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 | * | |
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)) { | |
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 | |
35b09d82 | 80 | #ifdef CONFIG_F2FS_STAT_FS |
7bc09003 | 81 | sbi->bg_gc++; |
35b09d82 | 82 | #endif |
7bc09003 | 83 | |
43727527 JK |
84 | /* if return value is not zero, no victim was selected */ |
85 | if (f2fs_gc(sbi)) | |
b59d0bae | 86 | wait_ms = gc_th->no_gc_sleep_time; |
7bc09003 JK |
87 | } while (!kthread_should_stop()); |
88 | return 0; | |
89 | } | |
90 | ||
91 | int start_gc_thread(struct f2fs_sb_info *sbi) | |
92 | { | |
1042d60f | 93 | struct f2fs_gc_kthread *gc_th; |
ec7b1f2d | 94 | dev_t dev = sbi->sb->s_bdev->bd_dev; |
7a267f8d | 95 | int err = 0; |
7bc09003 | 96 | |
48600e44 | 97 | if (!test_opt(sbi, BG_GC)) |
7a267f8d | 98 | goto out; |
7bc09003 | 99 | gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL); |
7a267f8d NJ |
100 | if (!gc_th) { |
101 | err = -ENOMEM; | |
102 | goto out; | |
103 | } | |
7bc09003 | 104 | |
b59d0bae NJ |
105 | gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; |
106 | gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; | |
107 | gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; | |
108 | ||
d2dc095f NJ |
109 | gc_th->gc_idle = 0; |
110 | ||
7bc09003 JK |
111 | sbi->gc_thread = gc_th; |
112 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
113 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, | |
ec7b1f2d | 114 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 | 115 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
7a267f8d | 116 | err = PTR_ERR(gc_th->f2fs_gc_task); |
7bc09003 | 117 | kfree(gc_th); |
25718423 | 118 | sbi->gc_thread = NULL; |
7bc09003 | 119 | } |
7a267f8d NJ |
120 | |
121 | out: | |
122 | return err; | |
7bc09003 JK |
123 | } |
124 | ||
125 | void stop_gc_thread(struct f2fs_sb_info *sbi) | |
126 | { | |
127 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
128 | if (!gc_th) | |
129 | return; | |
130 | kthread_stop(gc_th->f2fs_gc_task); | |
131 | kfree(gc_th); | |
132 | sbi->gc_thread = NULL; | |
133 | } | |
134 | ||
d2dc095f | 135 | static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type) |
7bc09003 | 136 | { |
d2dc095f NJ |
137 | int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY; |
138 | ||
139 | if (gc_th && gc_th->gc_idle) { | |
140 | if (gc_th->gc_idle == 1) | |
141 | gc_mode = GC_CB; | |
142 | else if (gc_th->gc_idle == 2) | |
143 | gc_mode = GC_GREEDY; | |
144 | } | |
145 | return gc_mode; | |
7bc09003 JK |
146 | } |
147 | ||
148 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
149 | int type, struct victim_sel_policy *p) | |
150 | { | |
151 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
152 | ||
4ebefc44 | 153 | if (p->alloc_mode == SSR) { |
7bc09003 JK |
154 | p->gc_mode = GC_GREEDY; |
155 | p->dirty_segmap = dirty_i->dirty_segmap[type]; | |
156 | p->ofs_unit = 1; | |
157 | } else { | |
d2dc095f | 158 | p->gc_mode = select_gc_type(sbi->gc_thread, gc_type); |
7bc09003 JK |
159 | p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; |
160 | p->ofs_unit = sbi->segs_per_sec; | |
161 | } | |
162 | p->offset = sbi->last_victim[p->gc_mode]; | |
163 | } | |
164 | ||
165 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
166 | struct victim_sel_policy *p) | |
167 | { | |
