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" | |
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; | |
32 | wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; | |
33 | long wait_ms; | |
34 | ||
35 | wait_ms = GC_THREAD_MIN_SLEEP_TIME; | |
36 | ||
37 | do { | |
38 | if (try_to_freeze()) | |
39 | continue; | |
40 | else | |
41 | wait_event_interruptible_timeout(*wq, | |
42 | kthread_should_stop(), | |
43 | msecs_to_jiffies(wait_ms)); | |
44 | if (kthread_should_stop()) | |
45 | break; | |
46 | ||
d6212a5f CL |
47 | if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { |
48 | wait_ms = GC_THREAD_MAX_SLEEP_TIME; | |
49 | continue; | |
50 | } | |
51 | ||
7bc09003 JK |
52 | /* |
53 | * [GC triggering condition] | |
54 | * 0. GC is not conducted currently. | |
55 | * 1. There are enough dirty segments. | |
56 | * 2. IO subsystem is idle by checking the # of writeback pages. | |
57 | * 3. IO subsystem is idle by checking the # of requests in | |
58 | * bdev's request list. | |
59 | * | |
60 | * Note) We have to avoid triggering GCs too much frequently. | |
61 | * Because it is possible that some segments can be | |
62 | * invalidated soon after by user update or deletion. | |
63 | * So, I'd like to wait some time to collect dirty segments. | |
64 | */ | |
65 | if (!mutex_trylock(&sbi->gc_mutex)) | |
66 | continue; | |
67 | ||
68 | if (!is_idle(sbi)) { | |
69 | wait_ms = increase_sleep_time(wait_ms); | |
70 | mutex_unlock(&sbi->gc_mutex); | |
71 | continue; | |
72 | } | |
73 | ||
74 | if (has_enough_invalid_blocks(sbi)) | |
75 | wait_ms = decrease_sleep_time(wait_ms); | |
76 | else | |
77 | wait_ms = increase_sleep_time(wait_ms); | |
78 | ||
79 | sbi->bg_gc++; | |
80 | ||
43727527 JK |
81 | /* if return value is not zero, no victim was selected */ |
82 | if (f2fs_gc(sbi)) | |
7bc09003 JK |
83 | wait_ms = GC_THREAD_NOGC_SLEEP_TIME; |
84 | else if (wait_ms == GC_THREAD_NOGC_SLEEP_TIME) | |
85 | wait_ms = GC_THREAD_MAX_SLEEP_TIME; | |
86 | ||
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; |
7bc09003 | 95 | |
48600e44 CL |
96 | if (!test_opt(sbi, BG_GC)) |
97 | return 0; | |
7bc09003 JK |
98 | gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL); |
99 | if (!gc_th) | |
100 | return -ENOMEM; | |
101 | ||
102 | sbi->gc_thread = gc_th; | |
103 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
104 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, | |
ec7b1f2d | 105 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 JK |
106 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
107 | kfree(gc_th); | |
25718423 | 108 | sbi->gc_thread = NULL; |
7bc09003 JK |
109 | return -ENOMEM; |
110 | } | |
111 | return 0; | |
112 | } | |
113 | ||
114 | void stop_gc_thread(struct f2fs_sb_info *sbi) | |
115 | { | |
116 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
117 | if (!gc_th) | |
118 | return; | |
119 | kthread_stop(gc_th->f2fs_gc_task); | |
120 | kfree(gc_th); | |
121 | sbi->gc_thread = NULL; | |
122 | } | |
123 | ||
124 | static int select_gc_type(int gc_type) | |
125 | { | |
126 | return (gc_type == BG_GC) ? GC_CB : GC_GREEDY; | |
127 | } | |
128 | ||
129 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
130 | int type, struct victim_sel_policy *p) | |
131 | { | |
132 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
133 | ||
4ebefc44 | 134 | if (p->alloc_mode == SSR) { |
