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/init.h>
  11 #include <linux/f2fs_fs.h>
  12 #include <linux/kthread.h>
  13 #include <linux/delay.h>
  14 #include <linux/freezer.h>
  15 #include <linux/sched/signal.h>
  16 #include <linux/random.h>
  17 #include <linux/sched/mm.h>
  18
  19 #include "f2fs.h"
  20 #include "node.h"
  21 #include "segment.h"
  22 #include "gc.h"
  23 #include "iostat.h"
  24 #include <trace/events/f2fs.h>
  25
  26 static struct kmem_cache *victim_entry_slab;
  27
  28 static unsigned int count_bits(const unsigned long *addr,
  29                                 unsigned int offset, unsigned int len);
  30
  31 static int gc_thread_func(void *data)
  32 {
  33         struct f2fs_sb_info *sbi = data;
  34         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
  35         wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
  36         wait_queue_head_t *fggc_wq = &sbi->gc_thread->fggc_wq;
  37         unsigned int wait_ms;
  38         struct f2fs_gc_control gc_control = {
  39                 .victim_segno = NULL_SEGNO,
  40                 .should_migrate_blocks = false,
  41                 .err_gc_skipped = false };
  42
  43         wait_ms = gc_th->min_sleep_time;
  44
  45         set_freezable();
  46         do {
  47                 bool sync_mode, foreground = false;
  48
  49                 wait_event_interruptible_timeout(*wq,
  50                                 kthread_should_stop() || freezing(current) ||
  51                                 waitqueue_active(fggc_wq) ||
  52                                 gc_th->gc_wake,
  53                                 msecs_to_jiffies(wait_ms));
  54
  55                 if (test_opt(sbi, GC_MERGE) && waitqueue_active(fggc_wq))
  56                         foreground = true;
  57
  58                 /* give it a try one time */
  59                 if (gc_th->gc_wake)
  60                         gc_th->gc_wake = 0;
  61
  62                 if (try_to_freeze()) {
  63                         stat_other_skip_bggc_count(sbi);
  64                         continue;
  65                 }
  66                 if (kthread_should_stop())
  67                         break;
  68
  69                 if (sbi->sb->s_writers.frozen >= SB_FREEZE_WRITE) {
  70                         increase_sleep_time(gc_th, &wait_ms);
  71                         stat_other_skip_bggc_count(sbi);
  72                         continue;
  73                 }
  74
  75                 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
  76                         f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
  77                         f2fs_stop_checkpoint(sbi, false,
  78                                         STOP_CP_REASON_FAULT_INJECT);
  79                 }
  80
  81                 if (!sb_start_write_trylock(sbi->sb)) {
  82                         stat_other_skip_bggc_count(sbi);
  83                         continue;
  84                 }
  85
  86                 /*
  87                  * [GC triggering condition]
  88                  * 0. GC is not conducted currently.
  89                  * 1. There are enough dirty segments.
  90                  * 2. IO subsystem is idle by checking the # of writeback pages.
  91                  * 3. IO subsystem is idle by checking the # of requests in
  92                  *    bdev's request list.
  93                  *
  94                  * Note) We have to avoid triggering GCs frequently.
  95                  * Because it is possible that some segments can be
  96                  * invalidated soon after by user update or deletion.
  97                  * So, I'd like to wait some time to collect dirty segments.
  98                  */
  99                 if (sbi->gc_mode == GC_URGENT_HIGH ||
 100                                 sbi->gc_mode == GC_URGENT_MID) {
 101                         wait_ms = gc_th->urgent_sleep_time;
 102                         f2fs_down_write(&sbi->gc_lock);
 103                         goto do_gc;
 104                 }
 105
 106                 if (foreground) {
 107                         f2fs_down_write(&sbi->gc_lock);
 108                         goto do_gc;
 109                 } else if (!f2fs_down_write_trylock(&sbi->gc_lock)) {
 110                         stat_other_skip_bggc_count(sbi);
 111                         goto next;
 112                 }
 113
 114                 if (!is_idle(sbi, GC_TIME)) {
 115                         increase_sleep_time(gc_th, &wait_ms);
 116                         f2fs_up_write(&sbi->gc_lock);
 117                         stat_io_skip_bggc_count(sbi);
 118                         goto next;
 119                 }
 120
 121                 if (has_enough_invalid_blocks(sbi))
 122                         decrease_sleep_time(gc_th, &wait_ms);
 123                 else
 124                         increase_sleep_time(gc_th, &wait_ms);
 125 do_gc:
 126                 if (!foreground)
 127                         stat_inc_bggc_count(sbi->stat_info);
 128
 129                 sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
 130
 131                 /* foreground GC was been triggered via f2fs_balance_fs() */
 132                 if (foreground)
 133                         sync_mode = false;
 134
 135                 gc_control.init_gc_type = sync_mode ? FG_GC : BG_GC;
 136                 gc_control.no_bg_gc = foreground;
 137                 gc_control.nr_free_secs = foreground ? 1 : 0;
 138
 139                 /* if return value is not zero, no victim was selected */
 140                 if (f2fs_gc(sbi, &gc_control)) {
 141                         /* don't bother wait_ms by foreground gc */
 142                         if (!foreground)
 143                                 wait_ms = gc_th->no_gc_sleep_time;
 144                 }
 145
 146                 if (foreground)
 147                         wake_up_all(&gc_th->fggc_wq);
 148
 149                 trace_f2fs_background_gc(sbi->sb, wait_ms,
 150                                 prefree_segments(sbi), free_segments(sbi));
 151
 152                 /* balancing f2fs's metadata periodically */
 153                 f2fs_balance_fs_bg(sbi, true);
 154 next:
 155                 if (sbi->gc_mode == GC_URGENT_HIGH) {
 156                         spin_lock(&sbi->gc_urgent_high_lock);
 157                         if (sbi->gc_urgent_high_remaining) {
 158                                 sbi->gc_urgent_high_remaining--;
 159                                 if (!sbi->gc_urgent_high_remaining)
 160                                         sbi->gc_mode = GC_NORMAL;
 161                         }
 162                         spin_unlock(&sbi->gc_urgent_high_lock);
 163                 }
 164                 sb_end_write(sbi->sb);
 165
 166         } while (!kthread_should_stop());
 167         return 0;
 168 }
 169
 170 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
 171 {
 172         struct f2fs_gc_kthread *gc_th;
 173         dev_t dev = sbi->sb->s_bdev->bd_dev;
 174         int err = 0;
 175
 176         gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 177         if (!gc_th) {
 178                 err = -ENOMEM;
 179                 goto out;
 180         }
 181
 182         gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
 183         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 184         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 185         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 186
 187         gc_th->gc_wake = 0;
 188
 189         sbi->gc_thread = gc_th;
 190         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 191         init_waitqueue_head(&sbi->gc_thread->fggc_wq);
 192         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 193                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 194         if (IS_ERR(gc_th->f2fs_gc_task)) {
 195                 err = PTR_ERR(gc_th->f2fs_gc_task);
 196                 kfree(gc_th);
 197                 sbi->gc_thread = NULL;
 198         }
 199 out:
 200         return err;
 201 }
 202
 203 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
 204 {
 205         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 206
 207         if (!gc_th)
 208                 return;
 209         kthread_stop(gc_th->f2fs_gc_task);
 210         wake_up_all(&gc_th->fggc_wq);
 211         kfree(gc_th);
 212         sbi->gc_thread = NULL;
 213 }
 214
 215 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
 216 {
 217         int gc_mode;
 218
 219         if (gc_type == BG_GC) {
 220                 if (sbi->am.atgc_enabled)
 221                         gc_mode = GC_AT;
 222                 else
 223                         gc_mode = GC_CB;
 224         } else {
 225                 gc_mode = GC_GREEDY;
 226         }
 227
 228         switch (sbi->gc_mode) {
 229         case GC_IDLE_CB:
 230                 gc_mode = GC_CB;
 231                 break;
 232         case GC_IDLE_GREEDY:
 233         case GC_URGENT_HIGH:
 234                 gc_mode = GC_GREEDY;
 235                 break;
 236         case GC_IDLE_AT:
 237                 gc_mode = GC_AT;
 238                 break;
 239         }
 240
 241         return gc_mode;
 242 }
 243
 244 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 245                         int type, struct victim_sel_policy *p)
 246 {
 247         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 248
 249         if (p->alloc_mode == SSR) {
 250                 p->gc_mode = GC_GREEDY;
 251                 p->dirty_bitmap = dirty_i->dirty_segmap[type];
 252                 p->max_search = dirty_i->nr_dirty[type];
 253                 p->ofs_unit = 1;
 254         } else if (p->alloc_mode == AT_SSR) {
 255                 p->gc_mode = GC_GREEDY;
 256                 p->dirty_bitmap = dirty_i->dirty_segmap[type];
 257                 p->max_search = dirty_i->nr_dirty[type];
 258                 p->ofs_unit = 1;
 259         } else {
 260                 p->gc_mode = select_gc_type(sbi, gc_type);
 261                 p->ofs_unit = sbi->segs_per_sec;
 262                 if (__is_large_section(sbi)) {
 263                         p->dirty_bitmap = dirty_i->dirty_secmap;
 264                         p->max_search = count_bits(p->dirty_bitmap,
 265                                                 0, MAIN_SECS(sbi));
 266                 } else {
 267                         p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
 268                         p->max_search = dirty_i->nr_dirty[DIRTY];
 269                 }
 270         }
 271
 272         /*
 273          * adjust candidates range, should select all dirty segments for
 274          * foreground GC and urgent GC cases.