b7250d2d JK |
168 | /* SSR allocates in a segment unit */ |
169 | if (p->alloc_mode == SSR) | |
170 | return 1 << sbi->log_blocks_per_seg; | |
7bc09003 JK |
171 | if (p->gc_mode == GC_GREEDY) |
172 | return (1 << sbi->log_blocks_per_seg) * p->ofs_unit; | |
173 | else if (p->gc_mode == GC_CB) | |
174 | return UINT_MAX; | |
175 | else /* No other gc_mode */ | |
176 | return 0; | |
177 | } | |
178 | ||
179 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
180 | { | |
181 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f JK |
182 | unsigned int hint = 0; |
183 | unsigned int secno; | |
7bc09003 JK |
184 | |
185 | /* | |
186 | * If the gc_type is FG_GC, we can select victim segments | |
187 | * selected by background GC before. | |
188 | * Those segments guarantee they have small valid blocks. | |
189 | */ | |
5ec4e49f JK |
190 | next: |
191 | secno = find_next_bit(dirty_i->victim_secmap, TOTAL_SECS(sbi), hint++); | |
192 | if (secno < TOTAL_SECS(sbi)) { | |
193 | if (sec_usage_check(sbi, secno)) | |
194 | goto next; | |
195 | clear_bit(secno, dirty_i->victim_secmap); | |
196 | return secno * sbi->segs_per_sec; | |
7bc09003 JK |
197 | } |
198 | return NULL_SEGNO; | |
199 | } | |
200 | ||
201 | static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) | |
202 | { | |
203 | struct sit_info *sit_i = SIT_I(sbi); | |
204 | unsigned int secno = GET_SECNO(sbi, segno); | |
205 | unsigned int start = secno * sbi->segs_per_sec; | |
206 | unsigned long long mtime = 0; | |
207 | unsigned int vblocks; | |
208 | unsigned char age = 0; | |
209 | unsigned char u; | |
210 | unsigned int i; | |
211 | ||
212 | for (i = 0; i < sbi->segs_per_sec; i++) | |
213 | mtime += get_seg_entry(sbi, start + i)->mtime; | |
214 | vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
215 | ||
216 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
217 | vblocks = div_u64(vblocks, sbi->segs_per_sec); | |
218 | ||
219 | u = (vblocks * 100) >> sbi->log_blocks_per_seg; | |
220 | ||
221 | /* Handle if the system time is changed by user */ | |
222 | if (mtime < sit_i->min_mtime) | |
223 | sit_i->min_mtime = mtime; | |
224 | if (mtime > sit_i->max_mtime) | |
225 | sit_i->max_mtime = mtime; | |
226 | if (sit_i->max_mtime != sit_i->min_mtime) | |
227 | age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), | |
228 | sit_i->max_mtime - sit_i->min_mtime); | |
229 | ||
230 | return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); | |
231 | } | |
232 | ||
233 | static unsigned int get_gc_cost(struct f2fs_sb_info *sbi, unsigned int segno, | |
234 | struct victim_sel_policy *p) | |
235 | { | |
236 | if (p->alloc_mode == SSR) | |
237 | return get_seg_entry(sbi, segno)->ckpt_valid_blocks; | |
238 | ||
239 | /* alloc_mode == LFS */ | |
240 | if (p->gc_mode == GC_GREEDY) | |
241 | return get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
242 | else | |
243 | return get_cb_cost(sbi, segno); | |
244 | } | |
245 | ||
0a8165d7 | 246 | /* |
111d2495 | 247 | * This function is called from two paths. |
7bc09003 JK |
248 | * One is garbage collection and the other is SSR segment selection. |
249 | * When it is called during GC, it just gets a victim segment | |
250 | * and it does not remove it from dirty seglist. | |
251 | * When it is called from SSR segment selection, it finds a segment | |
