fs/f2fs/gc.c

   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/backing-dev.h>
  11 #include <linux/init.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/kthread.h>
  14 #include <linux/delay.h>
  15 #include <linux/freezer.h>
  16 #include <linux/sched/signal.h>
  17 #include <linux/random.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
  39         wait_ms = gc_th->min_sleep_time;
  40
  41         set_freezable();
  42         do {
  43                 bool sync_mode, foreground = false;
  44
  45                 wait_event_interruptible_timeout(*wq,
  46                                 kthread_should_stop() || freezing(current) ||
  47                                 waitqueue_active(fggc_wq) ||
  48                                 gc_th->gc_wake,
  49                                 msecs_to_jiffies(wait_ms));
  50
  51                 if (test_opt(sbi, GC_MERGE) && waitqueue_active(fggc_wq))
  52                         foreground = true;
  53
  54                 /* give it a try one time */
  55                 if (gc_th->gc_wake)
  56                         gc_th->gc_wake = 0;
  57
  58                 if (try_to_freeze()) {
  59                         stat_other_skip_bggc_count(sbi);
  60                         continue;
  61                 }
  62                 if (kthread_should_stop())
  63                         break;
  64
  65                 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
  66                         increase_sleep_time(gc_th, &wait_ms);
  67                         stat_other_skip_bggc_count(sbi);
  68                         continue;
  69                 }
  70
  71                 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
  72                         f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
  73                         f2fs_stop_checkpoint(sbi, false);
  74                 }
  75
  76                 if (!sb_start_write_trylock(sbi->sb)) {
  77                         stat_other_skip_bggc_count(sbi);
  78                         continue;
  79                 }
  80
  81                 /*
  82                  * [GC triggering condition]
  83                  * 0. GC is not conducted currently.
  84                  * 1. There are enough dirty segments.
  85                  * 2. IO subsystem is idle by checking the # of writeback pages.
  86                  * 3. IO subsystem is idle by checking the # of requests in
  87                  *    bdev's request list.
  88                  *
  89                  * Note) We have to avoid triggering GCs frequently.
  90                  * Because it is possible that some segments can be
  91                  * invalidated soon after by user update or deletion.
  92                  * So, I'd like to wait some time to collect dirty segments.
  93                  */
  94                 if (sbi->gc_mode == GC_URGENT_HIGH) {
  95                         wait_ms = gc_th->urgent_sleep_time;
  96                         down_write(&sbi->gc_lock);
  97                         goto do_gc;
  98                 }
  99
 100                 if (foreground) {
 101                         down_write(&sbi->gc_lock);
 102                         goto do_gc;
 103                 } else if (!down_write_trylock(&sbi->gc_lock)) {
 104                         stat_other_skip_bggc_count(sbi);
 105                         goto next;
 106                 }
 107
 108                 if (!is_idle(sbi, GC_TIME)) {
 109                         increase_sleep_time(gc_th, &wait_ms);
 110                         up_write(&sbi->gc_lock);
 111                         stat_io_skip_bggc_count(sbi);
 112                         goto next;
 113                 }
 114
 115                 if (has_enough_invalid_blocks(sbi))
 116                         decrease_sleep_time(gc_th, &wait_ms);
 117                 else
 118                         increase_sleep_time(gc_th, &wait_ms);
 119 do_gc:
 120                 if (!foreground)
 121                         stat_inc_bggc_count(sbi->stat_info);
 122
 123                 sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
 124
 125                 /* foreground GC was been triggered via f2fs_balance_fs() */
 126                 if (foreground)
 127                         sync_mode = false;
 128
 129                 /* if return value is not zero, no victim was selected */
 130                 if (f2fs_gc(sbi, sync_mode, !foreground, false, NULL_SEGNO))
 131                         wait_ms = gc_th->no_gc_sleep_time;
 132
 133                 if (foreground)
 134                         wake_up_all(&gc_th->fggc_wq);
 135
 136                 trace_f2fs_background_gc(sbi->sb, wait_ms,
 137                                 prefree_segments(sbi), free_segments(sbi));
 138
 139                 /* balancing f2fs's metadata periodically */
 140                 f2fs_balance_fs_bg(sbi, true);
 141 next:
 142                 sb_end_write(sbi->sb);
 143
 144         } while (!kthread_should_stop());
 145         return 0;
 146 }
 147
 148 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
 149 {
 150         struct f2fs_gc_kthread *gc_th;
 151         dev_t dev = sbi->sb->s_bdev->bd_dev;
 152         int err = 0;
 153
 154         gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 155         if (!gc_th) {
 156                 err = -ENOMEM;
 157                 goto out;
 158         }
 159
 160         gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
 161         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 162         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 163         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 164
 165         gc_th->gc_wake = 0;
 166
 167         sbi->gc_thread = gc_th;
 168         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 169         init_waitqueue_head(&sbi->gc_thread->fggc_wq);
 170         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 171                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 172         if (IS_ERR(gc_th->f2fs_gc_task)) {
 173                 err = PTR_ERR(gc_th->f2fs_gc_task);
 174                 kfree(gc_th);
 175                 sbi->gc_thread = NULL;
 176         }
 177 out:
 178         return err;
 179 }
 180
 181 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
 182 {
 183         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 184
 185         if (!gc_th)
 186                 return;
 187         kthread_stop(gc_th->f2fs_gc_task);
 188         wake_up_all(&gc_th->fggc_wq);
 189         kfree(gc_th);
 190         sbi->gc_thread = NULL;
 191 }
 192
 193 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
 194 {
 195         int gc_mode;
 196
 197         if (gc_type == BG_GC) {
 198                 if (sbi->am.atgc_enabled)
 199                         gc_mode = GC_AT;
 200                 else
 201                         gc_mode = GC_CB;
 202         } else {
 203                 gc_mode = GC_GREEDY;
 204         }
 205
 206         switch (sbi->gc_mode) {
 207         case GC_IDLE_CB:
 208                 gc_mode = GC_CB;
 209                 break;
 210         case GC_IDLE_GREEDY:
 211         case GC_URGENT_HIGH:
 212                 gc_mode = GC_GREEDY;
 213                 break;
 214         case GC_IDLE_AT:
 215                 gc_mode = GC_AT;
 216                 break;
 217         }
 218
 219         return gc_mode;
 220 }
 221
 222 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 223                         int type, struct victim_sel_policy *p)
 224 {
 225         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 226
 227         if (p->alloc_mode == SSR) {
 228                 p->gc_mode = GC_GREEDY;
 229                 p->dirty_bitmap = dirty_i->dirty_segmap[type];
 230                 p->max_search = dirty_i->nr_dirty[type];
 231                 p->ofs_unit = 1;
 232         } else if (p->alloc_mode == AT_SSR) {
 233                 p->gc_mode = GC_GREEDY;
 234                 p->dirty_bitmap = dirty_i->dirty_segmap[type];
 235                 p->max_search = dirty_i->nr_dirty[type];
 236                 p->ofs_unit = 1;
 237         } else {
 238                 p->gc_mode = select_gc_type(sbi, gc_type);
 239                 p->ofs_unit = sbi->segs_per_sec;
 240                 if (__is_large_section(sbi)) {
 241                         p->dirty_bitmap = dirty_i->dirty_secmap;
 242                         p->max_search = count_bits(p->dirty_bitmap,
 243                                                 0, MAIN_SECS(sbi));
 244                 } else {
 245                         p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
 246                         p->max_search = dirty_i->nr_dirty[DIRTY];
 247                 }
 248         }
 249
 250         /*
 251          * adjust candidates range, should select all dirty segments for
 252          * foreground GC and urgent GC cases.
