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_freezable_timeout(*wq,
  50                                 kthread_should_stop() ||
  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 (f2fs_readonly(sbi->sb)) {
  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                 stat_inc_gc_call_count(sbi, foreground ?
 125                                         FOREGROUND : BACKGROUND);
 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 bool f2fs_check_victim_tree(struct f2fs_sb_info *sbi,
 394                                 struct rb_root_cached *root)
 395 {
 396 #ifdef CONFIG_F2FS_CHECK_FS
 397         struct rb_node *cur = rb_first_cached(root), *next;
 398         struct victim_entry *cur_ve, *next_ve;
 399
 400         while (cur) {
 401                 next = rb_next(cur);
 402                 if (!next)
 403                         return true;
 404
 405                 cur_ve = rb_entry(cur, struct victim_entry, rb_node);
 406                 next_ve = rb_entry(next, struct victim_entry, rb_node);
 407
 408                 if (cur_ve->mtime > next_ve->mtime) {
 409                         f2fs_info(sbi, "broken victim_rbtree, "
 410                                 "cur_mtime(%llu) next_mtime(%llu)",
 411                                 cur_ve->mtime, next_ve->mtime);
 412                         return false;
 413                 }
 414                 cur = next;
 415         }
 416 #endif
 417         return true;
 418 }
 419
 420 static struct victim_entry *__lookup_victim_entry(struct f2fs_sb_info *sbi,
 421                                         unsigned long long mtime)
 422 {
 423         struct atgc_management *am = &sbi->am;
 424         struct rb_node *node = am->root.rb_root.rb_node;
 425         struct victim_entry *ve = NULL;
 426
 427         while (node) {
 428                 ve = rb_entry(node, struct victim_entry, rb_node);
 429
 430                 if (mtime < ve->mtime)
 431                         node = node->rb_left;
 432                 else
 433                         node = node->rb_right;
 434         }
 435         return ve;
 436 }
 437
 438 static struct victim_entry *__create_victim_entry(struct f2fs_sb_info *sbi,
 439                 unsigned long long mtime, unsigned int segno)
 440 {
 441         struct atgc_management *am = &sbi->am;
 442         struct victim_entry *ve;
 443
 444         ve =  f2fs_kmem_cache_alloc(victim_entry_slab, GFP_NOFS, true, NULL);
 445
 446         ve->mtime = mtime;
 447         ve->segno = segno;
 448
 449         list_add_tail(&ve->list, &am->victim_list);
 450         am->victim_count++;
 451
 452         return ve;
 453 }
 454
 455 static void __insert_victim_entry(struct f2fs_sb_info *sbi,
 456                                 unsigned long long mtime, unsigned int segno)
 457 {
 458         struct atgc_management *am = &sbi->am;
 459         struct rb_root_cached *root = &am->root;
 460         struct rb_node **p = &root->rb_root.rb_node;
 461         struct rb_node *parent = NULL;
 462         struct victim_entry *ve;
 463         bool left_most = true;
 464
 465         /* look up rb tree to find parent node */
 466         while (*p) {
 467                 parent = *p;
 468                 ve = rb_entry(parent, struct victim_entry, rb_node);
 469
 470                 if (mtime < ve->mtime) {
 471                         p = &(*p)->rb_left;
 472                 } else {
 473                         p = &(*p)->rb_right;
 474                         left_most = false;
 475                 }
 476         }
 477
 478         ve = __create_victim_entry(sbi, mtime, segno);
 479
 480         rb_link_node(&ve->rb_node, parent, p);
 481         rb_insert_color_cached(&ve->rb_node, root, left_most);
 482 }
 483
 484 static void add_victim_entry(struct f2fs_sb_info *sbi,
 485                                 struct victim_sel_policy *p, unsigned int segno)
 486 {
 487         struct sit_info *sit_i = SIT_I(sbi);
 488         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 489         unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 490         unsigned long long mtime = 0;
 491         unsigned int i;
 492
 493         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 494                 if (p->gc_mode == GC_AT &&
 495                         get_valid_blocks(sbi, segno, true) == 0)
 496                         return;
 497         }
 498
 499         for (i = 0; i < sbi->segs_per_sec; i++)
 500                 mtime += get_seg_entry(sbi, start + i)->mtime;
 501         mtime = div_u64(mtime, sbi->segs_per_sec);
 502
 503         /* Handle if the system time has changed by the user */
 504         if (mtime < sit_i->min_mtime)
 505                 sit_i->min_mtime = mtime;
 506         if (mtime > sit_i->max_mtime)
 507                 sit_i->max_mtime = mtime;
 508         if (mtime < sit_i->dirty_min_mtime)
 509                 sit_i->dirty_min_mtime = mtime;
 510         if (mtime > sit_i->dirty_max_mtime)
 511                 sit_i->dirty_max_mtime = mtime;
 512
 513         /* don't choose young section as candidate */
 514         if (sit_i->dirty_max_mtime - mtime < p->age_threshold)
 515                 return;
 516
 517         __insert_victim_entry(sbi, mtime, segno);
 518 }
 519
 520 static void atgc_lookup_victim(struct f2fs_sb_info *sbi,
 521                                                 struct victim_sel_policy *p)
 522 {
 523         struct sit_info *sit_i = SIT_I(sbi);
 524         struct atgc_management *am = &sbi->am;
 525         struct rb_root_cached *root = &am->root;
 526         struct rb_node *node;
 527         struct victim_entry *ve;
 528         unsigned long long total_time;
 529         unsigned long long age, u, accu;
 530         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 531         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 532         unsigned int sec_blocks = CAP_BLKS_PER_SEC(sbi);
 533         unsigned int vblocks;
 534         unsigned int dirty_threshold = max(am->max_candidate_count,
 535                                         am->candidate_ratio *
 536                                         am->victim_count / 100);
 537         unsigned int age_weight = am->age_weight;
 538         unsigned int cost;
 539         unsigned int iter = 0;
 540
 541         if (max_mtime < min_mtime)
 542                 return;
 543
 544         max_mtime += 1;
 545         total_time = max_mtime - min_mtime;
 546
 547         accu = div64_u64(ULLONG_MAX, total_time);
 548         accu = min_t(unsigned long long, div_u64(accu, 100),
 549                                         DEFAULT_ACCURACY_CLASS);
 550
 551         node = rb_first_cached(root);
 552 next:
 553         ve = rb_entry_safe(node, struct victim_entry, rb_node);
 554         if (!ve)
 555                 return;
 556
 557         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 558                 goto skip;
 559
 560         /* age = 10000 * x% * 60 */
 561         age = div64_u64(accu * (max_mtime - ve->mtime), total_time) *
 562                                                                 age_weight;
 563
 564         vblocks = get_valid_blocks(sbi, ve->segno, true);
 565         f2fs_bug_on(sbi, !vblocks || vblocks == sec_blocks);
 566
 567         /* u = 10000 * x% * 40 */
 568         u = div64_u64(accu * (sec_blocks - vblocks), sec_blocks) *
 569                                                         (100 - age_weight);
 570
 571         f2fs_bug_on(sbi, age + u >= UINT_MAX);
 572
 573         cost = UINT_MAX - (age + u);
 574         iter++;
 575
 576         if (cost < p->min_cost ||
 577                         (cost == p->min_cost && age > p->oldest_age)) {
 578                 p->min_cost = cost;
 579                 p->oldest_age = age;
 580                 p->min_segno = ve->segno;
 581         }
 582 skip:
 583         if (iter < dirty_threshold) {
 584                 node = rb_next(node);
 585                 goto next;
 586         }
 587 }
 588
 589 /*
 590  * select candidates around source section in range of
 591  * [target - dirty_threshold, target + dirty_threshold]
 592  */
 593 static void atssr_lookup_victim(struct f2fs_sb_info *sbi,
 594                                                 struct victim_sel_policy *p)
 595 {
 596         struct sit_info *sit_i = SIT_I(sbi);
 597         struct atgc_management *am = &sbi->am;
 598         struct victim_entry *ve;
 599         unsigned long long age;
 600         unsigned long long max_mtime = sit_i->dirty_max_mtime;
 601         unsigned long long min_mtime = sit_i->dirty_min_mtime;
 602         unsigned int seg_blocks = sbi->blocks_per_seg;
 603         unsigned int vblocks;
 604         unsigned int dirty_threshold = max(am->max_candidate_count,
 605                                         am->candidate_ratio *
 606                                         am->victim_count / 100);
 607         unsigned int cost, iter;
 608         int stage = 0;
