xref: /openbmc/linux/fs/f2fs/gc.c (revision 4cff79e9)
1 /*
2  * fs/f2fs/gc.c
3  *
4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5  *             http://www.samsung.com/
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #include <linux/fs.h>
12 #include <linux/module.h>
13 #include <linux/backing-dev.h>
14 #include <linux/init.h>
15 #include <linux/f2fs_fs.h>
16 #include <linux/kthread.h>
17 #include <linux/delay.h>
18 #include <linux/freezer.h>
19 
20 #include "f2fs.h"
21 #include "node.h"
22 #include "segment.h"
23 #include "gc.h"
24 #include <trace/events/f2fs.h>
25 
26 static int gc_thread_func(void *data)
27 {
28 	struct f2fs_sb_info *sbi = data;
29 	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
30 	wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
31 	unsigned int wait_ms;
32 
33 	wait_ms = gc_th->min_sleep_time;
34 
35 	set_freezable();
36 	do {
37 		wait_event_interruptible_timeout(*wq,
38 				kthread_should_stop() || freezing(current) ||
39 				gc_th->gc_wake,
40 				msecs_to_jiffies(wait_ms));
41 
42 		/* give it a try one time */
43 		if (gc_th->gc_wake)
44 			gc_th->gc_wake = 0;
45 
46 		if (try_to_freeze())
47 			continue;
48 		if (kthread_should_stop())
49 			break;
50 
51 		if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
52 			increase_sleep_time(gc_th, &wait_ms);
53 			continue;
54 		}
55 
56 #ifdef CONFIG_F2FS_FAULT_INJECTION
57 		if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
58 			f2fs_show_injection_info(FAULT_CHECKPOINT);
59 			f2fs_stop_checkpoint(sbi, false);
60 		}
61 #endif
62 
63 		if (!sb_start_write_trylock(sbi->sb))
64 			continue;
65 
66 		/*
67 		 * [GC triggering condition]
68 		 * 0. GC is not conducted currently.
69 		 * 1. There are enough dirty segments.
70 		 * 2. IO subsystem is idle by checking the # of writeback pages.
71 		 * 3. IO subsystem is idle by checking the # of requests in
72 		 *    bdev's request list.
73 		 *
74 		 * Note) We have to avoid triggering GCs frequently.
75 		 * Because it is possible that some segments can be
76 		 * invalidated soon after by user update or deletion.
77 		 * So, I'd like to wait some time to collect dirty segments.
78 		 */
79 		if (gc_th->gc_urgent) {
80 			wait_ms = gc_th->urgent_sleep_time;
81 			mutex_lock(&sbi->gc_mutex);
82 			goto do_gc;
83 		}
84 
85 		if (!mutex_trylock(&sbi->gc_mutex))
86 			goto next;
87 
88 		if (!is_idle(sbi)) {
89 			increase_sleep_time(gc_th, &wait_ms);
90 			mutex_unlock(&sbi->gc_mutex);
91 			goto next;
92 		}
93 
94 		if (has_enough_invalid_blocks(sbi))
95 			decrease_sleep_time(gc_th, &wait_ms);
96 		else
97 			increase_sleep_time(gc_th, &wait_ms);
98 do_gc:
99 		stat_inc_bggc_count(sbi);
100 
101 		/* if return value is not zero, no victim was selected */
102 		if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true, NULL_SEGNO))
103 			wait_ms = gc_th->no_gc_sleep_time;
104 
105 		trace_f2fs_background_gc(sbi->sb, wait_ms,
106 				prefree_segments(sbi), free_segments(sbi));
107 
108 		/* balancing f2fs's metadata periodically */
109 		f2fs_balance_fs_bg(sbi);
110 next:
111 		sb_end_write(sbi->sb);
112 
113 	} while (!kthread_should_stop());
114 	return 0;
115 }
116 
117 int start_gc_thread(struct f2fs_sb_info *sbi)
118 {
119 	struct f2fs_gc_kthread *gc_th;
120 	dev_t dev = sbi->sb->s_bdev->bd_dev;
121 	int err = 0;
122 
123 	gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
124 	if (!gc_th) {
125 		err = -ENOMEM;
126 		goto out;
127 	}
128 
129 	gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
130 	gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
131 	gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
132 	gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
133 
134 	gc_th->gc_idle = 0;
135 	gc_th->gc_urgent = 0;
136 	gc_th->gc_wake= 0;
137 
138 	sbi->gc_thread = gc_th;
139 	init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
140 	sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
141 			"f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
142 	if (IS_ERR(gc_th->f2fs_gc_task)) {
143 		err = PTR_ERR(gc_th->f2fs_gc_task);
144 		kfree(gc_th);
145 		sbi->gc_thread = NULL;
146 	}
147 out:
148 	return err;
149 }
150 
151 void stop_gc_thread(struct f2fs_sb_info *sbi)
152 {
153 	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
154 	if (!gc_th)
155 		return;
156 	kthread_stop(gc_th->f2fs_gc_task);
157 	kfree(gc_th);
158 	sbi->gc_thread = NULL;
159 }
160 
