs_is_pinned_file(inode)) {
804 		f2fs_pin_file_control(inode, true);
805 		err = -EAGAIN;
806 		goto out;
807 	}
808 
809 	set_new_dnode(&dn, inode, NULL, NULL, 0);
810 	err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
811 	if (err)
812 		goto out;
813 
814 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
815 		ClearPageUptodate(page);
816 		err = -ENOENT;
817 		goto put_out;
818 	}
819 
820 	/*
821 	 * don't cache encrypted data into meta inode until previous dirty
822 	 * data were writebacked to avoid racing between GC and flush.
823 	 */
824 	f2fs_wait_on_page_writeback(page, DATA, true, true);
825 
826 	f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
827 
828 	err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
829 	if (err)
830 		goto put_out;
831 
832 	set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
833 
834 	/* read page */
835 	fio.page = page;
836 	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
837 
838 	if (lfs_mode)
839 		down_write(&fio.sbi->io_order_lock);
840 
841 	mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
842 					fio.old_blkaddr, false);
843 	if (!mpage)
844 		goto up_out;
845 
846 	fio.encrypted_page = mpage;
847 
848 	/* read source block in mpage */
849 	if (!PageUptodate(mpage)) {
850 		err = f2fs_submit_page_bio(&fio);
851 		if (err) {
852 			f2fs_put_page(mpage, 1);
853 			goto up_out;
854 		}
855 
856 		f2fs_update_iostat(fio.sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
857 		f2fs_update_iostat(fio.sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
858 
859 		lock_page(mpage);
860 		if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
861 						!PageUptodate(mpage))) {
862 			err = -EIO;
863 			f2fs_put_page(mpage, 1);
864 			goto up_out;
865 		}
866 	}
867 
868 	f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
869 					&sum, CURSEG_COLD_DATA, NULL);
870 
871 	fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
872 				newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
873 	if (!fio.encrypted_page) {
874 		err = -ENOMEM;
875 		f2fs_put_page(mpage, 1);
876 		goto recover_block;
877 	}
878 
879 	/* write target block */
880 	f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
881 	memcpy(page_address(fio.encrypted_page),
882 				page_address(mpage), PAGE_SIZE);
883 	f2fs_put_page(mpage, 1);
884 	invalidate_mapping_pages(META_MAPPING(fio.sbi),
885 				fio.old_blkaddr, fio.old_blkaddr);
886 
887 	set_page_dirty(fio.encrypted_page);
888 	if (clear_page_dirty_for_io(fio.encrypted_page))
889 		dec_page_count(fio.sbi, F2FS_DIRTY_META);
890 
891 	set_page_writeback(fio.encrypted_page);
892 	ClearPageError(page);
893 
894 	/* allocate block address */
895 	f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
896 
897 	fio.op = REQ_OP_WRITE;
898 	fio.op_flags = REQ_SYNC;
899 	fio.new_blkaddr = newaddr;
900 	f2fs_submit_page_write(&fio);
901 	if (fio.retry) {
902 		err = -EAGAIN;
903 		if (PageWriteback(fio.encrypted_page))
904 			end_page_writeback(fio.encrypted_page);
905 		goto put_page_out;
906 	}
907 
908 	f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
909 
910 	f2fs_update_data_blkaddr(&dn, newaddr);
911 	set_inode_flag(inode, FI_APPEND_WRITE);
912 	if (page->index == 0)
913 		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
914 put_page_out:
915 	f2fs_put_page(fio.encrypted_page, 1);
916 recover_block:
917 	if (err)