 275          */
 276         if (gc_type != FG_GC &&
 277                         (sbi->gc_mode != GC_URGENT_HIGH) &&
 278                         (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
 279                         p->max_search > sbi->max_victim_search)
 280                 p->max_search = sbi->max_victim_search;
 281
 282         /* let's select beginning hot/small space first in no_heap mode*/
 283         if (f2fs_need_rand_seg(sbi))
 284                 p->offset = prandom_u32_max(MAIN_SECS(sbi) * sbi->segs_per_sec);
 285         else if (test_opt(sbi, NOHEAP) &&
 286                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
 287                 p->offset = 0;
 288         else
 289                 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
 290 }
 291
 292 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 293                                 struct victim_sel_policy *p)
 294 {
 295         /* SSR allocates in a segment unit */
 296         if (p->alloc_mode == SSR)
 297                 return sbi->blocks_per_seg;
 298         else if (p->alloc_mode == AT_SSR)
 299                 return UINT_MAX;
 300
 301         /* LFS */
 302         if (p->gc_mode == GC_GREEDY)
 303                 return 2 * sbi->blocks_per_seg * p->ofs_unit;
 304         else if (p->gc_mode == GC_CB)
 305                 return UINT_MAX;
 306         else if (p->gc_mode == GC_AT)
 307                 return UINT_MAX;
 308         else /* No other gc_mode */
 309                 return 0;
 310 }
 311
 312 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 313 {
 314         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 315         unsigned int secno;
 316
 317         /*
 318          * If the gc_type is FG_GC, we can select victim segments
 319          * selected by background GC before.
 320          * Those segments guarantee they have small valid blocks.
 321          */
 322         for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
 323                 if (sec_usage_check(sbi, secno))
 324                         continue;
 325                 clear_bit(secno, dirty_i->victim_secmap);
 326                 return GET_SEG_FROM_SEC(sbi, secno);
 327         }
 328         return NULL_SEGNO;
 329 }
 330
 331 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 332 {
 333         struct sit_info *sit_i = SIT_I(sbi);
 334         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 335         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 336         unsigned long long mtime = 0;
 337         unsigned int vblocks;
 338         unsigned char age = 0;
 339         unsigned char u;
 340         unsigned int i;
 341         unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
 342
 343         for (i = 0; i < usable_segs_per_sec; i++)
 344                 mtime += get_seg_entry(sbi, start + i)->mtime;
 345         vblocks = get_valid_blocks(sbi, segno, true);
 346
 347         mtime = div_u64(mtime, usable_segs_per_sec);
 348         vblocks = div_u64(vblocks, usable_segs_per_sec);
 349
 350         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 351
 352         /* Handle if the system time has changed by the user */
 353         if (mtime < sit_i->min_mtime)
 354                 sit_i->min_mtime = mtime;
 355         if (mtime > sit_i->max_mtime)
 356                 sit_i->max_mtime = mtime;
 357         if (sit_i->max_mtime != sit_i->min_mtime)
 358                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 359                                 sit_i->max_mtime - sit_i->min_mtime);
 360
 361         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 362 }
 363
 364 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 365                         unsigned int segno, struct victim_sel_policy *p)
 366 {
 367         if (p->alloc_mode == SSR)
 368                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 369
 370         /* alloc_mode == LFS */
 371         if (p->gc_mode == GC_GREEDY)
 372                 return get_valid_blocks(sbi, segno, true);
 373         else if (p->gc_mode == GC_CB)
 374                 return get_cb_cost(sbi, segno);
 375
 376         f2fs_bug_on(sbi, 1);
 377         return 0;
 378 }
 379
 380 static unsigned int count_bits(const unsigned long *addr,
 381                                 unsigned int offset, unsigned int len)
 382 {
 383         unsigned int end = offset + len, sum = 0;
 384
 385         while (offset < end) {
 386                 if (test_bit(offset++, addr))
 387                         ++sum;
 388         }
 389         return sum;
 390 }
 391
 392 static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi,
 393                                 unsigned long long mtime, unsigned int segno,
 394                                 struct rb_node *parent, struct rb_node **p,
 395                                 bool left_most)
 396 {
 397         struct atgc_management *am = &sbi->am;
 398         struct victim_entry *ve;
 399
 400         ve =  f2fs_kmem_cache_alloc(victim_entry_slab,
 401                                 GFP_NOFS, true, NULL);
 402
 403         ve->mtime = mtime;
 404         ve->segno = segno;
 405
 406         rb_link_node(&ve->rb_node, parent, p);
 407         rb_insert_color_cached(&ve->rb_node, &am->root, left_most);
 408
 409         list_add_tail(&ve->list, &am->victim_list);
 410
 411         am->victim_count++;
 412
 413         return ve;
 414 }
 415
 416 static void insert_victim_entry(struct f2fs_sb_info *sbi,
 417                                 unsigned long long mtime, unsigned int segno)
 418 {
 419         struct atgc_management *am = &sbi->am;
 420         struct rb_node **p;
 421         struct rb_node *parent = NULL;
 422         bool left_most = true;
 423
 424         p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most);
 425         attach_victim_entry(sbi, mtime, segno, parent, p, left_most);
 426 }
 427
 428 static void add_victim_entry(struct f2fs_sb_info *sbi,
 429                                 struct victim_sel_policy *p, unsigned int segno)
 430 {
 431         struct sit_info *sit_i = SIT_I(sbi);
 432         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 433         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 434         unsigned long long mtime = 0;
 435         unsigned int i;
 436
 437         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 438                 if (p->gc_mode == GC_AT &&
 439                         get_valid_blocks(sbi, segno, true) == 0)
 440                         return;
 441         }
 442
 443         for (i = 0; i < sbi->segs_per_sec; i++)
 444                 mtime += get_seg_entry(sbi, start + i)->mtime;
 445         mtime = div_u64(mtime, sbi->segs_per_sec);
 446
 447         /* Handle if the system time has changed by the user */
 448         if (mtime < sit_i->min_mtime)
 449                 sit_i->min_mtime = mtime;
 450         if (mtime > sit_i->max_mtime)
 451                 sit_i->max_mtime = mtime;
 452         if (mtime < sit_i->dirty_min_mtime)
 453                 sit_i->dirty_min_mtime = mtime;
 454         if (mtime > sit_i->dirty_max_mtime)
 455                 sit_i->dirty_max_mtime = mtime;
 456
 457         /* don't choose young section as candidate */
 458         if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
 459                 return;
 460
 461         insert_victim_entry(sbi, mtime, segno);
 462 }
 463
 464 static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi,
 465                                                 struct victim_sel_policy *p)
 466 {
 467         struct atgc_management *am = &sbi->am;
 468         struct rb_node *parent = NULL;
 469         bool left_most;
 470
 471         f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most);
 472
 473         return parent;
 474 }
 475
 476 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
 477                                                 struct victim_sel_policy *p)
 478 {
 479         struct sit_info *sit_i = SIT_I(sbi);
 480         struct atgc_management *am = &sbi->am;
 481         struct rb_root_cached *root = &am->root;
 482         struct rb_node *node;
 483         struct rb_entry *re;
 484         struct victim_entry *ve;
 485         unsigned long long total_time;
 486         unsigned long long age, u, accu;
 487         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 488         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 489         unsigned int sec_blocks = CAP_BLKS_PER_SEC(sbi);
 490         unsigned int vblocks;
 491         unsigned int dirty_threshold = max(am->max_candidate_count,
 492                                         am->candidate_ratio *
 493                                         am->victim_count / 100);
 494         unsigned int age_weight = am->age_weight;
 495         unsigned int cost;
 496         unsigned int iter = 0;
 497
 498         if (max_mtime < min_mtime)
 499                 return;
 500
 501         max_mtime += 1;
 502         total_time = max_mtime - min_mtime;
 503
 504         accu = div64_u64(ULLONG_MAX, total_time);
 505         accu = min_t(unsigned long long, div_u64(accu, 100),
 506                                         DEFAULT_ACCURACY_CLASS);
 507
 508         node = rb_first_cached(root);
 509 next:
 510         re = rb_entry_safe(node, struct rb_entry, rb_node);
 511         if (!re)
 512                 return;
 513
 514         ve = (struct victim_entry *)re;
 515
 516         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 517                 goto skip;
 518
 519         /* age = 10000 * x% * 60 */
 520         age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
 521                                                                 age_weight;
 522
 523         vblocks = get_valid_blocks(sbi, ve->segno, true);
 524         f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
 525
 526         /* u = 10000 * x% * 40 */
 527         u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
 528                                                         (100 - age_weight);
 529
 530         f2fs_bug_on(sbi, age + u >= UINT_MAX);
 531
 532         cost = UINT_MAX - (age + u);
 533         iter++;
 534
 535         if (cost < p->min_cost ||
 536                         (cost == p->min_cost && age > p->oldest_age)) {
 537                 p->min_cost = cost;
 538                 p->oldest_age = age;
