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                 }
  79
  80                 if (!sb_start_write_trylock(sbi->sb)) {
  81                         stat_other_skip_bggc_count(sbi);
  82                         continue;
  83                 }
  84
  85                 /*
  86                  * [GC triggering condition]
  87                  * 0. GC is not conducted currently.
  88                  * 1. There are enough dirty segments.
  89                  * 2. IO subsystem is idle by checking the # of writeback pages.
  90                  * 3. IO subsystem is idle by checking the # of requests in
  91                  *    bdev's request list.
  92                  *
  93                  * Note) We have to avoid triggering GCs frequently.
  94                  * Because it is possible that some segments can be
  95                  * invalidated soon after by user update or deletion.
  96                  * So, I'd like to wait some time to collect dirty segments.
  97                  */
  98                 if (sbi->gc_mode == GC_URGENT_HIGH) {
  99                         spin_lock(&sbi->gc_urgent_high_lock);
 100                         if (sbi->gc_urgent_high_limited) {
 101                                 if (!sbi->gc_urgent_high_remaining) {
 102                                         sbi->gc_urgent_high_limited = false;
 103                                         spin_unlock(&sbi->gc_urgent_high_lock);
 104                                         sbi->gc_mode = GC_NORMAL;
 105                                         continue;
 106                                 }
 107                                 sbi->gc_urgent_high_remaining--;
 108                         }
 109                         spin_unlock(&sbi->gc_urgent_high_lock);
 110                 }
 111
 112                 if (sbi->gc_mode == GC_URGENT_HIGH ||
 113                                 sbi->gc_mode == GC_URGENT_MID) {
 114                         wait_ms = gc_th->urgent_sleep_time;
 115                         f2fs_down_write(&sbi->gc_lock);
 116                         goto do_gc;
 117                 }
 118
 119                 if (foreground) {
 120                         f2fs_down_write(&sbi->gc_lock);
 121                         goto do_gc;
 122                 } else if (!f2fs_down_write_trylock(&sbi->gc_lock)) {
 123                         stat_other_skip_bggc_count(sbi);
 124                         goto next;
 125                 }
 126
 127                 if (!is_idle(sbi, GC_TIME)) {
 128                         increase_sleep_time(gc_th, &wait_ms);
 129                         f2fs_up_write(&sbi->gc_lock);
 130                         stat_io_skip_bggc_count(sbi);
 131                         goto next;
 132                 }
 133
 134                 if (has_enough_invalid_blocks(sbi))
 135                         decrease_sleep_time(gc_th, &wait_ms);
 136                 else
 137                         increase_sleep_time(gc_th, &wait_ms);
 138 do_gc:
 139                 if (!foreground)
 140                         stat_inc_bggc_count(sbi->stat_info);
 141
 142                 sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
 143
 144                 /* foreground GC was been triggered via f2fs_balance_fs() */
 145                 if (foreground)
 146                         sync_mode = false;
 147
 148                 gc_control.init_gc_type = sync_mode ? FG_GC : BG_GC;
 149                 gc_control.no_bg_gc = foreground;
 150                 gc_control.nr_free_secs = foreground ? 1 : 0;
 151
 152                 /* if return value is not zero, no victim was selected */
 153                 if (f2fs_gc(sbi, &gc_control))
 154                         wait_ms = gc_th->no_gc_sleep_time;
 155
 156                 if (foreground)
 157                         wake_up_all(&gc_th->fggc_wq);
 158
 159                 trace_f2fs_background_gc(sbi->sb, wait_ms,
 160                                 prefree_segments(sbi), free_segments(sbi));
 161
 162                 /* balancing f2fs's metadata periodically */
 163                 f2fs_balance_fs_bg(sbi, true);
 164 next:
 165                 sb_end_write(sbi->sb);
 166
 167         } while (!kthread_should_stop());
 168         return 0;
 169 }
 170
 171 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
 172 {
 173         struct f2fs_gc_kthread *gc_th;
 174         dev_t dev = sbi->sb->s_bdev->bd_dev;
 175         int err = 0;
 176
 177         gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 178         if (!gc_th) {
 179                 err = -ENOMEM;
 180                 goto out;
 181         }
 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                 err = PTR_ERR(gc_th->f2fs_gc_task);
 197                 kfree(gc_th);
 198                 sbi->gc_thread = NULL;
 199         }
 200 out:
 201         return err;
 202 }
 203
 204 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi)
 205 {
 206         struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 207
 208         if (!gc_th)
 209                 return;
 210         kthread_stop(gc_th->f2fs_gc_task);
 211         wake_up_all(&gc_th->fggc_wq);
 212         kfree(gc_th);
 213         sbi->gc_thread = NULL;
 214 }
 215
 216 static int select_gc_type(struct f2fs_sb_info *sbi, int gc_type)
 217 {
 218         int gc_mode;
 219
 220         if (gc_type == BG_GC) {
 221                 if (sbi->am.atgc_enabled)
 222                         gc_mode = GC_AT;
 223                 else
 224                         gc_mode = GC_CB;
 225         } else {
 226                 gc_mode = GC_GREEDY;
 227         }
 228
 229         switch (sbi->gc_mode) {
 230         case GC_IDLE_CB:
 231                 gc_mode = GC_CB;
 232                 break;
 233         case GC_IDLE_GREEDY:
 234         case GC_URGENT_HIGH:
 235                 gc_mode = GC_GREEDY;
 236                 break;
 237         case GC_IDLE_AT:
 238                 gc_mode = GC_AT;
 239                 break;
 240         }
 241
 242         return gc_mode;
 243 }
 244
 245 static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 246                         int type, struct victim_sel_policy *p)
 247 {
 248         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 249
 250         if (p->alloc_mode == SSR) {
 251                 p->gc_mode = GC_GREEDY;
 252                 p->dirty_bitmap = dirty_i->dirty_segmap[type];
 253                 p->max_search = dirty_i->nr_dirty[type];
 254                 p->ofs_unit = 1;
 255         } else if (p->alloc_mode == AT_SSR) {
 256                 p->gc_mode = GC_GREEDY;
 257                 p->dirty_bitmap = dirty_i->dirty_segmap[type];
 258                 p->max_search = dirty_i->nr_dirty[type];
 259                 p->ofs_unit = 1;
 260         } else {
 261                 p->gc_mode = select_gc_type(sbi, gc_type);
 262                 p->ofs_unit = sbi->segs_per_sec;
 263                 if (__is_large_section(sbi)) {
 264                         p->dirty_bitmap = dirty_i->dirty_secmap;
 265                         p->max_search = count_bits(p->dirty_bitmap,
 266                                                 0, MAIN_SECS(sbi));
 267                 } else {
 268                         p->dirty_bitmap = dirty_i->dirty_segmap[DIRTY];
 269                         p->max_search = dirty_i->nr_dirty[DIRTY];
 270                 }
 271         }
 272
 273         /*
 274          * adjust candidates range, should select all dirty segments for
 275          * foreground GC and urgent GC cases.
 276          */
 277         if (gc_type != FG_GC &&
 278                         (sbi->gc_mode != GC_URGENT_HIGH) &&
 279                         (p->gc_mode != GC_AT && p->alloc_mode != AT_SSR) &&
 280                         p->max_search > sbi->max_victim_search)
 281                 p->max_search = sbi->max_victim_search;
 282
 283         /* let's select beginning hot/small space first in no_heap mode*/
 284         if (f2fs_need_rand_seg(sbi))
 285                 p->offset = prandom_u32() % (MAIN_SECS(sbi) * sbi->segs_per_sec);
 286         else if (test_opt(sbi, NOHEAP) &&
 287                 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
 288                 p->offset = 0;
 289         else
 290                 p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
 291 }
 292
 293 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 294                                 struct victim_sel_policy *p)
 295 {
 296         /* SSR allocates in a segment unit */
 297         if (p->alloc_mode == SSR)
 298                 return sbi->blocks_per_seg;
 299         else if (p->alloc_mode == AT_SSR)
 300                 return UINT_MAX;
 301
 302         /* LFS */
 303         if (p->gc_mode == GC_GREEDY)
 304                 return 2 * sbi->blocks_per_seg * p->ofs_unit;
 305         else if (p->gc_mode == GC_CB)
 306                 return UINT_MAX;
 307         else if (p->gc_mode == GC_AT)
 308                 return UINT_MAX;
 309         else /* No other gc_mode */
 310                 return 0;
 311 }
 312
 313 static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 314 {
 315         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 316         unsigned int secno;
 317
 318         /*
 319          * If the gc_type is FG_GC, we can select victim segments
 320          * selected by background GC before.
