xref: /openbmc/linux/fs/f2fs/gc.c (revision 3b23dc52)
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 (sbi->gc_mode == 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 f2fs_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_wake= 0;
135 
136 	sbi->gc_thread = gc_th;
137 	init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
138 	sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
139 			"f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
140 	if (IS_ERR(gc_th->f2fs_gc_task)) {
141 		err = PTR_ERR(gc_th->f2fs_gc_task);
142 		kfree(gc_th);
143 		sbi->gc_thread = NULL;
144 	}
145 out:
146 	return err;
147 }
148 
149 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
150 {
151 	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
152 	if (!gc_th)
153 		return;
154 	kthread_stop(gc_th->f2fs_gc_task);
155 	kfree(gc_th);
156 	sbi->gc_thread = NULL;
157 }
158 
159 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
160 {
161 	int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
162 
163 	switch (sbi->gc_mode) {
164 	case GC_IDLE_CB:
165 		gc_mode = GC_CB;
166 		break;
167 	case GC_IDLE_GREEDY:
168 	case GC_URGENT:
169 		gc_mode = GC_GREEDY;
170 		break;
171 	}
172 	return gc_mode;
173 }
174 
175 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
176 			int type, struct victim_sel_policy *p)
177 {
178 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
179 
180 	if (p->alloc_mode == SSR) {
181 		p->gc_mode = GC_GREEDY;
182 		p->dirty_segmap = dirty_i->dirty_segmap[type];
183 		p->max_search = dirty_i->nr_dirty[type];
184 		p->ofs_unit = 1;
185 	} else {
186 		p->gc_mode = select_gc_type(sbi, gc_type);
187 		p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
188 		p->max_search = dirty_i->nr_dirty[DIRTY];
189 		p->ofs_unit = sbi->segs_per_sec;
190 	}
191 
192 	/* we need to check every dirty segments in the FG_GC case */
193 	if (gc_type != FG_GC &&
194 			(sbi->gc_mode != GC_URGENT) &&
195 			p->max_search > sbi->max_victim_search)
196 		p->max_search = sbi->max_victim_search;
197 
198 	/* let's select beginning hot/small space first in no_heap mode*/
199 	if (test_opt(sbi, NOHEAP) &&
200 		(type == CURSEG_HOT_DATA || IS_NODESEG(type)))
201 		p->offset = 0;
202 	else
203 		p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
204 }
205 
206 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
207 				struct victim_sel_policy *p)
208 {
209 	/* SSR allocates in a segment unit */
210 	if (p->alloc_mode == SSR)
211 		return sbi->blocks_per_seg;
212 	if (p->gc_mode == GC_GREEDY)
213 		return 2 * sbi->blocks_per_seg * p->ofs_unit;
214 	else if (p->gc_mode == GC_CB)
215 		return UINT_MAX;
216 	else /* No other gc_mode */
217 		return 0;
218 }
219 
220 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
221 {
222 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
223 	unsigned int secno;
224 
225 	/*
226 	 * If the gc_type is FG_GC, we can select victim segments
227 	 * selected by background GC before.
228 	 * Those segments guarantee they have small valid blocks.