252 | * which has minimum valid blocks and removes it from dirty seglist. | |
253 | */ | |
254 | static int get_victim_by_default(struct f2fs_sb_info *sbi, | |
255 | unsigned int *result, int gc_type, int type, char alloc_mode) | |
256 | { | |
257 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
258 | struct victim_sel_policy p; | |
b2b3460a | 259 | unsigned int secno, max_cost; |
7bc09003 JK |
260 | int nsearched = 0; |
261 | ||
262 | p.alloc_mode = alloc_mode; | |
263 | select_policy(sbi, gc_type, type, &p); | |
264 | ||
265 | p.min_segno = NULL_SEGNO; | |
b2b3460a | 266 | p.min_cost = max_cost = get_max_cost(sbi, &p); |
7bc09003 JK |
267 | |
268 | mutex_lock(&dirty_i->seglist_lock); | |
269 | ||
270 | if (p.alloc_mode == LFS && gc_type == FG_GC) { | |
271 | p.min_segno = check_bg_victims(sbi); | |
272 | if (p.min_segno != NULL_SEGNO) | |
273 | goto got_it; | |
274 | } | |
275 | ||
276 | while (1) { | |
277 | unsigned long cost; | |
5ec4e49f | 278 | unsigned int segno; |
7bc09003 JK |
279 | |
280 | segno = find_next_bit(p.dirty_segmap, | |
281 | TOTAL_SEGS(sbi), p.offset); | |
282 | if (segno >= TOTAL_SEGS(sbi)) { | |
283 | if (sbi->last_victim[p.gc_mode]) { | |
284 | sbi->last_victim[p.gc_mode] = 0; | |
285 | p.offset = 0; | |
286 | continue; | |
287 | } | |
288 | break; | |
289 | } | |
290 | p.offset = ((segno / p.ofs_unit) * p.ofs_unit) + p.ofs_unit; | |
5ec4e49f | 291 | secno = GET_SECNO(sbi, segno); |
7bc09003 | 292 | |
5ec4e49f | 293 | if (sec_usage_check(sbi, secno)) |
7bc09003 | 294 | continue; |
5ec4e49f | 295 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
7bc09003 JK |
296 | continue; |
297 | ||
298 | cost = get_gc_cost(sbi, segno, &p); | |
299 | ||
300 | if (p.min_cost > cost) { | |
301 | p.min_segno = segno; | |
302 | p.min_cost = cost; | |
303 | } | |
304 | ||
b2b3460a | 305 | if (cost == max_cost) |
7bc09003 JK |
306 | continue; |
307 | ||
308 | if (nsearched++ >= MAX_VICTIM_SEARCH) { | |
309 | sbi->last_victim[p.gc_mode] = segno; | |
310 | break; | |
311 | } | |
312 | } | |
7bc09003 | 313 | if (p.min_segno != NULL_SEGNO) { |
b2b3460a | 314 | got_it: |
7bc09003 | 315 | if (p.alloc_mode == LFS) { |
5ec4e49f JK |
316 | secno = GET_SECNO(sbi, p.min_segno); |
317 | if (gc_type == FG_GC) | |
318 | sbi->cur_victim_sec = secno; | |
319 | else | |
320 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 321 | } |
5ec4e49f | 322 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
8e46b3ed NJ |
323 | |
324 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, | |
325 | sbi->cur_victim_sec, | |
326 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 JK |
327 | } |
328 | mutex_unlock(&dirty_i->seglist_lock); | |
329 | ||
330 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
331 | } | |
332 | ||
333 | static const struct victim_selection default_v_ops = { | |
334 | .get_victim = get_victim_by_default, | |
335 | }; | |
336 | ||
337 | static struct inode *find_gc_inode(nid_t ino, struct list_head *ilist) | |
338 | { | |
7bc09003 JK |
339 | struct inode_entry *ie; |
340 | ||
6cc4af56 | 341 | list_for_each_entry(ie, ilist, list) |
7bc09003 JK |
342 | if (ie->inode->i_ino == ino) |
343 | return ie->inode; | |
7bc09003 JK |
344 | return NULL; |
345 | } | |
346 | ||
347 | static void add_gc_inode(struct inode *inode, struct list_head *ilist) | |
348 | { | |
6cc4af56 GZ |
349 | struct inode_entry *new_ie; |
