Commit | Line | Data |
---|---|---|
0a8165d7 | 1 | /* |
7bc09003 JK |
2 | * fs/f2fs/gc.c |
3 | * | |
4 | * Copyright (c) 2012 Samsung Electronics Co., Ltd. | |
5 | * http://www.samsung.com/ | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License version 2 as | |
9 | * published by the Free Software Foundation. | |
10 | */ | |
11 | #include <linux/fs.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/backing-dev.h> | |
7bc09003 JK |
14 | #include <linux/init.h> |
15 | #include <linux/f2fs_fs.h> | |
16 | #include <linux/kthread.h> | |
17 | #include <linux/delay.h> | |
18 | #include <linux/freezer.h> | |
19 | #include <linux/blkdev.h> | |
20 | ||
21 | #include "f2fs.h" | |
22 | #include "node.h" | |
23 | #include "segment.h" | |
24 | #include "gc.h" | |
8e46b3ed | 25 | #include <trace/events/f2fs.h> |
7bc09003 JK |
26 | |
27 | static struct kmem_cache *winode_slab; | |
28 | ||
29 | static int gc_thread_func(void *data) | |
30 | { | |
31 | struct f2fs_sb_info *sbi = data; | |
b59d0bae | 32 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; |
7bc09003 JK |
33 | wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; |
34 | long wait_ms; | |
35 | ||
b59d0bae | 36 | wait_ms = gc_th->min_sleep_time; |
7bc09003 JK |
37 | |
38 | do { | |
39 | if (try_to_freeze()) | |
40 | continue; | |
41 | else | |
42 | wait_event_interruptible_timeout(*wq, | |
43 | kthread_should_stop(), | |
44 | msecs_to_jiffies(wait_ms)); | |
45 | if (kthread_should_stop()) | |
46 | break; | |
47 | ||
d6212a5f | 48 | if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { |
b59d0bae | 49 | wait_ms = increase_sleep_time(gc_th, wait_ms); |
d6212a5f CL |
50 | continue; |
51 | } | |
52 | ||
7bc09003 JK |
53 | /* |
54 | * [GC triggering condition] | |
55 | * 0. GC is not conducted currently. | |
56 | * 1. There are enough dirty segments. | |
57 | * 2. IO subsystem is idle by checking the # of writeback pages. | |
58 | * 3. IO subsystem is idle by checking the # of requests in | |
59 | * bdev's request list. | |
60 | * | |
e1c42045 | 61 | * Note) We have to avoid triggering GCs frequently. |
7bc09003 JK |
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)) { | |
b59d0bae | 70 | wait_ms = increase_sleep_time(gc_th, wait_ms); |
7bc09003 JK |
71 | mutex_unlock(&sbi->gc_mutex); |
72 | continue; | |
73 | } | |
74 | ||
75 | if (has_enough_invalid_blocks(sbi)) | |
b59d0bae | 76 | wait_ms = decrease_sleep_time(gc_th, wait_ms); |
7bc09003 | 77 | else |
b59d0bae | 78 | wait_ms = increase_sleep_time(gc_th, wait_ms); |
7bc09003 | 79 | |
dcdfff65 | 80 | stat_inc_bggc_count(sbi); |
7bc09003 | 81 | |
43727527 JK |
82 | /* if return value is not zero, no victim was selected */ |
83 | if (f2fs_gc(sbi)) | |
b59d0bae | 84 | wait_ms = gc_th->no_gc_sleep_time; |
81eb8d6e | 85 | |
4660f9c0 JK |
86 | /* balancing f2fs's metadata periodically */ |
87 | f2fs_balance_fs_bg(sbi); | |
81eb8d6e | 88 | |
7bc09003 JK |
89 | } while (!kthread_should_stop()); |
90 | return 0; | |
91 | } | |
92 | ||
93 | int start_gc_thread(struct f2fs_sb_info *sbi) | |
94 | { | |
1042d60f | 95 | struct f2fs_gc_kthread *gc_th; |
ec7b1f2d | 96 | dev_t dev = sbi->sb->s_bdev->bd_dev; |
7a267f8d | 97 | int err = 0; |
7bc09003 | 98 | |
48600e44 | 99 | if (!test_opt(sbi, BG_GC)) |
7a267f8d | 100 | goto out; |
7bc09003 | 101 | gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL); |
7a267f8d NJ |