 609
 610         if (max_mtime < min_mtime)
 611                 return;
 612         max_mtime += 1;
 613 next_stage:
 614         iter = 0;
 615         ve = __lookup_victim_entry(sbi, p->age);
 616 next_node:
 617         if (!ve) {
 618                 if (stage++ == 0)
 619                         goto next_stage;
 620                 return;
 621         }
 622
 623         if (ve->mtime >= max_mtime || ve->mtime < min_mtime)
 624                 goto skip_node;
 625
 626         age = max_mtime - ve->mtime;
 627
 628         vblocks = get_seg_entry(sbi, ve->segno)->ckpt_valid_blocks;
 629         f2fs_bug_on(sbi, !vblocks);
 630
 631         /* rare case */
 632         if (vblocks == seg_blocks)
 633                 goto skip_node;
 634
 635         iter++;
 636
 637         age = max_mtime - abs(p->age - age);
 638         cost = UINT_MAX - vblocks;
 639
 640         if (cost < p->min_cost ||
 641                         (cost == p->min_cost && age > p->oldest_age)) {
 642                 p->min_cost = cost;
 643                 p->oldest_age = age;
 644                 p->min_segno = ve->segno;
 645         }
 646 skip_node:
 647         if (iter < dirty_threshold) {
 648                 ve = rb_entry(stage == 0 ? rb_prev(&ve->rb_node) :
 649                                         rb_next(&ve->rb_node),
 650                                         struct victim_entry, rb_node);
 651                 goto next_node;
 652         }
 653
 654         if (stage++ == 0)
 655                 goto next_stage;
 656 }
 657
 658 static void lookup_victim_by_age(struct f2fs_sb_info *sbi,
 659                                                 struct victim_sel_policy *p)
 660 {
 661         f2fs_bug_on(sbi, !f2fs_check_victim_tree(sbi, &sbi->am.root));
 662
 663         if (p->gc_mode == GC_AT)
 664                 atgc_lookup_victim(sbi, p);
 665         else if (p->alloc_mode == AT_SSR)
 666                 atssr_lookup_victim(sbi, p);
 667         else
 668                 f2fs_bug_on(sbi, 1);
 669 }
 670
 671 static void release_victim_entry(struct f2fs_sb_info *sbi)
 672 {
 673         struct atgc_management *am = &sbi->am;
 674         struct victim_entry *ve, *tmp;
 675
 676         list_for_each_entry_safe(ve, tmp, &am->victim_list, list) {
 677                 list_del(&ve->list);
 678                 kmem_cache_free(victim_entry_slab, ve);
 679                 am->victim_count--;
 680         }
 681
 682         am->root = RB_ROOT_CACHED;
 683
 684         f2fs_bug_on(sbi, am->victim_count);
 685         f2fs_bug_on(sbi, !list_empty(&am->victim_list));
 686 }
 687
 688 static bool f2fs_pin_section(struct f2fs_sb_info *sbi, unsigned int segno)
 689 {
 690         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 691         unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 692
 693         if (!dirty_i->enable_pin_section)
 694                 return false;
 695         if (!test_and_set_bit(secno, dirty_i->pinned_secmap))
 696                 dirty_i->pinned_secmap_cnt++;
 697         return true;
 698 }
 699
 700 static bool f2fs_pinned_section_exists(struct dirty_seglist_info *dirty_i)
 701 {
 702         return dirty_i->pinned_secmap_cnt;
 703 }
 704
 705 static bool f2fs_section_is_pinned(struct dirty_seglist_info *dirty_i,
 706                                                 unsigned int secno)
 707 {
 708         return dirty_i->enable_pin_section &&
 709                 f2fs_pinned_section_exists(dirty_i) &&
 710                 test_bit(secno, dirty_i->pinned_secmap);
 711 }
 712
 713 static void f2fs_unpin_all_sections(struct f2fs_sb_info *sbi, bool enable)
 714 {
 715         unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
 716
 717         if (f2fs_pinned_section_exists(DIRTY_I(sbi))) {
 718                 memset(DIRTY_I(sbi)->pinned_secmap, 0, bitmap_size);
 719                 DIRTY_I(sbi)->pinned_secmap_cnt = 0;
 720         }
 721         DIRTY_I(sbi)->enable_pin_section = enable;
 722 }
 723
 724 static int f2fs_gc_pinned_control(struct inode *inode, int gc_type,
 725                                                         unsigned int segno)
 726 {
 727         if (!f2fs_is_pinned_file(inode))
 728                 return 0;
 729         if (gc_type != FG_GC)
 730                 return -EBUSY;
 731         if (!f2fs_pin_section(F2FS_I_SB(inode), segno))
 732                 f2fs_pin_file_control(inode, true);
 733         return -EAGAIN;
 734 }
 735
 736 /*
 737  * This function is called from two paths.
 738  * One is garbage collection and the other is SSR segment selection.
 739  * When it is called during GC, it just gets a victim segment
 740  * and it does not remove it from dirty seglist.
 741  * When it is called from SSR segment selection, it finds a segment
 742  * which has minimum valid blocks and removes it from dirty seglist.
 743  */
 744 int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
 745                         int gc_type, int type, char alloc_mode,
 746                         unsigned long long age)
 747 {
 748         struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
 749         struct sit_info *sm = SIT_I(sbi);
 750         struct victim_sel_policy p;
 751         unsigned int secno, last_victim;
 752         unsigned int last_segment;
 753         unsigned int nsearched;
 754         bool is_atgc;
 755         int ret = 0;
 756
 757         mutex_lock(&dirty_i->seglist_lock);
 758         last_segment = MAIN_SECS(sbi) * sbi->segs_per_sec;
 759
 760         p.alloc_mode = alloc_mode;
 761         p.age = age;
 762         p.age_threshold = sbi->am.age_threshold;
 763
 764 retry:
 765         select_policy(sbi, gc_type, type, &p);
 766         p.min_segno = NULL_SEGNO;
 767         p.oldest_age = 0;
 768         p.min_cost = get_max_cost(sbi, &p);
 769
 770         is_atgc = (p.gc_mode == GC_AT || p.alloc_mode == AT_SSR);
 771         nsearched = 0;
 772
 773         if (is_atgc)
 774                 SIT_I(sbi)->dirty_min_mtime = ULLONG_MAX;
 775
 776         if (*result != NULL_SEGNO) {
 777                 if (!get_valid_blocks(sbi, *result, false)) {
 778                         ret = -ENODATA;
 779                         goto out;
 780                 }
 781
 782                 if (sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
 783                         ret = -EBUSY;
 784                 else
 785                         p.min_segno = *result;
 786                 goto out;
 787         }
 788
 789         ret = -ENODATA;
 790         if (p.max_search == 0)
 791                 goto out;
 792
 793         if (__is_large_section(sbi) && p.alloc_mode == LFS) {
 794                 if (sbi->next_victim_seg[BG_GC] != NULL_SEGNO) {
 795                         p.min_segno = sbi->next_victim_seg[BG_GC];
 796                         *result = p.min_segno;
 797                         sbi->next_victim_seg[BG_GC] = NULL_SEGNO;
 798                         goto got_result;
 799                 }
 800                 if (gc_type == FG_GC &&
 801                                 sbi->next_victim_seg[FG_GC] != NULL_SEGNO) {
 802                         p.min_segno = sbi->next_victim_seg[FG_GC];
 803                         *result = p.min_segno;
 804                         sbi->next_victim_seg[FG_GC] = NULL_SEGNO;
 805                         goto got_result;
 806                 }
 807         }
 808
 809         last_victim = sm->last_victim[p.gc_mode];
 810         if (p.alloc_mode == LFS && gc_type == FG_GC) {
 811                 p.min_segno = check_bg_victims(sbi);
 812                 if (p.min_segno != NULL_SEGNO)
 813                         goto got_it;
 814         }
 815
 816         while (1) {
 817                 unsigned long cost, *dirty_bitmap;
 818                 unsigned int unit_no, segno;
 819
 820                 dirty_bitmap = p.dirty_bitmap;
 821                 unit_no = find_next_bit(dirty_bitmap,
 822                                 last_segment / p.ofs_unit,
 823                                 p.offset / p.ofs_unit);
 824                 segno = unit_no * p.ofs_unit;
 825                 if (segno >= last_segment) {
 826                         if (sm->last_victim[p.gc_mode]) {
 827                                 last_segment =
 828                                         sm->last_victim[p.gc_mode];
 829                                 sm->last_victim[p.gc_mode] = 0;
 830                                 p.offset = 0;
 831                                 continue;
 832                         }
 833                         break;
 834                 }
 835
 836                 p.offset = segno + p.ofs_unit;
 837                 nsearched++;
 838
 839 #ifdef CONFIG_F2FS_CHECK_FS
 840                 /*
 841                  * skip selecting the invalid segno (that is failed due to block
 842                  * validity check failure during GC) to avoid endless GC loop in
 843                  * such cases.
