xref: /linux/fs/f2fs/gc.c (revision b6ebbac51bedf9e98e837688bc838f400196da5e)
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 	long wait_ms;
32 
33 	wait_ms = gc_th->min_sleep_time;
34 
35 	do {
36 		if (try_to_freeze())
37 			continue;
38 		else
39 			wait_event_interruptible_timeout(*wq,
40 						kthread_should_stop(),
41 						msecs_to_jiffies(wait_ms));
42 		if (kthread_should_stop())
43 			break;
44 
45 		if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
46 			increase_sleep_time(gc_th, &wait_ms);
47 			continue;
48 		}
49 
50 		/*
51 		 * [GC triggering condition]
52 		 * 0. GC is not conducted currently.
53 		 * 1. There are enough dirty segments.
54 		 * 2. IO subsystem is idle by checking the # of writeback pages.
55 		 * 3. IO subsystem is idle by checking the # of requests in
56 		 *    bdev's request list.
57 		 *
58 		 * Note) We have to avoid triggering GCs frequently.
59 		 * Because it is possible that some segments can be
60 		 * invalidated soon after by user update or deletion.
61 		 * So, I'd like to wait some time to collect dirty segments.
62 		 */
63 		if (!mutex_trylock(&sbi->gc_mutex))
64 			continue;
65 
66 		if (!is_idle(sbi)) {
67 			increase_sleep_time(gc_th, &wait_ms);
68 			mutex_unlock(&sbi->gc_mutex);
69 			continue;
70 		}
71 
72 		if (has_enough_invalid_blocks(sbi))
73 			decrease_sleep_time(gc_th, &wait_ms);
74 		else
75 			increase_sleep_time(gc_th, &wait_ms);
76 
77 		stat_inc_bggc_count(sbi);
78 
79 		/* if return value is not zero, no victim was selected */
80 		if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC)))
81 			wait_ms = gc_th->no_gc_sleep_time;
82 
83 		trace_f2fs_background_gc(sbi->sb, wait_ms,
84 				prefree_segments(sbi), free_segments(sbi));
85 
86 		/* balancing f2fs's metadata periodically */
87 		f2fs_balance_fs_bg(sbi);
88 
89 	} while (!kthread_should_stop());
90 	return 0;
91 }
92 
93 int start_gc_thread(struct f2fs_sb_info *sbi)
94 {
95 	struct f2fs_gc_kthread *gc_th;
96 	dev_t dev = sbi->sb->s_bdev->bd_dev;
97 	int err = 0;
98 
99 	gc_th = f2fs_kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
100 	if (!gc_th) {
101 		err = -ENOMEM;
102 		goto out;
103 	}
104 
105 	gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
106 	gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
107 	gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
108 
109 	gc_th->gc_idle = 0;
110 
111 	sbi->gc_thread = gc_th;
112 	init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
113 	sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
114 			"f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
115 	if (IS_ERR(gc_th->f2fs_gc_task)) {
116 		err = PTR_ERR(gc_th->f2fs_gc_task);
117 		kfree(gc_th);
118 		sbi->gc_thread = NULL;
119 	}
120 out:
121 	return err;
122 }
123 
124 void stop_gc_thread(struct f2fs_sb_info *sbi)
125 {
126 	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
127 	if (!gc_th)
128 		return;
129 	kthread_stop(gc_th->f2fs_gc_task);
130 	kfree(gc_th);
131 	sbi->gc_thread = NULL;
132 }
133 
134 static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
135 {
136 	int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
137 
138 	if (gc_th && gc_th->gc_idle) {
139 		if (gc_th->gc_idle == 1)
140 			gc_mode = GC_CB;
141 		else if (gc_th->gc_idle == 2)
142 			gc_mode = GC_GREEDY;
143 	}
144 	return gc_mode;
145 }
146 
147 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
148 			int type, struct victim_sel_policy *p)
149 {
150 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
151 
152 	if (p->alloc_mode == SSR) {
153 		p->gc_mode = GC_GREEDY;
154 		p->dirty_segmap = dirty_i->dirty_segmap[type];
155 		p->max_search = dirty_i->nr_dirty[type];
156 		p->ofs_unit = 1;
157 	} else {
158 		p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
159 		p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
160 		p->max_search = dirty_i->nr_dirty[DIRTY];
161 		p->ofs_unit = sbi->segs_per_sec;
162 	}
163 
164 	if (p->max_search > sbi->max_victim_search)
165 		p->max_search = sbi->max_victim_search;
166 
167 	p->offset = sbi->last_victim[p->gc_mode];
168 }
169 
170 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
171 				struct victim_sel_policy *p)
172 {
173 	/* SSR allocates in a segment unit */
174 	if (p->alloc_mode == SSR)
175 		return sbi->blocks_per_seg;
176 	if (p->gc_mode == GC_GREEDY)
177 		return sbi->blocks_per_seg * p->ofs_unit;
178 	else if (p->gc_mode == GC_CB)
179 		return UINT_MAX;
180 	else /* No other gc_mode */
181 		return 0;
182 }
183 
184 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
185 {
186 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
187 	unsigned int secno;
188 
189 	/*
190 	 * If the gc_type is FG_GC, we can select victim segments
191 	 * selected by background GC before.
