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                 } else {
 145                         /* reset wait_ms to default sleep time */
 146                         if (wait_ms == gc_th->no_gc_sleep_time)
 147                                 wait_ms = gc_th->min_sleep_time;
 148                 }
 149
 150                 if (foreground)
 151                         wake_up_all(&gc_th->fggc_wq);
 152
 153                 trace_f2fs_background_gc(sbi->sb, wait_ms,
 154                                 prefree_segments(sbi), free_segments(sbi));
 155
 156                 /* balancing f2fs's metadata periodically */
 157                 f2fs_balance_fs_bg(sbi, true);
 158 next:
 159                 if (sbi->gc_mode != GC_NORMAL) {
 160                         spin_lock(&sbi->gc_remaining_trials_lock);
 161                         if (sbi->gc_remaining_trials) {
 162                                 sbi->gc_remaining_trials--;
 163                                 if (!sbi->gc_remaining_trials)
 164                                         sbi->gc_mode = GC_NORMAL;
 165                         }
 166                         spin_unlock(&sbi->gc_remaining_trials_lock);
 167                 }
 168                 sb_end_write(sbi->sb);
 169
 170         } while (!kthread_should_stop());
 171         return 0;
 172 }
 173
 174 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
 175 {
 176         struct f2fs_gc_kthread *gc_th;
 177         dev_t dev = sbi->sb->s_bdev->bd_dev;
 178
 179         gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 180         if (!gc_th)
 181                 return -ENOMEM;
 182
 183         gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
 184         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 185         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 186         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 187
 188         gc_th->gc_wake = 0;
 189
 190         sbi->gc_thread = gc_th;
 191         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 192         init_waitqueue_head(&sbi->gc_thread->fggc_wq);
 193         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 194                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 195         if (IS_ERR(gc_th->f2fs_gc_task)) {
 196                 int err = PTR_ERR(gc_th->f2fs_gc_task);
 197
 198                 kfree(gc_th);
 199                 sbi->gc_thread = NULL;
 200                 return err;
 201         }
 202
 203         return 0;
 204 }
 205
 206 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
 207 {
 208         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 209
 210         if (!gc_th)
 211                 return;
 212         kthread_stop(gc_th->f2fs_gc_task);
 213         wake_up_all(&gc_th->fggc_wq);
 214         kfree(gc_th);
 215         sbi->gc_thread = NULL;
 216 }
 217
 218 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
 219 {
 220         int gc_mode;
 221
 222         if (gc_type == BG_GC) {
 223                 if (sbi->am.atgc_enabled)
 224                         gc_mode = GC_AT;
 225                 else
 226                         gc_mode = GC_CB;
 227         } else {
 228                 gc_mode = GC_GREEDY;
 229         }
 230
 231         switch (sbi->gc_mode) {
 232         case GC_IDLE_CB:
 233                 gc_mode = GC_CB;
 234                 break;
 235         case GC_IDLE_GREEDY:
 236         case GC_URGENT_HIGH:
 237                 gc_mode = GC_GREEDY;
 238                 break;
 239         case GC_IDLE_AT:
 240                 gc_mode = GC_AT;
 241                 break;
 242         }
 243
 244         return gc_mode;
 245 }
 246
 247 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 248                         int type, struct victim_sel_policy *p)
 249 {
 250         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 251
 252         if (p->alloc_mode == SSR) {
 253                 p->gc_mode = GC_GREEDY;
 254                 p->dirty_bitmap = dirty_i->dirty_segmap[type];
 255                 p->max_search = dirty_i->nr_dirty[type];
 256                 p->ofs_unit = 1;
 257         } else if (p->alloc_mode == AT_SSR) {
 258                 p->gc_mode = GC_GREEDY;
 259                 p->dirty_bitmap = dirty_i->dirty_segmap[type];
 260                 p->max_search = dirty_i->nr_dirty[type];
 261                 p->ofs_unit = 1;
 262         } else {
 263                 p->gc_mode = select_gc_type(sbi, gc_type);
 264                 p->ofs_unit = sbi->segs_per_sec;
 265                 if (__is_large_section(sbi)) {
 266                         p->dirty_bitmap = dirty_i->dirty_secmap;
 267                         p->max_search = count_bits(p->dirty_bitmap,
 268                                                 0, MAIN_SECS(sbi));
 269                 } else {
 270                         p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
 271                         p->max_search = dirty_i->nr_dirty[DIRTY];
 272                 }
 273         }
 274
 275         /*
 276          * adjust candidates range, should select all dirty segments for
 277          * foreground GC and urgent GC cases.
 278          */
 279         if (gc_type != FG_GC &&
 280                         (sbi->gc_mode != GC_URGENT_HIGH) &&
 281                         (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
 282                         p->max_search > sbi->max_victim_search)
 283                 p->max_search = sbi->max_victim_search;
 284
 285         /* let's select beginning hot/small space first in no_heap mode*/
 286         if (f2fs_need_rand_seg(sbi))
 287                 p->offset = get_random_u32_below(MAIN_SECS(sbi) * sbi->segs_per_sec);
 288         else if (test_opt(sbi, NOHEAP) &&
 289                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
 290                 p->offset = 0;
 291         else
 292                 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
 293 }
 294
 295 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 296                                 struct victim_sel_policy *p)
 297 {
 298         /* SSR allocates in a segment unit */
 299         if (p->alloc_mode == SSR)
 300                 return sbi->blocks_per_seg;
 301         else if (p->alloc_mode == AT_SSR)
 302                 return UINT_MAX;
 303
 304         /* LFS */
 305         if (p->gc_mode == GC_GREEDY)
 306                 return 2 * sbi->blocks_per_seg * p->ofs_unit;
 307         else if (p->gc_mode == GC_CB)
 308                 return UINT_MAX;
 309         else if (p->gc_mode == GC_AT)
 310                 return UINT_MAX;
 311         else /* No other gc_mode */
 312                 return 0;
 313 }
 314
 315 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 316 {
 317         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 318         unsigned int secno;
 319
 320         /*
 321          * If the gc_type is FG_GC, we can select victim segments
 322          * selected by background GC before.
 323          * Those segments guarantee they have small valid blocks.
