1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * fs/f2fs/gc.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
7 */
8 #include <linux/fs.h>
9 #include <linux/module.h>
10 #include <linux/init.h>
11 #include <linux/f2fs_fs.h>
12 #include <linux/kthread.h>
13 #include <linux/delay.h>
14 #include <linux/freezer.h>
15 #include <linux/sched/signal.h>
16 #include <linux/random.h>
17 #include <linux/sched/mm.h>
18
19 #include "f2fs.h"
20 #include "node.h"
21 #include "segment.h"
22 #include "gc.h"
23 #include "iostat.h"
24 #include <trace/events/f2fs.h>
25
26 static struct kmem_cache *victim_entry_slab;
27
28 static unsigned int count_bits(const unsigned long *addr,
29 unsigned int offset, unsigned int len);
30
31 static int gc_thread_func(void *data)
32 {
33 struct f2fs_sb_info *sbi = data;
34 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
35 wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
36 wait_queue_head_t *fggc_wq = &sbi->gc_thread->fggc_wq;
37 unsigned int wait_ms;
38 struct f2fs_gc_control gc_control = {
39 .victim_segno = NULL_SEGNO,
40 .should_migrate_blocks = false,
41 .err_gc_skipped = false };
42
43 wait_ms = gc_th->min_sleep_time;
44
45 set_freezable();
46 do {
47 bool sync_mode, foreground = false;
48
49 wait_event_freezable_timeout(*wq,
50 kthread_should_stop() ||
51 waitqueue_active(fggc_wq) ||
52 gc_th->gc_wake,
53 msecs_to_jiffies(wait_ms));
54
55 if (test_opt(sbi, GC_MERGE) && waitqueue_active(fggc_wq))
56 foreground = true;
57
58 /* give it a try one time */
59 if (gc_th->gc_wake)
60 gc_th->gc_wake = false;
61
62 if (f2fs_readonly(sbi->sb)) {
63 stat_other_skip_bggc_count(sbi);
64 continue;
65 }
66 if (kthread_should_stop())
67 break;
68
69 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
70 increase_sleep_time(gc_th, &wait_ms);
71 stat_other_skip_bggc_count(sbi);
72 continue;
73 }
74
75 if (time_to_inject(sbi, FAULT_CHECKPOINT))
76 f2fs_stop_checkpoint(sbi, false,
77 STOP_CP_REASON_FAULT_INJECT);
78
79 if (!sb_start_write_trylock(sbi->sb)) {
80 stat_other_skip_bggc_count(sbi);
81 continue;
82 }
83
84 gc_control.one_time = false;
85
86 /*
87 * [GC triggering condition]
88 * 0. GC is not conducted currently.
89 * 1. There are enough dirty segments.
90 * 2. IO subsystem is idle by checking the # of writeback pages.
91 * 3. IO subsystem is idle by checking the # of requests in
92 * bdev's request list.
93 *
94 * Note) We have to avoid triggering GCs frequently.
95 * Because it is possible that some segments can be
96 * invalidated soon after by user update or deletion.
97 * So, I'd like to wait some time to collect dirty segments.
98 */
99 if (sbi->gc_mode == GC_URGENT_HIGH ||
100 sbi->gc_mode == GC_URGENT_MID) {
101 wait_ms = gc_th->urgent_sleep_time;
102 f2fs_down_write(&sbi->gc_lock);
103 goto do_gc;
104 }
105
106 if (foreground) {
107 f2fs_down_write(&sbi->gc_lock);
108 goto do_gc;
109 } else if (!f2fs_down_write_trylock(&sbi->gc_lock)) {
110 stat_other_skip_bggc_count(sbi);
111 goto next;
112 }
113
114 if (!is_idle(sbi, GC_TIME)) {
115 increase_sleep_time(gc_th, &wait_ms);
116 f2fs_up_write(&sbi->gc_lock);
117 stat_io_skip_bggc_count(sbi);
118 goto next;
119 }
120
121 if (f2fs_sb_has_blkzoned(sbi)) {
122 if (has_enough_free_blocks(sbi, LIMIT_NO_ZONED_GC)) {
123 wait_ms = gc_th->no_gc_sleep_time;
124 f2fs_up_write(&sbi->gc_lock);
125 goto next;
126 }
127 if (wait_ms == gc_th->no_gc_sleep_time)
128 wait_ms = gc_th->max_sleep_time;
129 }
130
131 if (need_to_boost_gc(sbi)) {
132 decrease_sleep_time(gc_th, &wait_ms);
133 if (f2fs_sb_has_blkzoned(sbi))
134 gc_control.one_time = true;
135 } else {
136 increase_sleep_time(gc_th, &wait_ms);
137 }
138 do_gc:
139 stat_inc_gc_call_count(sbi, foreground ?
140 FOREGROUND : BACKGROUND);
141
142 sync_mode = (F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC) ||
143 gc_control.one_time;
144
145 /* foreground GC was been triggered via f2fs_balance_fs() */
146 if (foreground)
147 sync_mode = false;
148
149 gc_control.init_gc_type = sync_mode ? FG_GC : BG_GC;
150 gc_control.no_bg_gc = foreground;
151 gc_control.nr_free_secs = foreground ? 1 : 0;
152
153 /* if return value is not zero, no victim was selected */
154 if (f2fs_gc(sbi, &gc_control)) {
155 /* don't bother wait_ms by foreground gc */
156 if (!foreground)
157 wait_ms = gc_th->no_gc_sleep_time;
158 } else {
159 /* reset wait_ms to default sleep time */
160 if (wait_ms == gc_th->no_gc_sleep_time)
161 wait_ms = gc_th->min_sleep_time;
162 }
163
164 if (foreground)
165 wake_up_all(&gc_th->fggc_wq);
166
167 trace_f2fs_background_gc(sbi->sb, wait_ms,
168 prefree_segments(sbi), free_segments(sbi));
169
170 /* balancing f2fs's metadata periodically */
171 f2fs_balance_fs_bg(sbi, true);
172 next:
173 if (sbi->gc_mode != GC_NORMAL) {
174 spin_lock(&sbi->gc_remaining_trials_lock);
175 if (sbi->gc_remaining_trials) {
176 sbi->gc_remaining_trials--;
177 if (!sbi->gc_remaining_trials)
178 sbi->gc_mode = GC_NORMAL;
179 }
180 spin_unlock(&sbi->gc_remaining_trials_lock);
181 }
182 sb_end_write(sbi->sb);
183
184 } while (!kthread_should_stop());
185 return 0;
186 }
187
188 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
189 {
190 struct f2fs_gc_kthread *gc_th;
191 dev_t dev = sbi->sb->s_bdev->bd_dev;
192
193 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
194 if (!gc_th)
195 return -ENOMEM;
196
197 gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
198
199 if (f2fs_sb_has_blkzoned(sbi)) {
200 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME_ZONED;
201 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME_ZONED;
202 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME_ZONED;
203 } else {
204 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
205 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
206 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
207 }
208
209 gc_th->gc_wake = false;
210
211 sbi->gc_thread = gc_th;
212 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
213 init_waitqueue_head(&sbi->gc_thread->fggc_wq);
214 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
215 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
216 if (IS_ERR(gc_th->f2fs_gc_task)) {
217 int err = PTR_ERR(gc_th->f2fs_gc_task);
218
219 kfree(gc_th);
220 sbi->gc_thread = NULL;
221 return err;
222 }
223
224 return 0;
225 }
226
227 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
228 {
229 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
230
231 if (!gc_th)
232 return;
233 kthread_stop(gc_th->f2fs_gc_task);
234 wake_up_all(&gc_th->fggc_wq);
235 kfree(gc_th);
236 sbi->gc_thread = NULL;
237 }
238
239 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
240 {
241 int gc_mode;
242
243 if (gc_type == BG_GC) {
244 if (sbi->am.atgc_enabled)
245 gc_mode = GC_AT;
246 else
247 gc_mode = GC_CB;
248 } else {
249 gc_mode = GC_GREEDY;
250 }
251
252 switch (sbi->gc_mode) {
253 case GC_IDLE_CB:
254 gc_mode = GC_CB;
255 break;
256 case GC_IDLE_GREEDY:
257 case GC_URGENT_HIGH:
258 gc_mode = GC_GREEDY;
259 break;
260 case GC_IDLE_AT:
261 gc_mode = GC_AT;
262 break;
263 }
264
265 return gc_mode;
266 }
267
268 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
269 int type, struct victim_sel_policy *p)
270 {
271 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
272
273 if (p->alloc_mode == SSR) {
274 p->gc_mode = GC_GREEDY;
275 p->dirty_bitmap = dirty_i->dirty_segmap[type];
276 p->max_search = dirty_i->nr_dirty[type];
277 p->ofs_unit = 1;
278 } else if (p->alloc_mode == AT_SSR) {
279 p->gc_mode = GC_GREEDY;
280 p->dirty_bitmap = dirty_i->dirty_segmap[type];
281 p->max_search = dirty_i->nr_dirty[type];
282 p->ofs_unit = 1;
283 } else {
284 p->gc_mode = select_gc_type(sbi, gc_type);
285 p->ofs_unit = SEGS_PER_SEC(sbi);
286 if (__is_large_section(sbi)) {
287 p->dirty_bitmap = dirty_i->dirty_secmap;
288 p->max_search = count_bits(p->dirty_bitmap,
289 0, MAIN_SECS(sbi));
290 } else {
291 p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
292 p->max_search = dirty_i->nr_dirty[DIRTY];
293 }
294 }
295
296 /*
297 * adjust candidates range, should select all dirty segments for
298 * foreground GC and urgent GC cases.