7bc09003 JK |
135 | p->gc_mode = GC_GREEDY; |
136 | p->dirty_segmap = dirty_i->dirty_segmap[type]; | |
137 | p->ofs_unit = 1; | |
138 | } else { | |
139 | p->gc_mode = select_gc_type(gc_type); | |
140 | p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; | |
141 | p->ofs_unit = sbi->segs_per_sec; | |
142 | } | |
143 | p->offset = sbi->last_victim[p->gc_mode]; | |
144 | } | |
145 | ||
146 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
147 | struct victim_sel_policy *p) | |
148 | { | |
b7250d2d JK |
149 | /* SSR allocates in a segment unit */ |
150 | if (p->alloc_mode == SSR) | |
151 | return 1 << sbi->log_blocks_per_seg; | |
7bc09003 JK |
152 | if (p->gc_mode == GC_GREEDY) |
153 | return (1 << sbi->log_blocks_per_seg) * p->ofs_unit; | |
154 | else if (p->gc_mode == GC_CB) | |
155 | return UINT_MAX; | |
156 | else /* No other gc_mode */ | |
157 | return 0; | |
158 | } | |
159 | ||
160 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
161 | { | |
162 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f JK |
163 | unsigned int hint = 0; |
164 | unsigned int secno; | |
7bc09003 JK |
165 | |
166 | /* | |
167 | * If the gc_type is FG_GC, we can select victim segments | |
168 | * selected by background GC before. | |
169 | * Those segments guarantee they have small valid blocks. | |
170 | */ | |
5ec4e49f JK |
171 | next: |
172 | secno = find_next_bit(dirty_i->victim_secmap, TOTAL_SECS(sbi), hint++); | |
173 | if (secno < TOTAL_SECS(sbi)) { | |
174 | if (sec_usage_check(sbi, secno)) | |
175 | goto next; | |
176 | clear_bit(secno, dirty_i->victim_secmap); | |
177 | return secno * sbi->segs_per_sec; | |
7bc09003 JK |
178 | } |
179 | return NULL_SEGNO; | |
180 | } | |
181 | ||
182 | static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) | |
183 | { | |
184 | struct sit_info *sit_i = SIT_I(sbi); | |
185 | unsigned int secno = GET_SECNO(sbi, segno); | |
186 | unsigned int start = secno * sbi->segs_per_sec; | |
187 | unsigned long long mtime = 0; | |
188 | unsigned int vblocks; | |
189 | unsigned char age = 0; | |
190 | unsigned char u; | |
191 | unsigned int i; | |
192 | ||
193 | for (i = 0; i < sbi->segs_per_sec; i++) | |
194 | mtime += get_seg_entry(sbi, start + i)->mtime; | |
195 | vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
196 | ||
197 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
198 | vblocks = div_u64(vblocks, sbi->segs_per_sec); | |
199 | ||
200 | u = (vblocks * 100) >> sbi->log_blocks_per_seg; | |
201 | ||
202 | /* Handle if the system time is changed by user */ | |
203 | if (mtime < sit_i->min_mtime) | |
204 | sit_i->min_mtime = mtime; | |
205 | if (mtime > sit_i->max_mtime) | |
206 | sit_i->max_mtime = mtime; | |
207 | if (sit_i->max_mtime != sit_i->min_mtime) | |
208 | age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), | |
209 | sit_i->max_mtime - sit_i->min_mtime); | |
210 | ||
211 | return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); | |
212 | } | |
213 | ||
214 | static unsigned int get_gc_cost(struct f2fs_sb_info *sbi, unsigned int segno, | |
215 | struct victim_sel_policy *p) | |
216 | { | |
217 | if (p->alloc_mode == SSR) | |
218 | return get_seg_entry(sbi, segno)->ckpt_valid_blocks; | |
219 | ||
220 | /* alloc_mode == LFS */ | |
221 | if (p->gc_mode == GC_GREEDY) | |
222 | return get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
