blob: 76adbc3641f1534ead6151b06dd8257219f7911e [file] [log] [blame]
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +09001/*
Jaegeuk Kim7bc09002012-11-02 17:13:01 +09002 * 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>
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090014#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"
Namjae Jeon8e46b3e2013-04-23 16:42:53 +090025#include <trace/events/f2fs.h>
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090026
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090027static int gc_thread_func(void *data)
28{
29 struct f2fs_sb_info *sbi = data;
Namjae Jeonb59d0ba2013-08-04 23:09:40 +090030 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090031 wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
32 long wait_ms;
33
Namjae Jeonb59d0ba2013-08-04 23:09:40 +090034 wait_ms = gc_th->min_sleep_time;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090035
36 do {
37 if (try_to_freeze())
38 continue;
39 else
40 wait_event_interruptible_timeout(*wq,
41 kthread_should_stop(),
42 msecs_to_jiffies(wait_ms));
43 if (kthread_should_stop())
44 break;
45
Changman Leed6212a52013-01-29 18:30:07 +090046 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
Chao Yu88dd8932015-01-26 20:24:21 +080047 increase_sleep_time(gc_th, &wait_ms);
Changman Leed6212a52013-01-29 18:30:07 +090048 continue;
49 }
50
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090051 /*
52 * [GC triggering condition]
53 * 0. GC is not conducted currently.
54 * 1. There are enough dirty segments.
55 * 2. IO subsystem is idle by checking the # of writeback pages.
56 * 3. IO subsystem is idle by checking the # of requests in
57 * bdev's request list.
58 *
arter97e1c42042014-08-06 23:22:50 +090059 * Note) We have to avoid triggering GCs frequently.
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090060 * Because it is possible that some segments can be
61 * invalidated soon after by user update or deletion.
62 * So, I'd like to wait some time to collect dirty segments.
63 */
64 if (!mutex_trylock(&sbi->gc_mutex))
65 continue;
66
67 if (!is_idle(sbi)) {
Chao Yu88dd8932015-01-26 20:24:21 +080068 increase_sleep_time(gc_th, &wait_ms);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090069 mutex_unlock(&sbi->gc_mutex);
70 continue;
71 }
72
73 if (has_enough_invalid_blocks(sbi))
Chao Yu88dd8932015-01-26 20:24:21 +080074 decrease_sleep_time(gc_th, &wait_ms);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090075 else
Chao Yu88dd8932015-01-26 20:24:21 +080076 increase_sleep_time(gc_th, &wait_ms);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090077
Jaegeuk Kimdcdfff62013-10-22 20:56:10 +090078 stat_inc_bggc_count(sbi);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090079
Jaegeuk Kim43727522013-02-04 15:11:17 +090080 /* if return value is not zero, no victim was selected */
81 if (f2fs_gc(sbi))
Namjae Jeonb59d0ba2013-08-04 23:09:40 +090082 wait_ms = gc_th->no_gc_sleep_time;
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +090083
Jaegeuk Kim4660f9c2013-10-24 14:19:18 +090084 /* balancing f2fs's metadata periodically */
85 f2fs_balance_fs_bg(sbi);
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +090086
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090087 } while (!kthread_should_stop());
88 return 0;
89}
90
91int start_gc_thread(struct f2fs_sb_info *sbi)
92{
Namjae Jeon1042d602012-12-01 10:56:13 +090093 struct f2fs_gc_kthread *gc_th;
Namjae Jeonec7b1f22013-02-02 23:52:28 +090094 dev_t dev = sbi->sb->s_bdev->bd_dev;
Namjae Jeon7a267f82013-05-26 11:05:32 +090095 int err = 0;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090096
97 gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
Namjae Jeon7a267f82013-05-26 11:05:32 +090098 if (!gc_th) {
99 err = -ENOMEM;
100 goto out;
101 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900102
