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; | |
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 | ||
35b09d82 | 79 | #ifdef CONFIG_F2FS_STAT_FS |
7bc09003 | 80 | sbi->bg_gc++; |
35b09d82 | 81 | #endif |
7bc09003 | 82 | |
43727527 JK |
83 | /* if return value is not zero, no victim was selected */ |
84 | if (f2fs_gc(sbi)) | |
7bc09003 | 85 | wait_ms = GC_THREAD_NOGC_SLEEP_TIME; |
7bc09003 JK |
86 | } while (!kthread_should_stop()); |
87 | return 0; | |
88 | } | |
89 | ||
90 | int start_gc_thread(struct f2fs_sb_info *sbi) | |
91 | { | |
1042d60f | 92 | struct f2fs_gc_kthread *gc_th; |
ec7b1f2d | 93 | dev_t dev = sbi->sb->s_bdev->bd_dev; |
7a267f8d | 94 | int err = 0; |
7bc09003 | 95 | |
48600e44 | 96 | if (!test_opt(sbi, BG_GC)) |
7a267f8d | 97 | goto out; |
7bc09003 | 98 | gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL); |
7a267f8d NJ |
99 | if (!gc_th) { |
100 | err = -ENOMEM; | |
101 | goto out; | |
102 | } | |
7bc09003 JK |
103 | |
104 | sbi->gc_thread = gc_th; | |
105 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
106 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, | |
ec7b1f2d | 107 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 | 108 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
7a267f8d | 109 | err = PTR_ERR(gc_th->f2fs_gc_task); |
7bc09003 | 110 | kfree(gc_th); |
25718423 | 111 | sbi->gc_thread = NULL; |
7bc09003 | 112 | } |
7a267f8d NJ |
113 | |
114 | out: | |
115 | return err; | |
7bc09003 JK |
116 | } |
117 | ||
118 | void stop_gc_thread(struct f2fs_sb_info *sbi) | |
119 | { | |
120 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
121 | if (!gc_th) | |
122 | return; | |
123 | kthread_stop(gc_th->f2fs_gc_task); | |
124 | kfree(gc_th); | |
125 | sbi->gc_thread = NULL; | |
126 | } | |
127 | ||
128 | static int select_gc_type(int gc_type) | |
129 | { | |
130 | return (gc_type == BG_GC) ? GC_CB : GC_GREEDY; | |
131 | } | |
132 | ||
133 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
134 | int type, struct victim_sel_policy *p) | |
135 | { | |
136 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
137 | ||
4ebefc44 | 138 | if (p->alloc_mode == SSR) { |
7bc09003 JK |
139 | p->gc_mode = GC_GREEDY; |
140 | p->dirty_segmap = dirty_i->dirty_segmap[type]; | |
141 | p->ofs_unit = 1; | |
142 | } else { | |
143 | p->gc_mode = select_gc_type(gc_type); | |
144 | p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; | |
145 | p->ofs_unit = sbi->segs_per_sec; | |
146 | } | |
147 | p->offset = sbi->last_victim[p->gc_mode]; | |
148 | } | |
149 | ||
150 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
151 | struct victim_sel_policy *p) | |
152 | { | |
b7250d2d JK |
153 | /* SSR allocates in a segment unit */ |
154 | if (p->alloc_mode == SSR) | |
155 | return 1 << sbi->log_blocks_per_seg; | |
7bc09003 JK |
156 | if (p->gc_mode == GC_GREEDY) |
157 | return (1 << sbi->log_blocks_per_seg) * p->ofs_unit; | |
158 | else if (p->gc_mode == GC_CB) | |
159 | return UINT_MAX; | |
160 | else /* No other gc_mode */ | |
161 | return 0; | |
162 | } | |
163 | ||
164 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
165 | { | |
166 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f JK |
167 | unsigned int hint = 0; |
168 | unsigned int secno; | |