 253          */
 254         if (gc_type != FG_GC &&
 255                         (sbi->gc_mode != GC_URGENT_HIGH) &&
 256                         (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
 257                         p->max_search > sbi->max_victim_search)
 258                 p->max_search = sbi->max_victim_search;
 259
 260         /* let's select beginning hot/small space first in no_heap mode*/
 261         if (f2fs_need_rand_seg(sbi))
 262                 p->offset = prandom_u32() % (MAIN_SECS(sbi) * sbi->segs_per_sec);
 263         else if (test_opt(sbi, NOHEAP) &&
 264                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
 265                 p->offset = 0;
 266         else
 267                 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
 268 }
 269
 270 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 271                                 struct victim_sel_policy *p)
 272 {
 273         /* SSR allocates in a segment unit */
 274         if (p->alloc_mode == SSR)
 275                 return sbi->blocks_per_seg;
 276         else if (p->alloc_mode == AT_SSR)
 277                 return UINT_MAX;
 278
 279         /* LFS */
 280         if (p->gc_mode == GC_GREEDY)
 281                 return 2 * sbi->blocks_per_seg * p->ofs_unit;
 282         else if (p->gc_mode == GC_CB)
 283                 return UINT_MAX;
 284         else if (p->gc_mode == GC_AT)
 285                 return UINT_MAX;
 286         else /* No other gc_mode */
 287                 return 0;
 288 }
 289
 290 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 291 {
 292         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 293         unsigned int secno;
 294
 295         /*
 296          * If the gc_type is FG_GC, we can select victim segments
 297          * selected by background GC before.
 298          * Those segments guarantee they have small valid blocks.
 299          */
 300         for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
 301                 if (sec_usage_check(sbi, secno))
 302                         continue;
 303                 clear_bit(secno, dirty_i->victim_secmap);
 304                 return GET_SEG_FROM_SEC(sbi, secno);
 305         }
 306         return NULL_SEGNO;
 307 }
 308
 309 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 310 {
 311         struct sit_info *sit_i = SIT_I(sbi);
 312         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 313         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 314         unsigned long long mtime = 0;
 315         unsigned int vblocks;
 316         unsigned char age = 0;
 317         unsigned char u;
 318         unsigned int i;
 319         unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
 320
 321         for (i = 0; i < usable_segs_per_sec; i++)
 322                 mtime += get_seg_entry(sbi, start + i)->mtime;
 323         vblocks = get_valid_blocks(sbi, segno, true);
 324
 325         mtime = div_u64(mtime, usable_segs_per_sec);
 326         vblocks = div_u64(vblocks, usable_segs_per_sec);
 327
 328         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 329
 330         /* Handle if the system time has changed by the user */
 331         if (mtime < sit_i->min_mtime)
 332                 sit_i->min_mtime = mtime;
 333         if (mtime > sit_i->max_mtime)
 334                 sit_i->max_mtime = mtime;
 335         if (sit_i->max_mtime != sit_i->min_mtime)
 336                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 337                                 sit_i->max_mtime - sit_i->min_mtime);
 338
 339         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 340 }
 341
 342 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 343                         unsigned int segno, struct victim_sel_policy *p)
 344 {
 345         if (p->alloc_mode == SSR)
 346                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 347
 348         /* alloc_mode == LFS */
 349         if (p->gc_mode == GC_GREEDY)
 350                 return get_valid_blocks(sbi, segno, true);
 351         else if (p->gc_mode == GC_CB)
 352                 return get_cb_cost(sbi, segno);
 353
 354         f2fs_bug_on(sbi, 1);
 355         return 0;
 356 }
 357
 358 static unsigned int count_bits(const unsigned long *addr,
 359                                 unsigned int offset, unsigned int len)
 360 {
 361         unsigned int end = offset + len, sum = 0;
 362
 363         while (offset < end) {
 364                 if (test_bit(offset++, addr))
 365                         ++sum;
 366         }
 367         return sum;
 368 }
 369
 370 static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi,
 371                                 unsigned long long mtime, unsigned int segno,
 372                                 struct rb_node *parent, struct rb_node **p,
 373                                 bool left_most)
 374 {
 375         struct atgc_management *am = &sbi->am;
 376         struct victim_entry *ve;
 377
 378         ve =  f2fs_kmem_cache_alloc(victim_entry_slab,
 379                                 GFP_NOFS, true, NULL);
 380
 381         ve->mtime = mtime;
 382         ve->segno = segno;
 383
 384         rb_link_node(&ve->rb_node, parent, p);
 385         rb_insert_color_cached(&ve->rb_node, &am->root, left_most);
 386
 387         list_add_tail(&ve->list, &am->victim_list);
 388
 389         am->victim_count++;
 390
 391         return ve;
 392 }
 393
 394 static void insert_victim_entry(struct f2fs_sb_info *sbi,
 395                                 unsigned long long mtime, unsigned int segno)
 396 {
 397         struct atgc_management *am = &sbi->am;
 398         struct rb_node **p;
 399         struct rb_node *parent = NULL;
 400         bool left_most = true;
 401
 402         p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most);
 403         attach_victim_entry(sbi, mtime, segno, parent, p, left_most);
 404 }
 405
 406 static void add_victim_entry(struct f2fs_sb_info *sbi,
 407                                 struct victim_sel_policy *p, unsigned int segno)
 408 {
 409         struct sit_info *sit_i = SIT_I(sbi);
 410         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 411         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 412         unsigned long long mtime = 0;
 413         unsigned int i;
 414
 415         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 416                 if (p->gc_mode == GC_AT &&
 417                         get_valid_blocks(sbi, segno, true) == 0)
 418                         return;
 419         }
 420
 421         for (i = 0; i < sbi->segs_per_sec; i++)
 422                 mtime += get_seg_entry(sbi, start + i)->mtime;
 423         mtime = div_u64(mtime, sbi->segs_per_sec);
 424
 425         /* Handle if the system time has changed by the user */
 426         if (mtime < sit_i->min_mtime)
 427                 sit_i->min_mtime = mtime;
 428         if (mtime > sit_i->max_mtime)
 429                 sit_i->max_mtime = mtime;
 430         if (mtime < sit_i->dirty_min_mtime)
 431                 sit_i->dirty_min_mtime = mtime;
 432         if (mtime > sit_i->dirty_max_mtime)
 433                 sit_i->dirty_max_mtime = mtime;
 434
 435         /* don't choose young section as candidate */
 436         if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
 437                 return;
 438
 439         insert_victim_entry(sbi, mtime, segno);
 440 }
 441
 442 static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi,
 443                                                 struct victim_sel_policy *p)
 444 {
 445         struct atgc_management *am = &sbi->am;
 446         struct rb_node *parent = NULL;
 447         bool left_most;
 448
 449         f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most);
 450
 451         return parent;
 452 }
 453
 454 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
 455                                                 struct victim_sel_policy *p)
 456 {
 457         struct sit_info *sit_i = SIT_I(sbi);
 458         struct atgc_management *am = &sbi->am;
 459         struct rb_root_cached *root = &am->root;
 460         struct rb_node *node;
 461         struct rb_entry *re;
 462         struct victim_entry *ve;
 463         unsigned long long total_time;
 464         unsigned long long age, u, accu;
 465         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 466         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 467         unsigned int sec_blocks = BLKS_PER_SEC(sbi);
 468         unsigned int vblocks;
 469         unsigned int dirty_threshold = max(am->max_candidate_count,
 470                                         am->candidate_ratio *
 471                                         am->victim_count / 100);
 472         unsigned int age_weight = am->age_weight;
 473         unsigned int cost;
 474         unsigned int iter = 0;
 475
 476         if (max_mtime < min_mtime)
 477                 return;
 478
 479         max_mtime += 1;
 480         total_time = max_mtime - min_mtime;
 481
 482         accu = div64_u64(ULLONG_MAX, total_time);
 483         accu = min_t(unsigned long long, div_u64(accu, 100),
 484                                         DEFAULT_ACCURACY_CLASS);
 485
 486         node = rb_first_cached(root);
 487 next:
 488         re = rb_entry_safe(node, struct rb_entry, rb_node);
 489         if (!re)
 490                 return;
 491
 492         ve = (struct victim_entry *)re;
 493
 494         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 495                 goto skip;
 496
 497         /* age = 10000 * x% * 60 */
 498         age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
 499                                                                 age_weight;
 500
 501         vblocks = get_valid_blocks(sbi, ve->segno, true);
 502         f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
 503
 504         /* u = 10000 * x% * 40 */
 505         u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
 506                                                         (100 - age_weight);
 507
