]>
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 | 30 | wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; |
b8c502b8 | 31 | unsigned int wait_ms; |
7bc09003 | 32 | |
b59d0bae | 33 | wait_ms = gc_th->min_sleep_time; |
7bc09003 | 34 | |
1d7be270 | 35 | set_freezable(); |
7bc09003 | 36 | do { |
1d7be270 | 37 | wait_event_interruptible_timeout(*wq, |
d9872a69 JK |
38 | kthread_should_stop() || freezing(current) || |
39 | gc_th->gc_wake, | |
1d7be270 JK |
40 | msecs_to_jiffies(wait_ms)); |
41 | ||
d9872a69 JK |
42 | /* give it a try one time */ |
43 | if (gc_th->gc_wake) | |
44 | gc_th->gc_wake = 0; | |
45 | ||
7bc09003 JK |
46 | if (try_to_freeze()) |
47 | continue; | |
7bc09003 JK |
48 | if (kthread_should_stop()) |
49 | break; | |
50 | ||
d6212a5f | 51 | if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { |
88dd8934 | 52 | increase_sleep_time(gc_th, &wait_ms); |
d6212a5f CL |
53 | continue; |
54 | } | |
55 | ||
0f348028 | 56 | #ifdef CONFIG_F2FS_FAULT_INJECTION |
55523519 CY |
57 | if (time_to_inject(sbi, FAULT_CHECKPOINT)) { |
58 | f2fs_show_injection_info(FAULT_CHECKPOINT); | |
0f348028 | 59 | f2fs_stop_checkpoint(sbi, false); |
55523519 | 60 | } |
0f348028 CY |
61 | #endif |
62 | ||
dc6febb6 CY |
63 | if (!sb_start_write_trylock(sbi->sb)) |
64 | continue; | |
65 | ||
7bc09003 JK |
66 | /* |
67 | * [GC triggering condition] | |
68 | * 0. GC is not conducted currently. | |
69 | * 1. There are enough dirty segments. | |
70 | * 2. IO subsystem is idle by checking the # of writeback pages. | |
71 | * 3. IO subsystem is idle by checking the # of requests in | |
72 | * bdev's request list. | |
73 | * | |
e1c42045 | 74 | * Note) We have to avoid triggering GCs frequently. |
7bc09003 JK |
75 | * Because it is possible that some segments can be |
76 | * invalidated soon after by user update or deletion. | |
77 | * So, I'd like to wait some time to collect dirty segments. | |
78 | */ | |
79 | if (!mutex_trylock(&sbi->gc_mutex)) | |
dc6febb6 | 80 | goto next; |
7bc09003 | 81 | |
d9872a69 JK |
82 | if (gc_th->gc_urgent) { |
83 | wait_ms = gc_th->urgent_sleep_time; | |
84 | goto do_gc; | |
85 | } | |
86 | ||
7bc09003 | 87 | if (!is_idle(sbi)) { |
88dd8934 | 88 | increase_sleep_time(gc_th, &wait_ms); |
7bc09003 | 89 | mutex_unlock(&sbi->gc_mutex); |
dc6febb6 | 90 | goto next; |
7bc09003 JK |
91 | } |
92 | ||
93 | if (has_enough_invalid_blocks(sbi)) | |
88dd8934 | 94 | decrease_sleep_time(gc_th, &wait_ms); |
7bc09003 | 95 | else |
88dd8934 | 96 | increase_sleep_time(gc_th, &wait_ms); |
d9872a69 | 97 | do_gc: |
dcdfff65 | 98 | stat_inc_bggc_count(sbi); |
7bc09003 | 99 | |
43727527 | 100 | /* if return value is not zero, no victim was selected */ |
e066b83c | 101 | if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true, NULL_SEGNO)) |
b59d0bae | 102 | wait_ms = gc_th->no_gc_sleep_time; |
81eb8d6e | 103 | |
84e4214f JK |
104 | trace_f2fs_background_gc(sbi->sb, wait_ms, |
105 | prefree_segments(sbi), free_segments(sbi)); | |
106 | ||
4660f9c0 JK |
107 | /* balancing f2fs's metadata periodically */ |
108 | f2fs_balance_fs_bg(sbi); | |
dc6febb6 CY |
109 | next: |
110 | sb_end_write(sbi->sb); | |
81eb8d6e | 111 | |
7bc09003 JK |
112 | } while (!kthread_should_stop()); |
113 | return 0; | |
114 | } | |
115 | ||
116 | int start_gc_thread(struct f2fs_sb_info *sbi) | |
117 | { | |
1042d60f | 118 | struct f2fs_gc_kthread *gc_th; |
ec7b1f2d | 119 | dev_t dev = sbi->sb->s_bdev->bd_dev; |
7a267f8d | 120 | int err = 0; |
7bc09003 | 121 | |
1ecc0c5c | 122 | gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL); |
7a267f8d NJ |
123 | if (!gc_th) { |
124 | err = -ENOMEM; | |
125 | goto out; | |
126 | } | |
7bc09003 | 127 | |
d9872a69 | 128 | gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME; |
b59d0bae NJ |
129 | gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; |
130 | gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; | |
131 | gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; | |
132 | ||
d2dc095f | 133 | gc_th->gc_idle = 0; |
d9872a69 JK |
134 | gc_th->gc_urgent = 0; |
135 | gc_th->gc_wake= 0; | |
d2dc095f | 136 | |
7bc09003 JK |
137 | sbi->gc_thread = gc_th; |
138 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
139 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, | |
ec7b1f2d | 140 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 | 141 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
7a267f8d | 142 | err = PTR_ERR(gc_th->f2fs_gc_task); |
7bc09003 | 143 | kfree(gc_th); |
25718423 | 144 | sbi->gc_thread = NULL; |
7bc09003 | 145 | } |
7a267f8d NJ |
146 | out: |
147 | return err; | |
7bc09003 JK |
148 | } |
149 | ||
150 | void stop_gc_thread(struct f2fs_sb_info *sbi) | |
151 | { | |
152 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
153 | if (!gc_th) | |
154 | return; | |
155 | kthread_stop(gc_th->f2fs_gc_task); | |
156 | kfree(gc_th); | |
157 | sbi->gc_thread = NULL; | |
158 | } | |
159 | ||
