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 = false;
  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_stop_checkpoint(sbi, false,
  77                                         STOP_CP_REASON_FAULT_INJECT);
  78
  79                 if (!sb_start_write_trylock(sbi->sb)) {
  80                         stat_other_skip_bggc_count(sbi);
  81                         continue;
  82                 }
  83
  84                 /*
  85                  * [GC triggering condition]
  86                  * 0. GC is not conducted currently.
  87                  * 1. There are enough dirty segments.
  88                  * 2. IO subsystem is idle by checking the # of writeback pages.
  89                  * 3. IO subsystem is idle by checking the # of requests in
  90                  *    bdev's request list.
  91                  *
  92                  * Note) We have to avoid triggering GCs frequently.
  93                  * Because it is possible that some segments can be
  94                  * invalidated soon after by user update or deletion.
  95                  * So, I'd like to wait some time to collect dirty segments.
  96                  */
  97                 if (sbi->gc_mode == GC_URGENT_HIGH ||
  98                                 sbi->gc_mode == GC_URGENT_MID) {
  99                         wait_ms = gc_th->urgent_sleep_time;
 100                         f2fs_down_write(&sbi->gc_lock);
 101                         goto do_gc;
 102                 }
 103
 104                 if (foreground) {
 105                         f2fs_down_write(&sbi->gc_lock);
 106                         goto do_gc;
 107                 } else if (!f2fs_down_write_trylock(&sbi->gc_lock)) {
 108                         stat_other_skip_bggc_count(sbi);
 109                         goto next;
 110                 }
 111
 112                 if (!is_idle(sbi, GC_TIME)) {
 113                         increase_sleep_time(gc_th, &wait_ms);
 114                         f2fs_up_write(&sbi->gc_lock);
 115                         stat_io_skip_bggc_count(sbi);
 116                         goto next;
 117                 }
 118
 119                 if (has_enough_invalid_blocks(sbi))
 120                         decrease_sleep_time(gc_th, &wait_ms);
 121                 else
 122                         increase_sleep_time(gc_th, &wait_ms);
 123 do_gc:
 124                 if (!foreground)
 125                         stat_inc_bggc_count(sbi->stat_info);
 126
 127                 sync_mode = F2FS_OPTION(sbi).bggc_mode == BGGC_MODE_SYNC;
 128
 129                 /* foreground GC was been triggered via f2fs_balance_fs() */
 130                 if (foreground)
 131                         sync_mode = false;
 132
 133                 gc_control.init_gc_type = sync_mode ? FG_GC : BG_GC;
 134                 gc_control.no_bg_gc = foreground;
 135                 gc_control.nr_free_secs = foreground ? 1 : 0;
 136
 137                 /* if return value is not zero, no victim was selected */
 138                 if (f2fs_gc(sbi, &gc_control)) {
 139                         /* don't bother wait_ms by foreground gc */
 140                         if (!foreground)
 141                                 wait_ms = gc_th->no_gc_sleep_time;
 142                 } else {
 143                         /* reset wait_ms to default sleep time */
 144                         if (wait_ms == gc_th->no_gc_sleep_time)
 145                                 wait_ms = gc_th->min_sleep_time;
 146                 }
 147
 148                 if (foreground)
 149                         wake_up_all(&gc_th->fggc_wq);
 150
 151                 trace_f2fs_background_gc(sbi->sb, wait_ms,
 152                                 prefree_segments(sbi), free_segments(sbi));
 153
 154                 /* balancing f2fs's metadata periodically */
 155                 f2fs_balance_fs_bg(sbi, true);
 156 next:
 157                 if (sbi->gc_mode != GC_NORMAL) {
 158                         spin_lock(&sbi->gc_remaining_trials_lock);
 159                         if (sbi->gc_remaining_trials) {
 160                                 sbi->gc_remaining_trials--;
 161                                 if (!sbi->gc_remaining_trials)
 162                                         sbi->gc_mode = GC_NORMAL;
 163                         }
 164                         spin_unlock(&sbi->gc_remaining_trials_lock);
 165                 }
 166                 sb_end_write(sbi->sb);
 167
 168         } while (!kthread_should_stop());
 169         return 0;
 170 }
 171
 172 int f2fs_start_gc_thread(struct f2fs_sb_info *sbi)
 173 {
 174         struct f2fs_gc_kthread *gc_th;
 175         dev_t dev = sbi->sb->s_bdev->bd_dev;
 176
 177         gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 178         if (!gc_th)
 179                 return -ENOMEM;
 180
 181         gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
 182         gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 183         gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 184         gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 185
 186         gc_th->gc_wake = false;
 187
 188         sbi->gc_thread = gc_th;
 189         init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
 190         init_waitqueue_head(&sbi->gc_thread->fggc_wq);
 191         sbi->gc_thread->f2fs_gc_task = kthread_run(gc_thread_func, sbi,
 192                         "f2fs_gc-%u:%u", MAJOR(dev), MINOR(dev));
 193         if (IS_ERR(gc_th->f2fs_gc_task)) {
 194                 int err = PTR_ERR(gc_th->f2fs_gc_task);
 195
 196                 kfree(gc_th);
 197                 sbi->gc_thread = NULL;
 198                 return err;
 199         }
 200
 201         return 0;
 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 = get_random_u32_below(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 = CAP_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, max_addrs, base;
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         if (IS_INODE(node_page)) {
1109                 base = offset_in_addr(F2FS_INODE(node_page));