Namjae Jeonb59d0ba2013-08-04 23:09:40 +0900103 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
104 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
105 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
106
Namjae Jeond2dc0952013-08-04 23:10:15 +0900107 gc_th->gc_idle = 0;
108
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900109 sbi->gc_thread = gc_th;
110 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
111 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
Namjae Jeonec7b1f22013-02-02 23:52:28 +0900112 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900113 if (IS_ERR(gc_th->f2fs_gc_task)) {
Namjae Jeon7a267f82013-05-26 11:05:32 +0900114 err = PTR_ERR(gc_th->f2fs_gc_task);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900115 kfree(gc_th);
Namjae Jeon25718422013-02-02 23:52:42 +0900116 sbi->gc_thread = NULL;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900117 }
Namjae Jeon7a267f82013-05-26 11:05:32 +0900118out:
119 return err;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900120}
121
122void stop_gc_thread(struct f2fs_sb_info *sbi)
123{
124 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
125 if (!gc_th)
126 return;
127 kthread_stop(gc_th->f2fs_gc_task);
128 kfree(gc_th);
129 sbi->gc_thread = NULL;
130}
131
Namjae Jeond2dc0952013-08-04 23:10:15 +0900132static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900133{
Namjae Jeond2dc0952013-08-04 23:10:15 +0900134 int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
135
136 if (gc_th && gc_th->gc_idle) {
137 if (gc_th->gc_idle == 1)
138 gc_mode = GC_CB;
139 else if (gc_th->gc_idle == 2)
140 gc_mode = GC_GREEDY;
141 }
142 return gc_mode;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900143}
144
145static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
146 int type, struct victim_sel_policy *p)
147{
148 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
149
Jaegeuk Kim4ebefc42013-03-31 13:49:18 +0900150 if (p->alloc_mode == SSR) {
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900151 p->gc_mode = GC_GREEDY;
152 p->dirty_segmap = dirty_i->dirty_segmap[type];
Jin Xua26b7c82013-09-05 12:45:26 +0800153 p->max_search = dirty_i->nr_dirty[type];
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900154 p->ofs_unit = 1;
155 } else {
Namjae Jeond2dc0952013-08-04 23:10:15 +0900156 p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900157 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
Jin Xua26b7c82013-09-05 12:45:26 +0800158 p->max_search = dirty_i->nr_dirty[DIRTY];
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900159 p->ofs_unit = sbi->segs_per_sec;
160 }
Jin Xua26b7c82013-09-05 12:45:26 +0800161
Jaegeuk Kimb1c57c12014-01-08 13:45:08 +0900162 if (p->max_search > sbi->max_victim_search)
163 p->max_search = sbi->max_victim_search;
Jin Xua26b7c82013-09-05 12:45:26 +0800164
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900165 p->offset = sbi->last_victim[p->gc_mode];
166}
167
168static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
169 struct victim_sel_policy *p)
170{
Jaegeuk Kimb7250d22013-02-05 13:19:28 +0900171 /* SSR allocates in a segment unit */
172 if (p->alloc_mode == SSR)
173 return 1 << sbi->log_blocks_per_seg;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900174 if (p->gc_mode == GC_GREEDY)
175 return (1 << sbi->log_blocks_per_seg) * p->ofs_unit;
176 else if (p->gc_mode == GC_CB)
177 return UINT_MAX;
178 else /* No other gc_mode */
179 return 0;
180}
181
182static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
183{
184 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900185 unsigned int secno;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900186
187 /*
188 * If the gc_type is FG_GC, we can select victim segments
189 * selected by background GC before.
190 * Those segments guarantee they have small valid blocks.
