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
  18 #include "f2fs.h"
  19 #include "node.h"
  20 #include "segment.h"
  21 #include "gc.h"
  22 #include <trace/events/f2fs.h>
  23
  24 static struct kmem_cache *victim_entry_slab;
  25
  26 static unsigned int count_bits(const unsigned long *addr,
  27                                 unsigned int offset, unsigned int len);
  28
  29 static int gc_thread_func(void *data)
  30 {
  31         struct f2fs_sb_info *sbi = data;
  32         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
  33         wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
  34         unsigned int wait_ms;
  35
  36         wait_ms = gc_th->min_sleep_time;
  37
  38         set_freezable();
  39         do {
  40                 bool sync_mode;
  41
  42                 wait_event_interruptible_timeout(*wq,
  43                                 kthread_should_stop() || freezing(current) ||
  44                                 gc_th->gc_wake,
  45                                 msecs_to_jiffies(wait_ms));
  46
  47                 /* give it a try one time */
  48                 if (gc_th->gc_wake)
  49                         gc_th->gc_wake = 0;
  50
  51                 if (try_to_freeze()) {
  52                         stat_other_skip_bggc_count(sbi);
  53                         continue;
  54                 }
  55                 if (kthread_should_stop())
  56                         break;
  57
  58                 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
  59                         increase_sleep_time(gc_th, &wait_ms);
  60                         stat_other_skip_bggc_count(sbi);
  61                         continue;
  62                 }
  63
  64                 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
  65                         f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
  66                         f2fs_stop_checkpoint(sbi, false);
  67                 }
  68
  69                 if (!sb_start_write_trylock(sbi->sb)) {
  70                         stat_other_skip_bggc_count(sbi);
  71                         continue;
  72                 }
  73
  74                 /*
  75                  * [GC triggering condition]
  76                  * 0. GC is not conducted currently.
  77                  * 1. There are enough dirty segments.
  78                  * 2. IO subsystem is idle by checking the # of writeback pages.
  79                  * 3. IO subsystem is idle by checking the # of requests in
  80                  *    bdev's request list.
  81                  *
  82                  * Note) We have to avoid triggering GCs frequently.
  83                  * Because it is possible that some segments can be
  84                  * invalidated soon after by user update or deletion.
  85                  * So, I'd like to wait some time to collect dirty segments.
  86                  */
  87                 if (sbi->gc_mode == GC_URGENT_HIGH) {
  88                         wait_ms = gc_th->urgent_sleep_time;
  89                         down_write(&sbi->gc_lock);
  90                         goto do_gc;
  91                 }
  92
  93                 if (!down_write_trylock(&sbi->gc_lock)) {
  94                         stat_other_skip_bggc_count(sbi);
  95                         goto next;
  96                 }
  97
  98                 if (!is_idle(sbi, GC_TIME)) {
  99                         increase_sleep_time(gc_th, &wait_ms);
 100                         up_write(&sbi->gc_lock);
 101                         stat_io_skip_bggc_count(sbi);
 102                         goto next;
 103                 }
 104
 105                 if (has_enough_invalid_blocks(sbi))
 106                         decrease_sleep_time(gc_th, &wait_ms);
 107                 else
 108                         increase_sleep_time(gc_th, &wait_ms);
 109 do_gc:
 110                 stat_inc_bggc_count(sbi->stat_info);
 111
 112                 sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
 113
 114                 /* if return value is not zero, no victim was selected */
 115                 if (f2fs_gc(sbi, sync_mode, true, false, NULL_SEGNO))
 116                         wait_ms = gc_th->no_gc_sleep_time;
 117
 118                 trace_f2fs_background_gc(sbi->sb, wait_ms,
 119                                 prefree_segments(sbi), free_segments(sbi));
 120
 121                 /* balancing f2fs's metadata periodically */
 122                 f2fs_balance_fs_bg(sbi, true);
 123 next:
 124                 sb_end_write(sbi->sb);
 125
 126         } while (!kthread_should_stop());
 127         return 0;
 128 }
 129
 130 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
 131 {
 132         struct f2fs_gc_kthread *gc_th;
 133         dev_t dev = sbi->sb->s_bdev->bd_dev;
 134         int err = 0;
 135
 136         gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 137         if (!gc_th) {
 138                 err = -ENOMEM;
 139                 goto out;
 140         }
 141
 142         gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
 143         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 144         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 145         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 146
 147         gc_th->gc_wake= 0;
 148
 149         sbi->gc_thread = gc_th;
 150         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 151         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 152                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 153         if (IS_ERR(gc_th->f2fs_gc_task)) {
 154                 err = PTR_ERR(gc_th->f2fs_gc_task);
 155                 kfree(gc_th);
 156                 sbi->gc_thread = NULL;
 157         }
 158 out:
 159         return err;
 160 }
 161
 162 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
 163 {
 164         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 165         if (!gc_th)
 166                 return;
 167         kthread_stop(gc_th->f2fs_gc_task);
 168         kfree(gc_th);
 169         sbi->gc_thread = NULL;
 170 }
 171
 172 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
 173 {
 174         int gc_mode;
 175
 176         if (gc_type == BG_GC) {
 177                 if (sbi->am.atgc_enabled)
 178                         gc_mode = GC_AT;
 179                 else
 180                         gc_mode = GC_CB;
 181         } else {
 182                 gc_mode = GC_GREEDY;
 183         }
 184
 185         switch (sbi->gc_mode) {
 186         case GC_IDLE_CB:
 187                 gc_mode = GC_CB;
 188                 break;
 189         case GC_IDLE_GREEDY:
 190         case GC_URGENT_HIGH:
 191                 gc_mode = GC_GREEDY;
 192                 break;
 193         case GC_IDLE_AT:
 194                 gc_mode = GC_AT;
 195                 break;
 196         }
 197
 198         return gc_mode;
 199 }
 200
 201 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 202                         int type, struct victim_sel_policy *p)
 203 {
 204         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 205
 206         if (p->alloc_mode == SSR) {
 207                 p->gc_mode = GC_GREEDY;
 208                 p->dirty_bitmap = dirty_i->dirty_segmap[type];
 209                 p->max_search = dirty_i->nr_dirty[type];
 210                 p->ofs_unit = 1;
 211         } else if (p->alloc_mode == AT_SSR) {
 212                 p->gc_mode = GC_GREEDY;
 213                 p->dirty_bitmap = dirty_i->dirty_segmap[type];
 214                 p->max_search = dirty_i->nr_dirty[type];
 215                 p->ofs_unit = 1;
 216         } else {
 217                 p->gc_mode = select_gc_type(sbi, gc_type);
 218                 p->ofs_unit = sbi->segs_per_sec;
 219                 if (__is_large_section(sbi)) {
 220                         p->dirty_bitmap = dirty_i->dirty_secmap;
 221                         p->max_search = count_bits(p->dirty_bitmap,
 222                                                 0, MAIN_SECS(sbi));
 223                 } else {
 224                         p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
 225                         p->max_search = dirty_i->nr_dirty[DIRTY];
 226                 }
 227         }
 228
 229         /*
 230          * adjust candidates range, should select all dirty segments for
 231          * foreground GC and urgent GC cases.
 232          */
 233         if (gc_type != FG_GC &&
 234                         (sbi->gc_mode != GC_URGENT_HIGH) &&
 235                         (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
 236                         p->max_search > sbi->max_victim_search)
 237                 p->max_search = sbi->max_victim_search;
 238
 239         /* let's select beginning hot/small space first in no_heap mode*/
 240         if (test_opt(sbi, NOHEAP) &&
 241                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
 242                 p->offset = 0;
 243         else
 244                 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
 245 }
 246
 247 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 248                                 struct victim_sel_policy *p)
 249 {
 250         /* SSR allocates in a segment unit */
 251         if (p->alloc_mode == SSR)
 252                 return sbi->blocks_per_seg;
 253         else if (p->alloc_mode == AT_SSR)
 254                 return UINT_MAX;
 255
 256         /* LFS */
 257         if (p->gc_mode == GC_GREEDY)
 258                 return 2 * sbi->blocks_per_seg * p->ofs_unit;
 259         else if (p->gc_mode == GC_CB)
 260                 return UINT_MAX;
 261         else if (p->gc_mode == GC_AT)
 262                 return UINT_MAX;
 263         else /* No other gc_mode */
 264                 return 0;
 265 }
 266
 267 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 268 {
 269         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 270         unsigned int secno;
 271
 272         /*
 273          * If the gc_type is FG_GC, we can select victim segments
 274          * selected by background GC before.