7bc09003 JK |
169 | |
170 | /* | |
171 | * If the gc_type is FG_GC, we can select victim segments | |
172 | * selected by background GC before. | |
173 | * Those segments guarantee they have small valid blocks. | |
174 | */ | |
5ec4e49f JK |
175 | next: |
176 | secno = find_next_bit(dirty_i->victim_secmap, TOTAL_SECS(sbi), hint++); | |
177 | if (secno < TOTAL_SECS(sbi)) { | |
178 | if (sec_usage_check(sbi, secno)) | |
179 | goto next; | |
180 | clear_bit(secno, dirty_i->victim_secmap); | |
181 | return secno * sbi->segs_per_sec; | |
7bc09003 JK |
182 | } |
183 | return NULL_SEGNO; | |
184 | } | |
185 | ||
186 | static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) | |
187 | { | |
188 | struct sit_info *sit_i = SIT_I(sbi); | |
189 | unsigned int secno = GET_SECNO(sbi, segno); | |
190 | unsigned int start = secno * sbi->segs_per_sec; | |
191 | unsigned long long mtime = 0; | |
192 | unsigned int vblocks; | |
193 | unsigned char age = 0; | |
194 | unsigned char u; | |
195 | unsigned int i; | |
196 | ||
197 | for (i = 0; i < sbi->segs_per_sec; i++) | |
198 | mtime += get_seg_entry(sbi, start + i)->mtime; | |
199 | vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
200 | ||
201 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
202 | vblocks = div_u64(vblocks, sbi->segs_per_sec); | |
203 | ||
204 | u = (vblocks * 100) >> sbi->log_blocks_per_seg; | |
205 | ||
206 | /* Handle if the system time is changed by user */ | |
207 | if (mtime < sit_i->min_mtime) | |
208 | sit_i->min_mtime = mtime; | |
209 | if (mtime > sit_i->max_mtime) | |
210 | sit_i->max_mtime = mtime; | |
211 | if (sit_i->max_mtime != sit_i->min_mtime) | |
212 | age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), | |
213 | sit_i->max_mtime - sit_i->min_mtime); | |
214 | ||
215 | return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); | |
216 | } | |
217 | ||
218 | static unsigned int get_gc_cost(struct f2fs_sb_info *sbi, unsigned int segno, | |
219 | struct victim_sel_policy *p) | |
220 | { | |
221 | if (p->alloc_mode == SSR) | |
222 | return get_seg_entry(sbi, segno)->ckpt_valid_blocks; | |
223 | ||
224 | /* alloc_mode == LFS */ | |
225 | if (p->gc_mode == GC_GREEDY) | |
226 | return get_valid_blocks(sbi, segno, sbi->segs_per_sec); | |
227 | else | |
228 | return get_cb_cost(sbi, segno); | |
229 | } | |
230 | ||
0a8165d7 | 231 | /* |
111d2495 | 232 | * This function is called from two paths. |
7bc09003 JK |
233 | * One is garbage collection and the other is SSR segment selection. |
234 | * When it is called during GC, it just gets a victim segment | |
235 | * and it does not remove it from dirty seglist. | |
236 | * When it is called from SSR segment selection, it finds a segment | |
237 | * which has minimum valid blocks and removes it from dirty seglist. | |
238 | */ | |
239 | static int get_victim_by_default(struct f2fs_sb_info *sbi, | |
240 | unsigned int *result, int gc_type, int type, char alloc_mode) | |
241 | { | |
242 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
243 | struct victim_sel_policy p; | |
5ec4e49f | 244 | unsigned int secno; |
7bc09003 JK |
245 | int nsearched = 0; |
246 | ||
247 | p.alloc_mode = alloc_mode; | |
248 | select_policy(sbi, gc_type, type, &p); | |
249 | ||
250 | p.min_segno = NULL_SEGNO; | |
