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> | |
14 | #include <linux/proc_fs.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/f2fs_fs.h> | |
17 | #include <linux/kthread.h> | |
18 | #include <linux/delay.h> | |
19 | #include <linux/freezer.h> | |
20 | #include <linux/blkdev.h> | |
21 | ||
22 | #include "f2fs.h" | |
23 | #include "node.h" | |
24 | #include "segment.h" | |
25 | #include "gc.h" | |
8e46b3ed | 26 | #include <trace/events/f2fs.h> |
7bc09003 JK |
27 | |
28 | static struct kmem_cache *winode_slab; | |
29 | ||
30 | static int gc_thread_func(void *data) | |
31 | { | |
32 | struct f2fs_sb_info *sbi = data; | |
33 | wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; | |
34 | long wait_ms; | |
35 | ||
36 | wait_ms = GC_THREAD_MIN_SLEEP_TIME; | |
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 CL |
48 | if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { |
49 | wait_ms = GC_THREAD_MAX_SLEEP_TIME; | |
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)) { | |
70 | wait_ms = increase_sleep_time(wait_ms); | |
71 | mutex_unlock(&sbi->gc_mutex); | |
72 | continue; | |
73 | } | |
74 | ||
75 | if (has_enough_invalid_blocks(sbi)) | |
76 | wait_ms = decrease_sleep_time(wait_ms); | |
77 | else | |
78 | wait_ms = increase_sleep_time(wait_ms); | |
79 | ||
80 | sbi->bg_gc++; | |
81 | ||
43727527 JK |
82 | /* if return value is not zero, no victim was selected */ |
83 | if (f2fs_gc(sbi)) | |
7bc09003 | 84 | wait_ms = GC_THREAD_NOGC_SLEEP_TIME; |
7bc09003 JK |
85 | } while (!kthread_should_stop()); |
86 | return 0; | |
87 | } | |
88 | ||
89 | int start_gc_thread(struct f2fs_sb_info *sbi) | |
90 | { | |
1042d60f | 91 | struct f2fs_gc_kthread *gc_th; |
ec7b1f2d | 92 | dev_t dev = sbi->sb->s_bdev->bd_dev; |
7bc09003 | 93 | |
48600e44 CL |
94 | if (!test_opt(sbi, BG_GC)) |
95 | return 0; | |
7bc09003 JK |
96 | gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL); |
97 | if (!gc_th) | |
98 | return -ENOMEM; | |
99 | ||
100 | sbi->gc_thread = gc_th; | |
101 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
102 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, | |
ec7b1f2d | 103 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 JK |
104 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
105 | kfree(gc_th); | |
25718423 | 106 | sbi->gc_thread = NULL; |
7bc09003 JK |
107 | return -ENOMEM; |
108 | } | |
109 | return 0; | |
110 | } | |
111 | ||
112 | void stop_gc_thread(struct f2fs_sb_info *sbi) | |
113 | { | |
114 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
115 | if (!gc_th) | |
116 | return; | |
117 | kthread_stop(gc_th->f2fs_gc_task); | |
118 | kfree(gc_th); | |
119 | sbi->gc_thread = NULL; | |
120 | } | |
121 | ||
122 | static int select_gc_type(int gc_type) | |
123 | { | |
124 | return (gc_type == BG_GC) ? GC_CB : GC_GREEDY; | |
125 | } | |
126 | ||
127 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
128 | int type, struct victim_sel_policy *p) | |
129 | { | |
130 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
131 | ||
4ebefc44 | 132 | if (p->alloc_mode == SSR) { |
7bc09003 JK |
133 | p->gc_mode = GC_GREEDY; |
134 | p->dirty_segmap = dirty_i->dirty_segmap[type]; | |
135 | p->ofs_unit = 1; | |
136 | } else { | |
137 | p->gc_mode = select_gc_type(gc_type); | |
138 | p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; | |
139 | p->ofs_unit = sbi->segs_per_sec; | |
140 | } | |