102 | if (!gc_th) { |
103 | err = -ENOMEM; | |
104 | goto out; | |
105 | } | |
7bc09003 | 106 | |
b59d0bae NJ |
107 | gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; |
108 | gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; | |
109 | gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; | |
110 | ||
d2dc095f NJ |
111 | gc_th->gc_idle = 0; |
112 | ||
7bc09003 JK |
113 | sbi->gc_thread = gc_th; |
114 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
115 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, | |
ec7b1f2d | 116 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 | 117 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
7a267f8d | 118 | err = PTR_ERR(gc_th->f2fs_gc_task); |
7bc09003 | 119 | kfree(gc_th); |
25718423 | 120 | sbi->gc_thread = NULL; |
7bc09003 | 121 | } |
7a267f8d NJ |
122 | out: |
123 | return err; | |
7bc09003 JK |
124 | } |
125 | ||
126 | void stop_gc_thread(struct f2fs_sb_info *sbi) | |
127 | { | |
128 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
129 | if (!gc_th) | |
130 | return; | |
131 | kthread_stop(gc_th->f2fs_gc_task); | |
132 | kfree(gc_th); | |
133 | sbi->gc_thread = NULL; | |
134 | } | |
135 | ||
d2dc095f | 136 | static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type) |
7bc09003 | 137 | { |
d2dc095f NJ |
138 | int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY; |
139 | ||
140 | if (gc_th && gc_th->gc_idle) { | |
141 | if (gc_th->gc_idle == 1) | |
142 | gc_mode = GC_CB; | |
143 | else if (gc_th->gc_idle == 2) | |
144 | gc_mode = GC_GREEDY; | |
145 | } | |
146 | return gc_mode; | |
7bc09003 JK |
147 | } |
148 | ||
149 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
150 | int type, struct victim_sel_policy *p) | |
151 | { | |
152 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
153 | ||
4ebefc44 | 154 | if (p->alloc_mode == SSR) { |
7bc09003 JK |
155 | p->gc_mode = GC_GREEDY; |
156 | p->dirty_segmap = dirty_i->dirty_segmap[type]; | |
a26b7c8a | 157 | p->max_search = dirty_i->nr_dirty[type]; |
7bc09003 JK |
158 | p->ofs_unit = 1; |
159 | } else { | |
d2dc095f | 160 | p->gc_mode = select_gc_type(sbi->gc_thread, gc_type); |
7bc09003 | 161 | p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; |
a26b7c8a | 162 | p->max_search = dirty_i->nr_dirty[DIRTY]; |
7bc09003 JK |
163 | p->ofs_unit = sbi->segs_per_sec; |
164 | } | |
a26b7c8a | 165 | |
b1c57c1c JK |
166 | if (p->max_search > sbi->max_victim_search) |
167 | p->max_search = sbi->max_victim_search; | |
a26b7c8a | 168 | |
7bc09003 JK |
169 | p->offset = sbi->last_victim[p->gc_mode]; |
170 | } | |
171 | ||
172 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
173 | struct victim_sel_policy *p) | |
174 | { | |
b7250d2d JK |
175 | /* SSR allocates in a segment unit */ |
176 | if (p->alloc_mode == SSR) | |
177 | return 1 << sbi->log_blocks_per_seg; | |
7bc09003 JK |
178 | if (p->gc_mode == GC_GREEDY) |
179 | return (1 << sbi->log_blocks_per_seg) * p->ofs_unit; | |
180 | else if (p->gc_mode == GC_CB) | |
181 | return UINT_MAX; | |
182 | else /* No other gc_mode */ | |
183 | return 0; | |
184 | } | |
185 | ||
186 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
187 | { | |
188 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 189 | unsigned int secno; |
7bc09003 JK |
190 | |
191 | /* | |
192 | * If the gc_type is FG_GC, we can select victim segments | |