1110                 max_addrs = DEF_ADDRS_PER_INODE;
1111         } else {
1112                 base = 0;
1113                 max_addrs = DEF_ADDRS_PER_BLOCK;
1114         }
1115
1116         if (base + ofs_in_node >= max_addrs) {
1117                 f2fs_err(sbi, "Inconsistent blkaddr offset: base:%u, ofs_in_node:%u, max:%u, ino:%u, nid:%u",
1118                         base, ofs_in_node, max_addrs, dni->ino, dni->nid);
1119                 f2fs_put_page(node_page, 1);
1120                 return false;
1121         }
1122
1123         *nofs = ofs_of_node(node_page);
1124         source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1125         f2fs_put_page(node_page, 1);
1126
1127         if (source_blkaddr != blkaddr) {
1128 #ifdef CONFIG_F2FS_CHECK_FS
1129                 unsigned int segno = GET_SEGNO(sbi, blkaddr);
1130                 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1131
1132                 if (unlikely(check_valid_map(sbi, segno, offset))) {
1133                         if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1134                                 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",
1135                                          blkaddr, source_blkaddr, segno);
1136                                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1137                         }
1138                 }
1139 #endif
1140                 return false;
1141         }
1142         return true;
1143 }
1144
1145 static int ra_data_block(struct inode *inode, pgoff_t index)
1146 {
1147         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1148         struct address_space *mapping = inode->i_mapping;
1149         struct dnode_of_data dn;
1150         struct page *page;
1151         struct f2fs_io_info fio = {
1152                 .sbi = sbi,
1153                 .ino = inode->i_ino,
1154                 .type = DATA,
1155                 .temp = COLD,
1156                 .op = REQ_OP_READ,
1157                 .op_flags = 0,
1158                 .encrypted_page = NULL,
1159                 .in_list = 0,
1160                 .retry = 0,
1161         };
1162         int err;
1163
1164         page = f2fs_grab_cache_page(mapping, index, true);
1165         if (!page)
1166                 return -ENOMEM;
1167
1168         if (f2fs_lookup_read_extent_cache_block(inode, index,
1169                                                 &dn.data_blkaddr)) {
1170                 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1171                                                 DATA_GENERIC_ENHANCE_READ))) {
1172                         err = -EFSCORRUPTED;
1173                         f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1174                         goto put_page;
1175                 }
1176                 goto got_it;
1177         }
1178
1179         set_new_dnode(&dn, inode, NULL, NULL, 0);
1180         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1181         if (err)
1182                 goto put_page;
1183         f2fs_put_dnode(&dn);
1184
1185         if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1186                 err = -ENOENT;
1187                 goto put_page;
1188         }
1189         if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1190                                                 DATA_GENERIC_ENHANCE))) {
1191                 err = -EFSCORRUPTED;
1192                 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1193                 goto put_page;
1194         }
1195 got_it:
1196         /* read page */
1197         fio.page = page;
1198         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1199
1200         /*
1201          * don't cache encrypted data into meta inode until previous dirty
1202          * data were writebacked to avoid racing between GC and flush.
1203          */
1204         f2fs_wait_on_page_writeback(page, DATA, true, true);
1205
1206         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1207
1208         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1209                                         dn.data_blkaddr,
1210                                         FGP_LOCK | FGP_CREAT, GFP_NOFS);
1211         if (!fio.encrypted_page) {
1212                 err = -ENOMEM;
1213                 goto put_page;
1214         }
1215
1216         err = f2fs_submit_page_bio(&fio);
1217         if (err)
1218                 goto put_encrypted_page;
1219         f2fs_put_page(fio.encrypted_page, 0);
1220         f2fs_put_page(page, 1);
1221
1222         f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
1223         f2fs_update_iostat(sbi, NULL, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1224
1225         return 0;
1226 put_encrypted_page:
1227         f2fs_put_page(fio.encrypted_page, 1);
1228 put_page:
1229         f2fs_put_page(page, 1);
1230         return err;
1231 }
1232
1233 /*
1234  * Move data block via META_MAPPING while keeping locked data page.
1235  * This can be used to move blocks, aka LBAs, directly on disk.
1236  */
1237 static int move_data_block(struct inode *inode, block_t bidx,
1238                                 int gc_type, unsigned int segno, int off)
1239 {
1240         struct f2fs_io_info fio = {
1241                 .sbi = F2FS_I_SB(inode),
1242                 .ino = inode->i_ino,
1243                 .type = DATA,
1244                 .temp = COLD,
1245                 .op = REQ_OP_READ,
1246                 .op_flags = 0,
1247                 .encrypted_page = NULL,
1248                 .in_list = 0,
1249                 .retry = 0,
1250         };
1251         struct dnode_of_data dn;
1252         struct f2fs_summary sum;
1253         struct node_info ni;
1254         struct page *page, *mpage;
1255         block_t newaddr;
1256         int err = 0;
1257         bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1258         int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&
1259                                 (fio.sbi->gc_mode != GC_URGENT_HIGH) ?
