]>
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> | |
7bc09003 JK |
19 | |
20 | #include "f2fs.h" | |
21 | #include "node.h" | |
22 | #include "segment.h" | |
23 | #include "gc.h" | |
8e46b3ed | 24 | #include <trace/events/f2fs.h> |
7bc09003 | 25 | |
7bc09003 JK |
26 | static int gc_thread_func(void *data) |
27 | { | |
28 | struct f2fs_sb_info *sbi = data; | |
b59d0bae | 29 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; |
7bc09003 JK |
30 | wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; |
31 | long wait_ms; | |
32 | ||
b59d0bae | 33 | wait_ms = gc_th->min_sleep_time; |
7bc09003 | 34 | |
1d7be270 | 35 | set_freezable(); |
7bc09003 | 36 | do { |
1d7be270 JK |
37 | wait_event_interruptible_timeout(*wq, |
38 | kthread_should_stop() || freezing(current), | |
39 | msecs_to_jiffies(wait_ms)); | |
40 | ||
7bc09003 JK |
41 | if (try_to_freeze()) |
42 | continue; | |
7bc09003 JK |
43 | if (kthread_should_stop()) |
44 | break; | |
45 | ||
d6212a5f | 46 | if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { |
88dd8934 | 47 | increase_sleep_time(gc_th, &wait_ms); |
d6212a5f CL |
48 | continue; |
49 | } | |
50 | ||
0f348028 | 51 | #ifdef CONFIG_F2FS_FAULT_INJECTION |
55523519 CY |
52 | if (time_to_inject(sbi, FAULT_CHECKPOINT)) { |
53 | f2fs_show_injection_info(FAULT_CHECKPOINT); | |
0f348028 | 54 | f2fs_stop_checkpoint(sbi, false); |
55523519 | 55 | } |
0f348028 CY |
56 | #endif |
57 | ||
7bc09003 JK |
58 | /* |
59 | * [GC triggering condition] | |
60 | * 0. GC is not conducted currently. | |
61 | * 1. There are enough dirty segments. | |
62 | * 2. IO subsystem is idle by checking the # of writeback pages. | |
63 | * 3. IO subsystem is idle by checking the # of requests in | |
64 | * bdev's request list. | |
65 | * | |
e1c42045 | 66 | * Note) We have to avoid triggering GCs frequently. |
7bc09003 JK |
67 | * Because it is possible that some segments can be |
68 | * invalidated soon after by user update or deletion. | |
69 | * So, I'd like to wait some time to collect dirty segments. | |
70 | */ | |
71 | if (!mutex_trylock(&sbi->gc_mutex)) | |
72 | continue; | |
73 | ||
74 | if (!is_idle(sbi)) { | |
88dd8934 | 75 | increase_sleep_time(gc_th, &wait_ms); |
7bc09003 JK |
76 | mutex_unlock(&sbi->gc_mutex); |
77 | continue; | |
78 | } | |
79 | ||
80 | if (has_enough_invalid_blocks(sbi)) | |
88dd8934 | 81 | decrease_sleep_time(gc_th, &wait_ms); |
7bc09003 | 82 | else |
88dd8934 | 83 | increase_sleep_time(gc_th, &wait_ms); |
7bc09003 | 84 | |
dcdfff65 | 85 | stat_inc_bggc_count(sbi); |
7bc09003 | 86 | |
43727527 | 87 | /* if return value is not zero, no victim was selected */ |
e066b83c | 88 | if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true, NULL_SEGNO)) |
b59d0bae | 89 | wait_ms = gc_th->no_gc_sleep_time; |
81eb8d6e | 90 | |
84e4214f JK |
91 | trace_f2fs_background_gc(sbi->sb, wait_ms, |
92 | prefree_segments(sbi), free_segments(sbi)); | |
93 | ||
4660f9c0 JK |
94 | /* balancing f2fs's metadata periodically */ |
95 | f2fs_balance_fs_bg(sbi); | |
81eb8d6e | 96 | |
7bc09003 JK |
97 | } while (!kthread_should_stop()); |
98 | return 0; | |
99 | } | |
100 | ||
101 | int start_gc_thread(struct f2fs_sb_info *sbi) | |
102 | { | |
1042d60f | 103 | struct f2fs_gc_kthread *gc_th; |
ec7b1f2d | 104 | dev_t dev = sbi->sb->s_bdev->bd_dev; |
7a267f8d | 105 | int err = 0; |
7bc09003 | 106 | |
1ecc0c5c | 107 | gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL); |
7a267f8d NJ |
108 | if (!gc_th) { |
109 | err = -ENOMEM; | |
110 | goto out; | |
111 | } | |
7bc09003 | 112 | |
b59d0bae NJ |
113 | gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; |
114 | gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; | |
115 | gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; | |
116 | ||
d2dc095f NJ |
117 | gc_th->gc_idle = 0; |
118 | ||
7bc09003 JK |
119 | sbi->gc_thread = gc_th; |
120 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
121 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, | |
ec7b1f2d | 122 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 | 123 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
7a267f8d | 124 | err = PTR_ERR(gc_th->f2fs_gc_task); |
7bc09003 | 125 | kfree(gc_th); |
25718423 | 126 | sbi->gc_thread = NULL; |
7bc09003 | 127 | } |
7a267f8d NJ |
128 | out: |
129 | return err; | |
7bc09003 JK |
130 | } |
131 | ||
132 | void stop_gc_thread(struct f2fs_sb_info *sbi) | |
133 | { | |
134 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
135 | if (!gc_th) | |
136 | return; | |
137 | kthread_stop(gc_th->f2fs_gc_task); | |
138 | kfree(gc_th); | |
139 | sbi->gc_thread = NULL; | |
140 | } | |
141 | ||
d2dc095f | 142 | static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type) |
7bc09003 | 143 | { |
d2dc095f NJ |
144 | int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY; |
145 | ||
146 | if (gc_th && gc_th->gc_idle) { | |
147 | if (gc_th->gc_idle == 1) | |
