blob: fa3d2e2df8e70e883011afc5e46c9989e0e834ff [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>
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090019
20#include "f2fs.h"
21#include "node.h"
22#include "segment.h"
23#include "gc.h"
Namjae Jeon8e46b3e2013-04-23 16:42:53 +090024#include <trace/events/f2fs.h>
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090025
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090026static int gc_thread_func(void *data)
27{
28 struct f2fs_sb_info *sbi = data;
Namjae Jeonb59d0ba2013-08-04 23:09:40 +090029 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090030 wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
31 long wait_ms;
32
Namjae Jeonb59d0ba2013-08-04 23:09:40 +090033 wait_ms = gc_th->min_sleep_time;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090034
Jaegeuk Kim1d7be272017-05-17 10:36:58 -070035 set_freezable();
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090036 do {
Jaegeuk Kim1d7be272017-05-17 10:36:58 -070037 wait_event_interruptible_timeout(*wq,
38 kthread_should_stop() || freezing(current),
39 msecs_to_jiffies(wait_ms));
40
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090041 if (try_to_freeze())
42 continue;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090043 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
Chao Yu0f348022016-09-26 19:45:55 +080051#ifdef CONFIG_F2FS_FAULT_INJECTION
Chao Yu55523512017-02-25 11:08:28 +080052 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
53 f2fs_show_injection_info(FAULT_CHECKPOINT);
Chao Yu0f348022016-09-26 19:45:55 +080054 f2fs_stop_checkpoint(sbi, false);
Chao Yu55523512017-02-25 11:08:28 +080055 }
Chao Yu0f348022016-09-26 19:45:55 +080056#endif
57
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090058 /*
59 * [GC triggering condition]
60 * 0. GC is not conducted currently.
61 * 1. There are enough dirty segments.
62 * 2. IO subsystem is idle by checking the # of writeback pages.
63 * 3. IO subsystem is idle by checking the # of requests in
64 * bdev's request list.
65 *
arter97e1c42042014-08-06 23:22:50 +090066 * Note) We have to avoid triggering GCs frequently.
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090067 * Because it is possible that some segments can be
68 * invalidated soon after by user update or deletion.
69 * So, I'd like to wait some time to collect dirty segments.
70 */
71 if (!mutex_trylock(&sbi->gc_mutex))
72 continue;
73
74 if (!is_idle(sbi)) {
Chao Yu88dd8932015-01-26 20:24:21 +080075 increase_sleep_time(gc_th, &wait_ms);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090076 mutex_unlock(&sbi->gc_mutex);
77 continue;
78 }
79
80 if (has_enough_invalid_blocks(sbi))
Chao Yu88dd8932015-01-26 20:24:21 +080081 decrease_sleep_time(gc_th, &wait_ms);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090082 else
Chao Yu88dd8932015-01-26 20:24:21 +080083 increase_sleep_time(gc_th, &wait_ms);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090084
Jaegeuk Kimdcdfff62013-10-22 20:56:10 +090085 stat_inc_bggc_count(sbi);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090086
Jaegeuk Kim43727522013-02-04 15:11:17 +090087 /* if return value is not zero, no victim was selected */
Jaegeuk Kime066b832017-04-13 15:17:00 -070088 if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true, NULL_SEGNO))
Namjae Jeonb59d0ba2013-08-04 23:09:40 +090089 wait_ms = gc_th->no_gc_sleep_time;
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +090090
Jaegeuk Kim84e42142015-10-13 10:00:53 -070091 trace_f2fs_background_gc(sbi->sb, wait_ms,
92 prefree_segments(sbi), free_segments(sbi));
93
Jaegeuk Kim4660f9c2013-10-24 14:19:18 +090094 /* balancing f2fs's metadata periodically */
95 f2fs_balance_fs_bg(sbi);
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +090096
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090097 } while (!kthread_should_stop());
98 return 0;
99}
100
101int start_gc_thread(struct f2fs_sb_info *sbi)
102{
Namjae Jeon1042d602012-12-01 10:56:13 +0900103 struct f2fs_gc_kthread *gc_th;
Namjae Jeonec7b1f22013-02-02 23:52:28 +0900104 dev_t dev = sbi->sb->s_bdev->bd_dev;
Namjae Jeon7a267f82013-05-26 11:05:32 +0900105 int err = 0;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900106
Chao Yu1ecc0c52016-09-23 21:30:09 +0800107 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
Namjae Jeon7a267f82013-05-26 11:05:32 +0900108 if (!gc_th) {
109 err = -ENOMEM;
110 goto out;
111 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900112
Namjae Jeonb59d0ba2013-08-04 23:09:40 +0900113 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
114 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
115 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
116
Namjae Jeond2dc0952013-08-04 23:10:15 +0900117 gc_th->gc_idle = 0;
118
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900119 sbi->gc_thread = gc_th;
120 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
121 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
Namjae Jeonec7b1f22013-02-02 23:52:28 +0900122 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900123 if (IS_ERR(gc_th->f2fs_gc_task)) {
Namjae Jeon7a267f82013-05-26 11:05:32 +0900124 err = PTR_ERR(gc_th->f2fs_gc_task);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900125 kfree(gc_th);
Namjae Jeon25718422013-02-02 23:52:42 +0900126 sbi->gc_thread = NULL;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900127 }
Namjae Jeon7a267f82013-05-26 11:05:32 +0900128out:
129 return err;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900130}
131
132void stop_gc_thread(struct f2fs_sb_info *sbi)
133{
134 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
135 if (!gc_th)
136 return;
137 kthread_stop(gc_th->f2fs_gc_task);
138 kfree(gc_th);
139 sbi->gc_thread = NULL;
140}
141
Namjae Jeond2dc0952013-08-04 23:10:15 +0900142static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900143{
Namjae Jeond2dc0952013-08-04 23:10:15 +0900144 int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY;
145
146 if (gc_th && gc_th->gc_idle) {
147 if (gc_th->gc_idle == 1)
148 gc_mode = GC_CB;
149 else if (gc_th->gc_idle == 2)
150 gc_mode = GC_GREEDY;