1260                                 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1261
1262         /* do not read out */
1263         page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1264         if (!page)
1265                 return -ENOMEM;
1266
1267         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1268                 err = -ENOENT;
1269                 goto out;
1270         }
1271
1272         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1273         if (err)
1274                 goto out;
1275
1276         set_new_dnode(&dn, inode, NULL, NULL, 0);
1277         err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1278         if (err)
1279                 goto out;
1280
1281         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1282                 ClearPageUptodate(page);
1283                 err = -ENOENT;
1284                 goto put_out;
1285         }
1286
1287         /*
1288          * don't cache encrypted data into meta inode until previous dirty
1289          * data were writebacked to avoid racing between GC and flush.
1290          */
1291         f2fs_wait_on_page_writeback(page, DATA, true, true);
1292
1293         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1294
1295         err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);
1296         if (err)
1297                 goto put_out;
1298
1299         /* read page */
1300         fio.page = page;
1301         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1302
1303         if (lfs_mode)
1304                 f2fs_down_write(&fio.sbi->io_order_lock);
1305
1306         mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1307                                         fio.old_blkaddr, false);
1308         if (!mpage) {
1309                 err = -ENOMEM;
1310                 goto up_out;
1311         }
1312
1313         fio.encrypted_page = mpage;
1314
1315         /* read source block in mpage */
1316         if (!PageUptodate(mpage)) {
1317                 err = f2fs_submit_page_bio(&fio);
1318                 if (err) {
1319                         f2fs_put_page(mpage, 1);
1320                         goto up_out;
1321                 }
1322
1323                 f2fs_update_iostat(fio.sbi, inode, FS_DATA_READ_IO,
1324                                                         F2FS_BLKSIZE);
1325                 f2fs_update_iostat(fio.sbi, NULL, FS_GDATA_READ_IO,
1326                                                         F2FS_BLKSIZE);
1327
1328                 lock_page(mpage);
1329                 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1330                                                 !PageUptodate(mpage))) {
1331                         err = -EIO;
1332                         f2fs_put_page(mpage, 1);
1333                         goto up_out;
1334                 }
1335         }
1336
1337         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1338
1339         /* allocate block address */
1340         f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1341                                 &sum, type, NULL);
1342
1343         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1344                                 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1345         if (!fio.encrypted_page) {
1346                 err = -ENOMEM;
1347                 f2fs_put_page(mpage, 1);
1348                 goto recover_block;
1349         }
1350
1351         /* write target block */
1352         f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1353         memcpy(page_address(fio.encrypted_page),
1354                                 page_address(mpage), PAGE_SIZE);
1355         f2fs_put_page(mpage, 1);
1356         invalidate_mapping_pages(META_MAPPING(fio.sbi),
1357                                 fio.old_blkaddr, fio.old_blkaddr);
1358         f2fs_invalidate_compress_page(fio.sbi, fio.old_blkaddr);
1359
1360         set_page_dirty(fio.encrypted_page);
1361         if (clear_page_dirty_for_io(fio.encrypted_page))
1362                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
1363
1364         set_page_writeback(fio.encrypted_page);
1365
1366         fio.op = REQ_OP_WRITE;
1367         fio.op_flags = REQ_SYNC;
1368         fio.new_blkaddr = newaddr;
1369         f2fs_submit_page_write(&fio);
1370         if (fio.retry) {
1371                 err = -EAGAIN;
1372                 if (PageWriteback(fio.encrypted_page))
1373                         end_page_writeback(fio.encrypted_page);
1374                 goto put_page_out;
1375         }
1376
1377         f2fs_update_iostat(fio.sbi, NULL, FS_GC_DATA_IO, F2FS_BLKSIZE);
1378
1379         f2fs_update_data_blkaddr(&dn, newaddr);
1380         set_inode_flag(inode, FI_APPEND_WRITE);
1381         if (page->index == 0)
1382                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1383 put_page_out:
1384         f2fs_put_page(fio.encrypted_page, 1);
1385 recover_block:
1386         if (err)
1387                 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1388                                                         true, true, true);
1389 up_out:
1390         if (lfs_mode)
1391                 f2fs_up_write(&fio.sbi->io_order_lock);
1392 put_out:
1393         f2fs_put_dnode(&dn);
1394 out:
1395         f2fs_put_page(page, 1);
1396         return err;
1397 }
1398
1399 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1400                                                         unsigned int segno, int off)
1401 {
1402         struct page *page;
1403         int err = 0;
1404
1405         page = f2fs_get_lock_data_page(inode, bidx, true);
1406         if (IS_ERR(page))
1407                 return PTR_ERR(page);
1408
1409         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1410                 err = -ENOENT;
1411                 goto out;
1412         }
1413
1414         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1415         if (err)
1416                 goto out;
1417
1418         if (gc_type == BG_GC) {
1419                 if (PageWriteback(page)) {
1420                         err = -EAGAIN;
1421                         goto out;
1422                 }
1423                 set_page_dirty(page);
1424                 set_page_private_gcing(page);
1425         } else {
1426                 struct f2fs_io_info fio = {
1427                         .sbi = F2FS_I_SB(inode),
1428                         .ino = inode->i_ino,
1429                         .type = DATA,
1430                         .temp = COLD,
1431                         .op = REQ_OP_WRITE,
1432                         .op_flags = REQ_SYNC,
1433                         .old_blkaddr = NULL_ADDR,
1434                         .page = page,
1435                         .encrypted_page = NULL,
1436                         .need_lock = LOCK_REQ,
1437                         .io_type = FS_GC_DATA_IO,
1438                 };
1439                 bool is_dirty = PageDirty(page);
1440
1441 retry:
1442                 f2fs_wait_on_page_writeback(page, DATA, true, true);
1443
1444                 set_page_dirty(page);
1445                 if (clear_page_dirty_for_io(page)) {
1446                         inode_dec_dirty_pages(inode);
1447                         f2fs_remove_dirty_inode(inode);
1448                 }
1449
1450                 set_page_private_gcing(page);
1451
1452                 err = f2fs_do_write_data_page(&fio);
1453                 if (err) {
1454                         clear_page_private_gcing(page);
1455                         if (err == -ENOMEM) {
1456                                 memalloc_retry_wait(GFP_NOFS);
1457                                 goto retry;
1458                         }
1459                         if (is_dirty)
1460                                 set_page_dirty(page);
1461                 }
1462         }
1463 out:
1464         f2fs_put_page(page, 1);
1465         return err;
1466 }
1467
1468 /*
1469  * This function tries to get parent node of victim data block, and identifies
1470  * data block validity. If the block is valid, copy that with cold status and
1471  * modify parent node.
