2343 lines · c
1// SPDX-License-Identifier: GPL-2.02/*3 * fs/f2fs/gc.c4 *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 26static struct kmem_cache *victim_entry_slab;27 28static unsigned int count_bits(const unsigned long *addr,29 unsigned int offset, unsigned int len);30 31static 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 in92 * bdev's request list.93 *94 * Note) We have to avoid triggering GCs frequently.95 * Because it is possible that some segments can be96 * 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,123 gc_th->no_zoned_gc_percent)) {124 wait_ms = gc_th->no_gc_sleep_time;125 f2fs_up_write(&sbi->gc_lock);126 goto next;127 }128 if (wait_ms == gc_th->no_gc_sleep_time)129 wait_ms = gc_th->max_sleep_time;130 }131 132 if (need_to_boost_gc(sbi)) {133 decrease_sleep_time(gc_th, &wait_ms);134 if (f2fs_sb_has_blkzoned(sbi))135 gc_control.one_time = true;136 } else {137 increase_sleep_time(gc_th, &wait_ms);138 }139do_gc:140 stat_inc_gc_call_count(sbi, foreground ?141 FOREGROUND : BACKGROUND);142 143 sync_mode = (F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC) ||144 gc_control.one_time;145 146 /* foreground GC was been triggered via f2fs_balance_fs() */147 if (foreground)148 sync_mode = false;149 150 gc_control.init_gc_type = sync_mode ? FG_GC : BG_GC;151 gc_control.no_bg_gc = foreground;152 gc_control.nr_free_secs = foreground ? 1 : 0;153 154 /* if return value is not zero, no victim was selected */155 if (f2fs_gc(sbi, &gc_control)) {156 /* don't bother wait_ms by foreground gc */157 if (!foreground)158 wait_ms = gc_th->no_gc_sleep_time;159 } else {160 /* reset wait_ms to default sleep time */161 if (wait_ms == gc_th->no_gc_sleep_time)162 wait_ms = gc_th->min_sleep_time;163 }164 165 if (foreground)166 wake_up_all(&gc_th->fggc_wq);167 168 trace_f2fs_background_gc(sbi->sb, wait_ms,169 prefree_segments(sbi), free_segments(sbi));170 171 /* balancing f2fs's metadata periodically */172 f2fs_balance_fs_bg(sbi, true);173next:174 if (sbi->gc_mode != GC_NORMAL) {175 spin_lock(&sbi->gc_remaining_trials_lock);176 if (sbi->gc_remaining_trials) {177 sbi->gc_remaining_trials--;178 if (!sbi->gc_remaining_trials)179 sbi->gc_mode = GC_NORMAL;180 }181 spin_unlock(&sbi->gc_remaining_trials_lock);182 }183 sb_end_write(sbi->sb);184 185 } while (!kthread_should_stop());186 return 0;187}188 189int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)190{191 struct f2fs_gc_kthread *gc_th;192 dev_t dev = sbi->sb->s_bdev->bd_dev;193 194 gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);195 if (!gc_th)196 return -ENOMEM;197 198 gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;199 gc_th->valid_thresh_ratio = DEF_GC_THREAD_VALID_THRESH_RATIO;200 201 if (f2fs_sb_has_blkzoned(sbi)) {202 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME_ZONED;203 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME_ZONED;204 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME_ZONED;205 gc_th->no_zoned_gc_percent = LIMIT_NO_ZONED_GC;206 gc_th->boost_zoned_gc_percent = LIMIT_BOOST_ZONED_GC;207 } else {208 gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;209 gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;210 gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;211 gc_th->no_zoned_gc_percent = 0;212 gc_th->boost_zoned_gc_percent = 0;213 }214 215 gc_th->gc_wake = false;216 217 sbi->gc_thread = gc_th;218 init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);219 init_waitqueue_head(&sbi->gc_thread->fggc_wq);220 sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,221 "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));222 if (IS_ERR(gc_th->f2fs_gc_task)) {223 int err = PTR_ERR(gc_th->f2fs_gc_task);224 225 kfree(gc_th);226 sbi->gc_thread = NULL;227 return err;228 }229 230 return 0;231}232 233void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)234{235 struct f2fs_gc_kthread *gc_th = sbi->gc_thread;236 237 if (!gc_th)238 return;239 kthread_stop(gc_th->f2fs_gc_task);240 wake_up_all(&gc_th->fggc_wq);241 kfree(gc_th);242 sbi->gc_thread = NULL;243}244 245static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)246{247 int gc_mode;248 249 if (gc_type == BG_GC) {250 if (sbi->am.atgc_enabled)251 gc_mode = GC_AT;252 else253 gc_mode = GC_CB;254 } else {255 gc_mode = GC_GREEDY;256 }257 258 switch (sbi->gc_mode) {259 case GC_IDLE_CB:260 gc_mode = GC_CB;261 break;262 case GC_IDLE_GREEDY:263 case GC_URGENT_HIGH:264 gc_mode = GC_GREEDY;265 break;266 case GC_IDLE_AT:267 gc_mode = GC_AT;268 break;269 }270 271 return gc_mode;272}273 274static void select_policy(struct f2fs_sb_info *sbi, int gc_type,275 int type, struct victim_sel_policy *p)276{277 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);278 279 if (p->alloc_mode == SSR) {280 p->gc_mode = GC_GREEDY;281 p->dirty_bitmap = dirty_i->dirty_segmap[type];282 p->max_search = dirty_i->nr_dirty[type];283 p->ofs_unit = 1;284 } else if (p->alloc_mode == AT_SSR) {285 p->gc_mode = GC_GREEDY;286 p->dirty_bitmap = dirty_i->dirty_segmap[type];287 p->max_search = dirty_i->nr_dirty[type];288 p->ofs_unit = 1;289 } else {290 p->gc_mode = select_gc_type(sbi, gc_type);291 p->ofs_unit = SEGS_PER_SEC(sbi);292 if (__is_large_section(sbi)) {293 p->dirty_bitmap = dirty_i->dirty_secmap;294 p->max_search = count_bits(p->dirty_bitmap,295 0, MAIN_SECS(sbi));296 } else {297 p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];298 p->max_search = dirty_i->nr_dirty[DIRTY];299 }300 }301 302 /*303 * adjust candidates range, should select all dirty segments for304 * foreground GC and urgent GC cases.305 */306 if (gc_type != FG_GC &&307 (sbi->gc_mode != GC_URGENT_HIGH) &&308 (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&309 p->max_search > sbi->max_victim_search)310 p->max_search = sbi->max_victim_search;311 312 /* let's select beginning hot/small space first. */313 if (f2fs_need_rand_seg(sbi))314 p->offset = get_random_u32_below(MAIN_SECS(sbi) *315 SEGS_PER_SEC(sbi));316 else if (type == CURSEG_HOT_DATA || IS_NODESEG(type))317 p->offset = 0;318 else319 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];320}321 322static unsigned int get_max_cost(struct f2fs_sb_info *sbi,323 struct victim_sel_policy *p)324{325 /* SSR allocates in a segment unit */326 if (p->alloc_mode == SSR)327 return BLKS_PER_SEG(sbi);328 else if (p->alloc_mode == AT_SSR)329 return UINT_MAX;330 331 /* LFS */332 if (p->gc_mode == GC_GREEDY)333 return SEGS_TO_BLKS(sbi, 2 * p->ofs_unit);334 else if (p->gc_mode == GC_CB)335 return UINT_MAX;336 else if (p->gc_mode == GC_AT)337 return UINT_MAX;338 else /* No other gc_mode */339 return 0;340}341 342static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)343{344 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);345 unsigned int secno;346 347 /*348 * If the gc_type is FG_GC, we can select victim segments349 * selected by background GC before.350 * Those segments guarantee they have small valid blocks.351 */352 for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {353 if (sec_usage_check(sbi, secno))354 continue;355 clear_bit(secno, dirty_i->victim_secmap);356 return GET_SEG_FROM_SEC(sbi, secno);357 }358 return NULL_SEGNO;359}360 361static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)362{363 struct sit_info *sit_i = SIT_I(sbi);364 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);365 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);366 unsigned long long mtime = 0;367 unsigned int vblocks;368 unsigned char age = 0;369 unsigned char u;370 unsigned int i;371 unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi);372 373 for (i = 0; i < usable_segs_per_sec; i++)374 mtime += get_seg_entry(sbi, start + i)->mtime;375 