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 | */ | |
5b0e9539 | 79 | if (sbi->gc_mode == GC_URGENT) { |
d9872a69 | 80 | wait_ms = gc_th->urgent_sleep_time; |
69babac0 | 81 | mutex_lock(&sbi->gc_mutex); |
d9872a69 JK |
82 | goto do_gc; |
83 | } | |
84 | ||
69babac0 JK |
85 | if (!mutex_trylock(&sbi->gc_mutex)) |
86 | goto next; | |
87 | ||
7bc09003 | 88 | if (!is_idle(sbi)) { |
88dd8934 | 89 | increase_sleep_time(gc_th, &wait_ms); |
7bc09003 | 90 | mutex_unlock(&sbi->gc_mutex); |
dc6febb6 | 91 | goto next; |
7bc09003 JK |
92 | } |
93 | ||
94 | if (has_enough_invalid_blocks(sbi)) | |
88dd8934 | 95 | decrease_sleep_time(gc_th, &wait_ms); |
7bc09003 | 96 | else |
88dd8934 | 97 | increase_sleep_time(gc_th, &wait_ms); |
d9872a69 | 98 | do_gc: |
dcdfff65 | 99 | stat_inc_bggc_count(sbi); |
7bc09003 | 100 | |
43727527 | 101 | /* if return value is not zero, no victim was selected */ |
e066b83c | 102 | if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true, NULL_SEGNO)) |
b59d0bae | 103 | wait_ms = gc_th->no_gc_sleep_time; |
81eb8d6e | 104 | |
84e4214f JK |
105 | trace_f2fs_background_gc(sbi->sb, wait_ms, |
106 | prefree_segments(sbi), free_segments(sbi)); | |
107 | ||
4660f9c0 JK |
108 | /* balancing f2fs's metadata periodically */ |
109 | f2fs_balance_fs_bg(sbi); | |
dc6febb6 CY |
110 | next: |
111 | sb_end_write(sbi->sb); | |
81eb8d6e | 112 | |
7bc09003 JK |
113 | } while (!kthread_should_stop()); |
114 | return 0; | |
115 | } | |
116 | ||
117 | int start_gc_thread(struct f2fs_sb_info *sbi) | |
118 | { | |
1042d60f | 119 | struct f2fs_gc_kthread *gc_th; |
ec7b1f2d | 120 | dev_t dev = sbi->sb->s_bdev->bd_dev; |
7a267f8d | 121 | int err = 0; |
7bc09003 | 122 | |
1ecc0c5c | 123 | gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL); |
7a267f8d NJ |
124 | if (!gc_th) { |
125 | err = -ENOMEM; | |
126 | goto out; | |
127 | } | |
7bc09003 | 128 | |
d9872a69 | 129 | gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME; |
b59d0bae NJ |
130 | gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME; |
131 | gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME; | |
132 | gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME; | |
133 | ||
d9872a69 | 134 | gc_th->gc_wake= 0; |
d2dc095f | 135 | |
7bc09003 JK |
136 | sbi->gc_thread = gc_th; |
137 | init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head); | |
138 | sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi, | |
ec7b1f2d | 139 | "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev)); |
7bc09003 | 140 | if (IS_ERR(gc_th->f2fs_gc_task)) { |
7a267f8d | 141 | err = PTR_ERR(gc_th->f2fs_gc_task); |
7bc09003 | 142 | kfree(gc_th); |
25718423 | 143 | sbi->gc_thread = NULL; |
7bc09003 | 144 | } |
7a267f8d NJ |
145 | out: |
146 | return err; | |
7bc09003 JK |
147 | } |
148 | ||
149 | void stop_gc_thread(struct f2fs_sb_info *sbi) | |
150 | { | |
151 | struct f2fs_gc_kthread *gc_th = sbi->gc_thread; | |
152 | if (!gc_th) | |
153 | return; | |
154 | kthread_stop(gc_th->f2fs_gc_task); | |
155 | kfree(gc_th); | |
156 | sbi->gc_thread = NULL; | |
157 | } | |
158 | ||
5b0e9539 | 159 | static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type) |
7bc09003 | 160 | { |
d2dc095f NJ |
161 | int gc_mode = (gc_type == BG_GC) ? GC_CB : GC_GREEDY; |
162 | ||
5b0e9539 JK |
163 | switch (sbi->gc_mode) { |
164 | case GC_IDLE_CB: | |
165 | gc_mode = GC_CB; | |
166 | break; | |
167 | case GC_IDLE_GREEDY: | |
168 | case GC_URGENT: | |
b27bc809 | 169 | gc_mode = GC_GREEDY; |
5b0e9539 JK |
170 | break; |
171 | } | |
d2dc095f | 172 | return gc_mode; |
7bc09003 JK |
173 | } |
174 | ||
175 | static void select_policy(struct f2fs_sb_info *sbi, int gc_type, | |
176 | int type, struct victim_sel_policy *p) | |
177 | { | |
178 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
179 | ||
4ebefc44 | 180 | if (p->alloc_mode == SSR) { |
7bc09003 JK |
181 | p->gc_mode = GC_GREEDY; |
182 | p->dirty_segmap = dirty_i->dirty_segmap[type]; | |
a26b7c8a | 183 | p->max_search = dirty_i->nr_dirty[type]; |
7bc09003 JK |
184 | p->ofs_unit = 1; |
185 | } else { | |
5b0e9539 | 186 | p->gc_mode = select_gc_type(sbi, gc_type); |
7bc09003 | 187 | p->dirty_segmap = dirty_i->dirty_segmap[DIRTY]; |
a26b7c8a | 188 | p->max_search = dirty_i->nr_dirty[DIRTY]; |
7bc09003 JK |
189 | p->ofs_unit = sbi->segs_per_sec; |
190 | } | |
a26b7c8a | 191 | |