 539                 p->min_segno = ve->segno;
 540         }
 541 skip:
 542         if (iter < dirty_threshold) {
 543                 node = rb_next(node);
 544                 goto next;
 545         }
 546 }
 547
 548 /*
 549  * select candidates around source section in range of
 550  * [target - dirty_threshold, target + dirty_threshold]
 551  */
 552 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
 553                                                 struct victim_sel_policy *p)
 554 {
 555         struct sit_info *sit_i = SIT_I(sbi);
 556         struct atgc_management *am = &sbi->am;
 557         struct rb_node *node;
 558         struct rb_entry *re;
 559         struct victim_entry *ve;
 560         unsigned long long age;
 561         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 562         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 563         unsigned int seg_blocks = sbi->blocks_per_seg;
 564         unsigned int vblocks;
 565         unsigned int dirty_threshold = max(am->max_candidate_count,
 566                                         am->candidate_ratio *
 567                                         am->victim_count / 100);
 568         unsigned int cost;
 569         unsigned int iter = 0;
 570         int stage = 0;
 571
 572         if (max_mtime < min_mtime)
 573                 return;
 574         max_mtime += 1;
 575 next_stage:
 576         node = lookup_central_victim(sbi, p);
 577 next_node:
 578         re = rb_entry_safe(node, struct rb_entry, rb_node);
 579         if (!re) {
 580                 if (stage == 0)
 581                         goto skip_stage;
 582                 return;
 583         }
 584
 585         ve = (struct victim_entry *)re;
 586
 587         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 588                 goto skip_node;
 589
 590         age = max_mtime - ve->mtime;
 591
 592         vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
 593         f2fs_bug_on(sbi, !vblocks);
 594
 595         /* rare case */
 596         if (vblocks == seg_blocks)
 597                 goto skip_node;
 598
 599         iter++;
 600
 601         age = max_mtime - abs(p->age - age);
 602         cost = UINT_MAX - vblocks;
 603
 604         if (cost < p->min_cost ||
 605                         (cost == p->min_cost && age > p->oldest_age)) {
 606                 p->min_cost = cost;
 607                 p->oldest_age = age;
 608                 p->min_segno = ve->segno;
 609         }
 610 skip_node:
 611         if (iter < dirty_threshold) {
 612                 if (stage == 0)
 613                         node = rb_prev(node);
 614                 else if (stage == 1)
 615                         node = rb_next(node);
 616                 goto next_node;
 617         }
 618 skip_stage:
 619         if (stage < 1) {
 620                 stage++;
 621                 iter = 0;
 622                 goto next_stage;
 623         }
 624 }
 625 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
 626                                                 struct victim_sel_policy *p)
 627 {
 628         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
 629                                                 &sbi->am.root, true));
 630
 631         if (p->gc_mode == GC_AT)
 632                 atgc_lookup_victim(sbi, p);
 633         else if (p->alloc_mode == AT_SSR)
 634                 atssr_lookup_victim(sbi, p);
 635         else
 636                 f2fs_bug_on(sbi, 1);
 637 }
 638
 639 static void release_victim_entry(struct f2fs_sb_info *sbi)
 640 {
 641         struct atgc_management *am = &sbi->am;
 642         struct victim_entry *ve, *tmp;
 643
 644         list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
 645                 list_del(&ve->list);
 646                 kmem_cache_free(victim_entry_slab, ve);
 647                 am->victim_count--;
 648         }
 649
 650         am->root = RB_ROOT_CACHED;
 651
 652         f2fs_bug_on(sbi, am->victim_count);
 653         f2fs_bug_on(sbi, !list_empty(&am->victim_list));
 654 }
 655
 656 static bool f2fs_pin_section(struct f2fs_sb_info *sbi, unsigned int segno)
 657 {
 658         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 659         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 660
 661         if (!dirty_i->enable_pin_section)
 662                 return false;
 663         if (!test_and_set_bit(secno, dirty_i->pinned_secmap))
 664                 dirty_i->pinned_secmap_cnt++;
 665         return true;
 666 }
 667
 668 static bool f2fs_pinned_section_exists(struct dirty_seglist_info *dirty_i)
 669 {
 670         return dirty_i->pinned_secmap_cnt;
 671 }
 672
 673 static bool f2fs_section_is_pinned(struct dirty_seglist_info *dirty_i,
 674                                                 unsigned int secno)
 675 {
 676         return dirty_i->enable_pin_section &&
 677                 f2fs_pinned_section_exists(dirty_i) &&
 678                 test_bit(secno, dirty_i->pinned_secmap);
 679 }
 680
 681 static void f2fs_unpin_all_sections(struct f2fs_sb_info *sbi, bool enable)
 682 {
 683         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
 684
 685         if (f2fs_pinned_section_exists(DIRTY_I(sbi))) {
 686                 memset(DIRTY_I(sbi)->pinned_secmap, 0, bitmap_size);
 687                 DIRTY_I(sbi)->pinned_secmap_cnt = 0;
 688         }
 689         DIRTY_I(sbi)->enable_pin_section = enable;
 690 }
 691
 692 static int f2fs_gc_pinned_control(struct inode *inode, int gc_type,
 693                                                         unsigned int segno)
 694 {
 695         if (!f2fs_is_pinned_file(inode))
 696                 return 0;
 697         if (gc_type != FG_GC)
 698                 return -EBUSY;
 699         if (!f2fs_pin_section(F2FS_I_SB(inode), segno))
 700                 f2fs_pin_file_control(inode, true);
 701         return -EAGAIN;
 702 }
 703
 704 /*
 705  * This function is called from two paths.
 706  * One is garbage collection and the other is SSR segment selection.
 707  * When it is called during GC, it just gets a victim segment
 708  * and it does not remove it from dirty seglist.
 709  * When it is called from SSR segment selection, it finds a segment
 710  * which has minimum valid blocks and removes it from dirty seglist.
 711  */
 712 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 713                         unsigned int *result, int gc_type, int type,
 714                         char alloc_mode, unsigned long long age)
 715 {
 716         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 717         struct sit_info *sm = SIT_I(sbi);
 718         struct victim_sel_policy p;
 719         unsigned int secno, last_victim;
 720         unsigned int last_segment;
 721         unsigned int nsearched;
 722         bool is_atgc;
 723         int ret = 0;
 724
 725         mutex_lock(&dirty_i->seglist_lock);
 726         last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
 727
 728         p.alloc_mode = alloc_mode;
 729         p.age = age;
 730         p.age_threshold = sbi->am.age_threshold;
 731
 732 retry:
 733         select_policy(sbi, gc_type, type, &p);
 734         p.min_segno = NULL_SEGNO;
 735         p.oldest_age = 0;
 736         p.min_cost = get_max_cost(sbi, &p);
 737
 738         is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
 739         nsearched = 0;
 740
 741         if (is_atgc)
 742                 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
 743
 744         if (*result != NULL_SEGNO) {
 745                 if (!get_valid_blocks(sbi, *result, false)) {
 746                         ret = -ENODATA;
 747                         goto out;
 748                 }
 749
 750                 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
 751                         ret = -EBUSY;
 752                 else
 753                         p.min_segno = *result;
 754                 goto out;
 755         }
 756
 757         ret = -ENODATA;
 758         if (p.max_search == 0)
 759                 goto out;
 760
 761         if (__is_large_section(sbi) && p.alloc_mode == LFS) {
 762                 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
 763                         p.min_segno = sbi->next_victim_seg[BG_GC];
 764                         *result = p.min_segno;
 765                         sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
 766                         goto got_result;
 767                 }
 768                 if (gc_type == FG_GC &&
 769                                 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
 770                         p.min_segno = sbi->next_victim_seg[FG_GC];
 771                         *result = p.min_segno;
 772                         sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
 773                         goto got_result;
 774                 }
 775         }
 776
 777         last_victim = sm->last_victim[p.gc_mode];
 778         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 779                 p.min_segno = check_bg_victims(sbi);
 780                 if (p.min_segno != NULL_SEGNO)
 781                         goto got_it;
 782         }
 783
 784         while (1) {
 785                 unsigned long cost, *dirty_bitmap;
 786                 unsigned int unit_no, segno;
 787
 788                 dirty_bitmap = p.dirty_bitmap;
 789                 unit_no = find_next_bit(dirty_bitmap,
 790                                 last_segment / p.ofs_unit,
 791                                 p.offset / p.ofs_unit);
 792                 segno = unit_no * p.ofs_unit;
 793                 if (segno >= last_segment) {
 794                         if (sm->last_victim[p.gc_mode]) {
 795                                 last_segment =
 796                                         sm->last_victim[p.gc_mode];
 797                                 sm->last_victim[p.gc_mode] = 0;
 798                                 p.offset = 0;
 799                                 continue;
 800                         }
 801                         break;
 802                 }
 803
 804                 p.offset = segno + p.ofs_unit;
 805                 nsearched++;
 806
 807 #ifdef CONFIG_F2FS_CHECK_FS
 808                 /*
 809                  * skip selecting the invalid segno (that is failed due to block
 810                  * validity check failure during GC) to avoid endless GC loop in
 811                  * such cases.