 508         f2fs_bug_on(sbi, age + u >= UINT_MAX);
 509
 510         cost = UINT_MAX - (age + u);
 511         iter++;
 512
 513         if (cost < p->min_cost ||
 514                         (cost == p->min_cost && age > p->oldest_age)) {
 515                 p->min_cost = cost;
 516                 p->oldest_age = age;
 517                 p->min_segno = ve->segno;
 518         }
 519 skip:
 520         if (iter < dirty_threshold) {
 521                 node = rb_next(node);
 522                 goto next;
 523         }
 524 }
 525
 526 /*
 527  * select candidates around source section in range of
 528  * [target - dirty_threshold, target + dirty_threshold]
 529  */
 530 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
 531                                                 struct victim_sel_policy *p)
 532 {
 533         struct sit_info *sit_i = SIT_I(sbi);
 534         struct atgc_management *am = &sbi->am;
 535         struct rb_node *node;
 536         struct rb_entry *re;
 537         struct victim_entry *ve;
 538         unsigned long long age;
 539         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 540         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 541         unsigned int seg_blocks = sbi->blocks_per_seg;
 542         unsigned int vblocks;
 543         unsigned int dirty_threshold = max(am->max_candidate_count,
 544                                         am->candidate_ratio *
 545                                         am->victim_count / 100);
 546         unsigned int cost;
 547         unsigned int iter = 0;
 548         int stage = 0;
 549
 550         if (max_mtime < min_mtime)
 551                 return;
 552         max_mtime += 1;
 553 next_stage:
 554         node = lookup_central_victim(sbi, p);
 555 next_node:
 556         re = rb_entry_safe(node, struct rb_entry, rb_node);
 557         if (!re) {
 558                 if (stage == 0)
 559                         goto skip_stage;
 560                 return;
 561         }
 562
 563         ve = (struct victim_entry *)re;
 564
 565         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 566                 goto skip_node;
 567
 568         age = max_mtime - ve->mtime;
 569
 570         vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
 571         f2fs_bug_on(sbi, !vblocks);
 572
 573         /* rare case */
 574         if (vblocks == seg_blocks)
 575                 goto skip_node;
 576
 577         iter++;
 578
 579         age = max_mtime - abs(p->age - age);
 580         cost = UINT_MAX - vblocks;
 581
 582         if (cost < p->min_cost ||
 583                         (cost == p->min_cost && age > p->oldest_age)) {
 584                 p->min_cost = cost;
 585                 p->oldest_age = age;
 586                 p->min_segno = ve->segno;
 587         }
 588 skip_node:
 589         if (iter < dirty_threshold) {
 590                 if (stage == 0)
 591                         node = rb_prev(node);
 592                 else if (stage == 1)
 593                         node = rb_next(node);
 594                 goto next_node;
 595         }
 596 skip_stage:
 597         if (stage < 1) {
 598                 stage++;
 599                 iter = 0;
 600                 goto next_stage;
 601         }
 602 }
 603 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
 604                                                 struct victim_sel_policy *p)
 605 {
 606         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
 607                                                 &sbi->am.root, true));
 608
 609         if (p->gc_mode == GC_AT)
 610                 atgc_lookup_victim(sbi, p);
 611         else if (p->alloc_mode == AT_SSR)
 612                 atssr_lookup_victim(sbi, p);
 613         else
 614                 f2fs_bug_on(sbi, 1);
 615 }
 616
 617 static void release_victim_entry(struct f2fs_sb_info *sbi)
 618 {
 619         struct atgc_management *am = &sbi->am;
 620         struct victim_entry *ve, *tmp;
 621
 622         list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
 623                 list_del(&ve->list);
 624                 kmem_cache_free(victim_entry_slab, ve);
 625                 am->victim_count--;
 626         }
 627
 628         am->root = RB_ROOT_CACHED;
 629
 630         f2fs_bug_on(sbi, am->victim_count);
 631         f2fs_bug_on(sbi, !list_empty(&am->victim_list));
 632 }
 633
 634 /*
 635  * This function is called from two paths.
 636  * One is garbage collection and the other is SSR segment selection.
 637  * When it is called during GC, it just gets a victim segment
 638  * and it does not remove it from dirty seglist.
 639  * When it is called from SSR segment selection, it finds a segment
 640  * which has minimum valid blocks and removes it from dirty seglist.
 641  */
 642 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 643                         unsigned int *result, int gc_type, int type,
 644                         char alloc_mode, unsigned long long age)
 645 {
 646         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 647         struct sit_info *sm = SIT_I(sbi);
 648         struct victim_sel_policy p;
 649         unsigned int secno, last_victim;
 650         unsigned int last_segment;
 651         unsigned int nsearched;
 652         bool is_atgc;
 653         int ret = 0;
 654
 655         mutex_lock(&dirty_i->seglist_lock);
 656         last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
 657
 658         p.alloc_mode = alloc_mode;
 659         p.age = age;
 660         p.age_threshold = sbi->am.age_threshold;
 661
 662 retry:
 663         select_policy(sbi, gc_type, type, &p);
 664         p.min_segno = NULL_SEGNO;
 665         p.oldest_age = 0;
 666         p.min_cost = get_max_cost(sbi, &p);
 667
 668         is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
 669         nsearched = 0;
 670
 671         if (is_atgc)
 672                 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
 673
 674         if (*result != NULL_SEGNO) {
 675                 if (!get_valid_blocks(sbi, *result, false)) {
 676                         ret = -ENODATA;
 677                         goto out;
 678                 }
 679
 680                 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
 681                         ret = -EBUSY;
 682                 else
 683                         p.min_segno = *result;
 684                 goto out;
 685         }
 686
 687         ret = -ENODATA;
 688         if (p.max_search == 0)
 689                 goto out;
 690
 691         if (__is_large_section(sbi) && p.alloc_mode == LFS) {
 692                 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
 693                         p.min_segno = sbi->next_victim_seg[BG_GC];
 694                         *result = p.min_segno;
 695                         sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
 696                         goto got_result;
 697                 }
 698                 if (gc_type == FG_GC &&
 699                                 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
 700                         p.min_segno = sbi->next_victim_seg[FG_GC];
 701                         *result = p.min_segno;
 702                         sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
 703                         goto got_result;
 704                 }
 705         }
 706
 707         last_victim = sm->last_victim[p.gc_mode];
 708         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 709                 p.min_segno = check_bg_victims(sbi);
 710                 if (p.min_segno != NULL_SEGNO)
 711                         goto got_it;
 712         }
 713
 714         while (1) {
 715                 unsigned long cost, *dirty_bitmap;
 716                 unsigned int unit_no, segno;
 717
 718                 dirty_bitmap = p.dirty_bitmap;
 719                 unit_no = find_next_bit(dirty_bitmap,
 720                                 last_segment / p.ofs_unit,
 721                                 p.offset / p.ofs_unit);
 722                 segno = unit_no * p.ofs_unit;
 723                 if (segno >= last_segment) {
 724                         if (sm->last_victim[p.gc_mode]) {
 725                                 last_segment =
 726                                         sm->last_victim[p.gc_mode];
 727                                 sm->last_victim[p.gc_mode] = 0;
 728                                 p.offset = 0;
 729                                 continue;
 730                         }
 731                         break;
 732                 }
 733
 734                 p.offset = segno + p.ofs_unit;
 735                 nsearched++;
 736
 737 #ifdef CONFIG_F2FS_CHECK_FS
 738                 /*
 739                  * skip selecting the invalid segno (that is failed due to block
 740                  * validity check failure during GC) to avoid endless GC loop in
 741                  * such cases.
 742                  */
 743                 if (test_bit(segno, sm->invalid_segmap))
 744                         goto next;
 745 #endif
 746
 747                 secno = GET_SEC_FROM_SEG(sbi, segno);
 748
 749                 if (sec_usage_check(sbi, secno))
 750                         goto next;
 751
 752                 /* Don't touch checkpointed data */
 753                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 754                         if (p.alloc_mode == LFS) {
 755                                 /*
 756                                  * LFS is set to find source section during GC.
 757                                  * The victim should have no checkpointed data.
 758                                  */
 759                                 if (get_ckpt_valid_blocks(sbi, segno, true))
 760                                         goto next;
 761                         } else {
 762                                 /*
 763                                  * SSR | AT_SSR are set to find target segment
 764                                  * for writes which can be full by checkpointed
 765                                  * and newly written blocks.