229 	 */
230 	for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
231 		if (sec_usage_check(sbi, secno))
232 			continue;
233 		clear_bit(secno, dirty_i->victim_secmap);
234 		return GET_SEG_FROM_SEC(sbi, secno);
235 	}
236 	return NULL_SEGNO;
237 }
238 
239 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
240 {
241 	struct sit_info *sit_i = SIT_I(sbi);
242 	unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
243 	unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
244 	unsigned long long mtime = 0;
245 	unsigned int vblocks;
246 	unsigned char age = 0;
247 	unsigned char u;
248 	unsigned int i;
249 
250 	for (i = 0; i < sbi->segs_per_sec; i++)
251 		mtime += get_seg_entry(sbi, start + i)->mtime;
252 	vblocks = get_valid_blocks(sbi, segno, true);
253 
254 	mtime = div_u64(mtime, sbi->segs_per_sec);
255 	vblocks = div_u64(vblocks, sbi->segs_per_sec);
256 
257 	u = (vblocks * 100) >> sbi->log_blocks_per_seg;
258 
259 	/* Handle if the system time has changed by the user */
260 	if (mtime < sit_i->min_mtime)
261 		sit_i->min_mtime = mtime;
262 	if (mtime > sit_i->max_mtime)
263 		sit_i->max_mtime = mtime;
264 	if (sit_i->max_mtime != sit_i->min_mtime)
265 		age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
266 				sit_i->max_mtime - sit_i->min_mtime);
267 
268 	return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
269 }
270 
271 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
272 			unsigned int segno, struct victim_sel_policy *p)
273 {
274 	if (p->alloc_mode == SSR)
275 		return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
276 
277 	/* alloc_mode == LFS */
278 	if (p->gc_mode == GC_GREEDY)
279 		return get_valid_blocks(sbi, segno, true);
280 	else
281 		return get_cb_cost(sbi, segno);
282 }
283 
284 static unsigned int count_bits(const unsigned long *addr,
285 				unsigned int offset, unsigned int len)
286 {
287 	unsigned int end = offset + len, sum = 0;
288 
289 	while (offset < end) {
290 		if (test_bit(offset++, addr))
291 			++sum;
292 	}
293 	return sum;
294 }
295 
296 /*
297  * This function is called from two paths.
298  * One is garbage collection and the other is SSR segment selection.
299  * When it is called during GC, it just gets a victim segment
300  * and it does not remove it from dirty seglist.
301  * When it is called from SSR segment selection, it finds a segment
302  * which has minimum valid blocks and removes it from dirty seglist.
303  */
304 static int get_victim_by_default(struct f2fs_sb_info *sbi,
305 		unsigned int *result, int gc_type, int type, char alloc_mode)
306 {
307 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
308 	struct sit_info *sm = SIT_I(sbi);
309 	struct victim_sel_policy p;
310 	unsigned int secno, last_victim;
311 	unsigned int last_segment = MAIN_SEGS(sbi);
312 	unsigned int nsearched = 0;
313 
314 	mutex_lock(&dirty_i->seglist_lock);
315 
316 	p.alloc_mode = alloc_mode;
317 	select_policy(sbi, gc_type, type, &p);
318 
319 	p.min_segno = NULL_SEGNO;
320 	p.min_cost = get_max_cost(sbi, &p);
321 
322 	if (*result != NULL_SEGNO) {
323 		if (IS_DATASEG(get_seg_entry(sbi, *result)->type) &&
324 			get_valid_blocks(sbi, *result, false) &&
325 			!sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
326 			p.min_segno = *result;
327 		goto out;
328 	}
329 
330 	if (p.max_search == 0)
331 		goto out;
332 
333 	last_victim = sm->last_victim[p.gc_mode];
334 	if (p.alloc_mode == LFS && gc_type == FG_GC) {
335 		p.min_segno = check_bg_victims(sbi);
336 		if (p.min_segno != NULL_SEGNO)
337 			goto got_it;
338 	}
339 
340 	while (1) {
341 		unsigned long cost;
342 		unsigned int segno;
343 
344 		segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
345 		if (segno >= last_segment) {
346 			if (sm->last_victim[p.gc_mode]) {
347 				last_segment =
348 					sm->last_victim[p.gc_mode];
349 				sm->last_victim[p.gc_mode] = 0;
350 				p.offset = 0;
351 				continue;
352 			}
353 			break;
354 		}
355 
356 		p.offset = segno + p.ofs_unit;
357 		if (p.ofs_unit > 1) {
358 			p.offset -= segno % p.ofs_unit;
359 			nsearched += count_bits(p.dirty_segmap,
360 						p.offset - p.ofs_unit,
361 						p.ofs_unit);
362 		} else {
363 			nsearched++;
364 		}
365 
366 		secno = GET_SEC_FROM_SEG(sbi, segno);
367 
368 		if (sec_usage_check(sbi, secno))
369 			goto next;
370 		if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
371 			goto next;
372 
373 		cost = get_gc_cost(sbi, segno, &p);
374 
375 		if (p.min_cost > cost) {