251 | p.min_cost = get_max_cost(sbi, &p); | |
252 | ||
253 | mutex_lock(&dirty_i->seglist_lock); | |
254 | ||
255 | if (p.alloc_mode == LFS && gc_type == FG_GC) { | |
256 | p.min_segno = check_bg_victims(sbi); | |
257 | if (p.min_segno != NULL_SEGNO) | |
258 | goto got_it; | |
259 | } | |
260 | ||
261 | while (1) { | |
262 | unsigned long cost; | |
5ec4e49f | 263 | unsigned int segno; |
7bc09003 JK |
264 | |
265 | segno = find_next_bit(p.dirty_segmap, | |
266 | TOTAL_SEGS(sbi), p.offset); | |
267 | if (segno >= TOTAL_SEGS(sbi)) { | |
268 | if (sbi->last_victim[p.gc_mode]) { | |
269 | sbi->last_victim[p.gc_mode] = 0; | |
270 | p.offset = 0; | |
271 | continue; | |
272 | } | |
273 | break; | |
274 | } | |
275 | p.offset = ((segno / p.ofs_unit) * p.ofs_unit) + p.ofs_unit; | |
5ec4e49f | 276 | secno = GET_SECNO(sbi, segno); |
7bc09003 | 277 | |
5ec4e49f | 278 | if (sec_usage_check(sbi, secno)) |
7bc09003 | 279 | continue; |
5ec4e49f | 280 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
7bc09003 JK |
281 | continue; |
282 | ||
283 | cost = get_gc_cost(sbi, segno, &p); | |
284 | ||
285 | if (p.min_cost > cost) { | |
286 | p.min_segno = segno; | |
287 | p.min_cost = cost; | |
288 | } | |
289 | ||
290 | if (cost == get_max_cost(sbi, &p)) | |
291 | continue; | |
292 | ||
293 | if (nsearched++ >= MAX_VICTIM_SEARCH) { | |
294 | sbi->last_victim[p.gc_mode] = segno; | |
295 | break; | |
296 | } | |
297 | } | |
298 | got_it: | |
299 | if (p.min_segno != NULL_SEGNO) { | |
7bc09003 | 300 | if (p.alloc_mode == LFS) { |
5ec4e49f JK |
301 | secno = GET_SECNO(sbi, p.min_segno); |
302 | if (gc_type == FG_GC) | |
303 | sbi->cur_victim_sec = secno; | |
304 | else | |
305 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 306 | } |
5ec4e49f | 307 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
8e46b3ed NJ |
308 | |
309 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, | |
310 | sbi->cur_victim_sec, | |
311 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 JK |
312 | } |
313 | mutex_unlock(&dirty_i->seglist_lock); | |
314 | ||
315 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
316 | } | |
317 | ||
318 | static const struct victim_selection default_v_ops = { | |
319 | .get_victim = get_victim_by_default, | |
320 | }; | |
321 | ||
322 | static struct inode *find_gc_inode(nid_t ino, struct list_head *ilist) | |
323 | { | |
324 | struct list_head *this; | |
325 | struct inode_entry *ie; | |
326 | ||
327 | list_for_each(this, ilist) { | |
328 | ie = list_entry(this, struct inode_entry, list); | |
329 | if (ie->inode->i_ino == ino) | |
330 | return ie->inode; | |
331 | } | |
332 | return NULL; | |
333 | } | |
334 | ||
335 | static void add_gc_inode(struct inode *inode, struct list_head *ilist) | |
336 | { | |
337 | struct list_head *this; | |
338 | struct inode_entry *new_ie, *ie; | |
339 | ||
340 | list_for_each(this, ilist) { | |
341 | ie = list_entry(this, struct inode_entry, list); | |
342 | if (ie->inode == inode) { | |
343 | iput(inode); | |
344 | return; | |
345 | } | |
346 | } | |
347 | repeat: | |
348 | new_ie = kmem_cache_alloc(winode_slab, GFP_NOFS); | |
349 | if (!new_ie) { | |
350 | cond_resched(); | |
351 | goto repeat; | |
352 | } | |
353 | new_ie->inode = inode; | |