 766                                  */
 767                                 if (!f2fs_segment_has_free_slot(sbi, segno))
 768                                         goto next;
 769                         }
 770                 }
 771
 772                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 773                         goto next;
 774
 775                 if (is_atgc) {
 776                         add_victim_entry(sbi, &p, segno);
 777                         goto next;
 778                 }
 779
 780                 cost = get_gc_cost(sbi, segno, &p);
 781
 782                 if (p.min_cost > cost) {
 783                         p.min_segno = segno;
 784                         p.min_cost = cost;
 785                 }
 786 next:
 787                 if (nsearched >= p.max_search) {
 788                         if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
 789                                 sm->last_victim[p.gc_mode] =
 790                                         last_victim + p.ofs_unit;
 791                         else
 792                                 sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
 793                         sm->last_victim[p.gc_mode] %=
 794                                 (MAIN_SECS(sbi) * sbi->segs_per_sec);
 795                         break;
 796                 }
 797         }
 798
 799         /* get victim for GC_AT/AT_SSR */
 800         if (is_atgc) {
 801                 lookup_victim_by_age(sbi, &p);
 802                 release_victim_entry(sbi);
 803         }
 804
 805         if (is_atgc && p.min_segno == NULL_SEGNO &&
 806                         sm->elapsed_time < p.age_threshold) {
 807                 p.age_threshold = 0;
 808                 goto retry;
 809         }
 810
 811         if (p.min_segno != NULL_SEGNO) {
 812 got_it:
 813                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 814 got_result:
 815                 if (p.alloc_mode == LFS) {
 816                         secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
 817                         if (gc_type == FG_GC)
 818                                 sbi->cur_victim_sec = secno;
 819                         else
 820                                 set_bit(secno, dirty_i->victim_secmap);
 821                 }
 822                 ret = 0;
 823
 824         }
 825 out:
 826         if (p.min_segno != NULL_SEGNO)
 827                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 828                                 sbi->cur_victim_sec,
 829                                 prefree_segments(sbi), free_segments(sbi));
 830         mutex_unlock(&dirty_i->seglist_lock);
 831
 832         return ret;
 833 }
 834
 835 static const struct victim_selection default_v_ops = {
 836         .get_victim = get_victim_by_default,
 837 };
 838
 839 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
 840 {
 841         struct inode_entry *ie;
 842
 843         ie = radix_tree_lookup(&gc_list->iroot, ino);
 844         if (ie)
 845                 return ie->inode;
 846         return NULL;
 847 }
 848
 849 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
 850 {
 851         struct inode_entry *new_ie;
 852
 853         if (inode == find_gc_inode(gc_list, inode->i_ino)) {
 854                 iput(inode);
 855                 return;
 856         }
 857         new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab,
 858                                         GFP_NOFS, true, NULL);
 859         new_ie->inode = inode;
 860
 861         f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
 862         list_add_tail(&new_ie->list, &gc_list->ilist);
 863 }
 864
 865 static void put_gc_inode(struct gc_inode_list *gc_list)
 866 {
 867         struct inode_entry *ie, *next_ie;
 868
 869         list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
 870                 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
 871                 iput(ie->inode);
 872                 list_del(&ie->list);
 873                 kmem_cache_free(f2fs_inode_entry_slab, ie);
 874         }
 875 }
 876
 877 static int check_valid_map(struct f2fs_sb_info *sbi,
 878                                 unsigned int segno, int offset)
 879 {
 880         struct sit_info *sit_i = SIT_I(sbi);
 881         struct seg_entry *sentry;
 882         int ret;
 883
 884         down_read(&sit_i->sentry_lock);
 885         sentry = get_seg_entry(sbi, segno);
 886         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 887         up_read(&sit_i->sentry_lock);
 888         return ret;
 889 }
 890
 891 /*
 892  * This function compares node address got in summary with that in NAT.
 893  * On validity, copy that node with cold status, otherwise (invalid node)
 894  * ignore that.
 895  */
 896 static int gc_node_segment(struct f2fs_sb_info *sbi,
 897                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 898 {
 899         struct f2fs_summary *entry;
 900         block_t start_addr;
 901         int off;
 902         int phase = 0;
 903         bool fggc = (gc_type == FG_GC);
 904         int submitted = 0;
 905         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
 906
 907         start_addr = START_BLOCK(sbi, segno);
 908
 909 next_step:
 910         entry = sum;
 911
 912         if (fggc && phase == 2)
 913                 atomic_inc(&sbi->wb_sync_req[NODE]);
 914
 915         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
 916                 nid_t nid = le32_to_cpu(entry->nid);
 917                 struct page *node_page;
 918                 struct node_info ni;
 919                 int err;
 920
 921                 /* stop BG_GC if there is not enough free sections. */
 922                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 923                         return submitted;
 924
 925                 if (check_valid_map(sbi, segno, off) == 0)
 926                         continue;
 927
 928                 if (phase == 0) {
 929                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
 930                                                         META_NAT, true);
 931                         continue;
 932                 }
 933
 934                 if (phase == 1) {
 935                         f2fs_ra_node_page(sbi, nid);
 936                         continue;
 937                 }
 938
 939                 /* phase == 2 */
 940                 node_page = f2fs_get_node_page(sbi, nid);
 941                 if (IS_ERR(node_page))
 942                         continue;
 943
 944                 /* block may become invalid during f2fs_get_node_page */
 945                 if (check_valid_map(sbi, segno, off) == 0) {
 946                         f2fs_put_page(node_page, 1);
 947                         continue;
 948                 }
 949
 950                 if (f2fs_get_node_info(sbi, nid, &ni)) {
 951                         f2fs_put_page(node_page, 1);
 952                         continue;
 953                 }
 954
 955                 if (ni.blk_addr != start_addr + off) {
 956                         f2fs_put_page(node_page, 1);
 957                         continue;
 958                 }
 959
 960                 err = f2fs_move_node_page(node_page, gc_type);
 961                 if (!err && gc_type == FG_GC)
 962                         submitted++;
 963                 stat_inc_node_blk_count(sbi, 1, gc_type);
 964         }
 965
 966         if (++phase < 3)
 967                 goto next_step;
 968
 969         if (fggc)
 970                 atomic_dec(&sbi->wb_sync_req[NODE]);
 971         return submitted;
 972 }
 973
 974 /*
 975  * Calculate start block index indicating the given node offset.
 976  * Be careful, caller should give this node offset only indicating direct node
 977  * blocks. If any node offsets, which point the other types of node blocks such
 978  * as indirect or double indirect node blocks, are given, it must be a caller's
 979  * bug.
 980  */
 981 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
 982 {
 983         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
 984         unsigned int bidx;
 985
 986         if (node_ofs == 0)
 987                 return 0;
 988
 989         if (node_ofs <= 2) {
 990                 bidx = node_ofs - 1;
 991         } else if (node_ofs <= indirect_blks) {
 992                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
 993
 994                 bidx = node_ofs - 2 - dec;
 995         } else {
 996                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
 997
 998                 bidx = node_ofs - 5 - dec;
 999         }
1000         return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
1001 }
1002
1003 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1004                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
1005 {
1006         struct page *node_page;
1007         nid_t nid;
1008         unsigned int ofs_in_node;
1009         block_t source_blkaddr;
1010
1011         nid = le32_to_cpu(sum->nid);
1012         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
1013
1014         node_page = f2fs_get_node_page(sbi, nid);
1015         if (IS_ERR(node_page))
1016                 return false;
1017
1018         if (f2fs_get_node_info(sbi, nid, dni)) {
1019                 f2fs_put_page(node_page, 1);
1020                 return false;
1021         }
1022
1023         if (sum->version != dni->version) {
1024                 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
1025                           __func__);
1026                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1027         }
1028
1029         *nofs = ofs_of_node(node_page);
1030         source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1031         f2fs_put_page(node_page, 1);
1032
1033         if (source_blkaddr != blkaddr) {
1034 #ifdef CONFIG_F2FS_CHECK_FS
1035                 unsigned int segno = GET_SEGNO(sbi, blkaddr);
1036                 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1037
1038                 if (unlikely(check_valid_map(sbi, segno, offset))) {
1039                         if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1040                                 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",
1041                                          blkaddr, source_blkaddr, segno);
1042                                 f2fs_bug_on(sbi, 1);
1043                         }
1044                 }
1045 #endif
1046                 return false;
1047         }
1048         return true;
1049 }
1050
1051 static int ra_data_block(struct inode *inode, pgoff_t index)
1052 {
1053         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1054         struct address_space *mapping = inode->i_mapping;
1055         struct dnode_of_data dn;
1056         struct page *page;
1057         struct extent_info ei = {0, 0, 0};
1058         struct f2fs_io_info fio = {
1059                 .sbi = sbi,
1060                 .ino = inode->i_ino,
1061                 .type = DATA,
1062                 .temp = COLD,
1063                 .op = REQ_OP_READ,
1064                 .op_flags = 0,
1065                 .encrypted_page = NULL,
1066                 .in_list = false,
1067                 .retry = false,
1068         };
1069         int err;
1070
1071         page = f2fs_grab_cache_page(mapping, index, true);
1072         if (!page)
1073                 return -ENOMEM;
1074
1075         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1076                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1077                 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1078                                                 DATA_GENERIC_ENHANCE_READ))) {
1079                         err = -EFSCORRUPTED;
1080                         goto put_page;
1081                 }
1082                 goto got_it;
1083         }
1084
1085         set_new_dnode(&dn, inode, NULL, NULL, 0);
1086         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1087         if (err)
1088                 goto put_page;
1089         f2fs_put_dnode(&dn);
1090
1091         if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1092                 err = -ENOENT;
1093                 goto put_page;
1094         }
1095         if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1096                                                 DATA_GENERIC_ENHANCE))) {
1097                 err = -EFSCORRUPTED;
1098                 goto put_page;
1099         }
1100 got_it:
1101         /* read page */
1102         fio.page = page;
1103         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1104
1105         /*
1106          * don't cache encrypted data into meta inode until previous dirty
1107          * data were writebacked to avoid racing between GC and flush.