 321          * Those segments guarantee they have small valid blocks.
 322          */
 323         for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
 324                 if (sec_usage_check(sbi, secno))
 325                         continue;
 326                 clear_bit(secno, dirty_i->victim_secmap);
 327                 return GET_SEG_FROM_SEC(sbi, secno);
 328         }
 329         return NULL_SEGNO;
 330 }
 331
 332 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 333 {
 334         struct sit_info *sit_i = SIT_I(sbi);
 335         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 336         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 337         unsigned long long mtime = 0;
 338         unsigned int vblocks;
 339         unsigned char age = 0;
 340         unsigned char u;
 341         unsigned int i;
 342         unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi, segno);
 343
 344         for (i = 0; i < usable_segs_per_sec; i++)
 345                 mtime += get_seg_entry(sbi, start + i)->mtime;
 346         vblocks = get_valid_blocks(sbi, segno, true);
 347
 348         mtime = div_u64(mtime, usable_segs_per_sec);
 349         vblocks = div_u64(vblocks, usable_segs_per_sec);
 350
 351         u = (vblocks * 100) >> sbi->log_blocks_per_seg;
 352
 353         /* Handle if the system time has changed by the user */
 354         if (mtime < sit_i->min_mtime)
 355                 sit_i->min_mtime = mtime;
 356         if (mtime > sit_i->max_mtime)
 357                 sit_i->max_mtime = mtime;
 358         if (sit_i->max_mtime != sit_i->min_mtime)
 359                 age = 100 - div64_u64(100 * (mtime - sit_i->min_mtime),
 360                                 sit_i->max_mtime - sit_i->min_mtime);
 361
 362         return UINT_MAX - ((100 * (100 - u) * age) / (100 + u));
 363 }
 364
 365 static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 366                         unsigned int segno, struct victim_sel_policy *p)
 367 {
 368         if (p->alloc_mode == SSR)
 369                 return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 370
 371         /* alloc_mode == LFS */
 372         if (p->gc_mode == GC_GREEDY)
 373                 return get_valid_blocks(sbi, segno, true);
 374         else if (p->gc_mode == GC_CB)
 375                 return get_cb_cost(sbi, segno);
 376
 377         f2fs_bug_on(sbi, 1);
 378         return 0;
 379 }
 380
 381 static unsigned int count_bits(const unsigned long *addr,
 382                                 unsigned int offset, unsigned int len)
 383 {
 384         unsigned int end = offset + len, sum = 0;
 385
 386         while (offset < end) {
 387                 if (test_bit(offset++, addr))
 388                         ++sum;
 389         }
 390         return sum;
 391 }
 392
 393 static struct victim_entry *attach_victim_entry(struct f2fs_sb_info *sbi,
 394                                 unsigned long long mtime, unsigned int segno,
 395                                 struct rb_node *parent, struct rb_node **p,
 396                                 bool left_most)
 397 {
 398         struct atgc_management *am = &sbi->am;
 399         struct victim_entry *ve;
 400
 401         ve =  f2fs_kmem_cache_alloc(victim_entry_slab,
 402                                 GFP_NOFS, true, NULL);
 403
 404         ve->mtime = mtime;
 405         ve->segno = segno;
 406
 407         rb_link_node(&ve->rb_node, parent, p);
 408         rb_insert_color_cached(&ve->rb_node, &am->root, left_most);
 409
 410         list_add_tail(&ve->list, &am->victim_list);
 411
 412         am->victim_count++;
 413
 414         return ve;
 415 }
 416
 417 static void insert_victim_entry(struct f2fs_sb_info *sbi,
 418                                 unsigned long long mtime, unsigned int segno)
 419 {
 420         struct atgc_management *am = &sbi->am;
 421         struct rb_node **p;
 422         struct rb_node *parent = NULL;
 423         bool left_most = true;
 424
 425         p = f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, mtime, &left_most);
 426         attach_victim_entry(sbi, mtime, segno, parent, p, left_most);
 427 }
 428
 429 static void add_victim_entry(struct f2fs_sb_info *sbi,
 430                                 struct victim_sel_policy *p, unsigned int segno)
 431 {
 432         struct sit_info *sit_i = SIT_I(sbi);
 433         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 434         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 435         unsigned long long mtime = 0;
 436         unsigned int i;
 437
 438         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 439                 if (p->gc_mode == GC_AT &&
 440                         get_valid_blocks(sbi, segno, true) == 0)
 441                         return;
 442         }
 443
 444         for (i = 0; i < sbi->segs_per_sec; i++)
 445                 mtime += get_seg_entry(sbi, start + i)->mtime;
 446         mtime = div_u64(mtime, sbi->segs_per_sec);
 447
 448         /* Handle if the system time has changed by the user */
 449         if (mtime < sit_i->min_mtime)
 450                 sit_i->min_mtime = mtime;
 451         if (mtime > sit_i->max_mtime)
 452                 sit_i->max_mtime = mtime;
 453         if (mtime < sit_i->dirty_min_mtime)
 454                 sit_i->dirty_min_mtime = mtime;
 455         if (mtime > sit_i->dirty_max_mtime)
 456                 sit_i->dirty_max_mtime = mtime;
 457
 458         /* don't choose young section as candidate */
 459         if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
 460                 return;
 461
 462         insert_victim_entry(sbi, mtime, segno);
 463 }
 464
 465 static struct rb_node *lookup_central_victim(struct f2fs_sb_info *sbi,
 466                                                 struct victim_sel_policy *p)
 467 {
 468         struct atgc_management *am = &sbi->am;
 469         struct rb_node *parent = NULL;
 470         bool left_most;
 471
 472         f2fs_lookup_rb_tree_ext(sbi, &am->root, &parent, p->age, &left_most);
 473
 474         return parent;
 475 }
 476
 477 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
 478                                                 struct victim_sel_policy *p)
 479 {
 480         struct sit_info *sit_i = SIT_I(sbi);
 481         struct atgc_management *am = &sbi->am;
 482         struct rb_root_cached *root = &am->root;
 483         struct rb_node *node;
 484         struct rb_entry *re;
 485         struct victim_entry *ve;
 486         unsigned long long total_time;
 487         unsigned long long age, u, accu;
 488         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 489         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 490         unsigned int sec_blocks = BLKS_PER_SEC(sbi);
 491         unsigned int vblocks;
 492         unsigned int dirty_threshold = max(am->max_candidate_count,
 493                                         am->candidate_ratio *
 494                                         am->victim_count / 100);
 495         unsigned int age_weight = am->age_weight;
 496         unsigned int cost;
 497         unsigned int iter = 0;
 498
 499         if (max_mtime < min_mtime)
 500                 return;
 501
 502         max_mtime += 1;
 503         total_time = max_mtime - min_mtime;
 504
 505         accu = div64_u64(ULLONG_MAX, total_time);
 506         accu = min_t(unsigned long long, div_u64(accu, 100),
 507                                         DEFAULT_ACCURACY_CLASS);
 508
 509         node = rb_first_cached(root);
 510 next:
 511         re = rb_entry_safe(node, struct rb_entry, rb_node);
 512         if (!re)
 513                 return;
 514
 515         ve = (struct victim_entry *)re;
 516
 517         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 518                 goto skip;
 519
 520         /* age = 10000 * x% * 60 */
 521         age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
 522                                                                 age_weight;
 523
 524         vblocks = get_valid_blocks(sbi, ve->segno, true);
 525         f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
 526
 527         /* u = 10000 * x% * 40 */
 528         u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
 529                                                         (100 - age_weight);
 530
 531         f2fs_bug_on(sbi, age + u >= UINT_MAX);
 532
 533         cost = UINT_MAX - (age + u);
 534         iter++;
 535
 536         if (cost < p->min_cost ||
 537                         (cost == p->min_cost && age > p->oldest_age)) {
 538                 p->min_cost = cost;
 539                 p->oldest_age = age;
 540                 p->min_segno = ve->segno;
 541         }
 542 skip:
 543         if (iter < dirty_threshold) {
 544                 node = rb_next(node);
 545                 goto next;
 546         }
 547 }
 548
 549 /*
 550  * select candidates around source section in range of
 551  * [target - dirty_threshold, target + dirty_threshold]
 552  */
 553 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
 554                                                 struct victim_sel_policy *p)
 555 {
 556         struct sit_info *sit_i = SIT_I(sbi);
 557         struct atgc_management *am = &sbi->am;
 558         struct rb_node *node;
 559         struct rb_entry *re;
 560         struct victim_entry *ve;
 561         unsigned long long age;
 562         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 563         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 564         unsigned int seg_blocks = sbi->blocks_per_seg;