141 | p->offset = sbi->last_victim[p->gc_mode]; | |
142 | } | |
143 | ||
144 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
145 | struct victim_sel_policy *p) | |
146 | { | |
b7250d2d JK |
147 | /* SSR allocates in a segment unit */ |
148 | if (p->alloc_mode == SSR) | |
149 | return 1 << sbi->log_blocks_per_seg; | |
7bc09003 JK |
150 | if (p->gc_mode == GC_GREEDY) |
151 | return (1 << sbi->log_blocks_per_seg) * p->ofs_unit; | |
152 | else if (p->gc_mode == GC_CB) | |
153 | return UINT_MAX; | |
154 | else /* No other gc_mode */ | |
155 | return 0; | |
156 | } | |
157 | ||
158 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
159 | { | |
160 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f JK |
161 | unsigned int hint = 0; |
162 | unsigned int secno; | |
7bc09003 JK |
163 | |
164 | /* | |
165 | * If the gc_type is FG_GC, we can select victim segments | |
166 | * selected by background GC before. | |
167 | * Those segments guarantee they have small valid blocks. | |
168 | */ | |
5ec4e49f JK |
169 | next: |
170 | secno = find_next_bit(dirty_i->victim_secmap, TOTAL_SECS(sbi), hint++); | |
171 | if (secno < TOTAL_SECS(sbi)) { | |
172 | if (sec_usage_check(sbi, secno)) | |
173 | goto next; | |
174 | clear_bit(secno, dirty_i->victim_secmap); | |
175 | return secno * sbi->segs_per_sec; | |
7bc09003 JK |
176 | } |
177 | return NULL_SEGNO; | |
178 | } | |
179 | ||
180 | static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) | |
181 | { | |
182 | struct sit_info *sit_i = SIT_I(sbi); | |
183 | unsigned int secno = GET_SECNO(sbi, segno); | |
184 | unsigned int start = secno * sbi->segs_per_sec; | |
185 | unsigned long long mtime = 0; | |
186 | unsigned int vblocks; | |
187 | unsigned char age = 0; | |
188 | unsigned char u; | |
189 | unsigned int i; | |
190 | ||
191 | for (i = 0; i < sbi->segs_per_sec; i++) | |
192 | mtime += get_seg_entry(sbi, start + i)->mtime; | |
193 | vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
194 | ||
195 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
196 | vblocks = div_u64(vblocks, sbi->segs_per_sec); | |
197 | ||
198 | u = (vblocks * 100) >> sbi->log_blocks_per_seg; | |
199 | ||
200 | /* Handle if the system time is changed by user */ | |
201 | if (mtime < sit_i->min_mtime) | |
202 | sit_i->min_mtime = mtime; | |
203 | if (mtime > sit_i->max_mtime) | |
204 | sit_i->max_mtime = mtime; | |
205 | if (sit_i->max_mtime != sit_i->min_mtime) | |
206 | age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), | |
207 | sit_i->max_mtime - sit_i->min_mtime); | |
208 | ||
209 | return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); | |
210 | } | |
211 | ||
212 | static unsigned int get_gc_cost(struct f2fs_sb_info *sbi, unsigned int segno, | |
213 | struct victim_sel_policy *p) | |
214 | { | |
215 | if (p->alloc_mode == SSR) | |
216 | return get_seg_entry(sbi, segno)->ckpt_valid_blocks; | |
217 | ||
218 | /* alloc_mode == LFS */ | |
219 | if (p->gc_mode == GC_GREEDY) | |
220 | return get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
221 | else | |
222 | return get_cb_cost(sbi, segno); | |
223 | } | |
224 | ||
0a8165d7 | 225 | /* |
111d2495 | 226 | * This function is called from two paths. |
7bc09003 JK |
227 | * One is garbage collection and the other is SSR segment selection. |
228 | * When it is called during GC, it just gets a victim segment | |