299 */
300 if (gc_type != FG_GC &&
301 (sbi->gc_mode != GC_URGENT_HIGH) &&
302 (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
303 p->max_search > sbi->max_victim_search)
304 p->max_search = sbi->max_victim_search;
305
306 /* let's select beginning hot/small space first. */
307 if (f2fs_need_rand_seg(sbi))
308 p->offset = get_random_u32_below(MAIN_SECS(sbi) *
309 SEGS_PER_SEC(sbi));
310 else if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
311 p->offset = 0;
312 else
313 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
314 }
315
316 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
317 struct victim_sel_policy *p)
318 {
319 /* SSR allocates in a segment unit */
320 if (p->alloc_mode == SSR)
321 return BLKS_PER_SEG(sbi);
322 else if (p->alloc_mode == AT_SSR)
323 return UINT_MAX;
324
325 /* LFS */
326 if (p->gc_mode == GC_GREEDY)
327 return SEGS_TO_BLKS(sbi, 2 * p->ofs_unit);
328 else if (p->gc_mode == GC_CB)
329 return UINT_MAX;
330 else if (p->gc_mode == GC_AT)
331 return UINT_MAX;
332 else /* No other gc_mode */
333 return 0;
334 }
335
336 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
337 {
338 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
339 unsigned int secno;
340
341 /*
342 * If the gc_type is FG_GC, we can select victim segments
343 * selected by background GC before.
344 * Those segments guarantee they have small valid blocks.
345 */
346 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
347 if (sec_usage_check(sbi, secno))
348 continue;
349 clear_bit(secno, dirty_i->victim_secmap);
350 return GET_SEG_FROM_SEC(sbi, secno);
351 }
352 return NULL_SEGNO;
353 }
354
355 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
356 {
357 struct sit_info *sit_i = SIT_I(sbi);
358 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
359 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
360 unsigned long long mtime = 0;
361 unsigned int vblocks;
362 unsigned char age = 0;
363 unsigned char u;
364 unsigned int i;
365 unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
366
367 for (i = 0; i < usable_segs_per_sec; i++)
368 mtime += get_seg_entry(sbi, start + i)->mtime;
369 vblocks = get_valid_blocks(sbi, segno, true);
370
371 mtime = div_u64(mtime, usable_segs_per_sec);
372 vblocks = div_u64(vblocks, usable_segs_per_sec);
373
374 u = BLKS_TO_SEGS(sbi, vblocks * 100);
375
376 /* Handle if the system time has changed by the user */
377 if (mtime < sit_i->min_mtime)
378 sit_i->min_mtime = mtime;
379 if (mtime > sit_i->max_mtime)
380 sit_i->max_mtime = mtime;
381 if (sit_i->max_mtime != sit_i->min_mtime)
382 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
383 sit_i->max_mtime - sit_i->min_mtime);
384
385 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
386 }
387
388 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
389 unsigned int segno, struct victim_sel_policy *p)
390 {
391 if (p->alloc_mode == SSR)
392 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
393
394 /* alloc_mode == LFS */
395 if (p->gc_mode == GC_GREEDY)
396 return get_valid_blocks(sbi, segno, true);
397 else if (p->gc_mode == GC_CB)
398 return get_cb_cost(sbi, segno);
399
400 f2fs_bug_on(sbi, 1);
401 return 0;
402 }
403
404 static unsigned int count_bits(const unsigned long *addr,
405 unsigned int offset, unsigned int len)
406 {
407 unsigned int end = offset + len, sum = 0;
408
409 while (offset < end) {
410 if (test_bit(offset++, addr))
411 ++sum;
412 }
413 return sum;
414 }
415
416 static bool f2fs_check_victim_tree(struct f2fs_sb_info *sbi,
417 struct rb_root_cached *root)
418 {
419 #ifdef CONFIG_F2FS_CHECK_FS
420 struct rb_node *cur = rb_first_cached(root), *next;
421 struct victim_entry *cur_ve, *next_ve;
422
423 while (cur) {
424 next = rb_next(cur);
425 if (!next)
426 return true;
427
428 cur_ve = rb_entry(cur, struct victim_entry, rb_node);
429 next_ve = rb_entry(next, struct victim_entry, rb_node);
430
431 if (cur_ve->mtime > next_ve->mtime) {
432 f2fs_info(sbi, "broken victim_rbtree, "
433 "cur_mtime(%llu) next_mtime(%llu)",
434 cur_ve->mtime, next_ve->mtime);
435 return false;
436 }
437 cur = next;
438 }
439 #endif
440 return true;
441 }
442
443 static struct victim_entry *__lookup_victim_entry(struct f2fs_sb_info *sbi,
444 unsigned long long mtime)
445 {
446 struct atgc_management *am = &sbi->am;
447 struct rb_node *node = am->root.rb_root.rb_node;
448 struct victim_entry *ve = NULL;
449
450 while (node) {
451 ve = rb_entry(node, struct victim_entry, rb_node);
452
453 if (mtime < ve->mtime)
454 node = node->rb_left;
455 else
456 node = node->rb_right;
457 }
458 return ve;
459 }
460
461 static struct victim_entry *__create_victim_entry(struct f2fs_sb_info *sbi,
462 unsigned long long mtime, unsigned int segno)
463 {
464 struct atgc_management *am = &sbi->am;
465 struct victim_entry *ve;
466
467 ve = f2fs_kmem_cache_alloc(victim_entry_slab, GFP_NOFS, true, NULL);
468
469 ve->mtime = mtime;
470 ve->segno = segno;
471
472 list_add_tail(&ve->list, &am->victim_list);
473 am->victim_count++;
474
475 return ve;
476 }
477
478 static void __insert_victim_entry(struct f2fs_sb_info *sbi,
479 unsigned long long mtime, unsigned int segno)
480 {
481 struct atgc_management *am = &sbi->am;
482 struct rb_root_cached *root = &am->root;
483 struct rb_node **p = &root->rb_root.rb_node;
484 struct rb_node *parent = NULL;
485 struct victim_entry *ve;
486 bool left_most = true;
487
488 /* look up rb tree to find parent node */
489 while (*p) {
490 parent = *p;
491 ve = rb_entry(parent, struct victim_entry, rb_node);
492
493 if (mtime < ve->mtime) {
494 p = &(*p)->rb_left;
495 } else {
496 p = &(*p)->rb_right;
497 left_most = false;
498 }
499 }
500
501 ve = __create_victim_entry(sbi, mtime, segno);
502
503 rb_link_node(&ve->rb_node, parent, p);
504 rb_insert_color_cached(&ve->rb_node, root, left_most);
505 }
506
507 static void add_victim_entry(struct f2fs_sb_info *sbi,
508 struct victim_sel_policy *p, unsigned int segno)
509 {
510 struct sit_info *sit_i = SIT_I(sbi);
511 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
512 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
513 unsigned long long mtime = 0;
514 unsigned int i;
515
516 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
517 if (p->gc_mode == GC_AT &&
518 get_valid_blocks(sbi, segno, true) == 0)
519 return;
520 }
521
522 for (i = 0; i < SEGS_PER_SEC(sbi); i++)
523 mtime += get_seg_entry(sbi, start + i)->mtime;
524 mtime = div_u64(mtime, SEGS_PER_SEC(sbi));
525
526 /* Handle if the system time has changed by the user */
527 if (mtime < sit_i->min_mtime)
528 sit_i->min_mtime = mtime;
529 if (mtime > sit_i->max_mtime)
530 sit_i->max_mtime = mtime;
531 if (mtime < sit_i->dirty_min_mtime)
532 sit_i->dirty_min_mtime = mtime;
533 if (mtime > sit_i->dirty_max_mtime)
534 sit_i->dirty_max_mtime = mtime;
535
536 /* don't choose young section as candidate */
537 if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
538 return;
539
540 __insert_victim_entry(sbi, mtime, segno);
541 }
542
543 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
544 struct victim_sel_policy *p)
545 {
546 struct sit_info *sit_i = SIT_I(sbi);
547 struct atgc_management *am = &sbi->am;
548 struct rb_root_cached *root = &am->root;
549 struct rb_node *node;
550 struct victim_entry *ve;
551 unsigned long long total_time;
552 unsigned long long age, u, accu;
553 unsigned long long max_mtime = sit_i->dirty_max_mtime;
554 unsigned long long min_mtime = sit_i->dirty_min_mtime;
555 unsigned int sec_blocks = CAP_BLKS_PER_SEC(sbi);
556 unsigned int vblocks;
557 unsigned int dirty_threshold = max(am->max_candidate_count,
558 am->candidate_ratio *
559 am->victim_count / 100);
560 unsigned int age_weight = am->age_weight;
561 unsigned int cost;
562 unsigned int iter = 0;
563
564 if (max_mtime < min_mtime)
565 return;
566
567 max_mtime += 1;
568 total_time = max_mtime - min_mtime;
569
570 accu = div64_u64(ULLONG_MAX, total_time);
571 accu = min_t(unsigned long long, div_u64(accu, 100),
572 DEFAULT_ACCURACY_CLASS);
573
574 node = rb_first_cached(root);
575 next:
576 ve = rb_entry_safe(node, struct victim_entry, rb_node);
577 if (!ve)
578 return;
579
580 if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