350 | ||
351 | if (inode == find_gc_inode(inode->i_ino, ilist)) { | |
352 | iput(inode); | |
353 | return; | |
7bc09003 JK |
354 | } |
355 | repeat: | |
356 | new_ie = kmem_cache_alloc(winode_slab, GFP_NOFS); | |
357 | if (!new_ie) { | |
358 | cond_resched(); | |
359 | goto repeat; | |
360 | } | |
361 | new_ie->inode = inode; | |
362 | list_add_tail(&new_ie->list, ilist); | |
363 | } | |
364 | ||
365 | static void put_gc_inode(struct list_head *ilist) | |
366 | { | |
367 | struct inode_entry *ie, *next_ie; | |
368 | list_for_each_entry_safe(ie, next_ie, ilist, list) { | |
369 | iput(ie->inode); | |
370 | list_del(&ie->list); | |
371 | kmem_cache_free(winode_slab, ie); | |
372 | } | |
373 | } | |
374 | ||
375 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
376 | unsigned int segno, int offset) | |
377 | { | |
378 | struct sit_info *sit_i = SIT_I(sbi); | |
379 | struct seg_entry *sentry; | |
380 | int ret; | |
381 | ||
382 | mutex_lock(&sit_i->sentry_lock); | |
383 | sentry = get_seg_entry(sbi, segno); | |
384 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
385 | mutex_unlock(&sit_i->sentry_lock); | |
43727527 | 386 | return ret; |
7bc09003 JK |
387 | } |
388 | ||
0a8165d7 | 389 | /* |
7bc09003 JK |
390 | * This function compares node address got in summary with that in NAT. |
391 | * On validity, copy that node with cold status, otherwise (invalid node) | |
392 | * ignore that. | |
393 | */ | |
43727527 | 394 | static void gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
395 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
396 | { | |
397 | bool initial = true; | |
398 | struct f2fs_summary *entry; | |
399 | int off; | |
400 | ||
401 | next_step: | |
402 | entry = sum; | |
c718379b | 403 | |
7bc09003 JK |
404 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
405 | nid_t nid = le32_to_cpu(entry->nid); | |
406 | struct page *node_page; | |
7bc09003 | 407 | |
43727527 JK |
408 | /* stop BG_GC if there is not enough free sections. */ |
409 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
410 | return; | |
7bc09003 | 411 | |
43727527 | 412 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
413 | continue; |
414 | ||
415 | if (initial) { | |
416 | ra_node_page(sbi, nid); | |
417 | continue; | |
418 | } | |
419 | node_page = get_node_page(sbi, nid); | |
420 | if (IS_ERR(node_page)) | |
421 | continue; | |
422 | ||
423 | /* set page dirty and write it */ | |
4ebefc44 JK |
424 | if (gc_type == FG_GC) { |
425 | f2fs_submit_bio(sbi, NODE, true); | |
426 | wait_on_page_writeback(node_page); | |
7bc09003 | 427 | set_page_dirty(node_page); |
4ebefc44 JK |
428 | } else { |
429 | if (!PageWriteback(node_page)) | |
430 | set_page_dirty(node_page); | |
431 | } | |
7bc09003 JK |
432 | f2fs_put_page(node_page, 1); |
433 | stat_inc_node_blk_count(sbi, 1); | |
434 | } | |
c718379b | 435 | |
7bc09003 JK |
436 | if (initial) { |
437 | initial = false; | |
438 | goto next_step; | |
439 | } | |
440 | ||
441 | if (gc_type == FG_GC) { | |
442 | struct writeback_control wbc = { | |
443 | .sync_mode = WB_SYNC_ALL, | |
444 | .nr_to_write = LONG_MAX, | |
445 | .for_reclaim = 0, | |
446 | }; | |
447 | sync_node_pages(sbi, 0, &wbc); | |
4ebefc44 JK |
448 | |
449 | /* | |
450 | * In the case of FG_GC, it'd be better to reclaim this victim | |
451 | * completely. | |
452 | */ | |