229 | * and it does not remove it from dirty seglist. | |
230 | * When it is called from SSR segment selection, it finds a segment | |
231 | * which has minimum valid blocks and removes it from dirty seglist. | |
232 | */ | |
233 | static int get_victim_by_default(struct f2fs_sb_info *sbi, | |
234 | unsigned int *result, int gc_type, int type, char alloc_mode) | |
235 | { | |
236 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
237 | struct victim_sel_policy p; | |
5ec4e49f | 238 | unsigned int secno; |
7bc09003 JK |
239 | int nsearched = 0; |
240 | ||
241 | p.alloc_mode = alloc_mode; | |
242 | select_policy(sbi, gc_type, type, &p); | |
243 | ||
244 | p.min_segno = NULL_SEGNO; | |
245 | p.min_cost = get_max_cost(sbi, &p); | |
246 | ||
247 | mutex_lock(&dirty_i->seglist_lock); | |
248 | ||
249 | if (p.alloc_mode == LFS && gc_type == FG_GC) { | |
250 | p.min_segno = check_bg_victims(sbi); | |
251 | if (p.min_segno != NULL_SEGNO) | |
252 | goto got_it; | |
253 | } | |
254 | ||
255 | while (1) { | |
256 | unsigned long cost; | |
5ec4e49f | 257 | unsigned int segno; |
7bc09003 JK |
258 | |
259 | segno = find_next_bit(p.dirty_segmap, | |
260 | TOTAL_SEGS(sbi), p.offset); | |
261 | if (segno >= TOTAL_SEGS(sbi)) { | |
262 | if (sbi->last_victim[p.gc_mode]) { | |
263 | sbi->last_victim[p.gc_mode] = 0; | |
264 | p.offset = 0; | |
265 | continue; | |
266 | } | |
267 | break; | |
268 | } | |
269 | p.offset = ((segno / p.ofs_unit) * p.ofs_unit) + p.ofs_unit; | |
5ec4e49f | 270 | secno = GET_SECNO(sbi, segno); |
7bc09003 | 271 | |
5ec4e49f | 272 | if (sec_usage_check(sbi, secno)) |
7bc09003 | 273 | continue; |
5ec4e49f | 274 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
7bc09003 JK |
275 | continue; |
276 | ||
277 | cost = get_gc_cost(sbi, segno, &p); | |
278 | ||
279 | if (p.min_cost > cost) { | |
280 | p.min_segno = segno; | |
281 | p.min_cost = cost; | |
282 | } | |
283 | ||
284 | if (cost == get_max_cost(sbi, &p)) | |
285 | continue; | |
286 | ||
287 | if (nsearched++ >= MAX_VICTIM_SEARCH) { | |
288 | sbi->last_victim[p.gc_mode] = segno; | |
289 | break; | |
290 | } | |
291 | } | |
292 | got_it: | |
293 | if (p.min_segno != NULL_SEGNO) { | |
7bc09003 | 294 | if (p.alloc_mode == LFS) { |
5ec4e49f JK |
295 | secno = GET_SECNO(sbi, p.min_segno); |
296 | if (gc_type == FG_GC) | |
297 | sbi->cur_victim_sec = secno; | |
298 | else | |
299 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 300 | } |
5ec4e49f | 301 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
8e46b3ed NJ |
302 | |
303 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, | |
304 | sbi->cur_victim_sec, | |
305 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 JK |
306 | } |
307 | mutex_unlock(&dirty_i->seglist_lock); | |
308 | ||
309 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
310 | } | |
311 | ||
312 | static const struct victim_selection default_v_ops = { | |
313 | .get_victim = get_victim_by_default, | |
314 | }; | |
315 | ||
316 | static struct inode *find_gc_inode(nid_t ino, struct list_head *ilist) | |
317 | { | |
318 | struct list_head *this; | |
319 | struct inode_entry *ie; | |
320 | ||
321 | list_for_each(this, ilist) { | |
322 | ie = list_entry(this, struct inode_entry, list); | |
323 | if (ie->inode->i_ino == ino) | |
324 | return ie->inode; | |
325 | } | |
326 | return NULL; | |
327 | } | |
328 | ||
329 | static void add_gc_inode(struct inode *inode, struct list_head *ilist) | |