151 }
152 return gc_mode;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900153}
154
155static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
156 int type, struct victim_sel_policy *p)
157{
158 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
159
Jaegeuk Kim4ebefc42013-03-31 13:49:18 +0900160 if (p->alloc_mode == SSR) {
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900161 p->gc_mode = GC_GREEDY;
162 p->dirty_segmap = dirty_i->dirty_segmap[type];
Jin Xua26b7c82013-09-05 12:45:26 +0800163 p->max_search = dirty_i->nr_dirty[type];
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900164 p->ofs_unit = 1;
165 } else {
Namjae Jeond2dc0952013-08-04 23:10:15 +0900166 p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900167 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
Jin Xua26b7c82013-09-05 12:45:26 +0800168 p->max_search = dirty_i->nr_dirty[DIRTY];
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900169 p->ofs_unit = sbi->segs_per_sec;
170 }
Jin Xua26b7c82013-09-05 12:45:26 +0800171
Hou Pengyange93b9862017-02-16 12:34:31 +0000172 /* we need to check every dirty segments in the FG_GC case */
173 if (gc_type != FG_GC && p->max_search > sbi->max_victim_search)
Jaegeuk Kimb1c57c12014-01-08 13:45:08 +0900174 p->max_search = sbi->max_victim_search;
Jin Xua26b7c82013-09-05 12:45:26 +0800175
Jaegeuk Kim7a20b8a2017-03-24 20:41:45 -0400176 /* let's select beginning hot/small space first */
177 if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
178 p->offset = 0;
179 else
Jaegeuk Kime066b832017-04-13 15:17:00 -0700180 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900181}
182
183static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
184 struct victim_sel_policy *p)
185{
Jaegeuk Kimb7250d22013-02-05 13:19:28 +0900186 /* SSR allocates in a segment unit */
187 if (p->alloc_mode == SSR)
Chao Yu3519e3f2015-12-01 11:56:52 +0800188 return sbi->blocks_per_seg;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900189 if (p->gc_mode == GC_GREEDY)
Jaegeuk Kimc541a512017-03-25 00:03:02 -0700190 return 2 * sbi->blocks_per_seg * p->ofs_unit;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900191 else if (p->gc_mode == GC_CB)
192 return UINT_MAX;
193 else /* No other gc_mode */
194 return 0;
195}
196
197static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
198{
199 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900200 unsigned int secno;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900201
202 /*
203 * If the gc_type is FG_GC, we can select victim segments
204 * selected by background GC before.
205 * Those segments guarantee they have small valid blocks.
206 */
Jaegeuk Kim7cd85582014-09-23 11:23:01 -0700207 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900208 if (sec_usage_check(sbi, secno))
Chao Yub65ee142014-08-04 10:10:07 +0800209 continue;
Hou Pengyange93b9862017-02-16 12:34:31 +0000210
211 if (no_fggc_candidate(sbi, secno))
212 continue;
213
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900214 clear_bit(secno, dirty_i->victim_secmap);
Jaegeuk Kim4ddb1a42017-04-07 15:08:17 -0700215 return GET_SEG_FROM_SEC(sbi, secno);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900216 }
217 return NULL_SEGNO;
218}
219
220static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
221{
222 struct sit_info *sit_i = SIT_I(sbi);
Jaegeuk Kim4ddb1a42017-04-07 15:08:17 -0700223 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
224 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900225 unsigned long long mtime = 0;
226 unsigned int vblocks;
227 unsigned char age = 0;
228 unsigned char u;
229 unsigned int i;
230
231 for (i = 0; i < sbi->segs_per_sec; i++)
232 mtime += get_seg_entry(sbi, start + i)->mtime;
Jaegeuk Kim302bd342017-04-07 14:33:22 -0700233 vblocks = get_valid_blocks(sbi, segno, true);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900234
235 mtime = div_u64(mtime, sbi->segs_per_sec);
236 vblocks = div_u64(vblocks, sbi->segs_per_sec);
237
238 u = (vblocks * 100) >> sbi->log_blocks_per_seg;
239
arter97e1c42042014-08-06 23:22:50 +0900240 /* Handle if the system time has changed by the user */
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900241 if (mtime < sit_i->min_mtime)
242 sit_i->min_mtime = mtime;
243 if (mtime > sit_i->max_mtime)
244 sit_i->max_mtime = mtime;
245 if (sit_i->max_mtime != sit_i->min_mtime)
246 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
247 sit_i->max_mtime - sit_i->min_mtime);
248
249 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
250}
251
Hou Pengyangb9cd2062017-02-22 10:28:59 +0000252static unsigned int get_greedy_cost(struct f2fs_sb_info *sbi,
253 unsigned int segno)
254{
255 unsigned int valid_blocks =
Jaegeuk Kim302bd342017-04-07 14:33:22 -0700256 get_valid_blocks(sbi, segno, true);
Hou Pengyangb9cd2062017-02-22 10:28:59 +0000257
258 return IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
259 valid_blocks * 2 : valid_blocks;
260}
261
Yunlei Heb7b7c4c2017-05-17 17:22:51 +0800262static unsigned int get_ssr_cost(struct f2fs_sb_info *sbi,
263 unsigned int segno)
264{
265 struct seg_entry *se = get_seg_entry(sbi, segno);
266
267 return se->ckpt_valid_blocks > se->valid_blocks ?
268 se->ckpt_valid_blocks : se->valid_blocks;
269}
270
Jin Xua57e5642013-09-13 08:38:54 +0800271static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
272 unsigned int segno, struct victim_sel_policy *p)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900273{
274 if (p->alloc_mode == SSR)
Yunlei Heb7b7c4c2017-05-17 17:22:51 +0800275 return get_ssr_cost(sbi, segno);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900276
277 /* alloc_mode == LFS */
278 if (p->gc_mode == GC_GREEDY)
Hou Pengyangb9cd2062017-02-22 10:28:59 +0000279 return get_greedy_cost(sbi, segno);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900280 else
281 return get_cb_cost(sbi, segno);
282}
283
Fan Li688159b2016-02-03 16:21:57 +0800284static 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
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900296/*
Masanari Iida111d2492013-03-19 08:03:35 +0900297 * This function is called from two paths.
