]>
Commit | Line | Data |
---|---|---|
7c1a000d | 1 | // SPDX-License-Identifier: GPL-2.0 |
0a8165d7 | 2 | /* |
7bc09003 JK |
3 | * fs/f2fs/gc.c |
4 | * | |
5 | * Copyright (c) 2012 Samsung Electronics Co., Ltd. | |
6 | * http://www.samsung.com/ | |
7bc09003 JK |
7 | */ |
8 | #include <linux/fs.h> | |
9 | #include <linux/module.h> | |
10 | #include <linux/backing-dev.h> | |
7bc09003 JK |
11 | #include <linux/init.h> |
12 | #include <linux/f2fs_fs.h> | |
13 | #include <linux/kthread.h> | |
14 | #include <linux/delay.h> | |
15 | #include <linux/freezer.h> | |
7bc09003 JK |
16 | |
17 | #include "f2fs.h" | |
18 | #include "node.h" | |
19 | #include "segment.h" | |
20 | #include "gc.h" | |
8e46b3ed | 21 | #include <trace/events/f2fs.h> |
7bc09003 | 22 | |
7bc09003 JK |
23 | static int gc_thread_func(void *data) |
24 | { | |
25 | struct f2fs_sb_info *sbi = data; | |
b59d0bae | 26 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; |
7bc09003 | 27 | wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head; |
b8c502b8 | 28 | unsigned int wait_ms; |
7bc09003 | 29 | |
b59d0bae | 30 | wait_ms = gc_th->min_sleep_time; |
7bc09003 | 31 | |
1d7be270 | 32 | set_freezable(); |
7bc09003 | 33 | do { |
1d7be270 | 34 | wait_event_interruptible_timeout(*wq, |
d9872a69 JK |
35 | kthread_should_stop() || freezing(current) || |
36 | gc_th->gc_wake, | |
1d7be270 JK |
37 | msecs_to_jiffies(wait_ms)); |
38 | ||
d9872a69 JK |
39 | /* give it a try one time */ |
40 | if (gc_th->gc_wake) | |
41 | gc_th->gc_wake = 0; | |
42 | ||
7bc09003 JK |
43 | if (try_to_freeze()) |
44 | continue; | |
7bc09003 JK |
45 | if (kthread_should_stop()) |
46 | break; | |
47 | ||
d6212a5f | 48 | if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) { |
88dd8934 | 49 | increase_sleep_time(gc_th, &wait_ms); |
d6212a5f CL |
50 | continue; |
51 | } | |
52 | ||
55523519 CY |
53 | if (time_to_inject(sbi, FAULT_CHECKPOINT)) { |
54 | f2fs_show_injection_info(FAULT_CHECKPOINT); | |
0f348028 | 55 | f2fs_stop_checkpoint(sbi, false); |
55523519 | 56 | } |
0f348028 | 57 | |
dc6febb6 CY |
58 | if (!sb_start_write_trylock(sbi->sb)) |
59 | continue; | |
60 | ||
7bc09003 JK |
61 | /* |
62 | * [GC triggering condition] | |
63 | * 0. GC is not conducted currently. | |
64 | * 1. There are enough dirty segments. | |
65 | * 2. IO subsystem is idle by checking the # of writeback pages. | |
66 | * 3. IO subsystem is idle by checking the # of requests in | |
67 | * bdev's request list. | |
68 | * | |
e1c42045 | 69 | * Note) We have to avoid triggering GCs frequently. |
7bc09003 JK |
70 | * Because it is possible that some segments can be |
71 | * invalidated soon after by user update or deletion. | |
72 | * So, I'd like to wait some time to collect dirty segments. | |
73 | */ | |
5b0e9539 | 74 | if (sbi->gc_mode == GC_URGENT) { |
d9872a69 | 75 | wait_ms = gc_th->urgent_sleep_time; |
69babac0 | 76 | mutex_lock(&sbi->gc_mutex); |
d9872a69 JK |
77 | goto do_gc; |
78 | } | |
79 | ||
69babac0 JK |
80 | if (!mutex_trylock(&sbi->gc_mutex)) |
81 | goto next; | |
82 | ||
a7d10cf3 | 83 | if (!is_idle(sbi, GC_TIME)) { |
88dd8934 | 84 | increase_sleep_time(gc_th, &wait_ms); |
7bc09003 | 85 | mutex_unlock(&sbi->gc_mutex); |
dc6febb6 | 86 | goto next; |
7bc09003 JK |
87 | } |
88 | ||
89 | if (has_enough_invalid_blocks(sbi)) | |
88dd8934 | 90 | decrease_sleep_time(gc_th, &wait_ms); |
7bc09003 | 91 | else |
88dd8934 | 92 | increase_sleep_time(gc_th, &wait_ms); |
d9872a69 | 93 | do_gc: |
dcdfff65 | 94 | stat_inc_bggc_count(sbi); |
7bc09003 | 95 | |
43727527 | 96 | /* if return value is not zero, no victim was selected */ |
e066b83c | 97 | if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true, NULL_SEGNO)) |
b59d0bae | 98 | wait_ms = gc_th->no_gc_sleep_time; |
81eb8d6e | 99 | |
84e4214f JK |
100 | trace_f2fs_background_gc(sbi->sb, wait_ms, |
101 | prefree_segments(sbi), free_segments(sbi)); | |
102 | ||
4660f9c0 JK |
103 | /* balancing f2fs's metadata periodically */ |
104 | f2fs_balance_fs_bg(sbi); | |
dc6febb6 CY |
105 | next: |
106 | sb_end_write(sbi->sb); | |
81eb8d6e | 107 | |
7bc09003 JK |
108 | } while (!kthread_should_stop()); |
109 | return 0; | |
110 | } | |
111 | ||
4d57b86d | 112 | int f2fs_start_gc_thread(struct f2fs_sb_info *sbi) |
7bc09003 | 113 | { |
1042d60f | 114 | struct f2fs_gc_kthread *gc_th; |
ec7b1f2d | 115 | dev_t dev = sbi->sb->s_bdev->bd_dev; |
7a267f8d | 116 | int err = 0; |
7bc09003 | 117 | |
1ecc0c5c | 118 | gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL); |
7a267f8d NJ |
119 | if (!gc_th) { |
120 | err = -ENOMEM; | |
121 | goto out; | |
122 | } | |
7bc09003 | 123 | |
d9872a69 | 124 | gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME; |
b59d0bae NJ |
125 | gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; |
126 | gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; | |
127 | gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; | |
128 | ||
d9872a69 | 129 | gc_th->gc_wake= 0; |
d2dc095f | 130 | |
7bc09003 JK |
131 | sbi->gc_thread = gc_th; |
132 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
133 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, | |
ec7b1f2d | 134 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 | 135 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
7a267f8d | 136 | err = PTR_ERR(gc_th->f2fs_gc_task); |
7bc09003 | 137 | kfree(gc_th); |
25718423 | 138 | sbi->gc_thread = NULL; |
7bc09003 | 139 | } |
7a267f8d NJ |
140 | out: |
141 | return err; | |
7bc09003 JK |
142 | } |
143 | ||
4d57b86d | 144 | void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi) |
7bc09003 JK |
145 | { |
146 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
147 | if (!gc_th) | |
148 | return; | |
149 | kthread_stop(gc_th->f2fs_gc_task); | |
150 | kfree(gc_th); | |
151 | sbi->gc_thread = NULL; | |
152 | } | |
153 | ||
5b0e9539 | 154 | static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type) |
7bc09003 | 155 | { |
d2dc095f NJ |