e93b9865 | 192 | /* we need to check every dirty segments in the FG_GC case */ |
b27bc809 | 193 | if (gc_type != FG_GC && |
5b0e9539 | 194 | (sbi->gc_mode != GC_URGENT) && |
b27bc809 | 195 | p->max_search > sbi->max_victim_search) |
b1c57c1c | 196 | p->max_search = sbi->max_victim_search; |
a26b7c8a | 197 | |
b94929d9 YS |
198 | /* let's select beginning hot/small space first in no_heap mode*/ |
199 | if (test_opt(sbi, NOHEAP) && | |
200 | (type == CURSEG_HOT_DATA || IS_NODESEG(type))) | |
7a20b8a6 JK |
201 | p->offset = 0; |
202 | else | |
e066b83c | 203 | p->offset = SIT_I(sbi)->last_victim[p->gc_mode]; |
7bc09003 JK |
204 | } |
205 | ||
206 | static unsigned int get_max_cost(struct f2fs_sb_info *sbi, | |
207 | struct victim_sel_policy *p) | |
208 | { | |
b7250d2d JK |
209 | /* SSR allocates in a segment unit */ |
210 | if (p->alloc_mode == SSR) | |
3519e3f9 | 211 | return sbi->blocks_per_seg; |
7bc09003 | 212 | if (p->gc_mode == GC_GREEDY) |
c541a51b | 213 | return 2 * sbi->blocks_per_seg * p->ofs_unit; |
7bc09003 JK |
214 | else if (p->gc_mode == GC_CB) |
215 | return UINT_MAX; | |
216 | else /* No other gc_mode */ | |
217 | return 0; | |
218 | } | |
219 | ||
220 | static unsigned int check_bg_victims(struct f2fs_sb_info *sbi) | |
221 | { | |
222 | struct dirty_seglist_info *dirty_i = DIRTY_I(sbi); | |
5ec4e49f | 223 | unsigned int secno; |
7bc09003 JK |
224 | |
225 | /* | |
226 | * If the gc_type is FG_GC, we can select victim segments | |
227 | * selected by background GC before. | |
228 | * Those segments guarantee they have small valid blocks. | |
229 | */ | |
7cd8558b | 230 | for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) { |
5ec4e49f | 231 | if (sec_usage_check(sbi, secno)) |
b65ee148 | 232 | continue; |
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 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; |
7bc09003 JK |
372 | |
373 | cost = get_gc_cost(sbi, segno, &p); | |
374 | ||
375 | if (p.min_cost > cost) { | |
376 | p.min_segno = segno; | |
377 | p.min_cost = cost; | |
a57e564d | 378 | } |
688159b6 FL |
379 | next: |
380 | if (nsearched >= p.max_search) { | |
e066b83c JK |
381 | if (!sm->last_victim[p.gc_mode] && segno <= last_victim) |
382 | sm->last_victim[p.gc_mode] = last_victim + 1; | |
4ce53776 | 383 | else |
e066b83c JK |
384 | sm->last_victim[p.gc_mode] = segno + 1; |
385 | sm->last_victim[p.gc_mode] %= MAIN_SEGS(sbi); | |
7bc09003 JK |
386 | break; |
387 | } | |
388 | } | |
7bc09003 | 389 | if (p.min_segno != NULL_SEGNO) { |
b2b3460a | 390 | got_it: |
7bc09003 | 391 | if (p.alloc_mode == LFS) { |
4ddb1a4d | 392 | secno = GET_SEC_FROM_SEG(sbi, p.min_segno); |
5ec4e49f JK |
393 | if (gc_type == FG_GC) |
394 | sbi->cur_victim_sec = secno; | |
395 | else | |
396 | set_bit(secno, dirty_i->victim_secmap); | |
7bc09003 | 397 | } |
5ec4e49f | 398 | *result = (p.min_segno / p.ofs_unit) * p.ofs_unit; |
8e46b3ed NJ |
399 | |
400 | trace_f2fs_get_victim(sbi->sb, type, gc_type, &p, | |
401 | sbi->cur_victim_sec, | |
402 | prefree_segments(sbi), free_segments(sbi)); | |
7bc09003 | 403 | } |
3342bb30 | 404 | out: |
7bc09003 JK |
405 | mutex_unlock(&dirty_i->seglist_lock); |
406 | ||
407 | return (p.min_segno == NULL_SEGNO) ? 0 : 1; | |
408 | } | |
409 | ||
410 | static const struct victim_selection default_v_ops = { | |
411 | .get_victim = get_victim_by_default, | |
412 | }; | |
413 | ||
7dda2af8 | 414 | static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino) |
7bc09003 | 415 | { |
7bc09003 JK |
416 | struct inode_entry *ie; |
417 | ||
7dda2af8 CL |
418 | ie = radix_tree_lookup(&gc_list->iroot, ino); |
419 | if (ie) | |
420 | return ie->inode; | |
7bc09003 JK |
421 | return NULL; |
422 | } | |
423 | ||
7dda2af8 | 424 | static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode) |
7bc09003 | 425 | { |
6cc4af56 GZ |
426 | struct inode_entry *new_ie; |
427 | ||
7dda2af8 | 428 | if (inode == find_gc_inode(gc_list, inode->i_ino)) { |
6cc4af56 GZ |
429 | iput(inode); |
430 | return; | |
7bc09003 | 431 | } |
06292073 | 432 | new_ie = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); |
7bc09003 | 433 | new_ie->inode = inode; |
f28e5034 CY |
434 | |
435 | f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie); | |
7dda2af8 | 436 | list_add_tail(&new_ie->list, &gc_list->ilist); |
7bc09003 JK |
437 | } |
438 | ||
7dda2af8 | 439 | static void put_gc_inode(struct gc_inode_list *gc_list) |
7bc09003 JK |
