960626715647fbb443df60715a31426cb584c73e
[linux-3.10.git] / fs / f2fs / gc.c
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 #include <linux/blkdev.h>
20
21 #include "f2fs.h"
22 #include "node.h"
23 #include "segment.h"
24 #include "gc.h"
25 #include <trace/events/f2fs.h>
26
27 static struct kmem_cache *winode_slab;
28
29 static int gc_thread_func(void *data)
30 {
31         struct f2fs_sb_info *sbi = data;
32         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
33         wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
34         long wait_ms;
35
36         wait_ms = gc_th->min_sleep_time;
37
38         do {
39                 if (try_to_freeze())
40                         continue;
41                 else
42                         wait_event_interruptible_timeout(*wq,
43                                                 kthread_should_stop(),
44                                                 msecs_to_jiffies(wait_ms));
45                 if (kthread_should_stop())
46                         break;
47
48                 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
49                         wait_ms = gc_th->max_sleep_time;
50                         continue;
51                 }
52
53                 /*
54                  * [GC triggering condition]
55                  * 0. GC is not conducted currently.
56                  * 1. There are enough dirty segments.
57                  * 2. IO subsystem is idle by checking the # of writeback pages.
58                  * 3. IO subsystem is idle by checking the # of requests in
59                  *    bdev's request list.
60                  *
61                  * Note) We have to avoid triggering GCs too much frequently.
62                  * Because it is possible that some segments can be
63                  * invalidated soon after by user update or deletion.
64                  * So, I'd like to wait some time to collect dirty segments.
65                  */
66                 if (!mutex_trylock(&sbi->gc_mutex))
67                         continue;
68
69                 if (!is_idle(sbi)) {
70                         wait_ms = increase_sleep_time(gc_th, wait_ms);
71                         mutex_unlock(&sbi->gc_mutex);
72                         continue;
73                 }
74
75                 if (has_enough_invalid_blocks(sbi))
76                         wait_ms = decrease_sleep_time(gc_th, wait_ms);
77                 else
78                         wait_ms = increase_sleep_time(gc_th, wait_ms);
79
80                 stat_inc_bggc_count(sbi);
81
82                 /* if return value is not zero, no victim was selected */
83                 if (f2fs_gc(sbi))
84                         wait_ms = gc_th->no_gc_sleep_time;
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         if (!test_opt(sbi, BG_GC))
100                 goto out;
101         gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
102         if (!gc_th) {
103                 err = -ENOMEM;
104                 goto out;
105         }
106
107         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
108         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
109         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
110
111         gc_th->gc_idle = 0;
112
113         sbi->gc_thread = gc_th;
114         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
115         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
116                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
117         if (IS_ERR(gc_th->f2fs_gc_task)) {
118                 err = PTR_ERR(gc_th->f2fs_gc_task);
119                 kfree(gc_th);
120                 sbi->gc_thread = NULL;
121         }
122 out:
123         return err;
124 }
125
126 void stop_gc_thread(struct f2fs_sb_info *sbi)
127 {
128         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
129         if (!gc_th)
130                 return;
131         kthread_stop(gc_th->f2fs_gc_task);
132         kfree(gc_th);
133         sbi->gc_thread = NULL;
134 }
135
136 static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
137 {
138         int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
139
140         if (gc_th && gc_th->gc_idle) {
141                 if (gc_th->gc_idle == 1)
142                         gc_mode = GC_CB;
143                 else if (gc_th->gc_idle == 2)
144                         gc_mode = GC_GREEDY;
145         }
146         return gc_mode;
147 }
148
149 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
150                         int type, struct victim_sel_policy *p)
151 {
152         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
153
154         if (p->alloc_mode == SSR) {
155                 p->gc_mode = GC_GREEDY;
156                 p->dirty_segmap = dirty_i->dirty_segmap[type];
157                 p->max_search = dirty_i->nr_dirty[type];
158                 p->ofs_unit = 1;
159         } else {
160                 p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
161                 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
162                 p->max_search = dirty_i->nr_dirty[DIRTY];
163                 p->ofs_unit = sbi->segs_per_sec;
164         }
165
166         if (p->max_search > sbi->max_victim_search)
167                 p->max_search = sbi->max_victim_search;
168
169         p->offset = sbi->last_victim[p->gc_mode];
170 }
171
172 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
173                                 struct victim_sel_policy *p)
174 {
175         /* SSR allocates in a segment unit */
176         if (p->alloc_mode == SSR)
177                 return 1 << sbi->log_blocks_per_seg;
178         if (p->gc_mode == GC_GREEDY)
179                 return (1 << sbi->log_blocks_per_seg) * p->ofs_unit;
180         else if (p->gc_mode == GC_CB)
181                 return UINT_MAX;
182         else /* No other gc_mode */
183                 return 0;
184 }
185
186 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
187 {
188         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
189         unsigned int secno;
190
191         /*
192          * If the gc_type is FG_GC, we can select victim segments
193          * selected by background GC before.