376 			p.min_segno = segno;
377 			p.min_cost = cost;
378 		}
379 next:
380 		if (nsearched >= p.max_search) {
381 			if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
382 				sm->last_victim[p.gc_mode] = last_victim + 1;
383 			else
384 				sm->last_victim[p.gc_mode] = segno + 1;
385 			sm->last_victim[p.gc_mode] %= MAIN_SEGS(sbi);
386 			break;
387 		}
388 	}
389 	if (p.min_segno != NULL_SEGNO) {
390 got_it:
391 		if (p.alloc_mode == LFS) {
392 			secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
393 			if (gc_type == FG_GC)
394 				sbi->cur_victim_sec = secno;
395 			else
396 				set_bit(secno, dirty_i->victim_secmap);
397 		}
398 		*result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
399 
400 		trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
401 				sbi->cur_victim_sec,
402 				prefree_segments(sbi), free_segments(sbi));
403 	}
404 out:
405 	mutex_unlock(&dirty_i->seglist_lock);
406 
407 	return (p.min_segno == NULL_SEGNO) ? 0 : 1;
408 }
409 
410 static const struct victim_selection default_v_ops = {
411 	.get_victim = get_victim_by_default,
412 };
413 
414 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
415 {
416 	struct inode_entry *ie;
417 
418 	ie = radix_tree_lookup(&gc_list->iroot, ino);
419 	if (ie)
420 		return ie->inode;
421 	return NULL;
422 }
423 
424 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
425 {
426 	struct inode_entry *new_ie;
427 
428 	if (inode == find_gc_inode(gc_list, inode->i_ino)) {
429 		iput(inode);
430 		return;
431 	}
432 	new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS);
433 	new_ie->inode = inode;
434 
435 	f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
436 	list_add_tail(&new_ie->list, &gc_list->ilist);
437 }
438 
439 static void put_gc_inode(struct gc_inode_list *gc_list)
440 {
441 	struct inode_entry *ie, *next_ie;
442 	list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
443 		radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
444 		iput(ie->inode);
445 		list_del(&ie->list);
446 		kmem_cache_free(f2fs_inode_entry_slab, ie);
447 	}
448 }
449 
450 static int check_valid_map(struct f2fs_sb_info *sbi,
451 				unsigned int segno, int offset)
452 {
453 	struct sit_info *sit_i = SIT_I(sbi);
454 	struct seg_entry *sentry;
455 	int ret;
456 
457 	down_read(&sit_i->sentry_lock);
458 	sentry = get_seg_entry(sbi, segno);
459 	ret = f2fs_test_bit(offset, sentry->cur_valid_map);
460 	up_read(&sit_i->sentry_lock);
461 	return ret;
462 }
463 
464 /*
465  * This function compares node address got in summary with that in NAT.
466  * On validity, copy that node with cold status, otherwise (invalid node)
467  * ignore that.
468  */
469 static void gc_node_segment(struct f2fs_sb_info *sbi,
470 		struct f2fs_summary *sum, unsigned int segno, int gc_type)
471 {
472 	struct f2fs_summary *entry;
473 	block_t start_addr;
474 	int off;
475 	int phase = 0;
476 	bool fggc = (gc_type == FG_GC);
477 
478 	start_addr = START_BLOCK(sbi, segno);
479 
480 next_step:
481 	entry = sum;
482 
483 	if (fggc && phase == 2)
484 		atomic_inc(&sbi->wb_sync_req[NODE]);
485 
486 	for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
487 		nid_t nid = le32_to_cpu(entry->nid);
488 		struct page *node_page;
489 		struct node_info ni;
490 
491 		/* stop BG_GC if there is not enough free sections. */
492 		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
493 			return;
494 
495 		if (check_valid_map(sbi, segno, off) == 0)
496 			continue;
497 
498 		if (phase == 0) {
499 			f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
500 							META_NAT, true);
501 			continue;
502 		}
503 
504 		if (phase == 1) {
505 			f2fs_ra_node_page(sbi, nid);
506 			continue;
507 		}
508 
509 		/* phase == 2 */
510 		node_page = f2fs_get_node_page(sbi, nid);
511 		if (IS_ERR(node_page))
512 			continue;
513 
514 		/* block may become invalid during f2fs_get_node_page */
515 		if (check_valid_map(sbi, segno, off) == 0) {
516 			f2fs_put_page(node_page, 1);
517 			continue;
518 		}
519 
520 		f2fs_get_node_info(sbi, nid, &ni);
521 		if (ni.blk_addr != start_addr + off) {
522 			f2fs_put_page(node_page, 1);
523 			continue;
524 		}
525 
526 		f2fs_move_node_page(node_page, gc_type);
527 		stat_inc_node_blk_count(sbi, 1, gc_type);
528 	}
529 
530 	if (++phase < 3)
531 		goto next_step;
532 
533 	if (fggc)
534 		atomic_dec(&sbi->wb_sync_req[NODE]);
535 }
536 
537 /*
538  * Calculate start block index indicating the given node offset.