581 goto skip;
582
583 /* age = 10000 * x% * 60 */
584 age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
585 age_weight;
586
587 vblocks = get_valid_blocks(sbi, ve->segno, true);
588 f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
589
590 /* u = 10000 * x% * 40 */
591 u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
592 (100 - age_weight);
593
594 f2fs_bug_on(sbi, age + u >= UINT_MAX);
595
596 cost = UINT_MAX - (age + u);
597 iter++;
598
599 if (cost < p->min_cost ||
600 (cost == p->min_cost && age > p->oldest_age)) {
601 p->min_cost = cost;
602 p->oldest_age = age;
603 p->min_segno = ve->segno;
604 }
605 skip:
606 if (iter < dirty_threshold) {
607 node = rb_next(node);
608 goto next;
609 }
610 }
611
612 /*
613 * select candidates around source section in range of
614 * [target - dirty_threshold, target + dirty_threshold]
615 */
616 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
617 struct victim_sel_policy *p)
618 {
619 struct sit_info *sit_i = SIT_I(sbi);
620 struct atgc_management *am = &sbi->am;
621 struct victim_entry *ve;
622 unsigned long long age;
623 unsigned long long max_mtime = sit_i->dirty_max_mtime;
624 unsigned long long min_mtime = sit_i->dirty_min_mtime;
625 unsigned int vblocks;
626 unsigned int dirty_threshold = max(am->max_candidate_count,
627 am->candidate_ratio *
628 am->victim_count / 100);
629 unsigned int cost, iter;
630 int stage = 0;
631
632 if (max_mtime < min_mtime)
633 return;
634 max_mtime += 1;
635 next_stage:
636 iter = 0;
637 ve = __lookup_victim_entry(sbi, p->age);
638 next_node:
639 if (!ve) {
640 if (stage++ == 0)
641 goto next_stage;
642 return;
643 }
644
645 if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
646 goto skip_node;
647
648 age = max_mtime - ve->mtime;
649
650 vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
651 f2fs_bug_on(sbi, !vblocks);
652
653 /* rare case */
654 if (vblocks == BLKS_PER_SEG(sbi))
655 goto skip_node;
656
657 iter++;
658
659 age = max_mtime - abs(p->age - age);
660 cost = UINT_MAX - vblocks;
661
662 if (cost < p->min_cost ||
663 (cost == p->min_cost && age > p->oldest_age)) {
664 p->min_cost = cost;
665 p->oldest_age = age;
666 p->min_segno = ve->segno;
667 }
668 skip_node:
669 if (iter < dirty_threshold) {
670 ve = rb_entry(stage == 0 ? rb_prev(&ve->rb_node) :
671 rb_next(&ve->rb_node),
672 struct victim_entry, rb_node);
673 goto next_node;
674 }
675
676 if (stage++ == 0)
677 goto next_stage;
678 }
679
680 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
681 struct victim_sel_policy *p)
682 {
683 f2fs_bug_on(sbi, !f2fs_check_victim_tree(sbi, &sbi->am.root));
684
685 if (p->gc_mode == GC_AT)
686 atgc_lookup_victim(sbi, p);
687 else if (p->alloc_mode == AT_SSR)
688 atssr_lookup_victim(sbi, p);
689 else
690 f2fs_bug_on(sbi, 1);
691 }
692
693 static void release_victim_entry(struct f2fs_sb_info *sbi)
694 {
695 struct atgc_management *am = &sbi->am;
696 struct victim_entry *ve, *tmp;
697
698 list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
699 list_del(&ve->list);
700 kmem_cache_free(victim_entry_slab, ve);
701 am->victim_count--;
702 }
703
704 am->root = RB_ROOT_CACHED;
705
706 f2fs_bug_on(sbi, am->victim_count);
707 f2fs_bug_on(sbi, !list_empty(&am->victim_list));
708 }
709
710 static bool f2fs_pin_section(struct f2fs_sb_info *sbi, unsigned int segno)
711 {
712 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
713 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
714
715 if (!dirty_i->enable_pin_section)
716 return false;
717 if (!test_and_set_bit(secno, dirty_i->pinned_secmap))
718 dirty_i->pinned_secmap_cnt++;
719 return true;
720 }
721
722 static bool f2fs_pinned_section_exists(struct dirty_seglist_info *dirty_i)
723 {
724 return dirty_i->pinned_secmap_cnt;
725 }
726
727 static bool f2fs_section_is_pinned(struct dirty_seglist_info *dirty_i,
728 unsigned int secno)
729 {
730 return dirty_i->enable_pin_section &&
731 f2fs_pinned_section_exists(dirty_i) &&
732 test_bit(secno, dirty_i->pinned_secmap);
733 }
734
735 static void f2fs_unpin_all_sections(struct f2fs_sb_info *sbi, bool enable)
736 {
737 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
738
739 if (f2fs_pinned_section_exists(DIRTY_I(sbi))) {
740 memset(DIRTY_I(sbi)->pinned_secmap, 0, bitmap_size);
741 DIRTY_I(sbi)->pinned_secmap_cnt = 0;
742 }
743 DIRTY_I(sbi)->enable_pin_section = enable;
744 }
745
746 static int f2fs_gc_pinned_control(struct inode *inode, int gc_type,
747 unsigned int segno)
748 {
749 if (!f2fs_is_pinned_file(inode))
750 return 0;
751 if (gc_type != FG_GC)
752 return -EBUSY;
753 if (!f2fs_pin_section(F2FS_I_SB(inode), segno))
754 f2fs_pin_file_control(inode, true);
755 return -EAGAIN;
756 }
757
758 /*
759 * This function is called from two paths.
760 * One is garbage collection and the other is SSR segment selection.
761 * When it is called during GC, it just gets a victim segment
762 * and it does not remove it from dirty seglist.
763 * When it is called from SSR segment selection, it finds a segment
764 * which has minimum valid blocks and removes it from dirty seglist.
765 */
766 int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
767 int gc_type, int type, char alloc_mode,
768 unsigned long long age)
769 {
770 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
771 struct sit_info *sm = SIT_I(sbi);
772 struct victim_sel_policy p;
773 unsigned int secno, last_victim;
774 unsigned int last_segment;
775 unsigned int nsearched;
776 bool is_atgc;
777 int ret = 0;
778
779 mutex_lock(&dirty_i->seglist_lock);
780 last_segment = MAIN_SECS(sbi) * SEGS_PER_SEC(sbi);
781
782 p.alloc_mode = alloc_mode;
783 p.age = age;
784 p.age_threshold = sbi->am.age_threshold;
785
786 retry:
787 select_policy(sbi, gc_type, type, &p);
788 p.min_segno = NULL_SEGNO;
789 p.oldest_age = 0;
790 p.min_cost = get_max_cost(sbi, &p);
791
792 is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
793 nsearched = 0;
794
795 if (is_atgc)
796 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
797
798 if (*result != NULL_SEGNO) {
799 if (!get_valid_blocks(sbi, *result, false)) {
800 ret = -ENODATA;
801 goto out;
802 }
803
804 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
805 ret = -EBUSY;
806 else
807 p.min_segno = *result;
808 goto out;
809 }
810
811 ret = -ENODATA;
812 if (p.max_search == 0)
813 goto out;
814
815 if (__is_large_section(sbi) && p.alloc_mode == LFS) {
816 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
817 p.min_segno = sbi->next_victim_seg[BG_GC];
818 *result = p.min_segno;
819 sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
820 goto got_result;
821 }
822 if (gc_type == FG_GC &&
823 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
824 p.min_segno = sbi->next_victim_seg[FG_GC];
825 *result = p.min_segno;
826 sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
827 goto got_result;
828 }
829 }
830
831 last_victim = sm->last_victim[p.gc_mode];
832 if (p.alloc_mode == LFS && gc_type == FG_GC) {
833 p.min_segno = check_bg_victims(sbi);
834 if (p.min_segno != NULL_SEGNO)
835 goto got_it;
836 }
837
838 while (1) {
839 unsigned long cost, *dirty_bitmap;
840 unsigned int unit_no, segno;
841
842 dirty_bitmap = p.dirty_bitmap;
843 unit_no = find_next_bit(dirty_bitmap,
844 last_segment / p.ofs_unit,
845 p.offset / p.ofs_unit);
846 segno = unit_no * p.ofs_unit;
847 if (segno >= last_segment) {
848 if (sm->last_victim[p.gc_mode]) {
849 last_segment =
850 sm->last_victim[p.gc_mode];
851 sm->last_victim[p.gc_mode] = 0;
852 p.offset = 0;
853 continue;
854 }
855 break;
856 }
857
858 p.offset = segno + p.ofs_unit;
859 nsearched++;
860
861 #ifdef CONFIG_F2FS_CHECK_FS
862 /*
863 * skip selecting the invalid segno (that is failed due to block
864 * validity check failure during GC) to avoid endless GC loop in
865 * such cases.
866 */
867 if (test_bit(segno, sm->invalid_segmap))
868 goto next;
869 #endif
870
871 secno = GET_SEC_FROM_SEG(sbi, segno);
872
873 if (sec_usage_check(sbi, secno))
874 goto next;
875
876 /* Don't touch checkpointed data */
877 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
878 if (p.alloc_mode == LFS) {
879 /*
880 * LFS is set to find source section during GC.