 812                  */
 813                 if (test_bit(segno, sm->invalid_segmap))
 814                         goto next;
 815 #endif
 816
 817                 secno = GET_SEC_FROM_SEG(sbi, segno);
 818
 819                 if (sec_usage_check(sbi, secno))
 820                         goto next;
 821
 822                 /* Don't touch checkpointed data */
 823                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 824                         if (p.alloc_mode == LFS) {
 825                                 /*
 826                                  * LFS is set to find source section during GC.
 827                                  * The victim should have no checkpointed data.
 828                                  */
 829                                 if (get_ckpt_valid_blocks(sbi, segno, true))
 830                                         goto next;
 831                         } else {
 832                                 /*
 833                                  * SSR | AT_SSR are set to find target segment
 834                                  * for writes which can be full by checkpointed
 835                                  * and newly written blocks.
 836                                  */
 837                                 if (!f2fs_segment_has_free_slot(sbi, segno))
 838                                         goto next;
 839                         }
 840                 }
 841
 842                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 843                         goto next;
 844
 845                 if (gc_type == FG_GC && f2fs_section_is_pinned(dirty_i, secno))
 846                         goto next;
 847
 848                 if (is_atgc) {
 849                         add_victim_entry(sbi, &p, segno);
 850                         goto next;
 851                 }
 852
 853                 cost = get_gc_cost(sbi, segno, &p);
 854
 855                 if (p.min_cost > cost) {
 856                         p.min_segno = segno;
 857                         p.min_cost = cost;
 858                 }
 859 next:
 860                 if (nsearched >= p.max_search) {
 861                         if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
 862                                 sm->last_victim[p.gc_mode] =
 863                                         last_victim + p.ofs_unit;
 864                         else
 865                                 sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
 866                         sm->last_victim[p.gc_mode] %=
 867                                 (MAIN_SECS(sbi) * sbi->segs_per_sec);
 868                         break;
 869                 }
 870         }
 871
 872         /* get victim for GC_AT/AT_SSR */
 873         if (is_atgc) {
 874                 lookup_victim_by_age(sbi, &p);
 875                 release_victim_entry(sbi);
 876         }
 877
 878         if (is_atgc && p.min_segno == NULL_SEGNO &&
 879                         sm->elapsed_time < p.age_threshold) {
 880                 p.age_threshold = 0;
 881                 goto retry;
 882         }
 883
 884         if (p.min_segno != NULL_SEGNO) {
 885 got_it:
 886                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 887 got_result:
 888                 if (p.alloc_mode == LFS) {
 889                         secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
 890                         if (gc_type == FG_GC)
 891                                 sbi->cur_victim_sec = secno;
 892                         else
 893                                 set_bit(secno, dirty_i->victim_secmap);
 894                 }
 895                 ret = 0;
 896
 897         }
 898 out:
 899         if (p.min_segno != NULL_SEGNO)
 900                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 901                                 sbi->cur_victim_sec,
 902                                 prefree_segments(sbi), free_segments(sbi));
 903         mutex_unlock(&dirty_i->seglist_lock);
 904
 905         return ret;
 906 }
 907
 908 static const struct victim_selection default_v_ops = {
 909         .get_victim = get_victim_by_default,
 910 };
 911
 912 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
 913 {
 914         struct inode_entry *ie;
 915
 916         ie = radix_tree_lookup(&gc_list->iroot, ino);
 917         if (ie)
 918                 return ie->inode;
 919         return NULL;
 920 }
 921
 922 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
 923 {
 924         struct inode_entry *new_ie;
 925
 926         if (inode == find_gc_inode(gc_list, inode->i_ino)) {
 927                 iput(inode);
 928                 return;
 929         }
 930         new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab,
 931                                         GFP_NOFS, true, NULL);
 932         new_ie->inode = inode;
 933
 934         f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
 935         list_add_tail(&new_ie->list, &gc_list->ilist);
 936 }
 937
 938 static void put_gc_inode(struct gc_inode_list *gc_list)
 939 {
 940         struct inode_entry *ie, *next_ie;
 941
 942         list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
 943                 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
 944                 iput(ie->inode);
 945                 list_del(&ie->list);
 946                 kmem_cache_free(f2fs_inode_entry_slab, ie);
 947         }
 948 }
 949
 950 static int check_valid_map(struct f2fs_sb_info *sbi,
 951                                 unsigned int segno, int offset)
 952 {
 953         struct sit_info *sit_i = SIT_I(sbi);
 954         struct seg_entry *sentry;
 955         int ret;
 956
 957         down_read(&sit_i->sentry_lock);
 958         sentry = get_seg_entry(sbi, segno);
 959         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 960         up_read(&sit_i->sentry_lock);
 961         return ret;
 962 }
 963
 964 /*
 965  * This function compares node address got in summary with that in NAT.
 966  * On validity, copy that node with cold status, otherwise (invalid node)
 967  * ignore that.
 968  */
 969 static int gc_node_segment(struct f2fs_sb_info *sbi,
 970                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 971 {
 972         struct f2fs_summary *entry;
 973         block_t start_addr;
 974         int off;
 975         int phase = 0;
 976         bool fggc = (gc_type == FG_GC);
 977         int submitted = 0;
 978         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
 979
 980         start_addr = START_BLOCK(sbi, segno);
 981
 982 next_step:
 983         entry = sum;
 984
 985         if (fggc && phase == 2)
 986                 atomic_inc(&sbi->wb_sync_req[NODE]);
 987
 988         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
 989                 nid_t nid = le32_to_cpu(entry->nid);
 990                 struct page *node_page;
 991                 struct node_info ni;
 992                 int err;
 993
 994                 /* stop BG_GC if there is not enough free sections. */
 995                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 996                         return submitted;
 997
 998                 if (check_valid_map(sbi, segno, off) == 0)
 999                         continue;
1000
1001                 if (phase == 0) {
1002                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1003                                                         META_NAT, true);
1004                         continue;
1005                 }
1006
1007                 if (phase == 1) {
1008                         f2fs_ra_node_page(sbi, nid);
1009                         continue;
1010                 }
1011
1012                 /* phase == 2 */
1013                 node_page = f2fs_get_node_page(sbi, nid);
1014                 if (IS_ERR(node_page))
1015                         continue;
1016
1017                 /* block may become invalid during f2fs_get_node_page */
1018                 if (check_valid_map(sbi, segno, off) == 0) {
1019                         f2fs_put_page(node_page, 1);
1020                         continue;
1021                 }
1022
1023                 if (f2fs_get_node_info(sbi, nid, &ni, false)) {
1024                         f2fs_put_page(node_page, 1);
1025                         continue;
1026                 }
1027
1028                 if (ni.blk_addr != start_addr + off) {
1029                         f2fs_put_page(node_page, 1);
1030                         continue;
1031                 }
1032
1033                 err = f2fs_move_node_page(node_page, gc_type);
1034                 if (!err && gc_type == FG_GC)
1035                         submitted++;
1036                 stat_inc_node_blk_count(sbi, 1, gc_type);
1037         }
1038
1039         if (++phase < 3)
1040                 goto next_step;
1041
1042         if (fggc)
1043                 atomic_dec(&sbi->wb_sync_req[NODE]);
1044         return submitted;
1045 }
1046
1047 /*
1048  * Calculate start block index indicating the given node offset.
1049  * Be careful, caller should give this node offset only indicating direct node
1050  * blocks. If any node offsets, which point the other types of node blocks such
1051  * as indirect or double indirect node blocks, are given, it must be a caller's
1052  * bug.