223 | else | |
224 | return get_cb_cost(sbi, segno); | |
225 | } | |
226 | ||
0a8165d7 | 227 | /* |
111d2495 | 228 | * This function is called from two paths. |
7bc09003 JK |
229 | * One is garbage collection and the other is SSR segment selection. |
230 | * When it is called during GC, it just gets a victim segment | |
231 | * and it does not remove it from dirty seglist. | |
232 | * When it is called from SSR segment selection, it finds a segment | |
233 | * which has minimum valid blocks and removes it from dirty seglist. | |
234 | */ | |
235 | static int get_victim_by_default(struct f2fs_sb_info *sbi, | |
236 | unsigned int *result, int gc_type, int type, char alloc_mode) | |
237 | { | |
238 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
239 | struct victim_sel_policy p; | |
5ec4e49f | 240 | unsigned int secno; |
7bc09003 JK |
241 | int nsearched = 0; |
242 | ||
243 | p.alloc_mode = alloc_mode; | |
244 | select_policy(sbi, gc_type, type, &p); | |
245 | ||
246 | p.min_segno = NULL_SEGNO; | |
247 | p.min_cost = get_max_cost(sbi, &p); | |
248 | ||
249 | mutex_lock(&dirty_i->seglist_lock); | |
250 | ||
251 | if (p.alloc_mode == LFS && gc_type == FG_GC) { | |
252 | p.min_segno = check_bg_victims(sbi); | |
253 | if (p.min_segno != NULL_SEGNO) | |
254 | goto got_it; | |
255 | } | |
256 | ||
257 | while (1) { | |
258 | unsigned long cost; | |
5ec4e49f | 259 | unsigned int segno; |
7bc09003 JK |
260 | |
261 | segno = find_next_bit(p.dirty_segmap, | |
262 | TOTAL_SEGS(sbi), p.offset); | |
263 | if (segno >= TOTAL_SEGS(sbi)) { | |
264 | if (sbi->last_victim[p.gc_mode]) { | |
265 | sbi->last_victim[p.gc_mode] = 0; | |
266 | p.offset = 0; | |
267 | continue; | |
268 | } | |
269 | break; | |
270 | } | |
271 | p.offset = ((segno / p.ofs_unit) * p.ofs_unit) + p.ofs_unit; | |
5ec4e49f | 272 | secno = GET_SECNO(sbi, segno); |
7bc09003 | 273 | |
5ec4e49f | 274 | if (sec_usage_check(sbi, secno)) |
7bc09003 | 275 | continue; |
5ec4e49f | 276 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
7bc09003 JK |
277 | continue; |
278 | ||
279 | cost = get_gc_cost(sbi, segno, &p); | |
280 | ||
281 | if (p.min_cost > cost) { | |
282 | p.min_segno = segno; | |
283 | p.min_cost = cost; | |
284 | } | |
285 | ||
286 | if (cost == get_max_cost(sbi, &p)) | |
287 | continue; | |
288 | ||
289 | if (nsearched++ >= MAX_VICTIM_SEARCH) { | |
290 | sbi->last_victim[p.gc_mode] = segno; | |
291 | break; | |
292 | } | |
293 | } | |
294 | got_it: | |
295 | if (p.min_segno != NULL_SEGNO) { | |
7bc09003 | 296 | if (p.alloc_mode == LFS) { |
5ec4e49f JK |
297 | secno = GET_SECNO(sbi, p.min_segno); |
298 | if (gc_type == FG_GC) | |
299 | sbi->cur_victim_sec = secno; | |
300 | else | |
301 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 302 | } |
5ec4e49f | 303 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
7bc09003 JK |
304 | } |
305 | mutex_unlock(&dirty_i->seglist_lock); | |
306 | ||
307 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
308 | } | |
309 | ||
310 | static const struct victim_selection default_v_ops = { | |
311 | .get_victim = get_victim_by_default, | |
312 | }; | |
313 | ||
314 | static struct inode *find_gc_inode(nid_t ino, struct list_head *ilist) | |
315 | { | |
316 | struct list_head *this; | |
317 | struct inode_entry *ie; | |
318 | ||
319 | list_for_each(this, ilist) { | |
320 | ie = list_entry(this, struct inode_entry, list); | |