148 | gc_mode = GC_CB; | |
149 | else if (gc_th->gc_idle == 2) | |
150 | gc_mode = GC_GREEDY; | |
151 | } | |
152 | return gc_mode; | |
7bc09003 JK |
153 | } |
154 | ||
155 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
156 | int type, struct victim_sel_policy *p) | |
157 | { | |
158 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
159 | ||
4ebefc44 | 160 | if (p->alloc_mode == SSR) { |
7bc09003 JK |
161 | p->gc_mode = GC_GREEDY; |
162 | p->dirty_segmap = dirty_i->dirty_segmap[type]; | |
a26b7c8a | 163 | p->max_search = dirty_i->nr_dirty[type]; |
7bc09003 JK |
164 | p->ofs_unit = 1; |
165 | } else { | |
d2dc095f | 166 | p->gc_mode = select_gc_type(sbi->gc_thread, gc_type); |
7bc09003 | 167 | p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; |
a26b7c8a | 168 | p->max_search = dirty_i->nr_dirty[DIRTY]; |
7bc09003 JK |
169 | p->ofs_unit = sbi->segs_per_sec; |
170 | } | |
a26b7c8a | 171 | |
e93b9865 HP |
172 | /* we need to check every dirty segments in the FG_GC case */ |
173 | if (gc_type != FG_GC && p->max_search > sbi->max_victim_search) | |
b1c57c1c | 174 | p->max_search = sbi->max_victim_search; |
a26b7c8a | 175 | |
7a20b8a6 JK |
176 | /* let's select beginning hot/small space first */ |
177 | if (type == CURSEG_HOT_DATA || IS_NODESEG(type)) | |
178 | p->offset = 0; | |
179 | else | |
e066b83c | 180 | p->offset = SIT_I(sbi)->last_victim[p->gc_mode]; |
7bc09003 JK |
181 | } |
182 | ||
183 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
184 | struct victim_sel_policy *p) | |
185 | { | |
b7250d2d JK |
186 | /* SSR allocates in a segment unit */ |
187 | if (p->alloc_mode == SSR) | |
3519e3f9 | 188 | return sbi->blocks_per_seg; |
7bc09003 | 189 | if (p->gc_mode == GC_GREEDY) |
c541a51b | 190 | return 2 * sbi->blocks_per_seg * p->ofs_unit; |
7bc09003 JK |
191 | else if (p->gc_mode == GC_CB) |
192 | return UINT_MAX; | |
193 | else /* No other gc_mode */ | |
194 | return 0; | |
195 | } | |
196 | ||
197 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
198 | { | |
199 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 200 | unsigned int secno; |
7bc09003 JK |
201 | |
202 | /* | |
203 | * If the gc_type is FG_GC, we can select victim segments | |
204 | * selected by background GC before. | |
205 | * Those segments guarantee they have small valid blocks. | |
206 | */ | |
7cd8558b | 207 | for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { |
5ec4e49f | 208 | if (sec_usage_check(sbi, secno)) |
b65ee148 | 209 | continue; |
e93b9865 HP |
210 | |
211 | if (no_fggc_candidate(sbi, secno)) | |
212 | continue; | |
213 | ||
5ec4e49f | 214 | clear_bit(secno, dirty_i->victim_secmap); |
4ddb1a4d | 215 | return GET_SEG_FROM_SEC(sbi, secno); |
7bc09003 JK |
216 | } |
217 | return NULL_SEGNO; | |
218 | } | |
219 | ||
220 | static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) | |
221 | { | |
222 | struct sit_info *sit_i = SIT_I(sbi); | |
4ddb1a4d JK |
223 | unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); |
224 | unsigned int start = GET_SEG_FROM_SEC(sbi, secno); | |
7bc09003 JK |
225 | unsigned long long mtime = 0; |
226 | unsigned int vblocks; | |
227 | unsigned char age = 0; | |
228 | unsigned char u; | |
229 | unsigned int i; | |
230 | ||
231 | for (i = 0; i < sbi->segs_per_sec; i++) | |
232 | mtime += get_seg_entry(sbi, start + i)->mtime; | |
302bd348 | 233 | vblocks = get_valid_blocks(sbi, segno, true); |
7bc09003 JK |
234 | |
235 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
236 | vblocks = div_u64(vblocks, sbi->segs_per_sec); | |
237 | ||
238 | u = (vblocks * 100) >> sbi->log_blocks_per_seg; | |
239 | ||
e1c42045 | 240 | /* Handle if the system time has changed by the user */ |
7bc09003 JK |
241 | if (mtime < sit_i->min_mtime) |
242 | sit_i->min_mtime = mtime; | |
243 | if (mtime > sit_i->max_mtime) | |
244 | sit_i->max_mtime = mtime; | |
245 | if (sit_i->max_mtime != sit_i->min_mtime) | |
246 | age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), | |
247 | sit_i->max_mtime - sit_i->min_mtime); | |
248 | ||
249 | return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); | |
250 | } | |
251 | ||
b9cd2061 HP |
252 | static unsigned int get_greedy_cost(struct f2fs_sb_info *sbi, |
253 | unsigned int segno) | |
254 | { | |
255 | unsigned int valid_blocks = | |
302bd348 | 256 | get_valid_blocks(sbi, segno, true); |
b9cd2061 HP |
257 | |
258 | return IS_DATASEG(get_seg_entry(sbi, segno)->type) ? | |
259 | valid_blocks * 2 : valid_blocks; | |
260 | } | |
261 | ||
b7b7c4cf YH |
262 | static unsigned int get_ssr_cost(struct f2fs_sb_info *sbi, |
263 | unsigned int segno) | |
264 | { | |
265 | struct seg_entry *se = get_seg_entry(sbi, segno); | |
266 | ||
267 | return se->ckpt_valid_blocks > se->valid_blocks ? | |
268 | se->ckpt_valid_blocks : se->valid_blocks; | |
269 | } | |
270 | ||
a57e564d JX |
271 | static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, |