194          * Those segments guarantee they have small valid blocks.
195          */
196         for_each_set_bit(secno, dirty_i->victim_secmap, TOTAL_SECS(sbi)) {
197                 if (sec_usage_check(sbi, secno))
198                         continue;
199                 clear_bit(secno, dirty_i->victim_secmap);
200                 return secno * sbi->segs_per_sec;
201         }
202         return NULL_SEGNO;
203 }
204
205 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
206 {
207         struct sit_info *sit_i = SIT_I(sbi);
208         unsigned int secno = GET_SECNO(sbi, segno);
209         unsigned int start = secno * sbi->segs_per_sec;
210         unsigned long long mtime = 0;
211         unsigned int vblocks;
212         unsigned char age = 0;
213         unsigned char u;
214         unsigned int i;
215
216         for (i = 0; i < sbi->segs_per_sec; i++)
217                 mtime += get_seg_entry(sbi, start + i)->mtime;
218         vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec);
219
220         mtime = div_u64(mtime, sbi->segs_per_sec);
221         vblocks = div_u64(vblocks, sbi->segs_per_sec);
222
223         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
224
225         /* Handle if the system time is changed by user */
226         if (mtime < sit_i->min_mtime)
227                 sit_i->min_mtime = mtime;
228         if (mtime > sit_i->max_mtime)
229                 sit_i->max_mtime = mtime;
230         if (sit_i->max_mtime != sit_i->min_mtime)
231                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
232                                 sit_i->max_mtime - sit_i->min_mtime);
233
234         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
235 }
236
237 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
238                         unsigned int segno, struct victim_sel_policy *p)
239 {
240         if (p->alloc_mode == SSR)
241                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
242
243         /* alloc_mode == LFS */
244         if (p->gc_mode == GC_GREEDY)
245                 return get_valid_blocks(sbi, segno, sbi->segs_per_sec);
246         else
247                 return get_cb_cost(sbi, segno);
248 }
249
250 /*
251  * This function is called from two paths.
252  * One is garbage collection and the other is SSR segment selection.
253  * When it is called during GC, it just gets a victim segment
254  * and it does not remove it from dirty seglist.
255  * When it is called from SSR segment selection, it finds a segment
256  * which has minimum valid blocks and removes it from dirty seglist.
257  */
258 static int get_victim_by_default(struct f2fs_sb_info *sbi,
259                 unsigned int *result, int gc_type, int type, char alloc_mode)
260 {
261         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
262         struct victim_sel_policy p;
263         unsigned int secno, max_cost;
264         int nsearched = 0;
265
266         mutex_lock(&dirty_i->seglist_lock);
267
268         p.alloc_mode = alloc_mode;
269         select_policy(sbi, gc_type, type, &p);
270
271         p.min_segno = NULL_SEGNO;
272         p.min_cost = max_cost = get_max_cost(sbi, &p);
273
274         if (p.alloc_mode == LFS && gc_type == FG_GC) {
275                 p.min_segno = check_bg_victims(sbi);
276                 if (p.min_segno != NULL_SEGNO)
277                         goto got_it;
278         }
279
280         while (1) {
281                 unsigned long cost;
282                 unsigned int segno;
283
284                 segno = find_next_bit(p.dirty_segmap,
285                                                 TOTAL_SEGS(sbi), p.offset);
286                 if (segno >= TOTAL_SEGS(sbi)) {
287                         if (sbi->last_victim[p.gc_mode]) {
288                                 sbi->last_victim[p.gc_mode] = 0;
289                                 p.offset = 0;
290                                 continue;
291                         }
292                         break;
293                 }
294
295                 p.offset = segno + p.ofs_unit;
296                 if (p.ofs_unit > 1)
297                         p.offset -= segno % p.ofs_unit;
298
299                 secno = GET_SECNO(sbi, segno);
300
301                 if (sec_usage_check(sbi, secno))
302                         continue;
303                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
304                         continue;
305
306                 cost = get_gc_cost(sbi, segno, &p);
307
308                 if (p.min_cost > cost) {
309                         p.min_segno = segno;
310                         p.min_cost = cost;
311                 } else if (unlikely(cost == max_cost)) {
312                         continue;
313                 }
314
315                 if (nsearched++ >= p.max_search) {
316                         sbi->last_victim[p.gc_mode] = segno;
317                         break;
318                 }
319         }
320         if (p.min_segno != NULL_SEGNO) {