321 | if (ie->inode->i_ino == ino) | |
322 | return ie->inode; | |
323 | } | |
324 | return NULL; | |
325 | } | |
326 | ||
327 | static void add_gc_inode(struct inode *inode, struct list_head *ilist) | |
328 | { | |
329 | struct list_head *this; | |
330 | struct inode_entry *new_ie, *ie; | |
331 | ||
332 | list_for_each(this, ilist) { | |
333 | ie = list_entry(this, struct inode_entry, list); | |
334 | if (ie->inode == inode) { | |
335 | iput(inode); | |
336 | return; | |
337 | } | |
338 | } | |
339 | repeat: | |
340 | new_ie = kmem_cache_alloc(winode_slab, GFP_NOFS); | |
341 | if (!new_ie) { | |
342 | cond_resched(); | |
343 | goto repeat; | |
344 | } | |
345 | new_ie->inode = inode; | |
346 | list_add_tail(&new_ie->list, ilist); | |
347 | } | |
348 | ||
349 | static void put_gc_inode(struct list_head *ilist) | |
350 | { | |
351 | struct inode_entry *ie, *next_ie; | |
352 | list_for_each_entry_safe(ie, next_ie, ilist, list) { | |
353 | iput(ie->inode); | |
354 | list_del(&ie->list); | |
355 | kmem_cache_free(winode_slab, ie); | |
356 | } | |
357 | } | |
358 | ||
359 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
360 | unsigned int segno, int offset) | |
361 | { | |
362 | struct sit_info *sit_i = SIT_I(sbi); | |
363 | struct seg_entry *sentry; | |
364 | int ret; | |
365 | ||
366 | mutex_lock(&sit_i->sentry_lock); | |
367 | sentry = get_seg_entry(sbi, segno); | |
368 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
369 | mutex_unlock(&sit_i->sentry_lock); | |
43727527 | 370 | return ret; |
7bc09003 JK |
371 | } |
372 | ||
0a8165d7 | 373 | /* |
7bc09003 JK |
374 | * This function compares node address got in summary with that in NAT. |
375 | * On validity, copy that node with cold status, otherwise (invalid node) | |
376 | * ignore that. | |
377 | */ | |
43727527 | 378 | static void gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
379 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
380 | { | |
381 | bool initial = true; | |
382 | struct f2fs_summary *entry; | |
383 | int off; | |
384 | ||
385 | next_step: | |
386 | entry = sum; | |
387 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { | |
388 | nid_t nid = le32_to_cpu(entry->nid); | |
389 | struct page *node_page; | |
7bc09003 | 390 | |
43727527 JK |
391 | /* stop BG_GC if there is not enough free sections. */ |
392 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
393 | return; | |
7bc09003 | 394 | |
43727527 | 395 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
396 | continue; |
397 | ||
398 | if (initial) { | |
399 | ra_node_page(sbi, nid); | |
400 | continue; | |
401 | } | |
402 | node_page = get_node_page(sbi, nid); | |
403 | if (IS_ERR(node_page)) | |
404 | continue; | |
405 | ||
406 | /* set page dirty and write it */ | |
4ebefc44 JK |
407 | if (gc_type == FG_GC) { |
408 | f2fs_submit_bio(sbi, NODE, true); | |
409 | wait_on_page_writeback(node_page); | |
7bc09003 | 410 | set_page_dirty(node_page); |
4ebefc44 JK |
411 | } else { |
412 | if (!PageWriteback(node_page)) | |
413 | set_page_dirty(node_page); | |
414 | } | |
7bc09003 JK |
415 | f2fs_put_page(node_page, 1); |
416 | stat_inc_node_blk_count(sbi, 1); | |
417 | } | |
418 | if (initial) { | |
419 | initial = false; | |
420 | goto next_step; | |
421 | } | |
422 | ||
423 | if (gc_type == FG_GC) { | |
424 | struct writeback_control wbc = { | |