 844                  */
 845                 if (test_bit(segno, sm->invalid_segmap))
 846                         goto next;
 847 #endif
 848
 849                 secno = GET_SEC_FROM_SEG(sbi, segno);
 850
 851                 if (sec_usage_check(sbi, secno))
 852                         goto next;
 853
 854                 /* Don't touch checkpointed data */
 855                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
 856                         if (p.alloc_mode == LFS) {
 857                                 /*
 858                                  * LFS is set to find source section during GC.
 859                                  * The victim should have no checkpointed data.
 860                                  */
 861                                 if (get_ckpt_valid_blocks(sbi, segno, true))
 862                                         goto next;
 863                         } else {
 864                                 /*
 865                                  * SSR | AT_SSR are set to find target segment
 866                                  * for writes which can be full by checkpointed
 867                                  * and newly written blocks.
 868                                  */
 869                                 if (!f2fs_segment_has_free_slot(sbi, segno))
 870                                         goto next;
 871                         }
 872                 }
 873
 874                 if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
 875                         goto next;
 876
 877                 if (gc_type == FG_GC && f2fs_section_is_pinned(dirty_i, secno))
 878                         goto next;
 879
 880                 if (is_atgc) {
 881                         add_victim_entry(sbi, &p, segno);
 882                         goto next;
 883                 }
 884
 885                 cost = get_gc_cost(sbi, segno, &p);
 886
 887                 if (p.min_cost > cost) {
 888                         p.min_segno = segno;
 889                         p.min_cost = cost;
 890                 }
 891 next:
 892                 if (nsearched >= p.max_search) {
 893                         if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
 894                                 sm->last_victim[p.gc_mode] =
 895                                         last_victim + p.ofs_unit;
 896                         else
 897                                 sm->last_victim[p.gc_mode] = segno + p.ofs_unit;
 898                         sm->last_victim[p.gc_mode] %=
 899                                 (MAIN_SECS(sbi) * sbi->segs_per_sec);
 900                         break;
 901                 }
 902         }
 903
 904         /* get victim for GC_AT/AT_SSR */
 905         if (is_atgc) {
 906                 lookup_victim_by_age(sbi, &p);
 907                 release_victim_entry(sbi);
 908         }
 909
 910         if (is_atgc && p.min_segno == NULL_SEGNO &&
 911                         sm->elapsed_time < p.age_threshold) {
 912                 p.age_threshold = 0;
 913                 goto retry;
 914         }
 915
 916         if (p.min_segno != NULL_SEGNO) {
 917 got_it:
 918                 *result = (p.min_segno / p.ofs_unit) * p.ofs_unit;
 919 got_result:
 920                 if (p.alloc_mode == LFS) {
 921                         secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
 922                         if (gc_type == FG_GC)
 923                                 sbi->cur_victim_sec = secno;
 924                         else
 925                                 set_bit(secno, dirty_i->victim_secmap);
 926                 }
 927                 ret = 0;
 928
 929         }
 930 out:
 931         if (p.min_segno != NULL_SEGNO)
 932                 trace_f2fs_get_victim(sbi->sb, type, gc_type, &p,
 933                                 sbi->cur_victim_sec,
 934                                 prefree_segments(sbi), free_segments(sbi));
 935         mutex_unlock(&dirty_i->seglist_lock);
 936
 937         return ret;
 938 }
 939
 940 static struct inode *find_gc_inode(struct gc_inode_list *gc_list, nid_t ino)
 941 {
 942         struct inode_entry *ie;
 943
 944         ie = radix_tree_lookup(&gc_list->iroot, ino);
 945         if (ie)
 946                 return ie->inode;
 947         return NULL;
 948 }
 949
 950 static void add_gc_inode(struct gc_inode_list *gc_list, struct inode *inode)
 951 {
 952         struct inode_entry *new_ie;
 953
 954         if (inode == find_gc_inode(gc_list, inode->i_ino)) {
 955                 iput(inode);
 956                 return;
 957         }
 958         new_ie = f2fs_kmem_cache_alloc(f2fs_inode_entry_slab,
 959                                         GFP_NOFS, true, NULL);
 960         new_ie->inode = inode;
 961
 962         f2fs_radix_tree_insert(&gc_list->iroot, inode->i_ino, new_ie);
 963         list_add_tail(&new_ie->list, &gc_list->ilist);
 964 }
 965
 966 static void put_gc_inode(struct gc_inode_list *gc_list)
 967 {
 968         struct inode_entry *ie, *next_ie;
 969
 970         list_for_each_entry_safe(ie, next_ie, &gc_list->ilist, list) {
 971                 radix_tree_delete(&gc_list->iroot, ie->inode->i_ino);
 972                 iput(ie->inode);
 973                 list_del(&ie->list);
 974                 kmem_cache_free(f2fs_inode_entry_slab, ie);
 975         }
 976 }
 977
 978 static int check_valid_map(struct f2fs_sb_info *sbi,
 979                                 unsigned int segno, int offset)
 980 {
 981         struct sit_info *sit_i = SIT_I(sbi);
 982         struct seg_entry *sentry;
 983         int ret;
 984
 985         down_read(&sit_i->sentry_lock);
 986         sentry = get_seg_entry(sbi, segno);
 987         ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 988         up_read(&sit_i->sentry_lock);
 989         return ret;
 990 }
 991
 992 /*
 993  * This function compares node address got in summary with that in NAT.
 994  * On validity, copy that node with cold status, otherwise (invalid node)
 995  * ignore that.
 996  */
 997 static int gc_node_segment(struct f2fs_sb_info *sbi,
 998                 struct f2fs_summary *sum, unsigned int segno, int gc_type)
 999 {
1000         struct f2fs_summary *entry;
1001         block_t start_addr;
1002         int off;
1003         int phase = 0;
1004         bool fggc = (gc_type == FG_GC);
1005         int submitted = 0;
1006         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1007
1008         start_addr = START_BLOCK(sbi, segno);
1009
1010 next_step:
1011         entry = sum;
1012
1013         if (fggc && phase == 2)
1014                 atomic_inc(&sbi->wb_sync_req[NODE]);
1015
1016         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1017                 nid_t nid = le32_to_cpu(entry->nid);
1018                 struct page *node_page;
1019                 struct node_info ni;
1020                 int err;
1021
1022                 /* stop BG_GC if there is not enough free sections. */
1023                 if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
1024                         return submitted;
1025
1026                 if (check_valid_map(sbi, segno, off) == 0)
1027                         continue;
1028
1029                 if (phase == 0) {
1030                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1031                                                         META_NAT, true);
1032                         continue;
1033                 }
1034
1035                 if (phase == 1) {
1036                         f2fs_ra_node_page(sbi, nid);
1037                         continue;
1038                 }
1039
1040                 /* phase == 2 */
1041                 node_page = f2fs_get_node_page(sbi, nid);
1042                 if (IS_ERR(node_page))
1043                         continue;
1044
1045                 /* block may become invalid during f2fs_get_node_page */
1046                 if (check_valid_map(sbi, segno, off) == 0) {
1047                         f2fs_put_page(node_page, 1);
1048                         continue;
1049                 }
1050
1051                 if (f2fs_get_node_info(sbi, nid, &ni, false)) {
1052                         f2fs_put_page(node_page, 1);
1053                         continue;
1054                 }
1055
1056                 if (ni.blk_addr != start_addr + off) {
1057                         f2fs_put_page(node_page, 1);
1058                         continue;
1059                 }
1060
1061                 err = f2fs_move_node_page(node_page, gc_type);
1062                 if (!err && gc_type == FG_GC)
1063                         submitted++;
1064                 stat_inc_node_blk_count(sbi, 1, gc_type);
1065         }
1066
1067         if (++phase < 3)
1068                 goto next_step;
1069
1070         if (fggc)
1071                 atomic_dec(&sbi->wb_sync_req[NODE]);
1072         return submitted;
1073 }
1074
1075 /*
1076  * Calculate start block index indicating the given node offset.
1077  * Be careful, caller should give this node offset only indicating direct node
1078  * blocks. If any node offsets, which point the other types of node blocks such
1079  * as indirect or double indirect node blocks, are given, it must be a caller's
1080  * bug.