354 | list_add_tail(&new_ie->list, ilist); | |
355 | } | |
356 | ||
357 | static void put_gc_inode(struct list_head *ilist) | |
358 | { | |
359 | struct inode_entry *ie, *next_ie; | |
360 | list_for_each_entry_safe(ie, next_ie, ilist, list) { | |
361 | iput(ie->inode); | |
362 | list_del(&ie->list); | |
363 | kmem_cache_free(winode_slab, ie); | |
364 | } | |
365 | } | |
366 | ||
367 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
368 | unsigned int segno, int offset) | |
369 | { | |
370 | struct sit_info *sit_i = SIT_I(sbi); | |
371 | struct seg_entry *sentry; | |
372 | int ret; | |
373 | ||
374 | mutex_lock(&sit_i->sentry_lock); | |
375 | sentry = get_seg_entry(sbi, segno); | |
376 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
377 | mutex_unlock(&sit_i->sentry_lock); | |
43727527 | 378 | return ret; |
7bc09003 JK |
379 | } |
380 | ||
0a8165d7 | 381 | /* |
7bc09003 JK |
382 | * This function compares node address got in summary with that in NAT. |
383 | * On validity, copy that node with cold status, otherwise (invalid node) | |
384 | * ignore that. | |
385 | */ | |
43727527 | 386 | static void gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
387 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
388 | { | |
389 | bool initial = true; | |
390 | struct f2fs_summary *entry; | |
391 | int off; | |
392 | ||
393 | next_step: | |
394 | entry = sum; | |
c718379b | 395 | |
7bc09003 JK |
396 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
397 | nid_t nid = le32_to_cpu(entry->nid); | |
398 | struct page *node_page; | |
7bc09003 | 399 | |
43727527 JK |
400 | /* stop BG_GC if there is not enough free sections. */ |
401 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
402 | return; | |
7bc09003 | 403 | |
43727527 | 404 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
405 | continue; |
406 | ||
407 | if (initial) { | |
408 | ra_node_page(sbi, nid); | |
409 | continue; | |
410 | } | |
411 | node_page = get_node_page(sbi, nid); | |
412 | if (IS_ERR(node_page)) | |
413 | continue; | |
414 | ||
415 | /* set page dirty and write it */ | |
4ebefc44 JK |
416 | if (gc_type == FG_GC) { |
417 | f2fs_submit_bio(sbi, NODE, true); | |
418 | wait_on_page_writeback(node_page); | |
7bc09003 | 419 | set_page_dirty(node_page); |
4ebefc44 JK |
420 | } else { |
421 | if (!PageWriteback(node_page)) | |
422 | set_page_dirty(node_page); | |
423 | } | |
7bc09003 JK |
424 | f2fs_put_page(node_page, 1); |
425 | stat_inc_node_blk_count(sbi, 1); | |
426 | } | |
c718379b | 427 | |
7bc09003 JK |
428 | if (initial) { |
429 | initial = false; | |
430 | goto next_step; | |
431 | } | |
432 | ||
433 | if (gc_type == FG_GC) { | |
434 | struct writeback_control wbc = { | |
435 | .sync_mode = WB_SYNC_ALL, | |
436 | .nr_to_write = LONG_MAX, | |
437 | .for_reclaim = 0, | |
438 | }; | |
439 | sync_node_pages(sbi, 0, &wbc); | |
4ebefc44 JK |
440 | |
441 | /* | |
442 | * In the case of FG_GC, it'd be better to reclaim this victim | |
443 | * completely. | |
444 | */ | |
445 | if (get_valid_blocks(sbi, segno, 1) != 0) | |
446 | goto next_step; | |
7bc09003 | 447 | } |
7bc09003 JK |
448 | } |
449 | ||
0a8165d7 | 450 | /* |
9af45ef5 JK |
451 | * Calculate start block index indicating the given node offset. |