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900298 * 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 */
304static 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);
Jaegeuk Kime066b832017-04-13 15:17:00 -0700308 struct sit_info *sm = SIT_I(sbi);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900309 struct victim_sel_policy p;
Sheng Yong3fa56502016-09-29 18:37:31 +0800310 unsigned int secno, last_victim;
Chao Yua43f7ec2015-10-05 22:19:24 +0800311 unsigned int last_segment = MAIN_SEGS(sbi);
Fan Li688159b2016-02-03 16:21:57 +0800312 unsigned int nsearched = 0;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900313
Chao Yu210f41b2014-09-15 18:05:44 +0800314 mutex_lock(&dirty_i->seglist_lock);
315
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900316 p.alloc_mode = alloc_mode;
317 select_policy(sbi, gc_type, type, &p);
318
319 p.min_segno = NULL_SEGNO;
Sheng Yong3fa56502016-09-29 18:37:31 +0800320 p.min_cost = get_max_cost(sbi, &p);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900321
Jaegeuk Kime066b832017-04-13 15:17:00 -0700322 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
Chao Yu3342bb32015-10-05 22:20:40 +0800330 if (p.max_search == 0)
331 goto out;
332
Jaegeuk Kime066b832017-04-13 15:17:00 -0700333 last_victim = sm->last_victim[p.gc_mode];
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900334 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;
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900342 unsigned int segno;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900343
Chao Yua43f7ec2015-10-05 22:19:24 +0800344 segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
345 if (segno >= last_segment) {
Jaegeuk Kime066b832017-04-13 15:17:00 -0700346 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;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900350 p.offset = 0;
351 continue;
352 }
353 break;
354 }
Jin Xua57e5642013-09-13 08:38:54 +0800355
356 p.offset = segno + p.ofs_unit;
Fan Li688159b2016-02-03 16:21:57 +0800357 if (p.ofs_unit > 1) {
Jin Xua57e5642013-09-13 08:38:54 +0800358 p.offset -= segno % p.ofs_unit;
Fan Li688159b2016-02-03 16:21:57 +0800359 nsearched += count_bits(p.dirty_segmap,
360 p.offset - p.ofs_unit,
361 p.ofs_unit);
362 } else {
363 nsearched++;
364 }
365
Jaegeuk Kim4ddb1a42017-04-07 15:08:17 -0700366 secno = GET_SEC_FROM_SEG(sbi, segno);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900367
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900368 if (sec_usage_check(sbi, secno))
Fan Li688159b2016-02-03 16:21:57 +0800369 goto next;
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900370 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
Fan Li688159b2016-02-03 16:21:57 +0800371 goto next;
Hou Pengyange93b9862017-02-16 12:34:31 +0000372 if (gc_type == FG_GC && p.alloc_mode == LFS &&
373 no_fggc_candidate(sbi, secno))
374 goto next;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900375
376 cost = get_gc_cost(sbi, segno, &p);
377
378 if (p.min_cost > cost) {
379 p.min_segno = segno;
380 p.min_cost = cost;
Jin Xua57e5642013-09-13 08:38:54 +0800381 }
Fan Li688159b2016-02-03 16:21:57 +0800382next:
383 if (nsearched >= p.max_search) {
Jaegeuk Kime066b832017-04-13 15:17:00 -0700384 if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
385 sm->last_victim[p.gc_mode] = last_victim + 1;
Jaegeuk Kim4ce53772016-02-18 16:34:38 -0800386 else
Jaegeuk Kime066b832017-04-13 15:17:00 -0700387 sm->last_victim[p.gc_mode] = segno + 1;
388 sm->last_victim[p.gc_mode] %= MAIN_SEGS(sbi);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900389 break;
390 }
391 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900392 if (p.min_segno != NULL_SEGNO) {
Namjae Jeonb2b34602013-06-01 16:20:26 +0900393got_it:
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900394 if (p.alloc_mode == LFS) {
Jaegeuk Kim4ddb1a42017-04-07 15:08:17 -0700395 secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900396 if (gc_type == FG_GC)
397 sbi->cur_victim_sec = secno;
398 else
399 set_bit(secno, dirty_i->victim_secmap);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900400 }
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900401 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
Namjae Jeon8e46b3e2013-04-23 16:42:53 +0900402
403 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
404 sbi->cur_victim_sec,
405 prefree_segments(sbi), free_segments(sbi));
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900406 }
Chao Yu3342bb32015-10-05 22:20:40 +0800407out:
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900408 mutex_unlock(&dirty_i->seglist_lock);
409
410 return (p.min_segno == NULL_SEGNO) ? 0 : 1;
411}
412
413static const struct victim_selection default_v_ops = {
414 .get_victim = get_victim_by_default,
415};
416
Changman Lee7dda2af2014-11-28 15:49:40 +0000417static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900418{
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900419 struct inode_entry *ie;
420
Changman Lee7dda2af2014-11-28 15:49:40 +0000421 ie = radix_tree_lookup(&gc_list->iroot, ino);
422 if (ie)
423 return ie->inode;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900424 return NULL;
425}
426
Changman Lee7dda2af2014-11-28 15:49:40 +0000427static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900428{
Gu Zheng6cc4af52013-06-20 17:52:39 +0800429 struct inode_entry *new_ie;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900430
Changman Lee7dda2af2014-11-28 15:49:40 +0000431 if (inode == find_gc_inode(gc_list, inode->i_ino)) {
Gu Zheng6cc4af52013-06-20 17:52:39 +0800432 iput(inode);
433 return;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900434 }
Chao Yu06292072014-12-29 15:56:18 +0800435 new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900436 new_ie->inode = inode;
Chao Yuf28e5032015-01-23 20:37:53 +0800437
438 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
Changman Lee7dda2af2014-11-28 15:49:40 +0000439 list_add_tail(&new_ie->list, &gc_list->ilist);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900440}
441
Changman Lee7dda2af2014-11-28 15:49:40 +0000442static void put_gc_inode(struct gc_inode_list *gc_list)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900443{
444 struct inode_entry *ie, *next_ie;
Changman Lee7dda2af2014-11-28 15:49:40 +0000445 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
446 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900447 iput(ie->inode);
448 list_del(&ie->list);
Chao Yu06292072014-12-29 15:56:18 +0800449 kmem_cache_free(inode_entry_slab, ie);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900450 }
451}
452
453static int check_valid_map(struct f2fs_sb_info *sbi,
454 unsigned int segno, int offset)
455{
456 struct sit_info *sit_i = SIT_I(sbi);
457 struct seg_entry *sentry;
458 int ret;
459
460 mutex_lock(&sit_i->sentry_lock);
461 sentry = get_seg_entry(sbi, segno);
462 ret = f2fs_test_bit(offset, sentry->cur_valid_map);
463 mutex_unlock(&sit_i->sentry_lock);
Jaegeuk Kim43727522013-02-04 15:11:17 +0900464 return ret;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900465}
466
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900467/*
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900468 * This function compares node address got in summary with that in NAT.
469 * On validity, copy that node with cold status, otherwise (invalid node)
470 * ignore that.