d2dc095f | 160 | static int select_gc_type(struct f2fs_gc_kthread *gc_th, int gc_type) |
7bc09003 | 161 | { |
d2dc095f NJ |
162 | int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY; |
163 | ||
164 | if (gc_th && gc_th->gc_idle) { | |
165 | if (gc_th->gc_idle == 1) | |
166 | gc_mode = GC_CB; | |
167 | else if (gc_th->gc_idle == 2) | |
168 | gc_mode = GC_GREEDY; | |
169 | } | |
170 | return gc_mode; | |
7bc09003 JK |
171 | } |
172 | ||
173 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
174 | int type, struct victim_sel_policy *p) | |
175 | { | |
176 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
177 | ||
4ebefc44 | 178 | if (p->alloc_mode == SSR) { |
7bc09003 JK |
179 | p->gc_mode = GC_GREEDY; |
180 | p->dirty_segmap = dirty_i->dirty_segmap[type]; | |
a26b7c8a | 181 | p->max_search = dirty_i->nr_dirty[type]; |
7bc09003 JK |
182 | p->ofs_unit = 1; |
183 | } else { | |
d2dc095f | 184 | p->gc_mode = select_gc_type(sbi->gc_thread, gc_type); |
7bc09003 | 185 | p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; |
a26b7c8a | 186 | p->max_search = dirty_i->nr_dirty[DIRTY]; |
7bc09003 JK |
187 | p->ofs_unit = sbi->segs_per_sec; |
188 | } | |
a26b7c8a | 189 | |
e93b9865 HP |
190 | /* we need to check every dirty segments in the FG_GC case */ |
191 | if (gc_type != FG_GC && p->max_search > sbi->max_victim_search) | |
b1c57c1c | 192 | p->max_search = sbi->max_victim_search; |
a26b7c8a | 193 | |
482eb99a YS |
194 | /* let's select beginning hot/small space first in no_heap mode*/ |
195 | if (test_opt(sbi, NOHEAP) && | |
196 | (type == CURSEG_HOT_DATA || IS_NODESEG(type))) | |
7a20b8a6 JK |
197 | p->offset = 0; |
198 | else | |
e066b83c | 199 | p->offset = SIT_I(sbi)->last_victim[p->gc_mode]; |
7bc09003 JK |
200 | } |
201 | ||
202 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
203 | struct victim_sel_policy *p) | |
204 | { | |
b7250d2d JK |
205 | /* SSR allocates in a segment unit */ |
206 | if (p->alloc_mode == SSR) | |
3519e3f9 | 207 | return sbi->blocks_per_seg; |
7bc09003 | 208 | if (p->gc_mode == GC_GREEDY) |
c541a51b | 209 | return 2 * sbi->blocks_per_seg * p->ofs_unit; |
7bc09003 JK |
210 | else if (p->gc_mode == GC_CB) |
211 | return UINT_MAX; | |
212 | else /* No other gc_mode */ | |
213 | return 0; | |
214 | } | |
215 | ||
216 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
217 | { | |
218 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 219 | unsigned int secno; |
7bc09003 JK |
220 | |
221 | /* | |
222 | * If the gc_type is FG_GC, we can select victim segments | |
223 | * selected by background GC before. | |
224 | * Those segments guarantee they have small valid blocks. | |
225 | */ | |
7cd8558b | 226 | for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { |
5ec4e49f | 227 | if (sec_usage_check(sbi, secno)) |
b65ee148 | 228 | continue; |
e93b9865 HP |
229 | |
230 | if (no_fggc_candidate(sbi, secno)) | |
231 | continue; | |
232 | ||
5ec4e49f | 233 | clear_bit(secno, dirty_i->victim_secmap); |
4ddb1a4d | 234 | return GET_SEG_FROM_SEC(sbi, secno); |
7bc09003 JK |
235 | } |
236 | return NULL_SEGNO; | |
237 | } | |
238 | ||
239 | static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) | |
240 | { | |
241 | struct sit_info *sit_i = SIT_I(sbi); | |
4ddb1a4d JK |
242 | unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); |
243 | unsigned int start = GET_SEG_FROM_SEC(sbi, secno); | |
7bc09003 JK |
244 | unsigned long long mtime = 0; |
245 | unsigned int vblocks; | |
246 | unsigned char age = 0; | |
247 | unsigned char u; | |
248 | unsigned int i; | |
249 | ||
250 | for (i = 0; i < sbi->segs_per_sec; i++) | |
251 | mtime += get_seg_entry(sbi, start + i)->mtime; | |
302bd348 | 252 | vblocks = get_valid_blocks(sbi, segno, true); |
7bc09003 JK |
253 | |
254 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
255 | vblocks = div_u64(vblocks, sbi->segs_per_sec); | |
256 | ||
257 | u = (vblocks * 100) >> sbi->log_blocks_per_seg; | |
258 | ||
e1c42045 | 259 | /* Handle if the system time has changed by the user */ |
7bc09003 JK |
260 | if (mtime < sit_i->min_mtime) |
261 | sit_i->min_mtime = mtime; | |
262 | if (mtime > sit_i->max_mtime) | |
263 | sit_i->max_mtime = mtime; | |
264 | if (sit_i->max_mtime != sit_i->min_mtime) | |
265 | age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), | |
266 | sit_i->max_mtime - sit_i->min_mtime); | |
267 | ||
268 | return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); | |
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) | |
2afce76a | 275 | return get_seg_entry(sbi, segno)->ckpt_valid_blocks; |
7bc09003 JK |
276 | |
277 | /* alloc_mode == LFS */ | |
278 | if (p->gc_mode == GC_GREEDY) | |
91f4382b | 279 | return get_valid_blocks(sbi, segno, true); |
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 | 322 | if (*result != NULL_SEGNO) { |
c356f2ee | 323 | if (get_valid_blocks(sbi, *result, false) && |
e066b83c JK |
324 | !sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result))) |