918 		f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
919 								true, true);
920 up_out:
921 	if (lfs_mode)
922 		up_write(&fio.sbi->io_order_lock);
923 put_out:
924 	f2fs_put_dnode(&dn);
925 out:
926 	f2fs_put_page(page, 1);
927 	return err;
928 }
929 
930 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
931 							unsigned int segno, int off)
932 {
933 	struct page *page;
934 	int err = 0;
935 
936 	page = f2fs_get_lock_data_page(inode, bidx, true);
937 	if (IS_ERR(page))
938 		return PTR_ERR(page);
939 
940 	if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
941 		err = -ENOENT;
942 		goto out;
943 	}
944 
945 	if (f2fs_is_atomic_file(inode)) {
946 		F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
947 		F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
948 		err = -EAGAIN;
949 		goto out;
950 	}
951 	if (f2fs_is_pinned_file(inode)) {
952 		if (gc_type == FG_GC)
953 			f2fs_pin_file_control(inode, true);
954 		err = -EAGAIN;
955 		goto out;
956 	}
957 
958 	if (gc_type == BG_GC) {
959 		if (PageWriteback(page)) {
960 			err = -EAGAIN;
961 			goto out;
962 		}
963 		set_page_dirty(page);
964 		set_cold_data(page);
965 	} else {
966 		struct f2fs_io_info fio = {
967 			.sbi = F2FS_I_SB(inode),
968 			.ino = inode->i_ino,
969 			.type = DATA,
970 			.temp = COLD,
971 			.op = REQ_OP_WRITE,
972 			.op_flags = REQ_SYNC,
973 			.old_blkaddr = NULL_ADDR,
974 			.page = page,
975 			.encrypted_page = NULL,
976 			.need_lock = LOCK_REQ,
977 			.io_type = FS_GC_DATA_IO,
978 		};
979 		bool is_dirty = PageDirty(page);
980 
981 retry:
982 		f2fs_wait_on_page_writeback(page, DATA, true, true);
983 
984 		set_page_dirty(page);
985 		if (clear_page_dirty_for_io(page)) {
986 			inode_dec_dirty_pages(inode);
987 			f2fs_remove_dirty_inode(inode);
988 		}
989 
990 		set_cold_data(page);
991 
992 		err = f2fs_do_write_data_page(&fio);
993 		if (err) {
994 			clear_cold_data(page);
995 			if (err == -ENOMEM) {
996 				congestion_wait(BLK_RW_ASYNC,
997 						DEFAULT_IO_TIMEOUT);
998 				goto retry;
999 			}
1000 			if (is_dirty)
1001 				set_page_dirty(page);
1002 		}
1003 	}
1004 out:
1005 	f2fs_put_page(page, 1);
1006 	return err;
1007 }
1008 
1009 /*
1010  * This function tries to get parent node of victim data block, and identifies
1011  * data block validity. If the block is valid, copy that with cold status and
1012  * modify parent node.
1013  * If the parent node is not valid or the data block address is different,
1014  * the victim data block is ignored.
1015  */
1016 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1017 		struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
1018 {
1019 	struct super_block *sb = sbi->sb;
1020 	struct f2fs_summary *entry;
1021 	block_t start_addr;
1022 	int off;
1023 	int phase = 0;
1024 	int submitted = 0;
1025 
1026 	start_addr = START_BLOCK(sbi, segno);
1027 
1028 next_step:
1029 	entry = sum;
1030 
1031 	for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
1032 		struct page *data_page;
1033 		struct inode *inode;
1034 		struct node_info dni; /* dnode info for the data */
1035 		unsigned int ofs_in_node, nofs;
1036 		block_t start_bidx;
1037 		nid_t nid = le32_to_cpu(entry->nid);
1038 
1039 		/*
1040 		 * stop BG_GC if there is not enough free sections.
1041 		 * Or, stop GC if the segment becomes fully valid caused by
1042 		 * race condition along with SSR block allocation.