425 | .sync_mode = WB_SYNC_ALL, | |
426 | .nr_to_write = LONG_MAX, | |
427 | .for_reclaim = 0, | |
428 | }; | |
429 | sync_node_pages(sbi, 0, &wbc); | |
4ebefc44 JK |
430 | |
431 | /* | |
432 | * In the case of FG_GC, it'd be better to reclaim this victim | |
433 | * completely. | |
434 | */ | |
435 | if (get_valid_blocks(sbi, segno, 1) != 0) | |
436 | goto next_step; | |
7bc09003 | 437 | } |
7bc09003 JK |
438 | } |
439 | ||
0a8165d7 | 440 | /* |
9af45ef5 JK |
441 | * Calculate start block index indicating the given node offset. |
442 | * Be careful, caller should give this node offset only indicating direct node | |
443 | * blocks. If any node offsets, which point the other types of node blocks such | |
444 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
445 | * bug. | |
7bc09003 JK |
446 | */ |
447 | block_t start_bidx_of_node(unsigned int node_ofs) | |
448 | { | |
ce19a5d4 JK |
449 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
450 | unsigned int bidx; | |
7bc09003 | 451 | |
ce19a5d4 JK |
452 | if (node_ofs == 0) |
453 | return 0; | |
7bc09003 | 454 | |
ce19a5d4 | 455 | if (node_ofs <= 2) { |
7bc09003 JK |
456 | bidx = node_ofs - 1; |
457 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 458 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
459 | bidx = node_ofs - 2 - dec; |
460 | } else { | |
ce19a5d4 | 461 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
462 | bidx = node_ofs - 5 - dec; |
463 | } | |
ce19a5d4 | 464 | return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE; |
7bc09003 JK |
465 | } |
466 | ||
467 | static int check_dnode(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, | |
468 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) | |
469 | { | |
470 | struct page *node_page; | |
471 | nid_t nid; | |
472 | unsigned int ofs_in_node; | |
473 | block_t source_blkaddr; | |
474 | ||
475 | nid = le32_to_cpu(sum->nid); | |
476 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
477 | ||
478 | node_page = get_node_page(sbi, nid); | |
479 | if (IS_ERR(node_page)) | |
43727527 | 480 | return 0; |
7bc09003 JK |
481 | |
482 | get_node_info(sbi, nid, dni); | |
483 | ||
484 | if (sum->version != dni->version) { | |
485 | f2fs_put_page(node_page, 1); | |
43727527 | 486 | return 0; |
7bc09003 JK |
487 | } |
488 | ||
489 | *nofs = ofs_of_node(node_page); | |
490 | source_blkaddr = datablock_addr(node_page, ofs_in_node); | |
491 | f2fs_put_page(node_page, 1); | |
492 | ||
493 | if (source_blkaddr != blkaddr) | |
43727527 JK |
494 | return 0; |
495 | return 1; | |
7bc09003 JK |
496 | } |
497 | ||
498 | static void move_data_page(struct inode *inode, struct page *page, int gc_type) | |
499 | { | |
7bc09003 | 500 | if (gc_type == BG_GC) { |
4ebefc44 JK |
501 | if (PageWriteback(page)) |
502 | goto out; | |
7bc09003 JK |
503 | set_page_dirty(page); |
504 | set_cold_data(page); | |
505 | } else { | |
506 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
4ebefc44 JK |
507 | |
508 | if (PageWriteback(page)) { | |
509 | f2fs_submit_bio(sbi, DATA, true); | |
510 | wait_on_page_writeback(page); | |
511 | } | |
512 | ||
7bc09003 JK |
513 | if (clear_page_dirty_for_io(page) && |
514 | S_ISDIR(inode->i_mode)) { | |
515 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
516 | inode_dec_dirty_dents(inode); | |
517 | } | |
518 | set_cold_data(page); | |