191 */
Jaegeuk Kim7cd85582014-09-23 11:23:01 -0700192 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900193 if (sec_usage_check(sbi, secno))
Chao Yub65ee142014-08-04 10:10:07 +0800194 continue;
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900195 clear_bit(secno, dirty_i->victim_secmap);
196 return secno * sbi->segs_per_sec;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900197 }
198 return NULL_SEGNO;
199}
200
201static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
202{
203 struct sit_info *sit_i = SIT_I(sbi);
204 unsigned int secno = GET_SECNO(sbi, segno);
205 unsigned int start = secno * sbi->segs_per_sec;
206 unsigned long long mtime = 0;
207 unsigned int vblocks;
208 unsigned char age = 0;
209 unsigned char u;
210 unsigned int i;
211
212 for (i = 0; i < sbi->segs_per_sec; i++)
213 mtime += get_seg_entry(sbi, start + i)->mtime;
214 vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec);
215
216 mtime = div_u64(mtime, sbi->segs_per_sec);
217 vblocks = div_u64(vblocks, sbi->segs_per_sec);
218
219 u = (vblocks * 100) >> sbi->log_blocks_per_seg;
220
arter97e1c42042014-08-06 23:22:50 +0900221 /* Handle if the system time has changed by the user */
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900222 if (mtime < sit_i->min_mtime)
223 sit_i->min_mtime = mtime;
224 if (mtime > sit_i->max_mtime)
225 sit_i->max_mtime = mtime;
226 if (sit_i->max_mtime != sit_i->min_mtime)
227 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
228 sit_i->max_mtime - sit_i->min_mtime);
229
230 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
231}
232
Jin Xua57e5642013-09-13 08:38:54 +0800233static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
234 unsigned int segno, struct victim_sel_policy *p)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900235{
236 if (p->alloc_mode == SSR)
237 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
238
239 /* alloc_mode == LFS */
240 if (p->gc_mode == GC_GREEDY)
241 return get_valid_blocks(sbi, segno, sbi->segs_per_sec);
242 else
243 return get_cb_cost(sbi, segno);
244}
245
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900246/*
Masanari Iida111d2492013-03-19 08:03:35 +0900247 * This function is called from two paths.
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900248 * One is garbage collection and the other is SSR segment selection.
249 * When it is called during GC, it just gets a victim segment
250 * and it does not remove it from dirty seglist.
251 * When it is called from SSR segment selection, it finds a segment
252 * which has minimum valid blocks and removes it from dirty seglist.
253 */
254static int get_victim_by_default(struct f2fs_sb_info *sbi,
255 unsigned int *result, int gc_type, int type, char alloc_mode)
256{
257 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
258 struct victim_sel_policy p;
Namjae Jeonb2b34602013-06-01 16:20:26 +0900259 unsigned int secno, max_cost;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900260 int nsearched = 0;
261
Chao Yu210f41b2014-09-15 18:05:44 +0800262 mutex_lock(&dirty_i->seglist_lock);
263
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900264 p.alloc_mode = alloc_mode;
265 select_policy(sbi, gc_type, type, &p);
266
267 p.min_segno = NULL_SEGNO;
Namjae Jeonb2b34602013-06-01 16:20:26 +0900268 p.min_cost = max_cost = get_max_cost(sbi, &p);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900269
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900270 if (p.alloc_mode == LFS && gc_type == FG_GC) {
271 p.min_segno = check_bg_victims(sbi);
272 if (p.min_segno != NULL_SEGNO)
273 goto got_it;
274 }
275
276 while (1) {
277 unsigned long cost;
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900278 unsigned int segno;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900279
Jaegeuk Kim7cd85582014-09-23 11:23:01 -0700280 segno = find_next_bit(p.dirty_segmap, MAIN_SEGS(sbi), p.offset);
281 if (segno >= MAIN_SEGS(sbi)) {
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900282 if (sbi->last_victim[p.gc_mode]) {
283 sbi->last_victim[p.gc_mode] = 0;
284 p.offset = 0;
285 continue;
286 }
287 break;
288 }
Jin Xua57e5642013-09-13 08:38:54 +0800289
290 p.offset = segno + p.ofs_unit;
291 if (p.ofs_unit > 1)
292 p.offset -= segno % p.ofs_unit;
293
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900294 secno = GET_SECNO(sbi, segno);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900295
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900296 if (sec_usage_check(sbi, secno))