539  * Be careful, caller should give this node offset only indicating direct node
540  * blocks. If any node offsets, which point the other types of node blocks such
541  * as indirect or double indirect node blocks, are given, it must be a caller's
542  * bug.
543  */
544 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
545 {
546 	unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
547 	unsigned int bidx;
548 
549 	if (node_ofs == 0)
550 		return 0;
551 
552 	if (node_ofs <= 2) {
553 		bidx = node_ofs - 1;
554 	} else if (node_ofs <= indirect_blks) {
555 		int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
556 		bidx = node_ofs - 2 - dec;
557 	} else {
558 		int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
559 		bidx = node_ofs - 5 - dec;
560 	}
561 	return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode);
562 }
563 
564 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
565 		struct node_info *dni, block_t blkaddr, unsigned int *nofs)
566 {
567 	struct page *node_page;
568 	nid_t nid;
569 	unsigned int ofs_in_node;
570 	block_t source_blkaddr;
571 
572 	nid = le32_to_cpu(sum->nid);
573 	ofs_in_node = le16_to_cpu(sum->ofs_in_node);
574 
575 	node_page = f2fs_get_node_page(sbi, nid);
576 	if (IS_ERR(node_page))
577 		return false;
578 
579 	f2fs_get_node_info(sbi, nid, dni);
580 
581 	if (sum->version != dni->version) {
582 		f2fs_msg(sbi->sb, KERN_WARNING,
583 				"%s: valid data with mismatched node version.",
584 				__func__);
585 		set_sbi_flag(sbi, SBI_NEED_FSCK);
586 	}
587 
588 	*nofs = ofs_of_node(node_page);
589 	source_blkaddr = datablock_addr(NULL, node_page, ofs_in_node);
590 	f2fs_put_page(node_page, 1);
591 
592 	if (source_blkaddr != blkaddr)
593 		return false;
594 	return true;
595 }
596 
597 /*
598  * Move data block via META_MAPPING while keeping locked data page.
599  * This can be used to move blocks, aka LBAs, directly on disk.
600  */
601 static void move_data_block(struct inode *inode, block_t bidx,
602 				int gc_type, unsigned int segno, int off)
603 {
604 	struct f2fs_io_info fio = {
605 		.sbi = F2FS_I_SB(inode),
606 		.ino = inode->i_ino,
607 		.type = DATA,
608 		.temp = COLD,
609 		.op = REQ_OP_READ,
610 		.op_flags = 0,
611 		.encrypted_page = NULL,
612 		.in_list = false,
613 		.retry = false,
614 	};
615 	struct dnode_of_data dn;
616 	struct f2fs_summary sum;
617 	struct node_info ni;
618 	struct page *page;
619 	block_t newaddr;
620 	int err;
621 	bool lfs_mode = test_opt(fio.sbi, LFS);
622 
623 	/* do not read out */
624 	page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
625 	if (!page)
626 		return;
627 
628 	if (!check_valid_map(F2FS_I_SB(inode), segno, off))
629 		goto out;
630 
631 	if (f2fs_is_atomic_file(inode)) {
632 		F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
633 		F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
634 		goto out;
635 	}
636 
637 	if (f2fs_is_pinned_file(inode)) {
638 		f2fs_pin_file_control(inode, true);
639 		goto out;
640 	}
641 
642 	set_new_dnode(&dn, inode, NULL, NULL, 0);
643 	err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
644 	if (err)
645 		goto out;
646 
647 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
648 		ClearPageUptodate(page);
649 		goto put_out;
650 	}
651 
652 	/*
653 	 * don't cache encrypted data into meta inode until previous dirty
654 	 * data were writebacked to avoid racing between GC and flush.