 324          */
 325         for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
 326                 if (sec_usage_check(sbi, secno))
 327                         continue;
 328                 clear_bit(secno, dirty_i->victim_secmap);
 329                 return GET_SEG_FROM_SEC(sbi, secno);
 330         }
 331         return NULL_SEGNO;
 332 }
 333
 334 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 335 {
 336         struct sit_info *sit_i = SIT_I(sbi);
 337         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 338         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 339         unsigned long long mtime = 0;
 340         unsigned int vblocks;
 341         unsigned char age = 0;
 342         unsigned char u;
 343         unsigned int i;
 344         unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
 345
 346         for (i = 0; i < usable_segs_per_sec; i++)
 347                 mtime += get_seg_entry(sbi, start + i)->mtime;
 348         vblocks = get_valid_blocks(sbi, segno, true);
 349
 350         mtime = div_u64(mtime, usable_segs_per_sec);
 351         vblocks = div_u64(vblocks, usable_segs_per_sec);
 352
 353         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 354
 355         /* Handle if the system time has changed by the user */
 356         if (mtime < sit_i->min_mtime)
 357                 sit_i->min_mtime = mtime;
 358         if (mtime > sit_i->max_mtime)
 359                 sit_i->max_mtime = mtime;
 360         if (sit_i->max_mtime != sit_i->min_mtime)
 361                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 362                                 sit_i->max_mtime - sit_i->min_mtime);
 363
 364         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 365 }
 366
 367 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 368                         unsigned int segno, struct victim_sel_policy *p)
 369 {
 370         if (p->alloc_mode == SSR)
 371                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 372
 373         /* alloc_mode == LFS */
 374         if (p->gc_mode == GC_GREEDY)
 375                 return get_valid_blocks(sbi, segno, true);
 376         else if (p->gc_mode == GC_CB)
 377                 return get_cb_cost(sbi, segno);
 378
 379         f2fs_bug_on(sbi, 1);
 380         return 0;
 381 }
 382
 383 static unsigned int count_bits(const unsigned long *addr,
 384                                 unsigned int offset, unsigned int len)
 385 {
 386         unsigned int end = offset + len, sum = 0;
 387
 388         while (offset < end) {
 389                 if (test_bit(offset++, addr))
 390                         ++sum;
 391         }
 392         return sum;
 393 }
 394
 395 static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi,
 396                                 unsigned long long mtime, unsigned int segno,
 397                                 struct rb_node *parent, struct rb_node **p,
 398                                 bool left_most)
 399 {
 400         struct atgc_management *am = &sbi->am;
 401         struct victim_entry *ve;
 402
 403         ve =  f2fs_kmem_cache_alloc(victim_entry_slab,
 404                                 GFP_NOFS, true, NULL);
 405
 406         ve->mtime = mtime;
 407         ve->segno = segno;
 408
 409         rb_link_node(&ve->rb_node, parent, p);
 410         rb_insert_color_cached(&ve->rb_node, &am->root, left_most);
 411
 412         list_add_tail(&ve->list, &am->victim_list);
 413
 414         am->victim_count++;
 415
 416         return ve;
 417 }
 418
 419 static void insert_victim_entry(struct f2fs_sb_info *sbi,
 420                                 unsigned long long mtime, unsigned int segno)
 421 {
 422         struct atgc_management *am = &sbi->am;
 423         struct rb_node **p;
 424         struct rb_node *parent = NULL;
 425         bool left_most = true;
 426
 427         p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most);
 428         attach_victim_entry(sbi, mtime, segno, parent, p, left_most);
 429 }
 430
 431 static void add_victim_entry(struct f2fs_sb_info *sbi,
 432                                 struct victim_sel_policy *p, unsigned int segno)
 433 {
 434         struct sit_info *sit_i = SIT_I(sbi);
 435         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 436         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 437         unsigned long long mtime = 0;
 438         unsigned int i;
 439
 440         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 441                 if (p->gc_mode == GC_AT &&
 442                         get_valid_blocks(sbi, segno, true) == 0)
 443                         return;
 444         }
 445
 446         for (i = 0; i < sbi->segs_per_sec; i++)
 447                 mtime += get_seg_entry(sbi, start + i)->mtime;
 448         mtime = div_u64(mtime, sbi->segs_per_sec);
 449
 450         /* Handle if the system time has changed by the user */
 451         if (mtime < sit_i->min_mtime)
 452                 sit_i->min_mtime = mtime;
 453         if (mtime > sit_i->max_mtime)
 454                 sit_i->max_mtime = mtime;
 455         if (mtime < sit_i->dirty_min_mtime)
 456                 sit_i->dirty_min_mtime = mtime;
 457         if (mtime > sit_i->dirty_max_mtime)
 458                 sit_i->dirty_max_mtime = mtime;
 459
 460         /* don't choose young section as candidate */
 461         if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
 462                 return;
 463
 464         insert_victim_entry(sbi, mtime, segno);
 465 }
 466
 467 static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi,
 468                                                 struct victim_sel_policy *p)
 469 {
 470         struct atgc_management *am = &sbi->am;
 471         struct rb_node *parent = NULL;
 472         bool left_most;
 473
 474         f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most);
 475
 476         return parent;
 477 }
 478
 479 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
 480                                                 struct victim_sel_policy *p)
 481 {
 482         struct sit_info *sit_i = SIT_I(sbi);
 483         struct atgc_management *am = &sbi->am;
 484         struct rb_root_cached *root = &am->root;
 485         struct rb_node *node;
 486         struct rb_entry *re;
 487         struct victim_entry *ve;
 488         unsigned long long total_time;
 489         unsigned long long age, u, accu;
 490         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 491         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 492         unsigned int sec_blocks = CAP_BLKS_PER_SEC(sbi);
 493         unsigned int vblocks;
 494         unsigned int dirty_threshold = max(am->max_candidate_count,
 495                                         am->candidate_ratio *
 496                                         am->victim_count / 100);
 497         unsigned int age_weight = am->age_weight;
 498         unsigned int cost;
 499         unsigned int iter = 0;
 500
 501         if (max_mtime < min_mtime)
 502                 return;
 503
 504         max_mtime += 1;
 505         total_time = max_mtime - min_mtime;
 506
 507         accu = div64_u64(ULLONG_MAX, total_time);
 508         accu = min_t(unsigned long long, div_u64(accu, 100),
 509                                         DEFAULT_ACCURACY_CLASS);
 510
 511         node = rb_first_cached(root);
 512 next:
 513         re = rb_entry_safe(node, struct rb_entry, rb_node);
 514         if (!re)
 515                 return;
 516
 517         ve = (struct victim_entry *)re;
 518
 519         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 520                 goto skip;
 521
 522         /* age = 10000 * x% * 60 */
 523         age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
 524                                                                 age_weight;
 525
 526         vblocks = get_valid_blocks(sbi, ve->segno, true);
 527         f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
 528
 529         /* u = 10000 * x% * 40 */
 530         u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
 531                                                         (100 - age_weight);
 532
 533         f2fs_bug_on(sbi, age + u >= UINT_MAX);
 534
 535         cost = UINT_MAX - (age + u);
 536         iter++;
 537
 538         if (cost < p->min_cost ||
 539                         (cost == p->min_cost && age > p->oldest_age)) {
 540                 p->min_cost = cost;
 541                 p->oldest_age = age;
 542                 p->min_segno = ve->segno;
 543         }
 544 skip:
 545         if (iter < dirty_threshold) {
 546                 node = rb_next(node);
 547                 goto next;
 548         }
 549 }
 550
 551 /*
 552  * select candidates around source section in range of
 553  * [target - dirty_threshold, target + dirty_threshold]
 554  */
 555 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
 556                                                 struct victim_sel_policy *p)
 557 {
 558         struct sit_info *sit_i = SIT_I(sbi);
 559         struct atgc_management *am = &sbi->am;
 560         struct rb_node *node;
 561         struct rb_entry *re;
 562         struct victim_entry *ve;
 563         unsigned long long age;
 564         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 565         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 566         unsigned int seg_blocks = sbi->blocks_per_seg;
 567         unsigned int vblocks;
 568         unsigned int dirty_threshold = max(am->max_candidate_count,
 569                                         am->candidate_ratio *
 570                                         am->victim_count / 100);
 571         unsigned int cost;
 572         unsigned int iter = 0;
 573         int stage = 0;
 574
 575         if (max_mtime < min_mtime)
 576                 return;
 577         max_mtime += 1;
 578 next_stage:
 579         node = lookup_central_victim(sbi, p);
 580 next_node:
 581         re = rb_entry_safe(node, struct rb_entry, rb_node);
 582         if (!re) {
 583                 if (stage == 0)
 584                         goto skip_stage;
 585                 return;
 586         }
 587
 588         ve = (struct victim_entry *)re;
 589
 590         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 591                 goto skip_node;
 592
 593         age = max_mtime - ve->mtime;
 594
 595         vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
 596         f2fs_bug_on(sbi, !vblocks);
 597
 598         /* rare case */
 599         if (vblocks == seg_blocks)
 600                 goto skip_node;
 601
 602         iter++;
 603
 604         age = max_mtime - abs(p->age - age);
 605         cost = UINT_MAX - vblocks;
 606
 607         if (cost < p->min_cost ||
 608                         (cost == p->min_cost && age > p->oldest_age)) {
 609                 p->min_cost = cost;
 610                 p->oldest_age = age;
 611                 p->min_segno = ve->segno;
 612         }
 613 skip_node:
 614         if (iter < dirty_threshold) {
 615                 if (stage == 0)
 616                         node = rb_prev(node);
 617                 else if (stage == 1)
 618                         node = rb_next(node);
 619                 goto next_node;
 620         }
 621 skip_stage:
 622         if (stage < 1) {
 623                 stage++;
 624                 iter = 0;
 625                 goto next_stage;
 626         }
 627 }
 628 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
 629                                                 struct victim_sel_policy *p)
 630 {
 631         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
 632                                                 &sbi->am.root, true));
 633
 634         if (p->gc_mode == GC_AT)
 635                 atgc_lookup_victim(sbi, p);
 636         else if (p->alloc_mode == AT_SSR)
 637                 atssr_lookup_victim(sbi, p);
 638         else
 639                 f2fs_bug_on(sbi, 1);
 640 }
 641
 642 static void release_victim_entry(struct f2fs_sb_info *sbi)
 643 {
 644         struct atgc_management *am = &sbi->am;
 645         struct victim_entry *ve, *tmp;
 646
 647         list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
 648                 list_del(&ve->list);
 649                 kmem_cache_free(victim_entry_slab, ve);
 650                 am->victim_count--;
 651         }
 652
 653         am->root = RB_ROOT_CACHED;
 654
 655         f2fs_bug_on(sbi, am->victim_count);
 656         f2fs_bug_on(sbi, !list_empty(&am->victim_list));
 657 }
 658
 659 static bool f2fs_pin_section(struct f2fs_sb_info *sbi, unsigned int segno)
 660 {
 661         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 662         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 663
 664         if (!dirty_i->enable_pin_section)
 665                 return false;
 666         if (!test_and_set_bit(secno, dirty_i->pinned_secmap))
 667                 dirty_i->pinned_secmap_cnt++;
 668         return true;
 669 }
 670
 671 static bool f2fs_pinned_section_exists(struct dirty_seglist_info *dirty_i)
 672 {
 673         return dirty_i->pinned_secmap_cnt;
 674 }
 675
 676 static bool f2fs_section_is_pinned(struct dirty_seglist_info *dirty_i,
 677                                                 unsigned int secno)
 678 {
 679         return dirty_i->enable_pin_section &&
 680                 f2fs_pinned_section_exists(dirty_i) &&
 681                 test_bit(secno, dirty_i->pinned_secmap);
 682 }
 683
 684 static void f2fs_unpin_all_sections(struct f2fs_sb_info *sbi, bool enable)
 685 {
 686         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
 687
 688         if (f2fs_pinned_section_exists(DIRTY_I(sbi))) {
 689                 memset(DIRTY_I(sbi)->pinned_secmap, 0, bitmap_size);
 690                 DIRTY_I(sbi)->pinned_secmap_cnt = 0;
 691         }
 692         DIRTY_I(sbi)->enable_pin_section = enable;
 693 }
 694
 695 static int f2fs_gc_pinned_control(struct inode *inode, int gc_type,
 696                                                         unsigned int segno)
 697 {
 698         if (!f2fs_is_pinned_file(inode))
 699                 return 0;
 700         if (gc_type != FG_GC)
 701                 return -EBUSY;
 702         if (!f2fs_pin_section(F2FS_I_SB(inode), segno))
 703                 f2fs_pin_file_control(inode, true);
 704         return -EAGAIN;
 705 }
 706
 707 /*
 708  * This function is called from two paths.