193 | * selected by background GC before. | |
194 | * Those segments guarantee they have small valid blocks. | |
195 | */ | |
7cd8558b | 196 | for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { |
5ec4e49f | 197 | if (sec_usage_check(sbi, secno)) |
b65ee148 | 198 | continue; |
5ec4e49f JK |
199 | clear_bit(secno, dirty_i->victim_secmap); |
200 | return secno * sbi->segs_per_sec; | |
7bc09003 JK |
201 | } |
202 | return NULL_SEGNO; | |
203 | } | |
204 | ||
205 | static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) | |
206 | { | |
207 | struct sit_info *sit_i = SIT_I(sbi); | |
208 | unsigned int secno = GET_SECNO(sbi, segno); | |
209 | unsigned int start = secno * sbi->segs_per_sec; | |
210 | unsigned long long mtime = 0; | |
211 | unsigned int vblocks; | |
212 | unsigned char age = 0; | |
213 | unsigned char u; | |
214 | unsigned int i; | |
215 | ||
216 | for (i = 0; i < sbi->segs_per_sec; i++) | |
217 | mtime += get_seg_entry(sbi, start + i)->mtime; | |
218 | vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
219 | ||
220 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
221 | vblocks = div_u64(vblocks, sbi->segs_per_sec); | |
222 | ||
223 | u = (vblocks * 100) >> sbi->log_blocks_per_seg; | |
224 | ||
e1c42045 | 225 | /* Handle if the system time has changed by the user */ |
7bc09003 JK |
226 | if (mtime < sit_i->min_mtime) |
227 | sit_i->min_mtime = mtime; | |
228 | if (mtime > sit_i->max_mtime) | |
229 | sit_i->max_mtime = mtime; | |
230 | if (sit_i->max_mtime != sit_i->min_mtime) | |
231 | age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), | |
232 | sit_i->max_mtime - sit_i->min_mtime); | |
233 | ||
234 | return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); | |
235 | } | |
236 | ||
a57e564d JX |
237 | static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, |
238 | unsigned int segno, struct victim_sel_policy *p) | |
7bc09003 JK |
239 | { |
240 | if (p->alloc_mode == SSR) | |
241 | return get_seg_entry(sbi, segno)->ckpt_valid_blocks; | |
242 | ||
243 | /* alloc_mode == LFS */ | |
244 | if (p->gc_mode == GC_GREEDY) | |
245 | return get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
246 | else | |
247 | return get_cb_cost(sbi, segno); | |
248 | } | |
249 | ||
0a8165d7 | 250 | /* |
111d2495 | 251 | * This function is called from two paths. |
7bc09003 JK |
252 | * One is garbage collection and the other is SSR segment selection. |
253 | * When it is called during GC, it just gets a victim segment | |
254 | * and it does not remove it from dirty seglist. | |
255 | * When it is called from SSR segment selection, it finds a segment | |
256 | * which has minimum valid blocks and removes it from dirty seglist. | |
257 | */ | |
258 | static int get_victim_by_default(struct f2fs_sb_info *sbi, | |
259 | unsigned int *result, int gc_type, int type, char alloc_mode) | |
260 | { | |
261 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
262 | struct victim_sel_policy p; | |
b2b3460a | 263 | unsigned int secno, max_cost; |
7bc09003 JK |
264 | int nsearched = 0; |
265 | ||
210f41bc CY |
266 | mutex_lock(&dirty_i->seglist_lock); |
267 | ||
7bc09003 JK |
268 | p.alloc_mode = alloc_mode; |
269 | select_policy(sbi, gc_type, type, &p); | |
270 | ||
271 | p.min_segno = NULL_SEGNO; | |
b2b3460a | 272 | p.min_cost = max_cost = get_max_cost(sbi, &p); |
7bc09003 | 273 | |
7bc09003 JK |