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900297 continue;
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900298 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900299 continue;
300
301 cost = get_gc_cost(sbi, segno, &p);
302
303 if (p.min_cost > cost) {
304 p.min_segno = segno;
305 p.min_cost = cost;
Jin Xua57e5642013-09-13 08:38:54 +0800306 } else if (unlikely(cost == max_cost)) {
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900307 continue;
Jin Xua57e5642013-09-13 08:38:54 +0800308 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900309
Jin Xua26b7c82013-09-05 12:45:26 +0800310 if (nsearched++ >= p.max_search) {
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900311 sbi->last_victim[p.gc_mode] = segno;
312 break;
313 }
314 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900315 if (p.min_segno != NULL_SEGNO) {
Namjae Jeonb2b34602013-06-01 16:20:26 +0900316got_it:
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900317 if (p.alloc_mode == LFS) {
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900318 secno = GET_SECNO(sbi, p.min_segno);
319 if (gc_type == FG_GC)
320 sbi->cur_victim_sec = secno;
321 else
322 set_bit(secno, dirty_i->victim_secmap);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900323 }
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900324 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
Namjae Jeon8e46b3e2013-04-23 16:42:53 +0900325
326 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
327 sbi->cur_victim_sec,
328 prefree_segments(sbi), free_segments(sbi));
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900329 }
330 mutex_unlock(&dirty_i->seglist_lock);
331
332 return (p.min_segno == NULL_SEGNO) ? 0 : 1;
333}
334
335static const struct victim_selection default_v_ops = {
336 .get_victim = get_victim_by_default,
337};
338
Changman Lee7dda2af2014-11-28 15:49:40 +0000339static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900340{
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900341 struct inode_entry *ie;
342
Changman Lee7dda2af2014-11-28 15:49:40 +0000343 ie = radix_tree_lookup(&gc_list->iroot, ino);
344 if (ie)
345 return ie->inode;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900346 return NULL;
347}
348
Changman Lee7dda2af2014-11-28 15:49:40 +0000349static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900350{
Gu Zheng6cc4af52013-06-20 17:52:39 +0800351 struct inode_entry *new_ie;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900352
Changman Lee7dda2af2014-11-28 15:49:40 +0000353 if (inode == find_gc_inode(gc_list, inode->i_ino)) {
Gu Zheng6cc4af52013-06-20 17:52:39 +0800354 iput(inode);
355 return;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900356 }
Chao Yu06292072014-12-29 15:56:18 +0800357 new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900358 new_ie->inode = inode;
Chao Yuf28e5032015-01-23 20:37:53 +0800359
360 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
Changman Lee7dda2af2014-11-28 15:49:40 +0000361 list_add_tail(&new_ie->list, &gc_list->ilist);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900362}
363
Changman Lee7dda2af2014-11-28 15:49:40 +0000364static void put_gc_inode(struct gc_inode_list *gc_list)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900365{
366 struct inode_entry *ie, *next_ie;
Changman Lee7dda2af2014-11-28 15:49:40 +0000367 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
368 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900369 iput(ie->inode);
370 list_del(&ie->list);
Chao Yu06292072014-12-29 15:56:18 +0800371 kmem_cache_free(inode_entry_slab, ie);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900372 }
373}
374
375static int check_valid_map(struct f2fs_sb_info *sbi,
376 unsigned int segno, int offset)
377{
378 struct sit_info *sit_i = SIT_I(sbi);
379 struct seg_entry *sentry;
380 int ret;
381
382 mutex_lock(&sit_i->sentry_lock);
383 sentry = get_seg_entry(sbi, segno);
384 ret = f2fs_test_bit(offset, sentry->cur_valid_map);
385 mutex_unlock(&sit_i->sentry_lock);
Jaegeuk Kim43727522013-02-04 15:11:17 +0900386 return ret;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900387}
388
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900389/*
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900390 * This function compares node address got in summary with that in NAT.
391 * On validity, copy that node with cold status, otherwise (invalid node)
392 * ignore that.