1108          */
1109         f2fs_wait_on_page_writeback(page, DATA, true, true);
1110
1111         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1112
1113         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1114                                         dn.data_blkaddr,
1115                                         FGP_LOCK | FGP_CREAT, GFP_NOFS);
1116         if (!fio.encrypted_page) {
1117                 err = -ENOMEM;
1118                 goto put_page;
1119         }
1120
1121         err = f2fs_submit_page_bio(&fio);
1122         if (err)
1123                 goto put_encrypted_page;
1124         f2fs_put_page(fio.encrypted_page, 0);
1125         f2fs_put_page(page, 1);
1126
1127         f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1128         f2fs_update_iostat(sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1129
1130         return 0;
1131 put_encrypted_page:
1132         f2fs_put_page(fio.encrypted_page, 1);
1133 put_page:
1134         f2fs_put_page(page, 1);
1135         return err;
1136 }
1137
1138 /*
1139  * Move data block via META_MAPPING while keeping locked data page.
1140  * This can be used to move blocks, aka LBAs, directly on disk.
1141  */
1142 static int move_data_block(struct inode *inode, block_t bidx,
1143                                 int gc_type, unsigned int segno, int off)
1144 {
1145         struct f2fs_io_info fio = {
1146                 .sbi = F2FS_I_SB(inode),
1147                 .ino = inode->i_ino,
1148                 .type = DATA,
1149                 .temp = COLD,
1150                 .op = REQ_OP_READ,
1151                 .op_flags = 0,
1152                 .encrypted_page = NULL,
1153                 .in_list = false,
1154                 .retry = false,
1155         };
1156         struct dnode_of_data dn;
1157         struct f2fs_summary sum;
1158         struct node_info ni;
1159         struct page *page, *mpage;
1160         block_t newaddr;
1161         int err = 0;
1162         bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1163         int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&
1164                                 (fio.sbi->gc_mode != GC_URGENT_HIGH) ?
1165                                 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1166
1167         /* do not read out */
1168         page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1169         if (!page)
1170                 return -ENOMEM;
1171
1172         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1173                 err = -ENOENT;
1174                 goto out;
1175         }
1176
1177         if (f2fs_is_atomic_file(inode)) {
1178                 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1179                 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1180                 err = -EAGAIN;
1181                 goto out;
1182         }
1183
1184         if (f2fs_is_pinned_file(inode)) {
1185                 f2fs_pin_file_control(inode, true);
1186                 err = -EAGAIN;
1187                 goto out;
1188         }
1189
1190         set_new_dnode(&dn, inode, NULL, NULL, 0);
1191         err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1192         if (err)
1193                 goto out;
1194
1195         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1196                 ClearPageUptodate(page);
1197                 err = -ENOENT;
1198                 goto put_out;
1199         }
1200
1201         /*
1202          * don't cache encrypted data into meta inode until previous dirty
1203          * data were writebacked to avoid racing between GC and flush.
1204          */
1205         f2fs_wait_on_page_writeback(page, DATA, true, true);
1206
1207         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1208
1209         err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1210         if (err)
1211                 goto put_out;
1212
1213         /* read page */
1214         fio.page = page;
1215         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1216
1217         if (lfs_mode)
1218                 down_write(&fio.sbi->io_order_lock);
1219
1220         mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1221                                         fio.old_blkaddr, false);
1222         if (!mpage) {
1223                 err = -ENOMEM;
1224                 goto up_out;
1225         }
1226
1227         fio.encrypted_page = mpage;
1228
1229         /* read source block in mpage */
1230         if (!PageUptodate(mpage)) {
1231                 err = f2fs_submit_page_bio(&fio);
1232                 if (err) {
1233                         f2fs_put_page(mpage, 1);
1234                         goto up_out;
1235                 }
1236
1237                 f2fs_update_iostat(fio.sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1238                 f2fs_update_iostat(fio.sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1239
1240                 lock_page(mpage);
1241                 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1242                                                 !PageUptodate(mpage))) {
1243                         err = -EIO;
1244                         f2fs_put_page(mpage, 1);
1245                         goto up_out;
1246                 }
1247         }
1248
1249         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1250
1251         /* allocate block address */
1252         f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1253                                 &sum, type, NULL);
1254
1255         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1256                                 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1257         if (!fio.encrypted_page) {
1258                 err = -ENOMEM;
1259                 f2fs_put_page(mpage, 1);
1260                 goto recover_block;
1261         }
1262
1263         /* write target block */
1264         f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1265         memcpy(page_address(fio.encrypted_page),
1266                                 page_address(mpage), PAGE_SIZE);
1267         f2fs_put_page(mpage, 1);
1268         invalidate_mapping_pages(META_MAPPING(fio.sbi),
1269                                 fio.old_blkaddr, fio.old_blkaddr);
1270         f2fs_invalidate_compress_page(fio.sbi, fio.old_blkaddr);
1271
1272         set_page_dirty(fio.encrypted_page);
1273         if (clear_page_dirty_for_io(fio.encrypted_page))
1274                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
1275
1276         set_page_writeback(fio.encrypted_page);
1277         ClearPageError(page);
1278
1279         fio.op = REQ_OP_WRITE;
1280         fio.op_flags = REQ_SYNC;
1281         fio.new_blkaddr = newaddr;
1282         f2fs_submit_page_write(&fio);
1283         if (fio.retry) {
1284                 err = -EAGAIN;
1285                 if (PageWriteback(fio.encrypted_page))
1286                         end_page_writeback(fio.encrypted_page);
1287                 goto put_page_out;
1288         }
1289
1290         f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
1291
1292         f2fs_update_data_blkaddr(&dn, newaddr);
1293         set_inode_flag(inode, FI_APPEND_WRITE);
1294         if (page->index == 0)
1295                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1296 put_page_out:
1297         f2fs_put_page(fio.encrypted_page, 1);
1298 recover_block:
1299         if (err)
1300                 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1301                                                         true, true, true);
1302 up_out:
1303         if (lfs_mode)
1304                 up_write(&fio.sbi->io_order_lock);
1305 put_out:
1306         f2fs_put_dnode(&dn);
1307 out:
1308         f2fs_put_page(page, 1);
1309         return err;
1310 }
1311
1312 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1313                                                         unsigned int segno, int off)
1314 {
1315         struct page *page;
1316         int err = 0;
1317
1318         page = f2fs_get_lock_data_page(inode, bidx, true);
1319         if (IS_ERR(page))
1320                 return PTR_ERR(page);
1321
1322         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1323                 err = -ENOENT;
1324                 goto out;
1325         }
1326
1327         if (f2fs_is_atomic_file(inode)) {
1328                 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1329                 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1330                 err = -EAGAIN;
1331                 goto out;
1332         }
1333         if (f2fs_is_pinned_file(inode)) {
1334                 if (gc_type == FG_GC)
1335                         f2fs_pin_file_control(inode, true);
1336                 err = -EAGAIN;
1337                 goto out;
1338         }
1339
1340         if (gc_type == BG_GC) {
1341                 if (PageWriteback(page)) {
1342                         err = -EAGAIN;
1343                         goto out;
1344                 }
1345                 set_page_dirty(page);
1346                 set_page_private_gcing(page);
1347         } else {
1348                 struct f2fs_io_info fio = {
1349                         .sbi = F2FS_I_SB(inode),
1350                         .ino = inode->i_ino,
1351                         .type = DATA,
1352                         .temp = COLD,
1353                         .op = REQ_OP_WRITE,
1354                         .op_flags = REQ_SYNC,
1355                         .old_blkaddr = NULL_ADDR,
1356                         .page = page,
1357                         .encrypted_page = NULL,
1358                         .need_lock = LOCK_REQ,
1359                         .io_type = FS_GC_DATA_IO,
1360                 };
1361                 bool is_dirty = PageDirty(page);
1362
1363 retry:
1364                 f2fs_wait_on_page_writeback(page, DATA, true, true);
1365
1366                 set_page_dirty(page);
1367                 if (clear_page_dirty_for_io(page)) {
1368                         inode_dec_dirty_pages(inode);
1369                         f2fs_remove_dirty_inode(inode);
1370                 }
1371
1372                 set_page_private_gcing(page);
1373
1374                 err = f2fs_do_write_data_page(&fio);
1375                 if (err) {
1376                         clear_page_private_gcing(page);
1377                         if (err == -ENOMEM) {
1378                                 congestion_wait(BLK_RW_ASYNC,
1379                                                 DEFAULT_IO_TIMEOUT);
1380                                 goto retry;
1381                         }
1382                         if (is_dirty)
1383                                 set_page_dirty(page);
1384                 }
1385         }
1386 out:
1387         f2fs_put_page(page, 1);
1388         return err;
1389 }
1390
1391 /*
1392  * This function tries to get parent node of victim data block, and identifies
1393  * data block validity. If the block is valid, copy that with cold status and
1394  * modify parent node.