vblocks = get_valid_blocks(sbi, segno, true);376 377 mtime = div_u64(mtime, usable_segs_per_sec);378 vblocks = div_u64(vblocks, usable_segs_per_sec);379 380 u = BLKS_TO_SEGS(sbi, vblocks * 100);381 382 /* Handle if the system time has changed by the user */383 if (mtime < sit_i->min_mtime)384 sit_i->min_mtime = mtime;385 if (mtime > sit_i->max_mtime)386 sit_i->max_mtime = mtime;387 if (sit_i->max_mtime != sit_i->min_mtime)388 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),389 sit_i->max_mtime - sit_i->min_mtime);390 391 return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));392}393 394static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,395 unsigned int segno, struct victim_sel_policy *p)396{397 if (p->alloc_mode == SSR)398 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;399 400 if (p->one_time_gc && (get_valid_blocks(sbi, segno, true) >=401 CAP_BLKS_PER_SEC(sbi) * sbi->gc_thread->valid_thresh_ratio /402 100))403 return UINT_MAX;404 405 /* alloc_mode == LFS */406 if (p->gc_mode == GC_GREEDY)407 return get_valid_blocks(sbi, segno, true);408 else if (p->gc_mode == GC_CB)409 return get_cb_cost(sbi, segno);410 411 f2fs_bug_on(sbi, 1);412 return 0;413}414 415static unsigned int count_bits(const unsigned long *addr,416 unsigned int offset, unsigned int len)417{418 unsigned int end = offset + len, sum = 0;419 420 while (offset < end) {421 if (test_bit(offset++, addr))422 ++sum;423 }424 return sum;425}426 427static bool f2fs_check_victim_tree(struct f2fs_sb_info *sbi,428 struct rb_root_cached *root)429{430#ifdef CONFIG_F2FS_CHECK_FS431 struct rb_node *cur = rb_first_cached(root), *next;432 struct victim_entry *cur_ve, *next_ve;433 434 while (cur) {435 next = rb_next(cur);436 if (!next)437 return true;438 439 cur_ve = rb_entry(cur, struct victim_entry, rb_node);440 next_ve = rb_entry(next, struct victim_entry, rb_node);441 442 if (cur_ve->mtime > next_ve->mtime) {443 f2fs_info(sbi, "broken victim_rbtree, "444 "cur_mtime(%llu) next_mtime(%llu)",445 cur_ve->mtime, next_ve->mtime);446 return false;447 }448 cur = next;449 }450#endif451 return true;452}453 454static struct victim_entry *__lookup_victim_entry(struct f2fs_sb_info *sbi,455 unsigned long long mtime)456{457 struct atgc_management *am = &sbi->am;458 struct rb_node *node = am->root.rb_root.rb_node;459 struct victim_entry *ve = NULL;460 461 while (node) {462 ve = rb_entry(node, struct victim_entry, rb_node);463 464 if (mtime < ve->mtime)465 node = node->rb_left;466 else467 node = node->rb_right;468 }469 return ve;470}471 472static struct victim_entry *__create_victim_entry(struct f2fs_sb_info *sbi,473 unsigned long long mtime, unsigned int segno)474{475 struct atgc_management *am = &sbi->am;476 struct victim_entry *ve;477 478 ve = f2fs_kmem_cache_alloc(victim_entry_slab, GFP_NOFS, true, NULL);479 480 ve->mtime = mtime;481 ve->segno = segno;482 483 list_add_tail(&ve->list, &am->victim_list);484 am->victim_count++;485 486 return ve;487}488 489static void __insert_victim_entry(struct f2fs_sb_info *sbi,490 unsigned long long mtime, unsigned int segno)491{492 struct atgc_management *am = &sbi->am;493 struct rb_root_cached *root = &am->root;494 struct rb_node **p = &root->rb_root.rb_node;495 struct rb_node *parent = NULL;496 struct victim_entry *ve;497 bool left_most = true;498 499 /* look up rb tree to find parent node */500 while (*p) {501 parent = *p;502 ve = rb_entry(parent, struct victim_entry, rb_node);503 504 if (mtime < ve->mtime) {505 p = &(*p)->rb_left;506 } else {507 p = &(*p)->rb_right;508 left_most = false;509 }510 }511 512 ve = __create_victim_entry(sbi, mtime, segno);513 514 rb_link_node(&ve->rb_node, parent, p);515 rb_insert_color_cached(&ve->rb_node, root, left_most);516}517 518static void add_victim_entry(struct f2fs_sb_info *sbi,519 struct victim_sel_policy *p, unsigned int segno)520{521 struct sit_info *sit_i = SIT_I(sbi);522 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);523 unsigned int start = GET_SEG_FROM_SEC(sbi, secno);524 unsigned long long mtime = 0;525 unsigned int i;526 527 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {528 if (p->gc_mode == GC_AT &&529 get_valid_blocks(sbi, segno, true) == 0)530 return;531 }532 533 for (i = 0; i < SEGS_PER_SEC(sbi); i++)534 mtime += get_seg_entry(sbi, start + i)->mtime;535 mtime = div_u64(mtime, SEGS_PER_SEC(sbi));536 537 /* Handle if the system time has changed by the user */538 if (mtime < sit_i->min_mtime)539 sit_i->min_mtime = mtime;540 if (mtime > sit_i->max_mtime)541 sit_i->max_mtime = mtime;542 if (mtime < sit_i->dirty_min_mtime)543 sit_i->dirty_min_mtime = mtime;544 if (mtime > sit_i->dirty_max_mtime)545 sit_i->dirty_max_mtime = mtime;546 547 /* don't choose young section as candidate */548 if (sit_i->dirty_max_mtime - mtime < p->age_threshold)549 return;550 551 __insert_victim_entry(sbi, mtime, segno);552}553 554static void atgc_lookup_victim(struct f2fs_sb_info *sbi,555 struct victim_sel_policy *p)556{557 struct sit_info *sit_i = SIT_I(sbi);558 struct atgc_management *am = &sbi->am;559 struct rb_root_cached *root = &am->root;560 struct rb_node *node;561 struct victim_entry *ve;562 unsigned long long total_time;563 unsigned long long age, u, accu;564 unsigned long long max_mtime = sit_i->dirty_max_mtime;565 unsigned long long min_mtime = sit_i->dirty_min_mtime;566 unsigned int sec_blocks = CAP_BLKS_PER_SEC(sbi);567 unsigned int vblocks;568 unsigned int dirty_threshold = max(am->max_candidate_count,569 am->candidate_ratio *570 am->victim_count / 100);571 unsigned int age_weight = am->age_weight;572 unsigned int cost;573 unsigned int iter = 0;574 575 if (max_mtime < min_mtime)576 return;577 578 max_mtime += 1;579 total_time = max_mtime - min_mtime;580 581 accu = div64_u64(ULLONG_MAX, total_time);582 accu = min_t(unsigned long long, div_u64(accu, 100),583 DEFAULT_ACCURACY_CLASS);584 585 node = rb_first_cached(root);586next:587 ve = rb_entry_safe(node, struct victim_entry, rb_node);588 if (!ve)589 return;590 591 if (ve->mtime >= max_mtime || ve->mtime < min_mtime)592 goto skip;593 594 /* age = 10000 * x% * 60 */595 age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *596 age_weight;597 598 vblocks = get_valid_blocks(sbi, ve->segno, true);599 f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);600 601 /* u = 10000 * x% * 40 */602 u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *603 (100 - age_weight);604 605 f2fs_bug_on(sbi, age + u >= UINT_MAX);606 607 cost = UINT_MAX - (age + u);608 iter++;609 610 if (cost < p->min_cost ||611 (cost == p->min_cost && age > p->oldest_age)) {612 p->min_cost = cost;613 p->oldest_age = age;614 p->min_segno = ve->segno;615 }616skip:617 if (iter < dirty_threshold) {618 node = rb_next(node);619 goto next;620 }621}622 623/*624 * select candidates around source section in range of625 * [target - dirty_threshold, target + dirty_threshold]626 */627static void atssr_lookup_victim(struct f2fs_sb_info *sbi,628 struct victim_sel_policy *p)629{630 struct sit_info *sit_i = SIT_I(sbi);631 struct atgc_management *am = &sbi->am;632 struct victim_entry *ve;633 unsigned long long age;634 unsigned long long max_mtime = sit_i->dirty_max_mtime;635 unsigned long long min_mtime = sit_i->dirty_min_mtime;636 unsigned int vblocks;637 unsigned int dirty_threshold = max(am->max_candidate_count,638 am->candidate_ratio *639 am->victim_count / 100);640 unsigned int cost, iter;641 int stage = 0;642 643 if (max_mtime < min_mtime)644 return;645 max_mtime += 1;646next_stage:647 iter = 0;648 ve = __lookup_victim_entry(sbi, p->age);649next_node:650 if (!ve) {651 if (stage++ == 0)652 goto next_stage;653 return;654 }655 656 if (ve->mtime >= max_mtime || ve->mtime < min_mtime)657 goto skip_node;658 659 age = max_mtime - ve->mtime;660 661 vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;662 f2fs_bug_on(sbi, !vblocks);663 664 /* rare case */665 if (vblocks == BLKS_PER_SEG(sbi))666 goto skip_node;667 668 iter++;669 670 age = max_mtime - abs(p->age - age);671 cost = UINT_MAX - vblocks;672 673 if (cost < p->min_cost ||674 (cost == p->min_cost && age > p->oldest_age)) {675 p->min_cost = cost;676 p->oldest_age = age;677 p->min_segno = ve->segno;678 }679skip_node:680 if (iter < dirty_threshold) {681 ve = rb_entry(stage == 0 ? rb_prev(&ve->rb_node) :682 rb_next(&ve->rb_node),683 struct victim_entry, rb_node);684 goto next_node;685 }686 687 if (stage++ == 0)688 goto next_stage;689}690 691static void lookup_victim_by_age(struct f2fs_sb_info *sbi,692 struct victim_sel_policy *p)693{694 f2fs_bug_on(sbi, !f2fs_check_victim_tree(sbi, &sbi->am.root));695 696 if (p->gc_mode == GC_AT)697 atgc_lookup_victim(sbi, p);698 else if (p->alloc_mode == AT_SSR)699 atssr_lookup_victim(sbi, p);700 else701 f2fs_bug_on(sbi, 1);702}703 704static void release_victim_entry(struct f2fs_sb_info *sbi)705{706 struct atgc_management *am = &sbi->am;707 struct victim_entry *ve, *tmp;708 709 list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {710 list_del(&ve->list);711 kmem_cache_free(victim_entry_slab, ve);712 am->victim_count--;713 }714 715 am->root = RB_ROOT_CACHED;716 717 f2fs_bug_on(sbi, am->victim_count);718 f2fs_bug_on(sbi, !list_empty(&am->victim_list));719}720 721static bool f2fs_pin_section(struct f2fs_sb_info *sbi, unsigned int segno)722{723 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);724 unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);725 726 if (!dirty_i->enable_pin_section)727 return false;728 if (!test_and_set_bit(secno, dirty_i->pinned_secmap))729 dirty_i->pinned_secmap_cnt++;730 return true;731}732 733static bool f2fs_pinned_section_exists(struct dirty_seglist_info *dirty_i)734{735 return dirty_i->pinned_secmap_cnt;736}737 738static bool f2fs_section_is_pinned(struct dirty_seglist_info *dirty_i,739 unsigned int secno)740{741 return dirty_i->enable_pin_section &&742 f2fs_pinned_section_exists(dirty_i) &&743 test_bit(secno, dirty_i->pinned_secmap);744}745 746static void f2fs_unpin_all_sections(struct f2fs_sb_info *sbi, bool enable)747{748 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));749 750 if (f2fs_pinned_section_exists(DIRTY_I(sbi))) {751 memset(DIRTY_I(sbi)->pinned_secmap, 0, bitmap_size);752 DIRTY_I(sbi)->pinned_secmap_cnt = 0;753 }754 DIRTY_I(sbi)->enable_pin_section = enable;755}756 757static int f2fs_gc_pinned_control(struct inode *inode, int gc_type,758 unsigned int segno)759{760 if (!f2fs_is_pinned_file(inode))761 return 0;762 if (gc_type != FG_GC)763 return -EBUSY;764 if (!f2fs_pin_section(F2FS_I_SB(inode), segno))765 f2fs_pin_file_control(inode, true);766 return -EAGAIN;767}768 769/*770 * This function is called from two paths.771 * One is garbage collection and the other is SSR segment selection.772 * When it is called during GC, it just gets a victim segment773 * and it does not remove it from dirty seglist.774 * When it is called from SSR segment selection, it finds a segment775 * which has minimum valid blocks and removes it from dirty seglist.776 */777int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,778 int gc_type, int type, char alloc_mode,779 unsigned long long age, bool one_time)780{781 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);782 struct sit_info *sm = SIT_I(sbi);783 struct victim_sel_policy p;784 unsigned int secno, last_victim;785 unsigned int last_segment;786 unsigned int nsearched;787 bool is_atgc;788 int ret = 0;789 790 mutex_lock(&dirty_i->seglist_lock);791 last_segment = MAIN_SECS(sbi) * SEGS_PER_SEC(sbi);792 793 p.alloc_mode = alloc_mode;794 p.age = age;795 p.age_threshold = sbi->am.age_threshold;796 p.one_time_gc = one_time;797 798retry:799 select_policy(sbi, gc_type, type, &p);800 p.min_segno = NULL_SEGNO;801 p.oldest_age = 0;802 p.min_cost = get_max_cost(sbi, &p);803 804 is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);805 nsearched = 0;806 807 if (is_atgc)808 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;809 810 if (*result != NULL_SEGNO) {811 if (!get_valid_blocks(sbi, *result, false)) {812 ret = -ENODATA;813 goto out;814 }815 816 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))817 ret = -EBUSY;818 else819 p.min_segno = *result;820 goto out;821 }822 823 ret = -ENODATA;824 if (p.max_search == 0)825 goto out;826 827 if (__is_large_section(sbi) && p.alloc_mode == LFS) {828 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {829 p.min_segno = sbi->next_victim_seg[BG_GC];830 *result = p.min_segno;831 sbi->next_victim_seg[BG_GC] = NULL_SEGNO;832 goto got_result;833 }834 if (gc_type == FG_GC &&835 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {836 p.min_segno = sbi->next_victim_seg[FG_GC];837 *result = p.min_segno;838 sbi->next_victim_seg[FG_GC] = NULL_SEGNO;839 goto got_result;840 }841 }842 843 last_victim = sm->last_victim[p.gc_mode];844 if (p.alloc_mode == LFS && gc_type == FG_GC) {845 p.min_segno = check_bg_victims(sbi);846 if (p.min_segno != NULL_SEGNO)847 goto got_it;848 }849 850 while (1) {851 unsigned long cost, *dirty_bitmap;852 unsigned int unit_no, segno;853 854 dirty_bitmap = p.dirty_bitmap;855 unit_no = find_next_bit(dirty_bitmap,856 last_segment / p.ofs_unit,857 p.offset / p.ofs_unit);858 segno = unit_no * p.ofs_unit;859 if (segno >= last_segment) {860 if (sm->last_victim[p.gc_mode]) {861 last_segment =862 sm->last_victim[p.gc_mode];863 sm->last_victim[p.gc_mode] = 0;864 p.offset = 0;865 continue;866 }867 break;868 }869 870 p.offset = segno + p.ofs_unit;871 nsearched++;872 873#ifdef CONFIG_F2FS_CHECK_FS874 /*875 * skip selecting the invalid segno (that is failed due to block876 * validity check failure during GC) to avoid endless GC loop in877 * such cases.878 */879 if (test_bit(segno, sm->invalid_segmap))880 goto next;881#endif882 883 secno = GET_SEC_FROM_SEG(sbi, segno);884 885 if (sec_usage_check(sbi, secno))886 goto next;887 888 /* Don't touch checkpointed data */889 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {890 if (p.alloc_mode == LFS) {891 /*892 * LFS is set to find source section during GC.893 * The victim should have no checkpointed data.894 */895 if (get_ckpt_valid_blocks(sbi, segno, true))896 goto next;897 } else {898 /*899 * SSR | AT_SSR are set to find target segment900 * for writes which can be full by checkpointed901 * and newly written blocks.902 */903 if (!f2fs_segment_has_free_slot(sbi, segno))904 goto next;905 }906 }907 908 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))909 goto next;910 911 if (gc_type == FG_GC && f2fs_section_is_pinned(dirty_i, secno))912 goto next;913 914 if (is_atgc) {915 add_victim_entry(sbi, &p, segno);916 goto next;917 }918 919 cost = get_gc_cost(sbi, segno, &p);920 921 if (p.min_cost > cost) {922 p.min_segno = segno;923 p.min_cost = cost;924 }925next:926 if (nsearched >= p.max_search) {927 if (!sm->last_victim[p.gc_mode] && segno <= last_victim)928 sm->last_victim[p.gc_mode] =929 last_victim + p.ofs_unit;930 else931 sm->last_victim[p.gc_mode] = segno + p.ofs_unit;932 sm->last_victim[p.gc_mode] %=933 (MAIN_SECS(sbi) * SEGS_PER_SEC(sbi));934 break;935 }936 }937 938 /* get victim for GC_AT/AT_SSR */939 if (is_atgc) {940 lookup_victim_by_age(sbi, &p);941 release_victim_entry(sbi);942 }943 944 if (is_atgc && p.min_segno == NULL_SEGNO &&945 sm->elapsed_time < p.age_threshold) {946 p.age_threshold = 0;947 goto retry;948 }949 950 if (p.min_segno != NULL_SEGNO) {951got_it:952 