274 | if (p.alloc_mode == LFS && gc_type == FG_GC) { |
275 | p.min_segno = check_bg_victims(sbi); | |
276 | if (p.min_segno != NULL_SEGNO) | |
277 | goto got_it; | |
278 | } | |
279 | ||
280 | while (1) { | |
281 | unsigned long cost; | |
5ec4e49f | 282 | unsigned int segno; |
7bc09003 | 283 | |
7cd8558b JK |
284 | segno = find_next_bit(p.dirty_segmap, MAIN_SEGS(sbi), p.offset); |
285 | if (segno >= MAIN_SEGS(sbi)) { | |
7bc09003 JK |
286 | if (sbi->last_victim[p.gc_mode]) { |
287 | sbi->last_victim[p.gc_mode] = 0; | |
288 | p.offset = 0; | |
289 | continue; | |
290 | } | |
291 | break; | |
292 | } | |
a57e564d JX |
293 | |
294 | p.offset = segno + p.ofs_unit; | |
295 | if (p.ofs_unit > 1) | |
296 | p.offset -= segno % p.ofs_unit; | |
297 | ||
5ec4e49f | 298 | secno = GET_SECNO(sbi, segno); |
7bc09003 | 299 | |
5ec4e49f | 300 | if (sec_usage_check(sbi, secno)) |
7bc09003 | 301 | continue; |
5ec4e49f | 302 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
7bc09003 JK |
303 | continue; |
304 | ||
305 | cost = get_gc_cost(sbi, segno, &p); | |
306 | ||
307 | if (p.min_cost > cost) { | |
308 | p.min_segno = segno; | |
309 | p.min_cost = cost; | |
a57e564d | 310 | } else if (unlikely(cost == max_cost)) { |
7bc09003 | 311 | continue; |
a57e564d | 312 | } |
7bc09003 | 313 | |
a26b7c8a | 314 | if (nsearched++ >= p.max_search) { |
7bc09003 JK |
315 | sbi->last_victim[p.gc_mode] = segno; |
316 | break; | |
317 | } | |
318 | } | |
7bc09003 | 319 | if (p.min_segno != NULL_SEGNO) { |
b2b3460a | 320 | got_it: |
7bc09003 | 321 | if (p.alloc_mode == LFS) { |
5ec4e49f JK |
322 | secno = GET_SECNO(sbi, p.min_segno); |
323 | if (gc_type == FG_GC) | |
324 | sbi->cur_victim_sec = secno; | |
325 | else | |
326 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 327 | } |
5ec4e49f | 328 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
8e46b3ed NJ |
329 | |
330 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, | |
331 | sbi->cur_victim_sec, | |
332 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 JK |
333 | } |
334 | mutex_unlock(&dirty_i->seglist_lock); | |
335 | ||
336 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
337 | } | |
338 | ||
339 | static const struct victim_selection default_v_ops = { | |
340 | .get_victim = get_victim_by_default, | |
341 | }; | |
342 | ||
343 | static struct inode *find_gc_inode(nid_t ino, struct list_head *ilist) | |
344 | { | |
7bc09003 JK |
345 | struct inode_entry *ie; |
346 | ||
6cc4af56 | 347 | list_for_each_entry(ie, ilist, list) |
7bc09003 JK |
348 | if (ie->inode->i_ino == ino) |
349 | return ie->inode; | |
7bc09003 JK |
350 | return NULL; |
351 | } | |
352 | ||
353 | static void add_gc_inode(struct inode *inode, struct list_head *ilist) | |
354 | { | |
6cc4af56 GZ |
355 | struct inode_entry *new_ie; |
356 | ||
357 | if (inode == find_gc_inode(inode->i_ino, ilist)) { | |
358 | iput(inode); | |
359 | return; | |
7bc09003 | 360 | } |
7bd59381 GZ |
361 | |
362 | new_ie = f2fs_kmem_cache_alloc(winode_slab, GFP_NOFS); | |
7bc09003 JK |
363 | new_ie->inode = inode; |
364 | list_add_tail(&new_ie->list, ilist); | |
365 | } | |
366 | ||
367 | static void put_gc_inode(struct list_head *ilist) | |
368 | { | |
369 | struct inode_entry *ie, *next_ie; | |
370 | list_for_each_entry_safe(ie, next_ie, ilist, list) { | |
371 | iput(ie->inode); | |
372 | list_del(&ie->list); | |