1395  * If the parent node is not valid or the data block address is different,
1396  * the victim data block is ignored.
1397  */
1398 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1399                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1400                 bool force_migrate)
1401 {
1402         struct super_block *sb = sbi->sb;
1403         struct f2fs_summary *entry;
1404         block_t start_addr;
1405         int off;
1406         int phase = 0;
1407         int submitted = 0;
1408         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1409
1410         start_addr = START_BLOCK(sbi, segno);
1411
1412 next_step:
1413         entry = sum;
1414
1415         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1416                 struct page *data_page;
1417                 struct inode *inode;
1418                 struct node_info dni; /* dnode info for the data */
1419                 unsigned int ofs_in_node, nofs;
1420                 block_t start_bidx;
1421                 nid_t nid = le32_to_cpu(entry->nid);
1422
1423                 /*
1424                  * stop BG_GC if there is not enough free sections.
1425                  * Or, stop GC if the segment becomes fully valid caused by
1426                  * race condition along with SSR block allocation.
1427                  */
1428                 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1429                         (!force_migrate && get_valid_blocks(sbi, segno, true) ==
1430                                                         BLKS_PER_SEC(sbi)))
1431                         return submitted;
1432
1433                 if (check_valid_map(sbi, segno, off) == 0)
1434                         continue;
1435
1436                 if (phase == 0) {
1437                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1438                                                         META_NAT, true);
1439                         continue;
1440                 }
1441
1442                 if (phase == 1) {
1443                         f2fs_ra_node_page(sbi, nid);
1444                         continue;
1445                 }
1446
1447                 /* Get an inode by ino with checking validity */
1448                 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1449                         continue;
1450
1451                 if (phase == 2) {
1452                         f2fs_ra_node_page(sbi, dni.ino);
1453                         continue;
1454                 }
1455
1456                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1457
1458                 if (phase == 3) {
1459                         inode = f2fs_iget(sb, dni.ino);
1460                         if (IS_ERR(inode) || is_bad_inode(inode))
1461                                 continue;
1462
1463                         if (!down_write_trylock(
1464                                 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1465                                 iput(inode);
1466                                 sbi->skipped_gc_rwsem++;
1467                                 continue;
1468                         }
1469
1470                         start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1471                                                                 ofs_in_node;
1472
1473                         if (f2fs_post_read_required(inode)) {
1474                                 int err = ra_data_block(inode, start_bidx);
1475
1476                                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1477                                 if (err) {
1478                                         iput(inode);
1479                                         continue;
1480                                 }
1481                                 add_gc_inode(gc_list, inode);
1482                                 continue;
1483                         }
1484
1485                         data_page = f2fs_get_read_data_page(inode,
1486                                                 start_bidx, REQ_RAHEAD, true);
1487                         up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1488                         if (IS_ERR(data_page)) {
1489                                 iput(inode);
1490                                 continue;
1491                         }
1492
1493                         f2fs_put_page(data_page, 0);
1494                         add_gc_inode(gc_list, inode);
1495                         continue;
1496                 }
1497
1498                 /* phase 4 */
1499                 inode = find_gc_inode(gc_list, dni.ino);
1500                 if (inode) {
1501                         struct f2fs_inode_info *fi = F2FS_I(inode);
1502                         bool locked = false;
1503                         int err;
1504
1505                         if (S_ISREG(inode->i_mode)) {
1506                                 if (!down_write_trylock(&fi->i_gc_rwsem[READ])) {
1507                                         sbi->skipped_gc_rwsem++;
1508                                         continue;
1509                                 }
1510                                 if (!down_write_trylock(
1511                                                 &fi->i_gc_rwsem[WRITE])) {
1512                                         sbi->skipped_gc_rwsem++;
1513                                         up_write(&fi->i_gc_rwsem[READ]);
1514                                         continue;
1515                                 }
1516                                 locked = true;
1517
1518                                 /* wait for all inflight aio data */
1519                                 inode_dio_wait(inode);
1520                         }
1521
1522                         start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1523                                                                 + ofs_in_node;
1524                         if (f2fs_post_read_required(inode))
1525                                 err = move_data_block(inode, start_bidx,
1526                                                         gc_type, segno, off);
1527                         else
1528                                 err = move_data_page(inode, start_bidx, gc_type,
1529                                                                 segno, off);
1530
1531                         if (!err && (gc_type == FG_GC ||
1532                                         f2fs_post_read_required(inode)))
1533                                 submitted++;
1534
1535                         if (locked) {
1536                                 up_write(&fi->i_gc_rwsem[WRITE]);
1537                                 up_write(&fi->i_gc_rwsem[READ]);
1538                         }
1539
1540                         stat_inc_data_blk_count(sbi, 1, gc_type);
1541                 }
1542         }
1543
1544         if (++phase < 5)
1545                 goto next_step;
1546
1547         return submitted;
1548 }
1549
1550 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1551                         int gc_type)
1552 {
1553         struct sit_info *sit_i = SIT_I(sbi);
1554         int ret;
1555
1556         down_write(&sit_i->sentry_lock);
1557         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1558                                               NO_CHECK_TYPE, LFS, 0);
1559         up_write(&sit_i->sentry_lock);
1560         return ret;
1561 }
1562
1563 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1564                                 unsigned int start_segno,
1565                                 struct gc_inode_list *gc_list, int gc_type,
1566                                 bool force_migrate)
1567 {
1568         struct page *sum_page;
1569         struct f2fs_summary_block *sum;
1570         struct blk_plug plug;
1571         unsigned int segno = start_segno;
1572         unsigned int end_segno = start_segno + sbi->segs_per_sec;
1573         int seg_freed = 0, migrated = 0;
1574         unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1575                                                 SUM_TYPE_DATA : SUM_TYPE_NODE;
1576         int submitted = 0;
1577
1578         if (__is_large_section(sbi))
1579                 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1580
1581         /*
1582          * zone-capacity can be less than zone-size in zoned devices,
1583          * resulting in less than expected usable segments in the zone,
1584          * calculate the end segno in the zone which can be garbage collected
1585          */
1586         if (f2fs_sb_has_blkzoned(sbi))
1587                 end_segno -= sbi->segs_per_sec -
1588                                         f2fs_usable_segs_in_sec(sbi, segno);
1589
1590         sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1591
1592         /* readahead multi ssa blocks those have contiguous address */
1593         if (__is_large_section(sbi))
1594                 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1595                                         end_segno - segno, META_SSA, true);
1596
1597         /* reference all summary page */
1598         while (segno < end_segno) {
1599                 sum_page = f2fs_get_sum_page(sbi, segno++);
1600                 if (IS_ERR(sum_page)) {
1601                         int err = PTR_ERR(sum_page);
1602
1603                         end_segno = segno - 1;
1604                         for (segno = start_segno; segno < end_segno; segno++) {
1605                                 sum_page = find_get_page(META_MAPPING(sbi),
1606                                                 GET_SUM_BLOCK(sbi, segno));
1607                                 f2fs_put_page(sum_page, 0);
1608                                 f2fs_put_page(sum_page, 0);
1609                         }
1610                         return err;
1611                 }
1612                 unlock_page(sum_page);
1613         }
1614
1615         blk_start_plug(&plug);