272 | unsigned int segno, struct victim_sel_policy *p) | |
7bc09003 JK |
273 | { |
274 | if (p->alloc_mode == SSR) | |
b7b7c4cf | 275 | return get_ssr_cost(sbi, segno); |
7bc09003 JK |
276 | |
277 | /* alloc_mode == LFS */ | |
278 | if (p->gc_mode == GC_GREEDY) | |
b9cd2061 | 279 | return get_greedy_cost(sbi, segno); |
7bc09003 JK |
280 | else |
281 | return get_cb_cost(sbi, segno); | |
282 | } | |
283 | ||
688159b6 FL |
284 | static unsigned int count_bits(const unsigned long *addr, |
285 | unsigned int offset, unsigned int len) | |
286 | { | |
287 | unsigned int end = offset + len, sum = 0; | |
288 | ||
289 | while (offset < end) { | |
290 | if (test_bit(offset++, addr)) | |
291 | ++sum; | |
292 | } | |
293 | return sum; | |
294 | } | |
295 | ||
0a8165d7 | 296 | /* |
111d2495 | 297 | * This function is called from two paths. |
7bc09003 JK |
298 | * One is garbage collection and the other is SSR segment selection. |
299 | * When it is called during GC, it just gets a victim segment | |
300 | * and it does not remove it from dirty seglist. | |
301 | * When it is called from SSR segment selection, it finds a segment | |
302 | * which has minimum valid blocks and removes it from dirty seglist. | |
303 | */ | |
304 | static int get_victim_by_default(struct f2fs_sb_info *sbi, | |
305 | unsigned int *result, int gc_type, int type, char alloc_mode) | |
306 | { | |
307 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
e066b83c | 308 | struct sit_info *sm = SIT_I(sbi); |
7bc09003 | 309 | struct victim_sel_policy p; |
3fa56503 | 310 | unsigned int secno, last_victim; |
a43f7ec3 | 311 | unsigned int last_segment = MAIN_SEGS(sbi); |
688159b6 | 312 | unsigned int nsearched = 0; |
7bc09003 | 313 | |
210f41bc CY |
314 | mutex_lock(&dirty_i->seglist_lock); |
315 | ||
7bc09003 JK |
316 | p.alloc_mode = alloc_mode; |
317 | select_policy(sbi, gc_type, type, &p); | |
318 | ||
319 | p.min_segno = NULL_SEGNO; | |
3fa56503 | 320 | p.min_cost = get_max_cost(sbi, &p); |
7bc09003 | 321 | |
e066b83c JK |
322 | if (*result != NULL_SEGNO) { |
323 | if (IS_DATASEG(get_seg_entry(sbi, *result)->type) && | |
324 | get_valid_blocks(sbi, *result, false) && | |
325 | !sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result))) | |
326 | p.min_segno = *result; | |
327 | goto out; | |
328 | } | |
329 | ||
3342bb30 CY |
330 | if (p.max_search == 0) |
331 | goto out; | |
332 | ||
e066b83c | 333 | last_victim = sm->last_victim[p.gc_mode]; |
7bc09003 JK |
334 | if (p.alloc_mode == LFS && gc_type == FG_GC) { |
335 | p.min_segno = check_bg_victims(sbi); | |
336 | if (p.min_segno != NULL_SEGNO) | |
337 | goto got_it; | |
338 | } | |
339 | ||
340 | while (1) { | |
341 | unsigned long cost; | |
5ec4e49f | 342 | unsigned int segno; |
7bc09003 | 343 | |
a43f7ec3 CY |
344 | segno = find_next_bit(p.dirty_segmap, last_segment, p.offset); |
345 | if (segno >= last_segment) { | |
e066b83c JK |
346 | if (sm->last_victim[p.gc_mode]) { |
347 | last_segment = | |
348 | sm->last_victim[p.gc_mode]; | |
349 | sm->last_victim[p.gc_mode] = 0; | |
7bc09003 JK |
350 | p.offset = 0; |
351 | continue; | |
352 | } | |
353 | break; | |
354 | } | |
a57e564d JX |
355 | |
356 | p.offset = segno + p.ofs_unit; | |
688159b6 | 357 | if (p.ofs_unit > 1) { |
a57e564d | 358 | p.offset -= segno % p.ofs_unit; |
688159b6 FL |
359 | nsearched += count_bits(p.dirty_segmap, |
360 | p.offset - p.ofs_unit, | |
361 | p.ofs_unit); | |
362 | } else { | |
363 | nsearched++; | |
364 | } | |
365 | ||
4ddb1a4d | 366 | secno = GET_SEC_FROM_SEG(sbi, segno); |
7bc09003 | 367 | |
5ec4e49f | 368 | if (sec_usage_check(sbi, secno)) |
688159b6 | 369 | goto next; |
5ec4e49f | 370 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
688159b6 | 371 | goto next; |
e93b9865 HP |
372 | if (gc_type == FG_GC && p.alloc_mode == LFS && |
373 | no_fggc_candidate(sbi, secno)) | |
374 | goto next; | |
7bc09003 JK |
375 | |
376 | cost = get_gc_cost(sbi, segno, &p); | |
377 | ||
378 | if (p.min_cost > cost) { | |
379 | p.min_segno = segno; | |
380 | p.min_cost = cost; | |
a57e564d | 381 | } |
688159b6 FL |
382 | next: |
383 | if (nsearched >= p.max_search) { | |
e066b83c JK |
384 | if (!sm->last_victim[p.gc_mode] && segno <= last_victim) |
385 | sm->last_victim[p.gc_mode] = last_victim + 1; | |
4ce53776 | 386 | else |
e066b83c JK |
387 | sm->last_victim[p.gc_mode] = segno + 1; |
388 | sm->last_victim[p.gc_mode] %= MAIN_SEGS(sbi); | |
7bc09003 JK |
389 | break; |
390 | } | |
391 | } | |
7bc09003 | 392 | if (p.min_segno != NULL_SEGNO) { |
b2b3460a | 393 | got_it: |
7bc09003 | 394 | if (p.alloc_mode == LFS) { |
4ddb1a4d | 395 | secno = GET_SEC_FROM_SEG(sbi, p.min_segno); |
5ec4e49f JK |
396 | if (gc_type == FG_GC) |
397 | sbi->cur_victim_sec = secno; | |
398 | else | |
399 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 400 | } |
5ec4e49f | 401 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
8e46b3ed NJ |
402 | |
