]>
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 | 26 | |
7bc09003 JK |
27 | static int gc_thread_func(void *data) |
28 | { | |
29 | struct f2fs_sb_info *sbi = data; | |
b59d0bae | 30 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; |
7bc09003 JK |
31 | wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; |
32 | long wait_ms; | |
33 | ||
b59d0bae | 34 | wait_ms = gc_th->min_sleep_time; |
7bc09003 JK |
35 | |
36 | do { | |
37 | if (try_to_freeze()) | |
38 | continue; | |
39 | else | |
40 | wait_event_interruptible_timeout(*wq, | |
41 | kthread_should_stop(), | |
42 | msecs_to_jiffies(wait_ms)); | |
43 | if (kthread_should_stop()) | |
44 | break; | |
45 | ||
d6212a5f | 46 | if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { |
b59d0bae | 47 | wait_ms = increase_sleep_time(gc_th, wait_ms); |
d6212a5f CL |
48 | continue; |
49 | } | |
50 | ||
7bc09003 JK |
51 | /* |
52 | * [GC triggering condition] | |
53 | * 0. GC is not conducted currently. | |
54 | * 1. There are enough dirty segments. | |
55 | * 2. IO subsystem is idle by checking the # of writeback pages. | |
56 | * 3. IO subsystem is idle by checking the # of requests in | |
57 | * bdev's request list. | |
58 | * | |
e1c42045 | 59 | * Note) We have to avoid triggering GCs frequently. |
7bc09003 JK |
60 | * Because it is possible that some segments can be |
61 | * invalidated soon after by user update or deletion. | |
62 | * So, I'd like to wait some time to collect dirty segments. | |
63 | */ | |
64 | if (!mutex_trylock(&sbi->gc_mutex)) | |
65 | continue; | |
66 | ||
67 | if (!is_idle(sbi)) { | |
b59d0bae | 68 | wait_ms = increase_sleep_time(gc_th, wait_ms); |
7bc09003 JK |
69 | mutex_unlock(&sbi->gc_mutex); |
70 | continue; | |
71 | } | |
72 | ||
73 | if (has_enough_invalid_blocks(sbi)) | |
b59d0bae | 74 | wait_ms = decrease_sleep_time(gc_th, wait_ms); |
7bc09003 | 75 | else |
b59d0bae | 76 | wait_ms = increase_sleep_time(gc_th, wait_ms); |
7bc09003 | 77 | |
dcdfff65 | 78 | stat_inc_bggc_count(sbi); |
7bc09003 | 79 | |
43727527 JK |
80 | /* if return value is not zero, no victim was selected */ |
81 | if (f2fs_gc(sbi)) | |
b59d0bae | 82 | wait_ms = gc_th->no_gc_sleep_time; |
81eb8d6e | 83 | |
4660f9c0 JK |
84 | /* balancing f2fs's metadata periodically */ |
85 | f2fs_balance_fs_bg(sbi); | |
81eb8d6e | 86 | |
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 JK |
96 | |
97 | gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL); | |
7a267f8d NJ |
98 | if (!gc_th) { |
99 | err = -ENOMEM; | |
100 | goto out; | |
101 | } | |
7bc09003 | 102 | |
b59d0bae NJ |
103 | gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; |
104 | gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; | |
105 | gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; | |
106 | ||
d2dc095f NJ |
107 | gc_th->gc_idle = 0; |
108 | ||
7bc09003 JK |
109 | sbi->gc_thread = gc_th; |
110 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
111 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, | |
ec7b1f2d | 112 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 | 113 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
7a267f8d | 114 | err = PTR_ERR(gc_th->f2fs_gc_task); |
7bc09003 | 115 | kfree(gc_th); |
25718423 | 116 | sbi->gc_thread = NULL; |
7bc09003 | 117 | } |
7a267f8d NJ |
118 | out: |
119 | return err; | |
7bc09003 JK |
120 | } |
121 | ||
122 | void stop_gc_thread(struct f2fs_sb_info *sbi) | |