440 | { |
441 | struct inode_entry *ie, *next_ie; | |
7dda2af8 CL |
442 | list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) { |
443 | radix_tree_delete(&gc_list->iroot, ie->inode->i_ino); | |
7bc09003 JK |
444 | iput(ie->inode); |
445 | list_del(&ie->list); | |
06292073 | 446 | kmem_cache_free(inode_entry_slab, ie); |
7bc09003 JK |
447 | } |
448 | } | |
449 | ||
450 | static int check_valid_map(struct f2fs_sb_info *sbi, | |
451 | unsigned int segno, int offset) | |
452 | { | |
453 | struct sit_info *sit_i = SIT_I(sbi); | |
454 | struct seg_entry *sentry; | |
455 | int ret; | |
456 | ||
3d26fa6b | 457 | down_read(&sit_i->sentry_lock); |
7bc09003 JK |
458 | sentry = get_seg_entry(sbi, segno); |
459 | ret = f2fs_test_bit(offset, sentry->cur_valid_map); | |
3d26fa6b | 460 | up_read(&sit_i->sentry_lock); |
43727527 | 461 | return ret; |
7bc09003 JK |
462 | } |
463 | ||
0a8165d7 | 464 | /* |
7bc09003 JK |
465 | * This function compares node address got in summary with that in NAT. |
466 | * On validity, copy that node with cold status, otherwise (invalid node) | |
467 | * ignore that. | |
468 | */ | |
718e53fa | 469 | static void gc_node_segment(struct f2fs_sb_info *sbi, |
7bc09003 JK |
470 | struct f2fs_summary *sum, unsigned int segno, int gc_type) |
471 | { | |
7bc09003 | 472 | struct f2fs_summary *entry; |
26d58599 | 473 | block_t start_addr; |
7bc09003 | 474 | int off; |
7ea984b0 | 475 | int phase = 0; |
7bc09003 | 476 | |
26d58599 JK |
477 | start_addr = START_BLOCK(sbi, segno); |
478 | ||
7bc09003 JK |
479 | next_step: |
480 | entry = sum; | |
c718379b | 481 | |
7bc09003 JK |
482 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
483 | nid_t nid = le32_to_cpu(entry->nid); | |
484 | struct page *node_page; | |
26d58599 | 485 | struct node_info ni; |
7bc09003 | 486 | |
43727527 | 487 | /* stop BG_GC if there is not enough free sections. */ |
7f3037a5 | 488 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) |
718e53fa | 489 | return; |
7bc09003 | 490 | |
43727527 | 491 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
492 | continue; |
493 | ||
7ea984b0 CY |
494 | if (phase == 0) { |
495 | ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, | |
496 | META_NAT, true); | |
497 | continue; | |
498 | } | |
499 | ||
500 | if (phase == 1) { | |
7bc09003 JK |
501 | ra_node_page(sbi, nid); |
502 | continue; | |
503 | } | |
7ea984b0 CY |
504 | |
505 | /* phase == 2 */ | |
7bc09003 JK |
506 | node_page = get_node_page(sbi, nid); |
507 | if (IS_ERR(node_page)) | |
508 | continue; | |
509 | ||
9a01b56b HY |
510 | /* block may become invalid during get_node_page */ |
511 | if (check_valid_map(sbi, segno, off) == 0) { | |
512 | f2fs_put_page(node_page, 1); | |
513 | continue; | |
26d58599 JK |
514 | } |
515 | ||
516 | get_node_info(sbi, nid, &ni); | |
517 | if (ni.blk_addr != start_addr + off) { | |
518 | f2fs_put_page(node_page, 1); | |
519 | continue; | |
9a01b56b HY |
520 | } |
521 | ||
da011cc0 | 522 | move_node_page(node_page, gc_type); |
e1235983 | 523 | stat_inc_node_blk_count(sbi, 1, gc_type); |
7bc09003 | 524 | } |
c718379b | 525 | |
7ea984b0 | 526 | if (++phase < 3) |
7bc09003 | 527 | goto next_step; |
7bc09003 JK |
528 | } |
529 | ||
0a8165d7 | 530 | /* |
9af45ef5 JK |
531 | * Calculate start block index indicating the given node offset. |
532 | * Be careful, caller should give this node offset only indicating direct node | |
533 | * blocks. If any node offsets, which point the other types of node blocks such | |
534 | * as indirect or double indirect node blocks, are given, it must be a caller's | |
535 | * bug. | |
7bc09003 | 536 | */ |
81ca7350 | 537 | block_t start_bidx_of_node(unsigned int node_ofs, struct inode *inode) |
7bc09003 | 538 | { |
ce19a5d4 JK |
539 | unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4; |
540 | unsigned int bidx; | |
7bc09003 | 541 | |
ce19a5d4 JK |
542 | if (node_ofs == 0) |
543 | return 0; | |
7bc09003 | 544 | |
ce19a5d4 | 545 | if (node_ofs <= 2) { |
7bc09003 JK |
546 | bidx = node_ofs - 1; |
547 | } else if (node_ofs <= indirect_blks) { | |
ce19a5d4 | 548 | int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
549 | bidx = node_ofs - 2 - dec; |
550 | } else { | |
ce19a5d4 | 551 | int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1); |
7bc09003 JK |
552 | bidx = node_ofs - 5 - dec; |
553 | } | |
81ca7350 | 554 | return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode); |