393 */
Jaegeuk Kim43727522013-02-04 15:11:17 +0900394static void gc_node_segment(struct f2fs_sb_info *sbi,
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900395 struct f2fs_summary *sum, unsigned int segno, int gc_type)
396{
397 bool initial = true;
398 struct f2fs_summary *entry;
399 int off;
400
401next_step:
402 entry = sum;
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900403
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900404 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
405 nid_t nid = le32_to_cpu(entry->nid);
406 struct page *node_page;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900407
Jaegeuk Kim43727522013-02-04 15:11:17 +0900408 /* stop BG_GC if there is not enough free sections. */
409 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
410 return;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900411
Jaegeuk Kim43727522013-02-04 15:11:17 +0900412 if (check_valid_map(sbi, segno, off) == 0)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900413 continue;
414
415 if (initial) {
416 ra_node_page(sbi, nid);
417 continue;
418 }
419 node_page = get_node_page(sbi, nid);
420 if (IS_ERR(node_page))
421 continue;
422
Huang Ying9a01b562014-09-07 11:05:20 +0800423 /* block may become invalid during get_node_page */
424 if (check_valid_map(sbi, segno, off) == 0) {
425 f2fs_put_page(node_page, 1);
426 continue;
427 }
428
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900429 /* set page dirty and write it */
Jaegeuk Kim4ebefc42013-03-31 13:49:18 +0900430 if (gc_type == FG_GC) {
Yuan Zhong5514f0a2014-01-10 07:26:14 +0000431 f2fs_wait_on_page_writeback(node_page, NODE);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900432 set_page_dirty(node_page);
Jaegeuk Kim4ebefc42013-03-31 13:49:18 +0900433 } else {
434 if (!PageWriteback(node_page))
435 set_page_dirty(node_page);
436 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900437 f2fs_put_page(node_page, 1);
438 stat_inc_node_blk_count(sbi, 1);
439 }
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900440
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900441 if (initial) {
442 initial = false;
443 goto next_step;
444 }
445
446 if (gc_type == FG_GC) {
447 struct writeback_control wbc = {
448 .sync_mode = WB_SYNC_ALL,
449 .nr_to_write = LONG_MAX,
450 .for_reclaim = 0,
451 };
452 sync_node_pages(sbi, 0, &wbc);
Jaegeuk Kim4ebefc42013-03-31 13:49:18 +0900453
454 /*
455 * In the case of FG_GC, it'd be better to reclaim this victim
456 * completely.
457 */
458 if (get_valid_blocks(sbi, segno, 1) != 0)
459 goto next_step;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900460 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900461}
462
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900463/*
Jaegeuk Kim9af45ef2013-01-21 17:34:21 +0900464 * Calculate start block index indicating the given node offset.
465 * Be careful, caller should give this node offset only indicating direct node
466 * blocks. If any node offsets, which point the other types of node blocks such
467 * as indirect or double indirect node blocks, are given, it must be a caller's
468 * bug.
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900469 */
Jaegeuk Kimde936532013-08-12 21:08:03 +0900470block_t start_bidx_of_node(unsigned int node_ofs, struct f2fs_inode_info *fi)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900471{
Jaegeuk Kimce19a5d2012-12-26 12:03:22 +0900472 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
473 unsigned int bidx;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900474
Jaegeuk Kimce19a5d2012-12-26 12:03:22 +0900475 if (node_ofs == 0)
476 return 0;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900477
Jaegeuk Kimce19a5d2012-12-26 12:03:22 +0900478 if (node_ofs <= 2) {
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900479 bidx = node_ofs - 1;
480 } else if (node_ofs <= indirect_blks) {
Jaegeuk Kimce19a5d2012-12-26 12:03:22 +0900481 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900482 bidx = node_ofs - 2 - dec;
483 } else {