192 	 * Those segments guarantee they have small valid blocks.
193 	 */
194 	for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
195 		if (sec_usage_check(sbi, secno))
196 			continue;
197 		clear_bit(secno, dirty_i->victim_secmap);
198 		return secno * sbi->segs_per_sec;
199 	}
200 	return NULL_SEGNO;
201 }
202 
203 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
204 {
205 	struct sit_info *sit_i = SIT_I(sbi);
206 	unsigned int secno = GET_SECNO(sbi, segno);
207 	unsigned int start = secno * sbi->segs_per_sec;
208 	unsigned long long mtime = 0;
209 	unsigned int vblocks;
210 	unsigned char age = 0;
211 	unsigned char u;
212 	unsigned int i;
213 
214 	for (i = 0; i < sbi->segs_per_sec; i++)
215 		mtime += get_seg_entry(sbi, start + i)->mtime;
216 	vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec);
217 
218 	mtime = div_u64(mtime, sbi->segs_per_sec);
219 	vblocks = div_u64(vblocks, sbi->segs_per_sec);
220 
221 	u = (vblocks * 100) >> sbi->log_blocks_per_seg;
222 
223 	/* Handle if the system time has changed by the user */
224 	if (mtime < sit_i->min_mtime)
225 		sit_i->min_mtime = mtime;
226 	if (mtime > sit_i->max_mtime)
227 		sit_i->max_mtime = mtime;
228 	if (sit_i->max_mtime != sit_i->min_mtime)
229 		age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
230 				sit_i->max_mtime - sit_i->min_mtime);
231 
232 	return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
233 }
234 
235 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
236 			unsigned int segno, struct victim_sel_policy *p)
237 {
238 	if (p->alloc_mode == SSR)
239 		return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
240 
241 	/* alloc_mode == LFS */
242 	if (p->gc_mode == GC_GREEDY)
243 		return get_valid_blocks(sbi, segno, sbi->segs_per_sec);
244 	else
245 		return get_cb_cost(sbi, segno);
246 }
247 
248 static unsigned int count_bits(const unsigned long *addr,
249 				unsigned int offset, unsigned int len)
250 {
251 	unsigned int end = offset + len, sum = 0;
252 
253 	while (offset < end) {
254 		if (test_bit(offset++, addr))
255 			++sum;
256 	}
257 	return sum;
258 }
259 
260 /*
261  * This function is called from two paths.
262  * One is garbage collection and the other is SSR segment selection.
263  * When it is called during GC, it just gets a victim segment
264  * and it does not remove it from dirty seglist.
265  * When it is called from SSR segment selection, it finds a segment
266  * which has minimum valid blocks and removes it from dirty seglist.