1043 		 */
1044 		if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1045 				get_valid_blocks(sbi, segno, true) ==
1046 							BLKS_PER_SEC(sbi))
1047 			return submitted;
1048 
1049 		if (check_valid_map(sbi, segno, off) == 0)
1050 			continue;
1051 
1052 		if (phase == 0) {
1053 			f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1054 							META_NAT, true);
1055 			continue;
1056 		}
1057 
1058 		if (phase == 1) {
1059 			f2fs_ra_node_page(sbi, nid);
1060 			continue;
1061 		}
1062 
1063 		/* Get an inode by ino with checking validity */
1064 		if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1065 			continue;
1066 
1067 		if (phase == 2) {
1068 			f2fs_ra_node_page(sbi, dni.ino);
1069 			continue;
1070 		}
1071 
1072 		ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1073 
1074 		if (phase == 3) {
1075 			inode = f2fs_iget(sb, dni.ino);
1076 			if (IS_ERR(inode) || is_bad_inode(inode)) {
1077 				set_sbi_flag(sbi, SBI_NEED_FSCK);
1078 				continue;
1079 			}
1080 
1081 			if (!down_write_trylock(
1082 				&F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1083 				iput(inode);
1084 				sbi->skipped_gc_rwsem++;
1085 				continue;
1086 			}
1087 
1088 			start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1089 								ofs_in_node;
1090 
1091 			if (f2fs_post_read_required(inode)) {
1092 				int err = ra_data_block(inode, start_bidx);
1093 
1094 				up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1095 				if (err) {
1096 					iput(inode);
1097 					continue;
1098 				}
1099 				add_gc_inode(gc_list, inode);
1100 				continue;
1101 			}
1102 
1103 			data_page = f2fs_get_read_data_page(inode,
1104 						start_bidx, REQ_RAHEAD, true);
1105 			up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1106 			if (IS_ERR(data_page)) {
1107 				iput(inode);
1108 				continue;
1109 			}
1110 
1111 			f2fs_put_page(data_page, 0);
1112 			add_gc_inode(gc_list, inode);
1113 			continue;
1114 		}
1115 
1116 		/* phase 4 */
1117 		inode = find_gc_inode(gc_list, dni.ino);
1118 		if (inode) {
1119 			struct f2fs_inode_info *fi = F2FS_I(inode);
1120 			bool locked = false;
1121 			int err;
1122 
1123 			if (S_ISREG(inode->i_mode)) {
1124 				if (!down_write_trylock(&fi->i_gc_rwsem[READ]))
1125 					continue;
1126 				if (!down_write_trylock(
1127 						&fi->i_gc_rwsem[WRITE])) {
1128 					sbi->skipped_gc_rwsem++;
1129 					up_write(&fi->i_gc_rwsem[READ]);
1130 					continue;
1131 				}
1132 				locked = true;
1133 
1134 				/* wait for all inflight aio data */
1135 				inode_dio_wait(inode);
1136 			}
1137 
1138 			start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1139 								+ ofs_in_node;
1140 			if (f2fs_post_read_required(inode))
1141 				err = move_data_block(inode, start_bidx,
1142 							gc_type, segno, off);
1143 			else
1144 				err = move_data_page(inode, start_bidx, gc_type,
1145 								segno, off);
1146 
1147 			if (!err && (gc_type == FG_GC ||
1148 					f2fs_post_read_required(inode)))
1149 				submitted++;
1150 
1151 			if (locked) {
1152 				up_write(&fi->i_gc_rwsem[WRITE]);
1153 				up_write(&fi->i_gc_rwsem[READ]);
1154 			}
1155 
1156 			stat_inc_data_blk_count(sbi, 1, gc_type);
1157 		}
1158 	}
1159 
1160 	if (++phase < 5)
1161 		goto next_step;
1162 
1163 	return submitted;
1164 }
1165 
1166 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1167 			int gc_type)
1168 {
1169 	struct sit_info *sit_i = SIT_I(sbi);
1170 	int ret;
1171 
1172 	down_write(&sit_i->sentry_lock);
1173 	ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1174 					      NO_CHECK_TYPE, LFS);
1175 	up_write(&sit_i->sentry_lock);
1176 	return ret;
1177 }
1178 
1179 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1180 				unsigned int start_segno,
1181 				struct gc_inode_list *gc_list, int gc_type)
1182 {
1183 	struct page *sum_page;
1184 	struct f2fs_summary_block *sum;
1185 	struct blk_plug plug;
1186 	unsigned int segno = start_segno;
1187 	unsigned int end_segno = start_segno + sbi->segs_per_sec;
1188 	int seg_freed = 0, migrated = 0;
1189 	unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1190 						SUM_TYPE_DATA : SUM_TYPE_NODE;
1191 	int submitted = 0;
1192 
1193 	if (__is_large_section(sbi))
1194 		end_segno = rounddown(end_segno, sbi->segs_per_sec);
1195 