123 | { | |
124 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
125 | if (!gc_th) | |
126 | return; | |
127 | kthread_stop(gc_th->f2fs_gc_task); | |
128 | kfree(gc_th); | |
129 | sbi->gc_thread = NULL; | |
130 | } | |
131 | ||
d2dc095f | 132 | static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type) |
7bc09003 | 133 | { |
d2dc095f NJ |
134 | int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY; |
135 | ||
136 | if (gc_th && gc_th->gc_idle) { | |
137 | if (gc_th->gc_idle == 1) | |
138 | gc_mode = GC_CB; | |
139 | else if (gc_th->gc_idle == 2) | |
140 | gc_mode = GC_GREEDY; | |
141 | } | |
142 | return gc_mode; | |
7bc09003 JK |
143 | } |
144 | ||
145 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
146 | int type, struct victim_sel_policy *p) | |
147 | { | |
148 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
149 | ||
4ebefc44 | 150 | if (p->alloc_mode == SSR) { |
7bc09003 JK |
151 | p->gc_mode = GC_GREEDY; |
152 | p->dirty_segmap = dirty_i->dirty_segmap[type]; | |
a26b7c8a | 153 | p->max_search = dirty_i->nr_dirty[type]; |
7bc09003 JK |
154 | p->ofs_unit = 1; |
155 | } else { | |
d2dc095f | 156 | p->gc_mode = select_gc_type(sbi->gc_thread, gc_type); |
7bc09003 | 157 | p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; |
a26b7c8a | 158 | p->max_search = dirty_i->nr_dirty[DIRTY]; |
7bc09003 JK |
159 | p->ofs_unit = sbi->segs_per_sec; |
160 | } | |
a26b7c8a | 161 | |
b1c57c1c JK |
162 | if (p->max_search > sbi->max_victim_search) |
163 | p->max_search = sbi->max_victim_search; | |
a26b7c8a | 164 | |
7bc09003 JK |
165 | p->offset = sbi->last_victim[p->gc_mode]; |
166 | } | |
167 | ||
168 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
169 | struct victim_sel_policy *p) | |
170 | { | |
b7250d2d JK |
171 | /* SSR allocates in a segment unit */ |
172 | if (p->alloc_mode == SSR) | |
173 | return 1 << sbi->log_blocks_per_seg; | |
7bc09003 JK |
174 | if (p->gc_mode == GC_GREEDY) |
175 | return (1 << sbi->log_blocks_per_seg) * p->ofs_unit; | |
176 | else if (p->gc_mode == GC_CB) | |
177 | return UINT_MAX; | |
178 | else /* No other gc_mode */ | |
179 | return 0; | |
180 | } | |
181 | ||
182 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
183 | { | |
184 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 185 | unsigned int secno; |
7bc09003 JK |
186 | |
187 | /* | |
188 | * If the gc_type is FG_GC, we can select victim segments | |
189 | * selected by background GC before. | |
190 | * Those segments guarantee they have small valid blocks. | |
191 | */ | |
7cd8558b | 192 | for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { |
5ec4e49f | 193 | if (sec_usage_check(sbi, secno)) |
b65ee148 | 194 | continue; |
5ec4e49f JK |
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 | ||
e1c42045 | 221 | /* Handle if the system time has changed by the user */ |
7bc09003 JK |
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 | ||
a57e564d JX |
233 | static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, |
234 | unsigned int segno, struct victim_sel_policy *p) | |
7bc09003 JK |
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 | ||
210f41bc CY |
262 | mutex_lock(&dirty_i->seglist_lock); |
263 | ||
7bc09003 JK |
264 | p.alloc_mode = alloc_mode; |
265 | select_policy(sbi, gc_type, type, &p); | |
266 | ||
267 | p.min_segno = NULL_SEGNO; | |
b2b3460a | 268 | p.min_cost = max_cost = get_max_cost(sbi, &p); |