267  */
268 static int get_victim_by_default(struct f2fs_sb_info *sbi,
269 		unsigned int *result, int gc_type, int type, char alloc_mode)
270 {
271 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
272 	struct victim_sel_policy p;
273 	unsigned int secno, max_cost, last_victim;
274 	unsigned int last_segment = MAIN_SEGS(sbi);
275 	unsigned int nsearched = 0;
276 
277 	mutex_lock(&dirty_i->seglist_lock);
278 
279 	p.alloc_mode = alloc_mode;
280 	select_policy(sbi, gc_type, type, &p);
281 
282 	p.min_segno = NULL_SEGNO;
283 	p.min_cost = max_cost = get_max_cost(sbi, &p);
284 
285 	if (p.max_search == 0)
286 		goto out;
287 
288 	last_victim = sbi->last_victim[p.gc_mode];
289 	if (p.alloc_mode == LFS && gc_type == FG_GC) {
290 		p.min_segno = check_bg_victims(sbi);
291 		if (p.min_segno != NULL_SEGNO)
292 			goto got_it;
293 	}
294 
295 	while (1) {
296 		unsigned long cost;
297 		unsigned int segno;
298 
299 		segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
300 		if (segno >= last_segment) {
301 			if (sbi->last_victim[p.gc_mode]) {
302 				last_segment = sbi->last_victim[p.gc_mode];
303 				sbi->last_victim[p.gc_mode] = 0;
304 				p.offset = 0;
305 				continue;
306 			}
307 			break;
308 		}
309 
310 		p.offset = segno + p.ofs_unit;
311 		if (p.ofs_unit > 1) {
312 			p.offset -= segno % p.ofs_unit;
313 			nsearched += count_bits(p.dirty_segmap,
314 						p.offset - p.ofs_unit,
315 						p.ofs_unit);
316 		} else {
317 			nsearched++;
318 		}
319 
320 
321 		secno = GET_SECNO(sbi, segno);
322 
323 		if (sec_usage_check(sbi, secno))
324 			goto next;
325 		if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
326 			goto next;
327 
328 		cost = get_gc_cost(sbi, segno, &p);
329 
330 		if (p.min_cost > cost) {
331 			p.min_segno = segno;
332 			p.min_cost = cost;
333 		}
334 next:
335 		if (nsearched >= p.max_search) {
336 			if (!sbi->last_victim[p.gc_mode] && segno <= last_victim)
337 				sbi->last_victim[p.gc_mode] = last_victim + 1;
338 			else
339 				sbi->last_victim[p.gc_mode] = segno + 1;
340 			break;
341 		}
342 	}
343 	if (p.min_segno != NULL_SEGNO) {
344 got_it:
345 		if (p.alloc_mode == LFS) {
346 			secno = GET_SECNO(sbi, p.min_segno);
347 			if (gc_type == FG_GC)
348 				sbi->cur_victim_sec = secno;
349 			else
350 				set_bit(secno, dirty_i->victim_secmap);
351 		}
352 		*result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
353 
354 		trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
355 				sbi->cur_victim_sec,
356 				prefree_segments(sbi), free_segments(sbi));
357 	}
358 out:
359 	mutex_unlock(&dirty_i->seglist_lock);
360 
361 	return (p.min_segno == NULL_SEGNO) ? 0 : 1;
362 }
363 
364 static const struct victim_selection default_v_ops = {
365 	.get_victim = get_victim_by_default,
366 };
367 
368 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
369 {
370 	struct inode_entry *ie;
371 
372 	ie = radix_tree_lookup(&gc_list->iroot, ino);
373 	if (ie)
374 		return ie->inode;
375 	return NULL;
376 }
377 
378 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
379 {
380 	struct inode_entry *new_ie;
381 
382 	if (inode == find_gc_inode(gc_list, inode->i_ino)) {
383 		iput(inode);
384 		return;
385 	}
386 	new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
387 	new_ie->inode = inode;
388 
389 	f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
390 	list_add_tail(&new_ie->list, &gc_list->ilist);
391 }
392 
393 static void put_gc_inode(struct gc_inode_list *gc_list)
394 {
395 	struct inode_entry *ie, *next_ie;
396 	list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
397 		radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
398 		iput(ie->inode);
399 		list_del(&ie->list);
400 		kmem_cache_free(inode_entry_slab, ie);
401 	}
402 }
403 
404 static int check_valid_map(struct f2fs_sb_info *sbi,
405 				unsigned int segno, int offset)
406 {
407 	struct sit_info *sit_i = SIT_I(sbi);
408 	struct seg_entry *sentry;
409 	int ret;
410 
411 	mutex_lock(&sit_i->sentry_lock);
412 	sentry = get_seg_entry(sbi, segno);
413 	ret = f2fs_test_bit(offset, sentry->cur_valid_map);
414 	mutex_unlock(&sit_i->sentry_lock);
415 	return ret;
416 }
417 
418 /*
419  * This function compares node address got in summary with that in NAT.
420  * On validity, copy that node with cold status, otherwise (invalid node)
421  * ignore that.