 565         unsigned int vblocks;
 566         unsigned int dirty_threshold = max(am->max_candidate_count,
 567                                         am->candidate_ratio *
 568                                         am->victim_count / 100);
 569         unsigned int cost;
 570         unsigned int iter = 0;
 571         int stage = 0;
 572
 573         if (max_mtime < min_mtime)
 574                 return;
 575         max_mtime += 1;
 576 next_stage:
 577         node = lookup_central_victim(sbi, p);
 578 next_node:
 579         re = rb_entry_safe(node, struct rb_entry, rb_node);
 580         if (!re) {
 581                 if (stage == 0)
 582                         goto skip_stage;
 583                 return;
 584         }
 585
 586         ve = (struct victim_entry *)re;
 587
 588         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 589                 goto skip_node;
 590
 591         age = max_mtime - ve->mtime;
 592
 593         vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
 594         f2fs_bug_on(sbi, !vblocks);
 595
 596         /* rare case */
 597         if (vblocks == seg_blocks)
 598                 goto skip_node;
 599
 600         iter++;
 601
 602         age = max_mtime - abs(p->age - age);
 603         cost = UINT_MAX - vblocks;
 604
 605         if (cost < p->min_cost ||
 606                         (cost == p->min_cost && age > p->oldest_age)) {
 607                 p->min_cost = cost;
 608                 p->oldest_age = age;
 609                 p->min_segno = ve->segno;
 610         }
 611 skip_node:
 612         if (iter < dirty_threshold) {
 613                 if (stage == 0)
 614                         node = rb_prev(node);
 615                 else if (stage == 1)
 616                         node = rb_next(node);
 617                 goto next_node;
 618         }
 619 skip_stage:
 620         if (stage < 1) {
 621                 stage++;
 622                 iter = 0;
 623                 goto next_stage;
 624         }
 625 }
 626 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
 627                                                 struct victim_sel_policy *p)
 628 {
 629         f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
 630                                                 &sbi->am.root, true));
 631
 632         if (p->gc_mode == GC_AT)
 633                 atgc_lookup_victim(sbi, p);
 634         else if (p->alloc_mode == AT_SSR)
 635                 atssr_lookup_victim(sbi, p);
 636         else
 637                 f2fs_bug_on(sbi, 1);
 638 }
 639
 640 static void release_victim_entry(struct f2fs_sb_info *sbi)
 641 {
 642         struct atgc_management *am = &sbi->am;
 643         struct victim_entry *ve, *tmp;
 644
 645         list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
 646                 list_del(&ve->list);
 647                 kmem_cache_free(victim_entry_slab, ve);
 648                 am->victim_count--;
 649         }
 650
 651         am->root = RB_ROOT_CACHED;
 652
 653         f2fs_bug_on(sbi, am->victim_count);
 654         f2fs_bug_on(sbi, !list_empty(&am->victim_list));
 655 }
 656
 657 static bool f2fs_pin_section(struct f2fs_sb_info *sbi, unsigned int segno)
 658 {
 659         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 660         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 661
 662         if (!dirty_i->enable_pin_section)
 663                 return false;
 664         if (!test_and_set_bit(secno, dirty_i->pinned_secmap))
 665                 dirty_i->pinned_secmap_cnt++;
 666         return true;
 667 }
 668
 669 static bool f2fs_pinned_section_exists(struct dirty_seglist_info *dirty_i)
 670 {
 671         return dirty_i->pinned_secmap_cnt;
 672 }
 673
 674 static bool f2fs_section_is_pinned(struct dirty_seglist_info *dirty_i,
 675                                                 unsigned int secno)
 676 {
 677         return dirty_i->enable_pin_section &&
 678                 f2fs_pinned_section_exists(dirty_i) &&
 679                 test_bit(secno, dirty_i->pinned_secmap);
 680 }
 681
 682 static void f2fs_unpin_all_sections(struct f2fs_sb_info *sbi, bool enable)
 683 {
 684         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
 685
 686         if (f2fs_pinned_section_exists(DIRTY_I(sbi))) {
 687                 memset(DIRTY_I(sbi)->pinned_secmap, 0, bitmap_size);
 688                 DIRTY_I(sbi)->pinned_secmap_cnt = 0;
 689         }
 690         DIRTY_I(sbi)->enable_pin_section = enable;
 691 }
 692
 693 static int f2fs_gc_pinned_control(struct inode *inode, int gc_type,
 694                                                         unsigned int segno)
 695 {
 696         if (!f2fs_is_pinned_file(inode))
 697                 return 0;
 698         if (gc_type != FG_GC)
 699                 return -EBUSY;
 700         if (!f2fs_pin_section(F2FS_I_SB(inode), segno))
 701                 f2fs_pin_file_control(inode, true);
 702         return -EAGAIN;
 703 }
 704
 705 /*
 706  * This function is called from two paths.
 707  * One is garbage collection and the other is SSR segment selection.
 708  * When it is called during GC, it just gets a victim segment
 709  * and it does not remove it from dirty seglist.
 710  * When it is called from SSR segment selection, it finds a segment
 711  * which has minimum valid blocks and removes it from dirty seglist.
 712  */
 713 static int get_victim_by_default(struct f2fs_sb_info *sbi,
 714                         unsigned int *result, int gc_type, int type,
 715                         char alloc_mode, unsigned long long age)
 716 {
 717         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 718         struct sit_info *sm = SIT_I(sbi);
 719         struct victim_sel_policy p;
 720         unsigned int secno, last_victim;
 721         unsigned int last_segment;
 722         unsigned int nsearched;
 723         bool is_atgc;
 724         int ret = 0;
 725
 726         mutex_lock(&dirty_i->seglist_lock);
 727         last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
 728
 729         p.alloc_mode = alloc_mode;
 730         p.age = age;
 731         p.age_threshold = sbi->am.age_threshold;
 732
 733 retry:
 734         select_policy(sbi, gc_type, type, &p);
 735         p.min_segno = NULL_SEGNO;
 736         p.oldest_age = 0;
 737         p.min_cost = get_max_cost(sbi, &p);
 738
 739         is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
 740         nsearched = 0;
 741
 742         if (is_atgc)
 743                 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
 744
 745         if (*result != NULL_SEGNO) {
 746                 if (!get_valid_blocks(sbi, *result, false)) {
 747                         ret = -ENODATA;
 748                         goto out;
 749                 }
 750
 751                 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
 752                         ret = -EBUSY;
 753                 else
 754                         p.min_segno = *result;
 755                 goto out;
 756         }
 757
 758         ret = -ENODATA;
 759         if (p.max_search == 0)
 760                 goto out;
 761
 762         if (__is_large_section(sbi) && p.alloc_mode == LFS) {
 763                 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
 764                         p.min_segno = sbi->next_victim_seg[BG_GC];
 765                         *result = p.min_segno;
 766                         sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
 767                         goto got_result;
 768                 }
 769                 if (gc_type == FG_GC &&
 770                                 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
 771                         p.min_segno = sbi->next_victim_seg[FG_GC];
 772                         *result = p.min_segno;
 773                         sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
 774                         goto got_result;
 775                 }
 776         }
 777
 778         last_victim = sm->last_victim[p.gc_mode];
 779         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 780                 p.min_segno = check_bg_victims(sbi);
 781                 if (p.min_segno != NULL_SEGNO)
 782                         goto got_it;
 783         }
 784
 785         while (1) {
 786                 unsigned long cost, *dirty_bitmap;
 787                 unsigned int unit_no, segno;
 788
 789                 dirty_bitmap = p.dirty_bitmap;
 790                 unit_no = find_next_bit(dirty_bitmap,
 791                                 last_segment / p.ofs_unit,
 792                                 p.offset / p.ofs_unit);
 793                 segno = unit_no * p.ofs_unit;
 794                 if (segno >= last_segment) {
 795                         if (sm->last_victim[p.gc_mode]) {
 796                                 last_segment =
 797                                         sm->last_victim[p.gc_mode];
 798                                 sm->last_victim[p.gc_mode] = 0;
 799                                 p.offset = 0;
 800                                 continue;
 801                         }
 802                         break;
 803                 }
 804
 805                 p.offset = segno + p.ofs_unit;
 806                 nsearched++;
 807
 808 #ifdef CONFIG_F2FS_CHECK_FS
 809                 /*
 810                  * skip selecting the invalid segno (that is failed due to block
 811                  * validity check failure during GC) to avoid endless GC loop in
 812                  * such cases.