7bc09003 JK |
555 | } |
556 | ||
c1079892 | 557 | static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7bc09003 JK |
558 | struct node_info *dni, block_t blkaddr, unsigned int *nofs) |
559 | { | |
560 | struct page *node_page; | |
561 | nid_t nid; | |
562 | unsigned int ofs_in_node; | |
563 | block_t source_blkaddr; | |
564 | ||
565 | nid = le32_to_cpu(sum->nid); | |
566 | ofs_in_node = le16_to_cpu(sum->ofs_in_node); | |
567 | ||
568 | node_page = get_node_page(sbi, nid); | |
569 | if (IS_ERR(node_page)) | |
c1079892 | 570 | return false; |
7bc09003 JK |
571 | |
572 | get_node_info(sbi, nid, dni); | |
573 | ||
574 | if (sum->version != dni->version) { | |
c13ff37e JK |
575 | f2fs_msg(sbi->sb, KERN_WARNING, |
576 | "%s: valid data with mismatched node version.", | |
577 | __func__); | |
578 | set_sbi_flag(sbi, SBI_NEED_FSCK); | |
7bc09003 JK |
579 | } |
580 | ||
581 | *nofs = ofs_of_node(node_page); | |
7a2af766 | 582 | source_blkaddr = datablock_addr(NULL, node_page, ofs_in_node); |
7bc09003 JK |
583 | f2fs_put_page(node_page, 1); |
584 | ||
585 | if (source_blkaddr != blkaddr) | |
c1079892 NK |
586 | return false; |
587 | return true; | |
7bc09003 JK |
588 | } |
589 | ||
d4c759ee JK |
590 | /* |
591 | * Move data block via META_MAPPING while keeping locked data page. | |
592 | * This can be used to move blocks, aka LBAs, directly on disk. | |
593 | */ | |
594 | static void move_data_block(struct inode *inode, block_t bidx, | |
2ef79ecb | 595 | int gc_type, unsigned int segno, int off) |
4375a336 JK |
596 | { |
597 | struct f2fs_io_info fio = { | |
598 | .sbi = F2FS_I_SB(inode), | |
39d787be | 599 | .ino = inode->i_ino, |
4375a336 | 600 | .type = DATA, |
a912b54d | 601 | .temp = COLD, |
04d328de | 602 | .op = REQ_OP_READ, |
70fd7614 | 603 | .op_flags = 0, |
4375a336 | 604 | .encrypted_page = NULL, |
fb830fc5 | 605 | .in_list = false, |
4375a336 JK |
606 | }; |
607 | struct dnode_of_data dn; | |
608 | struct f2fs_summary sum; | |
609 | struct node_info ni; | |
610 | struct page *page; | |
4356e48e | 611 | block_t newaddr; |
4375a336 | 612 | int err; |
107a805d | 613 | bool lfs_mode = test_opt(fio.sbi, LFS); |
4375a336 JK |
614 | |
615 | /* do not read out */ | |
a56c7c6f | 616 | page = f2fs_grab_cache_page(inode->i_mapping, bidx, false); |
4375a336 JK |
617 | if (!page) |
618 | return; | |
619 | ||
20614711 YH |
620 | if (!check_valid_map(F2FS_I_SB(inode), segno, off)) |
621 | goto out; | |
622 | ||
2ef79ecb CY |
623 | if (f2fs_is_atomic_file(inode)) { |
624 | F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++; | |
625 | F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++; | |
5fe45743 | 626 | goto out; |
2ef79ecb | 627 | } |
5fe45743 | 628 | |
1ad71a27 JK |
629 | if (f2fs_is_pinned_file(inode)) { |
630 | f2fs_pin_file_control(inode, true); | |
631 | goto out; | |
632 | } | |
633 | ||
4375a336 JK |
634 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
635 | err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE); | |
636 | if (err) | |
637 | goto out; | |
638 | ||
08b39fbd CY |
639 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) { |
640 | ClearPageUptodate(page); | |
4375a336 | 641 | goto put_out; |
08b39fbd CY |
642 | } |
643 | ||
644 | /* | |
645 | * don't cache encrypted data into meta inode until previous dirty | |
646 | * data were writebacked to avoid racing between GC and flush. | |
647 | */ | |
fec1d657 | 648 | f2fs_wait_on_page_writeback(page, DATA, true); |
4375a336 JK |
649 | |
650 | get_node_info(fio.sbi, dn.nid, &ni); | |
651 | set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); | |
652 | ||
653 | /* read page */ | |
654 | fio.page = page; | |
7a9d7548 | 655 | fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; |
4375a336 | 656 | |
107a805d CY |
657 | if (lfs_mode) |
658 | down_write(&fio.sbi->io_order_lock); | |
659 | ||
4356e48e | 660 | allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr, |
fb830fc5 | 661 | &sum, CURSEG_COLD_DATA, NULL, false); |
4356e48e | 662 | |
01eccef7 CY |
663 | fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi), |
664 | newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS); | |
4356e48e CY |
665 | if (!fio.encrypted_page) { |
666 | err = -ENOMEM; | |
667 | goto recover_block; | |
668 | } | |
4375a336 | 669 | |
548aedac JK |
670 | err = f2fs_submit_page_bio(&fio); |