422  */
423 static void gc_node_segment(struct f2fs_sb_info *sbi,
424 		struct f2fs_summary *sum, unsigned int segno, int gc_type)
425 {
426 	bool initial = true;
427 	struct f2fs_summary *entry;
428 	block_t start_addr;
429 	int off;
430 
431 	start_addr = START_BLOCK(sbi, segno);
432 
433 next_step:
434 	entry = sum;
435 
436 	for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
437 		nid_t nid = le32_to_cpu(entry->nid);
438 		struct page *node_page;
439 		struct node_info ni;
440 
441 		/* stop BG_GC if there is not enough free sections. */
442 		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
443 			return;
444 
445 		if (check_valid_map(sbi, segno, off) == 0)
446 			continue;
447 
448 		if (initial) {
449 			ra_node_page(sbi, nid);
450 			continue;
451 		}
452 		node_page = get_node_page(sbi, nid);
453 		if (IS_ERR(node_page))
454 			continue;
455 
456 		/* block may become invalid during get_node_page */
457 		if (check_valid_map(sbi, segno, off) == 0) {
458 			f2fs_put_page(node_page, 1);
459 			continue;
460 		}
461 
462 		get_node_info(sbi, nid, &ni);
463 		if (ni.blk_addr != start_addr + off) {
464 			f2fs_put_page(node_page, 1);
465 			continue;
466 		}
467 
468 		move_node_page(node_page, gc_type);
469 		stat_inc_node_blk_count(sbi, 1, gc_type);
470 	}
471 
472 	if (initial) {
473 		initial = false;
474 		goto next_step;
475 	}
476 }
477 
478 /*
479  * Calculate start block index indicating the given node offset.
480  * Be careful, caller should give this node offset only indicating direct node
481  * blocks. If any node offsets, which point the other types of node blocks such
482  * as indirect or double indirect node blocks, are given, it must be a caller's
483  * bug.
484  */
485 block_t start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
486 {
487 	unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
488 	unsigned int bidx;
489 
490 	if (node_ofs == 0)
491 		return 0;
492 
493 	if (node_ofs <= 2) {
494 		bidx = node_ofs - 1;
495 	} else if (node_ofs <= indirect_blks) {
496 		int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
497 		bidx = node_ofs - 2 - dec;
498 	} else {
499 		int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
500 		bidx = node_ofs - 5 - dec;
501 	}
502 	return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode);
503 }
504 
505 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
506 		struct node_info *dni, block_t blkaddr, unsigned int *nofs)
507 {
508 	struct page *node_page;
509 	nid_t nid;
510 	unsigned int ofs_in_node;
511 	block_t source_blkaddr;
512 
513 	nid = le32_to_cpu(sum->nid);
514 	ofs_in_node = le16_to_cpu(sum->ofs_in_node);
515 
516 	node_page = get_node_page(sbi, nid);
517 	if (IS_ERR(node_page))
518 		return false;
519 
520 	get_node_info(sbi, nid, dni);
521 
522 	if (sum->version != dni->version) {
523 		f2fs_put_page(node_page, 1);
524 		return false;
525 	}
526 
527 	*nofs = ofs_of_node(node_page);
528 	source_blkaddr = datablock_addr(node_page, ofs_in_node);
529 	f2fs_put_page(node_page, 1);
530 
531 	if (source_blkaddr != blkaddr)
532 		return false;
533 	return true;
534 }
535 
536 static void move_encrypted_block(struct inode *inode, block_t bidx)
537 {
538 	struct f2fs_io_info fio = {
539 		.sbi = F2FS_I_SB(inode),
540 		.type = DATA,
541 		.op = REQ_OP_READ,
542 		.op_flags = READ_SYNC,
543 		.encrypted_page = NULL,
544 	};
545 	struct dnode_of_data dn;
546 	struct f2fs_summary sum;
547 	struct node_info ni;
548 	struct page *page;
549 	block_t newaddr;
550 	int err;
551 
552 	/* do not read out */
553 	page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
554 	if (!page)
555 		return;
556 
557 	set_new_dnode(&dn, inode, NULL, NULL, 0);
558 	err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
559 	if (err)
560 		goto out;
561 
562 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
563 		ClearPageUptodate(page);
564 		goto put_out;
565 	}
566 
567 	/*
568 	 * don't cache encrypted data into meta inode until previous dirty
569 	 * data were writebacked to avoid racing between GC and flush.