1616
1617         for (segno = start_segno; segno < end_segno; segno++) {
1618
1619                 /* find segment summary of victim */
1620                 sum_page = find_get_page(META_MAPPING(sbi),
1621                                         GET_SUM_BLOCK(sbi, segno));
1622                 f2fs_put_page(sum_page, 0);
1623
1624                 if (get_valid_blocks(sbi, segno, false) == 0)
1625                         goto freed;
1626                 if (gc_type == BG_GC && __is_large_section(sbi) &&
1627                                 migrated >= sbi->migration_granularity)
1628                         goto skip;
1629                 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1630                         goto skip;
1631
1632                 sum = page_address(sum_page);
1633                 if (type != GET_SUM_TYPE((&sum->footer))) {
1634                         f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1635                                  segno, type, GET_SUM_TYPE((&sum->footer)));
1636                         set_sbi_flag(sbi, SBI_NEED_FSCK);
1637                         f2fs_stop_checkpoint(sbi, false);
1638                         goto skip;
1639                 }
1640
1641                 /*
1642                  * this is to avoid deadlock:
1643                  * - lock_page(sum_page)         - f2fs_replace_block
1644                  *  - check_valid_map()            - down_write(sentry_lock)
1645                  *   - down_read(sentry_lock)     - change_curseg()
1646                  *                                  - lock_page(sum_page)
1647                  */
1648                 if (type == SUM_TYPE_NODE)
1649                         submitted += gc_node_segment(sbi, sum->entries, segno,
1650                                                                 gc_type);
1651                 else
1652                         submitted += gc_data_segment(sbi, sum->entries, gc_list,
1653                                                         segno, gc_type,
1654                                                         force_migrate);
1655
1656                 stat_inc_seg_count(sbi, type, gc_type);
1657                 sbi->gc_reclaimed_segs[sbi->gc_mode]++;
1658                 migrated++;
1659
1660 freed:
1661                 if (gc_type == FG_GC &&
1662                                 get_valid_blocks(sbi, segno, false) == 0)
1663                         seg_freed++;
1664
1665                 if (__is_large_section(sbi) && segno + 1 < end_segno)
1666                         sbi->next_victim_seg[gc_type] = segno + 1;
1667 skip:
1668                 f2fs_put_page(sum_page, 0);
1669         }
1670
1671         if (submitted)
1672                 f2fs_submit_merged_write(sbi,
1673                                 (type == SUM_TYPE_NODE) ? NODE : DATA);
1674
1675         blk_finish_plug(&plug);
1676
1677         stat_inc_call_count(sbi->stat_info);
1678
1679         return seg_freed;
1680 }
1681
1682 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1683                         bool background, bool force, unsigned int segno)
1684 {
1685         int gc_type = sync ? FG_GC : BG_GC;
1686         int sec_freed = 0, seg_freed = 0, total_freed = 0;
1687         int ret = 0;
1688         struct cp_control cpc;
1689         unsigned int init_segno = segno;
1690         struct gc_inode_list gc_list = {
1691                 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1692                 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1693         };
1694         unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1695         unsigned long long first_skipped;
1696         unsigned int skipped_round = 0, round = 0;
1697
1698         trace_f2fs_gc_begin(sbi->sb, sync, background,
1699                                 get_pages(sbi, F2FS_DIRTY_NODES),
1700                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1701                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1702                                 free_sections(sbi),
1703                                 free_segments(sbi),
1704                                 reserved_segments(sbi),
1705                                 prefree_segments(sbi));
1706
1707         cpc.reason = __get_cp_reason(sbi);
1708         sbi->skipped_gc_rwsem = 0;
1709         first_skipped = last_skipped;
1710 gc_more:
1711         if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1712                 ret = -EINVAL;
1713                 goto stop;
1714         }
1715         if (unlikely(f2fs_cp_error(sbi))) {
1716                 ret = -EIO;
1717                 goto stop;
1718         }
1719
1720         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1721                 /*
1722                  * For example, if there are many prefree_segments below given
1723                  * threshold, we can make them free by checkpoint. Then, we
1724                  * secure free segments which doesn't need fggc any more.
1725                  */
1726                 if (prefree_segments(sbi) &&
1727                                 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1728                         ret = f2fs_write_checkpoint(sbi, &cpc);
1729                         if (ret)
1730                                 goto stop;
1731                 }
1732                 if (has_not_enough_free_secs(sbi, 0, 0))
1733                         gc_type = FG_GC;
1734         }
1735
1736         /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1737         if (gc_type == BG_GC && !background) {
1738                 ret = -EINVAL;
1739                 goto stop;
1740         }
1741         ret = __get_victim(sbi, &segno, gc_type);
1742         if (ret)
1743                 goto stop;
1744
1745         seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type, force);
1746         if (gc_type == FG_GC &&
1747                 seg_freed == f2fs_usable_segs_in_sec(sbi, segno))
1748                 sec_freed++;
1749         total_freed += seg_freed;
1750
1751         if (gc_type == FG_GC) {
1752                 if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
1753                                                 sbi->skipped_gc_rwsem)
1754                         skipped_round++;
1755                 last_skipped = sbi->skipped_atomic_files[FG_GC];
1756                 round++;
1757         }
1758
1759         if (gc_type == FG_GC)
1760                 sbi->cur_victim_sec = NULL_SEGNO;
1761
1762         if (sync)
1763                 goto stop;
1764
1765         if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1766                 if (skipped_round <= MAX_SKIP_GC_COUNT ||
1767                                         skipped_round * 2 < round) {
1768                         segno = NULL_SEGNO;
1769                         goto gc_more;
1770                 }
1771
1772                 if (first_skipped < last_skipped &&
1773                                 (last_skipped - first_skipped) >
1774                                                 sbi->skipped_gc_rwsem) {
1775                         f2fs_drop_inmem_pages_all(sbi, true);
1776                         segno = NULL_SEGNO;
1777                         goto gc_more;
1778                 }
1779                 if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1780                         ret = f2fs_write_checkpoint(sbi, &cpc);
1781         }
1782 stop:
1783         SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1784         SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1785
1786         trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1787                                 get_pages(sbi, F2FS_DIRTY_NODES),
1788                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1789                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1790                                 free_sections(sbi),
1791                                 free_segments(sbi),
1792                                 reserved_segments(sbi),
1793                                 prefree_segments(sbi));
1794
1795         up_write(&sbi->gc_lock);
1796
1797         put_gc_inode(&gc_list);
1798
1799         if (sync && !ret)
1800                 ret = sec_freed ? 0 : -EAGAIN;
1801         return ret;
1802 }
1803
1804 int __init f2fs_create_garbage_collection_cache(void)
1805 {
1806         victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1807                                         sizeof(struct victim_entry));
1808         if (!victim_entry_slab)
1809                 return -ENOMEM;
1810         return 0;
1811 }
1812
1813 void f2fs_destroy_garbage_collection_cache(void)
1814 {
1815         kmem_cache_destroy(victim_entry_slab);
1816 }
1817
1818 static void init_atgc_management(struct f2fs_sb_info *sbi)
1819 {
1820         struct atgc_management *am = &sbi->am;
1821
1822         if (test_opt(sbi, ATGC) &&
1823                 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1824                 am->atgc_enabled = true;
1825
1826         am->root = RB_ROOT_CACHED;
1827         INIT_LIST_HEAD(&am->victim_list);
1828         am->victim_count = 0;
1829
1830         am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1831         am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1832         am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1833         am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
1834 }
1835
1836 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1837 {
1838         DIRTY_I(sbi)->v_ops = &default_v_ops;
1839
1840         sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1841
1842         /* give warm/cold data area from slower device */
1843         if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1844                 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1845                                 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1846