156 | int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY; |
157 | ||
5b0e9539 JK |
158 | switch (sbi->gc_mode) { |
159 | case GC_IDLE_CB: | |
160 | gc_mode = GC_CB; | |
161 | break; | |
162 | case GC_IDLE_GREEDY: | |
163 | case GC_URGENT: | |
b27bc809 | 164 | gc_mode = GC_GREEDY; |
5b0e9539 JK |
165 | break; |
166 | } | |
d2dc095f | 167 | return gc_mode; |
7bc09003 JK |
168 | } |
169 | ||
170 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
171 | int type, struct victim_sel_policy *p) | |
172 | { | |
173 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
174 | ||
4ebefc44 | 175 | if (p->alloc_mode == SSR) { |
7bc09003 JK |
176 | p->gc_mode = GC_GREEDY; |
177 | p->dirty_segmap = dirty_i->dirty_segmap[type]; | |
a26b7c8a | 178 | p->max_search = dirty_i->nr_dirty[type]; |
7bc09003 JK |
179 | p->ofs_unit = 1; |
180 | } else { | |
5b0e9539 | 181 | p->gc_mode = select_gc_type(sbi, gc_type); |
7bc09003 | 182 | p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; |
a26b7c8a | 183 | p->max_search = dirty_i->nr_dirty[DIRTY]; |
7bc09003 JK |
184 | p->ofs_unit = sbi->segs_per_sec; |
185 | } | |
a26b7c8a | 186 | |
e93b9865 | 187 | /* we need to check every dirty segments in the FG_GC case */ |
b27bc809 | 188 | if (gc_type != FG_GC && |
5b0e9539 | 189 | (sbi->gc_mode != GC_URGENT) && |
b27bc809 | 190 | p->max_search > sbi->max_victim_search) |
b1c57c1c | 191 | p->max_search = sbi->max_victim_search; |
a26b7c8a | 192 | |
b94929d9 YS |
193 | /* let's select beginning hot/small space first in no_heap mode*/ |
194 | if (test_opt(sbi, NOHEAP) && | |
195 | (type == CURSEG_HOT_DATA || IS_NODESEG(type))) | |
7a20b8a6 JK |
196 | p->offset = 0; |
197 | else | |
e066b83c | 198 | p->offset = SIT_I(sbi)->last_victim[p->gc_mode]; |
7bc09003 JK |
199 | } |
200 | ||
201 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
202 | struct victim_sel_policy *p) | |
203 | { | |
b7250d2d JK |
204 | /* SSR allocates in a segment unit */ |
205 | if (p->alloc_mode == SSR) | |
3519e3f9 | 206 | return sbi->blocks_per_seg; |
7bc09003 | 207 | if (p->gc_mode == GC_GREEDY) |
c541a51b | 208 | return 2 * sbi->blocks_per_seg * p->ofs_unit; |
7bc09003 JK |
209 | else if (p->gc_mode == GC_CB) |
210 | return UINT_MAX; | |
211 | else /* No other gc_mode */ | |
212 | return 0; | |
213 | } | |
214 | ||
215 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
216 | { | |
217 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 218 | unsigned int secno; |
7bc09003 JK |
219 | |
220 | /* | |
221 | * If the gc_type is FG_GC, we can select victim segments | |
222 | * selected by background GC before. | |
223 | * Those segments guarantee they have small valid blocks. | |
224 | */ | |
7cd8558b | 225 | for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { |
5ec4e49f | 226 | if (sec_usage_check(sbi, secno)) |
b65ee148 | 227 | continue; |
5ec4e49f | 228 | clear_bit(secno, dirty_i->victim_secmap); |
4ddb1a4d | 229 | return GET_SEG_FROM_SEC(sbi, secno); |
7bc09003 JK |
230 | } |
231 | return NULL_SEGNO; | |
232 | } | |
233 | ||
234 | static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno) | |
235 | { | |
236 | struct sit_info *sit_i = SIT_I(sbi); | |
4ddb1a4d JK |
237 | unsigned int secno = GET_SEC_FROM_SEG(sbi, segno); |
238 | unsigned int start = GET_SEG_FROM_SEC(sbi, secno); | |
7bc09003 JK |
239 | unsigned long long mtime = 0; |
240 | unsigned int vblocks; | |
241 | unsigned char age = 0; | |
242 | unsigned char u; | |
243 | unsigned int i; | |
244 | ||
245 | for (i = 0; i < sbi->segs_per_sec; i++) | |
246 | mtime += get_seg_entry(sbi, start + i)->mtime; | |
302bd348 | 247 | vblocks = get_valid_blocks(sbi, segno, true); |
7bc09003 JK |
248 | |
249 | mtime = div_u64(mtime, sbi->segs_per_sec); | |
250 | vblocks = div_u64(vblocks, sbi->segs_per_sec); | |
251 | ||
252 | u = (vblocks * 100) >> sbi->log_blocks_per_seg; | |
253 | ||
e1c42045 | 254 | /* Handle if the system time has changed by the user */ |
7bc09003 JK |
255 | if (mtime < sit_i->min_mtime) |
256 | sit_i->min_mtime = mtime; | |
257 | if (mtime > sit_i->max_mtime) | |
258 | sit_i->max_mtime = mtime; | |
259 | if (sit_i->max_mtime != sit_i->min_mtime) | |
260 | age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime), | |
261 | sit_i->max_mtime - sit_i->min_mtime); | |
262 | ||
263 | return UINT_MAX - ((100 * (100 - u) * age) / (100 + u)); | |
264 | } | |
265 | ||
a57e564d JX |
266 | static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi, |
267 | unsigned int segno, struct victim_sel_policy *p) | |
7bc09003 JK |
268 | { |
269 | if (p->alloc_mode == SSR) | |
2afce76a | 270 | return get_seg_entry(sbi, segno)->ckpt_valid_blocks; |
7bc09003 JK |
271 | |
272 | /* alloc_mode == LFS */ | |
273 | if (p->gc_mode == GC_GREEDY) | |
91f4382b | 274 | return get_valid_blocks(sbi, segno, true); |
7bc09003 JK |
275 | else |
276 | return get_cb_cost(sbi, segno); | |
277 | } | |
278 | ||
688159b6 FL |
279 | static unsigned int count_bits(const unsigned long *addr, |
280 | unsigned int offset, unsigned int len) | |
281 | { | |
282 | unsigned int end = offset + len, sum = 0; | |
283 | ||
284 | while (offset < end) { | |
285 | if (test_bit(offset++, addr)) | |
286 | ++sum; | |
287 | } | |
288 | return sum; | |
289 | } | |
290 | ||
0a8165d7 | 291 | /* |
111d2495 | 292 | * This function is called from two paths. |
7bc09003 JK |
293 | * One is garbage collection and the other is SSR segment selection. |
294 | * When it is called during GC, it just gets a victim segment | |
295 | * and it does not remove it from dirty seglist. | |
296 | * When it is called from SSR segment selection, it finds a segment | |
297 | * which has minimum valid blocks and removes it from dirty seglist. | |
298 | */ | |
299 | static int get_victim_by_default(struct f2fs_sb_info *sbi, | |
300 | unsigned int *result, int gc_type, int type, char alloc_mode) | |
301 | { | |
302 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