 709  * One is garbage collection and the other is SSR segment selection.
 710  * When it is called during GC, it just gets a victim segment
 711  * and it does not remove it from dirty seglist.
 712  * When it is called from SSR segment selection, it finds a segment
 713  * which has minimum valid blocks and removes it from dirty seglist.
 714  */
 715 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 716                         unsigned int *result, int gc_type, int type,
 717                         char alloc_mode, unsigned long long age)
 718 {
 719         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 720         struct sit_info *sm = SIT_I(sbi);
 721         struct victim_sel_policy p;
 722         unsigned int secno, last_victim;
 723         unsigned int last_segment;
 724         unsigned int nsearched;
 725         bool is_atgc;
 726         int ret = 0;
 727
 728         mutex_lock(&dirty_i->seglist_lock);
 729         last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
 730
 731         p.alloc_mode = alloc_mode;
 732         p.age = age;
 733         p.age_threshold = sbi->am.age_threshold;
 734
 735 retry:
 736         select_policy(sbi, gc_type, type, &p);
 737         p.min_segno = NULL_SEGNO;
 738         p.oldest_age = 0;
 739         p.min_cost = get_max_cost(sbi, &p);
 740
 741         is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
 742         nsearched = 0;
 743
 744         if (is_atgc)
 745                 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
 746
 747         if (*result != NULL_SEGNO) {
 748                 if (!get_valid_blocks(sbi, *result, false)) {
 749                         ret = -ENODATA;
 750                         goto out;
 751                 }
 752
 753                 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
 754                         ret = -EBUSY;
 755                 else
 756                         p.min_segno = *result;
 757                 goto out;
 758         }
 759
 760         ret = -ENODATA;
 761         if (p.max_search == 0)
 762                 goto out;
 763
 764         if (__is_large_section(sbi) && p.alloc_mode == LFS) {
 765                 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
 766                         p.min_segno = sbi->next_victim_seg[BG_GC];
 767                         *result = p.min_segno;
 768                         sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
 769                         goto got_result;
 770                 }
 771                 if (gc_type == FG_GC &&
 772                                 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
 773                         p.min_segno = sbi->next_victim_seg[FG_GC];
 774                         *result = p.min_segno;
 775                         sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
 776                         goto got_result;
 777                 }
 778         }
 779
 780         last_victim = sm->last_victim[p.gc_mode];
 781         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 782                 p.min_segno = check_bg_victims(sbi);
 783                 if (p.min_segno != NULL_SEGNO)
 784                         goto got_it;
 785         }
 786
 787         while (1) {
 788                 unsigned long cost, *dirty_bitmap;
 789                 unsigned int unit_no, segno;
 790
 791                 dirty_bitmap = p.dirty_bitmap;
 792                 unit_no = find_next_bit(dirty_bitmap,
 793                                 last_segment / p.ofs_unit,
 794                                 p.offset / p.ofs_unit);
 795                 segno = unit_no * p.ofs_unit;
 796                 if (segno >= last_segment) {
 797                         if (sm->last_victim[p.gc_mode]) {
 798                                 last_segment =
 799                                         sm->last_victim[p.gc_mode];
 800                                 sm->last_victim[p.gc_mode] = 0;
 801                                 p.offset = 0;
 802                                 continue;
 803                         }
 804                         break;
 805                 }
 806
 807                 p.offset = segno + p.ofs_unit;
 808                 nsearched++;
 809
 810 #ifdef CONFIG_F2FS_CHECK_FS
 811                 /*
 812                  * skip selecting the invalid segno (that is failed due to block
 813                  * validity check failure during GC) to avoid endless GC loop in
 814                  * such cases.
 815                  */
 816                 if (test_bit(segno, sm->invalid_segmap))
 817                         goto next;
 818 #endif
 819
 820                 secno = GET_SEC_FROM_SEG(sbi, segno);
 821
 822                 if (sec_usage_check(sbi, secno))
 823                         goto next;
 824
 825                 /* Don't touch checkpointed data */
 826                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 827                         if (p.alloc_mode == LFS) {
 828                                 /*
 829                                  * LFS is set to find source section during GC.
 830                                  * The victim should have no checkpointed data.
 831                                  */
 832                                 if (get_ckpt_valid_blocks(sbi, segno, true))
 833                                         goto next;
 834                         } else {
 835                                 /*
 836                                  * SSR | AT_SSR are set to find target segment
 837                                  * for writes which can be full by checkpointed
 838                                  * and newly written blocks.