1196 	/* readahead multi ssa blocks those have contiguous address */
1197 	if (__is_large_section(sbi))
1198 		f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1199 					end_segno - segno, META_SSA, true);
1200 
1201 	/* reference all summary page */
1202 	while (segno < end_segno) {
1203 		sum_page = f2fs_get_sum_page(sbi, segno++);
1204 		if (IS_ERR(sum_page)) {
1205 			int err = PTR_ERR(sum_page);
1206 
1207 			end_segno = segno - 1;
1208 			for (segno = start_segno; segno < end_segno; segno++) {
1209 				sum_page = find_get_page(META_MAPPING(sbi),
1210 						GET_SUM_BLOCK(sbi, segno));
1211 				f2fs_put_page(sum_page, 0);
1212 				f2fs_put_page(sum_page, 0);
1213 			}
1214 			return err;
1215 		}
1216 		unlock_page(sum_page);
1217 	}
1218 
1219 	blk_start_plug(&plug);
1220 
1221 	for (segno = start_segno; segno < end_segno; segno++) {
1222 
1223 		/* find segment summary of victim */
1224 		sum_page = find_get_page(META_MAPPING(sbi),
1225 					GET_SUM_BLOCK(sbi, segno));
1226 		f2fs_put_page(sum_page, 0);
1227 
1228 		if (get_valid_blocks(sbi, segno, false) == 0)
1229 			goto freed;
1230 		if (gc_type == BG_GC && __is_large_section(sbi) &&
1231 				migrated >= sbi->migration_granularity)
1232 			goto skip;
1233 		if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1234 			goto skip;
1235 
1236 		sum = page_address(sum_page);
1237 		if (type != GET_SUM_TYPE((&sum->footer))) {
1238 			f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1239 				 segno, type, GET_SUM_TYPE((&sum->footer)));
1240 			set_sbi_flag(sbi, SBI_NEED_FSCK);
1241 			f2fs_stop_checkpoint(sbi, false);
1242 			goto skip;
1243 		}
1244 
1245 		/*
1246 		 * this is to avoid deadlock:
1247 		 * - lock_page(sum_page)         - f2fs_replace_block
1248 		 *  - check_valid_map()            - down_write(sentry_lock)
1249 		 *   - down_read(sentry_lock)     - change_curseg()
1250 		 *                                  - lock_page(sum_page)
1251 		 */
1252 		if (type == SUM_TYPE_NODE)
1253 			submitted += gc_node_segment(sbi, sum->entries, segno,
1254 								gc_type);
1255 		else
1256 			submitted += gc_data_segment(sbi, sum->entries, gc_list,
1257 							segno, gc_type);
1258 
1259 		stat_inc_seg_count(sbi, type, gc_type);
1260 		migrated++;
1261 
1262 freed:
1263 		if (gc_type == FG_GC &&
1264 				get_valid_blocks(sbi, segno, false) == 0)
1265 			seg_freed++;
1266 
1267 		if (__is_large_section(sbi) && segno + 1 < end_segno)
1268 			sbi->next_victim_seg[gc_type] = segno + 1;
1269 skip:
1270 		f2fs_put_page(sum_page, 0);
1271 	}
1272 
1273 	if (submitted)
1274 		f2fs_submit_merged_write(sbi,
1275 				(type == SUM_TYPE_NODE) ? NODE : DATA);
1276 
1277 	blk_finish_plug(&plug);
1278 
1279 	stat_inc_call_count(sbi->stat_info);
1280 
1281 	return seg_freed;
1282 }
1283 
1284 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1285 			bool background, unsigned int segno)
1286 {
1287 	int gc_type = sync ? FG_GC : BG_GC;
1288 	int sec_freed = 0, seg_freed = 0, total_freed = 0;
1289 	int ret = 0;
1290 	struct cp_control cpc;
1291 	unsigned int init_segno = segno;
1292 	struct gc_inode_list gc_list = {
1293 		.ilist = LIST_HEAD_INIT(gc_list.ilist),
1294 		.iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1295 	};
1296 	unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1297 	unsigned long long first_skipped;
1298 	unsigned int skipped_round = 0, round = 0;
1299 
1300 	trace_f2fs_gc_begin(sbi->sb, sync, background,
1301 				get_pages(sbi, F2FS_DIRTY_NODES),
1302 				get_pages(sbi, F2FS_DIRTY_DENTS),
1303 				get_pages(sbi, F2FS_DIRTY_IMETA),
1304 				free_sections(sbi),
1305 				free_segments(sbi),
1306 				reserved_segments(sbi),
1307 				prefree_segments(sbi));
1308 
1309 	cpc.reason = __get_cp_reason(sbi);
1310 	sbi->skipped_gc_rwsem = 0;
1311 	first_skipped = last_skipped;
1312 gc_more:
1313 	if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1314 		ret = -EINVAL;
1315 		goto stop;
1316 	}
1317 	if (unlikely(f2fs_cp_error(sbi))) {
1318 		ret = -EIO;
1319 		goto stop;
1320 	}
1321 
1322 	if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1323 		/*
1324 		 * For example, if there are many prefree_segments below given
1325 		 * threshold, we can make them free by checkpoint. Then, we
1326 		 * secure free segments which doesn't need fggc any more.