161 static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
162 {
163 	int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
164 
165 	if (!gc_th)
166 		return gc_mode;
167 
168 	if (gc_th->gc_idle) {
169 		if (gc_th->gc_idle == 1)
170 			gc_mode = GC_CB;
171 		else if (gc_th->gc_idle == 2)
172 			gc_mode = GC_GREEDY;
173 	}
174 	if (gc_th->gc_urgent)
175 		gc_mode = GC_GREEDY;
176 	return gc_mode;
177 }
178 
179 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
180 			int type, struct victim_sel_policy *p)
181 {
182 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
183 
184 	if (p->alloc_mode == SSR) {
185 		p->gc_mode = GC_GREEDY;
186 		p->dirty_segmap = dirty_i->dirty_segmap[type];
187 		p->max_search = dirty_i->nr_dirty[type];
188 		p->ofs_unit = 1;
189 	} else {
190 		p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
191 		p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
192 		p->max_search = dirty_i->nr_dirty[DIRTY];
193 		p->ofs_unit = sbi->segs_per_sec;
194 	}
195 
196 	/* we need to check every dirty segments in the FG_GC case */
197 	if (gc_type != FG_GC &&
198 			(sbi->gc_thread && !sbi->gc_thread->gc_urgent) &&
199 			p->max_search > sbi->max_victim_search)
200 		p->max_search = sbi->max_victim_search;
201 
202 	/* let's select beginning hot/small space first in no_heap mode*/
203 	if (test_opt(sbi, NOHEAP) &&
204 		(type == CURSEG_HOT_DATA || IS_NODESEG(type)))
205 		p->offset = 0;
206 	else
207 		p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
208 }
209 
210 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
211 				struct victim_sel_policy *p)
212 {
213 	/* SSR allocates in a segment unit */
214 	if (p->alloc_mode == SSR)
215 		return sbi->blocks_per_seg;
216 	if (p->gc_mode == GC_GREEDY)
217 		return 2 * sbi->blocks_per_seg * p->ofs_unit;
218 	else if (p->gc_mode == GC_CB)
219 		return UINT_MAX;
220 	else /* No other gc_mode */
221 		return 0;
222 }
223 
224 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
225 {
226 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
227 	unsigned int secno;
228 
229 	/*
230 	 * If the gc_type is FG_GC, we can select victim segments
231 	 * selected by background GC before.
232 	 * Those segments guarantee they have small valid blocks.
233 	 */
234 	for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
235 		if (sec_usage_check(sbi, secno))
236 			continue;
237 
238 		if (no_fggc_candidate(sbi, secno))
239 			continue;
240 
241 		clear_bit(secno, dirty_i->victim_secmap);
242 		return GET_SEG_FROM_SEC(sbi, secno);
243 	}
244 	return NULL_SEGNO;
245 }
246 
247 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
248 {
249 	struct sit_info *sit_i = SIT_I(sbi);
250 	unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
251 	unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
252 	unsigned long long mtime = 0;
253 	unsigned int vblocks;
254 	unsigned char age = 0;
255 	unsigned char u;
256 	unsigned int i;
257 
258 	for (i = 0; i < sbi->segs_per_sec; i++)
259 		mtime += get_seg_entry(sbi, start + i)->mtime;
260 	vblocks = get_valid_blocks(sbi, segno, true);
261 
262 	mtime = div_u64(mtime, sbi->segs_per_sec);
263 	vblocks = div_u64(vblocks, sbi->segs_per_sec);
264 
265 	u = (vblocks * 100) >> sbi->log_blocks_per_seg;
266 
267 	/* Handle if the system time has changed by the user */
268 	if (mtime < sit_i->min_mtime)
269 		sit_i->min_mtime = mtime;
270 	if (mtime > sit_i->max_mtime)
271 		sit_i->max_mtime = mtime;
272 	if (sit_i->max_mtime != sit_i->min_mtime)
273 		age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
274 				sit_i->max_mtime - sit_i->min_mtime);
275 
276 	return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
277 }
278 
279 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
280 			unsigned int segno, struct victim_sel_policy *p)
281 {
282 	if (p->alloc_mode == SSR)
283 		return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
284 
285 	/* alloc_mode == LFS */
286 	if (p->gc_mode == GC_GREEDY)
287 		return get_valid_blocks(sbi, segno, true);
288 	else
289 		return get_cb_cost(sbi, segno);
290 }
291 
292 static unsigned int count_bits(const unsigned long *addr,
293 				unsigned int offset, unsigned int len)
294 {
295 	unsigned int end = offset + len, sum = 0;
296 
297 	while (offset < end) {
298 		if (test_bit(offset++, addr))
299 			++sum;
300 	}
301 	return sum;
302 }
303 
304 /*
305  * This function is called from two paths.