471 */
Chao Yu718e53f2016-01-23 16:23:55 +0800472static void gc_node_segment(struct f2fs_sb_info *sbi,
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900473 struct f2fs_summary *sum, unsigned int segno, int gc_type)
474{
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900475 struct f2fs_summary *entry;
Jaegeuk Kim26d58592015-08-14 14:37:50 -0700476 block_t start_addr;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900477 int off;
Chao Yu7ea984b2016-08-27 00:14:31 +0800478 int phase = 0;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900479
Jaegeuk Kim26d58592015-08-14 14:37:50 -0700480 start_addr = START_BLOCK(sbi, segno);
481
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900482next_step:
483 entry = sum;
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900484
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900485 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
486 nid_t nid = le32_to_cpu(entry->nid);
487 struct page *node_page;
Jaegeuk Kim26d58592015-08-14 14:37:50 -0700488 struct node_info ni;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900489
Jaegeuk Kim43727522013-02-04 15:11:17 +0900490 /* stop BG_GC if there is not enough free sections. */
Jaegeuk Kim7f3037a2016-09-01 12:02:51 -0700491 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
Chao Yu718e53f2016-01-23 16:23:55 +0800492 return;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900493
Jaegeuk Kim43727522013-02-04 15:11:17 +0900494 if (check_valid_map(sbi, segno, off) == 0)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900495 continue;
496
Chao Yu7ea984b2016-08-27 00:14:31 +0800497 if (phase == 0) {
498 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
499 META_NAT, true);
500 continue;
501 }
502
503 if (phase == 1) {
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900504 ra_node_page(sbi, nid);
505 continue;
506 }
Chao Yu7ea984b2016-08-27 00:14:31 +0800507
508 /* phase == 2 */
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900509 node_page = get_node_page(sbi, nid);
510 if (IS_ERR(node_page))
511 continue;
512
Huang Ying9a01b562014-09-07 11:05:20 +0800513 /* block may become invalid during get_node_page */
514 if (check_valid_map(sbi, segno, off) == 0) {
515 f2fs_put_page(node_page, 1);
516 continue;
517 }
518
Jaegeuk Kim26d58592015-08-14 14:37:50 -0700519 get_node_info(sbi, nid, &ni);
520 if (ni.blk_addr != start_addr + off) {
521 f2fs_put_page(node_page, 1);
522 continue;
523 }
524
Chao Yuda011cc2016-04-27 21:40:15 +0800525 move_node_page(node_page, gc_type);
Changman Leee1235982014-12-23 08:37:39 +0900526 stat_inc_node_blk_count(sbi, 1, gc_type);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900527 }
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900528
Chao Yu7ea984b2016-08-27 00:14:31 +0800529 if (++phase < 3)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900530 goto next_step;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900531}
532
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900533/*
Jaegeuk Kim9af45ef2013-01-21 17:34:21 +0900534 * Calculate start block index indicating the given node offset.
535 * Be careful, caller should give this node offset only indicating direct node
536 * blocks. If any node offsets, which point the other types of node blocks such
537 * as indirect or double indirect node blocks, are given, it must be a caller's
538 * bug.
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900539 */
Chao Yu81ca7352016-01-26 15:39:35 +0800540block_t start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900541{
Jaegeuk Kimce19a5d2012-12-26 12:03:22 +0900542 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
543 unsigned int bidx;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900544
Jaegeuk Kimce19a5d2012-12-26 12:03:22 +0900545 if (node_ofs == 0)
546 return 0;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900547
Jaegeuk Kimce19a5d2012-12-26 12:03:22 +0900548 if (node_ofs <= 2) {
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900549 bidx = node_ofs - 1;
550 } else if (node_ofs <= indirect_blks) {
Jaegeuk Kimce19a5d2012-12-26 12:03:22 +0900551 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900552 bidx = node_ofs - 2 - dec;
553 } else {
Jaegeuk Kimce19a5d2012-12-26 12:03:22 +0900554 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900555 bidx = node_ofs - 5 - dec;
556 }
Chao Yu81ca7352016-01-26 15:39:35 +0800557 return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900558}
559
Nicholas Krausec1079892015-06-30 21:37:21 -0400560static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900561 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
562{
563 struct page *node_page;
564 nid_t nid;
565 unsigned int ofs_in_node;
566 block_t source_blkaddr;
567
568 nid = le32_to_cpu(sum->nid);
569 ofs_in_node = le16_to_cpu(sum->ofs_in_node);
570
571 node_page = get_node_page(sbi, nid);
572 if (IS_ERR(node_page))
Nicholas Krausec1079892015-06-30 21:37:21 -0400573 return false;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900574
575 get_node_info(sbi, nid, dni);
576
577 if (sum->version != dni->version) {
Jaegeuk Kimc13ff372017-03-21 10:59:50 -0400578 f2fs_msg(sbi->sb, KERN_WARNING,
579 "%s: valid data with mismatched node version.",
580 __func__);
581 set_sbi_flag(sbi, SBI_NEED_FSCK);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900582 }
583
584 *nofs = ofs_of_node(node_page);
585 source_blkaddr = datablock_addr(node_page, ofs_in_node);
586 f2fs_put_page(node_page, 1);
587
588 if (source_blkaddr != blkaddr)
Nicholas Krausec1079892015-06-30 21:37:21 -0400589 return false;
590 return true;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900591}
592
Yunlei He20614712016-11-07 21:22:31 +0800593static void move_encrypted_block(struct inode *inode, block_t bidx,
594 unsigned int segno, int off)
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700595{
596 struct f2fs_io_info fio = {
597 .sbi = F2FS_I_SB(inode),
598 .type = DATA,
Jaegeuk Kima912b542017-05-10 11:18:25 -0700599 .temp = COLD,
Mike Christie04d328d2016-06-05 14:31:55 -0500600 .op = REQ_OP_READ,
Christoph Hellwig70fd7612016-11-01 07:40:10 -0600601 .op_flags = 0,
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700602 .encrypted_page = NULL,
Chao Yufb830fc2017-05-19 23:37:01 +0800603 .in_list = false,
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700604 };
605 struct dnode_of_data dn;
606 struct f2fs_summary sum;
607 struct node_info ni;
608 struct page *page;
Chao Yu4356e482016-02-23 17:52:43 +0800609 block_t newaddr;
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700610 int err;
611
612 /* do not read out */
Jaegeuk Kima56c7c62015-10-09 15:11:38 -0700613 page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700614 if (!page)
615 return;
616
Yunlei He20614712016-11-07 21:22:31 +0800617 if (!check_valid_map(F2FS_I_SB(inode), segno, off))
618 goto out;
619
Chao Yu5fe45742017-01-07 18:50:26 +0800620 if (f2fs_is_atomic_file(inode))
621 goto out;
622
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700623 set_new_dnode(&dn, inode, NULL, NULL, 0);
624 err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
625 if (err)
626 goto out;
627
Chao Yu08b39fb2015-10-08 13:27:34 +0800628 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
629 ClearPageUptodate(page);
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700630 goto put_out;
Chao Yu08b39fb2015-10-08 13:27:34 +0800631 }
632
633 /*
634 * don't cache encrypted data into meta inode until previous dirty
635 * data were writebacked to avoid racing between GC and flush.