881 * The victim should have no checkpointed data.
882 */
883 if (get_ckpt_valid_blocks(sbi, segno, true))
884 goto next;
885 } else {
886 /*
887 * SSR | AT_SSR are set to find target segment
888 * for writes which can be full by checkpointed
889 * and newly written blocks.
890 */
891 if (!f2fs_segment_has_free_slot(sbi, segno))
892 goto next;
893 }
894 }
895
896 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
897 goto next;
898
899 if (gc_type == FG_GC && f2fs_section_is_pinned(dirty_i, secno))
900 goto next;
901
902 if (is_atgc) {
903 add_victim_entry(sbi, &p, segno);
904 goto next;
905 }
906
907 cost = get_gc_cost(sbi, segno, &p);
908
909 if (p.min_cost > cost) {
910 p.min_segno = segno;
911 p.min_cost = cost;
912 }
913 next:
914 if (nsearched >= p.max_search) {
915 if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
916 sm->last_victim[p.gc_mode] =
917 last_victim + p.ofs_unit;
918 else
919 sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
920 sm->last_victim[p.gc_mode] %=
921 (MAIN_SECS(sbi) * SEGS_PER_SEC(sbi));
922 break;
923 }
924 }
925
926 /* get victim for GC_AT/AT_SSR */
927 if (is_atgc) {
928 lookup_victim_by_age(sbi, &p);
929 release_victim_entry(sbi);
930 }
931
932 if (is_atgc && p.min_segno == NULL_SEGNO &&
933 sm->elapsed_time < p.age_threshold) {
934 p.age_threshold = 0;
935 goto retry;
936 }
937
938 if (p.min_segno != NULL_SEGNO) {
939 got_it:
940 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
941 got_result:
942 if (p.alloc_mode == LFS) {
943 secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
944 if (gc_type == FG_GC)
945 sbi->cur_victim_sec = secno;
946 else
947 set_bit(secno, dirty_i->victim_secmap);
948 }
949 ret = 0;
950
951 }
952 out:
953 if (p.min_segno != NULL_SEGNO)
954 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
955 sbi->cur_victim_sec,
956 prefree_segments(sbi), free_segments(sbi));
957 mutex_unlock(&dirty_i->seglist_lock);
958
959 return ret;
960 }
961
962 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
963 {
964 struct inode_entry *ie;
965
966 ie = radix_tree_lookup(&gc_list->iroot, ino);
967 if (ie)
968 return ie->inode;
969 return NULL;
970 }
971
972 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
973 {
974 struct inode_entry *new_ie;
975
976 if (inode == find_gc_inode(gc_list, inode->i_ino)) {
977 iput(inode);
978 return;
979 }
980 new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab,
981 GFP_NOFS, true, NULL);
982 new_ie->inode = inode;
983
984 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
985 list_add_tail(&new_ie->list, &gc_list->ilist);
986 }
987
988 static void put_gc_inode(struct gc_inode_list *gc_list)
989 {
990 struct inode_entry *ie, *next_ie;
991
992 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
993 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
994 iput(ie->inode);
995 list_del(&ie->list);
996 kmem_cache_free(f2fs_inode_entry_slab, ie);
997 }
998 }
999
1000 static int check_valid_map(struct f2fs_sb_info *sbi,
1001 unsigned int segno, int offset)
1002 {
1003 struct sit_info *sit_i = SIT_I(sbi);
1004 struct seg_entry *sentry;
1005 int ret;
1006
1007 down_read(&sit_i->sentry_lock);
1008 sentry = get_seg_entry(sbi, segno);
1009 ret = f2fs_test_bit(offset, sentry->cur_valid_map);
1010 up_read(&sit_i->sentry_lock);
1011 return ret;
1012 }
1013
1014 /*
1015 * This function compares node address got in summary with that in NAT.
1016 * On validity, copy that node with cold status, otherwise (invalid node)
1017 * ignore that.
1018 */
1019 static int gc_node_segment(struct f2fs_sb_info *sbi,
1020 struct f2fs_summary *sum, unsigned int segno, int gc_type)
1021 {
1022 struct f2fs_summary *entry;
1023 block_t start_addr;
1024 int off;
1025 int phase = 0;
1026 bool fggc = (gc_type == FG_GC);
1027 int submitted = 0;
1028 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1029
1030 start_addr = START_BLOCK(sbi, segno);
1031
1032 next_step:
1033 entry = sum;
1034
1035 if (fggc && phase == 2)
1036 atomic_inc(&sbi->wb_sync_req[NODE]);
1037
1038 for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1039 nid_t nid = le32_to_cpu(entry->nid);
1040 struct page *node_page;
1041 struct node_info ni;
1042 int err;
1043
1044 /* stop BG_GC if there is not enough free sections. */
1045 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
1046 return submitted;
1047
1048 if (check_valid_map(sbi, segno, off) == 0)
1049 continue;
1050
1051 if (phase == 0) {
1052 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1053 META_NAT, true);
1054 continue;
1055 }
1056
1057 if (phase == 1) {
1058 f2fs_ra_node_page(sbi, nid);
1059 continue;
1060 }
1061
1062 /* phase == 2 */
1063 node_page = f2fs_get_node_page(sbi, nid);
1064 if (IS_ERR(node_page))
1065 continue;
1066
1067 /* block may become invalid during f2fs_get_node_page */
1068 if (check_valid_map(sbi, segno, off) == 0) {
1069 f2fs_put_page(node_page, 1);
1070 continue;
1071 }
1072
1073 if (f2fs_get_node_info(sbi, nid, &ni, false)) {
1074 f2fs_put_page(node_page, 1);
1075 continue;
1076 }
1077
1078 if (ni.blk_addr != start_addr + off) {
1079 f2fs_put_page(node_page, 1);
1080 continue;
1081 }
1082
1083 err = f2fs_move_node_page(node_page, gc_type);
1084 if (!err && gc_type == FG_GC)
1085 submitted++;
1086 stat_inc_node_blk_count(sbi, 1, gc_type);
1087 }
1088
1089 if (++phase < 3)
1090 goto next_step;
1091
1092 if (fggc)
1093 atomic_dec(&sbi->wb_sync_req[NODE]);
1094 return submitted;
1095 }
1096
1097 /*
1098 * Calculate start block index indicating the given node offset.
1099 * Be careful, caller should give this node offset only indicating direct node
1100 * blocks. If any node offsets, which point the other types of node blocks such
1101 * as indirect or double indirect node blocks, are given, it must be a caller's
1102 * bug.
1103 */
1104 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
1105 {
1106 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
1107 unsigned int bidx;
1108
1109 if (node_ofs == 0)
1110 return 0;
1111
1112 if (node_ofs <= 2) {
1113 bidx = node_ofs - 1;
1114 } else if (node_ofs <= indirect_blks) {
1115 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
1116
1117 bidx = node_ofs - 2 - dec;
1118 } else {
1119 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
1120
1121 bidx = node_ofs - 5 - dec;
1122 }
1123 return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
1124 }
1125
1126 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1127 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
1128 {
1129 struct page *node_page;
1130 nid_t nid;
1131 unsigned int ofs_in_node, max_addrs, base;
1132 block_t source_blkaddr;
1133
1134 nid = le32_to_cpu(sum->nid);
1135 ofs_in_node = le16_to_cpu(sum->ofs_in_node);
1136
1137 node_page = f2fs_get_node_page(sbi, nid);
1138 if (IS_ERR(node_page))
1139 return false;
1140
1141 if (f2fs_get_node_info(sbi, nid, dni, false)) {
1142 f2fs_put_page(node_page, 1);
1143 return false;
1144 }
1145
1146 if (sum->version != dni->version) {
1147 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
1148 __func__);
1149 set_sbi_flag(sbi, SBI_NEED_FSCK);
1150 }
1151
1152 if (f2fs_check_nid_range(sbi, dni->ino)) {
1153 f2fs_put_page(node_page, 1);
1154 return false;
1155 }
1156
1157 if (IS_INODE(node_page)) {
1158 base = offset_in_addr(F2FS_INODE(node_page));
1159 max_addrs = DEF_ADDRS_PER_INODE;
1160 } else {
1161 base = 0;
1162 max_addrs = DEF_ADDRS_PER_BLOCK;
1163 }
1164
1165 if (base + ofs_in_node >= max_addrs) {
1166 f2fs_err(sbi, "Inconsistent blkaddr offset: base:%u, ofs_in_node:%u, max:%u, ino:%u, nid:%u",
1167 base, ofs_in_node, max_addrs, dni->ino, dni->nid);
1168 f2fs_put_page(node_page, 1);
1169 return false;
1170 }
1171
1172 *nofs = ofs_of_node(node_page);
1173 source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1174 f2fs_put_page(node_page, 1);
1175
1176 if (source_blkaddr != blkaddr) {
1177 #ifdef CONFIG_F2FS_CHECK_FS
1178 unsigned int segno = GET_SEGNO(sbi, blkaddr);
1179 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1180
1181 if (unlikely(check_valid_map(sbi, segno, offset))) {
1182 if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1183 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",
1184 blkaddr, source_blkaddr, segno);
1185 set_sbi_flag(sbi, SBI_NEED_FSCK);
1186 }
1187 }
1188 #endif
1189 return false;
1190 }
1191 return true;
1192 }
1193
1194 static int ra_data_block(struct inode *inode, pgoff_t index)
1195 {
1196 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1197 struct address_space *mapping = f2fs_is_cow_file(inode) ?