373 | kmem_cache_free(winode_slab, ie); | |
374 | } | |
375 | } | |
376 | ||
377 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
378 | unsigned int segno, int offset) | |
379 | { | |
380 | struct sit_info *sit_i = SIT_I(sbi); | |
381 | struct seg_entry *sentry; | |
382 | int ret; | |
383 | ||
384 | mutex_lock(&sit_i->sentry_lock); | |
385 | sentry = get_seg_entry(sbi, segno); | |
386 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
387 | mutex_unlock(&sit_i->sentry_lock); | |
43727527 | 388 | return ret; |
7bc09003 JK |
389 | } |
390 | ||
0a8165d7 | 391 | /* |
7bc09003 JK |
392 | * This function compares node address got in summary with that in NAT. |
393 | * On validity, copy that node with cold status, otherwise (invalid node) | |
394 | * ignore that. | |
395 | */ | |
43727527 | 396 | static void gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
397 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
398 | { | |
399 | bool initial = true; | |
400 | struct f2fs_summary *entry; | |
401 | int off; | |
402 | ||
403 | next_step: | |
404 | entry = sum; | |
c718379b | 405 | |
7bc09003 JK |
406 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
407 | nid_t nid = le32_to_cpu(entry->nid); | |
408 | struct page *node_page; | |
7bc09003 | 409 | |
43727527 JK |
410 | /* stop BG_GC if there is not enough free sections. */ |
411 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
412 | return; | |
7bc09003 | 413 | |
43727527 | 414 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
415 | continue; |
416 | ||
417 | if (initial) { | |
418 | ra_node_page(sbi, nid); | |
419 | continue; | |
420 | } | |
421 | node_page = get_node_page(sbi, nid); | |
422 | if (IS_ERR(node_page)) | |
423 | continue; | |
424 | ||
9a01b56b HY |
425 | /* block may become invalid during get_node_page */ |
426 | if (check_valid_map(sbi, segno, off) == 0) { | |
427 | f2fs_put_page(node_page, 1); | |
428 | continue; | |
429 | } | |
430 | ||
7bc09003 | 431 | /* set page dirty and write it */ |
4ebefc44 | 432 | if (gc_type == FG_GC) { |
5514f0aa | 433 | f2fs_wait_on_page_writeback(node_page, NODE); |
7bc09003 | 434 | set_page_dirty(node_page); |
4ebefc44 JK |
435 | } else { |
436 | if (!PageWriteback(node_page)) | |
437 | set_page_dirty(node_page); | |
438 | } | |
7bc09003 JK |
439 | f2fs_put_page(node_page, 1); |
440 | stat_inc_node_blk_count(sbi, 1); | |
441 | } | |
c718379b | 442 | |
7bc09003 JK |
443 | if (initial) { |
444 | initial = false; | |
445 | goto next_step; | |
446 | } | |
447 | ||
448 | if (gc_type == FG_GC) { | |
449 | struct writeback_control wbc = { | |
450 | .sync_mode = WB_SYNC_ALL, | |
451 | .nr_to_write = LONG_MAX, | |
452 | .for_reclaim = 0, | |
453 | }; | |
454 | sync_node_pages(sbi, 0, &wbc); | |
4ebefc44 JK |
455 | |
456 | /* | |
457 | * In the case of FG_GC, it'd be better to reclaim this victim | |
458 | * completely. | |
459 | */ | |
460 | if (get_valid_blocks(sbi, segno, 1) != 0) | |
461 | goto next_step; | |
7bc09003 | 462 | } |
7bc09003 JK |
463 | } |
464 | ||
0a8165d7 | 465 | /* |
9af45ef5 JK |
466 | * Calculate start block index indicating the given node offset. |
467 | * Be careful, caller should give this node offset only indicating direct node | |
468 | * blocks. If any node offsets, which point the other types of node blocks such | |
469 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
470 | * bug. | |