403 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, | |
404 | sbi->cur_victim_sec, | |
405 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 | 406 | } |
3342bb30 | 407 | out: |
7bc09003 JK |
408 | mutex_unlock(&dirty_i->seglist_lock); |
409 | ||
410 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
411 | } | |
412 | ||
413 | static const struct victim_selection default_v_ops = { | |
414 | .get_victim = get_victim_by_default, | |
415 | }; | |
416 | ||
7dda2af8 | 417 | static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) |
7bc09003 | 418 | { |
7bc09003 JK |
419 | struct inode_entry *ie; |
420 | ||
7dda2af8 CL |
421 | ie = radix_tree_lookup(&gc_list->iroot, ino); |
422 | if (ie) | |
423 | return ie->inode; | |
7bc09003 JK |
424 | return NULL; |
425 | } | |
426 | ||
7dda2af8 | 427 | static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) |
7bc09003 | 428 | { |
6cc4af56 GZ |
429 | struct inode_entry *new_ie; |
430 | ||
7dda2af8 | 431 | if (inode == find_gc_inode(gc_list, inode->i_ino)) { |
6cc4af56 GZ |
432 | iput(inode); |
433 | return; | |
7bc09003 | 434 | } |
06292073 | 435 | new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); |
7bc09003 | 436 | new_ie->inode = inode; |
f28e5034 CY |
437 | |
438 | f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); | |
7dda2af8 | 439 | list_add_tail(&new_ie->list, &gc_list->ilist); |
7bc09003 JK |
440 | } |
441 | ||
7dda2af8 | 442 | static void put_gc_inode(struct gc_inode_list *gc_list) |
7bc09003 JK |
443 | { |
444 | struct inode_entry *ie, *next_ie; | |
7dda2af8 CL |
445 | list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { |
446 | radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); | |
7bc09003 JK |
447 | iput(ie->inode); |
448 | list_del(&ie->list); | |
06292073 | 449 | kmem_cache_free(inode_entry_slab, ie); |
7bc09003 JK |
450 | } |
451 | } | |
452 | ||
453 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
454 | unsigned int segno, int offset) | |
455 | { | |
456 | struct sit_info *sit_i = SIT_I(sbi); | |
457 | struct seg_entry *sentry; | |
458 | int ret; | |
459 | ||
460 | mutex_lock(&sit_i->sentry_lock); | |
461 | sentry = get_seg_entry(sbi, segno); | |
462 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
463 | mutex_unlock(&sit_i->sentry_lock); | |
43727527 | 464 | return ret; |
7bc09003 JK |
465 | } |
466 | ||
0a8165d7 | 467 | /* |
7bc09003 JK |
468 | * This function compares node address got in summary with that in NAT. |
469 | * On validity, copy that node with cold status, otherwise (invalid node) | |
470 | * ignore that. | |
471 | */ | |
718e53fa | 472 | static void gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
473 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
474 | { | |
7bc09003 | 475 | struct f2fs_summary *entry; |
26d58599 | 476 | block_t start_addr; |
7bc09003 | 477 | int off; |
7ea984b0 | 478 | int phase = 0; |
7bc09003 | 479 | |
26d58599 JK |
480 | start_addr = START_BLOCK(sbi, segno); |
481 | ||
7bc09003 JK |
482 | next_step: |
483 | entry = sum; | |
c718379b | 484 | |
7bc09003 JK |
485 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
486 | nid_t nid = le32_to_cpu(entry->nid); | |
487 | struct page *node_page; | |
26d58599 | 488 | struct node_info ni; |
7bc09003 | 489 | |
43727527 | 490 | /* stop BG_GC if there is not enough free sections. */ |
7f3037a5 | 491 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) |
718e53fa | 492 | return; |
7bc09003 | 493 | |
43727527 | 494 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
495 | continue; |
496 | ||
7ea984b0 CY |
497 | if (phase == 0) { |
498 | ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, | |
499 | META_NAT, true); | |
500 | continue; | |
501 | } | |
502 | ||
503 | if (phase == 1) { | |
7bc09003 JK |
504 | ra_node_page(sbi, nid); |
505 | continue; | |
506 | } | |
7ea984b0 CY |
507 | |
508 | /* phase == 2 */ | |
7bc09003 JK |
509 | node_page = get_node_page(sbi, nid); |
510 | if (IS_ERR(node_page)) | |
511 | continue; | |
512 | ||
9a01b56b HY |
513 | /* block may become invalid during get_node_page */ |
514 | if (check_valid_map(sbi, segno, off) == 0) { | |
515 | f2fs_put_page(node_page, 1); | |
516 | continue; | |
26d58599 JK |
517 | } |
518 | ||
519 | get_node_info(sbi, nid, &ni); | |
520 | if (ni.blk_addr != start_addr + off) { | |
521 | f2fs_put_page(node_page, 1); | |
522 | continue; | |
9a01b56b HY |
523 | } |
524 | ||
da011cc0 | 525 | move_node_page(node_page, gc_type); |
e1235983 | 526 | stat_inc_node_blk_count(sbi, 1, gc_type); |
7bc09003 | 527 | } |
c718379b | 528 | |
7ea984b0 | 529 | if (++phase < 3) |
7bc09003 | 530 | goto next_step; |
7bc09003 JK |
531 | } |
532 | ||
0a8165d7 | 533 | /* |
9af45ef5 JK |
534 | * Calculate start block index indicating the given node offset. |
535 | * Be careful, caller should give this node offset only indicating direct node | |
536 | * blocks. If any node offsets, which point the other types of node blocks such | |