306  * One is garbage collection and the other is SSR segment selection.
307  * When it is called during GC, it just gets a victim segment
308  * and it does not remove it from dirty seglist.
309  * When it is called from SSR segment selection, it finds a segment
310  * which has minimum valid blocks and removes it from dirty seglist.
311  */
312 static int get_victim_by_default(struct f2fs_sb_info *sbi,
313 		unsigned int *result, int gc_type, int type, char alloc_mode)
314 {
315 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
316 	struct sit_info *sm = SIT_I(sbi);
317 	struct victim_sel_policy p;
318 	unsigned int secno, last_victim;
319 	unsigned int last_segment = MAIN_SEGS(sbi);
320 	unsigned int nsearched = 0;
321 
322 	mutex_lock(&dirty_i->seglist_lock);
323 
324 	p.alloc_mode = alloc_mode;
325 	select_policy(sbi, gc_type, type, &p);
326 
327 	p.min_segno = NULL_SEGNO;
328 	p.min_cost = get_max_cost(sbi, &p);
329 
330 	if (*result != NULL_SEGNO) {
331 		if (IS_DATASEG(get_seg_entry(sbi, *result)->type) &&
332 			get_valid_blocks(sbi, *result, false) &&
333 			!sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
334 			p.min_segno = *result;
335 		goto out;
336 	}
337 
338 	if (p.max_search == 0)
339 		goto out;
340 
341 	last_victim = sm->last_victim[p.gc_mode];
342 	if (p.alloc_mode == LFS && gc_type == FG_GC) {
343 		p.min_segno = check_bg_victims(sbi);
344 		if (p.min_segno != NULL_SEGNO)
345 			goto got_it;
346 	}
347 
348 	while (1) {
349 		unsigned long cost;
350 		unsigned int segno;
351 
352 		segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
353 		if (segno >= last_segment) {
354 			if (sm->last_victim[p.gc_mode]) {
355 				last_segment =
356 					sm->last_victim[p.gc_mode];
357 				sm->last_victim[p.gc_mode] = 0;
358 				p.offset = 0;
359 				continue;
360 			}
361 			break;
362 		}
363 
364 		p.offset = segno + p.ofs_unit;
365 		if (p.ofs_unit > 1) {
366 			p.offset -= segno % p.ofs_unit;
367 			nsearched += count_bits(p.dirty_segmap,
368 						p.offset - p.ofs_unit,
369 						p.ofs_unit);
370 		} else {
371 			nsearched++;
372 		}
373 
374 		secno = GET_SEC_FROM_SEG(sbi, segno);
375 
376 		if (sec_usage_check(sbi, secno))
377 			goto next;
378 		if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
379 			goto next;
380 		if (gc_type == FG_GC && p.alloc_mode == LFS &&
381 					no_fggc_candidate(sbi, secno))
382 			goto next;
383 
384 		cost = get_gc_cost(sbi, segno, &p);
385 
386 		if (p.min_cost > cost) {
387 			p.min_segno = segno;
388 			p.min_cost = cost;
389 		}
390 next:
391 		if (nsearched >= p.max_search) {
392 			if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
393 				sm->last_victim[p.gc_mode] = last_victim + 1;
394 			else
395 				sm->last_victim[p.gc_mode] = segno + 1;
396 			sm->last_victim[p.gc_mode] %= MAIN_SEGS(sbi);
397 			break;
398 		}
399 	}
400 	if (p.min_segno != NULL_SEGNO) {
401 got_it:
402 		if (p.alloc_mode == LFS) {
403 			secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
404 			if (gc_type == FG_GC)
405 				sbi->cur_victim_sec = secno;
406 			else
407 				set_bit(secno, dirty_i->victim_secmap);
408 		}
409 		*result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
410 
411 		trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
412 				sbi->cur_victim_sec,
413 				prefree_segments(sbi), free_segments(sbi));
414 	}
415 out:
416 	mutex_unlock(&dirty_i->seglist_lock);
417 
418 	return (p.min_segno == NULL_SEGNO) ? 0 : 1;
419 }
420 