Jaegeuk Kimce19a5d2012-12-26 12:03:22 +0900484 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900485 bidx = node_ofs - 5 - dec;
486 }
Jaegeuk Kimde936532013-08-12 21:08:03 +0900487 return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(fi);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900488}
489
490static int check_dnode(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
491 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
492{
493 struct page *node_page;
494 nid_t nid;
495 unsigned int ofs_in_node;
496 block_t source_blkaddr;
497
498 nid = le32_to_cpu(sum->nid);
499 ofs_in_node = le16_to_cpu(sum->ofs_in_node);
500
501 node_page = get_node_page(sbi, nid);
502 if (IS_ERR(node_page))
Jaegeuk Kim43727522013-02-04 15:11:17 +0900503 return 0;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900504
505 get_node_info(sbi, nid, dni);
506
507 if (sum->version != dni->version) {
508 f2fs_put_page(node_page, 1);
Jaegeuk Kim43727522013-02-04 15:11:17 +0900509 return 0;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900510 }
511
512 *nofs = ofs_of_node(node_page);
513 source_blkaddr = datablock_addr(node_page, ofs_in_node);
514 f2fs_put_page(node_page, 1);
515
516 if (source_blkaddr != blkaddr)
Jaegeuk Kim43727522013-02-04 15:11:17 +0900517 return 0;
518 return 1;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900519}
520
521static void move_data_page(struct inode *inode, struct page *page, int gc_type)
522{
Jaegeuk Kim458e6192013-12-11 13:54:01 +0900523 struct f2fs_io_info fio = {
524 .type = DATA,
525 .rw = WRITE_SYNC,
Fan Li63a0b7cb2013-12-09 16:09:00 +0800526 };
527
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900528 if (gc_type == BG_GC) {
Jaegeuk Kim4ebefc42013-03-31 13:49:18 +0900529 if (PageWriteback(page))
530 goto out;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900531 set_page_dirty(page);
532 set_cold_data(page);
533 } else {
Yuan Zhong5514f0a2014-01-10 07:26:14 +0000534 f2fs_wait_on_page_writeback(page, DATA);
Jaegeuk Kim4ebefc42013-03-31 13:49:18 +0900535
Jaegeuk Kim1fe54f92014-02-07 10:00:06 +0900536 if (clear_page_dirty_for_io(page))
Jaegeuk Kima7ffdbe2014-09-12 15:53:45 -0700537 inode_dec_dirty_pages(inode);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900538 set_cold_data(page);
Jaegeuk Kim458e6192013-12-11 13:54:01 +0900539 do_write_data_page(page, &fio);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900540 clear_cold_data(page);
541 }
542out:
543 f2fs_put_page(page, 1);
544}
545
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900546/*
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900547 * This function tries to get parent node of victim data block, and identifies
548 * data block validity. If the block is valid, copy that with cold status and
549 * modify parent node.
550 * If the parent node is not valid or the data block address is different,
551 * the victim data block is ignored.
552 */
Jaegeuk Kim43727522013-02-04 15:11:17 +0900553static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
Changman Lee7dda2af2014-11-28 15:49:40 +0000554 struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900555{
556 struct super_block *sb = sbi->sb;
557 struct f2fs_summary *entry;
558 block_t start_addr;
Jaegeuk Kim43727522013-02-04 15:11:17 +0900559 int off;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900560 int phase = 0;
561
562 start_addr = START_BLOCK(sbi, segno);
563
564next_step:
565 entry = sum;
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900566
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900567 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
568 struct page *data_page;
569 struct inode *inode;
570 struct node_info dni; /* dnode info for the data */
571 unsigned int ofs_in_node, nofs;
572 block_t start_bidx;
573
Jaegeuk Kim43727522013-02-04 15:11:17 +0900574 /* stop BG_GC if there is not enough free sections. */
575 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
576 return;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900577
Jaegeuk Kim43727522013-02-04 15:11:17 +0900578 if (check_valid_map(sbi, segno, off) == 0)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900579 continue;