655 	 */
656 	f2fs_wait_on_page_writeback(page, DATA, true);
657 
658 	f2fs_get_node_info(fio.sbi, dn.nid, &ni);
659 	set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
660 
661 	/* read page */
662 	fio.page = page;
663 	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
664 
665 	if (lfs_mode)
666 		down_write(&fio.sbi->io_order_lock);
667 
668 	f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
669 					&sum, CURSEG_COLD_DATA, NULL, false);
670 
671 	fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
672 				newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
673 	if (!fio.encrypted_page) {
674 		err = -ENOMEM;
675 		goto recover_block;
676 	}
677 
678 	err = f2fs_submit_page_bio(&fio);
679 	if (err)
680 		goto put_page_out;
681 
682 	/* write page */
683 	lock_page(fio.encrypted_page);
684 
685 	if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) {
686 		err = -EIO;
687 		goto put_page_out;
688 	}
689 	if (unlikely(!PageUptodate(fio.encrypted_page))) {
690 		err = -EIO;
691 		goto put_page_out;
692 	}
693 
694 	set_page_dirty(fio.encrypted_page);
695 	f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true);
696 	if (clear_page_dirty_for_io(fio.encrypted_page))
697 		dec_page_count(fio.sbi, F2FS_DIRTY_META);
698 
699 	set_page_writeback(fio.encrypted_page);
700 	ClearPageError(page);
701 
702 	/* allocate block address */
703 	f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
704 
705 	fio.op = REQ_OP_WRITE;
706 	fio.op_flags = REQ_SYNC;
707 	fio.new_blkaddr = newaddr;
708 	f2fs_submit_page_write(&fio);
709 	if (fio.retry) {
710 		if (PageWriteback(fio.encrypted_page))
711 			end_page_writeback(fio.encrypted_page);
712 		goto put_page_out;
713 	}
714 
715 	f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
716 
717 	f2fs_update_data_blkaddr(&dn, newaddr);
718 	set_inode_flag(inode, FI_APPEND_WRITE);
719 	if (page->index == 0)
720 		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
721 put_page_out:
722 	f2fs_put_page(fio.encrypted_page, 1);
723 recover_block:
724 	if (lfs_mode)
725 		up_write(&fio.sbi->io_order_lock);
726 	if (err)
727 		f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
728 								true, true);
729 put_out:
730 	f2fs_put_dnode(&dn);
731 out:
732 	f2fs_put_page(page, 1);
733 }
734 
735 static void move_data_page(struct inode *inode, block_t bidx, int gc_type,
736 							unsigned int segno, int off)
737 {
738 	struct page *page;
739 
740 	page = f2fs_get_lock_data_page(inode, bidx, true);
741 	if (IS_ERR(page))
742 		return;
743 
744 	if (!check_valid_map(F2FS_I_SB(inode), segno, off))
745 		goto out;
746 
747 	if (f2fs_is_atomic_file(inode)) {
748 		F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
749 		F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
750 		goto out;
751 	}
752 	if (f2fs_is_pinned_file(inode)) {
753 		if (gc_type == FG_GC)
754 			f2fs_pin_file_control(inode, true);
755 		goto out;
756 	}
757 
758 	if (gc_type == BG_GC) {
759 		if (PageWriteback(page))
760 			goto out;
761 		set_page_dirty(page);
762 		set_cold_data(page);
763 	} else {
764 		struct f2fs_io_info fio = {
765 			.sbi = F2FS_I_SB(inode),
766 			.ino = inode->i_ino,
767 			.type = DATA,
768 			.temp = COLD,
769 			.op = REQ_OP_WRITE,
770 			.op_flags = REQ_SYNC,
771 			.old_blkaddr = NULL_ADDR,
772 			.page = page,
773 			.encrypted_page = NULL,
774 			.need_lock = LOCK_REQ,
775 			.io_type = FS_GC_DATA_IO,
776 		};
777 		bool is_dirty = PageDirty(page);
778 		int err;
779 
780 retry:
781 		set_page_dirty(page);
782 		f2fs_wait_on_page_writeback(page, DATA, true);
783 		if (clear_page_dirty_for_io(page)) {
784 			inode_dec_dirty_pages(inode);
785 			f2fs_remove_dirty_inode(inode);
786 		}
787 
788 		set_cold_data(page);
789 
790 		err = f2fs_do_write_data_page(&fio);
791 		if (err) {
792 			clear_cold_data(page);
793 			if (err == -ENOMEM) {
794 				congestion_wait(BLK_RW_ASYNC, HZ/50);
795 				goto retry;
796 			}
797 			if (is_dirty)
798 				set_page_dirty(page);
799 		}
800 	}
801 out:
802 	f2fs_put_page(page, 1);
803 }
804 
805 /*
806  * This function tries to get parent node of victim data block, and identifies
807  * data block validity. If the block is valid, copy that with cold status and
808  * modify parent node.