570 	 */
571 	f2fs_wait_on_page_writeback(page, DATA, true);
572 
573 	get_node_info(fio.sbi, dn.nid, &ni);
574 	set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
575 
576 	/* read page */
577 	fio.page = page;
578 	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
579 
580 	allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
581 							&sum, CURSEG_COLD_DATA);
582 
583 	fio.encrypted_page = pagecache_get_page(META_MAPPING(fio.sbi), newaddr,
584 					FGP_LOCK | FGP_CREAT, GFP_NOFS);
585 	if (!fio.encrypted_page) {
586 		err = -ENOMEM;
587 		goto recover_block;
588 	}
589 
590 	err = f2fs_submit_page_bio(&fio);
591 	if (err)
592 		goto put_page_out;
593 
594 	/* write page */
595 	lock_page(fio.encrypted_page);
596 
597 	if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) {
598 		err = -EIO;
599 		goto put_page_out;
600 	}
601 	if (unlikely(!PageUptodate(fio.encrypted_page))) {
602 		err = -EIO;
603 		goto put_page_out;
604 	}
605 
606 	set_page_dirty(fio.encrypted_page);
607 	f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true);
608 	if (clear_page_dirty_for_io(fio.encrypted_page))
609 		dec_page_count(fio.sbi, F2FS_DIRTY_META);
610 
611 	set_page_writeback(fio.encrypted_page);
612 
613 	/* allocate block address */
614 	f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
615 
616 	fio.op = REQ_OP_WRITE;
617 	fio.op_flags = WRITE_SYNC;
618 	fio.new_blkaddr = newaddr;
619 	f2fs_submit_page_mbio(&fio);
620 
621 	f2fs_update_data_blkaddr(&dn, newaddr);
622 	set_inode_flag(inode, FI_APPEND_WRITE);
623 	if (page->index == 0)
624 		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
625 put_page_out:
626 	f2fs_put_page(fio.encrypted_page, 1);
627 recover_block:
628 	if (err)
629 		__f2fs_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
630 								true, true);
631 put_out:
632 	f2fs_put_dnode(&dn);
633 out:
634 	f2fs_put_page(page, 1);
635 }
636 
637 static void move_data_page(struct inode *inode, block_t bidx, int gc_type)
638 {
639 	struct page *page;
640 
641 	page = get_lock_data_page(inode, bidx, true);
642 	if (IS_ERR(page))
643 		return;
644 
645 	if (gc_type == BG_GC) {
646 		if (PageWriteback(page))
647 			goto out;
648 		set_page_dirty(page);
649 		set_cold_data(page);
650 	} else {
651 		struct f2fs_io_info fio = {
652 			.sbi = F2FS_I_SB(inode),
653 			.type = DATA,
654 			.op = REQ_OP_WRITE,
655 			.op_flags = WRITE_SYNC,
656 			.page = page,
657 			.encrypted_page = NULL,
658 		};
659 		bool is_dirty = PageDirty(page);
660 		int err;
661 
662 retry:
663 		set_page_dirty(page);
664 		f2fs_wait_on_page_writeback(page, DATA, true);
665 		if (clear_page_dirty_for_io(page))
666 			inode_dec_dirty_pages(inode);
667 
668 		set_cold_data(page);
669 
670 		err = do_write_data_page(&fio);
671 		if (err == -ENOMEM && is_dirty) {
672 			congestion_wait(BLK_RW_ASYNC, HZ/50);
673 			goto retry;
674 		}
675 
676 		clear_cold_data(page);
677 	}
678 out:
679 	f2fs_put_page(page, 1);
680 }
681 
682 /*
683  * This function tries to get parent node of victim data block, and identifies
684  * data block validity. If the block is valid, copy that with cold status and
685  * modify parent node.
686  * If the parent node is not valid or the data block address is different,
687  * the victim data block is ignored.