636 */
Jaegeuk Kimfec1d652016-01-20 23:43:51 +0800637 f2fs_wait_on_page_writeback(page, DATA, true);
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700638
639 get_node_info(fio.sbi, dn.nid, &ni);
640 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
641
642 /* read page */
643 fio.page = page;
Chao Yu7a9d7542016-02-22 18:36:38 +0800644 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700645
Chao Yu4356e482016-02-23 17:52:43 +0800646 allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
Chao Yufb830fc2017-05-19 23:37:01 +0800647 &sum, CURSEG_COLD_DATA, NULL, false);
Chao Yu4356e482016-02-23 17:52:43 +0800648
649 fio.encrypted_page = pagecache_get_page(META_MAPPING(fio.sbi), newaddr,
650 FGP_LOCK | FGP_CREAT, GFP_NOFS);
651 if (!fio.encrypted_page) {
652 err = -ENOMEM;
653 goto recover_block;
654 }
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700655
Jaegeuk Kim548aeda2015-07-13 17:44:14 -0700656 err = f2fs_submit_page_bio(&fio);
657 if (err)
658 goto put_page_out;
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700659
660 /* write page */
661 lock_page(fio.encrypted_page);
Jaegeuk Kim548aeda2015-07-13 17:44:14 -0700662
Chao Yu1563ac72016-07-03 22:05:12 +0800663 if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) {
Chao Yu4356e482016-02-23 17:52:43 +0800664 err = -EIO;
Jaegeuk Kim548aeda2015-07-13 17:44:14 -0700665 goto put_page_out;
Chao Yu4356e482016-02-23 17:52:43 +0800666 }
Chao Yu1563ac72016-07-03 22:05:12 +0800667 if (unlikely(!PageUptodate(fio.encrypted_page))) {
Chao Yu4356e482016-02-23 17:52:43 +0800668 err = -EIO;
Jaegeuk Kim548aeda2015-07-13 17:44:14 -0700669 goto put_page_out;
Chao Yu4356e482016-02-23 17:52:43 +0800670 }
Jaegeuk Kim548aeda2015-07-13 17:44:14 -0700671
Jaegeuk Kim6282adb2015-07-25 00:29:17 -0700672 set_page_dirty(fio.encrypted_page);
Jaegeuk Kimfec1d652016-01-20 23:43:51 +0800673 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true);
Jaegeuk Kim6282adb2015-07-25 00:29:17 -0700674 if (clear_page_dirty_for_io(fio.encrypted_page))
675 dec_page_count(fio.sbi, F2FS_DIRTY_META);
676
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700677 set_page_writeback(fio.encrypted_page);
Jaegeuk Kim548aeda2015-07-13 17:44:14 -0700678
679 /* allocate block address */
Jaegeuk Kimfec1d652016-01-20 23:43:51 +0800680 f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
Chao Yu4356e482016-02-23 17:52:43 +0800681
Mike Christie04d328d2016-06-05 14:31:55 -0500682 fio.op = REQ_OP_WRITE;
Christoph Hellwig70fd7612016-11-01 07:40:10 -0600683 fio.op_flags = REQ_SYNC;
Chao Yu4356e482016-02-23 17:52:43 +0800684 fio.new_blkaddr = newaddr;
Jaegeuk Kimb9109b02017-05-10 11:28:38 -0700685 f2fs_submit_page_write(&fio);
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700686
Chao Yuf28b3432016-02-24 17:16:47 +0800687 f2fs_update_data_blkaddr(&dn, newaddr);
Jaegeuk Kim91942322016-05-20 10:13:22 -0700688 set_inode_flag(inode, FI_APPEND_WRITE);
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700689 if (page->index == 0)
Jaegeuk Kim91942322016-05-20 10:13:22 -0700690 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
Jaegeuk Kim548aeda2015-07-13 17:44:14 -0700691put_page_out:
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700692 f2fs_put_page(fio.encrypted_page, 1);
Chao Yu4356e482016-02-23 17:52:43 +0800693recover_block:
694 if (err)
695 __f2fs_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
696 true, true);
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700697put_out:
698 f2fs_put_dnode(&dn);
699out:
700 f2fs_put_page(page, 1);
701}
702
Yunlei He20614712016-11-07 21:22:31 +0800703static void move_data_page(struct inode *inode, block_t bidx, int gc_type,
704 unsigned int segno, int off)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900705{
Jaegeuk Kimc879f902015-04-24 14:34:30 -0700706 struct page *page;
707
Jaegeuk Kima56c7c62015-10-09 15:11:38 -0700708 page = get_lock_data_page(inode, bidx, true);
Jaegeuk Kimc879f902015-04-24 14:34:30 -0700709 if (IS_ERR(page))
710 return;
Fan Li63a0b7cb2013-12-09 16:09:00 +0800711
Yunlei He20614712016-11-07 21:22:31 +0800712 if (!check_valid_map(F2FS_I_SB(inode), segno, off))
713 goto out;
714
Chao Yu5fe45742017-01-07 18:50:26 +0800715 if (f2fs_is_atomic_file(inode))
716 goto out;
717
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900718 if (gc_type == BG_GC) {
Jaegeuk Kim4ebefc42013-03-31 13:49:18 +0900719 if (PageWriteback(page))
720 goto out;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900721 set_page_dirty(page);
722 set_cold_data(page);
723 } else {
Jaegeuk Kimc879f902015-04-24 14:34:30 -0700724 struct f2fs_io_info fio = {
725 .sbi = F2FS_I_SB(inode),
726 .type = DATA,
Jaegeuk Kima912b542017-05-10 11:18:25 -0700727 .temp = COLD,
Mike Christie04d328d2016-06-05 14:31:55 -0500728 .op = REQ_OP_WRITE,
Christoph Hellwig70fd7612016-11-01 07:40:10 -0600729 .op_flags = REQ_SYNC,
Hou Pengyange959c8f2017-04-25 12:45:13 +0000730 .old_blkaddr = NULL_ADDR,
Jaegeuk Kimc879f902015-04-24 14:34:30 -0700731 .page = page,
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700732 .encrypted_page = NULL,
Jaegeuk Kimcc156202017-05-12 13:51:34 -0700733 .need_lock = LOCK_REQ,
Jaegeuk Kimc879f902015-04-24 14:34:30 -0700734 };
Chao Yu72e1c792016-07-03 22:05:13 +0800735 bool is_dirty = PageDirty(page);
736 int err;
737
738retry:
Jaegeuk Kim6282adb2015-07-25 00:29:17 -0700739 set_page_dirty(page);
Jaegeuk Kimfec1d652016-01-20 23:43:51 +0800740 f2fs_wait_on_page_writeback(page, DATA, true);
Chao Yu933439c2016-10-11 22:57:01 +0800741 if (clear_page_dirty_for_io(page)) {
Jaegeuk Kima7ffdbe2014-09-12 15:53:45 -0700742 inode_dec_dirty_pages(inode);
Chao Yu933439c2016-10-11 22:57:01 +0800743 remove_dirty_inode(inode);
744 }
Chao Yu72e1c792016-07-03 22:05:13 +0800745
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900746 set_cold_data(page);
Chao Yu72e1c792016-07-03 22:05:13 +0800747
748 err = do_write_data_page(&fio);
749 if (err == -ENOMEM && is_dirty) {
750 congestion_wait(BLK_RW_ASYNC, HZ/50);
751 goto retry;
752 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900753 }
754out:
755 f2fs_put_page(page, 1);
756}
757
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900758/*
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900759 * This function tries to get parent node of victim data block, and identifies
760 * data block validity. If the block is valid, copy that with cold status and
761 * modify parent node.