809  * If the parent node is not valid or the data block address is different,
810  * the victim data block is ignored.
811  */
812 static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
813 		struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
814 {
815 	struct super_block *sb = sbi->sb;
816 	struct f2fs_summary *entry;
817 	block_t start_addr;
818 	int off;
819 	int phase = 0;
820 
821 	start_addr = START_BLOCK(sbi, segno);
822 
823 next_step:
824 	entry = sum;
825 
826 	for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
827 		struct page *data_page;
828 		struct inode *inode;
829 		struct node_info dni; /* dnode info for the data */
830 		unsigned int ofs_in_node, nofs;
831 		block_t start_bidx;
832 		nid_t nid = le32_to_cpu(entry->nid);
833 
834 		/* stop BG_GC if there is not enough free sections. */
835 		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
836 			return;
837 
838 		if (check_valid_map(sbi, segno, off) == 0)
839 			continue;
840 
841 		if (phase == 0) {
842 			f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
843 							META_NAT, true);
844 			continue;
845 		}
846 
847 		if (phase == 1) {
848 			f2fs_ra_node_page(sbi, nid);
849 			continue;
850 		}
851 
852 		/* Get an inode by ino with checking validity */
853 		if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
854 			continue;
855 
856 		if (phase == 2) {
857 			f2fs_ra_node_page(sbi, dni.ino);
858 			continue;
859 		}
860 
861 		ofs_in_node = le16_to_cpu(entry->ofs_in_node);
862 
863 		if (phase == 3) {
864 			inode = f2fs_iget(sb, dni.ino);
865 			if (IS_ERR(inode) || is_bad_inode(inode))
866 				continue;
867 
868 			/* if inode uses special I/O path, let's go phase 3 */
869 			if (f2fs_post_read_required(inode)) {
870 				add_gc_inode(gc_list, inode);
871 				continue;
872 			}
873 
874 			if (!down_write_trylock(
875 				&F2FS_I(inode)->i_gc_rwsem[WRITE])) {
876 				iput(inode);
877 				continue;
878 			}
879 
880 			start_bidx = f2fs_start_bidx_of_node(nofs, inode);
881 			data_page = f2fs_get_read_data_page(inode,
882 					start_bidx + ofs_in_node, REQ_RAHEAD,
883 					true);
884 			up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
885 			if (IS_ERR(data_page)) {
886 				iput(inode);
887 				continue;
888 			}
889 
890 			f2fs_put_page(data_page, 0);
891 			add_gc_inode(gc_list, inode);
892 			continue;
893 		}
894 
895 		/* phase 4 */
896 		inode = find_gc_inode(gc_list, dni.ino);
897 		if (inode) {
898 			struct f2fs_inode_info *fi = F2FS_I(inode);
899 			bool locked = false;
900 
901 			if (S_ISREG(inode->i_mode)) {
902 				if (!down_write_trylock(&fi->i_gc_rwsem[READ]))
903 					continue;
904 				if (!down_write_trylock(
905 						&fi->i_gc_rwsem[WRITE])) {
906 					up_write(&fi->i_gc_rwsem[READ]);
907 					continue;
908 				}
909 				locked = true;
910 
911 				/* wait for all inflight aio data */
912 				inode_dio_wait(inode);
913 			}
914 
915 			start_bidx = f2fs_start_bidx_of_node(nofs, inode)
916 								+ ofs_in_node;
917 			if (f2fs_post_read_required(inode))
918 				move_data_block(inode, start_bidx, gc_type,
919 								segno, off);
920 			else
921 				move_data_page(inode, start_bidx, gc_type,
922 								segno, off);
923 
924 			if (locked) {
925 				up_write(&fi->i_gc_rwsem[WRITE]);
926 				up_write(&fi->i_gc_rwsem[READ]);