537 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
538 | * bug. | |
7bc09003 | 539 | */ |
81ca7350 | 540 | block_t start_bidx_of_node(unsigned int node_ofs, struct inode *inode) |
7bc09003 | 541 | { |
ce19a5d4 JK |
542 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
543 | unsigned int bidx; | |
7bc09003 | 544 | |
ce19a5d4 JK |
545 | if (node_ofs == 0) |
546 | return 0; | |
7bc09003 | 547 | |
ce19a5d4 | 548 | if (node_ofs <= 2) { |
7bc09003 JK |
549 | bidx = node_ofs - 1; |
550 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 551 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
552 | bidx = node_ofs - 2 - dec; |
553 | } else { | |
ce19a5d4 | 554 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
555 | bidx = node_ofs - 5 - dec; |
556 | } | |
81ca7350 | 557 | return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode); |
7bc09003 JK |
558 | } |
559 | ||
c1079892 | 560 | static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
561 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) |
562 | { | |
563 | struct page *node_page; | |
564 | nid_t nid; | |
565 | unsigned int ofs_in_node; | |
566 | block_t source_blkaddr; | |
567 | ||
568 | nid = le32_to_cpu(sum->nid); | |
569 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
570 | ||
571 | node_page = get_node_page(sbi, nid); | |
572 | if (IS_ERR(node_page)) | |
c1079892 | 573 | return false; |
7bc09003 JK |
574 | |
575 | get_node_info(sbi, nid, dni); | |
576 | ||
577 | if (sum->version != dni->version) { | |
c13ff37e JK |
578 | f2fs_msg(sbi->sb, KERN_WARNING, |
579 | "%s: valid data with mismatched node version.", | |
580 | __func__); | |
581 | set_sbi_flag(sbi, SBI_NEED_FSCK); | |
7bc09003 JK |
582 | } |
583 | ||
584 | *nofs = ofs_of_node(node_page); | |
585 | source_blkaddr = datablock_addr(node_page, ofs_in_node); | |
586 | f2fs_put_page(node_page, 1); | |
587 | ||
588 | if (source_blkaddr != blkaddr) | |
c1079892 NK |
589 | return false; |
590 | return true; | |
7bc09003 JK |
591 | } |
592 | ||
20614711 YH |
593 | static void move_encrypted_block(struct inode *inode, block_t bidx, |
594 | unsigned int segno, int off) | |
4375a336 JK |
595 | { |
596 | struct f2fs_io_info fio = { | |
597 | .sbi = F2FS_I_SB(inode), | |
598 | .type = DATA, | |
a912b54d | 599 | .temp = COLD, |
04d328de | 600 | .op = REQ_OP_READ, |
70fd7614 | 601 | .op_flags = 0, |
4375a336 | 602 | .encrypted_page = NULL, |
fb830fc5 | 603 | .in_list = false, |
4375a336 JK |
604 | }; |
605 | struct dnode_of_data dn; | |
606 | struct f2fs_summary sum; | |
607 | struct node_info ni; | |
608 | struct page *page; | |
4356e48e | 609 | block_t newaddr; |
4375a336 JK |
610 | int err; |
611 | ||
612 | /* do not read out */ | |
a56c7c6f | 613 | page = f2fs_grab_cache_page(inode->i_mapping, bidx, false); |
4375a336 JK |
614 | if (!page) |
615 | return; | |
616 | ||
20614711 YH |
617 | if (!check_valid_map(F2FS_I_SB(inode), segno, off)) |
618 | goto out; | |
619 | ||
5fe45743 CY |
620 | if (f2fs_is_atomic_file(inode)) |
621 | goto out; | |
622 | ||
4375a336 JK |
623 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
624 | err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE); | |
625 | if (err) | |
626 | goto out; | |
627 | ||
08b39fbd CY |
628 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) { |
629 | ClearPageUptodate(page); | |
4375a336 | 630 | goto put_out; |
08b39fbd CY |
631 | } |
632 | ||
633 | /* | |
634 | * don't cache encrypted data into meta inode until previous dirty | |
635 | * data were writebacked to avoid racing between GC and flush. | |
636 | */ | |
fec1d657 | 637 | f2fs_wait_on_page_writeback(page, DATA, true); |
4375a336 JK |
638 | |
639 | get_node_info(fio.sbi, dn.nid, &ni); | |
640 | set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); | |
641 | ||
642 | /* read page */ | |
643 | fio.page = page; | |
7a9d7548 | 644 | fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; |
4375a336 | 645 | |
4356e48e | 646 | allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr, |
fb830fc5 | 647 | &sum, CURSEG_COLD_DATA, NULL, false); |
4356e48e CY |
648 | |
649 | fio.encrypted_page = pagecache_get_page(META_MAPPING(fio.sbi), newaddr, | |
650 | FGP_LOCK | FGP_CREAT, GFP_NOFS); | |
651 | if (!fio.encrypted_page) { | |
652 | err = -ENOMEM; | |
653 | goto recover_block; | |
654 | } | |
4375a336 | 655 | |
548aedac JK |
656 | err = f2fs_submit_page_bio(&fio); |
657 | if (err) | |
658 | goto put_page_out; | |
659 | ||
660 | /* write page */ | |
661 | lock_page(fio.encrypted_page); | |
662 | ||
1563ac75 | 663 | if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) { |
4356e48e | 664 | err = -EIO; |
548aedac | 665 | goto put_page_out; |
4356e48e | 666 | } |
1563ac75 | 667 | if (unlikely(!PageUptodate(fio.encrypted_page))) { |
4356e48e | 668 | err = -EIO; |
548aedac | 669 | goto put_page_out; |
4356e48e | 670 | } |
548aedac | 671 | |