453 | if (get_valid_blocks(sbi, segno, 1) != 0) | |
454 | goto next_step; | |
7bc09003 | 455 | } |
7bc09003 JK |
456 | } |
457 | ||
0a8165d7 | 458 | /* |
9af45ef5 JK |
459 | * Calculate start block index indicating the given node offset. |
460 | * Be careful, caller should give this node offset only indicating direct node | |
461 | * blocks. If any node offsets, which point the other types of node blocks such | |
462 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
463 | * bug. | |
7bc09003 JK |
464 | */ |
465 | block_t start_bidx_of_node(unsigned int node_ofs) | |
466 | { | |
ce19a5d4 JK |
467 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
468 | unsigned int bidx; | |
7bc09003 | 469 | |
ce19a5d4 JK |
470 | if (node_ofs == 0) |
471 | return 0; | |
7bc09003 | 472 | |
ce19a5d4 | 473 | if (node_ofs <= 2) { |
7bc09003 JK |
474 | bidx = node_ofs - 1; |
475 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 476 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
477 | bidx = node_ofs - 2 - dec; |
478 | } else { | |
ce19a5d4 | 479 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
480 | bidx = node_ofs - 5 - dec; |
481 | } | |
ce19a5d4 | 482 | return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE; |
7bc09003 JK |
483 | } |
484 | ||
485 | static int check_dnode(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, | |
486 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) | |
487 | { | |
488 | struct page *node_page; | |
489 | nid_t nid; | |
490 | unsigned int ofs_in_node; | |
491 | block_t source_blkaddr; | |
492 | ||
493 | nid = le32_to_cpu(sum->nid); | |
494 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
495 | ||
496 | node_page = get_node_page(sbi, nid); | |
497 | if (IS_ERR(node_page)) | |
43727527 | 498 | return 0; |
7bc09003 JK |
499 | |
500 | get_node_info(sbi, nid, dni); | |
501 | ||
502 | if (sum->version != dni->version) { | |
503 | f2fs_put_page(node_page, 1); | |
43727527 | 504 | return 0; |
7bc09003 JK |
505 | } |
506 | ||
507 | *nofs = ofs_of_node(node_page); | |
508 | source_blkaddr = datablock_addr(node_page, ofs_in_node); | |
509 | f2fs_put_page(node_page, 1); | |
510 | ||
511 | if (source_blkaddr != blkaddr) | |
43727527 JK |
512 | return 0; |
513 | return 1; | |
7bc09003 JK |
514 | } |
515 | ||
516 | static void move_data_page(struct inode *inode, struct page *page, int gc_type) | |
517 | { | |
7bc09003 | 518 | if (gc_type == BG_GC) { |
4ebefc44 JK |
519 | if (PageWriteback(page)) |
520 | goto out; | |
7bc09003 JK |
521 | set_page_dirty(page); |
522 | set_cold_data(page); | |
523 | } else { | |
524 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
4ebefc44 JK |
525 | |
526 | if (PageWriteback(page)) { | |
527 | f2fs_submit_bio(sbi, DATA, true); | |
528 | wait_on_page_writeback(page); | |
529 | } | |
530 | ||
7bc09003 JK |
531 | if (clear_page_dirty_for_io(page) && |
532 | S_ISDIR(inode->i_mode)) { | |
533 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
534 | inode_dec_dirty_dents(inode); | |
535 | } | |
536 | set_cold_data(page); | |
537 | do_write_data_page(page); | |
7bc09003 JK |
538 | clear_cold_data(page); |
539 | } | |
540 | out: | |
541 | f2fs_put_page(page, 1); | |
542 | } | |
543 | ||
0a8165d7 | 544 | /* |
7bc09003 JK |
545 | * This function tries to get parent node of victim data block, and identifies |