421 static const struct victim_selection default_v_ops = {
422 	.get_victim = get_victim_by_default,
423 };
424 
425 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
426 {
427 	struct inode_entry *ie;
428 
429 	ie = radix_tree_lookup(&gc_list->iroot, ino);
430 	if (ie)
431 		return ie->inode;
432 	return NULL;
433 }
434 
435 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
436 {
437 	struct inode_entry *new_ie;
438 
439 	if (inode == find_gc_inode(gc_list, inode->i_ino)) {
440 		iput(inode);
441 		return;
442 	}
443 	new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
444 	new_ie->inode = inode;
445 
446 	f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
447 	list_add_tail(&new_ie->list, &gc_list->ilist);
448 }
449 
450 static void put_gc_inode(struct gc_inode_list *gc_list)
451 {
452 	struct inode_entry *ie, *next_ie;
453 	list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
454 		radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
455 		iput(ie->inode);
456 		list_del(&ie->list);
457 		kmem_cache_free(inode_entry_slab, ie);
458 	}
459 }
460 
461 static int check_valid_map(struct f2fs_sb_info *sbi,
462 				unsigned int segno, int offset)
463 {
464 	struct sit_info *sit_i = SIT_I(sbi);
465 	struct seg_entry *sentry;
466 	int ret;
467 
468 	down_read(&sit_i->sentry_lock);
469 	sentry = get_seg_entry(sbi, segno);
470 	ret = f2fs_test_bit(offset, sentry->cur_valid_map);
471 	up_read(&sit_i->sentry_lock);
472 	return ret;
473 }
474 
475 /*
476  * This function compares node address got in summary with that in NAT.
477  * On validity, copy that node with cold status, otherwise (invalid node)
478  * ignore that.
479  */
480 static void gc_node_segment(struct f2fs_sb_info *sbi,
481 		struct f2fs_summary *sum, unsigned int segno, int gc_type)
482 {
483 	struct f2fs_summary *entry;
484 	block_t start_addr;
485 	int off;
486 	int phase = 0;
487 
488 	start_addr = START_BLOCK(sbi, segno);
489 
490 next_step:
491 	entry = sum;
492 
493 	for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
494 		nid_t nid = le32_to_cpu(entry->nid);
495 		struct page *node_page;
496 		struct node_info ni;
497 
498 		/* stop BG_GC if there is not enough free sections. */
499 		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
500 			return;
501 
502 		if (check_valid_map(sbi, segno, off) == 0)
503 			continue;
504 
505 		if (phase == 0) {
506 			ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
507 							META_NAT, true);
508 			continue;
509 		}
510 
511 		if (phase == 1) {
512 			ra_node_page(sbi, nid);
513 			continue;
514 		}
515 
516 		/* phase == 2 */
517 		node_page = get_node_page(sbi, nid);
518 		if (IS_ERR(node_page))
519 			continue;
520 
521 		/* block may become invalid during get_node_page */
522 		if (check_valid_map(sbi, segno, off) == 0) {
523 			f2fs_put_page(node_page, 1);
524 			continue;
525 		}
526 
527 		get_node_info(sbi, nid, &ni);
528 		if (ni.blk_addr != start_addr + off) {
529 			f2fs_put_page(node_page, 1);
530 			continue;
531 		}
532 
533 		move_node_page(node_page, gc_type);
534 		stat_inc_node_blk_count(sbi, 1, gc_type);
535 	}
536 
537 	if (++phase < 3)
538 		goto next_step;
539 }
540 
541 /*
542  * Calculate start block index indicating the given node offset.
543  * Be careful, caller should give this node offset only indicating direct node
544  * blocks. If any node offsets, which point the other types of node blocks such
545  * as indirect or double indirect node blocks, are given, it must be a caller's
546  * bug.