1053  */
1054 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
1055 {
1056         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
1057         unsigned int bidx;
1058
1059         if (node_ofs == 0)
1060                 return 0;
1061
1062         if (node_ofs <= 2) {
1063                 bidx = node_ofs - 1;
1064         } else if (node_ofs <= indirect_blks) {
1065                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
1066
1067                 bidx = node_ofs - 2 - dec;
1068         } else {
1069                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
1070
1071                 bidx = node_ofs - 5 - dec;
1072         }
1073         return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
1074 }
1075
1076 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1077                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
1078 {
1079         struct page *node_page;
1080         nid_t nid;
1081         unsigned int ofs_in_node, max_addrs, base;
1082         block_t source_blkaddr;
1083
1084         nid = le32_to_cpu(sum->nid);
1085         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
1086
1087         node_page = f2fs_get_node_page(sbi, nid);
1088         if (IS_ERR(node_page))
1089                 return false;
1090
1091         if (f2fs_get_node_info(sbi, nid, dni, false)) {
1092                 f2fs_put_page(node_page, 1);
1093                 return false;
1094         }
1095
1096         if (sum->version != dni->version) {
1097                 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
1098                           __func__);
1099                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1100         }
1101
1102         if (f2fs_check_nid_range(sbi, dni->ino)) {
1103                 f2fs_put_page(node_page, 1);
1104                 return false;
1105         }
1106
1107         if (IS_INODE(node_page)) {
1108                 base = offset_in_addr(F2FS_INODE(node_page));
1109                 max_addrs = DEF_ADDRS_PER_INODE;
1110         } else {
1111                 base = 0;
1112                 max_addrs = DEF_ADDRS_PER_BLOCK;
1113         }
1114
1115         if (base + ofs_in_node >= max_addrs) {
1116                 f2fs_err(sbi, "Inconsistent blkaddr offset: base:%u, ofs_in_node:%u, max:%u, ino:%u, nid:%u",
1117                         base, ofs_in_node, max_addrs, dni->ino, dni->nid);
1118                 f2fs_put_page(node_page, 1);
1119                 return false;
1120         }
1121
1122         *nofs = ofs_of_node(node_page);
1123         source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1124         f2fs_put_page(node_page, 1);
1125
1126         if (source_blkaddr != blkaddr) {
1127 #ifdef CONFIG_F2FS_CHECK_FS
1128                 unsigned int segno = GET_SEGNO(sbi, blkaddr);
1129                 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1130
1131                 if (unlikely(check_valid_map(sbi, segno, offset))) {
1132                         if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1133                                 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",
1134                                          blkaddr, source_blkaddr, segno);
1135                                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1136                         }
1137                 }
1138 #endif
1139                 return false;
1140         }
1141         return true;
1142 }
1143
1144 static int ra_data_block(struct inode *inode, pgoff_t index)
1145 {
1146         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1147         struct address_space *mapping = inode->i_mapping;
1148         struct dnode_of_data dn;
1149         struct page *page;
1150         struct extent_info ei = {0, 0, 0};
1151         struct f2fs_io_info fio = {
1152                 .sbi = sbi,
1153                 .ino = inode->i_ino,
1154                 .type = DATA,
1155                 .temp = COLD,
1156                 .op = REQ_OP_READ,
1157                 .op_flags = 0,
1158                 .encrypted_page = NULL,
1159                 .in_list = false,
1160                 .retry = false,
1161         };
1162         int err;
1163
1164         page = f2fs_grab_cache_page(mapping, index, true);
1165         if (!page)
1166                 return -ENOMEM;
1167
1168         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1169                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1170                 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1171                                                 DATA_GENERIC_ENHANCE_READ))) {
1172                         err = -EFSCORRUPTED;
1173                         f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1174                         goto put_page;
1175                 }
1176                 goto got_it;
1177         }
1178
1179         set_new_dnode(&dn, inode, NULL, NULL, 0);
1180         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1181         if (err)
1182                 goto put_page;
1183         f2fs_put_dnode(&dn);
1184
1185         if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1186                 err = -ENOENT;
1187                 goto put_page;
1188         }
1189         if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1190                                                 DATA_GENERIC_ENHANCE))) {
1191                 err = -EFSCORRUPTED;
1192                 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1193                 goto put_page;
1194         }
1195 got_it:
1196         /* read page */
1197         fio.page = page;
1198         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1199
1200         /*
1201          * don't cache encrypted data into meta inode until previous dirty
1202          * data were writebacked to avoid racing between GC and flush.
1203          */
1204         f2fs_wait_on_page_writeback(page, DATA, true, true);
1205
1206         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1207
1208         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1209                                         dn.data_blkaddr,
1210                                         FGP_LOCK | FGP_CREAT, GFP_NOFS);
1211         if (!fio.encrypted_page) {
1212                 err = -ENOMEM;
1213                 goto put_page;
1214         }
1215
1216         err = f2fs_submit_page_bio(&fio);
1217         if (err)
1218                 goto put_encrypted_page;
1219         f2fs_put_page(fio.encrypted_page, 0);
1220         f2fs_put_page(page, 1);
1221
1222         f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
1223         f2fs_update_iostat(sbi, NULL, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1224
1225         return 0;
1226 put_encrypted_page:
1227         f2fs_put_page(fio.encrypted_page, 1);
1228 put_page:
1229         f2fs_put_page(page, 1);
1230         return err;
1231 }
1232
1233 /*
1234  * Move data block via META_MAPPING while keeping locked data page.
1235  * This can be used to move blocks, aka LBAs, directly on disk.
1236  */
1237 static int move_data_block(struct inode *inode, block_t bidx,
1238                                 int gc_type, unsigned int segno, int off)
1239 {
1240         struct f2fs_io_info fio = {
1241                 .sbi = F2FS_I_SB(inode),
1242                 .ino = inode->i_ino,
1243                 .type = DATA,
1244                 .temp = COLD,
1245                 .op = REQ_OP_READ,
1246                 .op_flags = 0,
1247                 .encrypted_page = NULL,
1248                 .in_list = false,
1249                 .retry = false,
1250         };
1251         struct dnode_of_data dn;
1252         struct f2fs_summary sum;
1253         struct node_info ni;
1254         struct page *page, *mpage;
1255         block_t newaddr;
1256         int err = 0;
1257         bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1258         int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&
1259                                 (fio.sbi->gc_mode != GC_URGENT_HIGH) ?
1260                                 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1261
1262         /* do not read out */
1263         page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1264         if (!page)
1265                 return -ENOMEM;
1266
1267         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1268                 err = -ENOENT;
1269                 goto out;
1270         }
1271
1272         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1273         if (err)
1274                 goto out;
1275
1276         set_new_dnode(&dn, inode, NULL, NULL, 0);
1277         err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1278         if (err)
1279                 goto out;
1280
1281         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1282                 ClearPageUptodate(page);
1283                 err = -ENOENT;
1284                 goto put_out;
1285         }
1286
1287         /*
1288          * don't cache encrypted data into meta inode until previous dirty
1289          * data were writebacked to avoid racing between GC and flush.
1290          */
1291         f2fs_wait_on_page_writeback(page, DATA, true, true);
1292
1293         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1294
1295         err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);
1296         if (err)
1297                 goto put_out;
1298
1299         /* read page */
1300         fio.page = page;
1301         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1302
1303         if (lfs_mode)
1304                 f2fs_down_write(&fio.sbi->io_order_lock);
1305
1306         mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1307                                         fio.old_blkaddr, false);
1308         if (!mpage) {
1309                 err = -ENOMEM;
1310                 goto up_out;
1311         }
1312
1313         fio.encrypted_page = mpage;
1314
1315         /* read source block in mpage */
1316         if (!PageUptodate(mpage)) {
1317                 err = f2fs_submit_page_bio(&fio);
1318                 if (err) {
1319                         f2fs_put_page(mpage, 1);
1320                         goto up_out;
1321                 }
1322
1323                 f2fs_update_iostat(fio.sbi, inode, FS_DATA_READ_IO,
1324                                                         F2FS_BLKSIZE);
1325                 f2fs_update_iostat(fio.sbi, NULL, FS_GDATA_READ_IO,
1326                                                         F2FS_BLKSIZE);
1327
1328                 lock_page(mpage);
1329                 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1330                                                 !PageUptodate(mpage))) {
1331                         err = -EIO;
1332                         f2fs_put_page(mpage, 1);
1333                         goto up_out;
1334                 }
1335         }
1336
1337         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1338
1339         /* allocate block address */
1340         f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1341                                 &sum, type, NULL);
1342
1343         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1344                                 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1345         if (!fio.encrypted_page) {
1346                 err = -ENOMEM;
1347                 f2fs_put_page(mpage, 1);
1348                 goto recover_block;
1349         }
1350
1351         /* write target block */
1352         f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1353         memcpy(page_address(fio.encrypted_page),
1354                                 page_address(mpage), PAGE_SIZE);