 275          * Those segments guarantee they have small valid blocks.
 276          */
 277         for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
 278                 if (sec_usage_check(sbi, secno))
 279                         continue;
 280                 clear_bit(secno, dirty_i->victim_secmap);
 281                 return GET_SEG_FROM_SEC(sbi, secno);
 282         }
 283         return NULL_SEGNO;
 284 }
 285
 286 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 287 {
 288         struct sit_info *sit_i = SIT_I(sbi);
 289         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 290         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 291         unsigned long long mtime = 0;
 292         unsigned int vblocks;
 293         unsigned char age = 0;
 294         unsigned char u;
 295         unsigned int i;
 296         unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
 297
 298         for (i = 0; i < usable_segs_per_sec; i++)
 299                 mtime += get_seg_entry(sbi, start + i)->mtime;
 300         vblocks = get_valid_blocks(sbi, segno, true);
 301
 302         mtime = div_u64(mtime, usable_segs_per_sec);
 303         vblocks = div_u64(vblocks, usable_segs_per_sec);
 304
 305         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 306
 307         /* Handle if the system time has changed by the user */
 308         if (mtime < sit_i->min_mtime)
 309                 sit_i->min_mtime = mtime;
 310         if (mtime > sit_i->max_mtime)
 311                 sit_i->max_mtime = mtime;
 312         if (sit_i->max_mtime != sit_i->min_mtime)
 313                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 314                                 sit_i->max_mtime - sit_i->min_mtime);
 315
 316         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 317 }
 318
 319 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 320                         unsigned int segno, struct victim_sel_policy *p)
 321 {
 322         if (p->alloc_mode == SSR)
 323                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 324
 325         /* alloc_mode == LFS */
 326         if (p->gc_mode == GC_GREEDY)
 327                 return get_valid_blocks(sbi, segno, true);
 328         else if (p->gc_mode == GC_CB)
 329                 return get_cb_cost(sbi, segno);
 330
 331         f2fs_bug_on(sbi, 1);
 332         return 0;
 333 }
 334
 335 static unsigned int count_bits(const unsigned long *addr,
 336                                 unsigned int offset, unsigned int len)
 337 {
 338         unsigned int end = offset + len, sum = 0;
 339
 340         while (offset < end) {
 341                 if (test_bit(offset++, addr))
 342                         ++sum;
 343         }
 344         return sum;
 345 }
 346
 347 static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi,
 348                                 unsigned long long mtime, unsigned int segno,
 349                                 struct rb_node *parent, struct rb_node **p,
 350                                 bool left_most)
 351 {
 352         struct atgc_management *am = &sbi->am;
 353         struct victim_entry *ve;
 354
 355         ve =  f2fs_kmem_cache_alloc(victim_entry_slab, GFP_NOFS);
 356
 357         ve->mtime = mtime;
 358         ve->segno = segno;
 359
 360         rb_link_node(&ve->rb_node, parent, p);
 361         rb_insert_color_cached(&ve->rb_node, &am->root, left_most);
 362
 363         list_add_tail(&ve->list, &am->victim_list);
 364
 365         am->victim_count++;
 366
 367         return ve;
 368 }
 369
 370 static void insert_victim_entry(struct f2fs_sb_info *sbi,
 371                                 unsigned long long mtime, unsigned int segno)
 372 {
 373         struct atgc_management *am = &sbi->am;
 374         struct rb_node **p;
 375         struct rb_node *parent = NULL;
 376         bool left_most = true;
 377
 378         p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most);
 379         attach_victim_entry(sbi, mtime, segno, parent, p, left_most);
 380 }
 381
 382 static void add_victim_entry(struct f2fs_sb_info *sbi,
 383                                 struct victim_sel_policy *p, unsigned int segno)
 384 {
 385         struct sit_info *sit_i = SIT_I(sbi);
 386         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 387         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 388         unsigned long long mtime = 0;
 389         unsigned int i;
 390
 391         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 392                 if (p->gc_mode == GC_AT &&
 393                         get_valid_blocks(sbi, segno, true) == 0)
 394                         return;
 395
 396                 if (p->alloc_mode == AT_SSR &&
 397                         get_seg_entry(sbi, segno)->ckpt_valid_blocks == 0)
 398                         return;
 399         }
 400
 401         for (i = 0; i < sbi->segs_per_sec; i++)
 402                 mtime += get_seg_entry(sbi, start + i)->mtime;
 403         mtime = div_u64(mtime, sbi->segs_per_sec);
 404
 405         /* Handle if the system time has changed by the user */
 406         if (mtime < sit_i->min_mtime)
 407                 sit_i->min_mtime = mtime;
 408         if (mtime > sit_i->max_mtime)
 409                 sit_i->max_mtime = mtime;
 410         if (mtime < sit_i->dirty_min_mtime)
 411                 sit_i->dirty_min_mtime = mtime;
 412         if (mtime > sit_i->dirty_max_mtime)
 413                 sit_i->dirty_max_mtime = mtime;
 414
 415         /* don't choose young section as candidate */
 416         if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
 417                 return;
 418
 419         insert_victim_entry(sbi, mtime, segno);
 420 }
 421
 422 static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi,
 423                                                 struct victim_sel_policy *p)
 424 {
 425         struct atgc_management *am = &sbi->am;
 426         struct rb_node *parent = NULL;
 427         bool left_most;
 428
 429         f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most);
 430
 431         return parent;
 432 }
 433
 434 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
 435                                                 struct victim_sel_policy *p)
 436 {
 437         struct sit_info *sit_i = SIT_I(sbi);
 438         struct atgc_management *am = &sbi->am;
 439         struct rb_root_cached *root = &am->root;
 440         struct rb_node *node;
 441         struct rb_entry *re;
 442         struct victim_entry *ve;
 443         unsigned long long total_time;
 444         unsigned long long age, u, accu;
 445         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 446         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 447         unsigned int sec_blocks = BLKS_PER_SEC(sbi);
 448         unsigned int vblocks;
 449         unsigned int dirty_threshold = max(am->max_candidate_count,
 450                                         am->candidate_ratio *
 451                                         am->victim_count / 100);
 452         unsigned int age_weight = am->age_weight;
 453         unsigned int cost;
 454         unsigned int iter = 0;
 455
 456         if (max_mtime < min_mtime)
 457                 return;
 458
 459         max_mtime += 1;
 460         total_time = max_mtime - min_mtime;
 461
 462         accu = div64_u64(ULLONG_MAX, total_time);
 463         accu = min_t(unsigned long long, div_u64(accu, 100),
 464                                         DEFAULT_ACCURACY_CLASS);
 465
 466         node = rb_first_cached(root);
 467 next:
 468         re = rb_entry_safe(node, struct rb_entry, rb_node);
 469         if (!re)
 470                 return;
 471
 472         ve = (struct victim_entry *)re;
 473
 474         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 475                 goto skip;
 476
 477         /* age = 10000 * x% * 60 */
 478         age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
 479                                                                 age_weight;
 480
 481         vblocks = get_valid_blocks(sbi, ve->segno, true);
 482         f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
 483
 484         /* u = 10000 * x% * 40 */
 485         u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
 486                                                         (100 - age_weight);
 487
 488         f2fs_bug_on(sbi, age + u >= UINT_MAX);
 489
 490         cost = UINT_MAX - (age + u);
 491         iter++;
 492
 493         if (cost < p->min_cost ||
 494                         (cost == p->min_cost && age > p->oldest_age)) {
 495                 p->min_cost = cost;