1472  * If the parent node is not valid or the data block address is different,
1473  * the victim data block is ignored.
1474  */
1475 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1476                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1477                 bool force_migrate)
1478 {
1479         struct super_block *sb = sbi->sb;
1480         struct f2fs_summary *entry;
1481         block_t start_addr;
1482         int off;
1483         int phase = 0;
1484         int submitted = 0;
1485         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1486
1487         start_addr = START_BLOCK(sbi, segno);
1488
1489 next_step:
1490         entry = sum;
1491
1492         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1493                 struct page *data_page;
1494                 struct inode *inode;
1495                 struct node_info dni; /* dnode info for the data */
1496                 unsigned int ofs_in_node, nofs;
1497                 block_t start_bidx;
1498                 nid_t nid = le32_to_cpu(entry->nid);
1499
1500                 /*
1501                  * stop BG_GC if there is not enough free sections.
1502                  * Or, stop GC if the segment becomes fully valid caused by
1503                  * race condition along with SSR block allocation.
1504                  */
1505                 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1506                         (!force_migrate && get_valid_blocks(sbi, segno, true) ==
1507                                                         CAP_BLKS_PER_SEC(sbi)))
1508                         return submitted;
1509
1510                 if (check_valid_map(sbi, segno, off) == 0)
1511                         continue;
1512
1513                 if (phase == 0) {
1514                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1515                                                         META_NAT, true);
1516                         continue;
1517                 }
1518
1519                 if (phase == 1) {
1520                         f2fs_ra_node_page(sbi, nid);
1521                         continue;
1522                 }
1523
1524                 /* Get an inode by ino with checking validity */
1525                 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1526                         continue;
1527
1528                 if (phase == 2) {
1529                         f2fs_ra_node_page(sbi, dni.ino);
1530                         continue;
1531                 }
1532
1533                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1534
1535                 if (phase == 3) {
1536                         int err;
1537
1538                         inode = f2fs_iget(sb, dni.ino);
1539                         if (IS_ERR(inode) || is_bad_inode(inode) ||
1540                                         special_file(inode->i_mode))
1541                                 continue;
1542
1543                         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1544                         if (err == -EAGAIN) {
1545                                 iput(inode);
1546                                 return submitted;
1547                         }
1548
1549                         if (!f2fs_down_write_trylock(
1550                                 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1551                                 iput(inode);
1552                                 sbi->skipped_gc_rwsem++;
1553                                 continue;
1554                         }
1555
1556                         start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1557                                                                 ofs_in_node;
1558
1559                         if (f2fs_post_read_required(inode)) {
1560                                 int err = ra_data_block(inode, start_bidx);
1561
1562                                 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1563                                 if (err) {
1564                                         iput(inode);
1565                                         continue;
1566                                 }
1567                                 add_gc_inode(gc_list, inode);
1568                                 continue;
1569                         }
1570
1571                         data_page = f2fs_get_read_data_page(inode, start_bidx,
1572                                                         REQ_RAHEAD, true, NULL);
1573                         f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1574                         if (IS_ERR(data_page)) {
1575                                 iput(inode);
1576                                 continue;
1577                         }
1578
1579                         f2fs_put_page(data_page, 0);
1580                         add_gc_inode(gc_list, inode);
1581                         continue;
1582                 }
1583
1584                 /* phase 4 */
1585                 inode = find_gc_inode(gc_list, dni.ino);
1586                 if (inode) {
1587                         struct f2fs_inode_info *fi = F2FS_I(inode);
1588                         bool locked = false;
1589                         int err;
1590
1591                         if (S_ISREG(inode->i_mode)) {
1592                                 if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[READ])) {
1593                                         sbi->skipped_gc_rwsem++;
1594                                         continue;
1595                                 }
1596                                 if (!f2fs_down_write_trylock(
1597                                                 &fi->i_gc_rwsem[WRITE])) {
1598                                         sbi->skipped_gc_rwsem++;
1599                                         f2fs_up_write(&fi->i_gc_rwsem[READ]);
1600                                         continue;
1601                                 }
1602                                 locked = true;
1603
1604                                 /* wait for all inflight aio data */
1605                                 inode_dio_wait(inode);