547  */
548 block_t start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
549 {
550 	unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
551 	unsigned int bidx;
552 
553 	if (node_ofs == 0)
554 		return 0;
555 
556 	if (node_ofs <= 2) {
557 		bidx = node_ofs - 1;
558 	} else if (node_ofs <= indirect_blks) {
559 		int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
560 		bidx = node_ofs - 2 - dec;
561 	} else {
562 		int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
563 		bidx = node_ofs - 5 - dec;
564 	}
565 	return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode);
566 }
567 
568 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
569 		struct node_info *dni, block_t blkaddr, unsigned int *nofs)
570 {
571 	struct page *node_page;
572 	nid_t nid;
573 	unsigned int ofs_in_node;
574 	block_t source_blkaddr;
575 
576 	nid = le32_to_cpu(sum->nid);
577 	ofs_in_node = le16_to_cpu(sum->ofs_in_node);
578 
579 	node_page = get_node_page(sbi, nid);
580 	if (IS_ERR(node_page))
581 		return false;
582 
583 	get_node_info(sbi, nid, dni);
584 
585 	if (sum->version != dni->version) {
586 		f2fs_msg(sbi->sb, KERN_WARNING,
587 				"%s: valid data with mismatched node version.",
588 				__func__);
589 		set_sbi_flag(sbi, SBI_NEED_FSCK);
590 	}
591 
592 	*nofs = ofs_of_node(node_page);
593 	source_blkaddr = datablock_addr(NULL, node_page, ofs_in_node);
594 	f2fs_put_page(node_page, 1);
595 
596 	if (source_blkaddr != blkaddr)
597 		return false;
598 	return true;
599 }
600 
601 /*
602  * Move data block via META_MAPPING while keeping locked data page.
603  * This can be used to move blocks, aka LBAs, directly on disk.
604  */
605 static void move_data_block(struct inode *inode, block_t bidx,
606 					unsigned int segno, int off)
607 {
608 	struct f2fs_io_info fio = {
609 		.sbi = F2FS_I_SB(inode),
610 		.ino = inode->i_ino,
611 		.type = DATA,
612 		.temp = COLD,
613 		.op = REQ_OP_READ,
614 		.op_flags = 0,
615 		.encrypted_page = NULL,
616 		.in_list = false,
617 	};
618 	struct dnode_of_data dn;
619 	struct f2fs_summary sum;
620 	struct node_info ni;
621 	struct page *page;
622 	block_t newaddr;
623 	int err;
624 
625 	/* do not read out */
626 	page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
627 	if (!page)
628 		return;
629 
630 	if (!check_valid_map(F2FS_I_SB(inode), segno, off))
631 		goto out;
632 
633 	if (f2fs_is_atomic_file(inode))
634 		goto out;
635 
636 	if (f2fs_is_pinned_file(inode)) {
637 		f2fs_pin_file_control(inode, true);
638 		goto out;
639 	}
640 
641 	set_new_dnode(&dn, inode, NULL, NULL, 0);
642 	err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
643 	if (err)
644 		goto out;
645 
646 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
647 		ClearPageUptodate(page);
648 		goto put_out;
649 	}
650 
651 	/*
652 	 * don't cache encrypted data into meta inode until previous dirty
653 	 * data were writebacked to avoid racing between GC and flush.
654 	 */
655 	f2fs_wait_on_page_writeback(page, DATA, true);
656 
657 	get_node_info(fio.sbi, dn.nid, &ni);
658 	set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
659 
660 	/* read page */
661 	fio.page = page;
662 	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
663 
664 	allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
665 					&sum, CURSEG_COLD_DATA, NULL, false);
666 
667 	fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
668 				newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
669 	if (!fio.encrypted_page) {
670 		err = -ENOMEM;
671 		goto recover_block;
672 	}
673 
674 	err = f2fs_submit_page_bio(&fio);
675 	if (err)
676 		goto put_page_out;
677 
678 	/* write page */
679 	lock_page(fio.encrypted_page);
680 
681 	if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) {
682 		err = -EIO;
683 		goto put_page_out;
684 	}
685 	if (unlikely(!PageUptodate(fio.encrypted_page))) {
686 		err = -EIO;
687 		goto put_page_out;
688 	}
689 
690 	set_page_dirty(fio.encrypted_page);
691 	f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true);
692 	if (clear_page_dirty_for_io(fio.encrypted_page))
693 		dec_page_count(fio.sbi, F2FS_DIRTY_META);
694 