452 | * Be careful, caller should give this node offset only indicating direct node | |
453 | * blocks. If any node offsets, which point the other types of node blocks such | |
454 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
455 | * bug. | |
7bc09003 JK |
456 | */ |
457 | block_t start_bidx_of_node(unsigned int node_ofs) | |
458 | { | |
ce19a5d4 JK |
459 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
460 | unsigned int bidx; | |
7bc09003 | 461 | |
ce19a5d4 JK |
462 | if (node_ofs == 0) |
463 | return 0; | |
7bc09003 | 464 | |
ce19a5d4 | 465 | if (node_ofs <= 2) { |
7bc09003 JK |
466 | bidx = node_ofs - 1; |
467 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 468 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
469 | bidx = node_ofs - 2 - dec; |
470 | } else { | |
ce19a5d4 | 471 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
472 | bidx = node_ofs - 5 - dec; |
473 | } | |
ce19a5d4 | 474 | return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE; |
7bc09003 JK |
475 | } |
476 | ||
477 | static int check_dnode(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, | |
478 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) | |
479 | { | |
480 | struct page *node_page; | |
481 | nid_t nid; | |
482 | unsigned int ofs_in_node; | |
483 | block_t source_blkaddr; | |
484 | ||
485 | nid = le32_to_cpu(sum->nid); | |
486 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
487 | ||
488 | node_page = get_node_page(sbi, nid); | |
489 | if (IS_ERR(node_page)) | |
43727527 | 490 | return 0; |
7bc09003 JK |
491 | |
492 | get_node_info(sbi, nid, dni); | |
493 | ||
494 | if (sum->version != dni->version) { | |
495 | f2fs_put_page(node_page, 1); | |
43727527 | 496 | return 0; |
7bc09003 JK |
497 | } |
498 | ||
499 | *nofs = ofs_of_node(node_page); | |
500 | source_blkaddr = datablock_addr(node_page, ofs_in_node); | |
501 | f2fs_put_page(node_page, 1); | |
502 | ||
503 | if (source_blkaddr != blkaddr) | |
43727527 JK |
504 | return 0; |
505 | return 1; | |
7bc09003 JK |
506 | } |
507 | ||
508 | static void move_data_page(struct inode *inode, struct page *page, int gc_type) | |
509 | { | |
7bc09003 | 510 | if (gc_type == BG_GC) { |
4ebefc44 JK |
511 | if (PageWriteback(page)) |
512 | goto out; | |
7bc09003 JK |
513 | set_page_dirty(page); |
514 | set_cold_data(page); | |
515 | } else { | |
516 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
4ebefc44 JK |
517 | |
518 | if (PageWriteback(page)) { | |
519 | f2fs_submit_bio(sbi, DATA, true); | |
520 | wait_on_page_writeback(page); | |
521 | } | |
522 | ||
7bc09003 JK |
523 | if (clear_page_dirty_for_io(page) && |
524 | S_ISDIR(inode->i_mode)) { | |
525 | dec_page_count(sbi, F2FS_DIRTY_DENTS); | |
526 | inode_dec_dirty_dents(inode); | |
527 | } | |
528 | set_cold_data(page); | |
529 | do_write_data_page(page); | |
7bc09003 JK |
530 | clear_cold_data(page); |
531 | } | |
532 | out: | |
533 | f2fs_put_page(page, 1); | |
534 | } | |
535 | ||
0a8165d7 | 536 | /* |
7bc09003 JK |
537 | * This function tries to get parent node of victim data block, and identifies |
538 | * data block validity. If the block is valid, copy that with cold status and | |
539 | * modify parent node. | |
540 | * If the parent node is not valid or the data block address is different, | |