6282adbf | 672 | set_page_dirty(fio.encrypted_page); |
fec1d657 | 673 | f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true); |
6282adbf JK |
674 | if (clear_page_dirty_for_io(fio.encrypted_page)) |
675 | dec_page_count(fio.sbi, F2FS_DIRTY_META); | |
676 | ||
548aedac | 677 | set_page_writeback(fio.encrypted_page); |
4375a336 JK |
678 | |
679 | /* allocate block address */ | |
fec1d657 | 680 | f2fs_wait_on_page_writeback(dn.node_page, NODE, true); |
4356e48e | 681 | |
04d328de | 682 | fio.op = REQ_OP_WRITE; |
70fd7614 | 683 | fio.op_flags = REQ_SYNC; |
4356e48e | 684 | fio.new_blkaddr = newaddr; |
b9109b0e | 685 | f2fs_submit_page_write(&fio); |
4375a336 | 686 | |
f28b3434 | 687 | f2fs_update_data_blkaddr(&dn, newaddr); |
91942321 | 688 | set_inode_flag(inode, FI_APPEND_WRITE); |
4375a336 | 689 | if (page->index == 0) |
91942321 | 690 | set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); |
548aedac | 691 | put_page_out: |
4375a336 | 692 | f2fs_put_page(fio.encrypted_page, 1); |
4356e48e CY |
693 | recover_block: |
694 | if (err) | |
695 | __f2fs_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr, | |
696 | true, true); | |
4375a336 JK |
697 | put_out: |
698 | f2fs_put_dnode(&dn); | |
699 | out: | |
700 | f2fs_put_page(page, 1); | |
701 | } | |
702 | ||
20614711 YH |
703 | static void move_data_page(struct inode *inode, block_t bidx, int gc_type, |
704 | unsigned int segno, int off) | |
7bc09003 | 705 | { |
c879f90d JK |
706 | struct page *page; |
707 | ||
a56c7c6f | 708 | page = get_lock_data_page(inode, bidx, true); |
c879f90d JK |
709 | if (IS_ERR(page)) |
710 | return; | |
63a0b7cb | 711 | |
20614711 YH |
712 | if (!check_valid_map(F2FS_I_SB(inode), segno, off)) |
713 | goto out; | |
714 | ||
5fe45743 CY |
715 | if (f2fs_is_atomic_file(inode)) |
716 | goto out; | |
717 | ||
7bc09003 | 718 | if (gc_type == BG_GC) { |
4ebefc44 JK |
719 | if (PageWriteback(page)) |
720 | goto out; | |
7bc09003 JK |
721 | set_page_dirty(page); |
722 | set_cold_data(page); | |
723 | } else { | |
c879f90d JK |
724 | struct f2fs_io_info fio = { |
725 | .sbi = F2FS_I_SB(inode), | |
726 | .type = DATA, | |
a912b54d | 727 | .temp = COLD, |
04d328de | 728 | .op = REQ_OP_WRITE, |
70fd7614 | 729 | .op_flags = REQ_SYNC, |
e959c8f5 | 730 | .old_blkaddr = NULL_ADDR, |
c879f90d | 731 | .page = page, |
4375a336 | 732 | .encrypted_page = NULL, |
cc15620b | 733 | .need_lock = LOCK_REQ, |
c879f90d | 734 | }; |
72e1c797 CY |
735 | bool is_dirty = PageDirty(page); |
736 | int err; | |
737 | ||
738 | retry: | |
6282adbf | 739 | set_page_dirty(page); |
fec1d657 | 740 | f2fs_wait_on_page_writeback(page, DATA, true); |
933439c8 | 741 | if (clear_page_dirty_for_io(page)) { |
a7ffdbe2 | 742 | inode_dec_dirty_pages(inode); |
933439c8 CY |
743 | remove_dirty_inode(inode); |
744 | } | |
72e1c797 | 745 | |
7bc09003 | 746 | set_cold_data(page); |
72e1c797 CY |
747 | |
748 | err = do_write_data_page(&fio); | |
749 | if (err == -ENOMEM && is_dirty) { | |
750 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
751 | goto retry; | |
752 | } | |
7bc09003 JK |
753 | } |
754 | out: | |
755 | f2fs_put_page(page, 1); | |
756 | } | |
757 | ||
0a8165d7 | 758 | /* |
7bc09003 JK |
759 | * This function tries to get parent node of victim data block, and identifies |
760 | * data block validity. If the block is valid, copy that with cold status and | |
761 | * modify parent node. | |
762 | * If the parent node is not valid or the data block address is different, | |
763 | * the victim data block is ignored. | |
764 | */ | |
718e53fa | 765 | static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7dda2af8 | 766 | struct gc_inode_list *gc_list, unsigned int segno, int gc_type) |
7bc09003 JK |
767 | { |
768 | struct super_block *sb = sbi->sb; | |
769 | struct f2fs_summary *entry; | |
770 | block_t start_addr; | |
43727527 | 771 | int off; |
7bc09003 JK |
772 | int phase = 0; |
773 | ||
774 | start_addr = START_BLOCK(sbi, segno); | |
775 | ||
776 | next_step: | |
777 | entry = sum; | |
c718379b | 778 | |
7bc09003 JK |
779 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
780 | struct page *data_page; | |
781 | struct inode *inode; | |
782 | struct node_info dni; /* dnode info for the data */ | |
783 | unsigned int ofs_in_node, nofs; | |
784 | block_t start_bidx; | |
7ea984b0 | 785 | nid_t nid = le32_to_cpu(entry->nid); |
7bc09003 | 786 | |
43727527 | 787 | /* stop BG_GC if there is not enough free sections. */ |
7f3037a5 | 788 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) |
718e53fa | 789 | return; |
7bc09003 | 790 | |
43727527 | 791 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
792 | continue; |
793 | ||
794 | if (phase == 0) { | |
7ea984b0 CY |
795 | ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, |
796 | META_NAT, true); | |
797 | continue; | |
798 | } | |
799 | ||
800 | if (phase == 1) { | |
801 | ra_node_page(sbi, nid); | |
7bc09003 JK |