325 | p.min_segno = *result; | |
326 | goto out; | |
327 | } | |
328 | ||
3342bb30 CY |
329 | if (p.max_search == 0) |
330 | goto out; | |
331 | ||
e066b83c | 332 | last_victim = sm->last_victim[p.gc_mode]; |
7bc09003 JK |
333 | if (p.alloc_mode == LFS && gc_type == FG_GC) { |
334 | p.min_segno = check_bg_victims(sbi); | |
335 | if (p.min_segno != NULL_SEGNO) | |
336 | goto got_it; | |
337 | } | |
338 | ||
339 | while (1) { | |
340 | unsigned long cost; | |
5ec4e49f | 341 | unsigned int segno; |
7bc09003 | 342 | |
a43f7ec3 CY |
343 | segno = find_next_bit(p.dirty_segmap, last_segment, p.offset); |
344 | if (segno >= last_segment) { | |
e066b83c JK |
345 | if (sm->last_victim[p.gc_mode]) { |
346 | last_segment = | |
347 | sm->last_victim[p.gc_mode]; | |
348 | sm->last_victim[p.gc_mode] = 0; | |
7bc09003 JK |
349 | p.offset = 0; |
350 | continue; | |
351 | } | |
352 | break; | |
353 | } | |
a57e564d JX |
354 | |
355 | p.offset = segno + p.ofs_unit; | |
688159b6 | 356 | if (p.ofs_unit > 1) { |
a57e564d | 357 | p.offset -= segno % p.ofs_unit; |
688159b6 FL |
358 | nsearched += count_bits(p.dirty_segmap, |
359 | p.offset - p.ofs_unit, | |
360 | p.ofs_unit); | |
361 | } else { | |
362 | nsearched++; | |
363 | } | |
364 | ||
4ddb1a4d | 365 | secno = GET_SEC_FROM_SEG(sbi, segno); |
7bc09003 | 366 | |
5ec4e49f | 367 | if (sec_usage_check(sbi, secno)) |
688159b6 | 368 | goto next; |
5ec4e49f | 369 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
688159b6 | 370 | goto next; |
e93b9865 HP |
371 | if (gc_type == FG_GC && p.alloc_mode == LFS && |
372 | no_fggc_candidate(sbi, secno)) | |
373 | goto next; | |
7bc09003 JK |
374 | |
375 | cost = get_gc_cost(sbi, segno, &p); | |
376 | ||
377 | if (p.min_cost > cost) { | |
378 | p.min_segno = segno; | |
379 | p.min_cost = cost; | |
a57e564d | 380 | } |
688159b6 FL |
381 | next: |
382 | if (nsearched >= p.max_search) { | |
e066b83c JK |
383 | if (!sm->last_victim[p.gc_mode] && segno <= last_victim) |
384 | sm->last_victim[p.gc_mode] = last_victim + 1; | |
4ce53776 | 385 | else |
e066b83c JK |
386 | sm->last_victim[p.gc_mode] = segno + 1; |
387 | sm->last_victim[p.gc_mode] %= MAIN_SEGS(sbi); | |
7bc09003 JK |
388 | break; |
389 | } | |
390 | } | |
7bc09003 | 391 | if (p.min_segno != NULL_SEGNO) { |
b2b3460a | 392 | got_it: |
7bc09003 | 393 | if (p.alloc_mode == LFS) { |
4ddb1a4d | 394 | secno = GET_SEC_FROM_SEG(sbi, p.min_segno); |
5ec4e49f JK |
395 | if (gc_type == FG_GC) |
396 | sbi->cur_victim_sec = secno; | |
397 | else | |
398 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 399 | } |
5ec4e49f | 400 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
8e46b3ed NJ |
401 | |
402 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, | |
403 | sbi->cur_victim_sec, | |
404 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 | 405 | } |
3342bb30 | 406 | out: |
7bc09003 JK |
407 | mutex_unlock(&dirty_i->seglist_lock); |
408 | ||
409 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
410 | } | |
411 | ||
412 | static const struct victim_selection default_v_ops = { | |
413 | .get_victim = get_victim_by_default, | |
414 | }; | |
415 | ||
7dda2af8 | 416 | static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) |
7bc09003 | 417 | { |
7bc09003 JK |
418 | struct inode_entry *ie; |
419 | ||
7dda2af8 CL |
420 | ie = radix_tree_lookup(&gc_list->iroot, ino); |
421 | if (ie) | |
422 | return ie->inode; | |
7bc09003 JK |
423 | return NULL; |
424 | } | |
425 | ||
7dda2af8 | 426 | static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) |
7bc09003 | 427 | { |
6cc4af56 GZ |
428 | struct inode_entry *new_ie; |
429 | ||
7dda2af8 | 430 | if (inode == find_gc_inode(gc_list, inode->i_ino)) { |
6cc4af56 GZ |
431 | iput(inode); |
432 | return; | |
7bc09003 | 433 | } |
06292073 | 434 | new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); |
7bc09003 | 435 | new_ie->inode = inode; |
f28e5034 CY |
436 | |
437 | f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); | |
7dda2af8 | 438 | list_add_tail(&new_ie->list, &gc_list->ilist); |
7bc09003 JK |
439 | } |
440 | ||
7dda2af8 | 441 | static void put_gc_inode(struct gc_inode_list *gc_list) |
7bc09003 JK |
442 | { |
443 | struct inode_entry *ie, *next_ie; | |
7dda2af8 CL |
444 | list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { |
445 | radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); | |
7bc09003 JK |
446 | iput(ie->inode); |
447 | list_del(&ie->list); | |
06292073 | 448 | kmem_cache_free(inode_entry_slab, ie); |
7bc09003 JK |
449 | } |
450 | } | |
451 | ||
452 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
453 | unsigned int segno, int offset) | |
454 | { | |
455 | struct sit_info *sit_i = SIT_I(sbi); | |
456 | struct seg_entry *sentry; | |
457 | int ret; | |
458 | ||
3d26fa6b | 459 | down_read(&sit_i->sentry_lock); |
7bc09003 JK |
460 | sentry = get_seg_entry(sbi, segno); |
461 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