7bc09003 | 269 | |
7bc09003 JK |
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 | 279 | |
7cd8558b JK |
280 | segno = find_next_bit(p.dirty_segmap, MAIN_SEGS(sbi), p.offset); |
281 | if (segno >= MAIN_SEGS(sbi)) { | |
7bc09003 JK |
282 | if (sbi->last_victim[p.gc_mode]) { |
283 | sbi->last_victim[p.gc_mode] = 0; | |
284 | p.offset = 0; | |
285 | continue; | |
286 | } | |
287 | break; | |
288 | } | |
a57e564d JX |
289 | |
290 | p.offset = segno + p.ofs_unit; | |
291 | if (p.ofs_unit > 1) | |
292 | p.offset -= segno % p.ofs_unit; | |
293 | ||
5ec4e49f | 294 | secno = GET_SECNO(sbi, segno); |
7bc09003 | 295 | |
5ec4e49f | 296 | if (sec_usage_check(sbi, secno)) |
7bc09003 | 297 | continue; |
5ec4e49f | 298 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
7bc09003 JK |
299 | continue; |
300 | ||
301 | cost = get_gc_cost(sbi, segno, &p); | |
302 | ||
303 | if (p.min_cost > cost) { | |
304 | p.min_segno = segno; | |
305 | p.min_cost = cost; | |
a57e564d | 306 | } else if (unlikely(cost == max_cost)) { |
7bc09003 | 307 | continue; |
a57e564d | 308 | } |
7bc09003 | 309 | |
a26b7c8a | 310 | if (nsearched++ >= p.max_search) { |
7bc09003 JK |
311 | sbi->last_victim[p.gc_mode] = segno; |
312 | break; | |
313 | } | |
314 | } | |
7bc09003 | 315 | if (p.min_segno != NULL_SEGNO) { |
b2b3460a | 316 | got_it: |
7bc09003 | 317 | if (p.alloc_mode == LFS) { |
5ec4e49f JK |
318 | secno = GET_SECNO(sbi, p.min_segno); |
319 | if (gc_type == FG_GC) | |
320 | sbi->cur_victim_sec = secno; | |
321 | else | |
322 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 323 | } |
5ec4e49f | 324 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
8e46b3ed NJ |
325 | |
326 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, | |
327 | sbi->cur_victim_sec, | |
328 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 JK |
329 | } |
330 | mutex_unlock(&dirty_i->seglist_lock); | |
331 | ||
332 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
333 | } | |
334 | ||
335 | static const struct victim_selection default_v_ops = { | |
336 | .get_victim = get_victim_by_default, | |
337 | }; | |
338 | ||
7dda2af8 | 339 | static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) |
7bc09003 | 340 | { |
7bc09003 JK |
341 | struct inode_entry *ie; |
342 | ||
7dda2af8 CL |
343 | ie = radix_tree_lookup(&gc_list->iroot, ino); |
344 | if (ie) | |
345 | return ie->inode; | |
7bc09003 JK |
346 | return NULL; |
347 | } | |
348 | ||
7dda2af8 | 349 | static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) |
7bc09003 | 350 | { |
6cc4af56 GZ |
351 | struct inode_entry *new_ie; |
352 | ||
7dda2af8 | 353 | if (inode == find_gc_inode(gc_list, inode->i_ino)) { |
6cc4af56 GZ |
354 | iput(inode); |
355 | return; | |
7bc09003 | 356 | } |
06292073 | 357 | new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); |
7bc09003 | 358 | new_ie->inode = inode; |
9be32d72 | 359 | retry: |
769ec6e5 | 360 | if (radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie)) { |
9be32d72 | 361 | cond_resched(); |
7dda2af8 CL |
362 | goto retry; |
363 | } | |
364 | list_add_tail(&new_ie->list, &gc_list->ilist); | |
7bc09003 JK |
365 | } |
366 | ||
7dda2af8 | 367 | static void put_gc_inode(struct gc_inode_list *gc_list) |
7bc09003 JK |
368 | { |
369 | struct inode_entry *ie, *next_ie; | |
7dda2af8 CL |
370 | list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { |
371 | radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); | |
7bc09003 JK |
372 | iput(ie->inode); |
373 | list_del(&ie->list); | |
06292073 | 374 | kmem_cache_free(inode_entry_slab, ie); |
7bc09003 JK |
375 | } |
376 | } | |
377 | ||
378 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
379 | unsigned int segno, int offset) | |
380 | { | |
381 | struct sit_info *sit_i = SIT_I(sbi); | |
382 | struct seg_entry *sentry; | |
383 | int ret; | |
384 | ||
385 | mutex_lock(&sit_i->sentry_lock); | |
386 | sentry = get_seg_entry(sbi, segno); | |
387 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
388 | mutex_unlock(&sit_i->sentry_lock); | |
43727527 | 389 | return ret; |
7bc09003 JK |
390 | } |
391 | ||
0a8165d7 | 392 | /* |
7bc09003 JK |
393 | * This function compares node address got in summary with that in NAT. |
394 | * On validity, copy that node with cold status, otherwise (invalid node) | |
395 | * ignore that. | |
396 | */ | |
43727527 | 397 | static void gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
398 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
399 | { | |
400 | bool initial = true; | |
401 | struct f2fs_summary *entry; | |
402 | int off; | |
403 | ||
404 | next_step: | |
405 | entry = sum; | |
c718379b | 406 | |
7bc09003 JK |
407 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
408 | nid_t nid = le32_to_cpu(entry->nid); | |
409 | struct page *node_page; | |
7bc09003 | 410 | |
43727527 JK |
411 | /* stop BG_GC if there is not enough free sections. */ |
412 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
413 | return; | |
7bc09003 | 414 | |
43727527 | 415 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
416 | continue; |
417 | ||
418 | if (initial) { | |
419 | ra_node_page(sbi, nid); | |
420 | continue; | |
421 | } | |
422 | node_page = get_node_page(sbi, nid); | |
423 | if (IS_ERR(node_page)) | |
424 | continue; | |
425 | ||
9a01b56b HY |
426 | /* block may become invalid during get_node_page */ |
427 | if (check_valid_map(sbi, segno, off) == 0) { | |
428 | f2fs_put_page(node_page, 1); | |
429 | continue; | |
430 | } | |
431 | ||
7bc09003 | 432 | /* set page dirty and write it */ |
4ebefc44 | 433 | if (gc_type == FG_GC) { |
5514f0aa | 434 | f2fs_wait_on_page_writeback(node_page, NODE); |
7bc09003 | 435 | set_page_dirty(node_page); |
4ebefc44 JK |
436 | } else { |
437 | if (!PageWriteback(node_page)) | |
438 | set_page_dirty(node_page); | |
439 | } | |
7bc09003 JK |
440 | f2fs_put_page(node_page, 1); |
441 | stat_inc_node_blk_count(sbi, 1); | |
442 | } | |
c718379b | 443 | |
7bc09003 JK |
444 | if (initial) { |
445 | initial = false; | |
446 | goto next_step; | |
447 | } | |
448 | ||
449 | if (gc_type == FG_GC) { | |
450 | struct writeback_control wbc = { | |
451 | .sync_mode = WB_SYNC_ALL, | |
452 | .nr_to_write = LONG_MAX, | |
453 | .for_reclaim = 0, | |
454 | }; | |
455 | sync_node_pages(sbi, 0, &wbc); | |
4ebefc44 JK |
456 | |
457 | /* | |
458 | * In the case of FG_GC, it'd be better to reclaim this victim | |
459 | * completely. | |
460 | */ | |
461 | if (get_valid_blocks(sbi, segno, 1) != 0) | |
462 | goto next_step; | |
7bc09003 | 463 | } |
7bc09003 JK |
464 | } |
465 | ||
0a8165d7 | 466 | /* |
9af45ef5 JK |
467 | * Calculate start block index indicating the given node offset. |
468 | * Be careful, caller should give this node offset only indicating direct node | |