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;953got_result:954 if (p.alloc_mode == LFS) {955 secno = GET_SEC_FROM_SEG(sbi, p.min_segno);956 if (gc_type == FG_GC)957 sbi->cur_victim_sec = secno;958 else959 set_bit(secno, dirty_i->victim_secmap);960 }961 ret = 0;962 963 }964out:965 if (p.min_segno != NULL_SEGNO)966 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,967 sbi->cur_victim_sec,968 prefree_segments(sbi), free_segments(sbi));969 mutex_unlock(&dirty_i->seglist_lock);970 971 return ret;972}973 974static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)975{976 struct inode_entry *ie;977 978 ie = radix_tree_lookup(&gc_list->iroot, ino);979 if (ie)980 return ie->inode;981 return NULL;982}983 984static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)985{986 struct inode_entry *new_ie;987 988 if (inode == find_gc_inode(gc_list, inode->i_ino)) {989 iput(inode);990 return;991 }992 new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab,993 GFP_NOFS, true, NULL);994 new_ie->inode = inode;995 996 f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);997 list_add_tail(&new_ie->list, &gc_list->ilist);998}999 1000static void put_gc_inode(struct gc_inode_list *gc_list)1001{1002 struct inode_entry *ie, *next_ie;1003 1004 list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {1005 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);1006 iput(ie->inode);1007 list_del(&ie->list);1008 kmem_cache_free(f2fs_inode_entry_slab, ie);1009 }1010}1011 1012static int check_valid_map(struct f2fs_sb_info *sbi,1013 unsigned int segno, int offset)1014{1015 struct sit_info *sit_i = SIT_I(sbi);1016 struct seg_entry *sentry;1017 int ret;1018 1019 down_read(&sit_i->sentry_lock);1020 sentry = get_seg_entry(sbi, segno);1021 ret = f2fs_test_bit(offset, sentry->cur_valid_map);1022 up_read(&sit_i->sentry_lock);1023 return ret;1024}1025 1026/*1027 * This function compares node address got in summary with that in NAT.1028 * On validity, copy that node with cold status, otherwise (invalid node)1029 * ignore that.1030 */1031static int gc_node_segment(struct f2fs_sb_info *sbi,1032 struct f2fs_summary *sum, unsigned int segno, int gc_type)1033{1034 struct f2fs_summary *entry;1035 block_t start_addr;1036 int off;1037 int phase = 0;1038 bool fggc = (gc_type == FG_GC);1039 int submitted = 0;1040 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);1041 1042 start_addr = START_BLOCK(sbi, segno);1043 1044next_step:1045 entry = sum;1046 1047 if (fggc && phase == 2)1048 atomic_inc(&sbi->wb_sync_req[NODE]);1049 1050 for (off = 0; off < usable_blks_in_seg; off++, entry++) {1051 nid_t nid = le32_to_cpu(entry->nid);1052 struct page *node_page;1053 struct node_info ni;1054 int err;1055 1056 /* stop BG_GC if there is not enough free sections. */1057 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))1058 return submitted;1059 1060 if (check_valid_map(sbi, segno, off) == 0)1061 continue;1062 1063 if (phase == 0) {1064 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,1065 META_NAT, true);1066 continue;1067 }1068 1069 if (phase == 1) {1070 f2fs_ra_node_page(sbi, nid);1071 continue;1072 }1073 1074 /* phase == 2 */1075 node_page = f2fs_get_node_page(sbi, nid);1076 if (IS_ERR(node_page))1077 continue;1078 1079 /* block may become invalid during f2fs_get_node_page */1080 if (check_valid_map(sbi, segno, off) == 0) {1081 f2fs_put_page(node_page, 1);1082 continue;1083 }1084 1085 if (f2fs_get_node_info(sbi, nid, &ni, false)) {1086 f2fs_put_page(node_page, 1);1087 continue;1088 }1089 1090 if (ni.blk_addr != start_addr + off) {1091 f2fs_put_page(node_page, 1);1092 continue;1093 }1094 1095 err = f2fs_move_node_page(node_page, gc_type);1096 if (!err && gc_type == FG_GC)1097 submitted++;1098 stat_inc_node_blk_count(sbi, 1, gc_type);1099 }1100 1101 if (++phase < 3)1102 goto next_step;1103 1104 if (fggc)1105 atomic_dec(&sbi->wb_sync_req[NODE]);1106 return submitted;1107}1108 1109/*1110 * Calculate start block index indicating the given node offset.1111 * Be careful, caller should give this node offset only indicating direct node1112 * blocks. If any node offsets, which point the other types of node blocks such1113 * as indirect or double indirect node blocks, are given, it must be a caller's1114 * bug.1115 */1116block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)1117{1118 unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;1119 unsigned int bidx;1120 1121 if (node_ofs == 0)1122 return 0;1123 1124 if (node_ofs <= 2) {1125 bidx = node_ofs - 1;1126 } else if (node_ofs <= indirect_blks) {1127 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);1128 1129 bidx = node_ofs - 2 - dec;1130 } else {1131 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);1132 1133 bidx = node_ofs - 5 - dec;1134 }1135 return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);1136}1137 1138static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,1139 struct node_info *dni, block_t blkaddr, unsigned int *nofs)1140{1141 struct page *node_page;1142 nid_t nid;1143 unsigned int ofs_in_node, max_addrs, base;1144 block_t source_blkaddr;1145 1146 nid = le32_to_cpu(sum->nid);1147 ofs_in_node = le16_to_cpu(sum->ofs_in_node);1148 1149 node_page = f2fs_get_node_page(sbi, nid);1150 if (IS_ERR(node_page))1151 return false;1152 1153 if (f2fs_get_node_info(sbi, nid, dni, false)) {1154 f2fs_put_page(node_page, 1);1155 return false;1156 }1157 1158 if (sum->version != dni->version) {1159 f2fs_warn(sbi, "%s: valid data with mismatched node version.",1160 __func__);1161 set_sbi_flag(sbi, SBI_NEED_FSCK);1162 }1163 1164 if (f2fs_check_nid_range(sbi, dni->ino)) {1165 f2fs_put_page(node_page, 1);1166 return false;1167 }1168 1169 if (IS_INODE(node_page)) {1170 base = offset_in_addr(F2FS_INODE(node_page));1171 max_addrs = DEF_ADDRS_PER_INODE;1172 } else {1173 base = 0;1174 max_addrs = DEF_ADDRS_PER_BLOCK;1175 }1176 1177 if (base + ofs_in_node >= max_addrs) {1178 f2fs_err(sbi, "Inconsistent blkaddr offset: base:%u, ofs_in_node:%u, max:%u, ino:%u, nid:%u",1179 base, ofs_in_node, max_addrs, dni->ino, dni->nid);1180 f2fs_put_page(node_page, 1);1181 return false;1182 }1183 1184 *nofs = ofs_of_node(node_page);1185 source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);1186 f2fs_put_page(node_page, 1);1187 1188 if (source_blkaddr != blkaddr) {1189#ifdef CONFIG_F2FS_CHECK_FS1190 unsigned int segno = GET_SEGNO(sbi, blkaddr);1191 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);1192 1193 if (unlikely(check_valid_map(sbi, segno, offset))) {1194 if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {1195 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",1196 blkaddr, source_blkaddr, segno);1197 set_sbi_flag(sbi, SBI_NEED_FSCK);1198 }1199 }1200#endif1201 return false;1202 }1203 return true;1204}1205 1206static int ra_data_block(struct inode *inode, pgoff_t index)1207{1208 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);1209 struct address_space *mapping = f2fs_is_cow_file(inode) ?1210 F2FS_I(inode)->atomic_inode->i_mapping : inode->i_mapping;1211 struct dnode_of_data dn;1212 struct page *page;1213 struct f2fs_io_info fio = {1214 .sbi = sbi,1215 .ino = inode->i_ino,1216 .type = DATA,1217 .temp = COLD,1218 .op = REQ_OP_READ,1219 .op_flags = 0,1220 .encrypted_page = NULL,1221 .in_list = 0,1222 };1223 int err;1224 1225 page = f2fs_grab_cache_page(mapping, index, true);1226 if (!page)1227 return -ENOMEM;1228 1229 if (f2fs_lookup_read_extent_cache_block(inode, index,1230 &dn.data_blkaddr)) {1231 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,1232 DATA_GENERIC_ENHANCE_READ))) {1233 err = -EFSCORRUPTED;1234 goto put_page;1235 }1236 goto got_it;1237 }1238 1239 