1081  */
1082 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
1083 {
1084         unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
1085         unsigned int bidx;
1086
1087         if (node_ofs == 0)
1088                 return 0;
1089
1090         if (node_ofs <= 2) {
1091                 bidx = node_ofs - 1;
1092         } else if (node_ofs <= indirect_blks) {
1093                 int dec = (node_ofs - 4) / (NIDS_PER_BLOCK + 1);
1094
1095                 bidx = node_ofs - 2 - dec;
1096         } else {
1097                 int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
1098
1099                 bidx = node_ofs - 5 - dec;
1100         }
1101         return bidx * ADDRS_PER_BLOCK(inode) + ADDRS_PER_INODE(inode);
1102 }
1103
1104 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1105                 struct node_info *dni, block_t blkaddr, unsigned int *nofs)
1106 {
1107         struct page *node_page;
1108         nid_t nid;
1109         unsigned int ofs_in_node, max_addrs, base;
1110         block_t source_blkaddr;
1111
1112         nid = le32_to_cpu(sum->nid);
1113         ofs_in_node = le16_to_cpu(sum->ofs_in_node);
1114
1115         node_page = f2fs_get_node_page(sbi, nid);
1116         if (IS_ERR(node_page))
1117                 return false;
1118
1119         if (f2fs_get_node_info(sbi, nid, dni, false)) {
1120                 f2fs_put_page(node_page, 1);
1121                 return false;
1122         }
1123
1124         if (sum->version != dni->version) {
1125                 f2fs_warn(sbi, "%s: valid data with mismatched node version.",
1126                           __func__);
1127                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1128         }
1129
1130         if (f2fs_check_nid_range(sbi, dni->ino)) {
1131                 f2fs_put_page(node_page, 1);
1132                 return false;
1133         }
1134
1135         if (IS_INODE(node_page)) {
1136                 base = offset_in_addr(F2FS_INODE(node_page));
1137                 max_addrs = DEF_ADDRS_PER_INODE;
1138         } else {
1139                 base = 0;
1140                 max_addrs = DEF_ADDRS_PER_BLOCK;
1141         }
1142
1143         if (base + ofs_in_node >= max_addrs) {
1144                 f2fs_err(sbi, "Inconsistent blkaddr offset: base:%u, ofs_in_node:%u, max:%u, ino:%u, nid:%u",
1145                         base, ofs_in_node, max_addrs, dni->ino, dni->nid);
1146                 f2fs_put_page(node_page, 1);
1147                 return false;
1148         }
1149
1150         *nofs = ofs_of_node(node_page);
1151         source_blkaddr = data_blkaddr(NULL, node_page, ofs_in_node);
1152         f2fs_put_page(node_page, 1);
1153
1154         if (source_blkaddr != blkaddr) {
1155 #ifdef CONFIG_F2FS_CHECK_FS
1156                 unsigned int segno = GET_SEGNO(sbi, blkaddr);
1157                 unsigned long offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1158
1159                 if (unlikely(check_valid_map(sbi, segno, offset))) {
1160                         if (!test_and_set_bit(segno, SIT_I(sbi)->invalid_segmap)) {
1161                                 f2fs_err(sbi, "mismatched blkaddr %u (source_blkaddr %u) in seg %u",
1162                                          blkaddr, source_blkaddr, segno);
1163                                 set_sbi_flag(sbi, SBI_NEED_FSCK);
1164                         }
1165                 }
1166 #endif
1167                 return false;
1168         }
1169         return true;
1170 }
1171
1172 static int ra_data_block(struct inode *inode, pgoff_t index)
1173 {
1174         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1175         struct address_space *mapping = inode->i_mapping;
1176         struct dnode_of_data dn;
1177         struct page *page;
1178         struct f2fs_io_info fio = {
1179                 .sbi = sbi,
1180                 .ino = inode->i_ino,
1181                 .type = DATA,
1182                 .temp = COLD,
1183                 .op = REQ_OP_READ,
1184                 .op_flags = 0,
1185                 .encrypted_page = NULL,
1186                 .in_list = 0,
1187                 .retry = 0,
1188         };
1189         int err;
1190
1191         page = f2fs_grab_cache_page(mapping, index, true);
1192         if (!page)
1193                 return -ENOMEM;
1194
1195         if (f2fs_lookup_read_extent_cache_block(inode, index,
1196                                                 &dn.data_blkaddr)) {
1197                 if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1198                                                 DATA_GENERIC_ENHANCE_READ))) {
1199                         err = -EFSCORRUPTED;
1200                         f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1201                         goto put_page;
1202                 }
1203                 goto got_it;
1204         }
1205
1206         set_new_dnode(&dn, inode, NULL, NULL, 0);
1207         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1208         if (err)
1209                 goto put_page;
1210         f2fs_put_dnode(&dn);
1211
1212         if (!__is_valid_data_blkaddr(dn.data_blkaddr)) {
1213                 err = -ENOENT;
1214                 goto put_page;
1215         }
1216         if (unlikely(!f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr,
1217                                                 DATA_GENERIC_ENHANCE))) {
1218                 err = -EFSCORRUPTED;
1219                 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1220                 goto put_page;
1221         }
1222 got_it:
1223         /* read page */
1224         fio.page = page;
1225         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1226
1227         /*
1228          * don't cache encrypted data into meta inode until previous dirty
1229          * data were writebacked to avoid racing between GC and flush.
1230          */
1231         f2fs_wait_on_page_writeback(page, DATA, true, true);
1232
1233         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1234
1235         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(sbi),
1236                                         dn.data_blkaddr,
1237                                         FGP_LOCK | FGP_CREAT, GFP_NOFS);
1238         if (!fio.encrypted_page) {
1239                 err = -ENOMEM;
1240                 goto put_page;
1241         }
1242
1243         err = f2fs_submit_page_bio(&fio);
1244         if (err)
1245                 goto put_encrypted_page;
1246         f2fs_put_page(fio.encrypted_page, 0);
1247         f2fs_put_page(page, 1);
1248
1249         f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
1250         f2fs_update_iostat(sbi, NULL, FS_GDATA_READ_IO, F2FS_BLKSIZE);
1251
1252         return 0;
1253 put_encrypted_page:
1254         f2fs_put_page(fio.encrypted_page, 1);
1255 put_page:
1256         f2fs_put_page(page, 1);
1257         return err;
1258 }
1259
1260 /*
1261  * Move data block via META_MAPPING while keeping locked data page.
1262  * This can be used to move blocks, aka LBAs, directly on disk.
1263  */
1264 static int move_data_block(struct inode *inode, block_t bidx,
1265                                 int gc_type, unsigned int segno, int off)
1266 {
1267         struct f2fs_io_info fio = {
1268                 .sbi = F2FS_I_SB(inode),
1269                 .ino = inode->i_ino,
1270                 .type = DATA,
1271                 .temp = COLD,
1272                 .op = REQ_OP_READ,
1273                 .op_flags = 0,
1274                 .encrypted_page = NULL,
1275                 .in_list = 0,
1276                 .retry = 0,
1277         };
1278         struct dnode_of_data dn;
1279         struct f2fs_summary sum;
1280         struct node_info ni;
1281         struct page *page, *mpage;
1282         block_t newaddr;
1283         int err = 0;
1284         bool lfs_mode = f2fs_lfs_mode(fio.sbi);
1285         int type = fio.sbi->am.atgc_enabled && (gc_type == BG_GC) &&
1286                                 (fio.sbi->gc_mode != GC_URGENT_HIGH) ?
1287                                 CURSEG_ALL_DATA_ATGC : CURSEG_COLD_DATA;
1288
1289         /* do not read out */
1290         page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
1291         if (!page)
1292                 return -ENOMEM;
1293
1294         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1295                 err = -ENOENT;
1296                 goto out;
1297         }
1298
1299         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1300         if (err)
1301                 goto out;
1302
1303         set_new_dnode(&dn, inode, NULL, NULL, 0);
1304         err = f2fs_get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
1305         if (err)
1306                 goto out;
1307
1308         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1309                 ClearPageUptodate(page);
1310                 err = -ENOENT;
1311                 goto put_out;
1312         }
1313
1314         /*
1315          * don't cache encrypted data into meta inode until previous dirty
1316          * data were writebacked to avoid racing between GC and flush.