695 	set_page_writeback(fio.encrypted_page);
696 
697 	/* allocate block address */
698 	f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
699 
700 	fio.op = REQ_OP_WRITE;
701 	fio.op_flags = REQ_SYNC;
702 	fio.new_blkaddr = newaddr;
703 	err = f2fs_submit_page_write(&fio);
704 	if (err) {
705 		if (PageWriteback(fio.encrypted_page))
706 			end_page_writeback(fio.encrypted_page);
707 		goto put_page_out;
708 	}
709 
710 	f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
711 
712 	f2fs_update_data_blkaddr(&dn, newaddr);
713 	set_inode_flag(inode, FI_APPEND_WRITE);
714 	if (page->index == 0)
715 		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
716 put_page_out:
717 	f2fs_put_page(fio.encrypted_page, 1);
718 recover_block:
719 	if (err)
720 		__f2fs_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
721 								true, true);
722 put_out:
723 	f2fs_put_dnode(&dn);
724 out:
725 	f2fs_put_page(page, 1);
726 }
727 
728 static void move_data_page(struct inode *inode, block_t bidx, int gc_type,
729 							unsigned int segno, int off)
730 {
731 	struct page *page;
732 
733 	page = get_lock_data_page(inode, bidx, true);
734 	if (IS_ERR(page))
735 		return;
736 
737 	if (!check_valid_map(F2FS_I_SB(inode), segno, off))
738 		goto out;
739 
740 	if (f2fs_is_atomic_file(inode))
741 		goto out;
742 	if (f2fs_is_pinned_file(inode)) {
743 		if (gc_type == FG_GC)
744 			f2fs_pin_file_control(inode, true);
745 		goto out;
746 	}
747 
748 	if (gc_type == BG_GC) {
749 		if (PageWriteback(page))
750 			goto out;
751 		set_page_dirty(page);
752 		set_cold_data(page);
753 	} else {
754 		struct f2fs_io_info fio = {
755 			.sbi = F2FS_I_SB(inode),
756 			.ino = inode->i_ino,
757 			.type = DATA,
758 			.temp = COLD,
759 			.op = REQ_OP_WRITE,
760 			.op_flags = REQ_SYNC,
761 			.old_blkaddr = NULL_ADDR,
762 			.page = page,
763 			.encrypted_page = NULL,
764 			.need_lock = LOCK_REQ,
765 			.io_type = FS_GC_DATA_IO,
766 		};
767 		bool is_dirty = PageDirty(page);
768 		int err;
769 
770 retry:
771 		set_page_dirty(page);
772 		f2fs_wait_on_page_writeback(page, DATA, true);
773 		if (clear_page_dirty_for_io(page)) {
774 			inode_dec_dirty_pages(inode);
775 			remove_dirty_inode(inode);
776 		}
777 
778 		set_cold_data(page);
779 
780 		err = do_write_data_page(&fio);
781 		if (err == -ENOMEM && is_dirty) {
782 			congestion_wait(BLK_RW_ASYNC, HZ/50);
783 			goto retry;
784 		}
785 	}
786 out:
787 	f2fs_put_page(page, 1);
788 }
789 
790 /*
791  * This function tries to get parent node of victim data block, and identifies
792  * data block validity. If the block is valid, copy that with cold status and
793  * modify parent node.
794  * If the parent node is not valid or the data block address is different,
795  * the victim data block is ignored.
796  */
797 static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
798 		struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
799 {
800 	struct super_block *sb = sbi->sb;
801 	struct f2fs_summary *entry;
802 	block_t start_addr;
803 	int off;
804 	int phase = 0;
805 
806 	start_addr = START_BLOCK(sbi, segno);
807 
808 next_step:
809 	entry = sum;
810 
811 	for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
812 		struct page *data_page;
813 		struct inode *inode;
814 		struct node_info dni; /* dnode info for the data */
815 		unsigned int ofs_in_node, nofs;
816 		block_t start_bidx;
817 		nid_t nid = le32_to_cpu(entry->nid);
818 
819 		/* stop BG_GC if there is not enough free sections. */
820 		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
821 			return;
822 
823 		if (check_valid_map(sbi, segno, off) == 0)
824 			continue;
825 
826 		if (phase == 0) {
827 			ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
828 							META_NAT, true);
829 			continue;
830 		}
831 
832 		if (phase == 1) {
833 			ra_node_page(sbi, nid);
834 			continue;
835 		}
836 
837 		/* Get an inode by ino with checking validity */
838 		if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
839 			continue;
840 
841 		if (phase == 2) {
842 			ra_node_page(sbi, dni.ino);
843 			continue;
844 		}
845 
846 		ofs_in_node = le16_to_cpu(entry->ofs_in_node);