e066b83c | 303 | struct sit_info *sm = SIT_I(sbi); |
7bc09003 | 304 | struct victim_sel_policy p; |
3fa56503 | 305 | unsigned int secno, last_victim; |
a43f7ec3 | 306 | unsigned int last_segment = MAIN_SEGS(sbi); |
688159b6 | 307 | unsigned int nsearched = 0; |
7bc09003 | 308 | |
210f41bc CY |
309 | mutex_lock(&dirty_i->seglist_lock); |
310 | ||
7bc09003 JK |
311 | p.alloc_mode = alloc_mode; |
312 | select_policy(sbi, gc_type, type, &p); | |
313 | ||
314 | p.min_segno = NULL_SEGNO; | |
3fa56503 | 315 | p.min_cost = get_max_cost(sbi, &p); |
7bc09003 | 316 | |
e066b83c JK |
317 | if (*result != NULL_SEGNO) { |
318 | if (IS_DATASEG(get_seg_entry(sbi, *result)->type) && | |
319 | get_valid_blocks(sbi, *result, false) && | |
320 | !sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result))) | |
321 | p.min_segno = *result; | |
322 | goto out; | |
323 | } | |
324 | ||
3342bb30 CY |
325 | if (p.max_search == 0) |
326 | goto out; | |
327 | ||
e066b83c | 328 | last_victim = sm->last_victim[p.gc_mode]; |
7bc09003 JK |
329 | if (p.alloc_mode == LFS && gc_type == FG_GC) { |
330 | p.min_segno = check_bg_victims(sbi); | |
331 | if (p.min_segno != NULL_SEGNO) | |
332 | goto got_it; | |
333 | } | |
334 | ||
335 | while (1) { | |
336 | unsigned long cost; | |
5ec4e49f | 337 | unsigned int segno; |
7bc09003 | 338 | |
a43f7ec3 CY |
339 | segno = find_next_bit(p.dirty_segmap, last_segment, p.offset); |
340 | if (segno >= last_segment) { | |
e066b83c JK |
341 | if (sm->last_victim[p.gc_mode]) { |
342 | last_segment = | |
343 | sm->last_victim[p.gc_mode]; | |
344 | sm->last_victim[p.gc_mode] = 0; | |
7bc09003 JK |
345 | p.offset = 0; |
346 | continue; | |
347 | } | |
348 | break; | |
349 | } | |
a57e564d JX |
350 | |
351 | p.offset = segno + p.ofs_unit; | |
688159b6 | 352 | if (p.ofs_unit > 1) { |
a57e564d | 353 | p.offset -= segno % p.ofs_unit; |
688159b6 FL |
354 | nsearched += count_bits(p.dirty_segmap, |
355 | p.offset - p.ofs_unit, | |
356 | p.ofs_unit); | |
357 | } else { | |
358 | nsearched++; | |
359 | } | |
360 | ||
4ddb1a4d | 361 | secno = GET_SEC_FROM_SEG(sbi, segno); |
7bc09003 | 362 | |
5ec4e49f | 363 | if (sec_usage_check(sbi, secno)) |
688159b6 | 364 | goto next; |
5ec4e49f | 365 | if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap)) |
688159b6 | 366 | goto next; |
7bc09003 JK |
367 | |
368 | cost = get_gc_cost(sbi, segno, &p); | |
369 | ||
370 | if (p.min_cost > cost) { | |
371 | p.min_segno = segno; | |
372 | p.min_cost = cost; | |
a57e564d | 373 | } |
688159b6 FL |
374 | next: |
375 | if (nsearched >= p.max_search) { | |
e066b83c JK |
376 | if (!sm->last_victim[p.gc_mode] && segno <= last_victim) |
377 | sm->last_victim[p.gc_mode] = last_victim + 1; | |
4ce53776 | 378 | else |
e066b83c JK |
379 | sm->last_victim[p.gc_mode] = segno + 1; |
380 | sm->last_victim[p.gc_mode] %= MAIN_SEGS(sbi); | |
7bc09003 JK |
381 | break; |
382 | } | |
383 | } | |
7bc09003 | 384 | if (p.min_segno != NULL_SEGNO) { |
b2b3460a | 385 | got_it: |
7bc09003 | 386 | if (p.alloc_mode == LFS) { |
4ddb1a4d | 387 | secno = GET_SEC_FROM_SEG(sbi, p.min_segno); |
5ec4e49f JK |
388 | if (gc_type == FG_GC) |
389 | sbi->cur_victim_sec = secno; | |
390 | else | |
391 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 392 | } |
5ec4e49f | 393 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
8e46b3ed NJ |
394 | |
395 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, | |
396 | sbi->cur_victim_sec, | |
397 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 | 398 | } |
3342bb30 | 399 | out: |
7bc09003 JK |
400 | mutex_unlock(&dirty_i->seglist_lock); |
401 | ||
402 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
403 | } | |
404 | ||
405 | static const struct victim_selection default_v_ops = { | |
406 | .get_victim = get_victim_by_default, | |
407 | }; | |
408 | ||
7dda2af8 | 409 | static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) |
7bc09003 | 410 | { |
7bc09003 JK |
411 | struct inode_entry *ie; |
412 | ||
7dda2af8 CL |
413 | ie = radix_tree_lookup(&gc_list->iroot, ino); |
414 | if (ie) | |
415 | return ie->inode; | |
7bc09003 JK |
416 | return NULL; |
417 | } | |
418 | ||
7dda2af8 | 419 | static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) |
7bc09003 | 420 | { |
6cc4af56 GZ |
421 | struct inode_entry *new_ie; |
422 | ||
7dda2af8 | 423 | if (inode == find_gc_inode(gc_list, inode->i_ino)) { |
6cc4af56 GZ |
424 | iput(inode); |
425 | return; | |
7bc09003 | 426 | } |
4d57b86d | 427 | new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS); |
7bc09003 | 428 | new_ie->inode = inode; |
f28e5034 CY |
429 | |
430 | f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); | |
7dda2af8 | 431 | list_add_tail(&new_ie->list, &gc_list->ilist); |
7bc09003 JK |
432 | } |
433 | ||
7dda2af8 | 434 | static void put_gc_inode(struct gc_inode_list *gc_list) |
7bc09003 JK |
435 | { |
436 | struct inode_entry *ie, *next_ie; | |
7dda2af8 CL |
437 | list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { |
438 | radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); | |
7bc09003 JK |
439 | iput(ie->inode); |
440 | list_del(&ie->list); | |
4d57b86d | 441 | kmem_cache_free(f2fs_inode_entry_slab, ie); |
7bc09003 JK |
442 | } |
443 | } | |
444 | ||
445 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
446 | unsigned int segno, int offset) | |
447 | { | |
448 | struct sit_info *sit_i = SIT_I(sbi); | |
449 | struct seg_entry *sentry; | |
450 | int ret; | |
451 | ||
3d26fa6b | 452 | down_read(&sit_i->sentry_lock); |
7bc09003 JK |
453 | sentry = get_seg_entry(sbi, segno); |
454 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
3d26fa6b | 455 | up_read(&sit_i->sentry_lock); |
43727527 | 456 | return ret; |
7bc09003 JK |
457 | } |
458 | ||
0a8165d7 | 459 | /* |
7bc09003 JK |
460 | * This function compares node address got in summary with that in NAT. |
461 | * On validity, copy that node with cold status, otherwise (invalid node) | |