671 | if (err) | |
672 | goto put_page_out; | |
673 | ||
674 | /* write page */ | |
675 | lock_page(fio.encrypted_page); | |
676 | ||
1563ac75 | 677 | if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) { |
4356e48e | 678 | err = -EIO; |
548aedac | 679 | goto put_page_out; |
4356e48e | 680 | } |
1563ac75 | 681 | if (unlikely(!PageUptodate(fio.encrypted_page))) { |
4356e48e | 682 | err = -EIO; |
548aedac | 683 | goto put_page_out; |
4356e48e | 684 | } |
548aedac | 685 | |
6282adbf | 686 | set_page_dirty(fio.encrypted_page); |
fec1d657 | 687 | f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true); |
6282adbf JK |
688 | if (clear_page_dirty_for_io(fio.encrypted_page)) |
689 | dec_page_count(fio.sbi, F2FS_DIRTY_META); | |
690 | ||
548aedac | 691 | set_page_writeback(fio.encrypted_page); |
17c50035 | 692 | ClearPageError(page); |
4375a336 JK |
693 | |
694 | /* allocate block address */ | |
fec1d657 | 695 | f2fs_wait_on_page_writeback(dn.node_page, NODE, true); |
4356e48e | 696 | |
04d328de | 697 | fio.op = REQ_OP_WRITE; |
70fd7614 | 698 | fio.op_flags = REQ_SYNC; |
4356e48e | 699 | fio.new_blkaddr = newaddr; |
a9d572c7 SY |
700 | err = f2fs_submit_page_write(&fio); |
701 | if (err) { | |
702 | if (PageWriteback(fio.encrypted_page)) | |
703 | end_page_writeback(fio.encrypted_page); | |
704 | goto put_page_out; | |
705 | } | |
4375a336 | 706 | |
b0af6d49 CY |
707 | f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE); |
708 | ||
f28b3434 | 709 | f2fs_update_data_blkaddr(&dn, newaddr); |
91942321 | 710 | set_inode_flag(inode, FI_APPEND_WRITE); |
4375a336 | 711 | if (page->index == 0) |
91942321 | 712 | set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); |
548aedac | 713 | put_page_out: |
4375a336 | 714 | f2fs_put_page(fio.encrypted_page, 1); |
4356e48e | 715 | recover_block: |
107a805d CY |
716 | if (lfs_mode) |
717 | up_write(&fio.sbi->io_order_lock); | |
4356e48e CY |
718 | if (err) |
719 | __f2fs_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr, | |
720 | true, true); | |
4375a336 JK |
721 | put_out: |
722 | f2fs_put_dnode(&dn); | |
723 | out: | |
724 | f2fs_put_page(page, 1); | |
725 | } | |
726 | ||
20614711 YH |
727 | static void move_data_page(struct inode *inode, block_t bidx, int gc_type, |
728 | unsigned int segno, int off) | |
7bc09003 | 729 | { |
c879f90d JK |
730 | struct page *page; |
731 | ||
a56c7c6f | 732 | page = get_lock_data_page(inode, bidx, true); |
c879f90d JK |
733 | if (IS_ERR(page)) |
734 | return; | |
63a0b7cb | 735 | |
20614711 YH |
736 | if (!check_valid_map(F2FS_I_SB(inode), segno, off)) |
737 | goto out; | |
738 | ||
2ef79ecb CY |
739 | if (f2fs_is_atomic_file(inode)) { |
740 | F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++; | |
741 | F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++; | |
5fe45743 | 742 | goto out; |
2ef79ecb | 743 | } |
1ad71a27 JK |
744 | if (f2fs_is_pinned_file(inode)) { |
745 | if (gc_type == FG_GC) | |
746 | f2fs_pin_file_control(inode, true); | |
747 | goto out; | |
748 | } | |
5fe45743 | 749 | |
7bc09003 | 750 | if (gc_type == BG_GC) { |
4ebefc44 JK |
751 | if (PageWriteback(page)) |
752 | goto out; | |
7bc09003 JK |
753 | set_page_dirty(page); |
754 | set_cold_data(page); | |
755 | } else { | |
c879f90d JK |
756 | struct f2fs_io_info fio = { |
757 | .sbi = F2FS_I_SB(inode), | |
39d787be | 758 | .ino = inode->i_ino, |
c879f90d | 759 | .type = DATA, |
a912b54d | 760 | .temp = COLD, |
04d328de | 761 | .op = REQ_OP_WRITE, |
70fd7614 | 762 | .op_flags = REQ_SYNC, |
e959c8f5 | 763 | .old_blkaddr = NULL_ADDR, |
c879f90d | 764 | .page = page, |
4375a336 | 765 | .encrypted_page = NULL, |
cc15620b | 766 | .need_lock = LOCK_REQ, |
b0af6d49 | 767 | .io_type = FS_GC_DATA_IO, |
c879f90d | 768 | }; |
72e1c797 CY |
769 | bool is_dirty = PageDirty(page); |
770 | int err; | |
771 | ||
772 | retry: | |
6282adbf | 773 | set_page_dirty(page); |
fec1d657 | 774 | f2fs_wait_on_page_writeback(page, DATA, true); |
933439c8 | 775 | if (clear_page_dirty_for_io(page)) { |
a7ffdbe2 | 776 | inode_dec_dirty_pages(inode); |
933439c8 CY |
777 | remove_dirty_inode(inode); |
778 | } | |
72e1c797 | 779 | |
7bc09003 | 780 | set_cold_data(page); |
72e1c797 CY |
781 | |
782 | err = do_write_data_page(&fio); | |
783 | if (err == -ENOMEM && is_dirty) { | |
784 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
785 | goto retry; | |
786 | } | |
7bc09003 JK |