 496                 p->oldest_age = age;
 497                 p->min_segno = ve->segno;
 498         }
 499 skip:
 500         if (iter < dirty_threshold) {
 501                 node = rb_next(node);
 502                 goto next;
 503         }
 504 }
 505
 506 /*
 507  * select candidates around source section in range of
 508  * [target - dirty_threshold, target + dirty_threshold]
 509  */
 510 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
 511                                                 struct victim_sel_policy *p)
 512 {
 513         struct sit_info *sit_i = SIT_I(sbi);
 514         struct atgc_management *am = &sbi->am;
 515         struct rb_node *node;
 516         struct rb_entry *re;
 517         struct victim_entry *ve;
 518         unsigned long long age;
 519         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 520         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 521         unsigned int seg_blocks = sbi->blocks_per_seg;
 522         unsigned int vblocks;
 523         unsigned int dirty_threshold = max(am->max_candidate_count,
 524                                         am->candidate_ratio *
 525                                         am->victim_count / 100);
 526         unsigned int cost;
 527         unsigned int iter = 0;
 528         int stage = 0;
 529
 530         if (max_mtime < min_mtime)
 531                 return;
 532         max_mtime += 1;
 533 next_stage:
 534         node = lookup_central_victim(sbi, p);
 535 next_node:
 536         re = rb_entry_safe(node, struct rb_entry, rb_node);
 537         if (!re) {
 538                 if (stage == 0)
 539                         goto skip_stage;
 540                 return;
 541         }
 542
 543         ve = (struct victim_entry *)re;
 544
 545         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 546                 goto skip_node;
 547
 548         age = max_mtime - ve->mtime;
 549
 550         vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
 551         f2fs_bug_on(sbi, !vblocks);
 552
 553         /* rare case */
 554         if (vblocks == seg_blocks)
 555                 goto skip_node;
 556
 557         iter++;
 558
 559         age = max_mtime - abs(p->age - age);
 560         cost = UINT_MAX - vblocks;
 561
 562         if (cost < p->min_cost ||
 563                         (cost == p->min_cost && age > p->oldest_age)) {
 564                 p->min_cost = cost;
 565                 p->oldest_age = age;
 566                 p->min_segno = ve->segno;
 567         }
 568 skip_node:
 569         if (iter < dirty_threshold) {
 570                 if (stage == 0)
 571                         node = rb_prev(node);
 572                 else if (stage == 1)
 573                         node = rb_next(node);
 574                 goto next_node;
 575         }
 576 skip_stage:
 577         if (stage < 1) {
 578                 stage++;
 579                 iter = 0;
 580                 goto next_stage;
 581         }
 582 }
 583 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
 584                                                 struct victim_sel_policy *p)
 585 {
 586         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
 587                                                 &sbi->am.root, true));
 588
 589         if (p->gc_mode == GC_AT)
 590                 atgc_lookup_victim(sbi, p);
 591         else if (p->alloc_mode == AT_SSR)
 592                 atssr_lookup_victim(sbi, p);
 593         else
 594                 f2fs_bug_on(sbi, 1);
 595 }
 596
 597 static void release_victim_entry(struct f2fs_sb_info *sbi)
 598 {
 599         struct atgc_management *am = &sbi->am;
 600         struct victim_entry *ve, *tmp;
 601
 602         list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
 603                 list_del(&ve->list);
 604                 kmem_cache_free(victim_entry_slab, ve);
 605                 am->victim_count--;
 606         }
 607
 608         am->root = RB_ROOT_CACHED;
 609
 610         f2fs_bug_on(sbi, am->victim_count);
 611         f2fs_bug_on(sbi, !list_empty(&am->victim_list));
 612 }
 613
 614 /*
 615  * This function is called from two paths.
 616  * One is garbage collection and the other is SSR segment selection.
 617  * When it is called during GC, it just gets a victim segment
 618  * and it does not remove it from dirty seglist.
 619  * When it is called from SSR segment selection, it finds a segment
 620  * which has minimum valid blocks and removes it from dirty seglist.
 621  */
 622 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 623                         unsigned int *result, int gc_type, int type,
 624                         char alloc_mode, unsigned long long age)
 625 {
 626         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 627         struct sit_info *sm = SIT_I(sbi);
 628         struct victim_sel_policy p;
 629         unsigned int secno, last_victim;
 630         unsigned int last_segment;
 631         unsigned int nsearched;
 632         bool is_atgc;
 633         int ret = 0;
 634
 635         mutex_lock(&dirty_i->seglist_lock);
 636         last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
 637
 638         p.alloc_mode = alloc_mode;
 639         p.age = age;
 640         p.age_threshold = sbi->am.age_threshold;
 641
 642 retry:
 643         select_policy(sbi, gc_type, type, &p);
 644         p.min_segno = NULL_SEGNO;
 645         p.oldest_age = 0;
 646         p.min_cost = get_max_cost(sbi, &p);
 647
 648         is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
 649         nsearched = 0;
 650
 651         if (is_atgc)
 652                 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
 653
 654         if (*result != NULL_SEGNO) {
 655                 if (!get_valid_blocks(sbi, *result, false)) {
 656                         ret = -ENODATA;
 657                         goto out;
 658                 }
 659
 660                 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
 661                         ret = -EBUSY;
 662                 else
 663                         p.min_segno = *result;
 664                 goto out;
 665         }
 666
 667         ret = -ENODATA;
 668         if (p.max_search == 0)
 669                 goto out;
 670
 671         if (__is_large_section(sbi) && p.alloc_mode == LFS) {
 672                 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
 673                         p.min_segno = sbi->next_victim_seg[BG_GC];
 674                         *result = p.min_segno;
 675                         sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
 676                         goto got_result;
 677                 }
 678                 if (gc_type == FG_GC &&
 679                                 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
 680                         p.min_segno = sbi->next_victim_seg[FG_GC];
 681                         *result = p.min_segno;
 682                         sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
 683                         goto got_result;
 684                 }
 685         }
 686
 687         last_victim = sm->last_victim[p.gc_mode];
 688         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 689                 p.min_segno = check_bg_victims(sbi);
 690                 if (p.min_segno != NULL_SEGNO)
 691                         goto got_it;
 692         }
 693
 694         while (1) {
 695                 unsigned long cost, *dirty_bitmap;
 696                 unsigned int unit_no, segno;
 697
 698                 dirty_bitmap = p.dirty_bitmap;
 699                 unit_no = find_next_bit(dirty_bitmap,
 700                                 last_segment / p.ofs_unit,
 701                                 p.offset / p.ofs_unit);
 702                 segno = unit_no * p.ofs_unit;
 703                 if (segno >= last_segment) {
 704                         if (sm->last_victim[p.gc_mode]) {
 705                                 last_segment =
 706                                         sm->last_victim[p.gc_mode];
 707                                 sm->last_victim[p.gc_mode] = 0;
 708                                 p.offset = 0;
 709                                 continue;
 710                         }
 711                         break;
 712                 }
 713
 714                 p.offset = segno + p.ofs_unit;
 715                 nsearched++;
 716
 717 #ifdef CONFIG_F2FS_CHECK_FS
 718                 /*
 719                  * skip selecting the invalid segno (that is failed due to block
 720                  * validity check failure during GC) to avoid endless GC loop in
 721                  * such cases.