set_new_dnode(&dn, inode, NULL, NULL, 0);1240 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);1241 if (err)1242 goto put_page;1243 f2fs_put_dnode(&dn);1244 1245 if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {1246 err = -ENOENT;1247 goto put_page;1248 }1249 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,1250 DATA_GENERIC_ENHANCE))) {1251 err = -EFSCORRUPTED;1252 goto put_page;1253 }1254got_it:1255 /* read page */1256 fio.page = page;1257 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;1258 1259 /*1260 * don't cache encrypted data into meta inode until previous dirty1261 * data were writebacked to avoid racing between GC and flush.1262 */1263 f2fs_wait_on_page_writeback(page, DATA, true, true);1264 1265 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);1266 1267 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),1268 dn.data_blkaddr,1269 FGP_LOCK | FGP_CREAT, GFP_NOFS);1270 if (!fio.encrypted_page) {1271 err = -ENOMEM;1272 goto put_page;1273 }1274 1275 err = f2fs_submit_page_bio(&fio);1276 if (err)1277 goto put_encrypted_page;1278 f2fs_put_page(fio.encrypted_page, 0);1279 f2fs_put_page(page, 1);1280 1281 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);1282 f2fs_update_iostat(sbi, NULL, FS_GDATA_READ_IO, F2FS_BLKSIZE);1283 1284 return 0;1285put_encrypted_page:1286 f2fs_put_page(fio.encrypted_page, 1);1287put_page:1288 f2fs_put_page(page, 1);1289 return err;1290}1291 1292/*1293 * Move data block via META_MAPPING while keeping locked data page.1294 * This can be used to move blocks, aka LBAs, directly on disk.1295 */1296static int move_data_block(struct inode *inode, block_t bidx,1297 int gc_type, unsigned int segno, int off)1298{1299 struct address_space *mapping = f2fs_is_cow_file(inode) ?1300 F2FS_I(inode)->atomic_inode->i_mapping : inode->i_mapping;1301 struct f2fs_io_info fio = {1302 .sbi = F2FS_I_SB(inode),1303 .ino = inode->i_ino,1304 .type = DATA,1305 .temp = COLD,1306 .op = REQ_OP_READ,1307 .op_flags = 0,1308 .encrypted_page = NULL,1309 .in_list = 0,1310 };1311 struct dnode_of_data dn;1312 struct f2fs_summary sum;1313 struct node_info ni;1314 struct page *page, *mpage;1315 block_t newaddr;1316 int err = 0;1317 bool lfs_mode = f2fs_lfs_mode(fio.sbi);1318 int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&1319 (fio.sbi->gc_mode != GC_URGENT_HIGH) ?1320 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;1321 1322 /* do not read out */1323 page = f2fs_grab_cache_page(mapping, bidx, false);1324 if (!page)1325 return -ENOMEM;1326 1327 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {1328 err = -ENOENT;1329 goto out;1330 }1331 1332 err = f2fs_gc_pinned_control(inode, gc_type, segno);1333 if (err)1334 goto out;1335 1336 set_new_dnode(&dn, inode, NULL, NULL, 0);1337 err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);1338 if (err)1339 goto out;1340 1341 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {1342 ClearPageUptodate(page);1343 err = -ENOENT;1344 goto put_out;1345 }1346 1347 /*1348 * don't cache encrypted data into meta inode until previous dirty1349 * data were writebacked to avoid racing between GC and flush.1350 */1351 f2fs_wait_on_page_writeback(page, DATA, true, true);1352 1353 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);1354 1355 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);1356 if (err)1357 goto put_out;1358 1359 /* read page */1360 fio.page = page;1361 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;1362 1363 if (lfs_mode)1364 f2fs_down_write(&fio.sbi->io_order_lock);1365 1366 mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),1367 fio.old_blkaddr, false);1368 if (!mpage) {1369 err = -ENOMEM;1370 goto up_out;1371 }1372 1373 fio.encrypted_page = mpage;1374 1375 /* read source block in mpage */1376 if (!PageUptodate(mpage)) {1377 err = f2fs_submit_page_bio(&fio);1378 if (err) {1379 f2fs_put_page(mpage, 1);1380 goto up_out;1381 }1382 1383 f2fs_update_iostat(fio.sbi, inode, FS_DATA_READ_IO,1384 F2FS_BLKSIZE);1385 f2fs_update_iostat(fio.sbi, NULL, FS_GDATA_READ_IO,1386 F2FS_BLKSIZE);1387 1388 lock_page(mpage);1389 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||1390 !PageUptodate(mpage))) {1391 err = -EIO;1392 f2fs_put_page(mpage, 1);1393 goto up_out;1394 }1395 }1396 1397 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);1398 1399 /* allocate block address */1400 err = f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,1401 &sum, type, NULL);1402 if (err) {1403 f2fs_put_page(mpage, 1);1404 /* filesystem should shutdown, no need to recovery block */1405 goto up_out;1406 }1407 1408 fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),1409 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);1410 if (!fio.encrypted_page) {1411 err = -ENOMEM;1412 f2fs_put_page(mpage, 1);1413 goto recover_block;1414 }1415 1416 /* write target block */1417 f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);1418 memcpy(page_address(fio.encrypted_page),1419 page_address(mpage), PAGE_SIZE);1420 f2fs_put_page(mpage, 1);1421 1422 f2fs_invalidate_internal_cache(fio.sbi, fio.old_blkaddr);1423 1424 set_page_dirty(fio.encrypted_page);1425 if (clear_page_dirty_for_io(fio.encrypted_page))1426 dec_page_count(fio.sbi, F2FS_DIRTY_META);1427 1428 set_page_writeback(fio.encrypted_page);1429 1430 fio.op = REQ_OP_WRITE;1431 fio.op_flags = REQ_SYNC;1432 fio.new_blkaddr = newaddr;1433 f2fs_submit_page_write(&fio);1434 1435 f2fs_update_iostat(fio.sbi, NULL, FS_GC_DATA_IO, F2FS_BLKSIZE);1436 1437 f2fs_update_data_blkaddr(&dn, newaddr);1438 set_inode_flag(inode, FI_APPEND_WRITE);1439 1440 f2fs_put_page(fio.encrypted_page, 1);1441recover_block:1442 if (err)1443 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,1444 true, true, true);1445up_out:1446 if (lfs_mode)1447 f2fs_up_write(&fio.sbi->io_order_lock);1448put_out:1449 f2fs_put_dnode(&dn);1450out:1451 f2fs_put_page(page, 1);1452 return err;1453}1454 1455static int move_data_page(struct inode *inode, block_t bidx, int gc_type,1456 unsigned int segno, int off)1457{1458 struct page *page;1459 int err = 0;1460 1461 page = f2fs_get_lock_data_page(inode, bidx, true);1462 if (IS_ERR(page))1463 return PTR_ERR(page);1464 1465 if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {1466 err = -ENOENT;1467 goto out;1468 }1469 1470 err = f2fs_gc_pinned_control(inode, gc_type, segno);1471 if (err)1472 goto out;1473 1474 if (gc_type == BG_GC) {1475 if (folio_test_writeback(page_folio(page))) {1476 err = -EAGAIN;1477 goto out;1478 }1479 set_page_dirty(page);1480 set_page_private_gcing(page);1481 } else {1482 struct f2fs_io_info fio = {1483 .sbi = F2FS_I_SB(inode),1484 .ino = inode->i_ino,1485 .type = DATA,1486 .temp = COLD,1487 .op = REQ_OP_WRITE,1488 .op_flags = REQ_SYNC,1489 .old_blkaddr = NULL_ADDR,1490 .page = page,1491 .encrypted_page = NULL,1492 .need_lock = LOCK_REQ,1493 .io_type = FS_GC_DATA_IO,1494 };1495 bool is_dirty = PageDirty(page);1496 1497retry:1498 f2fs_wait_on_page_writeback(page, DATA, true, true);1499 1500 set_page_dirty(page);1501 if (clear_page_dirty_for_io(page)) {1502 inode_dec_dirty_pages(inode);1503 f2fs_remove_dirty_inode(inode);1504 }1505 1506 set_page_private_gcing(page);1507 1508 err = f2fs_do_write_data_page(&fio);1509 if (err) {1510 clear_page_private_gcing(page);1511 if (err == -ENOMEM) {1512 memalloc_retry_wait(GFP_NOFS);1513 goto retry;1514 }1515 if (is_dirty)1516 set_page_dirty(page);1517 }1518 }1519out:1520 f2fs_put_page(page, 1);1521 return err;1522}1523 1524/*1525 * This function tries to get parent node of victim data block, and identifies1526 * data block validity. If the block is valid, copy that with cold status and1527 * modify parent node.1528 * If