1198 F2FS_I(inode)->atomic_inode->i_mapping : inode->i_mapping;
1199 struct dnode_of_data dn;
1200 struct page *page;
1201 struct f2fs_io_info fio = {
1202 .sbi = sbi,
1203 .ino = inode->i_ino,
1204 .type = DATA,
1205 .temp = COLD,
1206 .op = REQ_OP_READ,
1207 .op_flags = 0,
1208 .encrypted_page = NULL,
1209 .in_list = 0,
1210 };
1211 int err;
1212
1213 page = f2fs_grab_cache_page(mapping, index, true);
1214 if (!page)
1215 return -ENOMEM;
1216
1217 if (f2fs_lookup_read_extent_cache_block(inode, index,
1218 &dn.data_blkaddr)) {
1219 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1220 DATA_GENERIC_ENHANCE_READ))) {
1221 err = -EFSCORRUPTED;
1222 goto put_page;
1223 }
1224 goto got_it;
1225 }
1226
1227 set_new_dnode(&dn, inode, NULL, NULL, 0);
1228 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1229 if (err)
1230 goto put_page;
1231 f2fs_put_dnode(&dn);
1232
1233 if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1234 err = -ENOENT;
1235 goto put_page;
1236 }
1237 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1238 DATA_GENERIC_ENHANCE))) {
1239 err = -EFSCORRUPTED;
1240 goto put_page;
1241 }
1242 got_it:
1243 /* read page */
1244 fio.page = page;
1245 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1246
1247 /*
1248 * don't cache encrypted data into meta inode until previous dirty
1249 * data were writebacked to avoid racing between GC and flush.
1250 */
1251 f2fs_wait_on_page_writeback(page, DATA, true, true);
1252
1253 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1254
1255 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1256 dn.data_blkaddr,
1257 FGP_LOCK | FGP_CREAT, GFP_NOFS);
1258 if (!fio.encrypted_page) {
1259 err = -ENOMEM;
1260 goto put_page;
1261 }
1262
1263 err = f2fs_submit_page_bio(&fio);
1264 if (err)
1265 goto put_encrypted_page;
1266 f2fs_put_page(fio.encrypted_page, 0);
1267 f2fs_put_page(page, 1);
1268
1269 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
1270 f2fs_update_iostat(sbi, NULL, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1271
1272 return 0;
1273 put_encrypted_page:
1274 f2fs_put_page(fio.encrypted_page, 1);
1275 put_page:
1276 f2fs_put_page(page, 1);
1277 return err;
1278 }
1279
1280 /*
1281 * Move data block via META_MAPPING while keeping locked data page.
1282 * This can be used to move blocks, aka LBAs, directly on disk.
1283 */
1284 static int move_data_block(struct inode *inode, block_t bidx,
1285 int gc_type, unsigned int segno, int off)
1286 {
1287 struct address_space *mapping = f2fs_is_cow_file(inode) ?
1288 F2FS_I(inode)->atomic_inode->i_mapping : inode->i_mapping;
1289 struct f2fs_io_info fio = {
1290 .sbi = F2FS_I_SB(inode),
1291 .ino = inode->i_ino,
1292 .type = DATA,
1293 .temp = COLD,
1294 .op = REQ_OP_READ,
1295 .op_flags = 0,
1296 .encrypted_page = NULL,
1297 .in_list = 0,
1298 };
1299 struct dnode_of_data dn;
1300 struct f2fs_summary sum;
1301 struct node_info ni;
1302 struct page *page, *mpage;
1303 block_t newaddr;
1304 int err = 0;
1305 bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1306 int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&
1307 (fio.sbi->gc_mode != GC_URGENT_HIGH) ?
1308 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1309
1310 /* do not read out */
1311 page = f2fs_grab_cache_page(mapping, bidx, false);
1312 if (!page)
1313 return -ENOMEM;
1314
1315 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1316 err = -ENOENT;
1317 goto out;
1318 }
1319
1320 err = f2fs_gc_pinned_control(inode, gc_type, segno);
1321 if (err)
1322 goto out;
1323
1324 set_new_dnode(&dn, inode, NULL, NULL, 0);
1325 err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1326 if (err)
1327 goto out;
1328
1329 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1330 ClearPageUptodate(page);
1331 err = -ENOENT;
1332 goto put_out;
1333 }
1334
1335 /*
1336 * don't cache encrypted data into meta inode until previous dirty
1337 * data were writebacked to avoid racing between GC and flush.
1338 */
1339 f2fs_wait_on_page_writeback(page, DATA, true, true);
1340
1341 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1342
1343 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);
1344 if (err)
1345 goto put_out;
1346
1347 /* read page */
1348 fio.page = page;
1349 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1350
1351 if (lfs_mode)
1352 f2fs_down_write(&fio.sbi->io_order_lock);
1353
1354 mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1355 fio.old_blkaddr, false);
1356 if (!mpage) {
1357 err = -ENOMEM;
1358 goto up_out;
1359 }
1360
1361 fio.encrypted_page = mpage;
1362
1363 /* read source block in mpage */
1364 if (!PageUptodate(mpage)) {
1365 err = f2fs_submit_page_bio(&fio);
1366 if (err) {
1367 f2fs_put_page(mpage, 1);
1368 goto up_out;
1369 }
1370
1371 f2fs_update_iostat(fio.sbi, inode, FS_DATA_READ_IO,
1372 F2FS_BLKSIZE);
1373 f2fs_update_iostat(fio.sbi, NULL, FS_GDATA_READ_IO,
1374 F2FS_BLKSIZE);
1375
1376 lock_page(mpage);
1377 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1378 !PageUptodate(mpage))) {
1379 err = -EIO;
1380 f2fs_put_page(mpage, 1);
1381 goto up_out;
1382 }
1383 }
1384
1385 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1386
1387 /* allocate block address */
1388 err = f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1389 &sum, type, NULL);
1390 if (err) {
1391 f2fs_put_page(mpage, 1);
1392 /* filesystem should shutdown, no need to recovery block */
1393 goto up_out;
1394 }
1395
1396 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1397 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1398 if (!fio.encrypted_page) {
1399 err = -ENOMEM;
1400 f2fs_put_page(mpage, 1);
1401 goto recover_block;
1402 }
1403
1404 /* write target block */
1405 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1406 memcpy(page_address(fio.encrypted_page),
1407 page_address(mpage), PAGE_SIZE);
1408 f2fs_put_page(mpage, 1);
1409
1410 f2fs_invalidate_internal_cache(fio.sbi, fio.old_blkaddr);
1411
1412 set_page_dirty(fio.encrypted_page);
1413 if (clear_page_dirty_for_io(fio.encrypted_page))
1414 dec_page_count(fio.sbi, F2FS_DIRTY_META);
1415
1416 set_page_writeback(fio.encrypted_page);
1417
1418 fio.op = REQ_OP_WRITE;
1419 fio.op_flags = REQ_SYNC;
1420 fio.new_blkaddr = newaddr;
1421 f2fs_submit_page_write(&fio);
1422
1423 f2fs_update_iostat(fio.sbi, NULL, FS_GC_DATA_IO, F2FS_BLKSIZE);
1424
1425 f2fs_update_data_blkaddr(&dn, newaddr);
1426 set_inode_flag(inode, FI_APPEND_WRITE);
1427
1428 f2fs_put_page(fio.encrypted_page, 1);
1429 recover_block:
1430 if (err)
1431 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1432 true, true, true);
1433 up_out:
1434 if (lfs_mode)
1435 f2fs_up_write(&fio.sbi->io_order_lock);
1436 put_out:
1437 f2fs_put_dnode(&dn);
1438 out:
1439 f2fs_put_page(page, 1);
1440 return err;
1441 }
1442
1443 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1444 unsigned int segno, int off)
1445 {
1446 struct page *page;
1447 int err = 0;
1448
1449 page = f2fs_get_lock_data_page(inode, bidx, true);
1450 if (IS_ERR(page))
1451 return PTR_ERR(page);
1452
1453 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1454 err = -ENOENT;
1455 goto out;
1456 }
1457
1458 err = f2fs_gc_pinned_control(inode, gc_type, segno);
1459 if (err)
1460 goto out;
1461
1462 if (gc_type == BG_GC) {
1463 if (folio_test_writeback(page_folio(page))) {
1464 err = -EAGAIN;
1465 goto out;
1466 }
1467 set_page_dirty(page);
1468 set_page_private_gcing(page);
1469 } else {
1470 struct f2fs_io_info fio = {
1471 .sbi = F2FS_I_SB(inode),
1472 .ino = inode->i_ino,
1473 .type = DATA,
1474 .temp = COLD,
1475 .op = REQ_OP_WRITE,
1476 .op_flags = REQ_SYNC,
1477 .old_blkaddr = NULL_ADDR,
1478 .page = page,
1479 .encrypted_page = NULL,
1480 .need_lock = LOCK_REQ,
1481 .io_type = FS_GC_DATA_IO,
1482 };
1483 bool is_dirty = PageDirty(page);
1484
1485 retry:
1486 f2fs_wait_on_page_writeback(page, DATA, true, true);
1487
1488 set_page_dirty(page);
1489 if (clear_page_dirty_for_io(page)) {
1490 inode_dec_dirty_pages(inode);
1491 f2fs_remove_dirty_inode(inode);
1492 }
1493
1494 set_page_private_gcing(page);
1495
1496 err = f2fs_do_write_data_page(&fio);
1497 if (err) {
1498 clear_page_private_gcing(page);
1499 if (err == -ENOMEM) {
1500 memalloc_retry_wait(GFP_NOFS);
1501 goto retry;
1502 }
1503 if (is_dirty)
1504 set_page_dirty(page);
1505 }
1506 }
1507 out:
1508 f2fs_put_page(page, 1);
1509 return err;
1510 }
1511
1512 /*
1513 * This function tries to get parent node of victim data block, and identifies
1514 * data block validity. If the block is valid, copy that with cold status and
1515 * modify parent node.