321 got_it:
322                 if (p.alloc_mode == LFS) {
323                         secno = GET_SECNO(sbi, p.min_segno);
324                         if (gc_type == FG_GC)
325                                 sbi->cur_victim_sec = secno;
326                         else
327                                 set_bit(secno, dirty_i->victim_secmap);
328                 }
329                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
330
331                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
332                                 sbi->cur_victim_sec,
333                                 prefree_segments(sbi), free_segments(sbi));
334         }
335         mutex_unlock(&dirty_i->seglist_lock);
336
337         return (p.min_segno == NULL_SEGNO) ? 0 : 1;
338 }
339
340 static const struct victim_selection default_v_ops = {
341         .get_victim = get_victim_by_default,
342 };
343
344 static struct inode *find_gc_inode(nid_t ino, struct list_head *ilist)
345 {
346         struct inode_entry *ie;
347
348         list_for_each_entry(ie, ilist, list)
349                 if (ie->inode->i_ino == ino)
350                         return ie->inode;
351         return NULL;
352 }
353
354 static void add_gc_inode(struct inode *inode, struct list_head *ilist)
355 {
356         struct inode_entry *new_ie;
357
358         if (inode == find_gc_inode(inode->i_ino, ilist)) {
359                 iput(inode);
360                 return;
361         }
362
363         new_ie = f2fs_kmem_cache_alloc(winode_slab, GFP_NOFS);
364         new_ie->inode = inode;
365         list_add_tail(&new_ie->list, ilist);
366 }
367
368 static void put_gc_inode(struct list_head *ilist)
369 {
370         struct inode_entry *ie, *next_ie;
371         list_for_each_entry_safe(ie, next_ie, ilist, list) {
372                 iput(ie->inode);
373                 list_del(&ie->list);
374                 kmem_cache_free(winode_slab, ie);
375         }
376 }
377
378 static int check_valid_map(struct f2fs_sb_info *sbi,
379                                 unsigned int segno, int offset)
380 {
381         struct sit_info *sit_i = SIT_I(sbi);
382         struct seg_entry *sentry;
383         int ret;
384
385         mutex_lock(&sit_i->sentry_lock);
386         sentry = get_seg_entry(sbi, segno);
387         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
388         mutex_unlock(&sit_i->sentry_lock);
389         return ret;
390 }
391
392 /*
393  * This function compares node address got in summary with that in NAT.
394  * On validity, copy that node with cold status, otherwise (invalid node)
395  * ignore that.
396  */
397 static void gc_node_segment(struct f2fs_sb_info *sbi,
398                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
399 {
400         bool initial = true;
401         struct f2fs_summary *entry;
402         int off;
403
404 next_step:
405         entry = sum;
406
407         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
408                 nid_t nid = le32_to_cpu(entry->nid);
409                 struct page *node_page;
410
411                 /* stop BG_GC if there is not enough free sections. */
412                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
413                         return;
414
415                 if (check_valid_map(sbi, segno, off) == 0)
416                         continue;
417
418                 if (initial) {
419                         ra_node_page(sbi, nid);
420                         continue;
421                 }
422                 node_page = get_node_page(sbi, nid);
423                 if (IS_ERR(node_page))
424                         continue;
425
426                 /* block may become invalid during get_node_page */
427                 if (check_valid_map(sbi, segno, off) == 0) {
428                         f2fs_put_page(node_page, 1);
429                         continue;
430                 }
431
432                 /* set page dirty and write it */
433                 if (gc_type == FG_GC) {
434                         f2fs_wait_on_page_writeback(node_page, NODE);
435                         set_page_dirty(node_page);
436                 } else {
437                         if (!PageWriteback(node_page))
438                                 set_page_dirty(node_page);
439                 }
440                 f2fs_put_page(node_page, 1);
441                 stat_inc_node_blk_count(sbi, 1);
442         }
443
444         if (initial) {
445                 initial = false;
446                 goto next_step;
447         }
448
449         if (gc_type == FG_GC) {
450                 struct writeback_control wbc = {
451                         .sync_mode = WB_SYNC_ALL,
452                         .nr_to_write = LONG_MAX,
453                         .for_reclaim = 0,
454                 };
455                 sync_node_pages(sbi, 0, &wbc);
456
457                 /*
458                  * In the case of FG_GC, it'd be better to reclaim this victim
459                  * completely.