330 | { | |
331 | struct list_head *this; | |
332 | struct inode_entry *new_ie, *ie; | |
333 | ||
334 | list_for_each(this, ilist) { | |
335 | ie = list_entry(this, struct inode_entry, list); | |
336 | if (ie->inode == inode) { | |
337 | iput(inode); | |
338 | return; | |
339 | } | |
340 | } | |
341 | repeat: | |
342 | new_ie = kmem_cache_alloc(winode_slab, GFP_NOFS); | |
343 | if (!new_ie) { | |
344 | cond_resched(); | |
345 | goto repeat; | |
346 | } | |
347 | new_ie->inode = inode; | |
348 | list_add_tail(&new_ie->list, ilist); | |
349 | } | |
350 | ||
351 | static void put_gc_inode(struct list_head *ilist) | |
352 | { | |
353 | struct inode_entry *ie, *next_ie; | |
354 | list_for_each_entry_safe(ie, next_ie, ilist, list) { | |
355 | iput(ie->inode); | |
356 | list_del(&ie->list); | |
357 | kmem_cache_free(winode_slab, ie); | |
358 | } | |
359 | } | |
360 | ||
361 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
362 | unsigned int segno, int offset) | |
363 | { | |
364 | struct sit_info *sit_i = SIT_I(sbi); | |
365 | struct seg_entry *sentry; | |
366 | int ret; | |
367 | ||
368 | mutex_lock(&sit_i->sentry_lock); | |
369 | sentry = get_seg_entry(sbi, segno); | |
370 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
371 | mutex_unlock(&sit_i->sentry_lock); | |
43727527 | 372 | return ret; |
7bc09003 JK |
373 | } |
374 | ||
0a8165d7 | 375 | /* |
7bc09003 JK |
376 | * This function compares node address got in summary with that in NAT. |
377 | * On validity, copy that node with cold status, otherwise (invalid node) | |
378 | * ignore that. | |
379 | */ | |
43727527 | 380 | static void gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
381 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
382 | { | |
383 | bool initial = true; | |
384 | struct f2fs_summary *entry; | |
385 | int off; | |
386 | ||
387 | next_step: | |
388 | entry = sum; | |
c718379b | 389 | |
7bc09003 JK |
390 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
391 | nid_t nid = le32_to_cpu(entry->nid); | |
392 | struct page *node_page; | |
7bc09003 | 393 | |
43727527 JK |
394 | /* stop BG_GC if there is not enough free sections. */ |
395 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
396 | return; | |
7bc09003 | 397 | |
43727527 | 398 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
399 | continue; |
400 | ||
401 | if (initial) { | |
402 | ra_node_page(sbi, nid); | |
403 | continue; | |
404 | } | |
405 | node_page = get_node_page(sbi, nid); | |
406 | if (IS_ERR(node_page)) | |
407 | continue; | |
408 | ||
409 | /* set page dirty and write it */ | |
4ebefc44 JK |
410 | if (gc_type == FG_GC) { |
411 | f2fs_submit_bio(sbi, NODE, true); | |
412 | wait_on_page_writeback(node_page); | |
7bc09003 | 413 | set_page_dirty(node_page); |
4ebefc44 JK |
414 | } else { |
415 | if (!PageWriteback(node_page)) | |
416 | set_page_dirty(node_page); | |
417 | } | |
7bc09003 JK |
418 | f2fs_put_page(node_page, 1); |
419 | stat_inc_node_blk_count(sbi, 1); | |
420 | } | |
c718379b | 421 | |
7bc09003 JK |
422 | if (initial) { |
423 | initial = false; | |
424 | goto next_step; | |
425 | } | |
426 | ||
427 | if (gc_type == FG_GC) { | |
428 | struct writeback_control wbc = { | |
429 | .sync_mode = WB_SYNC_ALL, | |
430 | .nr_to_write = LONG_MAX, | |
431 | .for_reclaim = 0, | |
432 | }; | |
433 | sync_node_pages(sbi, 0, &wbc); | |
4ebefc44 JK |