1847         init_atgc_management(sbi);
1848 }
1849
1850 static int free_segment_range(struct f2fs_sb_info *sbi,
1851                                 unsigned int secs, bool gc_only)
1852 {
1853         unsigned int segno, next_inuse, start, end;
1854         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1855         int gc_mode, gc_type;
1856         int err = 0;
1857         int type;
1858
1859         /* Force block allocation for GC */
1860         MAIN_SECS(sbi) -= secs;
1861         start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1862         end = MAIN_SEGS(sbi) - 1;
1863
1864         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1865         for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1866                 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1867                         SIT_I(sbi)->last_victim[gc_mode] = 0;
1868
1869         for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1870                 if (sbi->next_victim_seg[gc_type] >= start)
1871                         sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1872         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1873
1874         /* Move out cursegs from the target range */
1875         for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1876                 f2fs_allocate_segment_for_resize(sbi, type, start, end);
1877
1878         /* do GC to move out valid blocks in the range */
1879         for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1880                 struct gc_inode_list gc_list = {
1881                         .ilist = LIST_HEAD_INIT(gc_list.ilist),
1882                         .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1883                 };
1884
1885                 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
1886                 put_gc_inode(&gc_list);
1887
1888                 if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1889                         err = -EAGAIN;
1890                         goto out;
1891                 }
1892                 if (fatal_signal_pending(current)) {
1893                         err = -ERESTARTSYS;
1894                         goto out;
1895                 }
1896         }
1897         if (gc_only)
1898                 goto out;
1899
1900         err = f2fs_write_checkpoint(sbi, &cpc);
1901         if (err)
1902                 goto out;
1903
1904         next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1905         if (next_inuse <= end) {
1906                 f2fs_err(sbi, "segno %u should be free but still inuse!",
1907                          next_inuse);
1908                 f2fs_bug_on(sbi, 1);
1909         }
1910 out:
1911         MAIN_SECS(sbi) += secs;
1912         return err;
1913 }
1914
1915 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1916 {
1917         struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1918         int section_count;
1919         int segment_count;
1920         int segment_count_main;
1921         long long block_count;
1922         int segs = secs * sbi->segs_per_sec;
1923
1924         down_write(&sbi->sb_lock);
1925
1926         section_count = le32_to_cpu(raw_sb->section_count);
1927         segment_count = le32_to_cpu(raw_sb->segment_count);
1928         segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
1929         block_count = le64_to_cpu(raw_sb->block_count);
1930
1931         raw_sb->section_count = cpu_to_le32(section_count + secs);
1932         raw_sb->segment_count = cpu_to_le32(segment_count + segs);
1933         raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
1934         raw_sb->block_count = cpu_to_le64(block_count +
1935                                         (long long)segs * sbi->blocks_per_seg);
1936         if (f2fs_is_multi_device(sbi)) {
1937                 int last_dev = sbi->s_ndevs - 1;
1938                 int dev_segs =
1939                         le32_to_cpu(raw_sb->devs[last_dev].total_segments);
1940
1941                 raw_sb->devs[last_dev].total_segments =
1942                                                 cpu_to_le32(dev_segs + segs);
1943         }
1944
1945         up_write(&sbi->sb_lock);
1946 }
1947
1948 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
1949 {
1950         int segs = secs * sbi->segs_per_sec;
1951         long long blks = (long long)segs * sbi->blocks_per_seg;
1952         long long user_block_count =
1953                                 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
1954
1955         SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
1956         MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
1957         MAIN_SECS(sbi) += secs;
1958         FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
1959         FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
1960         F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
1961
1962         if (f2fs_is_multi_device(sbi)) {
1963                 int last_dev = sbi->s_ndevs - 1;
1964
1965                 FDEV(last_dev).total_segments =
1966                                 (int)FDEV(last_dev).total_segments + segs;
1967                 FDEV(last_dev).end_blk =
1968                                 (long long)FDEV(last_dev).end_blk + blks;
1969 #ifdef CONFIG_BLK_DEV_ZONED
1970                 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
1971                                         (int)(blks >> sbi->log_blocks_per_blkz);
1972 #endif
1973         }
1974 }
1975
1976 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
1977 {
1978         __u64 old_block_count, shrunk_blocks;
1979         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1980         unsigned int secs;
1981         int err = 0;
1982         __u32 rem;
1983
1984         old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
1985         if (block_count > old_block_count)
1986                 return -EINVAL;
1987
1988         if (f2fs_is_multi_device(sbi)) {
1989                 int last_dev = sbi->s_ndevs - 1;
1990                 __u64 last_segs = FDEV(last_dev).total_segments;
1991
1992                 if (block_count + last_segs * sbi->blocks_per_seg <=
1993                                                                 old_block_count)
1994                         return -EINVAL;
1995         }
1996
1997         /* new fs size should align to section size */
1998         div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
1999         if (rem)
2000                 return -EINVAL;
2001
2002         if (block_count == old_block_count)
2003                 return 0;
2004
2005         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2006                 f2fs_err(sbi, "Should run fsck to repair first.");
2007                 return -EFSCORRUPTED;
2008         }
2009
2010         if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2011                 f2fs_err(sbi, "Checkpoint should be enabled.");
2012                 return -EINVAL;
2013         }
2014
2015         shrunk_blocks = old_block_count - block_count;
2016         secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2017
2018         /* stop other GC */
2019         if (!down_write_trylock(&sbi->gc_lock))
2020                 return -EAGAIN;
2021
2022         /* stop CP to protect MAIN_SEC in free_segment_range */
2023         f2fs_lock_op(sbi);
2024
2025         spin_lock(&sbi->stat_lock);
2026         if (shrunk_blocks + valid_user_blocks(sbi) +
2027                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2028                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2029                 err = -ENOSPC;
2030         spin_unlock(&sbi->stat_lock);
2031
2032         if (err)
2033                 goto out_unlock;
2034
2035         err = free_segment_range(sbi, secs, true);
2036
2037 out_unlock:
2038         f2fs_unlock_op(sbi);
2039         up_write(&sbi->gc_lock);
2040         if (err)
2041                 return err;
2042
2043         set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2044
2045         freeze_super(sbi->sb);
2046         down_write(&sbi->gc_lock);
2047         down_write(&sbi->cp_global_sem);
2048
2049         spin_lock(&sbi->stat_lock);
2050         if (shrunk_blocks + valid_user_blocks(sbi) +
2051                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2052                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2053                 err = -ENOSPC;
2054         else
2055                 sbi->user_block_count -= shrunk_blocks;
2056         spin_unlock(&sbi->stat_lock);
2057         if (err)
2058                 goto out_err;
2059
2060         err = free_segment_range(sbi, secs, false);
2061         if (err)
2062                 goto recover_out;
2063
2064         update_sb_metadata(sbi, -secs);
2065
2066         err = f2fs_commit_super(sbi, false);
2067         if (err) {
2068                 update_sb_metadata(sbi, secs);
2069                 goto recover_out;
2070         }
2071
2072         update_fs_metadata(sbi, -secs);
2073         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2074         set_sbi_flag(sbi, SBI_IS_DIRTY);
2075
2076         err = f2fs_write_checkpoint(sbi, &cpc);
2077         if (err) {
2078                 update_fs_metadata(sbi, secs);
2079                 update_sb_metadata(sbi, secs);
2080                 f2fs_commit_super(sbi, false);
2081         }
2082 recover_out:
2083         if (err) {
2084                 set_sbi_flag(sbi, SBI_NEED_FSCK);
2085                 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2086
2087                 spin_lock(&sbi->stat_lock);
2088                 sbi->user_block_count += shrunk_blocks;
2089                 spin_unlock(&sbi->stat_lock);
2090         }
2091 out_err:
2092         up_write(&sbi->cp_global_sem);
2093         up_write(&sbi->gc_lock);
2094         thaw_super(sbi->sb);
2095         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2096         return err;
2097 }