541 | * the victim data block is ignored. | |
542 | */ | |
43727527 | 543 | static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
544 | struct list_head *ilist, unsigned int segno, int gc_type) |
545 | { | |
546 | struct super_block *sb = sbi->sb; | |
547 | struct f2fs_summary *entry; | |
548 | block_t start_addr; | |
43727527 | 549 | int off; |
7bc09003 JK |
550 | int phase = 0; |
551 | ||
552 | start_addr = START_BLOCK(sbi, segno); | |
553 | ||
554 | next_step: | |
555 | entry = sum; | |
c718379b | 556 | |
7bc09003 JK |
557 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
558 | struct page *data_page; | |
559 | struct inode *inode; | |
560 | struct node_info dni; /* dnode info for the data */ | |
561 | unsigned int ofs_in_node, nofs; | |
562 | block_t start_bidx; | |
563 | ||
43727527 JK |
564 | /* stop BG_GC if there is not enough free sections. */ |
565 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
566 | return; | |
7bc09003 | 567 | |
43727527 | 568 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
569 | continue; |
570 | ||
571 | if (phase == 0) { | |
572 | ra_node_page(sbi, le32_to_cpu(entry->nid)); | |
573 | continue; | |
574 | } | |
575 | ||
576 | /* Get an inode by ino with checking validity */ | |
43727527 | 577 | if (check_dnode(sbi, entry, &dni, start_addr + off, &nofs) == 0) |
7bc09003 JK |
578 | continue; |
579 | ||
580 | if (phase == 1) { | |
581 | ra_node_page(sbi, dni.ino); | |
582 | continue; | |
583 | } | |
584 | ||
585 | start_bidx = start_bidx_of_node(nofs); | |
586 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); | |
587 | ||
588 | if (phase == 2) { | |
d4686d56 | 589 | inode = f2fs_iget(sb, dni.ino); |
7bc09003 JK |
590 | if (IS_ERR(inode)) |
591 | continue; | |
592 | ||
593 | data_page = find_data_page(inode, | |
c718379b | 594 | start_bidx + ofs_in_node, false); |
7bc09003 JK |
595 | if (IS_ERR(data_page)) |
596 | goto next_iput; | |
597 | ||
598 | f2fs_put_page(data_page, 0); | |
599 | add_gc_inode(inode, ilist); | |
600 | } else { | |
601 | inode = find_gc_inode(dni.ino, ilist); | |
602 | if (inode) { | |
603 | data_page = get_lock_data_page(inode, | |
604 | start_bidx + ofs_in_node); | |
605 | if (IS_ERR(data_page)) | |
606 | continue; | |
607 | move_data_page(inode, data_page, gc_type); | |
608 | stat_inc_data_blk_count(sbi, 1); | |
609 | } | |
610 | } | |
611 | continue; | |
612 | next_iput: | |
613 | iput(inode); | |
614 | } | |
c718379b | 615 | |
7bc09003 JK |
616 | if (++phase < 4) |
617 | goto next_step; | |
43727527 | 618 | |
4ebefc44 | 619 | if (gc_type == FG_GC) { |
7bc09003 | 620 | f2fs_submit_bio(sbi, DATA, true); |
4ebefc44 JK |
621 | |
622 | /* | |
623 | * In the case of FG_GC, it'd be better to reclaim this victim | |
624 | * completely. | |
625 | */ | |
626 | if (get_valid_blocks(sbi, segno, 1) != 0) { | |
627 | phase = 2; | |
628 | goto next_step; | |
629 | } | |
630 | } | |
7bc09003 JK |
631 | } |
632 | ||
633 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
634 | int gc_type, int type) | |
635 | { | |
636 | struct sit_info *sit_i = SIT_I(sbi); | |
637 | int ret; | |
638 | mutex_lock(&sit_i->sentry_lock); | |
639 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, type, LFS); | |
640 | mutex_unlock(&sit_i->sentry_lock); | |