1516 * If the parent node is not valid or the data block address is different,
1517 * the victim data block is ignored.
1518 */
1519 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1520 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1521 bool force_migrate)
1522 {
1523 struct super_block *sb = sbi->sb;
1524 struct f2fs_summary *entry;
1525 block_t start_addr;
1526 int off;
1527 int phase = 0;
1528 int submitted = 0;
1529 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1530
1531 start_addr = START_BLOCK(sbi, segno);
1532
1533 next_step:
1534 entry = sum;
1535
1536 for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1537 struct page *data_page;
1538 struct inode *inode;
1539 struct node_info dni; /* dnode info for the data */
1540 unsigned int ofs_in_node, nofs;
1541 block_t start_bidx;
1542 nid_t nid = le32_to_cpu(entry->nid);
1543
1544 /*
1545 * stop BG_GC if there is not enough free sections.
1546 * Or, stop GC if the segment becomes fully valid caused by
1547 * race condition along with SSR block allocation.
1548 */
1549 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1550 (!force_migrate && get_valid_blocks(sbi, segno, true) ==
1551 CAP_BLKS_PER_SEC(sbi)))
1552 return submitted;
1553
1554 if (check_valid_map(sbi, segno, off) == 0)
1555 continue;
1556
1557 if (phase == 0) {
1558 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1559 META_NAT, true);
1560 continue;
1561 }
1562
1563 if (phase == 1) {
1564 f2fs_ra_node_page(sbi, nid);
1565 continue;
1566 }
1567
1568 /* Get an inode by ino with checking validity */
1569 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1570 continue;
1571
1572 if (phase == 2) {
1573 f2fs_ra_node_page(sbi, dni.ino);
1574 continue;
1575 }
1576
1577 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1578
1579 if (phase == 3) {
1580 int err;
1581
1582 inode = f2fs_iget(sb, dni.ino);
1583 if (IS_ERR(inode))
1584 continue;
1585
1586 if (is_bad_inode(inode) ||
1587 special_file(inode->i_mode)) {
1588 iput(inode);
1589 continue;
1590 }
1591
1592 if (f2fs_has_inline_data(inode)) {
1593 iput(inode);
1594 set_sbi_flag(sbi, SBI_NEED_FSCK);
1595 f2fs_err_ratelimited(sbi,
1596 "inode %lx has both inline_data flag and "
1597 "data block, nid=%u, ofs_in_node=%u",
1598 inode->i_ino, dni.nid, ofs_in_node);
1599 continue;
1600 }
1601
1602 err = f2fs_gc_pinned_control(inode, gc_type, segno);
1603 if (err == -EAGAIN) {
1604 iput(inode);
1605 return submitted;
1606 }
1607
1608 if (!f2fs_down_write_trylock(
1609 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1610 iput(inode);
1611 sbi->skipped_gc_rwsem++;
1612 continue;
1613 }
1614
1615 start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1616 ofs_in_node;
1617
1618 if (f2fs_meta_inode_gc_required(inode)) {
1619 int err = ra_data_block(inode, start_bidx);
1620
1621 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1622 if (err) {
1623 iput(inode);
1624 continue;
1625 }
1626 add_gc_inode(gc_list, inode);
1627 continue;
1628 }
1629
1630 data_page = f2fs_get_read_data_page(inode, start_bidx,
1631 REQ_RAHEAD, true, NULL);
1632 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1633 if (IS_ERR(data_page)) {
1634 iput(inode);
1635 continue;
1636 }
1637
1638 f2fs_put_page(data_page, 0);
1639 add_gc_inode(gc_list, inode);
1640 continue;
1641 }
1642
1643 /* phase 4 */
1644 inode = find_gc_inode(gc_list, dni.ino);
1645 if (inode) {
1646 struct f2fs_inode_info *fi = F2FS_I(inode);
1647 bool locked = false;
1648 int err;
1649
1650 if (S_ISREG(inode->i_mode)) {
1651 if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[WRITE])) {
1652 sbi->skipped_gc_rwsem++;
1653 continue;
1654 }
1655 if (!f2fs_down_write_trylock(
1656 &fi->i_gc_rwsem[READ])) {
1657 sbi->skipped_gc_rwsem++;
1658 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
1659 continue;
1660 }
1661 locked = true;
1662
1663 /* wait for all inflight aio data */
1664 inode_dio_wait(inode);
1665 }
1666
1667 start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1668 + ofs_in_node;
1669 if (f2fs_meta_inode_gc_required(inode))
1670 err = move_data_block(inode, start_bidx,
1671 gc_type, segno, off);
1672 else
1673 err = move_data_page(inode, start_bidx, gc_type,
1674 segno, off);
1675
1676 if (!err && (gc_type == FG_GC ||
1677 f2fs_meta_inode_gc_required(inode)))
1678 submitted++;
1679
1680 if (locked) {
1681 f2fs_up_write(&fi->i_gc_rwsem[READ]);
1682 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
1683 }
1684
1685 stat_inc_data_blk_count(sbi, 1, gc_type);
1686 }
1687 }
1688
1689 if (++phase < 5)
1690 goto next_step;
1691
1692 return submitted;
1693 }
1694
1695 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1696 int gc_type)
1697 {
1698 struct sit_info *sit_i = SIT_I(sbi);
1699 int ret;
1700
1701 down_write(&sit_i->sentry_lock);
1702 ret = f2fs_get_victim(sbi, victim, gc_type, NO_CHECK_TYPE, LFS, 0);
1703 up_write(&sit_i->sentry_lock);
1704 return ret;
1705 }
1706
1707 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1708 unsigned int start_segno,
1709 struct gc_inode_list *gc_list, int gc_type,
1710 bool force_migrate, bool one_time)
1711 {
1712 struct page *sum_page;
1713 struct f2fs_summary_block *sum;
1714 struct blk_plug plug;
1715 unsigned int segno = start_segno;
1716 unsigned int end_segno = start_segno + SEGS_PER_SEC(sbi);
1717 unsigned int sec_end_segno;
1718 int seg_freed = 0, migrated = 0;
1719 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1720 SUM_TYPE_DATA : SUM_TYPE_NODE;
1721 unsigned char data_type = (type == SUM_TYPE_DATA) ? DATA : NODE;
1722 int submitted = 0;
1723
1724 if (__is_large_section(sbi)) {
1725 sec_end_segno = rounddown(end_segno, SEGS_PER_SEC(sbi));
1726
1727 /*
1728 * zone-capacity can be less than zone-size in zoned devices,
1729 * resulting in less than expected usable segments in the zone,
1730 * calculate the end segno in the zone which can be garbage
1731 * collected
1732 */
1733 if (f2fs_sb_has_blkzoned(sbi))
1734 sec_end_segno -= SEGS_PER_SEC(sbi) -
1735 f2fs_usable_segs_in_sec(sbi, segno);
1736
1737 if (gc_type == BG_GC || one_time) {
1738 unsigned int window_granularity =
1739 sbi->migration_window_granularity;
1740
1741 if (f2fs_sb_has_blkzoned(sbi) &&
1742 !has_enough_free_blocks(sbi,
1743 LIMIT_BOOST_ZONED_GC))
1744 window_granularity *= BOOST_GC_MULTIPLE;
1745
1746 end_segno = start_segno + window_granularity;
1747 }
1748
1749 if (end_segno > sec_end_segno)
1750 end_segno = sec_end_segno;
1751 }
1752
1753 sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1754
1755 /* readahead multi ssa blocks those have contiguous address */
1756 if (__is_large_section(sbi))
1757 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1758 end_segno - segno, META_SSA, true);
1759
1760 /* reference all summary page */
1761 while (segno < end_segno) {
1762 sum_page = f2fs_get_sum_page(sbi, segno++);
1763 if (IS_ERR(sum_page)) {
1764 int err = PTR_ERR(sum_page);
1765
1766 end_segno = segno - 1;
1767 for (segno = start_segno; segno < end_segno; segno++) {
1768 sum_page = find_get_page(META_MAPPING(sbi),
1769 GET_SUM_BLOCK(sbi, segno));
1770 f2fs_put_page(sum_page, 0);
1771 f2fs_put_page(sum_page, 0);
1772 }
1773 return err;
1774 }
1775 unlock_page(sum_page);
1776 }
1777
1778 blk_start_plug(&plug);
1779
1780 for (segno = start_segno; segno < end_segno; segno++) {
1781
1782 /* find segment summary of victim */
1783 sum_page = find_get_page(META_MAPPING(sbi),
1784 GET_SUM_BLOCK(sbi, segno));
1785 f2fs_put_page(sum_page, 0);
1786
1787 if (get_valid_blocks(sbi, segno, false) == 0)
1788 goto freed;
1789 if (gc_type == BG_GC && __is_large_section(sbi) &&
1790 migrated >= sbi->migration_granularity)
1791 goto skip;
1792 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1793 goto skip;
1794
1795 sum = page_address(sum_page);
1796 if (type != GET_SUM_TYPE((&sum->footer))) {
1797 f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1798 segno, type, GET_SUM_TYPE((&sum->footer)));
1799 f2fs_stop_checkpoint(sbi, false,
1800 STOP_CP_REASON_CORRUPTED_SUMMARY);
1801 goto skip;
1802 }
1803
1804 /*
1805 * this is to avoid deadlock:
1806 * - lock_page(sum_page) - f2fs_replace_block
1807 * - check_valid_map() - down_write(sentry_lock)
1808 * - down_read(sentry_lock) - change_curseg()
1809 * - lock_page(sum_page)
1810 */
1811 if (type == SUM_TYPE_NODE)
1812 submitted += gc_node_segment(sbi, sum->entries, segno,
1813 gc_type);
1814 else
1815 submitted += gc_data_segment(sbi, sum->entries, gc_list,
1816 segno, gc_type,
1817 force_migrate);
1818
1819 stat_inc_gc_seg_count(sbi, data_type, gc_type);
1820 sbi->gc_reclaimed_segs[sbi->gc_mode]++;
1821 migrated++;
1822
1823 freed:
1824 if (gc_type == FG_GC &&
1825 get_valid_blocks(sbi, segno, false) == 0)
1826 seg_freed++;
1827
1828 if (__is_large_section(sbi))
1829 sbi->next_victim_seg[gc_type] =
1830 (segno + 1 < sec_end_segno) ?
