blob: b90dbe55403a4dbd0c1504c399a2285d416b6063 [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
27static struct kmem_cache *winode_slab;
28
29static int gc_thread_func(void *data)
30{
31 struct f2fs_sb_info *sbi = data;
Namjae Jeonb59d0ba2013-08-04 23:09:40 +090032 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090033 wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
34 long wait_ms;
35
Namjae Jeonb59d0ba2013-08-04 23:09:40 +090036 wait_ms = gc_th->min_sleep_time;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090037
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
Changman Leed6212a52013-01-29 18:30:07 +090048 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
Namjae Jeonb59d0ba2013-08-04 23:09:40 +090049 wait_ms = increase_sleep_time(gc_th, wait_ms);
Changman Leed6212a52013-01-29 18:30:07 +090050 continue;
51 }
52
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090053 /*
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)) {
Namjae Jeonb59d0ba2013-08-04 23:09:40 +090070 wait_ms = increase_sleep_time(gc_th, wait_ms);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090071 mutex_unlock(&sbi->gc_mutex);
72 continue;
73 }
74
75 if (has_enough_invalid_blocks(sbi))
Namjae Jeonb59d0ba2013-08-04 23:09:40 +090076 wait_ms = decrease_sleep_time(gc_th, wait_ms);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090077 else
Namjae Jeonb59d0ba2013-08-04 23:09:40 +090078 wait_ms = increase_sleep_time(gc_th, wait_ms);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090079
Jaegeuk Kimdcdfff62013-10-22 20:56:10 +090080 stat_inc_bggc_count(sbi);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090081
Jaegeuk Kim43727522013-02-04 15:11:17 +090082 /* if return value is not zero, no victim was selected */
83 if (f2fs_gc(sbi))
Namjae Jeonb59d0ba2013-08-04 23:09:40 +090084 wait_ms = gc_th->no_gc_sleep_time;
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +090085
Jaegeuk Kim4660f9c2013-10-24 14:19:18 +090086 /* balancing f2fs's metadata periodically */
87 f2fs_balance_fs_bg(sbi);
Jaegeuk Kim81eb8d62013-10-24 13:31:34 +090088
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090089 } while (!kthread_should_stop());
90 return 0;
91}
92
93int start_gc_thread(struct f2fs_sb_info *sbi)
94{
Namjae Jeon1042d602012-12-01 10:56:13 +090095 struct f2fs_gc_kthread *gc_th;
Namjae Jeonec7b1f22013-02-02 23:52:28 +090096 dev_t dev = sbi->sb->s_bdev->bd_dev;
Namjae Jeon7a267f82013-05-26 11:05:32 +090097 int err = 0;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +090098
Changman Lee48600e42013-02-04 10:05:09 +090099 if (!test_opt(sbi, BG_GC))
Namjae Jeon7a267f82013-05-26 11:05:32 +0900100 goto out;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900101 gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
Namjae Jeon7a267f82013-05-26 11:05:32 +0900102 if (!gc_th) {
103 err = -ENOMEM;
104 goto out;
105 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900106
Namjae Jeonb59d0ba2013-08-04 23:09:40 +0900107 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
Namjae Jeond2dc0952013-08-04 23:10:15 +0900111 gc_th->gc_idle = 0;
112
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900113 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,
Namjae Jeonec7b1f22013-02-02 23:52:28 +0900116 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900117 if (IS_ERR(gc_th->f2fs_gc_task)) {
Namjae Jeon7a267f82013-05-26 11:05:32 +0900118 err = PTR_ERR(gc_th->f2fs_gc_task);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900119 kfree(gc_th);
Namjae Jeon25718422013-02-02 23:52:42 +0900120 sbi->gc_thread = NULL;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900121 }
Namjae Jeon7a267f82013-05-26 11:05:32 +0900122out:
123 return err;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900124}
125
126void 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
Namjae Jeond2dc0952013-08-04 23:10:15 +0900136static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900137{
Namjae Jeond2dc0952013-08-04 23:10:15 +0900138 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;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900147}
148
149static 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
Jaegeuk Kim4ebefc42013-03-31 13:49:18 +0900154 if (p->alloc_mode == SSR) {