 839                                  */
 840                                 if (!f2fs_segment_has_free_slot(sbi, segno))
 841                                         goto next;
 842                         }
 843                 }
 844
 845                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 846                         goto next;
 847
 848                 if (gc_type == FG_GC && f2fs_section_is_pinned(dirty_i, secno))
 849                         goto next;
 850
 851                 if (is_atgc) {
 852                         add_victim_entry(sbi, &p, segno);
 853                         goto next;
 854                 }
 855
 856                 cost = get_gc_cost(sbi, segno, &p);
 857
 858                 if (p.min_cost > cost) {
 859                         p.min_segno = segno;
 860                         p.min_cost = cost;
 861                 }
 862 next:
 863                 if (nsearched >= p.max_search) {
 864                         if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
 865                                 sm->last_victim[p.gc_mode] =
 866                                         last_victim + p.ofs_unit;
 867                         else
 868                                 sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
 869                         sm->last_victim[p.gc_mode] %=
 870                                 (MAIN_SECS(sbi) * sbi->segs_per_sec);
 871                         break;
 872                 }
 873         }
 874
 875         /* get victim for GC_AT/AT_SSR */
 876         if (is_atgc) {
 877                 lookup_victim_by_age(sbi, &p);
 878                 release_victim_entry(sbi);
 879         }
 880
 881         if (is_atgc && p.min_segno == NULL_SEGNO &&
 882                         sm->elapsed_time < p.age_threshold) {
 883                 p.age_threshold = 0;
 884                 goto retry;
 885         }
 886
 887         if (p.min_segno != NULL_SEGNO) {
 888 got_it:
 889                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 890 got_result:
 891                 if (p.alloc_mode == LFS) {
 892                         secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
 893                         if (gc_type == FG_GC)
 894                                 sbi->cur_victim_sec = secno;
 895                         else
 896                                 set_bit(secno, dirty_i->victim_secmap);
 897                 }
 898                 ret = 0;
 899
 900         }
 901 out:
 902         if (p.min_segno != NULL_SEGNO)
 903                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 904                                 sbi->cur_victim_sec,
 905                                 prefree_segments(sbi), free_segments(sbi));
 906         mutex_unlock(&dirty_i->seglist_lock);
 907
 908         return ret;
 909 }
 910
 911 static const struct victim_selection default_v_ops = {
 912         .get_victim = get_victim_by_default,
 913 };
 914
 915 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
 916 {
 917         struct inode_entry *ie;
 918
 919         ie = radix_tree_lookup(&gc_list->iroot, ino);
 920         if (ie)
 921                 return ie->inode;
 922         return NULL;
 923 }
 924
 925 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
 926 {
 927         struct inode_entry *new_ie;
 928
 929         if (inode == find_gc_inode(gc_list, inode->i_ino)) {
 930                 iput(inode);
 931                 return;
 932         }
 933         new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab,
 934                                         GFP_NOFS, true, NULL);
 935         new_ie->inode = inode;
 936
 937         f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
 938         list_add_tail(&new_ie->list, &gc_list->ilist);
 939 }
 940
 941 static void put_gc_inode(struct gc_inode_list *gc_list)
 942 {
 943         struct inode_entry *ie, *next_ie;
 944
 945         list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
 946                 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
 947                 iput(ie->inode);
 948                 list_del(&ie->list);
 949                 kmem_cache_free(f2fs_inode_entry_slab, ie);
 950         }
 951 }
 952
 953 static int check_valid_map(struct f2fs_sb_info *sbi,
 954                                 unsigned int segno, int offset)
 955 {
 956         struct sit_info *sit_i = SIT_I(sbi);
 957         struct seg_entry *sentry;
 958         int ret;
 959
 960         down_read(&sit_i->sentry_lock);
 961         sentry = get_seg_entry(sbi, segno);
 962         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 963         up_read(&sit_i->sentry_lock);
 964         return ret;
 965 }
 966
 967 /*
 968  * This function compares node address got in summary with that in NAT.
 969  * On validity, copy that node with cold status, otherwise (invalid node)
 970  * ignore that.
 971  */
 972 static int gc_node_segment(struct f2fs_sb_info *sbi,
 973                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 974 {
 975         struct f2fs_summary *entry;
 976         block_t start_addr;
 977         int off;
 978         int phase = 0;
 979         bool fggc = (gc_type == FG_GC);
 980         int submitted = 0;
 981         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
 982
 983         start_addr = START_BLOCK(sbi, segno);
 984
 985 next_step:
 986         entry = sum;
 987
 988         if (fggc && phase == 2)
 989                 atomic_inc(&sbi->wb_sync_req[NODE]);
 990
 991         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
 992                 nid_t nid = le32_to_cpu(entry->nid);
 993                 struct page *node_page;
 994                 struct node_info ni;
 995                 int err;
 996
 997                 /* stop BG_GC if there is not enough free sections. */
 998                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 999                         return submitted;
1000
1001                 if (check_valid_map(sbi, segno, off) == 0)
1002                         continue;
1003
1004                 if (phase == 0) {
1005                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1006                                                         META_NAT, true);
1007                         continue;
1008                 }
1009
1010                 if (phase == 1) {
1011                         f2fs_ra_node_page(sbi, nid);
1012                         continue;
1013                 }
1014
1015                 /* phase == 2 */
1016                 node_page = f2fs_get_node_page(sbi, nid);
1017                 if (IS_ERR(node_page))
1018                         continue;
1019
1020                 /* block may become invalid during f2fs_get_node_page */
1021                 if (check_valid_map(sbi, segno, off) == 0) {
1022                         f2fs_put_page(node_page, 1);
1023                         continue;
1024                 }
1025
1026                 if (f2fs_get_node_info(sbi, nid, &ni, false)) {
1027                         f2fs_put_page(node_page, 1);
1028                         continue;
1029                 }
1030
1031                 if (ni.blk_addr != start_addr + off) {
1032                         f2fs_put_page(node_page, 1);
1033                         continue;
1034                 }
1035
1036                 err = f2fs_move_node_page(node_page, gc_type);
1037                 if (!err && gc_type == FG_GC)
1038                         submitted++;
1039                 stat_inc_node_blk_count(sbi, 1, gc_type);
1040         }
1041
1042         if (++phase < 3)
1043                 goto next_step;
1044
1045         if (fggc)
1046                 atomic_dec(&sbi->wb_sync_req[NODE]);
1047         return submitted;
1048 }
1049
1050 /*
1051  * Calculate start block index indicating the given node offset.
1052  * Be careful, caller should give this node offset only indicating direct node
1053  * blocks. If any node offsets, which point the other types of node blocks such
1054  * as indirect or double indirect node blocks, are given, it must be a caller's
1055  * bug.
1056  */
1057 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
1058 {
1059         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
1060         unsigned int bidx;
1061
1062         if (node_ofs == 0)
1063                 return 0;
1064
1065         if (node_ofs <= 2) {
1066                 bidx = node_ofs - 1;
1067         } else if (node_ofs <= indirect_blks) {
1068                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
1069
1070                 bidx = node_ofs - 2 - dec;
1071         } else {
1072                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
1073
1074                 bidx = node_ofs - 5 - dec;
1075         }
1076         return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
1077 }
1078
1079 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1080                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
1081 {
1082         struct page *node_page;
1083         nid_t nid;
1084         unsigned int ofs_in_node, max_addrs, base;
1085         block_t source_blkaddr;
1086
1087         nid = le32_to_cpu(sum->nid);
1088         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
1089
1090         node_page = f2fs_get_node_page(sbi, nid);
1091         if (IS_ERR(node_page))
1092                 return false;
1093
1094         if (f2fs_get_node_info(sbi, nid, dni, false)) {
1095                 f2fs_put_page(node_page, 1);
1096                 return false;
1097         }
1098
1099         if (sum->version != dni->version) {
1100                 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
1101                           __func__);
1102                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1103         }
1104
1105         if (f2fs_check_nid_range(sbi, dni->ino)) {
1106                 f2fs_put_page(node_page, 1);
1107                 return false;
1108         }
1109
1110         if (IS_INODE(node_page)) {
1111                 base = offset_in_addr(F2FS_INODE(node_page));
1112                 max_addrs = DEF_ADDRS_PER_INODE;
1113         } else {
1114                 base = 0;
1115                 max_addrs = DEF_ADDRS_PER_BLOCK;
1116         }
1117
1118         if (base + ofs_in_node >= max_addrs) {
1119                 f2fs_err(sbi, "Inconsistent blkaddr offset: base:%u, ofs_in_node:%u, max:%u, ino:%u, nid:%u",
1120                         base, ofs_in_node, max_addrs, dni->ino, dni->nid);
1121                 f2fs_put_page(node_page, 1);
1122                 return false;
1123         }
1124
1125         *nofs = ofs_of_node(node_page);
1126         source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1127         f2fs_put_page(node_page, 1);
1128
1129         if (source_blkaddr != blkaddr) {
1130 #ifdef CONFIG_F2FS_CHECK_FS
1131                 unsigned int segno = GET_SEGNO(sbi, blkaddr);
1132                 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1133
1134                 if (unlikely(check_valid_map(sbi, segno, offset))) {
1135                         if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1136                                 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",
1137                                          blkaddr, source_blkaddr, segno);
1138                                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1139                         }
1140                 }
1141 #endif
1142                 return false;
1143         }
1144         return true;
1145 }
1146
1147 static int ra_data_block(struct inode *inode, pgoff_t index)
1148 {
1149         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1150         struct address_space *mapping = inode->i_mapping;
1151         struct dnode_of_data dn;
1152         struct page *page;
1153         struct extent_info ei = {0, };
1154         struct f2fs_io_info fio = {
1155                 .sbi = sbi,
1156                 .ino = inode->i_ino,
1157                 .type = DATA,
1158                 .temp = COLD,
1159                 .op = REQ_OP_READ,
1160                 .op_flags = 0,
1161                 .encrypted_page = NULL,
1162                 .in_list = false,
1163                 .retry = false,
1164         };
1165         int err;
1166
1167         page = f2fs_grab_cache_page(mapping, index, true);
1168         if (!page)
1169                 return -ENOMEM;
1170
1171         if (f2fs_lookup_read_extent_cache(inode, index, &ei)) {
1172                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1173                 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1174                                                 DATA_GENERIC_ENHANCE_READ))) {
1175                         err = -EFSCORRUPTED;
1176                         f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1177                         goto put_page;
1178                 }
1179                 goto got_it;
1180         }
1181
1182         set_new_dnode(&dn, inode, NULL, NULL, 0);
1183         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1184         if (err)
1185                 goto put_page;
1186         f2fs_put_dnode(&dn);
1187
1188         if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1189                 err = -ENOENT;
1190                 goto put_page;
1191         }
1192         if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1193                                                 DATA_GENERIC_ENHANCE))) {
1194                 err = -EFSCORRUPTED;
1195                 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1196                 goto put_page;
1197         }
1198 got_it:
1199         /* read page */
1200         fio.page = page;
1201         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1202
1203         /*
1204          * don't cache encrypted data into meta inode until previous dirty
1205          * data were writebacked to avoid racing between GC and flush.