434 | |
435 | /* | |
436 | * In the case of FG_GC, it'd be better to reclaim this victim | |
437 | * completely. | |
438 | */ | |
439 | if (get_valid_blocks(sbi, segno, 1) != 0) | |
440 | goto next_step; | |
7bc09003 | 441 | } |
7bc09003 JK |
442 | } |
443 | ||
0a8165d7 | 444 | /* |
9af45ef5 JK |
445 | * Calculate start block index indicating the given node offset. |
446 | * Be careful, caller should give this node offset only indicating direct node | |
447 | * blocks. If any node offsets, which point the other types of node blocks such | |
448 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
449 | * bug. | |
7bc09003 JK |
450 | */ |
451 | block_t start_bidx_of_node(unsigned int node_ofs) | |
452 | { | |
ce19a5d4 JK |
453 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
454 | unsigned int bidx; | |
7bc09003 | 455 | |
ce19a5d4 JK |
456 | if (node_ofs == 0) |
457 | return 0; | |
7bc09003 | 458 | |
ce19a5d4 | 459 | if (node_ofs <= 2) { |
7bc09003 JK |
460 | bidx = node_ofs - 1; |
461 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 462 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
463 | bidx = node_ofs - 2 - dec; |
464 | } else { | |
ce19a5d4 | 465 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
466 | bidx = node_ofs - 5 - dec; |
467 | } | |
ce19a5d4 | 468 | return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE; |
7bc09003 JK |
469 | } |
470 | ||
471 | static int check_dnode(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, | |
472 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) | |
473 | { | |
474 | struct page *node_page; | |
475 | nid_t nid; | |
476 | unsigned int ofs_in_node; | |
477 | block_t source_blkaddr; | |
478 | ||
479 | nid = le32_to_cpu(sum->nid); | |
480 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
481 | ||
482 | node_page = get_node_page(sbi, nid); | |
483 | if (IS_ERR(node_page)) | |
43727527 | 484 | return 0; |
7bc09003 JK |
485 | |
486 | get_node_info(sbi, nid, dni); | |
487 | ||
488 | if (sum->version != dni->version) { | |
489 | f2fs_put_page(node_page, 1); | |
43727527 | 490 | return 0; |
7bc09003 JK |
491 | } |
492 | ||
493 | *nofs = ofs_of_node(node_page); | |
494 | source_blkaddr = datablock_addr(node_page, ofs_in_node); | |
495 | f2fs_put_page(node_page, 1); | |
496 | ||
497 | if (source_blkaddr != blkaddr) | |
43727527 JK |
498 | return 0; |
499 | return 1; | |
7bc09003 JK |
500 | } |
501 | ||
502 | static void move_data_page(struct inode *inode, struct page *page, int gc_type) | |
503 | { | |
7bc09003 | 504 | if (gc_type == BG_GC) { |
4ebefc44 JK |
505 | if (PageWriteback(page)) |
506 | goto out; | |
7bc09003 JK |
507 | set_page_dirty(page); |
508 | set_cold_data(page); | |
509 | } else { | |
510 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
4ebefc44 JK |
511 | |
512 | if (PageWriteback(page)) { | |
513 | f2fs_submit_bio(sbi, DATA, true); | |
514 | wait_on_page_writeback(page); | |
515 | } | |
516 | ||
7bc09003 JK |
517 | if (clear_page_dirty_for_io(page) && |
518 | S_ISDIR(inode->i_mode)) { | |
519 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
520 | inode_dec_dirty_dents(inode); | |
521 | } | |
522 | set_cold_data(page); | |
523 | do_write_data_page(page); | |
7bc09003 JK |
524 | clear_cold_data(page); |
525 | } | |
526 | out: | |
527 | f2fs_put_page(page, 1); | |
528 | } | |
529 | ||
0a8165d7 | 530 | /* |
7bc09003 JK |