546 | * data block validity. If the block is valid, copy that with cold status and | |
547 | * modify parent node. | |
548 | * If the parent node is not valid or the data block address is different, | |
549 | * the victim data block is ignored. | |
550 | */ | |
43727527 | 551 | static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
552 | struct list_head *ilist, unsigned int segno, int gc_type) |
553 | { | |
554 | struct super_block *sb = sbi->sb; | |
555 | struct f2fs_summary *entry; | |
556 | block_t start_addr; | |
43727527 | 557 | int off; |
7bc09003 JK |
558 | int phase = 0; |
559 | ||
560 | start_addr = START_BLOCK(sbi, segno); | |
561 | ||
562 | next_step: | |
563 | entry = sum; | |
c718379b | 564 | |
7bc09003 JK |
565 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
566 | struct page *data_page; | |
567 | struct inode *inode; | |
568 | struct node_info dni; /* dnode info for the data */ | |
569 | unsigned int ofs_in_node, nofs; | |
570 | block_t start_bidx; | |
571 | ||
43727527 JK |
572 | /* stop BG_GC if there is not enough free sections. */ |
573 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
574 | return; | |
7bc09003 | 575 | |
43727527 | 576 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
577 | continue; |
578 | ||
579 | if (phase == 0) { | |
580 | ra_node_page(sbi, le32_to_cpu(entry->nid)); | |
581 | continue; | |
582 | } | |
583 | ||
584 | /* Get an inode by ino with checking validity */ | |
43727527 | 585 | if (check_dnode(sbi, entry, &dni, start_addr + off, &nofs) == 0) |
7bc09003 JK |
586 | continue; |
587 | ||
588 | if (phase == 1) { | |
589 | ra_node_page(sbi, dni.ino); | |
590 | continue; | |
591 | } | |
592 | ||
593 | start_bidx = start_bidx_of_node(nofs); | |
594 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); | |
595 | ||
596 | if (phase == 2) { | |
d4686d56 | 597 | inode = f2fs_iget(sb, dni.ino); |
7bc09003 JK |
598 | if (IS_ERR(inode)) |
599 | continue; | |
600 | ||
601 | data_page = find_data_page(inode, | |
c718379b | 602 | start_bidx + ofs_in_node, false); |
7bc09003 JK |
603 | if (IS_ERR(data_page)) |
604 | goto next_iput; | |
605 | ||
606 | f2fs_put_page(data_page, 0); | |
607 | add_gc_inode(inode, ilist); | |
608 | } else { | |
609 | inode = find_gc_inode(dni.ino, ilist); | |
610 | if (inode) { | |
611 | data_page = get_lock_data_page(inode, | |
612 | start_bidx + ofs_in_node); | |
613 | if (IS_ERR(data_page)) | |
614 | continue; | |
615 | move_data_page(inode, data_page, gc_type); | |
616 | stat_inc_data_blk_count(sbi, 1); | |
617 | } | |
618 | } | |
619 | continue; | |
620 | next_iput: | |
621 | iput(inode); | |
622 | } | |
c718379b | 623 | |
7bc09003 JK |
624 | if (++phase < 4) |
625 | goto next_step; | |
43727527 | 626 | |
4ebefc44 | 627 | if (gc_type == FG_GC) { |
7bc09003 | 628 | f2fs_submit_bio(sbi, DATA, true); |
4ebefc44 JK |
629 | |
630 | /* | |
631 | * In the case of FG_GC, it'd be better to reclaim this victim | |
632 | * completely. | |
633 | */ | |
634 | if (get_valid_blocks(sbi, segno, 1) != 0) { | |
635 | phase = 2; | |
636 | goto next_step; | |
637 | } | |
638 | } | |
7bc09003 JK |
639 | } |
640 | ||
641 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
642 | int gc_type, int type) | |
643 | { | |
644 | struct sit_info *sit_i = SIT_I(sbi); | |
645 | int ret; | |
646 | mutex_lock(&sit_i->sentry_lock); | |