462 | * ignore that. | |
463 | */ | |
718e53fa | 464 | static void gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
465 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
466 | { | |
7bc09003 | 467 | struct f2fs_summary *entry; |
26d58599 | 468 | block_t start_addr; |
7bc09003 | 469 | int off; |
7ea984b0 | 470 | int phase = 0; |
c29fd0c0 | 471 | bool fggc = (gc_type == FG_GC); |
7bc09003 | 472 | |
26d58599 JK |
473 | start_addr = START_BLOCK(sbi, segno); |
474 | ||
7bc09003 JK |
475 | next_step: |
476 | entry = sum; | |
c718379b | 477 | |
c29fd0c0 CY |
478 | if (fggc && phase == 2) |
479 | atomic_inc(&sbi->wb_sync_req[NODE]); | |
480 | ||
7bc09003 JK |
481 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
482 | nid_t nid = le32_to_cpu(entry->nid); | |
483 | struct page *node_page; | |
26d58599 | 484 | struct node_info ni; |
7bc09003 | 485 | |
43727527 | 486 | /* stop BG_GC if there is not enough free sections. */ |
7f3037a5 | 487 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) |
718e53fa | 488 | return; |
7bc09003 | 489 | |
43727527 | 490 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
491 | continue; |
492 | ||
7ea984b0 | 493 | if (phase == 0) { |
4d57b86d | 494 | f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, |
7ea984b0 CY |
495 | META_NAT, true); |
496 | continue; | |
497 | } | |
498 | ||
499 | if (phase == 1) { | |
4d57b86d | 500 | f2fs_ra_node_page(sbi, nid); |
7bc09003 JK |
501 | continue; |
502 | } | |
7ea984b0 CY |
503 | |
504 | /* phase == 2 */ | |
4d57b86d | 505 | node_page = f2fs_get_node_page(sbi, nid); |
7bc09003 JK |
506 | if (IS_ERR(node_page)) |
507 | continue; | |
508 | ||
4d57b86d | 509 | /* block may become invalid during f2fs_get_node_page */ |
9a01b56b HY |
510 | if (check_valid_map(sbi, segno, off) == 0) { |
511 | f2fs_put_page(node_page, 1); | |
512 | continue; | |
26d58599 JK |
513 | } |
514 | ||
7735730d CY |
515 | if (f2fs_get_node_info(sbi, nid, &ni)) { |
516 | f2fs_put_page(node_page, 1); | |
517 | continue; | |
518 | } | |
519 | ||
26d58599 JK |
520 | if (ni.blk_addr != start_addr + off) { |
521 | f2fs_put_page(node_page, 1); | |
522 | continue; | |
9a01b56b HY |
523 | } |
524 | ||
4d57b86d | 525 | f2fs_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; |
c29fd0c0 CY |
531 | |
532 | if (fggc) | |
533 | atomic_dec(&sbi->wb_sync_req[NODE]); | |
7bc09003 JK |
534 | } |
535 | ||
0a8165d7 | 536 | /* |
9af45ef5 JK |
537 | * Calculate start block index indicating the given node offset. |
538 | * Be careful, caller should give this node offset only indicating direct node | |
539 | * blocks. If any node offsets, which point the other types of node blocks such | |
540 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
541 | * bug. | |
7bc09003 | 542 | */ |
4d57b86d | 543 | block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode) |
7bc09003 | 544 | { |
ce19a5d4 JK |
545 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
546 | unsigned int bidx; | |
7bc09003 | 547 | |
ce19a5d4 JK |
548 | if (node_ofs == 0) |
549 | return 0; | |
7bc09003 | 550 | |
ce19a5d4 | 551 | if (node_ofs <= 2) { |
7bc09003 JK |
552 | bidx = node_ofs - 1; |
553 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 554 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
555 | bidx = node_ofs - 2 - dec; |
556 | } else { | |
ce19a5d4 | 557 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
558 | bidx = node_ofs - 5 - dec; |
559 | } | |
81ca7350 | 560 | return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode); |
7bc09003 JK |
561 | } |
562 | ||
c1079892 | 563 | static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
564 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) |
565 | { | |
566 | struct page *node_page; | |
567 | nid_t nid; | |
568 | unsigned int ofs_in_node; | |
569 | block_t source_blkaddr; | |
570 | ||
571 | nid = le32_to_cpu(sum->nid); | |
572 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
573 | ||
4d57b86d | 574 | node_page = f2fs_get_node_page(sbi, nid); |
7bc09003 | 575 | if (IS_ERR(node_page)) |
c1079892 | 576 | return false; |
7bc09003 | 577 | |
7735730d CY |
578 | if (f2fs_get_node_info(sbi, nid, dni)) { |
579 | f2fs_put_page(node_page, 1); | |
580 | return false; | |
581 | } | |
7bc09003 JK |
582 | |
583 | if (sum->version != dni->version) { | |
c13ff37e JK |
584 | f2fs_msg(sbi->sb, KERN_WARNING, |
585 | "%s: valid data with mismatched node version.", | |
586 | __func__); | |
587 | set_sbi_flag(sbi, SBI_NEED_FSCK); | |
7bc09003 JK |
588 | } |
589 | ||
590 | *nofs = ofs_of_node(node_page); | |
7a2af766 | 591 | source_blkaddr = datablock_addr(NULL, node_page, ofs_in_node); |
7bc09003 JK |
592 | f2fs_put_page(node_page, 1); |
593 | ||
594 | if (source_blkaddr != blkaddr) | |
c1079892 NK |
595 | return false; |
596 | return true; | |
7bc09003 JK |
597 | } |
598 | ||
6aa58d8a CY |
599 | static int ra_data_block(struct inode *inode, pgoff_t index) |
600 | { | |
601 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); | |
602 | struct address_space *mapping = inode->i_mapping; | |
603 | struct dnode_of_data dn; | |
604 | struct page *page; | |
605 | struct extent_info ei = {0, 0, 0}; | |
606 | struct f2fs_io_info fio = { | |
607 | .sbi = sbi, | |
608 | .ino = inode->i_ino, | |
609 | .type = DATA, | |
610 | .temp = COLD, | |
611 | .op = REQ_OP_READ, | |
612 | .op_flags = 0, | |
613 | .encrypted_page = NULL, | |
614 | .in_list = false, | |
615 | .retry = false, | |
616 | }; | |
617 | int err; | |
618 | ||
619 | page = f2fs_grab_cache_page(mapping, index, true); | |
620 | if (!page) | |
621 | return -ENOMEM; | |
622 | ||
623 | if (f2fs_lookup_extent_cache(inode, index, &ei)) { | |
624 | dn.data_blkaddr = ei.blk + index - ei.fofs; | |
625 | goto got_it; | |
626 | } | |
627 | ||
628 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
629 | err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); | |
630 | if (err) | |
631 | goto put_page; | |