469 | * blocks. If any node offsets, which point the other types of node blocks such | |
470 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
471 | * bug. | |
7bc09003 | 472 | */ |
de93653f | 473 | block_t start_bidx_of_node(unsigned int node_ofs, struct f2fs_inode_info *fi) |
7bc09003 | 474 | { |
ce19a5d4 JK |
475 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
476 | unsigned int bidx; | |
7bc09003 | 477 | |
ce19a5d4 JK |
478 | if (node_ofs == 0) |
479 | return 0; | |
7bc09003 | 480 | |
ce19a5d4 | 481 | if (node_ofs <= 2) { |
7bc09003 JK |
482 | bidx = node_ofs - 1; |
483 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 484 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
485 | bidx = node_ofs - 2 - dec; |
486 | } else { | |
ce19a5d4 | 487 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
488 | bidx = node_ofs - 5 - dec; |
489 | } | |
de93653f | 490 | return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(fi); |
7bc09003 JK |
491 | } |
492 | ||
493 | static int check_dnode(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, | |
494 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) | |
495 | { | |
496 | struct page *node_page; | |
497 | nid_t nid; | |
498 | unsigned int ofs_in_node; | |
499 | block_t source_blkaddr; | |
500 | ||
501 | nid = le32_to_cpu(sum->nid); | |
502 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
503 | ||
504 | node_page = get_node_page(sbi, nid); | |
505 | if (IS_ERR(node_page)) | |
43727527 | 506 | return 0; |
7bc09003 JK |
507 | |
508 | get_node_info(sbi, nid, dni); | |
509 | ||
510 | if (sum->version != dni->version) { | |
511 | f2fs_put_page(node_page, 1); | |
43727527 | 512 | return 0; |
7bc09003 JK |
513 | } |
514 | ||
515 | *nofs = ofs_of_node(node_page); | |
516 | source_blkaddr = datablock_addr(node_page, ofs_in_node); | |
517 | f2fs_put_page(node_page, 1); | |
518 | ||
519 | if (source_blkaddr != blkaddr) | |
43727527 JK |
520 | return 0; |
521 | return 1; | |
7bc09003 JK |
522 | } |
523 | ||
524 | static void move_data_page(struct inode *inode, struct page *page, int gc_type) | |
525 | { | |
458e6197 JK |
526 | struct f2fs_io_info fio = { |
527 | .type = DATA, | |
528 | .rw = WRITE_SYNC, | |
63a0b7cb FL |
529 | }; |
530 | ||
7bc09003 | 531 | if (gc_type == BG_GC) { |
4ebefc44 JK |
532 | if (PageWriteback(page)) |
533 | goto out; | |
7bc09003 JK |
534 | set_page_dirty(page); |
535 | set_cold_data(page); | |
536 | } else { | |
5514f0aa | 537 | f2fs_wait_on_page_writeback(page, DATA); |
4ebefc44 | 538 | |
1fe54f9d | 539 | if (clear_page_dirty_for_io(page)) |
a7ffdbe2 | 540 | inode_dec_dirty_pages(inode); |
7bc09003 | 541 | set_cold_data(page); |
458e6197 | 542 | do_write_data_page(page, &fio); |
7bc09003 JK |
543 | clear_cold_data(page); |
544 | } | |
545 | out: | |
546 | f2fs_put_page(page, 1); | |
547 | } | |
548 | ||
0a8165d7 | 549 | /* |
7bc09003 JK |
550 | * This function tries to get parent node of victim data block, and identifies |
551 | * data block validity. If the block is valid, copy that with cold status and | |
552 | * modify parent node. | |
553 | * If the parent node is not valid or the data block address is different, | |
554 | * the victim data block is ignored. | |
555 | */ | |
43727527 | 556 | static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7dda2af8 | 557 | struct gc_inode_list *gc_list, unsigned int segno, int gc_type) |
7bc09003 JK |
558 | { |
559 | struct super_block *sb = sbi->sb; | |