688  */
689 static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
690 		struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
691 {
692 	struct super_block *sb = sbi->sb;
693 	struct f2fs_summary *entry;
694 	block_t start_addr;
695 	int off;
696 	int phase = 0;
697 
698 	start_addr = START_BLOCK(sbi, segno);
699 
700 next_step:
701 	entry = sum;
702 
703 	for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
704 		struct page *data_page;
705 		struct inode *inode;
706 		struct node_info dni; /* dnode info for the data */
707 		unsigned int ofs_in_node, nofs;
708 		block_t start_bidx;
709 
710 		/* stop BG_GC if there is not enough free sections. */
711 		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
712 			return;
713 
714 		if (check_valid_map(sbi, segno, off) == 0)
715 			continue;
716 
717 		if (phase == 0) {
718 			ra_node_page(sbi, le32_to_cpu(entry->nid));
719 			continue;
720 		}
721 
722 		/* Get an inode by ino with checking validity */
723 		if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
724 			continue;
725 
726 		if (phase == 1) {
727 			ra_node_page(sbi, dni.ino);
728 			continue;
729 		}
730 
731 		ofs_in_node = le16_to_cpu(entry->ofs_in_node);
732 
733 		if (phase == 2) {
734 			inode = f2fs_iget(sb, dni.ino);
735 			if (IS_ERR(inode) || is_bad_inode(inode))
736 				continue;
737 
738 			/* if encrypted inode, let's go phase 3 */
739 			if (f2fs_encrypted_inode(inode) &&
740 						S_ISREG(inode->i_mode)) {
741 				add_gc_inode(gc_list, inode);
742 				continue;
743 			}
744 
745 			start_bidx = start_bidx_of_node(nofs, inode);
746 			data_page = get_read_data_page(inode,
747 					start_bidx + ofs_in_node, REQ_RAHEAD,
748 					true);
749 			if (IS_ERR(data_page)) {
750 				iput(inode);
751 				continue;
752 			}
753 
754 			f2fs_put_page(data_page, 0);
755 			add_gc_inode(gc_list, inode);
756 			continue;
757 		}
758 
759 		/* phase 3 */
760 		inode = find_gc_inode(gc_list, dni.ino);
761 		if (inode) {
762 			struct f2fs_inode_info *fi = F2FS_I(inode);
763 			bool locked = false;
764 
765 			if (S_ISREG(inode->i_mode)) {
766 				if (!down_write_trylock(&fi->dio_rwsem[READ]))
767 					continue;
768 				if (!down_write_trylock(
769 						&fi->dio_rwsem[WRITE])) {
770 					up_write(&fi->dio_rwsem[READ]);
771 					continue;
772 				}
773 				locked = true;
774 			}
775 
776 			start_bidx = start_bidx_of_node(nofs, inode)
777 								+ ofs_in_node;
778 			if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
779 				move_encrypted_block(inode, start_bidx);
780 			else
781 				move_data_page(inode, start_bidx, gc_type);
782 
783 			if (locked) {
784 				up_write(&fi->dio_rwsem[WRITE]);
785 				up_write(&fi->dio_rwsem[READ]);
786 			}
787 
788 			stat_inc_data_blk_count(sbi, 1, gc_type);
789 		}
790 	}
791 
792 	if (++phase < 4)
793 		goto next_step;
794 }
795 
796 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
797 			int gc_type)
798 {
799 	struct sit_info *sit_i = SIT_I(sbi);
800 	int ret;
801 
802 	mutex_lock(&sit_i->sentry_lock);
803 	ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
804 					      NO_CHECK_TYPE, LFS);
805 	mutex_unlock(&sit_i->sentry_lock);
806 	return ret;
807 }
808 
809 static int do_garbage_collect(struct f2fs_sb_info *sbi,
810 				unsigned int start_segno,
811 				struct gc_inode_list *gc_list, int gc_type)
812 {
813 	struct page *sum_page;
814 	struct f2fs_summary_block *sum;
815 	struct blk_plug plug;
816 	unsigned int segno = start_segno;
817 	unsigned int end_segno = start_segno + sbi->segs_per_sec;
818 	int seg_freed = 0;
819 	unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
820 						SUM_TYPE_DATA : SUM_TYPE_NODE;
821 
822 	/* readahead multi ssa blocks those have contiguous address */