847 
848 		if (phase == 3) {
849 			inode = f2fs_iget(sb, dni.ino);
850 			if (IS_ERR(inode) || is_bad_inode(inode))
851 				continue;
852 
853 			/* if encrypted inode, let's go phase 3 */
854 			if (f2fs_encrypted_file(inode)) {
855 				add_gc_inode(gc_list, inode);
856 				continue;
857 			}
858 
859 			if (!down_write_trylock(
860 				&F2FS_I(inode)->dio_rwsem[WRITE])) {
861 				iput(inode);
862 				continue;
863 			}
864 
865 			start_bidx = start_bidx_of_node(nofs, inode);
866 			data_page = get_read_data_page(inode,
867 					start_bidx + ofs_in_node, REQ_RAHEAD,
868 					true);
869 			up_write(&F2FS_I(inode)->dio_rwsem[WRITE]);
870 			if (IS_ERR(data_page)) {
871 				iput(inode);
872 				continue;
873 			}
874 
875 			f2fs_put_page(data_page, 0);
876 			add_gc_inode(gc_list, inode);
877 			continue;
878 		}
879 
880 		/* phase 4 */
881 		inode = find_gc_inode(gc_list, dni.ino);
882 		if (inode) {
883 			struct f2fs_inode_info *fi = F2FS_I(inode);
884 			bool locked = false;
885 
886 			if (S_ISREG(inode->i_mode)) {
887 				if (!down_write_trylock(&fi->dio_rwsem[READ]))
888 					continue;
889 				if (!down_write_trylock(
890 						&fi->dio_rwsem[WRITE])) {
891 					up_write(&fi->dio_rwsem[READ]);
892 					continue;
893 				}
894 				locked = true;
895 
896 				/* wait for all inflight aio data */
897 				inode_dio_wait(inode);
898 			}
899 
900 			start_bidx = start_bidx_of_node(nofs, inode)
901 								+ ofs_in_node;
902 			if (f2fs_encrypted_file(inode))
903 				move_data_block(inode, start_bidx, segno, off);
904 			else
905 				move_data_page(inode, start_bidx, gc_type,
906 								segno, off);
907 
908 			if (locked) {
909 				up_write(&fi->dio_rwsem[WRITE]);
910 				up_write(&fi->dio_rwsem[READ]);
911 			}
912 
913 			stat_inc_data_blk_count(sbi, 1, gc_type);
914 		}
915 	}
916 
917 	if (++phase < 5)
918 		goto next_step;
919 }
920 
921 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
922 			int gc_type)
923 {
924 	struct sit_info *sit_i = SIT_I(sbi);
925 	int ret;
926 
927 	down_write(&sit_i->sentry_lock);
928 	ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
929 					      NO_CHECK_TYPE, LFS);
930 	up_write(&sit_i->sentry_lock);
931 	return ret;
932 }
933 
934 static int do_garbage_collect(struct f2fs_sb_info *sbi,
935 				unsigned int start_segno,
936 				struct gc_inode_list *gc_list, int gc_type)
937 {
938 	struct page *sum_page;
939 	struct f2fs_summary_block *sum;
940 	struct blk_plug plug;
941 	unsigned int segno = start_segno;
942 	unsigned int end_segno = start_segno + sbi->segs_per_sec;
943 	int seg_freed = 0;
944 	unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
945 						SUM_TYPE_DATA : SUM_TYPE_NODE;
946 
947 	/* readahead multi ssa blocks those have contiguous address */
948 	if (sbi->segs_per_sec > 1)
949 		ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
950 					sbi->segs_per_sec, META_SSA, true);
951 
952 	/* reference all summary page */
953 	while (segno < end_segno) {
954 		sum_page = get_sum_page(sbi, segno++);
955 		unlock_page(sum_page);
956 	}
957 
958 	blk_start_plug(&plug);
959 
960 	for (segno = start_segno; segno < end_segno; segno++) {
961 
962 		/* find segment summary of victim */
963 		sum_page = find_get_page(META_MAPPING(sbi),
964 					GET_SUM_BLOCK(sbi, segno));
965 		f2fs_put_page(sum_page, 0);
966 
967 		if (get_valid_blocks(sbi, segno, false) == 0 ||
968 				!PageUptodate(sum_page) ||
969 				unlikely(f2fs_cp_error(sbi)))
970 			goto next;
971 
972 		sum = page_address(sum_page);
973 		f2fs_bug_on(sbi, type != GET_SUM_TYPE((&sum->footer)));
974 
975 		/*
976 		 * this is to avoid deadlock:
977 		 * - lock_page(sum_page)         - f2fs_replace_block
978 		 *  - check_valid_map()            - down_write(sentry_lock)
979 		 *   - down_read(sentry_lock)     - change_curseg()
980 		 *                                  - lock_page(sum_page)
981 		 */
982 		if (type == SUM_TYPE_NODE)
983 			gc_node_segment(sbi, sum->entries, segno, gc_type);
984 		else
985 			gc_data_segment(sbi, sum->entries, gc_list, segno,
986 								gc_type);
987 
988 		stat_inc_seg_count(sbi, type, gc_type);
989 
990 		if (gc_type == FG_GC &&