1831 segno + 1 : NULL_SEGNO;
1832 skip:
1833 f2fs_put_page(sum_page, 0);
1834 }
1835
1836 if (submitted)
1837 f2fs_submit_merged_write(sbi, data_type);
1838
1839 blk_finish_plug(&plug);
1840
1841 if (migrated)
1842 stat_inc_gc_sec_count(sbi, data_type, gc_type);
1843
1844 return seg_freed;
1845 }
1846
1847 int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control)
1848 {
1849 int gc_type = gc_control->init_gc_type;
1850 unsigned int segno = gc_control->victim_segno;
1851 int sec_freed = 0, seg_freed = 0, total_freed = 0, total_sec_freed = 0;
1852 int ret = 0;
1853 struct cp_control cpc;
1854 struct gc_inode_list gc_list = {
1855 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1856 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1857 };
1858 unsigned int skipped_round = 0, round = 0;
1859 unsigned int upper_secs;
1860
1861 trace_f2fs_gc_begin(sbi->sb, gc_type, gc_control->no_bg_gc,
1862 gc_control->nr_free_secs,
1863 get_pages(sbi, F2FS_DIRTY_NODES),
1864 get_pages(sbi, F2FS_DIRTY_DENTS),
1865 get_pages(sbi, F2FS_DIRTY_IMETA),
1866 free_sections(sbi),
1867 free_segments(sbi),
1868 reserved_segments(sbi),
1869 prefree_segments(sbi));
1870
1871 cpc.reason = __get_cp_reason(sbi);
1872 gc_more:
1873 sbi->skipped_gc_rwsem = 0;
1874 if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1875 ret = -EINVAL;
1876 goto stop;
1877 }
1878 if (unlikely(f2fs_cp_error(sbi))) {
1879 ret = -EIO;
1880 goto stop;
1881 }
1882
1883 /* Let's run FG_GC, if we don't have enough space. */
1884 if (has_not_enough_free_secs(sbi, 0, 0)) {
1885 gc_type = FG_GC;
1886
1887 /*
1888 * For example, if there are many prefree_segments below given
1889 * threshold, we can make them free by checkpoint. Then, we
1890 * secure free segments which doesn't need fggc any more.
1891 */
1892 if (prefree_segments(sbi)) {
1893 stat_inc_cp_call_count(sbi, TOTAL_CALL);
1894 ret = f2fs_write_checkpoint(sbi, &cpc);
1895 if (ret)
1896 goto stop;
1897 /* Reset due to checkpoint */
1898 sec_freed = 0;
1899 }
1900 }
1901
1902 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1903 if (gc_type == BG_GC && gc_control->no_bg_gc) {
1904 ret = -EINVAL;
1905 goto stop;
1906 }
1907 retry:
1908 ret = __get_victim(sbi, &segno, gc_type);
1909 if (ret) {
1910 /* allow to search victim from sections has pinned data */
1911 if (ret == -ENODATA && gc_type == FG_GC &&
1912 f2fs_pinned_section_exists(DIRTY_I(sbi))) {
1913 f2fs_unpin_all_sections(sbi, false);
1914 goto retry;
1915 }
1916 goto stop;
1917 }
1918
1919 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type,
1920 gc_control->should_migrate_blocks,
1921 gc_control->one_time);
1922 if (seg_freed < 0)
1923 goto stop;
1924
1925 total_freed += seg_freed;
1926
1927 if (seg_freed == f2fs_usable_segs_in_sec(sbi, segno)) {
1928 sec_freed++;
1929 total_sec_freed++;
1930 }
1931
1932 if (gc_control->one_time)
1933 goto stop;
1934
1935 if (gc_type == FG_GC) {
1936 sbi->cur_victim_sec = NULL_SEGNO;
1937
1938 if (has_enough_free_secs(sbi, sec_freed, 0)) {
1939 if (!gc_control->no_bg_gc &&
1940 total_sec_freed < gc_control->nr_free_secs)
1941 goto go_gc_more;
1942 goto stop;
1943 }
1944 if (sbi->skipped_gc_rwsem)
1945 skipped_round++;
1946 round++;
1947 if (skipped_round > MAX_SKIP_GC_COUNT &&
1948 skipped_round * 2 >= round) {
1949 stat_inc_cp_call_count(sbi, TOTAL_CALL);
1950 ret = f2fs_write_checkpoint(sbi, &cpc);
1951 goto stop;
1952 }
1953 } else if (has_enough_free_secs(sbi, 0, 0)) {
1954 goto stop;
1955 }
1956
1957 __get_secs_required(sbi, NULL, &upper_secs, NULL);
1958
1959 /*
1960 * Write checkpoint to reclaim prefree segments.
1961 * We need more three extra sections for writer's data/node/dentry.
1962 */
1963 if (free_sections(sbi) <= upper_secs + NR_GC_CHECKPOINT_SECS &&
1964 prefree_segments(sbi)) {
1965 stat_inc_cp_call_count(sbi, TOTAL_CALL);
1966 ret = f2fs_write_checkpoint(sbi, &cpc);
1967 if (ret)
1968 goto stop;
1969 /* Reset due to checkpoint */
1970 sec_freed = 0;
1971 }
1972 go_gc_more:
1973 segno = NULL_SEGNO;
1974 goto gc_more;
1975
1976 stop:
1977 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1978 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno;
1979
1980 if (gc_type == FG_GC)
1981 f2fs_unpin_all_sections(sbi, true);
1982
1983 trace_f2fs_gc_end(sbi->sb, ret, total_freed, total_sec_freed,
1984 get_pages(sbi, F2FS_DIRTY_NODES),
1985 get_pages(sbi, F2FS_DIRTY_DENTS),
1986 get_pages(sbi, F2FS_DIRTY_IMETA),
1987 free_sections(sbi),
1988 free_segments(sbi),
1989 reserved_segments(sbi),
1990 prefree_segments(sbi));
1991
1992 f2fs_up_write(&sbi->gc_lock);
1993
1994 put_gc_inode(&gc_list);
1995
1996 if (gc_control->err_gc_skipped && !ret)
1997 ret = total_sec_freed ? 0 : -EAGAIN;
1998 return ret;
1999 }
2000
2001 int __init f2fs_create_garbage_collection_cache(void)
2002 {
2003 victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
2004 sizeof(struct victim_entry));
2005 return victim_entry_slab ? 0 : -ENOMEM;
2006 }
2007
2008 void f2fs_destroy_garbage_collection_cache(void)
2009 {
2010 kmem_cache_destroy(victim_entry_slab);
2011 }
2012
2013 static void init_atgc_management(struct f2fs_sb_info *sbi)
2014 {
2015 struct atgc_management *am = &sbi->am;
2016
2017 if (test_opt(sbi, ATGC) &&
2018 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
2019 am->atgc_enabled = true;
2020
2021 am->root = RB_ROOT_CACHED;
2022 INIT_LIST_HEAD(&am->victim_list);
2023 am->victim_count = 0;
2024
2025 am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
2026 am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
2027 am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
2028 am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
2029 }
2030
2031 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
2032 {
2033 sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
2034
2035 /* give warm/cold data area from slower device */
2036 if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
2037 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
2038 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
2039
2040 init_atgc_management(sbi);
2041 }
2042
2043 int f2fs_gc_range(struct f2fs_sb_info *sbi,
2044 unsigned int start_seg, unsigned int end_seg,
2045 bool dry_run, unsigned int dry_run_sections)
2046 {
2047 unsigned int segno;
2048 unsigned int gc_secs = dry_run_sections;
2049
2050 if (unlikely(f2fs_cp_error(sbi)))
2051 return -EIO;
2052
2053 for (segno = start_seg; segno <= end_seg; segno += SEGS_PER_SEC(sbi)) {
2054 struct gc_inode_list gc_list = {
2055 .ilist = LIST_HEAD_INIT(gc_list.ilist),
2056 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
2057 };
2058
2059 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true, false);
2060 put_gc_inode(&gc_list);
2061
2062 if (!dry_run && get_valid_blocks(sbi, segno, true))
2063 return -EAGAIN;
2064 if (dry_run && dry_run_sections &&
2065 !get_valid_blocks(sbi, segno, true) && --gc_secs == 0)
2066 break;
2067
2068 if (fatal_signal_pending(current))
2069 return -ERESTARTSYS;
2070 }
2071
2072 return 0;
2073 }
2074
2075 static int free_segment_range(struct f2fs_sb_info *sbi,
2076 unsigned int secs, bool dry_run)
2077 {
2078 unsigned int next_inuse, start, end;
2079 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
2080 int gc_mode, gc_type;
2081 int err = 0;
2082 int type;
2083
2084 /* Force block allocation for GC */
2085 MAIN_SECS(sbi) -= secs;
2086 start = MAIN_SECS(sbi) * SEGS_PER_SEC(sbi);