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900155 p->gc_mode = GC_GREEDY;
156 p->dirty_segmap = dirty_i->dirty_segmap[type];
Jin Xua26b7c82013-09-05 12:45:26 +0800157 p->max_search = dirty_i->nr_dirty[type];
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900158 p->ofs_unit = 1;
159 } else {
Namjae Jeond2dc0952013-08-04 23:10:15 +0900160 p->gc_mode = select_gc_type(sbi->gc_thread, gc_type);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900161 p->dirty_segmap = dirty_i->dirty_segmap[DIRTY];
Jin Xua26b7c82013-09-05 12:45:26 +0800162 p->max_search = dirty_i->nr_dirty[DIRTY];
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900163 p->ofs_unit = sbi->segs_per_sec;
164 }
Jin Xua26b7c82013-09-05 12:45:26 +0800165
Jaegeuk Kimb1c57c12014-01-08 13:45:08 +0900166 if (p->max_search > sbi->max_victim_search)
167 p->max_search = sbi->max_victim_search;
Jin Xua26b7c82013-09-05 12:45:26 +0800168
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900169 p->offset = sbi->last_victim[p->gc_mode];
170}
171
172static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
173 struct victim_sel_policy *p)
174{
Jaegeuk Kimb7250d22013-02-05 13:19:28 +0900175 /* SSR allocates in a segment unit */
176 if (p->alloc_mode == SSR)
177 return 1 << sbi->log_blocks_per_seg;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900178 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
186static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
187{
188 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900189 unsigned int hint = 0;
190 unsigned int secno;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900191
192 /*
193 * If the gc_type is FG_GC, we can select victim segments
194 * selected by background GC before.
195 * Those segments guarantee they have small valid blocks.
196 */
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900197next:
198 secno = find_next_bit(dirty_i->victim_secmap, TOTAL_SECS(sbi), hint++);
199 if (secno < TOTAL_SECS(sbi)) {
200 if (sec_usage_check(sbi, secno))
201 goto next;
202 clear_bit(secno, dirty_i->victim_secmap);
203 return secno * sbi->segs_per_sec;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900204 }
205 return NULL_SEGNO;
206}
207
208static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
209{
210 struct sit_info *sit_i = SIT_I(sbi);
211 unsigned int secno = GET_SECNO(sbi, segno);
212 unsigned int start = secno * sbi->segs_per_sec;
213 unsigned long long mtime = 0;
214 unsigned int vblocks;
215 unsigned char age = 0;
216 unsigned char u;
217 unsigned int i;
218
219 for (i = 0; i < sbi->segs_per_sec; i++)
220 mtime += get_seg_entry(sbi, start + i)->mtime;
221 vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec);
222
223 mtime = div_u64(mtime, sbi->segs_per_sec);
224 vblocks = div_u64(vblocks, sbi->segs_per_sec);
225
226 u = (vblocks * 100) >> sbi->log_blocks_per_seg;
227
228 /* Handle if the system time is changed by user */
229 if (mtime < sit_i->min_mtime)
230 sit_i->min_mtime = mtime;
231 if (mtime > sit_i->max_mtime)
232 sit_i->max_mtime = mtime;
233 if (sit_i->max_mtime != sit_i->min_mtime)
234 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
235 sit_i->max_mtime - sit_i->min_mtime);
236
237 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
238}
239
Jin Xua57e5642013-09-13 08:38:54 +0800240static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
241 unsigned int segno, struct victim_sel_policy *p)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900242{
243 if (p->alloc_mode == SSR)
244 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
245
246 /* alloc_mode == LFS */
247 if (p->gc_mode == GC_GREEDY)
248 return get_valid_blocks(sbi, segno, sbi->segs_per_sec);
249 else
250 return get_cb_cost(sbi, segno);
251}
252
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900253/*
Masanari Iida111d2492013-03-19 08:03:35 +0900254 * This function is called from two paths.
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900255 * One is garbage collection and the other is SSR segment selection.
256 * When it is called during GC, it just gets a victim segment
257 * and it does not remove it from dirty seglist.
258 * When it is called from SSR segment selection, it finds a segment
259 * which has minimum valid blocks and removes it from dirty seglist.