632 | f2fs_put_dnode(&dn); | |
633 | ||
634 | if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, | |
635 | DATA_GENERIC))) { | |
636 | err = -EFAULT; | |
637 | goto put_page; | |
638 | } | |
639 | got_it: | |
640 | /* read page */ | |
641 | fio.page = page; | |
642 | fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; | |
643 | ||
644 | fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi), | |
645 | dn.data_blkaddr, | |
646 | FGP_LOCK | FGP_CREAT, GFP_NOFS); | |
647 | if (!fio.encrypted_page) { | |
648 | err = -ENOMEM; | |
649 | goto put_page; | |
650 | } | |
651 | ||
652 | err = f2fs_submit_page_bio(&fio); | |
653 | if (err) | |
654 | goto put_encrypted_page; | |
655 | f2fs_put_page(fio.encrypted_page, 0); | |
656 | f2fs_put_page(page, 1); | |
657 | return 0; | |
658 | put_encrypted_page: | |
659 | f2fs_put_page(fio.encrypted_page, 1); | |
660 | put_page: | |
661 | f2fs_put_page(page, 1); | |
662 | return err; | |
663 | } | |
664 | ||
d4c759ee JK |
665 | /* |
666 | * Move data block via META_MAPPING while keeping locked data page. | |
667 | * This can be used to move blocks, aka LBAs, directly on disk. | |
668 | */ | |
669 | static void move_data_block(struct inode *inode, block_t bidx, | |
2ef79ecb | 670 | int gc_type, unsigned int segno, int off) |
4375a336 JK |
671 | { |
672 | struct f2fs_io_info fio = { | |
673 | .sbi = F2FS_I_SB(inode), | |
39d787be | 674 | .ino = inode->i_ino, |
4375a336 | 675 | .type = DATA, |
a912b54d | 676 | .temp = COLD, |
04d328de | 677 | .op = REQ_OP_READ, |
70fd7614 | 678 | .op_flags = 0, |
4375a336 | 679 | .encrypted_page = NULL, |
fb830fc5 | 680 | .in_list = false, |
fe16efe6 | 681 | .retry = false, |
4375a336 JK |
682 | }; |
683 | struct dnode_of_data dn; | |
684 | struct f2fs_summary sum; | |
685 | struct node_info ni; | |
6aa58d8a | 686 | struct page *page, *mpage; |
4356e48e | 687 | block_t newaddr; |
4375a336 | 688 | int err; |
107a805d | 689 | bool lfs_mode = test_opt(fio.sbi, LFS); |
4375a336 JK |
690 | |
691 | /* do not read out */ | |
a56c7c6f | 692 | page = f2fs_grab_cache_page(inode->i_mapping, bidx, false); |
4375a336 JK |
693 | if (!page) |
694 | return; | |
695 | ||
20614711 YH |
696 | if (!check_valid_map(F2FS_I_SB(inode), segno, off)) |
697 | goto out; | |
698 | ||
2ef79ecb CY |
699 | if (f2fs_is_atomic_file(inode)) { |
700 | F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++; | |
701 | F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++; | |
5fe45743 | 702 | goto out; |
2ef79ecb | 703 | } |
5fe45743 | 704 | |
1ad71a27 JK |
705 | if (f2fs_is_pinned_file(inode)) { |
706 | f2fs_pin_file_control(inode, true); | |
707 | goto out; | |
708 | } | |
709 | ||
4375a336 | 710 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
4d57b86d | 711 | err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE); |
4375a336 JK |
712 | if (err) |
713 | goto out; | |
714 | ||
08b39fbd CY |
715 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) { |
716 | ClearPageUptodate(page); | |
4375a336 | 717 | goto put_out; |
08b39fbd CY |
718 | } |
719 | ||
720 | /* | |
721 | * don't cache encrypted data into meta inode until previous dirty | |
722 | * data were writebacked to avoid racing between GC and flush. | |
723 | */ | |
fec1d657 | 724 | f2fs_wait_on_page_writeback(page, DATA, true); |
4375a336 | 725 | |
7735730d CY |
726 | err = f2fs_get_node_info(fio.sbi, dn.nid, &ni); |
727 | if (err) | |
728 | goto put_out; | |
729 | ||
4375a336 JK |
730 | set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); |
731 | ||
732 | /* read page */ | |
733 | fio.page = page; | |
7a9d7548 | 734 | fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; |
4375a336 | 735 | |
107a805d CY |
736 | if (lfs_mode) |
737 | down_write(&fio.sbi->io_order_lock); | |
738 | ||
4d57b86d | 739 | f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr, |
fb830fc5 | 740 | &sum, CURSEG_COLD_DATA, NULL, false); |
4356e48e | 741 | |
01eccef7 CY |
742 | fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi), |
743 | newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS); | |
4356e48e CY |
744 | if (!fio.encrypted_page) { |
745 | err = -ENOMEM; | |
746 | goto recover_block; | |
747 | } | |
4375a336 | 748 | |
6aa58d8a CY |
749 | mpage = f2fs_pagecache_get_page(META_MAPPING(fio.sbi), |
750 | fio.old_blkaddr, FGP_LOCK, GFP_NOFS); | |
751 | if (mpage) { | |
752 | bool updated = false; | |
753 | ||
754 | if (PageUptodate(mpage)) { | |
755 | memcpy(page_address(fio.encrypted_page), | |
756 | page_address(mpage), PAGE_SIZE); | |
757 | updated = true; | |
758 | } | |
759 | f2fs_put_page(mpage, 1); | |
760 | invalidate_mapping_pages(META_MAPPING(fio.sbi), | |
761 | fio.old_blkaddr, fio.old_blkaddr); | |
762 | if (updated) | |
763 | goto write_page; | |
764 | } | |
765 | ||
548aedac JK |
766 | err = f2fs_submit_page_bio(&fio); |
767 | if (err) | |
768 | goto put_page_out; | |
769 | ||
770 | /* write page */ | |
771 | lock_page(fio.encrypted_page); | |
772 | ||
1563ac75 | 773 | if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) { |
4356e48e | 774 | err = -EIO; |
548aedac | 775 | goto put_page_out; |
4356e48e | 776 | } |
1563ac75 | 777 | if (unlikely(!PageUptodate(fio.encrypted_page))) { |
4356e48e | 778 | err = -EIO; |
548aedac | 779 | goto put_page_out; |
4356e48e | 780 | } |
548aedac | 781 | |
6aa58d8a | 782 | write_page: |
6282adbf | 783 | set_page_dirty(fio.encrypted_page); |
fec1d657 | 784 | f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true); |
6282adbf JK |
785 | if (clear_page_dirty_for_io(fio.encrypted_page)) |
786 | dec_page_count(fio.sbi, F2FS_DIRTY_META); | |
787 | ||
548aedac | 788 | set_page_writeback(fio.encrypted_page); |
17c50035 | 789 | ClearPageError(page); |
4375a336 JK |
790 | |
791 | /* allocate block address */ | |
fec1d657 | 792 | f2fs_wait_on_page_writeback(dn.node_page, NODE, true); |
4356e48e | 793 | |
04d328de | 794 | fio.op = REQ_OP_WRITE; |
70fd7614 | 795 | fio.op_flags = REQ_SYNC; |