1606                         }
1607
1608                         start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1609                                                                 + ofs_in_node;
1610                         if (f2fs_post_read_required(inode))
1611                                 err = move_data_block(inode, start_bidx,
1612                                                         gc_type, segno, off);
1613                         else
1614                                 err = move_data_page(inode, start_bidx, gc_type,
1615                                                                 segno, off);
1616
1617                         if (!err && (gc_type == FG_GC ||
1618                                         f2fs_post_read_required(inode)))
1619                                 submitted++;
1620
1621                         if (locked) {
1622                                 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
1623                                 f2fs_up_write(&fi->i_gc_rwsem[READ]);
1624                         }
1625
1626                         stat_inc_data_blk_count(sbi, 1, gc_type);
1627                 }
1628         }
1629
1630         if (++phase < 5)
1631                 goto next_step;
1632
1633         return submitted;
1634 }
1635
1636 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1637                         int gc_type)
1638 {
1639         struct sit_info *sit_i = SIT_I(sbi);
1640         int ret;
1641
1642         down_write(&sit_i->sentry_lock);
1643         ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
1644                                               NO_CHECK_TYPE, LFS, 0);
1645         up_write(&sit_i->sentry_lock);
1646         return ret;
1647 }
1648
1649 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1650                                 unsigned int start_segno,
1651                                 struct gc_inode_list *gc_list, int gc_type,
1652                                 bool force_migrate)
1653 {
1654         struct page *sum_page;
1655         struct f2fs_summary_block *sum;
1656         struct blk_plug plug;
1657         unsigned int segno = start_segno;
1658         unsigned int end_segno = start_segno + sbi->segs_per_sec;
1659         int seg_freed = 0, migrated = 0;
1660         unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1661                                                 SUM_TYPE_DATA : SUM_TYPE_NODE;
1662         int submitted = 0;
1663
1664         if (__is_large_section(sbi))
1665                 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1666
1667         /*
1668          * zone-capacity can be less than zone-size in zoned devices,
1669          * resulting in less than expected usable segments in the zone,
1670          * calculate the end segno in the zone which can be garbage collected
1671          */
1672         if (f2fs_sb_has_blkzoned(sbi))
1673                 end_segno -= sbi->segs_per_sec -
1674                                         f2fs_usable_segs_in_sec(sbi, segno);
1675
1676         sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1677
1678         /* readahead multi ssa blocks those have contiguous address */
1679         if (__is_large_section(sbi))
1680                 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1681                                         end_segno - segno, META_SSA, true);
1682
1683         /* reference all summary page */
1684         while (segno < end_segno) {
1685                 sum_page = f2fs_get_sum_page(sbi, segno++);
1686                 if (IS_ERR(sum_page)) {
1687                         int err = PTR_ERR(sum_page);
1688
1689                         end_segno = segno - 1;
1690                         for (segno = start_segno; segno < end_segno; segno++) {
1691                                 sum_page = find_get_page(META_MAPPING(sbi),
1692                                                 GET_SUM_BLOCK(sbi, segno));
1693                                 f2fs_put_page(sum_page, 0);
1694                                 f2fs_put_page(sum_page, 0);
1695                         }
1696                         return err;
1697                 }
1698                 unlock_page(sum_page);
1699         }
1700
1701         blk_start_plug(&plug);
1702
1703         for (segno = start_segno; segno < end_segno; segno++) {
1704
1705                 /* find segment summary of victim */
1706                 sum_page = find_get_page(META_MAPPING(sbi),
1707                                         GET_SUM_BLOCK(sbi, segno));
1708                 f2fs_put_page(sum_page, 0);
1709
1710                 if (get_valid_blocks(sbi, segno, false) == 0)
1711                         goto freed;
1712                 if (gc_type == BG_GC && __is_large_section(sbi) &&
1713                                 migrated >= sbi->migration_granularity)
1714                         goto skip;
1715                 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1716                         goto skip;
1717
1718                 sum = page_address(sum_page);
1719                 if (type != GET_SUM_TYPE((&sum->footer))) {
1720                         f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1721                                  segno, type, GET_SUM_TYPE((&sum->footer)));
1722                         set_sbi_flag(sbi, SBI_NEED_FSCK);
1723                         f2fs_stop_checkpoint(sbi, false,
1724                                 STOP_CP_REASON_CORRUPTED_SUMMARY);
1725                         goto skip;
1726                 }
1727
1728                 /*
1729                  * this is to avoid deadlock:
1730                  * - lock_page(sum_page)         - f2fs_replace_block
1731                  *  - check_valid_map()            - down_write(sentry_lock)
1732                  *   - down_read(sentry_lock)     - change_curseg()
1733                  *                                  - lock_page(sum_page)
1734                  */
1735                 if (type == SUM_TYPE_NODE)
1736                         submitted += gc_node_segment(sbi, sum->entries, segno,
1737                                                                 gc_type);