787 | } |
788 | out: | |
789 | f2fs_put_page(page, 1); | |
790 | } | |
791 | ||
0a8165d7 | 792 | /* |
7bc09003 JK |
793 | * This function tries to get parent node of victim data block, and identifies |
794 | * data block validity. If the block is valid, copy that with cold status and | |
795 | * modify parent node. | |
796 | * If the parent node is not valid or the data block address is different, | |
797 | * the victim data block is ignored. | |
798 | */ | |
718e53fa | 799 | static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum, |
7dda2af8 | 800 | struct gc_inode_list *gc_list, unsigned int segno, int gc_type) |
7bc09003 JK |
801 | { |
802 | struct super_block *sb = sbi->sb; | |
803 | struct f2fs_summary *entry; | |
804 | block_t start_addr; | |
43727527 | 805 | int off; |
7bc09003 JK |
806 | int phase = 0; |
807 | ||
808 | start_addr = START_BLOCK(sbi, segno); | |
809 | ||
810 | next_step: | |
811 | entry = sum; | |
c718379b | 812 | |
7bc09003 JK |
813 | for (off = 0; off < sbi->blocks_per_seg; off++, entry++) { |
814 | struct page *data_page; | |
815 | struct inode *inode; | |
816 | struct node_info dni; /* dnode info for the data */ | |
817 | unsigned int ofs_in_node, nofs; | |
818 | block_t start_bidx; | |
7ea984b0 | 819 | nid_t nid = le32_to_cpu(entry->nid); |
7bc09003 | 820 | |
43727527 | 821 | /* stop BG_GC if there is not enough free sections. */ |
7f3037a5 | 822 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) |
718e53fa | 823 | return; |
7bc09003 | 824 | |
43727527 | 825 | if (check_valid_map(sbi, segno, off) == 0) |
7bc09003 JK |
826 | continue; |
827 | ||
828 | if (phase == 0) { | |
7ea984b0 CY |
829 | ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1, |
830 | META_NAT, true); | |
831 | continue; | |
832 | } | |
833 | ||
834 | if (phase == 1) { | |
835 | ra_node_page(sbi, nid); | |
7bc09003 JK |
836 | continue; |
837 | } | |
838 | ||
839 | /* Get an inode by ino with checking validity */ | |
c1079892 | 840 | if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs)) |
7bc09003 JK |
841 | continue; |
842 | ||
7ea984b0 | 843 | if (phase == 2) { |
7bc09003 JK |
844 | ra_node_page(sbi, dni.ino); |
845 | continue; | |
846 | } | |
847 | ||
7bc09003 JK |
848 | ofs_in_node = le16_to_cpu(entry->ofs_in_node); |
849 | ||
7ea984b0 | 850 | if (phase == 3) { |
d4686d56 | 851 | inode = f2fs_iget(sb, dni.ino); |
b73e5282 | 852 | if (IS_ERR(inode) || is_bad_inode(inode)) |
7bc09003 JK |
853 | continue; |
854 | ||
6dbb1796 EB |
855 | /* if inode uses special I/O path, let's go phase 3 */ |
856 | if (f2fs_post_read_required(inode)) { | |
4375a336 JK |
857 | add_gc_inode(gc_list, inode); |
858 | continue; | |
859 | } | |
860 | ||
bb06664a | 861 | if (!down_write_trylock( |
b2532c69 | 862 | &F2FS_I(inode)->i_gc_rwsem[WRITE])) { |
bb06664a CY |
863 | iput(inode); |
864 | continue; | |
865 | } | |
866 | ||
81ca7350 | 867 | start_bidx = start_bidx_of_node(nofs, inode); |
43f3eae1 | 868 | data_page = get_read_data_page(inode, |
70246286 CH |
869 | start_bidx + ofs_in_node, REQ_RAHEAD, |
870 | true); | |
b2532c69 | 871 | up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); |
31a32688 CL |
872 | if (IS_ERR(data_page)) { |
873 | iput(inode); | |
874 | continue; | |
875 | } | |
7bc09003 JK |
876 | |
877 | f2fs_put_page(data_page, 0); | |
7dda2af8 | 878 | add_gc_inode(gc_list, inode); |
31a32688 CL |
879 | continue; |
880 | } | |
881 | ||
7ea984b0 | 882 | /* phase 4 */ |
7dda2af8 | 883 | inode = find_gc_inode(gc_list, dni.ino); |
31a32688 | 884 | if (inode) { |
82e0a5aa CY |
885 | struct f2fs_inode_info *fi = F2FS_I(inode); |
886 | bool locked = false; | |
887 | ||
888 | if (S_ISREG(inode->i_mode)) { | |
b2532c69 | 889 | if (!down_write_trylock(&fi->i_gc_rwsem[READ])) |
82e0a5aa CY |
890 | continue; |
891 | if (!down_write_trylock( | |
b2532c69 CY |
892 | &fi->i_gc_rwsem[WRITE])) { |
893 | up_write(&fi->i_gc_rwsem[READ]); | |
82e0a5aa CY |
894 | continue; |
895 | } | |
896 | locked = true; | |
73ac2f4e CY |
897 | |
898 | /* wait for all inflight aio data */ | |
899 | inode_dio_wait(inode); | |
82e0a5aa CY |
900 | } |
901 | ||
81ca7350 | 902 | start_bidx = start_bidx_of_node(nofs, inode) |
c879f90d | 903 | + ofs_in_node; |
6dbb1796 | 904 | if (f2fs_post_read_required(inode)) |
2ef79ecb CY |
905 | move_data_block(inode, start_bidx, gc_type, |
906 | segno, off); | |
4375a336 | 907 | else |