7bc09003 | 471 | */ |
de93653f | 472 | block_t start_bidx_of_node(unsigned int node_ofs, struct f2fs_inode_info *fi) |
7bc09003 | 473 | { |
ce19a5d4 JK |
474 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
475 | unsigned int bidx; | |
7bc09003 | 476 | |
ce19a5d4 JK |
477 | if (node_ofs == 0) |
478 | return 0; | |
7bc09003 | 479 | |
ce19a5d4 | 480 | if (node_ofs <= 2) { |
7bc09003 JK |
481 | bidx = node_ofs - 1; |
482 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 483 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
484 | bidx = node_ofs - 2 - dec; |
485 | } else { | |
ce19a5d4 | 486 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
487 | bidx = node_ofs - 5 - dec; |
488 | } | |
de93653f | 489 | return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(fi); |
7bc09003 JK |
490 | } |
491 | ||
492 | static int check_dnode(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, | |
493 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) | |
494 | { | |
495 | struct page *node_page; | |
496 | nid_t nid; | |
497 | unsigned int ofs_in_node; | |
498 | block_t source_blkaddr; | |
499 | ||
500 | nid = le32_to_cpu(sum->nid); | |
501 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
502 | ||
503 | node_page = get_node_page(sbi, nid); | |
504 | if (IS_ERR(node_page)) | |
43727527 | 505 | return 0; |
7bc09003 JK |
506 | |
507 | get_node_info(sbi, nid, dni); | |
508 | ||
509 | if (sum->version != dni->version) { | |
510 | f2fs_put_page(node_page, 1); | |
43727527 | 511 | return 0; |
7bc09003 JK |
512 | } |
513 | ||
514 | *nofs = ofs_of_node(node_page); | |
515 | source_blkaddr = datablock_addr(node_page, ofs_in_node); | |
516 | f2fs_put_page(node_page, 1); | |
517 | ||
518 | if (source_blkaddr != blkaddr) | |
43727527 JK |
519 | return 0; |
520 | return 1; | |
7bc09003 JK |
521 | } |
522 | ||
523 | static void move_data_page(struct inode *inode, struct page *page, int gc_type) | |
524 | { | |
458e6197 JK |
525 | struct f2fs_io_info fio = { |
526 | .type = DATA, | |
527 | .rw = WRITE_SYNC, | |
63a0b7cb FL |
528 | }; |
529 | ||
7bc09003 | 530 | if (gc_type == BG_GC) { |
4ebefc44 JK |
531 | if (PageWriteback(page)) |
532 | goto out; | |
7bc09003 JK |
533 | set_page_dirty(page); |
534 | set_cold_data(page); | |
535 | } else { | |
5514f0aa | 536 | f2fs_wait_on_page_writeback(page, DATA); |
4ebefc44 | 537 | |
1fe54f9d | 538 | if (clear_page_dirty_for_io(page)) |
a7ffdbe2 | 539 | inode_dec_dirty_pages(inode); |
7bc09003 | 540 | set_cold_data(page); |
458e6197 | 541 | do_write_data_page(page, &fio); |
7bc09003 JK |
542 | clear_cold_data(page); |
543 | } | |
544 | out: | |
545 | f2fs_put_page(page, 1); | |
546 | } | |
547 | ||
0a8165d7 | 548 | /* |
7bc09003 JK |
549 | * This function tries to get parent node of victim data block, and identifies |
550 | * data block validity. If the block is valid, copy that with cold status and | |
551 | * modify parent node. | |
552 | * If the parent node is not valid or the data block address is different, | |
553 | * the victim data block is ignored. | |
554 | */ | |
43727527 | 555 | static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
556 | struct list_head *ilist, unsigned int segno, int gc_type) |
557 | { | |
558 | struct super_block *sb = sbi->sb; | |
559 | struct f2fs_summary *entry; | |
560 | block_t start_addr; | |
43727527 | 561 | int off; |
7bc09003 JK |