1327 		 */
1328 		if (prefree_segments(sbi) &&
1329 				!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1330 			ret = f2fs_write_checkpoint(sbi, &cpc);
1331 			if (ret)
1332 				goto stop;
1333 		}
1334 		if (has_not_enough_free_secs(sbi, 0, 0))
1335 			gc_type = FG_GC;
1336 	}
1337 
1338 	/* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1339 	if (gc_type == BG_GC && !background) {
1340 		ret = -EINVAL;
1341 		goto stop;
1342 	}
1343 	if (!__get_victim(sbi, &segno, gc_type)) {
1344 		ret = -ENODATA;
1345 		goto stop;
1346 	}
1347 
1348 	seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type);
1349 	if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec)
1350 		sec_freed++;
1351 	total_freed += seg_freed;
1352 
1353 	if (gc_type == FG_GC) {
1354 		if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
1355 						sbi->skipped_gc_rwsem)
1356 			skipped_round++;
1357 		last_skipped = sbi->skipped_atomic_files[FG_GC];
1358 		round++;
1359 	}
1360 
1361 	if (gc_type == FG_GC && seg_freed)
1362 		sbi->cur_victim_sec = NULL_SEGNO;
1363 
1364 	if (sync)
1365 		goto stop;
1366 
1367 	if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1368 		if (skipped_round <= MAX_SKIP_GC_COUNT ||
1369 					skipped_round * 2 < round) {
1370 			segno = NULL_SEGNO;
1371 			goto gc_more;
1372 		}
1373 
1374 		if (first_skipped < last_skipped &&
1375 				(last_skipped - first_skipped) >
1376 						sbi->skipped_gc_rwsem) {
1377 			f2fs_drop_inmem_pages_all(sbi, true);
1378 			segno = NULL_SEGNO;
1379 			goto gc_more;
1380 		}
1381 		if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1382 			ret = f2fs_write_checkpoint(sbi, &cpc);
1383 	}
1384 stop:
1385 	SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1386 	SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1387 
1388 	trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1389 				get_pages(sbi, F2FS_DIRTY_NODES),
1390 				get_pages(sbi, F2FS_DIRTY_DENTS),
1391 				get_pages(sbi, F2FS_DIRTY_IMETA),
1392 				free_sections(sbi),
1393 				free_segments(sbi),
1394 				reserved_segments(sbi),
1395 				prefree_segments(sbi));
1396 
1397 	up_write(&sbi->gc_lock);
1398 
1399 	put_gc_inode(&gc_list);
1400 
1401 	if (sync && !ret)
1402 		ret = sec_freed ? 0 : -EAGAIN;
1403 	return ret;
1404 }
1405 
1406 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1407 {
1408 	DIRTY_I(sbi)->v_ops = &default_v_ops;
1409 
1410 	sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1411 
1412 	/* give warm/cold data area from slower device */
1413 	if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1414 		SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1415 				GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1416 }
1417 
1418 static int free_segment_range(struct f2fs_sb_info *sbi,
1419 				unsigned int secs, bool gc_only)
1420 {
1421 	unsigned int segno, next_inuse, start, end;
1422 	struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1423 	int gc_mode, gc_type;
1424 	int err = 0;
1425 	int type;
1426 
1427 	/* Force block allocation for GC */
1428 	MAIN_SECS(sbi) -= secs;
1429 	start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1430 	end = MAIN_SEGS(sbi) - 1;
1431 
1432 	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1433 	for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1434 		if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1435 			SIT_I(sbi)->last_victim[gc_mode] = 0;
1436 