1738                 else
1739                         submitted += gc_data_segment(sbi, sum->entries, gc_list,
1740                                                         segno, gc_type,
1741                                                         force_migrate);
1742
1743                 stat_inc_seg_count(sbi, type, gc_type);
1744                 sbi->gc_reclaimed_segs[sbi->gc_mode]++;
1745                 migrated++;
1746
1747 freed:
1748                 if (gc_type == FG_GC &&
1749                                 get_valid_blocks(sbi, segno, false) == 0)
1750                         seg_freed++;
1751
1752                 if (__is_large_section(sbi))
1753                         sbi->next_victim_seg[gc_type] =
1754                                 (segno + 1 < end_segno) ? segno + 1 : NULL_SEGNO;
1755 skip:
1756                 f2fs_put_page(sum_page, 0);
1757         }
1758
1759         if (submitted)
1760                 f2fs_submit_merged_write(sbi,
1761                                 (type == SUM_TYPE_NODE) ? NODE : DATA);
1762
1763         blk_finish_plug(&plug);
1764
1765         stat_inc_call_count(sbi->stat_info);
1766
1767         return seg_freed;
1768 }
1769
1770 int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control)
1771 {
1772         int gc_type = gc_control->init_gc_type;
1773         unsigned int segno = gc_control->victim_segno;
1774         int sec_freed = 0, seg_freed = 0, total_freed = 0;
1775         int ret = 0;
1776         struct cp_control cpc;
1777         struct gc_inode_list gc_list = {
1778                 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1779                 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1780         };
1781         unsigned int skipped_round = 0, round = 0;
1782
1783         trace_f2fs_gc_begin(sbi->sb, gc_type, gc_control->no_bg_gc,
1784                                 gc_control->nr_free_secs,
1785                                 get_pages(sbi, F2FS_DIRTY_NODES),
1786                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1787                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1788                                 free_sections(sbi),
1789                                 free_segments(sbi),
1790                                 reserved_segments(sbi),
1791                                 prefree_segments(sbi));
1792
1793         cpc.reason = __get_cp_reason(sbi);
1794         sbi->skipped_gc_rwsem = 0;
1795 gc_more:
1796         if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1797                 ret = -EINVAL;
1798                 goto stop;
1799         }
1800         if (unlikely(f2fs_cp_error(sbi))) {
1801                 ret = -EIO;
1802                 goto stop;
1803         }
1804
1805         if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
1806                 /*
1807                  * For example, if there are many prefree_segments below given
1808                  * threshold, we can make them free by checkpoint. Then, we
1809                  * secure free segments which doesn't need fggc any more.
1810                  */
1811                 if (prefree_segments(sbi)) {
1812                         ret = f2fs_write_checkpoint(sbi, &cpc);
1813                         if (ret)
1814                                 goto stop;
1815                 }
1816                 if (has_not_enough_free_secs(sbi, 0, 0))
1817                         gc_type = FG_GC;
1818         }
1819
1820         /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1821         if (gc_type == BG_GC && gc_control->no_bg_gc) {
1822                 ret = -EINVAL;
1823                 goto stop;
1824         }
1825 retry:
1826         ret = __get_victim(sbi, &segno, gc_type);
1827         if (ret) {
1828                 /* allow to search victim from sections has pinned data */
1829                 if (ret == -ENODATA && gc_type == FG_GC &&
1830                                 f2fs_pinned_section_exists(DIRTY_I(sbi))) {
1831                         f2fs_unpin_all_sections(sbi, false);
1832                         goto retry;
1833                 }
1834                 goto stop;
1835         }
1836
1837         seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type,
1838                                 gc_control->should_migrate_blocks);
1839         total_freed += seg_freed;
1840
1841         if (seg_freed == f2fs_usable_segs_in_sec(sbi, segno))
1842                 sec_freed++;
1843
1844         if (gc_type == FG_GC)
1845                 sbi->cur_victim_sec = NULL_SEGNO;
1846
1847         if (gc_control->init_gc_type == FG_GC ||
1848             !has_not_enough_free_secs(sbi,
1849                                 (gc_type == FG_GC) ? sec_freed : 0, 0)) {
1850                 if (gc_type == FG_GC && sec_freed < gc_control->nr_free_secs)
1851                         goto go_gc_more;
1852                 goto stop;
1853         }
1854
1855         /* FG_GC stops GC by skip_count */
1856         if (gc_type == FG_GC) {
1857                 if (sbi->skipped_gc_rwsem)
1858                         skipped_round++;
1859                 round++;
1860                 if (skipped_round > MAX_SKIP_GC_COUNT &&
1861                                 skipped_round * 2 >= round) {
1862                         ret = f2fs_write_checkpoint(sbi, &cpc);
1863                         goto stop;
1864                 }
1865         }
1866
1867         /* Write checkpoint to reclaim prefree segments */
1868         if (free_sections(sbi) < NR_CURSEG_PERSIST_TYPE &&
1869                                 prefree_segments(sbi)) {
1870                 ret = f2fs_write_checkpoint(sbi, &cpc);
1871                 if (ret)
1872                         goto stop;
1873         }
1874 go_gc_more:
1875         segno = NULL_SEGNO;
1876         goto gc_more;
1877
1878 stop:
1879         SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1880         SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno;
1881
1882         if (gc_type == FG_GC)
1883                 f2fs_unpin_all_sections(sbi, true);
1884
1885         trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