 813                  */
 814                 if (test_bit(segno, sm->invalid_segmap))
 815                         goto next;
 816 #endif
 817
 818                 secno = GET_SEC_FROM_SEG(sbi, segno);
 819
 820                 if (sec_usage_check(sbi, secno))
 821                         goto next;
 822
 823                 /* Don't touch checkpointed data */
 824                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 825                         if (p.alloc_mode == LFS) {
 826                                 /*
 827                                  * LFS is set to find source section during GC.
 828                                  * The victim should have no checkpointed data.
 829                                  */
 830                                 if (get_ckpt_valid_blocks(sbi, segno, true))
 831                                         goto next;
 832                         } else {
 833                                 /*
 834                                  * SSR | AT_SSR are set to find target segment
 835                                  * for writes which can be full by checkpointed
 836                                  * and newly written blocks.
 837                                  */
 838                                 if (!f2fs_segment_has_free_slot(sbi, segno))
 839                                         goto next;
 840                         }
 841                 }
 842
 843                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 844                         goto next;
 845
 846                 if (gc_type == FG_GC && f2fs_section_is_pinned(dirty_i, secno))
 847                         goto next;
 848
 849                 if (is_atgc) {
 850                         add_victim_entry(sbi, &p, segno);
 851                         goto next;
 852                 }
 853
 854                 cost = get_gc_cost(sbi, segno, &p);
 855
 856                 if (p.min_cost > cost) {
 857                         p.min_segno = segno;
 858                         p.min_cost = cost;
 859                 }
 860 next:
 861                 if (nsearched >= p.max_search) {
 862                         if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
 863                                 sm->last_victim[p.gc_mode] =
 864                                         last_victim + p.ofs_unit;
 865                         else
 866                                 sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
 867                         sm->last_victim[p.gc_mode] %=
 868                                 (MAIN_SECS(sbi) * sbi->segs_per_sec);
 869                         break;
 870                 }
 871         }
 872
 873         /* get victim for GC_AT/AT_SSR */
 874         if (is_atgc) {
 875                 lookup_victim_by_age(sbi, &p);
 876                 release_victim_entry(sbi);
 877         }
 878
 879         if (is_atgc && p.min_segno == NULL_SEGNO &&
 880                         sm->elapsed_time < p.age_threshold) {
 881                 p.age_threshold = 0;
 882                 goto retry;
 883         }
 884
 885         if (p.min_segno != NULL_SEGNO) {
 886 got_it:
 887                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 888 got_result:
 889                 if (p.alloc_mode == LFS) {
 890                         secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
 891                         if (gc_type == FG_GC)
 892                                 sbi->cur_victim_sec = secno;
 893                         else
 894                                 set_bit(secno, dirty_i->victim_secmap);
 895                 }
 896                 ret = 0;
 897
 898         }
 899 out:
 900         if (p.min_segno != NULL_SEGNO)
 901                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 902                                 sbi->cur_victim_sec,
 903                                 prefree_segments(sbi), free_segments(sbi));
 904         mutex_unlock(&dirty_i->seglist_lock);
 905
 906         return ret;
 907 }
 908
 909 static const struct victim_selection default_v_ops = {
 910         .get_victim = get_victim_by_default,
 911 };
 912
 913 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
 914 {
 915         struct inode_entry *ie;
 916
 917         ie = radix_tree_lookup(&gc_list->iroot, ino);
 918         if (ie)
 919                 return ie->inode;
 920         return NULL;
 921 }
 922
 923 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
 924 {
 925         struct inode_entry *new_ie;
 926
 927         if (inode == find_gc_inode(gc_list, inode->i_ino)) {
 928                 iput(inode);
 929                 return;
 930         }
 931         new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab,
 932                                         GFP_NOFS, true, NULL);
 933         new_ie->inode = inode;
 934
 935         f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
 936         list_add_tail(&new_ie->list, &gc_list->ilist);
 937 }
 938
 939 static void put_gc_inode(struct gc_inode_list *gc_list)
 940 {
 941         struct inode_entry *ie, *next_ie;
 942
 943         list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
 944                 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
 945                 iput(ie->inode);
 946                 list_del(&ie->list);
 947                 kmem_cache_free(f2fs_inode_entry_slab, ie);
 948         }
 949 }
 950
 951 static int check_valid_map(struct f2fs_sb_info *sbi,
 952                                 unsigned int segno, int offset)
 953 {
 954         struct sit_info *sit_i = SIT_I(sbi);
 955         struct seg_entry *sentry;
 956         int ret;
 957
 958         down_read(&sit_i->sentry_lock);
 959         sentry = get_seg_entry(sbi, segno);
 960         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 961         up_read(&sit_i->sentry_lock);
 962         return ret;
 963 }
 964
 965 /*
 966  * This function compares node address got in summary with that in NAT.
 967  * On validity, copy that node with cold status, otherwise (invalid node)
 968  * ignore that.
 969  */
 970 static int gc_node_segment(struct f2fs_sb_info *sbi,
 971                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 972 {
 973         struct f2fs_summary *entry;
 974         block_t start_addr;
 975         int off;
 976         int phase = 0;
 977         bool fggc = (gc_type == FG_GC);
 978         int submitted = 0;
 979         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
 980
 981         start_addr = START_BLOCK(sbi, segno);
 982
 983 next_step:
 984         entry = sum;
 985
 986         if (fggc && phase == 2)
 987                 atomic_inc(&sbi->wb_sync_req[NODE]);
 988
 989         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
 990                 nid_t nid = le32_to_cpu(entry->nid);
 991                 struct page *node_page;
 992                 struct node_info ni;
 993                 int err;
 994
 995                 /* stop BG_GC if there is not enough free sections. */
 996                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
 997                         return submitted;
 998
 999                 if (check_valid_map(sbi, segno, off) == 0)
1000                         continue;
1001
1002                 if (phase == 0) {
1003                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1004                                                         META_NAT, true);
1005                         continue;
1006                 }
1007
1008                 if (phase == 1) {
1009                         f2fs_ra_node_page(sbi, nid);
1010                         continue;
1011                 }
1012
1013                 /* phase == 2 */
1014                 node_page = f2fs_get_node_page(sbi, nid);
1015                 if (IS_ERR(node_page))
1016                         continue;
1017
1018                 /* block may become invalid during f2fs_get_node_page */
1019                 if (check_valid_map(sbi, segno, off) == 0) {
1020                         f2fs_put_page(node_page, 1);
1021                         continue;
1022                 }
1023
1024                 if (f2fs_get_node_info(sbi, nid, &ni, false)) {
1025                         f2fs_put_page(node_page, 1);
1026                         continue;
1027                 }
1028
1029                 if (ni.blk_addr != start_addr + off) {
1030                         f2fs_put_page(node_page, 1);
1031                         continue;
1032                 }
1033
1034                 err = f2fs_move_node_page(node_page, gc_type);
1035                 if (!err && gc_type == FG_GC)
1036                         submitted++;
1037                 stat_inc_node_blk_count(sbi, 1, gc_type);
1038         }
1039
1040         if (++phase < 3)
1041                 goto next_step;
1042
1043         if (fggc)
1044                 atomic_dec(&sbi->wb_sync_req[NODE]);
1045         return submitted;
1046 }
1047
1048 /*
1049  * Calculate start block index indicating the given node offset.
1050  * Be careful, caller should give this node offset only indicating direct node
1051  * blocks. If any node offsets, which point the other types of node blocks such
1052  * as indirect or double indirect node blocks, are given, it must be a caller's
1053  * bug.