1206          */
1207         f2fs_wait_on_page_writeback(page, DATA, true, true);
1208
1209         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1210
1211         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1212                                         dn.data_blkaddr,
1213                                         FGP_LOCK | FGP_CREAT, GFP_NOFS);
1214         if (!fio.encrypted_page) {
1215                 err = -ENOMEM;
1216                 goto put_page;
1217         }
1218
1219         err = f2fs_submit_page_bio(&fio);
1220         if (err)
1221                 goto put_encrypted_page;
1222         f2fs_put_page(fio.encrypted_page, 0);
1223         f2fs_put_page(page, 1);
1224
1225         f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
1226         f2fs_update_iostat(sbi, NULL, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1227
1228         return 0;
1229 put_encrypted_page:
1230         f2fs_put_page(fio.encrypted_page, 1);
1231 put_page:
1232         f2fs_put_page(page, 1);
1233         return err;
1234 }
1235
1236 /*
1237  * Move data block via META_MAPPING while keeping locked data page.
1238  * This can be used to move blocks, aka LBAs, directly on disk.
1239  */
1240 static int move_data_block(struct inode *inode, block_t bidx,
1241                                 int gc_type, unsigned int segno, int off)
1242 {
1243         struct f2fs_io_info fio = {
1244                 .sbi = F2FS_I_SB(inode),
1245                 .ino = inode->i_ino,
1246                 .type = DATA,
1247                 .temp = COLD,
1248                 .op = REQ_OP_READ,
1249                 .op_flags = 0,
1250                 .encrypted_page = NULL,
1251                 .in_list = false,
1252                 .retry = false,
1253         };
1254         struct dnode_of_data dn;
1255         struct f2fs_summary sum;
1256         struct node_info ni;
1257         struct page *page, *mpage;
1258         block_t newaddr;
1259         int err = 0;
1260         bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1261         int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&
1262                                 (fio.sbi->gc_mode != GC_URGENT_HIGH) ?
1263                                 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1264
1265         /* do not read out */
1266         page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1267         if (!page)
1268                 return -ENOMEM;
1269
1270         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1271                 err = -ENOENT;
1272                 goto out;
1273         }
1274
1275         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1276         if (err)
1277                 goto out;
1278
1279         set_new_dnode(&dn, inode, NULL, NULL, 0);
1280         err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1281         if (err)
1282                 goto out;
1283
1284         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1285                 ClearPageUptodate(page);
1286                 err = -ENOENT;
1287                 goto put_out;
1288         }
1289
1290         /*
1291          * don't cache encrypted data into meta inode until previous dirty
1292          * data were writebacked to avoid racing between GC and flush.
1293          */
1294         f2fs_wait_on_page_writeback(page, DATA, true, true);
1295
1296         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1297
1298         err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);
1299         if (err)
1300                 goto put_out;
1301
1302         /* read page */
1303         fio.page = page;
1304         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1305
1306         if (lfs_mode)
1307                 f2fs_down_write(&fio.sbi->io_order_lock);
1308
1309         mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1310                                         fio.old_blkaddr, false);
1311         if (!mpage) {
1312                 err = -ENOMEM;
1313                 goto up_out;
1314         }
1315
1316         fio.encrypted_page = mpage;
1317
1318         /* read source block in mpage */
1319         if (!PageUptodate(mpage)) {
1320                 err = f2fs_submit_page_bio(&fio);
1321                 if (err) {
1322                         f2fs_put_page(mpage, 1);
1323                         goto up_out;
1324                 }
1325
1326                 f2fs_update_iostat(fio.sbi, inode, FS_DATA_READ_IO,
1327                                                         F2FS_BLKSIZE);
1328                 f2fs_update_iostat(fio.sbi, NULL, FS_GDATA_READ_IO,
1329                                                         F2FS_BLKSIZE);
1330
1331                 lock_page(mpage);
1332                 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1333                                                 !PageUptodate(mpage))) {
1334                         err = -EIO;
1335                         f2fs_put_page(mpage, 1);
1336                         goto up_out;
1337                 }
1338         }
1339
1340         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1341
1342         /* allocate block address */
1343         f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1344                                 &sum, type, NULL);
1345
1346         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1347                                 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1348         if (!fio.encrypted_page) {
1349                 err = -ENOMEM;
1350                 f2fs_put_page(mpage, 1);
1351                 goto recover_block;
1352         }
1353
1354         /* write target block */
1355         f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1356         memcpy(page_address(fio.encrypted_page),
1357                                 page_address(mpage), PAGE_SIZE);
1358         f2fs_put_page(mpage, 1);
1359         invalidate_mapping_pages(META_MAPPING(fio.sbi),
1360                                 fio.old_blkaddr, fio.old_blkaddr);
1361         f2fs_invalidate_compress_page(fio.sbi, fio.old_blkaddr);
1362
1363         set_page_dirty(fio.encrypted_page);
1364         if (clear_page_dirty_for_io(fio.encrypted_page))
1365                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
1366
1367         set_page_writeback(fio.encrypted_page);
1368         ClearPageError(page);
1369
1370         fio.op = REQ_OP_WRITE;
1371         fio.op_flags = REQ_SYNC;
1372         fio.new_blkaddr = newaddr;
1373         f2fs_submit_page_write(&fio);
1374         if (fio.retry) {
1375                 err = -EAGAIN;
1376                 if (PageWriteback(fio.encrypted_page))
1377                         end_page_writeback(fio.encrypted_page);
1378                 goto put_page_out;
1379         }
1380
1381         f2fs_update_iostat(fio.sbi, NULL, FS_GC_DATA_IO, F2FS_BLKSIZE);
1382
1383         f2fs_update_data_blkaddr(&dn, newaddr);
1384         set_inode_flag(inode, FI_APPEND_WRITE);
1385         if (page->index == 0)
1386                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1387 put_page_out:
1388         f2fs_put_page(fio.encrypted_page, 1);
1389 recover_block:
1390         if (err)
1391                 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1392                                                         true, true, true);
1393 up_out:
1394         if (lfs_mode)
1395                 f2fs_up_write(&fio.sbi->io_order_lock);
1396 put_out:
1397         f2fs_put_dnode(&dn);
1398 out:
1399         f2fs_put_page(page, 1);
1400         return err;
1401 }
1402
1403 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1404                                                         unsigned int segno, int off)
1405 {
1406         struct page *page;
1407         int err = 0;
1408
1409         page = f2fs_get_lock_data_page(inode, bidx, true);
1410         if (IS_ERR(page))
1411                 return PTR_ERR(page);
1412
1413         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1414                 err = -ENOENT;
1415                 goto out;
1416         }
1417
1418         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1419         if (err)
1420                 goto out;
1421
1422         if (gc_type == BG_GC) {
1423                 if (PageWriteback(page)) {
1424                         err = -EAGAIN;
1425                         goto out;
1426                 }
1427                 set_page_dirty(page);
1428                 set_page_private_gcing(page);
1429         } else {
1430                 struct f2fs_io_info fio = {
1431                         .sbi = F2FS_I_SB(inode),
1432                         .ino = inode->i_ino,
1433                         .type = DATA,
1434                         .temp = COLD,
1435                         .op = REQ_OP_WRITE,
1436                         .op_flags = REQ_SYNC,
1437                         .old_blkaddr = NULL_ADDR,
1438                         .page = page,
1439                         .encrypted_page = NULL,
1440                         .need_lock = LOCK_REQ,
1441                         .io_type = FS_GC_DATA_IO,
1442                 };
1443                 bool is_dirty = PageDirty(page);
1444
1445 retry:
1446                 f2fs_wait_on_page_writeback(page, DATA, true, true);
1447
1448                 set_page_dirty(page);
1449                 if (clear_page_dirty_for_io(page)) {
1450                         inode_dec_dirty_pages(inode);
1451                         f2fs_remove_dirty_inode(inode);
1452                 }
1453
1454                 set_page_private_gcing(page);
1455
1456                 err = f2fs_do_write_data_page(&fio);
1457                 if (err) {
1458                         clear_page_private_gcing(page);
1459                         if (err == -ENOMEM) {
1460                                 memalloc_retry_wait(GFP_NOFS);
1461                                 goto retry;
1462                         }
1463                         if (is_dirty)
1464                                 set_page_dirty(page);
1465                 }
1466         }
1467 out:
1468         f2fs_put_page(page, 1);
1469         return err;
1470 }
1471
1472 /*
1473  * This function tries to get parent node of victim data block, and identifies
1474  * data block validity. If the block is valid, copy that with cold status and
1475  * modify parent node.