802 | continue; |
803 | } | |
804 | ||
805 | /* Get an inode by ino with checking validity */ | |
c1079892 | 806 | if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) |
7bc09003 JK |
807 | continue; |
808 | ||
7ea984b0 | 809 | if (phase == 2) { |
7bc09003 JK |
810 | ra_node_page(sbi, dni.ino); |
811 | continue; | |
812 | } | |
813 | ||
7bc09003 JK |
814 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); |
815 | ||
7ea984b0 | 816 | if (phase == 3) { |
d4686d56 | 817 | inode = f2fs_iget(sb, dni.ino); |
b73e5282 | 818 | if (IS_ERR(inode) || is_bad_inode(inode)) |
7bc09003 JK |
819 | continue; |
820 | ||
4375a336 JK |
821 | /* if encrypted inode, let's go phase 3 */ |
822 | if (f2fs_encrypted_inode(inode) && | |
823 | S_ISREG(inode->i_mode)) { | |
824 | add_gc_inode(gc_list, inode); | |
825 | continue; | |
826 | } | |
827 | ||
81ca7350 | 828 | start_bidx = start_bidx_of_node(nofs, inode); |
43f3eae1 | 829 | data_page = get_read_data_page(inode, |
70246286 CH |
830 | start_bidx + ofs_in_node, REQ_RAHEAD, |
831 | true); | |
31a32688 CL |
832 | if (IS_ERR(data_page)) { |
833 | iput(inode); | |
834 | continue; | |
835 | } | |
7bc09003 JK |
836 | |
837 | f2fs_put_page(data_page, 0); | |
7dda2af8 | 838 | add_gc_inode(gc_list, inode); |
31a32688 CL |
839 | continue; |
840 | } | |
841 | ||
7ea984b0 | 842 | /* phase 4 */ |
7dda2af8 | 843 | inode = find_gc_inode(gc_list, dni.ino); |
31a32688 | 844 | if (inode) { |
82e0a5aa CY |
845 | struct f2fs_inode_info *fi = F2FS_I(inode); |
846 | bool locked = false; | |
847 | ||
848 | if (S_ISREG(inode->i_mode)) { | |
849 | if (!down_write_trylock(&fi->dio_rwsem[READ])) | |
850 | continue; | |
851 | if (!down_write_trylock( | |
852 | &fi->dio_rwsem[WRITE])) { | |
853 | up_write(&fi->dio_rwsem[READ]); | |
854 | continue; | |
855 | } | |
856 | locked = true; | |
857 | } | |
858 | ||
81ca7350 | 859 | start_bidx = start_bidx_of_node(nofs, inode) |
c879f90d | 860 | + ofs_in_node; |
4375a336 | 861 | if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) |
20614711 | 862 | move_encrypted_block(inode, start_bidx, segno, off); |
4375a336 | 863 | else |
20614711 | 864 | move_data_page(inode, start_bidx, gc_type, segno, off); |
82e0a5aa CY |
865 | |
866 | if (locked) { | |
867 | up_write(&fi->dio_rwsem[WRITE]); | |
868 | up_write(&fi->dio_rwsem[READ]); | |
869 | } | |
870 | ||
e1235983 | 871 | stat_inc_data_blk_count(sbi, 1, gc_type); |
7bc09003 | 872 | } |
7bc09003 | 873 | } |
c718379b | 874 | |
7ea984b0 | 875 | if (++phase < 5) |
7bc09003 | 876 | goto next_step; |
7bc09003 JK |
877 | } |
878 | ||
879 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
8a2d0ace | 880 | int gc_type) |
7bc09003 JK |
881 | { |
882 | struct sit_info *sit_i = SIT_I(sbi); | |
883 | int ret; | |
8a2d0ace | 884 | |
7bc09003 | 885 | mutex_lock(&sit_i->sentry_lock); |
8a2d0ace GZ |
886 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, |
887 | NO_CHECK_TYPE, LFS); | |
7bc09003 JK |
888 | mutex_unlock(&sit_i->sentry_lock); |
889 | return ret; | |
890 | } | |
891 | ||
718e53fa CY |
892 | static int do_garbage_collect(struct f2fs_sb_info *sbi, |
893 | unsigned int start_segno, | |
7dda2af8 | 894 | struct gc_inode_list *gc_list, int gc_type) |
7bc09003 JK |
895 | { |
896 | struct page *sum_page; | |
897 | struct f2fs_summary_block *sum; | |
c718379b | 898 | struct blk_plug plug; |
718e53fa CY |
899 | unsigned int segno = start_segno; |
900 | unsigned int end_segno = start_segno + sbi->segs_per_sec; | |
43ced84e | 901 | int sec_freed = 0; |
718e53fa CY |
902 | unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ? |
903 | SUM_TYPE_DATA : SUM_TYPE_NODE; | |
7bc09003 | 904 | |
718e53fa CY |
905 | /* readahead multi ssa blocks those have contiguous address */ |
906 | if (sbi->segs_per_sec > 1) | |
907 | ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), | |
908 | sbi->segs_per_sec, META_SSA, true); | |
909 | ||
910 | /* reference all summary page */ | |
911 | while (segno < end_segno) { | |
912 | sum_page = get_sum_page(sbi, segno++); | |
913 | unlock_page(sum_page); | |
914 | } | |
7bc09003 | 915 | |
c718379b JK |
916 | blk_start_plug(&plug); |
917 | ||
718e53fa | 918 | for (segno = start_segno; segno < end_segno; segno++) { |
aa987273 | 919 | |
718e53fa CY |
920 | /* find segment summary of victim */ |
921 | sum_page = find_get_page(META_MAPPING(sbi), | |
922 | GET_SUM_BLOCK(sbi, segno)); | |
718e53fa | 923 | f2fs_put_page(sum_page, 0); |
7bc09003 | 924 | |
302bd348 | 925 | if (get_valid_blocks(sbi, segno, false) == 0 || |
de0dcc40 JK |
926 | !PageUptodate(sum_page) || |
927 | unlikely(f2fs_cp_error(sbi))) | |
928 | goto next; | |
929 | ||
718e53fa CY |
930 | sum = page_address(sum_page); |
931 | f2fs_bug_on(sbi, type != GET_SUM_TYPE((&sum->footer))); | |
932 | ||
933 | /* | |
934 | * this is to avoid deadlock: | |
935 | * - lock_page(sum_page) - f2fs_replace_block | |
936 | * - check_valid_map() - mutex_lock(sentry_lock) | |