562 | int phase = 0; |
563 | ||
564 | start_addr = START_BLOCK(sbi, segno); | |
565 | ||
566 | next_step: | |
567 | entry = sum; | |
c718379b | 568 | |
7bc09003 JK |
569 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
570 | struct page *data_page; | |
571 | struct inode *inode; | |
572 | struct node_info dni; /* dnode info for the data */ | |
573 | unsigned int ofs_in_node, nofs; | |
574 | block_t start_bidx; | |
575 | ||
43727527 JK |
576 | /* stop BG_GC if there is not enough free sections. */ |
577 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
578 | return; | |
7bc09003 | 579 | |
43727527 | 580 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
581 | continue; |
582 | ||
583 | if (phase == 0) { | |
584 | ra_node_page(sbi, le32_to_cpu(entry->nid)); | |
585 | continue; | |
586 | } | |
587 | ||
588 | /* Get an inode by ino with checking validity */ | |
43727527 | 589 | if (check_dnode(sbi, entry, &dni, start_addr + off, &nofs) == 0) |
7bc09003 JK |
590 | continue; |
591 | ||
592 | if (phase == 1) { | |
593 | ra_node_page(sbi, dni.ino); | |
594 | continue; | |
595 | } | |
596 | ||
7bc09003 JK |
597 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); |
598 | ||
599 | if (phase == 2) { | |
d4686d56 | 600 | inode = f2fs_iget(sb, dni.ino); |
b73e5282 | 601 | if (IS_ERR(inode) || is_bad_inode(inode)) |
7bc09003 JK |
602 | continue; |
603 | ||
de93653f JK |
604 | start_bidx = start_bidx_of_node(nofs, F2FS_I(inode)); |
605 | ||
7bc09003 | 606 | data_page = find_data_page(inode, |
c718379b | 607 | start_bidx + ofs_in_node, false); |
7bc09003 JK |
608 | if (IS_ERR(data_page)) |
609 | goto next_iput; | |
610 | ||
611 | f2fs_put_page(data_page, 0); | |
612 | add_gc_inode(inode, ilist); | |
613 | } else { | |
614 | inode = find_gc_inode(dni.ino, ilist); | |
615 | if (inode) { | |
de93653f JK |
616 | start_bidx = start_bidx_of_node(nofs, |
617 | F2FS_I(inode)); | |
7bc09003 JK |
618 | data_page = get_lock_data_page(inode, |
619 | start_bidx + ofs_in_node); | |
620 | if (IS_ERR(data_page)) | |
621 | continue; | |
622 | move_data_page(inode, data_page, gc_type); | |
623 | stat_inc_data_blk_count(sbi, 1); | |
624 | } | |
625 | } | |
626 | continue; | |
627 | next_iput: | |
628 | iput(inode); | |
629 | } | |
c718379b | 630 | |
7bc09003 JK |
631 | if (++phase < 4) |
632 | goto next_step; | |
43727527 | 633 | |
4ebefc44 | 634 | if (gc_type == FG_GC) { |
458e6197 | 635 | f2fs_submit_merged_bio(sbi, DATA, WRITE); |
4ebefc44 JK |
636 | |
637 | /* | |
638 | * In the case of FG_GC, it'd be better to reclaim this victim | |
639 | * completely. | |
640 | */ | |
641 | if (get_valid_blocks(sbi, segno, 1) != 0) { | |
642 | phase = 2; | |
643 | goto next_step; | |
644 | } | |
645 | } | |
7bc09003 JK |
646 | } |
647 | ||
648 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
649 | int gc_type, int type) | |
650 | { | |
651 | struct sit_info *sit_i = SIT_I(sbi); | |
652 | int ret; | |
653 | mutex_lock(&sit_i->sentry_lock); | |
654 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, type, LFS); | |
655 | mutex_unlock(&sit_i->sentry_lock); | |
656 | return ret; | |
657 | } | |
658 | ||
43727527 | 659 | static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno, |
7bc09003 JK |
660 | struct list_head *ilist, int gc_type) |
661 | { | |
662 | struct page *sum_page; | |
663 | struct f2fs_summary_block *sum; | |