531 | * This function tries to get parent node of victim data block, and identifies |
532 | * data block validity. If the block is valid, copy that with cold status and | |
533 | * modify parent node. | |
534 | * If the parent node is not valid or the data block address is different, | |
535 | * the victim data block is ignored. | |
536 | */ | |
43727527 | 537 | static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
538 | struct list_head *ilist, unsigned int segno, int gc_type) |
539 | { | |
540 | struct super_block *sb = sbi->sb; | |
541 | struct f2fs_summary *entry; | |
542 | block_t start_addr; | |
43727527 | 543 | int off; |
7bc09003 JK |
544 | int phase = 0; |
545 | ||
546 | start_addr = START_BLOCK(sbi, segno); | |
547 | ||
548 | next_step: | |
549 | entry = sum; | |
c718379b | 550 | |
7bc09003 JK |
551 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
552 | struct page *data_page; | |
553 | struct inode *inode; | |
554 | struct node_info dni; /* dnode info for the data */ | |
555 | unsigned int ofs_in_node, nofs; | |
556 | block_t start_bidx; | |
557 | ||
43727527 JK |
558 | /* stop BG_GC if there is not enough free sections. */ |
559 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
560 | return; | |
7bc09003 | 561 | |
43727527 | 562 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
563 | continue; |
564 | ||
565 | if (phase == 0) { | |
566 | ra_node_page(sbi, le32_to_cpu(entry->nid)); | |
567 | continue; | |
568 | } | |
569 | ||
570 | /* Get an inode by ino with checking validity */ | |
43727527 | 571 | if (check_dnode(sbi, entry, &dni, start_addr + off, &nofs) == 0) |
7bc09003 JK |
572 | continue; |
573 | ||
574 | if (phase == 1) { | |
575 | ra_node_page(sbi, dni.ino); | |
576 | continue; | |
577 | } | |
578 | ||
579 | start_bidx = start_bidx_of_node(nofs); | |
580 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); | |
581 | ||
582 | if (phase == 2) { | |
d4686d56 | 583 | inode = f2fs_iget(sb, dni.ino); |
7bc09003 JK |
584 | if (IS_ERR(inode)) |
585 | continue; | |
586 | ||
587 | data_page = find_data_page(inode, | |
c718379b | 588 | start_bidx + ofs_in_node, false); |
7bc09003 JK |
589 | if (IS_ERR(data_page)) |
590 | goto next_iput; | |
591 | ||
592 | f2fs_put_page(data_page, 0); | |
593 | add_gc_inode(inode, ilist); | |
594 | } else { | |
595 | inode = find_gc_inode(dni.ino, ilist); | |
596 | if (inode) { | |
597 | data_page = get_lock_data_page(inode, | |
598 | start_bidx + ofs_in_node); | |
599 | if (IS_ERR(data_page)) | |
600 | continue; | |
601 | move_data_page(inode, data_page, gc_type); | |
602 | stat_inc_data_blk_count(sbi, 1); | |
603 | } | |
604 | } | |
605 | continue; | |
606 | next_iput: | |
607 | iput(inode); | |
608 | } | |
c718379b | 609 | |
7bc09003 JK |
610 | if (++phase < 4) |
611 | goto next_step; | |
43727527 | 612 | |
4ebefc44 | 613 | if (gc_type == FG_GC) { |
7bc09003 | 614 | f2fs_submit_bio(sbi, DATA, true); |
4ebefc44 JK |
615 | |
616 | /* | |
617 | * In the case of FG_GC, it'd be better to reclaim this victim | |
618 | * completely. | |
619 | */ | |
620 | if (get_valid_blocks(sbi, segno, 1) != 0) { | |
621 | phase = 2; | |
622 | goto next_step; | |
623 | } | |
624 | } | |
7bc09003 JK |
625 | } |
626 | ||
627 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
628 | int gc_type, int type) | |
629 | { | |
630 | struct sit_info *sit_i = SIT_I(sbi); | |
631 | int ret; | |