560 | struct f2fs_summary *entry; | |
561 | block_t start_addr; | |
43727527 | 562 | int off; |
7bc09003 JK |
563 | int phase = 0; |
564 | ||
565 | start_addr = START_BLOCK(sbi, segno); | |
566 | ||
567 | next_step: | |
568 | entry = sum; | |
c718379b | 569 | |
7bc09003 JK |
570 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
571 | struct page *data_page; | |
572 | struct inode *inode; | |
573 | struct node_info dni; /* dnode info for the data */ | |
574 | unsigned int ofs_in_node, nofs; | |
575 | block_t start_bidx; | |
576 | ||
43727527 JK |
577 | /* stop BG_GC if there is not enough free sections. */ |
578 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0)) | |
579 | return; | |
7bc09003 | 580 | |
43727527 | 581 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
582 | continue; |
583 | ||
584 | if (phase == 0) { | |
585 | ra_node_page(sbi, le32_to_cpu(entry->nid)); | |
586 | continue; | |
587 | } | |
588 | ||
589 | /* Get an inode by ino with checking validity */ | |
43727527 | 590 | if (check_dnode(sbi, entry, &dni, start_addr + off, &nofs) == 0) |
7bc09003 JK |
591 | continue; |
592 | ||
593 | if (phase == 1) { | |
594 | ra_node_page(sbi, dni.ino); | |
595 | continue; | |
596 | } | |
597 | ||
7bc09003 JK |
598 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); |
599 | ||
600 | if (phase == 2) { | |
d4686d56 | 601 | inode = f2fs_iget(sb, dni.ino); |
b73e5282 | 602 | if (IS_ERR(inode) || is_bad_inode(inode)) |
7bc09003 JK |
603 | continue; |
604 | ||
de93653f JK |
605 | start_bidx = start_bidx_of_node(nofs, F2FS_I(inode)); |
606 | ||
7bc09003 | 607 | data_page = find_data_page(inode, |
c718379b | 608 | start_bidx + ofs_in_node, false); |
31a32688 CL |
609 | if (IS_ERR(data_page)) { |
610 | iput(inode); | |
611 | continue; | |
612 | } | |
7bc09003 JK |
613 | |
614 | f2fs_put_page(data_page, 0); | |
7dda2af8 | 615 | add_gc_inode(gc_list, inode); |
31a32688 CL |
616 | continue; |
617 | } | |
618 | ||
619 | /* phase 3 */ | |
7dda2af8 | 620 | inode = find_gc_inode(gc_list, dni.ino); |
31a32688 CL |
621 | if (inode) { |
622 | start_bidx = start_bidx_of_node(nofs, F2FS_I(inode)); | |
623 | data_page = get_lock_data_page(inode, | |
7bc09003 | 624 | start_bidx + ofs_in_node); |
31a32688 CL |
625 | if (IS_ERR(data_page)) |
626 | continue; | |
627 | move_data_page(inode, data_page, gc_type); | |
628 | stat_inc_data_blk_count(sbi, 1); | |
7bc09003 | 629 | } |
7bc09003 | 630 | } |
c718379b | 631 | |
7bc09003 JK |
632 | if (++phase < 4) |
633 | goto next_step; | |
43727527 | 634 | |
4ebefc44 | 635 | if (gc_type == FG_GC) { |
458e6197 | 636 | f2fs_submit_merged_bio(sbi, DATA, WRITE); |
4ebefc44 JK |
637 | |
638 | /* | |
639 | * In the case of FG_GC, it'd be better to reclaim this victim | |
640 | * completely. | |
641 | */ | |
642 | if (get_valid_blocks(sbi, segno, 1) != 0) { | |
643 | phase = 2; | |
644 | goto next_step; | |
645 | } | |
646 | } | |
7bc09003 JK |
647 | } |
648 | ||
649 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
8a2d0ace | 650 | int gc_type) |
7bc09003 JK |
651 | { |
652 | struct sit_info *sit_i = SIT_I(sbi); | |
653 | int ret; | |
8a2d0ace | 654 | |
7bc09003 | 655 | mutex_lock(&sit_i->sentry_lock); |
8a2d0ace GZ |
656 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, |
657 | NO_CHECK_TYPE, LFS); | |
7bc09003 JK |
658 | mutex_unlock(&sit_i->sentry_lock); |
659 | return ret; | |
660 | } | |
661 | ||