1437 	for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1438 		if (sbi->next_victim_seg[gc_type] >= start)
1439 			sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1440 	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1441 
1442 	/* Move out cursegs from the target range */
1443 	for (type = CURSEG_HOT_DATA; type < NR_CURSEG_TYPE; type++)
1444 		f2fs_allocate_segment_for_resize(sbi, type, start, end);
1445 
1446 	/* do GC to move out valid blocks in the range */
1447 	for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1448 		struct gc_inode_list gc_list = {
1449 			.ilist = LIST_HEAD_INIT(gc_list.ilist),
1450 			.iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1451 		};
1452 
1453 		do_garbage_collect(sbi, segno, &gc_list, FG_GC);
1454 		put_gc_inode(&gc_list);
1455 
1456 		if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1457 			err = -EAGAIN;
1458 			goto out;
1459 		}
1460 		if (fatal_signal_pending(current)) {
1461 			err = -ERESTARTSYS;
1462 			goto out;
1463 		}
1464 	}
1465 	if (gc_only)
1466 		goto out;
1467 
1468 	err = f2fs_write_checkpoint(sbi, &cpc);
1469 	if (err)
1470 		goto out;
1471 
1472 	next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1473 	if (next_inuse <= end) {
1474 		f2fs_err(sbi, "segno %u should be free but still inuse!",
1475 			 next_inuse);
1476 		f2fs_bug_on(sbi, 1);
1477 	}
1478 out:
1479 	MAIN_SECS(sbi) += secs;
1480 	return err;
1481 }
1482 
1483 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1484 {
1485 	struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1486 	int section_count;
1487 	int segment_count;
1488 	int segment_count_main;
1489 	long long block_count;
1490 	int segs = secs * sbi->segs_per_sec;
1491 
1492 	down_write(&sbi->sb_lock);
1493 
1494 	section_count = le32_to_cpu(raw_sb->section_count);
1495 	segment_count = le32_to_cpu(raw_sb->segment_count);
1496 	segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
1497 	block_count = le64_to_cpu(raw_sb->block_count);
1498 
1499 	raw_sb->section_count = cpu_to_le32(section_count + secs);
1500 	raw_sb->segment_count = cpu_to_le32(segment_count + segs);
1501 	raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
1502 	raw_sb->block_count = cpu_to_le64(block_count +
1503 					(long long)segs * sbi->blocks_per_seg);
1504 	if (f2fs_is_multi_device(sbi)) {
1505 		int last_dev = sbi->s_ndevs - 1;
1506 		int dev_segs =
1507 			le32_to_cpu(raw_sb->devs[last_dev].total_segments);
1508 
1509 		raw_sb->devs[last_dev].total_segments =
1510 						cpu_to_le32(dev_segs + segs);
1511 	}
1512 
1513 	up_write(&sbi->sb_lock);
1514 }
1515 
1516 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
1517 {
1518 	int segs = secs * sbi->segs_per_sec;
1519 	long long blks = (long long)segs * sbi->blocks_per_seg;
1520 	long long user_block_count =
1521 				le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
1522 
1523 	SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
1524 	MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
1525 	MAIN_SECS(sbi) += secs;
1526 	FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
1527 	FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
1528 	F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
1529 
1530 	if (f2fs_is_multi_device(sbi)) {
1531 		int last_dev = sbi->s_ndevs - 1;
1532 
1533 		FDEV(last_dev).total_segments =
1534 				(int)FDEV(last_dev).total_segments + segs;
1535 		FDEV(last_dev).end_blk =