3d26fa6b | 462 | up_read(&sit_i->sentry_lock); |
43727527 | 463 | return ret; |
7bc09003 JK |
464 | } |
465 | ||
0a8165d7 | 466 | /* |
7bc09003 JK |
467 | * This function compares node address got in summary with that in NAT. |
468 | * On validity, copy that node with cold status, otherwise (invalid node) | |
469 | * ignore that. | |
470 | */ | |
718e53fa | 471 | static void gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
472 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
473 | { | |
7bc09003 | 474 | struct f2fs_summary *entry; |
26d58599 | 475 | block_t start_addr; |
7bc09003 | 476 | int off; |
7ea984b0 | 477 | int phase = 0; |
7bc09003 | 478 | |
26d58599 JK |
479 | start_addr = START_BLOCK(sbi, segno); |
480 | ||
7bc09003 JK |
481 | next_step: |
482 | entry = sum; | |
c718379b | 483 | |
7bc09003 JK |
484 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
485 | nid_t nid = le32_to_cpu(entry->nid); | |
486 | struct page *node_page; | |
26d58599 | 487 | struct node_info ni; |
7bc09003 | 488 | |
43727527 | 489 | /* stop BG_GC if there is not enough free sections. */ |
7f3037a5 | 490 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) |
718e53fa | 491 | return; |
7bc09003 | 492 | |
43727527 | 493 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
494 | continue; |
495 | ||
7ea984b0 CY |
496 | if (phase == 0) { |
497 | ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, | |
498 | META_NAT, true); | |
499 | continue; | |
500 | } | |
501 | ||
502 | if (phase == 1) { | |
7bc09003 JK |
503 | ra_node_page(sbi, nid); |
504 | continue; | |
505 | } | |
7ea984b0 CY |
506 | |
507 | /* phase == 2 */ | |
7bc09003 JK |
508 | node_page = get_node_page(sbi, nid); |
509 | if (IS_ERR(node_page)) | |
510 | continue; | |
511 | ||
9a01b56b HY |
512 | /* block may become invalid during get_node_page */ |
513 | if (check_valid_map(sbi, segno, off) == 0) { | |
514 | f2fs_put_page(node_page, 1); | |
515 | continue; | |
26d58599 JK |
516 | } |
517 | ||
518 | get_node_info(sbi, nid, &ni); | |
519 | if (ni.blk_addr != start_addr + off) { | |
520 | f2fs_put_page(node_page, 1); | |
521 | continue; | |
9a01b56b HY |
522 | } |
523 | ||
da011cc0 | 524 | move_node_page(node_page, gc_type); |
e1235983 | 525 | stat_inc_node_blk_count(sbi, 1, gc_type); |
7bc09003 | 526 | } |
c718379b | 527 | |
7ea984b0 | 528 | if (++phase < 3) |
7bc09003 | 529 | goto next_step; |
7bc09003 JK |
530 | } |
531 | ||
0a8165d7 | 532 | /* |
9af45ef5 JK |
533 | * Calculate start block index indicating the given node offset. |
534 | * Be careful, caller should give this node offset only indicating direct node | |
535 | * blocks. If any node offsets, which point the other types of node blocks such | |
536 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
537 | * bug. | |
7bc09003 | 538 | */ |
81ca7350 | 539 | block_t start_bidx_of_node(unsigned int node_ofs, struct inode *inode) |
7bc09003 | 540 | { |
ce19a5d4 JK |
541 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
542 | unsigned int bidx; | |
7bc09003 | 543 | |
ce19a5d4 JK |
544 | if (node_ofs == 0) |
545 | return 0; | |
7bc09003 | 546 | |
ce19a5d4 | 547 | if (node_ofs <= 2) { |
7bc09003 JK |
548 | bidx = node_ofs - 1; |
549 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 550 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
551 | bidx = node_ofs - 2 - dec; |
552 | } else { | |
ce19a5d4 | 553 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
554 | bidx = node_ofs - 5 - dec; |
555 | } | |
81ca7350 | 556 | return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode); |
7bc09003 JK |
557 | } |
558 | ||
c1079892 | 559 | static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
560 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) |
561 | { | |
562 | struct page *node_page; | |
563 | nid_t nid; | |
564 | unsigned int ofs_in_node; | |
565 | block_t source_blkaddr; | |
566 | ||
567 | nid = le32_to_cpu(sum->nid); | |
568 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
569 | ||
570 | node_page = get_node_page(sbi, nid); | |
571 | if (IS_ERR(node_page)) | |
c1079892 | 572 | return false; |
7bc09003 JK |
573 | |
574 | get_node_info(sbi, nid, dni); | |
575 | ||
576 | if (sum->version != dni->version) { | |
c13ff37e JK |
577 | f2fs_msg(sbi->sb, KERN_WARNING, |
578 | "%s: valid data with mismatched node version.", | |
579 | __func__); | |
580 | set_sbi_flag(sbi, SBI_NEED_FSCK); | |
7bc09003 JK |
581 | } |
582 | ||
583 | *nofs = ofs_of_node(node_page); | |
7a2af766 | 584 | source_blkaddr = datablock_addr(NULL, node_page, ofs_in_node); |
7bc09003 JK |
585 | f2fs_put_page(node_page, 1); |
586 | ||
587 | if (source_blkaddr != blkaddr) | |
c1079892 NK |
588 | return false; |
589 | return true; | |
7bc09003 JK |
590 | } |
591 | ||
d4c759ee JK |
592 | /* |
593 | * Move data block via META_MAPPING while keeping locked data page. | |
594 | * This can be used to move blocks, aka LBAs, directly on disk. | |
595 | */ | |