1476  * If the parent node is not valid or the data block address is different,
1477  * the victim data block is ignored.
1478  */
1479 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1480                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1481                 bool force_migrate)
1482 {
1483         struct super_block *sb = sbi->sb;
1484         struct f2fs_summary *entry;
1485         block_t start_addr;
1486         int off;
1487         int phase = 0;
1488         int submitted = 0;
1489         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1490
1491         start_addr = START_BLOCK(sbi, segno);
1492
1493 next_step:
1494         entry = sum;
1495
1496         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1497                 struct page *data_page;
1498                 struct inode *inode;
1499                 struct node_info dni; /* dnode info for the data */
1500                 unsigned int ofs_in_node, nofs;
1501                 block_t start_bidx;
1502                 nid_t nid = le32_to_cpu(entry->nid);
1503
1504                 /*
1505                  * stop BG_GC if there is not enough free sections.
1506                  * Or, stop GC if the segment becomes fully valid caused by
1507                  * race condition along with SSR block allocation.
1508                  */
1509                 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1510                         (!force_migrate && get_valid_blocks(sbi, segno, true) ==
1511                                                         CAP_BLKS_PER_SEC(sbi)))
1512                         return submitted;
1513
1514                 if (check_valid_map(sbi, segno, off) == 0)
1515                         continue;
1516
1517                 if (phase == 0) {
1518                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1519                                                         META_NAT, true);
1520                         continue;
1521                 }
1522
1523                 if (phase == 1) {
1524                         f2fs_ra_node_page(sbi, nid);
1525                         continue;
1526                 }
1527
1528                 /* Get an inode by ino with checking validity */
1529                 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1530                         continue;
1531
1532                 if (phase == 2) {
1533                         f2fs_ra_node_page(sbi, dni.ino);
1534                         continue;
1535                 }
1536
1537                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1538
1539                 if (phase == 3) {
1540                         int err;
1541
1542                         inode = f2fs_iget(sb, dni.ino);
1543                         if (IS_ERR(inode) || is_bad_inode(inode) ||
1544                                         special_file(inode->i_mode))
1545                                 continue;
1546
1547                         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1548                         if (err == -EAGAIN) {
1549                                 iput(inode);
1550                                 return submitted;
1551                         }
1552
1553                         if (!f2fs_down_write_trylock(
1554                                 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1555                                 iput(inode);
1556                                 sbi->skipped_gc_rwsem++;
1557                                 continue;
1558                         }
1559
1560                         start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1561                                                                 ofs_in_node;
1562
1563                         if (f2fs_post_read_required(inode)) {
1564                                 int err = ra_data_block(inode, start_bidx);
1565
1566                                 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1567                                 if (err) {
1568                                         iput(inode);
1569                                         continue;
1570                                 }
1571                                 add_gc_inode(gc_list, inode);
1572                                 continue;
1573                         }
1574
1575                         data_page = f2fs_get_read_data_page(inode, start_bidx,
1576                                                         REQ_RAHEAD, true, NULL);
1577                         f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1578                         if (IS_ERR(data_page)) {
1579                                 iput(inode);
1580                                 continue;
1581                         }
1582
1583                         f2fs_put_page(data_page, 0);
1584                         add_gc_inode(gc_list, inode);
1585                         continue;
1586                 }
1587
1588                 /* phase 4 */
1589                 inode = find_gc_inode(gc_list, dni.ino);
1590                 if (inode) {
1591                         struct f2fs_inode_info *fi = F2FS_I(inode);
1592                         bool locked = false;
1593                         int err;
1594
1595                         if (S_ISREG(inode->i_mode)) {
1596                                 if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[READ])) {
1597                                         sbi->skipped_gc_rwsem++;
1598                                         continue;
1599                                 }
1600                                 if (!f2fs_down_write_trylock(
1601                                                 &fi->i_gc_rwsem[WRITE])) {
1602                                         sbi->skipped_gc_rwsem++;
1603                                         f2fs_up_write(&fi->i_gc_rwsem[READ]);
1604                                         continue;
1605                                 }
1606                                 locked = true;
1607
1608                                 /* wait for all inflight aio data */
1609                                 inode_dio_wait(inode);
1610                         }
1611
1612                         start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1613                                                                 + ofs_in_node;
1614                         if (f2fs_post_read_required(inode))
1615                                 err = move_data_block(inode, start_bidx,
1616                                                         gc_type, segno, off);
1617                         else
1618                                 err = move_data_page(inode, start_bidx, gc_type,
1619                                                                 segno, off);
1620
1621                         if (!err && (gc_type == FG_GC ||
1622                                         f2fs_post_read_required(inode)))
1623                                 submitted++;
1624
1625                         if (locked) {
1626                                 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
1627                                 f2fs_up_write(&fi->i_gc_rwsem[READ]);
1628                         }
1629
1630                         stat_inc_data_blk_count(sbi, 1, gc_type);
1631                 }
1632         }
1633
1634         if (++phase < 5)
1635                 goto next_step;
1636
1637         return submitted;
1638 }
1639
1640 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1641                         int gc_type)
1642 {
1643         struct sit_info *sit_i = SIT_I(sbi);
1644         int ret;
1645
1646         down_write(&sit_i->sentry_lock);
1647         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1648                                               NO_CHECK_TYPE, LFS, 0);
1649         up_write(&sit_i->sentry_lock);
1650         return ret;
1651 }
1652
1653 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1654                                 unsigned int start_segno,
1655                                 struct gc_inode_list *gc_list, int gc_type,
1656                                 bool force_migrate)
1657 {
1658         struct page *sum_page;
1659         struct f2fs_summary_block *sum;
1660         struct blk_plug plug;
1661         unsigned int segno = start_segno;
1662         unsigned int end_segno = start_segno + sbi->segs_per_sec;
1663         int seg_freed = 0, migrated = 0;
1664         unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1665                                                 SUM_TYPE_DATA : SUM_TYPE_NODE;
1666         int submitted = 0;
1667
1668         if (__is_large_section(sbi))
1669                 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1670
1671         /*
1672          * zone-capacity can be less than zone-size in zoned devices,
1673          * resulting in less than expected usable segments in the zone,
1674          * calculate the end segno in the zone which can be garbage collected
1675          */
1676         if (f2fs_sb_has_blkzoned(sbi))
1677                 end_segno -= sbi->segs_per_sec -
1678                                         f2fs_usable_segs_in_sec(sbi, segno);
1679
1680         sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1681
1682         /* readahead multi ssa blocks those have contiguous address */
1683         if (__is_large_section(sbi))
1684                 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1685                                         end_segno - segno, META_SSA, true);
1686
1687         /* reference all summary page */
1688         while (segno < end_segno) {
1689                 sum_page = f2fs_get_sum_page(sbi, segno++);
1690                 if (IS_ERR(sum_page)) {
1691                         int err = PTR_ERR(sum_page);
1692
1693                         end_segno = segno - 1;
1694                         for (segno = start_segno; segno < end_segno; segno++) {
1695                                 sum_page = find_get_page(META_MAPPING(sbi),
1696                                                 GET_SUM_BLOCK(sbi, segno));
1697                                 f2fs_put_page(sum_page, 0);
1698                                 f2fs_put_page(sum_page, 0);
1699                         }