991 				get_valid_blocks(sbi, segno, false) == 0)
992 			seg_freed++;
993 next:
994 		f2fs_put_page(sum_page, 0);
995 	}
996 
997 	if (gc_type == FG_GC)
998 		f2fs_submit_merged_write(sbi,
999 				(type == SUM_TYPE_NODE) ? NODE : DATA);
1000 
1001 	blk_finish_plug(&plug);
1002 
1003 	stat_inc_call_count(sbi->stat_info);
1004 
1005 	return seg_freed;
1006 }
1007 
1008 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1009 			bool background, unsigned int segno)
1010 {
1011 	int gc_type = sync ? FG_GC : BG_GC;
1012 	int sec_freed = 0, seg_freed = 0, total_freed = 0;
1013 	int ret = 0;
1014 	struct cp_control cpc;
1015 	unsigned int init_segno = segno;
1016 	struct gc_inode_list gc_list = {
1017 		.ilist = LIST_HEAD_INIT(gc_list.ilist),
1018 		.iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1019 	};
1020 
1021 	trace_f2fs_gc_begin(sbi->sb, sync, background,
1022 				get_pages(sbi, F2FS_DIRTY_NODES),
1023 				get_pages(sbi, F2FS_DIRTY_DENTS),
1024 				get_pages(sbi, F2FS_DIRTY_IMETA),
1025 				free_sections(sbi),
1026 				free_segments(sbi),
1027 				reserved_segments(sbi),
1028 				prefree_segments(sbi));
1029 
1030 	cpc.reason = __get_cp_reason(sbi);
1031 gc_more:
1032 	if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1033 		ret = -EINVAL;
1034 		goto stop;
1035 	}
1036 	if (unlikely(f2fs_cp_error(sbi))) {
1037 		ret = -EIO;
1038 		goto stop;
1039 	}
1040 
1041 	if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1042 		/*
1043 		 * For example, if there are many prefree_segments below given
1044 		 * threshold, we can make them free by checkpoint. Then, we
1045 		 * secure free segments which doesn't need fggc any more.
1046 		 */
1047 		if (prefree_segments(sbi)) {
1048 			ret = write_checkpoint(sbi, &cpc);
1049 			if (ret)
1050 				goto stop;
1051 		}
1052 		if (has_not_enough_free_secs(sbi, 0, 0))
1053 			gc_type = FG_GC;
1054 	}
1055 
1056 	/* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1057 	if (gc_type == BG_GC && !background) {
1058 		ret = -EINVAL;
1059 		goto stop;
1060 	}
1061 	if (!__get_victim(sbi, &segno, gc_type)) {
1062 		ret = -ENODATA;
1063 		goto stop;
1064 	}
1065 
1066 	seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type);
1067 	if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec)
1068 		sec_freed++;
1069 	total_freed += seg_freed;
1070 
1071 	if (gc_type == FG_GC)
1072 		sbi->cur_victim_sec = NULL_SEGNO;
1073 
1074 	if (!sync) {
1075 		if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1076 			segno = NULL_SEGNO;
1077 			goto gc_more;
1078 		}
1079 
1080 		if (gc_type == FG_GC)
1081 			ret = write_checkpoint(sbi, &cpc);
1082 	}
1083 stop:
1084 	SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1085 	SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1086 
1087 	trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1088 				get_pages(sbi, F2FS_DIRTY_NODES),
1089 				get_pages(sbi, F2FS_DIRTY_DENTS),
1090 				get_pages(sbi, F2FS_DIRTY_IMETA),
1091 				free_sections(sbi),
1092 				free_segments(sbi),
1093 				reserved_segments(sbi),
1094 				prefree_segments(sbi));
1095 
1096 	mutex_unlock(&sbi->gc_mutex);
1097 
1098 	put_gc_inode(&gc_list);
1099 
1100 	if (sync)
1101 		ret = sec_freed ? 0 : -EAGAIN;
1102 	return ret;
1103 }
1104 
1105 void build_gc_manager(struct f2fs_sb_info *sbi)
1106 {
1107 	u64 main_count, resv_count, ovp_count;
1108 
1109 	DIRTY_I(sbi)->v_ops = &default_v_ops;
1110 
1111 	/* threshold of # of valid blocks in a section for victims of FG_GC */
1112 	main_count = SM_I(sbi)->main_segments << sbi->log_blocks_per_seg;
1113 	resv_count = SM_I(sbi)->reserved_segments << sbi->log_blocks_per_seg;
1114 	ovp_count = SM_I(sbi)->ovp_segments << sbi->log_blocks_per_seg;
1115 
1116 	sbi->fggc_threshold = div64_u64((main_count - ovp_count) *
1117 				BLKS_PER_SEC(sbi), (main_count - resv_count));
1118 	sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1119 
1120 	/* give warm/cold data area from slower device */
1121 	if (sbi->s_ndevs && sbi->segs_per_sec == 1)
1122 		SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1123 				GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1124 }
1125