1317          */
1318         f2fs_wait_on_page_writeback(page, DATA, true, true);
1319
1320         f2fs_wait_on_block_writeback(inode, dn.data_blkaddr);
1321
1322         err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);
1323         if (err)
1324                 goto put_out;
1325
1326         /* read page */
1327         fio.page = page;
1328         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
1329
1330         if (lfs_mode)
1331                 f2fs_down_write(&fio.sbi->io_order_lock);
1332
1333         mpage = f2fs_grab_cache_page(META_MAPPING(fio.sbi),
1334                                         fio.old_blkaddr, false);
1335         if (!mpage) {
1336                 err = -ENOMEM;
1337                 goto up_out;
1338         }
1339
1340         fio.encrypted_page = mpage;
1341
1342         /* read source block in mpage */
1343         if (!PageUptodate(mpage)) {
1344                 err = f2fs_submit_page_bio(&fio);
1345                 if (err) {
1346                         f2fs_put_page(mpage, 1);
1347                         goto up_out;
1348                 }
1349
1350                 f2fs_update_iostat(fio.sbi, inode, FS_DATA_READ_IO,
1351                                                         F2FS_BLKSIZE);
1352                 f2fs_update_iostat(fio.sbi, NULL, FS_GDATA_READ_IO,
1353                                                         F2FS_BLKSIZE);
1354
1355                 lock_page(mpage);
1356                 if (unlikely(mpage->mapping != META_MAPPING(fio.sbi) ||
1357                                                 !PageUptodate(mpage))) {
1358                         err = -EIO;
1359                         f2fs_put_page(mpage, 1);
1360                         goto up_out;
1361                 }
1362         }
1363
1364         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
1365
1366         /* allocate block address */
1367         f2fs_allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
1368                                 &sum, type, NULL);
1369
1370         fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
1371                                 newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
1372         if (!fio.encrypted_page) {
1373                 err = -ENOMEM;
1374                 f2fs_put_page(mpage, 1);
1375                 goto recover_block;
1376         }
1377
1378         /* write target block */
1379         f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true, true);
1380         memcpy(page_address(fio.encrypted_page),
1381                                 page_address(mpage), PAGE_SIZE);
1382         f2fs_put_page(mpage, 1);
1383
1384         f2fs_invalidate_internal_cache(fio.sbi, fio.old_blkaddr);
1385
1386         set_page_dirty(fio.encrypted_page);
1387         if (clear_page_dirty_for_io(fio.encrypted_page))
1388                 dec_page_count(fio.sbi, F2FS_DIRTY_META);
1389
1390         set_page_writeback(fio.encrypted_page);
1391
1392         fio.op = REQ_OP_WRITE;
1393         fio.op_flags = REQ_SYNC;
1394         fio.new_blkaddr = newaddr;
1395         f2fs_submit_page_write(&fio);
1396         if (fio.retry) {
1397                 err = -EAGAIN;
1398                 if (PageWriteback(fio.encrypted_page))
1399                         end_page_writeback(fio.encrypted_page);
1400                 goto put_page_out;
1401         }
1402
1403         f2fs_update_iostat(fio.sbi, NULL, FS_GC_DATA_IO, F2FS_BLKSIZE);
1404
1405         f2fs_update_data_blkaddr(&dn, newaddr);
1406         set_inode_flag(inode, FI_APPEND_WRITE);
1407 put_page_out:
1408         f2fs_put_page(fio.encrypted_page, 1);
1409 recover_block:
1410         if (err)
1411                 f2fs_do_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
1412                                                         true, true, true);
1413 up_out:
1414         if (lfs_mode)
1415                 f2fs_up_write(&fio.sbi->io_order_lock);
1416 put_out:
1417         f2fs_put_dnode(&dn);
1418 out:
1419         f2fs_put_page(page, 1);
1420         return err;
1421 }
1422
1423 static int move_data_page(struct inode *inode, block_t bidx, int gc_type,
1424                                                         unsigned int segno, int off)
1425 {
1426         struct page *page;
1427         int err = 0;
1428
1429         page = f2fs_get_lock_data_page(inode, bidx, true);
1430         if (IS_ERR(page))
1431                 return PTR_ERR(page);
1432
1433         if (!check_valid_map(F2FS_I_SB(inode), segno, off)) {
1434                 err = -ENOENT;
1435                 goto out;
1436         }
1437
1438         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1439         if (err)
1440                 goto out;
1441
1442         if (gc_type == BG_GC) {
1443                 if (PageWriteback(page)) {
1444                         err = -EAGAIN;
1445                         goto out;
1446                 }
1447                 set_page_dirty(page);
1448                 set_page_private_gcing(page);
1449         } else {
1450                 struct f2fs_io_info fio = {
1451                         .sbi = F2FS_I_SB(inode),
1452                         .ino = inode->i_ino,
1453                         .type = DATA,
1454                         .temp = COLD,
1455                         .op = REQ_OP_WRITE,
1456                         .op_flags = REQ_SYNC,
1457                         .old_blkaddr = NULL_ADDR,
1458                         .page = page,
1459                         .encrypted_page = NULL,
1460                         .need_lock = LOCK_REQ,
1461                         .io_type = FS_GC_DATA_IO,
1462                 };
1463                 bool is_dirty = PageDirty(page);
1464
1465 retry:
1466                 f2fs_wait_on_page_writeback(page, DATA, true, true);
1467
1468                 set_page_dirty(page);
1469                 if (clear_page_dirty_for_io(page)) {
1470                         inode_dec_dirty_pages(inode);
1471                         f2fs_remove_dirty_inode(inode);
1472                 }
1473
1474                 set_page_private_gcing(page);
1475
1476                 err = f2fs_do_write_data_page(&fio);
1477                 if (err) {
1478                         clear_page_private_gcing(page);
1479                         if (err == -ENOMEM) {
1480                                 memalloc_retry_wait(GFP_NOFS);
1481                                 goto retry;
1482                         }
1483                         if (is_dirty)
1484                                 set_page_dirty(page);
1485                 }
1486         }
1487 out:
1488         f2fs_put_page(page, 1);
1489         return err;
1490 }
1491
1492 /*
1493  * This function tries to get parent node of victim data block, and identifies
1494  * data block validity. If the block is valid, copy that with cold status and
1495  * modify parent node.
1496  * If the parent node is not valid or the data block address is different,
1497  * the victim data block is ignored.
1498  */
1499 static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1500                 struct gc_inode_list *gc_list, unsigned int segno, int gc_type,
1501                 bool force_migrate)
1502 {
1503         struct super_block *sb = sbi->sb;
1504         struct f2fs_summary *entry;
1505         block_t start_addr;
1506         int off;
1507         int phase = 0;
1508         int submitted = 0;
1509         unsigned int usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
1510
1511         start_addr = START_BLOCK(sbi, segno);
1512
1513 next_step:
1514         entry = sum;
1515
1516         for (off = 0; off < usable_blks_in_seg; off++, entry++) {
1517                 struct page *data_page;
1518                 struct inode *inode;
1519                 struct node_info dni; /* dnode info for the data */
1520                 unsigned int ofs_in_node, nofs;
1521                 block_t start_bidx;
1522                 nid_t nid = le32_to_cpu(entry->nid);
1523
1524                 /*
1525                  * stop BG_GC if there is not enough free sections.
1526                  * Or, stop GC if the segment becomes fully valid caused by
1527                  * race condition along with SSR block allocation.