1886                                 get_pages(sbi, F2FS_DIRTY_NODES),
1887                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1888                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1889                                 free_sections(sbi),
1890                                 free_segments(sbi),
1891                                 reserved_segments(sbi),
1892                                 prefree_segments(sbi));
1893
1894         f2fs_up_write(&sbi->gc_lock);
1895
1896         put_gc_inode(&gc_list);
1897
1898         if (gc_control->err_gc_skipped && !ret)
1899                 ret = sec_freed ? 0 : -EAGAIN;
1900         return ret;
1901 }
1902
1903 int __init f2fs_create_garbage_collection_cache(void)
1904 {
1905         victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1906                                         sizeof(struct victim_entry));
1907         return victim_entry_slab ? 0 : -ENOMEM;
1908 }
1909
1910 void f2fs_destroy_garbage_collection_cache(void)
1911 {
1912         kmem_cache_destroy(victim_entry_slab);
1913 }
1914
1915 static void init_atgc_management(struct f2fs_sb_info *sbi)
1916 {
1917         struct atgc_management *am = &sbi->am;
1918
1919         if (test_opt(sbi, ATGC) &&
1920                 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1921                 am->atgc_enabled = true;
1922
1923         am->root = RB_ROOT_CACHED;
1924         INIT_LIST_HEAD(&am->victim_list);
1925         am->victim_count = 0;
1926
1927         am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1928         am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1929         am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1930         am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
1931 }
1932
1933 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1934 {
1935         DIRTY_I(sbi)->v_ops = &default_v_ops;
1936
1937         sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1938
1939         /* give warm/cold data area from slower device */
1940         if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1941                 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1942                                 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1943
1944         init_atgc_management(sbi);
1945 }
1946
1947 static int free_segment_range(struct f2fs_sb_info *sbi,
1948                                 unsigned int secs, bool gc_only)
1949 {
1950         unsigned int segno, next_inuse, start, end;
1951         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1952         int gc_mode, gc_type;
1953         int err = 0;
1954         int type;
1955
1956         /* Force block allocation for GC */
1957         MAIN_SECS(sbi) -= secs;
1958         start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1959         end = MAIN_SEGS(sbi) - 1;
1960
1961         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
1962         for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
1963                 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
1964                         SIT_I(sbi)->last_victim[gc_mode] = 0;
1965
1966         for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
1967                 if (sbi->next_victim_seg[gc_type] >= start)
1968                         sbi->next_victim_seg[gc_type] = NULL_SEGNO;
1969         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
1970
1971         /* Move out cursegs from the target range */
1972         for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
1973                 f2fs_allocate_segment_for_resize(sbi, type, start, end);
1974
1975         /* do GC to move out valid blocks in the range */
1976         for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
1977                 struct gc_inode_list gc_list = {
1978                         .ilist = LIST_HEAD_INIT(gc_list.ilist),
1979                         .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1980                 };
1981
1982                 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
1983                 put_gc_inode(&gc_list);
1984
1985                 if (!gc_only && get_valid_blocks(sbi, segno, true)) {
1986                         err = -EAGAIN;
1987                         goto out;
1988                 }
1989                 if (fatal_signal_pending(current)) {
1990                         err = -ERESTARTSYS;
1991                         goto out;
1992                 }
1993         }
1994         if (gc_only)
1995                 goto out;
1996
1997         err = f2fs_write_checkpoint(sbi, &cpc);
1998         if (err)
1999                 goto out;
2000
2001         next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
2002         if (next_inuse <= end) {
2003                 f2fs_err(sbi, "segno %u should be free but still inuse!",
2004                          next_inuse);
2005                 f2fs_bug_on(sbi, 1);
2006         }
2007 out:
2008         MAIN_SECS(sbi) += secs;
2009         return err;
2010 }
2011
2012 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
2013 {
2014         struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
2015         int section_count;
2016         int segment_count;
2017         int segment_count_main;
2018         long long block_count;
2019         int segs = secs * sbi->segs_per_sec;
2020
2021         f2fs_down_write(&sbi->sb_lock);
2022
2023         section_count = le32_to_cpu(raw_sb->section_count);
2024         segment_count = le32_to_cpu(raw_sb->segment_count);
2025         segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
2026         block_count = le64_to_cpu(raw_sb->block_count);
2027
2028         raw_sb->section_count = cpu_to_le32(section_count + secs);
2029         raw_sb->segment_count = cpu_to_le32(segment_count + segs);
2030         raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
2031         raw_sb->block_count = cpu_to_le64(block_count +
2032                                         (long long)segs * sbi->blocks_per_seg);
2033         if (f2fs_is_multi_device(sbi)) {
2034                 int last_dev = sbi->s_ndevs - 1;