c718379b | 664 | struct blk_plug plug; |
7bc09003 JK |
665 | |
666 | /* read segment summary of victim */ | |
667 | sum_page = get_sum_page(sbi, segno); | |
7bc09003 | 668 | |
c718379b JK |
669 | blk_start_plug(&plug); |
670 | ||
7bc09003 JK |
671 | sum = page_address(sum_page); |
672 | ||
673 | switch (GET_SUM_TYPE((&sum->footer))) { | |
674 | case SUM_TYPE_NODE: | |
43727527 | 675 | gc_node_segment(sbi, sum->entries, segno, gc_type); |
7bc09003 JK |
676 | break; |
677 | case SUM_TYPE_DATA: | |
43727527 | 678 | gc_data_segment(sbi, sum->entries, ilist, segno, gc_type); |
7bc09003 JK |
679 | break; |
680 | } | |
c718379b JK |
681 | blk_finish_plug(&plug); |
682 | ||
7bc09003 JK |
683 | stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer))); |
684 | stat_inc_call_count(sbi->stat_info); | |
685 | ||
b7473754 | 686 | f2fs_put_page(sum_page, 1); |
7bc09003 JK |
687 | } |
688 | ||
408e9375 | 689 | int f2fs_gc(struct f2fs_sb_info *sbi) |
7bc09003 | 690 | { |
7bc09003 | 691 | struct list_head ilist; |
408e9375 | 692 | unsigned int segno, i; |
7bc09003 | 693 | int gc_type = BG_GC; |
43727527 JK |
694 | int nfree = 0; |
695 | int ret = -1; | |
75ab4cb8 JK |
696 | struct cp_control cpc = { |
697 | .reason = CP_SYNC, | |
698 | }; | |
7bc09003 JK |
699 | |
700 | INIT_LIST_HEAD(&ilist); | |
701 | gc_more: | |
6bacf52f | 702 | if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE))) |
408e9375 | 703 | goto stop; |
1e968fdf | 704 | if (unlikely(f2fs_cp_error(sbi))) |
203681f6 | 705 | goto stop; |
7bc09003 | 706 | |
d64f8047 | 707 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, nfree)) { |
408e9375 | 708 | gc_type = FG_GC; |
75ab4cb8 | 709 | write_checkpoint(sbi, &cpc); |
d64f8047 | 710 | } |
7bc09003 | 711 | |
408e9375 JK |
712 | if (!__get_victim(sbi, &segno, gc_type, NO_CHECK_TYPE)) |
713 | goto stop; | |
43727527 | 714 | ret = 0; |
7bc09003 | 715 | |
81c1a0f1 CY |
716 | /* readahead multi ssa blocks those have contiguous address */ |
717 | if (sbi->segs_per_sec > 1) | |
718 | ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), sbi->segs_per_sec, | |
719 | META_SSA); | |
720 | ||
43727527 JK |
721 | for (i = 0; i < sbi->segs_per_sec; i++) |
722 | do_garbage_collect(sbi, segno + i, &ilist, gc_type); | |
723 | ||
5ec4e49f JK |
724 | if (gc_type == FG_GC) { |
725 | sbi->cur_victim_sec = NULL_SEGNO; | |
43727527 | 726 | nfree++; |
5ec4e49f JK |
727 | WARN_ON(get_valid_blocks(sbi, segno, sbi->segs_per_sec)); |
728 | } | |
43727527 JK |
729 | |
730 | if (has_not_enough_free_secs(sbi, nfree)) | |
731 | goto gc_more; | |
732 | ||
733 | if (gc_type == FG_GC) | |
75ab4cb8 | 734 | write_checkpoint(sbi, &cpc); |
408e9375 | 735 | stop: |
7bc09003 JK |
736 | mutex_unlock(&sbi->gc_mutex); |
737 | ||
738 | put_gc_inode(&ilist); | |
43727527 | 739 | return ret; |
7bc09003 JK |
740 | } |
741 | ||
742 | void build_gc_manager(struct f2fs_sb_info *sbi) | |
743 | { | |
744 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
745 | } | |
746 | ||
6e6093a8 | 747 | int __init create_gc_caches(void) |
7bc09003 JK |
748 | { |
749 | winode_slab = f2fs_kmem_cache_create("f2fs_gc_inodes", | |
e8512d2e | 750 | sizeof(struct inode_entry)); |
7bc09003 JK |
751 | if (!winode_slab) |
752 | return -ENOMEM; | |
753 | return 0; | |
754 | } | |
755 | ||
756 | void destroy_gc_caches(void) | |
757 | { | |
758 | kmem_cache_destroy(winode_slab); | |
759 | } |