460                  */
461                 if (get_valid_blocks(sbi, segno, 1) != 0)
462                         goto next_step;
463         }
464 }
465
466 /*
467  * Calculate start block index indicating the given node offset.
468  * Be careful, caller should give this node offset only indicating direct node
469  * blocks. If any node offsets, which point the other types of node blocks such
470  * as indirect or double indirect node blocks, are given, it must be a caller's
471  * bug.
472  */
473 block_t start_bidx_of_node(unsigned int node_ofs, struct f2fs_inode_info *fi)
474 {
475         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
476         unsigned int bidx;
477
478         if (node_ofs == 0)
479                 return 0;
480
481         if (node_ofs <= 2) {
482                 bidx = node_ofs - 1;
483         } else if (node_ofs <= indirect_blks) {
484                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
485                 bidx = node_ofs - 2 - dec;
486         } else {
487                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
488                 bidx = node_ofs - 5 - dec;
489         }
490         return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(fi);
491 }
492
493 static int check_dnode(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
494                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
495 {
496         struct page *node_page;
497         nid_t nid;
498         unsigned int ofs_in_node;
499         block_t source_blkaddr;
500
501         nid = le32_to_cpu(sum->nid);
502         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
503
504         node_page = get_node_page(sbi, nid);
505         if (IS_ERR(node_page))
506                 return 0;
507
508         get_node_info(sbi, nid, dni);
509
510         if (sum->version != dni->version) {
511                 f2fs_put_page(node_page, 1);
512                 return 0;
513         }
514
515         *nofs = ofs_of_node(node_page);
516         source_blkaddr = datablock_addr(node_page, ofs_in_node);
517         f2fs_put_page(node_page, 1);
518
519         if (source_blkaddr != blkaddr)
520                 return 0;
521         return 1;
522 }
523
524 static void move_data_page(struct inode *inode, struct page *page, int gc_type)
525 {
526         struct f2fs_io_info fio = {
527                 .type = DATA,
528                 .rw = WRITE_SYNC,
529         };
530
531         if (gc_type == BG_GC) {
532                 if (PageWriteback(page))
533                         goto out;
534                 set_page_dirty(page);
535                 set_cold_data(page);
536         } else {
537                 f2fs_wait_on_page_writeback(page, DATA);
538
539                 if (clear_page_dirty_for_io(page))
540                         inode_dec_dirty_pages(inode);
541                 set_cold_data(page);
542                 do_write_data_page(page, &fio);
543                 clear_cold_data(page);
544         }
545 out:
546         f2fs_put_page(page, 1);
547 }
548
549 /*
550  * This function tries to get parent node of victim data block, and identifies
551  * data block validity. If the block is valid, copy that with cold status and
552  * modify parent node.
553  * If the parent node is not valid or the data block address is different,
554  * the victim data block is ignored.
555  */
556 static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
557                 struct list_head *ilist, unsigned int segno, int gc_type)
558 {
559         struct super_block *sb = sbi->sb;
560         struct f2fs_summary *entry;
561         block_t start_addr;
562         int off;
563         int phase = 0;
564
565         start_addr = START_BLOCK(sbi, segno);
566
567 next_step:
568         entry = sum;
569
570         for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
571                 struct page *data_page;
572                 struct inode *inode;
573                 struct node_info dni; /* dnode info for the data */
574                 unsigned int ofs_in_node, nofs;
575                 block_t start_bidx;
576
577                 /* stop BG_GC if there is not enough free sections. */
578                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
579                         return;
580
581                 if (check_valid_map(sbi, segno, off) == 0)
582                         continue;
583
584                 if (phase == 0) {
585                         ra_node_page(sbi, le32_to_cpu(entry->nid));
586                         continue;
587                 }
588
589                 /* Get an inode by ino with checking validity */
590                 if (check_dnode(sbi, entry, &dni, start_addr + off, &nofs) == 0)
591                         continue;
592
593                 if (phase == 1) {
594                         ra_node_page(sbi, dni.ino);
595                         continue;
596                 }
597
598                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
599
600                 if (phase == 2) {
601                         inode = f2fs_iget(sb, dni.ino);
602                         if (IS_ERR(inode) || is_bad_inode(inode))
603                                 continue;
604
605                         start_bidx = start_bidx_of_node(nofs, F2FS_I(inode));
606
607                         data_page = find_data_page(inode,
608                                         start_bidx + ofs_in_node, false);
609                         if (IS_ERR(data_page))
610                                 goto next_iput;
611
612                         f2fs_put_page(data_page, 0);
613                         add_gc_inode(inode, ilist);
614                 } else {
615                         inode = find_gc_inode(dni.ino, ilist);
616                         if (inode) {
617                                 start_bidx = start_bidx_of_node(nofs,
618                                                                 F2FS_I(inode));
619                                 data_page = get_lock_data_page(inode,
620                                                 start_bidx + ofs_in_node);
621                                 if (IS_ERR(data_page))
622                                         continue;
623                                 move_data_page(inode, data_page, gc_type);
624                                 stat_inc_data_blk_count(sbi, 1);
625                         }
626                 }
627                 continue;
628 next_iput:
629                 iput(inode);
630         }
631
632         if (++phase < 4)
633                 goto next_step;
634
635         if (gc_type == FG_GC) {
636                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
637
638                 /*
639                  * In the case of FG_GC, it'd be better to reclaim this victim
640                  * completely.