647 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, type, LFS); | |
648 | mutex_unlock(&sit_i->sentry_lock); | |
649 | return ret; | |
650 | } | |
651 | ||
43727527 | 652 | static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno, |
7bc09003 JK |
653 | struct list_head *ilist, int gc_type) |
654 | { | |
655 | struct page *sum_page; | |
656 | struct f2fs_summary_block *sum; | |
c718379b | 657 | struct blk_plug plug; |
7bc09003 JK |
658 | |
659 | /* read segment summary of victim */ | |
660 | sum_page = get_sum_page(sbi, segno); | |
661 | if (IS_ERR(sum_page)) | |
43727527 | 662 | return; |
7bc09003 | 663 | |
c718379b JK |
664 | blk_start_plug(&plug); |
665 | ||
7bc09003 JK |
666 | sum = page_address(sum_page); |
667 | ||
668 | switch (GET_SUM_TYPE((&sum->footer))) { | |
669 | case SUM_TYPE_NODE: | |
43727527 | 670 | gc_node_segment(sbi, sum->entries, segno, gc_type); |
7bc09003 JK |
671 | break; |
672 | case SUM_TYPE_DATA: | |
43727527 | 673 | gc_data_segment(sbi, sum->entries, ilist, segno, gc_type); |
7bc09003 JK |
674 | break; |
675 | } | |
c718379b JK |
676 | blk_finish_plug(&plug); |
677 | ||
7bc09003 JK |
678 | stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer))); |
679 | stat_inc_call_count(sbi->stat_info); | |
680 | ||
b7473754 | 681 | f2fs_put_page(sum_page, 1); |
7bc09003 JK |
682 | } |
683 | ||
408e9375 | 684 | int f2fs_gc(struct f2fs_sb_info *sbi) |
7bc09003 | 685 | { |
7bc09003 | 686 | struct list_head ilist; |
408e9375 | 687 | unsigned int segno, i; |
7bc09003 | 688 | int gc_type = BG_GC; |
43727527 JK |
689 | int nfree = 0; |
690 | int ret = -1; | |
7bc09003 JK |
691 | |
692 | INIT_LIST_HEAD(&ilist); | |
693 | gc_more: | |
408e9375 JK |
694 | if (!(sbi->sb->s_flags & MS_ACTIVE)) |
695 | goto stop; | |
7bc09003 | 696 | |
d64f8047 | 697 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, nfree)) { |
408e9375 | 698 | gc_type = FG_GC; |
d64f8047 JK |
699 | write_checkpoint(sbi, false); |
700 | } | |
7bc09003 | 701 | |
408e9375 JK |
702 | if (!__get_victim(sbi, &segno, gc_type, NO_CHECK_TYPE)) |
703 | goto stop; | |
43727527 | 704 | ret = 0; |
7bc09003 | 705 | |
43727527 JK |
706 | for (i = 0; i < sbi->segs_per_sec; i++) |
707 | do_garbage_collect(sbi, segno + i, &ilist, gc_type); | |
708 | ||
5ec4e49f JK |
709 | if (gc_type == FG_GC) { |
710 | sbi->cur_victim_sec = NULL_SEGNO; | |
43727527 | 711 | nfree++; |
5ec4e49f JK |
712 | WARN_ON(get_valid_blocks(sbi, segno, sbi->segs_per_sec)); |
713 | } | |
43727527 JK |
714 | |
715 | if (has_not_enough_free_secs(sbi, nfree)) | |
716 | goto gc_more; | |
717 | ||
718 | if (gc_type == FG_GC) | |
719 | write_checkpoint(sbi, false); | |
408e9375 | 720 | stop: |
7bc09003 JK |
721 | mutex_unlock(&sbi->gc_mutex); |
722 | ||
723 | put_gc_inode(&ilist); | |
43727527 | 724 | return ret; |
7bc09003 JK |
725 | } |
726 | ||
727 | void build_gc_manager(struct f2fs_sb_info *sbi) | |
728 | { | |
729 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
730 | } | |
731 | ||
6e6093a8 | 732 | int __init create_gc_caches(void) |
7bc09003 JK |
733 | { |
734 | winode_slab = f2fs_kmem_cache_create("f2fs_gc_inodes", | |
735 | sizeof(struct inode_entry), NULL); | |
736 | if (!winode_slab) | |
737 | return -ENOMEM; | |
738 | return 0; | |
739 | } | |
740 | ||
741 | void destroy_gc_caches(void) | |
742 | { | |
743 | kmem_cache_destroy(winode_slab); | |
744 | } |