927 			}
928 
929 			stat_inc_data_blk_count(sbi, 1, gc_type);
930 		}
931 	}
932 
933 	if (++phase < 5)
934 		goto next_step;
935 }
936 
937 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
938 			int gc_type)
939 {
940 	struct sit_info *sit_i = SIT_I(sbi);
941 	int ret;
942 
943 	down_write(&sit_i->sentry_lock);
944 	ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
945 					      NO_CHECK_TYPE, LFS);
946 	up_write(&sit_i->sentry_lock);
947 	return ret;
948 }
949 
950 static int do_garbage_collect(struct f2fs_sb_info *sbi,
951 				unsigned int start_segno,
952 				struct gc_inode_list *gc_list, int gc_type)
953 {
954 	struct page *sum_page;
955 	struct f2fs_summary_block *sum;
956 	struct blk_plug plug;
957 	unsigned int segno = start_segno;
958 	unsigned int end_segno = start_segno + sbi->segs_per_sec;
959 	int seg_freed = 0;
960 	unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
961 						SUM_TYPE_DATA : SUM_TYPE_NODE;
962 
963 	/* readahead multi ssa blocks those have contiguous address */
964 	if (sbi->segs_per_sec > 1)
965 		f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
966 					sbi->segs_per_sec, META_SSA, true);
967 
968 	/* reference all summary page */
969 	while (segno < end_segno) {
970 		sum_page = f2fs_get_sum_page(sbi, segno++);
971 		unlock_page(sum_page);
972 	}
973 
974 	blk_start_plug(&plug);
975 
976 	for (segno = start_segno; segno < end_segno; segno++) {
977 
978 		/* find segment summary of victim */
979 		sum_page = find_get_page(META_MAPPING(sbi),
980 					GET_SUM_BLOCK(sbi, segno));
981 		f2fs_put_page(sum_page, 0);
982 
983 		if (get_valid_blocks(sbi, segno, false) == 0 ||
984 				!PageUptodate(sum_page) ||
985 				unlikely(f2fs_cp_error(sbi)))
986 			goto next;
987 
988 		sum = page_address(sum_page);
989 		f2fs_bug_on(sbi, type != GET_SUM_TYPE((&sum->footer)));
990 
991 		/*
992 		 * this is to avoid deadlock:
993 		 * - lock_page(sum_page)         - f2fs_replace_block
994 		 *  - check_valid_map()            - down_write(sentry_lock)
995 		 *   - down_read(sentry_lock)     - change_curseg()
996 		 *                                  - lock_page(sum_page)
997 		 */
998 		if (type == SUM_TYPE_NODE)
999 			gc_node_segment(sbi, sum->entries, segno, gc_type);
1000 		else
1001 			gc_data_segment(sbi, sum->entries, gc_list, segno,
1002 								gc_type);
1003 
1004 		stat_inc_seg_count(sbi, type, gc_type);
1005 
1006 		if (gc_type == FG_GC &&
1007 				get_valid_blocks(sbi, segno, false) == 0)
1008 			seg_freed++;
1009 next:
1010 		f2fs_put_page(sum_page, 0);
1011 	}
1012 
1013 	if (gc_type == FG_GC)
1014 		f2fs_submit_merged_write(sbi,
1015 				(type == SUM_TYPE_NODE) ? NODE : DATA);
1016 
1017 	blk_finish_plug(&plug);
1018 
1019 	stat_inc_call_count(sbi->stat_info);
1020 
1021 	return seg_freed;
1022 }
1023 
1024 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1025 			bool background, unsigned int segno)
1026 {
1027 	int gc_type = sync ? FG_GC : BG_GC;
1028 	int sec_freed = 0, seg_freed = 0, total_freed = 0;
1029 	int ret = 0;
1030 	struct cp_control cpc;
1031 	unsigned int init_segno = segno;
1032 	struct gc_inode_list gc_list = {
1033 		.ilist = LIST_HEAD_INIT(gc_list.ilist),
1034 		.iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1035 	};