641                  */
642                 if (get_valid_blocks(sbi, segno, 1) != 0) {
643                         phase = 2;
644                         goto next_step;
645                 }
646         }
647 }
648
649 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
650                                                 int gc_type, int type)
651 {
652         struct sit_info *sit_i = SIT_I(sbi);
653         int ret;
654         mutex_lock(&sit_i->sentry_lock);
655         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, type, LFS);
656         mutex_unlock(&sit_i->sentry_lock);
657         return ret;
658 }
659
660 static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno,
661                                 struct list_head *ilist, int gc_type)
662 {
663         struct page *sum_page;
664         struct f2fs_summary_block *sum;
665         struct blk_plug plug;
666
667         /* read segment summary of victim */
668         sum_page = get_sum_page(sbi, segno);
669
670         blk_start_plug(&plug);
671
672         sum = page_address(sum_page);
673
674         switch (GET_SUM_TYPE((&sum->footer))) {
675         case SUM_TYPE_NODE:
676                 gc_node_segment(sbi, sum->entries, segno, gc_type);
677                 break;
678         case SUM_TYPE_DATA:
679                 gc_data_segment(sbi, sum->entries, ilist, segno, gc_type);
680                 break;
681         }
682         blk_finish_plug(&plug);
683
684         stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer)));
685         stat_inc_call_count(sbi->stat_info);
686
687         f2fs_put_page(sum_page, 1);
688 }
689
690 int f2fs_gc(struct f2fs_sb_info *sbi)
691 {
692         struct list_head ilist;
693         unsigned int segno, i;
694         int gc_type = BG_GC;
695         int nfree = 0;
696         int ret = -1;
697         struct cp_control cpc = {
698                 .reason = CP_SYNC,
699         };
700
701         INIT_LIST_HEAD(&ilist);
702 gc_more:
703         if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE)))
704                 goto stop;
705         if (unlikely(f2fs_cp_error(sbi)))
706                 goto stop;
707
708         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, nfree)) {
709                 gc_type = FG_GC;
710                 write_checkpoint(sbi, &cpc);
711         }
712
713         if (!__get_victim(sbi, &segno, gc_type, NO_CHECK_TYPE))
714                 goto stop;
715         ret = 0;
716
717         /* readahead multi ssa blocks those have contiguous address */
718         if (sbi->segs_per_sec > 1)
719                 ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), sbi->segs_per_sec,
720                                                                 META_SSA);
721
722         for (i = 0; i < sbi->segs_per_sec; i++)
723                 do_garbage_collect(sbi, segno + i, &ilist, gc_type);
724
725         if (gc_type == FG_GC) {
726                 sbi->cur_victim_sec = NULL_SEGNO;
727                 nfree++;
728                 WARN_ON(get_valid_blocks(sbi, segno, sbi->segs_per_sec));
729         }
730
731         if (has_not_enough_free_secs(sbi, nfree))
732                 goto gc_more;
733
734         if (gc_type == FG_GC)
735                 write_checkpoint(sbi, &cpc);
736 stop:
737         mutex_unlock(&sbi->gc_mutex);
738
739         put_gc_inode(&ilist);
740         return ret;
741 }
742
743 void build_gc_manager(struct f2fs_sb_info *sbi)
744 {
745         DIRTY_I(sbi)->v_ops = &default_v_ops;
746 }
747
748 int __init create_gc_caches(void)
749 {
750         winode_slab = f2fs_kmem_cache_create("f2fs_gc_inodes",
751                         sizeof(struct inode_entry));
752         if (!winode_slab)
753                 return -ENOMEM;
754         return 0;
755 }
756
757 void destroy_gc_caches(void)
758 {
759         kmem_cache_destroy(winode_slab);
760 }