1355         f2fs_put_page(mpage, 1);
1356         invalidate_mapping_pages(META_MAPPING(fio.sbi),
1357                                 fio.old_blkaddr, fio.old_blkaddr);
1358         f2fs_invalidate_compress_page(fio.sbi, fio.old_blkaddr);
1359
1360         set_page_dirty(fio.encrypted_page);
1361         if (clear_page_dirty_for_io(fio.encrypted_page))
1362                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
1363
1364         set_page_writeback(fio.encrypted_page);
1365         ClearPageError(page);
1366
1367         fio.op = REQ_OP_WRITE;
1368         fio.op_flags = REQ_SYNC;
1369         fio.new_blkaddr = newaddr;
1370         f2fs_submit_page_write(&fio);
1371         if (fio.retry) {
1372                 err = -EAGAIN;
1373                 if (PageWriteback(fio.encrypted_page))
1374                         end_page_writeback(fio.encrypted_page);
1375                 goto put_page_out;
1376         }
1377
1378         f2fs_update_iostat(fio.sbi, NULL, FS_GC_DATA_IO, F2FS_BLKSIZE);
1379
1380         f2fs_update_data_blkaddr(&dn, newaddr);
1381         set_inode_flag(inode, FI_APPEND_WRITE);
1382         if (page->index == 0)
1383                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1384 put_page_out:
1385         f2fs_put_page(fio.encrypted_page, 1);
1386 recover_block:
1387         if (err)
1388                 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1389                                                         true, true, true);
1390 up_out:
1391         if (lfs_mode)
1392                 f2fs_up_write(&fio.sbi->io_order_lock);
1393 put_out:
1394         f2fs_put_dnode(&dn);
1395 out:
1396         f2fs_put_page(page, 1);
1397         return err;
1398 }
1399
1400 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1401                                                         unsigned int segno, int off)
1402 {
1403         struct page *page;
1404         int err = 0;
1405
1406         page = f2fs_get_lock_data_page(inode, bidx, true);
1407         if (IS_ERR(page))
1408                 return PTR_ERR(page);
1409
1410         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1411                 err = -ENOENT;
1412                 goto out;
1413         }
1414
1415         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1416         if (err)
1417                 goto out;
1418
1419         if (gc_type == BG_GC) {
1420                 if (PageWriteback(page)) {
1421                         err = -EAGAIN;
1422                         goto out;
1423                 }
1424                 set_page_dirty(page);
1425                 set_page_private_gcing(page);
1426         } else {
1427                 struct f2fs_io_info fio = {
1428                         .sbi = F2FS_I_SB(inode),
1429                         .ino = inode->i_ino,
1430                         .type = DATA,
1431                         .temp = COLD,
1432                         .op = REQ_OP_WRITE,
1433                         .op_flags = REQ_SYNC,
1434                         .old_blkaddr = NULL_ADDR,
1435                         .page = page,
1436                         .encrypted_page = NULL,
1437                         .need_lock = LOCK_REQ,
1438                         .io_type = FS_GC_DATA_IO,
1439                 };
1440                 bool is_dirty = PageDirty(page);
1441
1442 retry:
1443                 f2fs_wait_on_page_writeback(page, DATA, true, true);
1444
1445                 set_page_dirty(page);
1446                 if (clear_page_dirty_for_io(page)) {
1447                         inode_dec_dirty_pages(inode);
1448                         f2fs_remove_dirty_inode(inode);
1449                 }
1450
1451                 set_page_private_gcing(page);
1452
1453                 err = f2fs_do_write_data_page(&fio);
1454                 if (err) {
1455                         clear_page_private_gcing(page);
1456                         if (err == -ENOMEM) {
1457                                 memalloc_retry_wait(GFP_NOFS);
1458                                 goto retry;
1459                         }
1460                         if (is_dirty)
1461                                 set_page_dirty(page);
1462                 }
1463         }
1464 out:
1465         f2fs_put_page(page, 1);
1466         return err;
1467 }
1468
1469 /*
1470  * This function tries to get parent node of victim data block, and identifies
1471  * data block validity. If the block is valid, copy that with cold status and
1472  * modify parent node.
1473  * If the parent node is not valid or the data block address is different,
1474  * the victim data block is ignored.
1475  */
1476 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1477                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1478                 bool force_migrate)
1479 {
1480         struct super_block *sb = sbi->sb;
1481         struct f2fs_summary *entry;
1482         block_t start_addr;
1483         int off;
1484         int phase = 0;
1485         int submitted = 0;
1486         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1487
1488         start_addr = START_BLOCK(sbi, segno);
1489
1490 next_step:
1491         entry = sum;
1492
1493         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1494                 struct page *data_page;
1495                 struct inode *inode;
1496                 struct node_info dni; /* dnode info for the data */
1497                 unsigned int ofs_in_node, nofs;
1498                 block_t start_bidx;
1499                 nid_t nid = le32_to_cpu(entry->nid);
1500
1501                 /*
1502                  * stop BG_GC if there is not enough free sections.
1503                  * Or, stop GC if the segment becomes fully valid caused by
1504                  * race condition along with SSR block allocation.
1505                  */
1506                 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1507                         (!force_migrate && get_valid_blocks(sbi, segno, true) ==
1508                                                         CAP_BLKS_PER_SEC(sbi)))
1509                         return submitted;
1510
1511                 if (check_valid_map(sbi, segno, off) == 0)
1512                         continue;
1513
1514                 if (phase == 0) {
1515                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1516                                                         META_NAT, true);
1517                         continue;
1518                 }
1519
1520                 if (phase == 1) {
1521                         f2fs_ra_node_page(sbi, nid);
1522                         continue;
1523                 }
1524
1525                 /* Get an inode by ino with checking validity */
1526                 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1527                         continue;
1528
1529                 if (phase == 2) {
1530                         f2fs_ra_node_page(sbi, dni.ino);
1531                         continue;
1532                 }
1533
1534                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1535
1536                 if (phase == 3) {
1537                         int err;
1538
1539                         inode = f2fs_iget(sb, dni.ino);
1540                         if (IS_ERR(inode) || is_bad_inode(inode) ||
1541                                         special_file(inode->i_mode))
1542                                 continue;
1543
1544                         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1545                         if (err == -EAGAIN) {
1546                                 iput(inode);
1547                                 return submitted;
1548                         }
1549
1550                         if (!f2fs_down_write_trylock(
1551                                 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1552                                 iput(inode);
1553                                 sbi->skipped_gc_rwsem++;
1554                                 continue;
1555                         }
1556
1557                         start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1558                                                                 ofs_in_node;
1559
1560                         if (f2fs_post_read_required(inode)) {
1561                                 int err = ra_data_block(inode, start_bidx);
1562
1563                                 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1564                                 if (err) {
1565                                         iput(inode);
1566                                         continue;
1567                                 }
1568                                 add_gc_inode(gc_list, inode);
1569                                 continue;
1570                         }
1571
1572                         data_page = f2fs_get_read_data_page(inode,
1573                                                 start_bidx, REQ_RAHEAD, true);
1574                         f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1575                         if (IS_ERR(data_page)) {
1576                                 iput(inode);
1577                                 continue;
1578                         }
1579
1580                         f2fs_put_page(data_page, 0);
1581                         add_gc_inode(gc_list, inode);
1582                         continue;
1583                 }
1584
1585                 /* phase 4 */
1586                 inode = find_gc_inode(gc_list, dni.ino);
1587                 if (inode) {
1588                         struct f2fs_inode_info *fi = F2FS_I(inode);
1589                         bool locked = false;
1590                         int err;
1591
1592                         if (S_ISREG(inode->i_mode)) {
1593                                 if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[READ])) {
1594                                         sbi->skipped_gc_rwsem++;
1595                                         continue;
1596                                 }
1597                                 if (!f2fs_down_write_trylock(
1598                                                 &fi->i_gc_rwsem[WRITE])) {
1599                                         sbi->skipped_gc_rwsem++;
1600                                         f2fs_up_write(&fi->i_gc_rwsem[READ]);
1601                                         continue;
1602                                 }
1603                                 locked = true;
1604
1605                                 /* wait for all inflight aio data */
1606                                 inode_dio_wait(inode);
1607                         }
1608
1609                         start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1610                                                                 + ofs_in_node;
1611                         if (f2fs_post_read_required(inode))
1612                                 err = move_data_block(inode, start_bidx,
1613                                                         gc_type, segno, off);
1614                         else
1615                                 err = move_data_page(inode, start_bidx, gc_type,
1616                                                                 segno, off);
1617
1618                         if (!err && (gc_type == FG_GC ||
1619                                         f2fs_post_read_required(inode)))
1620                                 submitted++;
1621
1622                         if (locked) {
1623                                 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
1624                                 f2fs_up_write(&fi->i_gc_rwsem[READ]);
1625                         }
1626
1627                         stat_inc_data_blk_count(sbi, 1, gc_type);