1054  */
1055 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
1056 {
1057         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
1058         unsigned int bidx;
1059
1060         if (node_ofs == 0)
1061                 return 0;
1062
1063         if (node_ofs <= 2) {
1064                 bidx = node_ofs - 1;
1065         } else if (node_ofs <= indirect_blks) {
1066                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
1067
1068                 bidx = node_ofs - 2 - dec;
1069         } else {
1070                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
1071
1072                 bidx = node_ofs - 5 - dec;
1073         }
1074         return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
1075 }
1076
1077 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1078                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
1079 {
1080         struct page *node_page;
1081         nid_t nid;
1082         unsigned int ofs_in_node;
1083         block_t source_blkaddr;
1084
1085         nid = le32_to_cpu(sum->nid);
1086         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
1087
1088         node_page = f2fs_get_node_page(sbi, nid);
1089         if (IS_ERR(node_page))
1090                 return false;
1091
1092         if (f2fs_get_node_info(sbi, nid, dni, false)) {
1093                 f2fs_put_page(node_page, 1);
1094                 return false;
1095         }
1096
1097         if (sum->version != dni->version) {
1098                 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
1099                           __func__);
1100                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1101         }
1102
1103         if (f2fs_check_nid_range(sbi, dni->ino)) {
1104                 f2fs_put_page(node_page, 1);
1105                 return false;
1106         }
1107
1108         *nofs = ofs_of_node(node_page);
1109         source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1110         f2fs_put_page(node_page, 1);
1111
1112         if (source_blkaddr != blkaddr) {
1113 #ifdef CONFIG_F2FS_CHECK_FS
1114                 unsigned int segno = GET_SEGNO(sbi, blkaddr);
1115                 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1116
1117                 if (unlikely(check_valid_map(sbi, segno, offset))) {
1118                         if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1119                                 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",
1120                                          blkaddr, source_blkaddr, segno);
1121                                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1122                         }
1123                 }
1124 #endif
1125                 return false;
1126         }
1127         return true;
1128 }
1129
1130 static int ra_data_block(struct inode *inode, pgoff_t index)
1131 {
1132         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1133         struct address_space *mapping = inode->i_mapping;
1134         struct dnode_of_data dn;
1135         struct page *page;
1136         struct extent_info ei = {0, 0, 0};
1137         struct f2fs_io_info fio = {
1138                 .sbi = sbi,
1139                 .ino = inode->i_ino,
1140                 .type = DATA,
1141                 .temp = COLD,
1142                 .op = REQ_OP_READ,
1143                 .op_flags = 0,
1144                 .encrypted_page = NULL,
1145                 .in_list = false,
1146                 .retry = false,
1147         };
1148         int err;
1149
1150         page = f2fs_grab_cache_page(mapping, index, true);
1151         if (!page)
1152                 return -ENOMEM;
1153
1154         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1155                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1156                 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1157                                                 DATA_GENERIC_ENHANCE_READ))) {
1158                         err = -EFSCORRUPTED;
1159                         goto put_page;
1160                 }
1161                 goto got_it;
1162         }
1163
1164         set_new_dnode(&dn, inode, NULL, NULL, 0);
1165         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1166         if (err)
1167                 goto put_page;
1168         f2fs_put_dnode(&dn);
1169
1170         if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1171                 err = -ENOENT;
1172                 goto put_page;
1173         }
1174         if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1175                                                 DATA_GENERIC_ENHANCE))) {
1176                 err = -EFSCORRUPTED;
1177                 goto put_page;
1178         }
1179 got_it:
1180         /* read page */
1181         fio.page = page;
1182         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1183
1184         /*
1185          * don't cache encrypted data into meta inode until previous dirty
1186          * data were writebacked to avoid racing between GC and flush.
1187          */
1188         f2fs_wait_on_page_writeback(page, DATA, true, true);
1189
1190         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1191
1192         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1193                                         dn.data_blkaddr,
1194                                         FGP_LOCK | FGP_CREAT, GFP_NOFS);
1195         if (!fio.encrypted_page) {
1196                 err = -ENOMEM;
1197                 goto put_page;
1198         }
1199
1200         err = f2fs_submit_page_bio(&fio);
1201         if (err)
1202                 goto put_encrypted_page;
1203         f2fs_put_page(fio.encrypted_page, 0);
1204         f2fs_put_page(page, 1);
1205
1206         f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1207         f2fs_update_iostat(sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1208
1209         return 0;
1210 put_encrypted_page:
1211         f2fs_put_page(fio.encrypted_page, 1);
1212 put_page:
1213         f2fs_put_page(page, 1);
1214         return err;
1215 }
1216
1217 /*
1218  * Move data block via META_MAPPING while keeping locked data page.
1219  * This can be used to move blocks, aka LBAs, directly on disk.
1220  */
1221 static int move_data_block(struct inode *inode, block_t bidx,
1222                                 int gc_type, unsigned int segno, int off)
1223 {
1224         struct f2fs_io_info fio = {
1225                 .sbi = F2FS_I_SB(inode),
1226                 .ino = inode->i_ino,
1227                 .type = DATA,
1228                 .temp = COLD,
1229                 .op = REQ_OP_READ,
1230                 .op_flags = 0,
1231                 .encrypted_page = NULL,
1232                 .in_list = false,
1233                 .retry = false,
1234         };
1235         struct dnode_of_data dn;
1236         struct f2fs_summary sum;
1237         struct node_info ni;
1238         struct page *page, *mpage;
1239         block_t newaddr;
1240         int err = 0;
1241         bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1242         int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&
1243                                 (fio.sbi->gc_mode != GC_URGENT_HIGH) ?
1244                                 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1245
1246         /* do not read out */
1247         page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1248         if (!page)
1249                 return -ENOMEM;
1250
1251         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1252                 err = -ENOENT;
1253                 goto out;
1254         }
1255
1256         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1257         if (err)
1258                 goto out;
1259
1260         set_new_dnode(&dn, inode, NULL, NULL, 0);
1261         err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1262         if (err)
1263                 goto out;
1264
1265         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1266                 ClearPageUptodate(page);
1267                 err = -ENOENT;
1268                 goto put_out;
1269         }
1270
1271         /*
1272          * don't cache encrypted data into meta inode until previous dirty
1273          * data were writebacked to avoid racing between GC and flush.
1274          */
1275         f2fs_wait_on_page_writeback(page, DATA, true, true);
1276
1277         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1278
1279         err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);
1280         if (err)
1281                 goto put_out;
1282
1283         /* read page */
1284         fio.page = page;
1285         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1286
1287         if (lfs_mode)
1288                 f2fs_down_write(&fio.sbi->io_order_lock);
1289
1290         mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1291                                         fio.old_blkaddr, false);
1292         if (!mpage) {
1293                 err = -ENOMEM;
1294                 goto up_out;
1295         }
1296
1297         fio.encrypted_page = mpage;
1298
1299         /* read source block in mpage */
1300         if (!PageUptodate(mpage)) {
1301                 err = f2fs_submit_page_bio(&fio);
1302                 if (err) {
1303                         f2fs_put_page(mpage, 1);
1304                         goto up_out;
1305                 }
1306
1307                 f2fs_update_iostat(fio.sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1308                 f2fs_update_iostat(fio.sbi, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1309
1310                 lock_page(mpage);
1311                 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1312                                                 !PageUptodate(mpage))) {
1313                         err = -EIO;
1314                         f2fs_put_page(mpage, 1);
1315                         goto up_out;
1316                 }
1317         }
1318
1319         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1320
1321         /* allocate block address */
1322         f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1323                                 &sum, type, NULL);
1324
1325         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1326                                 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1327         if (!fio.encrypted_page) {
1328                 err = -ENOMEM;
1329                 f2fs_put_page(mpage, 1);
1330                 goto recover_block;
1331         }
1332
1333         /* write target block */
1334         f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1335         memcpy(page_address(fio.encrypted_page),
1336                                 page_address(mpage), PAGE_SIZE);
1337         f2fs_put_page(mpage, 1);
1338         invalidate_mapping_pages(META_MAPPING(fio.sbi),
1339                                 fio.old_blkaddr, fio.old_blkaddr);
1340         f2fs_invalidate_compress_page(fio.sbi, fio.old_blkaddr);
1341
1342         set_page_dirty(fio.encrypted_page);
1343         if (clear_page_dirty_for_io(fio.encrypted_page))
1344                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
1345
1346         set_page_writeback(fio.encrypted_page);
1347         ClearPageError(page);
1348
1349         fio.op = REQ_OP_WRITE;
1350         fio.op_flags = REQ_SYNC;
1351         fio.new_blkaddr = newaddr;
1352         f2fs_submit_page_write(&fio);
1353         if (fio.retry) {
1354                 err = -EAGAIN;
1355                 if (PageWriteback(fio.encrypted_page))
1356                         end_page_writeback(fio.encrypted_page);
1357                 goto put_page_out;
1358         }
1359
1360         f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
1361
1362         f2fs_update_data_blkaddr(&dn, newaddr);
1363         set_inode_flag(inode, FI_APPEND_WRITE);
1364         if (page->index == 0)
1365                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1366 put_page_out:
1367         f2fs_put_page(fio.encrypted_page, 1);
1368 recover_block:
1369         if (err)
1370                 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1371                                                         true, true, true);
1372 up_out:
1373         if (lfs_mode)
1374                 f2fs_up_write(&fio.sbi->io_order_lock);
1375 put_out:
1376         f2fs_put_dnode(&dn);
1377 out:
1378         f2fs_put_page(page, 1);
1379         return err;
1380 }
1381
1382 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1383                                                         unsigned int segno, int off)
1384 {
1385         struct page *page;
1386         int err = 0;
1387
1388         page = f2fs_get_lock_data_page(inode, bidx, true);
1389         if (IS_ERR(page))
1390                 return PTR_ERR(page);
1391
1392         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1393                 err = -ENOENT;
1394                 goto out;
1395         }
1396
1397         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1398         if (err)
1399                 goto out;
1400
1401         if (gc_type == BG_GC) {
1402                 if (PageWriteback(page)) {
1403                         err = -EAGAIN;
1404                         goto out;
1405                 }
1406                 set_page_dirty(page);
1407                 set_page_private_gcing(page);
1408         } else {
1409                 struct f2fs_io_info fio = {
1410                         .sbi = F2FS_I_SB(inode),
1411                         .ino = inode->i_ino,
1412                         .type = DATA,
1413                         .temp = COLD,
1414                         .op = REQ_OP_WRITE,
1415                         .op_flags = REQ_SYNC,
1416                         .old_blkaddr = NULL_ADDR,
1417                         .page = page,
1418                         .encrypted_page = NULL,
1419                         .need_lock = LOCK_REQ,
1420                         .io_type = FS_GC_DATA_IO,
1421                 };
1422                 bool is_dirty = PageDirty(page);
1423
1424 retry:
1425                 f2fs_wait_on_page_writeback(page, DATA, true, true);
1426
1427                 set_page_dirty(page);
1428                 if (clear_page_dirty_for_io(page)) {
1429                         inode_dec_dirty_pages(inode);
1430                         f2fs_remove_dirty_inode(inode);
1431                 }
1432
1433                 set_page_private_gcing(page);
1434
1435                 err = f2fs_do_write_data_page(&fio);
1436                 if (err) {
1437                         clear_page_private_gcing(page);
1438                         if (err == -ENOMEM) {
1439                                 memalloc_retry_wait(GFP_NOFS);
1440                                 goto retry;
1441                         }
1442                         if (is_dirty)
1443                                 set_page_dirty(page);
1444                 }
1445         }
1446 out:
1447         f2fs_put_page(page, 1);
1448         return err;
1449 }
1450
1451 /*
1452  * This function tries to get parent node of victim data block, and identifies
1453  * data block validity. If the block is valid, copy that with cold status and
1454  * modify parent node.