2035                 int dev_segs =
2036                         le32_to_cpu(raw_sb->devs[last_dev].total_segments);
2037
2038                 raw_sb->devs[last_dev].total_segments =
2039                                                 cpu_to_le32(dev_segs + segs);
2040         }
2041
2042         f2fs_up_write(&sbi->sb_lock);
2043 }
2044
2045 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
2046 {
2047         int segs = secs * sbi->segs_per_sec;
2048         long long blks = (long long)segs * sbi->blocks_per_seg;
2049         long long user_block_count =
2050                                 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
2051
2052         SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
2053         MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
2054         MAIN_SECS(sbi) += secs;
2055         FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
2056         FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
2057         F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
2058
2059         if (f2fs_is_multi_device(sbi)) {
2060                 int last_dev = sbi->s_ndevs - 1;
2061
2062                 FDEV(last_dev).total_segments =
2063                                 (int)FDEV(last_dev).total_segments + segs;
2064                 FDEV(last_dev).end_blk =
2065                                 (long long)FDEV(last_dev).end_blk + blks;
2066 #ifdef CONFIG_BLK_DEV_ZONED
2067                 FDEV(last_dev).nr_blkz = (int)FDEV(last_dev).nr_blkz +
2068                                         (int)(blks >> sbi->log_blocks_per_blkz);
2069 #endif
2070         }
2071 }
2072
2073 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
2074 {
2075         __u64 old_block_count, shrunk_blocks;
2076         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
2077         unsigned int secs;
2078         int err = 0;
2079         __u32 rem;
2080
2081         old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
2082         if (block_count > old_block_count)
2083                 return -EINVAL;
2084
2085         if (f2fs_is_multi_device(sbi)) {
2086                 int last_dev = sbi->s_ndevs - 1;
2087                 __u64 last_segs = FDEV(last_dev).total_segments;
2088
2089                 if (block_count + last_segs * sbi->blocks_per_seg <=
2090                                                                 old_block_count)
2091                         return -EINVAL;
2092         }
2093
2094         /* new fs size should align to section size */
2095         div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
2096         if (rem)
2097                 return -EINVAL;
2098
2099         if (block_count == old_block_count)
2100                 return 0;
2101
2102         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2103                 f2fs_err(sbi, "Should run fsck to repair first.");
2104                 return -EFSCORRUPTED;
2105         }
2106
2107         if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2108                 f2fs_err(sbi, "Checkpoint should be enabled.");
2109                 return -EINVAL;
2110         }
2111
2112         shrunk_blocks = old_block_count - block_count;
2113         secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2114
2115         /* stop other GC */
2116         if (!f2fs_down_write_trylock(&sbi->gc_lock))
2117                 return -EAGAIN;
2118
2119         /* stop CP to protect MAIN_SEC in free_segment_range */
2120         f2fs_lock_op(sbi);
2121
2122         spin_lock(&sbi->stat_lock);
2123         if (shrunk_blocks + valid_user_blocks(sbi) +
2124                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2125                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2126                 err = -ENOSPC;
2127         spin_unlock(&sbi->stat_lock);
2128
2129         if (err)
2130                 goto out_unlock;
2131
2132         err = free_segment_range(sbi, secs, true);
2133
2134 out_unlock:
2135         f2fs_unlock_op(sbi);
2136         f2fs_up_write(&sbi->gc_lock);
2137         if (err)
2138                 return err;
2139
2140         freeze_super(sbi->sb);
2141         f2fs_down_write(&sbi->gc_lock);
2142         f2fs_down_write(&sbi->cp_global_sem);
2143
2144         spin_lock(&sbi->stat_lock);
2145         if (shrunk_blocks + valid_user_blocks(sbi) +
2146                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2147                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2148                 err = -ENOSPC;
2149         else
2150                 sbi->user_block_count -= shrunk_blocks;
2151         spin_unlock(&sbi->stat_lock);
2152         if (err)
2153                 goto out_err;
2154
2155         set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2156         err = free_segment_range(sbi, secs, false);
2157         if (err)
2158                 goto recover_out;
2159
2160         update_sb_metadata(sbi, -secs);
2161
2162         err = f2fs_commit_super(sbi, false);
2163         if (err) {
2164                 update_sb_metadata(sbi, secs);
2165                 goto recover_out;
2166         }
2167
2168         update_fs_metadata(sbi, -secs);
2169         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2170         set_sbi_flag(sbi, SBI_IS_DIRTY);
2171
2172         err = f2fs_write_checkpoint(sbi, &cpc);
2173         if (err) {
2174                 update_fs_metadata(sbi, secs);
2175                 update_sb_metadata(sbi, secs);
2176                 f2fs_commit_super(sbi, false);
2177         }
2178 recover_out:
2179         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2180         if (err) {
2181                 set_sbi_flag(sbi, SBI_NEED_FSCK);
2182                 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2183
2184                 spin_lock(&sbi->stat_lock);
2185                 sbi->user_block_count += shrunk_blocks;
2186                 spin_unlock(&sbi->stat_lock);
2187         }
2188 out_err:
2189         f2fs_up_write(&sbi->cp_global_sem);
2190         f2fs_up_write(&sbi->gc_lock);
2191         thaw_super(sbi->sb);
2192         return err;
2193 }