596 | static void move_data_block(struct inode *inode, block_t bidx, | |
597 | unsigned int segno, int off) | |
4375a336 JK |
598 | { |
599 | struct f2fs_io_info fio = { | |
600 | .sbi = F2FS_I_SB(inode), | |
39d787be | 601 | .ino = inode->i_ino, |
4375a336 | 602 | .type = DATA, |
a912b54d | 603 | .temp = COLD, |
04d328de | 604 | .op = REQ_OP_READ, |
70fd7614 | 605 | .op_flags = 0, |
4375a336 | 606 | .encrypted_page = NULL, |
fb830fc5 | 607 | .in_list = false, |
4375a336 JK |
608 | }; |
609 | struct dnode_of_data dn; | |
610 | struct f2fs_summary sum; | |
611 | struct node_info ni; | |
612 | struct page *page; | |
4356e48e | 613 | block_t newaddr; |
4375a336 JK |
614 | int err; |
615 | ||
616 | /* do not read out */ | |
a56c7c6f | 617 | page = f2fs_grab_cache_page(inode->i_mapping, bidx, false); |
4375a336 JK |
618 | if (!page) |
619 | return; | |
620 | ||
20614711 YH |
621 | if (!check_valid_map(F2FS_I_SB(inode), segno, off)) |
622 | goto out; | |
623 | ||
5fe45743 CY |
624 | if (f2fs_is_atomic_file(inode)) |
625 | goto out; | |
626 | ||
4375a336 JK |
627 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
628 | err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE); | |
629 | if (err) | |
630 | goto out; | |
631 | ||
08b39fbd CY |
632 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) { |
633 | ClearPageUptodate(page); | |
4375a336 | 634 | goto put_out; |
08b39fbd CY |
635 | } |
636 | ||
637 | /* | |
638 | * don't cache encrypted data into meta inode until previous dirty | |
639 | * data were writebacked to avoid racing between GC and flush. | |
640 | */ | |
fec1d657 | 641 | f2fs_wait_on_page_writeback(page, DATA, true); |
4375a336 JK |
642 | |
643 | get_node_info(fio.sbi, dn.nid, &ni); | |
644 | set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); | |
645 | ||
646 | /* read page */ | |
647 | fio.page = page; | |
7a9d7548 | 648 | fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; |
4375a336 | 649 | |
4356e48e | 650 | allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr, |
fb830fc5 | 651 | &sum, CURSEG_COLD_DATA, NULL, false); |
4356e48e | 652 | |
01eccef7 CY |
653 | fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi), |
654 | newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS); | |
4356e48e CY |
655 | if (!fio.encrypted_page) { |
656 | err = -ENOMEM; | |
657 | goto recover_block; | |
658 | } | |
4375a336 | 659 | |
548aedac JK |
660 | err = f2fs_submit_page_bio(&fio); |
661 | if (err) | |
662 | goto put_page_out; | |
663 | ||
664 | /* write page */ | |
665 | lock_page(fio.encrypted_page); | |
666 | ||
1563ac75 | 667 | if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) { |
4356e48e | 668 | err = -EIO; |
548aedac | 669 | goto put_page_out; |
4356e48e | 670 | } |
1563ac75 | 671 | if (unlikely(!PageUptodate(fio.encrypted_page))) { |
4356e48e | 672 | err = -EIO; |
548aedac | 673 | goto put_page_out; |
4356e48e | 674 | } |
548aedac | 675 | |
6282adbf | 676 | set_page_dirty(fio.encrypted_page); |
fec1d657 | 677 | f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true); |
6282adbf JK |
678 | if (clear_page_dirty_for_io(fio.encrypted_page)) |
679 | dec_page_count(fio.sbi, F2FS_DIRTY_META); | |
680 | ||
548aedac | 681 | set_page_writeback(fio.encrypted_page); |
4375a336 JK |
682 | |
683 | /* allocate block address */ | |
fec1d657 | 684 | f2fs_wait_on_page_writeback(dn.node_page, NODE, true); |
4356e48e | 685 | |
04d328de | 686 | fio.op = REQ_OP_WRITE; |
70fd7614 | 687 | fio.op_flags = REQ_SYNC; |
4356e48e | 688 | fio.new_blkaddr = newaddr; |
4262d833 SY |
689 | err = f2fs_submit_page_write(&fio); |
690 | if (err) { | |
691 | if (PageWriteback(fio.encrypted_page)) | |
692 | end_page_writeback(fio.encrypted_page); | |
693 | goto put_page_out; | |
694 | } | |
4375a336 | 695 | |
b0af6d49 CY |
696 | f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE); |
697 | ||
f28b3434 | 698 | f2fs_update_data_blkaddr(&dn, newaddr); |
91942321 | 699 | set_inode_flag(inode, FI_APPEND_WRITE); |
4375a336 | 700 | if (page->index == 0) |
91942321 | 701 | set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); |
548aedac | 702 | put_page_out: |
4375a336 | 703 | f2fs_put_page(fio.encrypted_page, 1); |
4356e48e CY |
704 | recover_block: |
705 | if (err) | |
706 | __f2fs_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr, | |
707 | true, true); | |
4375a336 JK |
708 | put_out: |
709 | f2fs_put_dnode(&dn); | |
710 | out: | |
711 | f2fs_put_page(page, 1); | |
712 | } | |
713 | ||
20614711 YH |
714 | static void move_data_page(struct inode *inode, block_t bidx, int gc_type, |
715 | unsigned int segno, int off) | |
7bc09003 | 716 | { |
c879f90d JK |
717 | struct page *page; |
718 | ||
a56c7c6f | 719 | page = get_lock_data_page(inode, bidx, true); |
c879f90d JK |
720 | if (IS_ERR(page)) |
721 | return; | |
63a0b7cb | 722 | |
20614711 YH |
723 | if (!check_valid_map(F2FS_I_SB(inode), segno, off)) |
724 | goto out; | |
725 | ||