762 * If the parent node is not valid or the data block address is different,
763 * the victim data block is ignored.
764 */
Chao Yu718e53f2016-01-23 16:23:55 +0800765static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
Changman Lee7dda2af2014-11-28 15:49:40 +0000766 struct gc_inode_list *gc_list, unsigned int segno, int gc_type)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900767{
768 struct super_block *sb = sbi->sb;
769 struct f2fs_summary *entry;
770 block_t start_addr;
Jaegeuk Kim43727522013-02-04 15:11:17 +0900771 int off;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900772 int phase = 0;
773
774 start_addr = START_BLOCK(sbi, segno);
775
776next_step:
777 entry = sum;
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900778
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900779 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
780 struct page *data_page;
781 struct inode *inode;
782 struct node_info dni; /* dnode info for the data */
783 unsigned int ofs_in_node, nofs;
784 block_t start_bidx;
Chao Yu7ea984b2016-08-27 00:14:31 +0800785 nid_t nid = le32_to_cpu(entry->nid);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900786
Jaegeuk Kim43727522013-02-04 15:11:17 +0900787 /* stop BG_GC if there is not enough free sections. */
Jaegeuk Kim7f3037a2016-09-01 12:02:51 -0700788 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
Chao Yu718e53f2016-01-23 16:23:55 +0800789 return;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900790
Jaegeuk Kim43727522013-02-04 15:11:17 +0900791 if (check_valid_map(sbi, segno, off) == 0)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900792 continue;
793
794 if (phase == 0) {
Chao Yu7ea984b2016-08-27 00:14:31 +0800795 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
796 META_NAT, true);
797 continue;
798 }
799
800 if (phase == 1) {
801 ra_node_page(sbi, nid);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900802 continue;
803 }
804
805 /* Get an inode by ino with checking validity */
Nicholas Krausec1079892015-06-30 21:37:21 -0400806 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900807 continue;
808
Chao Yu7ea984b2016-08-27 00:14:31 +0800809 if (phase == 2) {
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900810 ra_node_page(sbi, dni.ino);
811 continue;
812 }
813
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900814 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
815
Chao Yu7ea984b2016-08-27 00:14:31 +0800816 if (phase == 3) {
Jaegeuk Kimd4686d562013-01-31 15:36:04 +0900817 inode = f2fs_iget(sb, dni.ino);
Chao Yub73e5282014-08-30 09:52:34 +0800818 if (IS_ERR(inode) || is_bad_inode(inode))
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900819 continue;
820
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700821 /* if encrypted inode, let's go phase 3 */
822 if (f2fs_encrypted_inode(inode) &&
823 S_ISREG(inode->i_mode)) {
824 add_gc_inode(gc_list, inode);
825 continue;
826 }
827
Chao Yu81ca7352016-01-26 15:39:35 +0800828 start_bidx = start_bidx_of_node(nofs, inode);
Jaegeuk Kim43f3eae2015-04-30 17:00:33 -0700829 data_page = get_read_data_page(inode,
Christoph Hellwig70246282016-07-19 11:28:41 +0200830 start_bidx + ofs_in_node, REQ_RAHEAD,
831 true);
Changman Lee31a32682014-11-27 16:03:08 +0900832 if (IS_ERR(data_page)) {
833 iput(inode);
834 continue;
835 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900836
837 f2fs_put_page(data_page, 0);
Changman Lee7dda2af2014-11-28 15:49:40 +0000838 add_gc_inode(gc_list, inode);
Changman Lee31a32682014-11-27 16:03:08 +0900839 continue;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900840 }
Changman Lee31a32682014-11-27 16:03:08 +0900841
Chao Yu7ea984b2016-08-27 00:14:31 +0800842 /* phase 4 */
Changman Lee7dda2af2014-11-28 15:49:40 +0000843 inode = find_gc_inode(gc_list, dni.ino);
Changman Lee31a32682014-11-27 16:03:08 +0900844 if (inode) {
Chao Yu82e0a5a2016-07-13 09:18:29 +0800845 struct f2fs_inode_info *fi = F2FS_I(inode);
846 bool locked = false;
847
848 if (S_ISREG(inode->i_mode)) {
849 if (!down_write_trylock(&fi->dio_rwsem[READ]))
850 continue;
851 if (!down_write_trylock(
852 &fi->dio_rwsem[WRITE])) {
853 up_write(&fi->dio_rwsem[READ]);
854 continue;
855 }
856 locked = true;
857 }
858
Chao Yu81ca7352016-01-26 15:39:35 +0800859 start_bidx = start_bidx_of_node(nofs, inode)
Jaegeuk Kimc879f902015-04-24 14:34:30 -0700860 + ofs_in_node;
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700861 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
Yunlei He20614712016-11-07 21:22:31 +0800862 move_encrypted_block(inode, start_bidx, segno, off);
Jaegeuk Kim4375a332015-04-23 12:04:33 -0700863 else
Yunlei He20614712016-11-07 21:22:31 +0800864 move_data_page(inode, start_bidx, gc_type, segno, off);
Chao Yu82e0a5a2016-07-13 09:18:29 +0800865
866 if (locked) {
867 up_write(&fi->dio_rwsem[WRITE]);
868 up_write(&fi->dio_rwsem[READ]);
869 }
870
Changman Leee1235982014-12-23 08:37:39 +0900871 stat_inc_data_blk_count(sbi, 1, gc_type);
Changman Lee31a32682014-11-27 16:03:08 +0900872 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900873 }
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900874
Chao Yu7ea984b2016-08-27 00:14:31 +0800875 if (++phase < 5)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900876 goto next_step;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900877}
878
879static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
Gu Zheng8a2d0ac2014-10-20 17:45:48 +0800880 int gc_type)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900881{
882 struct sit_info *sit_i = SIT_I(sbi);
883 int ret;
Gu Zheng8a2d0ac2014-10-20 17:45:48 +0800884
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900885 mutex_lock(&sit_i->sentry_lock);