1628                 }
1629         }
1630
1631         if (++phase < 5)
1632                 goto next_step;
1633
1634         return submitted;
1635 }
1636
1637 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1638                         int gc_type)
1639 {
1640         struct sit_info *sit_i = SIT_I(sbi);
1641         int ret;
1642
1643         down_write(&sit_i->sentry_lock);
1644         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1645                                               NO_CHECK_TYPE, LFS, 0);
1646         up_write(&sit_i->sentry_lock);
1647         return ret;
1648 }
1649
1650 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1651                                 unsigned int start_segno,
1652                                 struct gc_inode_list *gc_list, int gc_type,
1653                                 bool force_migrate)
1654 {
1655         struct page *sum_page;
1656         struct f2fs_summary_block *sum;
1657         struct blk_plug plug;
1658         unsigned int segno = start_segno;
1659         unsigned int end_segno = start_segno + sbi->segs_per_sec;
1660         int seg_freed = 0, migrated = 0;
1661         unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1662                                                 SUM_TYPE_DATA : SUM_TYPE_NODE;
1663         int submitted = 0;
1664
1665         if (__is_large_section(sbi))
1666                 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1667
1668         /*
1669          * zone-capacity can be less than zone-size in zoned devices,
1670          * resulting in less than expected usable segments in the zone,
1671          * calculate the end segno in the zone which can be garbage collected
1672          */
1673         if (f2fs_sb_has_blkzoned(sbi))
1674                 end_segno -= sbi->segs_per_sec -
1675                                         f2fs_usable_segs_in_sec(sbi, segno);
1676
1677         sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1678
1679         /* readahead multi ssa blocks those have contiguous address */
1680         if (__is_large_section(sbi))
1681                 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1682                                         end_segno - segno, META_SSA, true);
1683
1684         /* reference all summary page */
1685         while (segno < end_segno) {
1686                 sum_page = f2fs_get_sum_page(sbi, segno++);
1687                 if (IS_ERR(sum_page)) {
1688                         int err = PTR_ERR(sum_page);
1689
1690                         end_segno = segno - 1;
1691                         for (segno = start_segno; segno < end_segno; segno++) {
1692                                 sum_page = find_get_page(META_MAPPING(sbi),
1693                                                 GET_SUM_BLOCK(sbi, segno));
1694                                 f2fs_put_page(sum_page, 0);
1695                                 f2fs_put_page(sum_page, 0);
1696                         }
1697                         return err;
1698                 }
1699                 unlock_page(sum_page);
1700         }
1701
1702         blk_start_plug(&plug);
1703
1704         for (segno = start_segno; segno < end_segno; segno++) {
1705
1706                 /* find segment summary of victim */
1707                 sum_page = find_get_page(META_MAPPING(sbi),
1708                                         GET_SUM_BLOCK(sbi, segno));
1709                 f2fs_put_page(sum_page, 0);
1710
1711                 if (get_valid_blocks(sbi, segno, false) == 0)
1712                         goto freed;
1713                 if (gc_type == BG_GC && __is_large_section(sbi) &&
1714                                 migrated >= sbi->migration_granularity)
1715                         goto skip;
1716                 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1717                         goto skip;
1718
1719                 sum = page_address(sum_page);
1720                 if (type != GET_SUM_TYPE((&sum->footer))) {
1721                         f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1722                                  segno, type, GET_SUM_TYPE((&sum->footer)));
1723                         set_sbi_flag(sbi, SBI_NEED_FSCK);
1724                         f2fs_stop_checkpoint(sbi, false,
1725                                 STOP_CP_REASON_CORRUPTED_SUMMARY);
1726                         goto skip;
1727                 }
1728
1729                 /*
1730                  * this is to avoid deadlock:
1731                  * - lock_page(sum_page)         - f2fs_replace_block
1732                  *  - check_valid_map()            - down_write(sentry_lock)
1733                  *   - down_read(sentry_lock)     - change_curseg()
1734                  *                                  - lock_page(sum_page)
1735                  */
1736                 if (type == SUM_TYPE_NODE)
1737                         submitted += gc_node_segment(sbi, sum->entries, segno,
1738                                                                 gc_type);
1739                 else
1740                         submitted += gc_data_segment(sbi, sum->entries, gc_list,
1741                                                         segno, gc_type,
1742                                                         force_migrate);
1743
1744                 stat_inc_seg_count(sbi, type, gc_type);
1745                 sbi->gc_reclaimed_segs[sbi->gc_mode]++;
1746                 migrated++;
1747
1748 freed:
1749                 if (gc_type == FG_GC &&
1750                                 get_valid_blocks(sbi, segno, false) == 0)
1751                         seg_freed++;
1752
1753                 if (__is_large_section(sbi))
1754                         sbi->next_victim_seg[gc_type] =
1755                                 (segno + 1 < end_segno) ? segno + 1 : NULL_SEGNO;
1756 skip:
1757                 f2fs_put_page(sum_page, 0);
1758         }
1759
1760         if (submitted)
1761                 f2fs_submit_merged_write(sbi,
1762                                 (type == SUM_TYPE_NODE) ? NODE : DATA);
1763
1764         blk_finish_plug(&plug);
1765
1766         stat_inc_call_count(sbi->stat_info);
1767
1768         return seg_freed;
1769 }
1770
1771 int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control)
1772 {
1773         int gc_type = gc_control->init_gc_type;
1774         unsigned int segno = gc_control->victim_segno;
1775         int sec_freed = 0, seg_freed = 0, total_freed = 0;
1776         int ret = 0;
1777         struct cp_control cpc;
1778         struct gc_inode_list gc_list = {
1779                 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1780                 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1781         };
1782         unsigned int skipped_round = 0, round = 0;
1783
1784         trace_f2fs_gc_begin(sbi->sb, gc_type, gc_control->no_bg_gc,
1785                                 gc_control->nr_free_secs,
1786                                 get_pages(sbi, F2FS_DIRTY_NODES),
1787                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1788                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1789                                 free_sections(sbi),
1790                                 free_segments(sbi),
1791                                 reserved_segments(sbi),
1792                                 prefree_segments(sbi));
1793
1794         cpc.reason = __get_cp_reason(sbi);
1795 gc_more:
1796         sbi->skipped_gc_rwsem = 0;
1797         if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1798                 ret = -EINVAL;
1799                 goto stop;
1800         }
1801         if (unlikely(f2fs_cp_error(sbi))) {
1802                 ret = -EIO;
1803                 goto stop;
1804         }
1805
1806         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1807                 /*
1808                  * For example, if there are many prefree_segments below given
1809                  * threshold, we can make them free by checkpoint. Then, we
1810                  * secure free segments which doesn't need fggc any more.
1811                  */
1812                 if (prefree_segments(sbi)) {
1813                         ret = f2fs_write_checkpoint(sbi, &cpc);
1814                         if (ret)
1815                                 goto stop;
1816                 }
1817                 if (has_not_enough_free_secs(sbi, 0, 0))
1818                         gc_type = FG_GC;
1819         }
1820
1821         /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1822         if (gc_type == BG_GC && gc_control->no_bg_gc) {
1823                 ret = -EINVAL;
1824                 goto stop;
1825         }
1826 retry:
1827         ret = __get_victim(sbi, &segno, gc_type);
1828         if (ret) {
1829                 /* allow to search victim from sections has pinned data */
1830                 if (ret == -ENODATA && gc_type == FG_GC &&
1831                                 f2fs_pinned_section_exists(DIRTY_I(sbi))) {
1832                         f2fs_unpin_all_sections(sbi, false);
1833                         goto retry;
1834                 }
1835                 goto stop;
1836         }
1837
1838         seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type,
1839                                 gc_control->should_migrate_blocks);
1840         total_freed += seg_freed;
1841
1842         if (seg_freed == f2fs_usable_segs_in_sec(sbi, segno))
1843                 sec_freed++;
1844
1845         if (gc_type == FG_GC)
1846                 sbi->cur_victim_sec = NULL_SEGNO;
1847
1848         if (gc_control->init_gc_type == FG_GC ||
1849             !has_not_enough_free_secs(sbi,
1850                                 (gc_type == FG_GC) ? sec_freed : 0, 0)) {
1851                 if (gc_type == FG_GC && sec_freed < gc_control->nr_free_secs)
1852                         goto go_gc_more;
1853                 goto stop;
1854         }
1855
1856         /* FG_GC stops GC by skip_count */
1857         if (gc_type == FG_GC) {
1858                 if (sbi->skipped_gc_rwsem)
1859                         skipped_round++;
1860                 round++;
1861                 if (skipped_round > MAX_SKIP_GC_COUNT &&
1862                                 skipped_round * 2 >= round) {
1863                         ret = f2fs_write_checkpoint(sbi, &cpc);
1864                         goto stop;
1865                 }
1866         }
1867
1868         /* Write checkpoint to reclaim prefree segments */
1869         if (free_sections(sbi) < NR_CURSEG_PERSIST_TYPE &&
1870                                 prefree_segments(sbi)) {
1871                 ret = f2fs_write_checkpoint(sbi, &cpc);
1872                 if (ret)
1873                         goto stop;
1874         }
1875 go_gc_more:
1876         segno = NULL_SEGNO;
1877         goto gc_more;
1878
1879 stop:
1880         SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1881         SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno;
1882
1883         if (gc_type == FG_GC)
1884                 f2fs_unpin_all_sections(sbi, true);
1885
1886         trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1887                                 get_pages(sbi, F2FS_DIRTY_NODES),
1888                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1889                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1890                                 free_sections(sbi),
1891                                 free_segments(sbi),
1892                                 reserved_segments(sbi),
1893                                 prefree_segments(sbi));
1894
1895         f2fs_up_write(&sbi->gc_lock);
1896
1897         put_gc_inode(&gc_list);
1898