937 | * - mutex_lock(sentry_lock) - change_curseg() | |
938 | * - lock_page(sum_page) | |
939 | */ | |
718e53fa CY |
940 | if (type == SUM_TYPE_NODE) |
941 | gc_node_segment(sbi, sum->entries, segno, gc_type); | |
942 | else | |
943 | gc_data_segment(sbi, sum->entries, gc_list, segno, | |
944 | gc_type); | |
945 | ||
946 | stat_inc_seg_count(sbi, type, gc_type); | |
f6fe2be3 | 947 | next: |
718e53fa CY |
948 | f2fs_put_page(sum_page, 0); |
949 | } | |
950 | ||
da011cc0 | 951 | if (gc_type == FG_GC) |
b9109b0e JK |
952 | f2fs_submit_merged_write(sbi, |
953 | (type == SUM_TYPE_NODE) ? NODE : DATA); | |
c718379b | 954 | |
718e53fa | 955 | blk_finish_plug(&plug); |
7bc09003 | 956 | |
43ced84e | 957 | if (gc_type == FG_GC && |
302bd348 | 958 | get_valid_blocks(sbi, start_segno, true) == 0) |
43ced84e | 959 | sec_freed = 1; |
17d899df CY |
960 | |
961 | stat_inc_call_count(sbi->stat_info); | |
962 | ||
43ced84e | 963 | return sec_freed; |
7bc09003 JK |
964 | } |
965 | ||
e066b83c JK |
966 | int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, |
967 | bool background, unsigned int segno) | |
7bc09003 | 968 | { |
d530d4d8 | 969 | int gc_type = sync ? FG_GC : BG_GC; |
43ced84e | 970 | int sec_freed = 0; |
e5dbd956 | 971 | int ret; |
d5053a34 | 972 | struct cp_control cpc; |
e066b83c | 973 | unsigned int init_segno = segno; |
7dda2af8 CL |
974 | struct gc_inode_list gc_list = { |
975 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
769ec6e5 | 976 | .iroot = RADIX_TREE_INIT(GFP_NOFS), |
7dda2af8 | 977 | }; |
d5053a34 | 978 | |
119ee914 | 979 | cpc.reason = __get_cp_reason(sbi); |
7bc09003 | 980 | gc_more: |
e5dbd956 WG |
981 | if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE))) { |
982 | ret = -EINVAL; | |
408e9375 | 983 | goto stop; |
e5dbd956 | 984 | } |
6d5a1495 CY |
985 | if (unlikely(f2fs_cp_error(sbi))) { |
986 | ret = -EIO; | |
203681f6 | 987 | goto stop; |
6d5a1495 | 988 | } |
7bc09003 | 989 | |
19f4e688 | 990 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) { |
6e17bfbc | 991 | /* |
19f4e688 HP |
992 | * For example, if there are many prefree_segments below given |
993 | * threshold, we can make them free by checkpoint. Then, we | |
994 | * secure free segments which doesn't need fggc any more. | |
6e17bfbc | 995 | */ |
8fd5a37e JK |
996 | if (prefree_segments(sbi)) { |
997 | ret = write_checkpoint(sbi, &cpc); | |
998 | if (ret) | |
999 | goto stop; | |
1000 | } | |
19f4e688 HP |
1001 | if (has_not_enough_free_secs(sbi, 0, 0)) |
1002 | gc_type = FG_GC; | |
d64f8047 | 1003 | } |
7bc09003 | 1004 | |
e5dbd956 | 1005 | ret = -EINVAL; |
19f4e688 HP |
1006 | /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */ |
1007 | if (gc_type == BG_GC && !background) | |
1008 | goto stop; | |
05eeb118 | 1009 | if (!__get_victim(sbi, &segno, gc_type)) |
408e9375 | 1010 | goto stop; |
43727527 | 1011 | ret = 0; |
7bc09003 | 1012 | |
43ced84e CY |
1013 | if (do_garbage_collect(sbi, segno, &gc_list, gc_type) && |
1014 | gc_type == FG_GC) | |
45fe8492 | 1015 | sec_freed++; |
43727527 | 1016 | |
5ee5293c | 1017 | if (gc_type == FG_GC) |
5ec4e49f | 1018 | sbi->cur_victim_sec = NULL_SEGNO; |
43727527 | 1019 | |
d530d4d8 | 1020 | if (!sync) { |
e066b83c JK |
1021 | if (has_not_enough_free_secs(sbi, sec_freed, 0)) { |
1022 | segno = NULL_SEGNO; | |
d530d4d8 | 1023 | goto gc_more; |
e066b83c | 1024 | } |
43727527 | 1025 | |
d530d4d8 | 1026 | if (gc_type == FG_GC) |
2956e450 | 1027 | ret = write_checkpoint(sbi, &cpc); |
d530d4d8 | 1028 | } |
408e9375 | 1029 | stop: |
e066b83c JK |
1030 | SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0; |
1031 | SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno; | |
7bc09003 JK |
1032 | mutex_unlock(&sbi->gc_mutex); |
1033 | ||
7dda2af8 | 1034 | put_gc_inode(&gc_list); |
d530d4d8 CY |
1035 | |
1036 | if (sync) | |
1037 | ret = sec_freed ? 0 : -EAGAIN; | |
43727527 | 1038 | return ret; |
7bc09003 JK |
1039 | } |
1040 | ||
1041 | void build_gc_manager(struct f2fs_sb_info *sbi) | |
1042 | { | |
4ddb1a4d | 1043 | u64 main_count, resv_count, ovp_count; |
e93b9865 | 1044 | |
7bc09003 | 1045 | DIRTY_I(sbi)->v_ops = &default_v_ops; |
e93b9865 HP |
1046 | |
1047 | /* threshold of # of valid blocks in a section for victims of FG_GC */ | |
1048 | main_count = SM_I(sbi)->main_segments << sbi->log_blocks_per_seg; | |
1049 | resv_count = SM_I(sbi)->reserved_segments << sbi->log_blocks_per_seg; | |
1050 | ovp_count = SM_I(sbi)->ovp_segments << sbi->log_blocks_per_seg; | |
e93b9865 | 1051 | |
4ddb1a4d JK |
1052 | sbi->fggc_threshold = div64_u64((main_count - ovp_count) * |
1053 | BLKS_PER_SEC(sbi), (main_count - resv_count)); | |
d5793249 JK |
1054 | |
1055 | /* give warm/cold data area from slower device */ | |
1056 | if (sbi->s_ndevs && sbi->segs_per_sec == 1) | |
1057 | SIT_I(sbi)->last_victim[ALLOC_NEXT] = | |
1058 | GET_SEGNO(sbi, FDEV(0).end_blk) + 1; | |
7bc09003 | 1059 | } |