5fe45743 CY |
726 | if (f2fs_is_atomic_file(inode)) |
727 | goto out; | |
728 | ||
7bc09003 | 729 | if (gc_type == BG_GC) { |
4ebefc44 JK |
730 | if (PageWriteback(page)) |
731 | goto out; | |
7bc09003 JK |
732 | set_page_dirty(page); |
733 | set_cold_data(page); | |
734 | } else { | |
c879f90d JK |
735 | struct f2fs_io_info fio = { |
736 | .sbi = F2FS_I_SB(inode), | |
39d787be | 737 | .ino = inode->i_ino, |
c879f90d | 738 | .type = DATA, |
a912b54d | 739 | .temp = COLD, |
04d328de | 740 | .op = REQ_OP_WRITE, |
70fd7614 | 741 | .op_flags = REQ_SYNC, |
e959c8f5 | 742 | .old_blkaddr = NULL_ADDR, |
c879f90d | 743 | .page = page, |
4375a336 | 744 | .encrypted_page = NULL, |
cc15620b | 745 | .need_lock = LOCK_REQ, |
b0af6d49 | 746 | .io_type = FS_GC_DATA_IO, |
c879f90d | 747 | }; |
72e1c797 CY |
748 | bool is_dirty = PageDirty(page); |
749 | int err; | |
750 | ||
751 | retry: | |
6282adbf | 752 | set_page_dirty(page); |
fec1d657 | 753 | f2fs_wait_on_page_writeback(page, DATA, true); |
933439c8 | 754 | if (clear_page_dirty_for_io(page)) { |
a7ffdbe2 | 755 | inode_dec_dirty_pages(inode); |
933439c8 CY |
756 | remove_dirty_inode(inode); |
757 | } | |
72e1c797 | 758 | |
7bc09003 | 759 | set_cold_data(page); |
72e1c797 CY |
760 | |
761 | err = do_write_data_page(&fio); | |
727f9de7 CY |
762 | if (err) { |
763 | clear_cold_data(page); | |
764 | if (err == -ENOMEM) { | |
765 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
766 | goto retry; | |
767 | } | |
768 | if (is_dirty) | |
769 | set_page_dirty(page); | |
72e1c797 | 770 | } |
7bc09003 JK |
771 | } |
772 | out: | |
773 | f2fs_put_page(page, 1); | |
774 | } | |
775 | ||
0a8165d7 | 776 | /* |
7bc09003 JK |
777 | * This function tries to get parent node of victim data block, and identifies |
778 | * data block validity. If the block is valid, copy that with cold status and | |
779 | * modify parent node. | |
780 | * If the parent node is not valid or the data block address is different, | |
781 | * the victim data block is ignored. | |
782 | */ | |
718e53fa | 783 | static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7dda2af8 | 784 | struct gc_inode_list *gc_list, unsigned int segno, int gc_type) |
7bc09003 JK |
785 | { |
786 | struct super_block *sb = sbi->sb; | |
787 | struct f2fs_summary *entry; | |
788 | block_t start_addr; | |
43727527 | 789 | int off; |
7bc09003 JK |
790 | int phase = 0; |
791 | ||
792 | start_addr = START_BLOCK(sbi, segno); | |
793 | ||
794 | next_step: | |
795 | entry = sum; | |
c718379b | 796 | |
7bc09003 JK |
797 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
798 | struct page *data_page; | |
799 | struct inode *inode; | |
800 | struct node_info dni; /* dnode info for the data */ | |
801 | unsigned int ofs_in_node, nofs; | |
802 | block_t start_bidx; | |
7ea984b0 | 803 | nid_t nid = le32_to_cpu(entry->nid); |
7bc09003 | 804 | |
43727527 | 805 | /* stop BG_GC if there is not enough free sections. */ |
7f3037a5 | 806 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) |
718e53fa | 807 | return; |
7bc09003 | 808 | |
43727527 | 809 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
810 | continue; |
811 | ||
812 | if (phase == 0) { | |
7ea984b0 CY |
813 | ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, |
814 | META_NAT, true); | |
815 | continue; | |
816 | } | |
817 | ||
818 | if (phase == 1) { | |
819 | ra_node_page(sbi, nid); | |
7bc09003 JK |
820 | continue; |
821 | } | |
822 | ||
823 | /* Get an inode by ino with checking validity */ | |
c1079892 | 824 | if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) |
7bc09003 JK |
825 | continue; |
826 | ||
7ea984b0 | 827 | if (phase == 2) { |
7bc09003 JK |
828 | ra_node_page(sbi, dni.ino); |
829 | continue; | |
830 | } | |
831 | ||
7bc09003 JK |
832 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); |
833 | ||
7ea984b0 | 834 | if (phase == 3) { |
d4686d56 | 835 | inode = f2fs_iget(sb, dni.ino); |
b73e5282 | 836 | if (IS_ERR(inode) || is_bad_inode(inode)) |
7bc09003 JK |
837 | continue; |
838 | ||
4375a336 | 839 | /* if encrypted inode, let's go phase 3 */ |
1958593e | 840 | if (f2fs_encrypted_file(inode)) { |
4375a336 JK |
841 | add_gc_inode(gc_list, inode); |
842 | continue; | |
843 | } | |
844 | ||
bb06664a CY |
845 | if (!down_write_trylock( |
846 | &F2FS_I(inode)->dio_rwsem[WRITE])) { | |
847 | iput(inode); | |
848 | continue; | |
849 | } | |
850 | ||
81ca7350 | 851 | start_bidx = start_bidx_of_node(nofs, inode); |
43f3eae1 | 852 | data_page = get_read_data_page(inode, |
70246286 CH |
853 | start_bidx + ofs_in_node, REQ_RAHEAD, |
854 | true); | |
bb06664a | 855 | up_write(&F2FS_I(inode)->dio_rwsem[WRITE]); |
31a32688 CL |
856 | if (IS_ERR(data_page)) { |
857 | iput(inode); | |
858 | continue; | |
859 | } | |
7bc09003 JK |
860 | |
861 | f2fs_put_page(data_page, 0); | |
7dda2af8 | 862 | add_gc_inode(gc_list, inode); |
31a32688 CL |
863 | continue; |
864 | } | |
865 | ||
7ea984b0 | 866 | /* phase 4 */ |