580
581 if (phase == 0) {
582 ra_node_page(sbi, le32_to_cpu(entry->nid));
583 continue;
584 }
585
586 /* Get an inode by ino with checking validity */
Jaegeuk Kim43727522013-02-04 15:11:17 +0900587 if (check_dnode(sbi, entry, &dni, start_addr + off, &nofs) == 0)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900588 continue;
589
590 if (phase == 1) {
591 ra_node_page(sbi, dni.ino);
592 continue;
593 }
594
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900595 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
596
597 if (phase == 2) {
Jaegeuk Kimd4686d562013-01-31 15:36:04 +0900598 inode = f2fs_iget(sb, dni.ino);
Chao Yub73e5282014-08-30 09:52:34 +0800599 if (IS_ERR(inode) || is_bad_inode(inode))
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900600 continue;
601
Jaegeuk Kimde936532013-08-12 21:08:03 +0900602 start_bidx = start_bidx_of_node(nofs, F2FS_I(inode));
603
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900604 data_page = find_data_page(inode,
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900605 start_bidx + ofs_in_node, false);
Changman Lee31a32682014-11-27 16:03:08 +0900606 if (IS_ERR(data_page)) {
607 iput(inode);
608 continue;
609 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900610
611 f2fs_put_page(data_page, 0);
Changman Lee7dda2af2014-11-28 15:49:40 +0000612 add_gc_inode(gc_list, inode);
Changman Lee31a32682014-11-27 16:03:08 +0900613 continue;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900614 }
Changman Lee31a32682014-11-27 16:03:08 +0900615
616 /* phase 3 */
Changman Lee7dda2af2014-11-28 15:49:40 +0000617 inode = find_gc_inode(gc_list, dni.ino);
Changman Lee31a32682014-11-27 16:03:08 +0900618 if (inode) {
619 start_bidx = start_bidx_of_node(nofs, F2FS_I(inode));
620 data_page = get_lock_data_page(inode,
621 start_bidx + ofs_in_node);
622 if (IS_ERR(data_page))
623 continue;
624 move_data_page(inode, data_page, gc_type);
625 stat_inc_data_blk_count(sbi, 1);
626 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900627 }
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900628
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900629 if (++phase < 4)
630 goto next_step;
Jaegeuk Kim43727522013-02-04 15:11:17 +0900631
Jaegeuk Kim4ebefc42013-03-31 13:49:18 +0900632 if (gc_type == FG_GC) {
Jaegeuk Kim458e6192013-12-11 13:54:01 +0900633 f2fs_submit_merged_bio(sbi, DATA, WRITE);
Jaegeuk Kim4ebefc42013-03-31 13:49:18 +0900634
635 /*
636 * In the case of FG_GC, it'd be better to reclaim this victim
637 * completely.
638 */
639 if (get_valid_blocks(sbi, segno, 1) != 0) {
640 phase = 2;
641 goto next_step;
642 }
643 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900644}
645
646static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
Gu Zheng8a2d0ac2014-10-20 17:45:48 +0800647 int gc_type)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900648{
649 struct sit_info *sit_i = SIT_I(sbi);
650 int ret;
Gu Zheng8a2d0ac2014-10-20 17:45:48 +0800651
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900652 mutex_lock(&sit_i->sentry_lock);
Gu Zheng8a2d0ac2014-10-20 17:45:48 +0800653 ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
654 NO_CHECK_TYPE, LFS);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900655 mutex_unlock(&sit_i->sentry_lock);
656 return ret;
657}
658
Jaegeuk Kim43727522013-02-04 15:11:17 +0900659static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno,
Changman Lee7dda2af2014-11-28 15:49:40 +0000660 struct gc_inode_list *gc_list, int gc_type)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900661{
662 struct page *sum_page;
663 struct f2fs_summary_block *sum;
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900664 struct blk_plug plug;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900665
666 /* read segment summary of victim */
667 sum_page = get_sum_page(sbi, segno);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900668
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900669 blk_start_plug(&plug);
670