260 */
261static int get_victim_by_default(struct f2fs_sb_info *sbi,
262 unsigned int *result, int gc_type, int type, char alloc_mode)
263{
264 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
265 struct victim_sel_policy p;
Namjae Jeonb2b34602013-06-01 16:20:26 +0900266 unsigned int secno, max_cost;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900267 int nsearched = 0;
268
269 p.alloc_mode = alloc_mode;
270 select_policy(sbi, gc_type, type, &p);
271
272 p.min_segno = NULL_SEGNO;
Namjae Jeonb2b34602013-06-01 16:20:26 +0900273 p.min_cost = max_cost = get_max_cost(sbi, &p);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900274
275 mutex_lock(&dirty_i->seglist_lock);
276
277 if (p.alloc_mode == LFS && gc_type == FG_GC) {
278 p.min_segno = check_bg_victims(sbi);
279 if (p.min_segno != NULL_SEGNO)
280 goto got_it;
281 }
282
283 while (1) {
284 unsigned long cost;
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900285 unsigned int segno;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900286
287 segno = find_next_bit(p.dirty_segmap,
288 TOTAL_SEGS(sbi), p.offset);
289 if (segno >= TOTAL_SEGS(sbi)) {
290 if (sbi->last_victim[p.gc_mode]) {
291 sbi->last_victim[p.gc_mode] = 0;
292 p.offset = 0;
293 continue;
294 }
295 break;
296 }
Jin Xua57e5642013-09-13 08:38:54 +0800297
298 p.offset = segno + p.ofs_unit;
299 if (p.ofs_unit > 1)
300 p.offset -= segno % p.ofs_unit;
301
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900302 secno = GET_SECNO(sbi, segno);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900303
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900304 if (sec_usage_check(sbi, secno))
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900305 continue;
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900306 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900307 continue;
308
309 cost = get_gc_cost(sbi, segno, &p);
310
311 if (p.min_cost > cost) {
312 p.min_segno = segno;
313 p.min_cost = cost;
Jin Xua57e5642013-09-13 08:38:54 +0800314 } else if (unlikely(cost == max_cost)) {
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900315 continue;
Jin Xua57e5642013-09-13 08:38:54 +0800316 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900317
Jin Xua26b7c82013-09-05 12:45:26 +0800318 if (nsearched++ >= p.max_search) {
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900319 sbi->last_victim[p.gc_mode] = segno;
320 break;
321 }
322 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900323 if (p.min_segno != NULL_SEGNO) {
Namjae Jeonb2b34602013-06-01 16:20:26 +0900324got_it:
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900325 if (p.alloc_mode == LFS) {
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900326 secno = GET_SECNO(sbi, p.min_segno);
327 if (gc_type == FG_GC)
328 sbi->cur_victim_sec = secno;
329 else
330 set_bit(secno, dirty_i->victim_secmap);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900331 }
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900332 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
Namjae Jeon8e46b3e2013-04-23 16:42:53 +0900333
334 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
335 sbi->cur_victim_sec,
336 prefree_segments(sbi), free_segments(sbi));
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900337 }
338 mutex_unlock(&dirty_i->seglist_lock);
339
340 return (p.min_segno == NULL_SEGNO) ? 0 : 1;
341}
342
343static const struct victim_selection default_v_ops = {
344 .get_victim = get_victim_by_default,
345};
346
347static struct inode *find_gc_inode(nid_t ino, struct list_head *ilist)
348{
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900349 struct inode_entry *ie;
350
Gu Zheng6cc4af52013-06-20 17:52:39 +0800351 list_for_each_entry(ie, ilist, list)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900352 if (ie->inode->i_ino == ino)
353 return ie->inode;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900354 return NULL;
355}
356
357static void add_gc_inode(struct inode *inode, struct list_head *ilist)
358{
Gu Zheng6cc4af52013-06-20 17:52:39 +0800359 struct inode_entry *new_ie;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900360
Gu Zheng6cc4af52013-06-20 17:52:39 +0800361 if (inode == find_gc_inode(inode->i_ino, ilist)) {
362 iput(inode);
363 return;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900364 }
Gu Zheng7bd59382013-10-22 14:52:26 +0800365
366 new_ie = f2fs_kmem_cache_alloc(winode_slab, GFP_NOFS);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900367 new_ie->inode = inode;
368 list_add_tail(&new_ie->list, ilist);
369}
370
371static void put_gc_inode(struct list_head *ilist)
372{
373 struct inode_entry *ie, *next_ie;
374 list_for_each_entry_safe(ie, next_ie, ilist, list) {
375 iput(ie->inode);
376 list_del(&ie->list);
377 kmem_cache_free(winode_slab, ie);
378 }
379}
380
381static int check_valid_map(struct f2fs_sb_info *sbi,
382 unsigned int segno, int offset)
383{
384 struct sit_info *sit_i = SIT_I(sbi);
385 struct seg_entry *sentry;
386 int ret;
387
388 mutex_lock(&sit_i->sentry_lock);
389 sentry = get_seg_entry(sbi, segno);
390 ret = f2fs_test_bit(offset, sentry->cur_valid_map);
391 mutex_unlock(&sit_i->sentry_lock);
Jaegeuk Kim43727522013-02-04 15:11:17 +0900392 return ret;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900393}
394
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900395/*
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900396 * This function compares node address got in summary with that in NAT.