823 	if (sbi->segs_per_sec > 1)
824 		ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
825 					sbi->segs_per_sec, META_SSA, true);
826 
827 	/* reference all summary page */
828 	while (segno < end_segno) {
829 		sum_page = get_sum_page(sbi, segno++);
830 		unlock_page(sum_page);
831 	}
832 
833 	blk_start_plug(&plug);
834 
835 	for (segno = start_segno; segno < end_segno; segno++) {
836 
837 		if (get_valid_blocks(sbi, segno, 1) == 0)
838 			continue;
839 
840 		/* find segment summary of victim */
841 		sum_page = find_get_page(META_MAPPING(sbi),
842 					GET_SUM_BLOCK(sbi, segno));
843 		f2fs_bug_on(sbi, !PageUptodate(sum_page));
844 		f2fs_put_page(sum_page, 0);
845 
846 		sum = page_address(sum_page);
847 		f2fs_bug_on(sbi, type != GET_SUM_TYPE((&sum->footer)));
848 
849 		/*
850 		 * this is to avoid deadlock:
851 		 * - lock_page(sum_page)         - f2fs_replace_block
852 		 *  - check_valid_map()            - mutex_lock(sentry_lock)
853 		 *   - mutex_lock(sentry_lock)     - change_curseg()
854 		 *                                  - lock_page(sum_page)
855 		 */
856 
857 		if (type == SUM_TYPE_NODE)
858 			gc_node_segment(sbi, sum->entries, segno, gc_type);
859 		else
860 			gc_data_segment(sbi, sum->entries, gc_list, segno,
861 								gc_type);
862 
863 		stat_inc_seg_count(sbi, type, gc_type);
864 
865 		f2fs_put_page(sum_page, 0);
866 	}
867 
868 	if (gc_type == FG_GC)
869 		f2fs_submit_merged_bio(sbi,
870 				(type == SUM_TYPE_NODE) ? NODE : DATA, WRITE);
871 
872 	blk_finish_plug(&plug);
873 
874 	if (gc_type == FG_GC) {
875 		while (start_segno < end_segno)
876 			if (get_valid_blocks(sbi, start_segno++, 1) == 0)
877 				seg_freed++;
878 	}
879 
880 	stat_inc_call_count(sbi->stat_info);
881 
882 	return seg_freed;
883 }
884 
885 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync)
886 {
887 	unsigned int segno;
888 	int gc_type = sync ? FG_GC : BG_GC;
889 	int sec_freed = 0, seg_freed;
890 	int ret = -EINVAL;
891 	struct cp_control cpc;
892 	struct gc_inode_list gc_list = {
893 		.ilist = LIST_HEAD_INIT(gc_list.ilist),
894 		.iroot = RADIX_TREE_INIT(GFP_NOFS),
895 	};
896 
897 	cpc.reason = __get_cp_reason(sbi);
898 gc_more:
899 	segno = NULL_SEGNO;
900 
901 	if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE)))
902 		goto stop;
903 	if (unlikely(f2fs_cp_error(sbi))) {
904 		ret = -EIO;
905 		goto stop;
906 	}
907 
908 	if (gc_type == BG_GC && has_not_enough_free_secs(sbi, sec_freed)) {
909 		gc_type = FG_GC;
910 		/*
911 		 * If there is no victim and no prefree segment but still not
912 		 * enough free sections, we should flush dent/node blocks and do
913 		 * garbage collections.
914 		 */
915 		if (__get_victim(sbi, &segno, gc_type) ||
916 						prefree_segments(sbi)) {
917 			write_checkpoint(sbi, &cpc);
918 			segno = NULL_SEGNO;
919 		} else if (has_not_enough_free_secs(sbi, 0)) {
920 			write_checkpoint(sbi, &cpc);
921 		}
922 	}
923 
924 	if (segno == NULL_SEGNO && !__get_victim(sbi, &segno, gc_type))
925 		goto stop;
926 	ret = 0;
927 
928 	seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type);
929 
930 	if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec)
931 		sec_freed++;
932 
933 	if (gc_type == FG_GC)
934 		sbi->cur_victim_sec = NULL_SEGNO;
935 
936 	if (!sync) {
937 		if (has_not_enough_free_secs(sbi, sec_freed))
938 			goto gc_more;
939 
940 		if (gc_type == FG_GC)
941 			write_checkpoint(sbi, &cpc);
942 	}
943 stop:
944 	mutex_unlock(&sbi->gc_mutex);
945 
946 	put_gc_inode(&gc_list);
947 
948 	if (sync)
949 		ret = sec_freed ? 0 : -EAGAIN;
950 	return ret;
951 }
952 
953 void build_gc_manager(struct f2fs_sb_info *sbi)
954 {
955 	DIRTY_I(sbi)->v_ops = &default_v_ops;
956 }
957