Gu Zheng8a2d0ac2014-10-20 17:45:48 +0800886 ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
887 NO_CHECK_TYPE, LFS);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900888 mutex_unlock(&sit_i->sentry_lock);
889 return ret;
890}
891
Chao Yu718e53f2016-01-23 16:23:55 +0800892static int do_garbage_collect(struct f2fs_sb_info *sbi,
893 unsigned int start_segno,
Changman Lee7dda2af2014-11-28 15:49:40 +0000894 struct gc_inode_list *gc_list, int gc_type)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900895{
896 struct page *sum_page;
897 struct f2fs_summary_block *sum;
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900898 struct blk_plug plug;
Chao Yu718e53f2016-01-23 16:23:55 +0800899 unsigned int segno = start_segno;
900 unsigned int end_segno = start_segno + sbi->segs_per_sec;
Chao Yu43ced842016-08-19 23:13:46 +0800901 int sec_freed = 0;
Chao Yu718e53f2016-01-23 16:23:55 +0800902 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
903 SUM_TYPE_DATA : SUM_TYPE_NODE;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900904
Chao Yu718e53f2016-01-23 16:23:55 +0800905 /* readahead multi ssa blocks those have contiguous address */
906 if (sbi->segs_per_sec > 1)
907 ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
908 sbi->segs_per_sec, META_SSA, true);
909
910 /* reference all summary page */
911 while (segno < end_segno) {
912 sum_page = get_sum_page(sbi, segno++);
913 unlock_page(sum_page);
914 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900915
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900916 blk_start_plug(&plug);
917
Chao Yu718e53f2016-01-23 16:23:55 +0800918 for (segno = start_segno; segno < end_segno; segno++) {
Jaegeuk Kimaa987272016-06-06 18:49:54 -0700919
Chao Yu718e53f2016-01-23 16:23:55 +0800920 /* find segment summary of victim */
921 sum_page = find_get_page(META_MAPPING(sbi),
922 GET_SUM_BLOCK(sbi, segno));
Chao Yu718e53f2016-01-23 16:23:55 +0800923 f2fs_put_page(sum_page, 0);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900924
Jaegeuk Kim302bd342017-04-07 14:33:22 -0700925 if (get_valid_blocks(sbi, segno, false) == 0 ||
Jaegeuk Kimde0dcc42016-10-12 13:38:41 -0700926 !PageUptodate(sum_page) ||
927 unlikely(f2fs_cp_error(sbi)))
928 goto next;
929
Chao Yu718e53f2016-01-23 16:23:55 +0800930 sum = page_address(sum_page);
931 f2fs_bug_on(sbi, type != GET_SUM_TYPE((&sum->footer)));
Jaegeuk Kim9236cac2015-05-28 18:19:17 -0700932
Chao Yu718e53f2016-01-23 16:23:55 +0800933 /*
934 * this is to avoid deadlock:
935 * - lock_page(sum_page) - f2fs_replace_block
936 * - check_valid_map() - mutex_lock(sentry_lock)
937 * - mutex_lock(sentry_lock) - change_curseg()
938 * - lock_page(sum_page)
939 */
Chao Yu718e53f2016-01-23 16:23:55 +0800940 if (type == SUM_TYPE_NODE)
941 gc_node_segment(sbi, sum->entries, segno, gc_type);
942 else
943 gc_data_segment(sbi, sum->entries, gc_list, segno,
944 gc_type);
945
946 stat_inc_seg_count(sbi, type, gc_type);
Jaegeuk Kimf6fe2be2016-09-21 09:34:48 -0700947next:
Chao Yu718e53f2016-01-23 16:23:55 +0800948 f2fs_put_page(sum_page, 0);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900949 }
Chao Yu718e53f2016-01-23 16:23:55 +0800950
Chao Yuda011cc2016-04-27 21:40:15 +0800951 if (gc_type == FG_GC)
Jaegeuk Kimb9109b02017-05-10 11:28:38 -0700952 f2fs_submit_merged_write(sbi,
953 (type == SUM_TYPE_NODE) ? NODE : DATA);
Chao Yu718e53f2016-01-23 16:23:55 +0800954
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900955 blk_finish_plug(&plug);
956
Chao Yu43ced842016-08-19 23:13:46 +0800957 if (gc_type == FG_GC &&
Jaegeuk Kim302bd342017-04-07 14:33:22 -0700958 get_valid_blocks(sbi, start_segno, true) == 0)
Chao Yu43ced842016-08-19 23:13:46 +0800959 sec_freed = 1;
Chao Yu17d899d2016-02-22 18:32:13 +0800960
961 stat_inc_call_count(sbi->stat_info);
962
Chao Yu43ced842016-08-19 23:13:46 +0800963 return sec_freed;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900964}
965
Jaegeuk Kime066b832017-04-13 15:17:00 -0700966int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
967 bool background, unsigned int segno)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900968{
Chao Yud530d4d2015-10-05 22:22:44 +0800969 int gc_type = sync ? FG_GC : BG_GC;
Chao Yu43ced842016-08-19 23:13:46 +0800970 int sec_freed = 0;
Weichao Guoe5dbd952017-05-11 04:28:00 +0800971 int ret;
Jaegeuk Kimd5053a342014-10-30 22:47:03 -0700972 struct cp_control cpc;
Jaegeuk Kime066b832017-04-13 15:17:00 -0700973 unsigned int init_segno = segno;
Changman Lee7dda2af2014-11-28 15:49:40 +0000974 struct gc_inode_list gc_list = {
975 .ilist = LIST_HEAD_INIT(gc_list.ilist),
Jaegeuk Kim769ec6e2014-12-03 20:47:26 -0800976 .iroot = RADIX_TREE_INIT(GFP_NOFS),
Changman Lee7dda2af2014-11-28 15:49:40 +0000977 };
Jaegeuk Kimd5053a342014-10-30 22:47:03 -0700978
Jaegeuk Kim119ee912015-01-29 11:45:33 -0800979 cpc.reason = __get_cp_reason(sbi);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900980gc_more:
Weichao Guoe5dbd952017-05-11 04:28:00 +0800981 if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE))) {
982 ret = -EINVAL;
Jaegeuk Kim408e9372013-01-03 17:55:52 +0900983 goto stop;
Weichao Guoe5dbd952017-05-11 04:28:00 +0800984 }
Chao Yu6d5a1492015-12-24 18:04:56 +0800985 if (unlikely(f2fs_cp_error(sbi))) {
986 ret = -EIO;
Jaegeuk Kim203681f2014-02-05 13:03:57 +0900987 goto stop;
Chao Yu6d5a1492015-12-24 18:04:56 +0800988 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900989
Hou Pengyang19f4e682017-02-25 03:57:38 +0000990 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
Jaegeuk Kim6e17bfb2016-01-23 22:00:57 +0800991 /*
Hou Pengyang19f4e682017-02-25 03:57:38 +0000992 * For example, if there are many prefree_segments below given
993 * threshold, we can make them free by checkpoint. Then, we
994 * secure free segments which doesn't need fggc any more.