 722                  */
 723                 if (test_bit(segno, sm->invalid_segmap))
 724                         goto next;
 725 #endif
 726
 727                 secno = GET_SEC_FROM_SEG(sbi, segno);
 728
 729                 if (sec_usage_check(sbi, secno))
 730                         goto next;
 731                 /* Don't touch checkpointed data */
 732                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
 733                                         get_ckpt_valid_blocks(sbi, segno) &&
 734                                         p.alloc_mode == LFS))
 735                         goto next;
 736                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 737                         goto next;
 738
 739                 if (is_atgc) {
 740                         add_victim_entry(sbi, &p, segno);
 741                         goto next;
 742                 }
 743
 744                 cost = get_gc_cost(sbi, segno, &p);
 745
 746                 if (p.min_cost > cost) {
 747                         p.min_segno = segno;
 748                         p.min_cost = cost;
 749                 }
 750 next:
 751                 if (nsearched >= p.max_search) {
 752                         if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
 753                                 sm->last_victim[p.gc_mode] =
 754                                         last_victim + p.ofs_unit;
 755                         else
 756                                 sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
 757                         sm->last_victim[p.gc_mode] %=
 758                                 (MAIN_SECS(sbi) * sbi->segs_per_sec);
 759                         break;
 760                 }
 761         }
 762
 763         /* get victim for GC_AT/AT_SSR */
 764         if (is_atgc) {
 765                 lookup_victim_by_age(sbi, &p);
 766                 release_victim_entry(sbi);
 767         }
 768
 769         if (is_atgc && p.min_segno == NULL_SEGNO &&
 770                         sm->elapsed_time < p.age_threshold) {
 771                 p.age_threshold = 0;
 772                 goto retry;
 773         }
 774
 775         if (p.min_segno != NULL_SEGNO) {
 776 got_it:
 777                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 778 got_result:
 779                 if (p.alloc_mode == LFS) {
 780                         secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
 781                         if (gc_type == FG_GC)
 782                                 sbi->cur_victim_sec = secno;
 783                         else
 784                                 set_bit(secno, dirty_i->victim_secmap);
 785                 }
 786                 ret = 0;
 787
 788         }
 789 out:
 790         if (p.min_segno != NULL_SEGNO)
 791                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 792                                 sbi->cur_victim_sec,
 793                                 prefree_segments(sbi), free_segments(sbi));
 794         mutex_unlock(&dirty_i->seglist_lock);
 795
 796         return ret;
 797 }
 798
 799 static const struct victim_selection default_v_ops = {
 800         .get_victim = get_victim_by_default,
 801 };
 802
 803 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
 804 {
 805         struct inode_entry *ie;
 806
 807         ie = radix_tree_lookup(&gc_list->iroot, ino);
 808         if (ie)
 809                 return ie->inode;
 810         return NULL;
 811 }
 812
 813 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
 814 {
 815         struct inode_entry *new_ie;
 816
 817         if (inode == find_gc_inode(gc_list, inode->i_ino)) {
 818                 iput(inode);
 819                 return;
 820         }
 821         new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab, GFP_NOFS);
 822         new_ie->inode = inode;
 823
 824         f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
 825         list_add_tail(&new_ie->list, &gc_list->ilist);
 826 }
 827
 828 static void put_gc_inode(struct gc_inode_list *gc_list)
 829 {
 830         struct inode_entry *ie, *next_ie;
 831         list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
 832                 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
 833                 iput(ie->inode);
 834                 list_del(&ie->list);
 835                 kmem_cache_free(f2fs_inode_entry_slab, ie);
 836         }
 837 }
 838
 839 static int check_valid_map(struct f2fs_sb_info *sbi,
 840                                 unsigned int segno, int offset)
 841 {
 842         struct sit_info *sit_i = SIT_I(sbi);
 843         struct seg_entry *sentry;
 844         int ret;
 845
 846         down_read(&sit_i->sentry_lock);
 847         sentry = get_seg_entry(sbi, segno);
 848         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 849         up_read(&sit_i->sentry_lock);
 850         return ret;
 851 }
 852
 853 /*
 854  * This function compares node address got in summary with that in NAT.
 855  * On validity, copy that node with cold status, otherwise (invalid node)
 856  * ignore that.
 857  */
 858 static int gc_node_segment(struct f2fs_sb_info *sbi,
 859                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 860 {
 861         struct f2fs_summary *entry;
 862         block_t start_addr;
 863         int off;
 864         int phase = 0;
 865         bool fggc = (gc_type == FG_GC);
 866         int submitted = 0;
 867         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
 868
 869         start_addr = START_BLOCK(sbi, segno);
 870
 871 next_step:
 872         entry = sum;
 873
 874         if (fggc && phase == 2)
 875                 atomic_inc(&sbi->wb_sync_req[NODE]);
 876
 877         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
 878                 nid_t nid = le32_to_cpu(entry->nid);
 879                 struct page *node_page;
 880                 struct node_info ni;
 881                 int err;
 882
 883                 /* stop BG_GC if there is not enough free sections. */
 884                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 885                         return submitted;
 886
 887                 if (check_valid_map(sbi, segno, off) == 0)
 888                         continue;
 889
 890                 if (phase == 0) {
 891                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
 892                                                         META_NAT, true);
 893                         continue;
 894                 }
 895
 896                 if (phase == 1) {
 897                         f2fs_ra_node_page(sbi, nid);
 898                         continue;
 899                 }
 900
 901                 /* phase == 2 */
 902                 node_page = f2fs_get_node_page(sbi, nid);
 903                 if (IS_ERR(node_page))
 904                         continue;
 905
 906                 /* block may become invalid during f2fs_get_node_page */
 907                 if (check_valid_map(sbi, segno, off) == 0) {
 908                         f2fs_put_page(node_page, 1);
 909                         continue;
 910                 }
 911
 912                 if (f2fs_get_node_info(sbi, nid, &ni)) {
 913                         f2fs_put_page(node_page, 1);
 914                         continue;
 915                 }
 916
 917                 if (ni.blk_addr != start_addr + off) {
 918                         f2fs_put_page(node_page, 1);
 919                         continue;
 920                 }
 921
 922                 err = f2fs_move_node_page(node_page, gc_type);
 923                 if (!err && gc_type == FG_GC)
 924                         submitted++;
 925                 stat_inc_node_blk_count(sbi, 1, gc_type);
 926         }
 927
 928         if (++phase < 3)
 929                 goto next_step;
 930
 931         if (fggc)
 932                 atomic_dec(&sbi->wb_sync_req[NODE]);
 933         return submitted;
 934 }
 935
 936 /*
 937  * Calculate start block index indicating the given node offset.
 938  * Be careful, caller should give this node offset only indicating direct node
 939  * blocks. If any node offsets, which point the other types of node blocks such
 940  * as indirect or double indirect node blocks, are given, it must be a caller's
 941  * bug.
 942  */
 943 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
 944 {
 945         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
 946         unsigned int bidx;
 947
 948         if (node_ofs == 0)
 949                 return 0;
 950
 951         if (node_ofs <= 2) {
 952                 bidx = node_ofs - 1;
 953         } else if (node_ofs <= indirect_blks) {
 954                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
 955                 bidx = node_ofs - 2 - dec;
 956         } else {
 957                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
 958                 bidx = node_ofs - 5 - dec;
 959         }
 960         return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
 961 }
 962
 963 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 964                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
 965 {
 966         struct page *node_page;
 967         nid_t nid;
 968         unsigned int ofs_in_node;
 969         block_t source_blkaddr;
 970
 971         nid = le32_to_cpu(sum->nid);
 972         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
 973
 974         node_page = f2fs_get_node_page(sbi, nid);
 975         if (IS_ERR(node_page))
 976                 return false;
 977
 978         if (f2fs_get_node_info(sbi, nid, dni)) {
 979                 f2fs_put_page(node_page, 1);
 980                 return false;
 981         }
 982
 983         if (sum->version != dni->version) {
 984                 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
 985                           __func__);
 986                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 987         }
 988
 989         *nofs = ofs_of_node(node_page);
 990         source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
 991         f2fs_put_page(node_page, 1);
 992
 993         if (source_blkaddr != blkaddr) {
 994 #ifdef CONFIG_F2FS_CHECK_FS
 995                 unsigned int segno = GET_SEGNO(sbi, blkaddr);
 996                 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 997
 998                 if (unlikely(check_valid_map(sbi, segno, offset))) {
 999                         if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1000                                 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u\n",
1001                                                 blkaddr, source_blkaddr, segno);
1002                                 f2fs_bug_on(sbi, 1);
1003                         }
1004                 }
1005 #endif
1006                 return false;
1007         }
1008         return true;
1009 }
1010
1011 static int ra_data_block(struct inode *inode, pgoff_t index)
1012 {
1013         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1014         struct address_space *mapping = inode->i_mapping;
1015         struct dnode_of_data dn;
1016         struct page *page;
1017         struct extent_info ei = {0, 0, 0};
1018         struct f2fs_io_info fio = {
1019                 .sbi = sbi,
1020                 .ino = inode->i_ino,
1021                 .type = DATA,
1022                 .temp = COLD,
1023                 .op = REQ_OP_READ,
1024                 .op_flags = 0,
1025                 .encrypted_page = NULL,
1026                 .in_list = false,
1027                 .retry = false,
1028         };
1029         int err;
1030
1031         page = f2fs_grab_cache_page(mapping, index, true);
1032         if (!page)
1033                 return -ENOMEM;
1034
1035         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1036                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1037                 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1038                                                 DATA_GENERIC_ENHANCE_READ))) {
1039                         err = -EFSCORRUPTED;
1040                         goto put_page;
1041                 }
1042                 goto got_it;
1043         }
1044
1045         set_new_dnode(&dn, inode, NULL, NULL, 0);
1046         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1047         if (err)
1048                 goto put_page;
1049         f2fs_put_dnode(&dn);
1050
1051         if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1052                 err = -ENOENT;
1053                 goto put_page;
1054         }
1055         if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1056                                                 DATA_GENERIC_ENHANCE))) {
1057                 err = -EFSCORRUPTED;
1058                 goto put_page;
1059         }
1060 got_it:
1061         /* read page */
1062         fio.page = page;
1063         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1064
1065         /*
1066          * don't cache encrypted data into meta inode until previous dirty
1067          * data were writebacked to avoid racing between GC and flush.