d4c759ee JK |
908 | move_data_page(inode, start_bidx, gc_type, |
909 | segno, off); | |
82e0a5aa CY |
910 | |
911 | if (locked) { | |
b2532c69 CY |
912 | up_write(&fi->i_gc_rwsem[WRITE]); |
913 | up_write(&fi->i_gc_rwsem[READ]); | |
82e0a5aa CY |
914 | } |
915 | ||
e1235983 | 916 | stat_inc_data_blk_count(sbi, 1, gc_type); |
7bc09003 | 917 | } |
7bc09003 | 918 | } |
c718379b | 919 | |
7ea984b0 | 920 | if (++phase < 5) |
7bc09003 | 921 | goto next_step; |
7bc09003 JK |
922 | } |
923 | ||
924 | static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim, | |
8a2d0ace | 925 | int gc_type) |
7bc09003 JK |
926 | { |
927 | struct sit_info *sit_i = SIT_I(sbi); | |
928 | int ret; | |
8a2d0ace | 929 | |
3d26fa6b | 930 | down_write(&sit_i->sentry_lock); |
8a2d0ace GZ |
931 | ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type, |
932 | NO_CHECK_TYPE, LFS); | |
3d26fa6b | 933 | up_write(&sit_i->sentry_lock); |
7bc09003 JK |
934 | return ret; |
935 | } | |
936 | ||
718e53fa CY |
937 | static int do_garbage_collect(struct f2fs_sb_info *sbi, |
938 | unsigned int start_segno, | |
7dda2af8 | 939 | struct gc_inode_list *gc_list, int gc_type) |
7bc09003 JK |
940 | { |
941 | struct page *sum_page; | |
942 | struct f2fs_summary_block *sum; | |
c718379b | 943 | struct blk_plug plug; |
718e53fa CY |
944 | unsigned int segno = start_segno; |
945 | unsigned int end_segno = start_segno + sbi->segs_per_sec; | |
c56f16da | 946 | int seg_freed = 0; |
718e53fa CY |
947 | unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ? |
948 | SUM_TYPE_DATA : SUM_TYPE_NODE; | |
7bc09003 | 949 | |
718e53fa CY |
950 | /* readahead multi ssa blocks those have contiguous address */ |
951 | if (sbi->segs_per_sec > 1) | |
952 | ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), | |
953 | sbi->segs_per_sec, META_SSA, true); | |
954 | ||
955 | /* reference all summary page */ | |
956 | while (segno < end_segno) { | |
957 | sum_page = get_sum_page(sbi, segno++); | |
958 | unlock_page(sum_page); | |
959 | } | |
7bc09003 | 960 | |
c718379b JK |
961 | blk_start_plug(&plug); |
962 | ||
718e53fa | 963 | for (segno = start_segno; segno < end_segno; segno++) { |
aa987273 | 964 | |
718e53fa CY |
965 | /* find segment summary of victim */ |
966 | sum_page = find_get_page(META_MAPPING(sbi), | |
967 | GET_SUM_BLOCK(sbi, segno)); | |
718e53fa | 968 | f2fs_put_page(sum_page, 0); |
7bc09003 | 969 | |
302bd348 | 970 | if (get_valid_blocks(sbi, segno, false) == 0 || |
de0dcc40 JK |
971 | !PageUptodate(sum_page) || |
972 | unlikely(f2fs_cp_error(sbi))) | |
973 | goto next; | |
974 | ||
718e53fa CY |
975 | sum = page_address(sum_page); |
976 | f2fs_bug_on(sbi, type != GET_SUM_TYPE((&sum->footer))); | |
977 | ||
978 | /* | |
979 | * this is to avoid deadlock: | |
980 | * - lock_page(sum_page) - f2fs_replace_block | |
3d26fa6b CY |
981 | * - check_valid_map() - down_write(sentry_lock) |
982 | * - down_read(sentry_lock) - change_curseg() | |
718e53fa CY |
983 | * - lock_page(sum_page) |
984 | */ | |
718e53fa CY |
985 | if (type == SUM_TYPE_NODE) |
986 | gc_node_segment(sbi, sum->entries, segno, gc_type); | |
987 | else | |
988 | gc_data_segment(sbi, sum->entries, gc_list, segno, | |
989 | gc_type); | |
990 | ||
991 | stat_inc_seg_count(sbi, type, gc_type); | |
c56f16da CY |
992 | |
993 | if (gc_type == FG_GC && | |
994 | get_valid_blocks(sbi, segno, false) == 0) | |
995 | seg_freed++; | |
f6fe2be3 | 996 | next: |
718e53fa CY |
997 | f2fs_put_page(sum_page, 0); |
998 | } | |
999 | ||
da011cc0 | 1000 | if (gc_type == FG_GC) |
b9109b0e JK |
1001 | f2fs_submit_merged_write(sbi, |
1002 | (type == SUM_TYPE_NODE) ? NODE : DATA); | |
c718379b | 1003 | |
718e53fa | 1004 | blk_finish_plug(&plug); |
7bc09003 | 1005 | |
17d899df CY |
1006 | stat_inc_call_count(sbi->stat_info); |
1007 | ||
c56f16da | 1008 | return seg_freed; |
7bc09003 JK |
1009 | } |
1010 | ||
e066b83c JK |
1011 | int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, |
1012 | bool background, unsigned int segno) | |
7bc09003 | 1013 | { |
d530d4d8 | 1014 | int gc_type = sync ? FG_GC : BG_GC; |
c56f16da CY |
1015 | int sec_freed = 0, seg_freed = 0, total_freed = 0; |
1016 | int ret = 0; | |
d5053a34 | 1017 | struct cp_control cpc; |
e066b83c | 1018 | unsigned int init_segno = segno; |
7dda2af8 CL |
1019 | struct gc_inode_list gc_list = { |
1020 | .ilist = LIST_HEAD_INIT(gc_list.ilist), | |