1528                  */
1529                 if ((gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) ||
1530                         (!force_migrate && get_valid_blocks(sbi, segno, true) ==
1531                                                         CAP_BLKS_PER_SEC(sbi)))
1532                         return submitted;
1533
1534                 if (check_valid_map(sbi, segno, off) == 0)
1535                         continue;
1536
1537                 if (phase == 0) {
1538                         f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
1539                                                         META_NAT, true);
1540                         continue;
1541                 }
1542
1543                 if (phase == 1) {
1544                         f2fs_ra_node_page(sbi, nid);
1545                         continue;
1546                 }
1547
1548                 /* Get an inode by ino with checking validity */
1549                 if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
1550                         continue;
1551
1552                 if (phase == 2) {
1553                         f2fs_ra_node_page(sbi, dni.ino);
1554                         continue;
1555                 }
1556
1557                 ofs_in_node = le16_to_cpu(entry->ofs_in_node);
1558
1559                 if (phase == 3) {
1560                         int err;
1561
1562                         inode = f2fs_iget(sb, dni.ino);
1563                         if (IS_ERR(inode) || is_bad_inode(inode) ||
1564                                         special_file(inode->i_mode))
1565                                 continue;
1566
1567                         err = f2fs_gc_pinned_control(inode, gc_type, segno);
1568                         if (err == -EAGAIN) {
1569                                 iput(inode);
1570                                 return submitted;
1571                         }
1572
1573                         if (!f2fs_down_write_trylock(
1574                                 &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
1575                                 iput(inode);
1576                                 sbi->skipped_gc_rwsem++;
1577                                 continue;
1578                         }
1579
1580                         start_bidx = f2fs_start_bidx_of_node(nofs, inode) +
1581                                                                 ofs_in_node;
1582
1583                         if (f2fs_post_read_required(inode)) {
1584                                 int err = ra_data_block(inode, start_bidx);
1585
1586                                 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1587                                 if (err) {
1588                                         iput(inode);
1589                                         continue;
1590                                 }
1591                                 add_gc_inode(gc_list, inode);
1592                                 continue;
1593                         }
1594
1595                         data_page = f2fs_get_read_data_page(inode, start_bidx,
1596                                                         REQ_RAHEAD, true, NULL);
1597                         f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
1598                         if (IS_ERR(data_page)) {
1599                                 iput(inode);
1600                                 continue;
1601                         }
1602
1603                         f2fs_put_page(data_page, 0);
1604                         add_gc_inode(gc_list, inode);
1605                         continue;
1606                 }
1607
1608                 /* phase 4 */
1609                 inode = find_gc_inode(gc_list, dni.ino);
1610                 if (inode) {
1611                         struct f2fs_inode_info *fi = F2FS_I(inode);
1612                         bool locked = false;
1613                         int err;
1614
1615                         if (S_ISREG(inode->i_mode)) {
1616                                 if (!f2fs_down_write_trylock(&fi->i_gc_rwsem[WRITE])) {
1617                                         sbi->skipped_gc_rwsem++;
1618                                         continue;
1619                                 }
1620                                 if (!f2fs_down_write_trylock(
1621                                                 &fi->i_gc_rwsem[READ])) {
1622                                         sbi->skipped_gc_rwsem++;
1623                                         f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
1624                                         continue;
1625                                 }
1626                                 locked = true;
1627
1628                                 /* wait for all inflight aio data */
1629                                 inode_dio_wait(inode);
1630                         }
1631
1632                         start_bidx = f2fs_start_bidx_of_node(nofs, inode)
1633                                                                 + ofs_in_node;
1634                         if (f2fs_post_read_required(inode))
1635                                 err = move_data_block(inode, start_bidx,
1636                                                         gc_type, segno, off);
1637                         else
1638                                 err = move_data_page(inode, start_bidx, gc_type,
1639                                                                 segno, off);
1640
1641                         if (!err && (gc_type == FG_GC ||
1642                                         f2fs_post_read_required(inode)))
1643                                 submitted++;
1644
1645                         if (locked) {
1646                                 f2fs_up_write(&fi->i_gc_rwsem[READ]);
1647                                 f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
1648                         }
1649
1650                         stat_inc_data_blk_count(sbi, 1, gc_type);
1651                 }
1652         }
1653
1654         if (++phase < 5)
1655                 goto next_step;
1656
1657         return submitted;
1658 }
1659
1660 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
1661                         int gc_type)
1662 {
1663         struct sit_info *sit_i = SIT_I(sbi);
1664         int ret;
1665
1666         down_write(&sit_i->sentry_lock);
1667         ret = f2fs_get_victim(sbi, victim, gc_type, NO_CHECK_TYPE, LFS, 0);
1668         up_write(&sit_i->sentry_lock);
1669         return ret;
1670 }
1671
1672 static int do_garbage_collect(struct f2fs_sb_info *sbi,
1673                                 unsigned int start_segno,
1674                                 struct gc_inode_list *gc_list, int gc_type,
1675                                 bool force_migrate)
1676 {
1677         struct page *sum_page;
1678         struct f2fs_summary_block *sum;
1679         struct blk_plug plug;
1680         unsigned int segno = start_segno;
1681         unsigned int end_segno = start_segno + sbi->segs_per_sec;
1682         int seg_freed = 0, migrated = 0;
1683         unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
1684                                                 SUM_TYPE_DATA : SUM_TYPE_NODE;
1685         unsigned char data_type = (type == SUM_TYPE_DATA) ? DATA : NODE;
1686         int submitted = 0;
1687
1688         if (__is_large_section(sbi))
1689                 end_segno = rounddown(end_segno, sbi->segs_per_sec);
1690
1691         /*
1692          * zone-capacity can be less than zone-size in zoned devices,
1693          * resulting in less than expected usable segments in the zone,
1694          * calculate the end segno in the zone which can be garbage collected
1695          */
1696         if (f2fs_sb_has_blkzoned(sbi))
1697                 end_segno -= sbi->segs_per_sec -
1698                                         f2fs_usable_segs_in_sec(sbi, segno);
1699
1700         sanity_check_seg_type(sbi, get_seg_entry(sbi, segno)->type);
1701
1702         /* readahead multi ssa blocks those have contiguous address */
1703         if (__is_large_section(sbi))
1704                 f2fs_ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
1705                                         end_segno - segno, META_SSA, true);
1706
1707         /* reference all summary page */
1708         while (segno < end_segno) {
1709                 sum_page = f2fs_get_sum_page(sbi, segno++);
1710                 if (IS_ERR(sum_page)) {
1711                         int err = PTR_ERR(sum_page);
1712
1713                         end_segno = segno - 1;
1714                         for (segno = start_segno; segno < end_segno; segno++) {
1715                                 sum_page = find_get_page(META_MAPPING(sbi),
1716                                                 GET_SUM_BLOCK(sbi, segno));
1717                                 f2fs_put_page(sum_page, 0);
1718                                 f2fs_put_page(sum_page, 0);
1719                         }
1720                         return err;
1721                 }
1722                 unlock_page(sum_page);
1723         }
1724
1725         blk_start_plug(&plug);
1726
1727         for (segno = start_segno; segno < end_segno; segno++) {
1728
1729                 /* find segment summary of victim */
1730                 sum_page = find_get_page(META_MAPPING(sbi),
1731                                         GET_SUM_BLOCK(sbi, segno));
1732                 f2fs_put_page(sum_page, 0);
1733
1734                 if (get_valid_blocks(sbi, segno, false) == 0)
1735                         goto freed;
1736                 if (gc_type == BG_GC && __is_large_section(sbi) &&
1737                                 migrated >= sbi->migration_granularity)
1738                         goto skip;
1739                 if (!PageUptodate(sum_page) || unlikely(f2fs_cp_error(sbi)))
1740                         goto skip;
1741
1742                 sum = page_address(sum_page);
1743                 if (type != GET_SUM_TYPE((&sum->footer))) {
1744                         f2fs_err(sbi, "Inconsistent segment (%u) type [%d, %d] in SSA and SIT",
1745                                  segno, type, GET_SUM_TYPE((&sum->footer)));
1746                         set_sbi_flag(sbi, SBI_NEED_FSCK);
1747                         f2fs_stop_checkpoint(sbi, false,
1748                                 STOP_CP_REASON_CORRUPTED_SUMMARY);
1749                         goto skip;
1750                 }
1751
1752                 /*
1753                  * this is to avoid deadlock:
1754                  * - lock_page(sum_page)         - f2fs_replace_block
1755                  *  - check_valid_map()            - down_write(sentry_lock)
1756                  *   - down_read(sentry_lock)     - change_curseg()
1757                  *                                  - lock_page(sum_page)
1758                  */
1759                 if (type == SUM_TYPE_NODE)
1760                         submitted += gc_node_segment(sbi, sum->entries, segno,
1761                                                                 gc_type);
1762                 else
1763                         submitted += gc_data_segment(sbi, sum->entries, gc_list,
1764                                                         segno, gc_type,
1765                                                         force_migrate);
1766
1767                 stat_inc_gc_seg_count(sbi, data_type, gc_type);
1768                 sbi->gc_reclaimed_segs[sbi->gc_mode]++;
1769                 migrated++;
1770
1771 freed:
1772                 if (gc_type == FG_GC &&
1773                                 get_valid_blocks(sbi, segno, false) == 0)
1774                         seg_freed++;
1775
1776                 if (__is_large_section(sbi))
1777                         sbi->next_victim_seg[gc_type] =
1778                                 (segno + 1 < end_segno) ? segno + 1 : NULL_SEGNO;
1779 skip:
1780                 f2fs_put_page(sum_page, 0);
1781         }
1782
1783         if (submitted)
1784                 f2fs_submit_merged_write(sbi, data_type);
1785
1786         blk_finish_plug(&plug);
1787
1788         if (migrated)
1789                 stat_inc_gc_sec_count(sbi, data_type, gc_type);
1790
1791         return seg_freed;
1792 }
1793
1794 int f2fs_gc(struct f2fs_sb_info *sbi, struct f2fs_gc_control *gc_control)
1795 {
1796         int gc_type = gc_control->init_gc_type;
1797         unsigned int segno = gc_control->victim_segno;
1798         int sec_freed = 0, seg_freed = 0, total_freed = 0, total_sec_freed = 0;
1799         int ret = 0;
1800         struct cp_control cpc;
1801         struct gc_inode_list gc_list = {
1802                 .ilist = LIST_HEAD_INIT(gc_list.ilist),
1803                 .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
1804         };
1805         unsigned int skipped_round = 0, round = 0;
1806         unsigned int upper_secs;
1807
1808         trace_f2fs_gc_begin(sbi->sb, gc_type, gc_control->no_bg_gc,
1809                                 gc_control->nr_free_secs,
1810                                 get_pages(sbi, F2FS_DIRTY_NODES),
1811                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1812                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1813                                 free_sections(sbi),
1814                                 free_segments(sbi),
1815                                 reserved_segments(sbi),
1816                                 prefree_segments(sbi));
1817
1818         cpc.reason = __get_cp_reason(sbi);
1819 gc_more:
1820         sbi->skipped_gc_rwsem = 0;
1821         if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
1822                 ret = -EINVAL;
1823                 goto stop;
1824         }
1825         if (unlikely(f2fs_cp_error(sbi))) {
1826                 ret = -EIO;
1827                 goto stop;
1828         }
1829
1830         /* Let's run FG_GC, if we don't have enough space. */
1831         if (has_not_enough_free_secs(sbi, 0, 0)) {
1832                 gc_type = FG_GC;
1833
1834                 /*
1835                  * For example, if there are many prefree_segments below given
1836                  * threshold, we can make them free by checkpoint. Then, we
1837                  * secure free segments which doesn't need fggc any more.