1068          */
1069         f2fs_wait_on_page_writeback(page, DATA, true, true);
1070
1071         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1072
1073         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1074                                         dn.data_blkaddr,
1075                                         FGP_LOCK | FGP_CREAT, GFP_NOFS);
1076         if (!fio.encrypted_page) {
1077                 err = -ENOMEM;
1078                 goto put_page;
1079         }
1080
1081         err = f2fs_submit_page_bio(&fio);
1082         if (err)
1083                 goto put_encrypted_page;
1084         f2fs_put_page(fio.encrypted_page, 0);
1085         f2fs_put_page(page, 1);
1086
1087         f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1088         f2fs_update_iostat(sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1089
1090         return 0;
1091 put_encrypted_page:
1092         f2fs_put_page(fio.encrypted_page, 1);
1093 put_page:
1094         f2fs_put_page(page, 1);
1095         return err;
1096 }
1097
1098 /*
1099  * Move data block via META_MAPPING while keeping locked data page.
1100  * This can be used to move blocks, aka LBAs, directly on disk.
1101  */
1102 static int move_data_block(struct inode *inode, block_t bidx,
1103                                 int gc_type, unsigned int segno, int off)
1104 {
1105         struct f2fs_io_info fio = {
1106                 .sbi = F2FS_I_SB(inode),
1107                 .ino = inode->i_ino,
1108                 .type = DATA,
1109                 .temp = COLD,
1110                 .op = REQ_OP_READ,
1111                 .op_flags = 0,
1112                 .encrypted_page = NULL,
1113                 .in_list = false,
1114                 .retry = false,
1115         };
1116         struct dnode_of_data dn;
1117         struct f2fs_summary sum;
1118         struct node_info ni;
1119         struct page *page, *mpage;
1120         block_t newaddr;
1121         int err = 0;
1122         bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1123         int type = fio.sbi->am.atgc_enabled ?
1124                                 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1125
1126         /* do not read out */
1127         page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1128         if (!page)
1129                 return -ENOMEM;
1130
1131         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1132                 err = -ENOENT;
1133                 goto out;
1134         }
1135
1136         if (f2fs_is_atomic_file(inode)) {
1137                 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1138                 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1139                 err = -EAGAIN;
1140                 goto out;
1141         }
1142
1143         if (f2fs_is_pinned_file(inode)) {
1144                 f2fs_pin_file_control(inode, true);
1145                 err = -EAGAIN;
1146                 goto out;
1147         }
1148
1149         set_new_dnode(&dn, inode, NULL, NULL, 0);
1150         err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1151         if (err)
1152                 goto out;
1153
1154         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1155                 ClearPageUptodate(page);
1156                 err = -ENOENT;
1157                 goto put_out;
1158         }
1159
1160         /*
1161          * don't cache encrypted data into meta inode until previous dirty
1162          * data were writebacked to avoid racing between GC and flush.
1163          */
1164         f2fs_wait_on_page_writeback(page, DATA, true, true);
1165
1166         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1167
1168         err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1169         if (err)
1170                 goto put_out;
1171
1172         /* read page */
1173         fio.page = page;
1174         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1175
1176         if (lfs_mode)
1177                 down_write(&fio.sbi->io_order_lock);
1178
1179         mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1180                                         fio.old_blkaddr, false);
1181         if (!mpage) {
1182                 err = -ENOMEM;
1183                 goto up_out;
1184         }
1185
1186         fio.encrypted_page = mpage;
1187
1188         /* read source block in mpage */
1189         if (!PageUptodate(mpage)) {
1190                 err = f2fs_submit_page_bio(&fio);
1191                 if (err) {
1192                         f2fs_put_page(mpage, 1);
1193                         goto up_out;
1194                 }
1195
1196                 f2fs_update_iostat(fio.sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1197                 f2fs_update_iostat(fio.sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1198
1199                 lock_page(mpage);
1200                 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1201                                                 !PageUptodate(mpage))) {
1202                         err = -EIO;
1203                         f2fs_put_page(mpage, 1);
1204                         goto up_out;
1205                 }
1206         }
1207
1208         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1209
1210         /* allocate block address */
1211         f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1212                                 &sum, type, NULL);
1213
1214         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1215                                 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1216         if (!fio.encrypted_page) {
1217                 err = -ENOMEM;
1218                 f2fs_put_page(mpage, 1);
1219                 goto recover_block;
1220         }
1221
1222         /* write target block */
1223         f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1224         memcpy(page_address(fio.encrypted_page),
1225                                 page_address(mpage), PAGE_SIZE);
1226         f2fs_put_page(mpage, 1);
1227         invalidate_mapping_pages(META_MAPPING(fio.sbi),
1228                                 fio.old_blkaddr, fio.old_blkaddr);
1229
1230         set_page_dirty(fio.encrypted_page);
1231         if (clear_page_dirty_for_io(fio.encrypted_page))
1232                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
1233
1234         set_page_writeback(fio.encrypted_page);
1235         ClearPageError(page);
1236
1237         fio.op = REQ_OP_WRITE;
1238         fio.op_flags = REQ_SYNC;
1239         fio.new_blkaddr = newaddr;
1240         f2fs_submit_page_write(&fio);
1241         if (fio.retry) {
1242                 err = -EAGAIN;
1243                 if (PageWriteback(fio.encrypted_page))
1244                         end_page_writeback(fio.encrypted_page);
1245                 goto put_page_out;
1246         }
1247
1248         f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
1249
1250         f2fs_update_data_blkaddr(&dn, newaddr);
1251         set_inode_flag(inode, FI_APPEND_WRITE);
1252         if (page->index == 0)
1253                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1254 put_page_out:
1255         f2fs_put_page(fio.encrypted_page, 1);
1256 recover_block:
1257         if (err)
1258                 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1259                                                         true, true, true);
1260 up_out:
1261         if (lfs_mode)
1262                 up_write(&fio.sbi->io_order_lock);
1263 put_out:
1264         f2fs_put_dnode(&dn);
1265 out:
1266         f2fs_put_page(page, 1);
1267         return err;
1268 }
1269
1270 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1271                                                         unsigned int segno, int off)
1272 {
1273         struct page *page;
1274         int err = 0;
1275
1276         page = f2fs_get_lock_data_page(inode, bidx, true);
1277         if (IS_ERR(page))
1278                 return PTR_ERR(page);
1279
1280         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1281                 err = -ENOENT;
1282                 goto out;
1283         }
1284
1285         if (f2fs_is_atomic_file(inode)) {
1286                 F2FS_I(inode)->i_gc_failures[GC_FAILURE_ATOMIC]++;
1287                 F2FS_I_SB(inode)->skipped_atomic_files[gc_type]++;
1288                 err = -EAGAIN;
1289                 goto out;
1290         }
1291         if (f2fs_is_pinned_file(inode)) {
1292                 if (gc_type == FG_GC)
1293                         f2fs_pin_file_control(inode, true);
1294                 err = -EAGAIN;
1295                 goto out;
1296         }
1297
1298         if (gc_type == BG_GC) {
1299                 if (PageWriteback(page)) {
1300                         err = -EAGAIN;
1301                         goto out;
1302                 }
1303                 set_page_dirty(page);
1304                 set_cold_data(page);
1305         } else {
1306                 struct f2fs_io_info fio = {
1307                         .sbi = F2FS_I_SB(inode),
1308                         .ino = inode->i_ino,
1309                         .type = DATA,
1310                         .temp = COLD,
1311                         .op = REQ_OP_WRITE,
1312                         .op_flags = REQ_SYNC,
1313                         .old_blkaddr = NULL_ADDR,
1314                         .page = page,
1315                         .encrypted_page = NULL,
1316                         .need_lock = LOCK_REQ,
1317                         .io_type = FS_GC_DATA_IO,
1318                 };
1319                 bool is_dirty = PageDirty(page);
1320
1321 retry:
1322                 f2fs_wait_on_page_writeback(page, DATA, true, true);
1323
1324                 set_page_dirty(page);
1325                 if (clear_page_dirty_for_io(page)) {
1326                         inode_dec_dirty_pages(inode);
1327                         f2fs_remove_dirty_inode(inode);
1328                 }
1329
1330                 set_cold_data(page);
1331
1332                 err = f2fs_do_write_data_page(&fio);
1333                 if (err) {
1334                         clear_cold_data(page);
1335                         if (err == -ENOMEM) {
1336                                 congestion_wait(BLK_RW_ASYNC,
1337                                                 DEFAULT_IO_TIMEOUT);
1338                                 goto retry;
1339                         }
1340                         if (is_dirty)
1341                                 set_page_dirty(page);
1342                 }
1343         }
1344 out:
1345         f2fs_put_page(page, 1);
1346         return err;
1347 }
1348
1349 /*
1350  * This function tries to get parent node of victim data block, and identifies
1351  * data block validity. If the block is valid, copy that with cold status and
1352  * modify parent node.