519 | do_write_data_page(page); | |
7bc09003 JK |
520 | clear_cold_data(page); |
521 | } | |
522 | out: | |
523 | f2fs_put_page(page, 1); | |
524 | } | |
525 | ||
0a8165d7 | 526 | /* |
7bc09003 JK |
527 | * This function tries to get parent node of victim data block, and identifies |
528 | * data block validity. If the block is valid, copy that with cold status and | |
529 | * modify parent node. | |
530 | * If the parent node is not valid or the data block address is different, | |
531 | * the victim data block is ignored. | |
532 | */ | |
43727527 | 533 | static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
534 | struct list_head *ilist, unsigned int segno, int gc_type) |
535 | { | |
536 | struct super_block *sb = sbi->sb; | |
537 | struct f2fs_summary *entry; | |
538 | block_t start_addr; | |
43727527 | 539 | int off; |
7bc09003 JK |
540 | int phase = 0; |
541 | ||
542 | start_addr = START_BLOCK(sbi, segno); | |
543 | ||
544 | next_step: | |
545 | entry = sum; | |
546 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { | |
547 | struct page *data_page; | |
548 | struct inode *inode; | |
549 | struct node_info dni; /* dnode info for the data */ | |
550 | unsigned int ofs_in_node, nofs; | |
551 | block_t start_bidx; | |
552 | ||
43727527 JK |
553 | /* stop BG_GC if there is not enough free sections. */ |
554 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
555 | return; | |
7bc09003 | 556 | |
43727527 | 557 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
558 | continue; |
559 | ||
560 | if (phase == 0) { | |
561 | ra_node_page(sbi, le32_to_cpu(entry->nid)); | |
562 | continue; | |
563 | } | |
564 | ||
565 | /* Get an inode by ino with checking validity */ | |
43727527 | 566 | if (check_dnode(sbi, entry, &dni, start_addr + off, &nofs) == 0) |
7bc09003 JK |
567 | continue; |
568 | ||
569 | if (phase == 1) { | |
570 | ra_node_page(sbi, dni.ino); | |
571 | continue; | |
572 | } | |
573 | ||
574 | start_bidx = start_bidx_of_node(nofs); | |
575 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); | |
576 | ||
577 | if (phase == 2) { | |
d4686d56 | 578 | inode = f2fs_iget(sb, dni.ino); |
7bc09003 JK |
579 | if (IS_ERR(inode)) |
580 | continue; | |
581 | ||
582 | data_page = find_data_page(inode, | |
583 | start_bidx + ofs_in_node); | |
584 | if (IS_ERR(data_page)) | |
585 | goto next_iput; | |
586 | ||
587 | f2fs_put_page(data_page, 0); | |
588 | add_gc_inode(inode, ilist); | |
589 | } else { | |
590 | inode = find_gc_inode(dni.ino, ilist); | |
591 | if (inode) { | |
592 | data_page = get_lock_data_page(inode, | |
593 | start_bidx + ofs_in_node); | |
594 | if (IS_ERR(data_page)) | |
595 | continue; | |
596 | move_data_page(inode, data_page, gc_type); | |
597 | stat_inc_data_blk_count(sbi, 1); | |
598 | } | |
599 | } | |
600 | continue; | |
601 | next_iput: | |
602 | iput(inode); | |
603 | } | |
604 | if (++phase < 4) | |
605 | goto next_step; | |
43727527 | 606 | |
4ebefc44 | 607 | if (gc_type == FG_GC) { |
7bc09003 | 608 | f2fs_submit_bio(sbi, DATA, true); |
4ebefc44 JK |
609 | |
610 | /* | |
611 | * In the case of FG_GC, it'd be better to reclaim this victim | |
612 | * completely. | |
613 | */ | |
614 | if (get_valid_blocks(sbi, segno, 1) != 0) { | |
615 | phase = 2; | |
616 | goto next_step; | |
617 | } | |
618 | } | |
7bc09003 JK |
619 | } |
620 | ||
621 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