7dda2af8 | 867 | inode = find_gc_inode(gc_list, dni.ino); |
31a32688 | 868 | if (inode) { |
82e0a5aa CY |
869 | struct f2fs_inode_info *fi = F2FS_I(inode); |
870 | bool locked = false; | |
871 | ||
872 | if (S_ISREG(inode->i_mode)) { | |
873 | if (!down_write_trylock(&fi->dio_rwsem[READ])) | |
874 | continue; | |
875 | if (!down_write_trylock( | |
876 | &fi->dio_rwsem[WRITE])) { | |
877 | up_write(&fi->dio_rwsem[READ]); | |
878 | continue; | |
879 | } | |
880 | locked = true; | |
73ac2f4e CY |
881 | |
882 | /* wait for all inflight aio data */ | |
883 | inode_dio_wait(inode); | |
82e0a5aa CY |
884 | } |
885 | ||
81ca7350 | 886 | start_bidx = start_bidx_of_node(nofs, inode) |
c879f90d | 887 | + ofs_in_node; |
1958593e | 888 | if (f2fs_encrypted_file(inode)) |
d4c759ee | 889 | move_data_block(inode, start_bidx, segno, off); |
4375a336 | 890 | else |
d4c759ee JK |
891 | move_data_page(inode, start_bidx, gc_type, |
892 | segno, off); | |
82e0a5aa CY |
893 | |
894 | if (locked) { | |
895 | up_write(&fi->dio_rwsem[WRITE]); | |
896 | up_write(&fi->dio_rwsem[READ]); | |
897 | } | |
898 | ||
e1235983 | 899 | stat_inc_data_blk_count(sbi, 1, gc_type); |
7bc09003 | 900 | } |
7bc09003 | 901 | } |
c718379b | 902 | |
7ea984b0 | 903 | if (++phase < 5) |
7bc09003 | 904 | goto next_step; |
7bc09003 JK |
905 | } |
906 | ||
907 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
8a2d0ace | 908 | int gc_type) |
7bc09003 JK |
909 | { |
910 | struct sit_info *sit_i = SIT_I(sbi); | |
911 | int ret; | |
8a2d0ace | 912 | |
3d26fa6b | 913 | down_write(&sit_i->sentry_lock); |
8a2d0ace GZ |
914 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, |
915 | NO_CHECK_TYPE, LFS); | |
3d26fa6b | 916 | up_write(&sit_i->sentry_lock); |
7bc09003 JK |
917 | return ret; |
918 | } | |
919 | ||
718e53fa CY |
920 | static int do_garbage_collect(struct f2fs_sb_info *sbi, |
921 | unsigned int start_segno, | |
7dda2af8 | 922 | struct gc_inode_list *gc_list, int gc_type) |
7bc09003 JK |
923 | { |
924 | struct page *sum_page; | |
925 | struct f2fs_summary_block *sum; | |
c718379b | 926 | struct blk_plug plug; |
718e53fa CY |
927 | unsigned int segno = start_segno; |
928 | unsigned int end_segno = start_segno + sbi->segs_per_sec; | |
c56f16da | 929 | int seg_freed = 0; |
718e53fa CY |
930 | unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ? |
931 | SUM_TYPE_DATA : SUM_TYPE_NODE; | |
7bc09003 | 932 | |
718e53fa CY |
933 | /* readahead multi ssa blocks those have contiguous address */ |
934 | if (sbi->segs_per_sec > 1) | |
935 | ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), | |
936 | sbi->segs_per_sec, META_SSA, true); | |
937 | ||
938 | /* reference all summary page */ | |
939 | while (segno < end_segno) { | |
940 | sum_page = get_sum_page(sbi, segno++); | |
941 | unlock_page(sum_page); | |
942 | } | |
7bc09003 | 943 | |
c718379b JK |
944 | blk_start_plug(&plug); |
945 | ||
718e53fa | 946 | for (segno = start_segno; segno < end_segno; segno++) { |
aa987273 | 947 | |
718e53fa CY |
948 | /* find segment summary of victim */ |
949 | sum_page = find_get_page(META_MAPPING(sbi), | |
950 | GET_SUM_BLOCK(sbi, segno)); | |
718e53fa | 951 | f2fs_put_page(sum_page, 0); |
7bc09003 | 952 | |
d0cc7e9e YS |
953 | if (get_valid_blocks(sbi, segno, false) == 0) |
954 | goto freed; | |
955 | if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi))) | |
de0dcc40 JK |
956 | goto next; |
957 | ||
718e53fa | 958 | sum = page_address(sum_page); |
58a1d469 CY |
959 | if (type != GET_SUM_TYPE((&sum->footer))) { |
960 | f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent segment (%u) " | |
961 | "type [%d, %d] in SSA and SIT", | |
962 | segno, type, GET_SUM_TYPE((&sum->footer))); | |
963 | set_sbi_flag(sbi, SBI_NEED_FSCK); | |
964 | goto next; | |
965 | } | |
718e53fa CY |
966 | |
967 | /* | |
968 | * this is to avoid deadlock: | |
969 | * - lock_page(sum_page) - f2fs_replace_block | |
3d26fa6b CY |
970 | * - check_valid_map() - down_write(sentry_lock) |
971 | * - down_read(sentry_lock) - change_curseg() | |
718e53fa CY |
972 | * - lock_page(sum_page) |
973 | */ | |
718e53fa CY |
974 | if (type == SUM_TYPE_NODE) |
975 | gc_node_segment(sbi, sum->entries, segno, gc_type); | |
976 | else | |
977 | gc_data_segment(sbi, sum->entries, gc_list, segno, | |
978 | gc_type); | |
979 | ||
980 | stat_inc_seg_count(sbi, type, gc_type); | |
c56f16da | 981 | |
d0cc7e9e | 982 | freed: |
c56f16da CY |
983 | if (gc_type == FG_GC && |
984 | get_valid_blocks(sbi, segno, false) == 0) | |
985 | seg_freed++; | |
f6fe2be3 | 986 | next: |
718e53fa CY |
987 | f2fs_put_page(sum_page, 0); |
988 | } | |
989 | ||
da011cc0 | 990 | if (gc_type == FG_GC) |
b9109b0e JK |
991 | f2fs_submit_merged_write(sbi, |
992 | (type == SUM_TYPE_NODE) ? NODE : DATA); | |
c718379b | 993 | |
718e53fa | 994 | blk_finish_plug(&plug); |
7bc09003 | 995 | |
17d899df CY |
996 | stat_inc_call_count(sbi->stat_info); |
997 | ||
c56f16da | 998 | return seg_freed; |