4356e48e | 796 | fio.new_blkaddr = newaddr; |
fe16efe6 CY |
797 | f2fs_submit_page_write(&fio); |
798 | if (fio.retry) { | |
a9d572c7 SY |
799 | if (PageWriteback(fio.encrypted_page)) |
800 | end_page_writeback(fio.encrypted_page); | |
801 | goto put_page_out; | |
802 | } | |
4375a336 | 803 | |
b0af6d49 CY |
804 | f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE); |
805 | ||
f28b3434 | 806 | f2fs_update_data_blkaddr(&dn, newaddr); |
91942321 | 807 | set_inode_flag(inode, FI_APPEND_WRITE); |
4375a336 | 808 | if (page->index == 0) |
91942321 | 809 | set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); |
548aedac | 810 | put_page_out: |
4375a336 | 811 | f2fs_put_page(fio.encrypted_page, 1); |
4356e48e | 812 | recover_block: |
107a805d CY |
813 | if (lfs_mode) |
814 | up_write(&fio.sbi->io_order_lock); | |
4356e48e | 815 | if (err) |
4d57b86d | 816 | f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr, |
4356e48e | 817 | true, true); |
4375a336 JK |
818 | put_out: |
819 | f2fs_put_dnode(&dn); | |
820 | out: | |
821 | f2fs_put_page(page, 1); | |
822 | } | |
823 | ||
20614711 YH |
824 | static void move_data_page(struct inode *inode, block_t bidx, int gc_type, |
825 | unsigned int segno, int off) | |
7bc09003 | 826 | { |
c879f90d JK |
827 | struct page *page; |
828 | ||
4d57b86d | 829 | page = f2fs_get_lock_data_page(inode, bidx, true); |
c879f90d JK |
830 | if (IS_ERR(page)) |
831 | return; | |
63a0b7cb | 832 | |
20614711 YH |
833 | if (!check_valid_map(F2FS_I_SB(inode), segno, off)) |
834 | goto out; | |
835 | ||
2ef79ecb CY |
836 | if (f2fs_is_atomic_file(inode)) { |
837 | F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++; | |
838 | F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++; | |
5fe45743 | 839 | goto out; |
2ef79ecb | 840 | } |
1ad71a27 JK |
841 | if (f2fs_is_pinned_file(inode)) { |
842 | if (gc_type == FG_GC) | |
843 | f2fs_pin_file_control(inode, true); | |
844 | goto out; | |
845 | } | |
5fe45743 | 846 | |
7bc09003 | 847 | if (gc_type == BG_GC) { |
4ebefc44 JK |
848 | if (PageWriteback(page)) |
849 | goto out; | |
7bc09003 JK |
850 | set_page_dirty(page); |
851 | set_cold_data(page); | |
852 | } else { | |
c879f90d JK |
853 | struct f2fs_io_info fio = { |
854 | .sbi = F2FS_I_SB(inode), | |
39d787be | 855 | .ino = inode->i_ino, |
c879f90d | 856 | .type = DATA, |
a912b54d | 857 | .temp = COLD, |
04d328de | 858 | .op = REQ_OP_WRITE, |
70fd7614 | 859 | .op_flags = REQ_SYNC, |
e959c8f5 | 860 | .old_blkaddr = NULL_ADDR, |
c879f90d | 861 | .page = page, |
4375a336 | 862 | .encrypted_page = NULL, |
cc15620b | 863 | .need_lock = LOCK_REQ, |
b0af6d49 | 864 | .io_type = FS_GC_DATA_IO, |
c879f90d | 865 | }; |
72e1c797 CY |
866 | bool is_dirty = PageDirty(page); |
867 | int err; | |
868 | ||
869 | retry: | |
6282adbf | 870 | set_page_dirty(page); |
fec1d657 | 871 | f2fs_wait_on_page_writeback(page, DATA, true); |
933439c8 | 872 | if (clear_page_dirty_for_io(page)) { |
a7ffdbe2 | 873 | inode_dec_dirty_pages(inode); |
4d57b86d | 874 | f2fs_remove_dirty_inode(inode); |
933439c8 | 875 | } |
72e1c797 | 876 | |
7bc09003 | 877 | set_cold_data(page); |
72e1c797 | 878 | |
4d57b86d | 879 | err = f2fs_do_write_data_page(&fio); |
14a28559 CY |
880 | if (err) { |
881 | clear_cold_data(page); | |
882 | if (err == -ENOMEM) { | |
883 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
884 | goto retry; | |
885 | } | |
886 | if (is_dirty) | |
887 | set_page_dirty(page); | |
72e1c797 | 888 | } |
7bc09003 JK |
889 | } |
890 | out: | |
891 | f2fs_put_page(page, 1); | |
892 | } | |
893 | ||
0a8165d7 | 894 | /* |
7bc09003 JK |
895 | * This function tries to get parent node of victim data block, and identifies |
896 | * data block validity. If the block is valid, copy that with cold status and | |
897 | * modify parent node. | |
898 | * If the parent node is not valid or the data block address is different, | |
899 | * the victim data block is ignored. | |
900 | */ | |
718e53fa | 901 | static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7dda2af8 | 902 | struct gc_inode_list *gc_list, unsigned int segno, int gc_type) |
7bc09003 JK |
903 | { |
904 | struct super_block *sb = sbi->sb; | |
905 | struct f2fs_summary *entry; | |
906 | block_t start_addr; | |
43727527 | 907 | int off; |
7bc09003 JK |
908 | int phase = 0; |
909 | ||
910 | start_addr = START_BLOCK(sbi, segno); | |
911 | ||
912 | next_step: | |
913 | entry = sum; | |
c718379b | 914 | |
7bc09003 JK |
915 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
916 | struct page *data_page; | |
917 | struct inode *inode; | |
918 | struct node_info dni; /* dnode info for the data */ | |
919 | unsigned int ofs_in_node, nofs; | |
920 | block_t start_bidx; | |
7ea984b0 | 921 | nid_t nid = le32_to_cpu(entry->nid); |
7bc09003 | 922 | |
43727527 | 923 | /* stop BG_GC if there is not enough free sections. */ |
7f3037a5 | 924 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) |
718e53fa | 925 | return; |
7bc09003 | 926 | |
43727527 | 927 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
928 | continue; |
929 | ||
930 | if (phase == 0) { | |
4d57b86d | 931 | f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, |
7ea984b0 CY |
932 | META_NAT, true); |
933 | continue; | |
934 | } | |
935 | ||
936 | if (phase == 1) { | |
4d57b86d | 937 | f2fs_ra_node_page(sbi, nid); |
7bc09003 JK |
938 | continue; |
939 | } | |
940 | ||
941 | /* Get an inode by ino with checking validity */ | |
c1079892 | 942 | if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) |
7bc09003 JK |
943 | continue; |
944 | ||
7ea984b0 | 945 | if (phase == 2) { |
4d57b86d | 946 | f2fs_ra_node_page(sbi, dni.ino); |
7bc09003 JK |
947 | continue; |
948 | } | |
949 | ||
7bc09003 JK |
950 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); |
951 | ||
7ea984b0 | 952 | if (phase == 3) { |
d4686d56 | 953 | inode = f2fs_iget(sb, dni.ino); |
b73e5282 | 954 | if (IS_ERR(inode) || is_bad_inode(inode)) |