641 | return ret; | |
642 | } | |
643 | ||
43727527 | 644 | static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno, |
7bc09003 JK |
645 | struct list_head *ilist, int gc_type) |
646 | { | |
647 | struct page *sum_page; | |
648 | struct f2fs_summary_block *sum; | |
c718379b | 649 | struct blk_plug plug; |
7bc09003 JK |
650 | |
651 | /* read segment summary of victim */ | |
652 | sum_page = get_sum_page(sbi, segno); | |
653 | if (IS_ERR(sum_page)) | |
43727527 | 654 | return; |
7bc09003 | 655 | |
c718379b JK |
656 | blk_start_plug(&plug); |
657 | ||
7bc09003 JK |
658 | sum = page_address(sum_page); |
659 | ||
660 | switch (GET_SUM_TYPE((&sum->footer))) { | |
661 | case SUM_TYPE_NODE: | |
43727527 | 662 | gc_node_segment(sbi, sum->entries, segno, gc_type); |
7bc09003 JK |
663 | break; |
664 | case SUM_TYPE_DATA: | |
43727527 | 665 | gc_data_segment(sbi, sum->entries, ilist, segno, gc_type); |
7bc09003 JK |
666 | break; |
667 | } | |
c718379b JK |
668 | blk_finish_plug(&plug); |
669 | ||
7bc09003 JK |
670 | stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer))); |
671 | stat_inc_call_count(sbi->stat_info); | |
672 | ||
b7473754 | 673 | f2fs_put_page(sum_page, 1); |
7bc09003 JK |
674 | } |
675 | ||
408e9375 | 676 | int f2fs_gc(struct f2fs_sb_info *sbi) |
7bc09003 | 677 | { |
7bc09003 | 678 | struct list_head ilist; |
408e9375 | 679 | unsigned int segno, i; |
7bc09003 | 680 | int gc_type = BG_GC; |
43727527 JK |
681 | int nfree = 0; |
682 | int ret = -1; | |
7bc09003 JK |
683 | |
684 | INIT_LIST_HEAD(&ilist); | |
685 | gc_more: | |
408e9375 JK |
686 | if (!(sbi->sb->s_flags & MS_ACTIVE)) |
687 | goto stop; | |
7bc09003 | 688 | |
d64f8047 | 689 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, nfree)) { |
408e9375 | 690 | gc_type = FG_GC; |
d64f8047 JK |
691 | write_checkpoint(sbi, false); |
692 | } | |
7bc09003 | 693 | |
408e9375 JK |
694 | if (!__get_victim(sbi, &segno, gc_type, NO_CHECK_TYPE)) |
695 | goto stop; | |
43727527 | 696 | ret = 0; |
7bc09003 | 697 | |
43727527 JK |
698 | for (i = 0; i < sbi->segs_per_sec; i++) |
699 | do_garbage_collect(sbi, segno + i, &ilist, gc_type); | |
700 | ||
5ec4e49f JK |
701 | if (gc_type == FG_GC) { |
702 | sbi->cur_victim_sec = NULL_SEGNO; | |
43727527 | 703 | nfree++; |
5ec4e49f JK |
704 | WARN_ON(get_valid_blocks(sbi, segno, sbi->segs_per_sec)); |
705 | } | |
43727527 JK |
706 | |
707 | if (has_not_enough_free_secs(sbi, nfree)) | |
708 | goto gc_more; | |
709 | ||
710 | if (gc_type == FG_GC) | |
711 | write_checkpoint(sbi, false); | |
408e9375 | 712 | stop: |
7bc09003 JK |
713 | mutex_unlock(&sbi->gc_mutex); |
714 | ||
715 | put_gc_inode(&ilist); | |
43727527 | 716 | return ret; |
7bc09003 JK |
717 | } |
718 | ||
719 | void build_gc_manager(struct f2fs_sb_info *sbi) | |
720 | { | |
721 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
722 | } | |
723 | ||
6e6093a8 | 724 | int __init create_gc_caches(void) |
7bc09003 JK |
725 | { |
726 | winode_slab = f2fs_kmem_cache_create("f2fs_gc_inodes", | |
727 | sizeof(struct inode_entry), NULL); | |
728 | if (!winode_slab) | |
729 | return -ENOMEM; | |
730 | return 0; | |
731 | } | |
732 | ||
733 | void destroy_gc_caches(void) | |
734 | { | |
735 | kmem_cache_destroy(winode_slab); | |
736 | } |