1455  * If the parent node is not valid or the data block address is different,
1456  * the victim data block is ignored.
1457  */
1458 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1459                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1460                 bool force_migrate)
1461 {
1462         struct super_block *sb = sbi->sb;
1463         struct f2fs_summary *entry;
1464         block_t start_addr;
1465         int off;
1466         int phase = 0;
1467         int submitted = 0;
1468         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1469
1470         start_addr = START_BLOCK(sbi, segno);
1471
1472 next_step:
1473         entry = sum;
1474
1475         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1476                 struct page *data_page;
1477                 struct inode *inode;
1478                 struct node_info dni; /* dnode info for the data */
1479                 unsigned int ofs_in_node, nofs;
1480                 block_t start_bidx;
1481                 nid_t nid = le32_to_cpu(entry->nid);
1482
1483                 /*
1484                  * stop BG_GC if there is not enough free sections.
1485                  * Or, stop GC if the segment becomes fully valid caused by
1486                  * race condition along with SSR block allocation.
1487                  */
1488                 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1489                         (!force_migrate && get_valid_blocks(sbi, segno, true) ==
1490                                                         BLKS_PER_SEC(sbi)))
1491                         return submitted;
1492
1493                 if (check_valid_map(sbi, segno, off) == 0)
1494                         continue;
1495
1496                 if (phase == 0) {
1497                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1498                                                         META_NAT, true);
1499                         continue;
1500                 }
1501
1502                 if (phase == 1) {
1503                         f2fs_ra_node_page(sbi, nid);
1504                         continue;
1505                 }
1506
1507                 /* Get an inode by ino with checking validity */
1508                 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1509                         continue;
1510
1511                 if (phase == 2) {
1512                         f2fs_ra_node_page(sbi, dni.ino);
1513                         continue;
1514                 }
1515
1516                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1517
1518                 if (phase == 3) {
1519                         int err;
1520
1521                         inode = f2fs_iget(sb, dni.ino);
1522                         if (IS_ERR(inode) || is_bad_inode(inode) ||
1523                                         special_file(inode->i_mode))
1524                                 continue;
1525
1526                         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1527                         if (err == -EAGAIN) {
1528                                 iput(inode);
1529                                 return submitted;
1530                         }
1531
1532                         if (!f2fs_down_write_trylock(
1533                                 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1534                                 iput(inode);
1535                                 sbi->skipped_gc_rwsem++;
1536                                 continue;
1537                         }
1538
1539                         start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1540                                                                 ofs_in_node;
1541
1542                         if (f2fs_post_read_required(inode)) {
1543                                 int err = ra_data_block(inode, start_bidx);
1544
1545                                 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1546                                 if (err) {
1547                                         iput(inode);
1548                                         continue;
1549                                 }
1550                                 add_gc_inode(gc_list, inode);
1551                                 continue;
1552                         }
1553
1554                         data_page = f2fs_get_read_data_page(inode,
1555                                                 start_bidx, REQ_RAHEAD, true);
1556                         f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1557                         if (IS_ERR(data_page)) {
1558                                 iput(inode);
1559                                 continue;
1560                         }
1561
1562                         f2fs_put_page(data_page, 0);
1563                         add_gc_inode(gc_list, inode);
1564                         continue;
1565                 }
1566
1567                 /* phase 4 */
1568                 inode = find_gc_inode(gc_list, dni.ino);
1569                 if (inode) {
1570                         struct f2fs_inode_info *fi = F2FS_I(inode);
1571                         bool locked = false;
1572                         int err;
1573
1574                         if (S_ISREG(inode->i_mode)) {
1575                                 if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[READ])) {
1576                                         sbi->skipped_gc_rwsem++;
1577                                         continue;
1578                                 }
1579                                 if (!f2fs_down_write_trylock(
1580                                                 &fi->i_gc_rwsem[WRITE])) {
1581                                         sbi->skipped_gc_rwsem++;
1582                                         f2fs_up_write(&fi->i_gc_rwsem[READ]);
1583                                         continue;
1584                                 }
1585                                 locked = true;
1586
1587                                 /* wait for all inflight aio data */
1588                                 inode_dio_wait(inode);
1589                         }
1590
1591                         start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1592                                                                 + ofs_in_node;
1593                         if (f2fs_post_read_required(inode))
1594                                 err = move_data_block(inode, start_bidx,
1595                                                         gc_type, segno, off);
1596                         else
1597                                 err = move_data_page(inode, start_bidx, gc_type,
1598                                                                 segno, off);
1599
1600                         if (!err && (gc_type == FG_GC ||
1601                                         f2fs_post_read_required(inode)))
1602                                 submitted++;
1603
1604                         if (locked) {
1605                                 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
1606                                 f2fs_up_write(&fi->i_gc_rwsem[READ]);
1607                         }
1608
1609                         stat_inc_data_blk_count(sbi, 1, gc_type);
1610                 }
1611         }
1612
1613         if (++phase < 5)
1614                 goto next_step;
1615
1616         return submitted;
1617 }
1618
1619 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1620                         int gc_type)
1621 {
1622         struct sit_info *sit_i = SIT_I(sbi);
1623         int ret;
1624
1625         down_write(&sit_i->sentry_lock);
1626         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1627                                               NO_CHECK_TYPE, LFS, 0);
1628         up_write(&sit_i->sentry_lock);
1629         return ret;
1630 }
1631
1632 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1633                                 unsigned int start_segno,
1634                                 struct gc_inode_list *gc_list, int gc_type,
1635                                 bool force_migrate)
1636 {
1637         struct page *sum_page;
1638         struct f2fs_summary_block *sum;
1639         struct blk_plug plug;
1640         unsigned int segno = start_segno;
1641         unsigned int end_segno = start_segno + sbi->segs_per_sec;
1642         int seg_freed = 0, migrated = 0;
1643         unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1644                                                 SUM_TYPE_DATA : SUM_TYPE_NODE;
1645         int submitted = 0;
1646
1647         if (__is_large_section(sbi))
1648                 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1649
1650         /*
1651          * zone-capacity can be less than zone-size in zoned devices,
1652          * resulting in less than expected usable segments in the zone,
1653          * calculate the end segno in the zone which can be garbage collected
1654          */
1655         if (f2fs_sb_has_blkzoned(sbi))
1656                 end_segno -= sbi->segs_per_sec -
1657                                         f2fs_usable_segs_in_sec(sbi, segno);
1658
1659         sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1660
1661         /* readahead multi ssa blocks those have contiguous address */
1662         if (__is_large_section(sbi))
1663                 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1664                                         end_segno - segno, META_SSA, true);
1665
1666         /* reference all summary page */
1667         while (segno < end_segno) {
1668                 sum_page = f2fs_get_sum_page(sbi, segno++);
1669                 if (IS_ERR(sum_page)) {
1670                         int err = PTR_ERR(sum_page);
1671
1672                         end_segno = segno - 1;
1673                         for (segno = start_segno; segno < end_segno; segno++) {
1674                                 sum_page = find_get_page(META_MAPPING(sbi),
1675                                                 GET_SUM_BLOCK(sbi, segno));
1676                                 f2fs_put_page(sum_page, 0);
1677                                 f2fs_put_page(sum_page, 0);