1536 				(long long)FDEV(last_dev).end_blk + blks;
1537 #ifdef CONFIG_BLK_DEV_ZONED
1538 		FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
1539 					(int)(blks >> sbi->log_blocks_per_blkz);
1540 #endif
1541 	}
1542 }
1543 
1544 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
1545 {
1546 	__u64 old_block_count, shrunk_blocks;
1547 	struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1548 	unsigned int secs;
1549 	int err = 0;
1550 	__u32 rem;
1551 
1552 	old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
1553 	if (block_count > old_block_count)
1554 		return -EINVAL;
1555 
1556 	if (f2fs_is_multi_device(sbi)) {
1557 		int last_dev = sbi->s_ndevs - 1;
1558 		__u64 last_segs = FDEV(last_dev).total_segments;
1559 
1560 		if (block_count + last_segs * sbi->blocks_per_seg <=
1561 								old_block_count)
1562 			return -EINVAL;
1563 	}
1564 
1565 	/* new fs size should align to section size */
1566 	div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
1567 	if (rem)
1568 		return -EINVAL;
1569 
1570 	if (block_count == old_block_count)
1571 		return 0;
1572 
1573 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1574 		f2fs_err(sbi, "Should run fsck to repair first.");
1575 		return -EFSCORRUPTED;
1576 	}
1577 
1578 	if (test_opt(sbi, DISABLE_CHECKPOINT)) {
1579 		f2fs_err(sbi, "Checkpoint should be enabled.");
1580 		return -EINVAL;
1581 	}
1582 
1583 	shrunk_blocks = old_block_count - block_count;
1584 	secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
1585 
1586 	/* stop other GC */
1587 	if (!down_write_trylock(&sbi->gc_lock))
1588 		return -EAGAIN;
1589 
1590 	/* stop CP to protect MAIN_SEC in free_segment_range */
1591 	f2fs_lock_op(sbi);
1592 	err = free_segment_range(sbi, secs, true);
1593 	f2fs_unlock_op(sbi);
1594 	up_write(&sbi->gc_lock);
1595 	if (err)
1596 		return err;
1597 
1598 	set_sbi_flag(sbi, SBI_IS_RESIZEFS);
1599 
1600 	freeze_super(sbi->sb);
1601 	down_write(&sbi->gc_lock);
1602 	mutex_lock(&sbi->cp_mutex);
1603 
1604 	spin_lock(&sbi->stat_lock);
1605 	if (shrunk_blocks + valid_user_blocks(sbi) +
1606 		sbi->current_reserved_blocks + sbi->unusable_block_count +
1607 		F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
1608 		err = -ENOSPC;
1609 	else
1610 		sbi->user_block_count -= shrunk_blocks;
1611 	spin_unlock(&sbi->stat_lock);
1612 	if (err)
1613 		goto out_err;
1614 
1615 	err = free_segment_range(sbi, secs, false);
1616 	if (err)
1617 		goto recover_out;
1618 
1619 	update_sb_metadata(sbi, -secs);
1620 
1621 	err = f2fs_commit_super(sbi, false);
1622 	if (err) {
1623 		update_sb_metadata(sbi, secs);
1624 		goto recover_out;
1625 	}
1626 
1627 	update_fs_metadata(sbi, -secs);
1628 	clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
1629 	set_sbi_flag(sbi, SBI_IS_DIRTY);
1630 
1631 	err = f2fs_write_checkpoint(sbi, &cpc);
1632 	if (err) {
1633 		update_fs_metadata(sbi, secs);
1634 		update_sb_metadata(sbi, secs);
1635 		f2fs_commit_super(sbi, false);
1636 	}
1637 recover_out:
1638 	if (err) {
1639 		set_sbi_flag(sbi, SBI_NEED_FSCK);
1640 		f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
1641 
1642 		spin_lock(&sbi->stat_lock);
1643 		sbi->user_block_count += shrunk_blocks;
1644 		spin_unlock(&sbi->stat_lock);
1645 	}
1646 out_err:
1647 	mutex_unlock(&sbi->cp_mutex);
1648 	up_write(&sbi->gc_lock);
1649 	thaw_super(sbi->sb);
1650 	clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
1651 	return err;
1652 }
1653