1838                  */
1839                 if (prefree_segments(sbi)) {
1840                         stat_inc_cp_call_count(sbi, TOTAL_CALL);
1841                         ret = f2fs_write_checkpoint(sbi, &cpc);
1842                         if (ret)
1843                                 goto stop;
1844                         /* Reset due to checkpoint */
1845                         sec_freed = 0;
1846                 }
1847         }
1848
1849         /* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
1850         if (gc_type == BG_GC && gc_control->no_bg_gc) {
1851                 ret = -EINVAL;
1852                 goto stop;
1853         }
1854 retry:
1855         ret = __get_victim(sbi, &segno, gc_type);
1856         if (ret) {
1857                 /* allow to search victim from sections has pinned data */
1858                 if (ret == -ENODATA && gc_type == FG_GC &&
1859                                 f2fs_pinned_section_exists(DIRTY_I(sbi))) {
1860                         f2fs_unpin_all_sections(sbi, false);
1861                         goto retry;
1862                 }
1863                 goto stop;
1864         }
1865
1866         seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type,
1867                                 gc_control->should_migrate_blocks);
1868         if (seg_freed < 0)
1869                 goto stop;
1870
1871         total_freed += seg_freed;
1872
1873         if (seg_freed == f2fs_usable_segs_in_sec(sbi, segno)) {
1874                 sec_freed++;
1875                 total_sec_freed++;
1876         }
1877
1878         if (gc_type == FG_GC) {
1879                 sbi->cur_victim_sec = NULL_SEGNO;
1880
1881                 if (has_enough_free_secs(sbi, sec_freed, 0)) {
1882                         if (!gc_control->no_bg_gc &&
1883                             total_sec_freed < gc_control->nr_free_secs)
1884                                 goto go_gc_more;
1885                         goto stop;
1886                 }
1887                 if (sbi->skipped_gc_rwsem)
1888                         skipped_round++;
1889                 round++;
1890                 if (skipped_round > MAX_SKIP_GC_COUNT &&
1891                                 skipped_round * 2 >= round) {
1892                         stat_inc_cp_call_count(sbi, TOTAL_CALL);
1893                         ret = f2fs_write_checkpoint(sbi, &cpc);
1894                         goto stop;
1895                 }
1896         } else if (has_enough_free_secs(sbi, 0, 0)) {
1897                 goto stop;
1898         }
1899
1900         __get_secs_required(sbi, NULL, &upper_secs, NULL);
1901
1902         /*
1903          * Write checkpoint to reclaim prefree segments.
1904          * We need more three extra sections for writer's data/node/dentry.
1905          */
1906         if (free_sections(sbi) <= upper_secs + NR_GC_CHECKPOINT_SECS &&
1907                                 prefree_segments(sbi)) {
1908                 stat_inc_cp_call_count(sbi, TOTAL_CALL);
1909                 ret = f2fs_write_checkpoint(sbi, &cpc);
1910                 if (ret)
1911                         goto stop;
1912                 /* Reset due to checkpoint */
1913                 sec_freed = 0;
1914         }
1915 go_gc_more:
1916         segno = NULL_SEGNO;
1917         goto gc_more;
1918
1919 stop:
1920         SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
1921         SIT_I(sbi)->last_victim[FLUSH_DEVICE] = gc_control->victim_segno;
1922
1923         if (gc_type == FG_GC)
1924                 f2fs_unpin_all_sections(sbi, true);
1925
1926         trace_f2fs_gc_end(sbi->sb, ret, total_freed, total_sec_freed,
1927                                 get_pages(sbi, F2FS_DIRTY_NODES),
1928                                 get_pages(sbi, F2FS_DIRTY_DENTS),
1929                                 get_pages(sbi, F2FS_DIRTY_IMETA),
1930                                 free_sections(sbi),
1931                                 free_segments(sbi),
1932                                 reserved_segments(sbi),
1933                                 prefree_segments(sbi));
1934
1935         f2fs_up_write(&sbi->gc_lock);
1936
1937         put_gc_inode(&gc_list);
1938
1939         if (gc_control->err_gc_skipped && !ret)
1940                 ret = total_sec_freed ? 0 : -EAGAIN;
1941         return ret;
1942 }
1943
1944 int __init f2fs_create_garbage_collection_cache(void)
1945 {
1946         victim_entry_slab = f2fs_kmem_cache_create("f2fs_victim_entry",
1947                                         sizeof(struct victim_entry));
1948         return victim_entry_slab ? 0 : -ENOMEM;
1949 }
1950
1951 void f2fs_destroy_garbage_collection_cache(void)
1952 {
1953         kmem_cache_destroy(victim_entry_slab);
1954 }
1955
1956 static void init_atgc_management(struct f2fs_sb_info *sbi)
1957 {
1958         struct atgc_management *am = &sbi->am;
1959
1960         if (test_opt(sbi, ATGC) &&
1961                 SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
1962                 am->atgc_enabled = true;
1963
1964         am->root = RB_ROOT_CACHED;
1965         INIT_LIST_HEAD(&am->victim_list);
1966         am->victim_count = 0;
1967
1968         am->candidate_ratio = DEF_GC_THREAD_CANDIDATE_RATIO;
1969         am->max_candidate_count = DEF_GC_THREAD_MAX_CANDIDATE_COUNT;
1970         am->age_weight = DEF_GC_THREAD_AGE_WEIGHT;
1971         am->age_threshold = DEF_GC_THREAD_AGE_THRESHOLD;
1972 }
1973
1974 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
1975 {
1976         sbi->gc_pin_file_threshold = DEF_GC_FAILED_PINNED_FILES;
1977
1978         /* give warm/cold data area from slower device */
1979         if (f2fs_is_multi_device(sbi) && !__is_large_section(sbi))
1980                 SIT_I(sbi)->last_victim[ALLOC_NEXT] =
1981                                 GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
1982
1983         init_atgc_management(sbi);
1984 }
1985
1986 static int free_segment_range(struct f2fs_sb_info *sbi,
1987                                 unsigned int secs, bool gc_only)
1988 {
1989         unsigned int segno, next_inuse, start, end;
1990         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
1991         int gc_mode, gc_type;
1992         int err = 0;
1993         int type;
1994
1995         /* Force block allocation for GC */
1996         MAIN_SECS(sbi) -= secs;
1997         start = MAIN_SECS(sbi) * sbi->segs_per_sec;
1998         end = MAIN_SEGS(sbi) - 1;
1999
2000         mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2001         for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
2002                 if (SIT_I(sbi)->last_victim[gc_mode] >= start)
2003                         SIT_I(sbi)->last_victim[gc_mode] = 0;
2004
2005         for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
2006                 if (sbi->next_victim_seg[gc_type] >= start)
2007                         sbi->next_victim_seg[gc_type] = NULL_SEGNO;
2008         mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2009
2010         /* Move out cursegs from the target range */
2011         for (type = CURSEG_HOT_DATA; type < NR_CURSEG_PERSIST_TYPE; type++)
2012                 f2fs_allocate_segment_for_resize(sbi, type, start, end);
2013
2014         /* do GC to move out valid blocks in the range */
2015         for (segno = start; segno <= end; segno += sbi->segs_per_sec) {
2016                 struct gc_inode_list gc_list = {
2017                         .ilist = LIST_HEAD_INIT(gc_list.ilist),
2018                         .iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
2019                 };
2020
2021                 do_garbage_collect(sbi, segno, &gc_list, FG_GC, true);
2022                 put_gc_inode(&gc_list);
2023
2024                 if (!gc_only && get_valid_blocks(sbi, segno, true)) {
2025                         err = -EAGAIN;
2026                         goto out;
2027                 }
2028                 if (fatal_signal_pending(current)) {
2029                         err = -ERESTARTSYS;
2030                         goto out;
2031                 }
2032         }
2033         if (gc_only)
2034                 goto out;
2035
2036         stat_inc_cp_call_count(sbi, TOTAL_CALL);
2037         err = f2fs_write_checkpoint(sbi, &cpc);
2038         if (err)
2039                 goto out;
2040