1353  * If the parent node is not valid or the data block address is different,
1354  * the victim data block is ignored.
1355  */
1356 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1357                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1358                 bool force_migrate)
1359 {
1360         struct super_block *sb = sbi->sb;
1361         struct f2fs_summary *entry;
1362         block_t start_addr;
1363         int off;
1364         int phase = 0;
1365         int submitted = 0;
1366         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1367
1368         start_addr = START_BLOCK(sbi, segno);
1369
1370 next_step:
1371         entry = sum;
1372
1373         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1374                 struct page *data_page;
1375                 struct inode *inode;
1376                 struct node_info dni; /* dnode info for the data */
1377                 unsigned int ofs_in_node, nofs;
1378                 block_t start_bidx;
1379                 nid_t nid = le32_to_cpu(entry->nid);
1380
1381                 /*
1382                  * stop BG_GC if there is not enough free sections.
1383                  * Or, stop GC if the segment becomes fully valid caused by
1384                  * race condition along with SSR block allocation.
1385                  */
1386                 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1387                         (!force_migrate && get_valid_blocks(sbi, segno, true) ==
1388                                                         BLKS_PER_SEC(sbi)))
1389                         return submitted;
1390
1391                 if (check_valid_map(sbi, segno, off) == 0)
1392                         continue;
1393
1394                 if (phase == 0) {
1395                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1396                                                         META_NAT, true);
1397                         continue;
1398                 }
1399
1400                 if (phase == 1) {
1401                         f2fs_ra_node_page(sbi, nid);
1402                         continue;
1403                 }
1404
1405                 /* Get an inode by ino with checking validity */
1406                 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1407                         continue;
1408
1409                 if (phase == 2) {
1410                         f2fs_ra_node_page(sbi, dni.ino);
1411                         continue;
1412                 }
1413
1414                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1415
1416                 if (phase == 3) {
1417                         inode = f2fs_iget(sb, dni.ino);
1418                         if (IS_ERR(inode) || is_bad_inode(inode)) {
1419                                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1420                                 continue;
1421                         }
1422
1423                         if (!down_write_trylock(
1424                                 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1425                                 iput(inode);
1426                                 sbi->skipped_gc_rwsem++;
1427                                 continue;
1428                         }
1429
1430                         start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1431                                                                 ofs_in_node;
1432
1433                         if (f2fs_post_read_required(inode)) {
1434                                 int err = ra_data_block(inode, start_bidx);
1435
1436                                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1437                                 if (err) {
1438                                         iput(inode);
1439                                         continue;
1440                                 }
1441                                 add_gc_inode(gc_list, inode);
1442                                 continue;
1443                         }
1444
1445                         data_page = f2fs_get_read_data_page(inode,
1446                                                 start_bidx, REQ_RAHEAD, true);
1447                         up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1448                         if (IS_ERR(data_page)) {
1449                                 iput(inode);
1450                                 continue;
1451                         }
1452
1453                         f2fs_put_page(data_page, 0);
1454                         add_gc_inode(gc_list, inode);
1455                         continue;
1456                 }
1457
1458                 /* phase 4 */
1459                 inode = find_gc_inode(gc_list, dni.ino);
1460                 if (inode) {
1461                         struct f2fs_inode_info *fi = F2FS_I(inode);
1462                         bool locked = false;
1463                         int err;
1464
1465                         if (S_ISREG(inode->i_mode)) {
1466                                 if (!down_write_trylock(&fi->i_gc_rwsem[READ]))
1467                                         continue;
1468                                 if (!down_write_trylock(
1469                                                 &fi->i_gc_rwsem[WRITE])) {
1470                                         sbi->skipped_gc_rwsem++;
1471                                         up_write(&fi->i_gc_rwsem[READ]);
1472                                         continue;
1473                                 }
1474                                 locked = true;
1475
1476                                 /* wait for all inflight aio data */
1477                                 inode_dio_wait(inode);
1478                         }
1479
1480                         start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1481                                                                 + ofs_in_node;
1482                         if (f2fs_post_read_required(inode))
1483                                 err = move_data_block(inode, start_bidx,
1484                                                         gc_type, segno, off);
1485                         else
1486                                 err = move_data_page(inode, start_bidx, gc_type,
1487                                                                 segno, off);
1488
1489                         if (!err && (gc_type == FG_GC ||
1490                                         f2fs_post_read_required(inode)))
1491                                 submitted++;
1492
1493                         if (locked) {
1494                                 up_write(&fi->i_gc_rwsem[WRITE]);
1495                                 up_write(&fi->i_gc_rwsem[READ]);
1496                         }
1497
1498                         stat_inc_data_blk_count(sbi, 1, gc_type);
1499                 }
1500         }
1501
1502         if (++phase < 5)
1503                 goto next_step;
1504
1505         return submitted;
1506 }
1507
1508 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1509                         int gc_type)
1510 {
1511         struct sit_info *sit_i = SIT_I(sbi);
1512         int ret;
1513
1514         down_write(&sit_i->sentry_lock);
1515         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1516                                               NO_CHECK_TYPE, LFS, 0);
1517         up_write(&sit_i->sentry_lock);
1518         return ret;
1519 }
1520
1521 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1522                                 unsigned int start_segno,
1523                                 struct gc_inode_list *gc_list, int gc_type,
1524                                 bool force_migrate)
1525 {
1526         struct page *sum_page;
1527         struct f2fs_summary_block *sum;
1528         struct blk_plug plug;
1529         unsigned int segno = start_segno;
1530         unsigned int end_segno = start_segno + sbi->segs_per_sec;
1531         int seg_freed = 0, migrated = 0;
1532         unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1533                                                 SUM_TYPE_DATA : SUM_TYPE_NODE;
1534         int submitted = 0;
1535
1536         if (__is_large_section(sbi))
1537                 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1538
1539         /*
1540          * zone-capacity can be less than zone-size in zoned devices,
1541          * resulting in less than expected usable segments in the zone,
1542          * calculate the end segno in the zone which can be garbage collected
1543          */
1544         if (f2fs_sb_has_blkzoned(sbi))
1545                 end_segno -= sbi->segs_per_sec -
1546                                         f2fs_usable_segs_in_sec(sbi, segno);
1547
1548         sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1549
1550         /* readahead multi ssa blocks those have contiguous address */
1551         if (__is_large_section(sbi))
1552                 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1553                                         end_segno - segno, META_SSA, true);
1554
1555         /* reference all summary page */
1556         while (segno < end_segno) {
1557                 sum_page = f2fs_get_sum_page(sbi, segno++);
1558                 if (IS_ERR(sum_page)) {
1559                         int err = PTR_ERR(sum_page);
1560
1561                         end_segno = segno - 1;
1562                         for (segno = start_segno; segno < end_segno; segno++) {
1563                                 sum_page = find_get_page(META_MAPPING(sbi),
1564                                                 GET_SUM_BLOCK(sbi, segno));
1565                                 f2fs_put_page(sum_page, 0);
1566                                 f2fs_put_page(sum_page, 0);