397 * On validity, copy that node with cold status, otherwise (invalid node)
398 * ignore that.
399 */
Jaegeuk Kim43727522013-02-04 15:11:17 +0900400static void gc_node_segment(struct f2fs_sb_info *sbi,
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900401 struct f2fs_summary *sum, unsigned int segno, int gc_type)
402{
403 bool initial = true;
404 struct f2fs_summary *entry;
405 int off;
406
407next_step:
408 entry = sum;
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900409
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900410 for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
411 nid_t nid = le32_to_cpu(entry->nid);
412 struct page *node_page;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900413
Jaegeuk Kim43727522013-02-04 15:11:17 +0900414 /* stop BG_GC if there is not enough free sections. */
415 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
416 return;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900417
Jaegeuk Kim43727522013-02-04 15:11:17 +0900418 if (check_valid_map(sbi, segno, off) == 0)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900419 continue;
420
421 if (initial) {
422 ra_node_page(sbi, nid);
423 continue;
424 }
425 node_page = get_node_page(sbi, nid);
426 if (IS_ERR(node_page))
427 continue;
428
429 /* 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 Kim7bc09002012-11-02 17:13:01 +0900537 inode_dec_dirty_dents(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,
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900554 struct list_head *ilist, unsigned int segno, int gc_type)
555{
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);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900599 if (IS_ERR(inode))
600 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);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900606 if (IS_ERR(data_page))
607 goto next_iput;
608
609 f2fs_put_page(data_page, 0);
610 add_gc_inode(inode, ilist);
611 } else {
612 inode = find_gc_inode(dni.ino, ilist);
613 if (inode) {
Jaegeuk Kimde936532013-08-12 21:08:03 +0900614 start_bidx = start_bidx_of_node(nofs,
615 F2FS_I(inode));
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900616 data_page = get_lock_data_page(inode,
617 start_bidx + ofs_in_node);
618 if (IS_ERR(data_page))
619 continue;
620 move_data_page(inode, data_page, gc_type);
621 stat_inc_data_blk_count(sbi, 1);
622 }
623 }
624 continue;
625next_iput:
626 iput(inode);
627 }
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,
647 int gc_type, int type)
648{
649 struct sit_info *sit_i = SIT_I(sbi);
650 int ret;
651 mutex_lock(&sit_i->sentry_lock);
652 ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, type, LFS);
653 mutex_unlock(&sit_i->sentry_lock);
654 return ret;
655}
656
Jaegeuk Kim43727522013-02-04 15:11:17 +0900657static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno,
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900658 struct list_head *ilist, int gc_type)
659{
660 struct page *sum_page;
661 struct f2fs_summary_block *sum;
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900662 struct blk_plug plug;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900663
664 /* read segment summary of victim */
665 sum_page = get_sum_page(sbi, segno);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900666
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900667 blk_start_plug(&plug);
668
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900669 sum = page_address(sum_page);
670
671 switch (GET_SUM_TYPE((&sum->footer))) {
672 case SUM_TYPE_NODE:
Jaegeuk Kim43727522013-02-04 15:11:17 +0900673 gc_node_segment(sbi, sum->entries, segno, gc_type);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900674 break;
675 case SUM_TYPE_DATA:
Jaegeuk Kim43727522013-02-04 15:11:17 +0900676 gc_data_segment(sbi, sum->entries, ilist, segno, gc_type);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900677 break;
678 }
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900679 blk_finish_plug(&plug);
680
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900681 stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer)));
682 stat_inc_call_count(sbi->stat_info);
683
Jaegeuk Kimb7473752013-04-01 08:32:21 +0900684 f2fs_put_page(sum_page, 1);
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900685}
686
Jaegeuk Kim408e9372013-01-03 17:55:52 +0900687int f2fs_gc(struct f2fs_sb_info *sbi)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900688{
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900689 struct list_head ilist;
Jaegeuk Kim408e9372013-01-03 17:55:52 +0900690 unsigned int segno, i;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900691 int gc_type = BG_GC;
Jaegeuk Kim43727522013-02-04 15:11:17 +0900692 int nfree = 0;
693 int ret = -1;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900694
695 INIT_LIST_HEAD(&ilist);
696gc_more:
Jaegeuk Kim6bacf522013-12-06 15:00:58 +0900697 if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE)))
Jaegeuk Kim408e9372013-01-03 17:55:52 +0900698 goto stop;
Jaegeuk Kim203681f2014-02-05 13:03:57 +0900699 if (unlikely(is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)))
700 goto stop;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900701
Jaegeuk Kimd64f8042013-04-08 16:01:00 +0900702 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, nfree)) {
Jaegeuk Kim408e9372013-01-03 17:55:52 +0900703 gc_type = FG_GC;
Jaegeuk Kimd64f8042013-04-08 16:01:00 +0900704 write_checkpoint(sbi, false);
705 }
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900706
Jaegeuk Kim408e9372013-01-03 17:55:52 +0900707 if (!__get_victim(sbi, &segno, gc_type, NO_CHECK_TYPE))
708 goto stop;
Jaegeuk Kim43727522013-02-04 15:11:17 +0900709 ret = 0;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900710
Chao Yu81c1a0f12014-02-27 19:12:24 +0800711 /* readahead multi ssa blocks those have contiguous address */
712 if (sbi->segs_per_sec > 1)
713 ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), sbi->segs_per_sec,
714 META_SSA);
715
Jaegeuk Kim43727522013-02-04 15:11:17 +0900716 for (i = 0; i < sbi->segs_per_sec; i++)
717 do_garbage_collect(sbi, segno + i, &ilist, gc_type);
718
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900719 if (gc_type == FG_GC) {
720 sbi->cur_victim_sec = NULL_SEGNO;
Jaegeuk Kim43727522013-02-04 15:11:17 +0900721 nfree++;
Jaegeuk Kim5ec4e492013-03-31 13:26:03 +0900722 WARN_ON(get_valid_blocks(sbi, segno, sbi->segs_per_sec));
723 }
Jaegeuk Kim43727522013-02-04 15:11:17 +0900724
725 if (has_not_enough_free_secs(sbi, nfree))
726 goto gc_more;
727
728 if (gc_type == FG_GC)
729 write_checkpoint(sbi, false);
Jaegeuk Kim408e9372013-01-03 17:55:52 +0900730stop:
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900731 mutex_unlock(&sbi->gc_mutex);
732
733 put_gc_inode(&ilist);
Jaegeuk Kim43727522013-02-04 15:11:17 +0900734 return ret;
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900735}
736
737void build_gc_manager(struct f2fs_sb_info *sbi)
738{
739 DIRTY_I(sbi)->v_ops = &default_v_ops;
740}
741
Namjae Jeon6e6093a2013-01-17 00:08:30 +0900742int __init create_gc_caches(void)
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900743{
744 winode_slab = f2fs_kmem_cache_create("f2fs_gc_inodes",
Gu Zhenge8512d22014-03-07 18:43:28 +0800745 sizeof(struct inode_entry));
Jaegeuk Kim7bc09002012-11-02 17:13:01 +0900746 if (!winode_slab)
747 return -ENOMEM;
748 return 0;
749}
750
751void destroy_gc_caches(void)
752{
753 kmem_cache_destroy(winode_slab);
754}