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900671 sum = page_address(sum_page);
672
673 switch (GET_SUM_TYPE((&sum->footer))) {
674 case SUM_TYPE_NODE:
Jaegeuk Kim43727522013-02-04 15:11:17 +0900675 gc_node_segment(sbi, sum->entries, segno, gc_type);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900676 break;
677 case SUM_TYPE_DATA:
Changman Lee7dda2af2014-11-28 15:49:40 +0000678 gc_data_segment(sbi, sum->entries, gc_list, segno, gc_type);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900679 break;
680 }
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900681 blk_finish_plug(&plug);
682
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900683 stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer)));
684 stat_inc_call_count(sbi->stat_info);
685
Jaegeuk Kimb7473752013-04-01 08:32:21 +0900686 f2fs_put_page(sum_page, 1);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900687}
688
Jaegeuk Kim408e9372013-01-03 17:55:52 +0900689int f2fs_gc(struct f2fs_sb_info *sbi)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900690{
Jaegeuk Kim408e9372013-01-03 17:55:52 +0900691 unsigned int segno, i;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900692 int gc_type = BG_GC;
Jaegeuk Kim43727522013-02-04 15:11:17 +0900693 int nfree = 0;
694 int ret = -1;
Jaegeuk Kimd5053a342014-10-30 22:47:03 -0700695 struct cp_control cpc;
Changman Lee7dda2af2014-11-28 15:49:40 +0000696 struct gc_inode_list gc_list = {
697 .ilist = LIST_HEAD_INIT(gc_list.ilist),
Jaegeuk Kim769ec6e2014-12-03 20:47:26 -0800698 .iroot = RADIX_TREE_INIT(GFP_NOFS),
Changman Lee7dda2af2014-11-28 15:49:40 +0000699 };
Jaegeuk Kimd5053a342014-10-30 22:47:03 -0700700
Jaegeuk Kim119ee912015-01-29 11:45:33 -0800701 cpc.reason = __get_cp_reason(sbi);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900702gc_more:
Jaegeuk Kim6bacf522013-12-06 15:00:58 +0900703 if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE)))
Jaegeuk Kim408e9372013-01-03 17:55:52 +0900704 goto stop;
Jaegeuk Kim1e968fd2014-08-11 16:49:25 -0700705 if (unlikely(f2fs_cp_error(sbi)))
Jaegeuk Kim203681f2014-02-05 13:03:57 +0900706 goto stop;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900707
Jaegeuk Kimd64f8042013-04-08 16:01:00 +0900708 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, nfree)) {
Jaegeuk Kim408e9372013-01-03 17:55:52 +0900709 gc_type = FG_GC;
Jaegeuk Kim75ab4cb2014-09-20 21:57:51 -0700710 write_checkpoint(sbi, &cpc);
Jaegeuk Kimd64f8042013-04-08 16:01:00 +0900711 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900712
Gu Zheng8a2d0ac2014-10-20 17:45:48 +0800713 if (!__get_victim(sbi, &segno, gc_type))
Jaegeuk Kim408e9372013-01-03 17:55:52 +0900714 goto stop;
Jaegeuk Kim43727522013-02-04 15:11:17 +0900715 ret = 0;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900716
Chao Yu81c1a0f12014-02-27 19:12:24 +0800717 /* 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
Jaegeuk Kim43727522013-02-04 15:11:17 +0900722 for (i = 0; i < sbi->segs_per_sec; i++)
Changman Lee7dda2af2014-11-28 15:49:40 +0000723 do_garbage_collect(sbi, segno + i, &gc_list, gc_type);
Jaegeuk Kim43727522013-02-04 15:11:17 +0900724
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900725 if (gc_type == FG_GC) {
726 sbi->cur_victim_sec = NULL_SEGNO;
Jaegeuk Kim43727522013-02-04 15:11:17 +0900727 nfree++;
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900728 WARN_ON(get_valid_blocks(sbi, segno, sbi->segs_per_sec));
729 }
Jaegeuk Kim43727522013-02-04 15:11:17 +0900730
731 if (has_not_enough_free_secs(sbi, nfree))
732 goto gc_more;
733
734 if (gc_type == FG_GC)
Jaegeuk Kim75ab4cb2014-09-20 21:57:51 -0700735 write_checkpoint(sbi, &cpc);
Jaegeuk Kim408e9372013-01-03 17:55:52 +0900736stop:
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900737 mutex_unlock(&sbi->gc_mutex);
738
Changman Lee7dda2af2014-11-28 15:49:40 +0000739 put_gc_inode(&gc_list);
Jaegeuk Kim43727522013-02-04 15:11:17 +0900740 return ret;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900741}
742
743void build_gc_manager(struct f2fs_sb_info *sbi)
744{
745 DIRTY_I(sbi)->v_ops = &default_v_ops;
746}