the parent node is not valid or the data block address is different,1529 * the victim data block is ignored.1530 */1531static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,1532 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,1533 bool force_migrate)1534{1535 struct super_block *sb = sbi->sb;1536 struct f2fs_summary *entry;1537 block_t start_addr;1538 int off;1539 int phase = 0;1540 int submitted = 0;1541 unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);1542 1543 start_addr = START_BLOCK(sbi, segno);1544 1545next_step:1546 entry = sum;1547 1548 for (off = 0; off < usable_blks_in_seg; off++, entry++) {1549 struct page *data_page;1550 struct inode *inode;1551 struct node_info dni; /* dnode info for the data */1552 unsigned int ofs_in_node, nofs;1553 block_t start_bidx;1554 nid_t nid = le32_to_cpu(entry->nid);1555 1556 /*1557 * stop BG_GC if there is not enough free sections.1558 * Or, stop GC if the segment becomes fully valid caused by1559 * race condition along with SSR block allocation.1560 */1561 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||1562 (!force_migrate && get_valid_blocks(sbi, segno, true) ==1563 CAP_BLKS_PER_SEC(sbi)))1564 return submitted;1565 1566 if (check_valid_map(sbi, segno, off) == 0)1567 continue;1568 1569 if (phase == 0) {1570 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,1571 META_NAT, true);1572 continue;1573 }1574 1575 if (phase == 1) {1576 f2fs_ra_node_page(sbi, nid);1577 continue;1578 }1579 1580 /* Get an inode by ino with checking validity */1581 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))1582 continue;1583 1584 if (phase == 2) {1585 f2fs_ra_node_page(sbi, dni.ino);1586 continue;1587 }1588 1589 ofs_in_node = le16_to_cpu(entry->ofs_in_node);1590 1591 if (phase == 3) {1592 int err;1593 1594 inode = f2fs_iget(sb, dni.ino);1595 if (IS_ERR(inode))1596 continue;1597 1598 if (is_bad_inode(inode) ||1599 special_file(inode->i_mode)) {1600 iput(inode);1601 continue;1602 }1603 1604 if (f2fs_has_inline_data(inode)) {1605 iput(inode);1606 set_sbi_flag(sbi, SBI_NEED_FSCK);1607 f2fs_err_ratelimited(sbi,1608 "inode %lx has both inline_data flag and "1609 "data block, nid=%u, ofs_in_node=%u",1610 inode->i_ino, dni.nid, ofs_in_node);1611 continue;1612 }1613 1614 err = f2fs_gc_pinned_control(inode, gc_type, segno);1615 if (err == -EAGAIN) {1616 iput(inode);1617 return submitted;1618 }1619 1620 if (!f2fs_down_write_trylock(1621 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {1622 iput(inode);1623 sbi->skipped_gc_rwsem++;1624 continue;1625 }1626 1627 start_bidx = f2fs_start_bidx_of_node(nofs, inode) +1628 ofs_in_node;1629 1630 if (f2fs_meta_inode_gc_required(inode)) {1631 int err = ra_data_block(inode, start_bidx);1632 1633 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);1634 if (err) {1635 iput(inode);1636 continue;1637 }1638 add_gc_inode(gc_list, inode);1639 continue;1640 }1641 1642 data_page = f2fs_get_read_data_page(inode, start_bidx,1643 REQ_RAHEAD, true, NULL);1644 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);1645 if (IS_ERR(data_page)) {1646 iput(inode);1647 continue;1648 }1649 1650 f2fs_put_page(data_page, 0);1651 add_gc_inode(gc_list, inode);1652 continue;1653 }1654 1655 /* phase 4 */1656 inode = find_gc_inode(gc_list, dni.ino);1657 if (inode) {1658 struct f2fs_inode_info *fi = F2FS_I(inode);1659 bool locked = false;1660 int err;1661 1662 if (S_ISREG(inode->i_mode)) {1663 if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[WRITE])) {1664 sbi->skipped_gc_rwsem++;1665 continue;1666 }1667 if (!f2fs_down_write_trylock(1668 &fi->i_gc_rwsem[READ])) {1669 sbi->skipped_gc_rwsem++;1670 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);1671 continue;1672 }1673 locked = true;1674 1675 /* wait for all inflight aio data */1676 inode_dio_wait(inode);1677 }1678 1679 start_bidx = f2fs_start_bidx_of_node(nofs, inode)1680 + ofs_in_node;1681 if (f2fs_meta_inode_gc_required(inode))1682 err = move_data_block(inode, start_bidx,1683 gc_type, segno, off);1684 else1685 err = move_data_page(inode, start_bidx, gc_type,1686 segno, off);1687 1688 if (!err && (gc_type == FG_GC ||1689 f2fs_meta_inode_gc_required(inode)))1690 submitted++;1691 1692 if (locked) {1693 f2fs_up_write(&fi->i_gc_rwsem[READ]);1694 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);1695 }1696 1697 stat_inc_data_blk_count(sbi, 1, gc_type);1698 }1699 }1700 1701 if (++phase < 5)1702 goto next_step;1703 1704 return submitted;1705}1706 1707static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,1708 int gc_type, bool one_time)1709{1710 struct sit_info *sit_i = SIT_I(sbi);1711 int ret;1712 1713 down_write(&sit_i->sentry_lock);1714 ret = f2fs_get_victim(sbi, victim, gc_type, NO_CHECK_TYPE,1715 LFS, 0, one_time);1716 up_write(&sit_i->sentry_lock);1717 return ret;1718}1719 1720static int do_garbage_collect(struct f2fs_sb_info *sbi,1721 unsigned int start_segno,1722 struct gc_inode_list *gc_list, int gc_type,1723 bool force_migrate, bool one_time)1724{1725 struct page *sum_page;1726 struct f2fs_summary_block *sum;1727 struct blk_plug plug;1728 unsigned int segno = start_segno;1729 unsigned int end_segno = start_segno + SEGS_PER_SEC(sbi);1730 unsigned int sec_end_segno;1731 int seg_freed = 0, migrated = 0;1732 unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?1733 SUM_TYPE_DATA : SUM_TYPE_NODE;1734 unsigned char data_type = (type == SUM_TYPE_DATA) ? DATA : NODE;1735 int submitted = 0;1736 1737 if (__is_large_section(sbi)) {1738 sec_end_segno = rounddown(end_segno, SEGS_PER_SEC(sbi));1739 1740 /*1741 * zone-capacity can be less than zone-size in zoned devices,1742 * resulting in less than expected usable segments in the zone,1743 * calculate the end segno in the zone which can be garbage1744 * collected1745 */1746 if (f2fs_sb_has_blkzoned(sbi))1747 sec_end_segno -= SEGS_PER_SEC(sbi) -1748 f2fs_usable_segs_in_sec(sbi);1749 1750 if (gc_type == BG_GC || one_time) {1751 unsigned int window_granularity =1752 sbi->migration_window_granularity;1753 1754 if (f2fs_sb_has_blkzoned(sbi) &&1755 !has_enough_free_blocks(sbi,1756 sbi->gc_thread->boost_zoned_gc_percent))1757 window_granularity *=1758 BOOST_GC_MULTIPLE;1759 1760 end_segno = start_segno + window_granularity;1761 }1762 1763 if (end_segno > sec_end_segno)1764 end_segno = sec_end_segno;1765 }1766 1767 sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);1768 1769 /* readahead multi ssa blocks those have contiguous address */1770 if (__is_large_section(sbi))1771 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),1772 end_segno - segno, META_SSA, true);1773 1774 /* reference all summary page */1775 while (segno < end_segno) {1776 sum_page = f2fs_get_sum_page(sbi, segno++);1777 if (IS_ERR(sum_page)) {1778 int err = PTR_ERR(sum_page);1779 1780 end_segno = segno - 1;1781 for (segno = start_segno; segno < end_segno; segno++) {1782 sum_page = find_get_page(META_MAPPING(sbi),1783 GET_SUM_BLOCK(sbi, segno));1784 f2fs_put_page(sum_page, 0);1785 f2fs_put_page(sum_page, 0);1786 }1787 return err;1788 }1789 unlock_page(sum_page);1790 }1791 1792 blk_start_plug(&plug);1793 1794 for (segno = start_segno; segno < end_segno; segno++) {1795 1796 /* find segment summary of victim */1797 sum_page = find_get_page(META_MAPPING(sbi),1798 GET_SUM_BLOCK(sbi, segno));1799 f2fs_put_page(sum_page, 0);1800 1801 if (get_valid_blocks(sbi, segno, false) == 0)1802 goto freed;1803 if (gc_type == BG_GC && __is_large_section(sbi) &&1804 migrated >= sbi->migration_granularity)1805 goto skip;1806 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))1807 goto skip;1808 1809 sum = page_address(sum_page);1810 if (type != GET_SUM_TYPE((&sum->footer))) {1811 