43727527 | 662 | static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno, |
7dda2af8 | 663 | struct gc_inode_list *gc_list, int gc_type) |
7bc09003 JK |
664 | { |
665 | struct page *sum_page; | |
666 | struct f2fs_summary_block *sum; | |
c718379b | 667 | struct blk_plug plug; |
7bc09003 JK |
668 | |
669 | /* read segment summary of victim */ | |
670 | sum_page = get_sum_page(sbi, segno); | |
7bc09003 | 671 | |
c718379b JK |
672 | blk_start_plug(&plug); |
673 | ||
7bc09003 JK |
674 | sum = page_address(sum_page); |
675 | ||
676 | switch (GET_SUM_TYPE((&sum->footer))) { | |
677 | case SUM_TYPE_NODE: | |
43727527 | 678 | gc_node_segment(sbi, sum->entries, segno, gc_type); |
7bc09003 JK |
679 | break; |
680 | case SUM_TYPE_DATA: | |
7dda2af8 | 681 | gc_data_segment(sbi, sum->entries, gc_list, segno, gc_type); |
7bc09003 JK |
682 | break; |
683 | } | |
c718379b JK |
684 | blk_finish_plug(&plug); |
685 | ||
7bc09003 JK |
686 | stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer))); |
687 | stat_inc_call_count(sbi->stat_info); | |
688 | ||
b7473754 | 689 | f2fs_put_page(sum_page, 1); |
7bc09003 JK |
690 | } |
691 | ||
408e9375 | 692 | int f2fs_gc(struct f2fs_sb_info *sbi) |
7bc09003 | 693 | { |
408e9375 | 694 | unsigned int segno, i; |
7bc09003 | 695 | int gc_type = BG_GC; |
43727527 JK |
696 | int nfree = 0; |
697 | int ret = -1; | |
d5053a34 | 698 | struct cp_control cpc; |
7dda2af8 CL |
699 | struct gc_inode_list gc_list = { |
700 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
769ec6e5 | 701 | .iroot = RADIX_TREE_INIT(GFP_NOFS), |
7dda2af8 | 702 | }; |
d5053a34 JK |
703 | |
704 | cpc.reason = test_opt(sbi, FASTBOOT) ? CP_UMOUNT : CP_SYNC; | |
7bc09003 | 705 | |
7bc09003 | 706 | gc_more: |
6bacf52f | 707 | if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE))) |
408e9375 | 708 | goto stop; |
1e968fdf | 709 | if (unlikely(f2fs_cp_error(sbi))) |
203681f6 | 710 | goto stop; |
7bc09003 | 711 | |
d64f8047 | 712 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, nfree)) { |
408e9375 | 713 | gc_type = FG_GC; |
75ab4cb8 | 714 | write_checkpoint(sbi, &cpc); |
d64f8047 | 715 | } |
7bc09003 | 716 | |
8a2d0ace | 717 | if (!__get_victim(sbi, &segno, gc_type)) |
408e9375 | 718 | goto stop; |
43727527 | 719 | ret = 0; |
7bc09003 | 720 | |
81c1a0f1 CY |
721 | /* readahead multi ssa blocks those have contiguous address */ |
722 | if (sbi->segs_per_sec > 1) | |
723 | ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), sbi->segs_per_sec, | |
724 | META_SSA); | |
725 | ||
43727527 | 726 | for (i = 0; i < sbi->segs_per_sec; i++) |
7dda2af8 | 727 | do_garbage_collect(sbi, segno + i, &gc_list, gc_type); |
43727527 | 728 | |
5ec4e49f JK |
729 | if (gc_type == FG_GC) { |
730 | sbi->cur_victim_sec = NULL_SEGNO; | |
43727527 | 731 | nfree++; |
5ec4e49f JK |
732 | WARN_ON(get_valid_blocks(sbi, segno, sbi->segs_per_sec)); |
733 | } | |
43727527 JK |
734 | |
735 | if (has_not_enough_free_secs(sbi, nfree)) | |
736 | goto gc_more; | |
737 | ||
738 | if (gc_type == FG_GC) | |
75ab4cb8 | 739 | write_checkpoint(sbi, &cpc); |
408e9375 | 740 | stop: |
7bc09003 JK |
741 | mutex_unlock(&sbi->gc_mutex); |
742 | ||
7dda2af8 | 743 | put_gc_inode(&gc_list); |
43727527 | 744 | return ret; |
7bc09003 JK |
745 | } |
746 | ||
747 | void build_gc_manager(struct f2fs_sb_info *sbi) | |
748 | { | |
749 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
750 | } |