1899         if (gc_control->err_gc_skipped && !ret)
1900                 ret = sec_freed ? 0 : -EAGAIN;
1901         return ret;
1902 }
1903
1904 int __init f2fs_create_garbage_collection_cache(void)
1905 {
1906         victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1907                                         sizeof(struct victim_entry));
1908         if (!victim_entry_slab)
1909                 return -ENOMEM;
1910         return 0;
1911 }
1912
1913 void f2fs_destroy_garbage_collection_cache(void)
1914 {
1915         kmem_cache_destroy(victim_entry_slab);
1916 }
1917
1918 static void init_atgc_management(struct f2fs_sb_info *sbi)
1919 {
1920         struct atgc_management *am = &sbi->am;
1921
1922         if (test_opt(sbi, ATGC) &&
1923                 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1924                 am->atgc_enabled = true;
1925
1926         am->root = RB_ROOT_CACHED;
1927         INIT_LIST_HEAD(&am->victim_list);
1928         am->victim_count = 0;
1929
1930         am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1931         am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1932         am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1933         am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
1934 }
1935
1936 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1937 {
1938         DIRTY_I(sbi)->v_ops = &default_v_ops;
1939
1940         sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1941
1942         /* give warm/cold data area from slower device */
1943         if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1944                 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1945                                 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1946
1947         init_atgc_management(sbi);
1948 }
1949
1950 static int free_segment_range(struct f2fs_sb_info *sbi,
1951                                 unsigned int secs, bool gc_only)
1952 {
1953         unsigned int segno, next_inuse, start, end;
1954         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1955         int gc_mode, gc_type;
1956         int err = 0;
1957         int type;
1958
1959         /* Force block allocation for GC */
1960         MAIN_SECS(sbi) -= secs;
1961         start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1962         end = MAIN_SEGS(sbi) - 1;
1963
1964         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1965         for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1966                 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1967                         SIT_I(sbi)->last_victim[gc_mode] = 0;
1968
1969         for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1970                 if (sbi->next_victim_seg[gc_type] >= start)
1971                         sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1972         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1973
1974         /* Move out cursegs from the target range */
1975         for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1976                 f2fs_allocate_segment_for_resize(sbi, type, start, end);
1977
1978         /* do GC to move out valid blocks in the range */
1979         for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1980                 struct gc_inode_list gc_list = {
1981                         .ilist = LIST_HEAD_INIT(gc_list.ilist),
1982                         .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1983                 };
1984
1985                 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
1986                 put_gc_inode(&gc_list);
1987
1988                 if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1989                         err = -EAGAIN;
1990                         goto out;
1991                 }
1992                 if (fatal_signal_pending(current)) {
1993                         err = -ERESTARTSYS;
1994                         goto out;
1995                 }
1996         }
1997         if (gc_only)
1998                 goto out;
1999
2000         err = f2fs_write_checkpoint(sbi, &cpc);
2001         if (err)
2002                 goto out;
2003
2004         next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
2005         if (next_inuse <= end) {
2006                 f2fs_err(sbi, "segno %u should be free but still inuse!",
2007                          next_inuse);
2008                 f2fs_bug_on(sbi, 1);
2009         }
2010 out:
2011         MAIN_SECS(sbi) += secs;
2012         return err;
2013 }
2014
2015 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
2016 {
2017         struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
2018         int section_count;
2019         int segment_count;
2020         int segment_count_main;
2021         long long block_count;
2022         int segs = secs * sbi->segs_per_sec;
2023
2024         f2fs_down_write(&sbi->sb_lock);
2025
2026         section_count = le32_to_cpu(raw_sb->section_count);
2027         segment_count = le32_to_cpu(raw_sb->segment_count);
2028         segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
2029         block_count = le64_to_cpu(raw_sb->block_count);
2030
2031         raw_sb->section_count = cpu_to_le32(section_count + secs);
2032         raw_sb->segment_count = cpu_to_le32(segment_count + segs);
2033         raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
2034         raw_sb->block_count = cpu_to_le64(block_count +
2035                                         (long long)segs * sbi->blocks_per_seg);
2036         if (f2fs_is_multi_device(sbi)) {
2037                 int last_dev = sbi->s_ndevs - 1;
2038                 int dev_segs =
2039                         le32_to_cpu(raw_sb->devs[last_dev].total_segments);
2040
2041                 raw_sb->devs[last_dev].total_segments =
2042                                                 cpu_to_le32(dev_segs + segs);
2043         }
2044
2045         f2fs_up_write(&sbi->sb_lock);
2046 }
2047
2048 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
2049 {
2050         int segs = secs * sbi->segs_per_sec;
2051         long long blks = (long long)segs * sbi->blocks_per_seg;
2052         long long user_block_count =
2053                                 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
2054
2055         SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
2056         MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
2057         MAIN_SECS(sbi) += secs;
2058         FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
2059         FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
2060         F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
2061
2062         if (f2fs_is_multi_device(sbi)) {
2063                 int last_dev = sbi->s_ndevs - 1;
2064
2065                 FDEV(last_dev).total_segments =
2066                                 (int)FDEV(last_dev).total_segments + segs;
2067                 FDEV(last_dev).end_blk =
2068                                 (long long)FDEV(last_dev).end_blk + blks;
2069 #ifdef CONFIG_BLK_DEV_ZONED
2070                 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
2071                                         (int)(blks >> sbi->log_blocks_per_blkz);
2072 #endif
2073         }
2074 }
2075
2076 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
2077 {
2078         __u64 old_block_count, shrunk_blocks;
2079         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
2080         unsigned int secs;
2081         int err = 0;
2082         __u32 rem;
2083
2084         old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
2085         if (block_count > old_block_count)
2086                 return -EINVAL;
2087
2088         if (f2fs_is_multi_device(sbi)) {
2089                 int last_dev = sbi->s_ndevs - 1;
2090                 __u64 last_segs = FDEV(last_dev).total_segments;
2091
2092                 if (block_count + last_segs * sbi->blocks_per_seg <=
2093                                                                 old_block_count)
2094                         return -EINVAL;
2095         }
2096
2097         /* new fs size should align to section size */
2098         div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
2099         if (rem)
2100                 return -EINVAL;
2101
2102         if (block_count == old_block_count)
2103                 return 0;
2104
2105         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2106                 f2fs_err(sbi, "Should run fsck to repair first.");
2107                 return -EFSCORRUPTED;
2108         }
2109
2110         if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2111                 f2fs_err(sbi, "Checkpoint should be enabled.");
2112                 return -EINVAL;
2113         }
2114
2115         shrunk_blocks = old_block_count - block_count;
2116         secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2117
2118         /* stop other GC */
2119         if (!f2fs_down_write_trylock(&sbi->gc_lock))
2120                 return -EAGAIN;
2121
2122         /* stop CP to protect MAIN_SEC in free_segment_range */
2123         f2fs_lock_op(sbi);
2124
2125         spin_lock(&sbi->stat_lock);
2126         if (shrunk_blocks + valid_user_blocks(sbi) +
2127                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2128                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2129                 err = -ENOSPC;
2130         spin_unlock(&sbi->stat_lock);
2131
2132         if (err)
2133                 goto out_unlock;
2134
2135         err = free_segment_range(sbi, secs, true);
2136
2137 out_unlock:
2138         f2fs_unlock_op(sbi);
2139         f2fs_up_write(&sbi->gc_lock);
2140         if (err)
2141                 return err;
2142
2143         freeze_super(sbi->sb);
2144         f2fs_down_write(&sbi->gc_lock);
2145         f2fs_down_write(&sbi->cp_global_sem);
2146
2147         spin_lock(&sbi->stat_lock);
2148         if (shrunk_blocks + valid_user_blocks(sbi) +
2149                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2150                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2151                 err = -ENOSPC;
2152         else
2153                 sbi->user_block_count -= shrunk_blocks;
2154         spin_unlock(&sbi->stat_lock);
2155         if (err)
2156                 goto out_err;
2157
2158         set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2159         err = free_segment_range(sbi, secs, false);
2160         if (err)
2161                 goto recover_out;
2162
2163         update_sb_metadata(sbi, -secs);
2164
2165         err = f2fs_commit_super(sbi, false);
2166         if (err) {
2167                 update_sb_metadata(sbi, secs);
2168                 goto recover_out;
2169         }
2170
2171         update_fs_metadata(sbi, -secs);
2172         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2173         set_sbi_flag(sbi, SBI_IS_DIRTY);
2174
2175         err = f2fs_write_checkpoint(sbi, &cpc);
2176         if (err) {
2177                 update_fs_metadata(sbi, secs);
2178                 update_sb_metadata(sbi, secs);
2179                 f2fs_commit_super(sbi, false);
2180         }
2181 recover_out:
2182         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2183         if (err) {
2184                 set_sbi_flag(sbi, SBI_NEED_FSCK);
2185                 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2186
2187                 spin_lock(&sbi->stat_lock);
2188                 sbi->user_block_count += shrunk_blocks;
2189                 spin_unlock(&sbi->stat_lock);
2190         }
2191 out_err:
2192         f2fs_up_write(&sbi->cp_global_sem);
2193         f2fs_up_write(&sbi->gc_lock);
2194         thaw_super(sbi->sb);
2195         return err;
2196 }