1036 	unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1037 	unsigned int skipped_round = 0, round = 0;
1038 
1039 	trace_f2fs_gc_begin(sbi->sb, sync, background,
1040 				get_pages(sbi, F2FS_DIRTY_NODES),
1041 				get_pages(sbi, F2FS_DIRTY_DENTS),
1042 				get_pages(sbi, F2FS_DIRTY_IMETA),
1043 				free_sections(sbi),
1044 				free_segments(sbi),
1045 				reserved_segments(sbi),
1046 				prefree_segments(sbi));
1047 
1048 	cpc.reason = __get_cp_reason(sbi);
1049 gc_more:
1050 	if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1051 		ret = -EINVAL;
1052 		goto stop;
1053 	}
1054 	if (unlikely(f2fs_cp_error(sbi))) {
1055 		ret = -EIO;
1056 		goto stop;
1057 	}
1058 
1059 	if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1060 		/*
1061 		 * For example, if there are many prefree_segments below given
1062 		 * threshold, we can make them free by checkpoint. Then, we
1063 		 * secure free segments which doesn't need fggc any more.
1064 		 */
1065 		if (prefree_segments(sbi)) {
1066 			ret = f2fs_write_checkpoint(sbi, &cpc);
1067 			if (ret)
1068 				goto stop;
1069 		}
1070 		if (has_not_enough_free_secs(sbi, 0, 0))
1071 			gc_type = FG_GC;
1072 	}
1073 
1074 	/* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1075 	if (gc_type == BG_GC && !background) {
1076 		ret = -EINVAL;
1077 		goto stop;
1078 	}
1079 	if (!__get_victim(sbi, &segno, gc_type)) {
1080 		ret = -ENODATA;
1081 		goto stop;
1082 	}
1083 
1084 	seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type);
1085 	if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec)
1086 		sec_freed++;
1087 	total_freed += seg_freed;
1088 
1089 	if (gc_type == FG_GC) {
1090 		if (sbi->skipped_atomic_files[FG_GC] > last_skipped)
1091 			skipped_round++;
1092 		last_skipped = sbi->skipped_atomic_files[FG_GC];
1093 		round++;
1094 	}
1095 
1096 	if (gc_type == FG_GC)
1097 		sbi->cur_victim_sec = NULL_SEGNO;
1098 
1099 	if (!sync) {
1100 		if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1101 			if (skipped_round > MAX_SKIP_ATOMIC_COUNT &&
1102 				skipped_round * 2 >= round)
1103 				f2fs_drop_inmem_pages_all(sbi, true);
1104 			segno = NULL_SEGNO;
1105 			goto gc_more;
1106 		}
1107 
1108 		if (gc_type == FG_GC)
1109 			ret = f2fs_write_checkpoint(sbi, &cpc);
1110 	}
1111 stop:
1112 	SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1113 	SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1114 
1115 	trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1116 				get_pages(sbi, F2FS_DIRTY_NODES),
1117 				get_pages(sbi, F2FS_DIRTY_DENTS),
1118 				get_pages(sbi, F2FS_DIRTY_IMETA),
1119 				free_sections(sbi),
1120 				free_segments(sbi),
1121 				reserved_segments(sbi),
1122 				prefree_segments(sbi));
1123 
1124 	mutex_unlock(&sbi->gc_mutex);
1125 
1126 	put_gc_inode(&gc_list);
1127 
1128 	if (sync)
1129 		ret = sec_freed ? 0 : -EAGAIN;
1130 	return ret;
1131 }
1132 
1133 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1134 {
1135 	DIRTY_I(sbi)->v_ops = &default_v_ops;
1136 
1137 	sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1138 
1139 	/* give warm/cold data area from slower device */
1140 	if (sbi->s_ndevs && sbi->segs_per_sec == 1)
1141 		SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1142 				GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1143 }
1144