632 | mutex_lock(&sit_i->sentry_lock); | |
633 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, type, LFS); | |
634 | mutex_unlock(&sit_i->sentry_lock); | |
635 | return ret; | |
636 | } | |
637 | ||
43727527 | 638 | static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno, |
7bc09003 JK |
639 | struct list_head *ilist, int gc_type) |
640 | { | |
641 | struct page *sum_page; | |
642 | struct f2fs_summary_block *sum; | |
c718379b | 643 | struct blk_plug plug; |
7bc09003 JK |
644 | |
645 | /* read segment summary of victim */ | |
646 | sum_page = get_sum_page(sbi, segno); | |
647 | if (IS_ERR(sum_page)) | |
43727527 | 648 | return; |
7bc09003 | 649 | |
c718379b JK |
650 | blk_start_plug(&plug); |
651 | ||
7bc09003 JK |
652 | sum = page_address(sum_page); |
653 | ||
654 | switch (GET_SUM_TYPE((&sum->footer))) { | |
655 | case SUM_TYPE_NODE: | |
43727527 | 656 | gc_node_segment(sbi, sum->entries, segno, gc_type); |
7bc09003 JK |
657 | break; |
658 | case SUM_TYPE_DATA: | |
43727527 | 659 | gc_data_segment(sbi, sum->entries, ilist, segno, gc_type); |
7bc09003 JK |
660 | break; |
661 | } | |
c718379b JK |
662 | blk_finish_plug(&plug); |
663 | ||
7bc09003 JK |
664 | stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer))); |
665 | stat_inc_call_count(sbi->stat_info); | |
666 | ||
b7473754 | 667 | f2fs_put_page(sum_page, 1); |
7bc09003 JK |
668 | } |
669 | ||
408e9375 | 670 | int f2fs_gc(struct f2fs_sb_info *sbi) |
7bc09003 | 671 | { |
7bc09003 | 672 | struct list_head ilist; |
408e9375 | 673 | unsigned int segno, i; |
7bc09003 | 674 | int gc_type = BG_GC; |
43727527 JK |
675 | int nfree = 0; |
676 | int ret = -1; | |
7bc09003 JK |
677 | |
678 | INIT_LIST_HEAD(&ilist); | |
679 | gc_more: | |
408e9375 JK |
680 | if (!(sbi->sb->s_flags & MS_ACTIVE)) |
681 | goto stop; | |
7bc09003 | 682 | |
d64f8047 | 683 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, nfree)) { |
408e9375 | 684 | gc_type = FG_GC; |
d64f8047 JK |
685 | write_checkpoint(sbi, false); |
686 | } | |
7bc09003 | 687 | |
408e9375 JK |
688 | if (!__get_victim(sbi, &segno, gc_type, NO_CHECK_TYPE)) |
689 | goto stop; | |
43727527 | 690 | ret = 0; |
7bc09003 | 691 | |
43727527 JK |
692 | for (i = 0; i < sbi->segs_per_sec; i++) |
693 | do_garbage_collect(sbi, segno + i, &ilist, gc_type); | |
694 | ||
5ec4e49f JK |
695 | if (gc_type == FG_GC) { |
696 | sbi->cur_victim_sec = NULL_SEGNO; | |
43727527 | 697 | nfree++; |
5ec4e49f JK |
698 | WARN_ON(get_valid_blocks(sbi, segno, sbi->segs_per_sec)); |
699 | } | |
43727527 JK |
700 | |
701 | if (has_not_enough_free_secs(sbi, nfree)) | |
702 | goto gc_more; | |
703 | ||
704 | if (gc_type == FG_GC) | |
705 | write_checkpoint(sbi, false); | |
408e9375 | 706 | stop: |
7bc09003 JK |
707 | mutex_unlock(&sbi->gc_mutex); |
708 | ||
709 | put_gc_inode(&ilist); | |
43727527 | 710 | return ret; |
7bc09003 JK |
711 | } |
712 | ||
713 | void build_gc_manager(struct f2fs_sb_info *sbi) | |
714 | { | |
715 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
716 | } | |
717 | ||
6e6093a8 | 718 | int __init create_gc_caches(void) |
7bc09003 JK |
719 | { |
720 | winode_slab = f2fs_kmem_cache_create("f2fs_gc_inodes", | |
721 | sizeof(struct inode_entry), NULL); | |
722 | if (!winode_slab) | |
723 | return -ENOMEM; | |
724 | return 0; | |
725 | } | |
726 | ||
727 | void destroy_gc_caches(void) | |
728 | { | |
729 | kmem_cache_destroy(winode_slab); | |
730 | } |