2087 end = MAIN_SEGS(sbi) - 1;
2088
2089 mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2090 for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
2091 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
2092 SIT_I(sbi)->last_victim[gc_mode] = 0;
2093
2094 for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
2095 if (sbi->next_victim_seg[gc_type] >= start)
2096 sbi->next_victim_seg[gc_type] = NULL_SEGNO;
2097 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2098
2099 /* Move out cursegs from the target range */
2100 for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++) {
2101 err = f2fs_allocate_segment_for_resize(sbi, type, start, end);
2102 if (err)
2103 goto out;
2104 }
2105
2106 /* do GC to move out valid blocks in the range */
2107 err = f2fs_gc_range(sbi, start, end, dry_run, 0);
2108 if (err || dry_run)
2109 goto out;
2110
2111 stat_inc_cp_call_count(sbi, TOTAL_CALL);
2112 err = f2fs_write_checkpoint(sbi, &cpc);
2113 if (err)
2114 goto out;
2115
2116 next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
2117 if (next_inuse <= end) {
2118 f2fs_err(sbi, "segno %u should be free but still inuse!",
2119 next_inuse);
2120 f2fs_bug_on(sbi, 1);
2121 }
2122 out:
2123 MAIN_SECS(sbi) += secs;
2124 return err;
2125 }
2126
2127 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
2128 {
2129 struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
2130 int section_count;
2131 int segment_count;
2132 int segment_count_main;
2133 long long block_count;
2134 int segs = secs * SEGS_PER_SEC(sbi);
2135
2136 f2fs_down_write(&sbi->sb_lock);
2137
2138 section_count = le32_to_cpu(raw_sb->section_count);
2139 segment_count = le32_to_cpu(raw_sb->segment_count);
2140 segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
2141 block_count = le64_to_cpu(raw_sb->block_count);
2142
2143 raw_sb->section_count = cpu_to_le32(section_count + secs);
2144 raw_sb->segment_count = cpu_to_le32(segment_count + segs);
2145 raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
2146 raw_sb->block_count = cpu_to_le64(block_count +
2147 (long long)SEGS_TO_BLKS(sbi, segs));
2148 if (f2fs_is_multi_device(sbi)) {
2149 int last_dev = sbi->s_ndevs - 1;
2150 int dev_segs =
2151 le32_to_cpu(raw_sb->devs[last_dev].total_segments);
2152
2153 raw_sb->devs[last_dev].total_segments =
2154 cpu_to_le32(dev_segs + segs);
2155 }
2156
2157 f2fs_up_write(&sbi->sb_lock);
2158 }
2159
2160 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
2161 {
2162 int segs = secs * SEGS_PER_SEC(sbi);
2163 long long blks = SEGS_TO_BLKS(sbi, segs);
2164 long long user_block_count =
2165 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
2166
2167 SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
2168 MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
2169 MAIN_SECS(sbi) += secs;
2170 FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
2171 FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
2172 F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
2173
2174 if (f2fs_is_multi_device(sbi)) {
2175 int last_dev = sbi->s_ndevs - 1;
2176
2177 FDEV(last_dev).total_segments =
2178 (int)FDEV(last_dev).total_segments + segs;
2179 FDEV(last_dev).end_blk =
2180 (long long)FDEV(last_dev).end_blk + blks;
2181 #ifdef CONFIG_BLK_DEV_ZONED
2182 FDEV(last_dev).nr_blkz = FDEV(last_dev).nr_blkz +
2183 div_u64(blks, sbi->blocks_per_blkz);
2184 #endif
2185 }
2186 }
2187
2188 int f2fs_resize_fs(struct file *filp, __u64 block_count)
2189 {
2190 struct f2fs_sb_info *sbi = F2FS_I_SB(file_inode(filp));
2191 __u64 old_block_count, shrunk_blocks;
2192 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
2193 unsigned int secs;
2194 int err = 0;
2195 __u32 rem;
2196
2197 old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
2198 if (block_count > old_block_count)
2199 return -EINVAL;
2200
2201 if (f2fs_is_multi_device(sbi)) {
2202 int last_dev = sbi->s_ndevs - 1;
2203 __u64 last_segs = FDEV(last_dev).total_segments;
2204
2205 if (block_count + SEGS_TO_BLKS(sbi, last_segs) <=
2206 old_block_count)
2207 return -EINVAL;
2208 }
2209
2210 /* new fs size should align to section size */
2211 div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
2212 if (rem)
2213 return -EINVAL;
2214
2215 if (block_count == old_block_count)
2216 return 0;
2217
2218 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2219 f2fs_err(sbi, "Should run fsck to repair first.");
2220 return -EFSCORRUPTED;
2221 }
2222
2223 if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2224 f2fs_err(sbi, "Checkpoint should be enabled.");
2225 return -EINVAL;
2226 }
2227
2228 err = mnt_want_write_file(filp);
2229 if (err)
2230 return err;
2231
2232 shrunk_blocks = old_block_count - block_count;
2233 secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2234
2235 /* stop other GC */
2236 if (!f2fs_down_write_trylock(&sbi->gc_lock)) {
2237 err = -EAGAIN;
2238 goto out_drop_write;
2239 }
2240
2241 /* stop CP to protect MAIN_SEC in free_segment_range */
2242 f2fs_lock_op(sbi);
2243
2244 spin_lock(&sbi->stat_lock);
2245 if (shrunk_blocks + valid_user_blocks(sbi) +
2246 sbi->current_reserved_blocks + sbi->unusable_block_count +
2247 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2248 err = -ENOSPC;
2249 spin_unlock(&sbi->stat_lock);
2250
2251 if (err)
2252 goto out_unlock;
2253
2254 err = free_segment_range(sbi, secs, true);
2255
2256 out_unlock:
2257 f2fs_unlock_op(sbi);
2258 f2fs_up_write(&sbi->gc_lock);
2259 out_drop_write:
2260 mnt_drop_write_file(filp);
2261 if (err)
2262 return err;
2263
2264 err = freeze_super(sbi->sb, FREEZE_HOLDER_USERSPACE);
2265 if (err)
2266 return err;
2267
2268 if (f2fs_readonly(sbi->sb)) {
2269 err = thaw_super(sbi->sb, FREEZE_HOLDER_USERSPACE);
2270 if (err)
2271 return err;
2272 return -EROFS;
2273 }
2274
2275 f2fs_down_write(&sbi->gc_lock);
2276 f2fs_down_write(&sbi->cp_global_sem);
2277
2278 spin_lock(&sbi->stat_lock);
2279 if (shrunk_blocks + valid_user_blocks(sbi) +
2280 sbi->current_reserved_blocks + sbi->unusable_block_count +
2281 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2282 err = -ENOSPC;
2283 else
2284 sbi->user_block_count -= shrunk_blocks;
2285 spin_unlock(&sbi->stat_lock);
2286 if (err)
2287 goto out_err;
2288
2289 set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2290 err = free_segment_range(sbi, secs, false);
2291 if (err)
2292 goto recover_out;
2293
2294 update_sb_metadata(sbi, -secs);
2295
2296 err = f2fs_commit_super(sbi, false);
2297 if (err) {
2298 update_sb_metadata(sbi, secs);
2299 goto recover_out;
2300 }
2301
2302 update_fs_metadata(sbi, -secs);
2303 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2304 set_sbi_flag(sbi, SBI_IS_DIRTY);
2305
2306 stat_inc_cp_call_count(sbi, TOTAL_CALL);
2307 err = f2fs_write_checkpoint(sbi, &cpc);
2308 if (err) {
2309 update_fs_metadata(sbi, secs);
2310 update_sb_metadata(sbi, secs);
2311 f2fs_commit_super(sbi, false);
2312 }
2313 recover_out:
2314 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2315 if (err) {
2316 set_sbi_flag(sbi, SBI_NEED_FSCK);
2317 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2318
2319 spin_lock(&sbi->stat_lock);
2320 sbi->user_block_count += shrunk_blocks;
2321 spin_unlock(&sbi->stat_lock);
2322 }
2323 out_err:
2324 f2fs_up_write(&sbi->cp_global_sem);
2325 f2fs_up_write(&sbi->gc_lock);
2326 thaw_super(sbi->sb, FREEZE_HOLDER_USERSPACE);
2327 return err;
2328 }
2329
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.