f6bb2a2c | 1021 | .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS), |
7dda2af8 | 1022 | }; |
2ef79ecb CY |
1023 | unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC]; |
1024 | unsigned int skipped_round = 0, round = 0; | |
d5053a34 | 1025 | |
c56f16da CY |
1026 | trace_f2fs_gc_begin(sbi->sb, sync, background, |
1027 | get_pages(sbi, F2FS_DIRTY_NODES), | |
1028 | get_pages(sbi, F2FS_DIRTY_DENTS), | |
1029 | get_pages(sbi, F2FS_DIRTY_IMETA), | |
1030 | free_sections(sbi), | |
1031 | free_segments(sbi), | |
1032 | reserved_segments(sbi), | |
1033 | prefree_segments(sbi)); | |
1034 | ||
119ee914 | 1035 | cpc.reason = __get_cp_reason(sbi); |
7bc09003 | 1036 | gc_more: |
1751e8a6 | 1037 | if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) { |
e5dbd956 | 1038 | ret = -EINVAL; |
408e9375 | 1039 | goto stop; |
e5dbd956 | 1040 | } |
6d5a1495 CY |
1041 | if (unlikely(f2fs_cp_error(sbi))) { |
1042 | ret = -EIO; | |
203681f6 | 1043 | goto stop; |
6d5a1495 | 1044 | } |
7bc09003 | 1045 | |
19f4e688 | 1046 | if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) { |
6e17bfbc | 1047 | /* |
19f4e688 HP |
1048 | * For example, if there are many prefree_segments below given |
1049 | * threshold, we can make them free by checkpoint. Then, we | |
1050 | * secure free segments which doesn't need fggc any more. | |
6e17bfbc | 1051 | */ |
8fd5a37e JK |
1052 | if (prefree_segments(sbi)) { |
1053 | ret = write_checkpoint(sbi, &cpc); | |
1054 | if (ret) | |
1055 | goto stop; | |
1056 | } | |
19f4e688 HP |
1057 | if (has_not_enough_free_secs(sbi, 0, 0)) |
1058 | gc_type = FG_GC; | |
d64f8047 | 1059 | } |
7bc09003 | 1060 | |
19f4e688 | 1061 | /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */ |
c56f16da CY |
1062 | if (gc_type == BG_GC && !background) { |
1063 | ret = -EINVAL; | |
19f4e688 | 1064 | goto stop; |
c56f16da CY |
1065 | } |
1066 | if (!__get_victim(sbi, &segno, gc_type)) { | |
1067 | ret = -ENODATA; | |
408e9375 | 1068 | goto stop; |
c56f16da | 1069 | } |
7bc09003 | 1070 | |
c56f16da CY |
1071 | seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type); |
1072 | if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec) | |
45fe8492 | 1073 | sec_freed++; |
c56f16da | 1074 | total_freed += seg_freed; |
43727527 | 1075 | |
2ef79ecb CY |
1076 | if (gc_type == FG_GC) { |
1077 | if (sbi->skipped_atomic_files[FG_GC] > last_skipped) | |
1078 | skipped_round++; | |
1079 | last_skipped = sbi->skipped_atomic_files[FG_GC]; | |
1080 | round++; | |
1081 | } | |
1082 | ||
5ee5293c | 1083 | if (gc_type == FG_GC) |
5ec4e49f | 1084 | sbi->cur_victim_sec = NULL_SEGNO; |
43727527 | 1085 | |
d530d4d8 | 1086 | if (!sync) { |
e066b83c | 1087 | if (has_not_enough_free_secs(sbi, sec_freed, 0)) { |
2ef79ecb CY |
1088 | if (skipped_round > MAX_SKIP_ATOMIC_COUNT && |
1089 | skipped_round * 2 >= round) | |
1090 | drop_inmem_pages_all(sbi, true); | |
e066b83c | 1091 | segno = NULL_SEGNO; |
d530d4d8 | 1092 | goto gc_more; |
e066b83c | 1093 | } |
43727527 | 1094 | |
d530d4d8 | 1095 | if (gc_type == FG_GC) |
2956e450 | 1096 | ret = write_checkpoint(sbi, &cpc); |
d530d4d8 | 1097 | } |
408e9375 | 1098 | stop: |
e066b83c JK |
1099 | SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0; |
1100 | SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno; | |
c56f16da CY |
1101 | |
1102 | trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed, | |
1103 | get_pages(sbi, F2FS_DIRTY_NODES), | |
1104 | get_pages(sbi, F2FS_DIRTY_DENTS), | |
1105 | get_pages(sbi, F2FS_DIRTY_IMETA), | |
1106 | free_sections(sbi), | |
1107 | free_segments(sbi), | |
1108 | reserved_segments(sbi), | |
1109 | prefree_segments(sbi)); | |
1110 | ||
7bc09003 JK |
1111 | mutex_unlock(&sbi->gc_mutex); |
1112 | ||
7dda2af8 | 1113 | put_gc_inode(&gc_list); |
d530d4d8 CY |
1114 | |
1115 | if (sync) | |
1116 | ret = sec_freed ? 0 : -EAGAIN; | |
43727527 | 1117 | return ret; |
7bc09003 JK |
1118 | } |
1119 | ||
1120 | void build_gc_manager(struct f2fs_sb_info *sbi) | |
1121 | { | |
1122 | DIRTY_I(sbi)->v_ops = &default_v_ops; | |
e93b9865 | 1123 | |
1ad71a27 | 1124 | sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES; |
d5793249 JK |
1125 | |
1126 | /* give warm/cold data area from slower device */ | |
1127 | if (sbi->s_ndevs && sbi->segs_per_sec == 1) | |
1128 | SIT_I(sbi)->last_victim[ALLOC_NEXT] = | |
1129 | GET_SEGNO(sbi, FDEV(0).end_blk) + 1; | |
7bc09003 | 1130 | } |