1700                         return err;
1701                 }
1702                 unlock_page(sum_page);
1703         }
1704
1705         blk_start_plug(&plug);
1706
1707         for (segno = start_segno; segno < end_segno; segno++) {
1708
1709                 /* find segment summary of victim */
1710                 sum_page = find_get_page(META_MAPPING(sbi),
1711                                         GET_SUM_BLOCK(sbi, segno));
1712                 f2fs_put_page(sum_page, 0);
1713
1714                 if (get_valid_blocks(sbi, segno, false) == 0)
1715                         goto freed;
1716                 if (gc_type == BG_GC && __is_large_section(sbi) &&
1717                                 migrated >= sbi->migration_granularity)
1718                         goto skip;
1719                 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1720                         goto skip;
1721
1722                 sum = page_address(sum_page);
1723                 if (type != GET_SUM_TYPE((&sum->footer))) {
1724                         f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1725                                  segno, type, GET_SUM_TYPE((&sum->footer)));
1726                         set_sbi_flag(sbi, SBI_NEED_FSCK);
1727                         f2fs_stop_checkpoint(sbi, false,
1728                                 STOP_CP_REASON_CORRUPTED_SUMMARY);
1729                         goto skip;
1730                 }
1731
1732                 /*
1733                  * this is to avoid deadlock:
1734                  * - lock_page(sum_page)         - f2fs_replace_block
1735                  *  - check_valid_map()            - down_write(sentry_lock)
1736                  *   - down_read(sentry_lock)     - change_curseg()
1737                  *                                  - lock_page(sum_page)
1738                  */
1739                 if (type == SUM_TYPE_NODE)
1740                         submitted += gc_node_segment(sbi, sum->entries, segno,
1741                                                                 gc_type);
1742                 else
1743                         submitted += gc_data_segment(sbi, sum->entries, gc_list,
1744                                                         segno, gc_type,
1745                                                         force_migrate);
1746
1747                 stat_inc_seg_count(sbi, type, gc_type);
1748                 sbi->gc_reclaimed_segs[sbi->gc_mode]++;
1749                 migrated++;
1750
1751 freed:
1752                 if (gc_type == FG_GC &&
1753                                 get_valid_blocks(sbi, segno, false) == 0)
1754                         seg_freed++;
1755
1756                 if (__is_large_section(sbi))
1757                         sbi->next_victim_seg[gc_type] =
1758                                 (segno + 1 < end_segno) ? segno + 1 : NULL_SEGNO;
1759 skip:
1760                 f2fs_put_page(sum_page, 0);
1761         }
1762
1763         if (submitted)
1764                 f2fs_submit_merged_write(sbi,
1765                                 (type == SUM_TYPE_NODE) ? NODE : DATA);
1766
1767         blk_finish_plug(&plug);
1768
1769         stat_inc_call_count(sbi->stat_info);
1770
1771         return seg_freed;
1772 }
1773
1774 int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control)
1775 {
1776         int gc_type = gc_control->init_gc_type;
1777         unsigned int segno = gc_control->victim_segno;
1778         int sec_freed = 0, seg_freed = 0, total_freed = 0;
1779         int ret = 0;
1780         struct cp_control cpc;
1781         struct gc_inode_list gc_list = {
1782                 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1783                 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1784         };
1785         unsigned int skipped_round = 0, round = 0;
1786
1787         trace_f2fs_gc_begin(sbi->sb, gc_type, gc_control->no_bg_gc,
1788                                 gc_control->nr_free_secs,
1789                                 get_pages(sbi, F2FS_DIRTY_NODES),
1790                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1791                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1792                                 free_sections(sbi),
1793                                 free_segments(sbi),
1794                                 reserved_segments(sbi),
1795                                 prefree_segments(sbi));
1796
1797         cpc.reason = __get_cp_reason(sbi);
1798         sbi->skipped_gc_rwsem = 0;
1799 gc_more:
1800         if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1801                 ret = -EINVAL;
1802                 goto stop;
1803         }
1804         if (unlikely(f2fs_cp_error(sbi))) {
1805                 ret = -EIO;
1806                 goto stop;
1807         }
1808
1809         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1810                 /*
1811                  * For example, if there are many prefree_segments below given
1812                  * threshold, we can make them free by checkpoint. Then, we
1813                  * secure free segments which doesn't need fggc any more.
1814                  */
1815                 if (prefree_segments(sbi)) {
1816                         ret = f2fs_write_checkpoint(sbi, &cpc);
1817                         if (ret)
1818                                 goto stop;
1819                 }
1820                 if (has_not_enough_free_secs(sbi, 0, 0))
1821                         gc_type = FG_GC;
1822         }
1823
1824         /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1825         if (gc_type == BG_GC && gc_control->no_bg_gc) {
1826                 ret = -EINVAL;
1827                 goto stop;
1828         }
1829 retry:
1830         ret = __get_victim(sbi, &segno, gc_type);
1831         if (ret) {
1832                 /* allow to search victim from sections has pinned data */
1833                 if (ret == -ENODATA && gc_type == FG_GC &&
1834                                 f2fs_pinned_section_exists(DIRTY_I(sbi))) {
1835                         f2fs_unpin_all_sections(sbi, false);
1836                         goto retry;
1837                 }
1838                 goto stop;
1839         }
1840
1841         seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type,
1842                                 gc_control->should_migrate_blocks);
1843         total_freed += seg_freed;
1844
1845         if (seg_freed == f2fs_usable_segs_in_sec(sbi, segno))
1846                 sec_freed++;
1847
1848         if (gc_type == FG_GC)
1849                 sbi->cur_victim_sec = NULL_SEGNO;
1850
1851         if (gc_control->init_gc_type == FG_GC ||
1852             !has_not_enough_free_secs(sbi,
1853                                 (gc_type == FG_GC) ? sec_freed : 0, 0)) {
1854                 if (gc_type == FG_GC && sec_freed < gc_control->nr_free_secs)
1855                         goto go_gc_more;
1856                 goto stop;
1857         }
1858
1859         /* FG_GC stops GC by skip_count */
1860         if (gc_type == FG_GC) {
1861                 if (sbi->skipped_gc_rwsem)
1862                         skipped_round++;
1863                 round++;
1864                 if (skipped_round > MAX_SKIP_GC_COUNT &&
1865                                 skipped_round * 2 >= round) {
1866                         ret = f2fs_write_checkpoint(sbi, &cpc);
1867                         goto stop;
1868                 }
1869         }
1870
1871         /* Write checkpoint to reclaim prefree segments */
1872         if (free_sections(sbi) < NR_CURSEG_PERSIST_TYPE &&
1873                                 prefree_segments(sbi)) {
1874                 ret = f2fs_write_checkpoint(sbi, &cpc);
1875                 if (ret)
1876                         goto stop;
1877         }
1878 go_gc_more:
1879         segno = NULL_SEGNO;
1880         goto gc_more;
1881
1882 stop:
1883         SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1884         SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno;
1885
1886         if (gc_type == FG_GC)
1887                 f2fs_unpin_all_sections(sbi, true);
1888
1889         trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1890                                 get_pages(sbi, F2FS_DIRTY_NODES),
1891                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1892                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1893                                 free_sections(sbi),
1894                                 free_segments(sbi),
1895                                 reserved_segments(sbi),
1896                                 prefree_segments(sbi));
1897
1898         f2fs_up_write(&sbi->gc_lock);
1899
1900         put_gc_inode(&gc_list);
1901
1902         if (gc_control->err_gc_skipped && !ret)
1903                 ret = sec_freed ? 0 : -EAGAIN;
1904         return ret;
1905 }
1906
1907 int __init f2fs_create_garbage_collection_cache(void)
1908 {
1909         victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1910                                         sizeof(struct victim_entry));
1911         return victim_entry_slab ? 0 : -ENOMEM;
1912 }
1913
1914 void f2fs_destroy_garbage_collection_cache(void)
1915 {
1916         kmem_cache_destroy(victim_entry_slab);
1917 }
1918
1919 static void init_atgc_management(struct f2fs_sb_info *sbi)
1920 {
1921         struct atgc_management *am = &sbi->am;
1922
1923         if (test_opt(sbi, ATGC) &&
1924                 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1925                 am->atgc_enabled = true;
1926
1927         am->root = RB_ROOT_CACHED;
1928         INIT_LIST_HEAD(&am->victim_list);
1929         am->victim_count = 0;
1930
1931         am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1932         am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1933         am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1934         am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
1935 }
1936
1937 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1938 {
1939         DIRTY_I(sbi)->v_ops = &default_v_ops;
1940
1941         sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1942
1943         /* give warm/cold data area from slower device */