1567                         }
1568                         return err;
1569                 }
1570                 unlock_page(sum_page);
1571         }
1572
1573         blk_start_plug(&plug);
1574
1575         for (segno = start_segno; segno < end_segno; segno++) {
1576
1577                 /* find segment summary of victim */
1578                 sum_page = find_get_page(META_MAPPING(sbi),
1579                                         GET_SUM_BLOCK(sbi, segno));
1580                 f2fs_put_page(sum_page, 0);
1581
1582                 if (get_valid_blocks(sbi, segno, false) == 0)
1583                         goto freed;
1584                 if (gc_type == BG_GC && __is_large_section(sbi) &&
1585                                 migrated >= sbi->migration_granularity)
1586                         goto skip;
1587                 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1588                         goto skip;
1589
1590                 sum = page_address(sum_page);
1591                 if (type != GET_SUM_TYPE((&sum->footer))) {
1592                         f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1593                                  segno, type, GET_SUM_TYPE((&sum->footer)));
1594                         set_sbi_flag(sbi, SBI_NEED_FSCK);
1595                         f2fs_stop_checkpoint(sbi, false);
1596                         goto skip;
1597                 }
1598
1599                 /*
1600                  * this is to avoid deadlock:
1601                  * - lock_page(sum_page)         - f2fs_replace_block
1602                  *  - check_valid_map()            - down_write(sentry_lock)
1603                  *   - down_read(sentry_lock)     - change_curseg()
1604                  *                                  - lock_page(sum_page)
1605                  */
1606                 if (type == SUM_TYPE_NODE)
1607                         submitted += gc_node_segment(sbi, sum->entries, segno,
1608                                                                 gc_type);
1609                 else
1610                         submitted += gc_data_segment(sbi, sum->entries, gc_list,
1611                                                         segno, gc_type,
1612                                                         force_migrate);
1613
1614                 stat_inc_seg_count(sbi, type, gc_type);
1615                 migrated++;
1616
1617 freed:
1618                 if (gc_type == FG_GC &&
1619                                 get_valid_blocks(sbi, segno, false) == 0)
1620                         seg_freed++;
1621
1622                 if (__is_large_section(sbi) && segno + 1 < end_segno)
1623                         sbi->next_victim_seg[gc_type] = segno + 1;
1624 skip:
1625                 f2fs_put_page(sum_page, 0);
1626         }
1627
1628         if (submitted)
1629                 f2fs_submit_merged_write(sbi,
1630                                 (type == SUM_TYPE_NODE) ? NODE : DATA);
1631
1632         blk_finish_plug(&plug);
1633
1634         stat_inc_call_count(sbi->stat_info);
1635
1636         return seg_freed;
1637 }
1638
1639 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
1640                         bool background, bool force, unsigned int segno)
1641 {
1642         int gc_type = sync ? FG_GC : BG_GC;
1643         int sec_freed = 0, seg_freed = 0, total_freed = 0;
1644         int ret = 0;
1645         struct cp_control cpc;
1646         unsigned int init_segno = segno;
1647         struct gc_inode_list gc_list = {
1648                 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1649                 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1650         };
1651         unsigned long long last_skipped = sbi->skipped_atomic_files[FG_GC];
1652         unsigned long long first_skipped;
1653         unsigned int skipped_round = 0, round = 0;
1654
1655         trace_f2fs_gc_begin(sbi->sb, sync, background,
1656                                 get_pages(sbi, F2FS_DIRTY_NODES),
1657                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1658                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1659                                 free_sections(sbi),
1660                                 free_segments(sbi),
1661                                 reserved_segments(sbi),
1662                                 prefree_segments(sbi));
1663
1664         cpc.reason = __get_cp_reason(sbi);
1665         sbi->skipped_gc_rwsem = 0;
1666         first_skipped = last_skipped;
1667 gc_more:
1668         if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1669                 ret = -EINVAL;
1670                 goto stop;
1671         }
1672         if (unlikely(f2fs_cp_error(sbi))) {
1673                 ret = -EIO;
1674                 goto stop;
1675         }
1676
1677         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1678                 /*
1679                  * For example, if there are many prefree_segments below given
1680                  * threshold, we can make them free by checkpoint. Then, we
1681                  * secure free segments which doesn't need fggc any more.
1682                  */
1683                 if (prefree_segments(sbi) &&
1684                                 !is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
1685                         ret = f2fs_write_checkpoint(sbi, &cpc);
1686                         if (ret)
1687                                 goto stop;
1688                 }
1689                 if (has_not_enough_free_secs(sbi, 0, 0))
1690                         gc_type = FG_GC;
1691         }
1692
1693         /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1694         if (gc_type == BG_GC && !background) {
1695                 ret = -EINVAL;
1696                 goto stop;
1697         }
1698         ret = __get_victim(sbi, &segno, gc_type);
1699         if (ret)
1700                 goto stop;
1701
1702         seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type, force);
1703         if (gc_type == FG_GC &&
1704                 seg_freed == f2fs_usable_segs_in_sec(sbi, segno))
1705                 sec_freed++;
1706         total_freed += seg_freed;
1707
1708         if (gc_type == FG_GC) {
1709                 if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
1710                                                 sbi->skipped_gc_rwsem)
1711                         skipped_round++;
1712                 last_skipped = sbi->skipped_atomic_files[FG_GC];
1713                 round++;
1714         }
1715
1716         if (gc_type == FG_GC && seg_freed)
1717                 sbi->cur_victim_sec = NULL_SEGNO;
1718
1719         if (sync)
1720                 goto stop;
1721
1722         if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
1723                 if (skipped_round <= MAX_SKIP_GC_COUNT ||
1724                                         skipped_round * 2 < round) {
1725                         segno = NULL_SEGNO;
1726                         goto gc_more;
1727                 }
1728
1729                 if (first_skipped < last_skipped &&
1730                                 (last_skipped - first_skipped) >
1731                                                 sbi->skipped_gc_rwsem) {
1732                         f2fs_drop_inmem_pages_all(sbi, true);
1733                         segno = NULL_SEGNO;
1734                         goto gc_more;
1735                 }
1736                 if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1737                         ret = f2fs_write_checkpoint(sbi, &cpc);
1738         }
1739 stop:
1740         SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1741         SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
1742
1743         trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1744                                 get_pages(sbi, F2FS_DIRTY_NODES),
1745                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1746                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1747                                 free_sections(sbi),
1748                                 free_segments(sbi),
1749                                 reserved_segments(sbi),
1750                                 prefree_segments(sbi));
1751
1752         up_write(&sbi->gc_lock);
1753
1754         put_gc_inode(&gc_list);
1755
1756         if (sync && !ret)
1757                 ret = sec_freed ? 0 : -EAGAIN;
1758         return ret;
1759 }
1760
1761 int __init f2fs_create_garbage_collection_cache(void)
1762 {
1763         victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1764                                         sizeof(struct victim_entry));
1765         if (!victim_entry_slab)
1766                 return -ENOMEM;
1767         return 0;
1768 }
1769
1770 void f2fs_destroy_garbage_collection_cache(void)
1771 {
1772         kmem_cache_destroy(victim_entry_slab);
1773 }
1774
1775 static void init_atgc_management(struct f2fs_sb_info *sbi)
1776 {
1777         struct atgc_management *am = &sbi->am;
1778
1779         if (test_opt(sbi, ATGC) &&
1780                 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1781                 am->atgc_enabled = true;
1782
1783         am->root = RB_ROOT_CACHED;
1784         INIT_LIST_HEAD(&am->victim_list);
1785         am->victim_count = 0;
1786
1787         am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1788         am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1789         am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1790 }
1791
1792 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1793 {
1794         DIRTY_I(sbi)->v_ops = &default_v_ops;
1795
1796         sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1797
1798         /* give warm/cold data area from slower device */