7bc09003 JK |
999 | } |
1000 | ||
e066b83c JK |
1001 | int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, |
1002 | bool background, unsigned int segno) | |
7bc09003 | 1003 | { |
d530d4d8 | 1004 | int gc_type = sync ? FG_GC : BG_GC; |
c56f16da CY |
1005 | int sec_freed = 0, seg_freed = 0, total_freed = 0; |
1006 | int ret = 0; | |
d5053a34 | 1007 | struct cp_control cpc; |
e066b83c | 1008 | unsigned int init_segno = segno; |
7dda2af8 CL |
1009 | struct gc_inode_list gc_list = { |
1010 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
769ec6e5 | 1011 | .iroot = RADIX_TREE_INIT(GFP_NOFS), |
7dda2af8 | 1012 | }; |
d5053a34 | 1013 | |
c56f16da CY |
1014 | trace_f2fs_gc_begin(sbi->sb, sync, background, |
1015 | get_pages(sbi, F2FS_DIRTY_NODES), | |
1016 | get_pages(sbi, F2FS_DIRTY_DENTS), | |
1017 | get_pages(sbi, F2FS_DIRTY_IMETA), | |
1018 | free_sections(sbi), | |
1019 | free_segments(sbi), | |
1020 | reserved_segments(sbi), | |
1021 | prefree_segments(sbi)); | |
1022 | ||
119ee914 | 1023 | cpc.reason = __get_cp_reason(sbi); |
7bc09003 | 1024 | gc_more: |
1751e8a6 | 1025 | if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) { |
e5dbd956 | 1026 | ret = -EINVAL; |
408e9375 | 1027 | goto stop; |
e5dbd956 | 1028 | } |
6d5a1495 CY |
1029 | if (unlikely(f2fs_cp_error(sbi))) { |
1030 | ret = -EIO; | |
203681f6 | 1031 | goto stop; |
6d5a1495 | 1032 | } |
7bc09003 | 1033 | |
19f4e688 | 1034 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) { |
6e17bfbc | 1035 | /* |
19f4e688 HP |
1036 | * For example, if there are many prefree_segments below given |
1037 | * threshold, we can make them free by checkpoint. Then, we | |
1038 | * secure free segments which doesn't need fggc any more. | |
6e17bfbc | 1039 | */ |
8fd5a37e JK |
1040 | if (prefree_segments(sbi)) { |
1041 | ret = write_checkpoint(sbi, &cpc); | |
1042 | if (ret) | |
1043 | goto stop; | |
1044 | } | |
19f4e688 HP |
1045 | if (has_not_enough_free_secs(sbi, 0, 0)) |
1046 | gc_type = FG_GC; | |
d64f8047 | 1047 | } |
7bc09003 | 1048 | |
19f4e688 | 1049 | /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */ |
c56f16da CY |
1050 | if (gc_type == BG_GC && !background) { |
1051 | ret = -EINVAL; | |
19f4e688 | 1052 | goto stop; |
c56f16da CY |
1053 | } |
1054 | if (!__get_victim(sbi, &segno, gc_type)) { | |
1055 | ret = -ENODATA; | |
408e9375 | 1056 | goto stop; |
c56f16da | 1057 | } |
7bc09003 | 1058 | |
c56f16da CY |
1059 | seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type); |
1060 | if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec) | |
45fe8492 | 1061 | sec_freed++; |
c56f16da | 1062 | total_freed += seg_freed; |
43727527 | 1063 | |
5ee5293c | 1064 | if (gc_type == FG_GC) |
5ec4e49f | 1065 | sbi->cur_victim_sec = NULL_SEGNO; |
43727527 | 1066 | |
d530d4d8 | 1067 | if (!sync) { |
e066b83c JK |
1068 | if (has_not_enough_free_secs(sbi, sec_freed, 0)) { |
1069 | segno = NULL_SEGNO; | |
d530d4d8 | 1070 | goto gc_more; |
e066b83c | 1071 | } |
43727527 | 1072 | |
d530d4d8 | 1073 | if (gc_type == FG_GC) |
2956e450 | 1074 | ret = write_checkpoint(sbi, &cpc); |
d530d4d8 | 1075 | } |
408e9375 | 1076 | stop: |
e066b83c JK |
1077 | SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0; |
1078 | SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno; | |
c56f16da CY |
1079 | |
1080 | trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed, | |
1081 | get_pages(sbi, F2FS_DIRTY_NODES), | |
1082 | get_pages(sbi, F2FS_DIRTY_DENTS), | |
1083 | get_pages(sbi, F2FS_DIRTY_IMETA), | |
1084 | free_sections(sbi), | |
1085 | free_segments(sbi), | |
1086 | reserved_segments(sbi), | |
1087 | prefree_segments(sbi)); | |
1088 | ||
7bc09003 JK |
1089 | mutex_unlock(&sbi->gc_mutex); |
1090 | ||
7dda2af8 | 1091 | put_gc_inode(&gc_list); |
d530d4d8 | 1092 | |
0ed34142 | 1093 | if (sync && !ret) |
d530d4d8 | 1094 | ret = sec_freed ? 0 : -EAGAIN; |
43727527 | 1095 | return ret; |
7bc09003 JK |
1096 | } |
1097 | ||
1098 | void build_gc_manager(struct f2fs_sb_info *sbi) | |
1099 | { | |
4ddb1a4d | 1100 | u64 main_count, resv_count, ovp_count; |
e93b9865 | 1101 | |
7bc09003 | 1102 | DIRTY_I(sbi)->v_ops = &default_v_ops; |
e93b9865 HP |
1103 | |
1104 | /* threshold of # of valid blocks in a section for victims of FG_GC */ | |
1105 | main_count = SM_I(sbi)->main_segments << sbi->log_blocks_per_seg; | |
1106 | resv_count = SM_I(sbi)->reserved_segments << sbi->log_blocks_per_seg; | |
1107 | ovp_count = SM_I(sbi)->ovp_segments << sbi->log_blocks_per_seg; | |
e93b9865 | 1108 | |
4ddb1a4d JK |
1109 | sbi->fggc_threshold = div64_u64((main_count - ovp_count) * |
1110 | BLKS_PER_SEC(sbi), (main_count - resv_count)); | |
d5793249 JK |
1111 | |
1112 | /* give warm/cold data area from slower device */ | |
2ee2eeed | 1113 | if (f2fs_is_multi_device(sbi) && sbi->segs_per_sec == 1) |
d5793249 JK |
1114 | SIT_I(sbi)->last_victim[ALLOC_NEXT] = |
1115 | GET_SEGNO(sbi, FDEV(0).end_blk) + 1; | |
7bc09003 | 1116 | } |