f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",1812 segno, type, GET_SUM_TYPE((&sum->footer)));1813 f2fs_stop_checkpoint(sbi, false,1814 STOP_CP_REASON_CORRUPTED_SUMMARY);1815 goto skip;1816 }1817 1818 /*1819 * this is to avoid deadlock:1820 * - lock_page(sum_page) - f2fs_replace_block1821 * - check_valid_map() - down_write(sentry_lock)1822 * - down_read(sentry_lock) - change_curseg()1823 * - lock_page(sum_page)1824 */1825 if (type == SUM_TYPE_NODE)1826 submitted += gc_node_segment(sbi, sum->entries, segno,1827 gc_type);1828 else1829 submitted += gc_data_segment(sbi, sum->entries, gc_list,1830 segno, gc_type,1831 force_migrate);1832 1833 stat_inc_gc_seg_count(sbi, data_type, gc_type);1834 sbi->gc_reclaimed_segs[sbi->gc_mode]++;1835 migrated++;1836 1837freed:1838 if (gc_type == FG_GC &&1839 get_valid_blocks(sbi, segno, false) == 0)1840 seg_freed++;1841 1842 if (__is_large_section(sbi))1843 sbi->next_victim_seg[gc_type] =1844 (segno + 1 < sec_end_segno) ?1845 segno + 1 : NULL_SEGNO;1846skip:1847 f2fs_put_page(sum_page, 0);1848 }1849 1850 if (submitted)1851 f2fs_submit_merged_write(sbi, data_type);1852 1853 blk_finish_plug(&plug);1854 1855 if (migrated)1856 stat_inc_gc_sec_count(sbi, data_type, gc_type);1857 1858 return seg_freed;1859}1860 1861int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control)1862{1863 int gc_type = gc_control->init_gc_type;1864 unsigned int segno = gc_control->victim_segno;1865 int sec_freed = 0, seg_freed = 0, total_freed = 0, total_sec_freed = 0;1866 int ret = 0;1867 struct cp_control cpc;1868 struct gc_inode_list gc_list = {1869 .ilist = LIST_HEAD_INIT(gc_list.ilist),1870 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),1871 };1872 unsigned int skipped_round = 0, round = 0;1873 unsigned int upper_secs;1874 1875 trace_f2fs_gc_begin(sbi->sb, gc_type, gc_control->no_bg_gc,1876 gc_control->nr_free_secs,1877 get_pages(sbi, F2FS_DIRTY_NODES),1878 get_pages(sbi, F2FS_DIRTY_DENTS),1879 get_pages(sbi, F2FS_DIRTY_IMETA),1880 free_sections(sbi),1881 free_segments(sbi),1882 reserved_segments(sbi),1883 prefree_segments(sbi));1884 1885 cpc.reason = __get_cp_reason(sbi);1886gc_more:1887 sbi->skipped_gc_rwsem = 0;1888 if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {1889 ret = -EINVAL;1890 goto stop;1891 }1892 if (unlikely(f2fs_cp_error(sbi))) {1893 ret = -EIO;1894 goto stop;1895 }1896 1897 /* Let's run FG_GC, if we don't have enough space. */1898 if (has_not_enough_free_secs(sbi, 0, 0)) {1899 gc_type = FG_GC;1900 1901 /*1902 * For example, if there are many prefree_segments below given1903 * threshold, we can make them free by checkpoint. Then, we1904 * secure free segments which doesn't need fggc any more.1905 */1906 if (prefree_segments(sbi)) {1907 stat_inc_cp_call_count(sbi, TOTAL_CALL);1908 ret = f2fs_write_checkpoint(sbi, &cpc);1909 if (ret)1910 goto stop;1911 /* Reset due to checkpoint */1912 sec_freed = 0;1913 }1914 }1915 1916 /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */1917 if (gc_type == BG_GC && gc_control->no_bg_gc) {1918 ret = -EINVAL;1919 goto stop;1920 }1921retry:1922 ret = __get_victim(sbi, &segno, gc_type, gc_control->one_time);1923 if (ret) {1924 /* allow to search victim from sections has pinned data */1925 if (ret == -ENODATA && gc_type == FG_GC &&1926 f2fs_pinned_section_exists(DIRTY_I(sbi))) {1927 f2fs_unpin_all_sections(sbi, false);1928 goto retry;1929 }1930 goto stop;1931 }1932 1933 seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type,1934 gc_control->should_migrate_blocks,1935 gc_control->one_time);1936 if (seg_freed < 0)1937 goto stop;1938 1939 total_freed += seg_freed;1940 1941 if (seg_freed == f2fs_usable_segs_in_sec(sbi)) {1942 sec_freed++;1943 total_sec_freed++;1944 }1945 1946 if (gc_control->one_time)1947 goto stop;1948 1949 if (gc_type == FG_GC) {1950 sbi->cur_victim_sec = NULL_SEGNO;1951 1952 if (has_enough_free_secs(sbi, sec_freed, 0)) {1953 if (!gc_control->no_bg_gc &&1954 total_sec_freed < gc_control->nr_free_secs)1955 goto go_gc_more;1956 goto stop;1957 }1958 if (sbi->skipped_gc_rwsem)1959 skipped_round++;1960 round++;1961 if (skipped_round > MAX_SKIP_GC_COUNT &&1962 skipped_round * 2 >= round) {1963 stat_inc_cp_call_count(sbi, TOTAL_CALL);1964 ret = f2fs_write_checkpoint(sbi, &cpc);1965 goto stop;1966 }1967 } else if (has_enough_free_secs(sbi, 0, 0)) {1968 goto stop;1969 }1970 1971 __get_secs_required(sbi, NULL, &upper_secs, NULL);1972 1973 /*1974 * Write checkpoint to reclaim prefree segments.1975 * We need more three extra sections for writer's data/node/dentry.1976 */1977 if (free_sections(sbi) <= upper_secs + NR_GC_CHECKPOINT_SECS &&1978 prefree_segments(sbi)) {1979 stat_inc_cp_call_count(sbi, TOTAL_CALL);1980 ret = f2fs_write_checkpoint(sbi, &cpc);1981 if (ret)1982 goto stop;1983 /* Reset due to checkpoint */1984 sec_freed = 0;1985 }1986go_gc_more:1987 segno = NULL_SEGNO;1988 goto gc_more;1989 1990stop:1991 SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;1992 SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno;1993 1994 if (gc_type == FG_GC)1995 f2fs_unpin_all_sections(sbi, true);1996 1997 trace_f2fs_gc_end(sbi->sb, ret, total_freed, total_sec_freed,1998 get_pages(sbi, F2FS_DIRTY_NODES),1999 get_pages(sbi, F2FS_DIRTY_DENTS),2000 get_pages(sbi, F2FS_DIRTY_IMETA),2001 free_sections(sbi),2002 free_segments(sbi),2003 reserved_segments(sbi),2004 prefree_segments(sbi));2005 2006 f2fs_up_write(&sbi->gc_lock);2007 2008 put_gc_inode(&gc_list);2009 2010 if (gc_control->err_gc_skipped && !ret)2011 ret = total_sec_freed ? 0 : -EAGAIN;2012 return ret;2013}2014 2015int __init f2fs_create_garbage_collection_cache(void)2016{2017 victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",2018 sizeof(struct victim_entry));2019 return victim_entry_slab ? 0 : -ENOMEM;2020}2021 2022void f2fs_destroy_garbage_collection_cache(void)2023{2024 kmem_cache_destroy(victim_entry_slab);2025}2026 2027static void init_atgc_management(struct f2fs_sb_info *sbi)2028{2029 struct atgc_management *am = &sbi->am;2030 2031 if (test_opt(sbi, ATGC) &&2032 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)2033 am->atgc_enabled = true;2034 2035 am->root = RB_ROOT_CACHED;2036 INIT_LIST_HEAD(&am->victim_list);2037 am->victim_count = 0;2038 2039 am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;2040 am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;2041 am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;2042 am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;2043}2044 2045void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)2046{2047 sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;2048 2049 /* give warm/cold data area from slower device */2050 if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))2051 SIT_I(sbi)->last_victim[ALLOC_NEXT] =2052 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;2053 2054 init_atgc_management(sbi);2055}2056 2057int f2fs_gc_range(struct f2fs_sb_info *sbi,2058 unsigned int start_seg, unsigned int end_seg,2059 bool dry_run, unsigned int dry_run_sections)2060{2061 unsigned int segno;2062 unsigned int gc_secs = dry_run_sections;2063 2064 if (unlikely(f2fs_cp_error(sbi)))2065 return -EIO;2066 2067 for (segno = start_seg; segno <= end_seg; segno += SEGS_PER_SEC(sbi)) {2068 struct gc_inode_list gc_list = {2069 .ilist = LIST_HEAD_INIT(gc_list.ilist),2070 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),2071 };2072 2073 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true, false);2074 put_gc_inode(&gc_list);2075 2076 if (!dry_run && get_valid_blocks(sbi, segno, true))2077 return -EAGAIN;2078 if (dry_run && dry_run_sections &&2079 !get_valid_blocks(sbi, segno, true) && --gc_secs == 0)2080 break;2081 2082 if (fatal_signal_pending(current))2083 return -ERESTARTSYS;2084 }2085 2086 return 0;2087}2088 2089static int free_segment_range(struct f2fs_sb_info *sbi,2090 unsigned int secs, bool dry_run)2091{2092 unsigned int next_inuse, start, end;2093 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };2094 int gc_mode, gc_type;2095 int err = 0;2096 int type;2097 2098 /* Force block allocation for GC */2099 MAIN_SECS(sbi) -= secs;2100 start = MAIN_SECS(sbi) * SEGS_PER_SEC(sbi);2101 end = MAIN_SEGS(sbi) - 1;2102 2103 mutex_lock(&DIRTY_I(sbi)->seglist_lock);2104 for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)2105 if (SIT_I(sbi)->last_victim[gc_mode] >= start)2106 SIT_I(sbi)->last_victim[gc_mode] = 0;2107 2108 for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)2109 if (sbi->next_victim_seg[gc_type] >= start)2110 sbi->next_victim_seg[gc_type] = NULL_SEGNO;2111 mutex_unlock(&DIRTY_I(sbi)->seglist_lock);2112 2113 /* Move out cursegs from the target range */2114 for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++) {2115 err = f2fs_allocate_segment_for_resize(sbi, type, start, end);2116 if (err)2117 goto out;2118 }2119 2120 /* do GC to move out valid blocks in the range */2121 err = f2fs_gc_range(sbi, start, end, dry_run, 0);2122 if (err || dry_run)2123 goto out;2124 2125 stat_inc_cp_call_count(sbi, TOTAL_CALL);2126 err = f2fs_write_checkpoint(sbi, &cpc);2127 if (err)2128 goto out;2129 2130 next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);2131 if (next_inuse <= end) {2132 f2fs_err(sbi, "segno %u should be free but still inuse!",2133 next_inuse);2134 f2fs_bug_on(sbi, 1);2135 }2136out:2137 MAIN_SECS(sbi) += secs;2138 return err;2139}2140 2141static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)2142{2143 struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);2144 int section_count;2145 int segment_count;2146 int segment_count_main;2147 long long block_count;2148 int segs = secs * SEGS_PER_SEC(sbi);2149 2150 f2fs_down_write(&sbi->sb_lock);2151 2152 section_count = le32_to_cpu(raw_sb->section_count);2153 segment_count = le32_to_cpu(raw_sb->segment_count);2154 segment_count_main = le32_to_cpu(raw_sb->segment_count_main);2155 block_count = le64_to_cpu(raw_sb->block_count);2156 2157 raw_sb->section_count = cpu_to_le32(section_count + secs);2158 raw_sb->segment_count = cpu_to_le32(segment_count + segs);2159 raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);2160 raw_sb->block_count = cpu_to_le64(block_count +2161 (long long)SEGS_TO_BLKS(sbi, segs));2162 if (f2fs_is_multi_device(sbi)) {2163 int last_dev = sbi->s_ndevs - 1;2164 int dev_segs =2165 le32_to_cpu(raw_sb->devs[last_dev].total_segments);2166 2167 raw_sb->devs[last_dev].total_segments =2168 cpu_to_le32(dev_segs + segs);2169 }2170 2171 f2fs_up_write(&sbi->sb_lock);2172}2173 2174static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)2175{2176 int segs = secs * SEGS_PER_SEC(sbi);2177 long long blks = SEGS_TO_BLKS(sbi, segs);2178 long long user_block_count =2179 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);2180 2181 SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;2182 MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;2183 MAIN_SECS(sbi) += secs;2184 FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;2185 FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;2186 F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);2187 2188 if (f2fs_is_multi_device(sbi)) {2189 int last_dev = sbi->s_ndevs - 1;2190 2191 FDEV(last_dev).total_segments =2192 (int)FDEV(last_dev).total_segments + segs;2193 FDEV(last_dev).end_blk =2194 (long long)FDEV(last_dev).end_blk + blks;2195#ifdef CONFIG_BLK_DEV_ZONED2196 FDEV(last_dev).nr_blkz = FDEV(last_dev).nr_blkz +2197 div_u64(blks, sbi->blocks_per_blkz);2198#endif2199 }2200}2201 2202int f2fs_resize_fs(struct file *filp, __u64 block_count)2203{2204 struct f2fs_sb_info *sbi = F2FS_I_SB(file_inode(filp));2205 __u64 old_block_count, shrunk_blocks;2206 struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };2207 unsigned int secs;2208 int err = 0;2209 __u32 rem;2210 2211 old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);2212 if (block_count > old_block_count)2213 return -EINVAL;2214 2215 if (f2fs_is_multi_device(sbi)) {2216 int last_dev = sbi->s_ndevs - 1;2217 __u64 last_segs = FDEV(last_dev).total_segments;2218 2219 if (block_count + SEGS_TO_BLKS(sbi, last_segs) <=2220 old_block_count)2221 return -EINVAL;2222 }2223 2224 /* new fs size should align to section size */2225 div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);2226 if (rem)2227 return -EINVAL;2228 2229 if (block_count == old_block_count)2230 return 0;2231 2232 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {2233 f2fs_err(sbi, "Should run fsck to repair first.");2234 return -EFSCORRUPTED;2235 }2236 2237 if (test_opt(sbi, DISABLE_CHECKPOINT)) {2238 f2fs_err(sbi, "Checkpoint should be enabled.");2239 return -EINVAL;2240 }2241 2242 err = mnt_want_write_file(filp);2243 if (err)2244 return err;2245 2246 shrunk_blocks = old_block_count - block_count;2247 secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));2248 2249 /* stop other GC */2250 if (!f2fs_down_write_trylock(&sbi->gc_lock)) {2251 err = -EAGAIN;2252 goto out_drop_write;2253 }2254 2255 /* stop CP to protect MAIN_SEC in free_segment_range */2256 f2fs_lock_op(sbi);2257 2258 spin_lock(&sbi->stat_lock);2259 if (shrunk_blocks + valid_user_blocks(sbi) +2260 sbi->current_reserved_blocks + sbi->unusable_block_count +2261 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)2262 err = -ENOSPC;2263 spin_unlock(&sbi->stat_lock);2264 2265 if (err)2266 goto out_unlock;2267 2268 err = free_segment_range(sbi, secs, true);2269 2270out_unlock:2271 f2fs_unlock_op(sbi);2272 f2fs_up_write(&sbi->gc_lock);2273out_drop_write:2274 mnt_drop_write_file(filp);2275 if (err)2276 return err;2277 2278 err = freeze_super(sbi->sb, FREEZE_HOLDER_USERSPACE);2279 if (err)2280 return err;2281 2282 if (f2fs_readonly(sbi->sb)) {2283 err = thaw_super(sbi->sb, FREEZE_HOLDER_USERSPACE);2284 if (err)2285 return err;2286 return -EROFS;2287 }2288 2289 f2fs_down_write(&sbi->gc_lock);2290 f2fs_down_write(&sbi->cp_global_sem);2291 2292 spin_lock(&sbi->stat_lock);2293 if (shrunk_blocks + valid_user_blocks(sbi) +2294 sbi->current_reserved_blocks + sbi->unusable_block_count +2295 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)2296 err = -ENOSPC;2297 else2298 sbi->user_block_count -= shrunk_blocks;2299 spin_unlock(&sbi->stat_lock);2300 if (err)2301 goto out_err;2302 2303 set_sbi_flag(sbi, SBI_IS_RESIZEFS);2304 err = free_segment_range(sbi, secs, false);2305 if (err)2306 goto recover_out;2307 2308 update_sb_metadata(sbi, -secs);2309 2310 err = f2fs_commit_super(sbi, false);2311 if (err) {2312 update_sb_metadata(sbi, secs);2313 goto recover_out;2314 }2315 2316 update_fs_metadata(sbi, -secs);2317 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);2318 set_sbi_flag(sbi, SBI_IS_DIRTY);2319 2320 stat_inc_cp_call_count(sbi, TOTAL_CALL);2321 err = f2fs_write_checkpoint(sbi, &cpc);2322 if (err) {2323 update_fs_metadata(sbi, secs);2324 update_sb_metadata(sbi, secs);2325 f2fs_commit_super(sbi, false);2326 }2327recover_out:2328 clear_sbi_flag(sbi, SBI_IS_RESIZEFS);2329 if (err) {2330 set_sbi_flag(sbi, SBI_NEED_FSCK);2331 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");2332 2333 spin_lock(&sbi->stat_lock);2334 sbi->user_block_count += shrunk_blocks;2335 spin_unlock(&sbi->stat_lock);2336 }2337out_err:2338 f2fs_up_write(&sbi->cp_global_sem);2339 f2fs_up_write(&sbi->gc_lock);2340 thaw_super(sbi->sb, FREEZE_HOLDER_USERSPACE);2341 return err;2342}2343