1799         if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1800                 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1801                                 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1802
1803         init_atgc_management(sbi);
1804 }
1805
1806 static int free_segment_range(struct f2fs_sb_info *sbi,
1807                                 unsigned int secs, bool gc_only)
1808 {
1809         unsigned int segno, next_inuse, start, end;
1810         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1811         int gc_mode, gc_type;
1812         int err = 0;
1813         int type;
1814
1815         /* Force block allocation for GC */
1816         MAIN_SECS(sbi) -= secs;
1817         start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1818         end = MAIN_SEGS(sbi) - 1;
1819
1820         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1821         for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1822                 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1823                         SIT_I(sbi)->last_victim[gc_mode] = 0;
1824
1825         for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1826                 if (sbi->next_victim_seg[gc_type] >= start)
1827                         sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1828         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1829
1830         /* Move out cursegs from the target range */
1831         for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1832                 f2fs_allocate_segment_for_resize(sbi, type, start, end);
1833
1834         /* do GC to move out valid blocks in the range */
1835         for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1836                 struct gc_inode_list gc_list = {
1837                         .ilist = LIST_HEAD_INIT(gc_list.ilist),
1838                         .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1839                 };
1840
1841                 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
1842                 put_gc_inode(&gc_list);
1843
1844                 if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1845                         err = -EAGAIN;
1846                         goto out;
1847                 }
1848                 if (fatal_signal_pending(current)) {
1849                         err = -ERESTARTSYS;
1850                         goto out;
1851                 }
1852         }
1853         if (gc_only)
1854                 goto out;
1855
1856         err = f2fs_write_checkpoint(sbi, &cpc);
1857         if (err)
1858                 goto out;
1859
1860         next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1861         if (next_inuse <= end) {
1862                 f2fs_err(sbi, "segno %u should be free but still inuse!",
1863                          next_inuse);
1864                 f2fs_bug_on(sbi, 1);
1865         }
1866 out:
1867         MAIN_SECS(sbi) += secs;
1868         return err;
1869 }
1870
1871 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1872 {
1873         struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1874         int section_count;
1875         int segment_count;
1876         int segment_count_main;
1877         long long block_count;
1878         int segs = secs * sbi->segs_per_sec;
1879
1880         down_write(&sbi->sb_lock);
1881
1882         section_count = le32_to_cpu(raw_sb->section_count);
1883         segment_count = le32_to_cpu(raw_sb->segment_count);
1884         segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
1885         block_count = le64_to_cpu(raw_sb->block_count);
1886
1887         raw_sb->section_count = cpu_to_le32(section_count + secs);
1888         raw_sb->segment_count = cpu_to_le32(segment_count + segs);
1889         raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
1890         raw_sb->block_count = cpu_to_le64(block_count +
1891                                         (long long)segs * sbi->blocks_per_seg);
1892         if (f2fs_is_multi_device(sbi)) {
1893                 int last_dev = sbi->s_ndevs - 1;
1894                 int dev_segs =
1895                         le32_to_cpu(raw_sb->devs[last_dev].total_segments);
1896
1897                 raw_sb->devs[last_dev].total_segments =
1898                                                 cpu_to_le32(dev_segs + segs);
1899         }
1900
1901         up_write(&sbi->sb_lock);
1902 }
1903
1904 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
1905 {
1906         int segs = secs * sbi->segs_per_sec;
1907         long long blks = (long long)segs * sbi->blocks_per_seg;
1908         long long user_block_count =
1909                                 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
1910
1911         SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
1912         MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
1913         MAIN_SECS(sbi) += secs;
1914         FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
1915         FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
1916         F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
1917
1918         if (f2fs_is_multi_device(sbi)) {
1919                 int last_dev = sbi->s_ndevs - 1;
1920
1921                 FDEV(last_dev).total_segments =
1922                                 (int)FDEV(last_dev).total_segments + segs;
1923                 FDEV(last_dev).end_blk =
1924                                 (long long)FDEV(last_dev).end_blk + blks;
1925 #ifdef CONFIG_BLK_DEV_ZONED
1926                 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
1927                                         (int)(blks >> sbi->log_blocks_per_blkz);
1928 #endif
1929         }
1930 }
1931
1932 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
1933 {
1934         __u64 old_block_count, shrunk_blocks;
1935         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1936         unsigned int secs;
1937         int err = 0;
1938         __u32 rem;
1939
1940         old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
1941         if (block_count > old_block_count)
1942                 return -EINVAL;
1943
1944         if (f2fs_is_multi_device(sbi)) {
1945                 int last_dev = sbi->s_ndevs - 1;
1946                 __u64 last_segs = FDEV(last_dev).total_segments;
1947
1948                 if (block_count + last_segs * sbi->blocks_per_seg <=
1949                                                                 old_block_count)
1950                         return -EINVAL;
1951         }
1952
1953         /* new fs size should align to section size */
1954         div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
1955         if (rem)
1956                 return -EINVAL;
1957
1958         if (block_count == old_block_count)
1959                 return 0;
1960
1961         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1962                 f2fs_err(sbi, "Should run fsck to repair first.");
1963                 return -EFSCORRUPTED;
1964         }
1965
1966         if (test_opt(sbi, DISABLE_CHECKPOINT)) {
1967                 f2fs_err(sbi, "Checkpoint should be enabled.");
1968                 return -EINVAL;
1969         }
1970
1971         shrunk_blocks = old_block_count - block_count;
1972         secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
1973
1974         /* stop other GC */
1975         if (!down_write_trylock(&sbi->gc_lock))
1976                 return -EAGAIN;
1977
1978         /* stop CP to protect MAIN_SEC in free_segment_range */
1979         f2fs_lock_op(sbi);
1980         err = free_segment_range(sbi, secs, true);
1981         f2fs_unlock_op(sbi);
1982         up_write(&sbi->gc_lock);
1983         if (err)
1984                 return err;
1985
1986         set_sbi_flag(sbi, SBI_IS_RESIZEFS);
1987
1988         freeze_super(sbi->sb);
1989         down_write(&sbi->gc_lock);
1990         down_write(&sbi->cp_global_sem);
1991
1992         spin_lock(&sbi->stat_lock);
1993         if (shrunk_blocks + valid_user_blocks(sbi) +
1994                 sbi->current_reserved_blocks + sbi->unusable_block_count +
1995                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
1996                 err = -ENOSPC;
1997         else
1998                 sbi->user_block_count -= shrunk_blocks;
1999         spin_unlock(&sbi->stat_lock);
2000         if (err)
2001                 goto out_err;
2002
2003         err = free_segment_range(sbi, secs, false);
2004         if (err)
2005                 goto recover_out;
2006
2007         update_sb_metadata(sbi, -secs);
2008
2009         err = f2fs_commit_super(sbi, false);
2010         if (err) {
2011                 update_sb_metadata(sbi, secs);
2012                 goto recover_out;
2013         }
2014
2015         update_fs_metadata(sbi, -secs);
2016         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2017         set_sbi_flag(sbi, SBI_IS_DIRTY);
2018
2019         err = f2fs_write_checkpoint(sbi, &cpc);
2020         if (err) {
2021                 update_fs_metadata(sbi, secs);
2022                 update_sb_metadata(sbi, secs);
2023                 f2fs_commit_super(sbi, false);
2024         }
2025 recover_out:
2026         if (err) {
2027                 set_sbi_flag(sbi, SBI_NEED_FSCK);
2028                 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2029
2030                 spin_lock(&sbi->stat_lock);
2031                 sbi->user_block_count += shrunk_blocks;
2032                 spin_unlock(&sbi->stat_lock);
2033         }
2034 out_err:
2035         up_write(&sbi->cp_global_sem);
2036         up_write(&sbi->gc_lock);
2037         thaw_super(sbi->sb);
2038         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2039         return err;
2040 }