7bc09003 JK |
955 | continue; |
956 | ||
bb06664a | 957 | if (!down_write_trylock( |
b2532c69 | 958 | &F2FS_I(inode)->i_gc_rwsem[WRITE])) { |
bb06664a | 959 | iput(inode); |
6f8d4455 | 960 | sbi->skipped_gc_rwsem++; |
bb06664a CY |
961 | continue; |
962 | } | |
963 | ||
6aa58d8a CY |
964 | start_bidx = f2fs_start_bidx_of_node(nofs, inode) + |
965 | ofs_in_node; | |
966 | ||
967 | if (f2fs_post_read_required(inode)) { | |
968 | int err = ra_data_block(inode, start_bidx); | |
969 | ||
970 | up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); | |
971 | if (err) { | |
972 | iput(inode); | |
973 | continue; | |
974 | } | |
975 | add_gc_inode(gc_list, inode); | |
976 | continue; | |
977 | } | |
978 | ||
4d57b86d | 979 | data_page = f2fs_get_read_data_page(inode, |
6aa58d8a | 980 | start_bidx, REQ_RAHEAD, true); |
b2532c69 | 981 | up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); |
31a32688 CL |
982 | if (IS_ERR(data_page)) { |
983 | iput(inode); | |
984 | continue; | |
985 | } | |
7bc09003 JK |
986 | |
987 | f2fs_put_page(data_page, 0); | |
7dda2af8 | 988 | add_gc_inode(gc_list, inode); |
31a32688 CL |
989 | continue; |
990 | } | |
991 | ||
7ea984b0 | 992 | /* phase 4 */ |
7dda2af8 | 993 | inode = find_gc_inode(gc_list, dni.ino); |
31a32688 | 994 | if (inode) { |
82e0a5aa CY |
995 | struct f2fs_inode_info *fi = F2FS_I(inode); |
996 | bool locked = false; | |
997 | ||
998 | if (S_ISREG(inode->i_mode)) { | |
b2532c69 | 999 | if (!down_write_trylock(&fi->i_gc_rwsem[READ])) |
82e0a5aa CY |
1000 | continue; |
1001 | if (!down_write_trylock( | |
b2532c69 | 1002 | &fi->i_gc_rwsem[WRITE])) { |
6f8d4455 | 1003 | sbi->skipped_gc_rwsem++; |
b2532c69 | 1004 | up_write(&fi->i_gc_rwsem[READ]); |
82e0a5aa CY |
1005 | continue; |
1006 | } | |
1007 | locked = true; | |
73ac2f4e CY |
1008 | |
1009 | /* wait for all inflight aio data */ | |
1010 | inode_dio_wait(inode); | |
82e0a5aa CY |
1011 | } |
1012 | ||
4d57b86d | 1013 | start_bidx = f2fs_start_bidx_of_node(nofs, inode) |
c879f90d | 1014 | + ofs_in_node; |
6dbb1796 | 1015 | if (f2fs_post_read_required(inode)) |
2ef79ecb CY |
1016 | move_data_block(inode, start_bidx, gc_type, |
1017 | segno, off); | |
4375a336 | 1018 | else |
d4c759ee JK |
1019 | move_data_page(inode, start_bidx, gc_type, |
1020 | segno, off); | |
82e0a5aa CY |
1021 | |
1022 | if (locked) { | |
b2532c69 CY |
1023 | up_write(&fi->i_gc_rwsem[WRITE]); |
1024 | up_write(&fi->i_gc_rwsem[READ]); | |
82e0a5aa CY |
1025 | } |
1026 | ||
e1235983 | 1027 | stat_inc_data_blk_count(sbi, 1, gc_type); |
7bc09003 | 1028 | } |
7bc09003 | 1029 | } |
c718379b | 1030 | |
7ea984b0 | 1031 | if (++phase < 5) |
7bc09003 | 1032 | goto next_step; |
7bc09003 JK |
1033 | } |
1034 | ||
1035 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
8a2d0ace | 1036 | int gc_type) |
7bc09003 JK |
1037 | { |
1038 | struct sit_info *sit_i = SIT_I(sbi); | |
1039 | int ret; | |
8a2d0ace | 1040 | |
3d26fa6b | 1041 | down_write(&sit_i->sentry_lock); |
8a2d0ace GZ |
1042 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, |
1043 | NO_CHECK_TYPE, LFS); | |
3d26fa6b | 1044 | up_write(&sit_i->sentry_lock); |
7bc09003 JK |
1045 | return ret; |
1046 | } | |
1047 | ||
718e53fa CY |
1048 | static int do_garbage_collect(struct f2fs_sb_info *sbi, |
1049 | unsigned int start_segno, | |
7dda2af8 | 1050 | struct gc_inode_list *gc_list, int gc_type) |
7bc09003 JK |
1051 | { |
1052 | struct page *sum_page; | |
1053 | struct f2fs_summary_block *sum; | |
c718379b | 1054 | struct blk_plug plug; |
718e53fa CY |
1055 | unsigned int segno = start_segno; |
1056 | unsigned int end_segno = start_segno + sbi->segs_per_sec; | |
c56f16da | 1057 | int seg_freed = 0; |
718e53fa CY |
1058 | unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ? |
1059 | SUM_TYPE_DATA : SUM_TYPE_NODE; | |
7bc09003 | 1060 | |
718e53fa CY |
1061 | /* readahead multi ssa blocks those have contiguous address */ |
1062 | if (sbi->segs_per_sec > 1) | |
4d57b86d | 1063 | f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), |
718e53fa CY |
1064 | sbi->segs_per_sec, META_SSA, true); |
1065 | ||
1066 | /* reference all summary page */ | |
1067 | while (segno < end_segno) { | |
4d57b86d | 1068 | sum_page = f2fs_get_sum_page(sbi, segno++); |
718e53fa CY |
1069 | unlock_page(sum_page); |
1070 | } | |
7bc09003 | 1071 | |
c718379b JK |
1072 | blk_start_plug(&plug); |
1073 | ||
718e53fa | 1074 | for (segno = start_segno; segno < end_segno; segno++) { |
aa987273 | 1075 | |
718e53fa CY |
1076 | /* find segment summary of victim */ |
1077 | sum_page = find_get_page(META_MAPPING(sbi), | |
1078 | GET_SUM_BLOCK(sbi, segno)); | |
718e53fa | 1079 | f2fs_put_page(sum_page, 0); |
7bc09003 | 1080 | |
302bd348 | 1081 | if (get_valid_blocks(sbi, segno, false) == 0 || |
de0dcc40 JK |
1082 | !PageUptodate(sum_page) || |
1083 | unlikely(f2fs_cp_error(sbi))) | |
1084 | goto next; | |
1085 | ||
718e53fa | 1086 | sum = page_address(sum_page); |
10d255c3 CY |
1087 | if (type != GET_SUM_TYPE((&sum->footer))) { |
1088 | f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent segment (%u) " | |
1089 | "type [%d, %d] in SSA and SIT", | |
1090 | segno, type, GET_SUM_TYPE((&sum->footer))); | |
1091 | set_sbi_flag(sbi, SBI_NEED_FSCK); | |
1092 | goto next; | |
1093 | } | |
718e53fa CY |
1094 | |
1095 | /* | |
1096 | * this is to avoid deadlock: | |
1097 | * - lock_page(sum_page) - f2fs_replace_block | |
3d26fa6b CY |
1098 | * - check_valid_map() - down_write(sentry_lock) |
1099 | * - down_read(sentry_lock) - change_curseg() | |
718e53fa CY |
1100 | * - lock_page(sum_page) |
1101 | */ | |
718e53fa CY |
1102 | if (type == SUM_TYPE_NODE) |
1103 | gc_node_segment(sbi, sum->entries, segno, gc_type); | |
1104 | else | |
1105 | gc_data_segment(sbi, sum->entries, gc_list, segno, | |
1106 | gc_type); | |
1107 | ||
1108 | stat_inc_seg_count(sbi, type, gc_type); | |
c56f16da CY |
1109 | |
1110 | if (gc_type == FG_GC && | |