2041         next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
2042         if (next_inuse <= end) {
2043                 f2fs_err(sbi, "segno %u should be free but still inuse!",
2044                          next_inuse);
2045                 f2fs_bug_on(sbi, 1);
2046         }
2047 out:
2048         MAIN_SECS(sbi) += secs;
2049         return err;
2050 }
2051
2052 static void update_sb_metadata(struct f2fs_sb_info *sbi, int secs)
2053 {
2054         struct f2fs_super_block *raw_sb = F2FS_RAW_SUPER(sbi);
2055         int section_count;
2056         int segment_count;
2057         int segment_count_main;
2058         long long block_count;
2059         int segs = secs * sbi->segs_per_sec;
2060
2061         f2fs_down_write(&sbi->sb_lock);
2062
2063         section_count = le32_to_cpu(raw_sb->section_count);
2064         segment_count = le32_to_cpu(raw_sb->segment_count);
2065         segment_count_main = le32_to_cpu(raw_sb->segment_count_main);
2066         block_count = le64_to_cpu(raw_sb->block_count);
2067
2068         raw_sb->section_count = cpu_to_le32(section_count + secs);
2069         raw_sb->segment_count = cpu_to_le32(segment_count + segs);
2070         raw_sb->segment_count_main = cpu_to_le32(segment_count_main + segs);
2071         raw_sb->block_count = cpu_to_le64(block_count +
2072                                         (long long)segs * sbi->blocks_per_seg);
2073         if (f2fs_is_multi_device(sbi)) {
2074                 int last_dev = sbi->s_ndevs - 1;
2075                 int dev_segs =
2076                         le32_to_cpu(raw_sb->devs[last_dev].total_segments);
2077
2078                 raw_sb->devs[last_dev].total_segments =
2079                                                 cpu_to_le32(dev_segs + segs);
2080         }
2081
2082         f2fs_up_write(&sbi->sb_lock);
2083 }
2084
2085 static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
2086 {
2087         int segs = secs * sbi->segs_per_sec;
2088         long long blks = (long long)segs * sbi->blocks_per_seg;
2089         long long user_block_count =
2090                                 le64_to_cpu(F2FS_CKPT(sbi)->user_block_count);
2091
2092         SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
2093         MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
2094         MAIN_SECS(sbi) += secs;
2095         FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
2096         FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
2097         F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
2098
2099         if (f2fs_is_multi_device(sbi)) {
2100                 int last_dev = sbi->s_ndevs - 1;
2101
2102                 FDEV(last_dev).total_segments =
2103                                 (int)FDEV(last_dev).total_segments + segs;
2104                 FDEV(last_dev).end_blk =
2105                                 (long long)FDEV(last_dev).end_blk + blks;
2106 #ifdef CONFIG_BLK_DEV_ZONED
2107                 FDEV(last_dev).nr_blkz = FDEV(last_dev).nr_blkz +
2108                                         div_u64(blks, sbi->blocks_per_blkz);
2109 #endif
2110         }
2111 }
2112
2113 int f2fs_resize_fs(struct file *filp, __u64 block_count)
2114 {
2115         struct f2fs_sb_info *sbi = F2FS_I_SB(file_inode(filp));
2116         __u64 old_block_count, shrunk_blocks;
2117         struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
2118         unsigned int secs;
2119         int err = 0;
2120         __u32 rem;
2121
2122         old_block_count = le64_to_cpu(F2FS_RAW_SUPER(sbi)->block_count);
2123         if (block_count > old_block_count)
2124                 return -EINVAL;
2125
2126         if (f2fs_is_multi_device(sbi)) {
2127                 int last_dev = sbi->s_ndevs - 1;
2128                 __u64 last_segs = FDEV(last_dev).total_segments;
2129
2130                 if (block_count + last_segs * sbi->blocks_per_seg <=
2131                                                                 old_block_count)
2132                         return -EINVAL;
2133         }
2134
2135         /* new fs size should align to section size */
2136         div_u64_rem(block_count, BLKS_PER_SEC(sbi), &rem);
2137         if (rem)
2138                 return -EINVAL;
2139
2140         if (block_count == old_block_count)
2141                 return 0;
2142
2143         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2144                 f2fs_err(sbi, "Should run fsck to repair first.");
2145                 return -EFSCORRUPTED;
2146         }
2147
2148         if (test_opt(sbi, DISABLE_CHECKPOINT)) {
2149                 f2fs_err(sbi, "Checkpoint should be enabled.");
2150                 return -EINVAL;
2151         }
2152
2153         err = mnt_want_write_file(filp);
2154         if (err)
2155                 return err;
2156
2157         shrunk_blocks = old_block_count - block_count;
2158         secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
2159
2160         /* stop other GC */
2161         if (!f2fs_down_write_trylock(&sbi->gc_lock)) {
2162                 err = -EAGAIN;
2163                 goto out_drop_write;
2164         }
2165
2166         /* stop CP to protect MAIN_SEC in free_segment_range */
2167         f2fs_lock_op(sbi);
2168
2169         spin_lock(&sbi->stat_lock);
2170         if (shrunk_blocks + valid_user_blocks(sbi) +
2171                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2172                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2173                 err = -ENOSPC;
2174         spin_unlock(&sbi->stat_lock);
2175
2176         if (err)
2177                 goto out_unlock;
2178
2179         err = free_segment_range(sbi, secs, true);
2180
2181 out_unlock:
2182         f2fs_unlock_op(sbi);
2183         f2fs_up_write(&sbi->gc_lock);
2184 out_drop_write:
2185         mnt_drop_write_file(filp);
2186         if (err)
2187                 return err;
2188
2189         err = freeze_super(sbi->sb, FREEZE_HOLDER_USERSPACE);
2190         if (err)
2191                 return err;
2192
2193         if (f2fs_readonly(sbi->sb)) {
2194                 err = thaw_super(sbi->sb, FREEZE_HOLDER_USERSPACE);
2195                 if (err)
2196                         return err;
2197                 return -EROFS;
2198         }
2199
2200         f2fs_down_write(&sbi->gc_lock);
2201         f2fs_down_write(&sbi->cp_global_sem);
2202
2203         spin_lock(&sbi->stat_lock);
2204         if (shrunk_blocks + valid_user_blocks(sbi) +
2205                 sbi->current_reserved_blocks + sbi->unusable_block_count +
2206                 F2FS_OPTION(sbi).root_reserved_blocks > sbi->user_block_count)
2207                 err = -ENOSPC;
2208         else
2209                 sbi->user_block_count -= shrunk_blocks;
2210         spin_unlock(&sbi->stat_lock);
2211         if (err)
2212                 goto out_err;
2213
2214         set_sbi_flag(sbi, SBI_IS_RESIZEFS);
2215         err = free_segment_range(sbi, secs, false);
2216         if (err)
2217                 goto recover_out;
2218
2219         update_sb_metadata(sbi, -secs);
2220
2221         err = f2fs_commit_super(sbi, false);
2222         if (err) {
2223                 update_sb_metadata(sbi, secs);
2224                 goto recover_out;
2225         }
2226
2227         update_fs_metadata(sbi, -secs);
2228         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2229         set_sbi_flag(sbi, SBI_IS_DIRTY);
2230
2231         stat_inc_cp_call_count(sbi, TOTAL_CALL);
2232         err = f2fs_write_checkpoint(sbi, &cpc);
2233         if (err) {
2234                 update_fs_metadata(sbi, secs);
2235                 update_sb_metadata(sbi, secs);
2236                 f2fs_commit_super(sbi, false);
2237         }
2238 recover_out:
2239         clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
2240         if (err) {
2241                 set_sbi_flag(sbi, SBI_NEED_FSCK);
2242                 f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
2243
2244                 spin_lock(&sbi->stat_lock);
2245                 sbi->user_block_count += shrunk_blocks;
2246                 spin_unlock(&sbi->stat_lock);
2247         }
2248 out_err:
2249         f2fs_up_write(&sbi->cp_global_sem);
2250         f2fs_up_write(&sbi->gc_lock);
2251         thaw_super(sbi->sb, FREEZE_HOLDER_USERSPACE);
2252         return err;
2253 }