622 | int gc_type, int type) | |
623 | { | |
624 | struct sit_info *sit_i = SIT_I(sbi); | |
625 | int ret; | |
626 | mutex_lock(&sit_i->sentry_lock); | |
627 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, type, LFS); | |
628 | mutex_unlock(&sit_i->sentry_lock); | |
629 | return ret; | |
630 | } | |
631 | ||
43727527 | 632 | static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno, |
7bc09003 JK |
633 | struct list_head *ilist, int gc_type) |
634 | { | |
635 | struct page *sum_page; | |
636 | struct f2fs_summary_block *sum; | |
7bc09003 JK |
637 | |
638 | /* read segment summary of victim */ | |
639 | sum_page = get_sum_page(sbi, segno); | |
640 | if (IS_ERR(sum_page)) | |
43727527 | 641 | return; |
7bc09003 | 642 | |
7bc09003 JK |
643 | sum = page_address(sum_page); |
644 | ||
645 | switch (GET_SUM_TYPE((&sum->footer))) { | |
646 | case SUM_TYPE_NODE: | |
43727527 | 647 | gc_node_segment(sbi, sum->entries, segno, gc_type); |
7bc09003 JK |
648 | break; |
649 | case SUM_TYPE_DATA: | |
43727527 | 650 | gc_data_segment(sbi, sum->entries, ilist, segno, gc_type); |
7bc09003 JK |
651 | break; |
652 | } | |
653 | stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer))); | |
654 | stat_inc_call_count(sbi->stat_info); | |
655 | ||
b7473754 | 656 | f2fs_put_page(sum_page, 1); |
7bc09003 JK |
657 | } |
658 | ||
408e9375 | 659 | int f2fs_gc(struct f2fs_sb_info *sbi) |
7bc09003 | 660 | { |
7bc09003 | 661 | struct list_head ilist; |
408e9375 | 662 | unsigned int segno, i; |
7bc09003 | 663 | int gc_type = BG_GC; |
43727527 JK |
664 | int nfree = 0; |
665 | int ret = -1; | |
7bc09003 JK |
666 | |
667 | INIT_LIST_HEAD(&ilist); | |
668 | gc_more: | |
408e9375 JK |
669 | if (!(sbi->sb->s_flags & MS_ACTIVE)) |
670 | goto stop; | |
7bc09003 | 671 | |
d64f8047 | 672 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, nfree)) { |
408e9375 | 673 | gc_type = FG_GC; |
d64f8047 JK |
674 | write_checkpoint(sbi, false); |
675 | } | |
7bc09003 | 676 | |
408e9375 JK |
677 | if (!__get_victim(sbi, &segno, gc_type, NO_CHECK_TYPE)) |
678 | goto stop; | |
43727527 | 679 | ret = 0; |
7bc09003 | 680 | |
43727527 JK |
681 | for (i = 0; i < sbi->segs_per_sec; i++) |
682 | do_garbage_collect(sbi, segno + i, &ilist, gc_type); | |
683 | ||
5ec4e49f JK |
684 | if (gc_type == FG_GC) { |
685 | sbi->cur_victim_sec = NULL_SEGNO; | |
43727527 | 686 | nfree++; |
5ec4e49f JK |
687 | WARN_ON(get_valid_blocks(sbi, segno, sbi->segs_per_sec)); |
688 | } | |
43727527 JK |
689 | |
690 | if (has_not_enough_free_secs(sbi, nfree)) | |
691 | goto gc_more; | |
692 | ||
693 | if (gc_type == FG_GC) | |
694 | write_checkpoint(sbi, false); | |
408e9375 | 695 | stop: |
7bc09003 JK |
696 | mutex_unlock(&sbi->gc_mutex); |
697 | ||
698 | put_gc_inode(&ilist); | |
43727527 | 699 | return ret; |
7bc09003 JK |
700 | } |
701 | ||
702 | void build_gc_manager(struct f2fs_sb_info *sbi) | |
703 | { | |
704 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
705 | } | |
706 | ||
6e6093a8 | 707 | int __init create_gc_caches(void) |
7bc09003 JK |
708 | { |
709 | winode_slab = f2fs_kmem_cache_create("f2fs_gc_inodes", | |
710 | sizeof(struct inode_entry), NULL); | |
711 | if (!winode_slab) | |
712 | return -ENOMEM; | |
713 | return 0; | |
714 | } | |
715 | ||
716 | void destroy_gc_caches(void) | |
717 | { | |
718 | kmem_cache_destroy(winode_slab); | |
719 | } |