1678                         }
1679                         return err;
1680                 }
1681                 unlock_page(sum_page);
1682         }
1683
1684         blk_start_plug(&plug);
1685
1686         for (segno = start_segno; segno < end_segno; segno++) {
1687
1688                 /* find segment summary of victim */
1689                 sum_page = find_get_page(META_MAPPING(sbi),
1690                                         GET_SUM_BLOCK(sbi, segno));
1691                 f2fs_put_page(sum_page, 0);
1692
1693                 if (get_valid_blocks(sbi, segno, false) == 0)
1694                         goto freed;
1695                 if (gc_type == BG_GC && __is_large_section(sbi) &&
1696                                 migrated >= sbi->migration_granularity)
1697                         goto skip;
1698                 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1699                         goto skip;
1700
1701                 sum = page_address(sum_page);
1702                 if (type != GET_SUM_TYPE((&sum->footer))) {
1703                         f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1704                                  segno, type, GET_SUM_TYPE((&sum->footer)));
1705                         set_sbi_flag(sbi, SBI_NEED_FSCK);
1706                         f2fs_stop_checkpoint(sbi, false);
1707                         goto skip;
1708                 }
1709
1710                 /*
1711                  * this is to avoid deadlock:
1712                  * - lock_page(sum_page)         - f2fs_replace_block
1713                  *  - check_valid_map()            - down_write(sentry_lock)
1714                  *   - down_read(sentry_lock)     - change_curseg()
1715                  *                                  - lock_page(sum_page)
1716                  */
1717                 if (type == SUM_TYPE_NODE)
1718                         submitted += gc_node_segment(sbi, sum->entries, segno,
1719                                                                 gc_type);
1720                 else
1721                         submitted += gc_data_segment(sbi, sum->entries, gc_list,
1722                                                         segno, gc_type,
1723                                                         force_migrate);
1724
1725                 stat_inc_seg_count(sbi, type, gc_type);
1726                 sbi->gc_reclaimed_segs[sbi->gc_mode]++;
1727                 migrated++;
1728
1729 freed:
1730                 if (gc_type == FG_GC &&
1731                                 get_valid_blocks(sbi, segno, false) == 0)
1732                         seg_freed++;
1733
1734                 if (__is_large_section(sbi) && segno + 1 < end_segno)
1735                         sbi->next_victim_seg[gc_type] = segno + 1;
1736 skip:
1737                 f2fs_put_page(sum_page, 0);
1738         }
1739
1740         if (submitted)
1741                 f2fs_submit_merged_write(sbi,
1742                                 (type == SUM_TYPE_NODE) ? NODE : DATA);
1743
1744         blk_finish_plug(&plug);
1745
1746         stat_inc_call_count(sbi->stat_info);
1747
1748         return seg_freed;
1749 }
1750
1751 int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control)
1752 {
1753         int gc_type = gc_control->init_gc_type;
1754         unsigned int segno = gc_control->victim_segno;
1755         int sec_freed = 0, seg_freed = 0, total_freed = 0;
1756         int ret = 0;
1757         struct cp_control cpc;
1758         struct gc_inode_list gc_list = {
1759                 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1760                 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1761         };
1762         unsigned int skipped_round = 0, round = 0;
1763
1764         trace_f2fs_gc_begin(sbi->sb, gc_type, gc_control->no_bg_gc,
1765                                 gc_control->nr_free_secs,
1766                                 get_pages(sbi, F2FS_DIRTY_NODES),
1767                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1768                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1769                                 free_sections(sbi),
1770                                 free_segments(sbi),
1771                                 reserved_segments(sbi),
1772                                 prefree_segments(sbi));
1773
1774         cpc.reason = __get_cp_reason(sbi);
1775         sbi->skipped_gc_rwsem = 0;
1776 gc_more:
1777         if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1778                 ret = -EINVAL;
1779                 goto stop;
1780         }
1781         if (unlikely(f2fs_cp_error(sbi))) {
1782                 ret = -EIO;
1783                 goto stop;
1784         }
1785
1786         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1787                 /*
1788                  * For example, if there are many prefree_segments below given
1789                  * threshold, we can make them free by checkpoint. Then, we
1790                  * secure free segments which doesn't need fggc any more.
1791                  */
1792                 if (prefree_segments(sbi)) {
1793                         ret = f2fs_write_checkpoint(sbi, &cpc);
1794                         if (ret)
1795                                 goto stop;
1796                 }
1797                 if (has_not_enough_free_secs(sbi, 0, 0))
1798                         gc_type = FG_GC;
1799         }
1800
1801         /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1802         if (gc_type == BG_GC && gc_control->no_bg_gc) {
1803                 ret = -EINVAL;
1804                 goto stop;
1805         }
1806 retry:
1807         ret = __get_victim(sbi, &segno, gc_type);
1808         if (ret) {
1809                 /* allow to search victim from sections has pinned data */
1810                 if (ret == -ENODATA && gc_type == FG_GC &&
1811                                 f2fs_pinned_section_exists(DIRTY_I(sbi))) {
1812                         f2fs_unpin_all_sections(sbi, false);
1813                         goto retry;
1814                 }
1815                 goto stop;
1816         }
1817
1818         seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type,
1819                                 gc_control->should_migrate_blocks);
1820         total_freed += seg_freed;
1821
1822         if (seg_freed == f2fs_usable_segs_in_sec(sbi, segno))
1823                 sec_freed++;
1824
1825         if (gc_type == FG_GC)
1826                 sbi->cur_victim_sec = NULL_SEGNO;
1827
1828         if (gc_control->init_gc_type == FG_GC ||
1829             !has_not_enough_free_secs(sbi,
1830                                 (gc_type == FG_GC) ? sec_freed : 0, 0)) {
1831                 if (gc_type == FG_GC && sec_freed < gc_control->nr_free_secs)
1832                         goto go_gc_more;
1833                 goto stop;
1834         }
1835
1836         /* FG_GC stops GC by skip_count */
1837         if (gc_type == FG_GC) {
1838                 if (sbi->skipped_gc_rwsem)
1839                         skipped_round++;
1840                 round++;
1841                 if (skipped_round > MAX_SKIP_GC_COUNT &&
1842                                 skipped_round * 2 >= round) {
1843                         ret = f2fs_write_checkpoint(sbi, &cpc);
1844                         goto stop;
1845                 }
1846         }
1847
1848         /* Write checkpoint to reclaim prefree segments */
1849         if (free_sections(sbi) < NR_CURSEG_PERSIST_TYPE &&
1850                                 prefree_segments(sbi)) {
1851                 ret = f2fs_write_checkpoint(sbi, &cpc);
1852                 if (ret)
1853                         goto stop;
1854         }
1855 go_gc_more:
1856         segno = NULL_SEGNO;
1857         goto gc_more;
1858
1859 stop:
1860         SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1861         SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno;
1862
1863         if (gc_type == FG_GC)
1864                 f2fs_unpin_all_sections(sbi, true);
1865
1866         trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1867                                 get_pages(sbi, F2FS_DIRTY_NODES),
1868                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1869                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1870                                 free_sections(sbi),
1871                                 free_segments(sbi),
1872                                 reserved_segments(sbi),
1873                                 prefree_segments(sbi));
1874
1875         f2fs_up_write(&sbi->gc_lock);
1876
1877         put_gc_inode(&gc_list);
1878
1879         if (gc_control->err_gc_skipped && !ret)
1880                 ret = sec_freed ? 0 : -EAGAIN;
1881         return ret;
1882 }
1883
1884 int __init f2fs_create_garbage_collection_cache(void)
1885 {
1886         victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1887                                         sizeof(struct victim_entry));
1888         if (!victim_entry_slab)
1889                 return -ENOMEM;
1890         return 0;
1891 }
1892
1893 void f2fs_destroy_garbage_collection_cache(void)
1894 {
1895         kmem_cache_destroy(victim_entry_slab);
1896 }
1897
1898 static void init_atgc_management(struct f2fs_sb_info *sbi)
1899 {
1900         struct atgc_management *am = &sbi->am;
1901
1902         if (test_opt(sbi, ATGC) &&
1903                 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1904                 am->atgc_enabled = true;
1905
1906         am->root = RB_ROOT_CACHED;
1907         INIT_LIST_HEAD(&am->victim_list);
1908         am->victim_count = 0;
1909
1910         am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1911         am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1912         am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1913         am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
1914 }
1915
1916 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1917 {
1918         DIRTY_I(sbi)->v_ops = &default_v_ops;
1919
1920         sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1921
1922         /* give warm/cold data area from slower device */