Jaegeuk Kim6e17bfb2016-01-23 22:00:57 +0800995 */
Jaegeuk Kim8fd5a372017-04-07 17:25:54 -0700996 if (prefree_segments(sbi)) {
997 ret = write_checkpoint(sbi, &cpc);
998 if (ret)
999 goto stop;
1000 }
Hou Pengyang19f4e682017-02-25 03:57:38 +00001001 if (has_not_enough_free_secs(sbi, 0, 0))
1002 gc_type = FG_GC;
Jaegeuk Kimd64f8042013-04-08 16:01:00 +09001003 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +09001004
Weichao Guoe5dbd952017-05-11 04:28:00 +08001005 ret = -EINVAL;
Hou Pengyang19f4e682017-02-25 03:57:38 +00001006 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1007 if (gc_type == BG_GC && !background)
1008 goto stop;
Yunlei He05eeb112017-02-17 17:16:38 +08001009 if (!__get_victim(sbi, &segno, gc_type))
Jaegeuk Kim408e9372013-01-03 17:55:52 +09001010 goto stop;
Jaegeuk Kim43727522013-02-04 15:11:17 +09001011 ret = 0;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +09001012
Chao Yu43ced842016-08-19 23:13:46 +08001013 if (do_garbage_collect(sbi, segno, &gc_list, gc_type) &&
1014 gc_type == FG_GC)
Chao Yu45fe8492015-09-28 17:42:24 +08001015 sec_freed++;
Jaegeuk Kim43727522013-02-04 15:11:17 +09001016
Jaegeuk Kim5ee52932015-08-15 22:06:08 -07001017 if (gc_type == FG_GC)
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +09001018 sbi->cur_victim_sec = NULL_SEGNO;
Jaegeuk Kim43727522013-02-04 15:11:17 +09001019
Chao Yud530d4d2015-10-05 22:22:44 +08001020 if (!sync) {
Jaegeuk Kime066b832017-04-13 15:17:00 -07001021 if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1022 segno = NULL_SEGNO;
Chao Yud530d4d2015-10-05 22:22:44 +08001023 goto gc_more;
Jaegeuk Kime066b832017-04-13 15:17:00 -07001024 }
Jaegeuk Kim43727522013-02-04 15:11:17 +09001025
Chao Yud530d4d2015-10-05 22:22:44 +08001026 if (gc_type == FG_GC)
Jaegeuk Kim2956e452016-09-21 09:28:06 -07001027 ret = write_checkpoint(sbi, &cpc);
Chao Yud530d4d2015-10-05 22:22:44 +08001028 }
Jaegeuk Kim408e9372013-01-03 17:55:52 +09001029stop:
Jaegeuk Kime066b832017-04-13 15:17:00 -07001030 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1031 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +09001032 mutex_unlock(&sbi->gc_mutex);
1033
Changman Lee7dda2af2014-11-28 15:49:40 +00001034 put_gc_inode(&gc_list);
Chao Yud530d4d2015-10-05 22:22:44 +08001035
1036 if (sync)
1037 ret = sec_freed ? 0 : -EAGAIN;
Jaegeuk Kim43727522013-02-04 15:11:17 +09001038 return ret;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +09001039}
1040
1041void build_gc_manager(struct f2fs_sb_info *sbi)
1042{
Jaegeuk Kim4ddb1a42017-04-07 15:08:17 -07001043 u64 main_count, resv_count, ovp_count;
Hou Pengyange93b9862017-02-16 12:34:31 +00001044
Jaegeuk Kim7bc09002012-11-02 17:13:01 +09001045 DIRTY_I(sbi)->v_ops = &default_v_ops;
Hou Pengyange93b9862017-02-16 12:34:31 +00001046
1047 /* threshold of # of valid blocks in a section for victims of FG_GC */
1048 main_count = SM_I(sbi)->main_segments << sbi->log_blocks_per_seg;
1049 resv_count = SM_I(sbi)->reserved_segments << sbi->log_blocks_per_seg;
1050 ovp_count = SM_I(sbi)->ovp_segments << sbi->log_blocks_per_seg;
Hou Pengyange93b9862017-02-16 12:34:31 +00001051
Jaegeuk Kim4ddb1a42017-04-07 15:08:17 -07001052 sbi->fggc_threshold = div64_u64((main_count - ovp_count) *
1053 BLKS_PER_SEC(sbi), (main_count - resv_count));
Jaegeuk Kimd5793242017-04-18 15:03:15 -07001054
1055 /* give warm/cold data area from slower device */
1056 if (sbi->s_ndevs && sbi->segs_per_sec == 1)
1057 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1058 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +09001059}