1111 | get_valid_blocks(sbi, segno, false) == 0) | |
1112 | seg_freed++; | |
f6fe2be3 | 1113 | next: |
718e53fa CY |
1114 | f2fs_put_page(sum_page, 0); |
1115 | } | |
1116 | ||
da011cc0 | 1117 | if (gc_type == FG_GC) |
b9109b0e JK |
1118 | f2fs_submit_merged_write(sbi, |
1119 | (type == SUM_TYPE_NODE) ? NODE : DATA); | |
c718379b | 1120 | |
718e53fa | 1121 | blk_finish_plug(&plug); |
7bc09003 | 1122 | |
17d899df CY |
1123 | stat_inc_call_count(sbi->stat_info); |
1124 | ||
c56f16da | 1125 | return seg_freed; |
7bc09003 JK |
1126 | } |
1127 | ||
e066b83c JK |
1128 | int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, |
1129 | bool background, unsigned int segno) | |
7bc09003 | 1130 | { |
d530d4d8 | 1131 | int gc_type = sync ? FG_GC : BG_GC; |
c56f16da CY |
1132 | int sec_freed = 0, seg_freed = 0, total_freed = 0; |
1133 | int ret = 0; | |
d5053a34 | 1134 | struct cp_control cpc; |
e066b83c | 1135 | unsigned int init_segno = segno; |
7dda2af8 CL |
1136 | struct gc_inode_list gc_list = { |
1137 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
f6bb2a2c | 1138 | .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), |
7dda2af8 | 1139 | }; |
2ef79ecb | 1140 | unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC]; |
6f8d4455 | 1141 | unsigned long long first_skipped; |
2ef79ecb | 1142 | unsigned int skipped_round = 0, round = 0; |
d5053a34 | 1143 | |
c56f16da CY |
1144 | trace_f2fs_gc_begin(sbi->sb, sync, background, |
1145 | get_pages(sbi, F2FS_DIRTY_NODES), | |
1146 | get_pages(sbi, F2FS_DIRTY_DENTS), | |
1147 | get_pages(sbi, F2FS_DIRTY_IMETA), | |
1148 | free_sections(sbi), | |
1149 | free_segments(sbi), | |
1150 | reserved_segments(sbi), | |
1151 | prefree_segments(sbi)); | |
1152 | ||
119ee914 | 1153 | cpc.reason = __get_cp_reason(sbi); |
6f8d4455 JK |
1154 | sbi->skipped_gc_rwsem = 0; |
1155 | first_skipped = last_skipped; | |
7bc09003 | 1156 | gc_more: |
1751e8a6 | 1157 | if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) { |
e5dbd956 | 1158 | ret = -EINVAL; |
408e9375 | 1159 | goto stop; |
e5dbd956 | 1160 | } |
6d5a1495 CY |
1161 | if (unlikely(f2fs_cp_error(sbi))) { |
1162 | ret = -EIO; | |
203681f6 | 1163 | goto stop; |
6d5a1495 | 1164 | } |
7bc09003 | 1165 | |
19f4e688 | 1166 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) { |
6e17bfbc | 1167 | /* |
19f4e688 HP |
1168 | * For example, if there are many prefree_segments below given |
1169 | * threshold, we can make them free by checkpoint. Then, we | |
1170 | * secure free segments which doesn't need fggc any more. | |
6e17bfbc | 1171 | */ |
8fd5a37e | 1172 | if (prefree_segments(sbi)) { |
4d57b86d | 1173 | ret = f2fs_write_checkpoint(sbi, &cpc); |
8fd5a37e JK |
1174 | if (ret) |
1175 | goto stop; | |
1176 | } | |
19f4e688 HP |
1177 | if (has_not_enough_free_secs(sbi, 0, 0)) |
1178 | gc_type = FG_GC; | |
d64f8047 | 1179 | } |
7bc09003 | 1180 | |
19f4e688 | 1181 | /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */ |
c56f16da CY |
1182 | if (gc_type == BG_GC && !background) { |
1183 | ret = -EINVAL; | |
19f4e688 | 1184 | goto stop; |
c56f16da CY |
1185 | } |
1186 | if (!__get_victim(sbi, &segno, gc_type)) { | |
1187 | ret = -ENODATA; | |
408e9375 | 1188 | goto stop; |
c56f16da | 1189 | } |
7bc09003 | 1190 | |
c56f16da CY |
1191 | seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type); |
1192 | if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec) | |
45fe8492 | 1193 | sec_freed++; |
c56f16da | 1194 | total_freed += seg_freed; |
43727527 | 1195 | |
2ef79ecb | 1196 | if (gc_type == FG_GC) { |
6f8d4455 JK |
1197 | if (sbi->skipped_atomic_files[FG_GC] > last_skipped || |
1198 | sbi->skipped_gc_rwsem) | |
2ef79ecb CY |
1199 | skipped_round++; |
1200 | last_skipped = sbi->skipped_atomic_files[FG_GC]; | |
1201 | round++; | |
1202 | } | |
1203 | ||
5ee5293c | 1204 | if (gc_type == FG_GC) |
5ec4e49f | 1205 | sbi->cur_victim_sec = NULL_SEGNO; |
43727527 | 1206 | |
6f8d4455 JK |
1207 | if (sync) |
1208 | goto stop; | |
1209 | ||
1210 | if (has_not_enough_free_secs(sbi, sec_freed, 0)) { | |
1211 | if (skipped_round <= MAX_SKIP_GC_COUNT || | |
1212 | skipped_round * 2 < round) { | |
e066b83c | 1213 | segno = NULL_SEGNO; |
d530d4d8 | 1214 | goto gc_more; |
e066b83c | 1215 | } |
43727527 | 1216 | |
6f8d4455 JK |
1217 | if (first_skipped < last_skipped && |
1218 | (last_skipped - first_skipped) > | |
1219 | sbi->skipped_gc_rwsem) { | |
1220 | f2fs_drop_inmem_pages_all(sbi, true); | |
1221 | segno = NULL_SEGNO; | |
1222 | goto gc_more; | |
1223 | } | |
d530d4d8 | 1224 | if (gc_type == FG_GC) |
4d57b86d | 1225 | ret = f2fs_write_checkpoint(sbi, &cpc); |
d530d4d8 | 1226 | } |
408e9375 | 1227 | stop: |
e066b83c JK |
1228 | SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0; |
1229 | SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno; | |
c56f16da CY |
1230 | |
1231 | trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed, | |
1232 | get_pages(sbi, F2FS_DIRTY_NODES), | |
1233 | get_pages(sbi, F2FS_DIRTY_DENTS), | |
1234 | get_pages(sbi, F2FS_DIRTY_IMETA), | |
1235 | free_sections(sbi), | |
1236 | free_segments(sbi), | |
1237 | reserved_segments(sbi), | |
1238 | prefree_segments(sbi)); | |
1239 | ||
7bc09003 JK |
1240 | mutex_unlock(&sbi->gc_mutex); |
1241 | ||
7dda2af8 | 1242 | put_gc_inode(&gc_list); |
d530d4d8 CY |
1243 | |
1244 | if (sync) | |
1245 | ret = sec_freed ? 0 : -EAGAIN; | |
43727527 | 1246 | return ret; |
7bc09003 JK |
1247 | } |
1248 | ||
4d57b86d | 1249 | void f2fs_build_gc_manager(struct f2fs_sb_info *sbi) |
7bc09003 JK |
1250 | { |
1251 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
e93b9865 | 1252 | |
1ad71a27 | 1253 | sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES; |
d5793249 JK |
1254 | |
1255 | /* give warm/cold data area from slower device */ | |
1256 | if (sbi->s_ndevs && sbi->segs_per_sec == 1) | |
1257 | SIT_I(sbi)->last_victim[ALLOC_NEXT] = | |
1258 | GET_SEGNO(sbi, FDEV(0).end_blk) + 1; | |
7bc09003 | 1259 | } |