1923         if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1924                 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1925                                 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1926
1927         init_atgc_management(sbi);
1928 }
1929
1930 static int free_segment_range(struct f2fs_sb_info *sbi,
1931                                 unsigned int secs, bool gc_only)
1932 {
1933         unsigned int segno, next_inuse, start, end;
1934         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1935         int gc_mode, gc_type;
1936         int err = 0;
1937         int type;
1938
1939         /* Force block allocation for GC */
1940         MAIN_SECS(sbi) -= secs;
1941         start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1942         end = MAIN_SEGS(sbi) - 1;
1943
1944         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1945         for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1946                 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1947                         SIT_I(sbi)->last_victim[gc_mode] = 0;
1948
1949         for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1950                 if (sbi->next_victim_seg[gc_type] >= start)
1951                         sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1952         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1953
1954         /* Move out cursegs from the target range */
1955         for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1956                 f2fs_allocate_segment_for_resize(sbi, type, start, end);
1957
1958         /* do GC to move out valid blocks in the range */
1959         for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1960                 struct gc_inode_list gc_list = {
1961                         .ilist = LIST_HEAD_INIT(gc_list.ilist),
1962                         .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1963                 };
1964
1965                 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
1966                 put_gc_inode(&gc_list);
1967
1968                 if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1969                         err = -EAGAIN;
1970                         goto out;
1971                 }
1972                 if (fatal_signal_pending(current)) {
1973                         err = -ERESTARTSYS;
1974                         goto out;
1975                 }
1976         }
1977         if (gc_only)
1978                 goto out;
1979
1980         err = f2fs_write_checkpoint(sbi, &cpc);
1981         if (err)
1982                 goto out;
1983
1984         next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
1985         if (next_inuse <= end) {
1986                 f2fs_err(sbi, "segno %u should be free but still inuse!",
1987                          next_inuse);
1988                 f2fs_bug_on(sbi, 1);
1989         }
1990 out:
1991         MAIN_SECS(sbi) += secs;
1992         return err;
1993 }
1994
1995 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
1996 {
1997         struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
1998         int section_count;
1999         int segment_count;
2000         int segment_count_main;
2001         long long block_count;
2002         int segs = secs * sbi->segs_per_sec;
2003
2004         f2fs_down_write(&sbi->sb_lock);
2005
2006         section_count = le32_to_cpu(raw_sb->section_count);
2007         segment_count = le32_to_cpu(raw_sb->segment_count);
2008         segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
2009         block_count = le64_to_cpu(raw_sb->block_count);
2010
2011         raw_sb->section_count = cpu_to_le32(section_count + secs);
2012         raw_sb->segment_count = cpu_to_le32(segment_count + segs);
2013         raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
2014         raw_sb->block_count = cpu_to_le64(block_count +
2015                                         (long long)segs * sbi->blocks_per_seg);
2016         if (f2fs_is_multi_device(sbi)) {
2017                 int last_dev = sbi->s_ndevs - 1;
2018                 int dev_segs =
2019                         le32_to_cpu(raw_sb->devs[last_dev].total_segments);
2020
2021                 raw_sb->devs[last_dev].total_segments =
2022                                                 cpu_to_le32(dev_segs + segs);
2023         }
2024
2025         f2fs_up_write(&sbi->sb_lock);
2026 }
2027
2028 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
2029 {
2030         int segs = secs * sbi->segs_per_sec;
2031         long long blks = (long long)segs * sbi->blocks_per_seg;
2032         long long user_block_count =
2033                                 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
2034
2035         SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
2036         MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
2037         MAIN_SECS(sbi) += secs;
2038         FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
2039         FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
2040         F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
2041
2042         if (f2fs_is_multi_device(sbi)) {
2043                 int last_dev = sbi->s_ndevs - 1;
2044
2045                 FDEV(last_dev).total_segments =
2046                                 (int)FDEV(last_dev).total_segments + segs;
2047                 FDEV(last_dev).end_blk =
2048                                 (long long)FDEV(last_dev).end_blk + blks;
2049 #ifdef CONFIG_BLK_DEV_ZONED
2050                 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
2051                                         (int)(blks >> sbi->log_blocks_per_blkz);
2052 #endif
2053         }
2054 }
2055
2056 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
2057 {
2058         __u64 old_block_count, shrunk_blocks;
2059         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
2060         unsigned int secs;
2061         int err = 0;
2062         __u32 rem;
2063
2064         old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
2065         if (block_count > old_block_count)
2066                 return -EINVAL;
2067
2068         if (f2fs_is_multi_device(sbi)) {
2069                 int last_dev = sbi->s_ndevs - 1;
2070                 __u64 last_segs = FDEV(last_dev).total_segments;
2071
2072                 if (block_count + last_segs * sbi->blocks_per_seg <=
2073                                                                 old_block_count)
2074                         return -EINVAL;
2075         }
2076
2077         /* new fs size should align to section size */
2078         div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
2079         if (rem)
2080                 return -EINVAL;
2081
2082         if (block_count == old_block_count)
2083                 return 0;
2084
2085         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2086                 f2fs_err(sbi, "Should run fsck to repair first.");
2087                 return -EFSCORRUPTED;
2088         }
2089
2090         if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2091                 f2fs_err(sbi, "Checkpoint should be enabled.");
2092                 return -EINVAL;
2093         }
2094
2095         shrunk_blocks = old_block_count - block_count;
2096         secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2097
2098         /* stop other GC */
2099         if (!f2fs_down_write_trylock(&sbi->gc_lock))
2100                 return -EAGAIN;
2101
2102         /* stop CP to protect MAIN_SEC in free_segment_range */
2103         f2fs_lock_op(sbi);
2104
2105         spin_lock(&sbi->stat_lock);
2106         if (shrunk_blocks + valid_user_blocks(sbi) +
2107                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2108                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2109                 err = -ENOSPC;
2110         spin_unlock(&sbi->stat_lock);
2111
2112         if (err)
2113                 goto out_unlock;
2114
2115         err = free_segment_range(sbi, secs, true);
2116
2117 out_unlock:
2118         f2fs_unlock_op(sbi);
2119         f2fs_up_write(&sbi->gc_lock);
2120         if (err)
2121                 return err;
2122
2123         set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2124
2125         freeze_super(sbi->sb);
2126         f2fs_down_write(&sbi->gc_lock);
2127         f2fs_down_write(&sbi->cp_global_sem);
2128
2129         spin_lock(&sbi->stat_lock);
2130         if (shrunk_blocks + valid_user_blocks(sbi) +
2131                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2132                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2133                 err = -ENOSPC;
2134         else
2135                 sbi->user_block_count -= shrunk_blocks;
2136         spin_unlock(&sbi->stat_lock);
2137         if (err)
2138                 goto out_err;
2139
2140         err = free_segment_range(sbi, secs, false);
2141         if (err)
2142                 goto recover_out;
2143
2144         update_sb_metadata(sbi, -secs);
2145
2146         err = f2fs_commit_super(sbi, false);
2147         if (err) {
2148                 update_sb_metadata(sbi, secs);
2149                 goto recover_out;
2150         }
2151
2152         update_fs_metadata(sbi, -secs);
2153         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2154         set_sbi_flag(sbi, SBI_IS_DIRTY);
2155
2156         err = f2fs_write_checkpoint(sbi, &cpc);
2157         if (err) {
2158                 update_fs_metadata(sbi, secs);
2159                 update_sb_metadata(sbi, secs);
2160                 f2fs_commit_super(sbi, false);
2161         }
2162 recover_out:
2163         if (err) {
2164                 set_sbi_flag(sbi, SBI_NEED_FSCK);
2165                 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2166
2167                 spin_lock(&sbi->stat_lock);
2168                 sbi->user_block_count += shrunk_blocks;
2169                 spin_unlock(&sbi->stat_lock);
2170         }
2171 out_err:
2172         f2fs_up_write(&sbi->cp_global_sem);
2173         f2fs_up_write(&sbi->gc_lock);
2174         thaw_super(sbi->sb);
2175         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2176         return err;
2177 }