1944         if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1945                 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1946                                 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1947
1948         init_atgc_management(sbi);
1949 }
1950
1951 static int free_segment_range(struct f2fs_sb_info *sbi,
1952                                 unsigned int secs, bool gc_only)
1953 {
1954         unsigned int segno, next_inuse, start, end;
1955         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1956         int gc_mode, gc_type;
1957         int err = 0;
1958         int type;
1959
1960         /* Force block allocation for GC */
1961         MAIN_SECS(sbi) -= secs;
1962         start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1963         end = MAIN_SEGS(sbi) - 1;
1964
1965         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1966         for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1967                 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1968                         SIT_I(sbi)->last_victim[gc_mode] = 0;
1969
1970         for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1971                 if (sbi->next_victim_seg[gc_type] >= start)
1972                         sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1973         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1974
1975         /* Move out cursegs from the target range */
1976         for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1977                 f2fs_allocate_segment_for_resize(sbi, type, start, end);
1978
1979         /* do GC to move out valid blocks in the range */
1980         for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1981                 struct gc_inode_list gc_list = {
1982                         .ilist = LIST_HEAD_INIT(gc_list.ilist),
1983                         .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1984                 };
1985
1986                 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
1987                 put_gc_inode(&gc_list);
1988
1989                 if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1990                         err = -EAGAIN;
1991                         goto out;
1992                 }
1993                 if (fatal_signal_pending(current)) {
1994                         err = -ERESTARTSYS;
1995                         goto out;
1996                 }
1997         }
1998         if (gc_only)
1999                 goto out;
2000
2001         err = f2fs_write_checkpoint(sbi, &cpc);
2002         if (err)
2003                 goto out;
2004
2005         next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
2006         if (next_inuse <= end) {
2007                 f2fs_err(sbi, "segno %u should be free but still inuse!",
2008                          next_inuse);
2009                 f2fs_bug_on(sbi, 1);
2010         }
2011 out:
2012         MAIN_SECS(sbi) += secs;
2013         return err;
2014 }
2015
2016 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
2017 {
2018         struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
2019         int section_count;
2020         int segment_count;
2021         int segment_count_main;
2022         long long block_count;
2023         int segs = secs * sbi->segs_per_sec;
2024
2025         f2fs_down_write(&sbi->sb_lock);
2026
2027         section_count = le32_to_cpu(raw_sb->section_count);
2028         segment_count = le32_to_cpu(raw_sb->segment_count);
2029         segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
2030         block_count = le64_to_cpu(raw_sb->block_count);
2031
2032         raw_sb->section_count = cpu_to_le32(section_count + secs);
2033         raw_sb->segment_count = cpu_to_le32(segment_count + segs);
2034         raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
2035         raw_sb->block_count = cpu_to_le64(block_count +
2036                                         (long long)segs * sbi->blocks_per_seg);
2037         if (f2fs_is_multi_device(sbi)) {
2038                 int last_dev = sbi->s_ndevs - 1;
2039                 int dev_segs =
2040                         le32_to_cpu(raw_sb->devs[last_dev].total_segments);
2041
2042                 raw_sb->devs[last_dev].total_segments =
2043                                                 cpu_to_le32(dev_segs + segs);
2044         }
2045
2046         f2fs_up_write(&sbi->sb_lock);
2047 }
2048
2049 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
2050 {
2051         int segs = secs * sbi->segs_per_sec;
2052         long long blks = (long long)segs * sbi->blocks_per_seg;
2053         long long user_block_count =
2054                                 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
2055
2056         SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
2057         MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
2058         MAIN_SECS(sbi) += secs;
2059         FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
2060         FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
2061         F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
2062
2063         if (f2fs_is_multi_device(sbi)) {
2064                 int last_dev = sbi->s_ndevs - 1;
2065
2066                 FDEV(last_dev).total_segments =
2067                                 (int)FDEV(last_dev).total_segments + segs;
2068                 FDEV(last_dev).end_blk =
2069                                 (long long)FDEV(last_dev).end_blk + blks;
2070 #ifdef CONFIG_BLK_DEV_ZONED
2071                 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
2072                                         (int)(blks >> sbi->log_blocks_per_blkz);
2073 #endif
2074         }
2075 }
2076
2077 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
2078 {
2079         __u64 old_block_count, shrunk_blocks;
2080         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
2081         unsigned int secs;
2082         int err = 0;
2083         __u32 rem;
2084
2085         old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
2086         if (block_count > old_block_count)
2087                 return -EINVAL;
2088
2089         if (f2fs_is_multi_device(sbi)) {
2090                 int last_dev = sbi->s_ndevs - 1;
2091                 __u64 last_segs = FDEV(last_dev).total_segments;
2092
2093                 if (block_count + last_segs * sbi->blocks_per_seg <=
2094                                                                 old_block_count)
2095                         return -EINVAL;
2096         }
2097
2098         /* new fs size should align to section size */
2099         div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
2100         if (rem)
2101                 return -EINVAL;
2102
2103         if (block_count == old_block_count)
2104                 return 0;
2105
2106         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2107                 f2fs_err(sbi, "Should run fsck to repair first.");
2108                 return -EFSCORRUPTED;
2109         }
2110
2111         if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2112                 f2fs_err(sbi, "Checkpoint should be enabled.");
2113                 return -EINVAL;
2114         }
2115
2116         shrunk_blocks = old_block_count - block_count;
2117         secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2118
2119         /* stop other GC */
2120         if (!f2fs_down_write_trylock(&sbi->gc_lock))
2121                 return -EAGAIN;
2122
2123         /* stop CP to protect MAIN_SEC in free_segment_range */
2124         f2fs_lock_op(sbi);
2125
2126         spin_lock(&sbi->stat_lock);
2127         if (shrunk_blocks + valid_user_blocks(sbi) +
2128                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2129                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2130                 err = -ENOSPC;
2131         spin_unlock(&sbi->stat_lock);
2132
2133         if (err)
2134                 goto out_unlock;
2135
2136         err = free_segment_range(sbi, secs, true);
2137
2138 out_unlock:
2139         f2fs_unlock_op(sbi);
2140         f2fs_up_write(&sbi->gc_lock);
2141         if (err)
2142                 return err;
2143
2144         freeze_super(sbi->sb);
2145         f2fs_down_write(&sbi->gc_lock);
2146         f2fs_down_write(&sbi->cp_global_sem);
2147
2148         spin_lock(&sbi->stat_lock);
2149         if (shrunk_blocks + valid_user_blocks(sbi) +
2150                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2151                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2152                 err = -ENOSPC;
2153         else
2154                 sbi->user_block_count -= shrunk_blocks;
2155         spin_unlock(&sbi->stat_lock);
2156         if (err)
2157                 goto out_err;
2158
2159         set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2160         err = free_segment_range(sbi, secs, false);
2161         if (err)
2162                 goto recover_out;
2163
2164         update_sb_metadata(sbi, -secs);
2165
2166         err = f2fs_commit_super(sbi, false);
2167         if (err) {
2168                 update_sb_metadata(sbi, secs);
2169                 goto recover_out;
2170         }
2171
2172         update_fs_metadata(sbi, -secs);
2173         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2174         set_sbi_flag(sbi, SBI_IS_DIRTY);
2175
2176         err = f2fs_write_checkpoint(sbi, &cpc);
2177         if (err) {
2178                 update_fs_metadata(sbi, secs);
2179                 update_sb_metadata(sbi, secs);
2180                 f2fs_commit_super(sbi, false);
2181         }
2182 recover_out:
2183         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2184         if (err) {
2185                 set_sbi_flag(sbi, SBI_NEED_FSCK);
2186                 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2187
2188                 spin_lock(&sbi->stat_lock);
2189                 sbi->user_block_count += shrunk_blocks;
2190                 spin_unlock(&sbi->stat_lock);
2191         }
2192 out_err:
2193         f2fs_up_write(&sbi->cp_global_sem);
2194         f2fs_up_write(&sbi->gc_lock);
2195         thaw_super(sbi->sb);
2196         return err;
2197 }